Operator Theory: Advances and Applications Volume 218 Founded in 1979 by Israel Gohberg
Editors: Joseph A. Ball (Blacksburg, VA, USA) Harry Dym (Rehovot, Israel) Marinus A. Kaashoek (Amsterdam, The Netherlands) Heinz Langer (Vienna, Austria) Christiane Tretter (Bern, Switzerland) Associate Editors: Vadim Adamyan (Odessa, Ukraine) Albrecht Bรถttcher (Chemnitz, Germany) B. Malcolm Brown (Cardiff, UK) Raul Curto (Iowa, IA, USA) Fritz Gesztesy (Columbia, MO, USA) Pavel Kurasov (Lund, Sweden) Leonid E. Lerer (Haifa, Israel) Vern Paulsen (Houston, TX, USA) Mihai Putinar (Santa Barbara, CA, USA) Leiba Rodman (Williamsburg, VA, USA) Ilya M. Spitkovsky (Williamsburg, VA, USA)
Honorary and Advisory Editorial Board: Lewis A. Coburn (Buffalo, NY, USA) Ciprian Foias (College Station, TX, USA) J.William Helton (San Diego, CA, USA) Thomas Kailath (Stanford, CA, USA) Peter Lancaster (Calgary, Canada) Peter D. Lax (New York, NY, USA) Donald Sarason (Berkeley, CA, USA) Bernd Silbermann (Chemnitz, Germany) Harold Widom (Santa Cruz, CA, USA)
Harry Dym Marinus A. Kaashoek Peter Lancaster Heinz Langer Leonid Lerer Editors
A Panorama of Modern Operator Theory and Related Topics The Israel Gohberg Memorial Volume
Editors Harry Dym Department of Mathematics Weizmann Institute of Science Rehovot, Israel
Marinus A. Kaashoek Department of Mathematics VU University Amsterdam Amsterdam, The Netherlands
Peter Lancaster Department of Mathematics & Statistics University of Calgary Calgary, Alberta, Canada
Heinz Langer Institute of Analysis and Scientific Computing Vienna University of Technology Vienna, Austria
Leonid Lerer Department of Mathematics Technion Israel Institute of Technology Haifa, Israel
ISBN 978-3-0348-0220-8 e-ISBN 978-3-0348-0221-5 DOI 10.1007/978-3-0348-0221-5 Springer Basel Dordrecht Heidelberg London New York Library of Congress Control Number: 2012930973 Mathematics Subject Classification (2010): 47-XX, 46-XX, 32-XX, 15-XX, 93-XX ยฉ Springer Basel AG 2012 This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting, reproduction on microfilms or in other ways, and storage in data banks. For any kind of use, permission of the copyright owner must be obtained. Printed on acid-free paper
Springer Basel AG is part of Springer Science + Business Media (www.birkhauser-science.com)
Contents Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ix
D. Alpay and H. Attia An Interpolation Problem for Functions with Values in a Commutative Ring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
J. Arazy and H. Upmeier Minimal and Maximal Invariant Spaces of Holomorphic Functions on Bounded Symmetric Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
D.Z. Arov and H. Dym B-regular ๐ฝ-inner Matrix-valued Functions . . . . . . . . . . . . . . . . . . . . . . . . . .
51
J.A. Ball and V. Bolotnikov Canonical Transfer-function Realization for Schur-Agler-class Functions of the Polydisk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
H. Bart, T. Ehrhardt and B. Silbermann Spectral Regularity of Banach Algebras and Non-commutative Gelfand Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 W. Bauer and N. Vasilevski Banach Algebras of Commuting Toeplitz Operators on the Unit Ball via the Quasi-hyperbolic Group . . . . . . . . . . . . . . . . . . . . . . . 155 H. Bercovici, R.G. Douglas and C. Foias Canonical Models for Bi-isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 A. Bยจ ottcher, S. Grudsky, D. Huybrechs and A. Iserles First-order Trace Formulae for the Iterates of the FoxโLi Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 A. Brudnyi, L. Rodman and I.M. Spitkovsky Factorization Versus Invertibility of Matrix Functions on Compact Abelian Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
vi
Contents
P. Dewilde Banded Matrices, Banded Inverses and Polynomial Representations for Semi-separable Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 V.K. Dubovoy, B. Fritzsche and B. Kirstein Description of Helson-Szegห o Measures in Terms of the Schur Parameter Sequences of Associated Schur Functions . . . . . . . . . .
269
Y. Eidelman and I. Haimovici Divide and Conquer Method for Eigenstructure of Quasiseparable Matrices Using Zeroes of Rational Matrix Functions . . . . . . . . . . . . . . . . . 299 R.L. Ellis An Identity Satis๏ฌed by Certain Orthogonal Vector-valued Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
329
I. Feldman and N. Krupnik Invertibility of Certain Fredholm Operators . . . . . . . . . . . . . . . . . . . . . . . . . 345 F.L. Hernยด andez, Y. Raynaud and E.M. Semenov Bernstein Widths and Super Strictly Singular Inclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359 M.A. Kaashoek and F. van Schagen On Inversion of Certain Structured Linear Transformations Related to Block Toeplitz Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377 S. Koyuncu and H.J. Woerdeman The Inverse of a Two-level Positive De๏ฌnite Toeplitz Operator Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387 P. Lancaster and I. Zaballa Parametrizing Structure Preserving Transformations of Matrix Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
403
P. Lancaster and I. Zaballa A Review of Canonical Forms for Selfadjoint Matrix Polynomials . . . . 425 H. Langer, A. Markus and V. Matsaev Linearization, Factorization, and the Spectral Compression of a Self-adjoint Analytic Operator Function Under the Condition (VM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
445
J. Leiterer An Estimate for the Splitting of Holomorphic Cocycles. One Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
465
Contents
vii
L. Lerer and A.C.M. Ran The Discrete Algebraic Riccati Equation and Hermitian Block Toeplitz Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495 F. Oggier and A. Bruckstein On Cyclic and Nearly Cyclic Multiagent Interactions in the Plane . . . 513 ยจ J. Ostensson and D.R. Yafaev A Trace Formula for Di๏ฌerential Operators of Arbitrary Order . . . . . . 541 L. Rodman Jordan Structures and Lattices of Invariant Subspaces of Real Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571 J. Rovnyak and L.A. Sakhnovich Pseudospectral Functions for Canonical Di๏ฌerential Systems. II . . . . .
583
D. Xia Operator Identities for Subnormal Tuples of Operators . . . . . . . . . . . . . . 613
Israel Gohberg 1928โ2009
Preface Israel Gohberg, the founder of the Birkhยจauser OT series (and the journal Integral Equations and Operator Theory) passed away on October 12, 2009, a few months after his eighty-๏ฌrst birthday. This brought to a close more than sixty years of intense mathematical activity, some 25 in the former Soviet Union, where he was born, and the remaining 35 or so while living in Israel, but with many extended visits to collaborators in Europe (primarily Germany and The Netherlands), the US and Canada. A recent Birkhยจauser volume, Israel Gohberg and Friends, provides extensive documentation of Israelโs life, activities and interests. It includes biographical material, a list of his papers (458), books (25) and students (40) up to March 2008, as well as testimonials from his many collaborators, students and friends some of which are reprinted from the proceedings of the conferences that celebrated his sixtieth, seventieth and eightieth birthday, in Calgary, Groningen and Williamsburg, respectively. The journal Linear Algebra and its Applications also printed six short articles on Israel Gohberg by six of his collaborators in volume 433 (2010), 877โ 892. Obituaries appeared in various other journals, including the IEEE Control Systems Magazine (volume 30, December 2010). In spite of deteriorating health in his later years, Israel maintained a full schedule of activities and continued to generate a steady stream of ideas, plans for the future, articles and books and continued to exhibit a positive optimistic outlook. Even when he was hospitalized in the Intensive Care Unit of Meir Hospital in Kfar Saba, Israel, and forced to cancel the ๏ฌrst of a planned sequence of meetings in Germany, he expressed the hope of being able to participate in the second. Unfortunately, this was not to be. This volume is a collection of articles written to honor his memory by a number of his former students, collaborators, colleagues and friends on subjects that intersect with his many interests. A list of the key words that appear in the titles gives a partial indication of their scope: interpolation, transfer function, realization theory, Banach algebras, Toeplitz operators, factorization, (numerical) algorithms, semi-separable matrices and operators, Fredholm operators, block Toeplitz matrices, inversion of structured matrices, Riccati equations, trace formulas, matrix polynomials, linearization, analytic operator functions, Jordan structures, canonical di๏ฌerential systems, multivariable operator theory. December 2011
Harry Dym, Marinus A. Kaashoek, Peter Lancaster, Heinz Langer, Leonid Lerer
Operator Theory: Advances and Applications, Vol. 218, 1โ17 c 2012 Springer Basel AG โ
An Interpolation Problem for Functions with Values in a Commutative Ring Daniel Alpay and Haim Attia Dedicated to the memory of Israel Gohberg
Abstract. It was recently shown that the theory of linear stochastic systems can be viewed as a particular case of the theory of linear systems on a certain commutative ring of power series in a countable number of variables. In the present work we study an interpolation problem in this setting. A key tool is the principle of permanence of algebraic identities. Mathematics Subject Classi๏ฌcation (2000). 60H40, 93C05. Keywords. White noise space, stochastic distributions, linear systems on rings.
1. Introduction There are numerous connections between classical interpolation problems and optimal control and the theory of linear systems; see for instance [10, 1]. In these settings, the coe๏ฌcient space is the complex ๏ฌeld โ, or in the case of real systems, the real numbers โ. Furthermore, already from its inception, linear system theory was considered when the coe๏ฌcient space is a general (commutative) ๏ฌeld, or more generally a commutative ring; see [22, 25]. In [8, 6] a new approach to the theory of linear stochastic systems was developed, in which the coe๏ฌcient space is now a certain commutative ring โ (see Section 3 below). The results from [22, 25] do not seem to be directly applicable to this theory, and the speci๏ฌc properties of โ played a key role in the arguments in [8, 6]. We set โ0 = {0, 1, 2, 3, . . .} . The purpose of this work is to discuss the counterparts of classical interpolation problems in this new setting. To set the problems into perspective, we begin this D. Alpay thanks the Earl Katz family for endowing the chair which supported his research.
2
D. Alpay and H. Attia
introduction with a short discussion of the deterministic case. In the classical theory of linear systems, input-output relations of the form ๐ฆ๐ =
๐ โ
โ๐โ๐ ๐ข๐ ,
๐ = 0, 1, . . . ,
(1.1)
๐=0
where (๐ข๐ )๐โโ0 is called the input sequence, (๐ฆ๐ )๐โโ0 is the output sequence, and (โ๐ )๐โโ0 is the impulse response, play an important role. The sequence (โ๐ )๐โโ0 may consist of matrices (of common dimensions), and then the input and output sequences consist of vectors of appropriate dimensions. Similarly state space equations ๐ฅ๐+1 = ๐ด๐ฅ๐ + ๐ต๐ข๐ , ๐ฆ๐ = ๐ถ๐ฅ๐ + ๐ท๐ข๐ ,
๐ = 0, 1, . . .
play an important role. Here ๐ฅ๐ denotes the state at time ๐, and ๐ด, ๐ต, ๐ถ and ๐ท are matrices with complex entries. The transfer function of the system is โ(๐) =
โ โ
โ๐ ๐๐ ,
๐=0
in the case (1.1), and
โ(๐) = ๐ท + ๐๐ถ(๐ผ โ ๐๐ด)โ1 ๐ต
in the case of state space equations, when assuming the state at ๐ = 0 to be equal to 0. Classical interpolation problems bear various applications to the corresponding linear systems. See for instance [10, Part VI], [21]. To ๏ฌx ideas, we consider the case of bitangential interpolation problem for matrix-valued functions analytic and contractive in the open unit disk (Schur functions), and will even consider only the Nevanlinna-Pick interpolation problem in the sequel to keep notation simple, but it will be clear that the discussion extends to more general cases. Recall (see [10, ยง18.5 p. 409]) that the bitangential interpolation problem may be de๏ฌned in terms of a septuple of matrices ๐ = (๐ถ+ , ๐ถโ , ๐ด๐ , ๐ด๐ , ๐ต+ , ๐ตโ , ฮ) by the conditions โ Res๐=๐0 (๐๐ผ โ ๐ด๐ )โ1 ๐ต+ ๐(๐) = โ๐ตโ , ๐0 โ๐ป
โ
โ
Res๐=๐0 ๐(๐)๐ถโ (๐๐ผ โ ๐ด๐ )โ1 = ๐ถ+ ,
๐0 โ๐ป
Res๐=๐0 (๐๐ผ โ ๐ด๐ )โ1 ๐ต+ ๐(๐)๐ถโ (๐๐ผ โ ๐ด๐ )โ1 = ฮ,
๐0 โ๐ป
where ๐ด๐ and ๐ด๐ have their spectra in the open unit disk, where (๐ด๐ , ๐ต+ ) is a full range pair (that is, controllable) and where (๐ถโ , ๐ด๐ ) is a null kernel pair (that is, observable). We send the reader to [10] for the de๏ฌnitions. Moreover, ฮ satis๏ฌes the compatibility condition ฮ๐ด๐ โ ๐ด๐ ฮ = ๐ต+ ๐ถ+ + ๐ตโ ๐ถโ .
Interpolation Problem in a Commutative Ring Let ๐ be the matrix (see [10, p. 458]) ( ๐1 ๐ = ฮ
) ฮโ , ๐2
3
(1.2)
where ๐1 and ๐2 are the solutions of the Stein equations โ โ ๐ถโ โ ๐ถ+ ๐ถ+ , ๐1 โ ๐ดโ๐ ๐1 ๐ด๐ = ๐ถโ
โ โ ๐2 โ ๐ด๐ ๐2 ๐ดโ๐ = ๐ต+ ๐ต+ โ ๐ตโ ๐ตโ .
Furthermore, and assuming the unknown function ๐ to be โ๐ร๐ -valued, we set ( ) 0 ๐ผ๐ ๐ฝ= . 0 โ๐ผ๐ When ๐ is strictly positive, the solutions of the interpolation problem are given in terms of a linear fractional transformation based on a ๐ฝ-inner rational function ฮ built from the septuple ๐ via the formula (see [10, (18.5.6) p. 410]) )( ) ( โ (๐๐ผ โ ๐ด๐ )โ1 0 ๐ถ+ โ๐ต+ ฮ(๐) = ๐ผ + (๐ โ ๐0 ) โ 0 (๐ผ โ ๐๐ดโ๐ )โ1 ๐ถโ ๐ตโ (1.3) ( ) โ โ (๐ผ โ ๐0 ๐ดโ๐ )โ1 ๐ถ+ โ(๐ผ โ ๐0 ๐ดโ๐ )โ1 ๐ถโ , ร ๐ โ1 (๐ด๐ โ ๐0 ๐ผ)โ1 ๐ต+ (๐ด๐ โ ๐0 ๐ผ)โ1 ๐ตโ where ๐0 is ๏ฌxed on the unit circle and such that the various inverses exist in the above formula. An important fact is that the entries of ๐1 and ๐2 are rational functions of the entries of the matrices of ๐. The same holds for the entries of ๐ since ฮ belongs to the septuple ๐. As a consequence, there exists a rational function ๐ (๐), built from ๐ and such that the entries of ฮ are polynomials in ๐, with coe๏ฌcients which are themselves polynomials in the entries of the matrices of ๐ with coe๏ฌcients in the set of integers โค. This fact will allow us in the sequel to use the principle of permanence of identities (see [9, p. 456]), to extend interpolation problems to a more general setting. Allowing in (1.1) the input sequence (๐ข๐ )๐โโ0 to consist of random variables has been considered for a long time. On the other hand, allowing also the impulse response of the system to carry some randomness seems much more di๏ฌcult to tackle. Recently a new approach to the theory of linear stochastic systems was developed using Hidaโs white noise space theory [18], [19], [23], and Kondratievโs spaces of stochastic test functions and distributions [20]. In this approach, see [3], [5], [6], the complex numbers are replaced by random variables in the white noise space, or more generally, by stochastic distributions in the Kondratiev space, and the product of complex numbers is replaced by the Wick product. For instance, (1.1) now becomes ๐ โ ๐ฆ๐ = โ๐โ๐ โข๐ข๐ , ๐ = 0, 1, . . . (1.4) ๐=0
where the various quantities are in the white noise space, or more generally in the Kondratievโs space of stochastic distributions, and โข denotes the Wick product.
4
D. Alpay and H. Attia
An important role in this theory is played by a ring โ of power series in countably many variables with coe๏ฌcients in โ. This ring is endowed with a topology, which is that of the dual of a countably normed nuclear space. See Sections 2 and 3. Let us denote by โ r(๐ง) = ๐๐ผ ๐ง ๐ผ , (1.5) ๐ผโโ
an element of โ, where โ denotes the set of sequences (๐ผ1 , ๐ผ2 , . . .), whose entries are in โ0 , and for which ๐ผ๐ โ= 0 for only a ๏ฌnite number of indices ๐, and where we have used the multi-index notation ๐ง ๐ผ = ๐ง1๐ผ1 ๐ง2๐ผ2 โ
โ
โ
๐ผ โ โ.
The ring โ has the following properties: (P1 ) If r โ โ and r(0, 0, 0, . . . ) โ= 0, then r has an inverse in โ. (P2 ) If r โ โ๐ร๐ is such that r(0, 0, 0, . . . ) = 0๐ร๐ and if ๐ is a function of one complex variable, analytic in a neighborhood of the origin, with Taylor expansion โ โ ๐๐ ๐๐ , ๐ (๐) = ๐=0
then, the series def.
๐ (r) =
โ โ
๐๐ r๐
๐=0 ๐ร๐
converges in โ . Furthermore, if ๐ is another function of one complex variable, analytic in a neighborhood of the origin, we have (๐ ๐)(r) = ๐ (r)๐(r). (1.6) โ def. โ ๐ผ โ (P3 ) If r(๐ง) = ๐ผโโ ๐๐ผ ๐ง ๐ผ โ โ, then rโ (๐ง) = ๐ผโโ ๐๐ผ ๐ง โ โ, where ๐๐ผ denotes the conjugate of the complex number ๐๐ผ . โ
Property (P1 ) implies in particular that a matrix A โ โ๐ร๐ is invertible in โ๐ร๐ if and only if det A(0) โ= 0. This fact, together with (๐2 ), allows to de๏ฌne expressions such as โ โ ๐๐ CA๐โ1 B, (1.7) H (๐) = D + ๐C(๐ผ๐ โ ๐A)โ1 B = D + ๐=1
where A, B, C, and D are matrices of appropriate dimensions with entries in โ, and where ๐ is an independent complex variable. As explained in [6] this is the transfer function of some underlying linear systems, and is a rational function with coe๏ฌcients in โ. The purpose of this paper is to explain how to tackle in the present setting counterparts of some classical interpolation problems which appear in the theory of linear systems. To illustrate our strategy, we focus on the Nevanlinna-Pick interpolation problem, but our method works the same for the general bitangential
Interpolation Problem in a Commutative Ring
5
interpolation problem. The computations done in the classical theory (that is, when the coe๏ฌcient space consists of the complex numbers) extend to the case where โ is replaced by the ring โ. In some cases, such as Nevanlinna-Pick interpolation, this can be shown by direct computations. In the general case, one needs to use the principle of permanence of identities, see [9, p. 456]. We note that there are other commutative rings with properties (P1 ), (P2 ) and (P3 ) for which the above analysis is applicable. See [7]. The paper consists of ๏ฌve sections besides the present introduction. In the second section we review Hidaโs white noise space setting and the Kondratiev spaces of stochastic distributions. The de๏ฌnition and main properties of the ring โ are given in Section 3. In Section 4 we de๏ฌne and study analytic functions from an open set of โ with values in โ. In Section 5 we consider the Nevanlinna-Pick interpolation problem. In the last section we discuss the bitangential interpolation problem.
2. The white noise space We here review Hidaโs white noise space theory and the associated spaces of stochastic distributions introduced by Kondratiev. See [18], [19], [20], [23]. Let S (โ) denote the Schwartz space of smooth real-valued rapidly decreasing functions. It is a nuclear space, and by the Bochner-Minlos theorem (see [15, Thยดeor`eme 2, p. 342]), there exists a probability measure on the Borel sets โฌ of the dual space def.
S (โ)โฒ = ฮฉ such that ๐
โ
โฅ๐ โฅ2 L2 (โ) 2
โซ =
ฮฉ
๐๐โจ๐,๐ โฉ ๐๐ (๐),
โ๐ โ S (โ),
(2.8)
where the brackets โจโ
, โ
โฉ denote the duality between S (โ) and S (โ)โฒ . The probability space ๐ฒ = (ฮฉ, โฌ, ๐๐ ) is called the white noise probability space. We will be interested in particular in L2 (๐ฒ), called the white noise space. For ๐ โ S (โ), let ๐๐ denote the random variable ๐๐ (๐) = โจ๐, ๐ โฉ. It follows from (2.8) that โฅ๐ โฅL2 (โ) = โฅ๐๐ โฅL2 (๐ฒ) . Therefore, ๐๐ extends continuously to an isometry from L2 (โ) into L2 (๐ฒ). In the presentation of the Gelfand triple associated to the white noise space which we will use, we follow [20]. The white noise space L2 (๐ฒ) admits a special orthogonal basis (๐ป๐ผ )๐ผโโ , indexed by the set โ and built in terms of the Hermite functions ห โ๐
6
D. Alpay and H. Attia
and of the Hermite polynomials โ๐ de๏ฌned by โ โ ๐ป๐ผ (๐) = โ๐ผ๐ (๐หโ๐ (๐)). ๐=1
We refer to [20, De๏ฌnition 2.2.1 p. 19] for more information. In terms of this basis, any element of L2 (๐ฒ) can be written as โ ๐น = ๐๐ผ ๐ป๐ผ , ๐๐ผ โ โ, (2.9) ๐ผโโ
with
โฅ๐น โฅ2L2 (๐ฒ) =
โ
โฃ๐๐ผ โฃ2 ๐ผ! < โ.
๐ผโโ
There are quite a number of Gelfand triples associated to L2 (๐ฒ). In our previous works [2], [5], and in the present one, we focus on the one consisting of the Kondratiev space ๐1 of stochastic test functions, of L2 (๐ฒ), and of the Kondratiev space ๐โ1 of stochastic distributions. To de๏ฌne these spaces we ๏ฌrst introduce for ๐ โ โ the Hilbert space โ๐ which consists of series of the form (1.5) such that ( )1/2 โ def. 2 2 ๐๐ผ โฅ๐น โฅ๐ = (๐ผ!) โฃ๐๐ผ โฃ (2โ) < โ, (2.10) ๐ผโโ
and the Hilbert spaces โ๐โฒ consisting of sequences (๐๐ผ )๐ผโโ such that ( )1/2 โ โฒ def. 2 โ๐๐ผ โฅ๐น โฅ๐ = โฃ๐๐ผ โฃ (2โ) < โ. ๐ผโโ
We note that, for ๐น โ โ๐โฒ we have lim ๐โฅ๐ โฅ๐น โฅโฒ๐ = โฃ๐(0,0,0,...) โฃ2 , ๐โโ
(2.11)
as can be seen, for instance, by applying the dominated convergence theorem to an appropriate discrete measure. Following the usage inโthe literature, we will also write the elements of โ๐โฒ as formal power series ๐ผโโ ๐๐ผ ๐ป๐ผ . Note that (โ๐ )๐โโ forms a decreasing sequence of Hilbert spaces, with increasing norms, while (โ๐โฒ )๐โโ forms an increasing sequence of Hilbert spaces, with decreasing norms. The spaces ๐1 and ๐โ1 are de๏ฌned by the corresponding projective and inductive limits โ โ โฉ โช ๐1 = โ๐ and ๐โ1 = โ๐โฒ . ๐=1
๐=1
The Wick product is de๏ฌned on the basis (๐ป๐ผ )๐ผโโ by ๐ป๐ผ โข๐ป๐ฝ = ๐ป๐ผ+๐ฝ .
It extends to an everywhere de๏ฌned and continuous map from ๐1 ร ๐1 into itself and from ๐โ1 ร ๐โ1 into itself.1 Let ๐ > 0, and let ๐ > ๐ + 1. Consider โ โ โ๐โฒ and 1 The
continuity properties are proved in [7] for a more general family of rings.
Interpolation Problem in a Commutative Ring
7
๐ข โ โ๐โฒ . Then, Vห ageโs inequality holds: โฅโโข๐ขโฅโฒ๐ โค ๐ด(๐ โ ๐)โฅโโฅโฒ๐โฅ๐ขโฅโฒ๐ , where
( ๐ด(๐ โ ๐) =
โ
(2.12)
)1/2 (๐โ๐)๐ผ
(2โ)
< โ.
(2.13)
๐ผโโ
See [20, Proposition 3.3.2 p. 118]. The following result is a direct consequence of (2.13) and will be useful in the sequel. def.
โฒ Lemma 2.1. Let ๐น โ โ๐โฒ . Then, ๐น โข๐ = ๐น โข โ
โ
โ
โข๐น โ โ๐+2 and
๐ times
โฅ๐น โข๐ โฅโฒ๐+2
)๐ 1 ( โค ๐ด(2)โฅ๐น โฅโฒ๐ , ๐ด(2)
๐ = 1, 2, 3, . . .
(2.14)
Proof. We proceed by induction. The case ๐ = 1 holds since โฅ๐น โฅโฒ๐+2 โค โฅ๐น โฅโฒ๐ ,
for
๐น โ โ๐โฒ .
Assume now that (2.14) holds at rank ๐. Then, from (2.12) we have โฅ๐น โข(๐+1) โฅโฒ๐+2 โค ๐ด(2)โฅ๐น โฅโฒ๐ โฅ๐น โข๐ โฅโฒ๐+2 )๐ 1 ( โค ๐ด(2))โฅ๐น โฅโฒ๐ ๐ด(2)โฅ๐น โฅโฒ๐ ๐ด(2) )๐+1 1 ( = ๐ด(2)โฅ๐น โฅโฒ๐ . ๐ด(2)
โก
3. The ring ๐ฝ The Kondratiev space ๐โ1 endowed with the Wick product is a commutative ring of sequences (๐๐ผ )๐ผโโ , with properties (๐1 ), (๐2 ) and (๐3 ), where in (๐1 ) one understands by evaluation at the origin the ๏ฌrst coe๏ฌcient of the sequence. Using the Hermite transform (de๏ฌned below), we view ๐โ1 as a ring of powers series in in๏ฌnitely many variables. We point out that there are other commutative rings of sequences with properties (๐1 ), (๐2 ) and (๐3 ), and for which a counterpart of inequality (2.12) holds. See [7]. The Hermite transform is de๏ฌned by ๐ผ(๐ป๐ผ ) = ๐ง ๐ผ ,
with ๐ผ โ โ and ๐ง = (๐ง1 , ๐ง2 , . . .) โ โโ .
Then
๐ผ(๐ป๐ผ โ ๐ป๐ฝ ) = ๐ผ(๐ป๐ผ )๐ผ(๐ป๐ฝ ). โ โ It extends for ๐น = ๐ผโโ ๐๐ผ ๐ป๐ผ โ ๐โ1 by the formula ๐ผ(๐น )(๐ง) = ๐ผโโ ๐๐ผ ๐ง ๐ผ , and converges in sets of the form โง โซ โจ โฌ โ ๐พ๐ (๐
) = ๐ง โ โโ : โฃ๐งโฃ๐ผ (2โ)๐๐ผ < ๐
2 , โฉ โญ ๐ผโ=0
8
D. Alpay and H. Attia
where ๐ is such that ๐น โ โ๐โฒ . The Kondratiev space ๐โ1 is closed under the Wick product, and we have ๐ผ(๐น โ ๐บ)(๐ง) = ๐ผ(๐น )(๐ง)๐ผ(๐บ)(๐ง)
and
๐ผ(๐น + ๐บ)(๐ง) = ๐ผ(๐น )(๐ง) + ๐ผ(๐บ)(๐ง)
for any ๐น, ๐บ โ ๐โ1 . Therefore the image of the Kondratiev space ๐โ1 under the Hermite transform is a commutative ring, denoted by def
โ = Im(๐ผ(๐โ1 )). This ring was introduced in [6]. We transpose to it via the Hermite transform the properties of ๐โ1 . We have โ โช โ= ๐ผ(โ๐โฒ ). ๐=1
We de๏ฌne the adjoint Gโ = (h๐ ๐ก ) โ โ๐ร๐ of G = (g๐ก๐ ) โ โ๐ร๐ by h๐ ๐ก (๐ง) = โ g๐ก๐ (๐ง) (๐ก โ {1, . . . , ๐} and ๐ โ {1, . . . , ๐}). Then for A โ โ๐ร๐ and B โ โ๐ร๐ข we have (AB)โ = Bโ Aโ . (3.15) โ โ โ Note that G (0) = G(0) , where G(0) is the usual adjoint matrix. De๏ฌnition 3.1. An element A โ โ๐ร๐ will be said strictly positive, A > 0, if it can be written as A = GGโ , where G โ โ๐ร๐ is invertible. It will be said positive if G is not assumed to be invertible. Lemma 3.2. Let A โ โ๐ร๐ . Then, A is strictly positive if and only if A(0) โ โ๐ร๐ is a strictly positive matrix (in the usual sense). Proof. If A = GGโ with det G(0) โ= 0, then A(0) = G(0)G(0)โ is a strictly positive matrix. Conversely, assume that A โ โ๐ร๐ is such that A(0) > 0. We write A(๐ง) = A(0) + (A(๐ง) โ A(0)) โ โ โ โ = A(0){๐ผ๐ + ( A(0))โ1 (A(๐ง) โ A(0))( A(0))โ1 } A(0). โ โ Note that E(๐ง) = A(0))โ1 (A(๐ง) โ A(0))( A(0))โ1 vanishes at ๐ง = (0, 0, . . .). Property (๐2 ) with 1 1 2 1โ
3 3 ๐ (๐) = 1 + ๐ โ ๐ + ๐ + โ
โ
โ
= (1 + ๐)1/2 , โฃ๐โฃ < 1, 2 2โ
4 2โ
4โ
6 โ implies that A = CCโ , where C = A(0)๐ (E). โก Similarly, if A is positive, then A(0) is also positive, but the converse statement need not hold. Take for instance ๐ = 1 and A(๐ง) = ๐ง1 . Then it is readily seen that one cannot ๏ฌnd r โ โ such that ๐ง1 = rโ (๐ง)r(๐ง). We de๏ฌne the ring of polynomials with coe๏ฌcients in โ by โ[๐]. To avoid confusion between the variable ๐ and the variables ๐ง we introduce the notation I (r) = r(0),
r โ โ.
Interpolation Problem in a Commutative Ring
9
De๏ฌnition 3.3. A rational function with values in โ๐ร๐ is an expression of the form R(๐) = p(๐)(q(๐))โ1 (3.16) ๐ร๐ where p โ (โ[๐]) , and q โ โ[๐] is such that I (q(๐)) โโก 0. Let R โ โ๐ร๐ (๐). Then, I (R) โ โ๐ร๐ (๐), and it is readily seen that (I (R))(๐) = I (R(๐)).
(3.17)
It is proved in [6] that every rational function with values matrices with entries in โ and for which I (q(0)) โ= 0 can be written as (1.7). Example 3.4. Let r โ โ. The function ๐นr (๐) = (๐ โ r)(1 โ ๐rโ )โ1 โ โ(๐) is rational. It is de๏ฌned for ๐ โ โ such that 1 โ= ๐(I (r))โ . The next example of rational function need not be de๏ฌned for ๐ = 0. Example 3.5. Let r โ โ. The function ๐นr (๐) = (๐ โ r)(๐ โ rโ )โ1 โ โ(๐) is rational. It is de๏ฌned for ๐ โ โ such that ๐ โ= (I (r))โ .
4. Analytic functions with values in ๐ฝ It is possible to de๏ฌne analytic functions with values in a locally convex topological vector space (see for instance the discussion in [13, 14, 17, 16]). Here the structure of โ allows us to focus, locally, on the classical de๏ฌnition of Hilbert space-valued functions, as we now explain. Proposition 4.1. Let ฮฉ โ โ be an open set and let f : ฮฉ โ โ be a continuous function. Then, f is locally Hilbert space valued, that is, for every ๐0 โ ฮฉ, there is a compact neighborhood ๐พ of ๐0 and a number ๐0 such that f (๐พ) โ ๐ผ(โ๐โฒ 0 ). Proof. Every ๐0 โ ฮฉ has a neighborhood ๐พ of the form ๐ต๐ฟ = {๐ โ ฮฉ ; โฃ๐0 โ๐โฃ โค ๐ฟ} for some ๐ฟ > 0. Since ๐ต๐ฟ is a compact set and f is continuous, f (๐ต๐ฟ ) is compact in โ, and therefore strongly bounded. See [12, p. 54]. Thus there exists ๐0 โ โ such that f (๐ต๐ฟ ) โ ๐ผ(โ๐โฒ 0 ) and is bounded in the norm of ๐ผ(โ๐โฒ 0 ). See [12, Section 5.3 p. 45]. โก Therefore we can de๏ฌne an analytic function from ฮฉ to โ as a continuous function which locally admits a power expansion with coe๏ฌcients in one of the spaces ๐ผ(โ๐โฒ ). The following example shows that we cannot expect to have a ๏ฌxed ๐ in general. โ ๐ 2 Example 4.2. Let f (๐, ๐ง) = โ ๐=1 ๐ ๐ง๐ . Then f is continuous (as a function of ๐) from โ into โ, but there is no ๐ such that f (๐, ๐ง) (viewed now as a function of ๐ง) belongs to ๐ผ(โ๐โฒ ) for all ๐ โ โ.
10
D. Alpay and H. Attia Indeed, let ๐0 โ โ. We have (โฅf (๐0 )โฅโฒ๐ )2 =
โ โ
โฃ๐๐0 โฃ(2๐)โ๐ = 2โ๐
๐=1
โ โ
๐Re ๐0 โ๐ < โ,
๐=1
for ๐ > Re ๐0 + 1. To show continuity at a point ๐0 โ โ, we take ๐ > โฃ๐0 โฃ + 2, and restrict ๐ to be such that โฃ๐0 โ ๐โฃ < 1. Using the elementary estimate โฃ๐๐ง1 โ ๐๐ง2 โฃ โค โฃ๐ง1 โ ๐ง2 โฃ โ
max โฃ๐๐ง โฃ, ๐งโ[๐ง1 ,๐ง2 ]
(4.1)
for ๐ง1 , ๐ง2 โ โ, we have for ๐ = 2, 3, . . . ๐
โฃ๐ 2 โ ๐
๐0 2
โฃ โค (ln ๐)
โฃ๐ โ ๐0 โฃ โฃ๐0 โฃ+1 ln ๐ ๐ 2 2
and so (โฅf (๐) โ f (๐0 )โฅโฒ๐+2 )2 = 2โ๐โ2
โ โ
๐
โฃ๐ 2 โ ๐
๐0 2
โฃ2 ๐โ๐โ2
๐=2
โค 2โ๐โ2
โ โฃ๐ โ ๐0 โฃ2 โ (ln ๐)2 โฃ๐0 โฃ+1โ๐ ๐ , 4 ๐2 ๐=2
and hence the continuity at the point ๐0 in the norm โฅ โ
โฅโฒ๐+2 , and hence in โ. See in particular [12, p. 57] for the latter. Recall that, in the case of Hilbert space, weak and strong analyticity are equivalent, and can be expressed in terms of power series expansions. The argument uses the uniform boundedness theorem. See [24, Theorem VI.4, p. 189]. We de๏ฌne the evaluation of an โ-valued analytic function at a point r โ โ. We ๏ฌrst introduce โฮฉ = {r โ โ; I (r) โ ฮฉ}, where ฮฉ โ โ is open. Theorem 4.3. Let ฮฉ be an open subset of โ, and let f : ฮฉ โ โ be an analytic function. Let r โ โฮฉ , and let f (๐) =
โ โ
f๐ (๐ โ I (r))๐ ,
(4.2)
๐=0
be the Taylor expansion around I (r) โ ฮฉ, where the f๐ โ โ๐โฒ 0 for some ๐0 โ โ, and where the convergence is in โ๐โฒ 0 . The series f (r) =
โ โ ๐=0
converges in โ๐โฒ for some ๐ > ๐0 .
f๐ (r โ I (r))๐
(4.3)
Interpolation Problem in a Commutative Ring
11
Proof. Let ๐พ be a compact neighborhood of I (r), and let ๐0 โ โ be such that f (๐พ) โ โ๐โฒ 0 . Let furthermore ๐
be the radius of convergence of the โ๐โฒ 0 -valued power series (4.2). In view of (2.11), there exists ๐, which we can assume strictly larger than ๐0 , such that ๐ด(2)โฅr โ I (r)โฅโฒ๐ < ๐
. (4.4) On the other hand, using (2.14), we obtain โฅf๐ (r โ I (r))๐ โฅโฒ๐+2 โค ๐ด(2)โฅf๐ โฅโฒ๐0 โฅ(r โ I (r))๐ โฅโฒ๐+2 ( )๐ โค โฅf๐ โฅโฒ๐0 ๐ด(2)โฅr โ I (r)โฅโฒ๐ . In view of (4.4), the series โ โ ๐=0
converges and so the series
( )๐ โฅf๐ โฅโฒ๐0 ๐ด(2)โฅr โ I (r)โฅโฒ๐ โ โ
f๐ (r โ I (r))๐
๐=0
โฒ converges absolutely in ๐ผ(โ๐+2 ).
โก
The evaluation of f at r is de๏ฌned to be f (r) given by (4.3). Proposition 4.4. We can rewrite the evaluation at r as a Cauchy integral โฎ 1 f (๐) f (r) = ๐๐ 2๐๐ ๐ โr where the integration is along a circle centered at I (r) and of radius ๐ < ๐
and in ฮฉ. Proof. As in Theorem 4.3 we consider a compact neighborhood ๐พ of I (r), and let ๐0 be such that f (๐พ) โ โ๐โฒ 0 . We consider a circle at centered I (r) and which lies inside ๐พ. We have โฎ โฎ 1 f (๐) 1 f (๐) ๐๐ = ๐๐ 2๐๐ ๐ โr 2๐๐ ๐ โ I (r) + I (r) โ r { )๐ } โฎ โ ( โ 1 f (๐) r โ I (r) = ๐๐ 2๐๐ ๐ โ I (r) ๐=0 ๐ โ I (r) โ {โฎ } 1 โ f (๐) ๐ = (r โ I (r)) ๐๐ , 2๐๐ ๐=0 (๐ โ I (r))๐+1 where we have used the estimates as in the proof of Theorem 4.3 and the dominated convergence theorem to justify the interchange of integration and summation. โก Recall that a function ๐ analytic and contractive in the open unit disk is called a Schur function. Furthermore, by the maximum modulus principle, ๐ is in fact strictly contractive in ๐ป, unless it is identically equal to a unitary constant.
12
D. Alpay and H. Attia
We will call a function f analytic from the open unit disk ๐ป into โ a Schur function (notation: f โ ๐โ ) if the function ๐ โ I (f (๐)) is a Schur function. For instance, both 1 + ๐ง1 ๐ง3 and 0.5 + 10๐ง1 โ 3๐ง5 are Schur functions. We now de๏ฌne the analog of the open unit disk by โ๐ป = {r โ โ; I (r) โ ๐ป}, and the analog of strictly contractive Schur functions as the set of analytic functions f : ๐ป โ โ such that the function ๐ โ I (f (๐)) is a strictly contractive Schur function. Theorem 4.5. f โ ๐โ is a strictly contractive Schur function if and only if f : ๐ป โ โ๐ป is analytic. Proof. If f is analytic from ๐ป into โ, and such that the ๐ โ I (f (๐)) is a strictly contractive Schur function, it means by de๏ฌnition that the range of f lies inside โ๐ป . Conversely, let f : ๐ป โ โ be analytic and such that I (f ) is a strictly contractive Schur function. Then for every 0 < ๐ < 1, there exists ๐ โ โ (which may depend on ๐) such that f (โฃ๐โฃ โค ๐) โ ๐ผ(โโฒ ๐ ). We can write f as f (๐) =
โ โ
๐๐ fn ,
๐=0
โโ where โฃ๐โฃ < ๐ and fn โ Now I (f )(๐) = ๐=0 ๐๐ I (fn ) for โฃ๐โฃ < ๐. Since this holds for all ๐ โ (0, 1) the function ๐ โ f (๐) has range inside โ๐ป . โก ๐ผ(โ๐โฒ ).
5. Nevanlinna-Pick Interpolation In this section we solve the following interpolation problem (๐ผ๐ ). Problem 5.1. Given ๐ โ โ and points a1 , . . . , a๐ , b1 , . . . , b๐ โ โ๐ป , ๏ฌnd all Schur functions f with coe๏ฌcients in โ such that f (a๐ ) = b๐ for ๐ = 1, 2, . . . , ๐. The solution of this problem under the assumption that some matrix is strictly positive, is presented in Theorem 5.3 below. We ๏ฌrst give some preliminary arguments, and note that if f is a solution of the interpolation problem 5.1, then ๐ = I (f ) is a solution of the classical interpolation problem ๐ (๐๐ ) = ๐๐ ,
๐ = 1, . . . , ๐,
(5.5)
where we have set ๐๐ = I (a๐ ) and ๐๐ = I (b๐ ),
๐ = 1, . . . , ๐.
This last problem is solved as follows: let ๐ denote the ๐ ร ๐ Hermitian matrix with ๐๐ entry equal to 1 โ ๐๐ ๐โ๐ . (5.6) 1 โ ๐๐ ๐โ๐
Interpolation Problem in a Commutative Ring
13
A necessary and su๏ฌcient condition for (5.5) to have a solution in the family of Schur functions is that ๐ โฅ 0. We will assume ๐ > 0. Set, in the notation of the introduction, ๐ดโ๐ = ๐ด = diag (๐โ1 , ๐โ2 , . . . , ๐โ๐ ), ( ) ( ) ๐ต+ 1 1 โ
โ
โ
1 def โ = โ โ = ๐ถ, ๐ตโ ๐1 ๐2 โ
โ
โ
๐โ๐ ( ) 1 0 ๐ฝ= . 0 โ1
(5.7)
Furthermore, specializing the formula for ฮ given in the introduction with ๐ง0 = 1, or using the formula arising from the theory of reproducing kernel Hilbert spaces (see [11], [1]), set ( ) ๐(๐) ๐(๐) def ฮ(๐) = ๐ผ2 โ (1 โ ๐)๐ถ(๐ผ๐ โ ๐๐ด)โ1 ๐ โ1 (๐ผ โ ๐ด)โโ ๐ถ โ ๐ฝ = . ๐(๐) ๐(๐) We now gather the main properties of the matrix-valued function ฮ relevant to the present work. For proofs, we refer to [1], [10], [11]. Proposition 5.2. The following hold: (a) The matrix-valued function ฮ is ๐ฝ-inner with respect to the open unit disk. (b) ฮ has no poles in ๐ป and ๐(๐)๐ + ๐(๐) โ= 0 for all ๐ โ ๐ป and all ๐ in the closed unit disk. (c) The identity ( ) 1 โ๐๐ ฮ(๐๐ ) = 0, ๐ = 1, . . . , ๐. (5.8) is valid. (d) The linear fractional transformation def.
๐ฮ(๐) (๐(๐)) =
๐(๐)๐(๐) + ๐(๐) ๐(๐)๐(๐) + ๐(๐)
describes the set of all solutions of the problem (5.5) in the family of Schur functions when ๐ varies in the family of Schur functions. To solve the interpolation problem 5.1 we introduce the matrices A, C and P, with entries in โ, built by formulas (5.6) and (5.7), but with a1 , . . . , a๐ , b1 , . . . , b๐ instead of ๐1 , . . . , ๐๐ , ๐1 , . . . , ๐๐ . Note that P > 0 since ๐ > 0, and we can de๏ฌne the โ2ร2 -valued function ฮ as ฮ but with A, C and P instead of ๐ด, ๐ถ and ๐ . We have I (A) = ๐ด, I (C) = ๐ถ, and I (P) = ๐. Furthermore,
I (ฮ(๐)) = ฮ(๐).
(5.9)
Theorem 5.3. Assume P > 0. Then, there is a one-to-one correspondence between the solutions f of the problem 5.1 in ๐โ and the elements g โ ๐โ via the linear fractional transformation f = ๐ฮ (g).
14
D. Alpay and H. Attia
Proof. We ๏ฌrst claim that the matrix-valued function ฮ satis๏ฌes the counterparts of (5.8), that is, ( ) 1 โb๐ ฮ(a๐ ) = 0, ๐ = 1, . . . , ๐. (5.10) This is done using the permanence of algebraic identities. See [9, p. 456] for the latter. Indeed, the matrix-valued function ( โ ) โ โ (1 โ ๐โ ๐๐ ) ฮ(๐) (det(๐ผ๐ โ ๐๐ด))(det(๐ผ๐ โ ๐ด ))(det ๐ ) โ,๐=1,...,๐
is a polynomial in ๐ with coe๏ฌcients which are themselves polynomials in the ๐๐ and the ๐๐ , with entire coe๏ฌcients. Therefore, multiplying both sides of (5.8) by the polynomial function ( โ ) (det(๐ผ๐ โ ๐๐ด))(det(๐ผ๐ โ ๐ดโ ))(det ๐ ) (1 โ ๐โ ๐โ๐ ) โ,๐=1,...,๐
evaluated at ๐ = ๐๐ (๐ = 1, 2, . . . , ๐), and taking the real and imaginary part of the equalities (5.8), we obtain for each ๐ four polynomial identities in the 4๐ real variables Re ๐๐ , Re ๐๐ , Im ๐๐ , Im ๐๐ , with ๐ = 1, . . . , ๐, with entire coe๏ฌcients, namely { } โ ( ) โ โ Re 1 โ๐๐ det (๐ผ โ ๐๐ ๐ด) det (๐ผ โ ๐ด ) det ๐ (1 โ ๐โ ๐๐ )ฮ(๐๐ ) โ,๐=1,...,๐
{ Im
( = 0
} (1 โ ๐โ ๐โ๐ )ฮ(๐๐ )
โ
( ) 1 โ๐๐ det (๐ผ โ ๐๐ ๐ด) det (๐ผ โ ๐ดโ ) det ๐
) 0 ,
โ,๐=1,...,๐
( = 0
) 0 .
It follows (see [9, p. 456]) that these identities hold in any commutative rings, and in particular in โ: { } โ ( ) โ โ Re 1 โb๐ det (๐ผ โ a๐ A) det (๐ผ โ A ) det P (1 โ aโ a๐ )ฮ(a๐ ) โ,๐=1,...,๐
Im
{ ( 1
( = 0
โ
) โb๐ det (๐ผ โ a๐ A) det (๐ผ โ Aโ ) det P
) 0 ,
} (1 โ aโ aโ๐ )ฮ(a๐ )
โ,๐=1,...,๐
( = 0
) 0 .
We now use the fact that we are in the ring โ. Because of the choice of the a๐ , the element โ (1 โ aโ aโ๐ ) det (๐ผ โ a๐ A) det (๐ผ โ Aโ ) โ,๐=1,...,๐
Interpolation Problem in a Commutative Ring
15
is invertible in โ. When furthermore P > 0 we can divide both sides of the above equalities by โ det (๐ผ โ a๐ A) det (๐ผ โ Aโ ) (1 โ aโ aโ๐ ) det P โ,๐=1,...,๐
and obtain (5.10). Let now r โ ๐โ , and let u, v be analytic โ-valued functions de๏ฌned by ( ) ( ) ( ) u(๐) r(๐) a(๐)r(๐) + b(๐) = ฮ(๐) = . v(๐) 1 c(๐)r(๐) + d(๐) Using (5.10) we have that u(a๐ ) = b๐ v(a๐ ),
๐ = 1, . . . , ๐.
To conclude, we need to show that v(a๐ ) is invertible in โ for ๐ = 1, . . . , ๐. But we have I (v(a๐ )) = ๐(๐๐ )I (r)(๐๐ ) + ๐(๐๐ ), ๐ = 1, . . . , ๐. The function ฮ(๐) = I (ฮ(๐)) is ๐ฝ-unitary on the unit circle and has no poles there. Therefore, we have ๐(๐๐ )I (r)(๐๐ ) + ๐(๐๐ ) โ= 0 (see item (b) in Proposition 5.2), and hence v(a๐ ) is invertible in โ. Therefore uvโ1 = ๐ฮ (r) is a solution of the interpolation problem. Assume now that f is a solution. Then, we know from the discussion before the theorem that there exists a Schur function ๐(๐) such that I (f (๐)) = ๐I (ฮ(๐)) (๐(๐)).
(5.11)
De๏ฌne a โ-valued function r by f (๐) = ๐ฮ(๐) (r(๐)). Taking I on both sides of this expression we obtain I (f (๐)) = ๐I (ฮ(๐)) (I (r(๐))). Comparing with (5.11), we obtain I (r(๐)) = ๐(๐), and hence r โ ๐โ .
โก
6. More general interpolation problem The matrix-valued function ฮ de๏ฌned by (1.3) and describing the set of solutions of the bitangential problem satis๏ฌes the conditions โ ( ) Res๐=๐0 (๐๐ผ โ ๐ด๐ )โ1 ๐ต+ ๐ตโ ฮ(๐) = 0 ๐0 โ๐ป
โ
๐0 โ๐ป
Res๐=๐0 ฮ(1/๐โ )โ
( ) ๐ถโ (๐๐ผ โ ๐ด๐ )โ1 = 0. ๐ถ+
See also [4]. As for the Nevanlinna-Pick case, these conditions can be translated into a ๏ฌnite number of polynomial equations with coe๏ฌcients in โค, and the principle of permanence of identities allows to extend these properties in the case of
16
D. Alpay and H. Attia
a commutative ring. On the other hand, we do not know how to extend the third interpolation property, and so the method is not applicable to the most general bitangential interpolation problem. If we restrict the parameter to be a constant contractive matrix, the third condition also translates into a polynomial identity with entire coe๏ฌcients, and the same method can still be used. The case of functions with poles inside the open unit disk, or the degenerate cases, are more di๏ฌcult to treat, and will be considered elsewhere.
References [1] D. Alpay. The Schur algorithm, reproducing kernel spaces and system theory. American Mathematical Society, Providence, RI, 2001. Translated from the 1998 French original by Stephen S. Wilson, Panoramas et Synth`eses. [Panoramas and Syntheses]. [2] D. Alpay, H. Attia, and D. Levanony. Une gยดenยดeralisation de lโintยดegrale stochastique de Wick-Itห o. C. R. Math. Acad. Sci. Paris, 346(5-6):261โ265, 2008. [3] D. Alpay, H. Attia, and D. Levanony. On the characteristics of a class of gaussian processes within the white noise space setting. Stochastic processes and applications, 120:1074โ1104, 2010. [4] D. Alpay, P. Bruinsma, A. Dijksma, and H.S.V. de Snoo. Interpolation problems, extensions of symmetric operators and reproducing kernel spaces II. Integral Equations Operator Theory, 14:465โ500, 1991. [5] D. Alpay and D. Levanony. Linear stochastic systems: a white noise approach. Acta Applicandae Mathematicae, 110:545โ572, 2010. [6] D. Alpay, D. Levanony, and A. Pinhas. Linear stochastic state space theory in the white noise space setting. SIAM Journal of Control and Optimization, 48:5009โ5027, 2010. [7] D. Alpay and Guy Salomon. A family of commutative rings with a Vห ageโs inequality. Arxiv manuscript number http://arxiv.org/abs/1106.5746. [8] Daniel Alpay and David Levanony. Linear stochastic systems: a white noise approach. Acta Appl. Math., 110(2):545โ572, 2010. [9] Michael Artin. Algebra. Prentice Hall Inc., Englewood Cli๏ฌs, NJ, 1991. [10] J. Ball, I. Gohberg, and L. Rodman. Interpolation of rational matrix functions, volume 45 of Operator Theory: Advances and Applications. Birkhยจ auser Verlag, Basel, 1990. [11] H. Dym. ๐ฝ-contractive matrix functions, reproducing kernel Hilbert spaces and interpolation. Published for the Conference Board of the Mathematical Sciences, Washington, DC, 1989. [12] I.M. Gelfand and G.E. Shilov. Generalized functions. Volume 2. Academic Press. [13] A. Grothendieck. Sur certains espaces de fonctions holomorphes. I. J. Reine Angew. Math., 192:35โ64, 1953. [14] A. Grothendieck. Sur certains espaces de fonctions holomorphes. II. J. Reine Angew. Math., 192:78โ95, 1953.
Interpolation Problem in a Commutative Ring
17
[15] I.M. Guelfand and N.Y. Vilenkin. Les distributions. Tome 4: Applications de lโanalyse harmonique. Collection Universitaire de Mathยดematiques, No. 23. Dunod, Paris, 1967. [16] M. Hervยดe. Analytic and plurisubharmonic functions in ๏ฌnite and in๏ฌnite-dimensional spaces. Number 198 in Lecture Notes in Mathematics. Springer-Verlag, 1971. [17] M. Hervยดe. Analyticity in in๏ฌnite-dimensional spaces, volume 10 of de Gruyter Studies in Mathematics. Walter de Gruyter & Co., Berlin, 1989. [18] T. Hida, H. Kuo, J. Pottho๏ฌ, and L. Streit. White noise, volume 253 of Mathematics and its Applications. Kluwer Academic Publishers Group, Dordrecht, 1993. An in๏ฌnite-dimensional calculus. [19] T. Hida and Si Si. Lectures on white noise functionals. World Scienti๏ฌc Publishing Co. Pte. Ltd., Hackensack, NJ, 2008. [20] H. Holden, B. รksendal, J. Ubรธe, and T. Zhang. Stochastic partial di๏ฌerential equations. Probability and its Applications. Birkhยจ auser Boston Inc., Boston, MA, 1996. [21] M. Kaashoek. State space theory of rational matrix functions and applications. In P. Lancaster, editor, Lectures on operator theory and its applications, volume 3 of Fields Institute Monographs, pages 235โ333. American Mathematical Society, 1996. [22] R.E. Kalman, P.L. Falb, and M.A. Arbib. Topics in mathematical system theory. McGraw-Hill Book Co., New York, 1969. [23] Hui-Hsiung Kuo. White noise distribution theory. Probability and Stochastics Series. CRC Press, Boca Raton, FL, 1996. [24] M. Reed and B. Simon. Methods of modern mathematical physics. I. Functional analysis. Academic Press, New York, 1972. [25] E.D. Sontag. Linear systems over commutative rings: A survey. Ricerche di Automatica, 7:1โ34, 1976. Daniel Alpay Department of Mathematics Ben Gurion University of the Negev P.O.B. 653, Beโer Sheva 84105, Israel e-mail:
[email protected] Haim Attia Department of Mathematics Sami Shamoon College of Engineering Beโer Sheva 84100, Israel e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 19โ49 c 2012 Springer Basel AG โ
Minimal and Maximal Invariant Spaces of Holomorphic Functions on Bounded Symmetric Domains Jonathan Arazy and Harald Upmeier Dedicated to the memory of Israel Gohberg
Abstract. Let ๐ท be a Cartan domain in โ๐ and let ๐บ = ๐ด๐ข๐ก(๐ท) be the group of all biholomorphic automorphisms of ๐บ. Consider the projective representation of ๐บ on spaces of holomorphic functions on ๐ท (๐๐ (๐)๐ )(๐ง) := {๐ฝ(๐ โ1 )(๐ง)}๐/๐ ๐ (๐ โ1 (๐ง)),
๐ โ ๐บ,
๐ง โ ๐ท,
where ๐ is the genus of ๐ท and ๐ is in the Wallach set ๐ (๐ท). We identify the minimal and the maximal ๐๐ (๐บ)-invariant Banach spaces of holomorphic functions on ๐ท in a very explicit way: The minimal space ๐๐ is a Besov-1 space, and the maximal space โณ๐ is a weighted ๐ป โ -space. Moreover, with respect to the pairing under the (unique) ๐๐ (๐บ)invariant inner product we have ๐โ๐ = โณ๐ . In the second part of the paper we consider invariant Banach spaces of vector-valued holomorphic functions and obtain analogous descriptions of the unique maximal and minimal space, in particular for the important special case of โconstantโ partitions which arises naturally in connection with nontube type domains. Mathematics Subject Classi๏ฌcation (2000). 46E22, 32M15. Keywords. Banach spaces, holomorphic functions, symmetric domains.
1. Bounded symmetric domains and Jordan triples Let ๐ท be a Cartan domain in โ๐ , i.e., an irreducible bounded symmetric domain in its Harish-Chandra realization. Then ๐ = โ๐ is a hermitian Jordan triple. The main example is the matrix ball ๐ท = ๐ท(๐ผ๐,๐ ) = {๐ง โ ๐๐,๐ (โ), ๐ผ๐ โ ๐ง๐ง โ > 0},
1 โค ๐ โค ๐.
20
J. Arazy and H. Upmeier
with triple product
1 (๐ฅ๐ฆ โ ๐ง + ๐ง๐ฆ โ ๐ฅ). 2 In this paper we only sketch the necessary background on Cartan domains and hermitian Jordan triples, for more details cf. [U2], [L2], [FK2]. Let ๐บ = Aut (๐ท) be the group of holomorphic automorphisms, and let {๐ฅ, ๐ฆ, ๐ง} =
๐พ = {๐ โ ๐บ; ๐(0) = 0} be the maximal compact subgroup. Using Cartanโs linearity theorem, one proves that ๐พ consists of linear maps. Then ๐ท โก ๐บ/๐พ via the evaluation map ๐ โ ๐(0). The symmetries of ๐ท have the form ๐ 0 (๐ง) = โ๐ง and, more generally, ๐ ๐ง = ๐ ๐ 0 ๐ โ1 , where ๐ โ ๐บ satis๏ฌes ๐(0) = ๐ง. For each ๐ โ ๐ท there exists a unique midpoint symmetry ๐๐ ๏ฌxing the geodesic midpoint between 0 and ๐, and satisfying ๐๐ (0) = ๐. Example 1.1. For ๐ท = ๐ท(๐ผ๐,๐ ) we have { ( ) } ๐ผ๐ฝ โ ๐บ = ๐๐ (๐, ๐) = ๐ = โ ๐๐ฟ (โ, ๐ + ๐); ๐๐ฝ๐ = ๐ฝ ๐พ๐ฟ ( ) ๐ผ 0 where ๐ฝ = ๐ . The action is given by Mยจobius transformations 0 โ๐ผ๐ ๐ โ
๐ง = (๐ผ๐ง + ๐ฝ)(๐พ๐ง + ๐ฟ)โ1 and the midpoint symmetry is ๐๐ (๐ง) = (๐ผ๐ โ ๐๐โ )โ1/2 (๐ โ ๐ง)(๐ผ โ ๐โ ๐ง)โ1 (๐ผ โ ๐โ ๐)1/2 . In the 1-dimensional case, this reduces to ๐๐ (๐ง) =
๐โ๐ง . 1 โ ๐โ ๐ง
The group ๐พ โก ๐(๐ (๐) ร ๐ (๐)) acts via ๐(๐ง) = ๐ข๐ง๐ฃ, where ๐ข โ ๐ (๐), ๐ฃ โ ๐ (๐) and det (๐ข) det (๐ฃ) = 1. In general, the domain ๐ท is characterized by the dimension ๐, the genus ๐, and the rank ๐. Moreover we have characteristic multiplicities ๐, ๐ de๏ฌned via ๐ ๐ = ๐(๐ โ 1) + ๐ + ๐๐, 2 ๐ = (๐ โ 1) ๐ + 2 + ๐. In the matrix case ๐ท = ๐ท (๐ผ๐,๐ ) for 1 โค ๐ โค ๐, we have ๐ = ๐ โ
๐, ๐ = ๐ + ๐, ๐ = 2, ๐ = ๐ โ ๐. For any hermitian Jordan triple ๐ and ๐ข, ๐ฃ โ ๐, the Bergman operator ๐ต(๐ข, ๐ฃ) acting on ๐ is de๏ฌned by ๐ต(๐ข, ๐ฃ) ๐ง = ๐ง โ 2{๐ข ๐ฃ โ ๐ง} + ๐๐ข ๐๐ฃ ๐ง
Minimal and Maximal Invariant Spaces
21
where ๐๐ฃ ๐ง = {๐ฃ ๐ง โ ๐ฃ}. It is known that det ๐ต(๐ง, ๐ค) = โ(๐ง, ๐ค)๐ , where โ(๐ง, ๐ค) is a ๐พ-invariant sesqui-holomorphic polynomial determined by โ(๐ง, ๐ง) =
๐ โ
(1 โ ๐ 2๐ (๐ง)),
๐=1
where ๐ ๐ (๐ง) are the singular values of ๐ง. For matrices, we have โ(๐ง, ๐ค) = det (๐ผ โ ๐ง๐คโ ). If ๐ง, ๐ค โ ๐ and ๐ต(๐ง, ๐ค) is invertible, we de๏ฌne the quasi-inverse [L1], [L2] ๐ง ๐ค := ๐ต(๐ง, ๐ค)โ1 (๐ง โ ๐๐ง ๐ค). One can show [L2, p. 25, JP35] that ๐ต(๐ง, ๐ค)โ1 = ๐ต(๐ง ๐ค , โ๐ค). The โtransvectionโ ๐๐ โ ๐บ [L2, Proposition 9.8], de๏ฌned by ๐๐ (๐ง) = ๐ + ๐ต(๐, ๐)1/2 ๐ง โ๐ = ๐๐ (โ๐ง) for all ๐, ๐ง โ ๐ท, satis๏ฌes ๐๐โ1 = ๐โ๐ and ๐๐โฒ (๐ง) = ๐ต(๐, ๐)1/2 ๐ต(๐ง, โ๐)โ1 = ๐ต(๐, ๐)1/2 ๐ต(๐ง โ๐ , ๐).
2. Hilbert spaces of holomorphic functions Let ๐๐(๐ง) be the Lebesgue measure. The unique (up to a constant multiple) ๐บinvariant measure on ๐ท has the form โ(๐ง, ๐ง)โ๐ ๐๐(๐ง). Given a parameter ๐ > ๐ โ 1 we de๏ฌne a probability measure ๐๐๐ (๐ง) = ๐๐ โ
โ(๐ง, ๐ง)๐โ๐ ๐๐(๐ง) on ๐ท, which has the quasi-invariance property 2๐
๐๐๐ (๐(๐ง)) = โฃ๐ฝ(๐, ๐ง) ๐ โฃ ๐๐๐ (๐ง), โ ๐ โ ๐บ.
(2.1)
โฒ
Here ๐ฝ(๐, ๐ง) = det ๐ (๐ง) is the Jacobian of ๐ at ๐ง. (2.1) follows from ๐ต(๐(๐ง), ๐(๐ค)) = ๐ โฒ (๐ง) ๐ต(๐ง, ๐ค) ๐ โฒ (๐ค)โ
โ ๐ โ ๐บ, โ ๐ง, ๐ค โ ๐ท
(2.2)
which yields the quasi-invariance 1
1
โ(๐(๐ง), ๐(๐ค)) = ๐ฝ(๐, ๐ง) ๐ โ(๐ง, ๐ค) ๐ฝ(๐, ๐ค) ๐ , โ ๐ โ ๐บ
(2.3)
of โ. Proposition 2.1. Each ๐ โ ๐ท has a unique โpolar decompositionโ ๐ = ๐๐ โ
๐ with ๐ = ๐(0), ๐ โ ๐พ. Proof. De๏ฌne ๐ = ๐(0) and consider ๐ = ๐๐โ1 โ ๐. Then ๐ โ ๐บ and ๐(0) = 0. Therefore ๐ โ ๐พ and ๐ = ๐๐ โ ๐. โก
22
J. Arazy and H. Upmeier Using Proposition 2.1, we de๏ฌne a cocycle ๐ฝ๐ : ๐บ ร ๐ท โ โ by putting ๐ฝ๐ (๐๐ ๐, ๐ง) := โ(๐, ๐)๐/2 โ(๐๐ง, โ๐)โ๐ ,
(2.4)
using the sesqui-holomorphic branch of โ(๐ง, ๐ค)โ๐ on ๐ท ร ๐ท normalized by โ(0, 0)โ๐ = 1. Then โฃ๐ฝ๐ (๐, ๐ง)โฃ = โฃ๐ฝ (๐, ๐ง)โฃ๐/๐ . The Jacobian of ๐๐ has the form ๐ฝ(๐๐โ1 , ๐ง) = โ(๐, ๐)๐/2 โ(๐ง, ๐)โ๐ . Since ๐๐โ1 = ๐โ๐ , (2.4) implies ๐ฝ๐ (๐๐โ1 , ๐ง) = โ(๐, ๐)๐/2 โ(๐ง, ๐)โ๐ . Now consider the so-called Wallach set ๐ (๐ท) := {๐; (๐ง, ๐ค) โ โ(๐ง, ๐ค)โ๐ positive de๏ฌnite}
(2.5)
and, for ๐ โ ๐ (๐ท), de๏ฌne the reproducing kernel Hilbert space โ๐ = span {โ(โ
, ๐ค)โ๐ ; ๐ค โ ๐ท} with inner product determined by โจโ(โ
, ๐ค)โ๐ , โ(โ
, ๐ง)โ๐ โฉ๐ = โ(๐ง, ๐ค)โ๐ for the reproducing kernel of โ๐ . The corresponding group action (๐๐ (๐)๐ )(๐ง) := ๐ฝ๐ (๐ โ1 , ๐ง) ๐ ((๐ โ1 (๐ง))
(2.6)
on โ๐ acts projectively: ๐๐ (๐1 โ ๐2 ) = ๐(๐1 , ๐2 ) ๐๐ (๐1 ) ๐๐ (๐2 ) for a unimodular cocycle. Then ๐๐ (๐) : โ๐ โ โ๐ acts isometrically, โ ๐ โ ๐บ, because (2.3) implies ๐ฝ๐ (๐, ๐ง) โ(๐(๐ง), ๐(๐ค))โ๐ ๐ฝ๐ (๐, ๐ค) = โ(๐ง, ๐ค)โ๐ . One can show that โ๐ is irreducible for the action ๐๐ of ๐บ. The primary examples are the weighted Bergman space โ๐ = ๐ฟ2๐ (๐ท, ๐๐ ) for ๐ > ๐ โ 1, and the Hardy space โ ๐ = ๐ป 2 (๐, ๐), for ๐ = ๐๐ . Here ๐ is the Shilov ๐ boundary of ๐ท and ๐ is the unique ๐พ-invariant probability measure on ๐. For a deeper analysis of โ๐ , we need the ๏ฌne structure of the polynomial algebra ๐ซ of ๐. For 1 โค ๐ โค ๐ there exist Jordan theoretic minors ๐๐ (๐ง) generalizing the principal ๐ ร ๐-minors for matrices. In particular, ๐๐ = ๐ is the Jordan determinant. The conical polynomials, for any signature m = (๐1 , . . . , ๐๐ ) โ โ๐ satisfying ๐1 โฅ ๐2 โฅ โ
โ
โ
โฅ ๐๐ โฅ 0, are given by ๐m (๐ง) =
๐ โ ๐=1
๐๐ (๐ง)๐๐ โ๐๐+1 , ๐ง โ ๐,
Minimal and Maximal Invariant Spaces
23
where ๐๐+1 := 0. For diagonal matrices (including the rectangular case), we have โก โค ๐ก1 0 $ ๐ โข $ โฅ โ ๐ก2 โข $ โฅ ๐ ๐m โข 0 = ๐ก๐ ๐ = ๐กm . $ โฅ .. $ โฆ โฃ . 0
๐ก๐
๐=1
Denote by ๐ซm the span of {๐m โ ๐; ๐ โ ๐พ}. It is well known [S], [U1], [FK1] that the {๐ซm }mโฅ0 are ๐พ-irreducible and ๐พ-inequivalent, and there is a direct sum decomposition โ โ ๐ซ= ๐ซm . (2.7) mโฅ0
It follows that the {๐ซm }๐โฅ0 are pairwise orthogonal in any ๐พ-invariant inner product on ๐ซ. Consider the Fischer inner product โซ (โ ) โ 2 1 โจ๐, ๐น โฉโฑ = ๐ ๐ (๐ง) ๐น (๐ง) ๐โโฃ๐งโฃ ๐๐(๐ง) = ๐ (๐น )(0) (2.8) ๐ โ๐ง โ๐
on ๐ซ, where ๐น โ (๐ง) := ๐น (๐ง). De๏ฌne ๐พ m (๐ง, ๐ค) as the reproducing kernel for ๐ซm in the Fischer inner product. Then โ ๐โจ๐ง,๐คโฉ = ๐พ m (๐ง, ๐ค). (2.9) mโฅ0
For ๐ โ โ and ๐ง, ๐ค โ ๐ท there is a binomial expansion โ โ(๐ง, ๐ค)โ๐ = (๐)m ๐พ m (๐ง, ๐ค),
(2.10)
mโฅ0
where (๐)m =
๐ โ1 ( ๐ ๐โ โ
๐ + โ โ (๐ โ 1)
๐=1 โ=0
๐ ๐) โ ( ๐) = ๐ โ (๐ โ 1) 2 2 ๐๐ ๐=1
is the multi-variable โPochhammer symbolโ. As a consequence, one obtains a determination of the Wallach set { โ๐ }๐โ1 ( ) ๐ ๐ (๐ท) = {๐ โ โ; (๐)m โฅ 0 โ m} = โช (๐ โ 1) , โ 2 โ=0 2 as a union of a discrete and a continuous part [RV], [W], [LA], [FK1]. The multivariable hypergeometric functions are de๏ฌned as ๐ โ ( ) โ 1 (๐ผ๐ )m ๐ผ1 , . . . , ๐ผ๐ (๐ง, ๐ค) = ๐พ m (๐ง, ๐ค). ๐ ๐น๐ ๐ โ ๐ฝ1 , . . . , ๐ฝ๐ mโฅ0 (๐ฝ๐ )m 1
For example, we have 0 ๐น0 (๐ง, ๐ค) = exp โจ๐ง, ๐คโฉ by (2.9), and (2.10) yields 1 ๐น0
(๐)(๐ง, ๐ค) = โ(๐ง, ๐ค)โ๐ .
24
J. Arazy and H. Upmeier Let ๐ผ0 , ๐ผ1 , . . . , ๐ผ๐ ; ๐ฝ1 , . . . , ๐ฝ๐ > (๐ โ 1) ๐2 . Put ๐พ=
๐ โ
๐ผ๐ โ
0
๐ โ
๐ฝ๐ .
1
By [FK1], the hypergeometric functions have the following asymptotic behaviour, uniformly for ๐ง โ ๐ท: ( ) ๐ ๐ผ ๐พ > (๐ โ 1) =โ ๐+1 ๐น๐ (๐ง, ๐ง) โ โ(๐ง, ๐ง)โ๐พ (2.11) 2 ๐ฝ ( ) ๐ ๐ผ ๐พ < โ(๐ โ 1) =โ ๐+1 ๐น๐ (๐ง, ๐ง) โ 1. (2.12) 2 ๐ฝ Remark 2.1. For the unit ball (๐ = 1) and ๐พ = 0, we have ( ) ( 1 ) ๐ผ (๐ง, ๐ง) โ log . ๐+1 ๐น๐ ๐ฝ 1 โ โฃ๐งโฃ For the exact asymptotics if ๐ง is scalar, see [Y]. For ๐ = 2, exact asymptotics are given in [EZ]. In the following we consider Banach spaces of holomorphic functions on ๐ท which are โinvariantโ under the group action (2.6), with the aim to characterize the (unique) maximal and minimal invariant Banach spaces and describe them via explicit formulas. In later sections this study is extended to the case of vectorvalued holomorphic functions associated with the holomorphic discrete series of ๐บ. In this context our main result concerns symmetric domains which are โnot of tube typeโ. In this paper we only consider parameters ๐ belonging to the Wallach set (2.5). In a separate paper [AU4] we consider the so-called โpole setโ arising from analytic continuation, and show that our results concerning the maximal and minimal invariant space can be generalized to this situation via suitable intertwining operators.
3. Invariant Banach spaces of holomorphic functions In this section we assume that ๐ โ ๐ (๐ท) is a Wallach parameter and consider the weighted group action ๐๐ de๏ฌned in (2.6). For the unweighted action (๐ = 0) and the unit disk, the results of this section have been obtained in [AF], [AFP]. De๏ฌnition 3.1. Let ๐ be a non-trivial Banach space of holomorphic functions on ๐ท. We say that ๐ is ๐๐ (๐บ)-invariant if (i) ๐ โ ๐, ๐ โ ๐บ =โ ๐๐ (๐) ๐ โ ๐ and โฅ๐๐ (๐) ๐ โฅ๐ = โฅ๐ โฅ๐ . (ii) For any ๏ฌnite (complex) Borel measure ๐ on ๐พ, the linear operator (convolution by ๐) โซ (๐๐ ๐ )(๐ง) = maps ๐ continuously into itself.
๐พ
๐ (๐๐ง) ๐๐(๐)
Minimal and Maximal Invariant Spaces
25
(iii) For every ๐ง โ ๐ท, the evaluation functional ๐ โ ๐ฟ๐ง (๐ ) := ๐ (๐ง) is bounded on ๐ (it su๏ฌces to require the continuity of ๐ฟ0 ). Note that condition (ii) holds if ๐พ acts on ๐ strongly continuously via ๐(๐)๐ = ๐ โ ๐ โ1 . Proposition 3.1. ๐ contains the constant function 1 and, normalizing โฅ1โฅ๐ = 1, we have for ๐ โ ๐ โฃ๐ (0)โฃ โค โฅ๐ โฅ๐ /โฅ1โฅ๐ = โฅ๐ โฅ๐ . Proof. Since ๐ท is circular, we have by (ii) and (iii) for all ๐ง โ ๐ท 1 ๐ (0)1 = 2๐
โซ2๐
๐ (๐๐๐ ๐ง)๐๐.
โก
0
Corollary 3.1. For ๐ โ ๐ and ๐ โ ๐ท, we have โฃ๐ (๐)โฃ โค โ(๐, ๐)โ๐/2 โฅ๐ โฅ๐ . Hence convergence in ๐ implies uniform convergence on compact subsets of ๐ท. Proof. Use the formula โฃ(๐๐ (๐๐โ1 ) ๐ )(0)โฃ = โ(๐, ๐)๐/2 โฃ๐ (๐)โฃ โค โฅ๐๐ (๐๐โ1 ) ๐ โฅ๐ = โฅ๐ โฅ๐ . โก โ Corollary 3.2. If ๐ = ๐m โ ๐, then ๐m โ ๐ for all m, and the projections mโฅ0
๐ โ ๐m are continuous.
Proof. In terms of the character ๐m of ๐พ on ๐ซm , we have โซ ๐m (๐ง) = ๐ (๐ โ1 ๐ง) ๐m (๐) ๐๐.
โก
๐พ ๐ 2,
Corollary 3.3. If ๐ > (๐ โ 1) then ๐ซ is dense in ๐ in the topology of uniform convergence on compact subsets. If ๐ = โ๐ 2 , 0 โค โ โค ๐ โ 1, the same holds for โ โ ๐ซโ = ๐ซm . (3.1) ๐โ+1 =0 mโฅ0
Proof. From 1 โ ๐ (Proposition 3.1) it follows by (i) that ๐๐ (๐๐ ) 1 = const โ(โ, ๐)โ๐ โ ๐
for all ๐ โ ๐ท.
m
Applying Corollary 3.2, we obtain (๐)m ๐พ (โ, ๐) โ ๐, hence either (๐)m = 0 or else ๐ซm = span {๐พ m (โ, ๐) : ๐ โ ๐ท} โ ๐. โก Our main goal is to characterize the maximal and minimal invariant spaces. De๏ฌnition 3.2. Let โณ๐ = {๐ โ โ(๐ท); โฅ๐ โฅโณ๐ < โ}, where โฅ๐ โฅโณ๐ = sup โ(๐ง, ๐ง)๐/2 โฃ๐ (๐ง)โฃ = sup โฃ(๐๐ (๐) ๐ )(0)โฃ. ๐งโ๐ท
๐โ๐บ
26
J. Arazy and H. Upmeier
It is easy to see that โณ๐ satis๏ฌes (i), (ii) and (iii) of De๏ฌnition 3.1. Hence using the second expression for the norm, it follows that โณ๐ is ๐๐ (๐บ)-invariant. We remark that taking another base point ๐ โ ๐ท instead of 0 yields the same space with a norm proportional to โฅ โ
โฅโณ๐ . Proposition 3.2. If ๐ is ๐๐ (๐บ)-invariant, then ๐ โ โณ๐ and โฅ๐ โฅโณ๐ โค โฅ๐ โฅ๐ , โ ๐ โ ๐. Proof. In view of Proposition 3.1, we have โฃ(๐๐ (๐) ๐ )(0)โฃ โค โฅ๐ โฅ๐ , โ ๐ โ ๐.
โก
Corollary 3.4. โณ๐ is the unique maximal ๐๐ (๐บ)-invariant space, and it is a weighted ๐ป โ -space, with weight โ(๐ง, ๐ง)๐/2 . We remark that there exist spaces of holomorphic functions on ๐ท satisfying (i), (ii) of De๏ฌnition 3.1, but not (iii). For instance, let ๐ be any holomorphic function on ๐ท (possibly not in โณ๐ ). De๏ฌne ๐๐,๐ (๐ ) to be the space of all functions of the form โ โ ) ( ๐น (๐ง) = ๐๐ ๐๐ (๐๐ ) ๐ (๐ง), ๐=1
where ๐๐ โ ๐บ and
โ โ ๐=1
โฃ๐๐ โฃ < โ. For ๐น โ ๐๐,๐ (๐ ) we de๏ฌne โฅ๐น โฅ๐๐,๐ (๐ ) = inf
โ โ
โฃ๐๐ โฃ,
๐=1
where the in๏ฌmum is taken over all admissible representations of ๐น . Then it is easy to check that ๐๐,๐ (๐ ) is the smallest Banach space of holomorphic functions on ๐ท which contains ๐ and satis๏ฌes (i) and (ii) of De๏ฌnition 3.1. Proposition 3.3. The Banach space ๐๐,๐ (๐ ) satis๏ฌes condition (iii) if and only if ๐ โ โณ๐ . More generally, let ๐ be a Banach space of holomorphic functions on ๐ท satisfying (i) and (ii). Then ๐ satis๏ฌes (iii) if and only if ๐ โ โณ๐ continuously. Proof. If (iii) holds, then ๐๐,๐ (๐ ) (resp., ๐) is a ๐๐ (๐บ)-invariant Banach space and Proposition 3.2 implies ๐ โ โณ๐ (resp., ๐ โ โณ๐ continuously). Conversely, if ๐ โ โณ๐ , then sup โฃ๐๐ (๐) ๐ (0)โฃ < โ ๐โ๐บ
and hence ๐ฟ0 is continuous on ๐๐,๐ (๐ ). Similarly, ๐ โ โณ๐ continuously implies for all ๐ โ ๐ โฃ๐ (0)โฃ โค โฅ๐ โฅโณ๐ โค ๐ โฅ๐ โฅ๐ . Hence ๐ฟ0 is continuous on ๐. By (i), the continuity of ๐ฟ๐ง , ๐ง โ ๐ท, follows. De๏ฌnition 3.3. Let ๐๐ consist of all ๐ โ โ(๐ท) such that โซ ๐๐(๐) โ(๐, ๐)๐/2 โ(๐ง, ๐)โ๐ ๐ (๐ง) = ๐ท
โก
(3.2)
Minimal and Maximal Invariant Spaces
27
for some ๏ฌnite (complex) Borel measure ๐ on ๐ท. De๏ฌne the norm โฅ๐ โฅ๐๐ = inf {โฅ๐โฅ; ๐ satis๏ฌes (3.2)}. Proposition 3.4. We have ๐ โ ๐๐ if and only if โซ ๐ (๐ง) = ๐๐(๐) (๐๐ (๐) 1)(๐ง), โ ๐ง โ ๐ท
(3.3)
๐บ
for some ๏ฌnite Borel measure ๐ on ๐บ. Moreover โฅ๐ โฅ๐ = inf {โฅ๐โฅ; ๐
satis๏ฌes (3.3)}.
Hence ๐๐ is ๐๐ (๐บ)-invariant. The straightforward proof is omitted. Also, the condition โฅ1โฅ๐๐ = 1 is satis๏ฌed. Indeed, if 1=
โซ ๐ท
(3.4)
๐๐ (๐) โ(๐, ๐)๐/2 โ(๐ง, ๐)โ๐ , โ ๐ง โ ๐ท
then for ๐ง = 0 we have โซ โซ โซ ๐/2 ๐/2 1= ๐๐ (๐) โ(๐, ๐) โค ๐โฃ๐โฃ (๐) โ(๐, ๐) โค ๐โฃ๐โฃ (๐) = โฅ๐โฅ ๐ท
๐ท
๐ท
and therefore 1 โค โฅ1โฅ๐๐ = inf {โฅ๐โฅ : ๐ representing measure}. On the other hand, for ๐ = ๐ฟ0 we have โซ ๐ ๐ฟ0 (๐) โ(๐, ๐)๐/2 โ(๐ง, ๐)โ๐ = 1 ๐ท
so โฅ1โฅ๐๐ โค โฅ๐ฟ0 โฅ = 1. Hence (3.4) holds. Proposition 3.5. There is a canonical duality ๐โ๐ โก โณ๐ with respect to the pairing โจ๐, ๐น โฉ๐ of โ๐ . Proof. Let ๐น โ โณ๐ and ๐ โ ๐๐ , with representation (3.3). Since (๐๐ (๐) ๐น )(0) = โจ๐๐ (๐) ๐น, 1โฉ๐ = โจ๐น, ๐๐ (๐ โ1 ) 1โฉ๐ , it follows that
โซ
โจ๐, ๐น โฉ๐ =
๐๐(๐) โจ๐๐ (๐) 1, ๐น โฉ๐ ๐บ
โซ
=
๐๐(๐) โจ1, ๐๐ (๐ โ1 ) ๐น โฉ๐ =
๐บ
Hence
โซ ๐๐(๐) ๐๐ (๐ โ1 ) ๐น (0). ๐บ
โซ โฃโจ๐, ๐น โฉ๐ โฃ โค ๐บ
๐โฃ๐โฃ (๐) โฃ(๐๐ (๐ โ1 )๐น )(0)โฃ โค โฅ๐โฅโฅ๐น โฅโณ๐ .
28
J. Arazy and H. Upmeier
This holds for every representing measure ๐ for ๐ , hence โฃโจ๐, ๐น โฉ๐ โฃ โค โฅ๐ โฅ๐๐ โฅ โฅ๐น โฅโณ๐ .
(3.5)
Thus sup
โฅ๐ โฅ๐๐ โค1
โฃโจ๐, ๐น โฉ๐ โฃ โค โฅ๐น โฅโณ๐ .
The converse inequality follows from โฅ๐น โฅโณ๐ = sup โฃโจ๐น, ๐๐ (๐) 1โฉ๐ โฃ โค ๐โ๐บ
sup
โฅ๐ โฅ๐๐ โค1
โฃโจ๐, ๐น โฉ๐ โฃ.
This means that the operator ๐ : โณ๐ โ ๐โ๐ de๏ฌned by (๐ ๐น )(๐ ) = โจ๐, ๐น โฉ๐ is an isometry. We claim that ๐ is surjective. Indeed, let ฮฆ โ ๐โ๐ and de๏ฌne ๐น (๐ง) = ฮฆ(โ(โ
, ๐ง)โ๐ ). Then ๐น is holomorphic and โ(๐ง, ๐ง)๐/2 โฃ๐น (๐ง)โฃ = โฃฮฆ(โ(๐ง, ๐ง)๐/2 โ(โ
, ๐ง)โ๐ )โฃ = โฃฮฆ(๐๐ (๐๐งโ1 ) 1)โฃ โค โฅฮฆโฅ๐โ๐ . So ๐น โ โณ๐ and โฅ๐น โฅโณ๐ โค โฅฮฆโฅ๐โ๐ . Also, if ๐ โ ๐๐ is represented as in (3.2), then โซ โซ ๐/2 โ๐ ฮฆ(๐ ) = ๐๐(๐) โ(๐, ๐) ฮฆ(โ(โ
, ๐) ) = ๐๐(๐) โ(๐, ๐)๐/2 ๐น (๐) ๐ท ๐ท โซ = ๐๐(๐) โ(๐, ๐)๐/2 โจโ(โ
, ๐)โ๐ , ๐น โฉ๐ = โจ๐, ๐น โฉ๐ . ๐ท
It follows that ๐ (๐น ) = ฮฆ, and so ๐ is a surjective isometry.
โก
De๏ฌnition 3.4. Let ๐๐,๐ be the space of all ๐ โ ๐๐ which are represented with respect to a discrete measure, i.e., ๐ (๐ง) =
โ โ
๐๐ (๐๐ (๐๐ ) 1)(๐ง)
(3.6)
๐=1
with ๐๐ โ ๐บ and ๐๐ โ โ such that
โ ๐
โฃ๐๐ โฃ < โ, with the norm
โฅ๐ โฅ๐๐,๐ = inf
โ โ
โฃ๐๐ โฃ
๐=1
over all representations (3.6). Clearly, ๐๐,๐ is a closed subspace of ๐๐ and โฅ๐ โฅ๐๐ โค โฅ๐ โฅ๐๐,๐ for all ๐ โ ๐๐,๐ . Proposition 3.6. The dual space of ๐๐,๐ is identi๏ฌed isometrically with โณ๐ , with respect to the pairing โจ๐, ๐น โฉ๐ , ๐ โ ๐๐,๐ , ๐น โ โณ๐ . In particular, ๐๐,๐ = ๐๐ with equal norms.
Minimal and Maximal Invariant Spaces
29
Proof. The fact that ๐โ๐,๐ = โณ๐ isometrically is proved as in the proof of Proposition 3.5. This also yields that โฅ๐ โฅ๐๐ = โฅ๐ โฅ๐๐,๐ for all ๐ โ ๐๐,๐ . To prove that ๐๐ = ๐๐,๐ it su๏ฌces (by the Hahn-Banach theorem) to prove that if ฮฆ โ ๐โ๐ vanishes on ๐๐,๐ then it is zero. But this follows from the identi๏ฌcation of ๐โ๐ with โณ๐ . โก Proposition 3.7. If ๐ โ= 0 is ๐๐ (๐บ)-invariant, then ๐๐ โ ๐ and โฅ๐ โฅ๐ โค โฅ๐ โฅ๐๐ , โ ๐ โ ๐๐ . Hence ๐๐ is the unique minimal ๐๐ (๐บ)-invariant Banach space. Proof. Since 1 โ ๐ and โฅ1โฅ๐ = 1 we have โฅ๐๐ (๐) 1โฅ๐ = 1 for all ๐ โ ๐บ. Let โ โ ๐ โ ๐๐ = ๐๐,๐ , and let ๐ = ๐๐ ๐๐ (๐๐ ) 1 be an admissible representation. Then ๐=1
the series converges absolutely โ โ
โฅ๐๐ ๐๐ (๐๐ ) 1โฅ๐ =
๐=1
โ โ
โฃ๐๐ โฃ < โ,
๐=1
and the completeness of ๐๐ guarantees that the convergence is also in the norm of ๐. Therefore ๐ โ ๐ and โฅ๐ โฅ๐ โค
โ
โฅ๐๐ ๐๐ (๐๐ ) 1โฅ๐ =
๐
โ โ
โฃ๐๐ โฃ.
๐=1
This holds for all discrete representations of ๐ , hence โฅ๐ โฅ๐ โค โฅ๐ โฅ๐๐ .
โก
We remark that there exist functions ๐ โ โณ๐ for which the group action ๐ โ ๐๐ (๐) ๐ is not continuous in the norm of โณ๐ . This leads to the following (0)
De๏ฌnition 3.5. Let โณ๐ = {๐ โ โณ๐ ; ๐ โ ๐๐ (๐)๐ is continuous in the โณ๐ norm}. Proposition 3.8. (0) (i) โณ๐ is the maximal ๐๐ (๐บ)-invariant space ๐ for which ๐ โ ๐๐ (๐) ๐ is continuous in norm for all ๐ โ ๐; (0)โ (ii) โณ๐ = ๐๐ with respect to โจโ
, โ
โฉ๐ ; (0) (0)โโ (iii) The canonical embedding of โณ๐ in โณ๐ = โณ๐ is the inclusion map. These statements will not be proved here, since they are not needed for our main problem: to identify ๐๐ via concrete integral formulas (not as a quotient space of the ๏ฌnite Borel measures on ๐ท or ๐บ). De๏ฌnition 3.6. The shift operator ๐๐ผ๐พ on ๐ซ (โdi๏ฌerentiation of order ๐พ โ ๐ผโ) is de๏ฌned by (โ ) โ (๐พ)m ๐๐ผ๐พ ๐m = ๐m . (๐ผ)m mโฅ0
mโฅ0
30
J. Arazy and H. Upmeier In view of the Faraut-Korยดanyi-formula (2.10), we have (๐๐ผ๐พ ๐ )(๐ง) = โจ๐, โ(โ
, ๐ง)โ๐พ โฉ๐ผ ,
and the reproducing kernel identity yields ๐๐ผ๐พ (โ(โ
, ๐ง)โ๐ผ ) = โ(โ
, ๐ง)โ๐พ . It follows that ๐๐ผ๐พ (โ๐ผ ) = โ๐พ . Remark 3.1. If ๐ผ > (๐โ1) ๐2 , then ๐๐ผ๐พ is de๏ฌned on all of ๐ซ. If ๐ผ = then ๐๐ผ๐พ is de๏ฌned only on ๐ซโ (cf. (3.1)).
โ๐ 2 ,
0 โค โ โค ๐โ1,
Our ๏ฌrst main result is Theorem 3.1. Let ๐ โ ๐ (๐ท), ๐ > (๐ โ1) ๐. Choose ๐ฝ โ โ such that ๐ฝ + ๐2 > ๐ โ 1. Then there is a continuous embedding ๐๐๐+๐ฝ (๐๐ ) โ ๐ฟ1๐ (๐ท, ๐๐ฝ+ ๐2 ). Here ๐ฟ1๐ denotes the subspace of holomorphic functions in ๐ฟ1 . Proof. It is enough to consider the โatomsโ: ๐ = โ(๐, ๐)๐/2 โ(โ
, ๐)โ๐ for ๐ โ ๐ท. We have ๐
๐
(๐๐๐+๐ฝ ๐ )(๐ง) = โ(๐, ๐) 2 โจโ(โ
, ๐)โ๐ , โ(โ
, ๐ง)โ(๐+๐ฝ) โฉ๐ = โ(๐, ๐) 2 โ(๐ง, ๐)โ(๐+๐ฝ) . ๐ 2
Using the asymptotic behaviour of 2 ๐น1 , following from the assumption > (๐ โ 1) ๐2 , we obtain โซ $ $ ๐+๐ฝ $2 ๐ $ โฅ๐๐๐+๐ฝ ๐ โฅ๐ฟ1 (๐๐ฝ+ ๐ ) = ๐๐ฝ+๐/2 โ(๐, ๐) 2 $โ(๐ง, ๐)โ 2 $ โ(๐ง, ๐ง)๐ฝ+๐/2โ๐ ๐๐ง 2 ๐ท ( ๐+๐ฝ ๐+๐ฝ ) ๐ ๐ ๐ 2 2 = โ(๐, ๐) 2 2 ๐น1 (๐, ๐) โ โ(๐, ๐) 2 โ(๐, ๐)โ 2 = 1. โก ๐ ๐ฝ+ 2 Theorem 3.1 has the following converse
Theorem 3.2. Let ๐ โ ๐ (๐ท) be arbitrary. Choose ๐ฝ โ โ such that ๐ฝ + ๐2 > ๐ โ 1. Let ๐ be analytic on ๐ท such that ๐๐๐+๐ฝ ๐ โ ๐ฟ1๐ (๐ท, ๐๐ฝ+ ๐2 ). Then ๐ โ ๐๐ and โฅ๐ โฅ๐๐ โค
๐๐+๐ฝ โฅ๐๐๐+๐ฝ ๐ โฅ๐ฟ1 (๐๐ฝ+ ๐ ) . 2 ๐๐ฝ+๐/2
Proof. Consider the ๏ฌnite Borel measure ๐๐(๐) = (๐๐๐+๐ฝ ๐ )(๐) โ(๐, ๐)๐ฝ+๐/2โ๐ ๐๐.
Minimal and Maximal Invariant Spaces
31
Using the self-adjointness of ๐๐๐+๐ฝ with respect to ๐๐+๐ฝ and the reproducing property, we obtain โซ โซ ๐๐(๐) โ(๐, ๐)๐/2 โ(๐ง, ๐)โ๐ = ๐๐ โ(๐, ๐)๐+๐ฝโ๐ (๐๐๐+๐ฝ ๐ )(๐) โ(๐, ๐ง)โ๐ ๐ท ๐ท โซ = ๐๐ โ(๐, ๐)๐+๐ฝโ๐ ๐ (๐) ๐๐๐+๐ฝ (โ(โ
, ๐ง)โ๐ )(๐) โซ๐ท 1 = ๐๐ โ(๐, ๐)๐+๐ฝโ๐ ๐ (๐) โ(๐, ๐ง)โ(๐+๐ฝ) = ๐ (๐ง). ๐ ๐+๐ฝ ๐ท Hence ๐ โ ๐๐ and โฅ๐ โฅ๐๐ โค ๐๐+๐ฝ โฅ๐โฅ = Corollary 3.5. If
๐ 2
๐๐+๐ฝ ๐๐ฝ+๐/2
โฅ๐๐๐+๐ฝ ๐ โฅ๐ฟ1 (๐๐ฝ+ ๐ ) .
โก
2
> ๐ โ 1 we can choose ๐ฝ = 0. Hence ๐๐ = ๐ฟ1๐ (๐ท, ๐ ๐2 ).
Corollary 3.6. For each ๐ โ ๐๐ , the map ๐บ โ ๐ โ ๐๐ (๐) ๐ โ ๐๐ is continuous in the norm of ๐๐ . ๐ Proof. This follows by realizing ๐๐ as ๐๐+๐ฝ (๐ฟ1๐ (๐ท, ๐๐ฝ+ ๐2 )) with ๐ฝ+ ๐2 > ๐โ1.
Corollary 3.7. Let ๐ > (๐ โ 1) ๐ and choose ๐ฝ โ โ such that ๐ฝ + ๐ โ ๐๐ โโ
๐๐๐+๐ฝ
๐โ
๐ฟ1๐ (๐ท,
๐ 2
โก
> ๐ โ 1. Then
๐๐ฝ+ ๐2 ).
(3.7)
Specializing to rank ๐ = 1, we obtain Corollary 3.8. Let ๐ท be the open unit ball of โ๐ . Let ๐ be a holomorphic function on ๐ท and choose ๐ฝ such that ๐ฝ + ๐2 > ๐. Then (3.7) holds.
4. Invariant Banach spaces of vector-valued holomorphic functions We now turn to vector-valued holomorphic function spaces related to the holomorphic discrete series. In this section we describe the unique maximal space, and obtain a su๏ฌcient condition for membership in the unique minimal space. For any ๏ฌxed partition m = (๐1 , . . . , ๐๐ ) consider the m-th Peter-Weyl component ๐ซm (cf. (2.7)) and parameters ๐ โ โ such that the integral โซ ๐พ m (๐ต(๐, ๐) ๐, ๐) ๐โ1 = ๐๐ โ(๐, ๐)๐โ๐ (4.1) ๐,m ๐พ m (๐, ๐) ๐ท is ๏ฌnite. Here ๐ โ ๐ is a maximal tripotent. It is well known that ๐พ m (๐, ๐) =
๐m (๐/๐)m
where ๐m = dim ๐ซm . For example, in the rank 1 case (unit ball) we have ๐พ ๐ (๐ง, ๐ค) =
(๐งโฃ๐ค)๐ ๐!
32
J. Arazy and H. Upmeier
and (๐)m = ๐(๐ + 1) โ
โ
โ
(๐ + ๐ โ 1) =
(๐+๐โ1)! . (๐โ1)! ๐
On the other hand, the space ๐ซ๐ ( ) of homogeneous polynomials on ๐ = โ has dimension ๐+๐โ1 , the number of ๐ solutions of ๐1 + โ
โ
โ
+ ๐๐ = ๐ in integers ๐๐ โฅ 0. Thus, for ๐ = (1, 0, . . . , 0) we obtain ๐๐ (๐ + ๐ โ 1)! 1 1 = = = ๐พ ๐ (๐, ๐). (๐)๐ ๐!(๐ โ 1)! (๐)๐ ๐! Since ๐พ acts irreducibly on ๐ซm it follows that โซ $ โจ๐โฃ๐โฉโฑ = ๐๐,m ๐๐ โ(๐, ๐)๐โ๐ โจ๐ โ ๐ต(๐, ๐)1/2 $ ๐ โ ๐ต(๐, ๐)1/2 โฉโฑ ๐ท
for all ๐, ๐ โ ๐ซm . Here โจ๐โฃ๐โฉโฑ is the Fischer-Fock norm (2.8). Equivalently, โซ ๐(๐) = ๐๐,m ๐๐ โ(๐, ๐)๐โ๐ โ
๐(๐ต(๐, ๐)๐) (4.2) ๐ท
for all ๐ โ ๐ซm and ๐ โ ๐. Let โ๐,m denote the Hilbert space of all holomorphic functions ฮฆ : ๐ท โ ๐ซm , ๐ง โ ฮฆ๐ง (๐) = ฮฆ(๐ง, ๐) such that โฅฮฆโฅ2๐,m
โซ = ๐๐,m
๐ท
๐๐ง โ(๐ง, ๐ง)๐โ๐ โฅฮฆ๐ง โ ๐ต(๐ง, ๐ง)1/2 โฅ2โฑ < +โ.
Here we write ฮฆ๐ง (๐) = ฮฆ(๐ง, ๐) for ๐ง โ ๐ท, ๐ โ ๐, noting that ฮฆ(๐ง, โ) is a polynomial of type m in the ๐-variable. In this notation, ฮฆ๐ง โ ๐ต(๐ง, ๐ง)1/2 (๐) = ฮฆ(๐ง, ๐ต(๐ง, ๐ง)1/2 ๐). Moreover the scalar parameter ๐ is chosen large enough so that ๐๐,m > 0, and so โ๐,m contains all the โconstantโ functions (1 โ ๐)(๐ง, ๐) = ๐(๐) for ๐ โ ๐ซm . It is easily shown that (๐๐,m (๐ โ1 )ฮฆ)(๐ง, ๐) = ๐ฝ๐ (๐, ๐ง) ฮฆ(๐(๐ง), ๐ โฒ(๐ง)๐), with ๐ โ ๐บ, ฮฆ โ โ๐,m , ๐ง โ ๐ท and ๐ โ ๐, de๏ฌnes a unitary (projective) representation of ๐บ on โ๐,m belonging to the so-called holomorphic discrete series of ๐บ [AU3]. Proposition 4.1. For ฮฆ โ โ๐,m we have the reproducing property โซ ฮฆ๐ง (๐) = ๐๐,m ๐๐ โ(๐, ๐)๐โ๐ โ(๐ง, ๐)โ๐ โ
ฮฆ๐ (๐ต(๐, ๐) ๐ต(๐ง, ๐)โ1 ๐). ๐ท
(4.3)
Minimal and Maximal Invariant Spaces
33
Proof. The reproducing formula, for a suitable constant, is proved in [AU3]. Applying the formula to ๐ง = 0, we obtain โซ ฮฆ0 (๐) = ๐๐,m ๐๐ โ(๐, ๐)๐โ๐ ฮฆ๐ (๐ต(๐, ๐) ๐) (4.4) ๐ท
which reduces to (4.2) for ฮฆ = 1 โ ๐, and thus speci๏ฌes the constant.
โก
De๏ฌnition 4.1. Let ๐ โ ๐ช(๐ท, ๐ซm ) be a non-trivial Banach space of ๐ซm -valued holomorphic functions on ๐ท. We say that ๐ is ๐๐,m (๐บ)-invariant if (i) ฮฆ โ ๐, ๐ โ ๐บ =โ ๐๐,m (๐) ฮฆ โ ๐ and โฅ๐๐,m (๐) ฮฆโฅ๐ = โฅฮฆโฅ๐ . (ii) For any ๏ฌnite (complex) Borel measure ๐ on ๐พ, the linear operator (convolution by ๐) โซ (๐๐ ฮฆ)(๐ง, ๐) =
๐๐(๐) ฮฆ(๐๐ง, ๐๐) ๐พ
maps ๐ continuously into itself. (iii) For every ๐ง โ ๐ท, the evaluation map ฮฆ โ (๐ฟ๐ง โ ๐ผ) ฮฆ โ ๐ซm , de๏ฌned by (๐ฟ๐ง โ ๐ผ) ฮฆ(๐) := ฮฆ(๐ง, ๐), is bounded on ๐. As before, condition (ii) is satis๏ฌed if the unweighted representation of ๐พ on ๐ is strongly continuous. Proposition 4.2. Let ๐ โ= (0) be an invariant Banach space in ๐ช(๐ท, ๐ซm ). Then (i) 1 โ ๐ซm โ ๐, and there exists a constant ๐๐ such that for all ฮฆ โ ๐ (ii) โฅฮฆ0 โฅโฑ โค ๐๐ โฅฮฆโฅ๐ . Proof. Put ๐ := ๐1 + โ
โ
โ
+ ๐๐ , and consider the ๏ฌnite Borel measure ๐๐ ๐ ๐ก ๐๐ก/2๐. Since the polynomials in ๐ซm have total degree ๐, we have โซ2๐ โซ2๐ ๐๐ก ๐ ๐ ๐ก ๐๐ก ๐ ๐ ๐ก โ๐ ๐ ๐ก โ๐๐ก โ๐๐ก ๐ ฮฆ(๐ ๐ง, ๐ ๐) = ๐ ๐ ฮฆ(๐โ๐๐ก ๐ง, ๐) 2๐ 2๐ 0
0
โซ2๐ = 0
๐๐ก ฮฆ(๐โ๐๐ก ๐ง, ๐) = ฮฆ(0, ๐) = ฮฆ0 (๐). 2๐
Since the action ๐๐,m on ๐ is isometric and ๐๐ก/2๐ is a probability measure, it follows that โซ2๐ ๐๐ก ๐ ๐ ๐ก 1 โ ฮฆ0 = ๐ ๐๐,m (๐๐๐ก ) ฮฆ (4.5) 2๐ 0
belongs to ๐, and โฅ1 โ ฮฆ0 โฅ๐ โค โฅฮฆโฅ๐ . Choosing ฮฆ โ= 0, there exists ๐ง โ ๐ท such that ฮฆ๐ง (๐) = ฮฆ(๐ง, ๐) โโก 0. Applying a suitable ๐๐,m (๐)-transformation, we may assume ๐ง = 0, i.e., ฮฆ0 (๐) = ฮฆ(0, ๐) โโก 0. Since ๐พ acts irreducibly on ๐ซm , it follows from (4.5) that 1 โ ๐ โ ๐ for all ๐ โ ๐ซm , i.e., 1 โ ๐ซm โ ๐, and there exists ๐๐ > 0 such that โฅ๐โฅโฑ โค ๐๐ โฅ1 โ ๐โฅ๐ . โก
34
J. Arazy and H. Upmeier
De๏ฌnition 4.2. Let โณ๐,m โ ๐ช(๐ท, ๐ซm ) be the Banach space of all holomorphic functions ฮฆ : ๐ท โ ๐ซm such that โฅฮฆโฅโณ๐,m < +โ, where โฅฮฆโฅโณ๐,m = sup โ(๐ง, ๐ง)๐/2 โฅฮฆ๐ง โ ๐ต(๐ง, ๐ง)1/2 โฅโฑ = sup โฅ(๐๐,m (๐) ฮฆ)0 โฅโฑ . ๐งโ๐ท
๐โ๐บ
The requirements (ii) and (iii) in De๏ฌnition 4.1 are easily checked, and hence, with the second expression for the norm, it follows that โณ๐,m is ๐๐,m (๐บ)-invariant. Changing the ๐พ-invariant inner product on ๐ซm , or taking another โbase pointโ ๐ โ ๐ท instead of 0, changes the norm only by a proportionality constant. Theorem 4.1. Let ๐ โ ๐ช(๐ท, ๐ซm ) be a ๐๐,m -invariant Banach space. Then ๐ โ โณ๐,m continuously, i.e., โณ๐,m is the unique maximal invariant space. Proof. Let ฮฆ โ ๐. Then Proposition 4.2 implies โฅ(๐๐,m (๐) ฮฆ)0 โฅโฑ โค ๐๐ โ
โฅ๐๐,m (๐) ฮฆโฅ๐ = ๐๐ โฅฮฆโฅ๐ and hence
sup โฅ(๐๐,m (๐) ฮฆ)0 โฅโฑ โค ๐๐ โ
โฅฮฆโฅ๐ .
๐โ๐บ
The assertion follows.
โก
For ๐ โ ๐ซm and ๐ โ ๐บ, de๏ฌne ๐๐ := ๐๐,m (๐) (1 โ ๐) โ ๐ช(๐ท, ๐ซm ). For ๐ = ๐๐ , we put ๐๐ := ๐๐๐ and obtain (๐๐๐ง )(๐) = โ(๐, ๐)๐/2 โ(๐ง, ๐)โ๐ ๐ (๐ต(๐, ๐)1/2 ๐ต(๐ง, ๐)โ1 ๐).
(4.6)
More generally, ๐๐๐ง๐ ๐ (๐) = โ(๐, ๐)๐/2 โ(๐ง, ๐)โ๐ ๐ (๐ โ1 ๐ต(๐, ๐)1/2 ๐ต(๐ง, ๐)โ1 ๐). Lemma 4.1. For large enough parameters ๐ผ, ๐ฝ, ๐พ we have the change of variables formula โซ ๐๐ค โ(๐ค, ๐ค)๐ผโ๐ โ(๐๐ (๐ฅ), ๐ค)โ๐ฝ โ(๐ค, ๐๐ (๐ฆ))โ๐พ ๐ (๐๐โ1 (๐ค)) = โ(๐, ๐)๐ผโ๐ฝโ๐พ โ(๐ฅ, ๐)๐ฝ โ(๐, ๐ฆ)๐พ โซ โ
๐๐ค โ(๐ค, ๐ค)๐ผโ๐ โ(๐ฅ, ๐ค)โ๐ฝ โ(๐ค, ๐ฆ)โ๐พ โ(๐ค, ๐)๐พโ๐ผ โ(๐, ๐ค)๐ฝโ๐ผ ๐ (๐ค) ๐ท
for all ๐, ๐ฅ, ๐ฆ โ ๐ท and all ๐ โ ๐ฟ1 (๐ท, ๐๐ผ ). Proof. Since ๐๐ค โ(๐ค, ๐ค)โ๐ is ๐บ-invariant, it follows that โซ ๐๐ค โ(๐ค, ๐ค)๐ผโ๐ โ(๐๐ (๐ฅ), ๐ค)โ๐ฝ โ(๐ค, ๐๐ (๐ฆ))โ๐พ ๐ (๐๐โ1 (๐ค)) ๐ท โซ = ๐๐ค โ(๐ค, ๐ค)โ๐ โ(๐๐ (๐ค), ๐๐ (๐ค))๐ผ ๐ท
โ
โ(๐๐ (๐ฅ), ๐๐ (๐ค))โ๐ฝ โ(๐๐ (๐ค), ๐๐ (๐ฆ))โ๐พ ๐ (๐ค).
Minimal and Maximal Invariant Spaces
35
Now the assertion follows from โ(๐๐ (๐ค), ๐๐ (๐ค))๐ผ โ(๐๐ (๐ฅ), ๐๐ (๐ค))โ๐ฝ โ(๐๐ (๐ค), ๐๐ (๐ฆ))โ๐พ [ ]๐ผ = โ(๐, ๐) โ(๐ค, ๐)โ1 โ(๐ค, ๐ค) โ(๐, ๐ค)โ1 [ ]โ๐ฝ โ
โ(๐, ๐) โ(๐ฅ, ๐)โ1 โ(๐ฅ, ๐ค) โ(๐, ๐ค)โ1 [ ]โ๐พ โ
โ(๐, ๐) โ(๐ค, ๐)โ1 โ(๐ค, ๐ฆ) โ(๐, ๐ฆ)โ1 = โ(๐, ๐)๐ผโ๐ฝโ๐พ โ(๐ฅ, ๐)๐ฝ โ(๐, ๐ฆ)๐พ โ(๐ค, ๐ค)๐ผ โ(๐ฅ, ๐ค)โ๐ฝ โ
โ(๐ค, ๐ฆ)โ๐พ โ(๐ค, ๐)๐พโ๐ผ โ(๐, ๐ค)๐ฝโ๐ผ .
โก
Generalizing De๏ฌnition 3.6, we de๏ฌne the shift operator ๐๐๐+๐ฝ acting on ๐ช(๐ท, ๐ซm ) by โซ (๐๐๐+๐ฝ ฮฆ)๐ง (๐) = ๐๐,m ๐๐ค โ(๐ค, ๐ค)๐โ๐ โ(๐ง, ๐ค)โ(๐+๐ฝ) ฮฆ๐ค (๐ต(๐ค, ๐ค) ๐ต(๐ง, ๐ค)โ1 ๐) ๐ท
for all ๐ง โ ๐ท and ๐ โ ๐. The normalization is chosen so that ๐ฝ = 0 yields the identity. It is easily shown that ๐๐๐+๐ฝ commutes with the (unweighted) action of ๐พ on ๐ช(๐ท, ๐ซm ). Proposition 4.3. Let ๐ โ ๐ซm and ๐, ๐ง โ ๐ท. Then, using the notation (4.6), we have (๐๐๐+๐ฝ ๐๐ )๐ง = โ(๐ง, ๐)โ๐ฝ ๐๐๐ง . Proof. Using a ๐-rotation in the anti-holomorphic variable ๐ค yields โซ ๐๐ค โ(๐ค, ๐ค)๐โ๐ โ(๐๐ (๐ง), ๐ค)โ(๐+๐ฝ) โ(๐, ๐ค)๐ฝ ๐(๐ต(๐ค, ๐ค) ๐ต(๐๐ (๐ง), ๐ค)โ1 ๐๐โฒ (๐ง) ๐) ๐ท โซ โซ ๐๐ ๐โ๐ = ๐๐ค โ(๐ค, ๐ค) โ(๐๐ (๐ง), ๐๐ค)โ(๐+๐ฝ) โ(๐, ๐๐ค)๐ฝ 2๐ ๐ท ๐
โซ =
๐ท
โซ =
๐ท
โ
๐(๐ต(๐ค, ๐ค) ๐ต(๐๐ (๐ง), ๐๐ค)โ1 ๐๐โฒ (๐ง) ๐) ๐๐ค โ(๐ค, ๐ค)๐โ๐ โ(๐๐(๐ง), 0)โ(๐+๐ฝ) โ(๐, 0)๐ฝ โ
๐(๐ต(๐ค, ๐ค) ๐ต(๐๐ (๐ง), 0)โ1 ๐๐โฒ (๐ง) ๐) โฒ ๐๐ค โ(๐ค, ๐ค)๐โ๐ ๐(๐ต(๐ค, ๐ค) ๐๐โฒ (๐ง) ๐) = ๐โ1 ๐,m ๐(๐๐ (๐ง) ๐).
Applying Lemma 4.1 to ๐ฅ = ๐๐ (๐ง), ๐ฆ = 0 we obtain (๐๐๐+๐ฝ ๐๐ )๐ง (๐) โซ = ๐๐,m ๐๐ค โ(๐ค, ๐ค)๐โ๐ โ(๐ง, ๐ค) ๐๐๐ค (๐ต(๐ค, ๐ค)โ(๐+๐ฝ) ๐ต(๐ง, ๐ค)โ1 ๐) ๐ท โซ = ๐๐,m ๐๐ค โ(๐ค, ๐ค)๐โ๐ โ(๐ง, ๐ค)โ(๐+๐ฝ) โ(๐, ๐)๐/2 ๐ท
โ
โ(๐ค, ๐)โ๐ ๐(๐๐โฒ (๐ค) ๐ต(๐ค, ๐ค) ๐ต(๐ง, ๐ค)โ1 ๐)
36
J. Arazy and H. Upmeier = ๐๐,m โ(๐, ๐)๐/2
โซ ๐ท
๐๐ค โ(๐ค, ๐ค)๐โ๐ โ(๐ง, ๐ค)โ(๐+๐ฝ)
โ
โ(๐ค, ๐)โ๐ ๐(๐ต(๐๐ (๐ค), ๐๐ (๐ค)) ๐ต(๐๐ (๐ง), ๐๐ (๐ค))โ1 ๐๐โฒ (๐ง) ๐). The general transformation formula (2.2) specializes to ๐ต(๐๐ (๐ง), ๐๐ (๐ค)) = ๐๐โฒ (๐ง) ๐ต(๐ง, ๐ค) ๐๐โฒ (๐ค)โ = ๐ต(๐, ๐)1/2 ๐ต(๐ง, ๐)โ1 ๐ต(๐ง, ๐ค) ๐ต(๐, ๐ค)โ1 ๐ต(๐, ๐)1/2 . As a consequence, ๐ต(๐๐ (๐ค), ๐๐ (๐ค)) ๐ต(๐๐ (๐ง), ๐๐(๐ค))โ1 ๐๐โฒ (๐ง) = ๐๐โฒ (๐ค) ๐ต(๐ค, ๐ค) ๐ต(๐ง, ๐ค)โ1 . Hence (๐๐๐+๐ฝ ๐๐ )๐ง (๐) = ๐๐,m โ(๐, ๐)๐/2 โ(๐, ๐)โ(๐+๐ฝ) โ(๐๐ (๐ง), ๐)๐+๐ฝ โซ โ
๐๐ค โ(๐ค, ๐ค)๐โ๐ โ(๐๐ (๐ง), ๐ค)โ(๐+๐ฝ) โ(๐, ๐ค)๐ฝ ๐(๐ต(๐ค, ๐ค) ๐ต(๐๐ (๐ง), ๐ค)โ1 ๐๐โฒ (๐ง) ๐) ๐ท
= โ(๐, ๐)๐/2 โ(๐ง, ๐)โ(๐+๐ฝ) ๐(๐๐โฒ (๐ง) ๐) = โ(๐ง, ๐)โ๐ฝ ๐๐๐ง (๐).
โก
Proposition 4.4. The operators ๐๐๐พ are symmetric with respect to โจโ
, โ
โฉ๐,m , namely โจ๐๐๐พ ฮฆ, ฮจโฉ๐,m = โจฮฆ, ๐๐๐พ ฮจโฉ๐,m
(4.7)
for all ฮฆ, ฮจ โ โ๐,m for which ๐๐๐พ ฮฆ, ๐๐๐พ ฮจ โ โ๐,m . Proof. For convenience we denote ๐๐๐,m (๐ง) = ๐๐,m ๐๐ง โ(๐ง, ๐ง)๐โ๐ . Then we have
โซ โฉ โช โจ๐๐๐พ ฮฆ, ฮจโฉ๐,m = ๐๐๐,m (๐ง) (๐๐๐พ ฮฆ)(๐ง, ๐ต(๐ง, ๐ง)1/2 โ
), ฮจ(๐ง, ๐ต(๐ง, ๐ง)1/2โ
) โฑ ๐ท โฉโซ โซ = ๐๐๐,m (๐ง) ๐๐๐,m (๐ค) โ(๐ง, ๐ค)โ๐พ ฮฆ(๐ค, ๐ต(๐ค, ๐ค) ๐ต(๐ง, ๐ค)โ1 ๐ต(๐ง, ๐ง)1/2 โ
), ๐ท ๐ท โช ฮจ(๐ง, ๐ต(๐ง, ๐ง)1/2โ
) โฑ โฉ โซ = ๐๐๐,m (๐ค) ฮฆ(๐ค, ๐ต(๐ค, ๐ค) ๐ต(๐ง, ๐ค)โ1 ๐ต(๐ง, ๐ง)1/2 โ
), ๐ท โช โซ ๐๐๐,m (๐ง) โ(๐ค, ๐ง)โ๐พ ฮจ(๐ง, ๐ต(๐ง, ๐ง)1/2 โ
) . ๐ท
โฑ
Using the fact that for all ๐, ๐ โ ๐ซm and ๐ โ ๐พ
โ
โจ๐ โ ๐, ๐โฉโฑ = โจ๐, ๐ โ ๐ โ โฉโฑ
Minimal and Maximal Invariant Spaces
37
we obtain (with ๐ = ๐ต(๐ค, ๐ค)1/2 ๐ต(๐ง, ๐ค)โ1 ๐ต(๐ง, ๐ง)1/2 ) that the last integral is equal to โฉ โซ ๐๐๐,m (๐ค) ฮฆ(๐ค, ๐ต(๐ค, ๐ค)1/2 โ
), ๐ท โช โซ ๐๐๐,m (๐ง) โ(๐ค, ๐ง)๐พ ฮจ(๐ง, ๐ต(๐ง, ๐ง) ๐ต(๐ค, ๐ง)โ1 ๐ต(๐ค, ๐ค)1/2 โ
) ๐ท โฑ โซ โฉ โช 1/2 ๐พ 1/2 = ๐๐๐,m (๐ค) ฮฆ(๐ค, ๐ต(๐ค, ๐ค) โ
), (๐๐ ฮจ)(๐ค, ๐ต(๐ค, ๐ค) โ
) โฑ
๐ท
= โจฮฆ, ๐๐๐พ ฮจโฉ๐,m .
โก
The same arguments yield the following result. Proposition 4.5. For ๐, ๐พ โ โ let ฮฆ โ โ๐,m โฉ โ๐พ,m and ฮจ โ โ๐,m with ๐๐๐พ ฮจ โ โ๐พ,m . Then โจฮฆ, ฮจโฉ๐,m = โจฮฆ, ๐๐๐พ ฮจโฉ๐พ,m . Proof.
โซ
โฉ โช ๐๐๐พ,m (๐ง) ฮฆ(๐ง, ๐ต(๐ง, ๐ง)1/2 โ
), (๐๐๐พ ฮจ)(๐ง, ๐ต(๐ง, ๐ง)1/2 โ
) โฑ ๐ท โฉ โซ = ๐๐๐พ,m (๐ง) ฮฆ(๐ง, ๐ต(๐ง, ๐ง)1/2 โ
), ๐ท โช โซ โ๐พ โ1 1/2 ๐๐๐,m (๐ค) โ(๐ค, ๐ง) ฮจ(๐ค, ๐ต(๐ค, ๐ค) ๐ต(๐ง, ๐ค) ๐ต(๐ง, ๐ง) โ
) ๐ท โฑ โฉโซ โซ = ๐๐๐,m (๐ค) ๐๐๐พ,m (๐ง) ฮฆ(๐ง, ๐ต(๐ง, ๐ง) ๐ต(๐ค, ๐ง)โ1 ๐ต(๐ค, ๐ค)1/2 โ
) โ(๐ง, ๐ค)โ๐พ , ๐ท ๐ท โช 1/2 ฮจ(๐ค, ๐ต(๐ค, ๐ค) โ
) โฑ โซ โฉ โช 1/2 = ๐๐๐,m (๐ค) ฮฆ(๐ค, ๐ต(๐ค, ๐ค) โ
), ฮจ(๐ค, ๐ต(๐ค, ๐ค)1/2 โ
) = โจฮฆ, ฮจโฉ๐,m
โจฮฆ,
๐๐๐พ
ฮจโฉ๐พ,m =
โฑ
๐ท
where we have used the reproducing property.
โก
Corollary 4.1. Let ฮจ, ฮฆ โ โ๐,m โฉ โ๐พ,m satisfy ๐๐๐พ ฮจ, ๐๐๐พ ฮฆ โ โ๐พ,m . Then โจ๐๐๐พ ฮฆ, ฮจโฉ๐พ,m = โจฮฆ, ฮจโฉ๐,m = โจฮฆ, ๐๐๐พ ฮจโฉ๐พ,m . Proof. The second equality follows from Proposition 4.5. For the ๏ฌrst, โจ๐๐๐พ ฮฆ, ฮจโฉ๐พ,m = โจฮจ, ๐๐๐พ ฮฆโฉ๐พ,m = โจฮจ, ฮฆโฉ๐,m = โจฮฆ, ฮจโฉ๐,m . Proposition 4.6. We have โซ ฮฆ๐ง (๐) = ๐๐+๐ฝ,m ๐๐ โ(๐, ๐)๐+๐ฝโ๐ โ(๐ง, ๐)โ๐ (๐๐๐+๐ฝ ฮฆ)๐ (๐ต(๐, ๐) ๐ต(๐ง, ๐)โ1 ๐). ๐ท
โก
38
J. Arazy and H. Upmeier
Proof. Let ๐ง โ ๐ท and ๐ โ ๐ซm be ๏ฌxed. The reproducing formula (4.6) applied to ๐ + ๐ฝ yields ๐/2 (ฮฆ๐ง โฃ ๐ โ ๐ต(๐ง, ๐ง)1/2 )โฑ ๐โ1 ๐+๐ฝ,m โ
โ(๐ง, ๐ง) โซ = ๐๐ โ(๐, ๐)๐+๐ฝโ๐ โ(๐ง, ๐)โ(๐+๐ฝ) โ(๐ง, ๐ง)๐/2 ๐ท
โซ =
๐ท
โซ =
๐ท
โซ =
๐ท
โซ =
๐ท
โ
(ฮฆ๐ โ ๐ต(๐, ๐) ๐ต(๐ง, ๐)โ1 โฃ ๐ โ ๐ต(๐ง, ๐ง)1/2)โฑ ๐๐ โ(๐, ๐)๐+๐ฝโ๐ โ(๐ง, ๐)โ(๐+๐ฝ) โ(๐ง, ๐ง)๐/2 โ
(ฮฆ๐ โ ๐ต(๐, ๐) โฃ ๐ โ ๐ต(๐ง, ๐ง)1/2 ๐ต(๐, ๐ง)โ1 )โฑ ๐๐ โ(๐, ๐)๐+๐ฝโ๐ โ
(ฮฆ๐ โ ๐ต(๐, ๐) โฃ โ(๐, ๐ง)โ๐ฝ โ
โ(๐, ๐ง)โ๐ โ(๐ง, ๐ง)๐/2 ๐ โ ๐ต(๐ง, ๐ง)1/2 ๐ต(๐, ๐ง)โ1 )โฑ ๐๐ โ(๐, ๐)๐+๐ฝโ๐ (ฮฆ๐ โ ๐ต(๐, ๐) โฃ โ(๐, ๐ง)โ๐ฝ โ
๐๐ง๐ )โฑ ๐๐ โ(๐, ๐)๐+๐ฝโ๐ (ฮฆ๐ โ ๐ต(๐, ๐) โฃ (๐๐๐+๐ฝ ๐๐ง )๐ )โฑ .
Using Proposition 4.4 for the parameter ๐ + ๐ฝ, we obtain ๐/2 (ฮฆ๐ง โฃ ๐ โ ๐ต(๐ง, ๐ง)1/2 )โฑ ๐โ1 ๐+๐ฝ,m โ
โ(๐ง, ๐ง) โซ ( ) = ๐๐ โ(๐, ๐)๐+๐ฝโ๐ (๐๐๐+๐ฝ ฮฆ)๐ โ ๐ต(๐, ๐) โฃ ๐๐ง๐ โฑ โซ๐ท = ๐๐ โ(๐, ๐)๐+๐ฝโ๐ ๐ท ) ( โ
(๐๐๐+๐ฝ ฮฆ)๐ โ ๐ต(๐, ๐) โฃ โ(๐, ๐ง)โ๐ โ(๐ง, ๐ง)๐/2 ๐ โ ๐ต(๐ง, ๐ง)1/2 ๐ต(๐, ๐ง)โ1 โฑ โซ ๐/2 ๐+๐ฝโ๐ โ๐ = โ(๐ง, ๐ง) ๐๐ โ(๐, ๐) โ(๐ง, ๐) ๐ท ) ( โ
(๐๐๐+๐ฝ ฮฆ)๐ โ ๐ต(๐, ๐) โฃ ๐ โ ๐ต(๐ง, ๐ง)1/2 ๐ต(๐, ๐ง)โ1 โฑ โซ ๐/2 ๐+๐ฝโ๐ โ๐ = โ(๐ง, ๐ง) ๐๐ โ(๐, ๐) โ(๐ง, ๐) ๐ท ( ) โ
(๐๐๐+๐ฝ ฮฆ)๐ โ ๐ต(๐, ๐) ๐ต(๐ง, ๐)โ1 โฃ ๐ โ ๐ต(๐ง, ๐ง)1/2 . โฑ
Since any polynomial in ๐ซm has the form โ(๐ง, ๐ง)๐/2 ๐ โ ๐ต(๐ง, ๐ง)1/2 , the assertion follows. โก Remark 4.1. Proposition 4.6 can be written as ๐ ๐๐๐+๐ฝ ฮฆ = ฮฆ ๐๐+๐ฝ
Minimal and Maximal Invariant Spaces
39
for ฮฆ in a dense subspace of โ๐,m . Thus, formally, ๐๐พ๐ ๐๐๐พ = ๐ผ for all ๐, ๐พ โ โ large enough. Up to now, the polynomial ๐ โ ๐ซm was arbitrary. We now specialize to ๐ด(๐) = ๐พ๐m (๐) = ๐พ m (๐, ๐) where ๐ โ ๐ is a maximal tripotent. Then we have ๐ด๐๐ง๐ ๐ (๐) = โ(๐, ๐)๐/2 โ(๐ง, ๐)โ๐ ๐พ m (๐ โ1 ๐ต(๐, ๐)1/2 ๐ต(๐ง, ๐)โ1 ๐, ๐) = โ(๐, ๐)๐/2 โ(๐ง, ๐)โ๐ ๐พ m (๐ต(๐, ๐)1/2 ๐ต(๐ง, ๐)โ1 ๐, ๐๐).
(4.8)
De๏ฌnition 4.3. (i) Let ๐๐,m denote the Banach space of all holomorphic functions ฮฆ : ๐ท โ ๐ซm which have a representation โซ ฮฆ๐ง (๐) = ๐๐ (๐) ๐ด๐๐ง (๐) ๐บ
for some ๏ฌnite โ-valued Borel measure on ๐บ. The norm is de๏ฌned as the in๏ฌmum โฅฮฆโฅ๐๐,m = inf โฅ๐โฅ ๐
taken over all such representations. (ii) De๏ฌne a vector-valued ๐ฟ1 -space โ1๐พ to consist of all ฮฆ โ ๐ช(๐ท, ๐ซm ) such that โซ โฅฮฆโฅโ1๐พ := ๐๐พ,m ๐๐ง โ(๐ง, ๐ง)๐พโ๐ โฅฮฆ๐ง โ ๐ต(๐ง, ๐ง)1/2 โฅโฑ < โ. ๐ท
Here โฅ โ
โฅโฑ is the Fischer norm on ๐ซm . Our main theorem in this section is Theorem 4.2. Let ฮฆ โ ๐ช(๐ท, ๐ซm ) and suppose that ๐๐๐+๐ฝ ฮฆ โ โ1๐ฝ+๐/2 . Then ฮฆ โ ๐๐,m and 1/2 ๐๐+๐ฝ,m โฅฮฆโฅ๐๐,m โค (๐/๐)1/2 โฅ๐๐๐+๐ฝ ฮฆโฅโ1๐ฝ+๐/2 . m ๐m ๐๐ฝ+๐/2 Proof. De๏ฌne a complex measure ๐ on ๐บ by ๐๐ (๐๐ ๐) = ๐๐ ๐๐ โ(๐, ๐)๐ฝ+๐/2โ๐ (๐๐๐+๐ฝ ฮฆ)๐ (๐ต(๐, ๐)1/2 ๐๐). For each ๐ โ ๐พ the Cauchy-Schwarz inequality yields $ ๐+๐ฝ $ $ $ m $(๐๐ ฮฆ)๐ (๐ต(๐, ๐)1/2 ๐๐)$ = $((๐๐๐+๐ฝ ฮฆ)๐ โ ๐ต(๐, ๐)1/2 โฃ ๐พ๐๐ )โฑ $ โค โฅ(๐๐๐+๐ฝ ฮฆ)๐ โ ๐ต(๐, ๐)1/2 โฅโฑ โ
๐พ m (๐, ๐)1/2 =
1/2
๐m
1/2
(๐/๐)m
โฅ(๐๐๐+๐ฝ ฮฆ)๐ โ ๐ต(๐, ๐)1/2 โฅโฑ
๐+๐ฝ = ๐1/2 ฮฆ)๐ โ ๐ต(๐, ๐)1/2 โฅ๐/๐ . m โฅ(๐๐
40
J. Arazy and H. Upmeier
Hence
โซ โฅ๐โฅ =
โซ ๐๐
๐ท
๐พ
โค =
$ $ ๐๐ โ(๐, ๐)๐ฝ+๐/2โ๐ $(๐๐๐+๐ฝ ฮฆ)๐ (๐ต(๐, ๐)1/2 ๐๐)$ โซ
1/2
๐m
1/2
(๐/๐)m
๐ท
1
๐m
1 1 ๐๐ โ(๐, ๐)๐ฝ+๐/2โ๐ 1(๐๐๐+๐ฝ ฮฆ)๐ โ ๐ต(๐, ๐)1/2 1โฑ
1/2
๐๐ฝ+๐/2 (๐/๐)1/2 m
1 ๐+๐ฝ 1 1๐๐ ฮฆ1 1 โ
๐ฝ+๐/2
.
Hence ๐ is a ๏ฌnite measure on ๐บ. Moreover, (4.8) implies โซ ๐๐(๐) ๐ด๐๐ง (๐) ๐บ
=
โซ
๐๐โ(๐,๐)๐ฝ+๐/2โ๐ โ(๐,๐)๐/2 โ(๐ง,๐)โ๐ ๐ท โซ โ
๐๐ (๐๐๐+๐ฝ ฮฆ)๐ (๐ต(๐,๐)1/2 ๐๐) ๐พ m (๐ต(๐,๐)1/2 ๐ต(๐ง,๐)โ1 ๐, ๐๐)
โซ =
๐ท
โซ
๐พ
๐๐โ(๐,๐)
๐+๐ฝโ๐
โ๐
โซ
โ(๐ง,๐)
๐๐ (๐๐๐+๐ฝ ฮฆ)๐ (๐ต(๐,๐)1/2 ๐๐)
๐พ
โ
๐พ m (๐๐, ๐ต(๐,๐)1/2 ๐ต(๐ง,๐)โ1 ๐)
( ) m ๐๐โ(๐,๐)๐+๐ฝโ๐ โ(๐ง,๐)โ๐ (๐๐๐+๐ฝ ฮฆ)๐ โ ๐ต(๐,๐)1/2 โฃ๐พ๐ต(๐,๐) 1/2 ๐ต(๐ง,๐)โ1 ๐ ๐/๐ ๐ท โซ = (๐/๐)โ1 ๐๐โ(๐,๐)๐+๐ฝโ๐ โ(๐ง,๐)โ๐ m ๐ท ( ) m โ
(๐๐๐+๐ฝ ฮฆ)๐ โ ๐ต(๐,๐)1/2 โฃ๐พ๐ต(๐,๐) 1/2 ๐ต(๐ง,๐)โ1 ๐ โฑ โซ ( ) โ1 ๐+๐ฝโ๐ โ๐ ๐+๐ฝ = (๐/๐)m ๐๐โ(๐,๐) โ(๐ง,๐) (๐๐ ฮฆ)๐ ๐ต(๐,๐)1/2 ๐ต(๐,๐)1/2 ๐ต(๐ง,๐)โ1 ๐ โซ๐ท ( ) โ1 = (๐/๐)m ๐๐โ(๐,๐)๐+๐ฝโ๐ โ(๐ง,๐)โ๐ (๐๐๐+๐ฝ ฮฆ)๐ ๐ต(๐,๐) ๐ต(๐ง,๐)โ1 ๐ =
๐ท
โ1 = (๐/๐)โ1 m ๐๐+๐ฝ,m ฮฆ๐ง (๐)
using Proposition 4.4. Thus ฮฆ is represented by ๐, up to a constant.
โก
Minimal and Maximal Invariant Spaces
41
5. Minimal spaces for non-tube type domains In this section we obtain a โconverseโ of Theorem 4.2, and thus a complete characterization of the minimal space, for the special partitions s = (๐ , . . . , ๐ ), where ๐ โ โ. These โconstantโ partitions arise naturally in the study of highest quotients (Dirichlet spaces) for domains which are not of tube type (cf. [AU3]). The integration formulas developed here may be of independent interest. We consider the Peirce decomposition ( ) ๐1 ๐ = ๐1 โ ๐1/2 = (5.1) ๐1/2 of ๐ for a maximal tripotent ๐, and write ๐ง โ ๐ as ๐ง = ๐ง1 + ๐ง1/2 , with ๐ง1 โ ๐1 and ๐ง1/2 โ ๐1/2 . Lemma 5.1. For ๐ข โ ๐1 , ๐ฃ โ ๐1/2 the Bergman operator ๐ต(๐ข, ๐ฃ) has a blockmatrix decomposition ) ( ๐ผ โ2๐ข โก ๐ฃ โ (5.2) ๐ต(๐ข, ๐ฃ) = 1 0 ๐ผ1/2 with respect to (5.1). Here ๐ผ๐ denotes the identity operator on ๐๐ . Proof. For ๐ง โ ๐, we have {๐ข ๐ฃ โ ๐ง1 } โ ๐3/2 = (0) and ๐๐ฃ ๐ง1 โ ๐0 = (0), since ๐ is maximal. Moreover, ๐๐ฃ ๐ง1/2 โ ๐1/2 and hence ๐๐ข ๐๐ฃ ๐ง1/2 โ ๐3/2 = (0). Thus ๐ต(๐ข, ๐ฃ) ๐ง = ๐ง โ 2{๐ข ๐ฃ โ ๐ง} + ๐๐ข ๐๐ฃ ๐ง = ๐ง1 + ๐ง1/2 โ 2{๐ข ๐ฃโ (๐ง1 + ๐ง1/2 )} + ๐๐ข ๐๐ฃ (๐ง1 + ๐ง1/2 ) = ๐ง1 + ๐ง1/2 โ 2{๐ข ๐ฃโ ๐ง1/2 }, with ๐ง1 โ 2{๐ข ๐ฃ โ ๐ง1/2 } โ ๐1 . The assertion follows.
โก
Corollary 5.1. For ๐ข โ ๐1 , ๐ฃ โ ๐1/2 , we have det๐ ๐ต(๐ข, ๐ฃ) = 1. In particular, ๐ต(๐ข, ๐ฃ) is invertible, with inverse given by ) ( ๐ผ 2๐ข โก ๐ฃ โ ๐ต(๐ข, ๐ฃ)โ1 = ๐ต(๐ข, โ๐ฃ) = 1 0 ๐ผ1/2 Lemma 5.2. If ๐ต(๐ง, ๐ค) is invertible and ๐๐ง ๐ค = ๐๐ค ๐ง = 0, then ๐ง ๐ค = ๐ง. Proof. By assumption, we have ๐ต(๐ง,๐ค)๐ง = ๐ง โ 2{๐ง ๐คโ ๐ง} + ๐๐ง ๐๐ค ๐ง = ๐ง โ 2๐๐ง ๐ค + ๐๐ง ๐๐ค ๐ง = ๐ง = ๐ง โ ๐๐ง ๐ค = ๐ต(๐ง,๐ค)๐ง ๐ค . Since ๐ต(๐ง, ๐ค) is invertible, we conclude that ๐ง = ๐ง ๐ค .
โก
Proposition 5.1. Suppose ๐ฃ, ๐ข โ ๐ท and ๐๐ข ๐ฃ = ๐๐ฃ ๐ข = 0. Then we have
) ( ๐ต ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ, ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ = ๐ต(๐ข, ๐ข)1/2 ๐ต(๐ฃ, ๐ข) ๐ต(๐ฃ, ๐ฃ) ๐ต(๐ข, ๐ฃ) ๐ต(๐ข, ๐ข)1/2 .
(5.3)
(5.4)
42
J. Arazy and H. Upmeier
Proof. Since ๐ฃ โ๐ข = ๐ฃ by Lemma 5.2, we have ๐๐ข (๐ฃ) = ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ โ๐ข = ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ and ๐๐ขโฒ (๐ฃ) = ๐ต(๐ข, ๐ข)1/2 ๐ต(๐ฃ โ๐ข , ๐ข) = ๐ต(๐ข, ๐ข)1/2 ๐ต(๐ฃ, ๐ข). Now apply (2.2). โก For any tripotent, the Peirce spaces are hermitian Jordan subtriples of ๐, and ๐1 and ๐0 are always irreducible if ๐ is irreducible. One can show that in our case of a maximal tripotent (i.e., ๐0 = (0)) the Peirce 12 -space ๐1/2 is also irreducible. Let ๐ท1 = ๐ท โฉ ๐1 and ๐ท1/2 = ๐ท โฉ ๐1/2 denote the respective open unit balls. Corollary 5.2. Let ๐ข โ ๐ท1 and ๐ฃ โ ๐ท1/2 . Then (5.3) holds and, in addition, we have โ(๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ, ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ) = โ(๐ข, ๐ข) โ(๐ฃ, ๐ฃ). (5.5) Proof. By Lemma 5.1 and Lemma 5.2, the assumption of Proposition 5.1 is satis๏ฌed, showing that (5.4) holds. Moreover, โ(๐ข, ๐ฃ) = 1 = โ(๐ฃ, ๐ข) by Lemma 5.1. Therefore (5.5) follows from (5.4) by taking determinants. โก Proposition 5.2. For ๐ข โ ๐1 and ๐ฃ โ ๐1/2 , we have ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ โ ๐ท if and only if ๐ข โ ๐ท1 and ๐ฃ โ ๐ท1/2 . Proof. As a consequence of the spectral theorem for Jordan triples, we have โ(๐ง, ๐ง) > 0 for ๐ง โ ๐ท and โ(๐ง, ๐ง) = 0 for all ๐ง โ โ๐ท. Hence ๐ท is a connected component of ๐ := {๐ง โ ๐ : โ(๐ง, ๐ง) > 0}. De๏ฌne ๐ : ๐ท โ ๐1/2 by ๐(๐ค) := ๐ต(๐ค1 , ๐ค1 )โ1/2 ๐ค1/2 for all ๐ค = ๐ค1 + ๐ค1/2 โ ๐ท with ๐ค๐ โ ๐๐ . Since Peirce projections are contractive, we have โฅ๐ค1 โฅ โค โฅ๐คโฅ < 1. Therefore ๐ค1 โ ๐ท1 and ๐ต(๐ค1 , ๐ค1 ) is invertible. By Corollary 5.2, we have โ(๐ค1 , ๐ค1 ) โ(๐(๐ค), ๐(๐ค)) = โ(๐ค, ๐ค) โ= 0. It follows that โ(๐(๐ค), ๐(๐ค)) โ= 0 and therefore ๐(๐ค) โ ๐1/2 โฉ ๐ . Since ๐ is continuous and ๐ท is connected, it follows that ๐(๐ท) belongs to the 0-connected component of ๐ โฉ ๐1/2 , which coincides with ๐ท1/2 . This shows that ๐ค = ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ โ ๐ท implies ๐ข โ ๐ท1 and ๐ฃ = ๐(๐ค) โ ๐ท1/2 . Conversely, let ๐ข โ ๐ท1 . De๏ฌne ๐น๐ข : ๐1/2 โ ๐ by ๐น๐ข (๐ฃ) := ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ. Then Corollary 5.2 implies โ(๐น๐ข (๐ฃ), ๐น๐ข (๐ฃ)) = โ(๐ข, ๐ข) โ(๐ฃ, ๐ฃ). If ๐ฃ โ ๐ท1/2 , then โ(๐ฃ, ๐ฃ) โ= 0 and hence ๐น๐ข (๐ฃ) โ ๐ . Since ๐น๐ข (0) = ๐ข โ ๐ท1 โ ๐ท, ๐น๐ข (๐ท1/2 ) belongs to the ๐ข-connected component of ๐ , which coincides with ๐ท. Therefore ๐ค = ๐น๐ข (๐ฃ) โ ๐ท. โก
Minimal and Maximal Invariant Spaces
43
According to Proposition 5.2 the map ๐น (๐ข, ๐ฃ) := ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ de๏ฌnes a real-analytic isomorphism from ๐ท1 ร ๐ท1/2 onto ๐ท, with inverse ๐น โ1 (๐ค1 + ๐ค1/2 ) = ๐ค1 + ๐ต(๐ค1 , ๐ค1 )โ1/2 ๐ค1/2 . Put ๐ฝ(๐ข) := ๐ต(๐ข, ๐ข)1/2 โ End(๐). Then ๐น has the derivative ๐น โฒ (๐ข, ๐ฃ)(๐ฅ, ๐ฆ) = ๐ฅ + ๐ฝ(๐ข) ๐ฆ + (๐ฝ โฒ (๐ข) ๐ฅ) ๐ฃ for ๐ฅ โ ๐1 , ๐ฆ โ ๐1/2 . Since ๐ฝ(๐ข) preserves both Peirce spaces, the same is true for ๐ฝ โฒ (๐ข)๐ฅ โ End(๐). Thus we have a block-matrix decomposition ( ) ๐ผ1 ๐ ๐น โฒ (๐ข, ๐ฃ) = 0 ๐ต(๐ข, ๐ข)1/2 with respect to (5.1), where ๐ ๐ฅ := (๐ฝ โฒ (๐ข) ๐ฅ) ๐ฃ = It follows that
โ $$ $ ๐ฝ(๐ข + ๐ก๐ฅ) ๐ฃ. โ๐ก ๐ก=0
det๐ ๐น โฒ (๐ข, ๐ฃ) = det๐1/2 ๐ต(๐ข, ๐ข)1/2 = โ(๐ข, ๐ข)๐/2 . Hence ๐น โฒ (๐ข, ๐ข) has the โrealโ determinant $ $2 det ๐น โฒ (๐ข, ๐ฃ) = $det๐1/2 ๐ต(๐ข, ๐ข)1/2 $ = โ(๐ข, ๐ข)๐ .
(5.6)
Making the change of variables ๐ค = ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ (5.6) yields
(๐ข โ ๐ท1 , ๐ฃ โ ๐ท1/2 )
๐๐ค = โ(๐ข, ๐ข)๐ ๐๐ข ๐๐ฃ.
(5.7) (5.8)
Proposition 5.3. Let ๐ข โ ๐1 , ๐ฃ โ ๐1/2 and ๐ = ๐1 + ๐1/2 โ ๐ with ๐๐ โ ๐๐ . Suppose that ๐ต(๐1 , ๐ข) is invertible. Then โ(๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ, ๐) = โ(๐ข, ๐1 ) โ
โ(๐ฃ, ๐ต(๐ข, ๐ข)1/2 ๐ต(๐1 , ๐ข)โ1 ๐1/2 ).
(5.9)
Proof. Polarizing the identity (5.5) yields โ(๐ข + ๐ต(๐ข, ๐1 )1/2 ๐ฃ1 , ๐1 + ๐ต(๐1 , ๐ข)1/2 ๐ฃ2 ) = โ(๐ข, ๐1 ) โ(๐ฃ1 , ๐ฃ2 )
(5.10)
whenever ๐ฃ1 , ๐ฃ2 โ ๐1/2 . Putting ๐ฃ1 = ๐ต(๐ข, ๐1 )โ1/2 ๐ต(๐ข, ๐ข)1/2 ๐ฃ
and ๐ฃ2 = ๐ต(๐1 , ๐ข)โ1/2 ๐1/2 ,
the left-hand sides of (5.9) and (5.10) agree, whereas ( ) โ(๐ฃ1 , ๐ฃ2 ) = โ ๐ต(๐ข, ๐1 )โ1/2 ๐ต(๐ข, ๐ข)1/2 ๐ฃ, ๐ต(๐1 , ๐ข)โ1/2 ๐1/2 ( ) = โ ๐ต(๐ข, ๐ข)1/2 ๐ฃ, ๐ต(๐1 , ๐ข)โ1 ๐1/2 ( ) = โ ๐ฃ, ๐ต(๐ข, ๐ข)1/2 ๐ต(๐1 , ๐ข)โ1 ๐1/2 .
โก
44
J. Arazy and H. Upmeier
Lemma 5.3. Let ๐ข โ ๐ท1 and ๐ = ๐1 + ๐1/2 โ ๐ท with ๐๐ โ ๐๐ . Then ๐ต(๐1 , ๐ข) is invertible and ๐ต(๐ข, ๐ข)1/2 ๐ต(๐1 , ๐ข)โ1 ๐1/2 โ ๐ท1/2 . Proof. Since ๐1 โ ๐ท1 , it follows that ๐ต(๐1 , ๐ข) is invertible. Therefore the addition formula [L2, p.26] yields (๐ข๐1 )
๐๐ข = (๐1 + ๐1/2 )๐ข = ๐๐ข1 + ๐ต(๐1 , ๐ข)โ1 ๐1/2 (๐ข๐1 )
since ๐ข๐1 โ ๐1 and hence ๐1/2
= ๐๐ข1 + ๐ต(๐1 , ๐ข)โ1 ๐1/2
= ๐1/2 by Lemma 5.2. It follows that
๐โ๐ข (๐) = โ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐๐ข = โ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐๐ข1 + ๐ต(๐ข, ๐ข)1/2 ๐ต(๐1 , ๐ข)โ1 ๐1/2 . Since ๐ โ ๐ท, we have ๐โ๐ข (๐) โ ๐ท. Therefore the Peirce 12 -component ๐ต(๐ข, ๐ข)1/2 ๐ต(๐1 , ๐ข)โ1 ๐1/2 โ ๐ท1/2 .
โก
Let ๐1 : ๐ โ ๐1 denote the Peirce 1-projection. Lemma 5.4. For ๐ข โ ๐1 and ๐ฃ โ ๐1/2 , we have ๐1 ๐ต(๐ฃ, ๐ข) = ๐1 . Proof. Using Lemma 5.1 and ๐ต(๐ฃ, ๐ข) = ๐ต(๐ข, ๐ฃ)โ we write ( )( ) ( ๐ผ 0 ๐ผ1 0 ๐ผ ๐1 ๐ต(๐ฃ, ๐ข) = 1 = 1 0 0 โ2๐ฃ โก ๐ขโ ๐ผ1/2 0
0 0
) = ๐1 .
Here ๐ผ๐ is the identity map on ๐๐ .
โก
Lemma 5.5. Let s = (๐ , . . . , ๐ ) and ๐ค = ๐ค1 + ๐ค1/2 โ ๐ท with ๐ค๐ โ ๐๐ . Then ๐พ๐s (๐ต(๐ค, ๐ค) ๐) =
๐s โ(๐ค, ๐ค)๐ โ(๐ค1 , ๐ค1 )๐ . (๐/๐)s
(5.11)
Proof. Let ๐ be the Jordan algebra determinant of ๐1 , normalized by ๐ (๐) = 1. Then ๐s ๐พ๐s (๐ง) = ๐พ s (๐, ๐) ๐ (๐1 ๐ง)๐ = ๐ (๐1 ๐ง)๐ . (๐/๐)s Writing ๐ค = ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ with ๐ข โ ๐ท1 and ๐ฃ โ ๐ท1/2 , Proposition 5.1 and Lemma 5.4 imply ๐1 ๐ต(๐ค, ๐ค) ๐ = ๐1 ๐ต(๐ข, ๐ข)1/2 ๐ต(๐ฃ, ๐ข) ๐ต(๐ฃ, ๐ฃ) ๐ต(๐ข, ๐ฃ) ๐ต(๐ข, ๐ข)1/2 ๐ = ๐1 ๐ต(๐ข, ๐ข)1/2 ๐1 ๐ต(๐ฃ, ๐ข) ๐ต(๐ฃ, ๐ฃ) ๐ต(๐ข, ๐ฃ) ๐1 ๐ต(๐ข, ๐ข)1/2 ๐ = ๐1 ๐ต(๐ข, ๐ข)1/2 ๐1 ๐ต(๐ฃ, ๐ฃ) ๐1 ๐ต(๐ข, ๐ข)1/2 ๐ = ๐ต(๐ข, ๐ข)1/2 ๐ต(๐ฃ, ๐ฃ) ๐ต(๐ข, ๐ข)1/2 ๐. The invertible transformations ๐1 ๐ต(๐ข, ๐ข)1/2 ๐1 and ๐1 ๐ต(๐ฃ, ๐ฃ) ๐1 on ๐1 belong to the โstructure groupโ ๐พ1โ of ๐1 , and ๐ has the semi-invariance property ๐ (๐พ๐ง) = ๐ (๐พ๐) ๐ (๐ง) = (Det ๐พ)๐/๐1 ๐ (๐ง) for all ๐พ โ ๐พ1โ and ๐ง โ ๐1 .
Minimal and Maximal Invariant Spaces
45
It follows that ๐ (๐1 ๐ต(๐ค, ๐ค) ๐) = ๐ ((๐1 ๐ต(๐ข, ๐ข)1/2 ๐1 ) (๐1 ๐ต(๐ฃ, ๐ฃ) ๐1 ) (๐1 ๐ต(๐ข, ๐ข)1/2 ) ๐) = ๐ (๐ต(๐ข, ๐ข)1/2 ๐)2 ๐ (๐ต(๐ฃ, ๐ฃ) ๐) = โ(๐ข, ๐ข)2 โ(๐ฃ, ๐ฃ). Since โ(๐ค, ๐ค) = โ(๐ข, ๐ข) โ(๐ฃ, ๐ฃ) by (5.5), the assertion follows.
โก
Let ๐1 , ๐1 , ๐1 , ๐1 and ๐1/2 , ๐1/2 , ๐1/2 , ๐1/2 denote the respective invariants for the (irreducible) Jordan triples ๐1 and ๐1/2 . Theorem 5.1. The integral de๏ฌning ๐โ1 ๐,s is ๏ฌnite (i.e., ๐๐,s > 0) if and only if ๐ + ๐ > ๐ โ 1. In this case we have ฮฮฉ (2๐ + ๐) ฮฮฉ1/2 (๐ + ๐ โ ๐ + ๐1/2 ) ๐๐,s = . ๐1/2 ๐ ๐ ฮฮฉ (2๐ + ๐ โ ๐๐1 ) ฮฮฉ1/2 (๐ + ๐ โ ๐ + ๐1/2 โ ๐1/2 ) Proof. Combining (5.11), (5.5) and (5.7) we see that โซ โซ โซ ๐ +๐โ๐ ๐ 2๐ +๐+๐โ๐ ๐โ1 = ๐๐ค โ(๐ค, ๐ค) โ(๐ค , ๐ค ) = ๐๐ข โ(๐ข, ๐ข) ๐๐ฃ โ(๐ฃ, ๐ฃ)๐ +๐โ๐ . 1 1 ๐,s ๐ท
๐ท1
๐ท1/2
Since ๐ โ ๐ = ๐1 (the genus of ๐1 ), we have โซ ๐1 ๐๐ข โ(๐ข, ๐ข)2๐ +๐+๐โ๐ = ๐ ๐1 ฮฮฉ (2๐ + ๐ โ )/ฮฮฉ (2๐ + ๐) ๐ ๐ท1
which is ๏ฌnite if and only if 2๐ + ๐ > (๐ โ 1) ๐ + 1 = ๐1 โ 1. Also, โซ ( ๐1/2 ) ๐๐ฃ โ(๐ฃ, ๐ฃ)๐ +๐โ๐ = ๐ ๐1/2 ฮฮฉ1/2 ๐ + ๐ โ ๐ + ๐1/2 โ /ฮฮฉ1/2 (๐ + ๐ โ ๐ + ๐1/2 ) ๐1/2
๐ท1/2
which is ๏ฌnite if and only if ๐ + ๐ โ ๐ + ๐1/2 โ Since ๐1/2 โ
๐1/2 ๐1/2
= (๐1/2 โ 1)
๐1/2 2
๐1/2 ๐1/2 > (๐1/2 โ 1) . ๐1/2 2
+ 1, this is equivalent to ๐ + ๐ > ๐ โ 1.
โก
Proposition 5.4. Let ๐ โ ๐ท and ๐ โ ๐. Then โซ 1 โฅ๐๐๐+๐ฝ (๐พ๐m )๐ โฅโ1๐ฝ+๐/2 = ๐๐ง โ(๐ง, ๐ง)๐ฝ+๐/2โ๐ โฃโ(๐ง, ๐)โ๐ฝ โฃ โ
๐พ๐m (๐ต(๐ง, ๐ง) ๐)1/2 ๐๐ฝ+๐/2 ๐ท โซ = ๐๐ง โ(๐ง, ๐ง)๐ฝ+๐/2โ๐ โ
โฃโ(๐ง, ๐)โ๐ฝ โฃ โ
โฅ๐พ๐m โ ๐ต(๐ง, ๐ง)1/2 โฅโฑ (5.12) ๐ท
Proof. Proposition 4.3 and (4.8) imply (๐๐๐+๐ฝ (๐พ๐m )๐ )๐ง โ ๐ต(๐ง, ๐ง)1/2 = โ(๐ง, ๐)โ๐ฝ (๐พ๐m )๐๐ง โ ๐ต(๐ง, ๐ง)1/2 = โ(๐, ๐)๐/2 โ(๐ง, ๐)โ(๐ฝ+๐) ๐พ๐m โ ๐ต(๐, ๐)1/2 ๐ต(๐ง, ๐)โ1 ๐ต(๐ง, ๐ง)1/2 .
46
J. Arazy and H. Upmeier
Since โจ๐พ๐m โ ๐ต(๐, ๐)1/2 ๐ต(๐ง, ๐)โ1 ๐ต(๐ง, ๐ง)1/2 โฉ2โฑ โฒ โฒ = ๐พ๐m (๐โ๐ (๐ง) ๐ต(๐ง, ๐ง) ๐โ๐ (๐ง)โ ๐) = ๐พ๐m (๐ต(๐๐โ1 (๐ง), ๐๐โ1 (๐ง))๐),
it follows that โฅ(๐๐๐+๐ฝ (๐พ๐m )๐ )๐ง โ ๐ต(๐ง, ๐ง)1/2 โฅโฑ $ $ = โ(๐, ๐)๐/2 $โ(๐ง, ๐)โ(๐ฝ+๐) $ ๐พ๐m (๐ต(๐๐โ1 (๐ง), ๐๐โ1 (๐ง)) ๐)1/2 . Applying Lemma 4.1 to ๐ฅ = ๐ฆ = 0 yields 1
โฅ๐๐๐+๐ฝ (๐พ๐m )๐ โฅโ1๐ฝ+๐/2 ๐๐ฝ+๐/2 โซ = ๐๐ง โ(๐ง, ๐ง)๐ฝ+๐/2โ๐ โฅ(๐๐๐+๐ฝ (๐พ๐m )๐ )๐ง โ ๐ต(๐ง, ๐ง)1/2 โฅโฑ ๐ท โซ $ $ ๐/2 = โ(๐, ๐) ๐๐ง โ(๐ง, ๐ง)๐ฝ+๐/2โ๐ โ
$โ(๐ง, ๐)โ(๐ฝ+๐) $ ๐ท
โ
๐พ๐m (๐ต(๐๐โ1 (๐ง), ๐๐โ1 (๐ง)) ๐)1/2 = โ(๐, ๐)๐/2 โ(๐, ๐)โ๐/2 โซ โ
๐๐ง โ(๐ง, ๐ง)๐ฝ+๐/2โ๐ โ(๐ง, ๐)โ๐ฝ/2 โ(๐, ๐ง)โ๐ฝ/2 ๐พ๐m (๐ต(๐ง, ๐ง) ๐)1/2 ๐ท โซ $ $ = ๐๐ง โ(๐ง, ๐ง)๐ฝ+๐/2โ๐ $โ(๐ง, ๐)โ๐ฝ $ ๐พ๐m (๐ต(๐ง, ๐ง) ๐)1/2 โซ๐ท $ $ = ๐๐ง โ(๐ง, ๐ง)๐ฝ+๐/2โ๐ $โ(๐ง, ๐)โ๐ฝ $ โ
โฅ๐พ๐m โ ๐ต(๐ง, ๐ง)1/2 โฅโฑ . ๐ท
Our main result in this section is Theorem 5.2. Let ๐ โ โ and ๐ satisfy ๐ ๐ ๐ + > (๐ โ 1) 2 2 and ๐1/2 ๐ +๐ > (๐1/2 โ 1) + ๐ โ ๐1/2 . 2 2 Let ๐ฝ โ โ satisfy ๐ฝ + ๐+๐ 2 > ๐ โ 1. Then we have for ฮฆ โ ๐ช(๐ท, ๐ซs ) ฮฆ โ ๐๐,s โโ ๐๐๐+๐ฝ ฮฆ โ โ1๐ฝ+๐/2 . Proof. Let ๐1/2 be the genus of ๐1/2 , and put ๐ผ=๐ฝ+
๐+๐ + ๐1/2 โ ๐. 2
Then ๐ฝ โ ๐ผ = ๐ โ ๐1/2 โ
๐1/2 ๐+๐ <โ (๐1/2 โ 1) 2 2
โก
Minimal and Maximal Invariant Spaces by assumption. This implies ๐ถ (1/2) := sup
๐ฆโ๐ท1/2
(1/2)
2 ๐น1
47
( ) ๐ฝ/2 ๐ฝ/2 (๐ฆ, ๐ฆ) < +โ. ๐ผ
By Lemma 5.3, ๐ต(๐ข, ๐ข)1/2 ๐ต(๐1 , ๐ข)โ1 ๐1/2 โ ๐ท1/2 and hence ( ) (1/2) ๐ฝ/2 ๐ฝ/2 (๐ต(๐ข, ๐ข)1/2 ๐ต(๐1 , ๐ข)โ1 ๐1/2 , ๐ต(๐ข, ๐ข)1/2 ๐ต(๐1 , ๐ข)โ1 ๐1/2 ) 2 ๐น1 ๐ผ โค ๐ถ (1/2) for all ๐ข โ ๐ท1 and ๐ = ๐1 + ๐1/2 โ ๐ท. Now consider ฮฆ = ๐ด๐๐ = (๐พ๐s )๐ . Specializing Proposition 5.4 to the constant partition s = (๐ , . . . , ๐ ) and making the change of variables ๐ค = ๐ข + ๐ต(๐ข, ๐ข)1/2 ๐ฃ as in (5.7), we obtain with Proposition 5.3 and Lemma 5.5. 1/2
1
(๐/๐)s
๐๐ฝ+๐/2 =
1/2 ๐s 1/2 (๐/๐)s 1/2 ๐s
โซ = = =
โซ๐ท โซ๐ท
โฅ๐๐๐+๐ฝ ฮฆโฅโ1๐ฝ+๐/2
โซ ๐ท
$ $ ๐๐ค โ(๐ค, ๐ค)๐ฝ+๐/2โ๐ $โ(๐ค, ๐)โ๐ฝ $ ๐พ๐s (๐ต(๐ค, ๐ค) ๐)1/2
$ $ ๐๐ค โ(๐ค, ๐ค)๐ฝ+๐/2โ๐ $โ(๐ค, ๐)โ๐ฝ $ โ(๐ค, ๐ค)๐ /2 โ(๐ค1 , ๐ค1 )๐ /2 ๐๐ค โ(๐ค, ๐ค)๐ฝ+
๐+๐ 2 โ๐
$ $ โ(๐ค1 , ๐ค1 )๐ /2 $โ(๐ค, ๐)โ๐ฝ $
$ $ ๐ ๐๐ข โ(๐ข, ๐ข)๐ฝ+ 2 +๐ +๐โ๐ $โ(๐ข, ๐1 )โ๐ฝ $ ๐ท1 โซ $ $ ๐+๐ โ
๐๐ฃ โ(๐ฃ, ๐ฃ)๐ฝ+ 2 โ๐ $โ(๐ฃ, ๐ต(๐ข, ๐ข)1/2 ๐ต(๐1 , ๐ข)โ1 ๐1/2 )โ๐ฝ $ ๐ท1/2
( ) $ $ (1/2) ๐ฝ/2 ๐ฝ/2 ๐ฝ+๐ +๐/2โ๐1 $ โ๐ฝ $ ๐๐ข โ(๐ข, ๐ข) โ(๐ข, ๐ ) ๐น 1 2 1 (1/2) ๐ผ ๐ท1 ๐๐ผ ( ) 1/2 โ1 1/2 โ1 โ
๐ต(๐ข, ๐ข) ๐ต(๐1 , ๐ข) ๐1/2 , ๐ต(๐ข, ๐ข) ๐ต(๐1 , ๐ข) ๐1/2 โซ $ $ ๐ถ (1/2) โค (1/2) ๐๐ข โ(๐ข, ๐ข)๐ฝ+๐ +๐/2โ๐1 โ
$โ(๐ข, ๐1 )โ๐ฝ $ ๐ท1 ๐๐ผ ( ) (1/2) ๐ถ ๐ฝ/2 ๐ฝ/2 (1) = (1) ๐น (๐1 , ๐1 ) (1/2) 2 1 ๐ฝ + ๐ + ๐/2 ๐ ๐๐ผ =
โซ
1
๐ฝ+๐ +๐/2
โค
๐ถ (1/2) โ
๐ถ (1) (1)
(1/2)
๐๐ฝ+๐ +๐/2 ๐๐ผ
,
where ๐ถ
(1)
:= sup
๐1 โ๐ท1
(1) 2 ๐น1
(
) ๐ฝ/2 ๐ฝ/2 (๐1 , ๐1 ) < +โ ๐ฝ + ๐ + ๐/2
48
J. Arazy and H. Upmeier
since our assumption on the parameters implies
Every ฮฆ โ ๐๐,m
๐ฝ ๐ฝ ๐ ๐ ๐ + โ (๐ฝ + + ๐ ) = โ๐ โ < โ(๐ โ 1) . 2 2 2 2 2 has a representation โซ ฮฆ = ๐๐(๐) ๐ด๐ ๐บ
for a ๏ฌnite complex measure ๐ on ๐บ. Then โซ ๐๐๐+๐ฝ ฮฆ = ๐๐(๐) ๐๐๐+๐ฝ ๐ด๐ ๐บ
and the previous calculation shows โฅ๐๐๐+๐ฝ ฮฆโฅโ1๐ฝ+๐/2 โค โฅ๐โฅ โ
sup โฅ๐๐๐+๐ฝ ๐ด๐ โฅโ1๐ฝ+๐/2 ๐โ๐บ
โค
1/2 ๐s 1/2 (๐/๐)s
๐๐ฝ+๐/2 (1) (1/2) ๐๐ฝ+๐ +๐/2 ๐๐ผ
๐ถ (1/2) ๐ถ (1) โ
โฅ๐โฅ.
It follows that ๐๐๐+๐ฝ ฮฆ โ โ1๐ฝ+๐/2 , as asserted. Thus we obtain the implication ฮฆ โ ๐๐,s =โ ๐๐๐+๐ฝ ฮฆ โ โ1๐ฝ+๐/2 . The converse implication follows from Theorem 4.2, applied to the partition s = (๐ , . . . , ๐ ). โก
References [A1] [A2] [AF] [AFP] [AU1] [AU2] [AU3]
[AU4]
J. Arazy, A survey of invariant Hilbert spaces of analytic functions on bounded symmetric domains, Contemp. Math. 185 (1995), 7โ65. J. Arazy, Boundedness and compactness of generalized Hankel operators on bounded symmetric domains, J. Funct. Anal. 137 (1996), 97โ151. J. Arazy, S. Fisher, Some aspects of the minimal Mยจ obius invariant space of analytic functions on the unit disk, Springer Lect. Notes in Math. 1070 (1984), 24โ44. J. Arazy, S. Fisher and J. Peetre, Mยจ obius invariant function spaces, J. reine angew. Math. 363 (1985), 110โ145. J. Arazy and H. Upmeier, Invariant inner products in spaces of holomorphic functions on bounded symmetric domains, Documenta Math 2 (1997), 213โ261. J. Arazy and H. Upmeier, Boundary measures for symmetric domains and integral formulas for the discrete Wallach points, Int. Equ. Op. Th. 47 (2003), 375โ434. J. Arazy and H. Upmeier, Jordan Grassmann manifolds and intertwining operators for weighted Bergman spaces, Proceedings Cluj-Napoca (2007), 25โ53, ClujUniversity Press 2008. J. Arazy and H. Upmeier, Intertwining operators and invariant function spaces for pole set parameters, (in preparation).
Minimal and Maximal Invariant Spaces [EZ]
49
M. Englis and G. Zhang, On the Faraut-Korยด anyi hypergeometric function in rank two, Ann. Inst. Fourier 54 (2004), 1855โ1875. [FK1] J. Faraut and A. Koranyi, Function spaces and reproducing kernels on bounded symmetric domains, J. Funct. Anal. 88 (1990), 64โ89. [FK2] J. Faraut and A. Koranyi, Analysis on Symmetric Cones, Clarendon Press, Oxford (1994). [LA] M. Lassalle, Alg`ebres de Jordan et ensemble de Wallach, Invent. Math. 89 (1987), 375โ393. [L1] O. Loos, Jordan Pairs, Springer Lect. Notes in Math. 460 (1975). [L2] O. Loos, Bounded Symmetric Domains and Jordan Pairs, Univ. of California, Irvine (1977). [RT] L. Rubel and R. Timoney, An extremal property of the Bloch space, Proc. Amer. Math. Soc. 75 (1979), 45โ49. [RV] H. Rossi, M. Vergne, Analytic continuation of holomorphic discrete series of a semi-simple Lie group, Acta Math. 136 (1975), 1โ59. [S] W. Schmid, Die Randwerte holomorpher Funktionen auf hermitesch symmetrischen Rยจ aumen, Invent. Math. 9 (1969), 61โ80. [U1] H. Upmeier, Jordan algebras and harmonic analysis on symmetric spaces, Amer. J. Math. 108 (1986), 1โ25. [U2] H. Upmeier, Jordan Algebras in Analysis, Operator Theory and Quantum Mechanics, CBMS Series in Math. 67, Amer. Math. Soc. (1987). [W] N. Wallach, The analytic continuation of the discrete series I, II, Trans. Amer. Math. Soc. 251 (1979), 1โ17 and 19โ37. [Y] Z. Yan, A class of generalized hypergeometric functions in several variables, Can. J. Math. 44 (1992), 1317โ1338. Jonathan Arazy Department of Mathematics University of Haifa Haifa 31905, Israel e-mail:
[email protected] Harald Upmeier Fachbereich Mathematik Universitยจ at Marburg D-35032 Marburg, Germany e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 51โ73 c 2012 Springer Basel AG โ
B-regular ๐ฑ -inner Matrix-valued Functions Damir Z. Arov and Harry Dym Dedicated to the memory of our valued teacher, colleague and friend, Israel Gohberg ZโL.
Abstract. In the study of the class ๐ฐ(๐ฝ) of mvfโs (matrix-valued functions) that are ๐ฝ-inner with respect to the open upper half-plane โ+ and a given signature matrix ๐ฝ, special roles are played by the classes ๐ฐโ๐
(๐ฝ), ๐ฐ๐๐
(๐ฝ), ๐ฐโ๐ ๐
(๐ฝ) and ๐ฐ๐๐ ๐
(๐ฝ) of left regular, right regular, left strongly regular and right strongly regular ๐ฝ-inner mvfโs. These are discussed at length in [ArD08] and the references cited therein. Shorter introductions may be found in the survey articles [ArD05] and [ArD07]. In particular, these classes are characterized in terms of the RKHSโs (reproducing kernel Hilbert spaces) โ(๐ ) that are associated with each mvf ๐ โ ๐ฐ(๐ฝ). If ๐ = ๐1 ๐2 with ๐1 , ๐2 โ ๐ฐ(๐ฝ), then, by a theorem of L. de Branges, the RKHS โ(๐1 ) is contractively included in โ(๐ ); necessary and su๏ฌcient conditions for isometric inclusion are also given. In this paper we introduce the class ๐ฐ๐ต๐
(๐ฝ) of ๐ต-regular ๐ฝ-inner mvfโs. It is characterized by the fact that if ๐ = ๐1 ๐2 with ๐1 , ๐2 โ ๐ฐ(๐ฝ), then โ(๐1 ) is isometrically included in โ(๐ ). If ๐ โ ๐ฐ(๐ฝ) is the characteristic mvf of a Livsic-Brodskii operator node, i.e., if ๐ is holomorphic at the point ๐ = 0 and normalized by ๐ (0) = ๐ผ๐ , then, thanks to another theorem of L. de Branges, ๐ โ ๐ฐ๐ต๐
(๐ฝ) if and only if every normalized left divisor ๐1 of ๐ โ ๐ฐ(๐ฝ) is left regular in the Brodskii sense. We shall show that ๐ฐโ๐ ๐
(๐ฝ) โช ๐ฐ๐๐ ๐
(๐ฝ) โ ๐ฐ๐ต๐
(๐ฝ). We shall also discuss the inverse monodromy problem for canonical di๏ฌerential systems for monodromy matrices ๐ โ ๐ฐ๐ต๐
(๐ฝ) and shall present an example of a 2 ร 2 canonical di๏ฌerential system for which the matrizant (fundamental solution) ๐๐ฅ (๐) belongs to the class ๐ฐโ๐ ๐
(๐ฝ) for every ๐ฅ > 0, but does not belong to the class ๐ฐ๐๐ ๐
(๐ฝ). Mathematics Subject Classi๏ฌcation (2000). Primary 47B32, 46E22, 47A48; Secondary 93C15, 45xx. Keywords. Canonical systems, de Branges spaces, ๐ฝ-inner matrix-valued functions, reproducing kernel Hilbert spaces, Livsic-Brodskii nodes.
52
D.Z. Arov and H. Dym
1. Introduction The class ๐ฐ(๐ฝ) of ๐ฝ-inner mvfโs (matrix-valued functions) with respect to the open upper half-plane โ+ = {๐ โ โ : โ๐ > 0} is the set of meromorphic ๐ ร ๐ mvfโs ๐ (๐) in โ+ that are ๐ฝ-contractive on the set ๐ฅ+ ๐ = {๐ โ โ+ : at which ๐ (๐) is holomorphic} and have nontangential limits a.e. on the real axis โ that are ๐ฝ-unitary, i.e., and
๐ (๐)๐ฝ๐ (๐)โ โค ๐ฝ
for ๐ โ ๐ฅ+ ๐
๐ (๐)๐ฝ๐ (๐)โ = ๐ฝ
a.e. on โ,
(1.1) (1.2) โ
โ
respectively. Here ๐ฝ denotes an ๐ร๐ signature matrix, i.e., ๐ฝ = ๐ฝ and ๐ฝ ๐ฝ = ๐ผ๐ . The signature matrices ยฑ๐ผ๐ , ยฑ๐ฝ๐ and ยฑ๐ฅ๐ , where [ ] [ ] 0 โ๐ผ๐ 0 โ๐๐ผ๐ ๐ฝ๐ = and ๐ฅ๐ = , โ๐ผ๐ 0 ๐๐ผ๐ 0 will be of particular interest. The existence of nontangential boundary values a.e. on โ is a consequence of the fact that (1.1) guarantees that ๐ โ ๐ฐ(๐ฝ) belongs to the class ๐ฉ ๐ร๐ of ๐ ร ๐ mvfโs that are meromorphic in โ+ with bounded Nevanlinna characteristic there. Moreover, every such ๐ has a meromorphic pseudocontinuation with bounded Nevanlinna characteristic in the open lower half-plane โโ that may be de๏ฌned on the set + ฮฉโ and det ๐ (๐) โ= 0}. ๐ = {๐ โ โโ : ๐ โ ๐ฅ๐ by the formula
๐โ (๐) = ๐ฝ(๐ # (๐))โ1 ๐ฝ
for ๐ โ ฮฉโ ๐,
(1.3)
where ๐ # (๐) = ๐ (๐)โ
and, as will be needed later, ๐ โผ (๐) = ๐ (โ๐)โ .
(1.4)
Formulas (1.2) and (1.3) serve to guarantee that the nontangential boundary values of ๐ and ๐โ coincide a.e. on โ, i.e., ๐ (๐) = lim ๐ (๐ + ๐๐) = lim ๐โ (๐ โ ๐๐), ๐โ0
๐โ0
(1.5)
and hence that ๐โ is a pseudocontinuation of ๐ . From now on mvfโs ๐ โ ๐ฐ(๐ฝ) will be considered in the set โ 0 ๐ฅ๐ = ๐ฅ+ ๐ โช ๐ฅ๐ โช ๐ฅ๐ ,
where and
๐ฅโ ๐ = {๐ โ โโ : at which ๐ (๐) is holomorphic} ๐ฅ0๐ = {๐ โ โ : at which ๐ (๐) is holomorphic}.
B-regular ๐ฝ-inner Matrix-valued Functions
53
If ๐ โ ๐ฐ(๐ฝ) then the formula โง โ ๏ฃด โจ ๐ฝ โ ๐ (๐)๐ฝ๐ (๐) if ๐ โ= ๐ โ2๐๐(๐ โ ๐) ๐พ๐๐ (๐) = (1.6) โฒ โ ๏ฃด โฉ ๐ (๐๐ฝ๐ (๐) if ๐ = ๐ 2๐๐ de๏ฌnes a positive kernel on ๐ฅ๐ ร ๐ฅ๐ . Therefore, by the matrix version of a theorem of Aronszajn, there is an RKHS (reproducing kernel Hilbert space) โ(๐ ) with ๐พ๐๐ (๐) as its RK (reproducing kernel). This means that the following two conditions are met: (1) ๐พ๐๐ ๐ฃ โ โ(๐ ) for every choice of ๐ โ ๐ฅ๐ and ๐ฃ โ โ๐ . (2) If ๐ โ โ(๐ ), then ๐ฃ โ ๐ (๐) = โจ๐, ๐๐๐ ๐ฃโฉโ(๐)
for every ๐ โ โ(๐ ), ๐ โ ๐ฅ๐ and ๐ฃ โ โ๐ . ๐ฅ+ ๐
(1.7)
๐ฅโ ๐
The restrictions ๐+ and ๐โ of ๐ โ โ(๐ ) to and are holomorphic with bounded Nevanlinna characteristic in โ+ and โโ , respectively. Moreover (as shown in Theorem 5.49 of [ArD08]), ๐โ is the pseudocontinuation of ๐+ . Thus, if ๐ โ โ(๐ ), then ๐ (๐) = ๐+ (๐) = lim ๐ (๐ + ๐๐) = lim ๐โ (๐ โ ๐๐) = ๐โ (๐) a.e. on โ. ๐โ0
๐โ0
A mvf ๐ โ ๐ฐ(๐ฝ) belongs to the class ๐ฐ๐๐ ๐
(๐ฝ) of right strongly regular ๐ฝ-inner mvfโs if the nontangential boundary value ๐ (๐) belongs to ๐ฟ๐ 2 (โ) for every ๐ โ โ(๐ ). Thus, upon identifying ๐ โ โ(๐ ) with its boundary values, this can be reexpressed as ๐ฐ๐๐ ๐
(๐ฝ) = {๐ โ ๐ฐ(๐ฝ) : โ(๐ ) โ ๐ฟ๐ (1.8) 2 (โ)}. The class ๐ฐโ๐ ๐
(๐ฝ) of left strongly regular ๐ฝ-inner mvfโs may be de๏ฌned as ๐ฐโ๐ ๐
(๐ฝ) = {๐ โ ๐ฐ(๐ฝ) : ๐ โผ โ ๐ฐ๐๐ ๐
(๐ฝ)}.
(1.9)
A mvf ๐ โ ๐ฐ(๐ฝ) belongs to the class ๐ฐ๐ต๐
(๐ฝ) of B-regular ๐ฝ-inner mvfโs if for every factorization ๐ = ๐1 ๐2 with ๐1 , ๐2 โ ๐ฐ(๐ฝ) (1.10) the equality โ(๐1 ) โฉ ๐1 โ(๐2 ) = {0}
(1.11)
is in force. The importance of the class ๐ฐ๐ต๐
(๐ฝ) of B-regular ๐ฝ-inner mvfโs is exhibited by the following two theorems of L. de Branges that correspond to Theorems 5.52 and 5.50 in [ArD08]. The formulations there were in๏ฌuenced by the discussion in Section 5 of [AlD84], which in turn is based on [Br63] and [Br65]. Theorem 1.1. (L. de Branges) Let ๐ = ๐1 ๐2 , where ๐1 , ๐2 โ ๐ฐ(๐ฝ). Then โ(๐1 ) is contained contractively in โ(๐ ), i.e., โ(๐1 ) โ โ(๐ )
(as vector spaces)
54
D.Z. Arov and H. Dym
and โฅ๐ โฅโ(๐) โค โฅ๐ โฅโ(๐1 )
for ๐ โ โ(๐1 ).
Moreover, this inclusion is isometric if and only if (1.11) holds. Furthermore, โ(๐1 ) โฉ ๐1 โ(๐2 ) = {0} โโ โ(๐ ) = โ(๐1 ) โ ๐1 โ(๐2 ).
(1.12)
Theorem 1.2. (L. de Branges) Let ๐ โ ๐ฐ(๐ฝ) and let โ be a closed subspace of โ(๐ ) that is ๐
๐ผ invariant for every point ๐ผ โ ๐ฅ๐ . Then there exists a mvf ๐1 โ ๐ฐ(๐ฝ) such that ๐ฅ๐1 โ ๐ฅ๐ , โ = โ(๐1 ) and ๐1โ1 ๐ โ ๐ฐ (๐ฝ). Moreover, the space โ(๐1 ) is isometrically included in โ(๐ ), and โ(๐ ) = โ(๐1 ) โ ๐1 โ(๐2 ),
where
๐2 = ๐1โ1 ๐.
(1.13)
Remark 1.3. If ๐ โ โฐ โฉ ๐ฐ โ (๐ฝ), then โฅ๐
0๐ โฅ1/๐ tends to zero as ๐ โ โ since ๐
0 is a Volterra operator and therefore the identity ๐
๐ผ โ ๐
0 = ๐ผ๐
0 ๐
๐ผ =โ ๐
๐ผ =
โ โ
๐ผ๐โ1 ๐
0๐ .
๐=1
Thus, for such mvfโs ๐ a closed subspace โ of โ(๐ ) is invariant for every point ๐ผ โ โ if and only if it is invariant under ๐
0 . A simple example of a continuous family of mvfโs ๐๐ โโ ๐ฐ๐ต๐
(๐ฝ) may be constructed by ๏ฌxing a matrix ๐ โ โ๐ร๐ such that ๐ โ ๐ = ๐ผ๐ and ๐ โ ๐ฝ๐ โ = 0 and setting ๐๐ (๐) = exp{๐๐๐ ๐ ๐ โ ๐ฝ} = ๐ผ๐ + ๐๐๐ ๐ ๐ โ ๐ฝ
for ๐ โฅ 0.
Then ๐๐ โ โฐ โฉ ๐ฐ(๐ฝ) for every ๐ โฅ 0, ๐๐ ๐๐ก = ๐๐ +๐ก and, as follows readily from formula (1.6), the RK of the RKHS โ(๐๐ ) is given by the formula ๐พ๐๐๐ (๐) =
๐ ๐ ๐ โ, 2๐
which serves to exhibit โ(๐๐ ) as the ๐-dimensional subspace of โ๐ spanned by the columns of ๐ for every ๐ > 0. Thus, the spaces โ(๐๐ ) are all the same as vector spaces for ๐ > 0. However, the norms depend upon ๐ and ( )2 ( )2 2๐ ๐ก 2๐ โฅ๐ ๐ฅโฅ2โ(๐๐ก ) = โฅ ๐ ๐ โ ๐ฅโฅ2โ(๐๐ก ) = โฅ๐พ๐๐๐ก (๐ ๐ฅ)โฅ2โ(๐๐ก ) ๐ก 2๐ ๐ก ( )2 2๐ 2๐ โ = ๐ฅโ ๐ โ ๐พ๐๐๐ก (๐)๐ ๐ฅ = ๐ฅ ๐ฅ ๐ก ๐ก 2๐ โ < ๐ฅ ๐ฅ = โฅ๐ ๐ฅโฅ2โ(๐๐ ) for 0 < ๐ < ๐ก. ๐ An example of a canonical di๏ฌerential system with matrizant ๐๐ฅ (๐) that belongs to ๐ฐโ๐ ๐
(๐ฝ) but does not belong to ๐ฐ๐๐ ๐
(๐ฝ) will be furnished in Section 8.
B-regular ๐ฝ-inner Matrix-valued Functions
55
2. A unitary operator from ํ(๐ผ ) onto ํ(๐ผ โผ ) It is important to keep in mind that ๐ โ ๐ฐ(๐ฝ) โโ ๐ โผ โ ๐ฐ(๐ฝ). Lemma 2.1. Let ๐ โ ๐ฐ(๐ฝ) and let ๐ be the operator de๏ฌned on โ(๐ ) by the formula (๐ ๐ )(๐) = ๐ โผ (๐)๐ฝ๐ (โ๐) for ๐ โ ๐ฅ๐ โฉ ๐ฅ๐ โผ . (2.1) Then ๐ is a unitary operator from โ(๐ ) onto โ(๐ โผ ). Proof. Let ๐, ๐ โ ๐ฅ๐ โผ , โ๐, โ๐ โ ๐ฅ๐ and suppose that ๐ โ= ๐, det ๐ โผ (๐) โ= 0 and det ๐ โผ (๐) โ= 0. Then โผ
๐ ๐พ๐๐ (๐) = ๐ โผ (๐)๐ฝ๐พโ๐ (โ๐)๐ฝ๐ โผ (๐)โ .
(2.2)
Therefore, the operator ๐ maps the dense subspace โ1 of vvfโs ๐ โ โ(๐ ) of the form ๐ โ ๐ ๐ (๐) = ๐พโ๐ (๐)๐ฝ๐ โผ (๐๐ )โ ๐๐ with ๐๐ โ โ๐ and ๐ โฅ 1 (2.3) ๐ ๐=1
into the dense subspace โ2 of vvfโs ๐(๐) = (๐ ๐ )(๐) = ๐ โผ (๐)๐ฝ
๐ โ
๐ ๐พโ๐ (โ๐)๐ฝ๐ โผ (๐๐ )โ ๐๐ ๐
๐=1
=
๐ โ
โผ
๐พ๐๐๐
(๐)๐๐
(2.4)
with ๐๐ โ โ
๐
and ๐ โฅ 1.
๐=1
Moreover, if ๐ and ๐ are de๏ฌned by the above formulas, then โจ๐, ๐ โฉโ(๐) = =
๐ โ ๐ โ ๐=1 ๐=1 ๐ โ ๐ โ ๐=1 ๐=1
๐ ๐๐โ ๐ โผ (๐๐ )๐ฝ๐พโ๐ (โ๐๐ )๐ฝ๐ โผ (๐๐ )โ ๐๐ ๐
(2.5) โผ ๐๐โ ๐พ๐๐๐ (๐๐ )๐๐
= โจ๐, ๐โฉโ(๐ โผ ) .
Thus, ๐ maps โ1 isometrically onto โ2 . Moreover, if ๐ โ โ(๐ ), then there exists a sequence of vvfโs ๐๐ โ โ1 such that โฅ๐ โ ๐๐ โฅโ(๐) โ 0 as ๐ โ โ. But, as โ(๐ ) is a RKHS, this implies that ๐๐ (๐) โ ๐ (๐) at each point ๐ โ ๐ฅ๐ as ๐ โ โ. Thus, if ๐๐ = ๐ ๐๐ for ๐ = 1, 2, . . . , then ๐๐ (๐) = (๐ ๐๐ )(๐) = ๐ โผ (๐)๐ฝ๐๐ (โ๐) โ ๐ โผ (๐)๐ฝ๐ (โ๐) for each point ๐ โ ๐ฅ๐ โผ such that โ๐ โ ๐ฅ๐ . Since โฅ๐๐ โฅโ(๐ โผ ) = โฅ๐๐ โฅโ(๐ ) โ โฅ๐ โฅโ(๐)
as ๐ โ โ
and โฅ๐๐ โ ๐๐ โฅโ(๐ โผ ) = โฅ๐๐ โ ๐๐ โฅโ(๐) ,
as ๐ โ โ
56
D.Z. Arov and H. Dym
there exists a ๐ โ โ(๐ โผ ) such that โฅ๐๐ โ ๐โฅโ(๐ โผ ) โ 0 as ๐ โ โ. Therefore, since โ(๐ โผ ) is a RKHS and โฉ ๐ฅ๐ , ๐ฅ๐ โผ = ๐โโ(๐ โผ )
๐๐ (๐) โ ๐(๐) at each point ๐ โ ๐ฅ๐ โผ as ๐ โ โ. Consequently, ๐(๐) = ๐ โผ (๐)๐ฝ๐ (โ๐) = (๐ ๐ )(๐) for ๐ โ โ(๐ ), i.e., ๐ maps โ(๐ ) into โ(๐ โผ ). Therefore, since ๐ is an isometry on the full space โ(๐ ) and โ2 is dense in โ(๐ โผ ), ๐ maps โ(๐ ) onto โ(๐ โผ ). โก Theorem 2.2. ๐ฐโ๐ ๐
(๐ฝ) = {๐ โ ๐ฐ(๐ฝ) : ๐ ๐ โ ๐ฟ๐ 2 (โ)
for every ๐ โ โ(๐ )}.
Proof. This follows from Lemma 2.1 and formulas (1.8) and (1.9). ๐ฟ๐ 2 (โ)
if and only if ๐(โ๐) belongs to Remark 2.3. Since ๐(๐) belongs to the equality (2.6) is equivalent to the following equality ๐ฐโ๐ ๐
(๐ฝ) = {๐ โ ๐ฐ(๐ฝ) : ๐ # ๐ฝ๐ โ ๐ฟ๐ 2 (โ) for every ๐ โ โ(๐ )} = {๐ โ ๐ฐ(๐ฝ) : ๐ โ1 ๐ โ ๐ฟ๐ 2 (โ) for every ๐ โ โ(๐ )}.
(2.6) โก ๐ฟ๐ 2 (โ), (2.7)
3. Some properties of the class ํค๐ฉ๐น (๐ฑ ) Theorem 3.1. ๐ โ ๐ฐ๐ต๐
(๐ฝ) โโ ๐ โผ โ ๐ฐ๐ต๐
(๐ฝ). Proof. If ๐ โ ๐ฐ๐ต๐
(๐ฝ) and ๐ = ๐1 ๐2 is a factorization of ๐ with factors ๐1 , ๐2 โ ๐ฐ(๐ฝ), then ๐ โผ = ๐2โผ ๐1โผ . Let ๐ โ โ(๐2โผ ) โฉ ๐2โผ โ(๐1โผ ). Then, by Lemma 2.1, ๐ (๐) = ๐2โผ (๐)๐ฝ๐2 (โ๐) = ๐2โผ (๐)๐1โผ (๐)๐ฝ๐1 (โ๐), where ๐๐ โ โ(๐๐ ) for ๐ = 1, 2. Therefore, ๐ฝ๐2 (โ๐) = ๐1โผ (๐)๐ฝ๐1 (โ๐), i.e.,
๐2 (๐) = ๐ฝ๐1# (๐)๐ฝ๐1 (๐) = ๐1 (๐)โ1 ๐1 (๐).
Thus,
๐1 = ๐1 ๐2 ,
and hence
๐1 โ โ(๐1 ) โฉ ๐1 โ(๐2 ) = {0}.
Consequently, ๐ = 0, i.e., โ(๐1 ) โฉ ๐1 โ(๐2 ) = {0} =โ โ(๐2โผ ) โฉ ๐2โผ โ(๐1โผ ) = {0}. The converse implication then follows from the fact that (๐ โผ )โผ = ๐ .
โก
B-regular ๐ฝ-inner Matrix-valued Functions
57
Theorem 3.2. The following two inclusions are in force: and
๐ฐ๐๐ ๐
(๐ฝ) โ ๐ฐ๐ต๐
(๐ฝ)
(3.1)
๐ฐโ๐ ๐
(๐ฝ) โ ๐ฐ๐ต๐
(๐ฝ)
(3.2)
Proof. The inclusion (3.1) follows from Theorem 1.1 and Theorems 5.50 and 5.92 in [ArD08]. The inclusion (3.2) follows from the characterization (1.9), the inclusion (3.1) and Theorem 3.1. โก Theorem 3.3. If ๐1 , . . . , ๐๐ โ ๐ฐ(๐ฝ) and ๐ = ๐1 โ
โ
โ
๐๐ , then ๐ โ ๐ฐ๐ต๐
(๐ฝ) =โ ๐๐ โ ๐ฐ๐ต๐
(๐ฝ)
for ๐ = 1, . . . , ๐.
Proof. It su๏ฌces to consider the case ๐ = 2. Then if ๐ = ๐1 ๐2 , ๐1 = ๐๐ ๐๐ , with ๐๐ , ๐๐ , ๐2 โ ๐ฐ(๐ฝ) and ๐ โ ๐ฐ๐ต๐
(๐ฝ), the two factorizations ๐ = ๐1 ๐2 and ๐ = ๐๐ (๐๐ ๐2 ) imply that โฅ๐ โฅโ(๐1 ) = โฅ๐ โฅโ(๐)
for every ๐ โ โ(๐1 )
โฅ๐ โฅโ(๐๐ ) = โฅ๐ โฅโ(๐)
for every ๐ โ โ(๐๐ ),
and respectively. Therefore, โฅ๐ โฅโ(๐๐ ) = โฅ๐ โฅโ(๐1 )
for every ๐ โ โ(๐๐ ),
which proves that ๐1 โ ๐ฐ๐ต๐
(๐ฝ). The proof that ๐2 โ ๐ฐ๐
๐ต (๐ฝ) follows from formula ๐ โผ = ๐2โผ ๐1โผ and Theorem 3.1. โก
4. Canonical systems with B-regular matrizants Let
๐ฐ โ (๐ฝ) = {๐ โ ๐ฐ(๐ฝ) : 0 โ ๐ฅ๐
and ๐ (0) = ๐ผ๐ }
and
โฐ โฉ ๐ฐ(๐ฝ) = {๐ โ ๐ฐ(๐ฝ) : ๐ is an entire mvf}. A family of ๐ ร ๐ mvfโs ๐๐ฅ (๐), 0 โค ๐ฅ < โ, that is continuous with respect to ๐ฅ on [0, โ) for each ๐ โ โ and meets the conditions ๐๐ฅโ1 ๐๐ฅ2 โ โฐ โฉ ๐ฐ โ (๐ฝ) when 0 โค ๐ฅ1 โค ๐ฅ2 < โ and ๐0 (๐) โก ๐ผ๐ 1
(4.1)
will be called a normalized monotonic continuous chain of entire ๐ฝ-inner mvf โs. It is well known that if ๐ (๐ฅ) is a continuous nondecreasing ๐ ร ๐ mvf on [0, โ) with ๐ (0) = 0, then the matrizant (fundamental solution) of the canonical integral system โซ ๐ฅ ๐๐ (๐)๐๐ (๐ )๐ฝ, 0 โค ๐ฅ < โ, (4.2) ๐๐ฅ (๐) = ๐ผ๐ + ๐๐ 0
58
D.Z. Arov and H. Dym
is normalized monotonic continuous chain of entire ๐ฝ-inner mvfโs. There is a converse statement in the class โ โฐ โฉ ๐ฐ๐ต๐
(๐ฝ) = โฐ โฉ ๐ฐ โ (๐ฝ) โฉ ๐ฐ๐ต๐
(๐ฝ)
that will be presented below in Theorem 4.1. If โซ ๐ฅ ๐ (๐ฅ) = ๐ป(๐ )๐๐ for ๐ฅ โ [0, โ) 0
and some ๐ ร ๐ mvf ๐ป that meets the conditions ๐ป โ ๐ฟ๐ร๐ 1,๐๐๐ ([0, โ)) and
๐ป(๐ฅ) โฅ 0 a.e. on [0, โ),
(4.3)
then the matrizant ๐๐ฅ (๐) is a solution of the canonical di๏ฌerential system โ๐๐ฅ (๐) = ๐๐๐๐ฅ (๐)๐ป(๐ฅ)๐ฝ for 0 โค ๐ฅ < โ, with ๐0 (๐) = ๐ผ๐ , (4.4) โ๐ฅ wherein the Hermitian ๐ป(๐ฅ) is subject to (4.3). From time to time we shall also impose the normalization trace ๐ป(๐ฅ) = 1
a.e. on [0, โ].
(4.5)
Theorem 4.1. Each normalized monotonic continuous chain ๐๐ฅ (๐), 0 โค ๐ฅ < โ, of entire B-regular ๐ฝ-inner mvfrโs is the matrizant of exactly one canonical integral system (4.2) with a continuous nondecreasing mass function ๐ (๐ฅ), 0 โค ๐ฅ < โ, with ๐ (0) = 0 that may be obtained from ๐๐ฅ (๐) by the formula ( ) โ๐๐ฅ ๐ (๐ฅ) = โ๐ (0)๐ฝ. (4.6) โ๐ Proof. This follows from the de๏ฌnition of the class ๐ฐ๐ต๐
(๐ฝ) and Theorem 4.6 in [ArD97]. โก Theorem 4.2. Each normalized monotonic continuous chain ๐๐ฅ (๐), 0 โค ๐ฅ < โ, of entire right or left strongly regular ๐ฝ-inner mvf โs is the matrizant of exactly one canonical integral system (4.2) with a continuous nondecreasing mass function ๐ (๐ฅ), 0 โค ๐ฅ < โ, with ๐ (0) = 0 that may be obtained from ๐๐ฅ (๐) by the formula (4.6). Proof. This is an immediate consequence of Theorems 3.2 and 4.1.
โก
5. Direct and inverse monodromy problems A canonical integral system (4.1) is said to be a regular integral system if โ < โ and the mass function ๐ (๐ฅ) is a continuous non decreasing ๐ ร ๐ mvf on the closed interval [0, โ] with ๐ (0) = 0. In this case the matrizant ๐๐ฅ (๐), 0 โค ๐ฅ โค โ is a normalized monotonic continuous chain of entire ๐ฝ-inner mvfโs on the interval [0, โ] and the value ๐ (๐) = ๐โ (๐) of the matrizant at the right-hand end point โ of the interval is called the monodromy matrix.
B-regular ๐ฝ-inner Matrix-valued Functions
59
Similarly, a canonical di๏ฌerential system (4.4) is said to be a regular di๏ฌerential system if โ < โ and the Hermitian ๐ป(๐ฅ) meets the conditions ๐ป โ ๐ฟ๐ร๐ ([0, โ]) and 1
๐ป(๐ฅ) โฅ 0 a.e. on [0, โ].
(5.1)
The matrizant ๐๐ฅ (๐), 0 โค ๐ฅ โค โ, of such a system is a normalized monotonic continuous chain of entire ๐ฝ-inner mvfโs on the interval [0, โ] that is in fact absolutely continuous with respect to ๐ฅ on [0, โ]. The value ๐โ (๐) of the matrizant at the right-hand end point of the interval is called the monodromy matrix of the system. It is clear that the monodromy matrices of regular canonical integral and di๏ฌerential systems belong to the class โฐ โฉ ๐ฐ โ (๐ฝ). Moreover, by elementary estimates it may also be shown that they are of exponential type. A converse to these results was obtained by V.P. Potapov [Po60] as an application of his work on the multiplicative representation of meromorphic ๐ฝ-contractive mvfโs ๐ (๐) in โ+ with det ๐ (๐) โ= 0 for some ๐ โ ๐ฅ+ ๐. Theorem 5.1. (V.P. Potapov) If ๐ โ โฐโฉ๐ฐ โ (๐ฝ), then ๐ (๐) is the monodromy matrix of a regular canonical di๏ฌerential system on the interval [0, โ] with a Hermitian ๐ป(๐ฅ) that meets the conditions (4.5) and (5.1). Moreover, the length of this interval is uniquely speci๏ฌed by the formula [ ( ) ] โ๐ โ = โ๐ = trace โ๐ (0)๐ฝ . (5.2) โ๐ Remark 5.2. A mvf ๐ โ โฐ โฉ๐ฐ โ (๐ฝ) is automatically of exponential type. This follows from the fact that such a mvf ๐ has bounded Nevanlinna characteristic in both โ+ and โโ and a theorem of M.G. Krein; see, e.g., Theorem 3.108 in [ArD08] for the latter. In general, a mvf ๐ โ โฐ โฉ ๐ฐ โ (๐ฝ) is the monodromy matrix of more than one canonical di๏ฌerential system, i.e., it is not possible to recover ๐ป(๐ฅ) uniquely from ๐ (๐). However, if ๐ฝ = ยฑ๐ผ๐ , then ๐ is the monodromy matrix of exactly one canonical di๏ฌerential system subject to the normalization conditions (5.1) and (4.5) if and only if ๐ (๐) and det ๐ (๐) have the same exponential type. This criterion is due to Brodskii-Kisilevskii; see, e.g., [Bro72]. Let ๐๐ยฑ = lim sup ๐ โ1 ln โฅ๐ (ยฑ๐๐)โฅ ๐โโ
for entire mvfโs ๐ . โ โ Theorem 5.3. If ๐ โ โฐ โฉ (๐ฐ๐๐ ๐
(๐ฝ) โช ๐ฐโ๐ ๐
(๐ฝ)), then ๐ is the monodromy matrix of exactly one canonical di๏ฌerential system (4.4) with Hermitian ๐ป(๐ฅ) subject to the constraints (5.1) and (4.5) if and only if either
or
๐๐+ โค 0
and
โ ๐๐โ = ๐det ๐
๐๐โ โค 0
and
+ ๐๐+ = ๐det ๐.
Proof. A proof will be presented in Section 8.5 in [ArD12].
โก
60
D.Z. Arov and H. Dym
Remark 5.4. Theorem 5.3 is a generalization of the Brodskii-Kisilevski criterion, since by a theorem of M.G. Krein (see, e.g., Theorem 3.108 in [ArD08]) the exponential type of a mvf ๐ โ โฐ โฉ ๐ฐ โ (๐ฝ) is equal to max{๐๐+ , ๐๐โ }. A fundamental result of L. de Branges serves to establish uniqueness of the inverse monodromy problem for 2 ร 2 monodromy matrices in โฐ โฉ ๐ฐ(๐ฅ1 ). Theorem 5.5. (L. de Branges) If ๐ โ โฐ โฉ ๐ฐ โ (๐ฅ1 ) and det ๐ (๐) โก 1, then it is the monodromy matrix of exactly one canonical di๏ฌerential system with a real Hermitian ๐ป that is subject to the constraints (5.1) and (4.5). Proof. See [Br68a] and, for additional information,[DMc76].
โก
A mvf ๐ โ ๐ฐ(๐ฝ) is said to be symplectic if ๐ (๐)๐ ๐ฅ๐ ๐ (๐) = ๐ฅ๐
when ๐ โ ๐ฅ๐ .
๐
Here ๐ (๐) denotes the transpose of ๐ (๐). If ๐ โ โฐ โฉ ๐ฐ(๐ฝ) is symplectic, then ๐๐+ = ๐๐โ . If a 2 ร 2 mvf ๐ โ โฐ โฉ ๐ฐ(๐ฝ), then ๐ is symplectic โโ det ๐ (๐) โก 1 โโ ๐๐+ = ๐๐โ . A mvf ๐ โ โฐ โฉ ๐ฐ โ (๐ฝ) is said to be unicellular if every pair of left divisors ๐1 , ๐2 โ โฐ โฉ ๐ฐ โ (๐ฝ) of ๐ is ordered in the sense that either ๐1โ1 ๐2 โ โฐ โฉ ๐ฐ โ (๐ฝ)
or ๐2โ1 ๐1 โ โฐ โฉ ๐ฐ โ (๐ฝ).
It is readily seen that a mvf ๐ โ โฐ โฉ ๐ฐ โ (๐ฝ) is unicellular if and only if ๐ โผ is unicellular. The matrizant of a regular canonical integral system (4.2) with monodromy matrix ๐ is a family of ordered left divisors of ๐ that is maximal in a natural way. โ (๐ฝ), then any maximal family of ordered normalized Theorem 5.6. If ๐ โ โฐ โฉ ๐ฐ๐ต๐
left divisors of ๐ may be parametrized in such a way that it is the matrizant ๐๐ฅ (๐), 0 โค ๐ฅ โค โ๐ of a canonical di๏ฌerential system (4.4) with Hermitian ๐ป(๐ฅ) that meets the constraints (5.1) and (4.5).
Proof. Let ๐ฑ be a maximal family of ordered normalized left divisors of ๐ and for each ๐ โ ๐ฑ, let let ๐๐ฅ (๐) = ๐ (๐), where ๐ฅ = โ๐ = โ๐trace๐ โฒ (0)๐ฝ. Then 0 โค ๐ฅ โค โ๐ and the mvfโs ๐๐ฅ (๐) that are obtained by this parametrization satisfy the conditions of Theorem 4.1. Therefore, the conclusions of this theorem are valid. โก โ Theorem 5.7. If ๐ โ โฐ โฉ ๐ฐ๐ต๐
(๐ฝ), then the following assertions are equivalent: (1) ๐ is unicellular.
(2) ๐ is the monodromy matrix of exactly one canonical di๏ฌerential system (4.4) with Hermitian ๐ป(๐ฅ) that meets the constraints (5.1) and (4.5). (3) ๐ โผ is the monodromy matrix of exactly one canonical di๏ฌerential system (4.4) with Hermitian ๐ป(๐ฅ) that meets the constraints (5.1) and (4.5).
B-regular ๐ฝ-inner Matrix-valued Functions
61
Proof. This follows from Theorem 4.1 and the fact that any maximal monotone family of left divisors of ๐ in the class in ๐ฐ โ (๐ฝ) is a normalized monotonic continuous chain of entire ๐ฝ-inner mvfโs. In view of this (2) holds if and only if (1) holds. Finally, (3) is immediate from (2) and Theorem 3.1. โก
6. Connections with the theory of characteristic functions A mvf ๐ โ โฐ โฉ ๐ฐ โ (๐ฝ) may be identi๏ฌed as the characteristic mvf of an LB (LivsicBrodskii) Volterra ๐ฝ-node ฮฃ = (๐พ, ๐น ; ๐, โ๐ ; ๐ฝ): ๐ (๐) = ๐ฮฃ (๐) = ๐ผ๐ + ๐๐๐น (๐ผ โ ๐๐พ)โ1 ๐น โ ๐ฝ,
(6.1)
where ๐พ, the main operator of the node, is a Volterra operator in the Hilbert space ๐, ๐น is a bounded linear operator from ๐ into โ๐ , ๐ฝ is an ๐ ร ๐ signature matrix and ๐พ โ ๐พ โ = ๐๐น โ ๐ฝ๐น. (6.2) An LB Volterra ๐ฝ-node may be chosen to be simple, which means that โฉ ker ๐น ๐พ ๐ = {0}.
(6.3)
๐โฅ0
It is known that
โฉ
๐น ๐พ ๐ = {0} โโ ker ๐พ โฉ ker ๐น = {0};
(6.4)
๐โฅ0
and that if ฮฃ๐ = (๐พ๐ , ๐น๐ ; ๐๐ , โ๐ ; ๐ฝ), ๐ = 1, 2, is a pair of simple LB Volterra ๐ฝ-nodes with the same characteristic mvf ๐ , then they are unitarily equivalent, i.e., there exists a unitary operator ๐ from ๐1 onto ๐2 such that ๐พ2 = ๐ ๐พ1 ๐ โ1
and ๐น2 = ๐น1 ๐ โ1 ;
(6.5)
see, e.g., [Bro72] for details and additional information on the connections between the characteristic mvfโs of LB Volterra ๐ฝ-nodes and entire ๐ฝ-inner mvfโs. Remark 6.1. In this paper we focus on the class โฐ โฉ ๐ฐ โ (๐ฝ) because the matrizants of the canonical systems that we study belong to this class. However, it is also possible to characterize the larger class ๐ฐ โ (๐ฝ): A mvf ๐ โ ๐ฐ โ (๐ฝ) if and only if it is the characteristic mvf of a simple Livsic-Brodskii node ฮฃ = (๐พ, ๐น ; ๐, โ๐ ; ๐ฝ) for which the real part ๐พ๐
= (๐พ + ๐พ โ )/2 of the main operator ๐พ is a bounded selfadjoint operator with singular spectrum, i.e., โซ ๐ ๐พ๐
= ๐๐๐ธ๐ and ๐๐ฅ (๐) = โจ๐ธ๐ข ๐ฅ, ๐ฅโฉ are singular functions of ๐ (6.6) ๐
for every ๐ฅ โ ๐, where [๐, ๐] is a ๏ฌnite interval in โ. (Since โ ๐ ๐พ๐
๐น โ๐ = ๐, ๐โฅ0
62
D.Z. Arov and H. Dym
it is actually enough to require that (6.6) holds for every ๐ฅ โ ๐น โ๐ .) An equivalent requirement is that the mvf โซ ๐ ๐๐ธ๐ โ โ1 โ ๐(๐) = โ๐๐น (๐พ๐
โ ๐๐ผ) ๐น = โ๐๐น ๐น ๐ โ๐ ๐ (which belongs to the Carathยดeodory class) is purely singular, i.e., the nontangential limit ๐(๐) + ๐(๐)โ = 0 a.e. on โ; see, e.g., Lemma 6.3 and Theorem 6.4 in [ArD08] and pp. 28โ30 of [Bro72] for the last identi๏ฌcation. One basic model of a simple LB Volterra ๐ฝ-node with characteristic mvf ๐ โ โฐ โฉ ๐ฐ โ (๐ฝ) is due to L. de Branges: ฮฃ๐๐๐ (๐ ) = (๐
0 , ๐น0 ; โ(๐ ), โ๐ ; ๐ฝ), in which
โ ๐ (๐) โ ๐ (0) and ๐น0 ๐ = 2๐๐ (0) for ๐ โ โ(๐ ). (6.7) ๐ The veri๏ฌcation of simplicity is easily carried out with the help of the identities (๐
0 ๐ )(๐) =
โ ๐ (๐) (0) 2๐ for ๐ โ โ(๐ ) and ๐ = 0, 1, . . .. (6.8) ๐! Let ฮฃ1 = (๐พ1 , ๐น1 ; ๐1 , โ๐ ; ๐ฝ) and ฮฃ2 = (๐พ2 , ๐น2 ; ๐2 , โ๐ ; ๐ฝ) be a pair of LB Volterra ๐ฝ-nodes and let [ ] ๐พ1 ๐๐น1โ ๐ฝ๐น2 ๐พ= , ๐น = [๐น1 ๐น2 ] and ๐ = ๐1 โ ๐2 . 0 ๐พ2 ๐น0 ๐พ0๐ ๐ = (๐
๐๐ ๐ )(0) =
Then ฮฃ = (๐พ, ๐น ; ๐, ๐ถ ๐ ; ๐ฝ) is an LB Volterra ๐ฝ-node that is called the product of the nodes ฮฃ1 and ฮฃ2 and is denoted ฮฃ = ฮฃ1 ร ฮฃ2 . It is easy to see that in this product ๐1 is a closed subspace of ๐ that is invariant under ๐พ. An essential feature of this de๏ฌnition is that if ๐ = col (๐1 , ๐2 ) with ๐๐ โ ๐๐ for ๐ = 1, 2, then โฅ๐ โฅ2๐ = โฅ๐1 โฅ2๐1 + โฅ๐2 โฅ2๐2 . (6.9) Moreover, ๐พ1 = ๐พโฃ๐1 ,
๐น1 = ๐น โฃ๐1 ,
๐พ2 = ๐๐2 ๐พโฃ๐2
and ๐น2 = ๐น โฃ๐2 .
(6.10)
๐
Conversely, if ฮฃ = (๐พ, ๐น ; ๐, ๐ถ ; ๐ฝ) is an LB Volterra ๐ฝ-node and ๐1 is a closed subspace of ๐ that is invariant under ๐พ and ๐2 = ๐ โ ๐1 , then ฮฃ = ฮฃ1 ร ฮฃ2 ,
where ฮฃ๐ = (๐พ๐ , ๐น๐ ; ๐๐ , โ๐ ; ๐ฝ) for ๐ = 1, 2,
๐พ๐ and ๐น๐ are de๏ฌned as in (6.10) and ( [ ๐พ1 ๐ฮฃ = ๐ผ๐ + ๐๐[๐น1 ๐น2 ] ๐ผ โ ๐ 0
๐๐น1โ ๐ฝ๐น2 ๐พ2
])โ1 [ โ ] ๐น1 ๐ฝ = ๐ฮฃ1 ๐ฮฃ2 . ๐น2โ
It is known that the formula 1 (๐ฮฃ ๐ฅ)(๐) = โ ๐น (๐ผ โ ๐๐พ)โ1 ๐ฅ, 2๐
๐ฅ โ ๐,
(6.11)
B-regular ๐ฝ-inner Matrix-valued Functions
63
de๏ฌnes a unitary similarity from a simple LB Volterra ๐ฝ-node ฮฃ = (๐พ, ๐น ; ๐, โ๐ ; ๐ฝ) with characteristic mvf ๐ onto ฮฃ๐๐๐ (๐ ). Thus, if ๐ โ โฐ โฉ ๐ฐ โ (๐ฝ), then there exists a simple LB Volterra ๐ฝ-node ฮฃ = (๐พ, ๐น ; ๐, โ๐ ; ๐ฝ) such that ๐ฮฃ = ๐ , it is de๏ฌned up to unitary equivalence by ๐ . Moreover, every closed subspace ๐1 of ๐ that is invariant under ๐พ de๏ฌnes an LB Volterra ๐ฝ-node ฮฃ1 , as above, such that its characteristic mvf is a left divisor of ๐ . If a left divisor ๐1 of ๐ may be obtained in this way, i.e., as the characteristic mvf of a node ฮฃ1 = (๐พ1 , ๐น1 ; ๐1 , โ๐ ; ๐ฝ) that is related to the node ฮฃ = (๐พ, ๐น ; ๐, โ๐ ; ๐ฝ) with characteristic mvf ๐ as in (6.10), where ๐1 is a closed subspace of ๐ that is invariant under ๐พ, then it is called left regular in the Brodskii sense (or the Livsic-Brodskii sense). Equivalently, if ๐ = ๐1 ๐2 , then ๐1 is a regular left divisor in this sense if and only if the product ฮฃ1 ร ฮฃ2 of two simple LB Volterra J-nodes with characteristic mvfโs ๐1 and ๐2 is a simple node. Theorem 6.2. Let ๐ = ๐1 ๐2 , where ๐1 , ๐2 โ โฐ โฉ ๐ โ (๐ฝ). Then ๐1 is left regular divisor of ๐ in the Brodskii sense if and only if the L. de Branges condition (1.11) holds. Proof. Suppose ๏ฌrst that (1.11) holds. Then Theorem 1.1 holds. Let ฮฃ1 = ฮฃ๐๐๐ (๐1 ) and ฮฃ2 = ฮฃ๐๐๐ (๐2 ). Then both of these two nodes are simple. The equivalence (1.12) and the formulas for the operators in these nodes implies that ฮฃ = ฮฃ1 ร ฮฃ2 is unitarily equivalent to the simple node ฮฃ๐๐๐ (๐ ). Thus, ๐1 is left regular divisor of ๐ in the Brodskii sense. Conversely, if ๐1 is a left regular divisor of ๐ in the Brodskii sense, then it is the characteristic mvf of a node ฮฃ1 that is related to the simple LB Volterra ๐ฝnode ฮฃ๐๐๐ (๐ ) by (6.10). Therefore, the Hilbert space ๐1 in the node ฮฃ1 is a closed subspace of the Hilbert space โ(๐ ) = ๐ in the node ฮฃ๐๐๐ (๐ ) that is invariant ห1 ) for some under ๐
0 . Thus, in view of Theorem 1.2 and Remark 1.3, ๐1 = โ(๐ โ1 ห1 โ ๐ฐ โ (๐ฝ) for which ๐ ห1 ๐ โ ๐ฐ โ (๐ฝ) and โ(๐ ) = โ(๐ ห1 ) โ ๐ ห1 โ(๐2 ). mvf ๐ ห ห Consequently, as the characteristic mvf ๐1 of ฮฃ๐๐๐ (๐1 ) coincides with ๐1 , (1.11) holds by Theorem 1.1. โก Theorem 6.3. If ๐1 , ๐2 โ โฐ โฉ ๐ฐ โ (๐ฝ), then ฮฃ๐๐๐ (๐1 ) ร ฮฃ๐๐๐ (๐2 ) is simple if and only if (1.11) holds. Proof. This theorem follows from Theorem 6.2. However, we shall give an independent proof for the sake of added perspective. Let ๐๐ โ โ(๐๐ ) for ๐ = 1, 2 and let ๐ = col (๐1 , ๐2 ). Then ๐ โ ker ๐น โฉ ker ๐พ if and only if ๐น1 ๐1 + ๐น2 ๐2 = 0, i.e., if and only if ๐1 (0)+๐2 (0) = 0,
๐พ1 ๐1 + ๐๐น1โ ๐ฝ๐น2 ๐2 = 0 and
โ (๐
0 ๐1 )(๐)+๐ 2๐๐น1โ ๐ฝ๐2 (0) = 0
๐พ2 ๐2 = 0,
and (๐
๐ ๐2 )(0) = 0. (6.12)
64
D.Z. Arov and H. Dym
Therefore, since ๐น1โ ๐ฃ =
โ
2๐๐พ0๐ ๐ฃ =
for ๐ฃ โ โ๐ ,
๐ฝ โ ๐1 (๐)๐ฝ โ ๐ฃ โ๐ 2๐๐
๐1 (๐) โ ๐ (0) ๐ฝ โ ๐1 (๐)๐ฝ โ๐ ๐ฝ๐1 (0) ๐ โ๐๐ ๐1 (๐) โ ๐1 (๐)๐1 (0) = = 0. ๐ Thus, the three constraints in (6.12) imply that (๐
0 ๐1 )(๐) + ๐๐น1โ ๐ฝ๐น2 ๐2 =
๐1 (๐) = ๐1 (๐)๐1 (0) = ๐1 (๐)๐2 (0) = ๐1 (๐)๐2 (๐). But this implies that
๐1 โ โ(๐1 ) โฉ ๐1 โ(๐2 ), which is equal to zero if (1.11) is in force. Therefore, ๐2 = 0 and hence ๐ = 0. Thus, condition (1.11 implies that the product node ฮฃ๐๐๐ (๐1 )รฮฃ๐๐๐ (๐2 ) is simple. Conversely, if the product node ฮฃ๐๐๐ (๐1 ) ร ฮฃ๐๐๐ (๐2 ) is simple, then โ(๐1 ) must sit isometrically inside โ(๐1 ๐2 ), which implies that (1.11) holds, by another application of Theorem 1.1. โก A Volterra operator ๐พ in a Hilbert space ๐ is called unicellular if and only if the set of all closed subspaces of ๐ that are invariant under ๐พ are ordered by inclusion. โ Theorem 6.4. If ๐ โ โฐ โฉ ๐ฐ๐ต๐
(๐ฝ) is the characteristic mvf of a simple LB Volterra ๐ฝ-node with main operator ๐พ, then
๐
is unicellular
โโ
๐พ
is unicellular.
Proof. It su๏ฌces to verify this for the de Branges model, i.e., to show that ๐ is unicellular if and only if ๐
0 is unicellular on โ(๐ ). But, by Theorems 1.1 and 1.2, a closed subspace โ1 of โ(๐ ) is invariant under ๐
0 if and only if โ1 = โ(๐1 ) for some ๐1 โ โฐ โฉ๐ฐ โ (๐ฝ) that is a left divisor of ๐ and the factors in the corresponding factorization (1.10) meet the condition (1.11). โก There is another LB Volterra ๐ฝ-node associated with each mvf ๐ โ โฐ โฉ๐ฐ โ (๐ฝ) that is obtained after identifying ๐ as the monodromy matrix of a canonical differential system (4.4) with Hermitian ๐ป(๐ฅ), 0 โค ๐ฅ โค โ, that meets the constraints (5.1) and (4.5) that is de๏ฌned in terms of ๐ป as follows: ฮฃ๐ป = (๐พ๐ป , ๐น๐ป ; ๐๐ป , โ๐ ; ๐ฝ), where ๐๐ป =
{ } โซ โ measurable ๐ ร 1 vvfโs ๐ on [0, โ] : ๐ (๐ฅ)โ ๐ป(๐ฅ)๐ (๐ฅ)๐๐ฅ < โ , 0
โซ (๐พ๐ป ๐ )(๐ฅ) = ๐๐ฝ
โ 0
โซ ๐ป(๐ )๐ (๐ )๐๐
and ๐น๐ป ๐ =
โ 0
๐ป(๐ )๐ (๐ )๐๐
for ๐ โ ๐๐ป .
B-regular ๐ฝ-inner Matrix-valued Functions
65
For this node, formula (6.11) may be expressed in terms of the matrizant of the underlying canonical system: โซ โ 1 (๐ฮฃ๐ป ๐ )(๐) = โ ๐ (๐ , ๐)๐ (๐ )๐๐ . (6.13) 2๐ 0 Theorem 6.5. If ๐๐ฅ , 0 โค ๐ฅ โค โ, is the matrizant of the canonical system (4.4), then the node ฮฃ๐ป is simple if and only if the inclusions โ(๐๐ฅ ) โ โ(๐โ ) are isometric for every ๐ฅ โค โ. Proof. See, e.g., (5) in Theorem 8.26 in [ArD12].
โก
โ Theorem 6.6. If ๐ โ โฐ โฉ ๐ฐ๐ต๐
(๐ฝ), then there exists at least one simple LB Volterra ๐ฝ-node ฮฃ๐ป with characteristic mvf ๐ (๐) and a normalized Hermitian ๐ป(๐ฅ) that meets the constraints (5.1) and (4.5). There is exactly one such node if and only if ๐ is unicellular.
Proof. Let ฮฃ = (๐พ, ๐น ; ๐, โ๐ ; ๐ฝ) be any simple LB Volterra ๐ฝ-node with characteristic mvf ๐ . Then there exists a maximal chain of closed subspaces of ๐ that are invariant under ๐พ and are ordered by inclusion. The characteristic mvfโs of the projections of ฮฃ onto these subspaces is a maximal ordered chain of normalized left divisors of ๐ . By Theorem 5.6 this chain is the matrizant ๐๐ฅ (๐), 0 โค ๐ฅ โค โ, of a canonical di๏ฌerential system (4.4) with monodromy matrix ๐ and Hermitian ๐ป that meets the constraints (5.1) and (4.5). The node ฮฃ๐ป meets the claimed properties of the theorem, as follows with the help of Theorem 6.4. โก
7. de Branges spaces Let the ๐ ร 2๐ mvf ๐(๐) = [๐ธโ (๐) ๐ธ+ (๐)] with ๐ ร ๐ blocks ๐ธยฑ be de๏ฌned in terms of the bottom block row of ๐ด โ ๐ฐ (๐ฝ๐ ) by the formula [ ] โ 1 โ๐ผ๐ ๐ผ๐ ๐(๐) = 2[0 ๐ผ๐ ]๐ด(๐)๐, where ๐ = โ . 2 ๐ผ๐ ๐ผ๐ It is then readily checked that
[ ] โ โ 0 2[0 ๐ผ๐ ]{๐ฝ๐ โ ๐ด(๐)๐ฝ๐ ๐ด(๐)โ } 2 = ๐ธ+ (๐)๐ธ+ (๐)โ โ ๐ธโ (๐)๐ธโ (๐)โ ๐ผ๐
and hence that the kernel ๐พ๐๐ (๐)
[ ] ๐ธ+ (๐)๐ธ+ (๐)โ โ ๐ธโ (๐)๐ธโ (๐)โ 0 ๐ = = 2[0 ๐ผ๐ ]๐พ๐ (๐) ๐ผ ๐๐ (๐) ๐
is positive on ๐ฅ๐ ร ๐. Therefore, this kernel de๏ฌnes a RKHS (Reproducing kernel Hilbert space), which we shall refer to as the de Branges space โฌ(๐) based on the de Branges matrix ๐(๐); for additional information (including an intrinsic de๏ฌnition
66
D.Z. Arov and H. Dym
of โฌ(๐)) see, e.g., Sections 5.10 and 5.11 in [ArD08] (especially the ๏ฌrst and last equivalences in (5.115)). Moreover, it turns out that if [ ] โ โ ๐ ๐ = 1 โ โ(๐ด) and ๐2 ๐ = 2[0 ๐ผ๐ ]๐ = 2๐2 , ๐2 then โฅ๐ โฅ2โ(๐ด) = 2โฅ๐2 โฅ2โฌ(๐)
for ๐ โ โ(๐ด) โ ker ๐2 ,
i.e., the operator ๐2 is an isometry from โ(๐ด) โ ker ๐2 onto โฌ(๐). If ๐ = 1 and ๐ด โ โฐ โฉ ๐ฐ (๐ฝ1 ), then ker ๐2 = {0} if and only if lim ๐ โ1
๐โโ
๐11 (๐๐) + ๐12 (๐๐) = 0, ๐21 (๐๐) + ๐22 (๐๐)
see, e.g., Section 5.12 in [ArD08] for additional information. Alternate characterizations of scalar de Branges spaces โฌ(๐ธ) of entire functions based on an entire function ๐ธ that meets the condition โฃ๐ธ(๐)โฃ > โฃ๐ธ(๐)โฃ for ๐ โ โ+ are given in Sections 19โ23 of [Br68a]. (Theorem 4.1 in [Dy70] exhibits the consistency of de Brangesโ de๏ฌnitions with the characterization in terms of Hardy spaces in (5.115) of [ArD08].) Spaces of matrix- and operator-valued functions are considered in [Br68b].
8. An example In this section we shall consider the canonical di๏ฌerential system [ ] ๐ ๐ฅ 0 ๐ขโฒ (๐ฅ, ๐) = ๐๐๐ข(๐ฅ, ๐) ๐ฝ1 for 0 โค ๐ฅ < โ 0 ๐ฅโ๐ and โ1 < ๐ < 1, with matrizant [ ๐11 (๐ฅ, ๐) ๐ด๐ฅ (๐) = ๐ด(๐ฅ, ๐) = ๐21 (๐ฅ, ๐)
๐12 (๐ฅ, ๐)
(8.1)
] for 0 โค ๐ฅ < โ,
๐22 (๐ฅ, ๐)
(8.2)
and shall establish the following facts: (1) The matrizant can be expressed in terms of the Gamma function and Bessel functions ๐ฝ๐ (๐ฅ) of the ๏ฌrst kind as โก( )(1โ๐)/2 โค ๐ฅ๐ 1+๐ ฮ( ) 0 2 2 โฆ ๐ด(๐ฅ, ๐) = โฃ ( ๐ฅ๐ )(1+๐)/2 1โ๐ 0 ฮ( ) 2 2 [ ] ๐ฝ ๐โ1 (๐ฅ๐) โ๐๐ฅ๐ ๐ฝ 1+๐ (๐ฅ๐) 2 2 ร , (8.3) โ๐๐ฅโ๐ ๐ฝ 1โ๐ (๐ฅ๐) ๐ฝโ 1+๐ (๐ฅ๐) 2
2
B-regular ๐ฝ-inner Matrix-valued Functions or, equivalently, as [ ๐ด(๐ฅ, ๐) =
]
ฮ( 1+๐ ) 2
0
0
ฮ( 1โ๐ ) 2
โก
67
F ๐โ1 (๐ฅ๐)
2 รโฃ ๐ฅ๐ โ๐ โ๐ 2 ๐ฅ F 1โ๐ (๐ฅ๐)
๐ โ๐ ๐ฅ๐ 2 ๐ฅ F 1+๐ (๐ฅ๐)
2
2
Fโ 1+๐ (๐ฅ๐)
โค โฆ,
(8.4)
2
where F๐ (๐) = (๐/2)โ๐ ๐ฝ๐ (๐) =
โ โ ๐=0
(โ1)๐ (๐/2)2๐ . ฮ(๐ + 1)ฮ(๐ + 1 + ๐)
(8.5)
Moreover, ๐ด๐ฅ (๐) is real (i.e., ๐ด๐ฅ (โ๐) = ๐ด๐ฅ (๐)) and symplectic; ๐11 (๐ฅ, ๐) and ๐22 (๐ฅ, ๐) are even functions of ๐, ๐12 (๐ฅ, ๐) and ๐21 (๐ฅ, ๐) are odd functions of ๐ and โฃ๐๐๐ (๐ฅ, ๐)โฃ โค ๐๐๐ (๐ฅ, ๐โฃ๐โฃ)
๐ ๐๐ ๐ โ โ, ๐ฅ > 0 ๐๐๐ ๐, ๐ = 1, 2. (8.6) [ ] (2) The de Branges matrices ๐(๐ฅ, ๐) = ๐ธโ (๐ฅ, ๐) ๐ธ+ (๐ฅ, ๐) with components ๐ธยฑ (๐ฅ, ๐) = ๐22 (๐ฅ, ๐) ยฑ ๐21 (๐ฅ, ๐) ( )(1+๐)/2 } ๐ฅ๐ 1โ๐ { = ฮ( ) ๐ฝโ 1+๐ (๐ฅ๐) โ ๐๐ฅโ๐ ๐ฝ 1โ๐ (๐ฅ๐) 2 2 2 2
(8.7)
are entire functions of exponential type ๐ฅ, โฃ๐ธ+ (๐ฅ, ๐)โฃ = โฃ๐ธ+ (๐ฅ, โ๐)โฃ ๐ธโ (๐ฅ, ๐) =
# ๐ธ+ (๐ฅ, ๐)
for every point ๐ โ โ,
for every point ๐ โ โ
(8.8) (8.9)
and ๐ธยฑ (๐ฅ, ๐) โ= 0 for every point ๐ โ โยฑ . (3) The function ๐๐ฅ (๐) = ๐๐ด๐ฅ [1] =
๐11 (๐ฅ, ๐) + ๐12 (๐ฅ, ๐) ๐21 (๐ฅ, ๐) + ๐22 (๐ฅ, ๐)
(8.10)
belongs to the subclass of the Carathยดeodory class ๐ that is denoted by ๐๐ in [ArD08] and is characterized by the fact that it meets the condition lim ๐ โ1 ๐๐ฅ (๐๐) = 0
๐โโ
(8.11)
and its spectral function ๐๐ฅ (๐) is locally absolutely continuous; in fact } โซ โ{ 1 1 ๐ ๐๐ฅ (๐) = ๐๐ผ + โ โฃ๐ธ+ (๐ฅ, ๐)โฃโ2 ๐๐ (8.12) ๐๐ โโ ๐ โ ๐ 1 + ๐2 for ๐ โ โ+ and appropriate ๐ผ โ โ. (4) The function
ฮ๐ฅ (๐) = โฃ๐ธ+ (๐ฅ, ๐)โฃโ2 = ๐๐ฅโฒ (๐)
68
D.Z. Arov and H. Dym
satis๏ฌes the Muckenhoupt (๐ด2 ) condition { } โซ ๐ โซ ๐ 1 1 โ1 sup ฮ๐ฅ (๐)๐๐ ฮ๐ฅ (๐) ๐๐ < โ for every choice of ๐ < ๐ . ๐โ๐ ๐ ๐โ๐ ๐ โ (5) ๐ด๐ฅ โ โฐ โฉ ๐ฐโ๐ ๐
(๐ฝ1 ) for 0 โค ๐ฅ < โ. (6) The de Branges space โฌ(๐๐ฅ ) is the space of entire functions ๐ of exponential type less than or equal to ๐ฅ for which โซ โ โฃ๐ (๐)โฃ2 (8.13) ๐ ๐๐ < โ. โโ โฃ๐โฃ
(7) If ๐ = 0, then
[
] โ๐ sin(๐๐ฅ) , cos(๐๐ฅ)
cos(๐๐ฅ) ๐ด๐ฅ (๐) = โ๐ sin(๐๐ฅ)
๐๐ฅ (๐) = [๐๐๐๐ฅ
๐โ๐๐๐ฅ ]
and the de Branges space โฌ(๐๐ฅ ) is equal to the Paley-Wiener space {โซ ๐ฅ } โ(๐๐ฅ ) โ โโ (๐๐ฅ ) = ๐๐๐๐ก ๐ (๐ก)๐๐ก : ๐ โ ๐ฟ2 ([โ๐ฅ, ๐ฅ]) . โ๐ฅ
(8) If โ1 < ๐ < 0, then โฌ(๐๐ฅ ) is a proper subspace of โ(๐๐ฅ ) โ โโ (๐๐ฅ ). If 0 < ๐ < 1, then โ(๐๐ฅ ) โ โโ (๐๐ฅ ) is a proper subspace of โฌ(๐๐ฅ ). (9) If ๐ โ (โ1, 0) โช (0, 1) and ๐ฅ > 0, then ๐ด๐ฅ โโ ๐ฐ๐๐ ๐
(๐ฝ1 ). Proof. The proof is divided into steps. 1. Veri๏ฌcation of (1). The top row of the matrizant is a solution of the system [ โฒ ] [ ] ๐11 (๐ฅ, ๐) ๐โฒ12 (๐ฅ, ๐) = โ๐๐ ๐ฅโ๐ ๐12 (๐ฅ, ๐) ๐ฅ๐ ๐11 (๐ฅ, ๐) with initial condition
[
] [ ๐11 (0, ๐) ๐12 (0, ๐) = 1
] 0 .
Thus, ๐11 (๐ฅ, ๐) is a solution of the Bessel equation ๐ ๐โฒโฒ11 (๐ฅ, ๐) + ๐โฒ11 (๐ฅ, ๐) + ๐2 ๐11 (๐ฅ, ๐) = 0, ๐ฅ with initial conditions ๐11 (0, ๐) = 1 i.e.,
( ๐11 (๐ฅ, ๐) =
๐ฅ๐ 2
)(1โ๐)/2
( ฮ
0 โค ๐ฅ < โ,
and ๐โฒ11 (0, ๐) = 0,
1+๐ 2
) ๐ฝ(๐โ1)/2 (๐ฅ๐)
for 0 โค ๐ฅ < โ.
(8.14)
This justi๏ฌes the formula for ๐11 . The formula for ๐12 follows from the fact that ๐ ๐ ๐ โ๐ ๐ฅ ๐ฝ๐ (๐ฅ๐) = ๐๐ฅ๐ ๐ฝ๐โ1 (๐ฅ๐) and ๐ฅ ๐ฝ๐ (๐ฅ๐) = โ๐๐ฅโ๐ ๐ฝ๐+1 (๐ฅ๐). ๐๐ฅ ๐๐ฅ The remaining entries in (8.3) can be veri๏ฌed in much the same way, since [ โฒ ] [ ] ๐21 (๐ฅ, ๐) ๐โฒ22 (๐ฅ, ๐) = โ๐๐ ๐ฅโ๐ ๐22 (๐ฅ, ๐) ๐ฅ๐ ๐21 (๐ฅ, ๐)
B-regular ๐ฝ-inner Matrix-valued Functions with initial condition
[
] [ ๐21 (0, ๐) ๐22 (0, ๐) = 0
69
] 1 .
Furthermore, ๐ด๐ฅ (๐) is real and symplectic, since ๐ด๐ฅ (โ๐) = ๐ด๐ฅ (๐)
and
det ๐ด๐ฅ (๐) = 1.
Finally, the inequalities in (8.6) are immediate from from (8.4) and (8.5). 2. Formulas (8.8) and (8.9) hold and โฃ๐ธยฑ (๐ฅ, ๐)โฃ > 0 for every point ๐ โ โยฑ . Formulas (8.8) and (8.9 follow easily from (8.2), (8.4), (8.5) and the ๏ฌrst formula in (8.7). Next, since [ ] [ ] 0 0 1 ๐ด๐ฅ (๐)๐ฝ1 ๐ด๐ฅ (๐)โ ๐ด (๐) = 0 for ๐ โ โ, 1 ๐ฅ the identity
[
๐ธโ (๐ฅ, ๐)
] โ [ ๐ธ+ (๐ฅ, ๐) = 2 0
] 1 ๐ด๐ฅ (๐)๐
implies that โฃ๐ธ+ (๐ฅ, ๐)โฃ2 = โฃ๐ธโ (๐ฅ, ๐)โฃ2 . Thus, if ๐ โ โ, ๐ธ+ (๐ฅ, ๐) = 0 =โ ๐ธโ (๐ฅ, ๐) = 0 =โ rank ๐ด๐ฅ (๐) โค 1, which contradicts the invertibility of ๐ด๐ฅ (๐) on โ. Similarly, the inequality โฃ๐ธ+ (๐ฅ, ๐)โฃ โฅ โฃ๐ธโ (๐ฅ, ๐)โฃ
for ๐ โ โ+
(8.15)
implies that if ๐ธ+ (๐ฅ, ๐) = 0 for some point ๐ โ โ+ , then ๐ธโ (๐ฅ, ๐) = 0 and hence, det ๐ด๐ฅ (๐) = 0, which contradicts the already established fact that ๐ด๐ฅ (๐) is invertible for every point ๐ โ โ. Therefore, โฃ๐ธ+ (๐ฅ, ๐)โฃ > โฃ๐ธโ (๐ฅ, ๐)โฃ
for ๐ โ โ+
(8.16)
and the proof that โฃ๐ธ+ (๐ฅ, ๐)โฃ > 0 for ๐ โ โ+ is complete. In view of (8.9), this also justi๏ฌes the inequality โฃ๐ธโ (๐ฅ, ๐)โฃ > 0 for ๐ โ โโ . 3. If ๐ฅ > 0 and ๐ > 0, then there exist a pair of positive constants ๐พ1 and ๐พ2 that depend on ๐ฅ and ๐ such that ๐พ1 (1 + โฃ๐โฃ๐ ) โค โฃ๐ธ+ (๐ฅ, ๐)โฃ2 โค ๐พ2 (1 + โฃ๐โฃ๐ ) Since
โ ๐ฝ๐ (๐ฅ) โผ
it is readily checked that
for ๐ โ โ.
(8.17)
2 ๐ ๐ cos(๐ฅ โ โ ๐ ) as ๐ฅ โ โ, ๐๐ฅ 4 2
( )2 1 ( ๐ฅ๐ )๐ 1โ๐ โฃ๐ธ+ (๐ฅ, ๐)โฃ โผ ฮ ๐ 2 2 { 2 } ร cos (๐ฅ๐ + ๐(๐/4)) + ๐ฅโ2๐ sin2 (๐ฅ๐ + ๐(๐/4)) 2
Thus, if 1 ๐๐ฅ = ฮ ๐
(
1โ๐ 2
)2
โ2๐
min{1, ๐ฅ
}
and
1 ๐๐ฅ = ฮ ๐
(
1โ๐ 2
)2
as ๐ โ โ.
(8.18)
max{1, ๐ฅโ2๐ },
70 then
D.Z. Arov and H. Dym ( ๐ฅ๐ )๐ 2
๐๐ฅ (1 + ๐(1/๐)) โค โฃ๐ธ+ (๐ฅ, ๐)โฃ2 ( ๐ฅ๐ )๐ โค ๐๐ฅ (1 + ๐(1/๐)) 2
as ๐ โ โ. This serves to establish (8.17), since ๐ธ+ (๐ฅ, ๐) is an entire function of ๐ with no real zeros. ( )(1+๐)/2 ๐ฅ๐ ๐ โ๐ โ๐ 4. ๐ธ+ (๐ฅ, ๐๐) โผ ๐ฅ๐ ฮ( 1 โ 2 2 )(1 + ๐ฅ ) 2๐๐ฅ๐ ๐๐ ๐ โ โ. This follows from formula (8.7), and the relations ๐ฝ๐ (๐๐ฅ๐) = ๐๐ ๐ผ๐ (๐ฅ๐)
๐๐ฅ๐ and ๐ผ๐ (๐ฅ๐) โผ โ 2๐๐
as ๐ โ โ.
(8.19)
5. Veri๏ฌcation of (2). Assertion (2) follows from (1) and Step 2. Detailed type estimates are also furnished in [Dy70]. 6. Veri๏ฌcation of (3). In view of (8.3) and (8.19), ( ๐ฅ๐ )(1โ๐)/2 1 + ๐ { } ๐11 (๐ฅ, ๐๐) + ๐12 (๐ฅ, ๐๐) = ฮ( ) ๐ผ ๐โ1 (๐ฅ๐) + ๐ฅ๐ ๐ผ 1+๐ (๐ฅ๐) . 2 2 2 2 Therefore, by (8.10),
( ) } 1+๐ { ๐ ๐โ1 1+๐ ฮ ๐ผ (๐ฅ๐) + ๐ฅ ๐ผ (๐ฅ๐) ( ๐ฅ๐ )โ๐ 2 2 2 ( ){ ๐๐ฅ (๐๐) = } 1โ๐ 2 ฮ ๐ฅโ๐ ๐ผ 1โ๐ (๐ฅ๐) + ๐ผโ 1+๐ (๐ฅ๐) 2 2 2 ( ) 1+๐ (1 + ๐ฅ๐ ) ( ๐ฅ๐ )โ๐ ฮ 2 ( ) โผ 1โ๐ 2 ฮ (1 + ๐ฅโ๐ ) 2 ( ) 1+๐ ฮ ( ๐ )โ๐ 2 ( ) as ๐ฅ โ โ. = 1โ๐ 2 ฮ 2
Thus, (8.11) holds, since ๐ > โ1. The rest follows from the fact that ๐๐ฅ (๐) is holomorphic on โ and (โ๐๐ฅ )(๐) = โฃ๐ธ+ (๐ฅ, ๐)โฃโ2 for ๐ โ โ. 7. Veri๏ฌcation of (4) and (5). (4) follows from the bounds in (8.17); (5) follows from (3), (4) and Theorem 10.9 in [ArD08]. 8. Veri๏ฌcation of (6). If ๐ โ โฌ(๐๐ฅ ), then it is an entire function of exponential type at most ๐ฅ by Theorem 5.65 in [ArD08] and the bound โ โฃ๐ฃโ ๐ (๐)โฃ = โฃโจ๐, ๐พ๐๐ ๐ฃโฉโฌ(๐) โฃ โค โฅ๐ โฅโฌ(๐) ๐ฃโ ๐พ๐๐ (๐)๐ฃ ;
B-regular ๐ฝ-inner Matrix-valued Functions and
โซ
71
โ
โฃ๐ (๐)โฃ2 (8.20) 2 ๐๐ < โ. โโ โฃ๐ธ+ (๐ฅ, ๐)โฃ Thus, in view of the bounds in (8.17), (8.13) holds. In fact, since โซ โ 1 1 ๐๐ < โ (8.21) 2 2 โโ โฃ๐ธ+ (๐ฅ, ๐)โฃ 1 + ๐ and (8.9), (8.16) are in force, โฌ(๐๐ฅ ) can be characterized as the set of entire functions ๐ of exponential type โค ๐ฅ for which the integral in (8.20) is ๏ฌnite; see, e.g., Lemma 3.5 in [Dy70]. However, in view of (8.17), this is equivalent to (6). 9. Veri๏ฌcation of (7), (8) and (9). The asserted inclusions follow from the characterizations of โฌ(๐๐ฅ ) given in (10), the Paley-Wiener theorem (which serves to characterize the Paley-Wiener space by (10) with ๐ = 0) and the fact that: If ๐ is an entire function of exponential type at most ๐ฅ and โ1 < ๐ โค 0, then โซ โ โซ โ โฃ๐ (๐)โฃ2 โฃ๐ (๐)โฃ2 ๐๐ < โ, ๐ ๐๐ < โ =โ โโ โฃ๐โฃ โโ โฅ๐ โฅ2โฌ(๐๐ฅ)
=
i.e.,
โ1 < ๐ โค 0 =โ โฌ(๐๐ฅ ) โ โฌ([๐๐ฅ ๐โ๐ฅ ]). (8.22) On the other hand, if ๐ is an entire function of exponential type at most ๐ฅ and 0 โค ๐ < 1, then โซ โ โซ โ โฃ๐ (๐)โฃ2 โฃ๐ (๐)โฃ2 ๐๐ < โ =โ ๐ ๐๐ < โ, โโ โโ โฃ๐โฃ
i.e.,
0 โค ๐ < 1 =โ โฌ([๐๐ฅ ๐โ๐ฅ ]) โ โฌ(๐๐ฅ ). (8.23) โ โ Moreover, if 0 < ๐ < 1, then the function ๐ (๐) = sin ๐/ ๐ is an entire function of minimal exponential type that meets the condition (8.13). Therefore, ๐ โ โฌ(๐๐ฅ ) for every ๐ฅ > 0. However, ๐ โโ ๐ฟ2 (โ). Consequently the inclusion in (8.23) is proper when 0 < ๐ < 1. Next, to establish that the inclusion in (8.23) is proper when โ1 < ๐ < 0, it su๏ฌces to exhibit an entire function ๐ of exponential type at most ๐ฅ in ๐ฟ2 (โ) for which (8.13) fails. The formula โซ ๐/2 ๐ฮ(๐ โ 1) ) (1 ) (๐ > 1) (8.24) ๐ (๐) = (cos ๐ก)๐โ2 ๐๐๐๐ก ๐๐ก = ๐โ2 ( 1 1 1 2 ฮ ๐ โ๐/2 2 + 2๐ ฮ 2๐ โ 2๐ (which is taken from p. 186 of Titchmarsh [Ti62], who credits S. Ramanujan) exhibits the right-hand side as an entire function of exponential type ๐/2 when โซ ๐/2 (cos ๐ก)2(๐โ2) ๐๐ก < โ, ๐/2
i.e., when ๐ > 3/2. Moreover, since { (1 ) ( )}โ1 ฮ 2 ๐ + 12 ๐ ฮ 12 ๐ โ 12 ๐ = ๐(โฃ๐โฃ1โ๐ ),
(8.25)
72
D.Z. Arov and H. Dym
(8.13) fails for โ1 < ๐ < 0 when ๐ โค (3 โ ๐)/2. Therefore, since the function speci๏ฌed in (8.24) is of exponential type ๐/2, the inclusion (8.22) is proper for such choices of ๐ and ๐ฅ โฅ ๐/2. The same conclusion is obtained for 0 < ๐ฅ < ๐/2 by considering ๐ (๐๐) with 0 < ๐ < 1 in place of ๐ (๐). Finally (9) is immediate from Theorem 5.98 in [ArD08]. โก Additional features of this example that are connected with the interpretation of the mvfโs ๐ด๐ฅ as resolvent matrices of a class of related extension problems will be discussed in [ArD12]. Remark 8.1. It turns out that the function ๐๐ฅ (๐) de๏ฌned by (8.10) tends to a limit ๐โ (๐) as ๐ฅ โ โ and that this limit admits an integral representation โซ โ ๐โ (๐) = ๐2 ๐๐๐๐ก ๐โ (๐ก)๐๐ก for ๐ โ โ+ , (8.26) 0
where
๐โ (๐ก) = ๐๐ ๐ก๐+1 for ๐ก โฅ 0 and a constant ๐๐ . Thus, if โ1 < ๐ < 1/2, then โซ ๐ โฃ๐โ (๐ก)โฃ2 ๐๐ก = โ for every ๐ > 0,
(8.27)
0
which, in view of Theorem 8.39 in [ArD08], is a stronger conclusion than (9). We remark that variants of the di๏ฌerential system (8.1) considered in this section have been used for assorted purposes in [Dy70], [LLS], [Sak97] and presumably in many other places.
References [AlD84] D. Alpay and H. Dym, Hilbert spaces of analytic functions, inverse scattering and operator models, I., Integ. Equat. Oper. Th. 7 (1984), 589โ741. [ArD97] D.Z. Arov and H. Dym, ๐ฝ-inner matrix functions, interpolation and inverse problems for canonical systems, I: Foundations, Integ. Equat. Oper. Th. 29, (1997), 373โ454. [ArD05] D.Z. Arov and H. Dym, Strongly regular ๐ฝ-inner matrix-vaued functions and inverse problems for canonical systems, in: Recent Advances in Operator Theory and its Applications (M.A. Kaashoek, S. Seatzu and C. van der Mee, eds.), Oper. Theor. Adv. Appl., 160, Birkhยจ auser, Basel, 2005, pp. 101โ160. [ArD07] D.Z. Arov and H. Dym, Bitangential direct and inverse problems for systems of di๏ฌerential equations, in Probability, Geometry and Integrable Systems (M. Pinsky and B. Birnir, eds.) MSRI Publications, 55, Cambridge University Press, Cambridge, 2007, pp. 1โ28. [ArD08] D.Z. Arov and H. Dym, ๐ฝ-Contractive Matrix Valued Functions and Related Topics, Encyclopedia of Mathematics and its Applications, 116. Cambridge University Press, Cambridge, 2008.
B-regular ๐ฝ-inner Matrix-valued Functions
73
[ArD12] D.Z. Arov and H. Dym, Bitangential Direct and Inverse Problems for Systems of Di๏ฌerential Equations, Cambridge University Press, in press. [Br63] L. de Branges, Some Hilbert spaces of analytic functions I. Trans. Amer. Math. Soc. 106 (1963), 445โ668. [Br65] L. de Branges, Some Hilbert spaces of analytic functions II. J. Math. Anal. Appl. 11 (1965), 44โ72. [Br68a] L. de Branges, Hilbert Spaces of Entire Functions. Prentice-Hall, Englewood Cli๏ฌs, 1968. [Br68b] L. de Branges, The expansion theorem for Hilbert spaces of entire functions, in Entire Functions and Related Parts of Analysis Amer. Math. Soc., Providence, 1968, pp. 79โ148. [Bro72] M.S. Brodskii, Triangular and Jordan Representations of Linear Operators. Transl. Math Monographs, 32 Amer. Math. Soc. Providence, R.I., 1972. [Dy70] H. Dym, An introduction to de Branges spaces of entire functions with applications to di๏ฌerential equations of the Sturm-Liouville type. Adv. Math. 5 (1970), 395โ471. [DMc76] H. Dym and H.P. McKean, Gaussian Processes, Function Theory, and the Inverse Spectral Problem, Academic Press, New York, 1976; reprinted by Dover, New York, 2008. [LLS] H. Langer, M. Langer and Z. Sasvari, Continuation of Hermitian inde๏ฌnite functions and corresponding canonical systems: An example, Methods of Funct. Anal. Top., 10 (2004), no.1, 39โ53. [Po60] V.P. Potapov, The multiplicative structure of ๐ฝ-contractive matrix functions. Amer. Math. Soc. Transl. (2) 15 (1960) 131โ243. [Sak97] L.A. Sakhnovich, Interpolation Theory and its Applications. Mathematics and its Applications, 428, Kluwer Academic Publishers, Dordrecht, 1997. [Ti62] E.C. Titchmarsh, Introduction to the Theory of Fourier Integrals (Second Edition), Oxford University Press, Oxford, 1962. Damir Z. Arov Division of Informatics and Applied Mathematics South-Ukranian National Pedagogical University 65020 Odessa, Ukraine e-mail:
[email protected] Harry Dym Department of Mathematics The Weizmann Institute of Science Rehovot 76100, Israel e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 75โ122 c 2012 Springer Basel AG โ
Canonical Transfer-function Realization for Schur-Agler-class Functions of the Polydisk Joseph A. Ball and Vladimir Bolotnikov In memory of Israel Gohberg, a ๏ฌne teacher and colleague who will be missed
Abstract. Associated with any Schur-class function ๐(๐ง) (i.e., a contractive operator-valued holomorphic function on the unit disk) is the de BrangesRovnyak kernel ๐พ๐ (๐ง, ๐) = [๐ผ โ ๐(๐ง)๐(๐)โ ]/(1 โ ๐ง๐) and the reproducing kernel Hilbert space โ(๐พ๐ ) which serves as the canonical functional-model statespace for a coisometric transfer-function realization ๐(๐ง) = ๐ท+๐ง๐ถ(๐ผโ๐ง๐ด)โ1 ๐ต of ๐. To obtain a canonical functional-model unitary transfer-function realization, it is now well understood that one must work with a certain (2 ร 2)block matrix kernel and associated two-component reproducing kernel Hilbert space. In this paper we indicate how these ideas extend to the multivariable setting where the unit disk is replaced by the unit polydisk in ๐ complex variables. For the case ๐ > 2, one must replace the Schur class by the more restrictive Schur-Agler class (de๏ฌned in terms of the validity of a certain von Neumann inequality) in order to get a good realization theory paralleling the single-variable case. This work represents one contribution to the recent extension of the state-space method to multivariable settings, an area of research where Israel Gohberg was a prominent and leading practitioner. Mathematics Subject Classi๏ฌcation (2000). 47A57. Keywords. Operator-valued Schur-Agler functions, Agler decomposition, unitary realization.
1. Introduction 1.1. The classical setting and the legacy of Israel Gohberg For ๐ฐ and ๐ด Hilbert spaces, we let โ(๐ฐ, ๐ด) denote the space of bounded linear operators mapping ๐ฐ into ๐ด, abbreviated to โ(๐ฐ) in case ๐ฐ = ๐ด. We then de๏ฌne the operator-valued version of the classical Schur class ๐ฎ(๐ฐ, ๐ด) to consist of holomorphic functions ๐ on the unit disk ๐ป with values equal to contraction operators
76
J.A. Ball and V. Bolotnikov
between ๐ฐ and ๐ด. There is a close connection between Schur-class functions and dissipative discrete-time linear systems which we now explain. By a discrete time linear system we mean a system of equations of the form { ๐ฅ(๐ก + 1) = ๐ด๐ฅ(๐ก) + ๐ต๐ข(๐ก) (1.1) ฮฃ๐ ๐ฆ(๐ก) = ๐ถ๐ฅ(๐ก) + ๐ท๐ข(๐ก) where the evolution is along the nonnegative integers ๐ก โ โค+ := {0, 1, 2, . . . }. Here we view ๐ฐ as the input space, ๐ณ as the state space and ๐ด as the output space. Application of the โค-transform in the form ๐ฅ ห(๐ง) =
โ โ
๐ฅ(๐)๐ง ๐
๐=0
then results in an input-output relation of the form ๐ฆห(๐ง) = ๐ถ(๐ผ โ ๐ง๐ด)โ1 ๐ฅ(0) + ๐ฮฃ๐ (๐ง)ห ๐ข(๐ง) where
๐ฮฃ๐ (๐ง) = ๐ท + ๐ง๐ถ(๐ผ โ ๐ง๐ด)โ1 ๐ต
(1.2)
is a rational matrix function with no pole at the origin (in case ๐ฐ, ๐ณ , and ๐ด are all ๏ฌnite-dimensional), or, more generally, an operator-valued function, holomorphic on a neighborhood of the origin (in the in๏ฌnite-dimensional setting). A key discovery is that this procedure is reversible: any rational matrix-valued (or, more generally operator-valued) function holomorphic in a neighborhood of the origin can be represented in the form of a transfer function of a linear system (1.2); this is the starting point for the so-called state-space method in the analysis of all sorts of problems (e.g., Fredholm theory of Wiener-Hopf and singular integral operators, pole-zero structure and interpolation problems of tangential Lagrange-Sylvester and Nevanlinna-Pick type for rational matrix functions) of which Israel Gohberg was a leading ๏ฌgure (see [34, 47, 35, 36, 23, 45]). There is a special case of the discrete-time linear system (1.1) which is of special interest, namely the case where the system is conservative or dissipative, corresponding to the case where the linear spaces ๐ณ , ๐ฐ, ๐ด are all taken to be Hilbert ๐ด ๐ต ] is taken to be unitary (conservative spaces and the colligation matrix U = [ ๐ถ ๐ท case) or contractive (dissipative case). The contractivity of the colligation matrix U then results in the energy-balance relation โฅ๐ฅ(๐ก + 1)โฅ2 โ โฅ๐ฅ(๐ก)โฅ2 โค โฅ๐ข(๐ก)โฅ2 โ โฅ๐ฆ(๐ก)โฅ2 ,
(1.3)
i.e., the net change in the energy stored in the system from time ๐ก to time ๐ก + 1 cannot be more than the net energy supplied to the system through the absorption of the input signal ๐ข(๐ก) and the loss of the output signal ๐ฆ(๐ก). If we take ๐ฅ(0) = 0 and sum up over 0 โค ๐ก โค ๐ , we get 0 โค โฅ๐ฅ(๐ + 1)โฅ2 โค
๐ โ [ ] โฅ๐ข(๐ก)โฅ2 โ โฅ๐ฆ(๐ก)โฅ2 ๐ก=0
Canonical Realization
77
for all ๐ โฅ 0, which immediately implies that โฅ{๐ฆ(๐ก)}๐กโโค+ โฅโ2 โค โฅ{๐ข(๐ก)}๐กโโค+ โฅโ2 . Via the Plancherel theorem, it follows that โฅห ๐ฆโฅ2๐ป 2 โค โฅห ๐ขโฅ2๐ป 2 . Here, for any coe๏ฌcient ๐ด
๐ฐ
Hilbert space โฐ we use the notation ๐ปโฐ2 to denote the Hardy space of โฐ-valued functions on the unit disk ๐ป. As ๐ฆห = ๐ฮฃ๐ โ
๐ข ห, we then see that necessarily ๐ฮฃ๐ is in the Schur class ๐ฎ(๐ฐ, ๐ด). Such functions come up as scattering functions for conservative or dissipative linear circuits (see [37, 65]), as characteristic operator functions in the theory of canonical models for Hilbert space contraction operators (see [54, 58, 41]) as well as scattering functions for Lax-Phillips scattering systems (see [53, 1]). We mention that the Livหsic theory of characteristic functions (see [54, 55]) is really about modeling operators close to unitary (contractive or not), or after a Cayley transform change of variable, operators which are close to selfadjoint, where the state-space ๐ณ is allowed to carry an inde๏ฌnite inner-product rather than a positive-de๏ฌnite inner product; it was in this latter context that Israel Gohberg was also a participant in development of the theory (see [43]). Finer considerations can be used to characterize when ๐ is inner, i.e., in addition has unitary boundary values on the unit disk, but we do not go into the details of this point here. Su๏ฌce it to say that it is such energy-balance considerations which distinguish the earliest versions of the Livหsic theory of characteristic functions from the Kalman theory where the various spaces ๐ฐ, ๐ณ , ๐ด are just linear spaces and there is no consideration of any energy-balance relation as in (1.3) (see [51, 52]); it should be mentioned that energy-balance considerations were introduced into the control theory literature by Willems (see [63, 64]) with earlier foreshadowing in the circuit theory literature [37, 65]. The realization problem for Schur-class functions can be stated quite simply: given a function ๐ in the Schur class ๐ฎ(๐ฐ, ๐ด), can one ๏ฌnd a contractive (or even ๐ด ๐ต ] so that ๐(๐ง) is realized as the transfer funcunitary) colligation matrix U = [ ๐ถ ๐ท tion ๐ฮฃ๐ of the associated dissipative (or even conservative) discrete-time linear system (1.1)? For future reference, we state the following well-known but state-ofart re๏ฌned version of the solution of this problem; a good reference for this result and the single-variable results to follow in this section is the book [7] where the more general Pontryagin-space setting is handled. Theorem 1.1. Let ๐ : ๐ป โ โ(๐ฐ, ๐ด) be given. Then the following are equivalent: (1a) ๐ โ ๐ฎ(๐ฐ, ๐ด), i.e., ๐ is holomorphic on ๐ป with โฅ๐(๐ง)โฅ โค 1 for all ๐ง โ ๐ป. (1b) ๐ satis๏ฌes the von Neumann inequality: โฅ๐(๐ )โฅ โค 1 for any strictly contractive operator ๐ on a Hilbert space โ, where ๐(๐ ) is de๏ฌned by ๐(๐ ) =
โ โ
๐
๐๐ โ ๐ โ โ(๐ฐ โ โ, ๐ด โ โ)
if
๐=0
๐(๐ง) =
โ โ
๐๐ ๐ง ๐ .
๐=0
(2) The associated kernel function ๐พ๐ (๐ง, ๐) = is a positive kernel on ๐ป ร ๐ป.
๐ผ๐ด โ ๐(๐ง)๐(๐)โ 1 โ ๐ง๐
(1.4)
78
J.A. Ball and V. Bolotnikov
(3) There is an auxiliary Hilbert [ ] [ space ] ๐ณ [ and ] a unitary connecting operator (or ๐ด ๐ต ๐ณ ๐ณ colligation) U = : โ so that ๐(๐ง) can be expressed as ๐ถ ๐ท ๐ฐ ๐ด ๐(๐ง) = ๐ท + ๐ง๐ถ(๐ผ โ ๐ง๐ด)โ1 ๐ต.
(1.5)
(4) ๐(๐ง) has a realization as in (1.5) where the connecting operator U is any one of (i) isometric, (ii) coisometric, or (iii) contractive. We shall be interested in uniqueness issues in such a transfer-function re[ ] โฒ ๐ด ๐ต] : [๐ณ ] โ ๐ณ alization. Let us say that two colligations U = [ ๐ถ and U = ๐ด ๐ท ๐ฐ [ ๐ดโฒ ๐ต โฒ ] [ โฒ ] [ โฒ] ๐ณ ๐ณ : ๐ฐ โ ๐ด are unitarily equivalent if there is a unitary operator ๐ถ โฒ ๐ทโฒ ๐ : ๐ณ โ ๐ณ โฒ so that [ ][ ] [ โฒ ][ ] ๐ 0 ๐ด ๐ต ๐ด ๐ตโฒ ๐ 0 = . 0 ๐ผ๐ด ๐ถ ๐ท ๐ถ โฒ ๐ทโฒ 0 ๐ผ๐ฐ It is readily seen that if two colligations are unitarily equivalent, then their transfer functions are identical: ๐(๐ง) = ๐ โฒ (๐ง). The converse is true under appropriate โ minimality conditions which we now recall. In what follows, the symbol stands for the closed linear span. [ ] ๐ด ๐ต ] : [ ๐ณ ] โ ๐ณ is called De๏ฌnition 1.2. The colligation U = [ ๐ถ ๐ด ๐ท ๐ฐ 1. observable (or closely outer-connected) if the pair (๐ถ, ๐ด) is observable, i.e., if โ โ๐ โ Ran ๐ด ๐ถ = ๐ณ ; ๐โฅ0
2. controllable (or closely inner-connected) if the pair (๐ต, ๐ด) is controllable, i.e., โ if Ran ๐ด๐ ๐ต = ๐ณ ; ๐โฅ0
3. closely connected if
โ
{Ran ๐ด๐ ๐ต, Ran ๐ดโ๐ ๐ถ โ } = ๐ณ .
๐โฅ0
The kernel function ๐พ๐ (๐ง, ๐) given by (1.4) and the associated reproducing kernel Hilbert space โ(๐พ๐ ) is the classical de Branges-Rovnyak kernel and de Branges-Rovnyak reproducing kernel Hilbert space associated with the Schur-class function ๐ and has been much studied over the years, both as an object in itself and as a tool for other types of applications. The special role of the de BrangesRovnyak space in connection with the transfer-function realization for Schur-class functions is to provide a canonical functional-model realization for ๐, as illustrated in the following theorem. Theorem 1.3. Suppose that the function ๐ is in the Schur class ๐ฎ(๐ฐ, ๐ด) and let โ(๐พ๐ ) be the associated de Branges-Rovnyak space. De๏ฌne operators ๐ด, ๐ต, ๐ถ, ๐ท by ๐ (๐ง) โ ๐ (0) ๐(๐ง) โ ๐(0) ๐ด : ๐ (๐ง) โ , ๐ต : ๐ข โ ๐ข, (1.6) ๐ง ๐ง ๐ถ : ๐ (๐ง) โ ๐ (0), ๐ท : ๐ข โ ๐(0)๐ข.
Canonical Realization
79
๐ด ๐ต ] de๏ฌnes a coisometry from โ(๐พ ) โ ๐ฐ to Then the operator colligation U = [ ๐ถ ๐ ๐ท โ1 โ(๐พ๐ )โ๐ด. Moreover, U is observable and its transfer function ๐ท+๐ง๐ถ(๐ผ โ๐ง๐ด) [ ๐ดโฒ ๐ต โฒ ] [ ๐ต] ๐ณ equals ๐(๐ง). Finally, any observable coisometric colligation ๐ถ โฒ ๐ทโฒ : [ ๐ฐ ] โ ๐ณ ๐ด with transfer function equal to ๐ is unitarily equivalent to U.
It is easily seen from characterization (1a) in Theorem 1.1 that ๐ โ ๐ฎ(๐ฐ, ๐ด) โโ ๐ โฏ โ ๐ฎ(๐ด, ๐ฐ)
where ๐ โฏ (๐ง) := ๐(๐ง)โ .
Hence for a given Schur-class function ๐ there is also associated a dual de Brangesโ ๐(๐) Rovnyak space โ(๐พ๐ โฏ ) with reproducing kernel ๐พ๐ โฏ (๐ง, ๐) = ๐ผโ๐(๐ง) which 1โ๐ง๐ plays the same role of providing a canonical functional-model realization for isometric realizations of ๐ as โ(๐พ๐ ) plays for coisometric realizations, as illustrated in the next theorem. Theorem 1.4. Suppose that the function ๐ is in the Schur class ๐ฎ(๐ฐ, ๐ด) and let โ(๐พ๐ โฏ ) be the associated dual de Branges-Rovnyak space. De๏ฌne ห : ๐(๐ง) โ ๐ง๐(๐ง) โ ๐(๐ง)โ ห ห : ๐ข โ (๐ผ โ ๐(๐ง)โ ๐(0))๐ข, ๐ด ๐ (0), ๐ต (1.7) ห : ๐(๐ง) โ ๐ห(0), ห : ๐ข โ ๐(0)๐ข, ๐ถ ๐ท where ๐ห(0) is the unique vector in ๐ด such that โฉ โช ๐(๐ง)โ โ ๐(0)โ โจห ๐ (0), ๐ฆโฉ๐ฐ = ๐(๐ง), ๐ฆ for all ๐ฆ โ ๐ด. ๐ง โ(๐พ๐ โฏ ) [ ] ห = ๐ดห ๐ตห de๏ฌnes an isometry from โ(๐พ๐ โฏ ) โ ๐ฐ Then the operator colligation U ห ห ๐ถ ๐ท ห is controllable and its transfer function ๐ท ห + ๐งห๐ถ( ห ๐ผห โ to โ(๐พ๐ โฏ ) โ ๐ด. Moreover, U [ ๐ดโฒ ๐ต โฒ ] โ1 ห ห ๐งห๐ด) ๐ต equals ๐(๐ง). Finally any controllable isometric colligation ๐ถ โฒ ๐ทโฒ : [ ๐ณ ๐ฐ ]โ [๐ณ ] ห with transfer function equal to ๐ is unitarily equivalent to U. ๐ด
Remark 1.5. We note that Theorem 1.4 can be derived directly from Theorem 1.3 as follows: use formulas (1.6) with ๐ โฏ in place of ๐ and with ๐ฐ and ๐ด switched to de๏ฌne the observable coisometric realization U for the function ๐ โฏ . Then the ห := Uโ will be exactly as described in Theorem 1.4. Explicit formulas colligation U (1.7) for operators adjoint to those in (1.6) are obtained via standard routine calculations. ห ๐ which In addition to the kernels ๐พ๐ and ๐พ๐ โฏ , there is a positive kernel ๐พ combines these two and is de๏ฌned as follows: โก โค โก โค ๐(๐ง) โ ๐(๐) ๐ผ โ ๐(๐ง)๐(๐)โ ๐(๐ง) โ ๐(๐) โข ๐พ๐ (๐ง, ๐) โฅ โข โฅ ๐งโ๐ 1 โ ๐ง๐ ๐งโ๐ ห ๐) = โข โฅ=โข โฅ ๐พ(๐ง, โฃ ๐ โฏ (๐ง) โ ๐ โฏ (๐) โฆ โฃ ๐(๐ง)โ โ ๐(๐)โ ๐ผ โ ๐(๐ง)โ ๐(๐) โฆ . ๐พ๐ โฏ (๐ง, ๐) ๐งโ๐ ๐งโ๐ 1 โ ๐ง๐ (1.8) ห is also a positive kernel on ๐ปร๐ป and the associated reproducing It turns out that ๐พ ห ๐ ) serves as the canonical functional-model state space kernel Hilbert space โ(๐พ for unitary realizations of ๐, as summarized in the following theorem.
80
J.A. Ball and V. Bolotnikov
Theorem 1.6. Suppose that the function ๐ is in the Schur class ๐ฎ(๐ฐ, ๐ด) and let ห ๐) be the positive kernel on ๐ป given by (1.8). De๏ฌne operators ๐ด, ห ๐ต, ห ๐ถ, ห ๐ท ห by ๐พ(๐ง, [ ] [ ] [ ] ๐(๐ง)โ๐(0) ๐ (๐ง) [๐ (๐ง) โ ๐ (0)]/๐ง ๐ข ๐ง ๐ด: โ , ๐ต : ๐ข โ , ๐ง๐(๐ง) โ ๐(๐ง)๐ (0) (๐ผ โ ๐(๐ง)โ ๐(0))๐ข [๐(๐ง) ] ๐ (๐ง) ๐ถ: โ ๐ (0), ๐ท : ๐ข โ ๐(0)๐ข. ๐(๐ง) ๐ด ๐ต ] de๏ฌnes a unitary operator from โ(๐พ ห๐ ) โ Then the operator colligation U = [ ๐ถ ๐ท ห ๐ ) โ ๐ด. Moreover, U is closely connected and its transfer function ๐ฐ onto โ(๐พ ๐ท + [๐ง๐ถ(๐ผ โ] ๐ง๐ด)โ1 ๐ต equals ๐(๐ง). Finally, any closely connected unitary colligaโฒ ๐ต โฒ : (๐ณ โ ๐ฐ) โ (๐ณ โ ๐ด) with transfer function equal to ๐ is unitarily tion ๐ด ๐ถ โฒ ๐ทโฒ equivalent to U.
1.2. Extensions to several variables One of the further developments inspired by the success of the state-space method to problems in single-variable function theory as developed by Israel Gohberg and his collaborators was the introduction of state-space methods to several-variable function theory. One can start with the seminal work of Agler [2] where what is now called the Schur-Agler class on the unit polydisk ๐ป๐ and where the realization of a multivariable function as the transfer function of a certain type of multidimensional linear system (with evolution along a multidimensional lattice โค๐+ rather than along the nonnegative integers โค+ ) were introduced. This polydisk setting was later developed in ๏ฌner detail in [3, 28]. Parallel but di๏ฌerent results were then developed for the setting of the unit ball in โ๐ in [4, 29, 44, 20], general domains [10, 16], algebraic curves and Riemann surfaces [30, 31, 32, 22], and domains in โ๐ with matrix-polynomial de๏ฌning function [10, 16] (see [14] for a survey up to the year 2000), and now one can see transfer-function realizations appear for functions of noncommuting variables as well [33, 8, 24, 25, 50, 59, 60, 61, 56, 57, 15, 21]. We focus here on the case where the unit disk ๐ป is replaced by the unit polydisk ๐ป๐ = {๐ง = (๐ง1 , . . . , ๐ง๐ ) : โฃ๐ง๐ โฃ < 1}. We then recall the Schur-Agler class ๐ฎ๐๐ (๐ฐ, ๐ด) consisting of โ(๐ฐ, ๐ด)-valued functions analytic on ๐ป๐ and such that โฅ๐(๐ )โฅ โค 1 for any collection of ๐ commuting operators ๐ = (๐1 , . . . , ๐๐ ) on a Hilbert space ๐ฆ. The operator ๐(๐ ) is de๏ฌned by ๐(๐ ) =
โ โ
๐๐ โ ๐ ๐ โ โ(๐ฐ โ ๐ฆ, ๐ด โ ๐ฆ)
๐โโค๐ +
if ๐(๐ง) =
โ
๐๐ ๐ง ๐
๐โโค๐ +
with the standard multivariable notation ๐ง ๐ = ๐ง1๐1 โ
โ
โ
๐ง๐๐๐
and
๐ ๐ = ๐1๐1 โ
โ
โ
๐๐๐๐
if
๐ = (๐1 , . . . , ๐๐ ) โ โค๐+ .
The following result appears in [2, 3, 28] and is a multivariable analog of Theorem 1.1.
Canonical Realization
81
Theorem 1.7. Let ๐ be a โ(๐ฐ, ๐ด)-valued function de๏ฌned on ๐ป๐ . The following statements are equivalent: (1) ๐ belongs to the class ๐ฎ๐๐ (๐ฐ, ๐ด). (2) There exist positive kernels ๐พ1๐ฟ , . . . , ๐พ๐๐ฟ : ๐ป๐ ร ๐ป๐ โ โ(๐ด) such that for every ๐ง = (๐ง1 , . . . , ๐ง๐ ) and ๐ = (๐1 , . . . , ๐๐ ) in ๐ป๐ , ๐ผ๐ด โ ๐(๐ง)๐(๐)โ =
๐ โ (1 โ ๐ง๐ ๐ ๐ )๐พ๐๐ฟ (๐ง, ๐).
(1.9)
๐=1
(3) There exist positive kernels ๐พ1๐
, . . . , ๐พ๐๐
: ๐ป๐ ร ๐ป๐ โ โ(๐ฐ) such that for every ๐ง, ๐ โ ๐ป๐ , ๐ผ๐ฐ โ ๐(๐ง)โ ๐(๐) =
๐ โ
(1 โ ๐ง๐ ๐ ๐ )๐พ๐๐
(๐ง, ๐).
(1.10)
๐=1
(4) There exist positive kernels [ ๐ฟ ] ๐พ๐ ๐พ๐๐ฟ๐
๐พ๐ = : ๐ป๐ ร ๐ป๐ โ โ(๐ด โ ๐ฐ) ๐พ๐๐
๐ฟ ๐พ๐๐
for
๐ = 1, . . . , ๐
such that for every ๐ง, ๐ โ ๐ป๐ , [ ] ๐ผ๐ด โ ๐(๐ง)๐(๐)โ ๐(๐ง) โ ๐(๐) ๐(๐ง)โ โ ๐(๐)โ ๐ผ๐ฐ โ ๐(๐ง)โ ๐(๐) ] ๐ [ โ (1 โ ๐ง๐ ๐ ๐ )๐พ๐๐ฟ (๐ง, ๐) (๐ง๐ โ ๐ ๐ )๐พ๐๐ฟ๐
(๐ง, ๐) = . (๐ง๐ โ ๐ ๐ )๐พ๐๐
๐ฟ (๐ง, ๐) (1 โ ๐ง๐ ๐ ๐ )๐พ๐๐
(๐ง, ๐)
(1.11)
(1.12)
๐=1
(5) There exist Hilbert spaces ๐ณ1 , . . . , ๐ณ๐ and a unitary connecting operator (or colligation) U of the structured form โก โค โก โค โก โค ๐ด11 . . . ๐ด1๐ ๐ต1 ๐ณ1 ๐ณ1 [ ] โข . โฅ โข โฅ โข . . . . โฅ .. .. โฅ โข .. โฅ ๐ด ๐ต โข . โข . โฅ U= =โข . (1.13) โฅ : โข โฅโโข . โฅ ๐ถ ๐ท โฃ๐ด๐1 . . . ๐ด๐๐ ๐ต๐ โฆ โฃ๐ณ๐ โฆ โฃ๐ณ๐ โฆ ๐ถ1 . . . ๐ถ๐ ๐ท ๐ฐ ๐ด so that ๐(๐ง) can be realized in the form โ1
๐(๐ง) = ๐ท + ๐ถ (๐ผ โ ๐(๐ง)๐ด) where we have set
โก ๐ง1 ๐ผ๐ณ1 โข ๐(๐ง) = โฃ
๐(๐ง)๐ต
for all ๐ง โ ๐ป๐ ,
(1.14)
โค ..
.
โฅ โฆ.
(1.15)
๐ง๐ ๐ผ๐ณ๐
(6) ๐(๐ง) has a realization as in (1.14) where the connecting operator U is any one of (i) isometric, (ii) coisometric, or (iii) contractive.
82
J.A. Ball and V. Bolotnikov
In the sequel we shall use the terminology left Agler decomposition for ๐, right Agler decomposition for ๐, or simply Agler decomposition for ๐, for any collection of kernels {๐พ ๐ฟ , . . . , ๐พ๐๐ฟ }, {๐พ1๐
, . . . , ๐พ๐๐
}, or {๐พ1 , . . . , ๐พ๐ } of the form (1.11), such that the respective decomposition (1.9), (1.10), or (1.12) holds. A straightforward calculation (see, e.g., [17]) shows that for an operator colligation U of the form (1.13) and its transfer function ๐ (1.14), we have ๐ผ โ ๐(๐ง)๐(๐)โ = ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 (๐ผ โ ๐(๐ง)๐(๐)โ ) (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ (1.16) [ ] [ ] ๐(๐)โ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ โ โ1 + ๐ถ(๐ผ โ ๐(๐ง)๐ด) ๐(๐ง) ๐ผ (๐ผ โ UU ) ๐ผ and ๐ผ โ ๐(๐ง)โ ๐(๐) = ๐ต โ (๐ผ โ ๐(๐ง)โ ๐ดโ )โ1 (๐ผ โ ๐(๐ง)โ ๐(๐)) (๐ผ โ ๐ด๐(๐))โ1 ๐ต (1.17) ] [ ] [ ๐(๐)(๐ผ โ ๐ด๐(๐))โ1 ๐ต , + ๐ต โ (๐ผ โ ๐(๐ง)โ ๐ดโ )โ1 ๐(๐ง)โ ๐ผ (๐ผ โ Uโ U) ๐ผ from which it is clear that for a contractive U and any analytic โ(โ๐๐=1 ๐ณ๐ )-valued function ๐ in ๐ complex variables (i.e., not necessarily of the form (1.15)), formula (1.14) de๏ฌnes an โ(๐ฐ, ๐ด)-valued function ๐ analytic and contractive-valued on the domain {๐ง โ โ๐ : โฅ๐(๐ง)โฅ๐๐ < 1}. The relevance of formula (1.14) to the polydisk setting is mostly determined by the very special structure (1.15) of ๐. Extending most univariate notions to the polydisk setting we will often refer to these extensions as โstructuredโ ones meaning that for other choices of ๐, these notions will be substantially di๏ฌerent. Our purpose here is to describe the analogs of Theorems 1.3, 1.4 and 1.6, i.e., canonical model realizations for a Schur-Agler class function as in Theorem 1.7 with the speci๏ฌcation of a uniqueness criterion, for the polydisk setting. It turns out that coisometric, isometric and unitary realizations are not so useful for a viable theory in the multivariable setting, and it makes sense to work with weakly coisometric, weakly isometric and weakly unitary realizations instead โ see Section 2 below for precise de๏ฌnitions. Also one no longer has strict uniqueness, but the amount of freedom in the choice of canonical functional-model realization can be well described. Once the analog of Theorem 1.3 (weakly coisometric canonical functional-model realizations) is understood, the analog of Theorem 1.4 (canonical functional-model isometric realization) follows by a simple duality argument just as in the univariate case. We note that the polydisk analog of Theorem 1.3 already appears in [17]; here we also obtain converse characterizations of which families of kernels can arise as left Agler decompositions for some Schur-Agler class function (Theorems 3.7 and 3.9 below). The main new point of the present paper then is to develop the analog of Theorem 1.6 (weakly unitary canonical functional-model realizations โ see Theorem 5.9 below) for the polydisk setting. We also construct weakly coisometric realizations for a simple example (๐(๐ง) = ๐ง1 ๐ง2 ) to illustrate the main ideas โ see Example 3.6 below.
Canonical Realization
83
We note that the paper [17] also develops the analogs of canonical functionalmodel isometric and unitary realizations (Theorem 1.4 and Theorem 1.6) for the ball setting (more precisely, the contractive-multiplier class on the Drury-Arveson space as in [4, 29]), and for the general domain setting (more precisely, the SchurAgler class associated with a domain with matrix-polynomial de๏ฌning function as in [10, 16]). In addition, in our companion paper [19] we work out explicitly the analogs of Theorem 1.4 and 1.6 (canonical functional-model isometric and coisometric realizations) for the ball setting and discuss applications to the canonical model theory for commuting row contractions as initiated in the work of Bhattacharyya, Eschmeier and Sarkar ([38, 39]). Finally, the paper [18] extends the isometric/unitary canonical functional-model model theory to the general-domain setting, thereby providing a general setting containing the polydisk and ball versions as special cases. The main point of the present paper is to obtain weakly unitary realizations for a Schur-Agler class function on the polydisk with a given Agler decomposition in a canonical functional-model form. The paper [26] also obtains canonical functional-model realizations but by a quite di๏ฌerent approach involving the construction of a multievolution Lax-Phillips scattering system having the given Schur-class function ๐ as its scattering function. In [26] the goal is to obtain realizations for ๐ which are actually unitary (not just weakly unitary); the construction is rather complicated unless the scattering system satis๏ฌes some additional minimality conditions. Nevertheless this scattering approach is used in [49] to characterize Schur-class functions on the polydisk in terms of a generalized Agler decomposition. We mention that there is recent work of Arov-Sta๏ฌans (see [12, 13]) connected with single-variable de Branges-Rovnyak spaces which uses the term โcanonical modelโ is a somewhat di๏ฌerent sense from how we are using it here. In [12, 13] the authors take a behavioral approach whereby one does not distinguish between the input space ๐ฐ and the output space ๐ด but rather focuses on the โsignal spaceโ ๐ด โ ๐ฐ. The authors then obtain โcoordinate-freeโ versions of the de Brangesห ๐ ) which amount to geometric Rovnyak model spaces โ(๐พ๐ ), โ(๐พ๐ โฏ ) and โ(๐พ Kreหin-space descriptions of the graph of the Schur-class function ๐. In the present paper, we do distinguish between the input space ๐ฐ and the output space ๐ด and the point is to pick out a unique (to the extent possible in the multivariable setting) choice of state space and realization (among all that are equivalent in an appropriate sense) which gives rise to a transfer-function realization for ๐ in the desired class of realizations. 1.3. The plan of the paper The present paper is organized as follows. Following this Introduction, in Section 2 we introduce the modi๏ฌcations (weakly coisometric, weakly isometric, weakly unitary realization) of the notions of coisometric, isometric and unitary realizations appearing in Theorems 1.3, 1.4, 1.6 which have proved useful for a viable theory in
84
J.A. Ball and V. Bolotnikov
the polydisk setting. Section 3 then recalls the analog of Theorem 1.3 for the polydisk setting from [17], namely, the characterization of a weakly coisometric canonical functional-model realization for a Schur-Agler class function on the polydisk. Section 4 applies the duality argument to arrive at the parallel result concerning weakly isometric canonical functional-model realizations, and the ๏ฌnal Section 5 gives the complete results concerning weakly unitary canonical functional-model realizations for a Schur-Agler-class function on the polydisk.
2. Preliminaries Here we review a few preliminaries from [28] concerning colligation matrices realizing Schur-Agler-class functions over the polydisk. We say that a colligation U of the form (1.13) is unitarily equivalent to a colligation [ โฒ ] [ ๐ ] ] [ ๐ ๐ด ๐ตโฒ โ๐=1 ๐ณ๐โฒ โ๐=1 ๐ณ๐โฒ โฒ U = โ (2.1) : ๐ถ โฒ ๐ทโฒ ๐ฐ ๐ด if there are unitary operators ๐๐ : ๐ณ๐ โ ๐ณ๐โฒ such that [ ๐ ] [ โฒ ][ ][ โ๐=1 ๐๐ 0 ๐ด ๐ต โฒ โ๐๐=1 ๐๐ ๐ด ๐ต = ๐ถ โฒ ๐ทโฒ 0 ๐ผ๐ด ๐ถ ๐ท 0
] 0 . ๐ผ๐ฐ
(2.2)
The de๏ฌnition of unitary equivalence is certainly structured โ the block-diagonal structure of the unitary operator โ๐๐=1 ๐๐ is predetermined by the structure (1.15) of ๐. Equality (2.2) is what we need to guarantee (as in the univariate case) that the transfer functions of two unitarily equivalent colligations coincide. We next extend De๏ฌnition 1.2 to the polydisk setting. De๏ฌnition 2.1. Let ๐๐ณ๐ denote the orthogonal projection of ๐ณ = โ๐๐=1 ๐ณ๐ onto ๐ณ๐ . The structured colligation (1.13) is called observable if โ ๐ณ๐ = ๐๐ณ๐ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐ฆ for ๐ = 1, . . . , ๐. ๐โ๐ป๐ , ๐ฆโ๐ด
It is called controllable if โ ๐ณ๐ =
๐๐ณ๐ (๐ผ โ ๐ด๐(๐))โ1 ๐ต๐ข
for ๐ = 1, . . . , ๐,
๐โ๐ป๐ , ๐ขโ๐ฐ
and it is called closely connected if โ{ } ๐ณ๐ = ๐๐ณ๐ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐ฆ, ๐๐ณ๐ (๐ผ โ ๐ด๐(๐))โ1 ๐ต๐ข : ๐ โ ๐ป๐ , ๐ฆ โ ๐ด, ๐ข โ ๐ฐ for ๐ = 1, . . . , ๐. In analogy with the univariate case, a realization of the form (1.14) is called coisometric, isometric, unitary or contractive if the operator U is respectively, coisometric, isometric, unitary or just contractive. Note that Statement (6) in Theorem 1.7 concerning contractive realizations does not appear in [2, 3, 28]; however its equivalence to statements (1)โ(5) is quite simple (see [17]). It turns out that a more useful analog of โisometricโ and โcoisometricโ realizations appearing
Canonical Realization
85
in the classical univariate case is not that the whole connecting operator U (or Uโ ) be isometric, but rather that U (respectively, Uโ ) be isometric on a certain canonical subspace of ๐ณ โ ๐ฐ (of ๐ณ โ ๐ด, respectively). As it is natural to restrict to contractive connecting operators U to realize contractive functions ๐, we maintain the restriction that connecting operators be contractive. De๏ฌnition 2.2. The colligation U of the form (1.13) is called 1. weakly coisometric if the adjoint Uโ is contractive as an operator from ๐ณ โ ๐ด to ๐ณ โ ๐ฐ and is isometric when restricted to the subspace ๐ โ ๐ด; here ๐ โ ๐ณ is given by โ ๐(๐)โ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐ฆ โ ๐ณ ; (2.3) ๐ := ๐โ๐ป๐ , ๐ฆโ๐ด
2. weakly isometric if U is contractive as an operator from ๐ณ โ ๐ฐ to ๐ณ โ ๐ด and ห โ ๐ฐ; here the subspace ๐ หโ๐ณ is isometric when restricted to the subspace ๐ is given by โ ห := ๐ ๐(๐)(๐ผ โ ๐ด๐(๐))โ1 ๐ต๐ข โ ๐ณ ; (2.4) ๐โ๐ป๐ , ๐ขโ๐ฐ
3. weakly unitary if it is simultaneously weakly isometric and weakly coisometric. The notions of weakly coisometric/isometric/unitary colligations do not appear in the single-variable context for the simple reason that if U is observable (controllable, closely connected), then a weakly coisometric (weakly isometric, weakly unitary) colligation is automatically coisometric (isometric, unitary). Remark 2.3. From the identity (1.16) we see that what is needed for the validity of the identity ๐ผ โ ๐(๐ง)๐(๐)โ = ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 (๐ผ โ ๐(๐ง)๐(๐)โ ) (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ
(2.5)
is that Uโ be isometric when restricted to the subspace ] [ ] [ โ ๐ณ ๐(๐)โ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ; ๐ฆโ ๐0 := ๐ด ๐ผ ๐ ๐โโ , ๐ฆโ๐ด
by specializing ๐ to ๐ = 0 in โ๐ , we see that ๐0 contains
[
{0} ๐ด
]
as a subspace and
hence ๐0 coincides with ๐ โ๐ด where ๐ is as in (2.3). We conclude that the weakly coisometric property is exactly what is needed for the identity (2.5) to hold. By a similar analysis working with the identity (1.17), we see that the weak isometric property is what we need for the validity of the dual decomposition ๐ผ โ ๐(๐ง)โ ๐(๐) = ๐ต โ (๐ผ โ ๐(๐ง)โ ๐ดโ )โ1 (๐ผ โ ๐(๐ง)โ ๐(๐)) (๐ผ โ ๐ด๐(๐))โ1 ๐ต.
(2.6)
86
J.A. Ball and V. Bolotnikov
Remark 2.4. A perusal of the formulas for the operators ๐ด and ๐ต in Theorem 1.3 (see formulas (1.6)) reveals the key role of the di๏ฌerence-quotient transformation ๐ค(๐ง) โ ๐ค(0) ๐ง acting on a function analytic at 0. A key property of the space โ(๐พ๐ ) in Theorem 1.3 is that it is invariant under the di๏ฌerence-quotient transformation ๐
. It has been recognized for some time now that the multivariable analog of the di๏ฌerencequotient transformation is any solution of the Gleason problem (see [46, 5, 6, 9]). Given a space โ of holomorphic functions โ which are holomorphic in a neighborhood of the origin in ๐-dimensional complex Euclidean space โ๐ , we say that the operators ๐
1 , . . . , ๐
๐ mapping โ into itself solve the Gleason problem for โ if every function โ โ โ has a decomposition of the form ๐
: ๐ค(๐ง) โ
โ(๐ง) โ โ(0) =
๐ โ
๐ง๐ [๐
๐ โ](๐ง).
๐=1
We shall see that solutions of more structured Gleason problems enter into the de๏ฌnition of the colligation matrices for canonical functional models in the polydisk setting discussed below.
3. Weakly coisometric realizations For every function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) with a ๏ฌxed left Agler decomposition (1.9), one can construct a weakly coisometric realization in a certain canonical way; we now recall the construction from [17]. Let ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) be given along with the kernel collection {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } providing the left Agler decomposition (1.9) for ๐. We say that an operator-block ๐ matrix ๐ด = [๐ด๐๐ ]๐,๐=1 acting on โ๐๐=1 โ(๐พ๐๐ฟ ) solves the structured Gleason problem for the kernel collection {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } if the identity ๐1 (๐ง) + โ
โ
โ
+ ๐๐ (๐ง) โ [๐1 (0) + โ
โ
โ
+ ๐๐ (0)] =
๐ โ
๐ง๐ [๐ด๐ ]๐ (๐ง)
(3.1)
๐=1
holds for all ๐ = โ๐๐=1 ๐๐ โ โ๐๐=1 โ(๐พ๐๐ฟ ), where we write โก โค [๐ด๐ ]1 (๐ง) โข โฅ .. ๐ด๐ (๐ง) = โฃ โฆ โ โ๐๐=1 โ(๐พ๐๐ฟ ). . [๐ด๐ ]๐ (๐ง)
(We refer back to Remark 2.4 for a discussion of the โunstructuredโ Gleason problem.) We say that the operator ๐ต : ๐ฐ โ โ๐๐=1 โ(๐พ๐๐ฟ ) solves the structured Gleason problem for ๐ if the identity ๐(๐ง)๐ข โ ๐(0)๐ข =
๐ โ ๐=1
๐ง๐ [๐ต๐ข]๐ (๐ง) holds for all ๐ข โ ๐ฐ.
(3.2)
Canonical Realization
87
De๏ฌnition 3.1. Given ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด), we shall say that the block-operator matrix [ ] [ ๐ ] [ ๐ ] ๐ด ๐ต โ๐=1 โ(๐พ๐๐ฟ ) โ๐=1 โ(๐พ๐๐ฟ ) U= : โ (3.3) ๐ถ ๐ท ๐ฐ ๐ด is a canonical functional-model (abbreviated to c.f.m. in what follows) colligation associated with left Agler decomposition (1.9) for ๐ if 1. 2. 3. 4.
U is contractive. ๐ด : โ๐๐=1 โ(๐พ๐๐ฟ ) โ โ๐๐=1 โ(๐พ๐๐ฟ ) solves the structured Gleason problem (3.1). ๐ต : ๐ฐ โ โ๐๐=1 โ(๐พ๐๐ฟ ) solves the structured Gleason problem (3.2) for ๐. The operators ๐ถ : โ๐๐=1 โ(๐พ๐๐ฟ ) โ ๐ด and ๐ท : ๐ฐ โ ๐ด are given by ๐ถ : ๐ (๐ง) โ ๐1 (0) + โ
โ
โ
+ ๐๐ (0),
๐ท : ๐ข โ ๐(0)๐ข.
(3.4)
Equalities (3.1), (3.2) and (3.4) can be equivalently reformulated in terms of adjoint operators ๐ดโ , ๐ต โ , ๐ถ โ and ๐ทโ as follows. With a given left Agler decomposition {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } of a function ๐ โ ๐ฎ๐ด๐ (๐ฐ, ๐ด), we associate the kernel โก ๐ฟ โค ๐พ1 (๐ง, ๐) โข โฅ .. ๐ ๐ ๐ ๐๐ฟ (๐ง, ๐) := โฃ (3.5) โฆ : ๐ป ร ๐ป โ โ(๐ด, ๐ด ) . ๐พ๐๐ฟ (๐ง, ๐)
and the subspace
โก
โค ๐ 1 ๐พ1๐ฟ (โ
, ๐)๐ฆ โข โฅ .. ๐= ๐(๐)โ ๐๐ฟ (โ
, ๐)๐ฆ = โฃ โฆ โ โ๐๐=1 โ(๐พ๐๐ฟ ). (3.6) . ๐โ๐ป๐ , ๐ฆโ๐ด ๐โ๐ป๐ , ๐ฆโ๐ด ๐ ๐ ๐พ๐๐ฟ (โ
, ๐)๐ฆ โ
โ
Proposition 3.2. Given ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด), the block-operator matrix U of the form (3.3) is a c.f.m. colligation associated with left Agler decomposition {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } for ๐ if and only if U is contractive and [ โ ] [ ] [ ๐ฟ ] ๐ด ๐ถโ ๐(๐)โ ๐๐ฟ (โ
, ๐)๐ฆ ๐ (โ
, ๐)๐ฆ Uโ = : โ (3.7) ๐ต โ ๐ทโ ๐ฆ ๐(๐)โ ๐ฆ where ๐๐ฟ is de๏ฌned in (3.5). Proof. It follows by straightforward inner-product calculations (see Proposition 3.4 and Remark 3.6 in [17] for details) that identities (3.1), (3.2) and equalities (3.4) are equivalent respectively to ๐ดโ : ๐(๐)โ ๐๐ฟ (โ
, ๐)๐ฆ โ ๐๐ฟ (โ
, ๐)๐ฆ โ ๐๐ฟ (โ
, 0)๐ฆ, โ
โ
๐ฟ
โ
โ
๐ต : ๐(๐) ๐ (โ
, ๐)๐ฆ โ ๐(๐) ๐ฆ โ ๐(0) ๐ฆ, โ
๐ฟ
๐ถ : ๐ฆ โ ๐ (โ
, 0)๐ฆ,
โ
โ
๐ท : ๐ฆ โ ๐(0) ๐ฆ.
(3.8) (3.9) (3.10)
It remains to show that the last four equalities are equivalent to (3.7). Indeed, substituting the ๏ฌrst and the second equality from (3.10) into (3.8) and (3.9)
88
J.A. Ball and V. Bolotnikov
respectively, we get ๐ดโ ๐(๐)โ ๐๐ฟ (โ
, ๐)๐ฆ + ๐ถ โ ๐ฆ = ๐๐ฟ (โ
, ๐)๐ฆ, ๐ต โ ๐(๐)โ ๐๐ฟ (โ
, ๐)๐ฆ + ๐ท โ ๐ฆ = ๐(๐)โ ๐ฆ which express equalities of the top and of the bottom components in (3.7). Conversely, upon setting ๐ = 0 in (3.7) and taking into account that ๐(0) = 0 we get equalities (3.10). Substituting (3.10) back into (3.7) and comparing the top and the bottom components we arrive at (3.8) and (3.9). โก On the other hand, given an ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) with a ๏ฌxed left Agler decomposition {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ }, one can rearrange identity (1.9) as ๐ โ
๐ง๐ ๐ ๐ ๐พ๐๐ฟ (๐ง, ๐) + ๐ผ๐ด =
๐=1
๐ โ
๐พ๐๐ฟ (๐ง, ๐) + ๐(๐ง)๐(๐)โ
๐=1
and write the latter in the inner product form as ๐ โ ๐=1
โจ๐ยฏ๐ ๐พ๐๐ฟ (โ
, ๐)๐ฆ, ๐งยฏ๐ ๐พ๐๐ฟ (โ
, ๐ง)๐ฆ โฒ โฉโ(๐พ๐๐ฟ ) + โจ๐ฆ, ๐ฆ โฒ โฉ๐ด =
๐ โ ๐=1
or equivalently, as โฉ[
โจ๐พ๐๐ฟ (โ
, ๐)๐ฆ, ๐พ๐๐ฟ (โ
, ๐ง)๐ฆ โฒ โฉโ(๐พ๐๐ฟ ) + โจ๐(๐)โ ๐ฆ, ๐(๐ง)โ ๐ฆ โฒ โฉ๐ฐ ,
] [ ]โช ๐(๐)โ ๐๐ฟ (โ
, ๐)๐ฆ ๐(๐ง)โ ๐๐ฟ (โ
, ๐ง)๐ฆ โฒ , ๐ฆ ๐ฆโฒ ๐ฟ โ๐ ๐=1 โ(๐พ๐ )โ๐ด โฉ[ ๐ฟ ] [ ๐ฟ ]โช โฒ ๐ (โ
, ๐)๐ฆ ๐ (โ
, ๐ง)๐ฆ = , ๐(๐)โ ๐ฆ ๐(๐ง)โ ๐ฆ โฒ โ๐ โ(๐พ ๐ฟ )โ๐ฐ ๐=1
๐
where ๐๐ฟ is given in (3.5). It follows from the last identity that the linear map [ ] [ ๐ฟ ] ๐(๐)โ ๐๐ฟ (โ
, ๐)๐ฆ ๐ (โ
, ๐)๐ฆ ๐: โ (3.11) ๐ฆ ๐(๐)โ ๐ฆ extends to the isometry from [ ] [ ๐ ] โ ๐(๐)โ ๐๐ฟ (โ
, ๐)๐ฆ โ๐=1 โ(๐พ๐๐ฟ ) ๐๐ = โ ๐ฆ ๐ด ๐ ๐โ๐ป , ๐ฆโ๐ด
onto โ๐ =
โ ๐โ๐ป๐ , ๐ฆโ๐ด
[
๐๐ฟ (โ
, ๐)๐ฆ ๐(๐)โ ๐ฆ
]
[
] โ๐๐=1 โ(๐พ๐๐ฟ ) โ . ๐ฐ
By the same argument as used in Remark 2.3 for the splitting of the subspace ๐0 there, we see that the subspace ๐๐ splits as a direct sum ๐๐ = ๐ โ ๐ด where ๐ is given in (3.6) and that the defect spaces ๐โฅ := (โ๐ โ(๐พ ๐ฟ ) โ ๐ด) โ ๐๐ โผ = ๐โฅ and โโฅ := (โ๐ โ(๐พ ๐ฟ ) โ ๐ฐ ) โ โ๐ ๐
๐=1
๐
๐
๐=1
๐
Canonical Realization can be characterized as { โฅ
๐ =
๐=
๐ โ ๐=1
โโฅ ๐
๐๐ โ
๐ โ
โ(๐พ๐๐ฟ )
๐=1
:
๐ โ
89 }
๐ง๐ ๐๐ (๐ง) โก 0 ,
(3.12)
๐=1
{[ ] [ } ] โ ๐ ๐ โ๐๐=1 โ(๐พ๐๐ฟ ) โ : = ๐๐ (๐ง) + ๐(๐ง)๐ข โก 0 . ๐ข ๐ฐ
(3.13)
๐=1
Combining (3.11) with Proposition 3.2 we arrive at the following. Lemma 3.3. Given a left Agler decomposition {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } for a function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด), let ๐ be the isometric operator associated with this decomposition as in (3.11). A block-operator matrix U of the form (3.3) is a c.f.m. colligation associated with {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } if and only if Uโ is a contractive extension of ๐ to all of (โ๐๐=1 โ(๐พ๐๐ฟ )) โ ๐ด, i.e., Uโ โฃ๐โ๐ด = ๐
and
โฅUโ โฅ โค 1.
(3.14)
The following theorem is the multivariable counterpart of Theorem 1.3 for the polydisk setting. The ๏ฌrst statement is an immediate consequence of Lemma 3.3 and we refer to [17, Theorem 3.9] for the rest. Theorem 3.4. Let ๐ be a function in the Schur-Agler class ๐ฎ๐ด๐ (๐ฐ, ๐ด) with given left Agler decomposition {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ }. Then ๐ฟ ๐ฟ ๐ต 1. There exists a c.f.m. colligation U = [ ๐ด ๐ถ ๐ท ] associated with {๐พ1 , . . . , ๐พ๐ }. ๐ฟ ๐ฟ 2. Every c.f.m. colligation U associated with {๐พ1 , . . . , ๐พ๐ } is weakly coisometric and observable and furthermore, ๐(๐ง) = ๐ท + ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 ๐(๐ง)๐ต. 3. Any observable weakly coisometric colligation Uโฒ of the form (2.1) with its transfer function equal to ๐ is unitarily equivalent to some c.f.m. colligation U for ๐. We next describe all c.f.m. colligations associated with a given left Agler decomposition (1.9) of a function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด). By Lemma 3.3, it su๏ฌces to โ describe]all solutions [ ] contractive completion problem (3.14). Identifying [ ๐ U of๐ฟ the โฅ ๐ โ๐=1 โ(๐พ๐ ) we then can represent Uโ from (3.14) as with ๐ด ๐โ๐ด [ โฅ ] ] [ ๐ [ ] ๐ โ๐=1 โ(๐พ๐๐ฟ ) โ (3.15) โ U = ๐ ๐ : ๐ฐ ๐โ๐ด [ โ ] [ ๐ ] ๐ด โฃ๐ โฅ โ๐=1 โ(๐พ๐๐ฟ ) โฅ where ๐ = is unknown. โ : ๐ ๐ฐ ๐ต โ โฃ๐ โฅ Theorem 3.5. Let {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } be a ๏ฌxed Agler decomposition of a given function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด). Let ๐ be the associated isometry de๏ฌned in (3.11) with the defect spaces ๐โฅ and โโฅ ๐ de๏ฌned in (3.12), (3.13). Then all c.f.m. colligations associated with {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } are described by formula (3.15) where ๐ is an arbitrary contraction from ๐ โฅ into โโฅ ๐ . Moreover,
90
J.A. Ball and V. Bolotnikov
1. There exists an isometric c.f.m. colligation U associated with {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } if and only if dim ๐โฅ โฅ dim โโฅ ๐ . All such colligations are of the form (3.15) where ๐ : ๐โฅ โ โโฅ ๐ is a coisometry. 2. There exists a coisometric c.f.m. colligation U associated with {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } if and only if dim ๐โฅ โค dim โโฅ ๐ . All such colligations are of the form (3.15) where ๐ : ๐โฅ โ โโฅ ๐ is an isometry. 3. There exists a unitary c.f.m. colligation U associated with {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } if and only if dim ๐โฅ = dim โโฅ ๐ . All such colligations are of the form (3.15) where ๐ : ๐โฅ โ โโฅ is unitary. ๐ Proof. The operator Uโ of the form (3.15) is a contraction if and only if ๐๐ โ โค ๐ผ โ ๐ ๐ โ = ๐โโฅ ๐ which in turn, is equivalent to ๐ being a contraction from ๐โฅ into โโฅ ๐ . The operator Uโ is coisometric if and only if ๐๐ โ = ๐ผ โ ๐ ๐ โ = ๐โโฅ and statement ๐ (1) follows. Furthermore, Uโ of the form (3.15) is a coisometry if and only if ๐ โ ๐ = ๐ผ๐โ๐ด ,
๐ โ๐ = 0
and
๐ โ ๐ = ๐ผ๐ โฅ .
(3.16)
The ๏ฌrst equality in (3.16) holds since ๐ is isometric and the second equality holds since the ranges of ๐ and of ๐ are orthogonal. Thus, Uโ of the form (3.15) is coisometric if and only ๐ is isometric which completes the proof of (2). Statement (3) follows from (1) and (2). โก Example 3.6. If ๐1 , ๐2 : ๐ป โ ๐ป are Schur-class functions, then the function ๐(๐ง1 , ๐ง2 ) = ๐1 (๐ง1 ) โ
๐2 (๐ง2 ) belongs to the class ๐ฎ๐2 (โ, โ). Making use of coisometric functional-model realizations ห ๐ + ๐ง๐ ๐ถ ห๐ (๐ผโ(๐พ ) โ ๐ง๐ ๐ด ห๐ )โ1 ๐ต ห๐ ๐๐ (๐ง๐ ) = ๐ท ๐๐
(๐ = 1, 2)
for ๐1 and ๐2 provided by Theorem 1.3, one can easily get a coisometric realization (1.14) for ๐ with the state space equal โ(๐พ๐1 ) โ โ(๐พ๐2 ) and with operators ๐ด, ๐ต, ๐ถ, ๐ท given by [ ] [ ] [ ] ห1 ๐ต ห1 ๐ถ ห2 ๐ด ห 1๐ท ห 2, ๐ถ = ๐ถ ห1 ๐ท ห1๐ถ ห2 , ๐ต = ๐ต ห1 ๐ท ห2 ๐ต ห2 , ๐ด = ๐ท=๐ท ห2 . 0 ๐ด However, this realization is not a c.f.m. in the sense of De๏ฌnition 3.1 since the ห 1 = ๐1 (0) = 0). operator ๐ถ is not of the requested form (3.4) (unless ๐ท To construct a c.f.m. realization, we should choose an Agler decomposition for ๐. One possible decomposition is the following: 1 โ ๐(๐ง)๐(๐)โ = (1 โ ๐ง1 ๐ 1 )๐พ1 (๐ง, ๐) + (1 โ ๐ง2 ๐ 2 )๐พ2 (๐ง, ๐),
(3.17)
where ๐พ1 (๐ง, ๐) = ๐พ๐1 (๐ง1 , ๐1 ),
๐พ2 (๐ง, ๐) = ๐1 (๐ง1 )๐1 (๐1 )๐พ๐2 (๐ง2 , ๐2 )
(3.18)
Canonical Realization
91
and the kernels ๐พ๐๐ are de๏ฌned as in (1.4). Thus, โ(๐พ1 ) = โ(๐พ๐1 ) and โ(๐พ2 ) = ๐1 โ(๐พ๐2 ). Since the identity ๐ง1 โ1 (๐ง1 ) + ๐ง2 ๐1 (๐ง1 )โ2 (๐ง2 ) โก 0 implies โ1 โก โ2 โก 0, the subspace ๐โฅ (3.12) is trivial and therefore, there exists a unique c.f .m. colligation associated with decomposition (3.17). Explicit formulas for the entries ๐ด and ๐ต in terms of their adjoints follow (upon some straightforward manipulations) from (3.7): ๐พ๐1 (โ
, ๐1 ) โ ๐พ๐1 (โ
, 0) , ๐ดโ21 = 0, ๐1 ๐1 (๐1 ) โ ๐1 (0) ๐ดโ12 : ๐พ๐1 (โ
, ๐1 ) โ ๐1 โ
๐พ๐2 (โ
, 0) โ
, ๐1 ๐พ๐2 (โ
, ๐2 ) โ ๐พ๐2 (โ
, 0) ๐ดโ22 : ๐1 ๐พ๐2 (โ
, ๐2 ) โ ๐1 โ
, ๐2 ๐1 (๐1 ) โ ๐1 (0) ๐ต1โ : ๐พ๐1 (โ
, ๐1 ) โ โ
๐2 (0), ๐1 ๐2 (๐2 ) โ ๐2 (0) ๐ต2โ : ๐1 ๐พ๐2 (โ
, ๐2 ) โ ๐2 ๐ดโ11 : ๐พ๐1 (โ
, ๐1 ) โ
(3.19)
where the unspeci๏ฌed argument is either ๐ง1 or ๐ง2 depending on the context. Note that this colligation is coisometric and that ๐ด11 and ๐ด22 are backward shift operators on โ(๐พ1 ) and โ(๐พ2 ), respectively. On the other hand, one can consider a di๏ฌerent Agler decomposition (3.17) for ๐ where ๐พ1 (๐ง, ๐) = 12 ๐พ๐1 (๐ง1 , ๐1 ) โ
(1 + ๐2 (๐ง2 )๐2 (๐2 )), (3.20) ๐พ2 (๐ง, ๐) = 12 ๐พ๐2 (๐ง2 , ๐2 ) โ
(1 + ๐1 (๐ง1 )๐1 (๐1 )). With respect to this decomposition, the subspace ๐โฅ can be nontrivial, in which case there will exist a family of non-equivalent c.f.m. realizations. Here is a simple illustrative example. Let ๐(๐ง1 , ๐ง2 ) = ๐ง1 ๐ง2 . For the Agler representation (3.17), (3.18) we now have 1 โ ๐(๐ง)๐(๐)โ = (1 โ ๐ง1 ๐ 1 ) โ
1 + (1 โ ๐ง2 ๐ 2 ) โ
๐ง1 ๐ 1 .
(3.21)
The formulas (3.19) now take the form ๐ดโ11 = 0,
๐ดโ21 = 0,
๐ดโ22 = 0, ๐ดโ12 : 1 โ ๐ง1 , ๐ต1โ = 0, ๐ต2โ : ๐ง1 โ 1. {[ 1 ] [ 0 ] [ 0 ]} 0 , ๐ง1 , 0 With respect to the basis of โ(๐พ1 ) โ โ(๐พ2 ) โ โ, the matrix 0 1 0 of the c.f .m. colligation U has the form โก โค 0 1 0 U = โฃ0 0 1โฆ 1 0 0
92
J.A. Ball and V. Bolotnikov
and it is not hard to see that indeed ( [ ][ ])โ1 [ [ ] ๐ง1 0 0 1 ๐ง1 1 0 ๐ผ2 โ 0 ๐ง2 0 0 0
0 ๐ง2
][ ] [ [ ] 1 0 = 1 0 1 0
๐ง1 1
][ ] 0 = ๐ง1 ๐ง2 . ๐ง2
For the alternative Agler decomposition of ๐(๐ง) = ๐ง1 ๐ง2 in accordance to (3.20) we use the kernels ๐พ1 (๐ง, ๐) =
1 1 โ
(1 + ๐ง2 ๐ 2 ) and ๐พ2 (๐ง, ๐) = โ
(1 + ๐ง1 ๐ 1 ). 2 2
๐ง2 ๐ง1 We ๏ฌx orthonormal bases { โ12 , โ }, { โ12 , โ } and {1} for โ(๐พ1 ), โ(๐พ2 ) and โ, 2 2 โฅ respectively. Note that the subspaces ๐ (3.12) and โโฅ ๐ are given by {[ 1 ]} ๐ง2 โ1 ๐โฅ = span {[ โ๐ง ]} , โโฅ . (3.22) ๐ = span 1 0
In particular both ๐โฅ and โโฅ ๐ are nontrivial and of the same dimension 1. [ ] [ ] 1/2 โ(๐พ1 ) Formulas (3.10) give ๐ท โ = 0 and ๐ถ โ : 1 โ โ . Therefore the 1/2 โ(๐พ2 ) matrix representations of ๐ท and ๐ถ with respect to the ๏ฌxed choice of bases are [ ] ๐ท = 0, ๐ถ = โ12 0 โ12 0 . (3.23) Formulas (3.8), (3.9) amount to โก โ โค [ โก โค ] ๐ง2 ๐ 2 ๐ด11 ๐ดโ21 ๐ (1 + ๐ง ๐ ) 2 1 2 โฃ๐ดโ12 ๐ดโ22 โฆ : โ โฃ ๐ง1 ๐ 1 โฆ . ๐ (1 + ๐ง ๐ ) โ โ 1 2 1 ๐ต1 ๐ต2 2๐ 1 ๐ 2
(3.24)
Letting ๐1 = 0, ๐2 โ= 0 and then ๐2 = 0, ๐1 โ= 0 we get the action of all operators in (3.24) on constant functions: ๐ดโ11 : 1 โ 0, ๐ต1โ : 1 โ 0,
๐ดโ12 : 1 โ ๐ง1 ,
๐ต2โ : 1 โ 0.
๐ดโ21 : 1 โ ๐ง2 ,
๐ดโ22 : 1 โ 0,
Substituting these actions back into (3.24) we get the additional relation โก โ โค โก โค [ โ ] 0 ๐ด11 ๐ดโ21 โฃ๐ดโ12 ๐ดโ22 โฆ : ๐ง2 /โ2 โ โฃ 0 โฆ . โ ๐ง1 / 2 ๐ต1โ ๐ต2โ 2
(3.25)
(3.26)
From the characterization of the spaces ๐โฅ and โโฅ ๐ in (3.22), we see that the only freedom in the c.f.m. colligation U associated with this Agler decomposition is given by โก โ โค โก โ โค โ ] [ ๐ด11 ๐ดโ21 1/ โ2 โฃ๐ดโ12 ๐ดโ22 โฆ : ๐ง2 / โ2 โ ๐ โฃโ1/ 2โฆ (3.27) โ๐ง1 / 2 ๐ต1โ ๐ต2โ 0
Canonical Realization
93
where ๐ โ โ has โฃ๐โฃ โค 1. When we combine (3.26) and (3.27) we see that โ โค โก โ โค โก โ โค โก โ ]) [ ๐/2 โ2 ๐ด11 ๐ด11 ๐ดโ21 ( [ โ ] โ 1 1 ๐ง / 2 ๐ง / 2 2 โ 2 โ โฃ๐ดโ12 โฆ ๐ง2 / 2 = โฃ๐ดโ12 ๐ดโ22 โฆ โฆ + = โฃโ๐/2 โ 2 2 ๐ง2 / 2 2 โ๐ง1 / 2 โ โ โ ๐ต1 ๐ต1 ๐ต2 1/ 2 and, similarly, โก โ โค โก โ ๐ด21 ๐ด11 โ โฃ๐ดโ22 โฆ ๐ง1 / 2 = โฃ๐ดโ12 ๐ต2โ ๐ต1โ
โ โค โค โก โ ]) [ โ๐/2โ 2 ๐ดโ21 ( [ โ ] 1 1 ๐ง2 /โ2 ๐ง2 / โ2 ๐ดโ22 โฆ โ = โฃ ๐/2โ 2 โฆ . 2 ๐ง1 / 2 2 โ๐ง1 / 2 โ ๐ต2 1/ 2
Combining all these with the formulas (3.25) and (3.23), we now conclude that, with respect to the bases chosen as above, the matrix of the c.f .m. colligation U has the form โก โค 0 0 0 1 0โ โข ๐/2 0 โ๐/2 0 1/ 2 โฅ โข โฅ 1 0 0 0โ โฅ U=โข (3.28) โข 0 โฅ. โฃ โ๐/2 0 ๐/2 0 1/ 2 โฆ โ โ 1/ 2 0 1/ 2 0 0 For every choice of ๐ with โฃ๐โฃ โค 1 we have ( [ ] )โ1 [ ] ๐ง1 ๐ผ2 0 ๐ง1 ๐ผ2 0 ๐ท + ๐ถ ๐ผ4 โ ๐ด ๐ต 0 ๐ง2 ๐ผ2 0 ๐ง2 ๐ผ2 โก โคโ1 โก โค 1 0 0 โ๐ง1 0 ๐ โฅ โข๐ง1 โฅ ] โขโ ๐ ๐ง1 1[ 1 ๐ง 0 1 2 2 โฅ โข โฅ 1 0 1 0 โข = โฃ 0 โ๐ง2 1 0 โฆ โฃ0โฆ 2 ๐ ๐ง 0 โ ๐2 ๐ง2 1 ๐ง2 2 2 โก โคโก โค ๐ 2 โ ๐ง ๐ง โ ๐ง1 (1 + ๐2 ๐ง1 ๐ง2 ) 0 2 1 2 โฅ โข๐ง1 โฅ [ ] โขโ 1 โ โ โ โฅโข โฅ 1 0 1 0 โข = ๐ 2 โฃโ ๐ง2 (1 + ๐ ๐ง1 ๐ง2 ) โ โฆโฃ 0โฆ ๐ง ๐ง 2(1 + ๐๐ง1 ๐ง2 ) 2 2 1 2 โ โ โ โ ๐ง2 [ ][ ] ๐ ๐ 2 [ ] 1 ๐ง1 (1 + 2 ๐ง1 ๐ง2 ) ๐ง1 2 ๐ง1 ๐ง2 1 1 = = ๐ง1 ๐ง2 ๐ 2 ๐ง2 (1 + ๐2 ๐ง1 ๐ง2 ) ๐ง ๐ง ๐ง2 2(1 + ๐๐ง1 ๐ง2 ) 2 1 2 as expected. Note that the realization is weakly coisometric for any choice of ๐ with โฃ๐โฃ โค 1 and is unitary when โฃ๐โฃ = 1. Finally, we note that the more general example ๐(๐ง) = ๐ง1๐ ๐ง2๐ can be handled in much the same way; we welcome the reader to work out the details. We close this section with some discussion of various loose ends suggested by the results of this section. 3.1. Characterization of left Agler decompositions We have seen that construction of a c.f.m. for a Schur-Agler-class function ๐ requires knowledge of a left Agler decomposition for ๐. A natural question then is: which collections of kernels {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } arise as a left Agler decomposition
94
J.A. Ball and V. Bolotnikov
for some Schur-Agler class function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด)? The following result gives an intrinsic, although arguably not particularly easily checkable, characterization of such kernel collections. Theorem 3.7. Let {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } be a collection of ๐ โ(๐ด)-valued positive kernels on ๐ป๐ . Then {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } is a left Agler decomposition for some Schur-Aglerclass function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) (for some appropriate input space ๐ฐ) if and only if there exists a solution ๐ด = [๐ด๐๐ ]๐๐,๐=1 of the structured Gleason problem (3.1) which is contractive in the sense that ๐ โ
โฅ[๐ด๐ ]๐ โฅ2 โค
๐=1
[๐ ] 1 โ๐ for all ๐ = .. โ ๐=1 โ(๐พ๐๐ฟ ). .
๐ โ ๐=1
โฅ๐๐ โฅ2โ(๐พ ๐ฟ ) โ ๐
๐ โ
โฅ๐๐(0)โฅ2๐ด
(3.29)
๐=1
๐
๐ โ๐ ๐ฟ Moreover, if this is the case and if we de๏ฌne ๐ถ : ๐=1 โ(๐พ๐ ) โ ๐ด by [๐ ] 1 (3.30) ๐ถ : .. โ ๐1 (0) + โ
โ
โ
+ ๐๐ (0), .
๐๐
๐ต ] from ๐ฐ to then there exists a choice of operator [ ๐ท ๐ด ๐ต [ ๐ถ ๐ท ] is a c.f.m. for ๐.
โ๐
๐=1
โ(๐พ๐๐ฟ ) โ ๐ด so that
Proof. Necessity is immediate from Theorem 3.4 and the de๏ฌnition of c.f.m. Conversely suppose that we are given a collection {๐พ1๐ฟ, . . . , ๐พ๐๐ฟ } of โ(๐ด)valued positive kernels over ๐ป๐ for which there exists a contractive solution ๐ด = โ๐ ๐ฟ [๐ด๐๐ ]๐๐,๐=1 of the Gleason problem (3.1). De๏ฌne the operator ๐ถ : ๐=1 โ(๐พ๐ ) โ ๐ด as in (3.30). By the assumption that ๐ด is a contractive solution of the Gleason problem, ๐ด from โ๐๐=1 โ(๐พ๐๐ฟ ) to it follows that ) the block column matrix [ ๐ถ ] is contractive ( ๐ ๐ฟ ๐ต โ๐=1 โ(๐พ ( ๐ ) โ ๐ด. We ) may then construct an operator [ ๐ท ] from an input space ๐ฐ into โ๐๐=1 โ(๐พ๐๐ฟ ) โ ๐ด as a solution of the Cholesky factorization problem: ] [ ] [ [ ] ] ] ๐ด [ โ ๐ผ 0 ๐ต [ โ ๐ทโ = ๐ด ๐ถโ . ๐ต โ ๐ถ 0 ๐ผ ๐ท ๐ด ๐ต ] is coisometric. If we de๏ฌne ๐(๐ง) = It then follows that the colligation matrix [ ๐ถ ๐ท ๐ท + ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 ๐(๐ง)๐ต, then ๐ is in the Schur-Agler class ๐ฎ๐๐ (๐ฐ, ๐ด) and the identity (1.16) leads to the representation (2.5) for ๐ผ โ ๐(๐ง)๐(๐)โ . It is convenient to introduce the notation โ๐ : โ(๐พ๐ ) โ โ๐๐=1 โ(๐พ๐ ) for the inclusion map of โ(๐พ๐๐ฟ ) into the direct-sum space โ๐๐=1 โ(๐พ๐๐ฟ ) as the ๐th coordinate with the other coordinates equal to zero. We then have that the adjoint โ๐โ of โ๐ is given by [ ]
โ๐โ : ๐ =
๐1
.. .
๐๐
โ [๐ ]๐ := ๐๐ .
Canonical Realization
95
In addition we let ๐๐ be the projection operator on โ๐๐=1 โ(๐พ๐ ) given by ๐๐ = โ๐ โ๐โ . We next argue that we recover the kernel ๐พ๐๐ฟ (๐ง, ๐) as ๐พ๐๐ฟ (๐ง, ๐) = ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 ๐๐ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ .
(3.31)
By the de๏ฌnition of ๐ด solving the Gleason problem, we see that, for any ๐๐ โ โ(๐พ๐๐ฟ ), ๐ โ ๐ง๐1 [๐ดโ๐ ๐๐ ]๐1 (๐ง). ๐๐ (๐ง) = ๐๐ (0) + ๐1 =1
We then apply the Gleason-problem identity (3.1) to each โ๐1 [๐ดโ๐ ๐๐ ]๐1 and iterate to get โ ([ ๐ ๐ ] โ โ ๐๐ (๐ง) = ๐๐ (0) + ๐ง๐1 โ[๐ดโ๐ ๐๐ ]๐1 (0) + ๐ง๐2 ๐ดโ๐1 [๐ดโ๐ ๐๐ ]๐1 (0) ๐1 =1
+
๐ โ
๐ง๐3
๐3 =1
= ๐ถโ๐ ๐๐ +
๐ โ
๐2 ๐ 3
โ ๐ง๐1 โ๐ถ๐๐1 ๐ดโ๐ ๐๐ +
๐1 =1
+
๐ โ ๐3 =1
=
โ โ
๐ถ
๐=0
๐ โ
๐ง๐3 โ๐ถ๐๐3 ๐ด๐๐2 ๐ด๐๐1 ๐ดโ๐ ๐๐ + ๐ โ
๐4 =1
๐ง๐2 (๐ถ๐๐2 ๐ด๐๐1 ๐ดโ๐ ๐๐ +
๐2 =1
โ
(
๐2
๐2 =1
โโโ โ [ ๐ [ ] ] โ โ ๐ดโ๐2 ๐ดโ๐1 [๐ดโ๐ ๐๐ ] (0) + ๐ง๐4 โ
. . . โ โ โ ๐1
โโโ ๐ง๐4 โ
โ
โ
โ
โ โ โ
๐4 =1
)๐ ๐ง ๐ ๐๐ ๐ด
๐ โ
โ๐ ๐๐ = ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 โ๐ ๐๐
๐=1
which we summarize as ๐๐ (๐ง) = ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 โ๐ ๐๐ By the reproducing-kernel property of
๐พ๐๐ฟ (โ
, ๐)
for ๐๐ โ โ(๐พ๐๐ฟ ). for
โ(๐พ๐๐ฟ ), โ1
โจ๐๐ , ๐พ๐๐ฟ (โ
, ๐)๐ฆโฉโ(๐พ๐๐ฟ ) = โจ๐๐ (๐), ๐ฆโฉ๐ด = โจ๐ถ(๐ผ โ ๐(๐)๐ด) =
โจ๐๐ , โ๐โ (๐ผ
(3.32)
we also know that โ๐ ๐๐ , ๐ฆโฉ๐ด
โ
โ ๐ด ๐(๐)โ )โ1 ๐ถ โ ๐ฆโฉโ(๐พ๐๐ฟ )
from which we conclude that ๐พ๐๐ฟ (โ
, ๐)๐ฆ = โ๐โ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐ฆ.
(3.33)
Combining (3.33) with the general principle (3.32) applied to the case where ๐๐ = ๐พ๐ (โ
, ๐)๐ฆ then gives us ๐พ๐๐ฟ (๐ง, ๐) = ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 โ๐ โ๐โ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐ฆ and (3.31) follows as wanted.
(3.34)
96
J.A. Ball and V. Bolotnikov
๐ด ๐ต ] is coisometric by construction, we see Since the colligation matrix U = [ ๐ถ ๐ท from the identity (2.5) that ( ๐ ) โ โ โ1 ๐ผ โ ๐(๐ง)๐(๐) = ๐ถ(๐ผ โ ๐(๐ง)๐ด) (1 โ ๐ง๐ ๐ ๐ )๐๐ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐=1
=
๐ โ
(1 โ ๐ง๐ ๐ ๐ )๐พ๐๐ฟ (๐ง, ๐) (by (3.34))
๐=1
and we see that we recover {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } as a left Agler decomposition for ๐. It remains to verify the ๏ฌnal assertion in the statement of the theorem. By ๐ด ๐ต ] constructed above satis๏ฌes construction we see that the colligation matrix [ ๐ถ ๐ท properties (1), (2), and (4) in De๏ฌnition 3.1 for a c.f.m. of ๐. As for property (3), observe that (๐(๐ง) โ ๐(0))๐ข = ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 ๐(๐ง)๐ต๐ข =
๐ โ
๐ง๐ ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 ๐๐ ๐ต๐ข
๐=1
=
๐ โ
๐ง๐ [๐ต๐ข]๐ (๐ง)
๐=1 ๐ด ๐ต] where we used (3.32) for the last step. This completes the veri๏ฌcation that [ ๐ถ ๐ท is a c.f.m. for ๐. โก
Remark 3.8. A variant of Theorem 3.7 is Theorem 3.10 in [17] where it is assumed that {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } is a left Agler decomposition for a known ๐ in ๐ฎ๐๐ (๐ฐ, ๐ด) and then it is shown that any contractive solution ๐ด of the Gleason problem (3.1) can be embedded into a c.f.m. for ๐. A very similar argument as in the proof of Theorem 3.7 also occurs in the proofs of Theorem 2.2 and Theorem 3.1 in [20] where closely related results are proved but for contractive multipliers on the Drury-Arveson space rather than Schur-Agler-class functions on the polydisk. The univariate (๐ = 1) special case of Theorem 3.7 amounts essentially to Theorem 11 in [40] and can be viewed as a version of the Beurling-Lax theorem for backward-shift-invariant subspaces. We next o๏ฌer a second characterization of left Agler decompositions which may be easier to apply in some cases. Theorem 3.9. Suppose that we are given a collection {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } of โ(๐ด)-valued positive kernels. Then {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } is a left Agler decomposition for some SchurAgler-class function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) (for some appropriate input space ๐ฐ) if and โ๐ only if the kernel ๐ผ โ ๐=1 (1 โ ๐ง๐ ๐ ๐ )๐พ๐๐ฟ (๐ง, ๐) is a positive kernel.
Canonical Realization
97
Proof. If {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } is a left Agler decomposition for a Schur-Agler-class function ๐, it follows immediately from the de๏ฌning property (1.9) that ๐ผโ
๐ โ
(1 โ ๐ง๐ ๐ ๐ )๐พ๐๐ฟ (๐ง, ๐) = ๐(๐ง)๐(๐)โ
๐=1
is a positive kernel with Kolmogorov decomposition given by ๐(๐ง)๐(๐)โ . Conโ๐ versely, if ๐ผ โ ๐=1 (1 โ ๐ง๐ ๐ ๐ )๐พ๐๐ฟ (๐ง, ๐) is a positive kernel, it has a Kolmogorov decomposition ๐(๐ง)๐(๐)โ and then ๐ is a Schur-Agler-class function having {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } as a left Agler decomposition. โก 3.2. Weakly coisometric versus coisometric c.f.m.โs From the de๏ฌnitions we see that in case the subspace ๐ given by (3.6) is equal to โ๐ the whole space ๐=1 โ(๐พ๐๐ฟ ), then the weakly coisometric c.f.m. determined by the Agler decomposition {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } is unique and is automatically coisometric. In the univariate case (๐ = 1), we see ๐พ1๐ฟ = ๐พ ๐ฟ and the domain ๐ collapses to โ โ ๐= ๐๐พ ๐ฟ (โ
, ๐)๐ฆ = ๐พ ๐ฟ (โ
, ๐)๐ฆ = โ(๐พ ๐ฟ ) ๐โ๐ป ๐ฆโ๐ด
๐โ๐ป, ๐ฆโ๐ด
from which we see that weakly coisometric and coisometric c.f.m.โs coincide in the univariate case. As illustrated in Example 3.6, in the multivariate case it can happen that the containment in (3.6) holds with equality or that it is strict. In general not much is known about the actual construction and structure of Agler decompositions beyond ad hoc constructions as in Example 3.6; in particular, we do not know if, given a Schur-Agler-class function ๐, there exists a left Agler decomposition {๐พ1๐ฟ , . . . , ๐พ๐๐ฟ } which gives rise to a coisometric c.f.m. for ๐. We return to this topic in the context of two-component canonical functional models (the polydisk analog of Theorem 1.6) in Remark 5.12 below.
4. Weakly isometric realizations Using the strategy described in Remark 1.5, all the results concerning weakly isometric colligations/realizations associated with a ๏ฌxed right Agler decomposition (1.10) of a function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) can be obtained from their โcoisometricโ counterparts. Indeed, it follows from Theorem 1.7 that ๐ belongs to the class ๐ฎ๐๐ (๐ฐ, ๐ด) if and only if the associated function ๐ โฏ (๐ง) := ๐(ยฏ ๐ง )โ belongs to ๐ฎ๐๐ (๐ด, ๐ฐ) (we use the standard notation ๐งยฏ = (ยฏ ๐ง1 , . . . , ๐งยฏ๐ ) for ๐ง = (๐ง1 , . . . , ๐ง๐ ) โ โ๐ ). It is also clear from Theorem 1.7 that a right decomposition {๐พ1๐
, . . . , ๐พ๐๐
} for ๐ is at the same time a left decomposition for ๐ โฏ . Furthermore, ๐ is the transfer function of the colligation U of the form (1.13) if and only if ๐ โฏ is the transfer function of Uโ
98
J.A. Ball and V. Bolotnikov
which is readily seen upon taking adjoints in (1.14): ๐ โฏ (๐ง) = ๐(ยฏ ๐ง )โ = ๐ทโ + ๐ต โ ๐(ยฏ ๐ง )โ (๐ผ โ ๐ดโ ๐(ยฏ ๐ง )โ )โ1 ๐ถ โ = ๐ทโ + ๐ต โ ๐(๐ง)(๐ผ โ ๐ดโ ๐(๐ง))โ1 ๐ถ โ = ๐ทโ + ๐ต โ (๐ผ โ ๐(๐ง)๐ดโ )โ1 ๐(๐ง)๐ถ โ . Assume that we are given a function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) with a ๏ฌxed right Agler decomposition {๐พ1๐
, . . . , ๐พ๐๐
}. Let โก ๐
โค ๐พ1 (๐ง, ๐) โข โฅ .. ๐๐
(๐ง, ๐) := โฃ (4.1) โฆ : ๐ป๐ ร ๐ป๐ โ โ(๐ฐ, ๐ฐ ๐ ) . ๐พ๐๐
(๐ง, ๐)
and let
โก
โค ๐ 1 ๐พ1๐
(โ
, ๐)๐ข โข โฅ .. ห= ๐ ๐(๐)โ ๐๐
(โ
, ๐)๐ข = โฃ โฆ โ โ๐๐=1 โ(๐พ๐๐
). (4.2) . ๐ ๐ ๐โ๐ป , ๐ขโ๐ฐ ๐โ๐ป , ๐ขโ๐ฐ ๐ ๐ ๐พ๐๐
(โ
, ๐)๐ข โ
โ
De๏ฌnition 4.1. Given ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด), we shall say that the block-operator matrix [ ] [ ] [ ๐ ] ห ๐ต ห ๐ด โ๐๐=1 โ(๐พ๐๐
) โ๐=1 โ(๐พ๐๐
) ห = U : โ (4.3) ห ๐ท ห ๐ฐ ๐ด ๐ถ is a dual canonical functional-model (abbreviated to d.c.f.m. in what follows) colligation associated with right Agler decomposition (1.10) for ๐ if ห is contractive. 1. U ห and ๐ถ ห to the subspace ๐ ห โ โ๐ โ(๐พ ๐
) 2. The restrictions of operators ๐ด ๐=1 ๐ de๏ฌned in (4.3) have the following action on special kernel functions: ห ห : ๐(๐)โ ๐๐
(โ
, ๐)๐ข โ ๐๐
(โ
, ๐)๐ข โ ๐๐
(โ
, 0)๐ข, ๐ดโฃ ๐
(4.4)
ห ห : ๐(๐) ๐ (โ
, ๐)๐ข โ ๐(๐)๐ข โ ๐(0)๐ข. ๐ถโฃ ๐
(4.5)
โ
๐
ห : ๐ฐ โ โ๐ โ(๐พ ๐
) and ๐ท ห : ๐ฐ โ ๐ด are given by 3. The operators ๐ต ๐=1 ๐ ห : ๐ข โ ๐๐
(โ
, 0)๐ข, ๐ต
ห : ๐ข โ ๐(0)๐ข. ๐ท
(4.6)
The formulas (4.4)โ(4.6) look very much the same as formulas (3.8)โ(3.10) and reproducing the arguments from the proof of Proposition 3.2 we arrive at the following. ห of the form Proposition 4.2. Given ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด), the block-operator matrix U (4.3) is a d.c.f.m. colligation associated with right Agler decomposition {๐พ1๐
, . . . , ห is contractive and ๐พ๐๐
} for ๐ if and only if U [ ] [ ] [ ๐
] ห ๐ต ห ๐ (โ
, ๐)๐ข ๐ด ๐(๐)โ ๐๐
(โ
, ๐)๐ข ห U= ห ห : โ . (4.7) ๐ข ๐(๐)๐ข ๐ถ ๐ท
Canonical Realization
99
On the other hand, as a consequence of identity (1.10) we get (as in the previous section) that the formula [ ] [ ] [ ๐
] หห ๐ต ห ๐ (โ
, ๐)๐ข ๐ดโฃ ๐(๐)โ ๐๐
(โ
, ๐)๐ข ๐ ห ๐ = ห โ (4.8) ห : ๐ข ๐(๐)๐ข ๐ถโฃ๐ห ๐ท extends by continuity to de๏ฌne the isometry ๐ห : ๐๐ห โ โ๐ห where [ ๐
] [ ๐ ] โ ๐ (โ
, ๐)๐ฆ โ๐=1 โ(๐พ๐๐
) ห โ ๐ฐ and โ ห = ๐๐ห = ๐ท โ . ๐ ๐ฐ ๐(๐)๐ข ๐ ๐โ๐ป , ๐ขโ๐ฐ
The operator ๐ห is completely determined by the kernels {๐พ1๐
, . . . , ๐พ๐๐
} and it follows from (4.7) that a block-operator matrix U of the form (4.3) is a d.c.f.m. colligation associated with {๐พ1๐
, . . . , ๐พ๐๐
} if and only if U is a contractive extension of ๐ to all of (โ๐๐=1 โ(๐พ๐๐
)) โ ๐ด. This observation proves the ๏ฌrst statement in the following theorem which is the multivariable analog of Theorem 1.4. Theorem 4.3. Let ๐ be a function in the Schur-Agler class ๐ฎ๐ด๐ (๐ฐ, ๐ด) with given right Agler decomposition {๐พ1๐
, . . . , ๐พ๐๐
}. Then ๐ด ๐ต ] associated with {๐พ ๐
, . . . , ๐พ ๐
}. 1. There exists a d.c.f.m. colligation U = [ ๐ถ 1 ๐ ๐ท 2. Every d.c.f.m. colligation U associated with {๐พ1๐
, . . . , ๐พ๐๐
} is weakly isometric and controllable and furthermore, ๐(๐ง) = ๐ท + ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 ๐(๐ง)๐ต. 3. Any controllable weakly isometric colligation Uโฒ of the form (2.1) with the transfer function equal ๐ is unitarily equivalent to some d.c.f.m. colligation ห for ๐. U The latter theorem is a consequence of Theorem 3.4 so the proof will be omitted as well as the formulation of the theorem parallel to Theorem 3.5.
5. Weakly unitary realizations In this section we study unitary realizations of an ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) associated with a ๏ฌxed Agler decomposition (1.12). Following the streamlines of Section 2, we let โ(๐พ๐ ) to be the reproducing kernel Hilbert spaces associated with the kernels ๐พ๐ from decomposition (1.12). For functions ๐ โ โ๐๐=1 โ(๐พ๐ ), we will use representations and notation โก โค ๐1 [ ] [ ] ๐ ๐ โ ๐ ๐ด โข โฅ โ ๐= ๐๐ := โฃ ... โฆ โ โ(๐พ๐ ) where ๐๐ = ๐,+ : ๐ป๐ โ . (5.1) ๐๐,โ ๐ฐ ๐=1 ๐=1 ๐๐ We furthermore introduce the kernel โก โค ๐พ1 (๐ง, ๐) โข โฅ .. ๐ ๐ ๐ ๐(๐ง, ๐) := โฃ โฆ : ๐ป ร ๐ป โ โ(๐ด โ ๐ฐ, (๐ด โ ๐ฐ) ) . ๐พ๐ (๐ง, ๐)
(5.2)
100
J.A. Ball and V. Bolotnikov
and the subspaces [ ] [ ] } โ{ 0 ๐ฆ โ ๐ ๐(๐) ๐(โ
, ๐) ๐= : ๐ โ ๐ป , ๐ฆ โ ๐ด, ๐ข โ ๐ฐ , ๐(โ
, ๐) ๐ข 0 and
[ ] [ ] } โ{ ๐ฆ 0 โ ๐ , ๐(๐) ๐(โ
, ๐) : ๐ โ ๐ป , ๐ฆ โ ๐ด, ๐ข โ ๐ฐ ๐(โ
, ๐) โ= 0 ๐ข
(5.3)
(5.4)
of โ๐๐=1 โ(๐พ๐ ) whose orthogonal complements can be described as { } ] โ ๐ [ ๐ ๐ ๐ โ โ โ ๐๐,+ โฅ ๐ = ๐= โ(๐พ๐ ) : ๐ง๐ ๐๐,+ (๐ง) โก 0 & ๐๐,โ (๐ง) โก 0 (5.5) โ ๐๐,โ ๐=1
and
{ โฅ
โ =
๐=
๐=1
] ๐ [ โ ๐๐,+ ๐=1
๐๐,โ
โ
๐ โ
๐=1
โ(๐พ๐ ) :
๐=1
๐=1
๐ โ
๐๐,+ (๐ง) โก 0 &
๐=1
๐ โ
} ๐ง๐ ๐๐,โ (๐ง) โก 0 ,
๐=1
respectively. By the reproducing kernel property, we have [ ]โช โฉ ๐ฆ = โจ๐๐,+ (๐), ๐ฆโฉ๐ด , ๐๐ , ๐พ๐ (โ
, ๐) 0 โ(๐พ ) ๐ โฉ [ ]โช 0 ๐๐ , ๐พ๐ (โ
, ๐) = โจ๐๐,โ (๐), ๐ขโฉ๐ฐ . ๐ข โ(๐พ )
(5.6) (5.7) (5.8)
๐
We de๏ฌne the coisometric map s : โ๐๐=1 โ(๐พ๐ ) โ โ(๐พ1 + โ
โ
โ
+ ๐พ๐ ) by formula s๐ = ๐1 + โ
โ
โ
+ ๐๐
where ๐ =
๐ โ ๐=1
๐๐ โ
๐ โ
โ(๐พ๐ )
and observe that in view of (5.2), (5.7) and (5.8), โฉ [ ]โช ๐ฆ = โจ(s๐ )+ (๐), ๐ฆโฉ๐ด , ๐, ๐(โ
, ๐) 0 โ๐ โ(๐พ๐ ) ๐=1 โฉ [ ]โช 0 ๐, ๐(โ
, ๐) = โจ(s๐ )โ (๐), ๐ขโฉ๐ฐ . ๐ข โ๐ โ(๐พ ) ๐=1
(5.9)
๐=1
(5.10) (5.11)
๐
De๏ฌnition 5.1. A contractive colligation ] [ ๐ [ ] [ ๐ ] โ๐=1 โ(๐พ๐ ) ๐ด ๐ต โ๐=1 โ(๐พ๐ ) : U= โ ๐ถ ๐ท ๐ฐ ๐ด
(5.12)
will be called a two-component canonical functional-model (abbreviated to t.c.f.m. in what follows) realization associated with a ๏ฌxed Agler decomposition (1.12) of a given ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) if
Canonical Realization
101
1. The state space operator ๐ด solves the structured Gleason problem (s๐ )+ (๐ง) โ (s๐ )+ (0) =
๐ โ
๐ง๐ [๐ด๐ ]๐,+ (๐ง),
(5.13)
๐=1
whereas the adjoint operator ๐ดโ solves the dual structured Gleason problem (s๐ )โ (๐ง) โ (s๐ )โ (0) = 2. The operators ๐ถ : are of the form
โ๐
๐=1
๐ โ ๐=1
โ(๐พ๐ ) โ ๐ด, ๐ต โ :
๐ถ : ๐ โ (s๐ )+ (0),
๐ง๐ [๐ดโ ๐ ]๐,โ (๐ง). โ๐
๐=1
๐ต โ : ๐ โ (s๐ )โ (0),
(5.14)
โ(๐พ๐ ) โ ๐ฐ and ๐ท : ๐ฐ โ ๐ด ๐ท : ๐ข โ ๐(0)๐ข.
(5.15)
Proposition 5.2. Relations (5.13), (5.14) and (5.15) are equivalent respectively to equalities [ ] [ ] [ ] ๐ฆ ๐ฆ ๐ฆ โ โ , (5.16) โ ๐(โ
, 0) = ๐(โ
, ๐) ๐ด ๐(๐) ๐(โ
, ๐) 0 0 0 [ ] [ ] [ ] 0 0 0 , (5.17) โ ๐(โ
, 0) = ๐(โ
, ๐) ๐ด๐(๐)โ ๐(โ
, ๐) ๐ข ๐ข ๐ข [ ] [ ] 0 ๐ฆ ๐ถ โ ๐ฆ = ๐(โ
, 0) , and ๐ทโ ๐ฆ = ๐(0)โ ๐ฆ , ๐ต๐ข = ๐(โ
, 0) (5.18) ๐ข 0 holding for every ๐ โ ๐ป๐ , ๐ฆ โ ๐ด and ๐ข โ ๐ฐ. Proof. It follows from (5.10) that โจ(s๐ )+ (๐ง) โ (s๐ )+ (0) ๐ฆโฉ๐ด =
[ ]โช โฉ [ ] ๐ฆ ๐ฆ โ ๐(โ
, 0) ๐, ๐(โ
, ๐ง) 0 โ๐ 0
๐=1 โ(๐พ๐ )
and on the other hand, due to the diagonal structure (1.15) of ๐(๐ง), โช โฉ ๐ โฉ [ ]โช โ ๐ฆ ๐ง๐ [๐ด๐ ]๐,+ (๐ง), ๐ฆ = ๐(๐ง)๐ด๐, ๐(โ
, ๐ง) 0 โ๐ โ(๐พ ) ๐ ๐=1 ๐=1 ๐ด โฉ [ ]โช ๐ฆ = ๐, ๐ดโ ๐(๐ง)โ ๐(โ
, ๐ง) . 0 โ๐ โ(๐พ ) ๐=1
๐
โ๐๐=1 โ(๐พ๐ )
and ๐ฆ โ ๐ด, the Since the two latter equalities hold for every ๐ โ equivalence (5.13) โ (5.16) follows. The equivalence (5.14)โ (5.17) follows from (5.11) in much the same way; the formula for ๐ถ โ in (5.18) follows from โฉ [ ]โช ๐ฆ โจ๐, ๐ถ โ ๐ฆโฉ = โจ๐ถ๐, ๐ฆโฉ = โจ(s๐ )+ (0), ๐ฆโฉ = ๐, ๐(โ
, 0) 0 and the formula for ๐ต is a consequence of a similar computation. The formula for ๐ทโ is self-evident. โก
102
J.A. Ball and V. Bolotnikov
Proposition 5.3. Let ๐ต, ๐ถ and ๐ท be the operators de๏ฌned in (5.15). Then ๐ถ๐ถ โ + ๐ท๐ทโ = ๐ผ๐ด
and
๐ต โ ๐ต + ๐ทโ ๐ท = ๐ผ๐ด .
(5.19)
Furthermore ๐ต โ has the following action on kernel elements of the subspace ๐ de๏ฌned in (5.3): [ ] ๐ฆ ๐ต โ : ๐(๐)โ ๐(โ
, ๐) โ ๐(๐)โ ๐ฆ โ ๐(0)โ ๐ฆ, 0 [ ] 0 ยฏ ๐ต โ : ๐(โ
, ๐) โ ๐ข โ ๐(0)โ ๐(๐)๐ข, ๐ข
(5.20) (5.21)
for all ๐ โ ๐ป๐ , ๐ฆ โ ๐ด and ๐ข โ ๐ฐ, where ๐ is de๏ฌned in (5.2). Proof. We ๏ฌrst observe that โช 1 [ ]12 โฉโ ๐ 1 1 ๐ฆ 1 โฅ๐ถ โ ๐ฆโฅ2 = 1 ๐พ๐๐ฟ (0, 0)๐ฆ, ๐ฆ = โจ(๐ผ โ ๐(0)๐(0)โ )๐ฆ, ๐ฆโฉ , 1๐(โ
, 0) 0 1 = ๐=1
โช 1 [ ]12 โฉโ ๐ 1 1 0 ๐
1 โฅ๐ต๐ขโฅ = 1 ๐พ๐ (0, 0)๐ข, ๐ข = โจ(๐ผ โ ๐(0)โ ๐(0))๐ข, ๐ขโฉ , 1๐(โ
, 0) ๐ข 1 = 2
๐=1
where the ๏ฌrst equalities follow from formulas (5.18) for ๐ต and ๐ถ โ , the second equalities follow by reproducing kernel formulas (5.10), (5.11) along with de๏ฌnitions (5.9), (5.2) and (1.11) of s, ๐ and ๐พ๐ , and ๏ฌnally, the third equalities follow from the decomposition formula (1.12) evaluated at ๐ง = ๐ = 0. Taking into account the formulas (5.15) and (5.18) for ๐ท and ๐ทโ , we then have equalities โฅ๐ถ โ ๐ฆโฅ2 = โฅ๐ฆโฅ2 โ โฅ๐(0)โ ๐ฆโฅ2 = โฅ๐ฆโฅ2 โ โฅ๐ทโ ๐ฆโฅ2 ,
(5.22)
โฅ๐ต๐ขโฅ2 = โฅ๐ขโฅ2 โ โฅ๐(0)๐ขโฅ2 = โฅ๐ขโฅ2 โ โฅ๐ท๐ขโฅ2 holding for all ๐ฆ โ ๐ด and ๐ข โ ๐ฐ which are equivalent to operator equalities (5.19). To verify (5.20) and (5.21) we proceed as follows. By de๏ฌnitions (5.9), (1.11), (1.15) and (5.2) of s, ๐พ๐ , ๐(๐ง) and ๐, [ ( [ ])] [ [ ]] ๐ ๐ โ โ ๐ฆ ๐ฆ s ๐(๐)โ ๐(โ
, ๐) = ๐ ๐ ๐พ๐ (โ
, ๐) = ๐ ๐ ๐พ๐๐
๐ฟ (โ
, ๐)๐ฆ, 0 0 โ โ ๐=1
๐=1
[ ( [ ])] [ ]] ๐ [ ๐ โ โ 0 0 s ๐(โ
, ๐) = ๐พ๐ (โ
, ๐) = ๐พ๐๐
(โ
, ๐)๐ข. ๐ข ๐ข โ โ ๐=1
๐=1
Canonical Realization
103
Combining the de๏ฌnition (5.15) of ๐ต โ with the two last formulas evaluated at zero gives ๐ต โ ๐(๐)โ ๐(โ
, ๐)
[ ] [ ( [ ])] ๐ โ ๐ฆ ๐ฆ = s ๐(๐)โ ๐(โ
, ๐) (0) = ๐ ๐ ๐พ๐๐
๐ฟ (0, ๐)๐ฆ, 0 0 โ
(5.23)
๐=1
[ ] [ ( [ ])] ๐ โ 0 0 ๐ต ๐(โ
, ๐) = s ๐(โ
, ๐) (0) = ๐พ๐๐
(0.๐)๐ข. ๐ข ๐ข โ โ
(5.24)
๐=1
Upon letting ๐ง = 0 in (1.12) and equating the block entries in the bottom row we see that ๐(๐)โ โ ๐(0)โ =
๐ โ
๐ ๐ ๐พ๐๐
๐ฟ (0, ๐),
๐ผ๐ฐ โ ๐(0)โ ๐(๐) =
๐=1
๐ โ
๐พ๐๐
(0, ๐)
(5.25)
๐=1
and combining the two latter equalities with (5.23) and (5.24) gives (5.20), (5.21). โก Formulas (5.20), (5.21) describing the action of the operator ๐ต โ on elementary kernels of ๐ were easily obtained from the general formula (5.15) for ๐ต โ . Although the operator ๐ดโ is not de๏ฌned in De๏ฌnition 5.1 on the whole state space โ๐๐=1 โ(๐พ๐ ), it turns out that its action on elementary kernels of ๐ is completely determined by conditions (5.13) and (5.14). One half of the job is handled by formula (5.16) (which is equivalent to (5.13)). Another half is covered in the next proposition. ๐ด ๐ต ] be a t.c.f.m. colligation associated with the Agler Proposition 5.4. Let U = [ ๐ถ ๐ท decomposition (1.12) of a given ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) and let ๐ be given by (5.2). Then [ ] [ ] [ ] 0 0 ๐(๐)๐ข ๐ดโ : ๐(โ
, ๐) โ ๐(๐)โ ๐(โ
, ๐) โ ๐(โ
, 0) (5.26) ๐ข ๐ข 0
for all ๐ โ ๐ป๐ , ๐ฆ โ ๐ด and ๐ข โ ๐ฐ. Proof. We have to show that formula (5.26) follows from conditions in De๏ฌnition 5.1. To this end, we ๏ฌrst verify the equality 2
2
2
โฅ๐(๐)โ โ๐,๐ข โฅ โ โฅ๐ด๐(๐)โ โ๐,๐ข โฅ = โฅ๐ถ๐(๐)โ โ๐,๐ข โฅ๐ฐ
(5.27)
where the norms on the left-hand side are taken in โ๐๐=1 โ(๐พ๐ ) and where we have set for short [ ] 0 โ๐,๐ข := ๐(โ
, ๐) โ โ๐๐=1 โ(๐พ๐ ). (5.28) ๐ข
104
J.A. Ball and V. Bolotnikov
By the reproducing kernel property (5.11) and on account of (5.2) and (1.11), โจโ๐,๐ข , โ๐ง,๐ข โฉโ๐
๐=1 โ(๐พ๐ )
2
โฅ๐(๐)โ โ๐,๐ข โฅโ๐
๐=1 โ(๐พ๐ )
=
๐ โ
โจ๐พ๐๐
(๐ง, ๐)๐ข, ๐ขโฉ๐ฐ ,
(5.29)
โฃ๐๐ โฃ2 โ
โจ๐พ๐๐
(๐, ๐)๐ข, ๐ขโฉ๐ฐ .
(5.30)
๐=1
=
๐ โ ๐=1
Equality (5.17) holds by Proposition 5.2 and can be written as ๐ด๐(๐)โ โ๐,๐ข = โ๐,๐ข โ โ0,๐ข
(5.31)
in notation (5.28). This formula together with (5.29) leads us to 2
โฅ๐ด๐(๐)โ โ๐,๐ข โฅโ๐ =
๐ โ โฉ( ๐=1
๐=1 โ(๐พ๐ )
2
= โฅโ๐,๐ข โ โ0,๐ข โฅโ๐
๐=1 โ(๐พ๐ )
) โช ๐พ๐๐
(๐, ๐) โ ๐พ๐๐
(๐, 0) โ ๐พ๐๐
(0, ๐) + ๐พ๐๐
(0, 0) ๐ข, ๐ข ๐ฐ .
(5.32)
Upon letting ๐ง = ๐ in (1.12) we get the identity ๐ผ๐ฐ โ ๐(๐)โ ๐(๐) =
๐ โ
(1 โ โฃ๐๐ โฃ2 )๐พ๐๐
(๐, ๐)
(5.33)
๐=1
which together with the second relation in (5.25) implies ๐ โ ( ) (1 โ โฃ๐๐ โฃ2 โฃ)๐พ๐๐
(๐, ๐) โ ๐พ๐๐
(๐, 0) โ ๐พ๐๐
(0, ๐) + ๐พ๐๐
(0, 0) ๐=1
= ๐ผ๐ฐ โ ๐(๐)โ ๐(๐) โ (๐ผ๐ฐ โ ๐(๐)โ ๐(0)) โ (๐ผ๐ฐ โ ๐(0)โ ๐(๐)) + ๐ผ๐ฐ โ ๐(0)โ ๐(0) = โ(๐(๐)โ โ ๐(0)โ )(๐(๐) โ ๐(0)). Subtracting (5.32) from (5.30) and making use of the last identity gives us 1 12 2 2 โฅ๐(๐)โ โ๐,๐ข โฅ โ โฅ๐ด๐(๐)โ โ๐,๐ข โฅ = 1๐(๐)๐ข โ ๐(0)๐ข1๐ด . (5.34) On the other hand, it follows from the identity ๐(๐) โ ๐(0) =
๐ โ
๐ ๐ ๐พ๐๐ฟ๐
(0, ๐)
๐=1
(which is yet another consequence of the decomposition formula (1.12)), the explicit formula (5.15) for ๐ถ and de๏ฌnitions (5.9), (5.2) and (1.11), that [ [ ]] ๐ โ 0 ๐ถ๐(๐)โ โ๐,๐ข = [s (๐(๐)โ โ๐,๐ข )]+ (0) = ๐ ๐ ๐พ๐ (โ
, ๐)) (0) ๐ข + ๐=1
=
๐ โ ๐=1
๐ ๐ ๐พ๐๐ฟ๐
(0, ๐)๐ข = ๐(๐)๐ข โ ๐(0)๐ข. (5.35)
Canonical Realization
105
Substituting the latter equality into (5.34) completes the proof of (5.27). Writing (5.27) in the form โจ(๐ผ โ ๐ดโ ๐ด โ ๐ถ โ ๐ถ)๐(๐)โ โ๐,๐ข , ๐(๐)โ โ๐,๐ข โฉโ๐
๐=1 โ(๐พ๐ )
=0
and observing that the operator ๐ผ โ ๐ดโ ๐ด โ ๐ถ โ ๐ถ is positive semide๏ฌnite (since U is contractive by De๏ฌnition 5.1), we conclude that (๐ผ โ ๐ดโ ๐ด โ ๐ถ โ ๐ถ)๐(๐)โ โ๐,๐ข โก 0
for all ๐ โ ๐ป๐ , ๐ข โ ๐ฐ.
Applying the operator ๐ถ โ to both parts of (5.35) we get [ ] ๐(๐)๐ข โ ๐(0)๐ข ๐ถ โ ๐ถ๐(๐)โ โ๐,๐ข = ๐(โ
, 0) , 0
(5.36)
(5.37)
by the explicit formula (5.18) for ๐ถ โ . From the same formula and the formula (5.15) for ๐ท we get [ ] ๐(0)๐ข โ โ โ ๐ถ ๐ท๐ข = ๐ถ ๐(0) ๐ข = ๐(โ
, 0) . (5.38) 0 We next apply the operator ๐ดโ to both parts of equality (5.31): ๐ดโ ๐ด๐(๐)โ โ๐,๐ข = ๐ดโ โ๐,๐ข โ ๐ดโ โ0,๐ข .
(5.39)
Comparing (5.28) and the second formula in (5.18) (which holds by Proposition 5.2) convinces us that โ0,๐ข = ๐ต๐ข (5.40) so that (5.39) can be written as ๐ดโ โ๐,๐ข = ๐ดโ ๐ด๐(๐)โ โ๐,๐ข + ๐ดโ ๐ต๐ข.
(5.41)
Since U is contractive (by De๏ฌnition 5.1) and since ๐ต and ๐ท satisfy the second equality in (5.19), it then follows that ๐ดโ ๐ต + ๐ถ โ ๐ท = 0. Thus, [ ] ๐(0)๐ข ๐ดโ ๐ต๐ข = โ๐ถ โ ๐ท๐ข = โ๐ถ โ ๐(0)โ ๐ข = โ๐(โ
, 0) . 0 Taking the latter equality into account and making subsequent use of (5.36), (5.37) and (5.38) we then get from (5.41) ๐ดโ โ๐,๐ข = (๐ผ โ ๐ถ โ ๐ถ)๐(๐)โ โ๐,๐ข โ ๐ถ โ ๐ท๐ข [ ] [ ] ๐(0)๐ข ๐(๐)๐ข โ ๐(0)๐ข โ = ๐(๐) โ๐,๐ข โ ๐(โ
, 0) โ ๐(โ
, 0) 0 0 [ ] ๐(๐)๐ข = ๐(๐)โ โ๐,๐ข โ ๐(โ
, 0) . 0 Substituting (5.28) into the last identity we get (5.26) which completes the proof. โก
106
J.A. Ball and V. Bolotnikov
Remark 5.5. Since any t.c.f.m. colligation is contractive, we have in particular that ๐ด๐ดโ + ๐ต๐ต โ โค ๐ผ. Therefore, formulas (5.20), (5.21) and (5.26), (5.16) de๏ฌning the action of operators ๐ต โ and ๐ดโ on elementary kernels of the space ๐ (see (5.3)) can be extended by continuity to de๏ฌne these operators on the whole ๐. ๐ด ๐ต ] associated with a ๏ฌxed Agler Proposition 5.6. Any t.c.f.m. colligation U = [ ๐ถ ๐ท decomposition (1.12) of a given ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) is weakly unitary and closely connected. Furthermore,
๐(๐ง) = ๐ท + ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 ๐(๐ง)๐ต.
(5.42)
๐ด ๐ต ] be a t.c.f.m. colligation of ๐ associated with a ๏ฌxed Agler Proof. Let U = [ ๐ถ ๐ท decomposition (1.12). Then equalities (5.16)โ(5.18) and (5.26) hold (by Propositions 5.2 and 5.4) and can be solved for ๐(โ
, ๐) as follows: [ ] [ ] ๐ฆ ๐ฆ ๐(โ
, ๐) = (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐(โ
, 0) = (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐ฆ, (5.43) 0 0 [ ] [ ] 0 0 ยฏ โ1 ๐ต๐ข. ๐(โ
, ๐) = (๐ผ โ ๐ด๐(๐)โ )โ1 ๐(โ
, 0) = (๐ผ โ ๐ด๐(๐)) (5.44) ๐ข ๐ข
From (5.43) and (5.20) we conclude that equalities (๐ทโ + ๐ต โ ๐(๐)โ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ )๐ฆ = ๐(0)โ ๐ฆ + ๐ต โ ๐(๐ง)โ ๐(โ
, ๐)
[ ] ๐ฆ 0
= ๐(0)โ ๐ฆ + ๐(๐)โ ๐ฆ โ ๐(0)โ ๐ฆ = ๐(๐)โ ๐ฆ
(5.45)
hold for every ๐ โ ๐ป๐ and ๐ฆ โ ๐ด, which proves representation (5.42). Furthermore, in view of (5.2), โ{ } ๐โ(๐พ๐ ) (๐ผ โ ๐ดโ ๐(๐)โ ๐ถ โ ๐ฆ, ๐โ(๐พ๐ ) (๐ผ โ ๐ด๐(๐)๐ต๐ข : ๐ โ ๐ป๐ , ๐ฆ โ ๐ด, ๐ข โ ๐ฐ [ ] [ ] } โ{ ๐ฆ 0 = ๐โ(๐พ๐ ) ๐(โ
, ๐) , ๐โ(๐พ๐ ) ๐(โ
, ๐) : ๐ โ ๐ป๐ , ๐ฆ โ ๐ด, ๐ข โ ๐ฐ 0 ๐ข [ ] [ ] } โ{ ๐ฆ 0 = ๐พ๐ (โ
, ๐) , ๐พ๐ (โ
, ๐) : ๐ โ ๐ป๐ , ๐ฆ โ ๐ด, ๐ข โ ๐ฐ 0 ๐ข [ ] [ ] } โ{ ๐ฆ ๐ฆ = ๐พ๐ (โ
, ๐) : ๐ โ ๐ป๐ , โ ๐ด โ ๐ฐ = โ(๐พ๐ ) ๐ข ๐ข ๐ต and the colligation U = [ ๐ด ๐ถ ๐ท ] is closely connected by De๏ฌnition 2.1. To show that U is weakly unitary, let us rearrange the Agler decomposition (1.12) for ๐ as [ ] ] [ ] ๐ [ ] โ ๐ผ๐ด [ ๐ง๐ ๐ผ๐ด 0 ๐ยฏ๐ ๐ผ๐ด 0 ยฏ ๐ผ ๐( ๐) + ๐พ (๐ง, ๐) ๐ด ๐ ๐(ยฏ ๐ง )โ 0 ๐ผ๐ฐ 0 ๐ผ๐ฐ
[ ] ๐(๐ง) [ ๐(๐)โ = ๐ผ๐ฐ
๐=1
]
๐ผ๐ฐ +
๐ [ โ ๐ผ๐ด ๐=1
0
] [ 0 ๐ผ ๐พ๐ (๐ง, ๐) ๐ด ๐ง๐ ๐ผ๐ฐ 0
] 0 , ๐ยฏ๐ ๐ผ๐ฐ
Canonical Realization
107
which in turn can be written in the inner product form ยฏ โจ๐ฆ + ๐(๐)๐ข, ๐ฆ โฒ + ๐(ยฏ ๐ง )๐ขโฒ โฉ๐ด +
๐ โฉ โ
๐พ๐ (โ
, ๐)
[ยฏ ] ๐๐ ๐ฆ ๐ข
๐=1
= โจ๐(๐)โ ๐ฆ + ๐ข, ๐(๐ง)โ ๐ฆ โฒ + ๐ขโฒ โฉ๐ฐ +
๐ โฉ โ
, ๐พ๐ (โ
, ๐ง)
๐พ๐ (โ
, ๐)
๐=1
[
๐ฆ ] ๐ยฏ๐ ๐ข ,
[
๐ยฏ๐ ๐ฆ โฒ ๐ขโฒ
]โช
๐พ๐ (โ
, ๐ง)
โ(๐พ๐ )
[
๐ฆโฒ ๐ยฏ๐ ๐ขโฒ
]โช โ(๐พ๐ )
which is the same as [ ] ]โช โฉ[ ] [ [ โฒ] ๐(๐)โ ๐(โ
, ๐) [ ๐ฆ0 ] + ๐(โ
, ๐) [ ๐ข0 ] ๐(๐)โ ๐(โ
, ๐ง) ๐ฆ0 + ๐(โ
, ๐ง) ๐ข0โฒ , ยฏ ยฏ โฒ ๐ฆ + ๐(๐)๐ข ๐ฆ โฒ + ๐(๐)๐ข โฉ[ ] [ [ ]โช โฒ ] ๐ฆ ๐(โ
, ๐) [ 0 ] + ๐(๐)โ ๐(โ
, ๐) [ ๐ข0 ] ๐(โ
, ๐ง) ๐ฆ0 + ๐(๐)โ ๐(โ
, ๐) [ ๐ข0 ] = , ๐(๐)โ ๐ฆ + ๐ข ๐(๐ง)โ ๐ฆ โฒ + ๐ขโฒ (5.46) where the inner products are taken in (โ๐๐=1 โ(๐พ๐ )) โ ๐ด and (โ๐๐=1 โ(๐พ๐ )) โ ๐ฐ. Letting ๐ข = ๐ขโฒ = 0 and ๐ฆ = ๐ฆ โฒ in the latter equality gives 1[ ]1 1[ ]1 1 ๐(๐)โ ๐(โ
, ๐) [ ๐ฆ ] 1 1 ๐(โ
, ๐) [ ๐ฆ ] 1 0 0 1 1=1 1 1 1 1 1 ๐ฆ ๐(๐)โ ๐ฆ which on account of (5.43) can be written as 1[ ]1 1[ ]1 1 ๐(๐)โ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐ฆ 1 1 (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐ฆ 1 1 1=1 1. 1 1 1 1 ๐ฆ ๐(๐)โ ๐ฆ Since
[
๐ดโ ๐ตโ
๐ถโ ๐ทโ
][
๐(๐)โ (๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐ฆ ๐ฆ
]
[ =
(๐ผ โ ๐ดโ ๐(๐)โ )โ1 ๐ถ โ ๐ฆ ๐(๐)โ ๐ฆ
(5.47) ]
(the top components in the latter formula are equal automatically whereas the bottom components are equal due to (5.45)), equality (5.47) tells us that U is weakly coisometric by De๏ฌnition 2.2. Similarly letting ๐ข = ๐ขโฒ and ๐ฆ = ๐ฆ โฒ = 0 in (5.46) we get 1[ ]1 1[ ]1 1 ๐(โ
, ๐) [ 0 ] 1 1 ๐(๐)โ ๐(โ
, ๐) [ 0 ] 1 ๐ข ๐ข 1 1=1 1 ยฏ 1 1 1 1 ๐(๐)๐ข ๐ข which in view of (5.44) can be written as 1[ ]1 1[ ]1 ยฏ โ1 ๐ต๐ข 1 1 ๐(๐)(๐ผ ยฏ โ ๐ด๐(๐)) ยฏ โ1 ๐ต๐ข 1 1 (๐ผ โ ๐ด๐(๐)) 1 1=1 1 ยฏ 1 1 1 1 ๐(๐)๐ข ๐ข and since
[
๐ด ๐ถ
๐ต ๐ท
][
ยฏ โ ๐ด๐(๐)) ยฏ โ1 ๐ต๐ข ๐(๐)(๐ผ ๐ข
]
[ =
ยฏ โ1 ๐ต๐ข (๐ผ โ ๐ด๐(๐)) ยฏ ๐(๐)๐ข
]
(again, the top components are equal automatically and the bottom components are equal due to (5.42)), the colligation U is weakly isometric by De๏ฌnition 2.2. โก
108
J.A. Ball and V. Bolotnikov
Proposition 5.6 establishes common features of t.c.f.m. colligations leaving the question about the existence of at least one such colligation open. As was shown in the proof of Proposition 5.6, the Agler decomposition (1.12) can be written in the inner product form (5.46) from which we conclude that the linear map [ ] ] [ ๐ด๐ ๐ต๐ ๐(๐)โ ๐(โ
, ๐) [ ๐ฆ0 ] + ๐(โ
, ๐) [ ๐ข0 ] ๐ = : ยฏ ๐ฆ + ๐(๐)๐ข ๐ถ๐ ๐ท๐ [ ] ๐(โ
, ๐) [ ๐ฆ0 ] + ๐(๐)โ ๐(โ
, ๐) [ ๐ข0 ] โ (5.48) ๐(๐)โ ๐ฆ + ๐ข de๏ฌned completely in terms of a given Agler decomposition {๐พ1 , . . . , ๐พ๐ } of ๐, extends to the isometry from } โ {[ ๐(๐)โ ๐(โ
, ๐) [ ๐ฆ ] ] [ ๐(โ
, ๐) [ 0 ] ] ๐ 0 ๐ข ๐๐ = , : ๐ โ ๐ป , ๐ฆ โ ๐ด, ๐ข โ ๐ฐ ยฏ ๐ฆ ๐(๐)๐ข onto
} โ {[ ๐(โ
, ๐) [ ๐ฆ ] ] [ ๐(๐)โ ๐(โ
, ๐) [ 0 ] ] ๐ 0 ๐ข : ๐ โ ๐ป , ๐ฆ โ ๐ด, ๐ข โ ๐ฐ . , โ๐ = ๐(๐)โ ๐ฆ ๐ข
It is readily seen that ๐๐ and โ๐ contain respectively all vectors of the form [ ๐ฆ0 ] and [ ๐ข0 ] and therefore they are split into direct sums ๐๐ = ๐ โ ๐ด where the subspaces ๐ and โ of operators ๐ด๐ : ๐ โ โ,
and โ๐ = โ โ ๐ฐ
โ๐๐=1 โ(๐พ๐ )
๐ต๐ : ๐ฐ โ โ,
are de๏ฌned in (5.3), (5.4). For the
๐ถ๐ : ๐ โ ๐ด,
๐ท๐ : ๐ฐ โ ๐ด
we have from (5.48) the following relations: [ ] [ ] ๐ฆ ๐ฆ โ + ๐ต๐ ๐ฆ = ๐(โ
, ๐) , ๐ด๐ ๐(๐) ๐(โ
, ๐) 0 0 [ ] [ ] 0 0 โ ยฏ ๐ด๐ ๐(โ
, ๐) + ๐ต๐ ๐(๐)๐ข = ๐(๐) ๐(โ
, ๐) , ๐ข ๐ข [ ] ๐ฆ + ๐ท๐ ๐ฆ = ๐(๐)โ ๐ฆ, ๐ถ๐ ๐(๐)โ ๐(โ
, ๐) 0 [ ] 0 ยฏ = ๐ข. ๐ถ๐ ๐(โ
, ๐) + ๐ท๐ ๐(๐)๐ข ๐ข
(5.49) (5.50) (5.51) (5.52)
Equalities (5.49) and (5.50) are obtained upon comparing the top components in (5.48) with respectively, ๐ข = 0 and ๐ฆ = 0. Equalities (5.51) and (5.52) are obtained similarly upon comparing the bottom components in (5.48). Letting ๐ = 0 in (5.49) and (5.51) gives [ ] ๐ฆ and ๐ท๐ ๐ฆ = ๐(0)โ ๐ฆ. (5.53) ๐ต๐ ๐ฆ = ๐(โ
, 0) 0
Canonical Realization
109
Substituting the ๏ฌrst and the second formula in (5.53) respectively into (5.49), (5.50) and into (5.51) and (5.52) results in equalities [ ] [ ] [ ] ๐ฆ ๐ฆ ๐ฆ ๐ด๐ : ๐(๐)โ ๐(โ
, ๐) = ๐(โ
, ๐) โ ๐(โ
, 0) , (5.54) 0 0 0 [ ] [ ] [ ] 0 0 ๐(๐)๐ข ๐ด๐ : ๐(โ
, ๐) โ ๐(๐)โ ๐(โ
, ๐) โ ๐(โ
, 0) , (5.55) ๐ข ๐ข 0 [ ] ๐ฆ ๐ถ๐ : ๐(๐)โ ๐(โ
, ๐) โ ๐(๐)โ ๐ฆ โ ๐(0)โ ๐ฆ, (5.56) 0 [ ] 0 ยฏ ๐ถ๐ : ๐(โ
, ๐) โ ๐ข โ ๐(0)โ ๐(๐)๐ข (5.57) ๐ข holding for all ๐ โ ๐ป๐ , ๐ข โ ๐ฐ and ๐ฆ โ ๐ด and completely de๏ฌning the operators ๐ด๐ and ๐ถ๐ on the whole space ๐. Lemma 5.7. Given the Agler decomposition {๐พ1 , . . . , ๐พ๐ } for a function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด), let ๐ be the isometric operator associated with this decomposition as ๐ด ๐ต ] of the form (5.12) is a t.c.f.m. in (5.48). A block-operator matrix U = [ ๐ถ ๐ท colligation associated with {๐พ1 , . . . , ๐พ๐ } if and only if โฅUโ โฅ โค 1,
Uโ โฃ๐โ๐ด = ๐
๐ต โ โฃ๐ โฅ = 0,
(5.58)
that is, U is a contractive extension of ๐ from ๐ โ ๐ด to all of subject to condition ๐ต โ โฃ๐ โฅ = 0.
(โ๐๐=1 โ(๐พ๐ )) โ ๐ด
and
โ
๐ด ๐ต ] be a t.c.f.m. colligation associated with {๐พ , . . . , ๐พ }. Proof. Let U = [ ๐ถ 1 ๐ ๐ท Then U is contractive by de๏ฌnition and relations (5.16) and (5.18)โ(5.26) hold by Propositions 5.2 and 5.4. Comparing (5.16) and (5.26) with (5.54), (5.55) we see that ๐ดโ โฃ๐ = ๐ด๐ . Comparing (5.20), (5.21) with (5.56), (5.57) we conclude that ๐ต โ โฃ๐ = ๐ถ๐ . Also, it follows from (5.18) and (5.53) that ๐ถ โ = ๐ต๐ and ๐ทโ = ๐ท๐ . Finally, it is seen from formula (5.5) that for every ๐ = โ๐๐=1 ๐๐ โ ๐โฅ ,
(s๐ )โ (๐ง) =
๐ โ
๐๐,โ (๐ง) โก 0
๐=1
so that in particular, ๐ต โ ๐ = (s๐ )โ (0) = 0, which proves the last equality in (5.58). ๐ด ๐ต ] meets all the conConversely, let us assume that a colligation U = [ ๐ถ ๐ท ditions in (5.58). From the second relation in (5.58) we conclude the equalities (5.53)โ(5.57) hold with operators ๐ด๐ , ๐ต๐ , ๐ถ๐ and ๐ท๐ replaced by ๐ดโ , ๐ถ โ , ๐ต โ and ๐ทโ respectively. In other words, we conclude from (5.53) that ๐ถ โ and ๐ทโ are de๏ฌned exactly as in (5.18) which means (by Proposition (5.3)) that they are already of the requisite form. Equalities (5.56), (5.57) tell us that the operator ๐ต โ satis๏ฌes formulas (5.20), (5.21). As we have seen in the proof of Proposition 5.4, these formulas agree with the second formula in (5.15) de๏ฌning ๐ต โ on the whole โ๐๐=1 โ(๐พ๐ ). From the third condition in (3.14) we now conclude that ๐ต โ is de๏ฌned by formula (5.15) on the whole โ๐๐=1 โ(๐พ๐ ) and therefore, ๐ต is also of the requisite
110
J.A. Ball and V. Bolotnikov
form. The formula (5.54) (with ๐ดโ instead of ๐ด๐ ) leads us to (5.16) which means that ๐ด solves the Gleason problem (5.13). To complete the proof, it remains to show that ๐ดโ solves the dual Gleason problem (5.14) or equivalently, that (5.18) holds. Rather than (5.18), we have equality (5.50) (with ๐ดโ and ๐ถ โ instead of ๐ด๐ and ๐ต๐ respectively) which can be written in terms of notation (5.28) as ๐ดโ โ๐,๐ข = ๐(๐)โ โ๐,๐ข โ ๐ถ โ ๐(๐)๐ข
(5.59)
We use (5.59) to show that equality โฅโ๐,๐ข โฅ2โ๐
๐=1 โ(๐พ๐ )
โ โฅ๐ดโ โ๐,๐ข โฅ2โ๐
๐=1 โ(๐พ๐ )
= โฅ๐ต โ โ๐,๐ข โฅ2๐ฐ
(5.60)
holds for every ๐ โ ๐ป๐ and ๐ข โ ๐ฐ. Indeed, 1 1 2 2 2 ยฏ 12 โฅโ๐,๐ข โฅ โ โฅ๐ดโ โ๐,๐ข โฅ = โฅโ๐,๐ข โฅ โ 1๐(๐)โ โ๐,๐ข โ ๐ถ โ ๐(๐)๐ข โฉ โช 2 2 ยฏ = โฅโ๐,๐ข โฅ โ โฅ๐(๐)โ โ๐,๐ข โฅ + ๐ถ๐(๐)โ โ๐,๐ข , ๐(๐)๐ข 1 โฉ โช 1 ยฏ ๐ถ๐(๐)โ โ๐,๐ข โ 1๐ถ โ ๐(๐)๐ข ยฏ 12 . + ๐(๐)๐ข, (5.61) We next express all the terms on the right of (5.61) in terms of the function ๐: โฉ โช 2 2 โฅโ๐,๐ข โฅ โ โฅ๐(๐)โ โ๐,๐ข โฅ = (๐ผ๐ฐ โ ๐(๐)โ ๐(๐))๐ข, ๐ข , (5.62) โฉ โช โฉ โช โ โ ยฏ ยฏ (๐(๐) โ ๐(0))๐ข, ๐ข , ๐ถ๐(๐) โ๐,๐ข , ๐(๐)๐ข = ๐(๐) (5.63) โฉ โช โฉ( ) โช ยฏ ๐(๐)๐ข, ๐ถ๐(๐)โ โ๐,๐ข = ๐(๐)โ โ ๐(0)โ ๐(๐)๐ข, ๐ข , (5.64) 1 โ 12 2 โ 2 ยฏ 1 = โฅ๐(๐)๐ขโฅ โ โฅ๐(0) ๐(๐)๐ขโฅ ยฏ 1๐ถ ๐(๐)๐ข . (5.65) We mention that (5.62) follows from (5.29), (5.30) and (5.33); equality (5.62) is a consequence of (5.35). Taking adjoints in (5.63) gives (5.64) and equality (5.65) ยฏ in (5.22). We now substitute the four last is obtained upon letting ๐ฆ = ๐(๐)๐ข equalities into (5.61) to get โฅโ๐,๐ข โฅ2โ๐
๐=1 โ(๐พ๐ )
where
โ โฅ๐ดโ โ๐,๐ข โฅ2โ๐
๐=1 โ(๐พ๐ )
= โจ๐
(๐)๐ข, ๐ขโฉ๐ฐ
(5.66)
( ) ๐
(๐) = ๐ผ๐ฐ โ ๐(๐)โ ๐(๐) + ๐(๐)โ ๐(๐) โ ๐(0) ( ) + ๐(๐)โ โ ๐(0)โ ๐(๐) โ ๐(๐)โ ๐(๐) + ๐(๐)โ ๐(0)๐(0)โ ๐(๐) = ๐ผ๐ฐ โ ๐(๐)โ ๐(0) โ ๐(0)โ ๐(๐) + ๐(๐)โ ๐(0)๐(0)โ ๐(๐) ( )( ) = ๐ผ๐ฐ โ ๐(๐)โ ๐(0) ๐ผ๐ฐ โ ๐(0)โ ๐(๐) .
It is readily seen from (5.21) that ยฏ ๐ต โ โ๐,๐ข = ๐ข โ ๐(0)โ ๐(๐)๐ข and therefore
1 1 2 ยฏ 12 = โจ๐
(๐)๐ข, ๐ขโฉ , โฅ๐ต โ โ๐,๐ข โฅ๐ฐ = 1๐ข โ ๐(0)โ ๐(๐)๐ข ๐ฐ
(5.67)
Canonical Realization
111
which together with (5.66) completes the proof of (5.60). Writing (5.60) as โจ(๐ผ โ ๐ด๐ดโ โ ๐ต๐ต โ )โ๐,๐ข , โ๐,๐ข โฉ = 0 and observing that the operator ๐ผ โ ๐ด๐ดโ โ ๐ต๐ต โ is positive semide๏ฌnite (since ๐ด ๐ต ] is a contraction), we conclude that U = [๐ถ ๐ท (๐ผ โ ๐ด๐ดโ โ ๐ต๐ต โ )โ๐,๐ข = 0
for all ๐ โ ๐ป๐ , ๐ข โ ๐ฐ.
(5.68)
๐ด ๐ต] Since the operators ๐ต and ๐ท satisfy the ๏ฌrst equality (5.19) and since U = [ ๐ถ ๐ท โ โ is a contraction, we have ๐ด๐ถ + ๐ต๐ท = 0. We now combine this latter equality with (5.40), (5.67) and formula (5.18) for ๐ท โ to get ยฏ ยฏ โ0,๐ข = ๐ต๐ข = ๐ต(๐ต โ โ๐,๐ข + ๐(0)โ ๐(๐)๐ข) = ๐ต๐ต โ โ๐,๐ข + ๐ต๐ทโ ๐(๐)๐ข
ยฏ = ๐ต๐ต โ โ๐,๐ข โ ๐ด๐ถ โ ๐(๐)๐ข.
(5.69)
We now apply the operator ๐ด to both parts of (5.59): ยฏ ๐ด๐ดโ โ๐,๐ข = ๐ด๐(๐)โ โ๐,๐ข โ ๐ด๐ถ โ ๐(๐)๐ข and solve the obtained identity for ๐ด๐(๐)โ โ๐,๐ข with further simpli๏ฌcations based on (5.68) and (5.69): ยฏ = โ๐,๐ข โ ๐ต๐ต โ โ๐,๐ข โ ๐ต๐ทโ ๐(๐)๐ข ยฏ = โ๐,๐ข โ โ0,๐ข . ๐ด๐(๐)โ โ๐,๐ข = ๐ด๐ดโ โ๐,๐ข + ๐ด๐ถ โ ๐(๐)๐ข Substituting (5.28) into the latter equality we get (5.18) which completes the proof. โก As a consequence of Lemma 5.7 we get a description of all t.c.f.m. colligations associated with a given Agler decomposition of a Schur-Agler function. Lemma 5.8. Let {๐พ1 , . . . , ๐พ๐ } be a ๏ฌxed Agler decomposition of a function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด). Let ๐ be the associated isometry de๏ฌned in (5.48) with the defect spaces ๐โฅ and โโฅ de๏ฌned in (5.5), (5.6). Then all t.c.f.m. colligations associated with {๐พ1 , . . . , ๐พ๐ } are of the form [ โฅ ] ] [ โฅ ] [ ๐ 0 โ ๐ โ (5.70) : Uโ = ๐โ๐ด โโ๐ฐ 0 ๐ where we have identi๏ฌed [ ๐ ] โ๐=1 โ(๐พ๐ ) with ๐ด
[
๐โฅ ๐โ๐ด
]
[ and
โ๐๐=1 โ(๐พ๐ ) ๐ฐ
]
[ with
โโฅ โโ๐ฐ
]
and where ๐ is an arbitrary contraction from ๐โฅ into โโฅ . The colligation U is isometric (coisometric, unitary) if and only if ๐ is coisometric (isometric, unitary). For the proof, it is enough to recall that ๐ is unitary as an operator from ๐๐ = ๐ โ ๐ด onto โ๐ = โ โ ๐ฐ and then to refer to Lemma 5.7. The meaning of description (5.70) is clear: the operators ๐ต โ , ๐ถ โ , ๐ท โ and the restriction of ๐ดโ to the subspace ๐ in operator colligation Uโ are prescribed. The objective is to guarantee Uโ be contractive by suitable de๏ฌning ๐ดโ on ๐โฅ . Lemma 3.5 states that ๐ = ๐ดโ โฃ๐ โฅ must be a contraction with range contained in โโฅ .
112
J.A. Ball and V. Bolotnikov We now are ready to formulate the multivariable counterpart of Theorem 1.6.
Theorem 5.9. Let ๐ be a function in the Schur-Agler class ๐ฎ๐ด๐ (๐ฐ, ๐ด) with given Agler decomposition {๐พ1 , . . . , ๐พ๐ }. Then ๐ด ๐ต ] associated with {๐พ , . . . , ๐พ }. 1. There exists a t.c.f.m. colligation U = [ ๐ถ 1 ๐ ๐ท 2. Every t.c.f.m. colligation U associated with {๐พ1 , . . . , ๐พ๐ } is weakly unitary and closely connected and furthermore, ๐(๐ง) = ๐ท + ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 ๐(๐ง)๐ต. ห of the form (2.1) with the 3. Any weakly unitary closely connected colligation U transfer function equal ๐ is unitarily equivalent to some t.c.f.m. colligation U for ๐.
Proof. Part (1) is contained in Lemma 5.8.[ Part (2) proved in Proposition ] was ] [ ] [ ๐ ๐ ห ห ห ห ห = ๐ด ๐ต : โ๐=1 ๐ณ๐ โ โ๐=1 ๐ณ๐ be 5.6. To prove part (3) we assume that U ห ๐ท ๐ฐ ๐ด ๐ถ ๐ ห๐ and a closely connected weakly unitary colligation with the state space โ๐=1 ๐ ห Then ๐ admits Agler decomposition ห โ ๐(๐ง)๐ด) ห โ1 ๐(๐ง)๐ต. such that ๐(๐ง) = ๐ท + ๐ถ(๐ผ (1.12) with kernels ๐พ๐ de๏ฌned by: [ ๐ฟ ] ๐พ๐ (๐ง, ๐) ๐พ๐๐ฟ๐
(๐ง, ๐) ๐พ๐ (๐ง, ๐) = ๐พ๐๐
๐ฟ (๐ง, ๐) ๐พ๐๐
(๐ง, ๐) ] [ [ ] ห โ ๐(๐ง)๐ด) ห โ1 ๐ถ(๐ผ โ โ1 ห หโ ๐(๐)โ )โ1 ๐ถ ห หโ (๐ผ โ ๐ด๐(๐) ๐ = หโ (๐ผ โ ๐ด ) ๐ต ห ๐ โ โ1 ๐ ห ) ๐ต (๐ผ โ ๐(๐ง)๐ด for ๐ = 1, . . . , ๐. Let โ(๐พ๐ ) be the associated reproducing kernel Hilbert spaces and let โ๐ : ๐ณห๐ โ ๐ณห = โ๐๐=1 ๐ณห๐ be the inclusion maps โ๐ : ๐ฅ๐ โ 0 โ โ
โ
โ
โ 0 โ ๐ฅ๐ โ 0 โ โ
โ
โ
โ 0. Since the realization is closely connected, the operators ๐๐ : ๐ณห๐ โ โ(๐พ๐ ) given by [ ] ห โ ๐(๐ง)๐ด) ห โ1 ๐ถ(๐ผ (5.71) ๐๐ : ๐ฅ๐ โ ห โ หโ )โ1 โ๐ ๐ฅ๐ ๐ต (๐ผ โ ๐(๐ง)๐ด are unitary. Let us de๏ฌne ๐ด โ โ(โ๐๐=1 โ(๐พ๐ )) by ( ) ( ) ห ๐ด โ๐๐=1 ๐๐ = โ๐๐=1 ๐๐ ๐ด. In more detail: ๐ด = [๐ด๐๐ ]๐๐,๐=1 where [ [ ] ] ห โ ๐(๐ง)๐ด) ห โ1 ห โ ๐(๐ง)๐ด) ห โ1 ๐ถ(๐ผ ๐ถ(๐ผ ห ๐ด๐๐ : ห โ หโ )โ1 โ๐ ๐ฅ๐ โ ๐ต หโ )โ1 โ๐ ๐ด๐๐ ๐ฅ๐ . ห โ (๐ผ โ ๐(๐ง)๐ด ๐ต (๐ผ โ ๐(๐ง)๐ด Since the operators (5.71) are unitary, we have from (5.72) ( ) ( ) โ ห ๐ดโ โ๐๐=1 ๐๐ = โ๐๐=1 ๐๐ ๐ด
(5.72)
(5.73)
Canonical Realization
113
and therefore, [ [ ] ] ห โ ๐(๐ง)๐ด) ห โ1 ห โ ๐(๐ง)๐ด) ห โ1 ๐ถ(๐ผ ๐ถ(๐ผ โ หโ ๐ด๐๐ : ห โ หโ )โ1 โ๐ ๐ฅ๐ โ ๐ต หโ )โ1 โ๐ ๐ด๐๐ ๐ฅ๐ . ห โ (๐ผ โ ๐(๐ง)๐ด ๐ต (๐ผ โ ๐(๐ง)๐ด Take the generic element ๐ of โ๐๐=1 โ(๐พ๐ ) in the form [ ] ๐ ๐ โ โ ห โ ๐(๐ง)๐ด) ห โ1 ๐ถ(๐ผ ๐ฅ and let ๐ฅ := ๐ฅ๐ โ ๐ณห. โ ๐ (๐ง) = หโ )โ1 ๐ ๐ ห โ (๐ผ โ ๐(๐ง)๐ด ๐ต ๐=1 ๐=1
(5.74)
(5.75)
By (5.73) and (5.75), we have ([ ] ) ๐ โ ห โ ๐(๐ง)๐ด) ห โ1 ๐ถ(๐ผ [๐ด๐ ]๐ (๐ง) = ๐ด๐๐ (5.76) หโ )โ1 โ๐ ๐ฅ๐ หโ (๐ผ โ ๐(๐ง)๐ด ๐ต ๐=1 [ ] ] [ ๐ [ ] โ ห โ ๐(๐ง)๐ด) ห โ1 ห โ ๐(๐ง)๐ด) ห โ1 ๐ถ(๐ผ ๐ถ(๐ผ ห ห ๐ด โ โ ๐ด๐ฅ ๐ฅ = . = ๐ ๐๐ ๐ ๐ หโ (๐ผ โ ๐(๐ง)๐ด ห โ (๐ผ โ ๐(๐ง)๐ด หโ )โ1 หโ )โ1 ๐ ๐ต ๐ต ๐=1 Similarly, we get from (5.74) and (5.75) ] [ [ ] ห โ ๐(๐ง)๐ด) ห โ1 ๐ถ(๐ผ โ หโ ๐ฅ . โ๐ ๐ด [๐ด ๐ ]๐ (๐ง) = ห โ โ โ1 ห ๐ ๐ต (๐ผ โ ๐(๐ง)๐ด )
(5.77)
For ๐ and ๐ฅ as in (5.75), we have [ ] ๐ โ ห โ ๐(๐ง)๐ด) ห โ1 ๐ถ(๐ผ (s๐ )(๐ง) = หโ )โ1 โ๐ ๐ฅ๐ ห โ (๐ผ โ ๐(๐ง)๐ด ๐ต ๐=1 [ [ ] ๐ ] ห โ ๐(๐ง)๐ด) ห โ1 โ ห โ ๐(๐ง)๐ด) ห โ1 ๐ถ(๐ผ ๐ถ(๐ผ = หโ โ๐ ๐ฅ๐ = ห โ หโ )โ1 หโ )โ1 ๐ฅ ๐ต (๐ผ โ ๐(๐ง)๐ด ๐ต (๐ผ โ ๐(๐ง)๐ด ๐=1 which together with (5.76) and (5.77) gives ห โ ๐(๐ง)๐ด) ห โ1 ๐ฅ โ ๐ถ๐ฅ ห (s๐ )+ (๐ง) โ (s๐ )+ (0) = ๐ถ(๐ผ ห โ ๐(๐ง)๐ด) ห โ1 ๐(๐ง)๐ด๐ฅ ห = ๐ถ(๐ผ =
๐ โ ๐=1
๐ [ ] โ ห ห โ ๐(๐ง)๐ด) ห โ1 โ๐ ๐ด๐ฅ ๐ง๐ โ
๐ถ(๐ผ = ๐ง๐ โ
[๐ด๐ ]๐ (๐ง), ๐
๐=1
หโ (๐ผ โ ๐(๐ง)๐ด หโ )โ1 ๐ฅ โ ๐ต หโ๐ฅ (s๐ )โ (๐ง) โ (s๐ )โ (0) = ๐ต ห โ (๐ผ โ ๐(๐ง)๐ด หโ )โ1 ๐(๐ง)๐ด หโ ๐ฅ =๐ต =
๐ โ ๐=1
๐ [ ] โ หโ ๐ฅ = ห โ (๐ผ โ ๐(๐ง)๐ด หโ )โ1 โ๐ ๐ด ๐ง๐ โ
๐ต ๐ง๐ โ
[๐ดโ ๐ ]๐ (๐ง). ๐
โ๐๐=1 โ(๐พ๐๐ฟ ),
๐=1
the two latter equalities mean that Since ๐ is the generic element ๐ of the operators ๐ด and ๐ดโ solve Gleason problems (5.13) and (5.14), respectively. On
114
J.A. Ball and V. Bolotnikov
the other hand, for an ๐ฅ of the form (5.75), for operators ๐๐ de๏ฌned in (5.71) and for the operators ๐ถ and ๐ต โ de๏ฌned on โ๐๐=1 โ(๐พ๐ ) by formulas (5.15), we have ๐ถ(โ๐๐=1 ๐๐ )๐ฅ =
๐ โ
(๐๐ ๐ฅ๐ )+ (0) =
๐=1
๐ โ
ห โ ๐(0)๐ด) ห โ1 โ๐ ๐ฅ๐ = ๐ถ ห ๐ถ(๐ผ
๐=1
๐ โ
ห โ๐ ๐ฅ๐ = ๐ถ๐ฅ
๐=1
and quite similarly, ๐ต โ (โ๐๐=1 ๐๐ )๐ฅ =
๐ โ
(๐๐ ๐ฅ๐ )โ (0) =
๐=1
๐ โ
หโ )โ1 โ๐ ๐ฅ๐ = ๐ต หโ ห โ (๐ผ โ ๐(0)๐ด ๐ต
๐=1
๐ โ
ห โ ๐ฅ. โ๐ ๐ฅ๐ = ๐ต
๐=1
ห and ๐ต โ (โ๐ ๐๐ ) = ๐ต ห โ (which is equivalent to ๐ต = =๐ถ Thus, ๐=1 ห as the operator โ๐ ๐๐ is unitary). The two last equalities along with (โ๐๐=1 ๐๐ )๐ต ๐=1 ๐ด ๐ต ] is unitarily equivalent to the original (5.72) mean that[ the ]realization U = [ ๐ถ ๐ท ห ห ๐ด ๐ต ห = realization U via the unitary operator โ๐ ๐๐ . Therefore this realization ๐ถ(โ๐๐=1 ๐๐ )
๐=1
ห ๐ท ๐ถ
U is also weakly unitary. Also it is a functional-model realization since the state space ๐ณ is the functional-model state space โ๐๐=1 โ(๐พ๐ ), the operators ๐ต and ๐ถ are given by (5.15) and the state space operator ๐ด solves the Gleason problems in (5.13), (5.14). โก We next present the analog of Theorem 3.7 for the two-component setting for the case of unitary colligation matrices; the single-variable special case (๐ = 1) of this result amounts to Theorem 4 in [40]. Here we use notation as in (5.1) and (5.9). Theorem 5.10. Suppose that we are given a collection of โ(๐ด โ ๐ฐ)-valued positive kernels {[ ๐ฟ ๐ฟ๐
] [ ๐ฟ ๐ฟ๐
]} ๐พ1 ๐พ1 ๐พ๐ ๐พ๐ {๐พ1 , . . . , ๐พ๐ } = , . . . , . ๐พ ๐
๐ฟ ๐พ ๐
๐พ ๐
๐ฟ ๐พ ๐
1
1
๐
๐
Then {๐พ1 , . . . , ๐พ๐ } is the Agler decomposition for some unitary t.c.f.m. for some Schur-Agler-class function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) if and only if the following conditions hold: 1. The structured Gleason problem (5.13)โ(5.14) has an isometric solution ๐ด:
๐ โ
โ(๐พ๐ ) โ
๐=1
๐ โ
โ(๐พ๐ )
๐=1
in the sense that โฅ๐ด๐ โฅ2โ๐
๐=1
โฅ๐ดโ ๐ โฅ2โ๐
โ(๐พ๐ )
๐=1
= โฅ๐ โฅ2โ๐
โ(๐พ๐ )
๐=1
= โฅ๐ โฅ2โ๐
โ(๐พ๐ )
๐=1
โ โฅ(s๐ )+ (0)โฅ2๐ด ,
โ(๐พ๐ )
โ โฅ(s๐ )โ (0)โฅ2๐ฐ
(5.78)
โ๐ for all ๐ โ ๐=1 โ(๐พ๐ ). 2. The equality of range-defect dimensions
dim(Ran ๐ธ0 s+ )โฅ = dim(Ran ๐ธ0 sโ )โฅ
(5.79)
Canonical Realization
115
holds, where ๐ธ0 is the operator of evaluation at zero and where the maps ( ๐ ) ( ๐ ) ๐ ๐ โ โ โ โ ๐ฟ ๐
s+ : โ(๐พ๐ ) โ โ ๐พ๐ and sโ : โ(๐พ๐ ) โ โ ๐พ๐ ๐=1
๐=1
๐=1
๐=1
are given by s+ : ๐ โ
๐ โ
๐๐,+ ,
๐=1
sโ : ๐ โ
๐ โ
๐๐,โ .
๐=1
Moreover, if this is the case and if we de๏ฌne operators ๐ถ : โ๐ ๐ต : ๐ฐ โ ๐=1 โ(๐พ๐ ) by ๐ถ : ๐ โ (s๐ )+ (0),
โ๐
๐=1
โ(๐พ๐ ) โ ๐ด and
๐ต โ : ๐ โ (s๐ )โ (0),
(5.80)
๐ด ๐ต ] is a unitary t.c.f.m. then there exists an operator ๐ท : ๐ฐ โ ๐ด so that U = [ ๐ถ ๐ท for ๐ associated with the Agler decomposition {๐พ1 , . . . , ๐พ๐ } for ๐.
Proof. Necessity of the existence of a solution of the structured Gleason problem (5.13)โ(5.14) is immediate from the existence result, part (1) of Theorem 5.9, together with the de๏ฌnition of t.c.f.m. associated with a given Agler decomposition {๐พ1 , . . . , ๐พ๐ }. The additional conditions (5.78) (5.79) are a consequence of the ๐ด ๐ต ] is unitary. assumption that the t.c.f.m. colligation matrix U = [ ๐ถ ๐ท We next suppose that we are given a collection of kernels {๐พ1 , . . . , ๐พ๐ } as in the statement of the Theorem. De๏ฌne operators ๐ต and ๐ถ as in (5.80). The hypothesis (5.78) tells us that the block operators [ ] โ [ โ] โ ๐ ๐ ๐ ๐ โ โ ๐ด ๐ด : โ(๐พ๐ ) โ โ(๐พ๐ ) โ ๐ด and โ(๐พ๐ ) โ โ(๐พ๐ ) โ ๐ฐ โ : ๐ถ ๐ต ๐=1
๐=1
๐=1
๐=1
are isometric. We seek to de๏ฌne an operator ๐ท : ๐ฐ โ ๐ด in such a way that the ๐ด ๐ต ] is unitary. The isometric properties of [ ๐ด ] resulting matrix U = [ ๐ถ ๐ท ๐ถ [ colligation โ ] ๐ด and of ๐ต โ tell us that there exist isometries ๐ผ : ๐๐ด โ ๐ด and ๐ฝ : ๐๐ดโ โ ๐ฐ (where we have set ๐๐ด equal to the closure of the range of the operator ๐ท๐ด = (๐ผ โ๐ดโ ๐ด)1/2 and ๐๐ดโ equal to the closure of the range of the operator ๐ท๐ดโ := (๐ผ โ ๐ด๐ดโ )1/2 ) so that ๐ถ = ๐ผ๐ท๐ด and ๐ต โ = ๐ฝ๐ท๐ดโ . Note that Ran ๐ผ = Ran ๐ถ = Ran ๐ธ0 s+,
Ran ๐ฝ = Ran ๐ต โ = Ran ๐ธ0 sโ .
The dimension assumption (5.79) assures us that we can construct an isometry ๐พ from (Ran ๐ฝ)โฅ onto (Ran ๐ผ)โฅ . Let us now de๏ฌne an operator ๐ท : ๐ฐ โ ๐ด by { โ๐ผ๐ดโ ๐ฝ โ ๐ข if ๐ข โ Ran ๐ฝ, ๐ท๐ข = ๐พ๐ข if ๐ข โ (Ran ๐ฝ)โฅ
116
J.A. Ball and V. Bolotnikov
and extend it by linearity to all of ๐ฐ. Then it is easily checked that the colligation ๐ด ๐ต ] is unitary. We are now ready to de๏ฌne the Schur-Agler class matrix U = [ ๐ถ ๐ท function ๐(๐ง) by ๐(๐ง) = ๐ท + ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 ๐(๐ง)๐ต. ๐ We let โ๐ be the injection of [โ(๐พ๐ ) into of] [ ] โ๐=1 โ(๐พ๐ ) as in the proof ๐ฟ ๐ฟ๐
๐พ๐ ๐พ๐ ๐๐,+ Theorem 3.7 (but where now ๐พ๐ = ๐พ ๐
๐ฟ ๐พ ๐
and hence elements ๐๐ = ๐๐,โ ๐
๐
of โ(๐พ๐ ) consist of two components). We next argue that [ ] ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 ๐๐ (๐ง) = โ ๐ ๐ต โ (๐ผ โ ๐(๐ง)๐ด)โ1 ๐ ๐
(5.81)
or, equivalently, that ๐๐,+ (๐ง) = ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 โ๐ ๐๐ , โ
for all ๐๐ =
[
๐๐,+ ๐๐,โ
]
โ โ1
๐๐,โ (๐ง) = ๐ต (๐ผ โ ๐(๐ง)๐ด )
(5.82)
โ๐ ๐๐
(5.83)
in โ(๐พ๐ ). It su๏ฌces to note that (5.82) follows in the same way
as (3.32) in the proof of Theorem 3.7 based on the ๏ฌrst two-component Gleasonproblem identity (5.13). Similarly the second identity (5.83) follows in the same way by making use of the second two-component Gleason-problem identity (5.14), and hence (5.81) follows. We next make use of the reproducing-kernel property of ๐พ๐ to get โฉ[ ] [ ]โช ๐๐,+ (๐) ๐ฆ ๐ฆ โจ๐๐ , ๐พ๐ (โ
, ๐) [ ๐ข ]โฉโ(๐พ๐ ) = , ๐๐,โ (๐) ๐ข ๐ดโ๐ฐ โฉ[ ] [ ] [๐ฆ ]โช ๐ถ(๐ผ โ ๐(๐)๐ด)โ1 ๐๐,+ = (by (5.81)) โ ๐ ๐๐,โ , ๐ข ๐ต โ (๐ผ โ ๐(๐)๐ดโ )โ1 ๐ดโ๐ฐ [ ]โช ] โฉ[ ] ๐ฆ [ ๐๐,+ , โ๐โ (๐ผ โ ๐ดโ ๐(๐)โ ๐ถ โ (๐ผ โ ๐ด๐(๐)โ )โ1 ๐ต = ๐๐,โ ๐ข โ(๐พ ) ๐
from which we conclude that [ ] [ ๐ฆ ๐พ๐ (๐ง, ๐) = โ๐โ (๐ผ โ ๐ดโ ๐(๐)โ ๐ถ โ ๐ข
โ โ1
(๐ผ โ ๐ด๐(๐) )
[ ] ] ๐ฆ ๐ต . ๐ข
From the general identity (5.81) we conclude that [ ๐พ๐ (๐ง, ๐) =
] [ ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 โ โ โ1 โ ๐ถ โ โ โ โ1 ๐๐ (๐ผ โ ๐ด ๐(๐) ) ๐ต (๐ผ โ ๐(๐ง) ๐ด )
(๐ผ โ ๐ด๐(๐))โ1 ๐ต
]
(5.84) (where we have set ๐๐ = โ๐ โ๐โ ). ๐ด ๐ต ] is unitary (and hence in particular On the other hand, since U = [ ๐ถ ๐ท weakly unitary), it follows that ๐ admits an Agler decomposition (1.12) with
Canonical Realization kernels given by ] [ ห ๐ฟ (๐ง, ๐) ๐พ ห ๐ฟ๐
(๐ง, ๐) ๐พ ๐ ๐ ห ๐ (๐ง, ๐) = ๐พ ห ๐
(๐ง, ๐) ห ๐
๐ฟ (๐ง, ๐) ๐พ ๐พ ๐ ๐ ] [ [ ๐ถ(๐ผ โ ๐(๐ง)๐ด)โ1 โ โ โ1 โ ๐ถ = โ โ โ1 ๐๐ (๐ผ โ ๐ด ๐(๐) ) ๐ต (๐ผ โ ๐(๐ง)๐ด )
117
(๐ผ โ ๐ด๐(๐)โ )โ1 ๐ต
]
= ๐พ๐ (๐ง, ๐) (by (5.84)) and it follows that {๐พ1 (๐ง, ๐), . . . , ๐พ๐ (๐ง, ๐)} is an Agler decomposition for ๐.
โก
It is also possible to give a โweakly unitaryโ version of Theorem 5.10 Theorem 5.11. Given a collection of โ(๐ด โ ๐ฐ)-valued positive kernels {[ ๐ฟ ๐ฟ๐
] [ ๐ฟ ๐ฟ๐
]} ๐พ1 ๐พ1 ๐พ๐ ๐พ๐ {๐พ1 , . . . , ๐พ๐ } = , . . . , , ๐
๐ฟ ๐
๐พ1 ๐พ1 ๐พ๐๐
๐ฟ ๐พ๐๐
โ๐ let ๐ and โ be the subspaces of ๐=1 โ(๐พ๐ ) de๏ฌned in (5.3) and (5.4). Then {๐พ1 , . . . , ๐พ๐ } is the Agler decomposition for some t.c.f.m. for some function ๐ โ ๐ฎ๐๐ (๐ฐ, ๐ด) if and only if 1. The structured Gleason problem (5.13)โ(5.14) has a solution ๐ด:
๐ โ
โ(๐พ๐ ) โ
๐=1
๐ โ
โ(๐พ๐ )
๐=1
which is weakly unitary in the sense that the equalities in (5.78) hold with โค โ๐ in place of = for all ๐ โ ๐=1 โ(๐พ๐ ) and in addition the equalities โฅ๐๐ ๐ด๐ โฅ2โ๐
โ(๐พ๐ )
= โฅ๐ โฅ2โ๐
โ(๐พ๐ )
โ โฅ(s๐ )+ (0)โฅ2๐ด
for all ๐ โ โ,
โฅ๐โ ๐ดโ ๐โฅ2โ๐
โ(๐พ๐ )
= โฅ๐โฅ2โ๐
โ(๐พ๐ )
โ โฅ(s๐)โ (0)โฅ2๐ฐ
for all ๐ โ ๐
๐=1 ๐=1
๐=1
๐=1
(5.85)
where ๐๐ and ๐โ denote the orthogonal projections onto ๐ and โ. 2. The equality (5.79) of range-defect dimensions holds. Moreover, if this is the case and if we de๏ฌne operators ๐ถ and ๐ต as in (5.80), then ๐ด ๐ต ] is a t.c.f.m. for ๐ associated there exists an operator ๐ท : ๐ฐ โ ๐ด so that U = [ ๐ถ ๐ท with the Agler decomposition {๐พ1 , . . . , ๐พ๐ } for ๐. Proof. As the proof is mostly the same as that of Theorem 5.10, we just sketch the main ideas. The one di๏ฌerence from the proof of Theorem 5.10 is that in the su๏ฌciency part we start with isometric operators [ ] [ ] หโ ห ๐ด ๐ด : ๐ โโโ๐ฐ : โ โ ๐ โ ๐ด and ๐ตโ ๐ถ ห = ๐๐ ๐ดโฃโ . We use the same formulas as in the previous where we have set ๐ด ห theorem (with ๐ด instead of [๐ด) to construct ๐ท and then invoke Lemma 5.7 to ]
show that the operator U =
๐ 0 0
0 0 ห๐ต ๐ด ๐ถ ๐ท
is a unitary t.c.f.m. for ๐(๐ง) = ๐ท + ๐ถ(๐ผ โ
๐(๐ง)๐ด)โ1 ๐(๐ง)๐ต associated with the Agler decomposition {๐พ1 , . . . , ๐พ๐ }.
โก
118
J.A. Ball and V. Bolotnikov
Remark 5.12. As was the case for left Agler decompositions associated with c.f.m.โs (see Subsection 3.2), not much is known about the construction and structure of Agler decompositions associated with t.c.f.m.โs. However, in [26] there appears an example of an Agler decomposition (arising from an explicit closely connected unitary structured colligation matrix U) for which both ๐ and โ are proper โ๐ subspaces of ๐=1 โ(๐พ๐ ) of codimension 1. Left open there (and here) is whether โ๐ โ๐ there exists an example where ๐ โ= ๐=1 โ(๐พ๐ ) but โ = ๐=1 โ(๐พ๐ ) (or vice versa). More generally, we are lacking an example where ๐ and โ have unequal โ๐ codimensions in ๐=1 โ(๐พ๐ ), i.e., an example of an Agler decomposition for which no associated t.c.f.m. can be unitary.
References [1] V.M. Adamjan and D.Z. Arov, On unitary coupling of semi-unitary operators, Dokl. Akad. Nauk. Arm. SSR 43 (1966), no. 5, 257โ263. [2] J. Agler. On the representation of certain holomorphic functions de๏ฌned on a polydisk, in Topics in Operator Theory: Ernst D. Hellinger memorial Volume (eds. L. de Branges, I. Gohberg and J. Rovnyak), Oper. Theory Adv. Appl. OT 48, pp. 47โ66, Birkhยจ auser Verlag, Basel, 1990. [3] J. Agler and J.E. McCarthy, Nevanlinna-Pick interpolation on the bidisk, J. Reine Angew. Math. 506 (1999), 191โ204. [4] J. Agler and J.E. McCarthy, Complete Nevanlinna-Pick kernels, J. Funct. Anal. 175 (2000) no. 1, 111โ124. [5] D. Alpay and C. Dubi, Backward shift operator and ๏ฌnite-dimensional de BrangesRovnyak spaces in the ball, Linear Algebra Appl. 371 (2003), 277โ285. [6] D. Alpay and C. Dubi, A realization theorem for rational functions of several complex variables, Systems Control Lett. 49 (2003) no. 3, 225โ229. [7] D. Alpay, A. Dijksma, J. Rovnyak, and H. de Snoo, Schur functions, operator colligations, and reproducing kernel Pontryagin spaces, Oper. Theory Adv. Appl. OT 96, Birkhยจ auser Verlag, Basel, 1997. [8] D. Alpay and D.S. Kalyuzhnyi-Verbovetzkyi, Matrix-๐ฝ-unitary non-commutative rational formal power series, in The State Space Method: Generalizations and Applications pp. 49โ113 (eds. D. Alpay and I. Gohberg), Oper. Theory Adv. Appl. OT 161, Birkhยจ auser Verlag, Basel-Boston-Berlin, 2006. [9] D. Alpay and H.T. Kaptanoห glu, Gleasonโs problem and homogeneous interpolation in Hardy and Dirichlet-type spaces of the ball, J. Math. Anal. Appl. 276 (2002) no. 2, 654โ672. [10] C.-G. Ambrozie and D. Timotin, A von Neumann type inequality for certain domains in โ๐ , Proc. Amer. Math. Soc., 131 (2003), no. 3, 859โ869. [11] N. Aronszajn, Theory of reproducing kernels, Trans. Amer. Math. Soc., 68 (1950), 337โ404. [12] D.Z. Arov and O.J. Sta๏ฌans, A Kreหin-space coordinate-free version of the de BrangesRovnyak complementary space, J. Funct. Anal. 256 (2009) no. 12, 3892โ3915.
Canonical Realization
119
[13] D.Z. Arov and O.J. Sta๏ฌans, Two canonical passive state/signal shift realizations of passive discrete-time behaviors, J. Funct. Anal. 257 (2009) no. 8, 2573โ2634. [14] J.A. Ball, Linear systems, operator model theory and scattering: multivariable generalizations, in: Operator theory and its applications, 151โ178, Fields Inst. Commun., 25, Amer. Math. Soc., Providence, RI, 2000. [15] J.A. Ball, A. Biswas, Q. Fang and S. ter Horst, Multivariable generalizations of the Schur class: positive kernel characterization and transfer function realization, in Recent advances in operator theory and applications (eds. T. Ando, R.E. Curto, I.B. Jung, and W.Y. Lee), pp. 17โ79, Oper. Theory Adv. Appl. OT 187, Birkhยจ auser Verlag, 2009. [16] J.A. Ball and V. Bolotnikov, Realization and interpolation for Schur-Agler-class functions on domains with matrix polynomial de๏ฌning function in โ๐ , J. Funct. Anal. 213 (2004), no.1, 45โ87. [17] J.A. Ball and V. Bolotnikov, Canonical de Branges-Rovnyak model transfer-function realization for multivariable Schur-class functions, in Hilbert Spaces of Analytic Functions (eds. J. Mashreghi, T. Ransford, and K. Seip), CRM Proceedings & Lecture Notes 51, Amer. Math. Soc., Providence, 2010. [18] J.A. Ball and V. Bolotnikov, Canonical transfer-function realization for Schur-Aglerclass functions on domains with matrix polynomial de๏ฌning functions in โ๐ , in Recent Progress in Operator Theory and Its Applications, J.A. Ball, R. Curto, S. Grudsky, J.W. Helton, R. Quiroga-Barrancoi, and N. Vasilevski, eds., Proceedings of the International Workshop on Operator Theory and Applications (IWOTA), Guanajuato, Mexico, Oper. Theory Adv. Appl. OT 220, Birkhยจ auser, Springer Basel AG, 2012, 23โ55. [19] J.A. Ball and V. Bolotnikov, Canonical transfer-function realization for Schur multipliers on the Drury-Arveson space and models for commuting row contractions, Ind. U. Math. J., to appear. [20] J.A. Ball, V. Bolotnikov and Q. Fang, Transfer-function realization for multipliers of the Arveson space, J. Math. Anal. Appl. 333 (2007), no. 1, 68โ92. [21] J.A. Ball, V. Bolotnikov and Q. Fang, Schur-class multipliers on the Fock space: de Branges-Rovnyak reproducing kernel spaces and transfer-function realizations, in Operator Theory, Structured Matrices, and Dilations: Tiberiu Constantinescu Memorial Volume (eds. M. Bakonyi, A. Gheondea, M. Putinar and J. Rovnyak), pp. 85โ114, Theta Press, Bucharest, 2007. [22] J.A. Ball, K.F. Clancey, and V. Vinnikov, Concrete interpolation of meromorphic matrix functions on Riemann surfaces, in Reproducing kernel spaces and applications pp. 77โ134, Oper. Theory Adv. Appl. OT 143, Birkhยจ auser Verlag, 2003. [23] J.A. Ball, I. Gohberg, and L. Rodman, Interpolation of rational matrix functions, Oper. Theory Adv. Appl. OT 45, Birkhยจ auser Verlag, Basel, 1990. [24] J.A. Ball, G. Groenewald and T. Malakorn, Structured noncommutative multidimensional linear systems, SIAM J. Control Optim. 44 (2005), no. 4, 1474โ1528. [25] J.A. Ball, G. Groenewald and T. Malakorn, Conservative structured noncommutative multidimensional linear systems, in The State Space Method: Generalizations and Applications (eds. D. Alpay and I. Gohberg), pp. 179โ223, Oper. Theory Adv. Appl. OT 161, Birkhยจ auser Verlag, Basel, 2006.
120
J.A. Ball and V. Bolotnikov
[26] J.A. Ball, D.S. Kaliuzhnyi-Verbovetskyi, C. Sadosky, V. Vinnikov, Scattering systems with several evolutions and formal reproducing kernel Hilbert spaces, in preparation. [27] J.A. Ball, C. Sadosky, V. Vinnikov, Scattering systems with several evolutions and multidimensional input/.state/output systems, Integral Equations Operator Theory 52 (2005), no. 3, 323โ393. [28] J.A. Ball and T.T. Trent, Unitary colligations, reproducing kernel Hilbert spaces and NevanlinnaโPick interpolation in several variables, J. Funct. Anal., 157 (1998), no. 1, 1โ61. [29] J.A. Ball, T.T. Trent and V. Vinnikov, Interpolation and commutant lifting for multipliers on reproducing kernel Hilbert spaces, in Operator Theory and Analysis: The M.A. Kaashoek Anniversary Volume (eds. H. Bart, I. Gohberg and A.C.M. Ran), pp. 89โ138, Oper. Theory Adv. Appl. OT 122, Birkhยจ auser Verlag, Basel, 2001. [30] J.A. Ball and V. Vinnikov, Zero-pole interpolation for meromorphic matrix functions on an algebraic curve and transfer functions for 2D systems, Acta Appl. Math. 45 (1996) no. 3, 239โ316. [31] J.A. Ball and V. Vinnikov, Zero-pole interpolation for matrix meromorphic functions on a compact Riemann surface and a matrix Fay trisecant identity, Amer. J. Math. 121 (1999) no. 4, 841โ888. [32] J.A. Ball and V. Vinnikov, Overdetermined multidimensional systems: state space and frequency domain methods, in Mathematical systems theory in biology, communications, computation, and ๏ฌnance (eds. J. Rosenthal and D.S. Gilliam), pp. 63โ119, IMA Vol. Math. Appl. 134, Springer, New York, 2003. [33] J.A. Ball and V. Vinnikov, Lax-Phillips scattering and conservative linear systems: A Cuntz-algebra multidimensional setting, Memoirs of the American Mathematical Society, 178 no. 837, American Mathematical Society, Providence, 2005. [34] H. Bart, I. Gohberg, and M.A. Kaashoek, Minimal factorization of matrix and Operator functions, Oper. Theory Adv. Appl. OT 1, Birkhยจ auser Verlag, 1979. [35] H. Bart, I. Gohberg, M.A. Kaashoek, and A.C.M. Ran, Factorization of matrix and operator functions: the state space method, Oper. Theory Adv. Appl. OT 178, Birkhยจ auser Verlag, Basel, 2008. [36] H. Bart, I. Gohberg, M.A. Kaashoek, and A.C.M. Ran, A state space approach to canonical factorization with applications, Oper. Theory Adv. Appl. OT 200, Birkhยจ auser Verlag, Basel, 2010. [37] V. Belevitch, Classical Network Theory, Holden-Day, San Francisco, 1968. [38] T. Bhattacharyya, J. Eschmeier and J. Sarkar, Characteristic function of a pure commuting contractive tuple, Integral Equations Operator Theory 53 (2005), no. 1, 23โ32. [39] T. Bhattacharyya, J. Eschmeier and J. Sarkar, On c.n.c. commuting contractive tuples, Proc. Indian Acad. Sci. Math. Sci. 116 (2006), no. 3, 299โ316. [40] L. de Branges, Factorization and invariant subspaces, J. Math. Anal. Appl. 29 (1970), 163โ200. [41] L. de Branges and J. Rovnyak, Canonical models in quantum scattering theory, in: Perturbation Theory and its Applications in Quantum Mechanics (C. Wilcox, ed.) pp. 295โ392, Holt, Rinehart and Winston, New York, 1966.
Canonical Realization
121
[42] L. de Branges and J. Rovnyak, Square summable power series, Holt, Rinehart and Winston, New York, 1966. [43] V. Brodskiหฤฑ, I. Gohberg, and M.G. Kreหฤฑn, The characteristic function of an invertible operator, Acta Sci. Math. (Szeged) 32 (1971), 141โ164. [44] J. Eschmeier and M. Putinar, Spherical contractions and interpolation problems on the unit ball, J. Reine Angew. Math. 542 (2002), 219โ236. [45] C. Foiaยธs, A. Frazho, I. Gohberg and M.A. Kaashoek, Metric Constrained Interpolation, Commutant Lifting and Systems, Oper. Theory Adv. Appl. OT 100, Birkhยจ auser Verlag, Boston-Basel, 1998. [46] A.M. Gleason, Finitely generated ideals in Banach algebras, J. Math. Mech., 13 (1964), 125โ132. [47] I. Gohberg and M.A. Kaashoek (eds.), Constructive Methods of Wiener-Hopf Factorization, Oper. Theory Adv. Appl. OT 21, Birkhยจ auser Verlag, Basel, 1986. [48] I. Gohberg, P. Lancaster, and L. Rodman, Matrix polynomials, Academic Press, New York, 1982. [49] A Grinshpan, D.S. Kaliuzhnyi-Verbovetskyi, V. Vinnikov, and H.J. Woerdeman, Classes of tuples of commuting contractions satisfying the multivariable von Neumann inequality, J. Funct. Anal. 256 (2009), no. 9, 3035โ3054. [50] J.W. Helton, S. McCullough and V. Vinnikov, Noncommutative convexity arises from linear matrix inequalities, J. Functional Analysis 240 (2006) no. 1, 105โ191. [51] R.E. Kalman, Mathematical description of linear dynamical systems, J. SIAM Control Ser. A 1 (1963), 152โ192. [52] R.E. Kalman, P.L. Falb, and M.A. Arbib, Topics in mathematical system theory, McGraw-Hill, New York, 1969. [53] P.D. Lax and R.S. Phillips, Scattering Theory, Pure and Applied Math. 26, Academic Press, Boston, 1989. [54] M.S. Livหsic, On a class of linear operators in Hilbert space, Mat. Sbornik N.S. 19(61) (1946), 239โ262; English translation: Amer. Math. Soc. Transl. (2) 13 (1960), 61โ83. [55] M.S. Livหsic, Operators, oscillations, waves (Open Systems), Translations of Mathematical Monographs 34, Amer. Math. Soc., Providence, 1973. [56] P.S. Muhly and B. Solel, Hardy algebras, ๐ โ correspondences and interpolation theory, Math. Ann. 330 (2004), no. 2, 353โ415. [57] P.S. Muhly and B. Solel, Canonical models for representations of Hardy algebras, Integral Equations Operator Theory, 53 (2005), no. 3, 411โ452. [58] B. Sz.-Nagy and C. Foias, Harmonic analysis of operators on Hilbert space, North-Holland/American Elsevier, 1970; revised edition: B. Sz.-Nagy, C. Foias, H. Bercovici, and L. Kerchy, Harmonic analysis of operators on Hilbert space. Second edition. Revised and enlarged edition., Universitext, Springer, New York, 2010. [59] G. Popescu, Characteristic functions for in๏ฌnite sequences of noncommuting operators, J. Operator Theory 22 (1989), 51โ71. [60] G. Popescu, von Neumann inequality for (๐ต(โ)๐ )1 , Math. Scand. 68 (1991), 292โ 304. [61] G. Popescu, Multi-analytic operators on Fock spaces, Math. Ann. 303 (1995), 31โ46.
122
J.A. Ball and V. Bolotnikov
[62] D. Sarason, Sub-Hardy Hilbert Spaces in the Unit Disk, John Wiley and Sons Inc., New York, 1994. [63] J.C. Willems, Dissipative dynamical systems I: general theory, Arch. Rational Mech. Anal. 45 (1972), 321โ351. [64] J.C. Willems, Dissipative dynamical systems II: Linear systems with quadratic supply rates, Arch. Rational Mech. Anal. 45 (1972), 352โ393. [65] M.R. Wohlers, Lumped and distributed passive networks: a generalized and advanced viewpoint, Academic Press, New York, 1969. Joseph A. Ball Department of Mathematics Virginia Tech Blacksburg, VA 24061-0123, USA e-mail:
[email protected] Vladimir Bolotnikov Department of Mathematics The College of William and Mary Williamsburg, VA 23187-8795, USA e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 123โ153 c 2012 Springer Basel AG โ
Spectral Regularity of Banach Algebras and Non-commutative Gelfand Theory Harm Bart, Torsten Ehrhardt and Bernd Silbermann Dedicated to Israel Gohberg, in grateful recognition of his wonderful contributions to mathematics
Abstract. A new non-commutative Gelfand type criterion for spectrally regular behavior of vector-valued analytic functions is developed. Applications are given in situations that could not be handled with earlier methods. Some open problems are identi๏ฌed. Mathematics Subject Classi๏ฌcation (2000). Primary: 30G30, 46H99; Secondary: 47A56, 47L10. Keywords. Analytic vector-valued function, logarithmic residue, spectral regularity, polynomial identity algebra, radical, family of homomorphisms, family of matrix representations.
1. Introduction Let ฮ be a bounded Cauchy domain in the complex plane โ, let ๐ be a complex function de๏ฌned and analytic on an open neighborhood of the closure of ฮ, and suppose ๐ does not vanish on the boundary โฮ of ฮ. From complex function theory we know that the contour integral โซ 1 ๐ โฒ (๐) ๐๐ 2๐๐ โฮ ๐ (๐) is equal to the number of zeros of ๐ in ฮ. Hence it vanishes if and only if ๐ (๐) โ= 0 for each ๐ โ ฮ. The issue studied in the present paper is this: to what extent does the state of a๏ฌairs in the scalar case carry over to the more general Banach algebra setting? So the problem we investigate is the following. Let โฌ be a (nontrivial) unital (complex) Banach algebra, let ๐น be a โฌ-valued function de๏ฌned and analytic on an open neighborhood of the closure of a bounded Cauchy domain ฮ, and suppose ๐น takes invertible values on the boundary โฮ of ฮ. Does it follow (or under what extra conditions can one conclude) that ๐น takes invertible values on ฮ provided
124
H. Bart, T. Ehrhardt and B. Silbermann
it is given that the contour integral โซ 1 ๐น โฒ (๐)๐น (๐)โ1 ๐๐ 2๐๐ โฮ
(1)
vanishes? If, for the Banach algebra โฌ under consideration, the answer is always positive, then โฌ is called spectrally regular. Clearly, the archetypical example of such a Banach algebra is โ. A necessary condition for a Banach algebra to be spectrally regular is that it does not feature any nontrivial zero sum of idempotents (see [BES2]). In [BES1], a nontrivial zero sum of ๏ฌve idempotents is constructed for โฌ(โ2 ), the Banach algebra of bounded linear operators on the Hilbert space โ2 (cf. [E] and [PT]). Thus โฌ(โ2 ) is not spectrally regular. On the other hand, in the papers [Bar], [BES2], [BES3], [BES4] and [BES6], spectral regularity has been established for large classes of Banach algebras. Before we proceed, let us mention that in the present article, actually a somewhat stronger form of spectral regularity is adopted than described above. It is one that takes into account the phenomenon of quasinilpotency. Indeed, in the sequel we will call a unital Banach algebra โฌ spectrally regular if the following holds true: in the situation indicated above in which (1) is (well) de๏ฌned, the function ๐น has invertible values on ฮ provided (1) is quasinilpotent, i.e., has the singleton set {0} as its spectrum. The Banach algebras for which spectral regularity in the weaker sense was established in the papers [Bar], [BES2], [BES3], [BES4] and [BES6] are spectrally regular in the stronger sense too (see Section 2 below). The methods that have been used in the mentioned articles can be divided into two categories: those using trace arguments (in cases where Fredholm operators enter the picture), and those employing Gelfand type considerations (in situations where commutativity properties play a role). The approach via trace arguments has been systematically pursued in [BES5] and [BES6]. The present paper is devoted to a further exploration along the other line, where su๏ฌcient conditions for spectral regularity are established with the help of Gelfand type considerations. The ๏ฌrst step in this direction was taken in [Bar], dealing with the commutative case and using classical Gelfand theory; a second in [BES2], where (among others) polynomial identity algebras were considered and it was necessary to take recourse to non-commutative Gelfand theory, with matrix representations taking the place of the multiplicative linear functionals from classical Gelfand theory (see [Kr] or [P], Section 7.1). Here is a brief description of the contents of the present paper. Apart from the introduction (Section 1) and the list of references, the paper consists of four sections. Section 2 contains preliminaries on notation and terminology, as well as a review of earlier results serving as the proper context in which to position the material presented in the rest of the article. In Section 3, a new Gelfand type criterion for spectral regularity is derived, and, with an eye on applications later on in the paper, two corollaries are obtained. The results involve families of homomorphisms that are more general than the so-called su๏ฌcient families of
Spectral Regularity and Non-commutative Gelfand Theory
125
matrix representations that have been employed before in [BES2]. One of the two corollaries has a strong algebraic aspect in that it is formulated in terms of the radical of the underlying Banach algebra. The other corollary is concerned with Banach algebras that, in a certain sense, can be embedded in Banach algebras of bounded linear operators on a Banach space. Here the semigroup of Fredholm operators features as an important ingredient, and so in the background the ideal of the compact operators on a Banach space and the Calkin algebra play a role. The new criterion and (especially) its corollaries turn out to be e๏ฌective tools, enabling us to deal with a variety of situations which we could not handle earlier. This is illustrated by the material presented in Section 4. Here are three examples: 1) a unital Banach algebra โฌ is spectrally regular if and only if โฌ factored by its radical is; 2) the ๐ถ โ -algebra generated by the block Toeplitz operators having a piecewise continuous de๏ฌning (also called generating) function is spectrally regular; and 3) the same is true for the Banach subalgebra of โฌ(โ2 ) consisting of the bounded linear operators on โ2 having a block upper triangular matrix representation with respect to an orthonormal basis in โ2 . The analysis presented in this paper hinges on the use of certain families of Banach algebra homomorphisms having properties pertinent to the study of spectral regularity. Section 5 contains a few remarks about how the di๏ฌerent properties in question compare. In particular it is brie๏ฌy pointed out that the conceptual framework developed in Section 3 provides a genuine extension of the non-commutative Gelfand theory employed before in [BES2]. For a detailed analysis, see the forthcoming paper [BES7]. The expression (1) de๏ฌnes the left logarithmic residue of the function ๐น with respect to the Cauchy domain ฮ. There is also a right version obtained by replacing the left logarithmic derivative ๐น โฒ (๐)๐น (๐)โ1 by the right logarithmic derivative ๐น (๐)โ1 ๐น โฒ (๐). Accordingly one can make a distinction between left spectral regularity and right spectral regularity. For all results obtained in this paper, the left and right versions are analogous to one another. Therefore we will only consider the left version of the logarithmic residue and drop the quali๏ฌer โleftโ altogether. Note, however, that is not known whether a Banach algebra can be left spectrally regular while failing to be right spectrally regular. One ๏ฌnal remark. The Banach algebras considered in this paper are unital. They are nontrivial too, so their unit elements di๏ฌer from their zero elements. It is not assumed, however, that the unit elements have norm one. For an individual unital Banach algebra one can always renorm such that the unit element does have norm one. In working with families of Banach algebra homomorphisms the way we do here, a ๏ฌxation on unit elements with norm one would introduce an unnecessary and undesirable rigidity.
2. Preliminaries and review of earlier results In this section we review some earlier results. We also use the opportunity to ๏ฌx notations and to introduce terminology.
126
H. Bart, T. Ehrhardt and B. Silbermann
A spectral con๏ฌguration is a triple (โฌ, ฮ, ๐น ) where โฌ is a unital complex Banach algebra, ฮ is a bounded Cauchy domain in โ (see [TL] or [GGK1]) and ๐น is a โฌ-valued analytic function on an open neighborhood of the closure of ฮ which has invertible values on all of the boundary โฮ of ฮ. With such a spectral con๏ฌguration, taking โฮ to be positively oriented, one can associate the contour integral โซ 1 ๐ฟ๐
(๐น ; ฮ) = ๐น โฒ (๐)๐น (๐)โ1 ๐๐. 2๐๐ โฮ We call it the logarithmic residue associated with (โฌ, ฮ, ๐น ); sometimes the term logarithmic residue of ๐น with respect to ฮ is used as well. In the scalar case โฌ = โ, the logarithmic residue โซ 1 ๐ โฒ (๐) ๐๐ (2) 2๐๐ โฮ ๐ (๐) associated with a spectral con๏ฌguration (โ, ฮ, ๐ ) is equal to the number of zeros of ๐ in ฮ (multiplicities counted). This can be rephrased by saying that (2) is the winding number with respect to the origin of the curve {๐ (๐)}๐โโฮ , taken with the orientation induced by the one on โฮ. Thus ๐ฟ๐
(๐, ฮ) is a nonnegative integer which is zero if and only if ๐ does not vanish on ฮ. Motivated by these facts, and taking into account that in the general Banach algebra situation one can have nonzero quasinilpotent elements, we introduce the following terminology. The spectral con๏ฌguration (โฌ, ฮ, ๐น ) is said to be winding free when ๐ฟ๐
(๐น ; ฮ) = 0, spectrally winding free if ๐ฟ๐
(๐น ; ฮ) is quasinilpotent, and spectrally trivial in case ๐น takes invertible values on ฮ. By Cauchyโs theorem a spectral con๏ฌguration is winding free (and a fortiori spectrally winding free) provided it is spectrally trivial. As mentioned in the introduction, the converse of this is not generally true in the vector-valued situation. Under certain ๏ฌnite dimensionality conditions, positive results can be obtained. A bounded linear operator ๐ on a Banach space ๐ is called a Fredholm operator if its null space Ker ๐ is ๏ฌnite dimensional and its range Im ๐ has ๏ฌnite codimension in ๐ (and is therefore closed). The following theorem, extending Corollary 3.3 in [BES3], will serve as a key tool later on. Without going into details, we mention that the result allows for an extension to an abstract ๐ถ โ -algebra setting; see the forthcoming paper [BES8]. Theorem 2.1. Let ๐ be a Banach space, let (โฌ(๐), ฮ, ๐น ) be a spectral con๏ฌguration, and suppose ๐น is Fredholm operator valued on ฮ. The following statements are equivalent: (1) (โฌ(๐), ฮ, ๐น ) is spectrally trivial; (2) (โฌ(๐), ฮ, ๐น ) is winding free; (3) (โฌ(๐), ฮ, ๐น ) is spectrally winding free. Proof. Statements (1) and (2) are equivalent by Corollary 3.3 in [BES3]. Obviously (2) โ (3), and it remains to prove the implication (3) โ (2). Assume ๐ฟ๐
(๐น ; ฮ)
Spectral Regularity and Non-commutative Gelfand Theory
127
is quasinilpotent. The Fredholmness of ๐น implies that ๐ฟ๐
(๐น ; ฮ) is a ๏ฌnite rank operator on ๐ (see [GS]). Hence ๐ฟ๐
(๐น ; ฮ) is nilpotent. In particular its trace vanishes. By Proposition 3.2 in [BES3], the rank of the logarithmic residue ๐ฟ๐
(๐น ; ฮ) does not exceed its trace, and it follows that ๐ฟ๐
(๐น ; ฮ) = 0. โก Results as Theorem 2.1 are concerned with spectral con๏ฌgurations in which a given individual function has special properties. Without these additional properties, (spectrally) winding free spectral con๏ฌgurations with the same underlying Banach algebra might fail to be spectrally trivial. Lifting our conceptual framework to that of the underlying algebras, we call a unital Banach algebra โฌ spectrally regular if each spectrally winding free spectral con๏ฌguration (โฌ, ฮ, ๐น ) is spectrally trivial. Not every Banach algebra is spectrally regular. Indeed, from what was said in the introduction, it is clear that โฌ(โ2 ) is not. In [Bar], [BES2] and [BES4] positive results have been obtained, but these concern the (possibly) somewhat weaker type of spectral regularity featuring in those papers. That version of spectral regularity requires the triviality of a spectral con๏ฌguration (โฌ, ฮ, ๐น ) to follow from the con๏ฌguration being winding free instead of it being spectrally winding free. Nevertheless, all the Banach algebras that have been identi๏ฌed in [Bar], [BES2] and [BES4] as spectrally regular in this weaker sense are actually spectrally regular in the stronger sense considered in here. This can be seen by looking at the proofs given in [Bar], [BES2] and [BES4], but it will also become clear from the material to be presented below. We do not know whether the weak and the strong version of spectral regularity really di๏ฌer from each other or actually amount to the same. The matrix algebras โ๐ร๐ are spectrally regular. For the form of spectral regularity employed here (stronger than in our earlier publications), this conclusion can be obtained from Theorem 2.1 since matrices can be viewed as Fredholm operators. More generally, when ๐ is a Banach space, ๐ฆ(๐) stands for the ideal of the compact operators on ๐, and ๐ผ๐ denotes the identity operator on ๐, the Banach subalgebra โฌ๐ฆ (๐) = {๐๐ผ๐ + ๐ โฃ ๐ โ โ, ๐ โ ๐ฆ(๐)} of โฌ(๐) is spectrally regular. In case ๐ is ๏ฌnite dimensional, โฌ๐ฆ (๐) can be identi๏ฌed with the matrix algebra โ๐ร๐ where ๐ is the dimension of ๐. In case dim ๐ = โ, the result follows by combining Proposition 4.1 in [BES4] and Theorem 2.1. Commutative unital Banach algebras are spectrally regular too (see [Bar]). Such algebras belong to the wider class of polynomial identity algebras. A Banach algebra โฌ is called a polynomial identity (Banach) algebra, PI-algebra for short, if there exist a positive integer ๐ and a nontrivial polynomial ๐(๐ฅ1 , . . . , ๐ฅ๐ ) in ๐ noncommuting variables ๐ฅ1 , . . . , ๐ฅ๐ such that ๐(๐1 , . . . , ๐๐ ) = 0 for every choice of elements ๐1 , . . . , ๐๐ in โฌ. Clearly commutativity implies the property of being PI. Also, according to a celebrated result of Amitsur and Levitzky [AL], all algebras of the form โ๐ร๐ are PI-algebras. PI-algebras are spectrally regular (see [BES2] and below). PI-algebras have been investigated in [Kr], and we will now discuss material from there which is highly pertinent to the topic of the present paper (cf. Section 7.1 in [P]). For this, two more concepts are needed.
128
H. Bart, T. Ehrhardt and B. Silbermann
The ๏ฌrst is that of the radical. The de๏ฌnition and basic properties of this fundamental notion can be found, for instance, in [N], Section II.7.5 and [Kr], Section 13. For our purpose it is important to know that the radical โ(โฌ) of a unital Banach algebra โฌ is a closed two-sided ideal in โฌ which can be characterized as follows: an element ๐ in โฌ belongs to โ(โฌ) if and only if for each ๐ฅ โ โฌ both ๐ + ๐ฅ๐ and ๐ + ๐๐ฅ are invertible in โฌ. Here ๐ is the unit element in โฌ. We also need the concept of a su๏ฌcient family. A family {๐๐ : โฌ โ โฌ๐ }๐โฮฉ of continuous unital Banach algebra homomorphisms is said to be su๏ฌcient when an element ๐ โ โฌ is invertible in โฌ if and only if ๐๐ (๐) is invertible in โฌ๐ for all ๐ โ ฮฉ. Note that the โonly if partโ in this de๏ฌnition is automatically ful๏ฌlled. A continuous Banach algebra homomorphism into a matrix algebra โ๐ร๐ , ๐ a positive integer, is called a matrix representation. A family of matrix representations {๐๐ : โฌ โ โ๐๐ ร๐๐ }๐โฮฉ is said to be of ๏ฌnite order if the sizes of the matrices involved have a ๏ฌnite upper bound, i.e., if sup๐โฮฉ ๐๐ < โ. With this terminology, the following basic result holds: a unital Banach algebra โฌ possesses a su๏ฌcient family of matrix representations of ๏ฌnite order if and only if the quotient algebra โฌ/โ(โฌ) is a PI-algebra. The latter condition is obviously satis๏ฌed when โฌ itself is a PI-algebra. Hence, if โฌ is a PI-algebra, then โฌ possesses a su๏ฌcient family of matrix representations of ๏ฌnite order. For later reference (see Subsection 4.1), we also mention that, as a consequence, โฌ possesses a su๏ฌcient family of ๏ฌnite order if and only if so does the quotient algebra โฌ/โ(โฌ). We complete the exposition of this material by pointing out that the existence of a su๏ฌcient family of matrix representations (not necessarily of ๏ฌnite order) implies the spectral regularity of the underlying algebra. For the weaker form of spectral regularity used in our earlier papers, this result is contained in [BES2], Theorem 4.1. For the stronger form under consideration here, it is immediate from the spectral regularity of the matrix algebras and Corollary 3.5 below. At this point, we can make a connection with Problem 12 in [Kr], Section 29: characterize those Banach algebras which possess a su๏ฌcient family of matrix representations not necessarily of ๏ฌnite order. Spectral regularity is a necessary requirement for this; it is however not a su๏ฌcient condition (see the last paragraph in Section 5). There is one more class of spectrally regular Banach algebras that we want to mention: that of the Banach algebras covered by Theorem 4.2 in [BES2] and the remark made after that theorem. It is a subclass of a class of Banach algebras appearing in (numerically oriented) work by Hagen, Roch and the third author (see [Si] and [HRS]). The description of the class is somewhat involved, and we refrain from giving further details here. Theorems 4.1 and 4.2 in [BES2] referred to above are Gelfand type criteria in the sense that they are stated in terms of families of Banach algebra homomorphisms. In the next section we shall develop a new criterion of this type which turns out to be e๏ฌective for establishing spectral regularity in a variety of cases which we were not able to handle with the old tools.
Spectral Regularity and Non-commutative Gelfand Theory
129
3. A new Gelfand type criterion for spectral regularity In this section, we will extensively work with families of Banach algebra homomorphisms. These homomorphisms need not be unital. If ๐ is a Banach space, the identity operator on ๐ is denoted by ๐ผ๐ and the set of Fredholm operators on ๐ by โฑ (๐). Theorem 3.1. Let โฌ be a unital Banach algebra. For ๐ in an index set ฮฉ, let โฌ๐ be a spectrally regular Banach algebra and let ๐๐ : โฌ โ โฌ๐ be a continuous homomorphism. Further, for ๐ก in an index set ๐ , let ๐๐ก be a nontrivial Banach space and let ฮฆ๐ก : โฌ โ โฌ(๐๐ก ) be a continuous homomorphism. Assume the following two inclusions hold: โฉ โฉ (a) Ker ๐๐ โ ฮฆโ1 ๐ก [โฑ (๐๐ก ) โ {๐ผ๐๐ก }], ๐โฮฉ
(b)
โฉ
๐กโ๐
Ker ฮฆ๐ก โ โ(โฌ).
๐กโ๐
Then โฌ is spectrally regular. Proof. Let (โฌ, ฮ, ๐น ) be a spectral con๏ฌguration, and suppose it is spectrally winding free, i.e., ๐ฟ๐
(๐น ; ฮ) is quasinilpotent. We need to show that (โฌ, ฮ, ๐น ) is spectrally trivial, i.e., ๐น takes invertible values in โฌ on all of ฮ. The unit element in โฌ will be denoted by ๐, that in โฌ๐ by ๐๐ . Take ๐ โ ฮฉ, and put ๐๐ = ๐๐ (๐). Then ๐๐ is an idempotent in โฌ๐ and ๐๐ (๐) = ๐๐ ๐๐ (๐) = ๐๐ (๐)๐๐ for all ๐ โ โฌ. If ๐ โ โฌ is invertible in โฌ, then ๐๐ (๐) + ๐๐ โ ๐๐ is invertible in โฌ๐ , with inverse ๐๐ (๐โ1 ) + ๐๐ โ ๐๐ . Also, if ๐๐ (๐) + ๐๐ โ ๐๐ is invertible in โฌ๐ with inverse ๐๐ (๐1 ) + ๐๐ โ ๐๐ for some ๐1 โ โฌ, then ๐๐1 โ ๐ and ๐1 ๐ โ ๐ belong to Ker ๐๐ . In other words, ๐1 is an inverse of ๐ modulo the ideal Ker ๐๐ . Again let ( ๐ โ ) ฮฉ, and de๏ฌne the โฌ๐ -valued function ๐น๐ by stipulating that ๐น๐ (๐) = ๐๐ ๐น (๐) + ๐๐ โ ๐๐ . Along with ๐น , the function ๐น๐ is analytic on an open neighborhood of the closure of ฮ. As the function ๐น comes from the spectral con๏ฌguration (โฌ, ฮ, ๐น ), it takes invertible values on an open neighborhood ๐ of โฮ. Take ๐ โ (๐ . Then)๐น๐ (๐) is invertible in the Banach algebra โฌ๐ with inverse ๐น๐ (๐)โ1 = ๐๐ ๐น (๐)โ1 + ๐๐ โ ๐๐ . It follows, in particular, that (โฌ๐ , ฮ, ๐น๐ ) is a spectral con๏ฌguration. Next we compute ๐ฟ๐
(๐น๐ ; ฮ). Using that ๐น๐โฒ = ๐๐ โ ๐น โฒ , we get โซ 1 ๐ฟ๐
(๐น๐ ; ฮ) = ๐น โฒ (๐)๐น๐ (๐)โ1 ๐๐ 2๐๐ โฮ ๐ โซ ( )( ( ) ) 1 = ๐๐ ๐น โฒ (๐) ๐๐ ๐น (๐)โ1 + ๐๐ โ ๐๐ ๐๐ 2๐๐ โฮ โซ โซ ( โฒ ) ( ) ( ) 1 1 โ1 = ๐๐ ๐น (๐) ๐๐ ๐น (๐) ๐๐ + ๐๐ ๐น โฒ (๐) (๐๐ โ ๐๐ )๐๐. 2๐๐ โฮ 2๐๐ โฮ
130
H. Bart, T. Ehrhardt and B. Silbermann
( ) ( ) The last term vanishes because ๐๐ ๐น โฒ (๐) = ๐๐ ๐น โฒ (๐) ๐๐ , and we conclude that โซ ( ) ( ) 1 ๐ฟ๐
(๐น๐ ; ฮ) = ๐๐ ๐น โฒ (๐) ๐๐ ๐น (๐)โ1 ๐๐ 2๐๐ โฮ ( ) โซ 1 = ๐๐ ๐น โฒ (๐)๐น (๐)โ1 ๐๐ 2๐๐ โฮ ( ) = ๐๐ ๐ฟ๐
(๐น ; ฮ) . We proceed by proving that ๐ฟ๐
(๐น๐ ; ฮ) is quasinilpotent. Take ๐ โ โ โ {0}. As ๐ฟ๐
(๐น ; ฮ) is quasinilpotent, ๐๐ โ ๐ฟ๐
(๐น ; ฮ) is invertible in โฌ. Now ( ) ๐๐๐ โ ๐ฟ๐
(๐น๐ ; ฮ) = ๐(๐๐ โ ๐๐ ) + ๐๐ ๐๐ โ ๐ฟ๐
(๐น ; ฮ) , and the right-hand side of this identity is obviously invertible in โฌ๐ with inverse (( )โ1 ) ๐โ1 (๐๐ โ๐๐ )+๐๐ ๐๐โ๐ฟ๐
(๐น ; ฮ) . Thus ๐ is in the resolvent set of ๐ฟ๐
(๐น๐ ; ฮ), as desired. By hypothesis, the Banach algebra โฌ๐ is spectrally regular, and we have just proved that ๐ฟ๐
(๐น๐ ; ฮ) is quasinilpotent. Thus we may conclude that the spectral con๏ฌguration ๐ฟ๐
(๐น๐ ; ฮ) is spectrally trivial, i.e., the function ๐น๐ takes invertible values on ฮ. Put ๐ = ฮ โช ๐ . Then ๐ is an open neighborhood of the closure ฮ โช โฮ of the Cauchy domain ฮ and ๐น๐ takes invertible values on ๐ . Hence, by Cauchyโs integral formula, โซ 1 1 ๐น๐ (๐)โ1 = ๐น๐ (๐)โ1 ๐๐, ๐ โ ฮ. (3) 2๐๐ โฮ ๐ โ ๐ Let ๐ โ ฮ, and introduce 1 ๐บ(๐) = 2๐๐
โซ
1 ๐น (๐)โ1 ๐๐. (4) ๐ โ ๐ โฮ ( ) Then ๐บ(๐) โ โฌ and, using the identity ๐น๐ (๐)โ1 = ๐๐ ๐น (๐)โ1 + ๐๐ โ ๐๐ already obtained above, โซ ( ) ( ) 1 1 ๐๐ ๐บ(๐) = ๐๐ ๐น (๐)โ1 ๐๐ 2๐๐ โฮ ๐ โ ๐ โซ ) 1 1 ( = ๐น๐ (๐)โ1 โ (๐๐ โ ๐๐ ) ๐๐ 2๐๐ โฮ ๐ โ ๐ โซ โซ 1 1 1 1 = ๐น๐ (๐)โ1 ๐๐ โ (๐๐ โ ๐๐ )๐๐ 2๐๐ โฮ ๐ โ ๐ 2๐๐ โฮ ๐ โ ๐ = ๐น๐ (๐)โ1 โ (๐๐ โ ๐๐ ). ( ) ( ) Thus ๐น๐ (๐)โ1 = ๐๐ ๐บ(๐) + ๐๐ โ ๐๐ . As ๐น๐ (๐) = ๐๐ ๐น (๐) + ๐๐ โ ๐๐ (by de๏ฌnition), it follows that ๐บ(๐)๐น (๐) โ ๐ and ๐น (๐)๐บ(๐) โ ๐ belong to Ker ๐๐ .
Spectral Regularity and Non-commutative Gelfand Theory
131
Since ๐ โ ฮฉ was taken arbitrarily, we may conclude that, for ๐ โ ฮ as above, ๐บ(๐)๐น (๐) โ ๐ and ๐น (๐)๐บ(๐) โ ๐ are in the left-hand side of (a). Thus, taking into account the inclusion (a), ( ) ฮฆ๐ก ๐บ(๐)๐น (๐) โ ฮฆ๐ก (๐) โ โฑ(๐๐ก ) โ {๐ผ๐๐ก }, ๐ก โ ๐, (5) and, likewise,
( ) ฮฆ๐ก ๐น (๐)๐บ(๐) โ ฮฆ๐ก (๐) โ โฑ(๐๐ก ) โ {๐ผ๐๐ก },
๐ก โ ๐.
(6)
Take ๐ก โ ๐ , and put ๐๐ก = ฮฆ๐ก (๐). Then ๐๐ก is an idempotent in โฌ(๐๐ก ), in other words ๐๐ก is a projection of ๐๐ก , and ฮฆ๐ (๐) = ๐๐ก ฮฆ๐ก (๐) = ฮฆ๐ก (๐)๐๐ก for all ๐ โ โฌ. If ๐ โ โฌ is invertible in โฌ, then ฮฆ๐ก (๐) + ๐ผ๐๐ก โ ๐๐ก is invertible in โฌ(๐๐ก ), with inverse ฮฆ๐ก (๐โ1 ) + ๐ผ๐๐ก โ ๐๐ก . Also, if ฮฆ๐ก (๐) + ๐ผ๐๐ก โ ๐๐ก is invertible in โฌ(๐๐ก ) with inverse ฮฆ๐ก (๐1 ) + ๐ผ๐๐ก โ ๐๐ก for some ๐1 โ โฌ, then ๐๐1 โ ๐ and ๐1 ๐ โ ๐ belong to Ker ฮฆ๐ก . In other words, ๐1 is an inverse of ๐ modulo the ideal Ker ฮฆ๐ก . Again, let ๐ก โ ๐ , and introduce the โฌ(๐๐ก )-valued function ๐นห๐ก by putting ( ) ( ) ห ๐น๐ก (๐) = ฮฆ๐ก ๐น (๐) + ๐ผ๐๐ก โ ๐๐ก . Arguing as above, we see that โฌ(๐๐ก ), ฮ, ๐นห๐ก is ( ) a spectral con๏ฌguration. Also ๐ฟ๐
(๐นห๐ก ; ฮ) = ฮฆ๐ก ๐ฟ๐
(๐น ; ฮ) , and it follows that ๐ฟ๐
(๐นห๐ก ; ฮ) is quasinilpotent. Next observe that ( ( ) ) ( ( ) )( ( ) ) ฮฆ๐ก ๐บ(๐) + ๐ผ๐๐ก โ ๐๐ก ๐นห๐ก (๐) = ฮฆ๐ก ๐บ(๐) + ๐ผ๐๐ก โ ๐๐ก ฮฆ๐ก ๐น (๐) + ๐ผ๐๐ก โ ๐๐ก ( ( ) ) = ฮฆ๐ก ๐บ(๐)๐น (๐) โ ฮฆ๐ก (๐) + ๐ผ๐๐ก , ( ( ) ) and so ฮฆ๐ก ๐บ(๐) +๐ผ๐๐ก โ๐๐ก ๐นห๐ก (๐) โ โฑ(๐๐ก ) by (5). Similarly, by taking into account ( ( ) ) (6), we get ๐นห๐ก (๐) ฮฆ๐ก ๐บ(๐) + ๐ผ๐๐ก โ ๐๐ก โ โฑ(๐๐ก ). But then ๐นห๐ก (๐) is a Fredholm operator, and) we can apply Theorem 2.1 to see that the spectral con๏ฌguration ( โฌ(๐๐ก ), ฮ, ๐นห๐ก is spectrally trivial. Analogous to (3), we have โซ 1 1 ห ๐นห๐ก (๐)โ1 = ๐น๐ก (๐)โ1 ๐๐, ๐ โ ฮ, 2๐๐ โฮ ๐ โ ๐ ( ) and, by the same (sort of) reasoning as used before, ๐นห๐ก (๐)โ1 = ฮฆ๐ก ๐บ(๐) + ๐ผ๐๐ก โ ๐๐ก . Since ๐น๐ก (๐) = ฮฆ๐ก ๐น (๐) + ๐ผ๐๐ก โ ๐๐ก (by de๏ฌnition), it follows that ๐บ(๐)๐น (๐) โ ๐ and ๐น (๐)๐บ(๐) โ ๐ belong to Ker ฮฆ๐ก . As ๐ก โ ๐ was taken arbitrarily, we may conclude that ๐บ(๐)๐น (๐) โ ๐ and ๐น (๐)๐บ(๐) โ ๐ are in the left-hand side of (b) which, by assumption, is a subset of the radical of โฌ. So ๐บ(๐)๐น (๐) and ๐น (๐)๐บ(๐) are invertible. But then ๐น (๐) is both left and right invertible, hence invertible, as desired. The inverse of ๐น (๐) is ๐บ(๐) given by (4). โก Before drawing consequences from Theorem 3.1, we present some remarks on the conditionsโฉ (a) and (b) in the theorem. First we note that (a) in Theorem 3.1 is ful๏ฌlled when ๐โฮฉ Ker ๐๐ โ โ(โฌ). To see this, it is su๏ฌcient to prove that, with {ฮฆ๐ก : โฌ โ โฌ(๐๐ก)}๐กโ๐ [ of continuous ]homomorphisms as in Theorem 3.1, โฉ a family โ1 we have โ(โฌ) โ ฮฆ โฑ(๐๐ก ) โ {๐ผ๐๐ก } . The argument is as follows. Write ๐กโ๐ ๐ก
132
H. Bart, T. Ehrhardt and B. Silbermann
๐๐ก = ฮฆ๐ก (๐) with ๐ the unit element in โฌ. Then ๐๐ก is an idempotent in โฌ(๐๐ก ). Clearly ฮฆ๐ก (๐)๐๐ก = ๐๐ก ฮฆ๐ก (๐) = ฮฆ๐ก (๐) for all ๐ โ โฌ. Take ๐ โ โ(โฌ). Then ๐ + ๐ is invertible in โฌ, say with inverse ๐ . A straightforward computation now yields (ฮฆ๐ก (๐ + ๐) + ๐ผ๐๐ก โ ๐๐ก )(ฮฆ๐ก (๐ ) + ๐ผ๐๐ก โ ๐๐ก ) = ๐ผ๐๐ก , (ฮฆ๐ก (๐ ) + ๐ผ๐๐ก โ ๐๐ก )(ฮฆ๐ก (๐ + ๐) + ๐ผ๐๐ก โ ๐๐ก ) = ๐ผ๐๐ก . Thus, ฮฆ๐ก (๐) + ๐ผ๐๐ก = ฮฆ๐ก (๐ + ๐) + (๐ผ๐๐ก โ ๐๐ก ) is invertible in โฌ(๐๐ก ). Hence ฮฆ๐ก (๐) โ โฑ (๐๐ก ) โ {๐ผ๐๐ก }, as desired. Next we observe that (b) in Theorem 3.1 cannot be satis๏ฌed by the empty index family ๐ . Indeed, if so, the Banach algebra โฌ would coincide with its radical, and this can only happen in the (excluded) case when โฌ is trivial. Finally, in contrast to what we have for (b), it is possible to have (a) satis๏ฌed by the empty index family ฮฉ. The underlying fact (not di๏ฌcult to establish) is that the inclusion โฉ โฌ โ ฮฆโ1 (7) ๐ก [โฑ (๐๐ก ) โ {๐ผ๐๐ก }] ๐กโ๐
is satis๏ฌed if and only if for all ๐ก โ ๐ , the projection ๐๐ก = ฮฆ๐ก (๐) : ๐๐ก โ ๐๐ก has ๏ฌnite rank. So (7), which is trivially ful๏ฌlled when the index set ฮฉ is empty, basically means that the homomorphisms ฮฆ๐ก are (or rather can be identi๏ฌed with) matrix representations. Later on we will use two speci๏ฌc forms of Theorem 3.1. We give them as corollaries. In the ๏ฌrst โฌ is a closed subalgebra of a Banach algebra of the type โฌ(๐), unital but with unit element not necessarily equal to ๐ผ๐ . Corollary 3.2. Let ๐ be a nontrivial Banach space, and let โฌ be a closed subalgebra of โฌ(๐). For ๐ in an index set ฮฉ, let โฌ๐ be a spectrally regular Banach algebra, and let ๐๐ : โฌ โ โฌ๐ be a continuous homomorphism. Suppose โฉ Ker ๐๐ โ โฑ (๐) โ {๐ผ๐ }. (8) ๐โฮฉ
Then โฌ is spectrally regular. Proof. Take ๐ = {0}, put ๐0 = ๐, and let ฮฆ0 : โฌ โ โฌ(๐0 ) be the identical embedding of โฌ into โฌ(๐0 ). Then Ker ฮฆ0 = {0}, hence (b) in Theorem 3.1 is trivially ful๏ฌlled. From (8) it is obvious that (a) in Theorem 3.1 is satis๏ฌed too. โก Corollary 3.3. Let โฌ be a unital Banach algebra. For ๐ in an index set ฮฉ, let โฌ๐ be a spectrally regular Banach algebra, and let ๐๐ : โฌ โ โฌ๐ be a continuous homomorphism. Suppose โฉ Ker ๐๐ โ โ(โฌ). (9) Then โฌ is spectrally regular.
๐โฮฉ
Proof. Let ๐ be โฌ considered as a Banach space only. Then โฌ can be identi๏ฌed with a Banach subalgebra of โฌ(๐). The standard argument for this uses the left regular representation ฮจ of โฌ into โฌ(๐) de๏ฌned by ฮจ(๐)(๐ฅ) = ๐๐ฅ, ๐ฅ โ ๐, ๐ โ โฌ.
Spectral Regularity and Non-commutative Gelfand Theory
133
Having this identi๏ฌcation in mind, we need to prove that (9) implies (8). Take ๐ in the left-hand side of (9). Then ๐ โ โ(โฌ), and so ๐ + ๐ is invertible in โฌ. Here ๐ is the unit element in โฌ. Now under the left regular representation ฮจ, this unit element is identi๏ฌed with ๐ผ๐ . So ๐ + ๐ผ๐ is invertible in โฌ, hence invertible in โฌ(๐). It follows that ๐ โ โฑ (๐) โ {๐ผ๐ } as desired. โก A family {๐๐ : โฌ โ โฌ๐ }๐โฮฉ of Banach algebra homomorphisms for which (9) holds will be called radical-separating. This terminology is justi๏ฌed by the fact that the inclusion (9) holds if and only if the family {๐๐ : โฌ โ โฌ๐ }๐โฮฉ separates the points of โฌ modulo the radical of โฌ. If the stronger โฉ condition is satis๏ฌed that the family separates the points of โฌ or, equivalently, ๐โฮฉ Ker ๐๐ = {0}, we call the family separating. When the underlying Banach algebra โฌ is semisimple (i.e., its radical is trivial), the two concepts obviously amount to the same. The special situation where the โtest algebrasโ โฌ๐ are semisimple and the Banach algebra homomorphisms ๐๐ are surjective is of interest too (see Subsection 4.1, Lemma 4.5 and below). Indeed, in that case the family {๐๐ : โฌ โ โฌ๐ }๐โฮฉ is โฉ radical-separating if and only if the inclusion (9) is in fact an equality, i.e., ๐โฮฉ Ker ๐๐ = โ(โฌ). This is immediate from the following straightforward observation. If ๐ : โฌ โ ๐ is a surjective unital Banach algebra homomorphism, then ๐ maps โ(โฌ) into โ(๐); so when ๐ is semisimple, it ensues that โ(โฌ) โ Ker ๐. We now make a connection with material presented earlier in Section 2. Recall that a family {๐๐ : โฌ โ โฌ๐ }๐โฮฉ of continuous unital Banach algebra homomorphisms is said to be su๏ฌcient when an element ๐ โ โฌ is invertible in โฌ if (and only if) ๐๐ (๐) is invertible in โฌ๐ for all ๐ โ ฮฉ. Besides su๏ฌcient families, the books [RRS] and [RSS] also feature so-called weakly su๏ฌcient families. Inspired by the de๏ฌnitions given there, we introduce the notion of a partially weakly su๏ฌcient family of homomorphisms. Write โฅ.โฅ๐ for the norm in โฌ๐ and ๐๐ for the unit element in โฌ๐ . The family {๐๐ : โฌ โ โฌ๐ }๐โฮฉ of continuous unital Banach algebra homomorphisms is called partially weakly su๏ฌcient, or p.w. su๏ฌcient for short, provided that (a) sup๐โฮฉ โฅ๐๐ โฅ๐ < โ (recall from the last paragraph of the introduction that โฅ๐๐ โฅ๐ need not be equal to one), and (b) an element ๐ โ โฌ is invertible in โฌ if ๐๐ (๐) is invertible in โฌ๐ for all ๐ โ ฮฉ and sup๐โฮฉ โฅ๐๐ (๐)โ1 โฅ๐ < โ. In de๏ฌnitions of this type, conditions such as (b) are usually of the โif and only ifโ type. The fact that we do not impose this more restrictive requirement here is the reason for the use of the term โpartiallyโ in our terminology. A su๏ฌcient family of Banach algebra homomorphisms {๐๐ : โฌ โ โฌ๐ }๐โฮฉ is p.w. su๏ฌcient in the sense that it can be turned into a p.w. su๏ฌcient family by renorming the Banach algebras โฌ๐ with an appropriate equivalent norm. Indeed, for ๐ โ ฮฉ, just choose an equivalent Banach algebra norm for which the unit element ๐๐ in โฌ๐ has norm one. It is a standard fact from Banach algebra theory that this can be done.
134
H. Bart, T. Ehrhardt and B. Silbermann
In Section 2.2.5 of [RSS], a de๏ฌnition of weak su๏ฌciency is given which in its context, namely that of ๐ถ โ -algebras and โ -homomorphisms, amounts to the same as p.w. su๏ฌciency. Theorem 2.2.10 of [RSS] shows that the families in question are separating. In the general (non ๐ถ โ ) Banach algebra setting one has to be content with a weaker conclusion. Proposition 3.4. Let โฌ be a unital Banach algebra, and let {๐๐ : โฌ โ โฌ๐ }๐โฮฉ be a family of unital Banach algebra homomorphisms. If the family {๐๐ }๐โฮฉ is p.w. su๏ฌcient or su๏ฌcient, then it is radical-separating. Proof. Write ๐ for the unit element in โฌ, and let ๐๐ and โฅ.โฅ๐ stand for the unit element and norm in โฌ๐ , respectively. First, suppose that the family {๐๐ }๐โฮฉ is p.w. su๏ฌcient, thus, in particular, sup๐โฮฉ โฅ๐๐ โฅ๐ < โ. Take ๐ฅ in the left-hand side of (9). Then, for ๐ โ ฮฉ and ๐ โ โฌ, we have ๐๐ (๐ฅ) = 0 and ๐๐ (๐๐ฅ + ๐) = ๐๐ (๐)๐๐ (๐ฅ) + ๐๐ (๐) = ๐๐ (๐) = ๐๐ . So ๐๐ (๐๐ฅ+๐) is invertible in โฌ๐ and sup๐โฮฉ โฅ๐๐ (๐๐ฅ+๐)โ1 โฅ๐ = sup๐โฮฉ โฅ๐๐ โฅ๐ < โ. It follows that ๐๐ฅ + ๐ is invertible in โฌ. Similarly ๐ฅ๐ + ๐ is invertible in โฌ, and we conclude that ๐ฅ โ โ(โฌ). When the family {๐๐ }๐โฮฉ is su๏ฌcient instead of p.w. su๏ฌcient, the argument is even simpler (and left to the reader). One can also argue that, in the sense explained above, su๏ฌciency implies p.w. su๏ฌciency. โก Corollary 3.5. Let โฌ be a unital Banach algebra. For ๐ in an index set ฮฉ, let โฌ๐ be a spectrally regular Banach algebra and let ๐๐ : โฌ โ โฌ๐ be a continuous unital homomorphism. If the family {๐๐ }๐โฮฉ is su๏ฌcient or p.w. su๏ฌcient, then โฌ is spectrally regular. Proof. Combine Corollary 3.3 and Proposition 3.4.
โก
In connection with Corollary 3.5 note that the existence of a p.w. su๏ฌcient family can often be much more easily established than that of a su๏ฌcient family. For an example, consider the Banach algebra โโ . Comparing Corollaries 3.3 and 3.2, one is confronted with a striking di๏ฌerence between the conditions (9) and (8). In (9), both terms in the inclusion are ideals, and in fact ideals in one and the same given Banach algebra โฌ. In (8), however, the left-hand side is an ideal in a Banach subalgebra โฌ of the underlying Banach algebra โฌ(๐), whereas the right-hand side is a shifted semigroup of elements in โฌ(๐). Here are some comments meant to elucidate the situation. First let us look at (9). Evidently it follows from (9) that โฉ Ker ๐๐ โ ๐ข(โฌ) โ {๐}, (10) ๐โฮฉ
where (again) ๐ denotes the unit element in โฌ, and ๐ข(โฌ) stands for the group of invertible elements in โฌ. However, as the left-hand side of (10) is an ideal, (10) in turn implies (9). Thus (9) and (10) amount to the same.
Spectral Regularity and Non-commutative Gelfand Theory
135
Turning now to (8), observe that there is a certain analogy between the righthand sides of (8) and (10). Indeed, the set โฑ (๐) consists of the elements of โฌ(๐) that are invertible modulo the closed two-sided ideal ๐ฆ(๐) of the compact operators on ๐. On the other hand, the right-hand side of (8) need not be contained in โฌ. In fact, the circumstance that โฌ is embedded in the (generally) larger algebra โฌ(๐) of all bounded linear operators on ๐ is a key element in the proof of Corollary 3.2. Given the role the ideal ๐ฆ(๐) is playing in the background, one may wonder whether (8) can be reformulated in a form resembling (9), so with an ideal in the right-hand side of the inclusion. This is possible to the extent that Corollary[3.2( remains true)]when in (8) the right-hand side of the inclusion is replaced by ๐
โ1 โ โฌ(๐)/๐ฆ(๐) with ๐
being the canonical mapping of โฌ(๐) onto the Calkin [ ( )] algebra โฌ(๐)/๐ฆ(๐). In fact ๐ฆ(๐) โ ๐
โ1 โ โฌ(๐)/๐ฆ(๐) โ โฑ(๐) โ {๐ผ๐ }, and therefore each of the two inclusions โฉ [ ( )] Ker ๐๐ โ ๐
โ1 โ โฌ(๐)/๐ฆ(๐) , (11) ๐โฮฉ
โฉ
Ker ๐๐ โ ๐ฆ(๐),
(12)
๐โฮฉ
is su๏ฌcient for (8) to hold. Now (11) and (12) bear some resemblance to (9). However, unlike (10) and (9) which simply amount to the same, the relationship between the conditions (8) and (11), and the relationship between (8) and (12), are not so clear. The reason is that the set featuring in the left-hand sides of (8), (11) and (12), although it is an ideal in โฌ, need not be an ideal in โฌ(๐). The upshot of this discussion is that, although several modi๏ฌcations of Corollary 3.2 are possible, the formulation given above seems to be the optimal one. For completeness we add that in the previous two paragraphs, the ideal of the compact operators may be replaced by that of the strictly singular operators. For material on strictly singular operators, see Section III.2 in [Go] or Section 4.5 in [AA]. The new criterion for spectral regularity (Theorem 3.1) and its corollaries (Corollary 3.3 and Corollary 3.2) can be employed e๏ฌectively in so far as there is an adequate supply of spectrally regular test algebras โฌ๐ . Some classes of spectrally regular Banach algebras are described in Section 2. In this paper, the test algebras mostly employed are matrix algebras while an occasional use is made of an algebra which is not of that type. One of the classes of spectrally regular Banach algebras mentioned in Section 2 is that of the PI-algebras. However, the property of being PI is often di๏ฌcult or even impossible to check. On the other hand, there are many Banach algebras which become PI, hence spectrally regular, after factoring out the radical. In this connection it is fortunate that a Banach algebra โฌ is spectrally regular if and only โฌ/โ(โฌ) is (see Theorem 4.2 below). The Banach algebras in question are therefore suitable as test algebras too.
136
H. Bart, T. Ehrhardt and B. Silbermann
4. Applications In this section we present applications of Corollaries 3.3 and 3.2. In particular we establish the spectral regularity of certain Banach algebras for which this was hitherto impossible. Along the way some new results based on the older methods are obtained too. The material is divided into ๏ฌve subsections. 4.1. Subalgebras and quotients We begin with a special case of Corollary 3.3, worth to be stated in its own right. Corollary 4.1. Let โฌ and ๐ be unital Banach algebras, and let ฮฆ : โฌ โ ๐ be a continuous Banach algebra homomorphism. Assume Ker ฮฆ โ โ(โฌ) and ๐ is spectrally regular. Then โฌ is spectrally regular too. Proof. In Corollary 3.3, take for ฮฉ the singleton set {0}, for โฌ0 the Banach algebra ๐ (spectrally regular by assumption), and for ๐0 the homomorphism ฮฆ. โก The situation where the Banach algebra homomorphism ฮฆ in Corollary 4.1 happens to be injective is of particular interest. One can then view โฌ as a continuously embedded subalgebra of ๐. Thus, in particular, Corollary 4.1 implies that each closed unital subalgebra ๐ of a spectrally regular Banach algebra โฌ (where ๐ need not have the same unit element as โฌ) is spectrally regular again. Another immediate consequence is that a unital Banach algebra โฌ is spectrally regular provided it is ๏ฌnite dimensional. Indeed, if ๐ = dim โฌ, then โฌ can be identi๏ฌed with a Banach subalgebra of โ๐ร๐ . For this, use the left regular representation of โฌ into โฌ(๐), where ๐ is the ๐-dimensional Banach space obtained by considering โฌ as a Banach space only (cf. the proof of Corollary 3.3). Next we turn to quotient algebras. Here the situation is more involved. In fact there are two issues. First, is a quotient of a spectrally regular Banach algebra spectrally regular again? Second, if a quotient is spectrally regular, does it follow that the underlying algebra is spectrally regular too? As concerns the ๏ฌrst issue, in sharp contrast to what has just been observed for subalgebras, a quotient algebra of a spectrally regular Banach algebra need not be spectrally regular. The counterexample that we have uses elements developed in Subsection 4.2 below. For that reason it will be given there. Note that with this we also have an example of surjective Banach algebra homomorphism ฮจ : ๐ โ โฌ such that ๐ is spectrally regular while โฌ is not. Thus what might be called the dual of Corollary 4.1, taken with ฮฆ injective, does not hold. For the second issue, as might be expected, the answer is generally negative too. Here is a counterexample. Let โ stand for the set of positive integers, and consider the Banach space โ2 ({0} โช โ). Write it as a direct sum โ โ โ2 , where โ2 = โ2 (โ), and take for โฌ the Banach subalgebra of โฌ(โ โ โ2 ) consisting of all bounded linear operators from โ โ โ2 into โ โ โ2 having the diagonal form [ ] ๐ผ 0 , ๐ผ โ โ, ๐ โ โฌ(โ2 ). 0 ๐
Spectral Regularity and Non-commutative Gelfand Theory
137
Now let ๐ฅ be the set of all operators in โฌ of the type [ ] 0 0 , ๐ โ โฌ(โ2 ). 0 ๐ Then ๐ฅ is a closed two-sided ideal in โฌ. As โฌ/๐ฅ is isomorphic to โ, the quotient algebra โฌ/๐ฅ is spectrally regular. However, โฌ is not. In fact, along with โฌ(โ2 ), the Banach algebra โฌ features a nontrivial zero sum of idempotents, and this rules out the property of being spectrally regular (cf. the third paragraph of the introduction). Another way to see that โฌ is not spectrally regular is by ๏ฌrst noting that โฌ(โ2 ) can be viewed as a closed unital subalgebra of โฌ, and then taking into account the remark made after the proof of Corollary 4.1. In view of Corollary 4.3 below, we emphasize that in this counterexample the quotient algebra โฌ/๐ฅ is ๏ฌnite (in fact one) dimensional but that ๐ฅ is not contained in the radical of โฌ. In the remainder of this subsection we focus on the special situation where the ideal ๐ฅ which is factored out is contained in the radical of the underlying Banach algebra โฌ. Observe that in this situation invertibility modulo ๐ฅ and invertibility in โฌ amount to the same. This will be used several times later on. Theorem 4.2. Let โฌ be a unital Banach algebra, and let ๐ฅ be a closed two-sided ideal in โฌ which is contained in the radical of โฌ. Then โฌ is spectrally regular if and only if the quotient algebra โฌ/๐ฅ has this property. In particular, โฌ is spectrally regular if and only if so is โฌ/โ(โฌ). Proof. First suppose ๐ต/๐ฅ is spectrally regular. Take for ฮฉ the singleton set {0}, for โฌ0 the Banach algebra โฌ/๐ฅ , and for ๐0 the canonical mapping from โฌ onto โฌ0 . Then Ker ๐0 = ๐ฅ . By assumption ๐ฅ โ โ(โฌ). Hence the singleton family {๐0 } is radical-separating, and the desired result follows from Corollary 3.3. (Alternatively, one can use Corollary 3.5 after noting that the family {๐0 } is su๏ฌcient.) This proves the โif partโ of the theorem. Next we turn to the โonly if partโ and assume that โฌ is spectrally regular. Let (โฌ/๐ฅ , ฮ, ๐นห) be a spectral con๏ฌguration and suppose it is spectrally winding free. It must be shown that (โฌ/๐ฅ , ฮ, ๐นห ) is spectrally trivial. For this we shall use that the โฌ/๐ฅ -valued analytic functions can be lifted to โฌ. In other words, they can be written as the composition of an analytic โฌ-valued function with ๐
, the canonical mapping of โฌ onto the quotient space โฌ/๐ฅ . That this is indeed possible can be seen from the proof of Theorem 1a in [Gra] which is based on Grothendieckโs work on topological tensor products [Gro]; see also Section 3.0 in [ZKKP], [Ka], and Section 6.4 in [GL]. In the concrete situation that we have here, one can also proceed as follows, employing only lifting of continuous functions. Denote the domain of the function ๐นห by ๐ . Then ๐ is an open subset of the complex plane containing the closure ฮ of ฮ. Now let ฮ1 be another bounded Cauchy domain such that ฮ โ ฮ1 โ ฮ1 โ ๐ . Write ๐ห1 for the restriction of the function ๐นห to โฮ1 . Then ๐ห1 : โฮ1 โ โฌ/๐ฅ is a continuous function. There exists a continuous lifting of ๐ห1 , that is a function
138
H. Bart, T. Ehrhardt and B. Silbermann
๐1 : โฮ1 โ โฌ/๐ฅ such that ๐ห1 = ๐
โ ๐1 (see, e.g., [ZKKP], Section 1.0). De๏ฌne the function ๐น : ฮ1 โ โฌ by โซ 1 1 ๐น (๐) = ๐1 (๐)๐๐, ๐ โ ฮ1 . 2๐๐ โฮ1 ๐ โ ๐ Then ๐น is analytic on ฮ1 . Also, for ๐ โ ฮ1 , โซ 1 1 (๐
โ ๐น )(๐) = (๐
โ ๐1 )(๐)๐๐, 2๐๐ โฮ1 ๐ โ ๐ โซ 1 1 ห = ๐1 (๐)๐๐, 2๐๐ โฮ1 ๐ โ ๐ โซ 1 1 ห = ๐น (๐)๐๐, 2๐๐ โฮ1 ๐ โ ๐ and the latter expression is equal to ๐นห(๐) by the Cauchy integral formula. With ๐น we form a new spectral con๏ฌguration (โฌ, ฮ, ๐น ). That this is a( spectral ) con๏ฌguration indeed, can been as follows. For ๐ in โฮ we have that ๐
๐น (๐) is invertible in โฌห = โฌ/๐ฅ , so ๐น (๐) is invertible modulo the ideal ๐ฅ . But then, making use of on observation made earlier, ๐น (๐) is invertible in โฌ. ( ) Clearly ๐
๐ฟ๐
(๐น ; ฮ) = ๐ฟ๐
(๐นห ; ฮ), and the latter is quasinilpotent. Take ๐ in โ โ {0}. Then ๐๐
(๐) โ ๐ฟ๐
(๐นห ; ฮ) is invertible in โฌ/๐ฅ . Here ๐ stands for the unit element in โฌ. Now ๐
(๐๐ โ ๐ฟ๐
(๐น ; ฮ)) = ๐๐
(๐) โ ๐ฟ๐
(๐นห ; ฮ). Thus ๐๐ โ ๐ฟ๐
(๐น ; ฮ) is invertible modulo ๐ฅ , hence invertible in โฌ. Thus we have proved that ๐ฟ๐
(๐น ; ฮ) is quasinilpotent, i.e., the spectral con๏ฌguration (โฌ, ฮ, ๐น ) is spectrally winding free. As โฌ is assumed to be spectrally regular, we may conclude that (โฌ, ฮ, ๐น ) is ห ฮ, ๐นห ). spectrally trivial. But then so is the spectral con๏ฌguration (โฌ, โก The following result is a simple consequence of Theorem 4.2 and the remark made in the (second part of the) paragraph after the proof of Corollary 4.1 (see also the counterexample presented above). Corollary 4.3. Let โฌ be a unital Banach algebra, and let ๐ฅ be a closed two-sided ideal in โฌ which is contained in the radical of โฌ. Suppose the quotient algebra โฌ/๐ฅ is ๏ฌnite dimensional. Then โฌ is spectrally regular. In particular, โฌ is spectrally regular whenever โ(โฌ) has ๏ฌnite codimension in โฌ. In Section 2, the paragraph directly following the proof of Theorem 2.1, it was indicated that it is possible to work with a somewhat weaker form of spectral regularity than the one adopted here (cf. the ๏ฌrst four paragraphs of the introduction). For this weaker version (involving vanishing logarithmic residues instead of quasinilpotent ones), we have not been able to prove the โonly if partโ of Theorem 4.2; neither do we have a counterexample showing that it need not hold. Now, instead of looking at spectral regularity, we consider the (stronger) property of possessing a su๏ฌcient family matrix representations. We have the following analogue of Theorem 4.2.
Spectral Regularity and Non-commutative Gelfand Theory
139
Proposition 4.4. Let โฌ be a unital Banach algebra, and let ๐ฅ be a closed two-sided ideal in โฌ which is contained in the radical of โฌ. Then โฌ possesses a su๏ฌcient family of matrix representations if and only if so does the quotient algebra โฌ/๐ฅ . In particular, โฌ possesses a su๏ฌcient family of matrix representations if and only if this is the case for โฌ/โ(โฌ). In our review of known material presented in Section 2, we mentioned that the last statement in Proposition 4.4 is true when one works with su๏ฌcient families of matrix representations having the additional property of being of ๏ฌnite order. This additional property is not required here. To prove Proposition 4.4, need the following lemma. Lemma 4.5. Let โฌ be a unital Banach algebra. Then โฌ possesses a su๏ฌcient family of matrix representations if and only if โฌ possesses a su๏ฌcient family of surjective matrix representations. Proof. The โif partโ of the proposition is trivial. So we concentrate on the โonly if partโ. Let ๐ : โฌ โ โ๐ร๐ be a unital matrix representation of โฌ. It su๏ฌces to show that there exist a positive integer ๐, positive integers ๐1 , . . . , ๐๐ , and surjective unital matrix representations ๐๐ : โฌ โ โ๐๐ ร๐๐ ,
๐ = 1, . . . , ๐,
with the following properties: for ๐ โ โฌ, the matrix ๐(๐) is invertible in โ๐ร๐ if and only if ๐๐ (๐) is invertible in โ๐๐ ร๐๐ , ๐ = 1, . . . , ๐. The argument runs as follows. If the matrix representation ๐ itself is surjective, there is nothing to prove (case ๐ = 1). Assume it is not, so ๐[โฌ] is a proper subalgebra of โ๐ร๐ . Applying Burnsideโs Theorem (cf., [LR]), we see that ๐[โฌ] has a nontrivial invariant subspace, i.e., there is a nontrivial subspace ๐ of โ๐ร๐ such that ๐(๐)[๐ ] is contained in ๐ for all ๐ in โฌ. But then there exist an invertible ๐ ร ๐ matrix ๐, positive integers ๐โ and ๐+ , a unital matrix representation ๐โ : โฌ โ โ๐โ ร๐โ and a unital matrix representation ๐+ : โฌ โ โ๐+ ร๐+ such that ๐ has the form ] [ ๐โ (๐) โ โ1 ๐, ๐ โ โฌ. ๐(๐) = ๐ 0 ๐+ (๐) Clearly ๐(๐) is invertible in โ๐ร๐ if and only if ๐โ (๐) is invertible in โ๐โ ร๐โ and ๐+ (๐) is invertible in โ๐+ ร๐+ . If ๐โ and ๐+ are both surjective we are done (case ๐ = 2); if not we can again apply Burnsideโs Theorem and decompose further. This process terminates after at most ๐ steps. A completely rigorous argument can be given using induction. โก Proof of Proposition 4.4. To establish the โonly if partโ of the proposition, we may assume that โฌ possesses a su๏ฌcient family {๐๐ : โฌ โ โ๐๐ ร๐๐ }๐โฮฉ of surjective matrix representations (see Lemma 4.5). Take ๐ โ ฮฉ. As is well known, โ๐๐ ร๐๐ is (semi)simple. Thus the remark made in the second paragraph after the proof of Corollary 3.3 applies. It gives โ(โฌ) โ Ker ๐๐ . But then ๐ฅ โ Ker ๐๐ and ๐๐
140
H. Bart, T. Ehrhardt and B. Silbermann
induces a continuous unital Banach algebra homomorphism ฮฆ๐ from โฌ/๐ฅ into โ๐๐ ร๐๐ which satis๏ฌes ๐๐ = ฮฆ๐ โ ๐
. Here ๐
is the canonical homomorphism ๐๐ ร๐๐ of โฌ }๐โฮฉ is su๏ฌcient. Indeed, if ( onto ) โฌ/๐ฅ . The family {ฮฆ๐ : โฌ/๐ฅ โ โ ฮฆ๐ ๐
(๐) = ๐๐ (๐) is invertible for each ๐ โ ฮฉ, then ๐ is invertible in โฌ, hence ๐
(๐) is invertible in โฌ/๐ฅ . Next suppose that โฌ/๐ฅ possesses a su๏ฌcient family of matrix representations, say {ฮฆ๐ : โฌ/๐ฅ โ โ๐๐ ร๐๐ }๐โฮฉ . With ๐
as above, put ๐๐ (= ฮฆ๐) โ ๐
. Take ๐ โ โฌ, and assume ๐๐ (๐) is invertible for each ๐ โ ฮฉ. Then ฮฆ๐ ๐
(๐) is invertible for each ๐ โ ฮฉ, and we may conclude that ๐
(๐) is invertible in โฌ/๐ฅ . In other words, ๐ is invertible modulo the ideal ๐ฅ . As this ideal is contained in the radical of โฌ, it follows that ๐ is invertible in โฌ. Thus {๐๐ : โฌ โ โ๐๐ ร๐๐ }๐โฮฉ is a su๏ฌcient family of matrix representations, and the โif partโ of Proposition 4.4 has been proved. โก One may ask whether in Proposition 4.4 su๏ฌcient families can be replaced by (radical-)separating families. If โฌ/๐ฅ has a radical-separating family of matrix representations, then so has โฌ. The proof is a slight modi๏ฌcation of the argument given above to prove the โif partโ of Proposition 4.4 and employs the fact that โ(โฌ/๐ฅ ) = ๐
[โ(โฌ)], where ๐
is the canonical homomorphism from โฌ onto โฌ/๐ฅ . How about the converse? Here the situation is less clear. If โฌ possesses a (radical-) separating family of surjective matrix representations, then this is also the case for โฌ/๐ฅ . The proof is analogous to the reasoning presented above to prove the โonly if partโ of Proposition 4.4 and again employs the fact that โ(โฌ/๐ฅ ) = ๐
[โ(โฌ)]. However, we do not know whether the existence of a (radical-)separating family of matrix representations for โฌ implies the existence of such a family consisting of surjective homomorphisms. In other words, we do not know whether there is an analogue of Lemma 4.5 for families that are radical-separating (or even separating) instead of su๏ฌcient. Our conjecture is: there is not. Thus the question whether the existence of a (radical-)separating family of matrix representations for โฌ generally implies the existence of such a family for โฌ/๐ฅ is open. 4.2. Algebras of โโ -type Let ๐ be a nonempty set, and let B = {โฌ๐ก }๐กโ๐ be a family of unital Banach algebras for which it is assumed that sup๐กโ๐ โฅ๐๐ก โฅ๐ก < โ. Here ๐๐ก stands for the unit element in โฌ๐ก and โฅ.โฅ๐ก denotes the norm on โฌ๐ก . Write โBโ for the โโ -direct B product of the family โ B (cf. [P], Subsection 1.3.1). Thus โโ consists of all ๐ in the Cartesian product ๐กโ๐ โฌ๐ก such that โฃโฃโฃ๐ โฃโฃโฃ = sup โฅ๐ (๐ก)โฅ๐ก < โ. ๐กโ๐
With the operations of addition, scalar multiplication and multiplication de๏ฌned pointwise, and with โฃโฃโฃ.โฃโฃโฃ as norm, โBโ is a unital Banach algebra.
Spectral Regularity and Non-commutative Gelfand Theory
141
From Theorem 4.1 in [BES7] we know that (even when the constituting algebras โฌ๐ก are matrix algebras) โBโ need not possess a su๏ฌcient family of matrix representations. So, in general, the road to establishing spectral regularity for Banach algebras of the type โBโ via Theorem 4.1 in [BES2] is blocked, and Theorem 4.2 in [BES2], the other Gelfand type criterion in [BES2], does not seem to work either. Corollary 3.3 helps out in a surprisingly simple way. Theorem 4.6. Let ๐ be a nonempty set, and let B = {โฌ๐ก}๐กโ๐ be a family of unital Banach algebras. Then โB โ is spectrally regular if and only if so are all the Banach algebras โฌ๐ก , ๐ก โ ๐ . Proof. The family of point evaluations on โBโ is obviously separating the points of โBโ , so Corollary 3.3 gives the โif partโ of the theorem. The โonly if partโ is immediate from the remark made after the proof of Corollary 4.1. โก The โif partโ of Theorem 4.6 can also be obtained from Corollary 3.5. Indeed, the family of point evaluations on โBโ is easily seen to be p.w. su๏ฌcient. In general it is not su๏ฌcient, as can be seen by looking at โโ . Specializing to the case where the Banach algebras โฌ๐ก all coincide with a single Banach algebra โฌ, we write โโ (๐ ; โฌ) for the Banach algebra of all bounded functions from ๐ into โฌ, provided with the pointwise algebraic operations and the supremum norm. Corollary 4.7. Let ๐ be a nonempty set, and let โฌ be a unital Banach algebra. Then โโ (๐ ; โฌ) is spectrally regular if and only if so is โฌ. Combining Corollaries 4.7 and 4.1, one readily gets a variety of results. For instance, if ๐ is a compact topological space and โฌ is a spectrally regular Banach algebra, then the Banach algebra ๐(๐ ; โฌ) of all continuous functions from ๐ into โฌ (provided with the pointwise algebraic operations and the supremum norm) is spectrally regular. Another example is ๐๐ซ(โ; โฌ), the Banach algebra of continuous almost periodic functions from โ into โฌ (again provided with the pointwise algebraic operations and the supremum norm): if โฌ is spectrally regular, then so is ๐๐ซ(โ; โฌ). Finally, if โฌ is a spectrally regular Banach algebra, then the Wiener algebra ๐ฒ(๐; โฌ) of โฌ-valued functions on the unit circle ๐ is spectrally regular. This follows by noting that ๐ฒ(๐; โฌ) is continuously embedded in ๐(๐; โฌ). Taking advantage of Theorem 4.6, we close this subsection with an example of a spectrally regular ๐ถ โ -algebra ๐ having a closed two-sided ideal ๐ฅ (closed under the โ -operation) such that the quotient Banach algebra ๐/๐ฅ is not spectrally regular. As noted in the discussion after Corollary 4.1, the existence of such an example is in sharp contrast with the fact that each Banach subalgebra of a spectrally regular Banach algebra is spectrally regular again. ๐ร๐ To obtain the example we start with โM }๐โโ . This Banach โ with M = {โ algebra is spectrally regular by Theorem 4.6. We now pass to a ๐ถ โ -subalgebra of
142
H. Bart, T. Ehrhardt and B. Silbermann
๐ ๐ โM โ . For ๐ = 1, 2, 3, . . . , de๏ฌne ๐๐ : โ2 โ โ and ๐๐ : โ โ โ2 by โ โ ๐ฅ1 โ โ โ .. โ โ โ โ โ ๐ฅ1 โ . โ ๐ฅ1 ๐ฅ1 โ โ โ ๐ฅ2 โ ๐ฅ๐ โ โ .. โ โ โ โ โ .. โ โ. ๐๐ โ ๐ฅ3 โ = โ . โ , ๐๐ โ . โ = โ โ 0 โ โ โ โ โ .. ๐ฅ๐ ๐ฅ๐ โ 0 โ . โ โ .. . M Let โM โ,โ consist of all ๐ โ โโ such that the strong limits s-lim๐โโ ๐๐ ๐ (๐)๐๐ and โ โ M s-lim๐โโ ๐๐ ๐ (๐) ๐๐ exist in โฌ(โ2 ). Then โM โ,โ is a ๐ถ -subalgebra of โโ . Since M โ โM โ is spectrally regular, so is โโ,โ . Introduce the continuous ๐ถ -homomorphism M ฮจ : โโ,โ โ โฌ(โ2 ) by ฮจ(๐ ) = s-lim๐โโ ๐๐ ๐ (๐)๐๐ . Take ๐ โ โฌ(โ2 ), and let ๐ = (๐(1), ๐(2), ๐(3), . . .) be given by ๐(๐) = ๐๐ ๐ ๐๐ (so ๐ is built from the ๏ฌnite sections of ๐ ). Then s-lim๐โโ ๐๐ ๐(๐)๐๐ = ๐ and s-lim๐โโ ๐๐๐(๐)โ ๐๐ = ๐ โ . M Hence ๐ โ โM โ,โ and ฮจ(๐) = ๐ . We conclude that ฮจ : โโ,โ โ โฌ(โ2 ) is surjective. Put ๐ฅ = Ker ฮจ. Then ๐ฅ is a closed two-sided ideal in โM โ,โ (closed under the โ โ -operation) and the quotient space โM /๐ฅ is ๐ถ -isomorphic to โฌ(โ2 ). As โฌ(โ2 ) โ,โ M lacks the property of being spectrally regular, so does โโ,โ /๐ฅ .
4.3. Abstract matrix algebras Let โฌ be a unital Banach algebra, let ๐ be a positive integer, and let โฌ ๐ร๐ stand for the set of ๐ ร ๐ matrices with entries from โฌ. With the standard algebraic operations, and one of the usual norms (see, for instance, [P], Subsection 1.6.9), โฌ ๐ร๐ is again a unital Banach algebra. Clearly โฌ can be identi๏ฌed with the Banach subalgebra of โฌ ๐ร๐ consisting of all ๐ ร ๐ diagonal matrices in โฌ ๐ร๐ with constant diagonal. Thus โฌ is spectrally regular whenever โฌ ๐ร๐ has this property (see Corollary 4.1). What about the converse? Formulated in a more ๏ฌexible way: if โฌ is spectrally regular, under what additional conditions can one conclude that โฌ ๐ร๐ is spectrally regular too? The complete answer to this question is not known; two positive results that we have been able to obtain are presented below. To give the proper context for the ๏ฌrst, we recall that a Banach algebra is spectrally regular provided it possesses a radical-separating family of matrix representations (special case of Corollary 3.3). Proposition 4.8. Let โฌ be a unital Banach algebra, and let ๐ be a positive integer. Suppose โฌ possesses a radical-separating family of matrix representations (so โฌ is spectrally regular). Then the matrix algebra โฌ ๐ร๐ has a radical-separating family of matrix representations too, hence it is spectrally regular. Conversely, if โฌ ๐ร๐ has a radical-separating family of matrix representations, then so has โฌ. As will be clear from the proof, the proposition remains true when radicalseparating is replaced by separating. The modi๏ฌcation of the proposition involving unital matrix representations is correct also.
Spectral Regularity and Non-commutative Gelfand Theory
143
Proof. Let {๐๐ : โฌ โ โ๐๐ ร๐๐ }๐โฮฉ be a family of matrix representations. For ๐ โ ฮฉ, de๏ฌne ฮฆ๐ : โฌ ๐ร๐ โ โ๐๐๐ ร๐๐๐ by ( ) ฮฆ๐ [๐๐๐ ]๐๐,๐ =1 = [๐๐ (๐๐๐ )]๐๐,๐ =1 . Then ฮฆ๐ is a matrix representation (unital when ฮฆ๐ is). Clearly โฉ โฉ Ker ฮฆ๐ โ ๐๐๐ โ Ker ๐๐ , ๐, ๐ = 1, . . . , ๐. [๐๐๐ ]๐,๐ =1 โ ๐โฮฉ
โฉ
๐โฮฉ
โฉ If {๐๐ }๐โฮฉ is separating, then ๐โฮฉ Ker ๐๐ = {0}, hence ๐โฮฉ Ker ฮฆ๐ = {0}, so {ฮฆ๐ } is separating too. Next suppose {๐๐ }๐โฮฉ is radical-separating. Then โฉ ๐โฮฉ Ker ๐๐ โ โ(โฌ), and we see that โฉ Ker ฮฆ๐ โ ๐๐๐ โ โ(โฌ), ๐, ๐ = 1, . . . , ๐. [๐๐๐ ]๐,๐ =1 โ ๐โฮฉ
Now the radical of โฌ ๐ร๐ consists of all matrices in โฌ ๐ร๐ with entries in โ(โฌ). This well-known result can be found, for instance, as Proposition 5.14 in [CR]; cf. also Proposition 1.1.15 in [RSS] โฉ for a more general observation on ideals in matrix algebras. It follows that ๐โฮฉ Ker ฮฆ๐ โ โ(โฌ ๐ร๐ ), i.e., the family {ฮฆ๐ } is radical-separating. To start the argument for the second part, recall that โฌ can be identi๏ฌed with the inverse closed Banach subalgebra ๐ of โฌ ๐ร๐ consisting of all ๐ ร ๐ diagonal matrices in โฌ ๐ร๐ with constant diagonal. Let {ฮฆ๐ : โฌ ๐ร๐ โ โ๐๐ ร๐๐ }๐โฮฉ be a family of matrix representations, and, for ๐ โ ฮฉ, let ๐๐ be the restriction of ๐๐ ร๐๐ ฮฆ๐ to ๐. Then โฉ ๐๐ : ๐ โ โ โฉ is a matrix representation (unital when ฮฆ๐ is). Clearly Ker ๐ = ๐ โฉ ๐ ๐โฮฉ ๐โฮฉ Ker ฮฆ๐ . If {ฮฆ๐ }๐โฮฉ is separating, then โฉ โฉ Ker ฮฆ = {0}, hence Ker ๐๐ = {0}, ๐ ๐โฮฉ ๐โฮฉ โฉ so {๐๐ } is separating too. Next assume {ฮฆ๐ }โฉ๐โฮฉ is radical-separating. Thus ๐โฮฉ Ker ฮฆ๐ โ โ(โฌ ๐ร๐ ), and it follows that ๐โฮฉ Ker ๐๐ โ ๐ โฉ โ(โฌ ๐ร๐ ). The right-hand side of this inclusion is contained in the radical of ๐ because ๐ is inverse closed in โฌ ๐ร๐ . Hence {๐๐ }๐โฮฉ is radical-separating, as desired. โก Our next result is concerned with a special case of the situation covered by the ๏ฌrst part Proposition 4.8. However, the stronger condition that is imposed (cf. Proposition 3.4) allows for a correspondingly stronger conclusion. Anticipating on the proof to be given, we mention an important result due to Procesi and Small [PS] which will serve as an essential tool in the argument: if โฌ is a PI-algebra, then so is the matrix algebra โฌ ๐ร๐ (๐ a positive integer). For material on PI-algebras, see Section 2. Proposition 4.9. Let โฌ be a unital Banach algebra, let ๐ be a positive integer, and suppose โฌ possesses a su๏ฌcient family of matrix representations of ๏ฌnite order (so โฌ is spectrally regular). Then the matrix algebra algebra โฌ ๐ร๐ has a su๏ฌcient family of matrix representations of ๏ฌnite order too, hence it is spectrally regular. Proof. The hypothesis on โฌ amounts the requirement that the quotient algebra โฌ/โ(โฌ) is PI (see Section 2, and the references given there). Now apply
144
H. Bart, T. Ehrhardt and B. Silbermann
the result of Procesi and Small quoted above. It follows that the matrix algebra ( )๐ร๐ โฌ/โ(โฌ) is PI. Write ๐ for the canonical mapping of โฌ onto โฌ/โ(โฌ), and ( )๐ร๐ ( ) ๐ร๐ โ โฌ/โ(โฌ) by ฮฆ [๐๐๐ ]๐๐,๐ =1 = [๐(๐๐๐ )]๐๐,๐ =1 . Then ฮฆ is de๏ฌne ฮฆ : โฌ a surjective algebra homomorphism and, using Proposition 5.14 in [CR] again, its null space is โ(โฌ ๐ร๐ ). Thus โฌ ๐ร๐ /โ(โฌ ๐ร๐ ), being algebraically isomorphic ( )๐ร๐ to โฌ/โ(โฌ) , is a PI-algebra. But then, as we wanted to prove, โฌ ๐ร๐ has a su๏ฌcient family of matrix representations of ๏ฌnite order. โก As was noted before, โฌ can be identi๏ฌed with the inverse closed Banach subalgebra of โฌ ๐ร๐ consisting of all ๐ร ๐ diagonal matrices in โฌ ๐ร๐ with constant diagonal. Hence, if โฌ ๐ร๐ has a su๏ฌcient family of matrix representations of ๏ฌnite order, then so does โฌ. In combination with Proposition 4.9 this gives: the matrix algebra โฌ ๐ร๐ has a su๏ฌcient family of matrix representations of ๏ฌnite order if and only if so does โฌ. This bears a certain analogy to Proposition 4.4. The latter has no ๏ฌnite order condition on the su๏ฌcient family of matrix representations, however. We do not know whether one can do without this restriction here too. Let us ๏ฌnish this subsection with a simple observation concerning the Banach ๐ร๐ subalgebra โฌupper of โฌ ๐ร๐ consisting of the upper triangular ๐ ร ๐ matrices with entries in โฌ. ๐ร๐ Proposition 4.10. If the unital Banach algebra โฌ is spectrally regular, then โฌupper is spectrally regular too. ๐ร๐ Proof. The homomorphisms ๐1 , . . . , ๐๐ , with ๐๐ mapping a matrix from โฌupper into its ๐th diagonal element, form a su๏ฌcient family of Banach algebra homo๐ร๐ morphisms mapping โฌupper into the Banach algebra โฌ, and the latter is spectrally regular by assumption. โก
In Proposition 4.10, upper triangularity can of course be replaced by lower triangularity. For Banach algebras of operators, triangularity can be brought into connection with families of invariant subspaces. This line of thought is pursued in the next subsection. 4.4. Algebras of operators with prescribed invariant subspaces Let ๐ be a complex Banach space and let โณ be a family of closed nontrivial subspaces of ๐. By โฌ(๐; โณ) we denote the set of all operators ๐ โ โฌ(๐) such that ๐ [๐ ] โ ๐ for all ๐ โ โณ. Clearly โฌ(๐; โณ) is a Banach subalgebra of โฌ(๐). It is our aim to give su๏ฌcient conditions in order that โฌ(๐; โณ) is spectrally regular. An obvious condition of this type is that ๐ is ๏ฌnite dimensional so that โฌ(๐; โณ) can be identi๏ฌed with a subalgebra of โ๐ร๐ where ๐ is the dimension of ๐. Hence, from now on, we assume that ๐ is in๏ฌnite dimensional (so that โฌ(๐) and its Banach subalgebras need not be spectrally regular). Prominent of algebras of the type โฌ(๐; โณ) are the Banach sub( instances ) algebra of โฌ โ2 (โ) consisting of block upper triangular operators ) (with respect ( to a given orthonormal basis), the Banach subalgebra of โฌ โ2 (โ) consisting of
Spectral Regularity and Non-commutative Gelfand Theory
145
( ) block lower triangular operators, and the Banach subalgebra of โฌ โ2 (โค) consisting of block upper (or, alternatively, lower) triangular operators (all the time with ๏ฌnite but possibly variable block size). For these, spectral regularity can be established with the help of Corollary 3.2. However, basically the same argument as the one employed for these cases gives a more general result which shows that it makes sense to have โฑ (๐) โ {๐ผ๐ } in the right-hand side of (8); see the discussion involving the expressions (11) and (12) in Section 3. To facilitate the further exposition, we need some preparations. As before ๐ will be an in๏ฌnite-dimensional Banach space. We say that ๐ is almost included in ๐ , written ๐ โบ ๐ , if dim ๐/(๐ โฉ ๐ ) is ๏ฌnite. It is a well-known fact that dim ๐/(๐ โฉ ๐ ) = dim (๐ + ๐ )/๐ . Hence ๐ โบ ๐ if and only if dim (๐ + ๐ )/๐ < โ. If ๐ โ ๐ , then ๐ โบ ๐ if and only if ๐ has ๏ฌnite codimension in ๐ . Also ๐ โบ ๐ whenever ๐ โ ๐ . In particular ๐ โบ ๐ , so the relation โบ is re๏ฌexive. As is easily veri๏ฌed, it is also transitive. If ๐ is a linear operator on ๐ and ๐ โบ ๐ , then ๐ [๐ ] โบ ๐ [๐ ] too. The subspaces ๐ and ๐ are said to be almost equal, written ๐ โ ๐, if both ๐ โบ ๐ and ๐ โบ ๐ . This is equivalent to requiring that the quotient space (๐ + ๐ )/(๐ โฉ ๐ ) has ๏ฌnite dimension. Note that โ is an equivalence relation. Hence the collection of all closed subspaces of ๐ is the disjoint union of the equivalence classes modulo โ. A nonempty subset of such an equivalence class will be called a cluster. An example of a cluster is a nonempty family of ๏ฌnitedimensional subspaces of ๐. A nonempty family of closed ๏ฌnite codimensional subspaces of ๐ is a cluster as well. We are now ready to present our next theorem. Its proof will illustrate that in Corollary 3.2 it is important to have condition (8) instead of one of the possibly more restrictive requirements (11) or (12). Theorem 4.11. Let โณ1 , . . . , โณ๐ be an ๐-tuple of clusters of closed subspaces of the in๏ฌnite-dimensional Banach space X. Suppose the ๐-tuple is almost nested in the sense that โฉ โ ๐ โบ ๐, ๐ = 1, . . . , ๐ โ 1. (13) ๐โโณ๐
๐โโณ๐+1
โฉ Further assume that codim ๐โโณ1 ๐ < โ and dim ๐โโณ๐ ๐ < โ. Then, with โณ being the union of the clusters โณ1 โชโ
โ
โ
โชโณ๐ , the Banach algebra โฌ(๐; โณ) is spectrally regular. โ
The Banach algebras of block triangular operators mentioned earlier all correspond to situations where the requirements in the theorem are trivially ful๏ฌlled. For details, see Theorem 4.12 and the comments concerning it at the end of this subsection. Proof. Let ๐ and ๐ be closed subspaces of ๐, let ๐ be a bounded linear operator on ๐, and suppose ๐ [๐ ] โ ๐ and ๐ [๐ ] โ ๐ . Then ๐ [๐ + ๐ ] โ ๐ + ๐ and ๐ [๐ โฉ ๐ ] โ ๐ โฉ ๐ . Hence ๐ induces a bounded linear operator on the quotient
146
H. Bart, T. Ehrhardt and B. Silbermann
space (๐ + ๐ )/(๐ โฉ ๐ ) which we will denote by ๐๐,๐ . Clearly ๐๐,๐ is the zero operator on (๐ + ๐ )/(๐ โฉ ๐ ) if and only if ๐ [๐ + ๐ ] โ ๐ โฉ ๐ . Given the (assumed) inclusions ๐ [๐ ] โ ๐ and ๐ [๐ ] โ ๐ , this comes down to ๐ [๐ ] โ ๐ and ๐ [๐ ] โ ๐ . If ๐ is another bounded linear operator on ๐ leaving invariant ๐ and ๐ , then ๐ + ๐ leaves ๐ and ๐ invariant too and (๐ + ๐)๐,๐ = ๐๐,๐ + ๐๐,๐ . Similarly (๐ ๐)๐,๐ = ๐๐,๐ ๐๐,๐ and (๐ผ๐ )๐,๐ = ๐ผ๐๐,๐ , ๐ผ โ โ. Take ๐ โ {1, . . . , ๐} and ๐, ๐ โ โณ๐ with ๐ โ= ๐ . Then the quotient space (๐ + ๐ )/(๐ โฉ ๐ ) has positive ๏ฌnite dimension. For ๐ โ โฌ(๐; โณ), we have that ๐ leaves invariant ๐ and ๐ , and we can put ฮฆ๐;๐,๐ (๐ ) = ๐๐,๐ . In this way we get a continuous (unital) homomorphism ( ) ๐ +๐ ฮฆ๐;๐,๐ : โฌ(๐, โณ) โ โฌ . ๐ โฉ๐ In the sequel it will be identi๏ฌed with a matrix representation on โฌ(๐; โณ). Fix ๐ among the integers 1, . . . , ๐, and consider {ฮฆ๐;๐,๐ }๐,๐ โโณ๐ , ๐โ=๐ . This is a family of matrix representations on โฌ(๐; โณ). We claim that [ โ ] โฉ โฉ ๐ โ Ker ฮฆ๐;๐,๐ โ ๐ ๐ โ ๐, (14) ๐ โโณ๐
๐,๐ โโณ๐ , ๐โ=๐
โ
๐โโณ๐
where the symbol signals the operation of taking the closed linear span. The argument is as follows. To obtain the inclusion in the right-hand side of (14), we need to show that ๐ [๐ ] โ ๐ for all ๐, ๐ โ โณ๐ . Take ๐ in the left-hand side of (14) and ๐, ๐ โ โณ๐ . If ๐ = ๐ we have ๐ [๐ ] โ ๐ = ๐ because ๐ โ โฌ(๐; โณ). If ๐ โ= ๐ , we have ๐๐,๐ = ฮฆ๐;๐,๐ (๐ ) = 0, and so ๐ [๐ + ๐ ] โ ๐ โฉ ๐ , in particular ๐ [๐ ] โ ๐ . For convenience, write ๐ท0 = ๐, ๐๐+1 = {0} and โฉ โ ๐ท๐ = ๐, ๐๐ = ๐, ๐ = 1, . . . , ๐. ๐ โโณ๐
๐ โโณ๐
Then, by the hypotheses in the theorem, ๐ท๐ โบ ๐๐+1 for ๐ = 0, . . . , ๐. Hence ๐ [๐ท๐ ] โบ ๐ [๐๐+1 ],
๐ = 0, . . . , ๐,
(15)
where for ๐ one can take any linear operator on ๐. Next consider {ฮฆ๐;๐,๐ }๐,๐ โโณ๐ , ๐ โ=๐ ; โฉ ๐=1,...,๐ . This again is a family of matrix representations on โฌ(๐; โณ). Take ๐ in ๐,๐ โโณ๐ , ๐ โ=๐ ; ๐=1,...,๐ Ker ฮฆ๐;๐,๐ . Then we have from (14) that ๐ [๐๐ ] โ ๐ท๐ ,
๐ = 1, . . . , ๐ .
(16)
Combining (15) and (16), we get ๐ [๐ท๐ ] โบ ๐ท๐+1 , ๐ = 0, . . . , ๐ โ 1. But then, via (๏ฌnite) induction, ๐ ๐ [๐ท0 ] โบ ๐ท๐ , ๐ = 0, . . . , ๐. In particular ๐ ๐ [๐ท0 ] โบ ๐ท๐ . As ๐ท0 = ๐ and ๐ท๐ โบ ๐๐+1 = {0}, it follows that Im ๐ ๐ โบ {0}. Thus Im ๐ ๐ is ๏ฌnite dimensional, i.e., ๐ ๐ is a ๏ฌnite rank operator (hence compact). By standard Fredholm theory, we may conclude that ๐ผ๐ โ (โ๐ )๐ is a Fredholm operator, i.e., Ker(๐ผ๐ โ (โ๐ )๐ ) is ๏ฌnite dimensional and Im(๐ผ๐ โ (โ๐ )๐)
Spectral Regularity and Non-commutative Gelfand Theory
147
has ๏ฌnite codimension in ๐. Now ) ( ๐โ1 ) ( ๐โ1 โ โ ๐ ๐ ๐ = ๐ผ๐ โ (โ๐ ) = (๐ผ๐ + ๐ ) (โ๐ ) (โ๐ ) (๐ผ๐ + ๐ ), ๐=0
๐=0
therefore Im(๐ผ๐ โ (โ๐ )๐ ) โ Im(๐ผ๐ + ๐ ) and Ker(๐ผ๐ + ๐ ) โ Ker(๐ผ๐ โ (โ๐ )๐ ). ๐ So, along with ๐ผ๐ โ (โ๐ )โฉ , the operator ๐ผ๐ + ๐ is Fredholm. We conclude that ๐,๐ โโณ๐ , ๐โ=๐ ; ๐=1,...,๐ Ker ฮฆ๐;๐,๐ โ โฑ (๐) โ {๐ผ๐ }. Corollary 3.2 now gives that โฌ(๐; โณ) is spectrally regular. โก Elaborating on the proof, we note that the family of matrix representations {ฮฆ๐;๐,๐ }๐,๐ โโณ๐ , ๐โ=๐ ; ๐=1,...,๐ is nonempty. Suppose it is not. Then all the clusters โณ1 , . . . , โณ๐ are singletons and we get ๐ = ๐ท0 โบ ๐1 = ๐ท1 โบ ๐2 = ๐ท2 โบ โ
โ
โ
โบ ๐๐ = ๐ท๐ โบ ๐๐+1 = {0}. By transitivity this gives ๐ โบ {0}, contradicting the in๏ฌnite dimensionality of ๐. There is another elucidating observation to make. Suppose the ๐-tuple of clusters in Theorem 4.11 is nested (instead of only almost nested) in the sense that the almost inclusions in (13) are in fact genuine inclusions. Then all the almost inclusions in the above proof are genuine inclusions too. This leads to the stronger conclusion that ๐ is nilpotent; in fact ๐ ๐ = 0. We conclude this subsection by coming back to Theorem 4.11 for the case ๐ = 1. For that situation, the theorem reads as follows. Theorem 4.12. Let โณ be a cluster of closed subspaces of the in๏ฌnite-dimensional Banach space X. Assume โฉ โ ๐ < โ, dim ๐ < โ. (17) codim ๐โโณ
๐โโณ
Then โฌ(๐; โณ) is spectrally regular. Theorem 4.12 can be used to deal with the Banach algebras of triangular operators mentioned in the third paragraph of this subsection. Here are the details. (a) Let โณ be a nonempty family of ๏ฌnite-dimensional subspaces of the in๏ฌnitedimensional Banach space ๐. Then โณ is a cluster and it is clear that the second part of (17) is satis๏ฌed. If the ๏ฌrst part of (17) is ful๏ฌlled too, we may conclude that( โฌ(๐;) โณ) is spectrally regular. This covers the Banach subalgebra of โฌ โ2 (โ) consisting of block upper triangular operators where, for the appropriate choice of โณ, the ๏ฌrst part of (17) even amounts to โ ๐โโณ ๐ = โ2 (โ). (b) Let โณ be a nonempty family of ๏ฌnite codimensional subspaces of the in๏ฌnitedimensional Banach space ๐. Then โณ is a cluster and it is clear that the ๏ฌrst part of (17) is satis๏ฌed. If the second part of (17) is ful๏ฌlled too, we may conclude that ) โณ) is spectrally regular. This covers the Banach ( โฌ(๐; subalgebra of โฌ โ2 (โ) consisting of block lower triangular operators where,
148
H. Bart, T. Ehrhardt and B. Silbermann
for the appropriate choice of โณ, the second part of (17) even boils down to โฉ ๐โโณ ๐ = {0}. ) ( (c) Theorem 4.12 also covers the Banach subalgebra of โฌ โ2 (โค) consisting of block upper (or, alternatively lower) triangular operators. For the appropriate โ choice of โณ, the ๏ฌrst part of (17) boils down to ๐ = โ2 (โค) and the ๐โโณ โฉ second to ๐โโณ ๐ = {0}. In (a), (b) and (c), block triangularity is taken with respect to a given orthonormal basis in, respectively, โ2 (โ), โ2 (โ), and โ2 (โค). The blocks are allowed to be of variable (but ๏ฌnite) size. 4.5. Algebras of Toeplitz and singular integral operators We start with the following immediate consequence of Corollary 3.2. Corollary 4.13. Let ๐ be an in๏ฌnite-dimensional Banach space, and let โฌ be a Banach subalgebra of โฌ(๐). If the quotient Banach algebra โฌ/(๐ฆ(๐) โฉ โฌ) is spectrally regular, then so is โฌ. One may replace ๐ฆ(๐) by the generally larger ideal of the strictly singular operators on ๐; see the corresponding remark in Section 3. Proof. Consider the singleton family {๐
}, where ๐
: โฌ โ โฌ/(๐ฆ(๐) โฉ โฌ) is the canonical mapping, and apply Corollary 3.2. โก As a special case of Corollary 4.13, we have the following result. Let ๐ be an in๏ฌnite-dimensional Banach space, and let โฌ be a Banach subalgebra of โฌ(๐). Suppose the ideal ๐ฆ(๐) of the compact operators on ๐ is contained in โฌ. Then โฌ is spectrally regular provided the quotient โฌ/๐ฆ(๐) is. This means that in cases where ๐ฆ(๐) โ โฌ and โฌ/๐ฆ(๐) is a polynomial identity algebra or, more generally, โฌ/๐ฆ(๐) possesses a su๏ฌcient family of matrix representations, one can conclude that โฌ is spectrally regular. There is an abundance of such situations, especially in the theory of singular integral operators and Toeplitz operators: see, for instance, the books [BK], [BS], [Cor], [GGK2], [GK1], [GK2], [Kr], and the paper [GK3]. As a characteristic illustration, we consider the unital ๐ถ โ -algebras generated by block Toeplitz operators appearing in [GGK2], Sections XXXII.2 and XXXII.4. Depending on the continuity requirements imposed on the so-called de๏ฌning (or generating) function, the algebras in question are denoted there by ๐ฏ๐ (๐ถ) and ๐ฏ๐ (๐ ๐ถ). In fact, ๐ฏ๐ (๐ถ) and ๐ฏ๐ (๐ ๐ถ) are, respectively, the smallest closed subalgebra of โฌ(โ๐ 2 ) containing all block Toeplitz operators for which the de๏ฌning function is a continuous, respectively, a piecewise continuous, โ๐ร๐ -valued function. Theorem 4.14. The ๐ถ โ -algebras ๐ฏ๐ (๐ถ) and ๐ฏ๐ (๐ ๐ถ) are spectrally regular. Proof. Let ๐ฏ be one of the Banach algebras mentioned above. Then ๐ฏ is a Banach ๐ subalgebra of โฌ(โ๐ 2 ) where โ2 stands for the Hilbert space of square summable ๐ sequences with entries in โ . We now make use of the material presented in [GGK2], Chapter XXXII, in particular Theorems 2.1 and 4.2. The ๏ฌrst thing
Spectral Regularity and Non-commutative Gelfand Theory
149
to mention is that ๐ฏ contains the ideal ๐ฆ = ๐ฆ(โ๐ 2 ) of the compact operators on โ๐ 2 . The second is that ๐ฏ /๐ฆ can be identi๏ฌed with a Banach algebra of the type ๐(๐, โ๐ร๐ ) where ๐ is an appropriately chosen compact topological space. This Banach algebra is spectrally regular, a conclusion which has been drawn in Section 4.2 from Corollaries 4.1 and 4.7. Along with ๐(๐, โ๐ร๐ ), the quotient algebra ๐ฏ /๐ฆ is spectrally regular too. The spectral regularity of ๐ฏ now follows by applying Corollary 4.13. โก We add to the above argument that is also easy to see that the algebra ๐(๐, โ๐ร๐ ) is PI. Indeed, as the operations in ๐(๐, โ๐ร๐ ) are de๏ฌned pointwise, an annihilating polynomial for โ๐ร๐ is one for ๐(๐, โ๐ร๐ ) too. The property of being PI carries over to ๐ฏ /๐ฆ. Now let us specialize to the case ๐ = 1 and consider ๐ฏ (๐ถ) = ๐ฏ1 (๐ถ), the algebra generated by the Toeplitz operators on โ2 (โ) with continuous generating function. By a result of Coburn [Cob], the ๐ถ โ -algebra ๐ฏ (๐ถ) is โ -isomorphic to the so-called universal algebra generated by one nonunitary isometry. Hence this universal algebra, which can occur in many di๏ฌerent appearances, is spectrally regular (see [RR] and [GF]). For a further analysis, see the forthcoming paper [BES8] where related algebras are considered too. ( Toeplitz ) algebras can also be considered in the context of the spaces โ๐ (โค+ ), ๐ฟ๐ [0, โ) and ๐ป๐ (๐); see [BS]. Corollary 4.13 is then applicable too. Indeed, factoring out the compacts gives again a spectrally regular Banach algebra, in fact one that has a su๏ฌcient family of matrix representations of ๏ฌnite order. Recall that this does not automatically give that the quotient algebra is PI; it does when the quotient algebra is semisimple.
5. Concluding remarks In the above, we encountered families of Banach algebra homomorphisms having certain properties pertinent to the topic of this paper. Certain relationships between these properties are obvious, others, somewhat less trivial, have been established in Section 3. Restricting ourselves (in order to keep things tractable) to considering matrix representations only, the situation is as depicted in the following scheme PI โ su๏ฌcient, ๏ฌnite order โ su๏ฌcient โ p.w. su๏ฌcient โ radical-separating โ separating (where the third implication from the left has to be understood as being true modulo an appropriate renorming of the test algebras). Clearly, the overarching notion is that of a radical-separating family. Now the question arises, is it overarching in the strict sense? Or, in more precise terms, what about the converses of the implications in the above scheme?
150
H. Bart, T. Ehrhardt and B. Silbermann
This issue is addressed in [BES7]. Here we only mention that one of the main results presented there is that the converse of the implication su๏ฌcient โ radical-separating, (implicit in the above scheme) is not valid. In fact, not even the implication separating โ su๏ฌcient, holds. On the level of individual families, this has already been noted before: see the paragraph directly following the proof of Theorem 4.6. However, it is even true in the much stronger sense that a Banach algebra may possess a family of unital matrix representations which is separating and p.w. su๏ฌcient while it fails to ๐ร๐ have any su๏ฌcient one. An example is the ๐ถ โ -algebra โM }๐โโ , โ , with M = {โ featuring in the last paragraph of Subsection 4.2. For the proof of the fact that โM โ does not possess any su๏ฌcient family of matrix representations, one needs some โgraspโ on the collection of all unital matrix representations of the Banach algebra โM โ โ a highly nontrivial matter (cf. the situation for the relatively simple Banach algebra โโ ). One ๏ฌnal remark. As was mentioned in Section 2, spectral regularity is a necessary condition for a Banach algebra to have a su๏ฌcient family of matrix representations. The ๐ถ โ -algebra โM โ illustrates that it is not a su๏ฌcient condition. So additional requirements are needed to characterize the Banach algebras possessing a su๏ฌcient family of matrix representations (not necessarily of ๏ฌnite order), an issue posed as Problem 12 in Section 29 of [Kr]. Acknowledgement The second author (T.E.) was supported in part by NSF grant DMS-0901434.
References [AA]
[AL] [Bar] [BES1] [BES2] [BES3]
Y.A. Abramovich, C.D. Aliprantis, An Invitation to Operator Theory, Graduate Studies in Mathematics, Vol. 50, American Mathematical Society, Providence, Rhode Island 2002. S.A. Amitsur, J. Levitzky, Minimal identities for algebras, Proc. Amer. Math. Soc. 1 (1950), 449โ463. H. Bart, Spectral properties of locally holomorphic vector-valued functions, Paci๏ฌc J. Math. 52 (1974), 321โ329. H. Bart, T. Ehrhardt, B. Silbermann, Zero sums of idempotents in Banach algebras, Integral Equations and Operator Theory 19 (1994), 125โ134. H. Bart, T. Ehrhardt, B. Silbermann, Logarithmic residues in Banach algebras, Integral Equations and Operator Theory 19 (1994), 135โ152. H. Bart, T. Ehrhardt, B. Silbermann, Logarithmic residues of Fredholm operator-valued functions and sums of ๏ฌnite rank projections, In: Operator Theory: Advances and Applications, Vol. 130, Birkhยจ auser Verlag, Basel 2001, pp. 83โ106.
Spectral Regularity and Non-commutative Gelfand Theory [BES4]
151
H. Bart, T. Ehrhardt, B. Silbermann, Logarithmic residues in the Banach algebra generated by the compact operators and the identity, Mathematische Nachrichten 268 (2004), 3โ30. [BES5] H. Bart, T. Ehrhardt, B. Silbermann, Vector-Valued Logarithmic Residues and the Extraction of Elementary Factors, Econometric Institute Erasmus University Rotterdam, Report nr. EI 2007-31, 2007. [BES6] H. Bart, T. Ehrhardt, B. Silbermann, Trace conditions for regular spectral behavior of vector-valued analytic functions, Linear Algebra Appl. 430 (2009), 1945โ1965. [BES7] H. Bart, T. Ehrhardt, B. Silbermann, Families of homomorhisms in non-commutative Gelfand theory: comparisons and counterexamples, accepted for publication in the IWOTA 2010 Proceedings. In: W. Arendt, J.A. Ball, J. Behrndt, K.-H. Fยจ orster, V. Mehrmann, C. Trunk (eds.): Recent Advances in Operator Theory, Oper. Theory Adv. Appl. OT 221, Birkhยจ auser, Springer Basel AG, 2012. [BES8] H. Bart, T. Ehrhardt, B. Silbermann, Logarithmic Residues, Rouchยดeโs Theorem, Spectral Regularity, and Zero Sums of Idempotents: the ๐ถ โ -algebra Case, forthcoming. [BK] A. Bยจ ottcher, Yu. Karlovich, Carleson Curves, Muckenhaupt Weights, and Toeplitz Operators, Progress in Mathematics, Vol. 154, Birkhยจ auser Verlag, Basel 1997. [BS] A. Bยจ ottcher, B. Silbermann, Analysis of Toeplitz Operators, Springer Verlag, Berlin 1990. [Cob] L.A. Coburn, The ๐ถ โ -algebra generated by an isometry, Bull. Amer. Math. Soc. 73 (1967), 722โ726. [Cor] H.O. Cordes, Elliptic Pseudodi๏ฌerential Operators โ An Abstract Theory, Lecture Notes in Mathematics, Springer Verlag, Berlin 1995. [CR] C.W. Curtis, I. Reiner: Methods of representation theory, Vol. I. With applications to ๏ฌnite groups and orders, Wiley Classics Library. John Wiley and Sons, New York 1990. [E] T. Ehrhardt, Finite sums of idempotents and logarithmic residues on connected domains, Integral Equations and Operator Theory 21 (1995), 238โ242. [Go] S. Goldberg, Unbounded Linear Operators, McGraw-Hill, New York 1966. [GF] I. Gohberg, I.A. Feldman, Convolution Operators and Projection Methods for Their Solution, Translations of Mathematical Monographs, Vol. 41, Amer. Math. Soc., Providence, Rhode Island 1974. [GGK1] I. Gohberg, S. Goldberg, M.A. Kaashoek, Classes of Linear Operators, Vol. I, Operator Theory: Advances and Applications, Vol. 49, Birkhยจ auser Verlag, Basel 1990. [GGK2] I. Gohberg, S. Goldberg, M.A. Kaashoek, Classes of Linear Operators, Vol. II, Operator Theory: Advances and Applications, Vol. 63, Birkhยจ auser Verlag, Basel 1993. [GK1] I.C. Gohberg, N.Ya. Krupnik, One-Dimensional Linear Singular Integral Equations, Vol. 1, Operator Theory: Advances and Applications, Vol. 53, Birkhยจ auser Verlag, Basel 1992.
152
H. Bart, T. Ehrhardt and B. Silbermann
[GK2]
I.C. Gohberg, N.Ya. Krupnik, One-Dimensional Linear Singular Integral Equations, Vol. 2, Operator Theory: Advances and Applications, Vol. 54, Birkhยจ auser Verlag, Basel 1992.
[GK3]
I.C. Gohberg, N.Ya. Krupnik, On an algebra generated by Toeplitz matrices, Funk. Anal. i Priloz 3 (1969), 46โ56 (Russian); English Transl., Funct. Anal. Appl. 3 (1969), 119โ127.
[GL]
I. Gohberg, J. Leiterer, Holomorphic Operator Functions of One Variable and Applications, Operator Theory: Advances and Applications, Vol. 192, Birkhยจ auser Verlag, Basel 2009.
[GS]
I.C. Gohberg, E.I. Sigal, An operator generalization of the logarithmic residue theorem and the theorem of Rouchยดe, Mat. Sbornik 84 (126) (1971), 607โ629 (Russian); English Transl., Math. USSR Sbornik 13 (1971), 603โ625. B. Gramsch, Meromorphie in der Theorie der Fredholmoperatoren mit Anwendungen auf elliptische Di๏ฌerentialoperatoren, Math. Ann. 188 (1970), 97-112.
[Gra] [Gro]
[HRS]
[Ka] [Kr]
[LR] [N]
A. Grothendieck, Produits tensoriels topologiques et espaces nuclยดeaires, Mem. Amer. Math. Soc., No.16, American Mathematical Society, Providence, Rhode Island 1955 [French]. R. Hagen, S. Roch, B. Silbermann, Spectral Theory of Approximation Methods for Convolution Equations, Operator Theory: Advances and Applications, Vol. 74, Birkhยจ auser Verlag, Basel 1995. W. Kaballo, Lifting-Sยจ atze fยจ ur Vektorfunktionen und das ๐-Tensorprodukt, Habilitationsschrift, Kaiserslautern 1976. N.Ya. Krupnik, Banach Algebras with Symbol and Singular Integral Operators, Operator Theory: Advances and Applications, Vol. 26, Birkhยจ auser Verlag, Basel 1987. V. Lomonosov, P. Rosenthal, The simplest proof of Burnsideโs Theorem on matrix algebras, Linear Algebra Appl. 383 (2004), 45โ47. M.A. Naimark, Normed Rings, Wolters-Noordhof, Groningen 1970.
[RRS]
R. Rabinovich, S. Roch, B. Silbermann, Limit Operators and their Applications in Operator Theory, Operator Theory: Advances and Applications, Vol. 150, Birkhยจ auser Verlag, Basel 2004.
[P]
T.W. Palmer, Banach Algebras and The General Theory of *-Algebras, Volume I: Algebras and Banach Algebras, Cambridge University Press, Cambridge 1994.
[PS]
C. Procesi, L. Small, Endomorphism rings of modules over Pl-algebra, Math. Z. 106 (1968), 178โ180. C. Pearcy, D. Topping. Sums of small numbers of idempotents, Michigan Math. J. 14 (1967), 453โ465. M. Rosenblum, J. Rovnyak, Hardy Classes and Operator Theory, The Clarendon Press, Oxford University Press, New York 1985. S. Roch, P.A. Santos, B. Silbermann, Non-commutative Gelfand Theories, Springer Verlag, London Dordrecht, Heidelberg, New York 2011.
[PT] [RR] [RSS] [Si]
B. Silbermann, Symbol constructions and numerical analysis, In: Integral Equations and Inverse Problems (R. Lazarov, V. Petkov, eds.), Pitman Research Notes in Mathematics, Vol. 235, 1991, 241โ252.
Spectral Regularity and Non-commutative Gelfand Theory [TL]
153
A.E. Taylor, D.C. Lay, Introduction to Functional Analysis, Second Edition, John Wiley and Sons, New York 1980. [ZKKP] M.G. Zaหฤฑdenberg, S.G. Kreหฤฑn, P.A. Kuหcment, A.A. Pankov, Banach bundles and linear operators, Uspehi Mat. Nauk 30 no. 5(185) (1975), 101โ157 [Russian]; English Transl., Russian Math. Surveys 30 (1975), no. 5, 115โ175.
Harm Bart Econometric Institute Erasmus University Rotterdam P.O. Box 1738 NL-3000 DR Rotterdam, The Netherlands e-mail:
[email protected] Torsten Ehrhardt Mathematics Department University of California Santa Cruz, CA 95064, USA e-mail:
[email protected] Bernd Silbermann Fakultยจ at fยจ ur Mathematik Technische Universitยจ at Chemnitz D-09107 Chemnitz, Germany e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 155โ175 c 2012 Springer Basel AG โ
Banach Algebras of Commuting Toeplitz Operators on the Unit Ball via the Quasi-hyperbolic Group Wolfram Bauer and Nikolai Vasilevski To the memory of Professor I. Gohberg, a great mathematician and personality
Abstract. We continue the study of commutative algebras generated by Toeplitz operators acting on the weighted Bergman spaces over the unit ball ๐น๐ in โ๐ . As was observed recently, apart of the already known commutative Toeplitz ๐ถ โ -algebras, quite unexpectedly, there exist many others, not geometrically de๏ฌned, classes of symbols which generate commutative Toeplitz operator algebras on each weighted Bergman space. These classes of symbols were in a sense subordinated to the quasi-elliptic and quasi-parabolic groups of biholomorphisms of the unit ball. The corresponding commutative operator algebras were Banach, and being extended to the ๐ถ โ -algebras they became non-commutative. We consider here the case of symbols subordinated to the quasi-hyperbolic group and show that such classes of symbols are as well the sources for the commutative Banach algebras generated by Toeplitz operators. That is, together with the results of [11, 12], we cover the multidimensional extensions of all three model cases on the unit disk. Mathematics Subject Classi๏ฌcation (2000). Primary 47B35; Secondary 47L80, 32A36. Keywords. Toeplitz operator, weighted Bergman space, unit ball, commutative Banach algebra, quasi-hyperbolic group.
The ๏ฌrst named author has been supported by an โEmmy-Noether scholarshipโ of DFG (Deutsche Forschungsgemeinschaft). The second named author has been partially supported by CONACYT Project 102800, Mยด exico.
156
W. Bauer and N. Vasilevski
1. Introduction In the paper we continue the study of commutative algebras generated by Toeplitz operators acting on the weighted Bergman spaces over the unit ball ๐น๐ in โ๐ . The case of commutative ๐ถ โ -algebras was considered in [8], whose main result states that if the symbols of generating Toeplitz operators are invariant under the action of a maximal commutative subgroup of biholomorphisms of the unit ball, then the corresponding ๐ถ โ operator algebra is commutative on each commonly considered weighted Bergman space. There are ๏ฌve di๏ฌerent pairwise non-conjugate model classes of such subgroups: quasi-elliptic, quasi-parabolic, quasi-hyperbolic, nilpotent, and quasi-nilpotent (the last one depends on a parameter, giving in total ๐ + 2 model classes for the ๐-dimensional unit ball). In the case of the unit disk (๐ = 1), the above result is exact in a sense that (see for details [5]), under some technical assumption on โrichnessโ of the symbol classes, a ๐ถ โ -algebra generated by Toeplitz operators is commutative on each weighted Bergman space if and only if the symbols of generating Toeplitz operators are invariant under the action of a maximal commutative subgroup of the Mยจobius transformation of the unit disk. It was ๏ฌrmly expected that the multidimensional case preserves the regularities of the one-dimensional situation. That is, the invariance under the action of a maximal commutative subgroup of biholomorphisms for generating symbols is the only reason for appearing of Toeplitz operator algebras which are commutative on each weighted Bergman space. At the same time, quite unexpectedly it was observed in [12] that for ๐ > 1 there are many other, not geometrically de๏ฌned, classes of symbols which generate commutative Toeplitz operator algebras on each weighted Bergman space. These classes of symbols were in a sense originated from, or subordinated to the quasielliptic group, the corresponding commutative operator algebras were Banach, and being extended to ๐ถ โ -algebras they became non-commutative. Moreover, for ๐ = 1 all of them collapsed to the commutative ๐ถ โ -algebra generated by Toeplitz operators with radial symbols (one-dimensional quasi-elliptic case). It was shown then in [11] that the classes of symbols, subordinated to the quasi-parabolic group, as well generate via corresponding Toeplitz operators the Banach algebras which are commutative on each weighted Bergman space. Again being extended to ๐ถ โ algebras they became non-commutative, and for ๐ = 2 such algebras collapse to the single ๐ถ โ -algebra generated by Toeplitz operators with quasi-parabolic symbols. In this paper we consider the case of symbols subordinated to the quasihyperbolic group and show that such classes of symbols are the sources for the Banach algebras generated by Toeplitz operators which again are commutative on each weighted Bergman space. That is, together with [11, 12], we cover the multidimensional extensions of the (only) three model cases on the unit disk. The study of the last two model cases of maximal commutative subgroup of biholomorphisms of the unit ball, the nilpotent, and quasi-nilpotent groups (which appear only for ๐ > 1 and ๐ > 2, respectively), still remains as an important and interesting open question.
Banach Algebras of Commuting Toeplitz Operators
157
We mention as well that the commutativity properties of Toeplitz operators were studied in di๏ฌerent settings, for example, in [1, 2, 3, 4, 7] The paper is organized as follows. In Sections 2 and 3 we recall the notion of weighted Bergman spaces over the unit ball ๐น๐ in โ๐ and its unbounded realization as the Siegel domain ๐ท๐ . Via an explicitly given di๏ฌeomorphism from ๐ท๐ onto a half-space ๐ we identify the weighed Bergman space with a closed subspaces in ๐ฟ2 (๐, ๐๐ ) where ๐๐ is an induced measure depending on the weight parameter ๐ > โ1. In Sections 4, 5 and 6 we introduce polar type coordinates on ๐ and we explain the notion of quasi-hyperbolic symbols. Then an important result in [8] (cf. Theorem 6.1 of the present paper) establishes a unitary equivalence between Toeplitz operators acting on the weighted Bergman space over ๐ท๐ and certain explicitly given multiplication operators. Sections 7 and 8 provide the notion of hyperbolic ๐-quasi-radial and hyperbolic k-quasi-homogeneous symbols. Theorem 8.2 roughly speaking states that conjugation with the unitary operator of Theorem 6.1 transforms a Toeplitz operator having a hyperbolic k-quasi-homogeneous symbol into the product of a shift and a multiplication operator. In Section 9 we extend the results in [11, 12] to the case of Toeplitz operators with hyperbolic ๐-quasi-homogeneous symbols. In particular, we show that the Banach algebras generated by Toeplitz operators with the above hyperbolic ๐quasi-homogeneous symbols are commutative on each weighted Bergman space. A short appendix complements the text.
2. The domains ๐น๐ , ๐ซ๐ and ํ Let ๐น๐ := {๐ง = (๐ง1 , . . . , ๐ง๐ ) โ โ๐ : โฃ๐งโฃ2 = โฃ๐ง1 โฃ2 + โ
โ
โ
+ โฃ๐ง๐ โฃ2 < 1} be the unit ball in โ๐ . For points of โ๐ = โ๐โ1 ร โ we use the notation: ๐ง = (๐ง โฒ , ๐ง๐ ),
where ๐ง โฒ = (๐ง1 , . . . , ๐ง๐โ1 ) โ โ๐โ1 , ๐ง๐ โ โ.
By ๐ท๐ we denote the Siegel domain in โ๐ : { } ๐ท๐ := ๐ง = (๐ง โฒ , ๐ง๐ ) โ โ๐โ1 ร โ : Im ๐ง๐ โ โฃ๐ง โฒ โฃ2 > 0 . Recall that the Cayley transform ๐ : ๐น๐ โ ๐ท๐ is given by: ( ๐ง ) ๐ง๐โ1 1 โ ๐ง๐ ) ( 1 ๐(๐ง) = ๐ ,..., , = ๐1 , . . . , ๐๐โ1 , ๐๐ = ๐. 1 + ๐ง1 1 + ๐ง๐โ1 1 + ๐ง๐ The following result is well known: Lemma 2.1. The Cayley transform biholomorphically maps the unit ball ๐น๐ onto the Siegel Domain ๐ท๐ . The inverse transform ๐ โ1 : ๐ท๐ โ ๐น๐ is given by: ( 2๐๐1 2๐๐๐โ1 1 + ๐๐๐ ) ๐ โ1 (๐) = โ ,...,โ , . 1 โ ๐๐๐ 1 โ ๐๐๐ 1 โ ๐๐๐
158
W. Bauer and N. Vasilevski Consider the domain ๐ := โ๐โ1 ร โ ร โ+ . Then the mapping: ๐
: (๐ง โฒ , ๐ข, ๐ฃ) โ ๐ โ (๐ง โฒ , ๐ข + ๐๐ฃ + ๐โฃ๐ง โฒ โฃ2 ) โ ๐ท๐
(2.1) โฒ 2
de๏ฌnes a di๏ฌeomorphism between ๐ and ๐ท๐ . Note that Im (๐ข + ๐๐ฃ + ๐โฃ๐ง โฃ ) = ๐ฃ + โฃ๐ง โฒ โฃ2 > โฃ๐ง โฒ โฃ2 in the case of ๐ฃ โ โ+ . The inverse map ๐
โ1 : ๐ท๐ โ ๐ is given by: ( ) ๐
โ1 (๐ง โฒ , ๐ง๐ ) = ๐ง โฒ , Re ๐ง๐ , Im ๐ง๐ โ โฃ๐ง โฒ โฃ2 .
3. Weighted Bergman spaces over ๐น๐ , ๐ซ๐ , and ํ Let ๐ฃ be the standard Lebesgue measure on โ๐ โผ = โ2๐ . We write ๐ง๐ = ๐ฅ๐ + ๐๐ฆ๐ for ๐ ๐ = 1, . . . , ๐. On the ball ๐น and for ๐ > โ1 we consider the normalized weighted measure: ฮ(๐ + ๐ + 1) ๐๐๐ := ๐๐ (1 โ โฃ๐งโฃ2 )๐ ๐๐ฃ, ๐๐ := ๐ . ๐ ฮ(๐ + 1) Let ๐ be a function on ๐ท๐ , then integrals transform as follows: โซ โซ ๐ (๐) ๐ โ ๐(๐ง)๐๐ฃ(๐ง) = 22๐ ๐๐ฃ(๐). โฃ1 โ ๐๐๐ โฃ2๐+2 ๐ ๐น ๐ท๐ In particular, with ๐ โ ๐ฟ2 (๐น๐ , ๐๐ ) we have: โซ โฅ๐ โฅ2 = ๐๐ โฃ๐ (๐ง)โฃ2 (1 โ โฃ๐งโฃ2 )๐ ๐๐ฃ(๐ง) ๐น๐ โซ (1 โ โฃ๐ โ1 (๐)โฃ2 )๐ = 22๐ ๐๐ โฃ๐ โ ๐ โ1 (๐)โฃ2 ๐๐ฃ(๐) โฃ1 โ ๐๐๐ โฃ2๐+2 ๐ท๐ ( ) โซ โฒ 2 ๐ 2๐+2๐ โ1 2 Im ๐๐ โ โฃ๐ โฃ =2 ๐๐ โฃ๐ โ ๐ (๐)โฃ ๐๐ฃ(๐). โฃ1 โ ๐๐๐ โฃ2๐+2๐+2 ๐ท๐
(3.1)
We introduce the space ๐ฟ2 (๐ท๐ , ๐ ห๐ ), where the weight with respect to the Lebesgue measure is given by )๐ ๐๐ ( ๐ ห๐ (๐) = Im ๐๐ โ โฃ๐ โฒ โฃ2 . 4 From (3.1) we conclude: Corollary 3.1. The operator ๐ฐ๐ : ๐ฟ2 (๐น๐ , ๐๐ ) โ ๐ฟ2 (๐ท๐ , ๐ ห๐ ) de๏ฌned by: ( )๐+๐+1 ( ) 2 ๐ฐ๐ ๐ (๐) := ๐ โ ๐ โ1 (๐) 1 โ ๐๐๐ gives a unitary transformation of Hilbert spaces. Its inverse has the form: ( โ1 ) 1 ๐ฐ๐ ๐ (๐ง) = ๐ โ ๐(๐ง). (1 + ๐ง๐ )๐+๐+1 Proof. It is clear that ๐ฐ๐ is an isometry. The second assertion follows from: 1 1 1 = = (1 โ ๐๐๐ ). 1+๐๐ โ1 ๐ 1 + [๐ (๐)]๐ 2 1 + 1โ๐๐ ๐
โก
Banach Algebras of Commuting Toeplitz Operators
159
De๏ฌnition 3.2. With ๐ > โ1 we write ๐2๐ (๐น๐ ) and ๐2๐ (๐ท๐ ) for the weighted Bergman spaces of all analytic functions in ๐ฟ2 (๐น๐ , ๐๐ ) and ๐ฟ2 (๐ท๐ , ๐ ห๐ ), respectively. The restriction of ๐ฐ๐ to ๐2๐ (๐น๐ ) de๏ฌnes a unitary transformation onto furthermore the operator ๐ฐ๐ conjugates the corresponding weighted Bergman projections. Consider again the domain ๐ = โ๐โ1 ร โ ร โ+ . We write points ๐ค โ ๐ in the form ๐ค = (๐ง โฒ , ๐ข, ๐ฃ), where ๐ข โ โ and ๐ฃ โ โ+ . Let ๐ be a function on ๐ท๐ and ๐
: ๐ โ ๐ท๐ be the di๏ฌeomorphism (2.1). The determinant of the transformation ๐
is identically one, and therefore: โซ โซ ๐ โ ๐
(๐ง โฒ , ๐ข, ๐ฃ)๐๐ฃ(๐ค) = ๐ (๐ง)๐๐ฃ(๐ง). (3.2) ๐2๐ (๐ท๐ ),
๐
๐ท๐
De๏ฌnition 3.3. Let ๐ > โ1, then we consider the weighted space ๐ฟ2 (๐, ๐๐ ), where the weight function ๐๐ is de๏ฌned by: ๐๐ (๐ง โฒ , ๐ข, ๐ฃ) =
๐๐ ๐ ๐ฃ . 4
Moreover, let ๐0 : ๐ฟ2 (๐ท๐ , ๐ ห๐ ) โ ๐ฟ2 (๐, ๐๐ ) be the operator de๏ฌned by ๐0 ๐ := ๐ โ๐
. Let ๐ โ ๐ฟ2 (๐ท๐ , ๐ ห๐ ), then by (3.2): โซ ๐๐ โฅ๐0 ๐ โฅ2๐ฟ2 (๐,๐๐ ) = โฃ๐ โ ๐
(๐ง โฒ , ๐ข, ๐ฃ)โฃ2 ๐ฃ ๐ ๐๐ฃ(๐ค) 4 ๐ โซ ( )๐ ๐๐ = โฃ๐ (๐ง)โฃ2 Im ๐ง๐ โ โฃ๐ง โฒ โฃ2 ๐๐ฃ(๐ง) = โฅ๐ โฅ2๐ฟ2(๐ท๐ ,๐ห๐ ) . 4 ๐ท๐ It immediately follows: Lemma 3.4. The operator ๐0 is unitary with inverse ๐0โ1 = ๐0โ given by ๐0โ ๐ = ๐ โ ๐
โ1 . Consider the space ๐0 (๐) := ๐0 (๐2๐ (๐ท๐ )). It has been shown in [8] that ๐0 (๐) consists of all di๏ฌerentiable functions in ๐ฟ2 (๐, ๐๐ ) which satisfy the equations: ( ) ( ) 1 โ โ โ โ +๐ ๐=0 and โ ๐ง๐ ๐ = 0, ๐ = 1, . . . , ๐ โ 1, (3.3) 2 โ๐ข โ๐ฃ โ๐ง ๐ โ๐ฃ or the equations ( ) 1 โ โ +๐ ๐=0 2 โ๐ข โ๐ฃ
( and
) โ โ โ ๐ ๐ง๐ ๐ = 0, โ๐ง ๐ โ๐ข
๐ = 1, . . . , ๐โ 1. (3.4)
160
W. Bauer and N. Vasilevski
4. Polar-type coordinates in ํ Represent ๐ = โ๐โ1 ร โ ร โ+ in the form โ๐โ1 ร ฮ , where ฮ โ โ denotes the upper half-plane. We introduce in ๐ the non-isotropic upper semi-sphere: { } ฮฉ := (๐ง โฒ , ๐) โ โ๐โ1 ร ฮ : โฃ๐ง โฒ โฃ2 + โฃ๐โฃ = 1 . We use the following natural parametrization for points (๐ง โฒ , ๐) โ ฮฉ โ ๐: {( ) ฮฉ = ๐ 1 ๐ก1 , . . . , ๐ ๐โ1 ๐ก๐โ1 , ๐๐๐๐ : ๐ ๐ โ [0, 1), ๐ก๐ โ ๐1 , ๐ โ (0, 1], ๐ โ (0, ๐),
๐โ1 โ
} ๐ 2๐ + ๐ = 1 .
๐=1
This induces a representation of the points (๐ง โฒ , ๐) โ ๐ of the form: { 1 } ๐ = (๐ 2 ๐ง โฒ , ๐๐) : (๐ง โฒ , ๐) โ ฮฉ, ๐ โ โ+ , and we can write ๐ = ๐ (๐น๐โ1 ) ร ๐๐โ1 ร โ+ ร (0, ๐), where ๐ = ๐1 denotes the unit circle in โ and ๐ (๐น๐โ1 ) is the base of ๐น๐ in the sense of a Reinhardt domain: ๐โ1 { } โ ๐โ1 ๐โ1 ๐ (๐น ) := ๐ = (๐ 1 , . . . , ๐ ๐โ1 ) โ โ+ : ๐ 2๐ < 1 . ๐=1
Hence we can express points (๐ง โฒ , ๐) โ ๐ in the new coordinates (๐ , ๐ก, ๐, ๐) โ ๐ (๐น๐โ1 ) ร ๐๐โ1 ร โ+ ร (0, ๐), which are connected with the previous coordinates (๐ง โฒ , ๐ = ๐๐๐๐ ) by the formulas: โฃ๐ง๐ โฃ ๐ ๐ = โ , โฃ๐ง โฒ โฃ2 + โฃ๐โฃ ๐ = โฃ๐ง โฒ โฃ2 + โฃ๐โฃ,
๐ก๐ =
๐ง๐ , โฃ๐ง๐ โฃ
๐ = arg ๐,
1
or ๐ง๐ = ๐ 2 ๐ ๐ ๐ก๐ and ๐ = ๐(1 โ โฃ๐ โฃ2 )๐๐๐ , where ๐ = 1, . . . , ๐ โ 1. In these new coordinates we have: Theorem 4.1 ([8], Lemma 9.1). The equations (3.4) take the following form: for ๐ = 1, . . . , ๐ โ 1: โ๐ โ๐ 2๐ 2๐ โ๐ โ ๐ก๐ +๐ (sin ๐ + ๐ cos ๐ โ 1) , 2 โ๐ ๐ โ๐ก๐ 1 โ โฃ๐ โฃ โ๐ ๐โ1 [ ] โ๐ 2 โ๐ 1 โ โ๐ โฃ๐ โฃ 0=๐ โ ๐กโ +๐ 1+ (sin ๐ + ๐ cos ๐) . โ๐ 2 โ๐กโ 1 โ โฃ๐ โฃ2 โ๐
0 = ๐ ๐
โ=1
The space ๐0 (๐) = ๐0 (๐2๐ (๐ท๐ )) consists of all functions ๐ = ๐ (๐ , ๐ก, ๐, ๐) which satisfy the above equations and belong to: ( ) ๐ฟ2 (๐, ๐๐ ) = ๐ฟ2 ๐ (๐น๐โ1 ), (1 โ โฃ๐ โฃ2 )๐+1 ๐ ๐๐ ( ) ๐๐ โ ๐ฟ2 (๐๐โ1 ) โ ๐ฟ2 (โ+ , ๐๐+๐ ๐๐) โ ๐ฟ2 (0, ๐), sin๐ ๐๐๐ , 4 where we write ๐ ๐๐ := ๐ 1 ๐๐ 1 โ
โ
โ
๐ ๐โ1 ๐๐ ๐โ1 .
Banach Algebras of Commuting Toeplitz Operators
161
5. The operators ๐น0 and ๐น Let ๐ : ๐ฟ2 (โ+ , ๐๐+๐ ๐๐) โ ๐ฟ2 (โ) be the Mellin transform and by โฑ(๐โ1) = โฑ โ โ
โ
โ
โ โฑ we denote the (๐ โ 1)-dimensional discrete Fourier transform, where โฑ : ๐ฟ2 (๐1 ) โ โ2 (โค). More precisely, โซ [ ] ๐+๐โ1 1 ๐ ๐ (๐) : = โ ๐โ๐๐โ 2 ๐(๐) ๐๐ 2๐ โ+ โซ [ ] 1 ๐๐ก โฑ๐ (๐) : = โ ๐ (๐ก)๐กโ๐ . ๐๐ก 2๐ ๐1 Introduce the unitary operator ๐1 := ๐ผ โ โฑ(๐โ1) โ ๐ โ ๐ผ: ( ) ๐1 : ๐ฟ2 ๐ (๐น๐โ1 ), (1 โ โฃ๐ โฃ2 )๐+1 ๐ ๐๐ โ ๐ฟ2 (๐๐โ1 ) โ ๐ฟ2 (โ+ , ๐๐+๐ ๐๐) ( ) ๐๐ โ ๐ฟ2 (0, ๐), sin๐ ๐๐๐ 4 ( ) โโ ๐ฟ2 ๐ (๐น๐โ1 ), (1 โ โฃ๐ โฃ2 )๐+1 ๐ ๐๐ โ โ2 (โค๐โ1 ) โ ๐ฟ2 (โ) ( ) ๐๐ โ ๐ฟ2 (0, ๐), sin๐ ๐๐๐ . (5.1) 4 We identify the space on the right-hand side with: ( ) โฌ : = โ2 โค๐โ1 , โ where ( ) ( ) ๐๐ โ : = ๐ฟ2 (โ) โ ๐ฟ2 ๐ (๐น๐โ1 ), (1 โ โฃ๐ โฃ2 )๐+1 ๐ ๐๐ โ ๐ฟ2 (0, ๐), sin๐ ๐๐๐ . 4 Put ๐ := โค๐โ1 ร โ and ๐ := ๐ (๐น๐โ1 ) ร (0, ๐) and consider the spaces: ๐ฟ2 (๐, ๐) : = โ2 (โค๐โ1 ) โ ๐ฟ2 (โ)
( ) ๐๐ ๐ฟ2 (๐, ๐) : = ๐ฟ2 (๐ (๐น๐โ1 ), (1 โ โฃ๐ โฃ2 )๐+1 ๐ ๐๐ ) โ ๐ฟ2 (0, ๐), sin๐ ๐๐๐ . 4 2 2 De๏ฌnition 5.1. With ๐0 : ๐ฟ (๐ท๐ , ๐ ห๐ ) โ ๐ฟ (๐, ๐๐ ) this construction induces the unitary operator: ๐ := ๐1 ๐0 : ๐ฟ2 (๐ท๐ , ๐ ห๐ ) โ โฌ. Let ๐1 := โค๐โ1 ร โ โ ๐ and consider a function ๐0 = ๐0 (๐ฅ, ๐ฆ) on ๐1 ร ๐ + de๏ฌned by { } ๐0 (๐, ๐ , ๐) = ๐0 (๐, ๐, ๐ , ๐) , ๐โ1 ๐โโค+
where for ๐ โ
โค๐โ1 +
๐โ1
and (๐, ๐ , ๐) โ โ ร ๐ (๐น
) ร (0, ๐) we put
[ ]โ ๐+๐+โฃ๐โฃ+1 +๐๐ 2 ๐ฝ๐ (๐, ๐ , ๐) := ๐ ๐ 1 โ (1 + ๐)โฃ๐ โฃ2 ร๐
โ2(๐+๐
๐+๐+โฃ๐โฃ+1 2
) arctan
[( ) โฃ๐ โฃ2 1โ๐ 1โโฃ๐ โฃ2 tan
โฃ๐ โฃ2 ๐ 2 + 1โโฃ๐ โฃ2
]
, (5.2)
and we write: ๐0 (๐, ๐, ๐ , ๐) = ๐ผ๐ (๐)๐ฝ๐ (๐, ๐ , ๐) The following is shown in [8]:
and
๐ผ๐ (๐) := โฅ๐ฝ๐ (๐, โ
, โ
)โฅโ1 ๐ฟ2 (๐,๐) .
162
W. Bauer and N. Vasilevski
Proposition 5.2 ([8]). The function ๐0 has the properties (a)โ(c): (a) For each (๐, ๐) โ โค๐โ1 ร โ = ๐1 it holds ๐0 (๐, ๐, โ
, โ
) โ ๐ฟ2 (๐, ๐) and + โฅ๐0 (๐, ๐, โ
, โ
)โฅ๐ฟ2 (๐,๐) = 1. (b) ๐ maps the Bergman space ๐2๐ (๐ท๐ ) onto ๐0 ๐ฟ2 (๐1 , ๐) โ ๐ฟ2 (๐, ๐) โ ๐ฟ2 (๐, ๐): ( ) closed 2 ๐ : ๐2๐ (๐ท๐ ) โโ ๐0 ๐ฟ2 (๐1 , ๐) = ๐0 โ2 โค๐โ1 โ โฌ. + , ๐ฟ (โ) (c) For all ๐ โ ๐ฟ2 (๐1 , ๐) one has โฅ๐0 ๐ โฅ๐ฟ2 (๐,๐)โ๐ฟ2 (๐,๐) = โฅ๐ โฅ๐ฟ2 (๐1 ,๐) .
(5.3)
Now we introduce an isometric embedding:
( ๐โ1 2 2 ๐
0 : โ2 (โค๐โ1 , โ) + , ๐ฟ (โ)) โ โฌ = โ โค
by the rule:
{ } { } ๐
0 : ๐๐ (๐) ๐โโค๐โ1 โ ๐๐ (๐)๐0 (๐, ๐, ๐ , ๐) +
๐โโค๐โ1
,
(5.4)
where we put ๐๐ (๐)๐0 (๐, ๐, ๐ , ๐) = 0 if ๐ โ โค๐โ1 โ โค๐โ1 + . The adjoint operator 2 ๐
0โ : โฌ โ โ2 (โค๐โ1 , ๐ฟ (โ)) has the form: + { โซ { } โ ๐
0 : ๐๐ (๐, ๐ , ๐) ๐โโค๐โ1 โ ๐ผ๐ (๐) ๐ฝ๐ (๐, ๐ , ๐)๐๐ (๐, ๐ , ๐) ๐ (๐น๐โ1 )ร(0,๐)
} ๐๐ ร(1 โ โฃ๐ โฃ2 )๐+1 sin๐ ๐๐ ๐๐ ๐๐ . 4 ๐โโค๐โ1 +
(5.5)
One easily checks that ๐
0โ ๐
0 = ๐ผ : ๐ฟ2 (๐1 , ๐) โโ ๐ฟ2 (๐1 , ๐), ๐
0 ๐
0โ = ๐ : ๐ฟ2 (๐, ๐) โ ๐ฟ2 (๐, ๐) โโ ๐ (๐2๐ (๐ท๐ )) = ๐0 ๐ฟ2 (๐1 , ๐), where ๐ is the orthogonal projection onto the right-hand side. Theorem 5.3 ([8]). The operator ๐
:= ๐
0โ ๐ maps the Hilbert space ๐ป := ๐ฟ2 (๐ท๐ , ๐ ห๐ ) onto ๐ฟ2 (๐1 , ๐). The restriction and the adjoint operator: 2 ๐
โฃ๐ : ๐ : = ๐2๐ (๐ท๐ ) โโ ๐ฟ2 (๐1 , ๐) = โ2 (โค๐โ1 + , ๐ฟ (โ))
๐
โ = ๐ โ ๐
0 : ๐ฟ2 (๐1 , ๐) โโ ๐ โ ๐ป are isometric isomorphisms. Furthermore, ๐
๐
โ = ๐ผ : ๐ฟ2 (๐1 , ๐) โโ ๐ฟ2 (๐1 , ๐), ๐
โ ๐
= ๐ : ๐ป โโ ๐, where ๐ is the orthogonal projection of ๐ป onto ๐. Proof. Proposition 9.3. in [8].
โก
Banach Algebras of Commuting Toeplitz Operators
163
6. Toeplitz operators with quasi-hyperbolic symbols Recall that the Toeplitz operator ๐๐ with symbol ๐ โ ๐ฟโ (๐ท๐ ) acts on the weighted Bergman space ๐2๐ (๐ท๐ ) by the rule ๐๐ ๐ = ๐ต๐ท๐ ,๐ (๐๐), where ๐ต๐ท๐ ,๐ is the Bergman orthogonal projection of the space ๐ฟ2 (๐ท๐ , ๐ ห๐ ) onto the Bergman space ๐2๐ (๐ท๐ ). A bounded measurable symbol ๐ : ๐ท๐ โ โ is called quasi-hyperbolic if ๐ is invariant under the action of the quasi-hyperbolic group ๐๐โ1 ร โ+ acting on ๐ท๐ by: 1
๐๐โ1 ร โ+ โ (๐ก, ๐) : (๐ง โฒ , ๐ง๐ ) โ (๐ 2 ๐ก๐ง โฒ , ๐๐ง๐ ). Consider the group of non-isotropic dilations {๐ฟ๐ }๐โโ+ acting on โ๐โ1 รฮ + by the rule ( 1 ) 1 ๐ฟ๐ : (๐1 , . . . , ๐๐โ1 , ๐) โ ๐ 2 ๐1 , . . . , ๐ 2 ๐๐โ1 , ๐๐ . A function ๐ ห = ๐ ห(๐1 , . . . , ๐๐โ1 , ๐) is non-isotropic homogeneous of zero order on โ๐โ1 ร ฮ if it is invariant under ๐ฟ๐ , i.e., it can be recovered from its restriction to + the non-isotropic half-sphere ๐โ1 { } โ 2 ฮฉ+ := (๐1 , . . . , ๐๐โ1 , ๐) โ โ๐โ1 ร ฮ : ๐ + โฃ๐โฃ = 1 . + ๐ ๐=1
On the one hand, note that a function ๐ on ๐ท๐ is quasi-hyperbolic if and only if it has the form: ( ) ๐(๐ง โฒ , ๐ง๐ ) = ๐ห โ ๐
โ1 โฃ๐ง1 โฃ, . . . , โฃ๐ง๐โ1 โฃ, ๐ง๐ with a function ๐ ห which is non-isotropic homogeneous of zero order on โ๐โ1 + รฮ . On the other hand the non-isotropic homogeneous functions of zero order on โ๐โ1 รฮ + are of the type ( ) ๐1 ๐๐โ1 โฃ๐โฃ ๐๐ ๐ ห(๐1 , . . . , ๐๐โ1 , ๐) = ๐0 โ ,..., โ , ๐ โฃ๐โฃ2 + โฃ๐โฃ โฃ๐โฃ2 + โฃ๐โฃ โฃ๐โฃ2 + โฃ๐โฃ =ห ๐0 (๐ 1 , . . . , ๐ ๐โ1 , ๐) in our former coordinates (๐ 1 , ๐ 2 , . . . , ๐ ๐โ1 ) and ๐ and with a function ห ๐0 on ๐ (๐น๐โ1 ) ร (0, ๐). According to Theorem 10.5 in [8]: Theorem 6.1. Let ๐ โ ๐ฟโ (๐ท๐ ) be a quasi-hyperbolic function. Then the Toeplitz operator ๐๐ acting on ๐2๐ (๐ท๐ ) is unitary equivalent to the multiplication operator: ( ) ( ๐โ1 2 ) 2 2 ๐พ๐ ๐ผ = ๐
๐๐๐
โ : โ2 โค๐โ1 + , ๐ฟ (โ) โโ โ โค+ , ๐ฟ (โ) . The sequence ๐พ๐ = {๐พ๐ (๐, ๐)}๐โโค๐โ1 with ๐ โ โ is given by: + โซ ๐๐ ๐พ๐ (๐, ๐) = ๐ผ2๐ (๐) ๐(๐ , ๐)โฃ๐ฝ๐ (๐ , ๐, ๐)โฃ2 (1 โ โฃ๐ โฃ2 )๐+1 sin๐ ๐๐ ๐๐ ๐๐, 4 ๐ (๐น๐โ1 )ร(0,๐) where ๐ฝ๐ was de๏ฌned in (5.2).
164
W. Bauer and N. Vasilevski
7. Hyperbolic ๐-quasi-radial symbols Let ๐ = (๐1 , . . . , ๐๐ ) be a tuple of positive integers such that ๐1 โค ๐2 โค โ
โ
โ
โค ๐๐ and โฃ๐โฃ = ๐1 + โ
โ
โ
+ ๐๐ = ๐ โ 1. We arrange the coordinates of โ๐โ1 in ๐ groups: ๐ง(1) = (๐ง1,1 , . . . , ๐ง1,๐1 ), ๐ง(2) = (๐ง2,1 , . . . , ๐ง2,๐2 ), . . . , ๐ง(๐) = (๐ง๐,1 , . . . , ๐ง๐,๐๐ ). ๐(๐)
Represent each ๐ง(๐) = (๐ง๐,1 , . . . , ๐ง๐,๐๐ ) โ โ๐๐ in the form ๐ง(๐) = ๐๐ ๐(๐) , where โ ๐2๐๐ โ1 and โ ๐๐ = โฃ๐ง๐,1 โฃ2 + โ
โ
โ
+ โฃ๐ง๐,๐๐ โฃ2 .
De๏ฌnition 7.1. A function ๐ = ๐(๐ง โฒ , ๐ง๐ ) : ๐ท๐ โ โ is called hyperbolic k-quasiradial if ๐(๐ง โฒ , ๐ง๐ ) = ๐ ห(๐1 , . . . , ๐๐ , ๐ง๐ โ ๐โฃ๐ง โฒ โฃ2 )
(7.1) โ๐ +
with a function ๐ ห which is non-isotropic homogeneous of order zero on ร ฮ . In that case ๐ is, in particular, quasi-hyperbolic and ๐ ห can be represented in the form: ( ) ๐1 ๐๐ โฃ๐โฃ ๐ ห(๐1 , . . . , ๐๐ , ๐) = ๐0 โ ,..., โ , ๐๐๐ (7.2) โฃ๐โฃ2 + โฃ๐โฃ โฃ๐โฃ2 + โฃ๐โฃ โฃ๐โฃ2 + โฃ๐โฃ =๐ ห0 (๐ 1 , . . . , ๐ ๐ , ๐) , where ๐ = (๐1 , . . . , ๐๐ ) โ โ๐ ห0 is a function on ๐ (๐น๐ ) ร (0, ๐). + and ๐ By varying the tuple ๐ we have a collection of sets โ๐ of hyperbolic ๐-quasiradial functions. This collection is partially ordered by inclusion and we have: โ(๐โ1) โ โ๐ โ โ(1,...,1) . With a given multi-index ๐ผ = (๐ผ1 , . . . , ๐ผ๐โ1 ) โ โ๐โ1 we write: 0 ๐ผ(1) = (๐ผ1 , . . . , ๐ผ๐1 ), ๐ผ(2) = (๐ผ๐1 +1 , . . . , ๐ผ๐1 +๐2 ), . . . , ๐ผ(๐) = (๐ผ๐โ๐๐ , . . . , ๐ผ๐โ1 ). For a hyperbolic ๐-quasi-radial function ๐ we can further reduce the order of integration in the expression ๐พ๐ (๐, ๐), where ๐ โ โค๐โ1 + , in Theorem 6.1. With ๐ = (๐ 1 , . . . , ๐ ๐โ1 ) โ ๐น๐โ1 put ๐ ห := (โฃ๐ 1 โฃ, . . . , โฃ๐ ๐โ1 โฃ) โ ๐ (๐น๐โ1 ) and ๐ := (1, 1, . . . , 1). With a suitable function ๐ปโฃ๐โฃ : โ ร โ+ ร (0, ๐) โ โ and with ๐ โ โค๐โ1 we can write + ๐ฝ๐ (๐, ๐ ห, ๐)(1 โ โฃ๐ โฃ2 )
๐+1 2
( ) = ๐ ๐ โ
๐ปโฃ๐โฃ ๐, โฃ๐ โฃ, ๐ .
(7.3)
Banach Algebras of Commuting Toeplitz Operators
165
Hence one has: โซ
$ $2 ๐ ห0 (๐ , ๐) $๐ฝ๐ (๐, ๐ , ๐)$ (1 โ โฃ๐ โฃ2 )๐+1 ๐ ๐๐ ๐ (๐น๐โ1 ) โซ $ ( )$2 1 = ๐โ1 ๐ ห0 (ห ๐ , ๐) โฃ๐ 2๐+๐ โฃ$๐ปโฃ๐โฃ ๐, โฃ๐ โฃ, ๐ $ ๐๐ = (โ). 2 ๐น๐โ1 If ๐ is hyperbolic ๐-quasi-radial, we obtain: โซ โซ $ $2 1 (โ) = ๐โ1 ๐ ห0 (๐, ๐)โฃ๐ 2๐+๐ โฃ$๐ปโฃ๐โฃ (๐, โฃ๐โฃ, ๐)$ 2 ๐ (๐น๐ ) ๐๐1 โ1 รโ
โ
โ
ร๐๐๐ โ1 ร
๐ โ ๐=1
2โฃ๐(๐) โฃ+๐๐ โ1
๐๐
๐๐(๐ (1) ) โ
โ
โ
๐๐(๐ (๐) ) ๐๐.
Here and in what follows ๐๐ means the usual surface measure on the sphere. From Lemma A.1 we have: ( )โ1 ๐ โซ ๐ โ โ $ 2๐ +๐ $ ๐๐ + 1 ๐ (๐) (๐) $ $ ฮ๐ := ๐ ๐๐(๐ ) = 2 ๐! ฮ โฃ๐(๐) โฃ + , ๐๐ โ1 2 ๐=1 ๐ ๐=1 and it follows: Lemma 7.2. Let ๐ be hyperbolic ๐-quasi-radial and ๐ โ โค๐โ1 + , then: ( ) ๐น๐ โฃ๐(1) โฃ, . . . , โฃ๐(๐) โฃ, ๐ ( ), ๐พ๐ (๐, ๐) = ๐น๐ โฃ๐(1) โฃ, . . . , โฃ๐(๐) โฃ, ๐
(7.4)
where we have used the notation in (7.1) and (7.2) and we put ๐ โก 1. The function ๐น๐ is de๏ฌned by: โซ ( ) $ $2 ๐น๐ โฃ๐(1) โฃ, . . . , โฃ๐(๐) โฃ, ๐ := ๐ ห0 (๐, ๐)$๐ปโฃ๐โฃ (๐, โฃ๐โฃ, ๐)$ ๐ (๐น๐ )ร(0,๐) ๐ โ 2โฃ๐ โฃ+๐๐ โ1 ๐๐ ร ๐๐ (๐) 4 ๐=1
sin๐ ๐๐๐๐๐.
(7.5)
Proof. From our calculation before and with the notation (7.5) we have: ( )โ1 ( ๐ ) โ ๐๐ + 1 2 ๐โ๐+1 ๐พ๐ (๐, ๐) = ๐ผ๐ (๐)2 ๐! ฮ โฃ๐(๐) โฃ + ๐น๐ โฃ๐(1) โฃ, . . . , โฃ๐(๐) โฃ, ๐ . 2 ๐=1 Moreover, it holds 2 ๐ผโ2 ๐ (๐) = โฅ๐ฝ๐ (๐, โ
, โ
)โฅ๐ฟ2 (๐,๐) โซ ๐๐ = โฃ๐ฝ๐ (๐, ๐ , ๐)โฃ2 (1 โ โฃ๐ โฃ2 )๐+1 sin๐ ๐๐ ๐๐ ๐ 4 ๐ (๐น๐โ1 )ร(0,๐) ( )โ1 ๐ โ ( ) ๐๐ + 1 = 2๐โ๐+1 ๐! ฮ โฃ๐(๐) โฃ + ๐น๐ โฃ๐(1) โฃ, . . . , โฃ๐(๐) โฃ, ๐ , 2 ๐=1
which proves (7.4).
โก
166
W. Bauer and N. Vasilevski
8. Hyperbolic ๐-quasi-homogeneous functions โฒ ๐โ1 Let ๐ = (๐1 , . . . , ๐๐ ) โ โค๐ we write + . With a point ๐ง = (๐ง1 , . . . , ๐ง๐โ1 ) โ โ ๐ ๐ง(๐) = ๐๐ ๐(๐) , ๐ = 1, . . . , ๐, to de๏ฌne the vectors (๐1 , . . . , ๐๐ ) โ โ+ and ( ๐ := (๐1 , . . . , ๐๐โ1 ) = ๐(1) , ๐(2) , . . . , ๐(๐) ) โ ๐2๐1 โ1 ร ๐2๐2 โ1 ร โ
โ
โ
ร ๐2๐๐ โ1 .
A second representation of ๐ง โฒ has been given earlier: ( 1 ) 1 ๐ง โฒ = ๐ 2 ๐ 1 ๐ก1 , . . . , ๐ 2 ๐ ๐โ1 ๐ก๐โ1 , 1
๐ โ โ+ ,
1
where ๐ก = (๐ก1 , . . . , ๐ก๐โ1 ) โ ๐ ร โ
โ
โ
ร ๐ and ๐ = (๐ 1 , . . . , ๐ ๐โ1 ) โ ๐ (๐น๐โ1 ). Hence it follows: 1 ๐ง๐,โ = ๐๐ ๐๐,โ = ๐ 2 ๐ ๐,โ ๐ก๐,โ , for โ โ {1, . . . , ๐๐ }, and therefore 1
๐๐,โ =
1
๐2 ๐ ๐,โ ๐ก๐,โ ๐๐
and 1
Moreover, we have ๐๐ = โฃ๐ง(๐) โฃ = ๐ 2 ๐ (๐) โ= 0: ๐ ๐,โ ๐๐,โ = ๐ก๐,โ โฃ๐ (๐) โฃ
โ
โฃ๐๐,โ โฃ =
๐2 ๐ ๐,โ . ๐๐ 1
๐ 2๐,1 + โ
โ
โ
+ ๐ 2๐,๐๐ = ๐ 2 โฃ๐ (๐) โฃ, and in case of and
โฃ๐๐,โ โฃ =
๐ ๐,โ . โฃ๐ (๐) โฃ
(8.1)
De๏ฌnition 8.1. A function ๐ โ ๐ฟโ (๐ท๐ ) is called hyperbolic ๐-quasi-homogeneous if it has the form ( ) ๐ ๐(๐ง โฒ , ๐ง๐ ) = ๐ ห โฃ๐ง(1) โฃ, . . . , โฃ๐ง(๐) โฃ, ๐ง๐ โ ๐โฃ๐ง โฒ โฃ2 ๐ ๐ ๐ , where ๐ ห is non-isotropic homogeneous of order zero on โ๐ + ร ฮ . We call the pair (๐, ๐) โ โค๐โ1 ร โค๐โ1 the corresponding quasi-homogeneous degree. + + According to (8.1) we write: ๐ ( ) โ ๐ ๐(๐ง โฒ , ๐ง๐ ) = ๐ ห0 โฃ๐ (1) โฃ, . . . , โฃ๐ (๐) โฃ, arg(๐ง๐ โ ๐โฃ๐ง โฒ โฃ2 ) ๐ก๐ ๐ก ๐ ๐+๐ โฃ๐ (๐) โฃโโฃ๐(๐) โฃโโฃ๐(๐) โฃ ,
๐=1
=๐
where ๐ ห0 is a function on ๐ (๐น๐ ) ร (0, ๐) and for ๐ = 1, . . . , ๐ we have โฃ๐ (๐) โฃ = โ
โฃ๐ง(๐) โฃ โฃ๐ง โฒ โฃ2
+ โฃ๐ง๐ โ ๐โฃ๐ง โฒ โฃ2 โฃ
.
For a multi-index ๐ โ โค๐โ1 we denote by ๐ห๐ = {๐ฟ๐,๐ฝ }๐ฝโโค๐โ1 the ๐โs element + +
2 of the standard orthonormal basis in โ2 (โค๐โ1 + ). Given ๐(๐) โ ๐ฟ (โ), let ( ) { } ๐ห๐ ๐(๐) = ๐ห๐ โ ๐(๐) = ๐ฟ๐,๐ฝ ๐(๐) ๐ฝโโค๐โ1 +
be the corresponding one-component element of โ
2
2 (โค๐โ1 + , ๐ฟ (โ)).
Banach Algebras of Commuting Toeplitz Operators
167 ๐
Theorem 8.2. Given a hyperbolic ๐-quasi-homogeneous symbol ๐ = ๐๐ ๐ ๐ we have: โง 0, if there is an โ such that ๏ฃด ๏ฃด ๏ฃด โจ ๐โ + ๐โ โ ๐โ < 0, ๐
๐๐ ๐
โ : ๐ห๐ (๐(๐)) โ ๐ ๏ฃด ๐พ ห (๐)ห ๐ (๐(๐)), if for all โ one has ๐+๐โ๐ ๏ฃด ๏ฃด ๐,๐,๐,๐ โฉ ๐โ + ๐โ โ ๐โ โฅ 0, where ๐ ๐พห๐,๐,๐,๐ (๐)
= ฮ๐+๐ ๐ผ๐+๐โ๐ (๐)๐ผ๐ (๐) ร
๐ โ ๐=1
โซ
1 2๐โ1
2โฃ๐(๐) โฃ+โฃ๐(๐) โฃโโฃ๐(๐) โฃ+๐๐ โ1
๐๐
๐ (๐น๐ )ร(0,๐)
๐ ห0 (๐1 , . . . , ๐๐ , ๐)
[ ] ๐๐ ร ๐ปโฃ๐+๐โ๐โฃ โ
๐ปโฃ๐โฃ (๐, โฃ๐โฃ, ๐) sin๐ ๐๐๐๐๐. 4
Here we use the notation (7.3) and write as before: ๐ โซ โ 2๐ +2๐ +๐ ฮ๐+๐ = โฃ๐พ(๐)(๐) (๐) (๐) โฃ๐๐(๐พ(๐) ) ๐=1
๐๐๐ โ1
๐
= 2 (๐ + ๐)!
๐ โ ๐=1
( )โ1 ๐๐ + 1 ฮ โฃ๐(๐) โฃ + โฃ๐(๐) โฃ + . 2
Proof. Similar to the proof of Theorem 10.5 in [8] we have: ( ) ๐
๐๐ ๐
โ ๐ห๐ ๐(๐) ๐ โ ( ) ๐ ( ) โโฃ๐ +๐ โฃ = ๐
0โ ๐1 ๐ ห0 ๐ (1) , . . . , ๐ (๐) , ๐ ๐ก๐ ๐ก ๐ ๐+๐ ๐ (๐) (๐) (๐) ๐1โ1 ๐
0 ๐ห๐ ๐(๐)
=
๐
0โ ๐1 ๐ ห0
(
) ๐ ๐ (1) , . . . , ๐ (๐) , ๐ ๐ก๐ ๐ก ๐ ๐+๐ ร
๐ โ ๐=1
โโฃ๐
๐ (๐) (๐)
+๐(๐) โฃ
๐=1
{ } ๐1โ1 ๐ห๐ ๐(๐)๐ผ๐ (๐)๐ฝ๐ (๐, ๐ , ๐)
๐ { ( } โ ) โโฃ๐ +๐ โฃ = ๐
0โ ๐ห๐+๐โ๐ ๐ ห0 ๐ (1) , . . . , ๐ (๐) , ๐ ๐ ๐+๐ ๐ (๐) (๐) (๐) ๐(๐)๐ผ๐ (๐)๐ฝ๐ (๐, ๐ , ๐) , ๐=1
2 where ๐
0 : โ2 (โค๐โ1 + , ๐ฟ (โ)) โ โฌ and ๐1 = ๐ผ โ โฑ(๐โ1) โ ๐ โ ๐ผ have been de๏ฌned in (5.4) and (5.1), respectively. Hence we obtain from (5.5) that { ( ) ๐
๐๐๐ ๐ ๐ ๐ ๐
โ ๐ห๐ ๐(๐) = ๐ห๐+๐โ๐ ๐(๐)๐ผ๐+๐โ๐ (๐)๐ผ๐ (๐) โซ ( ) ร ๐ ห0 ๐ (1) , . . . , ๐ (๐) , ๐
ร ๐ฝ๐+๐โ๐ (๐, ๐ , ๐)
๐ (๐น๐โ1 )ร(0,๐) ๐ โ โโฃ๐ +๐ โฃ ๐+๐ ๐ฝ๐ (๐, ๐ , ๐)๐ ๐ (๐) (๐) (๐) (1 ๐=1
โ โฃ๐ โฃ2 )๐+1
} ๐๐ sin๐ ๐๐ ๐๐ ๐๐ . 4
168
W. Bauer and N. Vasilevski Again we de๏ฌne ๐ปโฃ๐โฃ by the relation ๐ฝ๐ (๐, ๐ , ๐)(1 โ โฃ๐ โฃ2 )
๐+1 2
= ๐ ๐ ๐ปโฃ๐โฃ (๐, โฃ๐ โฃ, ๐)
and we put ๐ = (1, . . . , 1) โ โค๐โ1 such that: + โซ ( ) ๐ ห0 โฃ๐ (1) โฃ, . . . , โฃ๐ (๐) โฃ, ๐ ๐ฝ๐+๐โ๐ (๐, ๐ , ๐)๐ฝ๐ (๐, ๐ , ๐)๐ ๐+๐ ๐ (๐น๐โ1 )
ร โซ =
โฃ๐ (๐) โฃโโฃ๐(๐) +๐(๐) โฃ (1 โ โฃ๐ โฃ2 )๐+1 ๐ ๐๐
๐=1
[ ] ) ๐ ห0 โฃ๐ (1) โฃ, . . . , โฃ๐ (๐) โฃ, ๐ ๐ 2๐+2๐+๐ ๐ปโฃ๐+๐โ๐โฃ โ
๐ปโฃ๐โฃ (๐, โฃ๐ โฃ, ๐) (
๐ (๐น๐โ1 )
๐ โ
ร
๐ โ
โฃ๐ (๐) โฃโโฃ๐(๐) +๐(๐) โฃ ๐๐ = (โ).
๐=1
It follows that: โซ ] $[ ( )$ 1 (โ) = ๐โ1 ๐ ห0 โฃ๐ (1) โฃ, . . . , โฃ๐ (๐) โฃ, ๐ $๐ 2๐+2๐+๐ $ ๐ปโฃ๐+๐โ๐โฃ โ
๐ปโฃ๐โฃ (๐, โฃ๐ โฃ, ๐) 2 ๐น๐โ1 ๐ โ ร โฃ๐ (๐) โฃโโฃ๐(๐) +๐(๐) โฃ ๐๐ ๐=1
=
1
โซ
2๐โ1 ร
๐ โ ๐=1
โซ
๐ (๐น๐ )
๐๐1 โ1 รโ
โ
โ
ร๐๐๐ โ1
๐ ห0 (๐1 , . . . , ๐๐ , ๐)
2โฃ๐(๐) โฃ+โฃ๐(๐) โฃโโฃ๐(๐) โฃ+๐๐ โ1 $ 2๐+2๐+๐ $ $ $
๐๐
๐พ
[ ] ร ๐ปโฃ๐+๐โ๐โฃ โ
๐ปโฃ๐โฃ (๐, โฃ๐โฃ, ๐) ๐๐(๐พ1 ) โ
โ
โ
๐๐(๐พ๐ )๐๐ โซ ๐ โ ฮ๐+๐ 2โฃ๐ โฃ+โฃ๐(๐) โฃโโฃ๐(๐) โฃ+๐๐ โ1 = ๐โ1 ๐ ห0 (๐1 , . . . , ๐๐ , ๐) ๐๐ (๐) 2 ๐ (๐น๐ ) ๐=1 [ ] ร ๐ปโฃ๐+๐โ๐โฃ โ
๐ปโฃ๐โฃ (๐, โฃ๐โฃ, ๐) ๐๐, which proves the assertion.
โก
9. Commutativity results Now we have collected all the tools to extend the results in [11, 12] to the case of Toeplitz operators with hyperbolic ๐-quasi-homogeneous symbols. Proposition 9.1. Let ๐ = (๐1 , ๐2 , . . . , ๐๐ ) โ โค๐ + and ๐, ๐ be a pair of orthogonal multi-indices. Then, (a) and (b) below are equivalent:
Banach Algebras of Commuting Toeplitz Operators
169
(a) For each pair of non identically zero hyperbolic ๐-quasi-radial functions ๐1 and ๐2 the Toeplitz operators ๐๐1 and ๐๐2 ๐ ๐ ๐ ๐ commute on each weighted Bergman space ๐2๐ (๐ท๐ ). (b) It holds โฃ๐(๐) โฃ = โฃ๐(๐) โฃ for each ๐ = 1, 2, . . . , ๐. Proof. We calculate the operator products in both orders using the Theorems 6.1 and 8.2. On the one hand and according to Theorem 5.3 we have ( )( ) ๐
๐๐2 ๐ ๐ ๐ ๐ ๐๐1 ๐
โ ๐ห๐ (๐(๐)) = ๐
๐๐2 ๐ ๐ ๐ ๐ ๐
โ ๐
๐๐1 ๐
โ ๐ห๐ (๐(๐)) ( ) = ๐
๐๐2 ๐ ๐ ๐ ๐ ๐
โ ๐ห๐ ๐พ๐1 (๐, ๐)๐(๐) ( ) ๐2 = ๐พห๐,๐,๐,๐ (๐)๐พ๐1 (๐, ๐)ห ๐๐+๐โ๐ ๐(๐) (for all โ such that ๐โ + ๐โ โ ๐โ โฅ 0). On the other hand: )( ) ( ( ) ( ) ๐
๐๐1 ๐๐2 ๐ ๐ ๐ ๐ ๐
โ ๐ห๐ ๐(๐) = ๐
๐๐1 ๐
โ ๐
๐๐2 ๐ ๐ ๐ ๐ ๐
โ ๐ห๐ ๐(๐) ( ) ( ) ๐2 = ๐
๐๐1 ๐
โ ๐ห๐+๐โ๐ ๐พห๐,๐,๐,๐ (๐)๐(๐) ( ) ๐2 = ๐พ๐1 (๐ + ๐ โ ๐, ๐)ห ๐พ๐,๐,๐,๐ (๐)ห ๐๐+๐โ๐ ๐(๐) . Hence both operators commute if and only if: ๐พ๐1 (๐, ๐) = ๐พ๐1 (๐ + ๐ โ ๐, ๐). According to Lemma 7.2 and with the notation (7.5) this means: ( ) ( ) ๐น๐1 โฃ๐(1) โฃ, . . . , โฃ๐(๐) โฃ, ๐ ๐น๐1 โฃ๐(1) + ๐(1) โ ๐(1) โฃ, . . . , โฃ๐(๐) + ๐(๐) โ ๐(๐) โฃ, ๐ ( ) = ( ). ๐น๐ โฃ๐(1) โฃ, . . . , โฃ๐(๐) โฃ, ๐ ๐น๐ โฃ๐(1) + ๐(1) โ ๐(1) โฃ, . . . , โฃ๐(๐) + ๐(๐) โ ๐(๐) โฃ, ๐ This relation is ful๏ฌlled for all possible symbols ๐1 if and only if (b) holds.
โก
Note that under the condition โฃ๐(๐) โฃ = โฃ๐(๐) โฃ, for each ๐ = 1, . . . , ๐, and with a hyperbolic ๐-quasi radial symbol ๐ we have: โซ ฮ๐+๐ ๐ผ๐+๐โ๐ (๐) ฮ๐ ๐ผ2๐ (๐) ๐ ๐พห๐,๐,๐,๐ (๐) = ๐ ห(๐, ๐) ฮ๐ ๐ผ๐ (๐) 2๐โ1 ๐ (๐น๐โ1 )ร(0,๐) ร
๐ โ ๐=1
=
2โฃ๐(๐) โฃ+๐๐ โ1 $
$ $๐ปโฃ๐โฃ (๐, โฃ๐โฃ, ๐)$2 ๐๐ sin๐ ๐๐๐๐๐ 4
๐๐
ฮ๐+๐ ๐ผ๐+๐โ๐ (๐) ๐พ๐ (๐, ๐). ฮ๐ ๐ผ๐ (๐)
Moreover,
โ ฮ๐+๐ ๐ผ๐+๐โ๐ (๐) ฮ๐+๐ ๐! โ = ฮ๐ ๐ผ๐ (๐) ฮ๐ (๐ + ๐ โ ๐)!
( ) ๐ +1 ฮ โฃ๐(๐) โฃ + ๐2 ( ). = โ ๐!(๐ + ๐ โ ๐)! ๐=1 ฮ โฃ๐(๐) + ๐(๐) โฃ + ๐๐ +1 2 (๐ + ๐)!
๐ โ
170
W. Bauer and N. Vasilevski
Thus we have: Lemma 9.2. Let ๐ = (๐1 , . . . , ๐๐ ) and ๐, ๐ โ โค๐โ1 be orthogonal multi-indices such + ๐โ1 that โฃ๐(๐) โฃ = โฃ๐(๐) โฃ for ๐ = 1, . . . , ๐. With ๐ โ โค+ and a hyperbolic ๐-quasi-radial symbols ๐ one has: ( ) ๐๐ +1 ๐ ฮ โฃ๐ โฃ + โ (๐) 2 (๐ + ๐)! ๐ ( ). ๐พห๐,๐,๐,๐ (๐) = ๐พ๐ (๐, ๐) โ
โ (9.1) ๐!(๐ + ๐ โ ๐)! ๐=1 ฮ โฃ๐(๐) + ๐(๐) โฃ + ๐๐ +1 2 From this result we conclude: Corollary 9.3. Let ๐ = (๐1 , ๐2 , . . . , ๐๐ ) โ โค๐ + be given. For each pair of orthogonal multi-indices ๐ and ๐ with โฃ๐(๐) โฃ = โฃ๐(๐) โฃ for all ๐ = 1, 2, . . . , ๐ and a hyperbolic ๐-quasi-radial function ๐ we have: ๐๐ ๐๐ ๐ ๐ ๐ = ๐๐ ๐ ๐ ๐ ๐๐ = ๐๐๐ ๐ ๐ ๐ . Proof. The ๏ฌrst equality directly follows from Proposition 9.1. Moreover, with the symbol ๐ โก 1 we have ๐พ๐ (๐, ๐) โก 1 for all multi-indices ๐ โ โค๐โ1 + . Thus by (9.1) ( ) ๐๐ +1 ๐ ฮ โฃ๐ โฃ + โ (๐) 2 (๐ + ๐)! ๐ ( ). ๐พห๐,๐,๐,๐ (๐) = โ (9.2) ๐!(๐ + ๐ โ ๐)! ๐=1 ฮ โฃ๐(๐) + ๐(๐) โฃ + ๐๐ +1 2 In other words, one has ๐ ๐ ๐พห๐,๐,๐,๐ (๐) = ๐พ๐ (๐, ๐) โ
๐พห๐,๐,๐,๐ (๐),
(9.3)
which together with the calculations in the proof of Proposition 9.1 implies the assertion. โก Given ๐ = (๐1 , ๐2 , . . . , ๐๐ ) and a pair of orthogonal multi-indices ๐ and ๐ with โฃ๐(๐) โฃ = โฃ๐(๐) โฃ, for all ๐ = 1, 2, . . . , ๐, put ๐ห(๐) := (0, . . . , 0, ๐(๐) , 0, . . . , 0)
and
๐ห(๐) := (0, . . . , 0, ๐(๐) , 0, . . . , 0).
Then of course ๐ = ๐ห(1) + ๐ห(2) + โ
โ
โ
+ ๐ห(๐) and ๐ = ๐ห(1) + ๐ห(2) + โ
โ
โ
+ ๐ห(๐) . For each ๐ = 1, 2, . . . , ๐ we introduce the Toeplitz operator ๐๐ := ๐๐ ๐ห(๐) ๐ ๐ห(๐) . Corollary 9.4. The operators ๐๐ for ๐ = 1, 2, . . . , ๐ mutually commute. Given an โ-tuple of indices (๐1 , ๐2 , . . . , ๐โ ) where 2 โค โ โค ๐ and let ๐หโ = ๐ห(๐1 ) + ๐ห(๐2 ) + โ
โ
โ
+ ๐ห(๐โ )
and
๐หโ = ๐ห(๐1 ) + ๐ห(๐2 ) + โ
โ
โ
+ ๐ห(๐โ ) .
Under the condition โฃ๐(๐) โฃ = โฃ๐(๐) โฃ, for all ๐ = 1, 2, . . . , ๐, it holds โ โ ๐=1
๐๐๐ = ๐๐ ๐หโ ๐ ๐หโ .
Banach Algebras of Commuting Toeplitz Operators
171
Proof. Let ๐ โก 1, then it is su๏ฌcient to show that for ๐ โ= โ: ๐ ๐ ๐พห๐,๐, ๐พ๐+ ๐ห(๐) ,ห ๐(๐) (๐)โ
ห ๐ห(๐) โห ๐(๐) ,๐,๐ห(โ) ,ห ๐(โ) (๐) ๐ ๐ =ห ๐พ๐,๐, ห๐+ ๐ห(โ) ,ห ๐(โ) (๐) โ
๐พ ๐ห(โ) โห ๐(โ) ,๐,๐ห(๐) ,ห ๐(๐) (๐) ๐ =ห ๐พ๐,๐, ๐ห(๐) +๐ห(โ) ,ห ๐(๐) +ห ๐(โ) (๐).
We calculate the ๏ฌrst product by using (9.2): ๐ ๐ ๐พห๐,๐, ๐พ๐+ ๐ห(๐) ,ห ๐(๐) (๐)ห ๐ห(๐) โห ๐(๐) ,๐,๐ห(โ) ,ห ๐(โ) (๐) ( ) ( ) ๐ +1 ฮ โฃ๐(๐) โฃ + ๐2 ฮ โฃ๐(โ) โฃ + ๐โ2+1 ) ( ) = ( ๐ +1 ฮ โฃ๐(๐) + ๐(๐) โฃ + ๐2 ฮ โฃ๐(โ) + ๐(โ) โฃ + ๐โ2+1
(๐ + ๐ห(๐) )! (๐ + ๐ห(๐) โ ๐ห(๐) + ๐ห(โ) )! โ รโ . ๐!(๐ + ๐ห(๐) โ ๐ห(๐) )! (๐ + ๐ห(๐) โ ๐ห(๐) )!(๐ + ๐ห(๐) โ ๐ห(๐) + ๐ห(โ) โ ๐ห(โ) )!
=:๐ด๐,โ
Note that (๐(๐) + ๐(๐) )!๐(โ) ! ๐ด๐,โ = ๐ถ โ ๐(๐) !(๐(๐) + ๐(๐) โ ๐(๐) )!๐(โ) !๐(โ) ! (๐(๐) + ๐(๐) โ ๐(๐) )!(๐(โ) + ๐(โ) )! รโ (๐(๐) + ๐(๐) โ ๐(๐) )!๐(โ) !(๐(๐) + ๐(๐) โ ๐(๐) )!(๐(โ) + ๐(โ) โ ๐(โ) )! (๐(๐) + ๐(๐) )! (๐(โ) + ๐(โ) )! โ = ๐ถโ . ๐(๐) !(๐(๐) + ๐(๐) โ ๐(๐) )! (๐(โ) + ๐(โ) โ ๐(โ) )!๐(โ) ! Here ๐ถ denotes a constant which is independent of ๐ห(๐) and ๐ห(๐) for an index ๐ โ {โ, ๐}. Finally, note that: (๐ + ๐ห(๐) + ๐ห(โ) )! ๐ ๐พห๐,๐, ๐ห(๐) +๐ห(โ) ,ห ๐(๐) +ห ๐(โ) (๐) = โ ๐!(๐ + ๐ห(๐) + ๐ห(โ) โ ๐ห(๐) โ ๐ห(โ) )! ( ) ( ) ๐ +1 ฮ โฃ๐(๐) โฃ + ๐2 ฮ โฃ๐(โ) โฃ + ๐โ2+1 ) ( ), ร ( ๐ +1 ฮ โฃ๐(๐) + ๐(๐) โฃ + ๐2 ฮ โฃ๐(โ) + ๐(โ) โฃ + ๐โ2+1 and the ๏ฌrst factor coincides with ๐ด๐,โ . The assertion is proven.
โก
๐ ยฏ๐ ๐ข ยฏ๐ฃ Fix a tuple ๐ = (๐1 , ๐2 , . . . , ๐๐ ) โ โค๐ + and let ๐1 = ๐1 ๐ ๐ and ๐2 = ๐2 ๐ ๐ be bounded measurable hyperbolic ๐-quasi-homogeneous symbols with ๐ โฅ ๐ and ๐ข โฅ ๐ฃ. Moreover, assume that โฃ๐(๐) โฃ = โฃ๐(๐) โฃ and โฃ๐ข(๐) โฃ = โฃ๐ฃ(๐) โฃ for all ๐ = 1, 2, . . . , ๐.
Theorem 9.5. The Toeplitz operators ๐๐1 and ๐๐2 commute on each weighted Bergman space ๐2๐ (๐ท๐ ) if and only if for each โ = 1, 2, . . . , ๐ โ 1 one of the
172
W. Bauer and N. Vasilevski
conditions (a)โ(d) is ful๏ฌlled: (a) ๐โ = ๐โ = 0, (b) ๐ขโ = ๐ฃโ = 0, (c) ๐โ = ๐ขโ = 0, (d) ๐โ = ๐ฃโ = 0. Proof. Let ๐ โ โ๐0 such that the following expressions are non-zero: ๐1 ๐2 (๐
๐๐1 ๐
โ ) (๐
๐๐2 ๐
โ ) ๐ห๐ (๐(๐)) = ๐พห๐+๐ขโ๐ฃ,๐,๐,๐ (๐)ห ๐พ๐,๐,๐ข,๐ฃ (๐)ห ๐๐+๐ข+๐โ๐ฃโ๐ (๐(๐)) ๐2 ๐1 (๐
๐๐2 ๐
โ ) (๐
๐๐1 ๐
โ ) ๐ห๐ (๐(๐)), = ๐พห๐+๐โ๐,๐,๐ข,๐ฃ (๐)ห ๐พ๐,๐,๐,๐ (๐)ห ๐๐+๐ข+๐โ๐ฃโ๐ (๐(๐)).
Hence, ๐๐1 and ๐๐2 commute if and only if for all ๐ (such that the expressions below are non-zero) we have: ๐1 ๐2 ๐2 ๐1 ๐พห๐+๐ขโ๐ฃ,๐,๐,๐ (๐)ห ๐พ๐,๐,๐ข,๐ฃ (๐) = ๐พห๐+๐โ๐,๐,๐ข,๐ฃ (๐)ห ๐พ๐,๐,๐,๐ (๐).
By (9.3) this is equivalent to ๐ ๐ ๐พ๐1 (๐ + ๐ข โ ๐ฃ, ๐)ห ๐พ๐+๐ขโ๐ฃ,๐,๐,๐ (๐)๐พ๐2 (๐, ๐)ห ๐พ๐,๐,๐ข,๐ฃ (๐)
๐ ๐ = ๐พ๐2 (๐ + ๐ โ ๐, ๐)ห ๐พ๐+๐โ๐,๐,๐ข,๐ฃ (๐)๐พ๐1 (๐, ๐)ห ๐พ๐,๐,๐,๐ (๐).
(9.4)
From โฃ๐(๐) โฃ = โฃ๐(๐) โฃ and โฃ๐ข(๐) โฃ = โฃ๐ฃ(๐) โฃ, for all ๐ = 1, . . . , ๐, together with Lemma 7.2 it follows that ๐พ๐1 (๐ + ๐ข โ ๐ฃ, ๐) = ๐พ๐1 (๐, ๐)
and
๐พ๐2 (๐ + ๐ โ ๐, ๐) = ๐พ๐2 (๐, ๐).
Hence, the relation (9.4) is equivalent to: ๐ ๐ ๐ ๐ ๐พห๐+๐ขโ๐ฃ,๐,๐,๐ (๐)ห ๐พ๐,๐,๐ข,๐ฃ (๐) = ๐พห๐+๐โ๐,๐,๐ข,๐ฃ (๐)ห ๐พ๐,๐,๐,๐ (๐).
We can write this equation more explicitly by using (9.2): (๐ + ๐ข โ ๐ฃ + ๐)! (๐ + ๐ข)! โ โ (๐ + ๐ข โ ๐ฃ)!(๐ + ๐ข โ ๐ฃ + ๐ โ ๐)! ๐!(๐ + ๐ข โ ๐ฃ)! ( ) ( ) ๐ +1 ๐ +1 ๐ ฮ โฃ๐(๐) + ๐ข(๐) โ ๐ฃ(๐) โฃ + ๐2 ฮ โฃ๐(๐) โฃ + ๐2 โ ( ) ( ) ร ๐๐ +1 ๐๐ +1 ฮ โฃ๐ + ๐ข โ ๐ฃ + ๐ โฃ + ฮ โฃ๐ + ๐ข โฃ + ๐=1 (๐) (๐) (๐) (๐) (๐) (๐) 2 2 (๐ + ๐ โ ๐ + ๐ข)! (๐ + ๐)! โ = โ (๐ + ๐ โ ๐)!(๐ + ๐ โ ๐ + ๐ข โ ๐ฃ)! ๐!(๐ + ๐ โ ๐)! ( ) ( ) ๐ +1 ๐ +1 ๐ ฮ โฃ๐(๐) + ๐(๐) โ ๐(๐) โฃ + ๐2 ฮ โฃ๐(๐) โฃ + ๐2 โ ( ) ( ) . ร ๐๐ +1 ๐ +1 ฮ โฃ๐(๐) + ๐(๐) โฃ + ๐2 ๐=1 ฮ โฃ๐(๐) + ๐(๐) โ ๐(๐) + ๐ข(๐) โฃ + 2 Since by assumption we have โฃ๐(๐) โฃ = โฃ๐(๐) โฃ and โฃ๐ข(๐) โฃ = โฃ๐ฃ(๐) โฃ for all ๐ = 1, . . . , ๐, this is equivalent to: (๐ + ๐ข โ ๐ฃ + ๐)!
(๐ + ๐ข)! (๐ + ๐)! = (๐ + ๐ โ ๐ + ๐ข)! . (๐ + ๐ข โ ๐ฃ)! (๐ + ๐ โ ๐)!
Varying ๐ one can check that this equality holds if and only if for each โ = 1, 2, . . . , ๐ โ 1 one of the conditions (a)โ(d) are ful๏ฌlled. โก
Banach Algebras of Commuting Toeplitz Operators
173
In the following we assume (i) and (ii): (i) For each ๐ with ๐๐ > 1 we have: ๐(๐) = (๐๐,1 , . . . , ๐๐,โ๐ , 0, . . . , 0)
(9.5)
๐(๐) = (0, . . . , 0, ๐๐,โ๐ +1 , . . . , ๐๐,๐๐ ) โฒ
(ii) if ๐๐ โฒ = ๐๐ โฒโฒ with ๐ < ๐ โฒโฒ , then โ๐ โฒ โค โ๐ โฒโฒ . Let ๐ = (๐1 , . . . , ๐๐ ) be a tuple as before and โ = (โ1 , . . . , โ๐ ) where { โ๐ = 0, if ๐๐ = 1, 1 โค โ๐ โค ๐๐ โ 1, if ๐๐ > 1. De๏ฌnition 9.6. We denote by โ๐ (โ) the linear space generated by all hyperbolic ๐ ๐-quasi-homogeneous functions ๐๐ ๐ ๐ , where the components ๐(๐) and ๐(๐) , ๐ = 1, . . . , ๐, of multi-indices ๐ and ๐ are of the form (9.5) with: ๐๐,1 + โ
โ
โ
+ ๐๐,โ๐ = ๐๐,โ๐ +1 + โ
โ
โ
+ ๐๐,๐๐ . and ๐๐,1 , . . . , ๐๐,โ๐ , ๐๐,โ๐ +1 , . . . , ๐๐,๐๐ โ โค+ . Note that โ๐ โ โ๐ (โ) and that the identity function ๐(๐ง) โก 1 belongs to โ๐ (โ). As an application of Theorem 9.5 we have: Corollary 9.7. The Banach algebra generated by Toeplitz operators with symbols from โ๐ (โ) is commutative. Finally we would like to note that: (a) For ๐ > 2 and ๐ โ= (1, 1, . . . , 1) these algebras are just Banach algebras, while the C*-algebras generated by them are non-commutative. (b) These Banach algebras are commutative for each weighted Bergman space ๐2๐ (๐ท๐ ) with ๐ > โ1. (c) For ๐ = 2 all these algebras collapse to the single ๐ถ โ -algebra generated by Toeplitz operators with quasi-hyperbolic symbols.
Appendix The following well-known relation is essentially used throughout the text. For convenience of the reader we add its short proof here. Lemma A.1. Let ๐๐ denote the usual surface measure on the (๐ โ 1)-dimensional sphere ๐๐โ1 and let ๐ผ โ โ๐0 . Then: { โซ 0, if some ๐ผ๐ is odd , ๐ผ ๐ฅ ๐๐ := 2ฮ(๐ฝ1 )ฮ(๐ฝ2 )โ
โ
โ
ฮ(๐ฝ๐ ) , if all ๐ผ๐ are even. ๐๐โ1 ฮ(๐ฝ1 +โ
โ
โ
+๐ฝ๐ ) where ๐ฝ๐ := 12 (๐ผ๐ + 1). Moreover, if ๐ผ โ โ๐0 then we have: ( ) ( ) โซ 2ฮ ๐ผ12+1 โ
โ
โ
ฮ ๐ผ๐2+1 ๐ผ ( ) โฃ๐ฆ โฃ๐๐ = . ๐๐โ1 ฮ ๐+โฃ๐ผโฃ 2
174
W. Bauer and N. Vasilevski
Proof. We only prove the second assertion which in particular implies the ๏ฌrst one. Consider: โซ 2 ๐ผ๐ผ := โฃ๐ฅ๐ผ โฃ๐โโฃ๐ฅโฃ ๐๐ฅ = =
โ๐ ๐ โโซ
๐=1 ๐ โ
โ
โซ
2
๐=1
=
๐ โ ๐=1
2
โฃ๐ฅ๐ผ๐ โฃ๐โ๐ฅ๐ ๐๐ฅ๐
0
( ฮ
โ
2
๐ฅ๐ผ๐ ๐โ๐ฅ๐ ๐๐ฅ๐
๐ผ๐ + 1 2
) .
By changing to polar coordinates we have โซ โซ โ 2 ๐ผ๐ผ = โฃ(๐๐ฆ)๐ผ โฃ๐โ๐ ๐๐โ1 ๐๐๐๐(๐ฆ) ๐โ1 โซ๐ โ 0 โซ 2 = ๐โฃ๐ผโฃ+๐โ1 ๐โ๐ ๐๐ โฃ๐ฆ ๐ผ โฃ ๐๐(๐ฆ) ๐โ1 0 ๐ ( )โซ 1 ๐ + โฃ๐ผโฃ = ฮ โฃ๐ฆ ๐ผ โฃ ๐๐(๐ฆ), 2 2 ๐โ1 ๐ and the assertion follows.
โก
References [1] W. Bauer, Y.L. Lee, Commuting Toeplitz operators on the Segal-Bargmann space, J. Funct. Anal. 260(2) (2011), 460โ489. [2] B.R. Choe, H. Koo and Y.J. Lee, Commuting Toeplitz operators on the polydisk, Trans. Amer. Math. Soc. 356 (2004), 1727โ1749. [3] B.R. Choe and Y.J. Lee, Pluriharmonic symbols of commuting Toeplitz operators, Illinois J. Math. 37 (1993), 424โ436. ห Cuห ห ckoviยดc and N.V. Rao, Mellin transform, monomial symbols and commuting [4] Z. Toeplitz operators, J. Funct. Anal. 154 (1998), 195โ214. [5] S. Grudsky, R. Quiroga-Barranco and N. Vasilevski, Commutative ๐ถ โ -algebras of Toeplitz operators and quantization on the unit disc, J. Funct. Anal. 234 (2006), 1โ44. [6] T. Le, The commutants of certain Toeplitz operators on weighted Bergman spaces, J. Math. Anal. Appl. 348(1) (2008), 1โ11. [7] Y.J. Lee, Commuting Toeplitz operators on the Hardy space of the polydisc, Proc. Amer. Math. Soc., vol. 138(1) (2010), 189โ197. [8] R. Quiroga-Barranco and N. Vasilevski, Commutative ๐ถ โ -algebras of Toeplitz operators on the unit ball, I. Bargmann-type transforms and spectral representations of Toeplitz operators, Integr. Equ. Oper. Theory 59(3) (2007), 379โ419.
Banach Algebras of Commuting Toeplitz Operators
175
[9] N. Vasilevski, Bergman space structure, commutative algebras of Toeplitz operators and hyperbolic geometry, Integr. Equ. Oper. Theory 46 (2003), 235โ251. , Commutative algebras of Toeplitz operators on the Bergman space, Birk[10] hยจ auser, Operator Theory: Advances and Applications, (2008). [11] , Parabolic quasi-radial quasi-homogeneous symbols and commutative algebras of Toeplitz operators, Operator Theory: Advances and Applications, v. 202 (2010), 553โ568. [12] , Quasi-radial quasi-homogeneous symbols and commutative Banach algebras of Toeplitz operators, Integr. Equ. Oper. Theory 66 (2010), 141โ152. Wolfram Bauer Mathematisches Institut Georg-August-Universitยจ at Bunsenstr. 3โ5 D-37073 Gยจ ottingen, Germany e-mail:
[email protected] Nikolai Vasilevski Departamento de Matemยด aticas CINVESTAV del I.P.N. Av. IPN 2508, Col. San Pedro Zacatenco Mยดexico D.F. 07360, Mยดexico e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 177โ205 c 2012 Springer Basel AG โ
Canonical Models for Bi-isometries H. Bercovici, R.G. Douglas and C. Foias We dedicate this paper to the memory of Israel Gohberg, great mathematician, wonderful human being, friend and teacher to us all
Abstract. A canonical model, analogous to the one for contraction operators, is introduced for bi-isometries, two commuting isometries on a Hilbert space. This model involves a contractive analytic operator-valued function on the unit disk. Various complete nonunitarity conditions are considered as well as bi-isometries for which both isometries are shifts. Several families of examples are introduced and classi๏ฌed. Mathematics Subject Classi๏ฌcation (2000). Primary: 47A45. Secondary: 47A15, 47B37. Keywords. Bi-isometry, characteristic function, functional model, pivotal operator, similarity.
1. Introduction It is di๏ฌcult to overestimate the importance of the von Neumann-Wold theorem on the structure of isometric operators on Hilbert space. Originally introduced in the study of symmetric operators by von Neumann, it became the foundation for Woldโs study of stationary stochastic processes. Later, it was the starting point for the study of contraction operators by Sz.-Nagy and the third author as well as a key ingredient in engineering systems theory. Thus it has had an important role in both pure mathematics and its applications. For nearly ๏ฌfty years, researchers have sought a similar structure theory for ๐-tuples of commuting isometries [4,11,12,15,16,17,19] with varying success. In [2] the authors rediscovered an earlier fundamental result of Berger, Coburn and Lebow [4] on a model for an ๐-tuple of commuting isometries and carried the analysis beyond what the latter researchers had done. In the course of this study, HB and RGD were supported in part by grants from the National Science Foundation.
178
H. Bercovici, R.G. Douglas and C. Foias
a very concrete canonical model for bi-isometries emerged; that is, for pairs of commuting isometries. This new model is related to the canonical functional model of a contraction, but it displays subtle di๏ฌerences and a new set of challenges. In this paper we take up the systematic presentation and development of this model. After some preliminaries, we begin in Section 3 by examining the passage from an ๐-isometry to an (๐ + 1)-isometry showing that essentially the main ingredient needed is a contraction in the commutant of a completely nonunitary ๐-isometry. In the case of a bi-isometry, this additional operator can be viewed as a contractive operator-valued analytic function in the unit disk. It is this function that is the heart of our canonical model. We relate the reducing subspaces of an ๐-isometry to this construction and investigate a variety of notions of complete nonunitarity which generalize the notion of completely nonunitary contractions and the results of several earlier researchers. (See Section 3 for the details.) In Section 4 we specialize to the case ๐ = 1, that is to the case of bi-isometries, and study the extension from the ๏ฌrst isometry to the pair. The analytic operator function mentioned above then is the characteristic function for the pair. Various relations between the bi-isometry and the characteristic function are investigated. In Section 5, this model is re-examined in the context of a functional model; that is, one in which the abstract Hilbert spaces are realized as Hardy spaces of vector-valued functions on the unit disk. This representation allows one to apply techniques from harmonic analysis in their study. In Section 6, we specialize to bi-shifts or bi-isometries for which both isometries are shift operators. (Note that this use of the term is not the same as that used by earlier authors.) In Section 7, we return to the functional model for bi-isometries obtaining unitary invariants for them. Finally, in Section 8, several families of bi-isometries are introduced and studied. The results here are not exhaustive but intended to illustrate various aspects of the earlier theory as well as the variety of possibilities presented by bi-isometries. At the ends of Sections 3 and 4, the connection between intertwining maps and common invariant subspaces for bi-isometries is discussed. This topic has already been considered in [3] and further results will be presented in another paper. The paper bene๏ฌtted from a thorough review by the referee who helped eliminate one serious error, along with numerous misprints in our original manuscript. The authors gratefully acknowledge his help in improving this work.
2. Preliminaries about commuting isometries We will study families ๐ = (๐๐ )๐โ๐ผ of commuting isometric operators on a complex Hilbert space โ. A (closed) subspace ๐ โ โ is invariant for ๐ if ๐๐ ๐ โ ๐ for ๐ โ ๐ผ; we write ๐โฃ๐ = (๐๐ โฃ๐)๐โ๐ผ if ๐ is invariant. The invariant subspace ๐ is reducing if ๐โฅ is invariant for ๐ as well. If ๐ is a reducing subspace, we have a
Canonical Models for Bi-isometries decomposition
179
๐ = (๐โฃ๐) โ (๐โฃ๐โฅ ),
and ๐โฃ๐ is called a direct summand of ๐. The family ๐ is said to be unitary if each ๐๐ , ๐ โ ๐ผ, is a unitary operator. We say that ๐ is completely nonunitary or cnu if it has no unitary direct summand acting on a space ๐ โ= {0}. The family ๐ is irreducible if it has no reducing subspaces other than {0} and โ. The following extension of the von Neumann-Wold decomposition was proved by I. Suciu [20]. Theorem 2.1. Let ๐ be a family of commuting isometries on โ. There exists a unique reducing subspace ๐ for ๐ with the following properties. (1) ๐โฃ๐ is unitary. (2) ๐โฃ๐โฅ is completely nonunitary. We recall, for the readerโs convenience, the construction of ๐. We simply set โก โค โ โฉ โฉ โฃ ๐= ๐๐1 ๐๐2 โ
โ
โ
๐๐๐ โโฆ . ๐ =1
๐1 ,๐2 ,...,๐๐ โ๐ผ
Obviously, ๐๐ ๐ โ ๐ for each ๐, and the commutativity of ๐ implies that ๐๐ ๐ โ ๐ as well. Thus ๐ reduces each ๐๐ to a unitary operator. It is then easily seen that ๐ is the largest invariant subspace for ๐ such that ๐โฃ๐ is unitary, and this immediately implies properties (1) and (2), as well as the uniqueness of ๐. Corollary 2.2. Consider a ๏ฌnite family ๐ = (๐0 , ๐1 , . . . , ๐๐ ) of commuting isometries on โ. Then ๐ is completely nonunitary if and only if the product ๐0 ๐1 โ
โ
โ
๐๐ is completely nonunitary. Proof. Indeed, the space ๐ in the preceding theorem can alternatively be described as โ โฉ ๐= ๐ ๐ โ, ๐=1
where ๐ = ๐0 ๐1 โ
โ
โ
๐๐ .
โก
More generally, given a subset ๐ฝ โ ๐ผ, we will say that ๐ is ๐ฝ-unitary if ๐๐ is a unitary operator for each ๐ โ ๐ฝ. The family ๐ is said to be ๐ฝ-cnu if it has no ๐ฝ-unitary direct summand acting on a nonzero space. Theorem 2.1 extends as follows. Theorem 2.3. Let ๐ = (๐๐ )๐โ๐ผ be a family of commuting isometries on a Hilbert space โ, and let ๐ฝ be a subset of ๐ผ. There exists a unique reducing subspace ๐๐ฝ for ๐ with the following properties. (1) ๐โฃ๐๐ฝ is ๐ฝ-unitary. (2) ๐โฃ๐โฅ ๐ฝ is ๐ฝ-cnu.
180
H. Bercovici, R.G. Douglas and C. Foias
Proof. Let us set ๐๐ฝ = (๐๐ )๐โ๐ฝ and apply Theorem 2.1 to this family. Thus we can write โ = ๐ โ ๐, where ๐ is reducing for ๐๐ฝ , ๐๐ฝ โฃ๐ is unitary, and ๐๐ฝ โฃ๐ is cnu. Denote by ๐๐ฝ the smallest reducing subspace for ๐ containing ๐, and set ๐๐ฝ = โ โ ๐๐ฝ . Since ๐๐ฝ reduces ๐๐ฝ โฃ๐, it follows immediately that (1) is satis๏ฌed. Moreover, if โ is any reducing subspace for ๐ such that ๐๐ฝ โฃโ is unitary, then โ โ ๐ so that โ โฅ ๐ and consequently โ โฅ ๐๐ฝ as well. We conclude that ๐๐ฝ is the largest reducing subspace for ๐ satisfying condition (1). Property (2), as well as the uniqueness of ๐๐ฝ , follow from this observation. โก Observe that ๐๐ผ is precisely the space ๐ in Theorem 2.1, and it is convenient to extend our notation so that ๐โ
= โ. We have then ๐๐ฝ1 โช๐ฝ2 = ๐๐ฝ1 โฉ ๐๐ฝ2 ,
๐ฝ1 , ๐ฝ2 โ ๐ผ.
(1) (๐๐ )๐โ๐ผ
(2)
Given two families ๐(1) = and ๐(2) = (๐ ๐ )๐โ๐ผ of commuting (1) (2) isometries on โ and โ , respectively, we denote by โ(๐(1) , ๐(2) ) the collection of all bounded linear operators ๐ : โ(1) โ โ(2) satisfying the intertwining relations (1) (2) ๐๐๐ = ๐ ๐ ๐ for every ๐ โ ๐ผ. In the special case ๐(1) = ๐(2) = ๐, we use the notation (๐)โฒ = โ(๐, ๐) for the commutant of ๐. Also, given ๐๐ โ โ(โ(๐) ) for ๐ = 1, 2, we denote by โ(๐1 , ๐2 ) the collection of all bounded linear operators ๐ : โ(1) โ โ(2) satisfying ๐๐1 = ๐2 ๐. (1)
(2)
Proposition 2.4. Consider two families ๐(1) = (๐๐ )๐โ๐ผ and ๐(2) = (๐ ๐ )๐โ๐ผ of commuting isometries on โ(1) and โ(2) . Denote by ๐(๐) the largest reducing subspace for ๐(๐) such that ๐(๐) โฃโ(๐) is unitary for ๐ = 1, 2. Then for every ๐ โ โ(๐(1) , ๐(2) ) we have ๐๐(1) โ ๐(2) . Proof. This follows immediately from the formulas de๏ฌning the spaces โ(๐) .
โก
The preceding result does not extend to the spaces ๐๐ฝ for ๐ฝ โ= ๐ผ. We illustrate this by a simple example. Denote by ๐ โ โ(๐ฟ2 ) the usual bilateral shift, and set ๐+ = ๐ โฃ๐ป 2 . We consider the Hilbert space โ = ๐ป 2 โ ๐ฟ2 โ ๐ฟ2 โ โ
โ
โ
, and the family ๐ = (๐0 , ๐1 ) de๏ฌned on โ by the formulas ๐0 (๐ฃ โ ๐ค0 โ ๐ค1 โ โ
โ
โ
) = ๐+ ๐ฃ โ ๐ ๐ค0 โ ๐ ๐ค1 โ โ
โ
โ
and
๐1 (๐ฃ โ ๐ค0 โ ๐ค1 โ โ
โ
โ
) = 0 โ ๐ฃ โ ๐ค0 โ ๐ค1 โ โ
โ
โ
. It is easy to verify that ๐ is {0}-cnu, but its restriction to the invariant subspace ๐ = 0 โ ๐ฟ2 โ ๐ฟ2 โ ๐ฟ2 โ โ
โ
โ
is {0}-unitary. Thus the inclusion operator from ๐ to โ does not map the {0}unitary part of ๐โฃ๐ to the {0}-unitary part of ๐. Another useful result is the existence of a unique minimal unitary extension for every family of commuting isometries [25, Chapter I] (see also [7] for a Banach space version). We review the result brie๏ฌy.
Canonical Models for Bi-isometries
181
Theorem 2.5. Let ๐ = (๐๐ )๐โ๐ผ be a family of commuting isometries on โ. There exists a family ๐ = (๐๐ )๐โ๐ผ of commuting unitary operators on a Hilbert space ๐ โ โ with the following properties. (1) โ is invariant for ๐ and ๐โฃโ = ๐. ] โโ [โ โ โ โ ๐ ๐ โ
โ
โ
๐ โ . (2) ๐ = ๐ =0 ๐๐ ๐1 ,๐2 ,...,๐๐ โ๐ผ ๐1 ๐2 If ๐โฒ is another family of commuting unitary operators on a space ๐โฒ โ โ satisfying the analogues of conditions (1) and (2), then there exists a surjective isometry ๐ : ๐ โ ๐โฒ such that ๐ โ = โ for โ โ โ, and ๐ ๐๐ = ๐๐โฒ ๐ for ๐ โ ๐ผ. In equation (2) above, we use the convention that ๐๐โ1 ๐๐โ2 โ
โ
โ
๐๐โ๐ โ = โ when ๐ = 0. The family ๐ is called the minimal unitary extension of ๐. In the sequel, we ห the minimal unitary extension of ๐, and by โ ห the space on which denote by ๐ it acts. It is easy to verify the following commutant extension result. This can be deduced from the results in [25, Chapter 1], and it is proved in [7] for isometric operators acting on a Banach space. (1)
Theorem 2.6. Let ๐(1) = (๐๐
(2)
)๐โ๐ผ and ๐(2) = (๐ ๐ )๐โ๐ผ be two families of comห ห (1) and ๐ (2) and โ(2) , respectively, and denote by ๐
muting isometries on โ(1) their minimal unitary extensions. The map ๐ โ ๐ = ๐ โฃโ(1) establishes an isoห ห (1) , ๐ (2) ) such that metric bijection between the collection of operators ๐ โ โ(๐ (1) (2) (1) (2) ๐ โ โ โ and โ(๐ , ๐ ).
Indeed, given ๐1 , ๐2 , . . . , ๐๐ โ ๐ผ, a given operator ๐ โ โ(๐(1) , ๐(2) ) easily โ โ โ ห ห (1) ห (1) (1) extends to the space ๐๐1 ๐๐2 โ
โ
โ
๐๐๐ โ(1) by setting โ โ โ โ โ โ ห ห ห ห (1) ห (1) (1) (2) ห (2) (2) ๐ ๐๐1 ๐๐2 โ
โ
โ
๐๐๐ โ = ๐๐1 ๐๐2 โ
โ
โ
๐๐๐ ๐โ,
โ โ โ ห ห (1) ห (1) (1) โ โ ๐๐1 ๐๐2 โ
โ
โ
๐๐๐ โ(1) ,
ห ห and the corresponding operator ๐ โ โ(๐, ๐โฒ ) is obtained by taking the closure of ห If ๐ is isometric this extension. This unique extension of ๐ will be denoted ๐. (1) (2) ห or unitary then so is ๐. In the particular case ๐ = ๐ = ๐, the operator ๐ ห โ (๐) ห โฒ satis๏ฌes belongs to the commutant of ๐, and its canonical extension ๐ ห โ โ. ๐โ Irreducible families of commuting isometries have special properties. Theorem 2.1 shows that they are either unitary or cnu. More precisely, we have the following result. Proposition 2.7. Let ๐ = (๐๐ )๐โ๐ผ be an irreducible family of commuting isometries on a nonzero Hilbert space โ. For every ๐0 โ ๐ผ, one of the following alternatives occurs. (1) ๐๐0 is a scalar multiple of the identity. (2) ๐ is {๐0 }-cnu.
182
H. Bercovici, R.G. Douglas and C. Foias
Proof. Assume that (2) does not occur. Theorem 2.3 implies then that ๐๐0 is unitary. Since the spectral projections of ๐๐0 reduce ๐, it follows that the spectrum of ๐๐0 is a singleton, and therefore (1) is true. โก The following result is a consequence of elementary facts about representations of ๐ถ โ -algebras. All the families of isometries in this statement are indexed by the same set ๐ผ. Proposition 2.8. Let ๐ be a family of commuting isometries on โ, and denote by โฑ a collection of mutually inequivalent irreducible families of commuting isometries such that every irreducible direct summand of ๐ is equivalent to an element of โฑ . (1) Fix ๐ โ โฑ, and let (๐๐ผ )๐ผโ๐ด be a maximal family of mutually equivalent reducing subspaces for ๐ such that ๐โฃ๐๐ผ is unitarily equivalent to ๐ for all ๐ผ โ ๐ด. Then the reducing space โ โ๐ = ๐๐ผ ๐ผโ๐ด
depends only on ๐. (2) If ๐1 , ๐2 โ โฑ are di๏ฌerent, then the spaces โ๐1 and โ๐2 are mutually orthogonal. (3) We have โ โ = โ0 โ โ๐ , ๐โโฑ
where โ0 is a reducing subspace for ๐ such that ๐โฃโ0 has no irreducible direct summand. When dim โ0 > 1, the family ๐โฃโ0 is obviously reducible; it just cannot be decomposed into a direct sum of irreducible families. However, it can be decomposed into a continuous direct integral of irreducibles if โ is separable. A concrete example of such a decomposition will be given in Section 8. Direct integrals are also useful in the proof of the following result, an early variant of which was proved in [20] when ๐ผ consists of two elements. We refer to [26] for the theory of direct integrals. Proposition 2.9. Let ๐ = (๐๐ )๐โ๐ผ be a ๏ฌnite family of commuting isometries on a Hilbert space โ. We can associate to each subset ๐ฝ โ ๐ผ a reducing space ๐๐ฝ for ๐ with the following properties. โ (1) โ = ๐ฝโ๐ผ ๐๐ฝ . (2) ๐๐ โฃ๐๐ฝ is unitary for each ๐ โ ๐ฝ. (3) ๐โฃ๐๐ฝ is {๐}-cnu for each ๐ โ / ๐ฝ. Proof. Since ๐ผ is ๏ฌnite, โ can be written as an orthogonal sum of separable reducing subspaces for ๐. Thus it is su๏ฌcient to consider the case of separable spaces โ. There exist a standard measurable space ฮฉ, a probability measure ๐ on ฮฉ, a measurable family (โ๐ก )๐กโฮฉ of Hilbert spaces, and a measurable collection
Canonical Models for Bi-isometries
183
(๐๐ก )๐กโฮฉ = ((๐๐ก๐ )๐โ๐ผ )๐กโฮฉ of irreducible families of commuting isometries on โ๐ก such that, up to unitary equivalence, โซ โ โซ โ โ= โ๐ก ๐๐(๐ก), ๐๐ = ๐๐ก๐ ๐๐(๐ก), ๐ โ ๐ผ. ฮฉ
ฮฉ
The reducing subspaces of ๐ are precisely the spaces of the form โซ โ ๐= ๐๐ก ๐๐(๐ก), ๐
where ๐ โ ฮฉ is measurable. Proposition 2.7 shows that for each ๐ก โ ฮฉ there exists a subset ๐ฝ(๐ก) โ ๐ผ such that ๐๐ก๐ is a scalar multiple of the identity if ๐ โ ๐ฝ(๐ก), while ๐๐ก is {๐}-cnu for ๐ โ / ๐ฝ(๐ก). We claim that the set ๐๐ = {๐ก โ ฮฉ : ๐๐ก๐ is a scalar multiple of the identity} is measurable for each ๐ โ ๐ผ. Indeed, consider measurable families of vectors ๐ก โ ๐๐๐ก โ โ๐ก , ๐ = 1, 2, . . . , such that the nonzero vectors in the set {๐๐๐ก : ๐ โฅ 1} form an orthonormal basis for โ๐ก for each ๐ก โ ฮฉ. Then the set ๐๐ is de๏ฌned by the countable family of equations โจ๐๐ก๐ ๐๐๐ก , ๐โ๐ก โฉ = 0,
โจ๐๐ก๐ ๐๐๐ก , ๐๐๐ก โฉ = โจ๐๐ก๐ ๐โ๐ก , ๐โ๐ก โฉ,
which must be satis๏ฌed when ๐ โ= โ and ๐๐๐ก โ= 0 โ= ๐โ๐ก . It follows that the set ๐๐ฝ = {๐ก โ [0, 1] : ๐ฝ(๐ก) = ๐ฝ} is measurable for each ๐ฝ โ ๐ผ. The spaces โซ โ ๐๐ฝ = โ๐ก ๐๐(๐ก), ๐๐ฝ
viewed as subspaces of โ, satisfy the conclusion of the proposition. This follows from the above description of the reducing subspaces of ๐. โก Some of the spaces ๐๐ฝ in the preceding proposition can equal {0}.
3. Inductive construction of commuting isometries In this section it will be convenient to index families of commuting isometries by ordinal numbers. Thus, given an ordinal number ๐, an ๐-isometry is simply a family ๐ = (๐๐ )0โค๐<๐ of commuting isometries on a Hilbert space. We consider a special construction which produces an (๐ + 1)-isometry starting from an ๐-isometry ๐ on โ and a contraction ๐ด โ (๐)โฒ ; that is, โฅ๐ดโฅ โค 1. ห โฒ on โ ห is then a contraction as well, ห โ (๐) Observe that the canonical extension ๐ด and therefore we can form the defect operator หโ ๐ด) ห 1/2 ๐ท๐ดห = (๐ผ โ ๐ด ห The space ๐ is reducing for ๐ ห because ๐ท ห commutes and the space ๐ = ๐ท๐ดหโ. ๐ด ห with ๐. We form the space ๐ = โ โ ๐โ ๐ โ โ
โ
โ
,
184
H. Bercovici, R.G. Douglas and C. Foias
and de๏ฌne an (๐ + 1)-isometry ๐๐ด = (๐๐ )0โค๐โค๐ on ๐ as follows. For 0 โค ๐ < ๐ we de๏ฌne ห๐ โฃ๐) โ (๐ ห๐ โฃ๐) โ โ
โ
โ
, ๐๐ = ๐๐ โ (๐ while ๐๐ (โ โ ๐0 โ ๐1 โ โ
โ
โ
) = ๐ดโ โ ๐ท๐ดหโ โ ๐0 โ ๐1 โ โ
โ
โ
if โ โ โ and ๐๐ โ ๐ for ๐ โ โ. It is easy to verify that ๐๐ด is in fact an (๐ + 1)-isometry. When the operator ๐ด is already isometric, we have ๐ = โ and ๐๐ด = (๐, ๐ด). In this trivial case, every (๐+1)-isometry is of the form ๐๐ด for some contraction ๐ด commuting with an ๐-isometry ๐. We give now a characterization of (๐ + 1)-isometries which are {0 โค ๐ < ๐}-cnu. Theorem 3.1. Let ๐ = (๐๐ )0โค๐โค๐ be an (๐ + 1)-isometry on ๐, where ๐ โฅ 1. The following conditions are equivalent. (1) ๐ is {0 โค ๐ < ๐}-cnu. (2) There exist a cnu ๐-isometry ๐, and a contraction ๐ด โ (๐)โฒ , such that ๐ is unitarily equivalent to ๐๐ด . Proof. Assume ๏ฌrst that ๐ = ๐๐ด , where ๐ด is a contraction in the commutant of the cnu ๐-isometry ๐ on โ. Let ๐ be a reducing subspace for ๐๐ด with the property that ๐๐ โฃ๐ is unitary for all ๐ < ๐. Since the cnu direct summand of the ๐-isometry (๐๐ )0โค๐<๐ is precisely ๐ viewed as acting on โ โ {0} โ {0} โ โ
โ
โ
, we conclude that ๐ โ {0} โ ๐ โ ๐ โ โ
โ
โ
and therefore ๐๐โ๐ โ โ {0} โ ๐ โ ๐ โ โ
โ
โ
for every โ โ ๐ and ๐ โฅ 1. This is not possible if โ โ= 0. Indeed, if โ = 0 โ 0 โ โ
โ
โ
โ 0 โ ๐๐ โ โ
โ
โ
, and the ๐ th component ๐๐ is the ๏ฌrst nonzero component of โ, then ๐๐โ๐ โ = ๐ท๐ดห๐๐ โ โ
โ
โ
โ / ๐ because ๐ท๐ดห๐๐ โ= 0. Conversely, assume that condition (1) is satis๏ฌed. Consider the ๐-isometry ๐โฒ = (๐๐ )0โค๐<๐ , and the decomposition ๐ = โ โ โโฅ into reducing subspaces for ๐โฒ such that ๐โฒ โฃโ is cnu and ๐โฒ โฃโโฅ is unitary. We denote by ๐ = ๐โฒ โฃโ the cnu direct summand of ๐โฒ , and de๏ฌne an operator ๐ด on โ by setting ๐ด = ๐โ ๐๐ โฃโ. Clearly ๐ด is a contraction, and the fact that ๐ด commutes with ๐ follows from the fact that the unitary component โโฅ is obviously invariant for ๐๐ , and หโฒ which can therefore ๐ดโ = ๐๐โ โฃโ. Consider next the minimal unitary extension ๐ be written as หโฒ = ๐ ห โ (๐โฒ โฃโโฅ ) ๐ ห ห โ โโฅ , and the unique isometric extension ๐ on the space โ ๐ of ๐๐ in the comหโฒ . Clearly, mutant of ๐ โฅ โฅ ห ๐ ๐ โฃโ = ๐๐ โฃโ , หห ห and the compression ๐โ ห ๐๐ โฃโ is precisely the contractive extension ๐ด of ๐ด in the ห ห commutant of ๐. We show next that ๐๐ is in fact the minimal isometric dilation
Canonical Models for Bi-isometries
185
ห ห In other words, the smallest invariant subspace ๐ for ๐ ห of ๐ด. ๐ containing โ is โฅ ห โ โ โ . To prove this, observe ๏ฌrst that, since โ ๐ ห ห ๐= ๐ ๐ โ ๐ โฅ0 โ ห is invariant for ๐ ห ห ห and โ ๐ , the space ๐ is actually reducing for ๐๐ . Moreover, ๐๐ โ ห๐ and ๐ ห is unitary for ๐ < ๐, and hence the operators ๐ ๐ also commute. Thus ๐ ห is also a reducing space for each ๐๐ if ๐ < ๐. We conclude that the space ๐โฅ โ โโฅ reduces ๐, and ๐โฒ โฃ๐โฅ is unitary. Hypothesis (1) implies that ๐โฅ = {0}. With this preparation out of the way, we ๏ฌnd ourselves in the familiar territory of minimal isometric dilations [25, Chapter II]. We recall that, up to unitary ห is the operator ๐ equivalence, the minimal isometric dilation of the contraction ๐ด de๏ฌned by ห โ ๐ท หโ โ ๐0 โ ๐1 โ โ
โ
โ
๐ (โ โ ๐0 โ ๐1 โ โ
โ
โ
) = ๐ดโ ๐ด
ห โ ๐ โ ๐ โ โ
โ
โ
, where ๐ = ๐ท หโ. ห We conclude that there exists a on the space โ ๐ด โฅ unitary operator ๐ : ๐ โ ๐ โ โ
โ
โ
โ โ such that (๐ผโ ห โ ๐ )๐ = ๐๐ (๐ผโ ห โ ๐ ). The reader will verify now without di๏ฌculty that the operator ๐ผโ โ ๐ provides a unitary equivalence between ๐๐ด and ๐. โก General (๐ + 1)-isometries are described using Theorem 2.3 with ๐ฝ = {0 โค ๐ < ๐}. We record the result below. Theorem 3.2. Let ๐ = (๐๐ )0โค๐โค๐ be an (๐ + 1)-isometry on ๐, where ๐ โฅ 1. There exist reducing subspaces ๐0 and ๐1 for ๐ with the following properties. (1) ๐0 โ ๐1 = ๐. (2) ๐๐ โฃ๐1 is unitary for every ๐ < ๐. (3) ๐โฃ๐0 is unitarily equivalent to ๐๐ด , where ๐ด is a contraction in the commutant of a cnu ๐-isometry ๐. In fact, the ๐-isometry ๐ on โ โ ๐ is the cnu part of ๐โฒ = (๐๐ )0โค๐<๐ , and the operator ๐ด is de๏ฌned by the equivalent relations ๐ด = ๐โ ๐๐ โฃโ,
๐ดโ = ๐๐โ โฃโ.
ห In particular, ๐๐ is an isometric dilation of ๐ด, and ๐ ๐ is an isometric dilation โฒ โฒ ) and ๐ด ห ห โฒ. ห where the extension ๐ ห ห of ๐ด, belongs to ( ๐ โ (๐) ๐ Thus, the space ๐0 is simply the {0 โค ๐ < ๐}-cnu summand of ๐. The operators which intertwine two (๐ + 1)-isometries can also be analyzed in the context of this inductive construction. Indeed, consider (๐ + 1)-isometries ๐(๐) acting on ๐(๐) for ๐ = 1, 2. Denote by โ(๐) the subspace of ๐(๐) on which the (๐)โฒ (๐) cnu part of the ๐-isometry ๐๐ = (๐๐ )0โค๐<๐ acts. The preceding results allow us to write the decompositions (๐)
(๐)
๐(๐) = ๐0 โ ๐1 ,
๐ = 1, 2,
186
H. Bercovici, R.G. Douglas and C. Foias (๐)
(๐)
(๐)
where ๐(๐) โฃ๐0 is {0 โค ๐ < ๐}-cnu, and ๐๐ โฃ๐1 (๐) Moreover, โ(๐) is contained in ๐0 , and we set ๐(๐) = ๐(๐)โฒ โฃโ(๐) ,
is unitary for 0 โค ๐ < ๐.
๐ด(๐) = ๐โ(๐) ๐๐(๐) โฃโ(๐) ,
๐ = 1, 2.
โฒ ห (๐)โฒ of the ๐-isometry ๐(๐) acts on the space The minimal unitary extension ๐
ห (๐) (๐) ห (๐) = ๐ ๐ 0 โ ๐1 , ห (๐) (๐) and we denote by ๐๐ the canonical extension of ๐๐ to this larger space. We have ห ห (๐) (๐) (๐) (๐) ๐๐ = (๐๐ โฃ๐0 ) โ (๐๐(๐) โฃ๐1 ) ห (๐) (๐) ห (๐) . and, as seen above, ๐๐ โฃ๐0 is the minimal isometric dilation of the operator ๐ด ห ห (๐) (๐) Observe that the space ๐0 contains the subspace โ0 of the minimal unitary ห (๐) of ๐(๐) . extension ๐ Any operator ๐ โ โ(๐(1) , ๐(2) ) can be represented as a matrix ] [ ๐00 ๐01 , ๐= ๐10 ๐11 (1)
(2)
where ๐๐๐ โ โ(๐(1) โฃ๐๐ , ๐(2) โฃ๐๐ ) for ๐, ๐ โ {0, 1}. Theorem 2.5 (applied to ห โ the entries in the ๏ฌrst column of ๐) implies the existence of an extension ๐ ห ห (1) (2) โ(๐ , ๐ ). This extension will be represented by a matrix of the form ] [ ห ๐01 ๐ 00 ห ๐= ห ๐10 ๐11 ห (๐) (๐) ห (๐) = ๐ relative to the decompositions ๐ 0 โ ๐1 . Proposition 3.3. With the above notation, the following statements are true. (1) The operator ๐ = ๐โ(2) ๐00 โฃโ(1) belongs to โ(๐(1) , ๐(2) ) and ๐๐ด(1) = ๐ด(2) ๐. ห ห ห ห (1) belongs to โ(๐ (1) , ๐ (2) ) โฉ โ(๐ด (1) , ๐ด (2) ), and ห ห (2) The operator ๐ต = ๐โห (2) ๐โฃโ ๐ตโ(1)โฅ = {0}. Proof. The intertwining properties of ๐ in part (1) follow from the fact that the space โ(๐) is reducing for ๐(๐)โฒ , invariant for ๐๐โ and invariant for ๐ โ by Proposition 2.4. In other words, we can use the fact that, relative to the decompositions ๐(๐) = โ(๐) โ โ(๐)โฅ , the relevant operators have matrices of the form ] ] [ ] [ (๐) [ (๐) ๐ 0 0 ๐ด (๐) 0 ๐๐ (๐) , 0 โค ๐ < ๐. , ๐๐ = ๐= , ๐๐ = โ โ โ โ 0 โ For part (2) we may assume that ๐(๐) , ๐ = 1, 2, are {0 โค ๐ < ๐}-cnu. Then we โก use again the fact that ๐โ(1)โฅ โ โ(2)โฅ and proceed as before.
Canonical Models for Bi-isometries
187
ห (1) = ๐ต. In the framework of [10, ห โ Observe that we have the equality ๐โฃ ห is said to be a lifting of ๐ต, and this lifting is conSec. II.1], the operator ๐ ห โค 1. A natural question arises: given a contraction ๐ต satisfying tractive if โฅ๐โฅ the requirements of Proposition 3.3(2), can one construct a contractive lifting ห ห (1) , ๐ (2) )? If one pursues the more modest goal of ๏ฌnding a contractive ห โ โ(๐ ๐ ห ห (2) ห โ โ(๐๐(1) , ๐ lifting ๐ ๐ ), the answer is in the a๏ฌrmative, and a parametrization of all such contractive liftings can be extracted from [10, Chapter VI]. We describe the result below, under the additional assumption that ๐(2) is {0 โค ๐ < ๐}-cnu. In the notation adopted in this section, this amounts to the requirement that (2) ๐1 = {0}. ห ห (1) , ๐ด (2) ) is Proposition 3.4. With the preceding notation, assume that ๐ต โ โ(๐ด ห ห ห โ โ(๐๐(1) , ๐๐(2) ) of an operator of norm โค 1. The set of contractive liftings ๐ ๐ต is parameterized by (that is, it is in a canonical bijection with) the set of all contractive analytic functions ๐
: ๐ป โ โ(๐, ๐โฒ ), where the spaces ๐ and ๐โฒ are given by the formulas ห (1) ห ห (1) โ ๐ท ๐ (1) , ๐ = ๐ท๐ต โ ๐ต ๐ โ [ ] ห (2) โฒ ห ห ห (2) (1) (1) ๐ = (๐๐ โ ๐ด )๐ต โ โ ๐ท๐ต โ ห (2) ห ห ห ห ห (2) )๐ต โ (1) โ ๐ท โ (1) : โ (1) โ โ (1) }, โ {(๐๐ โ ๐ด ๐ต and where ๐ท๐ต = (๐ผ โ ๐ต โ ๐ต)1/2 . One of the liftings considered above will yield an operator ๐ โ โ(๐(1) , ๐(2) ) ห ห (1) (2) ห ห๐ only when it also satis๏ฌes the conditions ๐ = ๐๐ ๐ for 0 โค ๐ < ๐, and ๐ต ๐ itself is subject to the supplementary conditions ห ห (1) , ๐ (2) ). ๐ตโ(1)โฅ = {0}, ๐ตโ(1) โ โ(2) , ๐ต โ โ(๐ (See Proposition 3.3 and its proof.) We continue the discussion now under the assumption that the operator ๐ต ห ห (1) , ๐ (2) ) is easily does satisfy these additional conditions. The fact that ๐ต โ โ(๐ ห ห (1) (2) seen to imply that ๐ท๐ต โ โ(๐ , ๐ ). Using the notation in Proposition 3.4, these intertwining conditions imply ห ห ห (1) (2) (1) ๐๐ ๐ โ ๐ and (๐๐ โ ๐๐ )๐โฒ โ ๐โฒ for 0 โค ๐ < ๐.
(3.1)
A routine application of techniques from of [10, Chapter VI] yields the following result. Proposition 3.5. With the above notation, assume that ๐(2) is {0 โค ๐ < ๐}-cnu. The set of contractions in โ(๐(1) , ๐(2) ) can be parameterized by pairs (๐ต, ๐
),
188
H. Bercovici, R.G. Douglas and C. Foias
ห ห (1) , ๐ด (2) ) is a contraction satisfying the conditions in Proposition where ๐ต โ โ(๐ด 3.3(2) and ๐
is a parameter as in Proposition 3.4 satisfying the additional conditions ห ห ห (1) (2) (1) (๐๐ โ ๐๐ )๐
(๐ง) = ๐
(๐ง)๐๐ โฃ๐, 0 โค ๐ < ๐, ๐ง โ ๐ป. These results enable one to begin a systematic study of the invariant subspaces of bi-isometries. This study was already started in [3] and it will be continued in a forthcoming paper. Remark 3.6. We emphasize again that the preceding result does not require that ๐(1) is {0 โค ๐ < ๐}-cnu.
4. The structure of bi-isometries For the remainder of this paper, we focus on bi-isometries ๐ = (๐0 , ๐1 ) on a Hilbert space ๐. Theorem 2.3 and its proof immediately yield the following result when ๐ฝ = {0}. Proposition 4.1. Consider a bi-isometry ๐ = (๐0 , ๐1 ) on ๐, let ๐ = โ โ โโฅ be the von Neumann-Wold decomposition relative to ๐0 , so that ๐0 = ๐0 โฃโ is a ห = ห0 = ๐ ห0 โ (๐0 โฃโโฅ ) โ โ(๐) unilateral shift and ๐0 โฃโโฅ is unitary. Denote by ๐ โฅ ห ห ห โ(โ โ โ ) the minimal unitary extension of ๐0 , and denote by ๐1 โ โ(๐) the ห0 . De๏ฌne unique isometric extension of ๐1 which commutes with ๐ โ โ ๐ ห ห โ ๐. ห1 โ, ๐= ๐ ๐=๐ ๐=0 โฅ
Then the subspace ๐ โ โ is reducing for ๐, and ๐0 โฃ๐ is unitary. Moreover, ๐ is the largest reducing subspace for ๐ with the property that ๐0 โฃ๐ is unitary. Corollary 4.2. With the notation of the preceding result, the following assertions are equivalent. (1) ๐ = {0}. (2) ๐ is {0}-cnu. โ โ ห1 is the minimal co-isometric extension of ๐ ห1 โฃโ. ห (3) The operator ๐ The particular case of Proposition 2.9 for bi-isometries can be proved by repeated application of Proposition 4.1. This result was obtained ๏ฌrst in the case of doubly commuting isometries in [19]; the general case appears in [11] (see also [17] for another proof). Corollary 4.3. Consider a bi-isometry ๐ = (๐0 , ๐1 ) on ๐. There exist unique reducing subspaces ๐00 , ๐11 , ๐01 , ๐11 for ๐ with the following properties. (1) ๐0 โฃ๐01 is a shift and ๐1 โฃ๐01 is unitary. (2) ๐0 โฃ๐10 is unitary and ๐1 โฃ๐10 is a shift. (3) ๐0 โฃ๐11 and ๐1 โฃ๐11 are unitary.
Canonical Models for Bi-isometries
189
(4) There is no nonzero reducing subspace ๐ โ ๐00 for ๐ such that either ๐0 โฃ๐ or ๐1 โฃ๐ is unitary. (5) ๐ = ๐00 โ ๐01 โ ๐10 โ ๐11 . Proof. Proposition 4.1 yields a decomposition ๐ = ๐โฅ โ๐ into reducing subspaces for ๐ such that ๐0 โฃ๐ is unitary and there is no reducing subspace ๐โฒ โ ๐โฅ for ๐ such that ๐0 โฃ๐โฒ is unitary. Apply this result with the pair ๐ replaced by (๐1 โฃ๐, ๐0 โฃ๐) and (๐1 โฃ๐โฅ , ๐0 โฃ๐โฅ ), respectively, to obtain decompositions ๐ = ๐10 โ ๐11 and ๐โฅ = ๐00 โ ๐01 , respectively, into sums of reducing subspaces such that ๐1 โฃ๐11 and ๐1 โฃ๐01 are unitary. Moreover, there is no nontrivial reducing subspace ๐ for ๐ contained in either ๐10 or ๐00 such that ๐1 โฃ๐ is unitary. We leave the remaining veri๏ฌcations to the interested reader. โก Consider a bi-isometry ๐ = (๐0 , ๐1 ) on the Hilbert space ๐. As in Proposition 4.1, we consider the Wold decomposition ๐ = โ โ โโฅ for ๐0 , with โ=
โ โ
๐0๐ ๐,
๐ = ker ๐0โ = ๐ โ ๐0 ๐,
๐=0
and we set ๐0 = ๐0 โฃโ and ๐ด = ๐โ ๐1 โฃโ. Thus, ๐0 is a unilateral shift and, as observed earlier, ๐ด is a contraction in the commutant of ๐0 . We will call (๐0 , ๐ด) the characteristic pair associated to the bi-isometry ๐. Thus, the characteristic pair is simply formed by a unilateral shift and a contraction in its commutant. The concept of unitary equivalence for these objects is the natural one: two such pairs are said to be unitarily equivalent if they are conjugated by a unitary operator (the same for the two operators of the pair). The pair (๐1 , ๐0 ) is also a bi-isometry, and the above procedure associates to it a characteristic pair. The characteristic pairs of (๐0 , ๐1 ) and (๐1 , ๐0 ) are not unitarily equivalent in general. For future reference, we restate Theorem 3.1 for the special case ๐ = 1; that is, the case of bi-isometries. Proposition 4.4. Let ๐0 โ โ(โ) be a unilateral shift, and ๐ด โ {๐0 }โฒ a contraction. ห0 ) the minimal unitary extension of ๐0 , let ๐ด ห0 โ โ(โ ห โ {๐ ห0 }โฒ be the Denote by ๐ โ ห 1/2 โ ห ห extension of ๐ด, and set ๐ท = (๐ผ โ ๐ด ๐ด) , ๐ = (๐ทโ0 ) . ห0 . (1) The space ๐ is reducing for ๐ (2) De๏ฌne the Hilbert space ๐ = โ โ ๐โ ๐ โ โ
โ
โ
, and the operators ๐0 , ๐1 โ โ(๐) by ห0 ๐0 โ ๐ ห0 ๐1 โ โ
โ
โ
, ๐0 (โ โ ๐0 โ ๐1 โ โ
โ
โ
) = ๐0 โ โ ๐ ๐1 (โ โ ๐0 โ ๐1 โ โ
โ
โ
) = ๐ดโ โ ๐ทโ โ ๐0 โ ๐1 โ โ
โ
โ
. Then (๐0 , ๐1 ) is a {0}-cnu bi-isometry whose characteristic pair is unitarily equivalent to (๐0 , ๐ด).
190
H. Bercovici, R.G. Douglas and C. Foias
We collect in the following statement some basic properties of the characteristic pair. These follow immediately from the results in Section 3. Proposition 4.5. Let ๐ = (๐0 , ๐1 ) and ๐โฒ = (๐0โฒ , ๐1โฒ ) be two bi-isometries with characteristic pairs (๐0 , ๐ด) and (๐0โฒ , ๐ดโฒ ), respectively. (1) The characteristic pair of ๐ โ ๐โฒ is (๐0 โ ๐0โฒ , ๐ด โ ๐ดโฒ ). (2) If ๐ is unitarily equivalent to ๐โฒ , then (๐0 , ๐ด) is unitarily equivalent to (๐0โฒ , ๐ดโฒ ). (3) Assume in addition that ๐ and ๐โฒ are {0}-cnu. If (๐0 , ๐ด) is unitarily equivalent to (๐0โฒ , ๐ดโฒ ), then ๐ is unitarily equivalent to ๐โฒ . (4) For every pair (๐0 , ๐ด), where ๐0 is a unilateral shift and ๐ด โ {๐0 }โฒ is a contraction, there exists a bi-isometry ๐ such that (๐0 , ๐ด) is the characteristic pair associated to ๐. This bi-isometry can be chosen to be {0}-cnu. The preceding proposition characterizes the reducing subspaces of a {0}cnu bi-isometry in terms of its characteristic pair. General invariant subspaces of a bi-isometry are not characterized as easily. One di๏ฌculty is the fact that the restriction of a {0}-cnu bi-isometry to an invariant subspace is not always {0}cnu. Assume then that we start with a {0}-cnu bi-isometry ๐ on ๐, ๐โฒ โ ๐ is an invariant subspace for ๐, and ๐โฒ = ๐โฃ๐โฒ . The inclusion operator ๐ โ โ(๐โฒ , ๐) is obviously an isometry in โ(๐โฒ , ๐). Conversely, given an isometric intertwining between bi-isometries ๐ โ โ(๐(1) , ๐), the range of ๐ is an invariant subspace for ๐. Thus the description of invariant subspaces for bi-isometries can be achieved by understanding the structure of isometric operators intertwining two bi-isometries. In the terminology of Proposition 3.5, one needs to ๏ฌnd the parameters ๐
which ห of a given contraction ๐ต. We presented in [3] give rise to isometric liftings ๐ some general results concerning this problem, and further results will appear in a forthcoming paper.
5. Functional representation The data in a characteristic pair (๐0 , ๐ด) on โ can alternately be encoded in a contractive analytic operator-valued function on the unit disk ๐ป. Set ๐ = โ โ ๐0 โ, and de๏ฌne operators ฮ๐ โ โ(๐) as follows: ฮ๐ = ๐๐ ๐0โ๐ ๐ดโฃ๐,
๐ โฅ 0.
We can then associate to the pair (๐0 , ๐ด) the operator-valued analytic function ฮ(๐ง) =
โ โ
๐ง ๐ ฮ๐ = ๐๐ (๐ผ โ ๐ง๐0โ )โ1 ๐ดโฃ๐,
โฃ๐งโฃ < 1.
๐=0
When (๐0 , ๐ด) is the characteristic pair of a bi-isometry ๐, ฮ will be called the characteristic function of ๐; we will use the notation ฮ = ฮ๐ when it is necessary. If ฮ is the characteristic function of ๐ = (๐0 , ๐1 ), then its coe๏ฌcients satisfy ฮ๐ = ๐๐ ๐0โ๐ ๐ดโฃ๐ = ๐๐ ๐0โ๐ ๐โ ๐1 โฃ๐ = ๐๐ ๐0โ๐ ๐1 โฃ๐,
๐ โฅ 0,
Canonical Models for Bi-isometries
191
since ๐๐ ๐0โ๐ ๐โ = ๐๐ ๐0โ๐ . In particular, the constant coe๏ฌcient ฮ0 = ฮ(0) = ๐๐ ๐1 โฃ๐ = (๐1โ โฃ๐)โ is precisely the pivotal operator associated with the pair (๐1 , ๐0 ), as de๏ฌned in [2]. For the convenience of the reader, we recall that the adjoint of the pivotal operator associated with a bi-isometry (๐0 , ๐1 ) is de๏ฌned as ๐0โ โฃ ker(๐1โ ). The operator ฮ(๐ง) is a contraction for โฃ๐งโฃ < 1; in fact sup โฅฮ(๐ง)โฅ = โฅ๐ดโฅ,
โฃ๐งโฃ<1
where ๐ดโ = ๐1โ โฃโ. Indeed, this equality follows easily from the fact that ๐ด is unitarily equivalent to the Toeplitz operator with symbol ฮ; see the discussion following Corollary 5.1 for a de๏ฌnition of Toeplitz operators. Unitary equivalence of characteristic functions is de๏ฌned in the natural way: ฮ is unitarily equivalent to ฮโฒ if ๐ ฮ(๐ง) = ฮโฒ (๐ง)๐, ๐ง โ ๐ป, for some unitary operator ๐ . It may be useful to contrast this notion of unitary equivalence with the weaker notion of coincidence, which is the appropriate concept in the study of functional models for contractions [25]. Two operator-valued analytic functions ฮ and ฮโฒ are said to coincide if there exist unitary operators ๐ and ๐ such that ๐ ฮ(๐ง) = ฮโฒ (๐ง)๐ for all ๐ง โ ๐ป. Proposition 4.5 can now be reformulated as follows. Corollary 5.1. Let ๐ and ๐โฒ be two bi-isometries with characteristic functions ฮ and ฮโฒ , respectively. (1) The characteristic function of ๐ โ ๐โฒ is given by ฮ(๐ง) โ ฮโฒ (๐ง) for ๐ง โ ๐ป. (2) If ๐ is unitarily equivalent to ๐โฒ , then ฮ is unitarily equivalent to ฮโฒ . (3) Assume in addition that ๐ and ๐โฒ are {0}-cnu. If ฮ is unitarily equivalent to ฮโฒ then ๐ is unitarily equivalent to ๐โฒ . (4) For every contractive analytic function ฮ : ๐ป โ โ(๐), there exists a {0}-cnu bi-isometry ๐ such that ฮ๐ is unitarily equivalent to ฮ. In order to translate the result of Proposition 4.4 into function theoretical terms we need some notation. First, given a separable, complex Hilbert space ๐, we denote as usual by ๐ป 2 (๐) the Hilbert space of all square summable power series with coe๏ฌcients in ๐. Given a contractive analytic function ฮ : ๐ป โ โ(๐), the analytic Toeplitz operator ๐ฮ โ โ(๐ป 2 (๐)) is de๏ฌned simply as pointwise multiplication by ฮ. The particular case ฮ(๐ง) = ๐ง๐ผ๐ yields the unilateral shift ๐๐ . The minimal unitary extension of ๐๐ is the bilateral shift ๐๐ on the Hilbert space ๐ฟ2 (๐) of all square summable Laurent series with coe๏ฌcients in ๐. The extension of ๐ฮ which commutes with ๐๐ is the Laurent operator ๐ฟฮ with symbol ฮ. Now, the space ๐ฟ2 (๐) can also be viewed as the space of square integrable ๐-valued functions ๐ : ๐ = โ๐ป โ ๐. When viewed in this manner, the operator ๐ฟฮ is given by (๐ฟฮ ๐ )(๐) = ฮ(๐)๐ (๐) for almost every ๐ โ ๐, where the strong operator limit ฮ(๐) = lim ฮ(๐๐) ๐โ1
192
H. Bercovici, R.G. Douglas and C. Foias
exists almost everywhere. Similarly, the operator ๐ท = (๐ผ โ ๐ฟโฮ ๐ฟฮ )1/2 is given as a multiplication operator by the strongly measurable operator-valued function ฮ(๐) = (๐ผ โ ฮ(๐)โ ฮ(๐))1/2 ,
๐ โ ๐.
The in๏ฌnite sum
ห โ โ (๐ทโ) ห โ โ (๐ทโ) ห โ โ โ
โ
โ
(๐ทโ) appearing in Proposition 4.4 can then be identi๏ฌed with ๐ป 2 ((๐ฟฮ ๐ฟ2 (๐)โ )). The elements in this space can be viewed as functions of two variables (๐ค, ๐) โ ๐ป ร ๐, analytic in ๐ค and measurable in ๐. We are now ready to reformulate Proposition 4.4. Proposition 5.2. Let ฮ : ๐ป โ โ(๐) be a contractive analytic function, and set ฮ(๐) = (๐ผ โ ฮ(๐)โ ฮ(๐))1/2 , ๐ โ ๐. (1) The space (๐ฟฮ ๐ฟ2 (๐))โ is reducing for ๐๐ . (2) De๏ฌne the Hilbert space ๐ = ๐ป 2 (๐) โ ๐ป 2 ((๐ฟฮ ๐ฟ2 (๐))โ ), and the operators ๐0 , ๐1 โ โ(โ) by ๐0 (๐ โ ๐) = ๐ โ ๐,
๐1 (๐ โ ๐) = ๐ โ ๐,
where
๐(๐ง) = ๐ง๐ (๐ง), ๐(๐ค, ๐) = ๐๐(๐ค, ๐), ๐(๐ง) = ฮ(๐ง)๐ (๐ง), ๐(๐ค, ๐) = ฮ(๐)๐ (๐) + ๐ค๐(๐ค, ๐) for ๐ง, ๐ค โ ๐ป and ๐ โ ๐. Then (๐0 , ๐1 ) is a {0}-cnu bi-isometry whose characteristic function is unitarily equivalent to ฮ. We will use the notation ๐(ฮ) = (๐0 , ๐1 ) for the bi-isometry described in the preceding statement. The mapping ฮ โ ๐(ฮ) establishes a bijection between unitary equivalence classes of contractive analytic functions ฮ : ๐ป โ โ(๐) and unitary equivalence classes of {0}-cnu bi-isometries ๐. The formulas given for ๐(ฮ) allow, in principle, explicit calculations. A ๏ฌrst instance is the following result. Proposition 5.3. Let ฮ : ๐ป โ โ(๐) be a contractive analytic function, and denote (๐0 , ๐1 ) = ๐(ฮ). (1) The operator ๐1 is unitary if and only if ฮ is a constant unitary operator, that is, ฮ(๐ง) โก ฮ(0), and ฮ(0) is a unitary operator in โ(๐). (2) The following conditions are equivalent: (a) ๐(ฮ) is {1}-cnu. (b) the contraction ฮ(0) is completely nonunitary. Proof. If ๐1 is unitary, ๐0 must be a unilateral shift, and therefore ๐ = ๐๐ , and ๐ = ๐ฮ . It is well known that ๐ฮ is unitary if and only if ฮ is a constant unitary operator. To prove (2), assume ๏ฌrst that ๐1 โฃ๐ is unitary for some nonzero reducing subspace ๐ of ๐(ฮ). Applying part (1) of Proposition 5.1 and part (1) of this
Canonical Models for Bi-isometries
193
proposition, which has already been proved, shows that we can write ฮ = ฮโฒ โ ฮโฒโฒ , with ฮโฒโฒ a constant unitary operator acting on a nonzero space. In particular ฮ(0) has a nontrivial unitary direct summand. Conversely, assume that ฮ(0) is not completely nonunitary, so that its restriction to some nonzero invariant subspace๐0 is a unitary operator. The contractive analytic function ฮ0 : ๐ป โ โ(๐0 ) de๏ฌned by ฮ0 (๐ง) = ๐๐0 ฮ(๐ง)โฃ๐0 is such that ฮ0 (0) is unitary. The maximum principle implies that ฮ0 is constant, and ๐0 reduces each ฮ(๐ง) to ฮ0 . A second application of part (1) of Proposition 5.1, as well as the already proved part (1) of this proposition, shows that ๐1 โฃ๐ is unitary for some nonzero reducing subspace ๐ of ๐(ฮ). โก ห = If ๐ = ๐(ฮ), one can also calculate the characteristic function of ๐ (๐1 , ๐0 ), whose coe๏ฌcients are (๐ผ โ ๐1 ๐1โ )๐1โ๐ ๐0 โฃran(๐ผ โ ๐1 ๐1โ ),
๐ โฅ 0.
Thus this function is given by (๐ผ โ ๐1 ๐1โ )(๐ผ โ ๐ง๐1โ )โ1 ๐0 โฃran(๐ผ โ ๐1 ๐1โ ),
๐ง โ ๐ป.
In these formulas we use the abbreviation โranโ for the range of an operator.
6. The structure of bi-shifts Consider a bi-isometry ๐ = (๐0 , ๐1 ). As seen earlier, the operators ๐0 and ๐1 do not need to be cnu, even if ๐ is {0}-cnu and {1}-cnu. In this section we study bi-isometries for which both ๐0 and ๐1 are cnu, and such bi-isometries will be called bi-shifts. Clearly bi-shifts are both {0}-cnu and {1}-cnu. Note that the bishifts described in [11] are, in our terminology, doubly commuting bi-shifts; see Proposition 6.4 below. Proposition 6.1. Assume that the bi-isometry ๐ is both {0}-cnu and {1)-cnu. The following conditions are equivalent. (1) ๐ is a bi-shift. (2) ๐0โ๐ โ 0 and ๐1โ๐ โ 0 as ๐ โ โ in the strong operator topology. (3) The characteristic function ฮ๐ is inner (that is, ฮ๐ (๐) โ โ(๐) is an isometry for almost every ๐ โ ๐) and it enjoys the following property: (โ) There exists no inner function ฮฉ : ๐ป โ โ(๐, ๐) such that ๐ โ= {0} and ฮ๐ (๐ง)ฮฉ(๐ง) = ฮฉ(๐ง)๐,
๐ง โ ๐ป,
with a unitary operator ๐ โ โ(๐). Proof. The proposition is almost immediate, but we provide the brief argument below in order to illustrate the use of the results in the preceding section. The equivalence between (1) and (2) follows from the fact that an isometry is cnu if and only if it is a unilateral shift. Assume next that (2) holds so that, in particular, ๐0 has no unitary part. With the notation of the preceding sections, ๐0 = ๐0 , and ๐ด = ๐1 , so that ๐ serves as its own characteristic pair. Passing to the functional model, we identify ๐0 with the unilateral shift ๐๐ , in which case ๐1 = ๐ฮ for
194
H. Bercovici, R.G. Douglas and C. Foias
some operator-valued function ฮ. The function ฮ must then be inner because ๐ฮ is an isometry. Assume now that a function ฮฉ exists with the properties in (โ). Then it follows that ๐ฮ โฃฮฉ๐ป 2 (๐) is a unitary operator, unitarily equivalent to ๐๐ โ โ(๐ป 2 (๐)). This contradicts the assumption that (2) holds, and we conclude that (3) is true. Finally, assume that (3) holds, but (2) does not. Since ๐๐ is completely nonunitary, the operator ๐ฮ must have a unitary part. The nonzero space โ โ โฉ โฉ ๐= ๐1๐ โ = ๐ฮ๐ ๐ป 2 (๐) ๐=0
๐=0
on which this unitary part acts is obviously invariant for ๐๐ , and the BeurlingLax-Halmos theorem implies that ๐ = ฮฉ๐ป 2 (๐) for some inner function ฮฉ : ๐ป โ โ(๐, ๐) with ๐ โ= {0}. The operator ๐ฮฉโ1๐ฮ ๐ฮฉ is then a unitary operator in the commutant of ๐๐ , and such operators are of the form ๐๐ for some unitary operator ๐ โ โ(๐). We conclude that ๐ฮ ๐ฮฉ = ๐ฮฉ ๐๐ , contrary to (3). โก Remark 6.2. The example ฮ(๐ง) โก ๐ผ๐ , ๐ง โ ๐ป, shows that condition (โ) is needed in the third statement of the preceding proposition. Proposition 6.1 shows that the construction of bi-shifts requires the construction of appropriate inner functions ฮ : ๐ป โ โ(๐). We start with some simple examples. Fix a nonzero Hilbert space ๐ and an inner function ๐ โ ๐ป โ . We can then form the bi-isometry ๐(๐ โ ๐ผ๐ ) = (๐๐ , ๐๐โ๐ผ๐ ). This is easily seen to be a bi-shift provided that ๐ is not constant. Proposition 6.3. Given two nonconstant inner functions ๐1 , ๐2 โ ๐ป โ , the bi-shifts ๐(๐1 โ ๐ผ๐ ) and ๐(๐2 โ ๐ผ๐ ) are quasi-similar if and only if ๐1 = ๐2 . Proof. Let ๐ โ โ(๐(๐1 โ ๐ผ๐ ), ๐(๐2 โ ๐ผ๐ )) be a quasi-a๏ฌnity. We have ๐ โ (๐๐ )โฒ , and therefore we have ๐ = ๐ฮ for some outer function ฮ : ๐ป โ โ(๐). The relation ๐๐๐1 โ๐ผ๐ = ๐๐2 โ๐ผ๐ ๐ implies that (๐1 (๐ง) โ ๐2 (๐ง))ฮ(๐ง) = ฮ(๐ง)(๐1 (๐ง) โ ๐ผ๐ ) โ (๐2 (๐ง) โ ๐ผ๐ )ฮ(๐ง),
๐ง โ ๐ป.
The operator ฮ(๐ง) has dense range for every ๐ง โ ๐ป, and we conclude that ๐1 = ๐2 . The converse is immediate. โก As pointed out earlier, the bi-isometries ๐(ฮ1 ) and ๐(ฮ2 ), ฮ1 , ฮ2 : ๐ป โ ๐, are unitarily equivalent if and only if the functions ฮ1 and ฮ2 are unitarily equivalent, that is, ๐ ฮ1 (๐ง) = ฮ2 (๐ง)๐ for a unitary operator ๐ independent of ๐ง โ ๐ป. Similarity of the two bi-isometries requires the existence of an invertible outer function ฮจ : ๐ป โ โ(๐) such that ฮจ(๐ง)ฮ1 (๐ง) = ฮ2 (๐ง)ฮจ(๐ง) for all ๐ง โ ๐ป. Another important family of bi-shifts is de๏ฌned on the Hardy space ๐ป 2 (๐ป2 )โ ๐ by the formula ๐๐ = (๐0 , ๐1 ), where (๐๐ ๐ )(๐ง0 , ๐ง1 ) = ๐ง๐ ๐ (๐ง0 , ๐ง1 ),
๐ โ ๐ป 2 (๐ป2 ) โ ๐, (๐ง0 , ๐ง1 ) โ ๐ป2 .
This class of bi-isometries has a simple characterization. Parts of the following proposition are known. We include a brief argument for the readerโs convenience.
Canonical Models for Bi-isometries
195
Proposition 6.4. Assume that the bi-isometry ๐ is both {0}-cnu and {1}-cnu. The following conditions are equivalent. (1) ๐ is unitarily equivalent to ๐๐ for some Hilbert space ๐. (2) ๐ is doubly commuting, that is, ๐0 ๐1โ = ๐1โ ๐0 . (3) The characteristic function ฮ๐ is a constant isometry. (4) The pivotal operator of (๐1 , ๐0 ) is an isometry. (5) The pivotal operator of ๐ is an isometry. Proof. It is immediate that (1) implies (2). For the remainder of the argument we identify ๐ with ๐(ฮ), where ฮ : ๐ป โ โ(๐) is a contractive analytic function. Thus ๐ acts on the space ๐ described in Proposition 5.3. Assume now that (2) holds. In this case the kernel of ๐0โ must be a reducing subspace for ๐1 . This kernel consists of functions in ๐ of the form ๐ โ 0 โ 0 โ โ
โ
โ
, with ๐ โ ๐ a constant. Since ๐1 (๐ โ 0 โ 0 โ โ
โ
โ
) = ฮ๐ โ ฮ๐ โ 0 โ โ
โ
โ
, we deduce immediately that ฮ is constant and ฮ = 0, so that (3) is true. Assume now that (3) holds, so that ฮ is a constant isometry. It follows that ฮ(0) is in particular an isometry. Condition (4) follows because ฮ(0) is the pivotal operator of the pair (๐1 , ๐0 ). Assume that (4) holds, so that ฮ(0) is an isometry. Then it follows from the maximum principle that ฮ(๐ง) = ฮ(0) for all ๐ง. In particular, the function ฮ is inner, and hence ๐0 = ๐๐ and ๐1 = ๐ฮ . Note that any orthogonal decomposition ฮ(0) = ฮ1 โ ฮ2 yields a decomposition ๐ฮ = ๐ฮ1 โ ๐ฮ2 . If ฮ1 is unitary, the operator ๐ฮ1 is unitary as well, and therefore ฮ1 must act on the space {0} because ๐ was assumed to be {1}-cnu. We deduce that ฮ(0) is cnu, and thus it is unitarily equivalent to ๐๐ for some Hilbert space ๐, and in this case ๐(ฮ) is unitarily equivalent to ๐๐ . So far we have proved that conditions (1โ4) are equivalent. The equivalence of (5) with these conditions follows from the symmetry of (2). โก The example of the constant function ฮ(๐ง) โก ๐ผ, ๐ง โ ๐ป, shows why the assumption that ๐ is both {0}-cnu and {1}-cnu is needed in the preceding proposition. If two isometries are quasi-similar and one of them is a shift, then the other one is a shift as well. It follows that a bi-isometry quasi-similar to a bi-shift must also be a bi-shift. We conclude this section with some simple properties of those bi-shifts which are similar to ๐๐ for some ๐. Proposition 6.5. Let ๐ = ๐(ฮ) be a bi-shift, where ฮ : ๐ป โ โ(๐) is an inner analytic function. Assume further that ๐ is similar to ๐๐ for some Hilbert space ๐. Then the following assertions are true. (1) The pivotal operator is similar to a unilateral shift. (2) There exists a bounded analytic function ฮฉ : ๐ป โ โ(๐) such that ฮฉ(๐ง)ฮ(๐ง) = ๐ผ,
๐ง โ ๐ป.
(3) The operator ฮ(๐ง) is similar to a unilateral shift for every ๐ง โ ๐ป.
196
H. Bercovici, R.G. Douglas and C. Foias
Proof. We argue ๏ฌrst that two similar bi-isometries have similar pivotal operators. Indeed, assume that ๐ โ โ(๐(1) , ๐(2) ) is an invertible operator. We have then (1) (2) (1)โ ๐ran๐0 = ran๐0 , and this implies that ๐ker ๐ (2)โ ๐โฃ ker ๐0 is an invertible 0 operator intertwining the two pivotal operators. Now, the pivotal operator of ๐๐ is a shift, and the preceding observation implies (1). By symmetry, we also deduce that ฮ(0) is similar to a shift, and then (2) follows from the main result of [24]. To verify (3), we observe that the bi-shift ๐๐ is unitarily equivalent to ๐(ฮ1 ), where ฮ1 (๐ง) โก ๐ for ๐ง โ ๐ป, with ๐ โ โ(๐) a unilateral shift. Let ๐ โ โ(๐(ฮ), ๐(ฮ1 )) be an invertible operator. We have ๐ โ (๐0 )โฒ , and therefore the operator ๐ is of the form ๐ = ๐ฮ for some bounded analytic function ฮ โ โ(๐). The fact that ๐ is invertible implies that ๐(๐ง) is invertible for every ๐ง โ ๐ป, and the relation ๐๐ฮ = ๐ฮ1 ๐ shows that ฮ(๐ง) is similar to ๐ = ฮ1 (๐ง). The proposition is proved. โก A di๏ฌerent approach to the similarity between a contraction and an isometry is described in [14]. This approach may also be useful in the study of similarities between bi-shifts. In contrast to certain results from the model theory of contractions, condition (2) in the above proposition is not su๏ฌcient to imply the similarity of ๐(ฮ) to a bi-shift of the form ๐๐ . This is illustrated by the following example. Example 6.6. De๏ฌne ฮ(๐ง) โ โ(โ2 ) using the in๏ฌnite matrix โก 3 โค 0 โ
โ
โ
5 ๐(๐ง) 0 0 โข 4๐ง 0 0 0 โ
โ
โ
โฅ โข 5 โฅ โข 0 1 0 0 โ
โ
โ
โฅ ฮ(๐ง) = โข โฅ , ๐ง โ ๐ป, โข 0 0 1 0 โ
โ
โ
โฅ โฃ โฆ .. .. .. . . . โ
โ
โ
. . . where ๐ โ ๐ป โ is an inner function such that ๐(0) โ= 0. The operator ฮ(0) has the eigenvalue 3๐(0)/5 and therefore it is not similar to a shift. However ฮ satis๏ฌes condition (2) in the preceding proposition. One left inverse is given by โก 5 โค ๐(๐ง) 0 0 โ
โ
โ
3๐(0) โข 0 0 1 0 โ
โ
โ
โฅ โข โฅ โข 0 0 0 1 โ
โ
โ
โฅ โฅ , ๐ง โ ๐ป, ฮฉ(๐ง) = โข โข .. โฅ โข 0 โฅ . 0 0 0 โฃ โฆ .. .. .. .. .. . . . . . with
[ ] 4 ๐(๐ง) ๐(๐ง) = 1โ , 5๐ง ๐(0)
๐ง โ ๐ป.
The reader will verify without di๏ฌculty that ๐(ฮ) is indeed a bi-shift.
Canonical Models for Bi-isometries
197
7. The unitary invariants of a functional model Bi-isometries ๐ = (๐0 , ๐1 ) with the property that the product ๐0 ๐1 is a shift were classi๏ฌed, up to unitary equivalence, in [4]; see also [2]. The parameters in that classi๏ฌcation are pairs (๐, ๐ ), where ๐ is a unitary operator on a Hilbert space ๐, and ๐ is an orthogonal projection on ๐. In this section we consider the characteristic functions of such bi-isometries. The bi-isometry ๐ = (๐0 , ๐1 ) associated to the pair (๐, ๐ ) acts on ๐ป 2 (๐) and is de๏ฌned by (๐0 ๐ )(๐ง) = ๐ (๐ง๐ +๐ โฅ )๐ (๐ง), (๐1 ๐ )(๐ง) = (๐ +๐ง๐ โฅ )๐ โ ๐ (๐ง),
๐ โ ๐ป 2 (๐), ๐ง โ ๐ป.
The space ๐ is identi๏ฌed with the space ker(๐0 ๐1 )โ of constant functions in ๐ป 2 (๐), while the range of ๐ โฅ is identi๏ฌed with ker ๐1โ . For a constant function ๐0 โ ๐ we have (7.1) ๐ ๐0 = ๐0 ๐0 , ๐0 โ ๐ โฅ ๐, while for ๐0 โ ๐ ๐ we have ๐0 ๐0 = ๐ง๐ ๐0 = ๐0 ๐1 ๐ ๐0 . Therefore the vector ๐0 = ๐1 ๐ ๐0 is in the range of ๐1 , and we ๏ฌnd that ๐ ๐0 = ๐1โ ๐0 ,
๐0 โ ๐ ๐.
(7.2)
From this we easily conclude that ker ๐0โ = ๐ ๐ ๐ = ๐1โ ๐ ๐. By reversing the order of these observations we easily deduce the following result. Proposition 7.1. Let ๐ = (๐0 , ๐1 ) be a bi-isometry on โ. De๏ฌne spaces ๐ = ker(๐0 ๐1 )โ ,
๐ = ker ๐0โ ,
๐ = ker ๐1โ .
(1) We have ๐ = ๐ โ ๐0 ๐ = ๐1 ๐ โ ๐. (2) The operator ๐ : ๐ โ ๐ de๏ฌned by ๐ (๐1 ๐ + ๐ ) = ๐ + ๐0 ๐
๐ โ ๐, ๐ โ ๐,
is unitary. (3) The bi-isometry associated with the pair (๐, ๐๐1 ๐ ) on ๐ is unitarily equivalent to the cnu part of ๐. For further calculation, it is convenient to replace the space ๐ by the external direct sum ๐โ๐ via the identi๏ฌcation ฮฆ : ๐โ๐ โ ๐1 ๐+๐ . With this identi๏ฌcation we obviously have [ ] ๐ผ๐ 0 โ . ฮฆ ๐ฮฆ = 0 0 Corollary 7.2. With the notation of Proposition 7.1, we have ] [ ๐1โ ๐0 โฃ๐ ๐1โ โฃ๐ โ ฮฆ ๐ฮฆ = . (๐ผ โ ๐1 ๐1โ )โฃ๐ (๐ผ โ ๐1 ๐1โ )๐0 โฃ๐
198
H. Bercovici, R.G. Douglas and C. Foias
Proof. For a vector ๐ โ ๐ we have ๐ ฮฆ(๐ โ 0) = ๐ ๐1 ๐ = ๐ = ๐1 ๐1โ ๐ + (๐ผ โ ๐1 ๐1โ )๐, and this is precisely the decomposition of this vector as an element of the space ๐1 ๐ โ ๐. Therefore ฮฆโ ๐ ฮฆ(๐ โ 0) = ๐1โ ๐ โ (๐ผ โ ๐1 ๐1โ )๐. To verify the identity involving the second column, we use a similar calculation: ๐ ฮฆ(0 โ ๐ ) = ๐ ๐ = ๐0 ๐ = ๐1 ๐1โ ๐0 ๐ + (๐ผ โ ๐1 ๐1โ )๐0 ๐, In these calculations we made use of (7.1) and (7.2).
๐ โ ๐. โก
Let us consider now a contractive analytic function ฮ : ๐ป โ โ(๐) and the functional model ๐(ฮ) = (๐0 , ๐1 ). In order to identify the space ๐, it will be useful to recall a few facts from the theory of functional models of contraction operators. Let us introduce the auxiliary space ๐ = ๐ป 2 (๐) โ (๐ฟฮ ๐ฟ2 (๐))โ , which can be viewed as a subspace of โ = ๐ป 2 (๐) โ ๐ป 2 ((๐ฟฮ ๐ฟ2 (๐))โ ). Obviously, the space ๐ is reducing for ๐0 . The space ๐ = {ฮ๐ข โ ฮ๐ข : ๐ข โ ๐ป 2 (๐)} is invariant for ๐0 , and therefore โ(ฮ) = ๐ โ ๐ is invariant for ๐0โ . The compression of ๐0 to this space is denoted ๐(ฮ), and it is called the functional model associated with ฮ. It is known that ๐(ฮ) is a completely nonunitary contraction, and the characteristic function of ๐(ฮ) coincides (in the sense de๏ฌned in [25]) with the purely contractive part of the function ฮ. A vector ๐ข โ ๐ฃ โ ๐ belongs to โ(ฮ) if and only if the measurable function ฮโ ๐ข + ฮ๐ฃ is orthogonal to ๐ป 2 (๐). In other words, we have a Fourier expansion โ โ
ฮโ ๐ข + ฮ๐ฃ =
๐ ๐ ๐๐ ,
๐=โ1 โ
with ๐๐ โ ๐. We will use the notation (ฮ ๐ข + ฮ๐ฃ)โ1 for ๐โ1 . Lemma 7.3. Viewed as a subspace of โ, we have โ(ฮ) = ๐. Moreover, ๐(ฮ) is precisely the pivotal operator associated with the bi-isometry ๐(ฮ): ๐(ฮ)โ = ๐0โ โฃ๐. Proof. In order to identify ๐, we consider its orthogonal complement which is easily calculated as ๐โฅ = ๐1 โ = ๐ โ ๐1 ๐ป 2 ((๐ฟฮ ๐ฟ2 (๐))โ ). The conclusion โ(ฮ) = ๐ then follows because โ = ๐ โ ๐1 ๐ป 2 ((๐ฟฮ ๐ฟ2 (๐))โ ).
Canonical Models for Bi-isometries
199
The identi๏ฌcation of the pivotal operator follows now from the fact that โ(ฮ) = ๐ is invariant for ๐0โ . โก Proposition 7.4. Let ฮ : ๐ป โ โ(๐) be a contractive analytic function, and ๐(ฮ) = (๐0 , ๐1 ) the corresponding model bi-isometry. Then ๐(ฮ) is unitarily equivalent to the bi-isometry associated with the pair (๐, ๐ ) of operators on ๐ โ โ(ฮ) de๏ฌned as follows: ๐ (๐ โ 0) = ฮ(0)โ ๐ โ [(๐ โ ฮฮ(0)โ ๐) โ (โฮฮ(0)โ ๐)], โ
๐ (0 โ (๐ข โ ๐ฃ)) = (ฮ ๐ข + ฮ๐ฃ)โ1 โ ๐(ฮ)(๐ข โ ๐ฃ), and
[ ๐ =
๐ผ๐ 0
0 0
๐ โ ๐,
๐ข โ ๐ฃ โ โ(ฮ),
(7.3)
] .
Proof. This proof amounts to an identi๏ฌcation of the matrix entries in Corollary 7.2. It is convenient to regard โ as an in๏ฌnite orthogonal sum โ = ๐ป 2 (๐) โ (๐ฟฮ ๐ฟ2 (๐))โ โ (๐ฟฮ ๐ฟ2 (๐))โ โ โ
โ
โ
, relative to which the operator ๐1 has the matrix โก 0 ๐ฮ โข ๐ฟฮ โฃ๐ป 2 (๐) 0 โข ๐1 = โข 0 ๐ผ (๐ฟฮ ๐ฟ2 (๐))โ โฃ .. .. . .
โค 0 โ
โ
โ
0 โ
โ
โ
โฅ โฅ . 0 โ
โ
โ
โฅ โฆ .. . . . .
We now apply the formulas in Corollary 7.2 to calculate the entries of the matrix ๐ explicitly. Thus, for ๐ โ ๐, which is viewed now as a subspace of โ, we obtain by applying the matrix above ๐1โ ๐ = ๐ฮโ ๐ = ๐๐ป 2 (๐) ฮโ ๐ = ฮ(0)โ ๐,
and (๐ผ โ ๐1 ๐1โ )๐ = ๐ โ ๐1 ฮ(0)โ ๐.
If ๐ข โ ๐ฃ โ โ(ฮ) then clearly (๐ผ โ ๐1 ๐1โ )๐0 (๐ข โ ๐ฃ) = ๐โ(ฮ) ๐0 (๐ข โ ๐ฃ) = ๐(ฮ)(๐ข โ ๐ฃ). For the ๏ฌrst direct summand in the right-hand side of (7.3), let us write ๐0 (๐ขโ๐ฃ) = ๐ขโฒ โ ๐ฃ โฒ and note that ๐1โ ๐0 (๐ข โ ๐ฃ) = ๐1โ (๐ขโฒ โ ๐ฃ โฒ ) = ๐๐ป 2 (๐) (ฮโ ๐ขโฒ + ฮ๐ฃ โฒ ). If we write the Fourier expansion ฮโ ๐ข + ฮ๐ฃ =
โ โ
๐ ๐ ๐๐ ,
๐=โ1
then ฮโ ๐ขโฒ + ฮ๐ฃ โฒ =
โ โ
๐ ๐+1 ๐๐ ,
๐=โ1
and the projection of this function onto ๐ป 2 (๐) is precisely ๐โ1 = (ฮโ ๐ข + ฮ๐ฃ)โ1 , as stated. โก
200
H. Bercovici, R.G. Douglas and C. Foias
Remark 7.5. The above results show us how to calculate the unitary invariants given the characteristic function ฮ. Conversely, one can write explicit formulas for the characteristic function of a cnu pair ๐ = (๐0 , ๐1 ) with given unitary invariants. We begin by noting a general formula for this characteristic function ฮ. We have ฮ(๐ง) = ๐๐ (๐ผ โ ๐ง๐0โ )โ1 ๐1 โฃ๐, ๐ง โ ๐ป, where ๐ = ker ๐0โ . Assume now that ๐ is a Hilbert space, ๐ โ โ(๐) is unitary, ๐ โ โ(๐) is an orthogonal projection, and ๐ is the bi-isometry on ๐ป 2 (๐) de๏ฌned using the pair (๐, ๐ ). Then the space ๐ = ker ๐0โ is identi๏ฌed with the range of ๐ ๐ ๐ โ , provided that we view the elements of ๐ as constant functions in ๐ป 2 (๐). It is easy then to see that ๐1 ๐ = ๐ โ ๐, ๐ โ ๐, so that ๐1 ๐ also consists of constant functions in ๐ป 2 (๐). Next we observe that the operator ๐0โ leaves invariant the space ๐ of constant functions, and in fact ๐0โ โฃ๐ = ๐ โฅ ๐ โ . We conclude that (๐ผ โ ๐ง๐0โ )โ1 โฃ๐ = (๐ผ โ ๐ง๐ โฅ ๐ โ )โ1 ,
๐ง โ ๐ป.
Note ๏ฌnally that ๐๐ โฃ๐ = ๐ ๐ ๐ โ . These observations allow us to write the formula ฮ(๐ง) = ๐ ๐ ๐ โ (๐ผ โ ๐ง๐ โฅ ๐ โ )โ1 ๐ โ โฃ๐,
๐ง โ ๐ป,
which involves only the unitary invariants of ๐. Note further that the unitary invariants of the pair (๐1 , ๐0 ) are (๐ โ , ๐ ๐ โฅ ๐ โ ). Thus, starting with a pair (๐0 , ๐1 ) whose characteristic function is known, one can use Proposition 3.4 to calculate the unitary invariants of (๐0 , ๐1 ), and then an application of the above formulas yield an explicit expression for the characteristic function of (๐1 , ๐0 ).
8. Examples of irreducible bi-isometries and direct integral decompositions For a single isometry, that is, when ๐ผ has only one element, it follows from the von NeumannโWold theorem that there is, up to unitary equivalence, only one nonunitary irreducible isometry. However, when ๐ผ has two or more elements there are many irreducible families of commuting isometries which do not consist of unitary operators. We will illustrate this in the case of bi-isometries ๐ = (๐0 , ๐1 ). These examples also illustrate the fact that the characteristic function is not always the best approach to the study of bi-isometries. Particular bi-isometries may be better understood using special methods. We recall that a complete unitary invariant of a completely nonunitary biisometry ๐ = (๐0 , ๐1 ) is given by a pair (๐, ๐ ), where ๐ is a unitary operator on
Canonical Models for Bi-isometries
201
some Hilbert space ๐, and ๐ is an orthogonal projection on ๐. The bi-isometry determined by (๐, ๐ ) acts on ๐ป 2 (๐) as follows: (๐0 ๐ )(๐ง) = ๐ (๐ง๐ +๐ โฅ )๐ (๐ง),
(๐1 ๐ )(๐ง) = (๐ +๐ง๐ โฅ )๐ โ ๐ (๐ง),
๐ โ ๐ป 2 (๐), ๐ง โ ๐ป,
where ๐ โฅ = ๐ผ๐ โ๐ . The bi-isometry ๐ is irreducible if and only if the pair (๐, ๐ ) is irreducible. Note for further use that the product ๐0 ๐1 is precisely multiplication by the variable ๐ง. (These unitary invariants classify more general bi-isometries than the completely nonunitary ones; see [4, 2].) For our illustration we will let ๐ be the bilateral shift on the space ๐ฟ2 of all square integrable functions on the unit circle ๐; thus (๐ ๐ )(๐) = ๐๐ (๐),
๐ โ ๐ฟ2 , ๐ โ ๐.
We will denote by ๐๐ (๐) = ๐ ๐ the standard orthonormal basis in ๐ฟ2 , and for every set ๐ด โ โค of integers we denote by ๐๐ด the orthogonal projection onto the space generated by {๐๐ : ๐ โ ๐ด}. In this case ๐0 and ๐1 are uniquely determined by the relations ๐1 ๐๐+1 = ๐๐ if ๐ โ ๐ด and ๐0 ๐๐ = ๐๐+1 if ๐ โ / ๐ด. Proposition 8.1. Two pairs (๐, ๐๐ด ), (๐, ๐๐ต ) are unitarily equivalent if and only if there exists ๐ โ โค such that ๐ต = {๐ + ๐ : ๐ โ ๐ด}. Proof. Su๏ฌciency is obvious: if ๐ต = ๐ด + ๐ then the operator ๐ ๐ implements the unitary equivalence of the two pairs. Conversely, assume that there is a unitary operator ฮฆ on ๐ฟ2 such that ฮฆ๐ = ๐ ฮฆ and ๐ ๐๐ด = ๐๐ต ๐ . There exists then a function ๐ โ ๐ฟโ such that โฃ๐โฃ = 1 almost everywhere and ฮฆ๐ = ๐๐ for every ๐ โ ๐ฟ2 . The fact that ๐๐๐ is in the range of ๐๐ต for ๐ โ ๐ด means that (๐, ๐๐โ๐ ) = (๐๐๐ , ๐๐ ) = 0, Similarly, ๐๐๐ is in the range of
๐โฅ ๐ต
๐ โ ๐ด, ๐ โ / ๐ต.
if ๐ โ / ๐ด, so that
(๐, ๐๐โ๐ ) = 0,
๐โ / ๐ด, ๐ โ ๐ต.
We deduce that there exists at least one integer ๐ not in the set {๐ โ ๐ : (๐, ๐) โ (๐ด ร (โค โ ๐ต)) โช ((โค โ ๐ด) ร ๐ต}. The function ๐๐ will then have the property that ๐๐+๐ = ๐๐ ๐๐ is in the range of ๐๐ต if ๐ โ ๐ด, and it is in the range of ๐โฅ / ๐ด. ๐ต if ๐ โ Therefore ๐ต = ๐ด + ๐. โก Corollary 8.2. The pair (๐, ๐๐ด ) is reducible if and only of ๐ด is a periodic set; that is, ๐ด = ๐ด + ๐ for some nonzero integer ๐. Proof. The pair (๐, ๐๐ด ) is reducible if and only if it commutes with a unitary which is not a scalar multiple of the identity. The argument in the proof of the preceding proposition shows that such a unitary can be chosen to be multiplication by ๐๐ for some ๐ โ โค โ {0}. โก We see therefore that there is a continuum of mutually inequivalent irreducible bi-isometries. Indeed, there is a continuum of subsets of โค, and only countably many of them are periodic.
202
H. Bercovici, R.G. Douglas and C. Foias
Quite interestingly, the bi-isometry associated with (๐, ๐๐ด ) can be described very explicitly. Consider the space ๐ฟ2 (๐2 ) = ๐ฟ2 โ ๐ฟ2 and its standard orthonormal basis ๐๐๐ (๐0 , ๐1 ) = ๐0๐ ๐1๐ , ๐, ๐ โ โค, ๐0 , ๐1 โ ๐. Multiplication by the two variables de๏ฌnes a bi-isometry ๐ = (๐0 , ๐1 ) on ๐ฟ2 (๐2 ); actually ๐0 and ๐1 are unitary. We will look at proper nonempty subsets ฮ โ โค2 with the property that the space โฮ generated by {๐๐๐ : (๐, ๐) โ ฮ} is invariant for ๐. In other words, (๐ + ๐, ๐ + ๐) โ ฮ if (๐, ๐) โ ฮ and ๐, ๐ โฅ 0 or, equivalently, ฮ + โ2 โ ฮ. We de๏ฌne the boundary โฮ of ฮ to consist of those pairs (๐, ๐) โ ฮ such that (๐ โ 1, ๐ โ 1) does not belong to ฮ. For each integer ๐, there exists a unique point ๐พ๐ = (๐๐ , ๐๐ ) โ โฮ such that ๐๐ โ ๐๐ = ๐. Uniqueness is obvious by the de๏ฌnition of โฮ; existence follows from the fact that โ
โ= ฮ โ= โค2 . The di๏ฌerence ๐พ๐+1 โ ๐พ๐ = (๐๐+1 โ ๐๐ , ๐๐+1 โ ๐๐ ) is either (1, 0) or (0, โ1). We can then de๏ฌne the set ๐ดฮ โ โค by ๐ดฮ = {๐ โ โค : ๐พ๐+1 โ ๐พ๐ = (0, โ1)}. Geometrically, ๐ดฮ is the union of the vertical segments in โฮ, omitting the lower endpoint of each one. The following result is an easy exercise. Proposition 8.3. For every subset ๐ด โ โค there exists a nonempty subset ฮ โ โค2 such that ฮ + โ2 โ ฮ and ๐ดฮ = ๐ด. We have ๐ดฮ+(๐,๐) = ๐ดฮ + ๐ โ ๐ for all (๐, ๐) โ โค2 . Proposition 8.4. Let ฮ be a nonempty proper subset of โค2 such that โฮ is invariant for ๐. The bi-isometry associated with the invariants (๐, ๐๐ดฮ ) is unitarily equivalent to ๐โฃโฮ . Proof. The space โโฮ = โฮ โ ๐ ๐ โฮ can be identi๏ฌed with ๐ฟ2 by mapping ๐๐พ๐ to ๐๐ . Denote by ๐0 the unitary operator on โโฮ which corresponds to the shift on ๐ฟ2 ; in other words, ๐0 ๐๐พ๐ = ๐๐พ๐+1 . Since ๐0 ๐1 corresponds with multiplication by ๐ง, it is clear that โโฮ can be identi๏ฌed with ๐ป 2 (โโฮ ). Therefore, we only need to show that ๐0 ๐๐พ๐ = ๐0 ๐1 ๐๐พ๐+1 if ๐ โ ๐ดฮ and ๐0 ๐๐พ๐ = ๐๐พ๐+1 if ๐ โ / ๐ดฮ . This however is immediate from the de๏ฌnition of ๐ดฮ and the remark preceding Proposition 8.1. โก A direct consequence of this proposition is the following: Corollary 8.5. Let ฮ and ฮโฒ be two nonempty proper subsets of โค2 such that โฮ and โฮโฒ are invariant for ๐. (1) The bi-isometries ๐โฃโฮ and ๐โฃโฮโฒ are unitarily equivalent if and only if ฮโฒ = ฮ + ๐พ for some ๐พ โ โค2 . (2) The bi-isometry ๐โฃโฮ is reducible if and only if ฮ = ฮ + ๐พ for some ๐พ โ โค2 โ {(0, 0)}. Two particular sets ฮ yielding irreducible bi-isometries were considered in [11, 17]. The ๏ฌrst is ฮ = โ2 , for which ๐ดฮ = {๐ : ๐ < 0}. The restriction ๐โฃโฮ is a doubly commuting bi-shift. The second is ฮ = (โค ร โ) โช (โ ร โค), for which
Canonical Models for Bi-isometries
203
๐ดฮ = โ. The corresponding restriction of ๐ was called a modi๏ฌed bi-shift in these works. The modi๏ฌed bi-shift can be seen to be the dual of the doubly commuting bi-shift in the sense of [6]. The bi-isometries of the form ๐โฃโฮ were considered earlier in [18]. They have the special property that the range projections of the isometries in the multiplicative semigroup they generate commute with each other. The case ฮ โ โ2 was also considered in [8] from the point of view of Hilbert modules over the bidisk algebra. We now illustrate the decomposition of a bi-isometry into a direct integral of irreducibles with the particular case provided by the set ๐ด = 2โค. In this case, the commutant of the pair (๐, ๐๐ด ) is the algebra generated by ๐ 2 , and this operator is a unitary operator with uniform multiplicity 2 relative to the usual arclength measure on ๐. This is realized upon using the identi๏ฌcation ฮฆ : ๐ฟ 2 โ ๐ฟ 2 โ ๐ฟ2 de๏ฌned by
(ฮฆ(๐ โ ๐))(๐) = ๐ (๐ 2 ) + ๐๐(๐ 2 ), ๐ โ ๐. The operator ฮฆโ ๐ ฮฆ is simply multiplication by the matrix-valued function ] [ 0 ๐ , ๐ โ ๐, ๐0 (๐) = 1 0 while ฮฆโ ๐๐ด ฮฆ is multiplication by the constant matrix ] [ 1 0 . ๐0 = 0 0 In other words, we have the decomposition โซ โ (๐, ๐ ) = (๐0 (๐), ๐0 )โฃ๐๐โฃ, ๐
and it is clear that the pairs (๐0 (๐), ๐0 ) are irreducible and mutually inequivalent. This corresponds with a direct integral decomposition of the corresponding bi-isometry. The reader will have no di๏ฌculty verifying that the bi-isometry associated with (๐0 (๐), ๐0 ) is of the form (๐๐, ๐), where ๐ is a unilateral shift of multiplicity one. The general case of a set ๐ด such that ๐ด = ๐ด + ๐, with ๐ > 2, lends itself to a similar analysis, with โค โก 0 0 0 โ
โ
โ
0 ๐ ๐โ1 โข 1 0 0 โ
โ
โ
0 0 โฅ โฅ โข โข 0 ๐ 0 โ
โ
โ
0 0 โฅ โฅ โข ๐0 (๐) = โข . . . . .. .. โฅ , ๐ โ ๐, .. โข .. .. .. . . โฅ โฅ โข โฃ 0 0 0 โ
โ
โ
0 0 โฆ 0 0 0 โ
โ
โ
๐ ๐โ2 0 and ๐0 a diagonal projection. The diagonal elements (๐ผ1 , ๐ผ2 , . . . , ๐ผ๐ ) of this projection are de๏ฌned by setting ๐ผ๐ = 1 if ๐ โ ๐ด and ๐ผ๐ = 0 otherwise. The pair (๐0 (๐), ๐0 ) is irreducible provided that ๐ is the smallest positive period of ๐ด.
204
H. Bercovici, R.G. Douglas and C. Foias
References [1] O.P. Agrawal, D.N. Clark and R.G. Douglas, Invariant subspaces in the polydisk, Paci๏ฌc J. Math. 121 (1986), 1โ11. [2] H. Bercovici, R.G. Douglas and C. Foias, On the classi๏ฌcation of multi-isometries, Acta Sci. Math. (Szeged) 72 (2006) no. 3-4, 639โ661. [3] โโโ, Bi-isometries and commutant lifting, Oper. Theory Adv. Appl. 197 (2010), 51โ76. [4] C.A. Berger, L. Coburn, and A. Lebow, Representation and index theory for ๐ถ โ algebras generated by commuting isometries, J. Funct. Anal. 27 (1978), 51โ99. [5] A. Beurling, On two problems concerning linear transformations in Hilbert space, Acta Math. 81 (1949), 239โ255. [6] J.B. Conway, The dual of a subnormal operator, J. Operator Theory 5 (1981), 195โ211. [7] R.G. Douglas, On extending commutative semigroups of isometries, Bull. London Math. Soc. 1 (1969), 157โ159. [8] R.G. Douglas, T. Nakazi, and M. Seto, Shift operators on the โ2 -valued Hardy space, Acta Sci. Math. (Szeged) 73 (2007), 729โ744. [9] P.L. Duren, Theory of ๐ป ๐ spaces, Academic Press, New York-London, 1970. [10] C. Foias, A.E. Frazho, I. Gohberg, and M.A. Kaashoek, Metric constrained interpolation, commutant lifting and systems, Birkhยจ auser Verlag, Basel, 1998. [11] D. Gaยธspar and P. Gaยธspar, Wold decompositions and the unitary model for biisometries, Integral Equations Operator Theory 49 (2004), 419โ433. [12] D. Gaยธspar and N. Suciu, Wold decompositions for commutative families of isometries, An. Univ. Timisoara Ser. Stint. Mat. 27 (1989), 31โ38. [13] P. Ghatage and V. Mandrekar, On Beurling type invariant subspaces of ๐ฟ2 (๐ 2 ) and their equivalence, J. Operator Theory 20 (1988), 83โ89. [14] H.-K. Kwon and S. Treil, Similarity of operators and geometry of eigenvector bundles, Publ. Mat. 53 (2009), 417โ438. [15] D. Popovici, On the structure of c.n.u. bi-isometries, Acta Sci. Math. (Szeged) 66 (2000), 719โ729. [16] โโโ, On the structure of c.n.u. bi-isometries. II, Acta Sci. Math. (Szeged) 68 (2002), 329โ347. [17] โโโ, A Wold-type decomposition for commuting isometric pairs, Proc. Amer. Math. Soc. 132 (2004), 2303โ2314. [18] H.N. Salas, Semigroups of isometries with commuting range properties, J. Operator Theory 14 (1985), 311โ346. [19] M. Sฬทlocinski, On Wold type decomposition of a pair of commuting isometries, Ann. Pol. Math. 37 (1980), 255โ262. [20] I. Suciu, On the semigroups of isometries, Studia Math. 30 (1968), 101โ110. [21] B. Sz.-Nagy, Unitary dilations of Hilbert space operators and related topics, Conference Board of Mathematical Sciences Regional Conference Series in Mathematics, No. 19, American Mathematical Society, Providence, R.I., 1974.
Canonical Models for Bi-isometries
205
[22] โโโ, Sur les contractions de lโespace de Hilbert, Acta Sci. Math. (Szeged) 15 (1953), 87โ92. [23] โโโ, Sur les contractions de lโespace de Hilbert. II, Acta Sci. Math. (Szeged) 18 (1957), 1โ14. [24] B. Sz.-Nagy and C. Foias, On contractions similar to isometries and Toeplitz operators, Ann. Acad. Sci. Fenn. Ser. A I Math. 2 (1976), 553โ564. [25] B. Sz.-Nagy, C. Foias, H. Bercovici, and L. Kยดerchy, Harmonic Analysis of Operators on Hilbert Spaces, Second Edition, Springer Verlag, New York, 2010. [26] M. Takesaki, Theory of operator algebras. I, Springer-Verlag, Berlin, 2002. [27] J. von Neumann, Allgemeine Eigenwerttheorie Hermitescher Funktionaloperatoren, Math. Ann. 102 (1929), 49โ131. [28] H. Wold, A study in the analysis of stationary time series, Stockholm, 1954. H. Bercovici Department of Mathematics Indiana University Bloomington, IN 47405, USA e-mail:
[email protected] R.G. Douglas and C. Foias Department of Mathematics Texas A&M University College Station, TX 77843, USA e-mail:
[email protected] [email protected]
Operator Theory: Advances and Applications, Vol. 218, 207โ224 c 2012 Springer Basel AG โ
First-order Trace Formulae for the Iterates of the FoxโLi Operator Albrecht Bยจottcher, Sergei Grudsky, Daan Huybrechs and Arieh Iserles To the memory of Israel Gohberg, one of the pioneers of contemporary WienerโHopf theory
Abstract. The paper is devoted to ๏ฌrst-order trace formulas for the iterates of the FoxโLi and related WienerโHopf integral operators. Such formulas provide ๏ฌrst insight into the asymptotic behaviour of the eigenvalues and can be used to test whether a speci๏ฌc guess for the eigenvalue distribution is acceptable or not. The main technical problem consists in obtaining the asymptotics of a multivariate oscillatory integral whose stationary points constitute a line. Mathematics Subject Classi๏ฌcation (2000). Primary 47B35; Secondary 45C05, 47B10, 78A60. Keywords. FoxโLi operator, WienerโHopf operator, eigenvalue, trace formula.
1. Introduction and main results The FoxโLi operator is the integral operator on ๐ฟ2 (โ1, 1) given by โ โซ 1 2 ๐ (๐น๐ ๐ )(๐ฅ) := ei๐(๐ฅโ๐ฆ) ๐ (๐ฆ) d๐ฆ, ๐ฅ โ (โ1, 1), ๐i โ1 โ where ๐ > 0 is a large parameter. Here and in the following, i stands for ei๐/4 . The spectrum of this operator is of great importance in laser engineering [12], [13], [15], [18], [19]. Physical aspects of the FoxโLi spectrum are also studied in the recent papers [2], [3], [4]. A very recent paper devoted to numerical methods for the approximation of the spectrum is [11]. These works indicate that the spectrum of ๐น๐ is composed of points on a spiral commencing at 1 and rotating clockwise to the origin. Sergei Grudsky acknowledges support of this work by a grant of the DAAD.
208
A. Bยจ ottcher, S. Grudsky, D. Huybrechs and A. Iserles
However, rigorous results are still very sparse. Landau and Widom [16], [20] established a second-order result for the asymptotic distribution of the singular values of ๐น๐ , that is, for the square roots of the eigenvalues of ๐น๐ ๐น๐โ . In [7], the regularized operator given by โ โซ 1 2 ๐ (๐น๐,๐ ๐ )(๐ฅ) := e(iโ๐)๐(๐ฅโ๐ฆ) ๐ (๐ฆ) d๐ฆ, ๐ฅ โ (โ1, 1), ๐i โ1 was considered, and it was proved that, for each ๏ฌxed ๐ > 0, the eigenvalues of ๐น๐,๐ converge to the logarithmic spiral { ( ) } 1 ๐๐ฅ2 ๐ฅ2 โ exp โ โ i : ๐ฅ โ (0, โ) 4(1 + ๐2 ) 4(1 + ๐2 ) 1 + ๐i in the Hausdor๏ฌ metric as ๐ โ โ. For ๐ = 0, this logarithmic spiral becomes the unit circle, and it is conjectured that the eigenvalues of ๐น๐ lie on spirals which eventually wind up closer and closer to the unit circle as ๐ goes to in๏ฌnity. We here prove the following ๏ฌrst-order trace formula for the eigenvalues of the FoxโLi operator ๐น๐ itself. Theorem 1.1. The operator ๐น๐ is a trace class operator, all eigenvalues are contained in the open unit disk, and, for each ๏ฌxed natural number ๐ โฅ 1, โ โ 2 ๐ tr ๐น๐๐ = โ + ๐( ๐) as ๐ โ โ. (1) ๐i๐ We do not know a rigorous argument that shows that ๐น๐ has in๏ฌnitely many eigenvalues. However, Theorem 1.1 for ๐ = 1 implies the following. Since it tells us that ๐น๐ is of trace class, we may compute the trace by integrating the kernel along the diagonal [14, Corollary III.10.2]. On the other hand, the trace is the sum of the eigenvalues ๐1 , ๐2 , . . . of ๐น๐ , repeated according to algebraic multiplicity. Consequently, โ โซ 1 โ โ 2 ๐ 2 ๐ ๐๐ = tr ๐น๐ = ei๐โ
0 d๐ฅ = โ , ๐i โ1 ๐i ๐ and since โฃ๐๐ โฃ < 1 for all ๐, it follows that the number ๐ of eigenvalues satis๏ฌes $ $ โ ๐ ๐ 2 ๐ $$ โ $$ โ โ =$ ๐๐ $ โค โฃ๐๐ โฃ < ๐, $ $ ๐ ๐=1 ๐=1 โ which leastwise reveals that ๐น๐ has at least 2 ๐/๐ eigenvalues. Given a family of functions ๐๐ : (0, โ) โ โ, we say that the eigenvalues of ๐น๐ are asymptotically distributed as the values of ๐๐ in the weak sense if, for each natural number ๐ โฅ 1, โซ โ โ ๐ tr ๐น๐ = ๐๐๐ (๐ฅ) d๐ฅ + ๐( ๐) as ๐ โ โ. (2) 0
Iterates of the FoxโLi Operator
209
Of course, saying so is motivated by the formulae โซ โ โ โ tr ๐น๐๐ = ๐๐๐ , ๐๐๐ (๐ฅ) d๐ฅ โ ๐๐๐ (๐). ๐
0
๐
We remark that equal asymptotic distribution in the strong sense would mean something like โซ โ โ tr ๐(๐น๐ ) = ๐(๐๐ (๐ฅ)) d๐ฅ + ๐( ๐) as ๐ โ โ 0
for every function ๐ โ ๐ถ โ (โ) satisfying ๐(0) = 0 or for every ๐ of the form ๐ = ๐ฮฉ , where ฮฉ โ โ โ {0} is a nice set and ๐ฮฉ stands for the characteristic function of ฮฉ. In the latter case, it would follow that โ #{๐ : ๐๐ โ ฮฉ} = โฃ{๐ฅ โ (0, โ) : ๐(๐ฅ) โ ฮฉ}โฃ + ๐( ๐) as ๐ โ โ, where #๐ and โฃ๐โฃ denote the cardinality and the Lebesgue measure of ๐, respectively. By Theorem 1.1, formula (2) is equivalent to saying that โ โซ โ โ 2 ๐ ๐๐๐ (๐ฅ) d๐ฅ = โ + ๐( ๐) as ๐ โ โ. (3) ๐i๐ 0 Using solely (3) we will show the following. Theorem 1.2. Let ๐๐ (๐ฅ) = exp(โ๐ผ(๐)๐ฅ๐ โ i๐ฝ(๐)๐ฅ๐ ) with positive real numbers ๐ผ(๐), ๐ฝ(๐), ๐. Then the eigenvalues of ๐น๐ are asymptotically distributed as the values of ๐๐ in the weak sense if and only if ( ) ( ) 1 ๐2 1 ๐ = 2, ๐ผ(๐) = ๐ , ๐ฝ(๐) = +๐ . ๐ 16๐ ๐ This result may be viewed as a ๏ฌrst step toward establishing with mathematical rigour Vainshteinโs formula ๐(1/2)๐ 3/2 ๐2 โ , ๐ฝ(๐) โ , 3/2 16๐ 16 2 ๐ which, to quote Cochran and Hinds [12], was obtained by Vainshtein [19] โusing a distinctly physical approach, based upon wave-guide theory.โ Here ๐(1/2) is Riemannโs zeta function at the point 1/2. ๐ = 2,
๐ผ(๐) โ
The FoxโLi operator ๐น๐ is easily seen to be unitarily similar to the operator โ on ๐ฟ2 (0, 2 ๐) that acts by the rule โซ 2โ๐ โ 2 1 (๐ ๐ )(๐ฅ) := โ ei(๐ฅโ๐ฆ) ๐ (๐ฆ) d๐ฆ, ๐ฅ โ (0, 2 ๐). (4) ๐i 0 In this way we are entering the realm of WienerโHopf operators. Given a function ๐ โ ๐ฟโ (โ), the so-called symbol, the convolution operator ๐ถ(๐) on ๐ฟ2 (โ) is de๏ฌned by โซ โ โซ โ 1 โi๐๐ฅ (๐ถ(๐)๐ )(๐ฅ) := e ๐(๐) ei๐๐ฆ ๐ (๐ฆ) d๐ฆ d๐, ๐ โ โ. 2๐ โโ โโ
210
A. Bยจ ottcher, S. Grudsky, D. Huybrechs and A. Iserles
Thus, ๐ถ(๐) takes the Fourier transform, multiplies the result by ๐, and then applies the inverse Fourier transform. The boundedness of ๐ guarantees (and is in fact equivalent to) the boundedness of ๐ถ(๐). The WienerโHopf operator ๐ (๐) is the compression of ๐ถ(๐) to ๐ฟ2 (0, โ), that is, ๐ (๐) = ๐ ๐ถ(๐)โฃ๐ฟ2 (0, โ), where (๐ ๐ )(๐ฅ) is zero for ๐ฅ < 0 and ๐ (๐ฅ) for ๐ฅ > 0. Finally, for a real number ๐ > 0, the truncated WienerโHopf operator ๐๐ (๐) is the compression of ๐ (๐) to ๐ฟ2 (0, ๐ ), i.e., ๐๐ (๐) = ๐๐ ๐ (๐)โฃ๐ฟ2 (0, ๐ ), where (๐๐ ๐ )(๐ฅ) = ๐ (๐ฅ) for 0 < ๐ฅ < ๐ and (๐๐ ๐ )(๐ฅ) = 0 for ๐ฅ > ๐ . If โซ โ ห ๐(๐) = โ(๐) := โ(๐ก)ei๐๐ก d๐ก, ๐ก โ โ, โโ
the Fourier transform being understood in the usual sense for โ โ ๐ฟ1 (โ) โช ๐ฟ2 (โ) and in the sense of distributions in more general situations, the convolution ๐ถ(๐) can be written as โซ โ (๐ถ(๐)๐ )(๐ฅ) = โ(๐ฅ โ ๐ฆ)๐ (๐ฆ) d๐ฆ, ๐ฅ โ โ, โโ
while the operators ๐ (๐) and ๐๐ (๐) are given by the same formula with integration over (โโ, โ) replaced by integration over (0, โ) and (0, ๐ ), respectively. Because โซ โ
โโ
2
ei๐ก ei๐๐ก d๐ก = eโi๐
2
/4
,
๐โโ
in the sense of distributions, we may identify the operator (3) as ๐2โ๐ (๐) with 2 ๐(๐) := eโi๐ /4 . We remark that ๐(๐) has oscillating discontinuities as ๐ โ ยฑโ and that this function does not belong to the classes of symbols with a well-developed ห + ๐ป โ (โ), ๐ ๐ถ(โ), ๐๐(โ), theory of their WienerโHopf operators, such as ๐ถ(โ) ๐๐ด๐ (โ); see [8] and [10]. In terms of WienerโHopf operators, Theorem 1.1 becomes formula (5) in the following result. 2
Theorem 1.3. Let ๐(๐) := eโi๐ /4 . The spectra of the operators ๐ถ(๐) and ๐ (๐) are the unit circle ๐ and the closed unit disc ๐ป, respectively. The spectrum of ๐๐ (๐) is contained in the open unit disc ๐ป, and for every natural number ๐ โฅ 1, the operators ๐๐๐ (๐) := [๐๐ (๐)]๐ and ๐๐ (๐ ๐ ) are trace class operators and tr ๐๐๐ (๐) = tr ๐๐ (๐ ๐ ) + ๐(๐ ) have
as
๐ โ โ.
(5)
Denoting by โ๐ (๐ก) the kernel of the convolution integral operator ๐๐ (๐ ๐ ), we โซ โ ๐ โ๐ (๐ฅ โ ๐ฅ) d๐ฅ = ๐ โ๐ (0) = ๐ ๐ (๐) d๐ 2๐ 0 โโ โ โซ โ 2 ๐ ๐ 4๐ ๐ = eโi๐๐ /4 d๐ = =โ , 2๐ โโ 2๐ i๐ ๐i๐
tr ๐๐ (๐ ๐ ) =
โซ
๐
and taking into account that ๐น๐ is unitarily similar to ๐2โ๐ (๐), we see that (5) is indeed the same as (1).
Iterates of the FoxโLi Operator
211
The discrete analogues of WienerโHopf operators are Toeplitz matrices. Given ๐ โ ๐ฟโ (๐), the ๐ ร ๐ Toeplitz matrix ๐๐ (๐) is the matrix (๐๐โ๐ )๐๐,๐=1 where ๐๐ is the ๐th Fourier coe๏ฌcient of ๐, โซ 2๐ 1 ๐๐ := ๐(ei๐ )eโi๐๐ d๐, ๐ โ โค. 2๐ 0 It is well known and not di๏ฌcult to prove (see, e.g., [9, Lemma 5.16 and Theorem 5.17]) that if ๐ is an arbitrary function in ๐ฟโ (๐), then tr ๐๐๐ (๐) = tr ๐๐ (๐๐ ) + ๐(๐) = (๐๐ )0 + ๐(๐)
as ๐ โ โ,
(6)
which is the discrete counterpart of (5). A ๏ฌnite Toeplitz matrix is automatically a trace class operator, but a truncated WienerโHopf operator need not be of trace class. Therefore the continuous analogue of (6) does not make sense for arbitrary ๐ โ ๐ฟโ (โ). What is known is the following, and we will include a proof for the readerโs convenience. Theorem 1.4. If ๐ โ ๐ฟโ (โ) โฉ ๐ฟ1 (โ), then the operators ๐๐๐ (๐) and ๐๐ (๐๐ ) are of trace class for every natural number ๐ โฅ 1 and every real number ๐ > 0, and โซ โ ๐ tr ๐๐๐ (๐) = tr ๐๐ (๐๐ ) + ๐(๐ ) = ๐๐ (๐) d๐ + ๐(๐ ) as ๐ โ โ. (7) 2๐ โโ 2
The function ๐(๐) = eโi๐ /4 in Theorem 1.3 is not in ๐ฟ1 (โ) and hence Theorem 1.3 cannot be deduced from Theorem 1.4. The actual value of Theorems 1.1 and 1.3 is that they show that (7) nevertheless remains true for ๐(๐) = ๐(๐) = 2 eโi๐ /4 . The following theorem unites (5) and (7). 2
Theorem 1.5. Let ๐(๐) = ๐(๐)๐(๐) where ๐(๐) := eโi๐ /4 and ๐ โ ๐ถ 3 (โ) is a function having ๏ฌnite limits ๐(โโ) = ๐(+โ) =: ๐(โ). Set ๐ข(๐) := ๐(๐) โ ๐(โ) and suppose that the functions ๐ 4 ๐ข(๐), ๐ 3 ๐ขโฒ (๐), ๐ 2 ๐ขโฒโฒ (๐), ๐ขโฒโฒโฒ (๐) belong to ๐ฟ1 (โ) and have zero limits as ๐ โ ยฑโ. Then, for every natural number ๐ โฅ 1 and every real number ๐ > 0, the operators ๐๐๐ (๐) and ๐๐ (๐๐ ) are of trace class and (7) holds. The remaining sections of the paper are devoted to the proofs of the theorems. In Section 2, we prove Theorem 1.4 and the portion of Theorem 1.3 concerning spectra. Proposition 2.4 addresses the pseudospectra of ๐น๐ and shows that, for each ๐ > 0, the ๐-pseudospectrum of ๐น๐ contains the closed unit disk ๐ป whenever ๐ is su๏ฌciently large. Theorem 1.1 and the (equivalent) trace formula of Theorem 1.3 are proved in Section 3 by determining the ๏ฌrst-order asymptotics of the oscillatory โซ1 multivariate integral โ1 ๐๐ (๐ฅ, ๐ฅ) d๐ฅ where ๐๐ (๐ฅ, ๐ฆ) is the kernel of the integral operator ๐น๐๐ ; note that ๐๐ (๐ฅ, ๐ฆ) is a (๐ โ 1)-fold integral. Sections 4 and 5 contain the proofs of Theorems 1.2 and 1.5, respectively.
212
A. Bยจ ottcher, S. Grudsky, D. Huybrechs and A. Iserles
2. WienerโHopf operators We begin with the proof of Theorem 1.4. Let ๐๐ denote the ๐th Schattenโvon Neumann class and โฅ โ
โฅ๐ the norm in ๐๐ , that is, the โ๐ norm of the singular values of the operator. In particular, โฅ โ
โฅ1 is the trace norm, โฅ โ
โฅ2 is the Hilbertโ Schmidt norm (= Frobenius norm), and โฅ โ
โฅโ coincides with the usual operator norm on ๐ฟ2 . It is well known that ๐๐ (๐) โ ๐1 whenever ๐ โ ๐ฟโ (โ) โฉ ๐ฟ1 (โ); see, e.g., [10, Section 10.83]. This implies that ๐๐๐ (๐) and ๐๐ (๐๐ ) are also in ๐1 for ๐ โฅ 1. Lemma 2.1. If ๐, ๐ โ ๐ฟโ (โ) โฉ ๐ฟ1 (โ), then โฅ๐๐ (๐)๐๐ (๐) โ ๐๐ (๐๐)โฅ1 = ๐(๐ )
as
๐ โ โ.
ห โฉ ๐ฟ2 (โ), Proof. If ๐ โ ๐ฟโ (โ) โฉ ๐ฟ1 (โ), then ๐ โ ๐ฟ2 (โ) and ๐ = โห with โ โ ๐ถ(โ) where โห is the one-point compacti๏ฌcation of โ. We denote by ๐ป(๐) the Hankel operator generated by ๐. This is the operator that acts on ๐ฟ2 (0, โ) by the rule โซ โ (๐ป(๐)๐ )(๐ฅ) := โ(๐ฅ + ๐ฆ)๐ (๐ฆ) d๐ฆ, ๐ฅ โ (0, โ). 0
Letting ห ๐(๐) := ๐(โ๐), we have โซ (๐ป(ห ๐)๐ )(๐ฅ) =
โ 0
โ(โ๐ฅ โ ๐ฆ)๐ (๐ฆ) d๐ฆ,
๐ฅ โ (0, โ).
A formula by Widom says that ๐๐ (๐๐) โ ๐๐ (๐)๐๐ (๐) = ๐๐ ๐ป(๐)๐ป(ห ๐)๐๐ + ๐
๐ ๐ป(ห๐)๐ป(๐)๐
๐ , 2
2
(8)
where ๐๐ is as in Section 1 and ๐
๐ : ๐ฟ (0, โ) โ ๐ฟ (0, ๐ ) is the operator that is given by (๐
๐ ๐ )(๐ฅ) := ๐ (๐ โ ๐ฅ) for 0 < ๐ฅ < ๐ and (๐
๐ ๐ )(๐ฅ) := 0 for ๐ฅ > ๐ ; see, for example, [10, Section 9.7(d)]. Since โฅ๐ต๐ถโฅ1 โค โฅ๐ตโฅ2 โฅ๐ถโฅ2 , it su๏ฌces to prove that โฅ๐๐ ๐ป(๐)โฅ22 โ 0 as ๐ โ โ ๐ for ๐ = โห โ ๐ฟโ (โ) โฉ ๐ฟ1 (โ). We have โซ โซ โซ โซ โฅ๐๐ ๐ป(๐)โฅ22 1 ๐ โ 1 ๐ โ = โฃโ(๐ฅ โ ๐ฆ)โฃ2 d๐ฆ d๐ฅ = โฃโ(๐ก)โฃ2 d๐ก d๐ฅ ๐ ๐ 0 0 ๐ 0 ๐ฅ โซ โซ โซ โซ 1 ๐ ๐ 1 ๐ โ = โฃโ(๐ก)โฃ2 d๐ก d๐ฅ + โฃโ(๐ก)โฃ2 d๐ก d๐ฅ, ๐ 0 ๐ฅ ๐ 0 ๐ and the second term on the right is โซ โ โฃโ(๐ก)โฃ2 d๐ก = ๐(1). ๐
The ๏ฌrst term equals 1 ๐
โซ 0
๐
โซ
๐ก 0
1 โฃโ(๐ก)โฃ d๐ฅ d๐ก = ๐ 2
โซ
๐ 0
๐กโฃโ(๐ก)โฃ2 d๐ก,
Iterates of the FoxโLi Operator and we write this as 1 ๐
โซ
๐0
0
1 ๐กโฃโ(๐ก)โฃ d๐ก + ๐ 2
โซ
๐ ๐0
213
๐กโฃโ(๐ก)โฃ2 d๐ก
(9)
where ๐0 = ๐0 (๐) is chosen so that โซ โซ ๐ โซ โ 1 ๐ ๐ ๐กโฃโ(๐ก)โฃ2 d๐ก โค โฃโ(๐ก)โฃ2 d๐ก โค โฃโ(๐ก)โฃ2 d๐ก < . ๐ ๐0 2 ๐0 ๐0 Since then 1 ๐
โซ 0
๐0
๐กโฃโ(๐ก)โฃ2 d๐ก โค
๐0 ๐
โซ
๐0 0
โฃโ(๐ก)โฃ2 d๐ก โค
๐0 ๐
โซ
โ 0
โฃโ(๐ก)โฃ2 d๐ก <
๐ 2
if only ๐ is large enough, we see that (9) is smaller than any prescribed ๐ > 0 whenever ๐ is su๏ฌciently large. โก Lemma 2.2. If ๐ โ ๐ฟโ (โ) โฉ ๐ฟ1 (โ) and ๐ โฅ 1 is a natural number, then โฅ๐๐๐ (๐) โ ๐๐ (๐๐ )โฅ1 = ๐(๐ )
as
๐ โ โ.
Proof. This is trivial for ๐ = 1. Assume that the assertion is true for some ๐ โฅ 1. We write ๐๐๐+1 (๐) โ ๐๐ (๐๐+1 ) as ( ) ๐๐๐ (๐) โ ๐๐ (๐๐ ) ๐๐ (๐) + ๐๐ (๐๐ )๐๐ (๐) โ ๐๐ (๐๐+1 ) and have 1( 1 ) 1 1 1 ๐๐๐ (๐) โ ๐๐ (๐๐ ) ๐๐ (๐)1 โค โฅ๐๐๐ (๐) โ ๐๐ (๐๐ )โฅ1 โฅ๐๐ (๐)โฅโ . 1
Clearly, โฅ๐๐ (๐)โฅโ โค โฅ๐โฅโ. Furthermore, โฅ๐๐๐ (๐) โ ๐๐ (๐๐ )โฅ1 = ๐(๐ ) by assumption, and โฅ๐๐ (๐๐ )๐๐ (๐) โ ๐๐ (๐๐+1 )โฅ1 = ๐(๐ ) due to Lemma 2.1. Thus, the assertion is valid for ๐ + 1. โก As โฃtr ๐ดโฃ โค โฅ๐ดโฅ1 for every trace class operator ๐ด, Theorem 1.4 is an obvious consequence of Lemma 2.2. The following result proves part of Theorem 1.3. We denote the spectrum of an operator ๐ด by sp ๐ด. The essential spectrum spess ๐ด is the set of all ๐ โ โ for which ๐ด โ ๐๐ผ is not Fredholm, that is, not invertible modulo compact operators. Clearly, spess ๐ด โ sp ๐ด. Proposition 2.3. If ๐(๐) := eโ๐๐ sp ๐ถ(๐)) = ๐,
2
/4
then
spess ๐ (๐) = sp ๐ (๐) = ๐ป,
sp ๐๐ (๐) โ ๐ป.
Proof. Throughout this proof, ๐ denotes an arbitrary function in ๐ฟโ (โ). The spectrum of ๐ถ(๐) is the essential range โ(๐) of ๐. Hence sp ๐ถ(๐) = ๐. To prove the assertion for the spectra of ๐ (๐), we have recourse to known results on Toeplitz operators. The passage from WienerโHopf operators on ๐ฟ2 (0, โ) to Toeplitz operators on the Hardy space ๐ป 2 (๐) and back can be performed by a standard unitary similarity; see, for example, Section 9.5(e) of [10]. The HartmanโWintner
214
A. Bยจ ottcher, S. Grudsky, D. Huybrechs and A. Iserles
and BrownโHalmos theorems, which can be found, for instance, as Theorems 2.30 and 2.33 in [10], yield the spectral inclusions โ(๐) โ sp ๐ (๐) โ conv โ(๐), where conv denotes the convex hull. Consequently, ๐ โ sp ๐ (๐) โ ๐ป. To show that spess ๐ (๐) is all of ๐ป, ๏ฌx some ๐ โ ๐ป. We have ๐(๐) โ ๐ = โฃ๐(๐) โ ๐โฃeโi๐(๐) with a function ๐ that can be written as ๐ = ๐ + ๐ฟ where ๐ โ ๐ถ(โ) โฉ ๐ฟโ (โ) and ๐ฟ โ ๐ถ(โ) is monotone on (โโ, 0 and (0, โ) with ๐ฟ(ยฑโ) = +โ. Now we can employ a result of [5], which is also cited and proved as Theorem 6.4 of [8] and, reduced to a necessary invertibility criterion, went as Proposition 2.26(d) into [10]. This result says that if ๐ โ ๐ has an argument as just described, then for ๐ (๐ โ ๐) to be Fredholm it is necessary that โฃ๐(๐)โฃ = ๐(log โฃ๐โฃ) as โฃ๐โฃ โ โ. Because in our case โฃ๐(๐)โฃ increases as โฃ๐โฃ2 , it follows that ๐ (๐) โ ๐๐ผ = ๐ (๐ โ ๐) cannot be Fredholm. Thus, ๐ โ spess ๐ (๐). Finally, in [7, Theorem 1.1], it is shown that โฅ๐๐ (๐)โฅโ < 1. We therefore arrive at the conclusion that sp ๐๐ (๐) โ ๐ป. โก As ๐๐ (๐) is not a normal operator, one could ask whether we should rather study the ๐-pseudospectrum sp๐ ๐๐ (๐) := {๐ โ โ : 1/๐ โค โฅ(๐๐ (๐) โ ๐๐ผ)โ1 โฅโ โค โ} than the spectrum sp ๐๐ (๐). See [18]. It is known that, for each ๐ > 0, the sets sp๐ ๐๐ (๐) converge to sp๐ ๐ (๐) as ๐ โ โ in the Hausdor๏ฌ metric if ๐ is piece2 wise continuous [6]. The symbol ๐(๐) = eโi๐ /4 is not piecewise continuous, but fortunately things are simple. Here is the result. Proposition 2.4. Given ๐ > 0, there is a ๐0 = ๐0 (๐) such that ๐ป โ sp๐ ๐๐ (๐) for all ๐ > ๐0 . Proof. Pick ๐ โ ๐ป. The operator ๐๐ (๐ โ ๐) and its adjoint ๐๐ (๐ โ ๐) converge strongly to ๐ (๐ โ ๐) and this operatorโs adjoint ๐ (๐ โ ๐). Thus, were the norms โฅ๐๐๐ (๐ โ ๐)โ1 โฅโ uniformly bounded for some sequence ๐๐ โ โ, ๐ (๐ โ ๐) would be invertible. As the latter is not the case due to Proposition 2.3, we conclude that โฅ๐๐ (๐ โ ๐)โ1 โฅโ โ โ for each ๐ โ ๐ป. This together with the compactness of ๐ป implies that for every ๐ > 0 there is a ๐0 (๐) such that โฅ๐๐ (๐ โ ๐)โ1 โฅโ โฅ 1/๐ for all ๐ > ๐0 (๐) and all ๐ โ ๐ป. โก Proposition 2.4 is equivalent to saying that given ๐ > 0 and ๐ โ ๐ป, there exists a number ๐0 = ๐0 (๐) such that for every ๐ > ๐0 we can ๏ฌnd ๐ โ ๐ฟ2 (0, ๐ ) satisfying โฅ๐ โฅ = 1 and โฅ๐๐ (๐)๐ โ ๐๐ โฅ โค ๐. This is in the spirit of Landauโs result [15]. He took ๐ from ๐ only but showed much more, namely that for ๐ > ๐0 there are at least 1000๐ orthonormal functions ๐ in ๐ฟ2 (0, ๐ ) such that โฅ๐๐ (๐)๐ โ ๐๐ โฅ โค ๐.
Iterates of the FoxโLi Operator
215
3. An oscillatory multivariate integral In this section we prove Theorem 1.1. Lemma 3.1. For every natural number ๐ โฅ 1, the operators ๐น๐๐ as well as the 2 operators ๐๐๐ (๐) and ๐๐ (๐ ๐ ) generated by ๐(๐) := eโi๐ /4 , are of trace class. Proof. Since ๐น๐ is unitarily similar to ๐2โ๐ (๐), it su๏ฌces to prove that ๐๐ (๐ ๐ ) is in the trace class. The operator ๐๐ (๐ ๐ ) acts by the rule โซ ๐ (๐๐ (๐ ๐ )๐ )(๐ฅ) = โ๐ (๐ฅ โ ๐ฆ)๐ (๐ฆ) d๐ฆ, ๐ฅ โ (0, ๐ ), (10) 0
where โ๐ (๐ก) =
1 2๐
โซ
โ
โโ
๐ ๐ (๐)eโi๐๐ก d๐ =
โซ
1 2๐
โ
โโ
eโi๐๐
2
/4โi๐๐ก
d๐ = โ
2 1 ei๐ก /๐ . ๐i๐
๐
Thus, ๐๐ (๐ ) is an integral operator over a ๏ฌnite interval with a smooth kernel. From [14, III.10.3] we therefore deduce that ๐๐ (๐ ๐ ) โ ๐1 . An alternative proof is as follows. Let โ๐,๐ be a ๐ถ 2 function on โ which coincides with โ๐ on (โ๐, ๐) and is identically zero outside (โ2๐, 2๐ ). As (10) does not depend on the values of โ๐ outside (โ๐, ๐ ), we have ๐๐ (๐ ๐ ) = ๐๐ (โห๐,๐ ). The function โห๐,๐ is in ๐ฟโ (โ) because โ๐,๐ โ ๐ฟ1 (โ), and twice integrating the integral โซ 2๐ โห๐,๐ (๐) = โ๐,๐ (๐ก)ei๐๐ก d๐ก โ2๐
by parts, we obtain โห๐,๐ (๐) =
1 (i๐)2
โซ
2๐ โ2๐
โโฒโฒ๐,๐ (๐ก)ei๐๐ก d๐ก
for ๐ โ= 0, which shows that โห๐,๐ โ ๐ฟ1 (โ). In the beginning of Section 2 we noticed that symbols in ๐ฟโ (โ) โฉ ๐ฟ1 (โ) generate truncated WienerโHopf operators in the trace class. Hence ๐๐ (โห๐,๐ ) โ ๐1 . โก We have (๐น๐๐ ๐ )(๐ฅ) = where
โซ
1
โ1
๐๐ (๐ฅ, ๐ฆ)๐ (๐ฆ) d๐ฆ,
๐ฅ โ (โ1, 1),
โ
(โ )2 โซ 1 2 2 ๐ i๐(๐ฅโ๐ฆ)2 ๐ e , ๐2 (๐ฅ, ๐ฆ) = ei๐(๐ฅโ๐ง) ei๐(๐งโ๐ฆ) d๐ง, ๐i ๐i โ1 (โ )3 โซ 1 โซ 1 2 2 2 ๐ ๐3 (๐ฅ, ๐ฆ) = ei๐(๐ฅโ๐ง) ei๐(๐งโ๐ค) ei๐(๐คโ๐ฆ) d๐ง d๐ค, ๐i โ1 โ1
๐1 (๐ฅ, ๐ฆ) =
and so on. Since ๐น๐๐ is of trace class by Lemma 3.1 and ๐๐ is continuous on [โ1, 1]2 , it follows that โซ 1 ๐ tr ๐น๐ = ๐๐ (๐ฅ, ๐ฅ) d๐ฅ; โ1
216
A. Bยจ ottcher, S. Grudsky, D. Huybrechs and A. Iserles
see [14, Corollary III.10.2]. Consequently, (โ )๐ ๐ tr ๐น๐๐ = ๐ผ๐ ๐i where
โซ ๐ผ๐ :=
1
โ1
โซ ...
1
โ1
โ exp โi๐
๐ โ
(11) โ
(๐ฅ๐ โ ๐ฅ๐+1 )2 โ d๐ฅ1 . . . d๐ฅ๐
(12)
๐=1
with ๐ฅ๐+1 := ๐ฅ1 . By virtue of some lucky circumstances, it is not di๏ฌcult to compute ๐ผ๐ straightforwardly for ๐ โค 4. Trivially, ๐ผ1 = 2. Letting โซ ๐ง 2 2 erf(๐ง) := โ eโ๐ d๐, ๐ง โ โ, ๐ 0 one almost immediately gets (โ ) โ 2 ๐i 8๐ i e8i๐ i ๐ผ2 = โ erf + โ 2 i 2 ๐ 2๐ ๐ (โ )2 โ โ ( ) ( ) 2 ๐i i 1 2 ๐i 1 ๐i 1 โ =โ โ +๐ = โ + ๐ , 2 ๐ 2๐ ๐ ๐ 2๐ ๐ ๐2 2 2 while with a little more labour, one obtains (โ )] โซ โ๐ (โ ) [ (โ ) 1 ๐i 6 2 2 โ ๐ผ3 = 3/2 โ erf ๐ฆ erf ๐ฆ + erf (2 ๐ โ ๐ฆ) d๐ฆ i i i ๐ 3 0 (โ )2 (โ )3 ( ) 2 ๐i 1 ๐i 1 =โ โ โ +๐ ๐ ๐ ๐ 3/2 3 ๐ 2 and
(โ )]2 โซ โ ๐ (โ ) [ (โ ) 1 (๐i)3/2 4 2 2 โ ๐ผ4 = 2 erf ๐ฆ erf ๐ฆ + erf (2 ๐ โ ๐ฆ) d๐ฆ ๐ 4 i i i 0 (โ )3 ( ) 2 ๐i 1 โ = +๐ . 2 ๐ ๐ 4
However, to tackle the general case we have to proceed di๏ฌerently. Theorem 3.2. As ๐ โ โ, 2 ๐ผ๐ = โ ๐
(โ
๐i ๐
)๐โ1 (1 + ๐(1)).
Proof. To establish the pattern for general ๐, we ๏ฌrst consider the case ๐ = 3. The oscillator function in (12) is ๐(๐ฅ1 , ๐ฅ2 , ๐ฅ3 ) := (๐ฅ1 โ ๐ฅ2 )2 + (๐ฅ2 โ ๐ฅ3 )2 + (๐ฅ3 โ ๐ฅ1 )2 ,
Iterates of the FoxโLi Operator
217
and its stationary points are on the straight line ๐ฅ1 = ๐ฅ2 = ๐ฅ3 . We make the change of variables ๐ก = ๐ฅ1 โ ๐ฅ2 ,
๐ข = ๐ฅ2 โ ๐ฅ3 ,
๐ฃ = ๐ฅ1 + ๐ฅ2 + ๐ฅ3
in (12). The determinant of the Jacobian is 1/3, hence โซ 1 ๐ผ3 = exp[i๐(๐ก2 + ๐ข2 + (๐ก + ๐ข)2 )] d๐ก d๐ข d๐ฃ 3 ฮ where ฮ is some polytope containing the origin in its interior. The new oscillator function โ(๐ก, ๐ข, ๐ฃ) = ๐ก2 + ๐ข2 + (๐ก + ๐ข)2 is independent of ๐ฃ, and as a function of ๐ก and ๐ข only, it has the single stationary point ๐ก = ๐ข = 0. The Hessian for โ, again thought of as a function of solely ๐ก and ๐ข, is ( ) 4 2 . 2 4 This is a positive de๏ฌnite matrix, and therefore โ can be written as ๐2 + ๐ 2 in suitable coordinates ๐ and ๐ . To ๏ฌnd the new coordinates, we try the ansatz ๐ = ๐๐ก + ๐๐ข, The equation is satis๏ฌed for
๐ = ๐๐ข.
(13)
(๐๐ก + ๐๐ข)2 + (๐๐ข)2 = ๐ก2 + ๐ข2 + (๐ก + ๐ข)2
โ 1 3 ๐ = 2, ๐ = โ , ๐ = . 2 2 The Jacobi โdeterminant of the substitution (13) with these coe๏ฌcients equals 1/(๐๐) = 1/ 3. Consequently, โซ 1 1 ๐ผ3 = โ exp[i๐(๐2 + ๐ 2 )] d๐ฃ d๐ d๐ 3 3 ฮฉ โ
where ฮฉ is again a polytope with the origin in its interior. Integrating over ๐ฃ we get ) โซ (โซ ๐ฃ2 (๐,๐ ) 1 1 ๐ผ3 = โ exp[i๐(๐2 + ๐ 2 )] d๐ฃ d๐ d๐ 3 3 ฮฉ1 ๐ฃ1 (๐,๐ ) โซ 1 1 = โ ๐ (๐, ๐ ) exp[i๐(๐2 + ๐ 2 )] d๐ d๐ 3 3 ฮฉ1 with ๐ (๐, ๐ ) := ๐ฃ2 (๐, ๐ ) โ ๐ฃ1 (๐, ๐ ) and some (planar) polytope ฮฉ1 with the origin in its interior. For ๐ = ๐ = 0, the variable ๐ฃ ranges from โ3 to 3. Hence ๐ (0, 0) = 6. The stationary phase formula โ โซ ๐ฝ ๐i i๐๐ฅ2 ๐ (๐ฅ)e d๐ฅ = ๐ (0) (1 + ๐(1)) ๐ โ๐ผ
218
A. Bยจ ottcher, S. Grudsky, D. Huybrechs and A. Iserles
can now be applied independently for ๐ and ๐ . The outcome is (โ )2 (โ )2 1 1 ๐i 2 ๐i ๐ผ3 = โ 6 (1 + ๐(1)) = โ (1 + ๐(1)). 3 3 ๐ ๐ 3 The pattern in the general case is now obvious. Substituting ๐ก๐ = ๐ฅ๐ โ ๐ฅ๐+1 in (12), we get ๐ผ๐ =
1 ๐
(1 โค ๐ โค ๐ โ 1),
๐ก๐ = ๐ฅ1 + ๐ฅ2 + โ
โ
โ
+ ๐ฅ๐
โซ ฮ
exp[i๐โ(๐ก1 , . . . , ๐ก๐โ1 )] d๐ก1 . . . d๐ก๐
with โ(๐ก1 , . . . , ๐ก๐โ1 ) =
๐โ1 โ ๐=1
โ
๐โ1 โ
๐ก2๐ + โ
โ2 ๐ก๐ โ .
๐=1
The Hessian of this function is the (๐ โ 1) ร (๐ โ 1) matrix โ โ 4 2 2 ... โ2 4 2 . . .โ โ. ๐ป := โ โ2 2 4 . . .โ ... ... ... ... The determinant of the ๐ ร ๐ matrix constituted by the ๏ฌrst ๐ rows and columns of ๐ป is 2๐ (๐ + 1). Thus, by Sylvesterโs theorem, ๐ป is positive de๏ฌnite. We look for a change of variables โ โ โ โโ โ ๐ 1 ๐11 ๐12 . . . ๐1,๐โ1 ๐ก1 โ ๐ 2 โ โ 0 ๐22 . . . โ โ ๐2,๐โ1 โ โ โ โ โ โ ๐ก2 โ โ . . . โ = โ. . . . . . . . . โ โ ... ... โ ๐ ๐โ1 0 0 . . . ๐๐โ1,๐โ1 ๐ก๐โ1 such that ๐ 21 + ๐ 22 + โ
โ
โ
+ ๐ 2๐โ1 = โ(๐ก1 , ๐ก2 , . . . , ๐ก๐โ1 ). It is easily seen that such a change of variables can be found with โ โ โ 3 ๐ ๐11 = 2, ๐22 = , . . . , ๐๐โ1,๐โ1 = . 2 ๐โ1 โ The Jacobi determinant equals 1/(๐11 ๐22 . . . ๐๐โ1,๐โ1 ) = 1/ ๐. We so arrive at the representation โซ 1 1 ๐ผ๐ = โ exp[i๐(๐ 21 + โ
โ
โ
+ ๐ 2๐โ1 )] d๐ก๐ d๐ 1 . . . d๐ ๐โ1 ๐ ๐ ฮฉ โซ 1 1 = โ ๐ (๐ 1 , . . . , ๐ ๐โ1 ) exp[i๐(๐ 21 + โ
โ
โ
+ ๐ 2๐โ1 )] d๐ 1 . . . d๐ ๐โ1 . ๐ ๐ ฮฉ1
Iterates of the FoxโLi Operator
219
As ๐ (0, . . . , 0) = 2๐, the usual stationary phase formula argument yields (โ )๐โ1 (โ )๐โ1 1 1 ๐i 2 ๐i ๐ผ๐ = โ 2๐ (1 + ๐(1)) = โ (1 + ๐(1)), ๐ ๐ ๐ ๐ ๐ as desired.
โก
Lemma 2.1 in conjunction with (11) and Theorem 3.2 proves Theorem 1.1. As already said, (5) is equivalent to (1). Thus, also the proof of Theorem 1.3 is at this point complete.
4. The logarithmic spiral ansatz We now prove Theorem 1.2. Letting ๐๐ be as in that theorem, we have โซ โ โซ โ ๐ ๐ ๐๐ (๐ฅ) d๐ฅ = eโ[๐ผ(๐)+i๐ฝ(๐)]๐๐ฅ d๐ฅ, 0
0
and hence (3) is true for some ๐ if and only if โ โซ โ ๐ โ[๐ผ(๐)+i๐ฝ(๐)]๐๐ฅ๐ โ1/2 e d๐ฅ = 2๐ (1 + ๐(1)), ๐i 0 or equivalently, after substituting ๐๐ฅ๐ โ ๐ฅ๐ , โ โซ โ ๐ โ1/๐ โ[๐ผ(๐)+i๐ฝ(๐)]๐ฅ๐ โ1/2 ๐ e d๐ฅ = 2๐ (1 + ๐(1)). ๐i 0
(14)
(15)
Taking (14) for ๐ = 1, we obtain โ โซ โ ๐ ๐ eโ[๐ผ(๐)+i๐ฝ(๐)]๐ฅ d๐ฅ = 2 (1 + ๐(1)), ๐i 0 whereas (15) for ๐ = 2 states that โ โซ โ ๐ ๐ eโ[๐ผ(๐)+i๐ฝ(๐)]๐ฅ d๐ฅ = 21+1/๐โ1/2 (1 + ๐(1)). ๐i 0 Comparing the last two formulas, we arrive at the conclusion that if (14) holds for ๐ = 1 and ๐ = 2, then necessarily ๐ = 2. Now consider (14) with ๐ = 2. Computing the integral, we obtain that (14) is equivalent to the statement that โ โ 1 ๐ 1 2 ๐ โ =โ (1 + ๐(1)), 2 ๐ผ(๐) + i๐ฝ(๐) ๐ ๐ ๐i which holds if and only if ๐ผ(๐) + i๐ฝ(๐) =
๐2 i (1 + ๐(1)). 16๐
(16)
220
A. Bยจ ottcher, S. Grudsky, D. Huybrechs and A. Iserles
Writing ๐(1) = ๐(1) + i๐(1) with two real ๐(1) on the right, we arrive at the conclusion that (16) is valid if and only if ( ) ( ) 1 ๐2 1 ๐ผ(๐) = ๐ , ๐ฝ(๐) = +๐ . ๐ 16๐ ๐ This completes the proof of Theorem 1.2.
5. Symbols with a FoxโLi discontinuity This section is devoted to the proof of Theorem 1.5. Lemma 5.1. (a) Let ๐ด๐ and ๐ต๐ be operators on ๐ฟ2 (0, ๐ ). If โฅ๐ด๐ โฅ1 = ๐(๐ ) and โฅ๐ต๐ โฅโ = ๐(1), then โฅ๐ด๐ ๐ต๐ โฅ1 = ๐(๐ ) and โฅ๐ต๐ ๐ด๐ โฅ1 = ๐(๐ ). (b) If ๐, ๐ โ ๐ฟโ (โ) and ๐ป(๐), ๐ป(ห๐) โ ๐1 , then โฅ๐๐ (๐)๐๐ (๐) โ ๐๐ (๐๐)โฅ1 = ๐(๐ ),
โฅ๐๐ (๐)๐๐ (๐) โ ๐๐ (๐๐)โฅ1 = ๐(๐ ).
(c) If ๐ โ ๐ฟโ (โ) and ๐ป(๐) โ ๐1 , then ๐ป(๐๐ ) โ ๐1 for every natural number ๐ โฅ 1. Proof. Part (a) follows from the inequalities โฅ๐ด๐ ๐ต๐ โฅ1 โค โฅ๐ด๐ โฅ1 โฅ๐ต๐ โฅโ ,
โฅ๐ต๐ ๐ด๐ โฅ1 โค โฅ๐ต๐ โฅโ โฅ๐ด๐ โฅ1 .
To prove (b) note that, by (8), ห ๐ โ ๐
๐ ๐ป(ห๐)๐ป(๐)๐
๐ ๐๐ (๐)๐๐ (๐) โ ๐๐ (๐๐) = โ๐๐ ๐ป(๐)๐ป(๐)๐ and that ห ๐ โฅ1 โค โฅ๐๐ โฅโ โฅ๐ป(๐)โฅ1 โฅ๐ป(๐)๐ ห ๐ โฅโ = ๐(1) = ๐(๐ ), โฅ๐๐ ๐ป(๐)๐ป(๐)๐ โฅ๐
๐ ๐ป(ห๐)๐ป(๐)๐
๐ โฅ1 โค โฅ๐
๐ โฅโ โฅ๐ป(ห๐)โฅ1 โฅ๐ป(๐)๐
๐ โฅโ = ๐(1) = ๐(๐ ). Finally, part (c) is obviously true for ๐ = 1. So suppose that ๐ป(๐๐ ) โ ๐1 for some ๐ โฅ 1. The identity ๐ป(๐๐+1 ) = ๐ป(๐๐ )๐ (ห๐) + ๐ (๐๐ )๐ป(๐), which is the continuous analogue of formula (2.19) in [10], shows that then ๐ป(๐๐+1 ) is also in ๐1 . โก 2
Proposition 5.2. Let ๐(๐) = ๐(๐)๐(๐) where ๐(๐) := eโi๐ /4 and ๐ is a function in ห such that ๐ป((๐ โ ๐(โ))๐ ๐ ) โ ๐1 and ๐ป((ห ๐ถ(โ) ๐ โ ๐(โ))๐ ๐ ) โ ๐1 for every integer ๐ ๐ ๐ โฅ 0. Then the operator ๐๐ (๐) and ๐๐ (๐ ) are of trace class for every natural number ๐ โฅ 1 and โซ โ ๐ ๐ ๐ tr ๐๐ (๐) = tr ๐๐ (๐ ) + ๐(๐ ) = ๐๐ (๐) d๐ + ๐(๐ ). (17) 2๐ โโ
Iterates of the FoxโLi Operator
221
Proof. Again by Widomโs formula (8), ๐๐ )๐ป(๐ ๐ )๐
๐ ; ๐๐ (๐๐ ) = ๐๐ (๐๐ )๐๐ (๐๐ ) + ๐๐ ๐ป(๐๐ )๐ป(๐ ๐ )๐๐ + ๐
๐ ๐ป(ห notice that ๐ ห(๐) := ๐(โ๐) = ๐(๐). Lemma 3.1 tells us that ๐๐ (๐๐ ) is in ๐1 . Let ๐ข := ๐ โ ๐(โ). The Hankel operator induced by a constant function is the zero operator. Hence ๐ป(๐) = ๐ป(๐ข) and ๐ป(ห ๐) = ๐ป(ห ๐ข). By our assumption, ๐ป(๐ข) and ๐ป(ห ๐ข) are in ๐1 . From Lemma 5.1(c) we therefore deduce that the operators ๐ป(๐๐ ) and ๐ป(ห ๐๐ ) are also in ๐1 . This shows that ๐๐ (๐๐ ) โ ๐1 and thus also that ๐ ๐๐ (๐) = [๐๐ (๐1 )]๐ โ ๐1 . In what follows we write ๐ด๐ โก ๐ต๐ if โฅ๐ด๐ โ ๐ต๐ โฅ1 = ๐(๐ ). Recall that ๐ข(๐) is de๏ฌned as ๐(๐) โ ๐(โ). Thus, ๐ = ๐ข๐ + ๐(โ)๐. We claim that for each natural number ๐ โฅ 1 it is true that ๐๐๐ (๐) โก ๐๐ [(๐ข๐ + ๐(โ)๐)๐ โ ๐(โ)๐ ๐ ๐ ] + ๐(โ)๐ ๐๐๐ (๐).
(18)
This is trivial for ๐ = 1. So assume the claim is true for some ๐ โฅ 1. Then [ ] ๐๐๐+1 (๐) โก ๐๐ (๐ข๐) + ๐(โ)๐๐ (๐) ร โก โค ๐ ( ) โ ๐ ๐(โ)๐โ๐ ๐๐ ((๐ข๐)๐ ๐ ๐โ๐ ) + ๐(โ)๐ ๐๐๐ (๐)โฆ . รโฃ ๐ ๐=1 ๐ข๐ ๐ ) are in ๐1 for all natural numbers ๐ โฅ 1 by our The operators ๐ป(๐ข๐ ๐ ) and ๐ป(ห assumption. From Lemma 5.1(b) we therefore obtain that ๐๐ (๐ข๐)๐๐ ((๐ข๐)๐ ๐ ๐โ๐ ) โก ๐๐ ((๐ข๐)(๐ข๐)๐ ๐ ๐โ๐ ) and using parts (a) and (b) of Lemma 5.1 we get ๐๐ (๐)๐๐ ((๐ข๐)๐ ๐ ๐โ๐ ) โก ๐๐ (๐)๐๐ (๐ข๐)๐๐ ((๐ข๐)๐โ1 ๐ ๐โ๐ ) โก ๐๐ (๐ข๐ 2 )๐๐ ((๐ข๐)๐โ1 ๐ ๐โ๐ ) โก ๐๐ (๐(๐ข๐)๐ ๐ ๐โ๐ ) and ๐๐ (๐ข๐)๐๐๐ (๐) โก ๐๐ (๐ข๐ 2 )๐๐๐โ1 (๐) โก ๐๐ (๐ข๐ 3 )๐๐๐โ2 (๐) โก โ
โ
โ
โก ๐๐ (๐ข๐ ๐+1 ). Consequently,
โก
โค ๐ ( ) โ ๐ ๐๐๐+1 (๐) โก ๐๐ โฃ(๐ข๐ + ๐(โ)๐) ๐(โ)๐โ๐ (๐ข๐)๐ ๐ ๐โ๐ โฆ ๐ ๐=1 [ ] + ๐๐ ๐(โ)๐ ๐ข๐ ๐+1 + ๐(โ)๐+1 ๐๐๐+1 (๐),
and the sum of the symbols in the brackets on the right is (๐ข๐ + ๐(โ)๐)[(๐ข๐ + ๐(โ)๐)๐ โ ๐(โ)๐ ๐ ๐ ] + ๐(โ)๐ ๐ข๐ ๐+1 = (๐ข๐ + ๐(โ)๐)๐+1 โ ๐(โ)๐+1 ๐ ๐+1 . This proves our claim (18) for ๐ + 1.
222
A. Bยจ ottcher, S. Grudsky, D. Huybrechs and A. Iserles
If ๐ด๐ โก ๐ต๐ , then tr ๐ด๐ = tr ๐ต๐ + ๐(๐ ). Since ๐ข๐ + ๐(โ)๐ = ๐, the trace of the ๏ฌrst term on the right of (18) equals โซ โ( ) 1 ๐๐ (๐) โ ๐(โ)๐ ๐ ๐ (๐) d๐, 2๐ โโ and from Theorem 1.3 we know that the trace of the second term on the right of (18) is โซ ๐(โ)๐ โ ๐ ๐ ๐ ๐ ๐ tr (๐(โ) ๐๐ (๐)) = ๐(โ) tr ๐๐ (๐ ) + ๐(๐ ) = ๐ (๐) d๐ + ๐(๐ ). 2๐ โโ Adding the two results we arrive at (17).
โก
The hypothesis of Proposition 5.2 stipulates that the Hankel operators ๐ป(๐ข๐ ๐ ) and ๐ป(ห ๐ข๐ ๐ ) are in ๐1 for every integer ๐ โฅ 0. Peller showed that the two Hankel operators ๐ป(๐) and ๐ป(ห๐) are of trace class if and only if ๐ is in the Besov space ๐ต11 (โ); see [17, p. 277]. Here is a simple su๏ฌcient condition for ๐ป(๐) and ๐ป(ห๐) to be in the trace class. Lemma 5.3. If ๐ โ ๐ถ 3 (โ) and the functions ๐ 2 ๐(๐), ๐ 2 ๐โฒ (๐), ๐ 2 ๐โฒโฒ (๐), ๐โฒโฒโฒ (๐) belong to ๐ฟ1 (โ) and have zero limits as ๐ โ ยฑโ, then ๐ป(๐) and ๐ป(ห๐) are trace class operators. Proof. Let โ(๐ก) =
1 2๐
โซ
โ
โโ
๐(๐)eโi๐๐ก d๐,
๐ก โ โ.
Since ๐๐(๐) and ๐ 2 ๐(๐) are in ๐ฟ1 (โ), we may twice di๏ฌerentiate the integral to see that โ is in ๐ถ 2 (โ) and โซ โ 1 โฒโฒ โ (๐ก) = (โi๐)2 ๐(๐)๐โi๐๐ก d๐. 2๐ โโ Using that (๐ 2 ๐(๐))โฒ = 2๐๐(๐) + ๐ 2 ๐โฒ (๐) and (๐ 2 ๐(๐))โฒโฒ = 2๐(๐) + 4๐๐โฒ (๐) + ๐ 2 ๐โฒโฒ (๐) are also in ๐ฟ1 (โ) and have zero limits at in๏ฌnity, we may twice integrate by parts to obtain that โซ 1 (โi)2 โ 2 โฒโฒ โ (๐ก) = (๐ ๐(๐))โฒโฒ eโi๐๐ก d๐, 2๐ (i๐ก)2 โโ which shows that โซ โ โโ
โฃ๐กโฃ โฃโโฒโฒ (๐ก)โฃ2 d๐ก < โ.
(19)
As ๐โฒ , ๐โฒโฒ , ๐โฒโฒโฒ are in ๐ฟ1 and have zero limits at in๏ฌnity, we have โซ โ 1 1 โ(๐ก) = ๐โฒโฒโฒ (๐)eโi๐๐ก d๐ 2๐ (i๐ก)3 โโ and hence
โซ
โ
โโ
โฃ๐กโฃ4 โฃโ(๐ก)โฃ2 d๐ก < โ.
(20)
Iterates of the FoxโLi Operator
223
Basor and Widom [1, p. 398] showed that ๐ป(๐) and ๐ป(ห๐) are of trace class if (19) and (20) hold. โก Corollary 5.4. If ๐ is as in Theorem 1.5, then the Hankel operators ๐ป(๐ข๐ ๐ ) and ๐ป(ห ๐ข๐๐ ) are in ๐1 for every real number ๐. Proof. The function ๐ := ๐ข๐ ๐ satis๏ฌes the hypothesis of Lemma 5.3.
โก
Combining Corollary 5.4 and Proposition 5.2, we arrive at Theorem 1.5.
References [1] E. Basor and H. Widom, Toeplitz and WienerโHopf determinants with piecewise continuous symbols. J. Funct. Analysis 50 (1983), 387โ413. [2] M. Berry, Fractal modes of unstable lasers with polygonal and circular mirrors. Optics Comm. 200 (2001), 321โ330. [3] M. Berry, Mode degeneracies and the Petermann excess-noise factor for unstable lasers. J. Modern Optics 50 (2003), 63โ81. [4] M. Berry, C. Storm, and W. van Saarlos, Theory of unstable laser modes: edge waves and fractality. Optics Comm. 197 (2001), 393โ402. [5] A. Bยจ ottcher, On Toeplitz operators generated by symbols with three essential cluster points. Preprint P-Math-04/86, Karl-Weierstrass-Institut, Berlin 1986. [6] A. Bยจ ottcher, Pseudospectra and singular values of large convolution operators. J. Integral Equations Appl. 6 (1994), 267โ301. [7] A. Bยจ ottcher, H. Brunner, A. Iserles, and S. Nรธrsett, On the singular values and eigenvalues of the FoxโLi and related operators. New York J. Math. 16 (2010), 539โ 561. [8] A. Bยจ ottcher and S. Grudsky, Toeplitz operators with discontinuous symbols: phenomena beyond piecewise continuity. Operator Theory: Adv. Appl. 90 (1996), 55โ118. [9] A. Bยจ ottcher and B. Silbermann, Introduction to Large Truncated Toeplitz Matrices. Springer-Verlag, New York, 1999. [10] A. Bยจ ottcher and B. Silbermann, Analysis of Toeplitz Operators. 2nd edition, Springer-Verlag, Berlin, Heidelberg, New York, 2006. [11] H. Brunner, A. Iserles, and S.P. Nรธrsett, The computation of the spectra of highly oscillatory Fredholm integral operators. J. Integral Equations Appl. To appear. [12] J.A. Cochran and E.W. Hinds, Eigensystems associated with the complex-symmetric kernels of laser theory. SIAM J. Appl. Math. 26 (1974), 776โ786. [13] A.G. Fox and T. Li, Resonance modes in a maser interferometer. Bell Systems Tech. J. 40 (1961), 453โ488. [14] I. Gohberg and M.G. Krein, Introduction to the Theory of Linear Nonselfadjoint Operators. Transl. Math. Monographs, Vol. 18, Amer. Math. Soc., Providence, RI, 1969. [15] H. Landau, The notion of approximate eigenvalues applied to an integral equation of laser theory. Quart. Appl. Math. 35 (1977/78), 165โ172.
224
A. Bยจottcher, S. Grudsky, D. Huybrechs and A. Iserles
[16] H. Landau and H. Widom, Eigenvalue distribution of time and frequency limiting. J. Math. Analysis Appl. 77 (1980), 469โ481. [17] V.V. Peller, Hankel Operators and Their Applications. Springer-Verlag, New York, Berlin, Heidelberg, 2003 [18] L.N. Trefethen and M. Embree, Spectra and Pseudospectra: The Behavior of Nonnormal Matrices and Operators. Princeton University Press, Princeton, NJ, 2005. [19] L.A. Vainshtein, Open resonators for lasers. Soviet Physics JETP 40 (1963), 709โ 719. [20] H. Widom, On a class of integral operators with discontinuous symbol. Operator Theory: Adv. Appl. 4 (1982), 477โ500. Albrecht Bยจ ottcher Fakultยจ at fยจ ur Mathematik Technische Universitยจ at Chemnitz D-09107 Chemnitz, Germany e-mail:
[email protected] Sergei Grudsky CINVESTAV del I.P.N. Departamento de Matemยด aticas Apartado Postal 14-740 07000 Ciudad de Mยดexico, Mยดexico e-mail:
[email protected] Daan Huybrechs Departement Computerwetenschappen Katholieke Universiteit Leuven Celestijnenlaan 200A B-3001 Leuven, Belgium e-mail:
[email protected] Arieh Iserles Department of Applied Mathematics and Theoretical Physics Centre for Mathematical Sciences University of Cambridge Cambridge CB3 0WA, United Kingdom e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 225โ239 c 2012 Springer Basel AG โ
Factorization Versus Invertibility of Matrix Functions on Compact Abelian Groups Alex Brudnyi, Leiba Rodman and Ilya M. Spitkovsky Dedicated to the memory of Israel Gohberg
Abstract. Open problems are stated and some new results are proved concerning the relationship between invertibility and factorization in various Banach algebras of matrix-valued functions on connected compact abelian groups. Mathematics Subject Classi๏ฌcation (2000). 47A56, 47A68. Keywords. Compact abelian groups, function algebras, factorization of Wiener-Hopf type.
1. Introduction Factorizations of Wiener-Hopf type have been widely recognized and studied as an important mathematical tool. The concept of factorization formed originally within the theory of systems of singular integral equations and boundary value problems, see the monographs [14, 7, 34, 28], for example. In the early development, the in๏ฌuential paper [27] played a major role. Since then, factorizations of WienerHopf type have been studied in various contexts, in particular, the state space method [4, 3]. Another direction is factorization of matrix functions on connected compact abelian groups, a topic that has been studied in [35, 36, 20, 8, 9, 42]. Besides providing a uni๏ฌed framework for Wiener-Hopf factorizations of various types, such as relative to the unit circle or torus [19], and relative to the real line or to โ๐ โ almost periodic factorization which has been extensively studied in recent years โ the abstract setting of connected compact abelian groups leads to new points of view, problems, and results. In this paper, we focus on some outstanding problems in this area. Let ๐บ be a (multiplicative) connected compact abelian group and let ฮ be its (additive) character group. Recall that ฮ consists of continuous homomorphisms
226
A. Brudnyi, L. Rodman and I.M. Spitkovsky
of ๐บ into the group ๐ of unimodular complex numbers. Since ๐บ is compact, ฮ is discrete (in the natural topology as a dual locally compact abelian group) [43, Theorem 1.2.5], and since ๐บ is connected, ฮ is torsion free [43, Theorem 2.5.6]. By duality, ๐บ is the character group of ฮ. Note that the character group of every torsion free abelian group with the discrete topology is connected and compact [43, Theorems 1.2.5, 2.5.6]. It is well known [43] that, because ๐บ is connected, ฮ can be made into a linearly ordered group. So let โชฏ be a ๏ฌxed linear order such that (ฮ, โชฏ) is an ordered group. Let ฮ+ = {๐ฅ โ ฮ : ๐ฅ เชฐ 0}, ฮโ = {๐ฅ โ ฮ : ๐ฅ โชฏ 0}. Standard widely used examples of ฮ are โค (the group of integers), โ (the group of rationals with the discrete topology), โ (the group of reals with the discrete topology), and โค๐ , โ๐ with lexicographic or other ordering (where ๐ is a positive integer). If ๐ is a unital ring, we denote by ๐ ๐ร๐ the ๐ ร ๐ matrix ring over ๐ , and by ๐บ๐ฟ(๐ ๐ร๐ ) the group of invertible elements of ๐ ๐ร๐ . Let ๐ถ(๐บ) be the unital Banach algebra of (complex-valued) continuous functions on ๐บ (in the uniform topology), and let ๐ (๐บ) be the (non-closed) subalgebra of ๐ถ(๐บ) of all ๏ฌnite linear combinations of functions โจ๐, โ
โฉ, ๐ โ ฮ, where โจ๐, ๐โฉ stands for the action of the character ๐ โ ฮ on the group element ๐ โ ๐บ (thus, โจ๐, ๐โฉ โ ๐). Note that ๐ (๐บ) is dense in ๐ถ(๐บ) (this fact is a corollary of the Stone-Weierstrass theorem). For ๐=
๐ โ
๐๐๐ โจ๐๐ , .โฉ โ ๐ (๐บ), ๐1 , . . . , ๐๐ โ ฮ are distinct; ๐๐๐ โ= 0, ๐ = 1, 2, . . . , ๐,
๐=1
the Bohr-Fourier spectrum is de๏ฌned as the ๏ฌnite set ๐(๐) := {๐1 , . . . , ๐๐ }. The notion of Bohr-Fourier spectrum is extended from functions in ๐ (๐บ) to ๐ถ(๐บ) by continuity; indeed, since the Bohr-Fourier coe๏ฌcients are continuous in the uniform topology, we can use approximations of a given element in ๐ถ(๐บ) by elements of ๐ (๐บ). The Bohr-Fourier spectrum of ๐ด = [๐๐๐ ]๐๐,๐โ1 โ ๐ถ(๐บ)๐ร๐ is, by de๏ฌnition, the union of the Bohr-Fourier spectra of the ๐๐๐ โs. Note that the Bohr-Fourier spectra of elements of ๐ถ(๐บ) are at most countable; a proof for the case ฮ = โ is found, for example, in [16, Theorem 1.15]; it can be easily extended to general connected compact abelian groups ๐บ. We say that a unital Banach algebra โฌ โ ๐ถ(๐บ) is admissible if the following properties are satis๏ฌed: (1) ๐ (๐บ) is dense in โฌ; (2) โฌ is inverse closed (i.e., ๐ โ โฌ โฉ ๐บ๐ฟ(๐ถ(๐บ)) implies ๐ โ ๐บ๐ฟ(โฌ)). Important examples of admissible algebras are ๐ถ(๐บ) itself and the Wiener algebra ๐ (๐บ) that consists of all functions ๐ on ๐บ of the form โ ๐๐ โจ๐, ๐โฉ, ๐ โ ๐บ, (1.1) ๐(๐) = ๐โฮ
Factorization Versus Invertibility where ๐๐ โ โ and
โ ๐โฮ
227
โฃ๐๐ โฃ < โ. The norm in ๐ (๐บ) is de๏ฌned by โ โฃ๐๐ โฃ. โฅ๐โฅ1 = ๐โฮ
The inverse closed property of ๐ (๐บ) follows from the Bochner-Philips theorem [6] (a generalization of the classical Wienerโs theorem for the case when ๐บ = ๐). Other examples of admissible algebras are weighted Wiener algebras. A function ๐ : ฮ โ [1, โ) is called a weight if ๐(๐พ1 + ๐พ2 ) โค ๐(๐พ1 )๐(๐พ2 ) for all ๐พ1 , ๐พ2 โ ฮ and lim๐โโ ๐โ1 log(๐(๐๐พ)) = 0 for every ๐พ โ ฮ. The weighted โ Wiener algebra ๐๐ (๐บ) consists of all functions ๐ on ๐บ of the form (1.1) where ๐โฮ ๐(๐)โฃ๐๐ โฃ < โ, with the norm โ โฅ๐โฅ๐ = ๐(๐)โฃ๐๐ โฃ. ๐โฮ
One veri๏ฌes that ๐๐ (๐บ) is indeed an inverse closed unital Banach algebra, see [2] for the inverse closedness property. For an admissible algebra โฌ, we denote by โฌยฑ the closed unital subalgebra of โฌ formed by elements of โฌ with the Bohr-Fourier spectrum in ฮยฑ . Thus, โฌยฑ = โฌ โฉ ๐ถ(๐บ)ยฑ . Also, ๐ร๐ ) โฉ โฌ ๐ร๐ = ๐บ๐ฟ(โฌยฑ ). (1.2) ๐บ๐ฟ(๐ถ(๐บ)๐ร๐ ยฑ Next, we recall the concept of factorization in the connected compact abelian group setting, see, e.g., [36, 35, 20]. Let โฌ be an admissible algebra, and let ๐ด โ โฌ๐ร๐ . A representation of the form ๐ด(๐) = ๐ด+ (๐) (diag (โจ๐1 , ๐โฉ, . . . , โจ๐๐ , ๐โฉ)) ๐ดโ (๐), ๐ดยฑ , ๐ดโ1 ยฑ
๐ร๐ โฌยฑ
๐ โ ๐บ,
(1.3)
โ and ๐1 , . . . , ๐๐ โ ฮ, is called a (left) โฌ-factorization of ๐ด where (with respect to the order โชฏ). It follows that the elements ๐1 , . . . , ๐๐ in (1.3) are uniquely determined by ๐ด, up to a permutation. Borrowing the terminology from the classical (ฮ = โค, ๐บ = ๐) setting, we call them the partial indices of ๐ด. The sum ๐1 + โ
โ
โ
+ ๐๐ is the total index of ๐ด. For ๐ = 1, the only partial index of ๐ด (therefore coinciding with its total index) is called simply the index of ๐ด. We say that ๐ด โ โฌ ๐ร๐ is โฌ-factorable if a โฌ-factorization of ๐ด exists. Denote by ๐บ๐ฟ๐น (โฌ ๐ร๐ ) the set of all โฌ-factorable ๐ ร ๐ matrix functions. Clearly, it is necessary that ๐ด โ ๐บ๐ฟ(โฌ ๐ร๐ ) for ๐ด to be โฌ-factorable. In this paper, we overview some available results and state open problems concerning the opposite direction: What can be said about the structure of the set of factorable matrix functions as a subset of invertible matrix functions? If ฮ = โค, then ๐บ = ๐, and ๐ (๐)-factorization is the classical Wiener-Hopf factorization on the unit circle. As it happens, in this case the above-mentioned necessary invertibility condition is su๏ฌcient as well. This result is due to GohbergKrein [27], and can also be found in many monographs, e.g., [14, 34], and a more recent survey [24]. On the other hand, it is well known that the condition ๐ด โ ๐บ๐ฟ(๐ถ(๐)๐ร๐ ) is not su๏ฌcient for ๐ถ(๐)-factorization even when ๐ = 1; an example can be found, e.g., in [30]. However, the set ๐บ๐ฟ๐น (๐ถ(๐)๐ร๐ ) is dense in ๐บ๐ฟ(๐ถ(๐)๐ร๐ ).
228
A. Brudnyi, L. Rodman and I.M. Spitkovsky
For ฮ = โ the dual group ๐บ is the Bohr compacti๏ฌcation โห of โ, so that ๐ถ(๐บ) is nothing but the algebra ๐ด๐ of Bohr almost periodic functions while ๐ (๐บ) is its (non-closed) subalgebra ๐ด๐ ๐ of ๐ด๐ functions with absolutely convergent Bohr-Fourier series. The โฌ-factorization corresponding to these cases, called ๐ด๐ and ๐ด๐ ๐ factorization, respectively, in the scalar case was considered in [15] and [29]. The matrix setting was ๏ฌrst treated in [31, 32]. It was then observed (see also [33] and [7, Section 15.1] for the full proofs) that already for ๐ = 2 there exist triangular matrix functions in ๐บ๐ฟ(๐ด๐ ๐ ๐ร๐ ) which are not even ๐ด๐ -factorable. These matrix functions have the form [ ] โจ๐ + ๐ฟ, ๐โฉ 0 ห ๐ด(๐) = , ๐ โ โ, (1.4) ๐1 โจโ๐, ๐โฉ + ๐2 + ๐3 โจ๐ฟ, ๐โฉ โจโ(๐ + ๐ฟ), ๐โฉ where ๐, ๐ฟ > 0, ๐ and ๐ฟ are not commensurable, and ๐1 , ๐2 , ๐3 are non-zero complex numbers such that (log โฃ๐3 โฃ)๐ + (log โฃ๐1 โฃ)๐ฟ = (log โฃ๐2 โฃ)(๐ + ๐ฟ).
(1.5)
In other words, the necessary invertibility condition in general is not su๏ฌcient โ a striking contrast with the scalar setting. The details can be found in [7], while more recent new classes are discussed in [13, 40].
2. Denseness of ๐ฎ๐ณ๐ญ (ํ) in ๐ฎ๐ณ(ํ) We start with the scalar case. An admissible algebra is said to be decomposing if โฌ+ + โฌโ = โฌ. For example, the weighted Wiener algebras are decomposing, but ๐ถ(๐บ) is not. Theorem 2.1. ([8]) Let โฌ โ ๐ถ(๐บ) be an admissible algebra, where ๐บ is a connected compact abelian group. Then: (a) The set ๐บ๐ฟ๐น (โฌ) of โฌ-factorable scalar functions is dense in ๐บ๐ฟ(โฌ); (b) The equality ๐บ๐ฟ๐น (โฌ) = ๐บ๐ฟ(โฌ) holds if and only if โฌ is decomposing. In the classical case ๐บ = ๐ part (b) is well known, see [28, Theorem 3.1], for example. Moreover, in this setting the results extend verbatim to the matrix case. Theorem 2.2. Let โฌ โ ๐ถ(๐) be an admissible algebra. Then: (a) The set ๐บ๐ฟ๐น (โฌ ๐ร๐ ) is dense in ๐บ๐ฟ(โฌ ๐ร๐ ); moreover, ๐บ๐ฟ๐น (โฌ ๐ร๐ ) is dense in โฌ ๐ร๐ ; (b) The equality ๐บ๐ฟ๐น (โฌ ๐ร๐ ) = ๐บ๐ฟ(โฌ ๐ร๐ ) holds if and only if โฌ is decomposing. Indeed, all trigonometric ๐ร๐ matrix polynomials which are invertible on the unit circle are โฌ-factorable (see, for example, the proof of Lemma VIII.2.1 in [30] or Section 2.4 in [34]); on the other hand, the set of invertible on ๐ trigonometric ๐ ร ๐ matrix polynomials is easily seen to be dense in ๐ (๐), hence it is also dense in โฌ ๐ร๐ . Part (b) was proved in [22] (see also [10, 11]). We do not know any other group ๐บ for which ๐บ๐ฟ๐น (โฌ ๐ร๐ ) = ๐บ๐ฟ(โฌ ๐ร๐ ) holds for every decomposing algebra โฌ. Thus:
Factorization Versus Invertibility
229
Open Problem 2.3. Identify those connected compact abelian groups ๐บ and their character groups ฮ with a linear order for which ๐บ๐ฟ๐น (โฌ ๐ร๐ ) = ๐บ๐ฟ(โฌ ๐ร๐ )
(2.1)
holds for every decomposing algebra โฌ. It was conjectured in [36] that (2.1) holds for โฌ = ๐ (๐บ) if and only if ฮ is isomorphic to a subgroup of the (additive) group of rational numbers โ. On the other hand, part (a) of Theorem 2.2 extends to some other groups: Theorem 2.4. The following statements are equivalent: (1) ๐บ๐ฟ๐น (โฌ ๐ร๐ ) is dense in โฌ ๐ร๐ , for every admissible algebra โฌ; ๐ร๐ ; (2) ๐บ๐ฟ๐น (๐ถ(๐บ)๐ร๐ ) is dense in ๐ถ(๐บ) (3) ฮ is (isomorphic to) a subgroup of โ. Proof. (1) =โ (2) is obvious, while (3) =โ (1) is proved in [8]. Suppose (2) holds; ๐ร๐ in particular, ๐บ๐ฟ(๐ถ(๐บ)๐ร๐ ) is dense in ๐ถ(๐บ) . We now use the well-known fact (see [18, 41, 39, 17]) that if ๐ is a compact Hausdor๏ฌ topological space, then ๐ถ(๐), the ๐ถ โ -algebra of continuous complex-valued functions on ๐, has dense group of invertible elements if and only if the covering dimension of ๐ is at most one; moreover, if ๐ถ(๐) has dense group of invertible elements, then so does ๐ถ(๐)๐ร๐ for every integer ๐ โฅ 1. Thus, ๐บ has covering dimension one. Since the two-dimensional torus ๐2 has covering dimension two, it follows that ฮ does not contain a subgroup isomorphic to โค2 . It is easy to see that any such ฮ is isomorphic to a subgroup of โ. โก
3. Nondenseness Let us return to example (1.4). As also was shown in [31, 33], the matrix (1.4) is ๐ด๐ ๐ factorable when the equality (1.5) does not hold. Therefore, the non-๐ด๐ ห factorable matrices delivered by (1.4), (1.5) are limits of ๐ (โ)-factorable ones. In all other concrete examples of non-factorability (the more recent of which can be found in [5, 1, 12]) the non-factorable matrix function always is a limit of factorable ones. In view of this situation, many researchers considered the following conjecture plausible: The set of ๐ (๐บ)-factorable matrix functions is dense in the group ๐บ๐ฟ(๐ (๐บ)๐ร๐ ). It turns out, however, that for ฮ = โ, as well as in many other cases, this conjecture fails for any admissible algebra โฌ. We describe the situation in a more general setting of triangularizable matrix functions. Let โฌ be an admissible algebra. An element ๐ด โ โฌ ๐ร๐ is said to be (left) โฌ-triangularizable if ๐ด admits a representation (1.3), where the middle term diag (โจ๐1 , ๐โฉ, . . . , โจ๐๐ , ๐โฉ) is replaced by a triangular matrix ๐ = [๐ก๐๐ ]๐๐,๐=1 with ๐ก๐๐ โ โฌ for ๐, ๐ = 1, . . . , ๐, ๐ก๐๐ = 0 if ๐ > ๐, and the diagonal elements ๐ก11 , . . . , ๐ก๐๐ belonging to ๐บ๐ฟ(โฌ). Denote by ๐บ๐ฟ๐ (โฌ ๐ร๐) the set of ๐ ร ๐ โฌ-triangularizable matrix functions. Clearly, ๐บ๐ฟ๐น (โฌ ๐ร๐ ) โ ๐บ๐ฟ๐ (โฌ ๐ร๐ ) โ ๐บ๐ฟ(โฌ ๐ร๐ ).
230
A. Brudnyi, L. Rodman and I.M. Spitkovsky
The following question arises naturally: Does ๐บ๐ฟ(โฌ ๐ร๐ ) = ๐บ๐ฟ๐ (โฌ ๐ร๐ ) hold for admissible algebras? The next result shows that generally the answer is no. Denote by ๐ฏ (โฌ ๐ร๐ ) the minimal closed subgroup of ๐บ๐ฟ(โฌ ๐ร๐ ) that contains ๐บ๐ฟ๐ (โฌ ๐ร๐). Theorem 3.1. ([8]) Let ฮ be a torsion free abelian group (in discrete topology) that contains a subgroup isomorphic to โค3 , and let โฌ be an admissible algebra of continuous functions on ๐บ, the dual of ฮ. Then, for every natural ๐ โฅ 2 there exist in๏ฌnitely many pathwise connected components of ๐บ๐ฟ(โฌ ๐ร๐ ) with the property that each one of these components does not intersect ๐ฏ (โฌ ๐ร๐ ). In particular, ๐ฏ (โฌ ๐ร๐ ), and a fortiori ๐บ๐ฟ๐น (โฌ ๐ร๐ ), is not dense in ๐บ๐ฟ(โฌ ๐ร๐ ). Theorem 3.1 naturally leads to Open Problem 3.2. (i) Describe all connected compact groups ๐บ (or their duals ฮ) such that ๐บ๐ฟ๐น (โฌ ๐ร๐ ) is dense in ๐บ๐ฟ(โฌ ๐ร๐ ), for any admissible algebra โฌ. (ii) Describe all connected compact groups ๐บ such that ๐ฏ (โฌ ๐ร๐ ) is dense in ๐บ๐ฟ(โฌ ๐ร๐ ), for any admissible algebra โฌ. Theorem 3.1 does not address the situation when the given ๐ด โ ๐บ๐ฟ(โฌ ๐ร๐ ) is already triangular. It is still a possibility that any such matrix can be approximated by โฌ-factorable ones. Thus, Open Problem 3.3. Prove or disprove that ๐บ๐ฟ๐น (โฌ ๐ร๐ ) is dense in ๐บ๐ฟ๐ (โฌ ๐ร๐). In relation with Open Problem 3.3 note that factorization of triangular matrix functions arises naturally in the consideration of convolution type equations on (unions of) intervals, see [21, 37, 38, 44, 7].
4. Topological properties of ๐ฎ๐ณ๐ญ (ํ๐ร๐ ) In this section we discuss brie๏ฌy some topological properties of the set ๐บ๐ฟ๐น (โฌ ๐ร๐ ). It will be assumed throughout the section that the admissible algebra โฌ is decomposing. A standard argument (see, for example, [23, Theorem XXIX.9.1]) shows that there exists an open neighborhood ฮ of identity in ๐บ๐ฟ(โฌ ๐ร๐ ) such that every element of ฮ admits a canonical โฌ-factorization, i.e., a โฌ-factorization with all partial indices equal to zero. As a consequence, we obtain: Proposition 4.1. The set of those ๐ด โ ๐บ๐ฟ๐น (โฌ ๐ร๐ ) that admit a canonical factorization is open in ๐บ๐ฟ(โฌ ๐ร๐ ). It is not known for a general ฮ whether or not the set ๐บ๐ฟ๐น (โฌ ๐ร๐ ) is open. Thus: Open Problem 4.2. Identify those connected compact abelian groups ๐บ for which ๐บ๐ฟ๐น (โฌ ๐ร๐ ) is open.
Factorization Versus Invertibility
231
For example, ๐บ = ๐ is such since then ๐บ๐ฟ๐น (โฌ ๐ร๐ ) = ๐บ๐ฟ(โฌ ๐ร๐ ). Proposition 4.1 leads to the following stability property of indices of scalar functions. Observe that in view of Theorem 2.1, every invertible element of โฌ is โฌ-factorable. Proposition 4.3. Let ๐ด โ ๐บ๐ฟ(โฌ). Then the index of every nearby (in the topology of โฌ) function ๐ต โ ๐บ๐ฟ(โฌ) is identical to that of ๐ด. Indeed, replacing ๐ด with โจโ๐, โ
โฉ๐ด(โ
), where ๐ is the index of ๐ด, we may assume that the latter equals zero, i.e., the โฌ-factorization of ๐ด is canonical. Now the result is immediate from Proposition 4.1. Next, consider the (pathwise) connected components of ๐บ๐ฟ๐น (โฌ ๐ร๐ ). Theorem 4.4. Every connected component of ๐บ๐ฟ๐น (โฌ ๐ร๐ ) has the form ๐ถ๐บ๐ฟ๐น๐ (โฌ ๐ร๐ ) := {๐ด โ ๐บ๐ฟ๐น (โฌ ๐ร๐ ) : the total index of ๐ด equals ๐}, where ๐ โ ฮ is ๏ฌxed. Thus, the connected component of ๐บ๐ฟ๐น (โฌ ๐ร๐ ) are parametrized by ๐ โ ฮ. For the Wiener algebra, Proposition 4.4 was proved in [9]. Proof. The proof of [8, Theorem 6.2] (see also [9, Section 6]) shows that every ๐ด โ ๐ถ๐บ๐ฟ๐น๐ (โฌ ๐ร๐ ) can be connected within ๐บ๐ฟ๐น (โฌ ๐ร๐ ) to diag (1, 1, . . . , 1, โจ๐, โ
โฉ). Conversely, assume there exists a continuous path from diag (1, 1, . . . , 1, โจ๐1 , โ
โฉ) to diag (1, 1, . . . , 1, โจ๐2 , โ
โฉ) within ๐บ๐ฟ๐น (โฌ ๐ร๐ ), where ๐1 , ๐2 โ ฮ. Passing to determinants, we obtain a path from โจ๐1 , โ
โฉ to โจ๐2 , โ
โฉ within ๐บ๐ฟ(โฌ). By Proposition 4.3 we must have ๐1 = ๐2 . โก In particular, the set of canonically โฌ-factorable ๐ ร ๐ matrix functions is connected. Canonically โฌ-factorable scalar functions can be described in several ways. Denote by ๐บ๐ฟ0 (โฌ) the connected component of ๐บ๐ฟ(โฌ) that contains the constant function 1. Proposition 4.5. The following statements are equivalent for ๐ด โ ๐บ๐ฟ(โฌ): (1) ๐ด admits canonical โฌ-factorization; (2) ๐ด has a logarithm in โฌ, i.e., ๐ด = ๐๐ต for some ๐ต โ โฌ; (3) ๐ด โ ๐บ๐ฟ0 (โฌ). Proof. The equivalence of (2) and (3) is well known for commutative unital Banach algebras. If (1) holds, then (3) holds in view of the connectivity of the set of canonically โฌ-factorable scalar functions. Finally, if (2) holds, then write ๐ต = ๐ต+ + ๐ตโ , where ๐ตยฑ โ โฌยฑ (the decomposing property of โฌ is used here). It follows that ๐ด = ๐๐ต+ ๐๐ตโ is a canonical โฌ-factorization. โก We conclude the section with some observations concerning triangular matrix functions. Since โฌ is decomposing, according to Theorem 2.1 factorability of such matrices implies factorability of all their diagonal entries. The converse is also
232
A. Brudnyi, L. Rodman and I.M. Spitkovsky
true, whenever ๐บ and ฮ are such that (2.1) holds. The next statement shows that condition (2.1) is irrelevant, provided that the factorization of diagonal elements is canonical. Proposition 4.6. Let ๐ด โ ๐บ๐ฟ(โฌ ๐ร๐ ) be upper (or lower) triangular such that the diagonal elements of ๐ด belong to ๐บ๐ฟ0 (โฌ). Then ๐ด admits a canonical โฌfactorization. Proof. The proof is by induction on ๐, the case ๐ = 1 being trivial. Let ๐ด โ ) be upper triangular with diagonal elements in ๐บ๐ฟ0 (โฌ). Write ๐ด = ๐บ๐ฟ(โฌ ๐ร๐] [ ๐ต ๐ถ , where ๐ต โ ๐บ๐ฟ(โฌ), ๐ท โ ๐บ๐ฟ(โฌ (๐โ1)ร(๐โ1) ). By the induction hypoth0 ๐ท esis, ๐ต and ๐ท admit canonical โฌ-factorizations ๐ต = ๐ต+ ๐ตโ , ๐ท = ๐ท+ ๐ทโ . Writing 1ร(๐โ1) โ1 โ1 ๐ต+ ๐ถ๐ทโ = ๐+ +๐โ , where ๐ยฑ โ โฌยฑ , we have a โฌ-canonical factorization ][ ][ ][ ] [ 0 0 1 ๐+ 1 ๐โ ๐ตโ ๐ต+ . โก ๐ด= 0 ๐ท+ 0 ๐ทโ 0 ๐ผ๐โ1 0 ๐ผ๐โ1 If the admissible algebra โฌ is not decomposing, then we can only assert that the group of upper triangular matrices in ๐บ๐ฟ(โฌ ๐ร๐ ) with diagonal elements in ๐บ๐ฟ0 (โฌ), is dense in the set of canonically โฌ-factorable upper triangular matrices in ๐บ๐ฟ(โฌ ๐ร๐ ). Note also that a triangular matrix ๐ด may admit a canonical โฌ-factorization while the factorization of its diagonal entries is non-canonical. The classical example of this phenomenon for ฮ = โค can be found in [27], while for ฮ = โ, e.g., matrices (1.4) with (1.5) not satis๏ฌed will do the job. Other examples of this nature are scattered throughout Chapters 14, 15 of [7].
5. Small Bohr-Fourier spectra Let ๐ด โ ๐บ๐ฟ(โฌ ๐ร๐ ). In view of Theorem 3.1 it is unlikely (if ฮ contains โค3 and ๐ โฅ 2) that ๐ด admits a โฌ-factorization. So one may consider imposing additional conditions on ๐ด to ensure โฌ-factorability. In this section, we consider small BohrFourier spectra. We denote by #๐(๐ด) the number of elements in the Bohr-Fourier spectrum of ๐ด. To start with an easy case, note that if #๐(๐ด) โค 2 then ๐ด is โฌ-factorable, for any admissible algebra โฌ. This can be proved without di๏ฌculty using the Kronecker form for two complex matrices. In the following, we need to distinguish archimedean and non-archimedean groups. The group ฮ (with the ๏ฌxed linear order โชฏ) is said to be archimedean if for every ๐, ๐ โป 0 there exists an integer ๐ such that ๐๐ โป ๐. A well-known Hยจolderโs theorem states that a linearly ordered abelian group is archimedean if and only if it is order isomorphic to a subgroup of โ. We have a non-factorability result, proved in [36, 35]:
Factorization Versus Invertibility
233
Theorem 5.1. Assume ฮ is non-archimedean (for example, ฮ = โค๐ , ๐ > 1, with the lexicographic order). Then for every ๐ โฅ 2 there is a ๐ (๐บ)-nonfactorable triangular ๐ด โ ๐บ๐ฟ(โฌ ๐ร๐ ) with #๐(๐ด) = 4. A concrete example is given in [35]: Assume 0 โบ ๐ โบ ๐, ๐, ๐ โ ฮ are such that ๐๐ โบ ๐ for all positive integers ๐. Let [ ] โจ๐, ๐โฉ๐ผ๐โ1 0 ๐ด(๐) = , ๐ โ ๐บ, (5.1) ๐ถ1 โ โจ๐, ๐โฉ๐ถ2 โจโ๐, ๐โฉ where ๐ถ1 = ๐ถ2 = [1 0 . . . 0] โ โ1ร(๐โ1) . Then ๐ด is not ๐ (๐บ)-factorable. This example is a particular case of a more general result: Theorem 5.2. Let ๐ด have the form [ ] โจ๐, ๐โฉ๐ผ๐ 0 ๐ด(๐) = , ๐ถ1 โจ๐, ๐โฉ โ ๐ถ2 โจ๐, ๐โฉ โจโ๐, ๐โฉ๐ผ๐
๐ โ ๐บ,
(5.2)
where ๐ถ1 , ๐ถ2 โ โ๐ร๐ . Assume that ๐ โป 0, ๐ โป ๐, and ๐๐ โบ ๐,
๐๐ โบ ๐
for all integers ๐.
(5.3)
Then for every admissible algebra โฌ, ๐ด admits a โฌ-factorization if rank (๐1 ๐ถ1 โ ๐2 ๐ถ2 ) = max{rank (๐ง1 ๐ถ1 โ ๐ง2 ๐ถ2 ) : ๐ง1 , ๐ง2 โ โ} for every ๐1 , ๐2 โ โ satisfying โฃ๐1 โฃ = โฃ๐2 โฃ = 1. (5.4) Moreover, in this case the factorization indices of ๐ด belong to the set {ยฑ๐, ยฑ๐, ยฑ๐, ยฑ(๐ โ (๐ โ ๐)), . . . , ยฑ(๐ โ min{๐, ๐}(๐ โ ๐))},
(5.5)
and if ๐ โ ๐๐ (๐ โ ๐), ๐ = 1, 2, . . . , ๐ , and โ(๐ โ โ๐ (๐ โ ๐)), ๐ = 1, 2, . . . , ๐ก, are the factorization indices of ๐ด other than ยฑ๐, ยฑ๐, ยฑ๐, then ๐1 + โ
โ
โ
+ ๐๐ + โ1 + โ
โ
โ
+ โ๐ก + ๐ โค ๐,
๐1 + โ
โ
โ
+ ๐๐ + โ1 + โ
โ
โ
+ โ๐ก + ๐ก โค ๐,
and ๐ก โ ๐ = ๐ โ ๐. Conversely, if ๐ด admits a ๐ (๐บ)-factorization, then (5.4) holds. Theorem 5.2 is proved in [35, 36] for the case when โฌ = ๐ (๐บ), with less explicit description of the factorization indices1 . We do not know whether the converse statement holds for any admissible algebra di๏ฌerent from ๐ (๐บ). We provide some details of the proof of Theorem 5.2 in a separate Section 6. In view of Theorem 5.1 we have the following open problem: Open Problem 5.3. Assume that the subgroup generated by ๐พ1 , ๐พ2 , ๐พ3 โ ฮ is not archimedean. Prove or disprove that every ๐ด โ ๐บ๐ฟ(โฌ ๐ร๐ ) with ๐(๐ด) = {๐พ1 , ๐พ2 , ๐พ3 } is โฌ-factorable, for any admissible algebra โฌ. 1 Note
that two ยฑ signs are inadvertently omitted in the statement of [35, Theorem 3].
234
A. Brudnyi, L. Rodman and I.M. Spitkovsky
For archimedean groups which are non-isomorphic to any subgroup of โ, an example exists of an invertible non-๐ (๐บ)-factorable ๐ ร ๐ matrix function ๐ด โ ๐ (๐บ) with #๐(๐ด) = 5 (see [35]); for 2 ร 2 matrices this example stems from (1.4). Thus: Open Problem 5.4. Assume that ฮ is archimedean and not isomorphic to a subgroup of โ. Prove or disprove that every ๐ด โ ๐บ๐ฟ(โฌ ๐ร๐ ) with #๐(๐ด) = 3 or #๐(๐ด) = 4 is โฌ-factorable.
6. Proof of Theorem 5.2 We follow the same approach as in [36, 35]. For the readersโ convenience, we provide some details. Assume that (5.4) holds. We have to prove that ๐ด given by (5.2) is โฌfactorable. Applying the transformation ๐ถ1 โโ ๐๐ถ1 ๐,
๐ถ2 โโ ๐๐ถ2 ๐,
for suitable invertible matrices ๐ and ๐ , we may assume that the pair (๐ถ1 , ๐ถ2 ) is in the Kronecker normal form; in other words, ๐ถ1 and ๐ถ2 are direct sums of blocks of the following types: (a) ๐ถ1 and ๐ถ2 are of size ๐ ร (๐ + 1) of the form ] [ [ ๐ถ1 = ๐ผ๐ 0๐ร1 , ๐ถ2 = 0๐ร1
๐ผ๐
]
.
(b) ๐ถ1 and ๐ถ2 are of size (๐ + 1) ร ๐ of the form [ [ ] ] ๐ผ๐ 01ร๐ ๐ถ1 = , ๐ถ2 = . 01ร๐ ๐ผ๐ (c) ๐ถ1 ๐ถ2 (d) ๐ถ1 (e) ๐ถ1 (f) ๐ถ1
is the ๐ ร ๐ upper triangular nilpotent Jordan block, denoted by ๐๐ , and = ๐ผ๐ . = ๐ผ๐ , and ๐ถ2 = ๐๐ . and ๐ถ2 are both invertible of the same size, say ๐ ร ๐. and ๐ถ2 are both zero matrices of the same size.
The proof thereby is reduced to the cases (a)โ(f). The case (f) is trivial. The case (a) is dealt with in full detail in [35, proof of Theorem 3] (using arguments similar to those in [31]); the indices in this case are โ๐ (๐ times), ๐ โ ๐(๐ โ ๐), and ๐ (๐ times), and the factors ๐ดโ1 ยฑ actually belong to ๐ (๐บ) โฉ โฌยฑ (such factorization was termed ๏ฌnite factorization in [36]). The case (b) is reduced to (a) as in [35], using the transformation [ [ ] ] 0 ๐ฝ๐ 0 ๐ฝ๐+1 ๐ด โโ ๐ดโ , ๐ฝ๐ 0 ๐ฝ๐+1 0 where ๐ฝ๐ is the ๐ร๐ matrix with 1โs along the top-right to the left-bottom diagonal and zeros elsewhere. The cases (c), (d), and (e) follow from the fact (proved in
Factorization Versus Invertibility
235
[36]) that, under the hypotheses of Theorem 5.2, if ๐ถ2 is invertible and the spectral radius of ๐ถ2โ1 ๐ถ1 is less than one, then a โฌ-factorization of ๐ด is given by formulas ] [ โโ โ ๐=0 (๐ถ2โ1 ๐ถ1 )๐ ๐ถ2โ1 โจ๐ โ ๐ โ ๐(๐ โ ๐), โ
โฉ ๐ผ ๐ด+ = , ๐ผ 0 ] [ โจ๐, โ
โฉ๐ผ 0 , ฮ= 0 โจโ๐, โ
โฉ๐ผ ] [ โจโ๐ โ ๐, โ
, โฉ๐ผ ๐ถ1 โจ๐ โ ๐, โ
โฉ โ ๐ถ2 , ๐ดโ = โโ โ1 ๐ โ1 0 ๐=0 (๐ถ2 ๐ถ1 ) ๐ถ2 โจโ๐(๐ โ ๐), โ
โฉ and if ๐ถ1 is invertible and the spectral radius of ๐ถ2 ๐ถ1โ1 is less than one, then a โฌ-factorization of ๐ด is given by formulas [ ] โโ โจ๐ โ ๐, โ
โฉ๐ผ โ๐ถ1โ1 ๐=0 (๐ถ2 ๐ถ1โ1 )๐ โจ๐(๐ โ ๐), โ
โฉ ๐ด+ = , 0 โ๐ถ2 ๐๐โ๐ + ๐ถ1 [ ] โจ๐, โ
โฉ๐ผ 0 ฮ= , 0 โจโ๐, โ
โฉ๐ผ [ ] โโ ๐ผ ๐ถ1โ1 ๐=0 (๐ถ2 ๐ถ1โ1 )๐ โจ๐(๐ โ ๐) โ ๐ โ ๐, โ
โฉ ๐ดโ = . 0 ๐ผ Note that using the Jordan form of ๐ถ2โ1 ๐ถ1 or of ๐ถ2 ๐ถ1โ1 , as the case may be, and using the inverse closed property of โฌ, one easily veri๏ฌes that the matrices in these formulas indeed belong to โฌ ๐ร๐ , for any admissible algebra โฌ. Assume now that ๐ด is ๐ (๐บ)-factorable. By [42, Theorem 1], there is a ๐ (๐บ)factorization of ๐ด with all factors having Bohr-Fourier coe๏ฌcients in the subgroup of ฮ generated by ๐, ๐, ๐. Thus, we may assume without loss of generality that ฮ = โค๐ , where ๐ = 2 or ๐ = 3. Now argue as in the โonly ifโ part of [35, Section 3]. โก
7. Wiener-Hopf equivalence Let ๐ด1 , ๐ด2 โ โฌ ๐ร๐ . We call ๐ด1 and ๐ด2 (left) Wiener-Hopf equivalent if there exist ๐ร๐ ๐ร๐ such that ๐ดโ1 and ๐ดยฑ โ โฌยฑ ยฑ โ โฌยฑ ๐ด2 (๐) = ๐ด+ (๐)๐ด1 (๐)๐ดโ (๐),
๐ โ ๐บ.
Clearly, this is indeed an equivalence relation. In the setting of operator polynomials and, more generally, analytic operator-valued functions, it was introduced in [25] and then investigated further in [3] and[26, Chapters XIII and XIV]. Of course, an invertible ๐ด โ โฌ ๐ร๐ is (left) โฌ-factorable if and only if it is (left) Wiener-Hopf equivalent to a diagonal matrix function, and two โฌ-factorable matrix functions are Wiener-Hopf equivalent if and only if the sets of their partial indices coincide. In the case of โฌ being a weighted Wiener algebra, the notion of Wiener-Hopf equivalence, along with the latter observation, are in [20].
236
A. Brudnyi, L. Rodman and I.M. Spitkovsky
If โฌ is such that invertibility in โฌ ๐ร๐ implies factorability (for example, if conditions of Theorem 2.2 b) hold), then the Wiener-Hopf equivalence classes are characterized completely by the sets of partial indices. In general, however, we arrive at Open Problem 7.1. For a given admissible algebra โฌ, describe the Wiener-Hopf equivalence classes (and their canonical representatives) of ๐บ๐ฟ(โฌ ๐ร๐). This problem is still open even for โฌ = ๐ด๐ or ๐ด๐ ๐ . Moreover, it is not even clear what are possible values of ๐ for which there exist โWiener-Hopf irreducibleโ ๐ด โ ๐บ๐ฟ(โฌ ๐ร๐ ), that is, ๐ด is not Wiener-Hopf equivalent to block diagonal matrices with at least two diagonal blocks. In all the constructions we are aware of (including [8]) only 1 ร 1 and 2 ร 2 blocks occur as Wiener-Hopf irreducibles, but there is no obvious reason why this should always be the case. Acknowledgment The research of Alex Brudnyi is supported in part by NSERC.
References [1] S. Avdonin, A. Bulanova, and W. Moran, Construction of sampling and interpolating sequences for multi-band signals. The two-band case, Int. J. Appl. Math. Comput. Sci. 17 (2007), no. 2, 143โ156. [2] R. Balan and I. Krishtal, An almost periodic noncommutative Wienerโs lemma, J. Math. Anal. Appl. 370 (2010), 339โ349. [3] H. Bart, I. Gohberg, and M.A. Kaashoek, Invariants for Wiener-Hopf equivalence of analytic operator functions, Constructive methods of Wiener-Hopf factorization, Operator Theory: Advances and Applications, vol. 21, Birkhยจ auser, Basel, 1986, pp. 317โ 355. [4] H. Bart, I. Gohberg, M.A. Kaashoek, and A.C.M. Ran, A state space approach to canonical factorization with applications, OT vol. 200, Birkhยจ auser Verlag, Basel and Boston, 2010. [5] M.A. Bastos, Yu.I. Karlovich, I.M. Spitkovsky, and P.M. Tishin, On a new algorithm for almost periodic factorization, Recent Progress in Operator Theory (Regensburg, 1995) (I. Gohberg, R. Mennicken, and C. Tretter, eds.), Operator Theory: Advances and Applications, vol. 103, Birkhยจ auser Verlag, Basel and Boston, 1998, pp. 53โ74. [6] S. Bochner and R.S. Phillips, Absolutely convergent Fourier expansion for noncommutative normed rings, Ann. of Math. 43 (1942), 409โ418. [7] A. Bยจ ottcher, Yu.I. Karlovich, and I.M. Spitkovsky, Convolution operators and factorization of almost periodic matrix functions, OT vol. 131, Birkhยจ auser Verlag, Basel and Boston, 2002. [8] A. Brudnyi, L. Rodman, and I.M. Spitkovsky, Non-denseness of factorable matrix functions, J. Functional Analysis 261 (2011), 1969โ1991. , Projective free algebras of continuous functions on compact abelian groups, [9] J. Functional Analysis 259 (2010), 918โ932.
Factorization Versus Invertibility
237
[10] M.S. Budjanu and I.C. Gohberg, General theorems on the factorization of matrixvalued functions, I. Fundamental theorems, Amer. Math. Soc. Transl. 102 (1973), 1โ14. [11] , General theorems on the factorization of matrix-valued functions, II. Some tests and their consequences, Amer. Math. Soc. Transl. 102 (1973), 15โ26. [12] M.C. Cห amara, C. Diogo, Yu.I. Karlovich, and I.M. Spitkovsky, Factorizations, Riemann-Hilbert problems and the corona theorem, arXiv:1103.1935v1 [math.FA] (2011), 1โ32. [13] M.C. Cห amara, Yu.I. Karlovich, and I.M. Spitkovsky, Almost periodic factorization of some triangular matrix functions, Modern Analysis and Applications. The Mark Krein Centenary Conference (V. Adamyan, Y. Berezansky, I. Gohberg, M. Gorbachuk, A. Kochubei, H. Langer, and G. Popov, eds.), Operator Theory: Advances and Applications, vol. 190, Birkhยจ auser Verlag, Basel and Boston, 2009, pp. 171โ190. [14] K.F. Clancey and I. Gohberg, Factorization of matrix functions and singular integral operators, OT vol. 3, Birkhยจ auser, Basel and Boston, 1981. [15] L. Coburn and R.G. Douglas, Translation operators on the half-line, Proc. Nat. Acad. Sci. USA 62 (1969), 1010โ1013. [16] C. Corduneanu, Almost periodic functions, J. Wiley & Sons, 1968. [17] H.G. Dales, Banach algebras and automatic continuity, London Mathematical Society Monographs. New Series, vol. 24, The Clarendon Press Oxford University Press, New York, 2000, Oxford Science Publications. [18] T.W. Dawson and J.F. Feinstein, On the denseness of the invertible group in Banach algebras, Proc. Amer. Math. Soc. 131 (2003), no. 9, 2831โ2839. [19] T. Ehrhardt and C.V.M. van der Mee, Canonical factorization of continuous functions on the ๐-torus, Proc. Amer. Math. Soc. 131 (2003), no. 3, 801โ813. [20] T. Ehrhardt, C.V.M. van der Mee, L. Rodman, and I.M. Spitkovsky, Factorizations in weighted Wiener algebras on ordered abelian groups, Integral Equations and Operator Theory 58 (2007), 65โ86. [21] M.P. Ganin, On a Fredholm integral equation whose kernel depends on the di๏ฌerence of the arguments, Izv. Vys. Uchebn. Zaved. Matematika (in Russian) (1963), no. 2 (33), 31โ43. [22] I. Gohberg, The factorization problem in normed rings, functions of isometric and symmetric operators, and singular integral equations, Uspehi Mat. Nauk 19 (1964), 71โ124 (in Russian). [23] I. Gohberg, S. Goldberg, and M.A. Kaashoek, Classes of linear operators. Vol. II, Birkhยจ auser Verlag, Basel and Boston, 1993. [24] I. Gohberg, M.A. Kaashoek, and I.M. Spitkovsky, An overview of matrix factorization theory and operator applications, Operator Theory: Advances and Applications 141 (2003), 1โ102. [25] I. Gohberg, M.A. Kaashoek, and F. van Schagen, Similarity of operator blocks and canonical forms. II. In๏ฌnite-dimensional case and Wiener-Hopf factorization, Topics in modern operator theory (Timiยธsoara/Herculane, 1980), Operator Theory: Advances and Applications, vol. 2, Birkhยจ auser, Basel, 1981, pp. 121โ170. , Partially speci๏ฌed matrices and operators: classi๏ฌcation, completion, appli[26] cations, OT vol. 79, Birkhยจ auser Verlag, Basel, 1995.
238
A. Brudnyi, L. Rodman and I.M. Spitkovsky
[27] I. Gohberg and M.G. Krein, Systems of integral equations on a half-line with kernel depending upon the di๏ฌerence of the arguments, Uspekhi Mat. Nauk 13 (1958), no. 2, 3โ72 (in Russian), English translation: Amer. Math. Soc. Transl. 14 (1960), no. 2, 217โ287. [28] I. Gohberg and N. Krupnik, One-dimensional linear singular integral equations. Introduction, OT 53, 54, vol. 1 and 2, Birkhยจ auser Verlag, Basel and Boston, 1992. [29] I.C. Gohberg and I.A. Feldman, Integro-di๏ฌerence Wiener-Hopf equations, Acta Sci. Math. Szeged 30 (1969), no. 3โ4, 199โ224 (in Russian). [30] , Convolution equations and projection methods for their solution, Nauka, Moscow, 1971 (in Russian), English translation Amer. Math. Soc. Transl. of Math. Monographs 41, Providence, R.I. 1974. [31] Yu.I. Karlovich and I.M. Spitkovsky, Factorization of almost periodic matrix functions and (semi) Fredholmness of some convolution type equations, No. 4421-85 dep., VINITI, Moscow, 1985, in Russian. [32] , On the theory of systems of equations of convolution type with semi-almostperiodic symbols in spaces of Bessel potentials, Soviet Math. Dokl. 33 (1986), 145โ149. [33] , Factorization of almost periodic matrix functions, J. Math. Anal. Appl. 193 (1995), 209โ232. [34] G.S. Litvinchuk and I.M. Spitkovsky, Factorization of measurable matrix functions, OT vol. 25, Birkhยจ auser Verlag, Basel and Boston, 1987. [35] C.V.M. van der Mee, L. Rodman, and I.M. Spitkovsky, Factorization of block triangular matrix functions with o๏ฌ diagonal binomials, Operator Theory: Advances and Applications 160 (2005), 423โ437. [36] C.V.M. van der Mee, L. Rodman, I.M. Spitkovsky, and H.J. Woerdeman, Factorization of block triangular matrix functions in Wiener algebras on ordered abelian groups, Operator Theory: Advances and Applications 149 (2004), 441โ465. [37] B.V. Palโcev, Convolution equations on a ๏ฌnite interval for a class of symbols having power asymptotics at in๏ฌnity, Izv. Akad. Nauk SSSR. Mat. 44 (1980), 322โ394 (in Russian), English translation: Math. USSR Izv. 16 (1981). [38] , A generalization of the Wiener-Hopf method for convolution equations on a ๏ฌnite interval with symbols having power asymptotics at in๏ฌnity, Mat. Sb. 113 (155) (1980), 355โ399 (in Russian), English translation: Math. USSR Sb. 41 (1982). [39] A.R. Pears, Dimension theory of general spaces, Cambridge University Press, Cambridge, England, 1975. [40] A. Rastogi, L. Rodman, and I.M. Spitkovsky, Almost periodic factorization of 2 ร 2 matrix functions: New cases of o๏ฌ diagonal spectrum, Recent Advances and New Directions in Applied and Pure Operator Theory (Williamsburg, 2008) (J.A. Ball, V. Bolotnikov, J.W. Helton, L. Rodman, and I.M. Spitkovsky, eds.), Operator Theory: Advances and Applications, vol. 202, Birkhยจ auser, Basel, 2010, pp. 469โ487. [41] G. Robertson, On the density of the invertible group in ๐ถ โ -algebras, Proc. Edinburgh Math. Soc. (2) 20 (1976), no. 2, 153โ157. [42] L. Rodman and I.M. Spitkovsky, Factorization of matrix functions with subgroup supported Fourier coe๏ฌcients, J. Math. Anal. Appl. 323 (2006), 604โ613. [43] W. Rudin, Fourier analysis on groups, John Wiley & Sons Inc., New York, 1990, Reprint of the 1962 original, a Wiley-Interscience Publication.
Factorization Versus Invertibility
239
[44] I.M. Spitkovsky, Factorization of several classes of semi-almost periodic matrix functions and applications to systems of convolution equations, Izvestiya VUZ., Mat. (1983), no. 4, 88โ94 (in Russian), English translation in Soviet Math. โ Iz. VUZ 27 (1983), 383โ388. Alex Brudnyi Department of Mathematics University of Calgary 2500 University Dr. NW Calgary, Alberta, Canada T2N 1N4 e-mail:
[email protected] Leiba Rodman and Ilya M. Spitkovsky Department of Mathematics College of William and Mary Williamsburg, VA 23187-8795, USA e-mail:
[email protected] [email protected]
Operator Theory: Advances and Applications, Vol. 218, 241โ268 c 2012 Springer Basel AG โ
Banded Matrices, Banded Inverses and Polynomial Representations for Semi-separable Operators Patrick Dewilde In fond memory of Israel Gohberg, towering mathematician and engaging friend
Abstract. The paper starts out with exploring properties of the URV factorization in the case of banded matrices or operators with banded inverse, showing that they result in factors with the same properties. Then it gives a derivation of representations for general semi-separable operators (matrices) as ratios of minimally banded matrices. It shows that under pretty general technical conditions (uniform reachability and/or controllability in ๏ฌnite time), left and right polynomial factorizations exist that are unique (canonical) when the factors are properly restrained. Next, it provides Bezout relations for these factors, explicit formulas for all the terms in these relations and an introduction to potential new applications such as Lยจ owner type interpolation theory for (general) matrices. Mathematics Subject Classi๏ฌcation (2000). 15A09, 15A21, 15A23, 15A60, 65F05, 65F20, 93B10, 93B20, 93B28, 93B50, 93B55. Keywords. Semi-separable systems, quasi-separable systems, URV decomposition, canonical polynomial forms, Bezout equations, Loewner interpolation, time varying dynamical systems.
1. Introduction: Semi-separable systems and the โone passโ URV method Semi-separable matrices were introduced in a famous paper by Israel Gohberg and two co-authors, Thomas Kailath and Israel Koltracht [15]. My contribution to the present โGohberg Memorial Issueโ is in honor, not only of Israel Gohberg, who has been a formidable leader in the development of mathematics in general and of applied operator theory and linear algebra in particular, but also of Israel Koltracht, who passed away in 2008, provided major ideas and had a strong in๏ฌuence on the
242
P. Dewilde
๏ฌeld. The idea of semi-separable systems, on which the present paper is based, has proved to be extremely valuable as a frame that provides the right kind of generality to treat problems in dynamical system theory, estimation theory and even just matrix algebra (the original work goes back to papers on Fredholm resolvents and the analysis of Gaussian Processes by Kailath [16] and Anderson and Kailath [17]; in recent times, this type of system has sometimes been called โquasi-separableโ, but I prefer to use the original terminology and see no need in introducing a new, confusing notion). Although the historical origin of semi-separability is in integral kernel theory, I restrict myself to the matrix algebra case, a case that was already contemplated in the paper just cited, be it in a restricted setting that does not allow to obtain the strongest possible results. Semi-separable theory kept interesting Israel Gohberg, in particular after the connections with time varying system theory were fully and independently developed, see [9]. In particular the inversion of linear system of equations with low numerical complexity (actually linear in the size of the matrix) got a new impetus when this connection was established, leading to a ๏ฌurry of new algorithms and papers in which various aspects of the theory were explored [13, 8, 14]. Recently, increased interest in banded operators with banded inverses was generated by Gilbert Strang [21]. The present paper is intended to show the connection between system theory and the theory of banded matrices, very much in the spirit originally set by Israel Gohberg and his co-workers. The approach I follow in this paper is what could be called โstructuralโ. The connection between computational or time discrete systems and their linear algebra stands central, operator theoretic arguments are relegated to the background, not because they are unimportant, but because the results presented are of a computational or system theoretical nature. There is a general, operator theoretic framework in which generalizations would properly ๏ฌt, namely the theory of Nest Algebras, originally proposed by Arveson [6]. I shall use a more limited framework, described in the next paragraph, that allows for a comfortable handling of block matrices of various sizes and the connection with discrete time system theory, actually the same framework of the book [9], with some variation to accommodate common practice in matrix algebra. In particular, in contrast to the book, causal matrices are (block) lower diagonal in this paper and numerical vectors are usually column vectors. To work comfortably with semi-separable systems, we need the use of sequences of indices and then indexed sequences of vectors. When โณ = [๐๐ ]โ ๐=โโ is a sequence of indices, then each ๐๐ is either a positive integer or zero, and a corresponding indexed sequence [๐ข๐ ] โ โโณ 2 will be a sequence of vectors such that each ๐ข๐ has dimension ๐๐ and the overall sum โ โ
โฅ๐ข๐ โฅ2
(1)
๐=โโ
is ๏ฌnite, the square root of which is then the quadratic norm of the sequence. When ๐๐ = 0, the corresponding entry just disappears (it is indicated as a mere โplace
Banded Matrices, Banded Inverses, Polynomial Representations
243
holderโ). A regular ๐-dimensional ๏ฌnite vector can so be considered as embedded in an in๏ฌnite sequence, whereby the entries from โโ to zero and ๐ + 1 to โ disappear, leaving just ๐ entries indexed by 1 โ
โ
โ
๐, corresponding, e.g., to the time points where they are being inputed into the system. On such sequences we may de๏ฌne a generic shift operator ๐, which does nothing else than shifting the position of the data in a column vector (the index) one notch forward, corresponding to the operation of a matrix whose ๏ฌrst subdiagonal is a block diagonal of unit matrices (๐๐,๐โ1 = ๐ผ, all other ๐๐,๐ = 0). It is also convenient to underline the zeroth element of a vector or the {0, 0}th element of a block matrix for orientation purposes. The shift ๐ has a transpose, indicated as ๐ โฒ , which is actually also its inverse (we write ๐ โโฒ = ๐). We use the prime to indicate transposition in general, in real arithmetic it corresponds to the usual transpose, in complex arithmetic to the Hermitian conjugate transpose. Hence (underlining the zeroth term in the series): [. . . , ๐ขโฒโ2 , ๐ขโฒโ1 , ๐ขโฒ0 , ๐ขโฒ1 , ๐ขโฒ2 , . . . ]๐ โฒ = [. . . , ๐ขโฒโ2 .๐ขโฒโ1 , ๐ขโฒ0 , ๐ขโฒ1 , . . . ]
(2)
๐ โฒ is hence a unitary shift represented as a strictly block upper unit matrix. Typically, a numerical analyst would handle only ๏ฌnite sequences of vectors, but the embedding in in๏ฌnite ones allows one to apply delays as desired and not worry about the precise index points. Similarly, we handle in this paper matrices in which the entries are matrices themselves. For example, ๐๐,๐ is a block of dimensions ๐๐ ร ๐๐ with [๐๐ ] = โณ and [๐๐ ] = ๐ฉ , and, again, indices with no entry are just placeholders, with the corresponding block entries disappearing โ also consisting just of place holders (interestingly, MATLAB now allows for such matrices, the lack of which was a major annoyance in previous versions. Place holders are very common in computer science, here they prove useful also in linear algebra). To complete the matrix algebra for this extension, only one extra rule is needed, namely that the product of an ๐ ร 0 matrix with a 0 ร ๐ matrix is a zero matrix of dimensions ๐ ร ๐. Block matrices usually represent maps from an indexed input sequence [๐ข๐ ] to an indexed output sequence [๐ฆ๐ ]. To de๏ฌne a semi-separable system, we need a more re๏ฌned structure, which we now introduce. We de๏ฌne a causal system (of computation) by a set of equations { ๐ฅ๐+1 = ๐ด๐ ๐ฅ๐ + ๐ต๐ ๐ข๐ (3) ๐ฆ๐ = ๐ถ๐ ๐ฅ๐ + ๐ท๐ ๐ข๐ in which we have introduced an intermediate (hidden) state sequence [๐ฅ๐ ], which is recursively computed (and acts as the memory of the computation), and matrices ๐ด๐ , ๐ต๐ , ๐ถ [ ] ๐ , ๐ท๐ at each index point ๐ representing the local linear computation. ๐ด๐ ๐ต๐ is called the system transition matrix at time point ๐ (๐ด๐ being called ๐ถ๐ ๐ท๐ the state transition matrix). What is the corresponding input/output matrix ๐ ? To obtain it, I follow the tradition in classical system theory, replace the local equations above with global equations on the (embedded) sequences ๐ข = [๐ข๐ ], ๐ฆ = [๐ฆ๐ ] and ๐ฅ = [๐ฅ๐ ], de๏ฌne โglobalโ block diagonal matrices ๐ด = diag(๐ด๐ ), ๐ต = diag(๐ต๐ ),
244 etc. and obtain
P. Dewilde {
๐ โฒ ๐ฅ = ๐ด๐ฅ + ๐ต๐ข ๐ฆ = ๐ถ๐ฅ + ๐ท๐ข and, after eliminating the state, the input-output matrix ๐ = ๐ท + ๐ถ๐(๐ผ โ ๐ด๐)โ1 ๐ต
(4) (5)
where I have assumed the inverse to exist. Hence, it must be given precise meaning. One way to do this is, is to assume that the spectral radius of ๐ด๐, ๐(๐ด๐) < 1, which is consistent with the boundedness of the operator. ๐ then represents a bounded, block lower matrix in semi-separable form. Another way would be to assume โone-sided expansionsโ, but this is a method that I do not pursue further in this paper, although it may have merit on its own, as I am mainly interested in stable numerics. A block upper matrix would have a similar representation, now with ๐ โฒ replacing ๐: ๐ = ๐ท + ๐ถ๐ โฒ (๐ผ โ ๐ด๐ โฒ )โ1 ๐ต. (6) For ease of reference, I indicate the transition matrix of an operator ๐ (whether causal or anti-causal) with the symbol โโโ as in [ ] ๐ด ๐ต ๐ โ . (7) ๐ถ ๐ท Such representations, often called realizations, produce in a nutshell the special structure of a semi-separable system. When, e.g., ๐ is block banded lower with two bands, then ๐ด = 0 and ๐ต = ๐ผ will do, the central band is represented by ๐ท[ and the ] choose ] three band, [ one can ] ๏ฌrst o๏ฌ band by ๐ถ. With [a block ] [ ๐ ๐ผ 0 0 because , with ๐ := , ๐ถ = ๐ถ1 ๐ถ2 and ๐ต = ๐ด= ๐[ 0 ๐ผ 0 ] ๐ 0 โ1 the state splits in two components. We ๏ฌnd, indeed, ๐(๐ผ โ ๐ด๐) := , ๐2 ๐ and hence ๐ = ๐ท + ๐ถ1 ๐ + ๐ถ2 ๐ 2 . This principle can easily be extended to yield representations for multi-banded matrices or matrix polynomials in ๐. State space realizations are not unique. The dimension chosen for ๐ฅ๐ at time point ๐ may be larger than necessary, in which case one calls the representation โnon minimalโ โ I shall not consider this case further. Assuming a minimal representation, one could also introduce a non singular state transformation ๐
๐ on the state at each time point, de๏ฌning a new state representation ๐ฅ ห๐ = ๐
๐โ1๐ฅ๐ . The transformed system transition matrix now becomes [ ] [ โ1 ] โ1 ห๐ ๐ดห๐ ๐ต ๐ต๐ ๐
๐+1 ๐ด๐ ๐
๐ ๐
๐+1 := (8) ๐ถ๐ ๐
๐ ๐ท๐ ๐ถห๐ ๐ท๐ for a lower system, and a similar, dual representation for an upper. Given a block lower matrix ๐ , what is a minimal semi-separable representation for it? This problem is known as the system realization problem, and was solved originally by Kronecker [18] in the context of rational functions, and then later by various authors in various circumstances, for the semi-separable case, see
Banded Matrices, Banded Inverses, Polynomial Representations
245
[9] for a complete treatment. An essential role in realization theory is played by the so-called ๐th Hankel operator ๐ป๐ de๏ฌned as โค โก ๐๐,๐โ1 โ
โ
โ
๐๐,๐โ2 โฅ โข (9) ๐ป๐ = โฃ โ
โ
โ
๐๐+1,๐โ2 ๐๐+1,๐โ1 โฆ , . . .. .. .. . i.e., a left lower corner matrix just West of the diagonal element ๐๐,๐ . It turns out that any minimal factorization of each ๐ป๐ yields a minimal realization [9], we have indeed โค โก ๐ถ๐ โฅ[ โข ๐ถ๐+1 ๐ด๐ ] โฅ โข ๐ป๐ = โข ๐ถ๐+2 ๐ด๐+1 ๐ด๐ โฅ โ
โ
โ
๐ด๐โ1 ๐ด๐โ2 ๐ต๐โ3 ๐ด๐โ1 ๐ต๐โ2 ๐ต๐โ1 (10) โฆ โฃ .. . where, as I explained before, entries may disappear when they reach the border of the matrix. This decomposition has an attractive physical meaning. We recognize โก โค ๐ถ๐ โข โฅ ๐ถ๐+1 ๐ด๐ โข โฅ ๐ช๐ = โข ๐ถ๐+2 ๐ด๐+1 ๐ด๐ โฅ (11) โฃ โฆ .. . as the ๐th observability operator, and [ โ๐ = โ
โ
โ
๐ด๐โ1 ๐ด๐โ2 ๐ต๐โ3
๐ด๐โ1 ๐ต๐โ2
๐ต๐โ1
]
(12)
as the ๐th reachability operator โ all these related to the (causal) lower operator we assumed. At any index point ๐, โ๐ maps inputs strictly before the time point ๐ to the state ๐ฅ๐ , while ๐ช๐ maps the state ๐ฅ๐ to the output at the present index point ๐ and outputs after that, giving its linear contribution to them. The rows of โ๐ form a basis for the rows of ๐ป๐ , while the columns of ๐ช๐ form a basis for the columns of ๐ป๐ in a minimal representation. When, e.g., the rows are chosen as an orthonormal basis for all the ๐ป๐ , then a realization will result for which ๐ด๐ ๐ดโฒ๐ + ๐ต๐ ๐ต๐โฒ = ๐ผ for all [ ] ๐. We call a realization in which ๐ด๐ ๐ต๐ has this property of being part of an orthogonal or unitary matrix, in input normal form. Dually, a realization is said to be in output normal form if for each index ๐, ๐ช๐โฒ ๐ช๐ = ๐ผ. A general matrix ๐ is in semi-separable form, when both the lower and upper parts have (in general di๏ฌerent) system realizations (all matrices shown are block diagonal and consisting typically of blocks of low dimensions): ๐ = ๐ถโ ๐(๐ผ โ ๐ดโ ๐)โ1 ๐ตโ + ๐ท + ๐ถ๐ข ๐ โฒ (๐ผ โ ๐ด๐ข ๐ โฒ )โ1 ๐ต๐ข .
(13)
In typical applications, all these matrices have low dimensions. Their value is that systems with semi-separable realizations can be inverted with a much lower order of numerical complexity than for the classical case of matrix inversion. I shall illustrate this principle soon.
246
P. Dewilde
It may seem laborious to ๏ฌnd realizations for common systems of equations like discretized partial di๏ฌerential equations or integral equations. Luckily, this is not the case. In many instances, realizations come with the physics of the problem. Very common are, besides block banded matrices, so-called smooth matrices [20], in which the Hankel matrices have natural low-rank approximations, and ratios of block banded matrices (which are in general full matrices), and, of course, systems derived from linearly coupled subsystems. The URV factorization The goal of an URV factorization is to represent (the block matrix) ๐ as a product of three (block) matrices, ๐ , ๐
, and ๐ , ๐ being isometric, ๐ co-isometric and ๐
upper and upper invertible. My goal in this section is to give the details of a method that computes the factorization in a numerically stable way directly on the semiseparable representation and in a โone passโ way, recursively computing the result for increasing indices. When ๐ = ๐ ๐
๐ and ๐ is invertible, then ๐ and ๐ will be plainly unitary, and ๐ โ1 = ๐ โฒ ๐
โ1 ๐ โฒ . However, when ๐ is general, then ๐ and ๐ are merely isometric, resp. co-isometric, and the solution of the least squares least squares solution for ๐ฆ = ๐ข๐ is given by ๐ข = ๐ โ ๐ฆ with ๐ โ = ๐ โฒ ๐
โ1 ๐ โฒ (the same would be true for ๐ฆ = ๐ข๐ , now with ๐ข = ๐ฆ๐ โฒ ๐
โ1 ๐ โฒ !) ๐ โ is called the โMoorePenrose inverseโ of ๐ . The solution to the ๐ ๐
๐ factorization problem in terms of system representations was originally given in [22], and was further elaborated in [9]. In [20] the factorization as a one pass recursive method was given. Remarkably, each of the factors has itself a simple semi-separable representation in terms of the original representation and of a complexity (as measured in the dimension of the intermediate state) that is at most equal to the original. The URV recursion starts with orthogonal operations on (block) columns, transforming ๏ฌrst the mixed lower-upper matrix ๐ to the upper form and then proceeding on an upper matrix โ in practice, one actually alternates (block) column operations that make the matrix upper with (block) row operations that reduce the upper form, to achieve the one pass solution. However, the block column operations turn out to be fully independent from the row operations, hence we can treat them ๏ฌrst and then complete with row operations (although in numerical practice [20] the operations are staggered). The (๏ฌrst) column phase of the URV factorization consists in getting rid of the lower or causal part in ๐ by post-multiplication with a unitary matrix, working on the semi-separable representation instead of on the original data. If one takes the lower part in input normal form, i.e., ๐ถหโ ๐(๐ผ โ หโ ๐ต หโ = ๐ถโ ๐(๐ผ โ ๐ดโ ๐)โ1 ๐ตโ such that ๐ดหโ ๐ดหโฒ + ๐ต ห โฒ = ๐ผ, then the realization ๐ดหโ ๐)โ1 ๐ต โ โ for (upper) ๐ is given by [ ] หโ ๐ดหโ ๐ต ๐ โ (14) ๐ถ๐ ๐ท๐ [ ] หโ where ๐ถ๐ and ๐ท๐ are formed by unitary completion of the co-isometric ๐ดหโ ๐ต (for an approach familiar to numerical analysts see [20]). ๐ โฒ is[ a minimal ] anti causal unitary operator, which pushes ๐ to upper from the right: ๐๐ข 0 := ๐ ๐ โฒ can
Banded Matrices, Banded Inverses, Polynomial Representations
247
be checked to be upper and a realization for ๐๐ข follows from the preceding as โก โค ๐ดหโฒโ 0 ๐ถ๐โฒ หโฒ โฆ. ๐๐ข โ โฃ (15) ๐ต๐ข ๐ต ๐ด๐ข ๐ต๐ข ๐ท๐โฒ โ โฒ โฒ โฒ โฒ ห ห ห ห ห ๐ถโ ๐ดโ + ๐ท ๐ตโ ๐ถ๐ข ๐ถโ ๐ถ๐ + ๐ท๐ท๐ As expected, the new transition matrix combines lower and upper parts and has become larger, but ๐๐ข is now (block) upper. Numerically, this step is executed as an LQ factorization as follows (for an introduction to QR and LQ factorizations, see the appendix). Let ๐ฅ๐ = ๐
๐ ๐ฅ ห๐ and let us assume we know ๐
๐ at step ๐, then [ ] ๐ดโ,๐ ๐
๐ ๐ตโ,๐ ๐ถโ,๐ ๐
๐ ๐ท๐ [ ][ ] (16) หโ,๐ ๐
๐+1 0 0 ๐ดหโ,๐ ๐ต = โฒ โฒ หโฒ ๐ถหโ,๐ ๐ดหโฒโ,๐ + ๐ท๐ ๐ต ๐ถหโ,๐ ๐ถห๐,๐ + ๐ท๐ ๐ท๐,๐ 0 ๐ถ๐,๐ ๐ท๐,๐ โ,๐ The LQ factorization of the left-handed matrix computes all the data of the righthand side, namely the transformation matrix, the data for the upper factor ๐๐ข and the new state transition matrix ๐
๐+1 , allowing the recursion to move on to the next index point. Because we have not assumed ๐ to be invertible, we have to allow for an LQ factorization that produces an echelon form rather than a strictly square lower triangular form, and allows for a kernel as well, represented by a block column of zeros. The next step is what is called an inner/outer factorization on the upper operator ๐๐ข to reduce it to an upper and upper invertible operator ๐๐ and an upper orthogonal operator ๐ such that ๐๐ข = ๐ ๐๐ . The idea is to ๏ฌnd an as large as possible upper and orthogonal operator ๐ such that ๐ โฒ ๐๐ข is still upper โ ๐ โฒ tries to push ๐๐ข back to lower, without destroying its โuppernessโ. When it does so maximally, an upper and upper invertible factor ๐๐ should result. There is a di๏ฌculty here that ๐๐ข might not be invertible to start with. This di๏ฌculty is not hard to surmount for the factorization to go through, but in order to avoid a too technical discussion, I start out by assuming invertibility and then remark that the procedure automatically produces the general formula needed. If the entries of ๐๐ข would be scalar, then I would already have reached the goal. Indeed, the inverse of ๐๐ข might have a lower part, which is to be captured by the inner operator ๐ that we shall now determine. When ๐๐ข = ๐ ๐๐ with ๐ upper and orthogonal, then also ๐๐ = ๐ โฒ ๐๐ข . Writing out the factorization in terms of the realization, and rede๏ฌning for brevity ๐๐ข := ๐ท + ๐ถ๐ โฒ (๐ผ โ ๐ด๐ โฒ )โ1 ๐ต we obtain [ โฒ ][ ] ๐๐ = ๐ท๐ + ๐ต๐โฒ (๐ผ โ ๐๐ดโฒ๐ )โ1 ๐๐ถ๐โฒ ๐ท + ๐ถ๐ โฒ (๐ผ โ ๐ด๐ โฒ )โ1 ๐ต โฒ โฒ = ๐ท๐ ๐ท + ๐ต๐โฒ (๐ผ โ ๐๐ดโฒ๐ )โ1 ๐๐ถ๐โฒ ๐ท + ๐ท๐ ๐ถ๐ โฒ (๐ผ โ ๐ด๐ โฒ )โ1 ๐ต (17) โฒ โฒ โ1 โฒ โฒ โฒ โ1 +๐ต๐ {(๐ผ โ ๐๐ด๐ ) ๐๐ถ๐ ๐ถ๐ (๐ผ โ ๐ด๐ ) }๐ต. This expression has the form: โdiagonal termโ + โstrictly lower termโ + โstrictly upper termโ + โmixed productโ. The last term has what is called โdichotomyโ, what
248
P. Dewilde
stands between {โ
} can again be split in three terms: (๐ผ โ ๐๐ดโฒ๐ )โ1 ๐๐ถ๐โฒ ๐ถ๐ โฒ (๐ผ โ ๐ด๐ โฒ )โ1 = (๐ผ โ ๐๐ดโฒ๐ )โ1 ๐๐ดโฒ๐ ๐ + ๐ + ๐ ๐ด๐ โฒ (๐ผ โ ๐ด๐ โฒ )โ1
(18)
with ๐ satisfying the โLyapunov-Stein equationโ ๐ โฒ ๐ ๐ = ๐ถ๐โฒ ๐ถ + ๐ดโฒ๐ ๐ ๐ด
(19)
โฒ ๐ถ๐ + ๐ดโฒ๐,๐ ๐๐ ๐ด๐ . The resulting strictly lower term or, with indices: ๐๐+1 = ๐ถ๐,๐ has to be annihilated, hence we should require ๐ถ๐โฒ ๐ท + ๐ดโฒ๐ ๐ ๐ต = 0, in fact ๐ should be chosen maximal with respect to this property (beware: Y depends on U!) Once these two equations are satis๏ฌed, the realization for ๐๐ results as ๐๐ = โฒ โฒ (๐ท๐ ๐ท + ๐ต๐โฒ ๐ ๐ต) + (๐ท๐ ๐ถ + ๐ต๐โฒ ๐ ๐ด)๐ โฒ (๐ผ โ ๐ด๐ โฒ )โ1 ๐ต โ we see that ๐๐ inherits ๐ด and ๐ต from ๐ and gets new values for the other constituents ๐ถ๐ and ๐ท๐ . Putting these operations together in one matrix equation and in a somewhat special order, we obtain ] ][ [ ] [ ๐ถ๐ ๐ท๐ ๐๐ต ๐๐ด ๐ต๐ ๐ด๐ . (20) = ๐ท๐ ๐ถ๐ 0 ๐ โฒ๐ ๐ ๐ท ๐ถ
Let us interpret this result without going into motivating theory (as in done in [9, 20]). We have a (block) QR factorization of the left-hand side. At stage ๐ one[ must assume knowledge of ๐๐ , and then perform a normal QR factorization ] ๐๐ ๐ต๐ ๐๐ ๐ด๐ of . ๐ท๐,๐ will be an invertible, upper triangular matrix, so its ๐ท๐ ๐ถ๐ dimensions are ๏ฌxed by the row dimension of ๐๐ . The remainder of the factorization produces ๐ถ๐,๐ and ๐๐+1 , and, of course, the โQ factorโ that gives a complete realization of ๐๐ . What if ๐ is actually singular? It turns out that then the QR factorization will produce just an upper staircase form with a number of zero rows. The precise result is โก โค ] [ ] ๐ท [ ๐ถ๐,๐ ๐ต๐,๐ ๐ด๐,๐ ๐ต๐,๐ โฃ ๐,๐ ๐๐ ๐ต๐ ๐๐ ๐ด๐ 0 ๐๐+1 โฆ , = (21) ๐ท๐ ๐ถ๐ ๐ท๐,๐ ๐ถ๐,๐ ๐ท๐,๐ 0 0 in which the extra columns represented by ๐ต๐ and ๐ท๐ de๏ฌne an isometric operator ๐ = ๐ท๐ + ๐ถ๐ ๐ โฒ (๐ผ โ ๐ด๐ ๐ โฒ )โ1 ๐ต๐ so that [ ] [ ] ๐๐ ๐ ๐ (22) ๐๐ข = 0 and ๐ characterizes the row kernel of ๐ . Another situation (of importance for the Lยจ owner interpolation theory treated in the last section of this paper) is when ๐ is right-outer (i.e., causal with causal right inverse). In that case ๐ should be empty for all index points and at each such point one then has the simpli๏ฌed QR factorization [ ] [ ] ๐ท ๐ถ = ๐ท๐ ๐ท๐ ๐ถ๐ . (23) Actually, one can then just choose ๐ท๐ = ๐ผ and nothing changes โ but the occurrence has to be tested of course. Whether this happens, is dependent on the past
Banded Matrices, Banded Inverses, Polynomial Representations
249
of ๐ , as we have to know ๐๐ at each step ๐. If the support for ๐ is only half in๏ฌnite (say with indices running ] from 1 on), it will be necessary and su๏ฌcient that all [ subsequent ๐ท๐ ๐ถ๐ have full row rank. Remarkably, the operations work on the rows of ๐๐ข in ascending index order, just as the earlier factorization worked in ascending index order on the columns. That means that the URV algorithm can be executed completely in ascending index order. The reader may wonder at this point (1) how to start the recursion and (2) whether the proposed recursive algorithm is numerically stable. On the ๏ฌrst point and with our convention of empty matrices, there is no problem starting out at the upper left[ corner of the ] matrix, both ๐ด1 and ๐0 are just empty, the ๏ฌrst QR is done on ๐ท1 ๐ถ1 . In case the original system does not start at some ๏ฌnite index, but has a system part that runs from โโ onwards, one must introduce knowledge of some initial condition on ๐ . This is provided, e.g., by an analysis of the LTI system running from โโ to 0 if that is indeed the case, see [10] for more details. On the matter of numerical stability, I o๏ฌer two remarks. First, propagating ๐๐ is numerically stable, one can show that a perturbation on any ๐๐ will die out exponentially if the propagating system is assumed exponentially stable. Second, one can show that the transition matrix ฮ of the inverse of the outer part will be exponentially stable as well, when certain conditions on the original system are satis๏ฌed [9], p. 367. Banded matrices with banded inverse A banded lower matrix will have a minimal semi-separable realization for which the transition operator ๐ด is such that ๐ด๐ is nilpotent (the โdegreeโ of nilpotency, which may be variable, determines the size of the band). Clearly, when ๐ด๐ is nilpotent, then so is ๐๐ด. Dually, an upper matrix with transition operator ๐ด shall be banded when ๐ด๐ โฒ or equivalently, ๐ โฒ ๐ด is nilpotent. Suppose ๐ is upper and upper invertible (๐ is outer) and banded, then an interesting question arises whether ๐ โ1 can be banded as well. If ๐ = ๐ท + ๐ถ๐ โฒ (๐ผ โ ๐ด๐ โฒ )โ1 ๐ต is a minimal upper realization for ๐ , then a minimal upper realization for ๐ โ1 is given by ๐ โ1 = ๐ทโ1 โ ๐ทโ1 ๐ถ๐ โฒ (๐ผ โ ฮ๐ โฒ )โ1 ๐ต๐ทโ1 , in which the transition matrix ฮ = ๐ด โ ๐ต๐ท โ1 ๐ถ. Typically, ฮ๐ โฒ will not be nilpotent when ๐ด is, but it can actually be, notably when ๐ต๐ทโ1 ๐ถ = 0. I call this case, in which the inverse has the same band as the original, a โstrictly banded inverseโ. It appears in major applications such as โlapped transformsโ, โHaar transformsโ and โwavelet representationsโ [21]. It is of course possible that the inverse is banded with a larger band than the original, but I do not know how to treat this more general case. In fact, all ๏ฌnite matrices belong to that more general class, so it does not really make sense for them. Theorem 1. Let ๐ be a double-sided, banded matrix with strictly banded inverse, then the URV factorization is such that the two inner factors ๐ and ๐ and the outer factor ๐๐ are all banded with strictly banded inverse.
250
P. Dewilde
Remarks. โ The factors can be obtained by the one-pass recursive algorithm described earlier and have each system realizations whose state complexity is at most equal to the state realization of the original. โ The notion of โbanded outer with strictly banded inverseโ is a generalization of the classical LTI notion of โunimodularโ. Proof. The theorem follows directly from the construction of the URV factorization given earlier. The factor ๐ is causal and inherits the transition matrix ๐ดโ of the lower part for which ๐ดโ ๐ is nilpotent. ๐ โฒ is automatically nilpotent also, as it has the same transition matrix conjugated. The transition matrix of the resulting upper ๐๐ข is from before [ ] 0 ๐ดหโฒโ ๐ด= (24) ห โฒ ๐ด๐ข . ๐ต๐ข ๐ต โ and is nilpotent, since both ๐ดหโฒโ ๐ โฒ and ๐ด๐ข ๐ โฒ are. The inverse ๐๐ขโ1 = ๐ ๐ โ1 exists by hypothesis and remains banded as a product of two banded matrices. The extraction of ๐ can now again be interpreted as an external factorization ๐๐ขโ1 = ๐๐โ1 ๐ โฒ , which is such that the upper, and necessarily banded ๐ (since ๐๐ขโ1 is banded) annihilates the lower part of ๐๐ขโ1 resulting in an upper ๐๐โ1 , which again has to be banded, and the same will be true for ๐๐ = ๐ โฒ ๐๐ข . More work is needed to show that the total size of the band does not increase by the procedure, but given the band structure of ๐ โฒ and ๐ โฒ , and the fact that ๐ โฒ is upper and ๐ โฒ lower, the stability of the band follows naturally. โก We are now ready to tackle the main topic of this paper: representations of semi-separable operators as ratios of polynomial matrices in the shift operator ๐.
2. Matrix polynomial representations Although there are complete theories for external and inner/outer factorizations (as somewhat described in the introduction), the polynomial representation the๐ฉ ory for general matrices or operators (viewed as maps โโณ 2 โ โ2 ) generalizing the complex matrix function theory to semi-separable matrices or time-varying systems has been elusive (a ๏ฌrst attempt can be found in [5], limited by the special problem treated in that paper). I tried to generalize the famous Popov construction to the semi-separable setting, but was unable to do so. When I encountered the paper of Paul Van Dooren on dead beat control [12], I stumbled on a feasible and attractive technique, which I am now presenting. Let ๐ be a causal (lower) operator (I shall assume ๐ to be bounded, although generalizations can be constructed). The goal is to ๏ฌnd minimal representations for ๐ of the type ๐ = ฮโ ๐ โ1 or ๐ = ๐โโฒ ฮโฒ๐ , in which ฮโ , ฮ๐ , ๐ , ๐ are all polynomials in ๐ of minimal degree. I shall show that under very mild conditions such representations do indeed exist and how they can be computed.
Banded Matrices, Banded Inverses, Polynomial Representations
251
Preliminaries The gist of the method that I shall present is the (recursive) calculation of preimages. To do this comfortably (as I shall have to modify bases recursively), I make a distinction between matrices and โabstractโ operators, the latter being basis free. I write abstract vectors in boldface or in Greek characters, while their concrete representation in a given basis is in normal font. โ๐So, when x is an abstract vector in a space with basis [๐๐ ]๐=1โ
โ
โ
๐ , we have x = ๐=1 ๐๐ ๐ฅ๐ with ๐ฅ๐ the components of x in the given basis. Following tradition of Di๏ฌerential Geometry (or we can just as well assemble the ๐๐ in a (row) vector stack [Quantum Mechanics), ] ๐1 โ
โ
โ
๐๐ and write x = ๐๐ฅ, in which now ๐ฅ is a column vector assembling the components of x in the basis ๐ โ and such a notation can accommodate any indexing scheme, of course. Suppose ๐ : ๐ณ โ ๐ด : x โ y = ๐x, if we have a basis stack ๐ in ๐ด, y = ๐๐ฆ and another ๐ in ๐ณ , then there is a matrix ๐ด so that ๐ฆ = ๐ด๐ฅ, because then y = ๐๐ฆ = ๐x = ๐๐๐ฅ, ๐๐ (assuming there are ๐ base [ vectors numbered ] 1 โ
โ
โ
๐) โ
โ
โ
๐๐ ๐๐ has the formal matrix calculus interpretation ๐๐ = and each 1 ๐ โ of these entries evaluates as ๐๐๐ = ๐ ๐๐ ๐ด๐๐ so that (again using matrix notation) ๐๐ฆ = ๐๐ด๐ฅ
(25)
and as ๐ forms a basis, necessarily ๐ฆ = ๐ด๐ฅ, a purely numerical expression. In the sequel I shall use spaces spanned by vectors that do โ not necessarily have to form a basis, in particular if ๐ is a stack of vectors I write ๐ for the space spanned by the vectors. โSuppose now that ๐ : โฌ โ ๐ด and ๐ : ๐ฐ โ ๐ด are operators to a same space ๐ด, let ๐0 de๏ฌne a subspace of โฌ and u a (row) stack of vectors in ๐ฐ (e.g., natural โ basis vectors), how do we know that the (๐๐ ) โ0 lies in the image of uโunder ๐ต or, more generally, what is the full pre-image of (๐u) under ๐? I claim: (๐๐0 ) lies in โ (๐u) i๏ฌ there exists a matrix ๐น such that ๐๐0 = ๐u๐น . The signi๏ฌcance of this is that the โinputโ โu๐น ๐ฅ is capable of annihilating ๐๐0 ๐ฅ (in control applications this is called a feedback loop). The existence de๏ฌnition โ of ๐น follows from the following โ of its entries: since each entry ๐๐0,๐ โ ๐u we can express ๐๐0,๐ = ๐ ๐u๐ ๐น๐,๐ โ. The next question is: ๏ฌnd a basis for the full subspace ๐ฎ โ โฌ that maps to (๐u), 0 โ โ i.e., for the pre-image of (u๐) under ๐, sometimes denoted ๐โ1 ( ๐u). I present algorithms to compute ๐น in the appendix. Dead beat control and the construction of a polynomial matrix Let us now assume that we are given the ๐ด and ๐ต operators of a causal semiseparable system, and that the pair {๐ด, ๐ต} is such that any state ๐ฅ๐ can be brought to zero in less than some ๏ฌxed ๏ฌnite time ๐ โ i.e., there exist inputs ๐ข๐ โ
โ
โ
๐ข๐+โ , โ โค ๐ that bring the ๐ฅ๐ to zero. Su๏ฌcient for that is that there exists a ๏ฌxed index ๐ such that the partial reachability operator [โ๐ ][๐โ๐:๐โ1] has full range for all ๐ (this means that every state at any time can be reached from zero or controlled to zero by an input sequence of length at most ๐. A weaker necessary and su๏ฌcient condition can
252
P. Dewilde
be formulated but is considerably more involved and hard to check. The su๏ฌcient condition is the common case, and easily satis๏ฌed in the ๏ฌnite-dimensional case). Our goal shall be to ๏ฌnd a diagonal operator ๐น such that (๐ดโ๐ต๐น )๐ is a minimally nilpotent operator. [ If that ]is the case, then the system de๏ฌned by the system ๐ด ๐ต transition matrix will be such that it has a causal, polynomial inverse ๐น ๐ผ ] [ ๐ด โ ๐ต๐น ๐ต . It is easy to check that this is given by the transition matrix โ๐น ๐ผ indeed the inverse system, just by solving the direct system for the input, given the output, and given the assumption that (๐ด โ ๐ต๐น )๐ is nilpotent, the inverse system becomes automatically polynomial in ๐ and hence causal. ๐น is found by the dead beat construction, which attempts to ๏ฌnd a feedback control that brings any state to zero in a minimum number of steps, and which I now introduce. Let us assume that we are standing at point ๐ in the state space recursion, โฌ๐ being the state space at point ๐. We assume (1) that the system is uniformly controllable in at most a ๏ฌxed ๏ฌnite time ๐ and (2) that we already know how to dead beat the state at point (๐ + 1). We materialize the latter assumption by assuming that we dispose of a basis ๐ for โฌ๐+1 which has a decomposition in subspaces ๐ฎ๐+1,0 โ ๐ฎ๐+1,1 โ โ
โ
โ
โ ๐ฎ๐+1,๐๐+1 = โฌ๐+1 , where ๐ฎ๐+1,๐ is de๏ฌned as the subspace of โฌ๐+1 that can be dead beat controlled in at most ๐ steps โ the 0th step being the control in step ๐ + 1. ๐ is a stack of bases ๐0 , ๐1 , . . . , ๐๐๐+1 such โ๐ that ๐ฎ๐+1,๐ = โ=0 ๐โ . For โฌ๐ we have possibly been given an original basis ๐ ๐ , the goal being to ๏ฌnd a dead beat decomposition for it, similar to the one we already have for โฌ๐+1 . Let ๐ด๐ and ๐ต๐ be the matrices in the current bases realizing ๐๐ : โฌ๐ โ โฌ๐+1 , respect. ๐๐ : ๐ฐ๐ โ โฌ๐+1 , ๐ฐ๐ having the basis u๐ . It should be clear that the state at stage ๐ can be dead beat controlled in at most ๐๐+1 + 1 steps, but it might be in less (certainly less than ๐), we denote the maximum number at stage ๐ by ๐๐ (of โcourse). Dropping indices wherever clear, we de๏ฌne ๐ฎ๐ โ โฌ๐ as the ๐ฎ๐ โ โฌ๐ as the pre-image under ๐๐ of ] [ โ pre-image of ๐๐ u๐ , and then recursively, (๐๐ u๐ , ๐ฎ๐+1,๐โ1 ) with ๐ โค ๐๐+1 + 1. As ๐0 ๐1 โ
โ
โ
๐๐๐+1 is the stack of basis vectors conformal to the decomposition [๐ฎ๐ ] of โฌ๐ and because of the pre-image relations just described we shall have [ ] ๐๐ ๐0 ๐1 ๐2 โ
โ
โ
๐๐๐+1 โค โก ๐น๐,0 ๐น๐,1 ๐น๐,2 โ
โ
โ
๐น๐,๐๐+1 ๐บ1,0 ๐บ1,1 โ
โ
โ
๐บ1,๐๐+1 โฅ [ ]โข โฅ (26) โข 0 = ๐๐ u๐ ๐0 ๐1 โ
โ
โ
๐๐๐ โข . โฅ .. .. . . . . . โฆ โฃ . . . . . 0
0
0
โ
โ
โ
๐บ๐๐ ,๐๐+1
for some matrices[๐น๐,๐ and ๐บโ,๐ (for algorithms to compute these matrices, see the ] appendix). ๐น๐ = ๐น๐,0 โ
โ
โ
๐น๐,๐๐+1 is the feedback matrix desired, at step ๐, in the bases just de๏ฌned (the ๐บโฒ ๐ produce a realization of the operator ๐๐ at step ๐ โ see also the appendix for more detail).
Banded Matrices, Banded Inverses, Polynomial Representations
253
2.1. Polynomial representations We start out with a causal (lower) matrix in output normal form ๐ = ๐ท + ๐ถ๐(๐ผ โ ๐ด๐)โ1 ๐ต : ๐ดโฒ ๐ด + ๐ถ โฒ ๐ถ = ๐ผ, and we assume the system to be uniformly strictly stable (๐(๐ด๐) < 1). We then know that the inner ๐ := ๐ท๐ + ๐ถ๐(๐ผ โ ๐ด๐)โ1 ๐ต๐ with observability pair {๐ด, ๐ถ} will be such that ๐ โฒ ๐ is causal (see ]Thm. 6.8 in [ ๐ด ๐ต๐ is unitary. [9]). The diagonal operators ๐ต๐ and ๐ท๐ are such that ๐ถ ๐ท๐ Using the dead beat control construction of the previous paragraphs based on [ ] ๐ด ๐ต๐ the matrices {๐ด, ๐ต๐ } we ๏ฌnd a feedback operator ๐น such that has a ๐น ๐ผ polynomial inverse. This leads to the following theorem: Theorem 2. There exist minimal polynomial operators ๐ and ๐ such that ๐ = ๐๐ โ1 = ๐โโฒ ๐ โฒ and realizations given by [ ] ๐ด ๐ต๐ โ1 โ ๐ ๐น ๐ผ [ ] ๐ด โ ๐ต๐ ๐น ๐ต๐ ๐ โ โ๐น ๐ผ ] [ (27) ๐ด โ ๐ต๐ ๐น ๐ต๐ ๐ โ ๐ถ โ ๐ท๐ ๐น ๐ท๐ [ ] ๐ดโฒ ๐ถโฒ โ1 โ ๐ ๐ต๐โฒ + ๐น ๐ดโฒ ๐ท๐โฒ + ๐น ๐ถ โฒ in which ๐ is polynomial with causal inverse, while ๐ is polynomial with anticausal inverse. Remark: ๐ characterizes what would be considered the โpolesโ of ๐ in a linear time invariant setting! Proof. We de๏ฌne ๐ โ1 by the dead beat construction based on {๐ด, ๐ต๐ }. As indicated there, the inverse ๐ then becomes automatically polynomial in ๐, as seen by direct evaluation of the output in terms of the input. Next, we obtain ๐โ1 from ๐โ1 = ๐ โ1 ๐ โฒ , which, using the property that {๐ด, ๐ต๐ } is in input normal form, evaluates to (๐ท๐โฒ + ๐น ๐ถ โฒ ) + (๐ต๐โฒ + ๐น ๐ดโฒ )(๐ผ โ ๐ โฒ ๐ดโฒ )โ1 ๐ โฒ ๐ถ โฒ and hence the (anticausal) realization given. A realization for ๐ is obtained directly from ๐ = ๐ ๐ by introducing the realizations for ๐ and ๐ . One veri๏ฌes that (๐ผ โ ๐ โฒ ๐ดโฒ )โ1 ๐ โฒ ๐ถ โฒ (๐ถ โ ๐ท๐ ๐น )๐(๐ผ โ ๐ด๐ ๐)โ1 = (๐ผ โ ๐ โฒ ๐ดโฒ )โ1 ๐ โฒ ๐ดโฒ + ๐ผ + ๐ด๐ ๐(๐ผ โ ๐ด๐ ๐)โ1 , so that, indeed, ๐ = ๐ โฒ ๐ and ๐โ1 ๐ = ๐ผ, with the given realization for ๐ as a causal polynomial. โก At this point I wish to introduce the notion of minimal lengths (causal) polynomial inverse based on a reachability pair {๐ด, ๐ต} (and a dual notion for the observability pair). For ease of discussion and without impairing generality, we normalize the instantaneous term to [๐ผ as before. ] Any minimal degree inverse will ๐ด ๐ต have a realization of the form ๐ โ1 โ for some suitable ๐น๐ , i.e., one for ๐น๐ ๐ผ
254
P. Dewilde
which ๐ด โ ๐ต๐น๐ is nilpotent. One such is when ๐น๐ is chosen equal to ๐น . I shall say the polynomial inverse has minimal lengths if ๐น๐ is chosen so that the rank of the nilpotent operator [๐(๐ด โ ๐ต๐น๐ )]๐ is minimal for each ๐. The following theorem, for which I give only a sketchy prove because the recursive proof is very technical, is valid. Theorem 3. All minimal lengths causal polynomials for which ๐ = ๐๐ ๐๐โ1 and for which ๐๐,0 = ๐ผ are of the form [ ] ๐ด ๐ต๐ โ1 (28) ๐๐ โ ๐น๐ ๐ผ with ๐น๐ = ๐น + ๐บ for some commensurable ๐บ in the kernel of ๐ต๐ , i.e., for which ๐ต๐ ๐บ = 0, and ๐น is determined by the dead beat construction. Moreover, ๐๐ = ๐๐ ๐ in which ๐๐ = ๐ผ โ ๐บ๐(๐ผ โ ๐ด๐)โ1 ๐ต๐ is outer with outer inverse ๐๐โ1 = ๐ผ + ๐บ๐(๐ผ โ ๐ด๐)โ1 ๐ต๐ . Proof (sketch). It is easily veri๏ฌed directly that, given ๐น๐ , ๐ = ๐๐ ๐๐โ1 with exactly the same construction for the polynomials ๐๐ and ๐๐ shown earlier in the case of ๐น . Conversely, to show the claimed unicity, let be given a factorization ๐ = ๐๐ ๐๐โ1 , then ๐๐โ1 as the right factor necessarily de๏ฌnes the controllability space of ๐ and since it is supposed to be minimal, its AB-pair can be chosen to be {๐ด, ๐ต๐ } (see the realization theory in, e.g., [9]). Hence, ๐๐โ1 must have a realization as given in the lemma, with some new ๐น๐ . To show then that ๐น๐ = ๐น + ๐บ, with ๐บ such that ๐ต๐ ๐บ = 0, one shows that this follows from the fact that the minimal dimensions of the ranges of ๐(๐ด โ ๐ต๐น๐ ) are precisely the dimensions of the spaces ๐ฎ๐,๐ of the dead beat construction, and the feedback operators realizing these dimensions all have to be of the form ๐น๐,๐ + ๐บ๐,๐ in which ๐บ๐,๐ belongs to the kernel of ๐ต๐ . This is the technical (recursive) part of the proof, which I just merely sketched. โก Next we have the minimal representations for ๐ as the ratio of two polynomials: Theorem 4. Let ๐ be a uniformly exponentially stable causal semi-separable operator, whose minimal realization is uniformly controllable and observable in ๏ฌnite time. Then ๐ has a minimal representation as a ratio of two polynomial operators ๐ = ๐โโฒ ฮโฒ = ๐๐โ1 ฮ๐ , in which ๐ is polynomial in ๐ with anticausal inverse, ฮ and ฮ๐ are polynomial in ๐ and ๐๐ is polynomial in ๐ with causal inverse. Moreover, ๐๐ will be a unique polynomial of minimal length with this property except for an invertible diagonal right factor. Remark. System representations for these polynomial matrices can be found through dead beat constructions based on reachability or observability pairs โ see the proof. Proof. Let ๐ be de๏ฌned as before. ๐ will also be uniformly reachable in ๏ฌnite time when it is uniformly observable in ๏ฌnite time, due to the fact that it has a uniformly
Banded Matrices, Banded Inverses, Polynomial Representations
255
stable unitary realization (the proof is a simple exercise in realization theory, as at each time point the rank of the reachability matrix equals that of the observability matrix). Let ๐ and ๐ be as derived above from ๐ and ๐ . The property โ๐ โฒ ๐ is causalโ translates into โฮ = ๐ โฒ ๐ is causalโ because of the causality properties of ๐ . This we verify directly. Since ๐ is polynomial, ฮ has to be polynomial as well, yielding the state space realization for ฮ (with ๐ด๐ := ๐ด โ ๐ต๐ ๐น ): ฮ = ๐ โฒ ๐ = ๐ทโฒ ๐ท๐ + ๐ท โฒ (๐ถ โ ๐ท๐ ๐น )(๐ผ โ ๐๐ด๐ )โ1 ๐๐ต๐ + ๐ต โฒ ๐ โฒ (๐ผ โ ๐ดโฒ ๐ โฒ )โ1 ๐ถ โฒ ๐ท๐ + ๐ต โฒ ๐ โฒ (๐ผ โ ๐ดโฒ ๐ โฒ )โ1 ๐ถโฒ(๐ถ โ ๐ท๐ ๐น )(๐ผ โ ๐๐ด๐ )โ1 ๐๐ต๐ โฒ
โฒ
(29)
โฒ
= (๐ท ๐ท๐ + ๐ต ๐ต๐ ) + [๐ท (๐ถ โ ๐ท๐ ๐น ) + ๐ต โฒ (๐ด โ ๐ต๐ ๐น )]๐(๐ผ โ ๐ด๐ ๐)โ1 ๐ต๐ or, with
[
๐ดโฒ ๐ตโฒ
๐ถโฒ ๐ทโฒ
][
๐ด ๐ถ
๐ต๐ ๐ท๐
]
[ =
๐ผ ๐ถ๐
0 ๐ท๐
] ,
ฮ = ๐ท๐ + [๐ถ๐ โ ๐ท๐ ๐น ]๐(๐ผ โ ๐ด๐ ๐)โ1 ๐ต๐
(30) (31)
which indeed exhibits ฮ as polynomial since ๐ด๐ ๐ is nilpotent. This proves the theorem for ๐ and ฮ. A further factorization is obtained with the same machinery. Let us de๏ฌne a kind of dual operator ๐๐โฒ = ๐ ๐ โฒ = (๐ท๐โฒ ๐ท + ๐ต๐โฒ ๐ต) + ๐ต๐โฒ (๐ผ โ ๐ โฒ ๐ดโฒ )โ1 ๐ โฒ (๐ถ โฒ ๐ท + ๐ดโฒ ๐ต), or, taking conjugates (the realization given may not be minimal, it will actually only be minimal when ๐ has no intrinsic inner left factor, i.e., an inner, degree reducing left factor!) ๐๐ := (๐ต โฒ ๐ต๐ + ๐ท โฒ ๐ท๐ ) + (๐ต โฒ ๐ด + ๐ทโฒ ๐ถ)๐(๐ผ โ ๐ด๐)โ1 ๐ต๐
(32)
โฒ
then we have the factorization ๐๐โฒ = ๐ โฒ ๐ = ๐ โ ๐โฒ ๐โโฒ ฮโฒ = ๐ โโฒ ฮโฒ , or ๐๐ = ฮ๐ โ1
(33)
and ๐ is now seen as the dead beat polynomial based on the input {๐ด, ๐ต}pair of ๐๐ . Essential uniqueness for ๐๐ in case ๐ด๐ is completely non-unitary, follows directly from the fact that in that case ๐ต๐ cannot have a non-zero kernel. โก The connection between ๐ and ๐๐ is ๐ท๐ + ๐ถ๐ ๐(๐ผ โ ๐ด๐)โ1 ๐ต๐ , we have [ ] [ ๐ด ๐ต๐ ๐ผ = โฒ ๐ถ ๐ท ๐ต ๐ ๐ ] [ [ ๐ด ๐ด ๐ต = ๐ถ ๐ท ๐ถ
in a sense โsymmetricalโ, with ๐๐ = ][ ] 0 ๐ด ๐ต๐ , ๐ทโฒ ] [๐ถ ๐ท๐ ] ๐ต๐ ๐ผ ๐ถ๐โฒ ๐ท๐ 0 ๐ท๐โฒ
(34)
showing that not only ๐ can be re-derived from (the non-minimal realization of) ๐๐ , but also that the relation is actually of the duality kind, reachability exchanged for observability. We have dual relations for ๐๐ . In particular, ๐ = ๐ ๐๐โฒ is causal,
256
P. Dewilde
and we have for the realizations, with ๐บ the dual of ๐น now based on the pair {๐ด, ๐ถ}, and ๐ด๐ = ๐ด โ ๐บ๐ถ nilpotent ] [ ๐ด ๐บ ๐๐โ1 โ [ ๐ถ ๐ผ ] ๐ด๐ โ๐บ ๐๐ โ [ โฒ ๐ถ โฒ ๐ผ โฒ ] ๐ด ๐ถ +๐ด๐บ โ1 ๐๐ โ (35) โฒ โฒ โฒ [ ๐ต๐ ๐ท๐ + ๐ต๐ ๐บ ] ๐ด๐ ๐ต๐ โ ๐บ๐ท๐ ๐๐ โ ๐ท๐ ] [๐ถ ๐ด๐ ๐ต โ ๐บ๐ท ฮ๐ โ ๐ถ ๐ท such that
โ1 ), (36) ๐๐ = ฮโฒ๐ ๐โโฒ ๐ (= ฮ๐ and there are of course a whole collection of similar conjugate relations. It is easily veri๏ฌed, by direct computation, that ๐ = ๐๐โ1 ฮ๐ . Hence, we have obtained a โrightโ polynomial representation of a uniformly exponentially stable causal matrix ๐ = ๐โโฒ ฮโฒ . ฮ need not be invertible, but remark that ๐โโฒ borrows the original observability pair as expected. Such a factorization is of course not unique, ฮ and ๐ could be replaced by any ฮ๐ , ๐๐ when ๐ is polynomial, invertible and ๐ โ1 is polynomial as well โ so that both ๐ and ๐๐ de๏ฌne the same observability kernel. ๐ is then a time varying or matrix version of a unimodular operator.
Bezout relations The previous development allows for the explicit determination of Bezout identities as well. We observe that ] [ ] [ ] [ ๐๐ ฮ๐ = ๐ผ ๐ท + ๐ถ๐(๐ผ โ ๐ด๐ ๐)โ1 โ๐บ ๐ต โ ๐บ๐ท . (37) Let now a minimal ๐ป be such that ๐ด โ ๐ต๐ป is nilpotent. ๐ป exists because we assumed the original system to be controllable in ๏ฌnite time and is obtained through the dead beat construction, then ] ][ [ [ ] ๐ผ โ๐ท ๐ถ ๐ด๐ โ โ๐บ ๐ต โ ๐บ๐ท = ๐ด โ ๐ต๐ป := ๐ดโ (38) ๐ป 0 ๐ผ and let now, for some new operators โฮ๐ and ๐๐ ] ] [ [ ] [ [ ] ๐ถ โ ๐ท๐ป ๐ โฮ๐ ๐ผ โ๐ท ๐(๐ผ โ ๐ดโ ๐)โ1 โ๐บ ๐ต , (39) โ = ๐ ๐๐ ๐ป 0 ๐ผ Then
[
]
[
๐ โฮ๐ ๐ ๐๐ and we have reached the Bezout identity ๐๐
ฮ๐
] =
๐๐ ๐ + ฮ๐ ๐ = ๐ผ,
[
๐ผ
0
]
(40) (41)
Banded Matrices, Banded Inverses, Polynomial Representations
257
with ๐ and ๐ polynomial in ๐, as well as a conjugate factorization ๐๐โ1 ฮ๐ = ฮ๐ ๐๐โ1 = ๐ . From this construction we have further [ ] ๐ดโ ๐ต ๐๐ โ [ โ๐ป ๐ผ ] ๐ด ๐ต โ1 ๐๐ โ (42) ๐ป ๐ผ ] [ ๐ต ๐ดโ ฮ๐ โ ๐ถ โ ๐ท๐ป ๐ท dual to the construction of ๐ and ฮ, where, again, ๐๐ is causal with causal inverse, ๐ป is the corresponding dead beat operator based on the pair {๐ด, ๐ต} (not necessarily in input normal form!), and ฮ๐ is polynomial. Finally, we also have two new polynomial operators ๐
and ๐ such that ] ] [ ] [ [ [ ] ๐ถ ๐ผ ๐ท ๐๐ ฮ๐ ๐(๐ผ โ ๐ด๐ ๐)โ1 โ๐บ ๐ต โ ๐บ๐ท , + = (43) ๐ป 0 ๐ผ ๐
๐ the dual Bezout identity
(44) (โ๐
)ฮ๐ + ๐๐๐ = ๐ผ, and the connection between the two completed, invertible polynomial matrices [ ] [ ]โ1 ๐ ๐ ฮ๐ ๐ โฮ๐ . (45) = ๐
๐ ๐ ๐๐
3. Polynomial interpolation theory for matrices: An approach Interpolation theory in the matrix context necessitates the notion of a โvaluationโ, introduced in [1] and further worked out in [2, 11, 7]. I quickly summarize the concepts in the present notation. Let ๐ = ๐0 + ๐๐1 + ๐ 2 ๐2 + โ
โ
โ
be a causal (lower) and bounded operator with the given diagonal expansion, and let ๐ be a (compatible) block diagonal operator such that ๐(๐ ๐ โฒ ) < 1. We de๏ฌne the value of ๐ at ๐ to be a diagonal operator, denoted ๐๐ (in the notation of the original paper it was denoted in a somewhat cumbersome way by ๐ โง (๐ )) which is such that ๐ = ๐๐ + (๐ โ ๐ )๐๐ for some bounded, causal (lower) ๐๐ . This is the socalled W-transform of ๐ , so called because of the resulting reproducing kernel, see [2], where it is also shown that ๐๐ is de๏ฌned by the strongly convergent series ๐๐ = ๐๐ + ๐ ๐1 + ๐ ๐ (โ1) ๐2 + ๐ ๐ (โ1) ๐ (โ2) ๐3 + โ
โ
โ
.
(46)
The notion clearly generalizes the valuation of a complex-valued matrix function ๐ (๐ง) at a point ๐ โ C as ๐ (๐). Because of the non-commutativity of the shift operator ๐, it does not have all the properties of the valuation in the complex plane. We do have the following properties. 1. ๐๐ is the ๏ฌrst anti-causal term in the expansion of (๐ โ ๐ )โ1 ๐ : (๐ โ ๐ )โ1 ๐ = ๐ โฒ2 (โ
โ
โ
) + ๐ โฒ ๐๐ + ๐๐ in which the โ
โ
โ
is purely anticausal.
(47)
258
P. Dewilde
Proof. Expand ๐ โฒ (๐ผ โ ๐ ๐ โฒ )โ1 ๐ ! 2. (Chain rule) For ๐ and ๐ anticausal we have (๐ ๐)๐ = [๐๐ ๐]๐ . If ๐๐ is (โ1) โ1 invertible, we have in addition (๐ ๐)๐ = ๐๐ ๐๐1 where ๐1 = ๐๐ ๐ ๐๐ . Proof. We have โ(โ1)
โ1 โ1 (๐ โ ๐ )โ1 ๐๐ = (๐๐ ๐ โ ๐๐ ๐ )โ1 = (๐๐๐ (โ1)
= ๐๐
โ1 โ ๐๐ ๐ )โ1
(๐ โ ๐1 )โ1 ,
and hence (๐ โ ๐ )โ1 ๐ ๐ = (๐ โ ๐ )โ1 ๐๐ ๐ + ๐๐ ๐ (โ1)
= ๐๐
(โ1)
(๐ โ ๐1 )โ1 ๐๐1 + ๐๐
๐๐ + ๐๐ ๐,
the last being equal again to (๐ โ ๐ )โ1 ๐๐ ๐๐1 + causal. 3. (Constants) Let ๐ท be a compatible diagonal operator, then (๐ ๐ท)๐ = ๐๐ ๐ท. If ๐ท is invertible and compatible, then (๐ท๐ )๐ = ๐ท(โ1) ๐๐1 , in which ๐1 = ๐ทโ1 ๐ ๐ท(โ1) . For addition we simply have (๐ + ๐ท)๐ = ๐๐ + ๐ท. 4. (State space formulas) Let ๐ = ๐ท + ๐ถ๐(๐ผ โ ๐ด๐)โ1 ๐ต be a realization for ๐ , assumed to be causal and such that ๐(๐ด๐) < 1. Then ๐๐ = ๐ท + ๐ ๐ ๐ต where ๐ solves the Lyapunov-Stein equation ๐ (1) = ๐ถ + ๐ ๐ ๐ด.
(48)
In fact, (โ1)
๐ = [๐ถ(๐ผ โ ๐๐ด)โ1 ]๐
= [๐ถ + ๐ ๐ถ (โ1) ๐ด + ๐ ๐ (โ1) ๐ถ (โ2) ๐ด(โ1) ๐ด + โ
โ
โ
](โ1)
(49)
and hence also
๐๐ = ๐ท + [๐ถ๐(๐ผ โ ๐ด๐)โ1 ]๐ ๐ต (50) in accordance with the previous rules. In the sequel we shall need still another property, given by the next Lemma, which follows by direct evaluation: Lemma 1. Suppose that for ๐ = 1, 2, ๐๐ = ๐ท๐ + ๐ถ๐ (๐ผ โ ๐ด๐)โ1 ๐ต, and that ๐2 is causally invertible, i.e., ๐2โ1 = ๐ท2โ1 โ ๐ท2โ1 ๐ถ2 ๐(๐ผ โ ๐ผ2 ๐)โ1 ๐ต๐ท2โ1 with ๐ผ2 := ๐ด โ ๐ต๐ท2โ1 ๐ถ2 , then ๐1 ๐2โ1 = ๐ท1 ๐ท2โ1 + (๐ถ1 โ ๐ท1 ๐ท2โ1 ๐ถ2 )๐(๐ผ โ ๐ผ2 ๐)โ1 ๐ต๐ท2โ1 .
(51)
The straight โLยจ owner typeโ directional interpolation problem for matrices can now be de๏ฌned as follows: given a block diagonal matrix ๐ and directional data ๐ and ๐, ๏ฌnd a causal operator ๐ such that (๐๐)๐ = ๐, or, to put it di๏ฌerently, such that ๐๐ interpolates ๐ at the (block diagonal) value ๐ . Note that ๐ cannot be taken out of the bracket! To somehow restrict the discussion to a โwell-posedโ case, we assume that the] [ interpolation data satis๏ฌes the property that the reachability pair ๐ ๐ โ๐
Banded Matrices, Banded Inverses, Polynomial Representations
259
is uniformly reachable in ๏ฌnite time. As a consequence we have that the interpolation data can be assumed to be input normalized ๐ ๐ โฒ +๐๐ โฒ + ๐๐โฒ = ๐ผ as well. Due to our previous theory, solutions can then be generated on the basis of polynomial representations of the operator [ ] ๐ โฒ = (๐ โ ๐ )โ1 ๐ โ๐ = ฮ๐โ1 (52) in which ฮ and ๐ are polynomial in ๐. From the previous theory of polynomial representations applied to ๐ , we have in sequence โก โค ๐โฒ ๐ผ โฃ ๐โฒ 0 โฆ ๐ โ โ๐ โฒ 0 โก โค โฒ ๐ ๐ ๐ก โฃ ๐ โฒ ๐11 ๐12 โฆ ๐ :โ โฒ โกโ๐ โฒ 0 ๐22โค (53) ๐ ๐ ๐ก โฃ ๐น1 ๐ผ 0 โฆ ๐ โ1 :โ ๐น2 0 ๐ผ โก โค ๐ โฒ โ (๐ ๐น1 + ๐ก๐น2 ) ๐ ๐ก ๐ :โ โฃ ๐ โฒ โ (๐11 ๐น1 + ๐12 ๐น2 ) ๐11 ๐12 โฆ , โ๐ โฒ โ ๐22 ๐น2 0 ๐22 [ ] in which the]unitary realization for ๐ is completed by ๐ต๐ :โ ๐ ๐ก and ๐ท๐ :โ [ ๐11 ๐12 , and ๐น is the feedback matrix belonging to the input reachability 0 ๐22 [ ] pair ๐ โฒ ๐ ๐ก (which by the way is in input normal form). Notice that ๐ is polynomial in ๐ with anti-causal inverse as before. Interpolations are now obtained by pulling ๐ to the left-hand side: [ ] [ ] ๐11 ๐12 โ1 ๐ โ๐ (๐ โ ๐ ) = ฮ. (54) ๐21 ๐22 Let now ๐ and ๐ be any causal, compatible operator (in particular diagonal constants), and ๐(1) := ๐11 ๐ + ๐12 ๐ as well as ๐(2) := ๐21 ๐ + ๐22 ๐, then we ๏ฌnd ( ) (๐ โ ๐ )โ1 ๐๐(1) โ ๐๐(2) โ causal (55) and hence If, in addition, ๐
(๐๐(1) )๐ = (๐๐(2) )๐ (2)
is causally invertible, we shall have, with ๐ = ๐ (๐๐)๐ = ๐
(56) (1)
(๐
(2) โ1
)
(57)
a solution of the stated Lยจ owner interpolation problem. This will be the case when ๏ฌnite matrices are concerned, because in that case, the invertibility of ๐22 is necessary and su๏ฌcient for the causal invertibility of ๐22 . However, the general case is much more involved and beyond the scope of the present paper. Actually, we
260
P. Dewilde
can prove the converse (we call ๐ regular when it has a polynomial representation ๐ = ๐๐โ1 with ๐ causally invertible): Theorem 5. Under the regularity conditions stated and ๐ as de๏ฌned in (53) we have that any causal and regular ๐ for which (๐๐)๐ = ๐ can be written as ๐ = ๐(1) (๐(2) )โ1 whereby ๐(1) := ๐11 ๐ + ๐12 ๐ and ๐(2) := ๐21 ๐ + ๐22 ๐ for causal operators ๐ and ๐, and ๐(2) is causally invertible. ] [ ๐ โ causal, and Proof. When (๐ โ ๐ )โ1 (๐๐ โ ๐) โ causal, then a fortiori ๐ โฒ ๐ผ since ๐ โฒ = ๐ ๐โ1 we must have ] ] [ [ ๐1 ๐ (58) =๐ ๐ผ ๐1 for some causal operators ๐1 and ๐1 . Let now ๐ = ๐๐โ1 be a polynomial representation for ๐ with ๐ causally invertible (as de๏ฌned in the previous section), then, by multiplying right with ๐, we conclude that there exist ๐ and ๐ such that ] [ (1) ] ] [ [ ๐ ๐ ๐ := =๐ (59) ๐ ๐ ๐(2) which makes ๐(2) causally invertible.
โก
The problem is hence โreducedโ to ๏ฌnding adequate ๐ and ๐, at least when one wants the interpolating function ๐ to be causally bounded. This is a di๏ฌerent problem than the one considered in the classical Lยจowner theory, where boundedness (or stability) does not play a role. Although I do not claim to have solved this part of the problem (at least not algorithmically), it is possible to test whether a given causal ๐ has a causal inverse, by computing an inner-outer decomposition, as explained in the ๏ฌrst section of this paper. If the inner factor turns out to be trivially constant (i.e., all ๐๐ are empty), then ๐ will have a causal inverse. Be that as it may, if one wants to proceed as in the LTI theory, then one can either work with an unstable (or formally causal) inverse, or assume that the factor to be inverted is indeed causally invertible. Lemma 1 then shows that the resulting interpolating operator is indeed of at most the same degree as ๐, given that the chosen ๐ and ๐ are mere diagonal operators.
4. Some further remarks Finding complete representations for semi-separable matrices as ratios of (minimal) banded matrices is new, to the best of my knowledge. A partial solution to the problem for the case of unitary matrices was given in [2] and involved quite a complex argument. I hope that the method presented in this paper greatly simpli๏ฌes the issue and provides for a complete set of representations. The classical, rational approach to Lยจ owner interpolation as initiated in [3] and very extensively treated in [4, 19] follows di๏ฌerent approaches that do not seem to
Banded Matrices, Banded Inverses, Polynomial Representations
261
generalize to the semi-separable case. However, the paper of Antoulas, Ball, Kang and Willems does clarify the role played by co-prime polynomial factorizations, which is also used in the theory presented here, although the factorizations are di๏ฌerent. The role played by controllability indices in the classical theory, is here taken over by the dead beat indices, which are closely related to them.
5. Appendix: Methods to compute pre-images and the numerical calculation of the ๐ญ matrix The most elementary operation needed to compute pre-images is the so-called QR factorization (and its duals) on a general matrix. Let ๐ด be an ๐ ร ๐ matrix of rank ๐ฟ, then a QR factorization compresses the rows of ๐ด into a new matrix ๐
whose ๏ฌrst ๐ฟ rows form a basis for the rows of ๐ด, and whose further rows are zero. The ๏ฌrst rows even[have a special form (which often is immaterial but numerically ] practical), namely 0 โ
โ
โ
0 ๐ โ
โ
โ
, where ๐ > 0 and all data is crowded to the North-East corner of the matrix. This is called an echelon form. It is achieved, e.g., through a sequence of elementary rotations acting on the rows of the matrix, compressing ๏ฌrst the data on the ๏ฌrst column to the top, as shown in the following schema: โค โก โค โค โก โก โ
โ
โ
โ
โ
โ
โ
โ
โ
โฒ โฒ โข โ
โ
โ
โฅ ๐1,2 โข 0 โ
โ
โฅ ๐1,3 โข 0 โ
โ
โฅ โฅ โข โฅ โฅ โข โข โฃ โ
โ
โ
โฆ โโ โฃ โ
โ
โ
โฆ โโ โฃ 0 โ
โ
โฆ โ
โ
โ
โ
โ
โ
โ
โ
โ
โค โค โค โก โก โก (60) โ
โ
โ
โ
โ
โ
โ
โ
โ
โฒ โฒ โฒ โฅ๐ โข โฅ โข ๐1,4 โข 0 ๐ โ
โ
โฅ โฅ 2,3 โข 0 โ
โ
โฅ 2,4 โข 0 โ
โ
โฅ โโ โข โฃ 0 โ
โ
โฆ โโ โฃ 0 0 โ
โฆ โโ โฃ 0 0 โ
โฆ 0 โ
โ
0 โ
โ
0 0 โ
to be followed by a ๏ฌnal rotation ๐โฒ3,4 on the third and fourth row. Here each ๐โฒ๐,๐ is the (transpose of) a Jacobi rotation matrix acting on elements of the ๐th and ๐th rows. Putting all these rotations together in a single matrix ๐, we obtain ๐โฒ ๐ด = ๐
or ๐ด = ๐๐
. When a zero column is encountered, it is skipped to the next, yielding not an upper triangular form with positive elements on the main diagonal, but a staircase form. The important issue here is that all the data in ๐ and ๐
are completely generated from the data in ๐ด, although there is no general formula known that expresses these elements in closed form โ in numerical engineering this is known as โarray processingโ, converting one array into others, and is maybe the most powerful numerical technique available in matrix calculus. A similar operation on the columns, often accomplished by compressing the columns of ๐ด in the South-East corner, and starting on the bottom row produces a stack of basis vectors in echelon form, crowded in the right-hand side of the matrix. Let us now move to the situation in the paper. Suppose bases ๐ for โฌ, ๐ for ๐ด and u for โฌ, respect. ๐ฐ have been chosen, and assume the realizations of the
262
P. Dewilde
operators ๐ : โฌ โ ๐ด and ๐ : ๐ฐ โ ๐ด in these bases to be the matrices ๐ด and ๐ต of dimensions respect. ๐พ ร ๐ฟ and ๐พ ร ๐. We perform a QR factorization on ๐ต = ๐ ๐
that determines ๐ and ๐
, and then an LQ factorization on ๐ โฒ ๐ด = ๐๐. ๐ and ๐ are orthogonal (unitary) matrices, ๐
is in top row-echelon form and ๐ in right column echelon form. [Note: the ๏ฌrst QR factorization compresses the rows of ๐ต to the top, while the next LQ factorization compresses columns to the right starting operations on the last row.] In block notation this produces [ ] ๐
๐ข ๐
= (61) 0 where the ๐ rows of ๐
๐ข are linearly independent, and [ ] 0 ๐11 ๐12 ๐= 0 0 ๐22
(62)
where the ๏ฌrst set of rows in ๐ is taken to have ๐ rows, equal to the number of rows in ๐
๐ข , the columns of ๐ (and in particular of ๐11 and ๐22 ) are linearly independent (de๏ฌning the dimensions of these matrices), entries may disappear, depending on the rank of the matrices involved (actually any entry may disappear). It follows from the respective staircase structures that the columns of ๐11 lie in the range of the columns of ๐
๐ข and also de๏ฌne the maximal (column) subspace with that property, for which they provide a basis thanks to [the echelon ]form. Hence, there exists a matrix ๐นห such that ๐11 = ๐
๐ข ๐นห . Let ๐ฅโฒ = 0 ๐ฅโฒ2 0 de๏ฌne a vector ๐ฅ conformal to ๐, then we have โก โค โก โค 0 0 [ ] ๐ โฒ ๐ด๐โฒ โฃ ๐ฅ2 โฆ = ๐
0 ๐นห 0 โฃ ๐ฅ2 โฆ (63) 0 0 for any vector ๐ฅ2 of appropriate dimensions (and, again, some entries may not be present), from which it follows that โก โค โก โค 0 0 [ ] ๐ด๐โฒ โฃ ๐ฅ2 โฆ = ๐ต 0 ๐นห 0 ๐๐โฒ โฃ ๐ฅ2 โฆ . (64) 0 0 Suppose the dimensions of ๐ฅ are ๐ฟ1 +๐ฟ2 +๐ฟ3 , then weโcan conclude that the columns of ๐๐โฒ from ๐ฟ1 + 1 to ๐ฟ1 + ๐ฟ2 span the pre-image ๐0 of ๐ต under ๐ด in the basis ๐. Let ๐2โฒ be that collection of columns (in MATLAB notation ๐2โฒ = ๐โฒ๐ฟ1 +1:๐ฟ1 +๐ฟ2 ,: ), then [๐0 = ๐๐2โฒ is] a choice of basis for ๐ฎ0 and we have, for any ๐ฅ = ๐2โฒ ๐ฅ2 and ๐น = 0 ๐นห 0 ๐ ๐๐๐ฅ = ๐u๐น ๐ฅ. (65) (Proof. As ๐ด๐2โฒ = ๐ต๐น ๐2โฒ , ๐๐ = ๐๐ด and ๐u = ๐๐ต, by pre-multiplication with ๐ and post-multiplication with x. Note that this does not assume orthogonality of the bases.) The algorithm can be enhanced numerically by using SVDโs, see the discussion further.
Banded Matrices, Banded Inverses, Polynomial Representations
263
Application to the computation of the feedback operator To compute the crucial feedback operator ๐น following the principle set out in the previous paragraph, we stack the vectors and perform a QR factorization on them, with ๐ธ๐ a stack of unit vectors corresponding to the {๐๐ }: ] [ ๐ต๐ ๐ธ๐ ๐ธ1 โ
โ
โ
๐ธ๐๐+1 = ๐ ๐
(66) ๐
is in row echelon form:
โก
๐
๐ต 0 .. .
โข โข โข ๐
=โข โข โฃ 0 0
๐
๐ต,0 ๐
00 .. . 0 0
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
๐
๐ต,๐๐+1 ๐
1,๐๐+1 .. .
๐
๐๐ ,๐๐+1 0
โค โฅ โฅ โฅ โฅ โฅ โฆ
(67)
in which the rows of ๐
๐ต are in column echelon form themselves (hence also โ linearly independent), ๐
00 is either empty (when ๐๐ต accidentally spans the space ๐0 ) and then a further staircase in row echelon form arises, ending in the block ๐
๐๐ ,๐๐+1 , which will only be on the main block diagonal when ๐๐ = ๐๐+1 + 1. Once ๐ is obtained, we perform a RQ factorization on ๐ ๐ด = ๐๐, producing โก โค 0 ๐00 โ
โ
โ
๐0,๐๐ โข 0 0 โ
โ
โ
๐1,๐๐ โฅ โข โฅ ๐=โข . (68) โฅ . .. .. โฃ .. โฆ โ
โ
โ
. 0
0
โ
โ
โ
๐๐๐+1 ,๐๐
in which ๐๐๐ ,๐๐+1 is either empty or in column echelon form, with the staircase mounting up leftwards till ๐00 , which actually may also be accidentally empty. It does not really matter that some or many of these main entries are empty, if so, they are ignored. The dimensions of the rows are chosen conformal to {๐0 , ๐1 , . . . , ๐๐๐+1 }, while the columns now follow a new indexing schema as a result from the factorization, and shall correspond to the requested basis (in an ultimate case it may be that ๐ด = 0, then ๐ = 0 and the whole matrix echelon structure disappears). Because the rows of ๐
are linearly independent, the ๏ฌrst block row of ๐ can be โkilledโ by ๐
๐ต (they are conformal), yielding the existence of a matrix ๐นห such that โค โก โค โก ๐
๐ต 0 ๐00 โ
โ
โ
๐0,๐๐ โฅ โข โข 0 0 โ
โ
โ
0 โฅ ] โฅ โข 0 โฅ[ โข (69) โข .. .. .. โฅ = โข .. โฅ 0 ๐นห0 โ
โ
โ
๐นห๐๐ . โฆ โฆ โฃ โฃ . . โ
โ
โ
. . 0 0 โ
โ
โ
0 0 Also, the basis for ๐ฎ๐ follows from ๐. If ๐00 is not empty, then the number of its columns plus the number of columns preceding it determine the dimension of the space ๐ฎ0 (we made the kernel of ๐ explicit here). If ๐11 is not empty, then its number of columns determines the dimension of ๐ฎ1 โ which again may be empty if the staircase does not make a jump in the rows corresponding to ๐1 etc. The
264
P. Dewilde
remainder of the matrix ๐, in column echelon form, is the feedback matrix ๐ดห๐ in the current bases ๐๐ and ๐๐ โฒ . Choosing the โdead beat basisโ for โฌ๐ and keeping the basis ๐ for โฌ๐+1 changes the matrix ๐ด to ๐ด๐โฒ , while ๐น = ๐นห ๐โฒ , and ๐ต just remains what it is. [ ]โฒ Example 1. We take ๐ต๐ = 1 0 0 1 , and โก โค โ1 0 0 2 โข 0 1 0 โฅ โข โฅ ๐ด๐ = โข โ1 (70) 1 โฅ โ 0 โฃ โ 2 2 โฆ โ1 0 1 2 and let us assume that ๐0 and ๐1 both have bases for โฌ๐+1 . We then have โก 1 [ ] โข 0 ๐ต๐ ๐ธ0 ๐ธ1 = โข โฃ 0 1 The ๏ฌrst QR factorization produces โก โค โก โ1 0 1 1 0 0 0 2 โข 0 0 1 0 0 โฅ โข 0 1 โข โฅ โข โฃ 0 0 0 1 0 โฆ=โฃ 0 0 โ1 1 0 0 0 1 0 2
0 0 1 0
dimension 2 and have been used as 1 0 0 0
0 1 0 0
0 0 1 0
โคโก โ โ โ12 2 โข 0 0 โฅ โฅโข 0 โฆโฃ 0 โ1 0 2
where we see the row echelon form appearing. Next, โก 1 โค โ 1 โ12 2 โข 0 1 0 โฅ โข โฅ ๐ โฒ ๐ด๐ = โข โ โ1 0 โ1 โฅ โฃ 2 2 โฆ โ โ12 0 โ12
โค 0 0 โฅ โฅ. 0 โฆ 1 โ 2 0 0 0
(71)
0 1 0 0
0 0 1 0
โ1 2
โค
0 โฅ โฅ 0 โฆ
(72)
โ1 2
(73)
and a LQ factorization produces the right column echelon form (starting operations from the last row and compressing the columns to the right): โก 1 โค โก โค โ 1 โ12 โค 1 1 0 โก โ1 2 โ1 โข 0 โฅ 2 2 โข โฅ 1 0 0 1 0 โฅโฃ โข โฅ 1 0 โฆ (74) โข โ โ1 0 โ1 โฅ = โข โฃ 0 0 1 โฆ โฃ 2 2 โฆ โ1 โ1 โ 2 2 0 0 1 โ โ1 0 โ1 2
2
the last matrix being the transition matrix in the bases ๐ห = ๐๐โฒ and ๐๐ , and it is in column echelon form. Comparing the row echelon form for the bases and ๐ดห๐ โ we see that ๐ห1 generates ๐ฎ0 , ๐ฎ1 = (๐ห1 , ๐ห2 ) and everything is ๐ฎ2 in โฌ๐ . The ๏ฌrst column of ๐ดห๐ can be annihilated by ๐ โฒ ๐ต๐ , hence [ ] ๐นห = โ12 โ12 0 . (75)
Banded Matrices, Banded Inverses, Polynomial Representations
265
ห while keeping the basis ๐ in โฌ๐+1 , we can denote If we change bases in โฌ๐ to ๐, โฒ ห ๐ด๐ = ๐ด๐ ๐ as the current transition matrix, ๐ต๐ stays what it is, ๐นห is the correct feedback matrix, and โก โค 0 0 โ โ12 โข 0 1 0 โฅ โฅ. ๐ดห๐ = ๐ด๐โฒ โ ๐ต๐ ๐นห = โข (76) โฃ 0 0 1 โฆ 1 โ 0 0 2 More advanced methods to determine pre-images. Consider again the situation with operators ๐ : โฌ โ ๐ด and ๐ : ๐ฐ โ ๐ด and realization matrices ๐ด and ๐ต respect. Let ๐ด = ๐ ฮฃ๐ โฒ be the SVD of the matrix ๐ด, with [ ] ฮฃ11 0 ฮฃ= (77) 0 0 in which ๐ and ๐ are orthogonal (or unitary in the complex case), ฮฃ contains the singular values in the classical canonical sense (๐1 โฅ ๐2 โฅ โ
โ
โ
โฅ ๐๐ ) with ๐๐ >[0 as the last ] signi๏ฌcant [ singular] value, ฮฃ11 = diag(๐1 , . . . , ๐๐ ). Partitioning โ ๐ = ๐1 ๐2 and ๐ = ๐1 ๐2 conformally to ฮฃ, we have that ๐2 (stack of columns) is the kernel of ๐ด, which shall always belong trivially to any pre-image. โ We also see that any image shall always belong to the range of ๐ด, namely ๐1 (in case one works directly on matrices, one assumes that the bases are just the natural ones, otherwise one just post multiplies with the actual bases, as done before). Let be given a (row) stack โ โ of (column) vectors y, each of dimension conformal to ๐ด. Then only ๐1 โฉ y can contain an image with pre-image y (if this space is โ zero, then there is no pre-image except for the trivial ๐2 ). The problem hence reduces to ๏ฌnding this intersection in a stable numerical way (the problem is that โ y may not be numerically well de๏ฌned, and there is also a problem with the intersection, which may only be approximate). โ One way to proceed is to remark that the intersection must be orthogonal to โ โ ๐2 . It is characterized by the kernel of ๐2โฒ y, we have more precisely, y๐ข โ ๐1 โฉ y i๏ฌ ๐2โฒ y๐ข = 0. The image can then be described as y๐ข with ๐ข in the kernel of ๐ด, and the pre-image is then โ โ โฒ โฒ ๐ดโ1 ( y) = ๐1 ฮฃโ1 ๐2 . (78) 11 ๐1 yker(๐2 y) + This expression shows the potential indeterminacy in a nutshell (one recognizes the Moore-Penrose inverse): there is the blow up of small singular values by ฮฃโ1 11 , and also the lack of precision in the dimension of the kernel of ๐2โฒ y, which can be taken minimal (strictly zero) or maximal (within some ๐). This can be done through another SVD ofโ๐2โฒ y. Alternatively, one can look for algorithms to determine the โ angle between y and ๐1 and take that part with angle zero โ this amounts more or less to the same as before, see the literature on computing angles between subspaces!
266 Example 2. This example is only LTI case. Let us take โก 1 ๐ด=โฃ 0 0
P. Dewilde intended to make a quick connection with the 1 1 0
โค โก โค 0 0 1 โฆ, ๐ต = โฃ 0 โฆ. 1 1
(79)
It is easily veri๏ฌed that the pair {๐ด, ๐ต} is reachable. We make use of the fact that ๐ด is invertible to determine the pre-images directly. In the current natural basis, a basis of ๐ฎ0 is the pre-image of ๐ต namely ๐ดโ1 ๐ต, of ๐ฎ1 one has to add the pre-image of ๐ดโ1 ๐ต, namely ๐ดโ2 ๐ต, and for ๐ฎ2 one adds its pre-image, namely ๐ดโ3 ๐ต. Hence the sought after dead beat basis is given by the columns of โก โค 1 3 6 [ โ1 ] ๐ = ๐ด ๐ต ๐ดโ2 ๐ต ๐ดโ3 ๐ต = โฃ โ1 โ2 โ3 โฆ . (80) 1 1 1 Transforming to the new basis we get (as ๐๐ฅ = ๐ห๐ฅห with ๐ห = ๐๐ the new basis) โค โค โก โก 3 1 0 3 ห = ๐ โ1 ๐ต = โฃ โ3 โฆ . (81) ๐ดห = ๐ โ1 ๐ด๐ = โฃ โ3 0 1 โฆ , ๐ต 1 0 0 1 โก โคโฒ 1 We see immediately that ๐นห = โฃ 0 โฆ . Transforming back we ๏ฌnd 0 โค โก 1 1 0 [ ] 1 1 โฆ, ๐น = 1 3 3 ๐ด๐ = ๐ด โ ๐ต๐น = โฃ 0 (82) โ1 โ3 โ2 and ๐ด๐ is indeed nilpotent as one should expect (to check, just calculate det(๐ง๐ผ โ ๐ด๐ )!). The more general LTI algorithms are extensions of this mechanism to the case where ๐ด is not invertible and the reachability base more complicated (Kronecker indices). In particular, in the MIMO case, one can determine stacks of pre-images based on the columns of ๐ต, so as to realize the polynomial inverse in a column-degree canonical form.
Banded Matrices, Banded Inverses, Polynomial Representations
267
References [1] D. Alpay and P. Dewilde. Time-varying signal approximation and estimation. In M.A. Kaashoek, J.H. van Schuppen, and A.C.M. Ran, editors, Signal Processing, Scattering and Operator Theory, and Numerical Methods, volume III of Proc. Int. Symp. MTNS-89, pages 1โ22. Birkhยจ auser Verlag, 1990. [2] D. Alpay, P. Dewilde, and H. Dym. Lossless Inverse Scattering and reproducing kernels for upper triangular operators. In I. Gohberg, editor, Extension and Interpolation of Linear Operators and Matrix Functions, volume 47 of Operator Theory, Advances and Applications, pages 61โ135. Birkhยจ auser Verlag, 1990. [3] A.C. Antoulas and B.D.O. Anderson. On the scalar rational interpolation problem. IMA J. Math. Control Inform., 3:61โ88, 1986. [4] A.C. Antoulas and J.A. Ball and J. Kang and J.C. Willems. On the solution of the minimal rational interpolation problem. Linear Algebra and its Applications, 137:511โ573, 1990. [5] D. Alpay and P. Dewilde and D. Volok. Interpolation and approximation of quasiseparable systems: the Schur-Takagi case. Calcolo, 42:139โ156, 2005. [6] W. Arveson. Interpolation problems in nest algebras. J. Functional Anal., 20:208โ 233, 1975. [7] J.A. Ball, I. Gohberg, and M.A. Kaashoek. Nevanlinna-Pick interpolation for timevarying input-output maps: the discrete case. In I. Gohberg, editor, Time-Variant Systems and Interpolation, volume 56 of Operator Theory: Advances and Applications, pages 1โ51. Birkhยจ auser Verlag, 1992. [8] S. Chandrasekaran, M. Gu, and T. Pals. A fast and stable solver for smooth recursively semi-separable systems. In SIAM Annual Conference, San Diego and SIAM Conference of Linear Algebra in Controls, Signals and Systems, Boston, 2001. [9] P. Dewilde and A.-J. van der Veen. Time-varying Systems and Computations. Kluwer, out of print but freely available at ens.ewi.tudelft.nl, 1998. [10] P. Dewilde and A.-J. van der Veen. Inner-outer factorization and the inversion of locally ๏ฌnite systems of equations. Linear Algebra and its Applications, 313:53โ100, 2000. [11] P.M. Dewilde. A course on the algebraic Schur and Nevanlinna-Pick interpolation problems. In Ed. F. Deprettere and A.J. van der Veen, editors, Algorithms and Parallel VLSI Architectures. Elsevier, 1991. [12] P. Van Dooren. A unitary method for deadbeat control. Proceedings MTNS, 1983. [13] Y. Eidelman and I. Gohberg. On a new class of structured matrices. Notes distributed at the 1999 AMS-IMS-SIAM Summer Research Conference, Structured Matrices in Operator Theory, Numerical Analysis, Control, Signal and Image Processing, 1999. [14] Y. Eidelman and I. Gohberg. A modi๏ฌcation of the Dewilde-van der Veen method for inversion of ๏ฌnite structured matrices. Linear Algebra and its Applications, 343-344, 2002. [15] I. Gohberg, T. Kailath, and I. Koltracht. Linear complexity algorithms for semiseparable matrices. Integral Equations and Operator Theory, 8:780โ804, 1985. [16] T. Kailath. Fredholm resolvents, Wiener-Hopf equations and Riccati di๏ฌerential equations. IEEE Trans. Information Theory, 15(6), November 1969.
268
P. Dewilde
[17] T. Kailath and B.D.O. Anderson. Some integral equations with nonsymmetric separable kernels. SIAM J. of Applied Math., 20 (4):659โ669, June 1971. [18] L. Kronecker. Algebraische Reduktion der Scharen bilinearer Formen. S.B. Akad. Berlin, pages 663โ776, 1890. [19] A. J. Mayo and A.C. Antoulas. A framework for the solution of the generalized realization problem. Linear Algebra and its Applications, 425:634โ662, 2007. [20] S. Chandrasekaran, P. Dewilde, M. Gu, T. Pals, A.-J. van der Veen and J. Xia. A fast backward stable solver for sequentially semi-separable matrices, volume HiPC202 of Lecture Notes in Computer Science, pages 545โ554. Springer Verlag, Berlin, 2002. [21] G. Strang. Banded matrices with banded inverses and ๐ = ๐๐๐ข. Linear Algebra and its Applications, to appear, 2011. [22] A.J. van der Veen. Time-Varying System Theory and Computational Modeling: Realization, Approximation, and Factorization. PhD thesis, Delft University of Technology, Delft, The Netherlands, June 1993. Patrick Dewilde Institute for Advance Study TU Mยจ unchen and Faculty of EEMCS TU Delft
Operator Theory: Advances and Applications, Vol. 218, 269โ297 c 2012 Springer Basel AG โ
Description of Helson-Szegห o Measures in Terms of the Schur Parameter Sequences of Associated Schur Functions Vladimir K. Dubovoy, Bernd Fritzsche and Bernd Kirstein Dedicated to the memory of Israel Gohberg
Abstract. Let ๐ be a probability measure on the Borelian ๐-algebra of the unit circle. Then we associate a Schur function ๐ in the unit disk with ๐ and give characterizations of the case that ๐ is a Helson-Szegห o measure in terms of the sequence of Schur parameters of ๐. Furthermore, we state some connections of these characterizations with the backward shift. Mathematics Subject Classi๏ฌcation (2000). Primary 30E05, 47A57. Keywords. Helson-Szegห o measures, Riesz projection, Schur functions, Schur parameters, unitary colligations.
1. Interrelated quadruples consisting of a probability measure, a normalized Carathยดeodory function, a Schur function and a sequence of contractive complex numbers Let ๐ป := {๐ โ โ : โฃ๐โฃ < 1} and ๐ := {๐ก โ โ : โฃ๐กโฃ = 1} be the unit disk and the unit circle in the complex plane โ, respectively. The central object in this paper is the class โณ+ (๐) of all ๏ฌnite nonnegative measures on the Borelian ๐-algebra ๐
on ๐. A measure ๐ โ โณ+ (๐) is called probability measure if ๐(๐) = 1. We denote by โณ1+ (๐) the subset of all probability measures which belong to โณ+ (๐). Now we are going to introduce the subset of Helson-Szegหo measures on ๐. For this reason, we denote by ๐ซ๐๐ the set of all trigonometric polynomials, i.e., the set of all functions ๐ : ๐ โ โ for which there exist a ๏ฌnite subset ๐ผ of the set โค of all integers and a sequence (๐๐ )๐โ๐ผ from โ such that โ ๐ (๐ก) = ๐๐ ๐ก๐ , ๐ก โ ๐. (1.1) ๐โ๐ผ
270
V.K. Dubovoy, B. Fritzsche and B. Kirstein
If ๐ โ ๐ซ๐๐ is given via (1.1), then the conjugation ๐ห of ๐ is de๏ฌned via โ ๐ห(๐ก) := โ๐ (sgn ๐)๐๐ ๐ก๐ , ๐ก โ ๐,
(1.2)
๐โ๐ผ
where sgn 0 := 0 and where sgn ๐ :=
๐ โฃ๐โฃ
for each ๐ โ โค โ {0}.
De๏ฌnition 1.1. A non-zero measure ๐ which belongs to โณ+ (๐) is called a HelsonSzegห o measure if there exists a positive real constant ๐ถ such that for all ๐ โ ๐ซ๐๐ the inequality โซ โซ โฃ๐ห(๐ก)โฃ2 ๐(๐๐ก) โค ๐ถ โฃ๐ (๐ก)โฃ2 ๐(๐๐ก) (1.3) ๐
๐
is satis๏ฌed. If ๐ โ โณ+ (๐), then ๐ is a Helson-Szegหo measure if and only if ๐ผ๐ is a Helson-Szegห o measure for each ๐ผ โ (0, +โ). Thus, the investigation of HelsonSzegหo measures can be restricted to the class โณ1+ (๐). The main goal of this paper is to describe all Helson-Szegห o measures ๐ belonging to โณ1+ (๐) in terms of the Schur parameter sequence of some Schur function ๐ which will be associated with ๐. Let ๐(๐ป) be the Carathยดeodory class of all functions ฮฆ : ๐ป โ โ which are holomorphic in ๐ป and which satisfy Re ฮฆ(๐) โฅ 0 for each ๐ โ ๐ป. Furthermore, let ๐ 0 (๐ป) := {ฮฆ โ ๐(๐ป) : ฮฆ(0) = 1}. The class ๐(๐ป) is intimately related with the class โณ+ (๐). According to the Riesz-Herglotz theorem (see, e.g., [14, Chapter 1]), for each function ฮฆ โ ๐(๐ป) there exist a unique measure ๐ โ โณ+ (๐) and a unique number ๐ฝ โ โ such that โซ ๐ก+๐ ฮฆ(๐) = ๐(๐๐ก) + ๐๐ฝ, ๐ โ ๐ป. (1.4) ๐ ๐กโ๐ Obviously, ๐ฝ = Im [ฮฆ(0)]. On the other hand, it can be easily checked that, for arbitrary ๐ โ โณ+ (๐) and ๐ฝ โ โ, the function ฮฆ which is de๏ฌned by the right-hand side of (1.4) belongs to ๐(๐ป). If we consider the Riesz-Herglotz representation (1.4) for a function ฮฆ โ ๐ 0 (๐ป), then ๐ฝ = 0 and ๐ belongs to the set โณ1+ (๐). Actually, in this way we obtain a bijective correspondence between the classes ๐ 0 (๐ป) and โณ1+ (๐). Let us now consider the Schur class ๐ฎ(๐ป) of all functions ฮ : ๐ป โ โ which are holomorphic in ๐ป and which satisfy ฮ(๐ป) โ ๐ป โช ๐. If ฮ โ ๐ฎ(๐ป), then the function ฮฆ : ๐ป โ โ de๏ฌned by ฮฆ(๐) :=
1 + ๐ฮ(๐) 1 โ ๐ฮ(๐)
(1.5)
belongs to the class ๐ 0 (๐ป). Note that from (1.5) it follows ๐ฮ(๐) =
ฮฆ(๐) โ 1 , ฮฆ(๐) + 1
๐ โ ๐ป.
(1.6)
Description of Helson-Szegห o Measures
271
Consequently, it can be easily veri๏ฌed that via (1.5) a bijective correspondence between the classes ๐ฎ(๐ป) and ๐ 0 (๐ป) is established. Let ๐ โ ๐ฎ. Following I. Schur [15], we set ๐0 := ๐ and ๐พ0 := ๐0 (0). Obviously, โฃ๐พ0 โฃ โค 1. If โฃ๐พ0 โฃ < 1, then we consider the function ๐1 : ๐ป โ โ de๏ฌned by ๐1 (๐) :=
1 ๐0 (๐) โ ๐พ0 โ
. ๐ 1 โ ๐พ0 ๐0 (๐)
In view of the lemma of H.A. Schwarz, we have ๐1 โ ๐ฎ. As above we set ๐พ1 := ๐1 (0) and if โฃ๐พ1 โฃ < 1, we consider the function ๐2 : ๐ป โ โ de๏ฌned by ๐2 (๐) :=
1 ๐1 (๐) โ ๐พ1 โ
. ๐ 1 โ ๐พ1 ๐1 (๐)
Further, we continue this procedure inductively. Namely, if in the ๐th step a function ๐๐ occurs for which the complex number ๐พ๐ := ๐๐ (0) ful๏ฌlls โฃ๐พ๐ โฃ < 1, we de๏ฌne ๐๐+1 : ๐ป โ โ by 1 ๐๐ (๐) โ ๐พ๐ ๐๐+1 (๐) := โ
๐ 1 โ ๐พ๐ ๐๐ (๐) and continue this procedure in the prescribed way. Let โ0 be the set of all nonnegative integers, and, for each ๐ผ โ โ and ๐ฝ โ โ โช {+โ}, let โ๐ผ,๐ฝ := {๐ โ โ0 : ๐ผ โค ๐ โค ๐ฝ}. Then two cases are possible: (1) The procedure can be carried out without end, i.e., โฃ๐พ๐ โฃ < 1 for each ๐ โ โ0 . (2) There exists an ๐ โ โ0 such that โฃ๐พ๐ โฃ = 1 and, if ๐ > 0, then โฃ๐พ๐ โฃ < 1 for each ๐ โ โ0,๐โ1 . Thus, for each function ๐ โ ๐ฎ, a sequence (๐พ๐ )๐ ๐=0 is associated with ๐. Here we have ๐ = โ (resp. ๐ = ๐) in the ๏ฌrst (resp. second) case. From I. Schurโs paper [15] it is known that the second case occurs if and only if ๐ is a ๏ฌnite Blaschke product of degree ๐. The above procedure is called Schur algorithm and the sequence (๐พ๐ )๐ ๐=0 obtained here is called the sequence of Schur parameters associated with the function ๐, whereas for each ๐ โ โ0,๐ the function ๐๐ is called the ๐th Schur transform of ๐. The symbol ฮ stands for the set of all sequences of Schur parameters associated with functions belonging to ๐ฎ. The following two properties established by I. Schur in [15] determine the particular role which Schur parameters play in the study of functions of class ๐ฎ. (a) Each sequence (๐พ๐ )โ ๐=0 of complex numbers with โฃ๐พ๐ โฃ < 1 for each ๐ โ โ0 belongs to ฮ. Furthermore, for each ๐ โ โ0 , a sequence (๐พ๐ )๐๐=0 of complex numbers with โฃ๐พ๐ โฃ = 1 and โฃ๐พ๐ โฃ < 1 for each ๐ โ โ0,๐โ1 belongs to ฮ. (b) There is a one-to-one correspondence between the sets ๐ฎ and ฮ. Thus, the Schur parameters are independent parameters which completely determine the functions of the class ๐ฎ. In the result of the above considerations we obtain special ordered quadruples [๐, ฮฆ, ฮ, ๐พ] consisting of a measure ๐ โ โณ1+ (๐), a function ฮฆ โ ๐ 0 (๐ป), a function ฮ โ ๐ฎ(๐ป), and Schur parameters ๐พ = (๐พ๐ )๐ ๐=0 โ ฮ, which are interrelated in such
272
V.K. Dubovoy, B. Fritzsche and B. Kirstein
way that each of these four objects uniquely determines the other three ones. For that reason, if one of the four objects is given, we will call the three others associated with it. The main goal of this paper is to derive a criterion which gives an answer to the question when a measure ๐ โ โณ1+ (๐) is a Helson-Szegห o measure (see Section 6). For this reason, we will need the properties of Helson-Szegหo measures listed below (see Theorem 1.2). For more information about Helson-Szegห o measures, we refer the reader, e.g., to [10, Chapter 7], [11, Chapter 5]. Let ๐ โ ๐ซ๐๐ be given by (1.1). Then we consider the Riesz projection ๐+ ๐ which is de๏ฌned by โ (๐+ ๐ )(๐ก) := ๐ ๐ ๐ก๐ , ๐ก โ ๐. ๐โ๐ผโฉโ0 ๐
Let ๐ซ๐๐+ := โ๐๐{๐ก : ๐ โ โ0 } and ๐ซ๐๐โ := โ๐๐{๐กโ๐ : ๐ โ โ} where โ is the set of all positive integers. Then, clearly, ๐+ is the projection which projects the linear space ๐ซ๐๐ onto ๐ซ๐๐+ parallel to ๐ซ๐๐โ . In view of a result due to Fatou (see, e.g., [14, Theorem 1.18]), we will use the following notation: If โ โ ๐ป 2 (๐ป), then the symbol โ stands for the radial boundary values of โ, which exist for ๐-a.e. ๐ก โ ๐ where ๐ is the normalized Lebesgue measure on ๐. If ๐ง โ โ, then the symbol ๐ง โ stands for the complex conjugate of ๐ง. Theorem 1.2. Let ๐ โ โณ1+(๐). Then the following statements are equivalent: (i) ๐ is a Helson-Szegห o measure. (ii) The Riesz projection ๐+ is bounded in ๐ฟ2๐ . (iii) The sequence (๐ก๐ )๐โโค is a (symmetric or nonsymmetric) basis of ๐ฟ2๐ . (iv) ๐ is absolutely continuous with respect to ๐ and there is an outer function ๐๐ โ โ ๐ป 2 (๐ป) such that ๐๐ = โฃโโฃ2 and ( ) dist โโ/โ, ๐ป โ (๐) < 1. Corollary 1.3. Let ๐ โ โณ1+ (๐) be a Helson-Szegห o measure. Then the Schur parameter sequence ๐พ = (๐พ๐ )๐ associated with ๐ is in๏ฌnite, i.e., ๐ = โ holds, and ๐=0 โโ 2 the series ๐=0 โฃ๐พ๐ โฃ converges, i.e., ๐พ โ ๐2 . Proof. Let ๐ โ ๐ฎ(๐ป) be the Schur function associated with ๐. Then it is known (see, e.g., [1, Chapter 3]) that ๐ {โซ } โ 2 (1 โ โฃ๐พ๐ โฃ ) = exp ln(1 โ โฃ๐(๐ก)โฃ2 )๐(๐๐ก) . (1.7) ๐
๐=0
We denote by ฮฆ the function from ๐ 0 (๐ป) associated with ๐. Using (1.4), assumption (iv) in Theorem 1.2, and Fatouโs theorem we obtain 1 โ โฃ๐(๐ก)โฃ2 =
4Re ฮฆ(๐ก) 4โฃโ(๐ก)โฃ2 = โฃฮฆ(๐ก) + 1โฃ2 โฃฮฆ(๐ก) + 1โฃ2
(1.8)
Description of Helson-Szegห o Measures
273
for ๐-a.e. ๐ก โ ๐. Thus, ln(1 โ โฃ๐(๐ก)โฃ2 ) = ln 4 + 2 ln โฃโ(๐ก)โฃ โ 2 ln โฃฮฆ(๐ก) + 1โฃ.
(1.9)
In view of Re ฮฆ(๐) > 0 for each ๐ โ ๐ป, the function ฮฆ is outer. Hence, ln โฃฮฆ + 1โฃ โ ๐ฟ1๐ (see, e.g., [14, Theorem 4.29 and Theorem 4.10]). Taking into account condition (iv) in Theorem 1.2, we obtain โ โ ๐ป 2 (๐ป). Thus, we infer from (1.9) that ln(1 โ โฃ๐(๐ก)โฃ2 ) โ ๐ฟ1๐ . Now the assertion follows from (1.7). โก Remark 1.4. Let ๐ โ โณ1+ (๐), and let the Lebesgue decomposition of ๐ with respect to ๐ be given by ๐(๐๐ก) = ๐ฃ(๐ก)๐(๐๐ก) + ๐๐ (๐๐ก),
(1.10)
where ๐๐ stands for the singular part of ๐ with respect to ๐. Then the relation Re ฮฆ = ๐ฃ holds ๐-a.e. on ๐. The identity (1.9) has now the form ln(1 โ โฃ๐(๐ก)โฃ2 ) = ln 4 + ln ๐ฃ(๐ก) โ 2 ln โฃฮฆ(๐ก) + 1โฃ for ๐-a.e. ๐ก โ ๐. From this and (1.7) now it follows a well-known result, namely, that ln ๐ฃ โ ๐ฟ1๐ (i.e., ๐ is a Szegหo measure) if and only if ๐ = โ and ๐พ โ ๐2 . In particular, a Helson-Szegห o measure is also a Szegหo measure. We ๏ฌrst wish to mention that our interest in describing the class of HelsonSzegหo measures in terms of Schur parameters was initiated by conversations with L.B. Golinskii and A.Ya. Kheifets who studied related questions in joint research with F. Peherstorfer and P.M. Yuditskii (see [7], [9], [12]). In Section 6 we will comment on some results in [7] which are similar to our own. The above-mentioned problem is of particular interest, even on its own. Solutions to this problem promise important applications and new results in scattering theory for CMV matrices (see [7], [9], [11]) and in nonlinear Fourier analysis (see [17]). Our approach to the description of Helson-Szegห o measures di๏ฌers from the one in [7] in that we investigate this question for CMV matrices in another basis (see [3, De๏ฌnition 2.2., Theorem 2.13]), namely the one for that CMV matrices have the full GGT representation (see Simon [16, pp. 261โ262, Remarks and Historical Notes]).
2. A unitary colligation associated with a Borelian probability measure on the unit circle The starting point of this section is the observation that a given Schur function ฮ โ ๐ฎ(๐ป) can be represented as characteristic function of some contraction in a Hilbert space. That means that there exists a separable complex Hilbert space โ and bounded linear operators ๐ : โ โ โ, ๐น : โ โ โ, ๐บ : โ โ โ, and ๐ : โ โ โ such that the block operator ( ) ๐ ๐น ๐ := :โโโ โโโโ (2.1) ๐บ ๐
274
V.K. Dubovoy, B. Fritzsche and B. Kirstein
is unitary and, moreover, that for each ๐ โ ๐ป the equality ฮ(๐) = ๐ + ๐๐บ(๐ผ โ ๐๐ )โ1 ๐น,
(2.2)
is ful๏ฌlled. Note that in (2.1) the complex plane โ is considered as the onedimensional complex Hilbert space with the usual inner product ( ) ๐ง, ๐ค โ = ๐ง โ ๐ค, ๐ง, ๐ค โ โ. The unitarity of the operator ๐ implies that the operator ๐ is contractive (i.e., โฅ๐ โฅ โค 1). Thus, for all ๐ โ ๐ป the operator ๐ผ โ ๐๐ is boundedly invertible. The unitarity of the operator ๐ means that the ordered tuple โณ := (โ, โ, โ; ๐, ๐น, ๐บ, ๐)
(2.3)
is a unitary colligation. In view of (2.2), the function ฮ is the characteristic operator function of the unitary colligation โณ. For a detailed treatment of unitary colligations and their characteristic functions we refer the reader to the landmark paper [2]. The following subspaces of โ will play an important role in the sequel โ๐ :=
โ โ
๐ ๐ ๐น (โ),
โ โ
โ๐ :=
๐=0
(๐ โ )๐ ๐บโ (โ).
(2.4)
๐=0
โ By the symbol โ ๐=0 ๐ด๐ we mean the smallest closed subspace generated by the subsets ๐ด๐ of the corresponding spaces. The subspaces โ๐ and โ๐ are called the subspaces of controllability and observability, respectively. We note that the unitary operator ๐ can be chosen such that โ = โ๐ โจ โ๐
(2.5)
holds. In this case the unitary colligation โณ is called simple. The simplicity of a unitary colligation means that there does not exist a nontrivial invariant subspace of โ on which the operator ๐ induces a unitary operator. Such kind of contractions ๐ are called completely nonunitary. Proposition 2.1. Let ๐ โ โณ1+ (๐) be a Szegห o measure. Let ฮ be the function belonging to ๐ฎ(๐ป) which is associated with ๐ and let โณ be the simple unitary colligation the characteristic operator function of which coincides with ฮ. Then the spaces โโฅ ๐ := โ โ โ๐ ,
โโฅ ๐ := โ โ โ๐
(2.6)
are nontrivial. Proof. Let ๐พ = (๐พ๐ )๐ ๐=0 โ ฮ be the Schur parameter sequence of ฮ. Then from Corollary 1.3 we infer that ๐ = โ and that ๐พ โ ๐2 . In this case it was proved in [3, Chapter 2] that both spaces (2.6) are nontrivial. โก โฅ Because of (2.4) and (2.6) it follows that the subspace โโฅ ๐ (resp. โ๐ ) is invariant with respect to ๐ (resp. ๐ โ ). It can be shown (see [3, Theorem 1.6]) that
๐๐ := Rstr.โโฅ ๐ ๐
and ๐๐ โ := Rstr.โโฅ ๐โ ๐
Description of Helson-Szegห o Measures
275
are both unilateral shifts. More precisely, ๐๐ (resp. ๐๐ โ ) is exactly the maximal unilateral shift contained in ๐ (resp. ๐ โ ). This means that an arbitrary invariant subspace with respect to ๐ (resp. ๐ โ ) on which ๐ (resp. ๐ โ ) induces a unilateral โฅ shift is contained in โโฅ ๐ (resp. โ๐ ). 1 Let ๐ โ โณ+ (๐). Then our subsequent considerations are concerned with the investigation of the unitary operator ๐๐ร : ๐ฟ2๐ โ ๐ฟ2๐ which is de๏ฌned for each ๐ โ ๐ฟ2๐ by (๐๐ร ๐ )(๐ก) := ๐กโ โ
๐ (๐ก),
๐ก โ ๐.
(2.7)
Denote by ๐ the embedding operator of โ into ๐ฟ2๐ , i.e., ๐ : โ โ ๐ฟ2๐ is such that for each ๐ โ โ the image ๐ (๐) of ๐ is the constant function on ๐ with value ๐. Denote by โ๐ the subspace of ๐ฟ2๐ which is generated by the constant functions and denote by 1 the constant function on ๐ with value 1. Then obviously ๐ (โ) = โ๐ and ๐ (1) = 1. We consider the subspace โ๐ := ๐ฟ2๐ โ โ๐ . Denote by ๐๐ร =
(
๐ร ๐บร
๐นร ๐ร
)
the block representation of the operator ๐๐ร with respect to the orthogonal decomposition ๐ฟ2๐ = โ๐ โ โ๐ . Then (see [3, Section 2.8]) the following result holds. Theorem 2.2. Let ๐ โ โณ1+ (๐). De๏ฌne ๐๐ := ๐ ร , ๐น๐ := ๐น ร ๐ , ๐บ๐ := ๐ โ ๐บร , and ๐๐ := ๐ โ ๐ ร ๐ . Then โณ๐ := (โ๐ , โ, โ; ๐๐, ๐น๐ , ๐บ๐ , ๐๐ )
(2.8)
is a simple unitary colligation the characteristic function ฮโณ๐ of which coincides with the Schur function ฮ associated with ๐. In view of Theorem 2.2, the operator ๐๐ is a completely nonunitary contraction and if the function ฮฆ is given by (1.4) with ๐ฝ = 0, then from (1.6) it follows ฮฆ(๐) โ 1 ๐ฮโณ๐ (๐) = , ๐ โ ๐ป. ฮฆ(๐) + 1 De๏ฌnition 2.3. Let ๐ โ โณ1+ (๐). Then the simple unitary colligation given by (2.8) is called the unitary colligation associated with ๐. Let ๐ โ โณ1+ (๐) be a Szegห o measure and let ๐พ = (๐พ๐ )๐ ๐=0 โ ฮ be the Schur parameter sequence associated with ๐. Then Remark 1.4 shows that ๐ = โ and ๐พ โ ๐2 . Furthermore, we use for all integers ๐ the setting ๐๐ : ๐ โ โ de๏ฌned by ๐๐ (๐ก) := ๐ก๐ .
(2.9)
276
V.K. Dubovoy, B. Fritzsche and B. Kirstein
Thus, we have ๐โ๐ = (๐๐ร )๐ 1, where ๐๐ร is the operator de๏ฌned by (2.7). We consider then the system {๐0 , ๐โ1 , ๐โ2 , . . .}. By the Gram-Schmidt orthogonalization method in the space ๐ฟ2๐ we get a unique sequence (๐๐ )โ ๐=0 of polynomials, where ๐๐ (๐ก) = ๐ผ๐,๐ ๐กโ๐ + ๐ผ๐,๐โ1 ๐กโ(๐โ1) + โ
โ
โ
+ ๐ผ๐,0 , ๐ก โ ๐, such that the conditions ๐ ๐ โ โ ๐๐ = (๐๐ร )๐ 1, ๐=0
๐=0
) ( ร ๐ (๐๐ ) 1, ๐๐ ๐ฟ2 > 0, ๐
๐ โ โ0 ,
๐ โ โ0 ,
(2.10)
(2.11)
are satis๏ฌed. We note that the second condition in (2.11) is equivalent to ( ) 1, ๐0 ๐ฟ2 > 0 and ๐ ( ร ) (2.12) ๐๐ ๐๐โ1 , ๐๐ ๐ฟ2 > 0, ๐ โ โ0 . ๐
In particular, since ๐(๐) = 1 holds, from the construction of ๐0 we see that ๐0 = 1.
(2.13)
We consider a simple unitary colligation โณ๐ of the type (2.8) associated with the measure ๐. The controllability and observability spaces (2.4) associated with the unitary colligation โณ๐ have the forms โ๐,๐ =
โ โ
๐๐๐ ๐น๐ (โ) and โ๐,๐ =
๐=0
โ โ
(๐๐โ )๐ ๐บโ๐ (โ),
(2.14)
๐=0
respectively. Let the sequence of functions (๐โฒ๐ )โ ๐=1 be de๏ฌned by ๐โฒ๐ := ๐๐๐โ1 ๐น๐ (1), In view of the formulas ๐ โ ๐=0
(๐๐ร )๐ 1
๐ โ โ.
(๐โ1 ) โ ๐ = (๐๐ ) ๐น๐ (1) โ โ๐ ,
(2.15)
๐ โ โ,
(2.16)
๐=0
it can be seen that the sequence (๐๐ )โ ๐=1 can be obtained by applying the GramSchmidt orthonormalization procedure to (๐โฒ๐ )โ ๐=1 with additional consideration of the normalization condition (2.12). Thus, we obtain the following result: Theorem 2.4. The system (๐๐ )โ ๐=1 of orthonormal polynomials is a basis in the space โ๐,๐ , and ) (โ โ โ๐,๐ = (๐กโ )๐ โ โ๐ . (2.17) ๐=0
This system can be obtained in the result of the application of the Gram-Schmidt orthogonalization procedure to the sequence (2.15) taking into account the normalization condition (2.12).
Description of Helson-Szegห o Measures
277
Remark 2.5. Analogously to (2.17) we have the equation (โ ) โ โ๐,๐ = ๐ก๐ โ โ ๐ .
(2.18)
๐=0
If ๐ is a contraction acting on some Hilbert space โ, then we use the setting ( ) ๐ฟ๐ := dim ๐๐ resp. ๐ฟ๐ โ := dim ๐๐ โ , where ๐๐ := ๐ท๐ (โ) (resp. ๐๐ โ := ๐ท๐ โ (โ) ) is โ the closure of the range of the โ defect operator ๐ท๐ := ๐ผโ โ ๐ โ ๐ (resp. ๐ท๐ โ := ๐ผโ โ ๐ ๐ โ ). In view of (2.6), let โโฅ ๐,๐ := โ๐ โ โ๐,๐ ,
โโฅ ๐,๐ := โ๐ โ โ๐,๐ .
If ๐ is a Szegหo measure, then we have ๐ฟ2๐ โ
โ โ
( ) ( ) ๐โ๐ = โ๐ โ โ๐ โ โ๐,๐ โ โ๐ = โ๐ โ โ๐,๐ = โโฅ ๐,๐ โ= {0}.
(2.19)
๐=0
So from Proposition 2.1 we obtain the known result that in this case the system 2 (๐๐ )โ ๐=0 is not complete in the space ๐ฟ๐ . In our case ( ) ๐๐๐ := Rstr.โโฅ ๐๐ resp. ๐๐๐โ := Rstr.โโฅ ๐โ ๐,๐ ๐,๐ ๐ is the maximal unilateral shift contained in ๐๐ (resp. ๐๐โ ) (see [3, Theorem 1.6]). In view of ๐ฟ๐๐ = ๐ฟ๐๐โ = 1 the multiplicity of the unilateral shift ๐๐๐ (resp. ๐๐๐โ ) is equal to 1. Proposition 2.6. The orthonormal system of the polynomials (๐๐ )โ ๐=0 is noncomplete in ๐ฟ2๐ if and only if the contraction ๐๐ (resp. ๐๐โ ) contains a maximal unilateral shift ๐๐๐ (resp. ๐๐๐โ ) of multiplicity 1.
3. On the connection between the Riesz projection ๐ท+ and the ๐ฑ projection ํ๐, ๐ฒ which projects ๐ณ๐ onto ๐ณ๐, ๐ฒ parallel to ๐ณ๐, ๐ฑ Let ๐ โ โณ1+ ( ๐ ). We consider the unitary colligation ฮ๐ of type (2.8) which is associated with the measure ๐. Then the following statement is true. Theorem 3.1. Let ๐ โ โณ1+ ( ๐ ) be a Szegห o measure. Then the Riesz projection ๐+ ๐ 2 is bounded in ๐ฟ๐ if and only if the projection ๐ซ๐, ๐ which projects โ๐ onto โ๐, ๐ parallel to โ๐, ๐ is bounded. Proof. For each ๐ โ โ0 we consider particular subspaces of the space โ๐ , namely (๐)
โ๐, ๐ :=
๐ โ ๐=0
๐๐๐ ๐น๐ ( โ ) ,
(๐)
โ๐, ๐ :=
๐ โ ( ๐=0
๐๐โ
)๐
๐บโ๐ ( โ ) ,
(3.1)
278
V.K. Dubovoy, B. Fritzsche and B. Kirstein
and (๐)
(๐)
โ๐(๐) := โ๐, ๐ โจ โ๐, ๐ ,
๐ฟ๐, ๐ := โ๐(๐) โ โ๐ .
(3.2)
Then from (2.15), (2.17), and (2.18) we obtain the relations โ๐, ๐ = โ๐ = (๐) โ๐, ๐
=
โ โ ๐=0 โ โ
โ๐, ๐ =
โ๐(๐) ,
๐ฟ2๐ =
๐=0 ( ๐ โ
) โ ๐
(๐ก )
(๐) โ๐, ๐
โ โ๐ ,
๐=0
and
( โ๐(๐)
(๐)
โ๐, ๐ ,
=
=
โ โ ๐=0 โ โ
๐ก
๐
โ โ๐ ,
(๐)
(3.3)
๐ฟ๐, ๐ ,
(3.4)
๐=0 ( ๐ โ
) ๐ก
๐
โ โ๐ ,
(3.5)
๐=0
)
๐ โ
โ๐, ๐ ,
๐ โ
๐ฟ๐, ๐ =
๐=โ๐
๐ก๐ .
(3.6)
๐=โ๐
Since ๐ is a Szegหo measure, for each ๐ โ โ0 we obtain (๐)
โ๐(๐) = โ๐, ๐
ห +
(๐)
โ๐, ๐ .
(3.7) (๐)
Suppose now that the Riesz projection ๐+ is bounded in ๐ฟ2๐ . Let โ โ โ๐ . Then, because of (3.6), the function โ has the form โ(๐ก) =
๐ โ
๐ ๐ ๐ก๐ ,
๐ก โ ๐.
(3.8)
๐=โ๐
From (3.5) and (3.7) we obtain โ = โ ๐ + โ๐ ,
(3.9)
where (๐)
(๐)
โ๐ โ โ๐, ๐ , โ๐ ( ๐ก ) = ๐0, ๐ +
โ๐ โ โ๐, ๐ , ๐ โ
๐โ๐ ๐กโ๐ ,
โ๐ ( ๐ก ) = ๐0, ๐ +
๐=1
๐ โ
๐ ๐ ๐ก๐ ,
(3.10)
๐=1
and ๐0 = ๐0, ๐ + ๐0, ๐ . Observe that ๐ ๐ซ๐, ๐ โ = โ๐ .
(3.11)
โ = โ + + โโ ,
(3.12)
On the other hand, we have
Description of Helson-Szegห o Measures
279
where, for each ๐ก โ ๐, โ+ ( ๐ก ) = ( ๐+ (โ) ) ( ๐ก ) =
๐ โ
๐ ๐ ๐ก๐ ,
โโ ( ๐ก ) =
๐=0
๐ โ
๐โ๐ ๐กโ๐ .
(3.13)
๐=1
For a polynomial โ๐ of the type (3.10) we set โ๐ ( 0 ) := ๐0, ๐ .
(3.14)
Then from (3.10)โ(3.14) we infer ๐ ๐+ โ = โ+ = โ๐ + โ๐ ( 0 ) โ
1 = ๐ซ๐, ๐ โ + โ๐ ( 0 ) โ
1 .
(3.15)
โ โ = โ๐ โ โ ๐ ( 0 ) โ
1 ,
(3.16)
Observe that where in view of (3.5) we see that โ๐ โฅ 1 . Let ๐โ๐ be the orthoprojection from ๐ฟ2๐ onto โ๐ . Then from (3.16) it follows that โ๐ ( 0 ) โ
1 = ๐โ๐ โโ = ๐โ๐ ( ๐ผ โ ๐+ ) โ . Inserting this expression into (3.15) we get ๐ ๐ซ๐, ๐ โ = ๐+ โ โ ๐โ๐ ( ๐ผ โ ๐+ ) โ .
From this and (3.4) it follows that the boundedness of the projection ๐+ in ๐ฟ2๐ ๐ implies the boundedness of the projection ๐ซ๐, ๐ in โ๐ . ๐ Conversely, suppose that the projection ๐ซ๐, ๐ is bounded in โ๐ . Let ๐ โ ๐ฟ๐, ๐ . (๐)
We denote by ๐โ(๐) the orthogonal projection from ๐ฟ2๐ onto โ๐ . We set ๐
โ := ๐โ(๐) ๐ ๐
and use for โ the notations introduced in (3.8)โ(3.10). Let ๐ = ๐+ + ๐ โ , where ๐+ := ๐+ ๐ . Then ๐ = ๐โ๐ ๐ + โ = ๐โ๐ ๐ + โ๐ + โ๐ . This implies ๐+ ๐ = ๐โ๐ ๐ + โ๐ + โ๐ ( 0 ) โ
1 . This means ๐ ๐+ ๐ = ๐โ๐ ๐ + ๐ซ๐, ๐ ๐โ(๐) ๐ + โ๐ ( 0 ) โ
1 . ๐
The mapping โ๐ โ โ๐ ( 0 ) is a linear functional on the set ๐ซ๐๐โค0 := โ๐๐{๐กโ๐ : ๐ โ โ0 }.
(3.17)
280
V.K. Dubovoy, B. Fritzsche and B. Kirstein
Since ๐ is a Szegหo measure, the Szegหo-Kolmogorov-Krein Theorem (see, e.g., [14, Theorem 4.31]) implies that this functional is bounded in ๐ฟ2๐ on the set ๐ซ๐๐ โค0 . Thus, there exists a constant ๐ถ โ ( 0, +โ ) such that โฅโ๐ ( 0 ) โ
1โฅ๐ฟ2 = โฃโ๐ ( 0 )โฃ ๐
โค ๐ถ โ
โฅโ๐ โฅ 1( ) 1 1 1 ๐ = ๐ถ โ
1 ๐ผ โ ๐ซ๐, ๐ โ1 1 1 1 ๐ 1 โค ๐ถ โ
1๐ผ โ ๐ซ๐, ๐ 1 โ
โฅโโฅ 1 1 1 ๐ 1 โค ๐ถ โ
1๐ผ โ ๐ซ๐, ๐ 1 โ
โฅ๐ โฅ .
(3.18)
From (3.17) and (3.18) we get
1 1 1 1 1 ๐ 1 1 ๐ 1 ๐ โฅ๐+ ๐ โฅ โค โฅ๐โ๐ ๐ โฅ + 1๐ซ๐, (๐) ๐ 1 + ๐ถ โ
1๐ผ โ ๐ซ ๐ โ๐ ๐, ๐ 1 โ
โฅ๐ โฅ 1 1 1 1 1 1 ๐ 1 ๐ 1 โค โฅ๐ โฅ + 1๐ซ๐, ๐ 1 โ
โฅ๐ โฅ + ๐ถ โ
1๐ผ โ ๐ซ๐, ๐ 1 โ
โฅ๐ โฅ .
Now considering the limit as ๐ โ โ and taking into account (3.4), we see that ๐ the boundedness of the projection ๐ซ๐, ๐ implies the boundedness of the Riesz 2 โก projection ๐+ in ๐ฟ๐ .
4. On the connection of the Riesz projection ๐ท+ with the orthoโฅ gonal projections from ๐ณ๐ onto the subspaces ๐ณโฅ ๐, ๐ฑ and ๐ณ๐, ๐ฒ Let ๐ โ โณ1+ ( ๐ ) be a Szegหo measure. As we did earlier, we consider the simple unitary colligation ฮ๐ of type (2.8) which is associated with the measure ๐. As was previously mentioned, we then have โโฅ ๐, ๐ โ= { 0 }
โโฅ ๐, ๐ โ= { 0 } .
and
We denote by ๐โโฅ and ๐โโฅ the orthogonal projections from โ๐ onto โโฅ ๐, ๐ and ๐, ๐ ๐, ๐
โโฅ ๐, ๐ , respectively. (๐)
Let โ โ โ๐ . Along with the decomposition (3.9) we consider the decomposition โโฅ โ=ห โ๐ + ห ๐ , where (๐) ห โ๐ โ โ๐,๐
and
(๐) (๐) ห โโฅ ๐ โ โ๐ โ โ๐, ๐ .
(4.1)
Description of Helson-Szegห o Measures
281
(๐)
From the shape (3.5) of the subspace โ๐, ๐ and the polynomial structure of the orthonormal basis ( ๐๐ )โ ๐=0 of the subspace โ๐, ๐ , it follows that ห โ. โโฅ ๐ = ๐โโฅ ๐, ๐ (๐) Since โ๐ (see (3.9)) and ห โ๐ belong to โ๐, ๐ , we get ๐ ๐ ห โ = ๐โโฅ โ๐, = ๐โโฅ ๐ซ๐, โโฅ ๐ = ๐โโฅ ๐ โ = ๐ต๐, ๐ ๐ซ๐, ๐ โ , ๐, ๐ ๐, ๐ ๐, ๐
(4.2)
๐ต๐,๐ := Rstr. โ๐,๐ ๐โโฅ : โ๐,๐ โ โโฅ ๐,๐ , ๐,๐
(4.3)
where i.e., we consider ๐ต๐, ๐ as an operator acting between the spaces โ๐, ๐ and โโฅ ๐, ๐ . ๐ Theorem 4.1. Let ๐ โ โณ1+ ( ๐ ) be a Szegห o measure. Then the projection ๐ซ๐, ๐ is bounded in โ๐ if and only if the operator ๐ต๐, ๐ de๏ฌned in (4.3) is boundedly invertible. (๐)
โ1 Proof. Suppose ๏ฌrst that ๐ต๐, ๐ has a bounded inverse ๐ต๐, ๐ . Then for โ โ โ๐ , in view of (4.1) and (4.2), we have ๐ โ1 โ. ๐ซ๐, ๐ โ = ๐ต๐, ๐ ๐โโฅ ๐, ๐ ๐ If ๐ โ โ, this gives us the boundedness of the projection ๐ซ๐, ๐ in โ๐ . (๐)
(๐)
๐ Conversely, suppose that the projection ๐ซ๐, ๐ is bounded in โ๐ . If โ โ โ๐ โ
โ๐, ๐ , then the decomposition (4.1) provides us โ=ห โโฅ ๐ and identity (4.2) yields ๐ โ = ๐ต๐, ๐ ๐ซ๐, ๐โ .
(4.4)
๐ Since ๐ซ๐, ๐ is bounded in โ๐ , Theorem 3.1 implies that the Riesz projection ๐+ is bounded in ๐ฟ2๐ . Then it follows from condition (iii) in Theorem 1.2 that
โ๐, ๐ โฉ โ๐, ๐ = { 0 } . Thus, from the shape (4.3) of the operator ๐ต๐, ๐ , we infer that ( โฅ) ๐ = โ๐ . ker ๐ต๐, ๐ = { 0 } and ๐ซ๐, ๐ โ๐ Now equation (4.4) can be rewritten in the form ๐ โ1 ๐ต๐, ๐ โ = ๐ซ๐, ๐ โ ,
(๐)
โ โ โ๐(๐) โ โ๐, ๐ .
The limit ๐ โ โ, (3.2) and (3.4) give us the desired result.
โก
The combination of Theorem 4.1 with Theorem 3.1 leads us to the following result.
282
V.K. Dubovoy, B. Fritzsche and B. Kirstein
Theorem 4.2. Let ๐ โ โณ1+ ( ๐ ) be a Szegห o measure. Then the Riesz projection ๐+ is bounded in ๐ฟ2๐ if and only if the operator ๐ต๐, ๐ de๏ฌned in (4.3) is boundedly invertible. Let ๐ โ ๐ซ๐๐ be given by (1.1). Along with the Riesz projection ๐+ , we consider the projection ๐โ , which is de๏ฌned by: โ ( ๐โ ) ( ๐ก ) := ๐ ๐ ๐ก๐ , ๐กโ๐. โ๐โ๐ผโฉโ0
Obviously, ๐โ = ๐+ ๐ ,
and
๐+ = ๐โ ๐ .
Thus, the boundedness of one of the projections ๐+ and ๐โ in ๐ฟ2๐ implies the boundedness of the other one. It is readily checked that the change from the projection ๐+ to ๐โ is connected with changing the roles of the spaces โ๐, ๐ and โ๐, ๐ . Thus we obtain the following result, which is dual to Theorem 4.2. Theorem 4.3. Let ๐ โ โณ1+ ( ๐ ) be a Szegห o measure. Then the Riesz projection ๐+ is bounded in ๐ฟ2๐ if and only if the operator ๐ต๐, ๐ : โ๐, ๐ โ โโฅ ๐, ๐ de๏ฌned by ๐ต๐, ๐ โ := ๐โโฅ โ ๐, ๐
(4.5)
is boundedly invertible. Here the symbol ๐โโฅ stand for the orthogonal projection ๐, ๐ from โ๐ onto โโฅ ๐, ๐ .
5. Matrix representation of the operator ๐ฉ๐, ๐ฒ in terms of the Schur parameters associated with the measure ๐ Let ๐ โ โณ1+ ( ๐ ) be a Szegหo measure. We consider the simple unitary colligation ฮ๐ of the type (2.8) which is associated with the measure ๐. In this case we have (see Section 2) โโฅ ๐, ๐ โ= { 0 }
and
โโฅ ๐, ๐ โ= { 0 }
The operator ๐ต๐, ๐ acts between the subspaces โ๐, ๐ and โโฅ ๐, ๐ . According to the matrix description of the operator ๐ต๐, ๐ we consider particular orthogonal bases in these subspaces. In the subspace โ๐, ๐ we have already considered one such โ basis, namely the basis consisting of the trigonometric polynomials ( ๐๐ )๐=1 (see Theorem 2.4). Regarding the construction of an orthonormal basis in โโฅ ๐, ๐ , we โ ๏ฌrst complete the system ( ๐๐ )๐=1 to an orthonormal basis in โ๐ . This procedure is described in more detail in [3]. We consider the orthogonal decomposition โ๐ = โ๐,๐ โ โโฅ ๐,๐ .
(5.1)
Description of Helson-Szegห o Measures
283
ห 0 the wandering subspace which generates the subspace associated Denote by ๐ ห 0 = 1 and, since with the unilateral shift ๐๐๐โ . Then (see Proposition 2.6) dim ๐ ๐๐๐โ is an isometric operator, we have (๐๐ร )โ . ๐๐๐โ = Rstr.โโฅ ๐,๐
(5.2)
Consequently, โโฅ ๐,๐ =
โ โ ๐=0
ห0) = ๐๐๐๐โ (๐
โ โ
ห0) = (๐๐โ )๐ (๐
๐=0
โ โ
ห 0 ). [(๐๐ร )โ ]๐ (๐
(5.3)
๐=0
ห 0 which ful๏ฌlls There exists (see [3, Corollary 1.10]) a unique unit function ๐1 โ ๐ ( โ ) ๐บ๐ (1), ๐1 ๐ฟ2 > 0. (5.4) ๐
Because of (5.2), (5.3), and (5.4) it follows that the sequence (๐๐ )โ ๐=1 , where ๐๐ := [(๐๐ร )โ ]๐โ1 ๐1 ,
๐ โ โ,
(5.5)
is the unique orthonormal basis of the space โโฅ ๐,๐ which satis๏ฌes the conditions ( โ ) (5.6) ๐บ๐ (1), ๐1 ๐ฟ2 > 0, ๐๐+1 = (๐๐ร )โ ๐๐ , ๐ โ โ, ๐
or equivalently ( โ ) ๐บ๐ (1), ๐1 ๐ฟ2 > 0, ๐
๐๐+1 (๐ก) = ๐ก๐ โ
๐1 (๐ก), ๐ก โ ๐,
๐ โ โ.
(5.7)
According to the considerations in [3] we introduce the following notion. De๏ฌnition 5.1. The constructed orthonormal basis ๐0 , ๐1 , ๐2 , . . . ; ๐1 , ๐2 , . . .
(5.8)
in the space ๐ฟ2๐ which satis๏ฌes the conditions (2.11) and (5.6) is called the canonical orthonormal basis in ๐ฟ2๐ . Note that the analytic structure of the system (๐๐ )โ ๐=1 is described in the paper [5]. Obviously, the canonical orthonormal basis (5.8) in ๐ฟ2๐ is uniquely determined by the conditions (2.11) and (5.6). Here the sequence (๐๐ )โ ๐=0 is an orthonormal system of polynomials (depending on ๐กโ ). The orthonormal system (๐๐ )โ ๐=1 is built with the aid of the operator ๐๐ร from the function ๐1 (see (5.5)) in a similar way as the system (๐๐ )โ ๐=0 was built from (the function )โ๐0 (see (2.10) and (2.11)). The only di๏ฌerence is that the system [(๐๐ร )โ ]๐ ๐1 ๐=0 is orthonormal, whereas )โ ( in the general case the system (๐๐ร )๐ ๐0 ๐=0 is not orthonormal. In this respect, the sequence (๐๐ )โ ๐=1 can be considered as a natural completion of the system of 2 orthonormal polynomials (๐๐ )โ ๐=0 to an orthonormal basis in ๐ฟ๐ .
284
V.K. Dubovoy, B. Fritzsche and B. Kirstein
Remark 5.2. The orthonormal system ๐1 , ๐2 , . . . ; ๐1 , ๐2 , . . .
(5.9)
is an orthonormal basis in the space โ๐ . We will call it the canonical orthonormal basis in โ๐ . It is well known (see, e.g., Brodskii [2]) that one can consider simultaneously together with the simple unitary colligation (2.8) the adjoint unitary colligation ห := (โ๐ , โ, โ; ๐ โ, ๐บโ , ๐น โ , ๐ โ ) (5.10) โณ ๐
๐
๐
๐
๐
which is also simple. Its characteristic function ฮโณ ห is for each ๐ง โ ๐ป given by ๐ โ โ ฮโณ ห (๐ง) = ฮโณ๐ (๐ง ). ๐
We note that the unitary colligation (5.10) is associated with the operator (๐๐ร )โ . It can be easily checked that the action of (๐๐ร )โ is given for each ๐ โ ๐ฟ2๐ by [(๐๐ร )โ ๐ ](๐ก) = ๐ก โ
๐ (๐ก),
๐ก โ ๐.
If we replace the operator ๐๐ร by (๐๐ร )โ in the preceding considerations, which have lead to the canonical orthonormal basis (5.8), then we obtain an orthonormal basis of the space ๐ฟ2๐ which consists of two sequences (๐ห๐ )โ ๐=0
and (๐ห๐ )โ ๐=1
(5.11)
of functions. From our treatments above it follows that the orthonormal basis (5.11) is uniquely determined by the following conditions: (a) The sequence (๐ห๐ )โ Gram-Schmidt orthogo๐=0 arises from the ( result of )the โ nalization procedure of the sequence [(๐๐ร )โ ]๐ 1 ๐=0 and additionally taking into account the normalization conditions ) ( ร โ๐ [(๐๐ ) ] 1, ๐ห๐ ๐ฟ2 > 0, ๐ โ โ0 . ๐
(b) The relations ) ( ๐น๐ (1), ๐ห1 ๐ฟ2 > 0 ๐
and ๐ห๐+1 = ๐๐ร ๐ห๐ ,
๐ โ โ,
hold. It can be easily checked that ๐ห๐ = ๐โ๐ , ๐ โ โ0 , and
๐ห๐ = ๐๐โ , ๐ โ โ. According to the paper [3] we introduce the following notion.
De๏ฌnition 5.3. The orthogonal basis ๐โ0 , ๐โ1 , ๐โ2 , . . . ; ๐1โ , ๐2โ , . . .
(5.12)
is called the conjugate canonical orthonormal basis with respect to the canonical orthonormal basis (5.8).
Description of Helson-Szegห o Measures
285
We note that ๐0 = ๐โ0 = 1. Similarly as (2.16) the identity (๐โ1 ) ๐ โ โ ร โ ๐ โ ๐ โ [(๐๐ ) ] 1 = (๐๐ ) ๐บ๐ (1) โ โ๐ ๐=0
(5.13)
๐=0
can be veri๏ฌed. Thus, โ๐,๐ = โโฅ ๐,๐ =
โ โ ๐=1 โ โ
๐๐ ,
โ๐,๐ = โโฅ ๐,๐ =
๐๐ ,
๐=1
โ โ ๐=1 โ โ
๐โ๐ ,
(5.14)
๐๐โ .
(5.15)
๐=1
In [3, Chapter 3] the unitary operator ๐ฐ was introduced which maps the elements of the canonical basis (5.8) onto the corresponding elements of the conjugate canonical basis (5.12). More precisely, we consider the operator ๐ฐ๐ ๐๐ = ๐โ๐ ,
๐ โ โ0 ,
๐ฐ๐ ๐๐ = ๐๐โ ,
and
The operator ๐ฐ๐ is related to the conjugation operator in and if โ โ โ โ ๐= ๐ผ๐ ๐๐ + ๐ฝ๐ ๐๐ , ๐=0
then โ
๐ =
โ โ
๐ผโ๐ ๐โ๐
+
๐=0
โ โ
๐ โ โ.
๐ฟ2๐ .
(5.16)
Namely, if ๐ โ ๐ฟ2๐
๐=1
๐ฝ๐โ ๐๐โ
=
๐=1
โ โ
๐ผโ๐ ๐ฐ๐๐
+
๐=0
From (5.16) it follows that
๐ฐ๐ : โ๐ โโ โ๐ ,
โ โ
๐ฝ๐โ ๐ฐ๐๐ .
๐=1
๐ฐ๐ (1) = 1 .
Let ๐ฐโ๐ := Rstr. โ๐ ๐ฐ๐ .
(5.17)
Then, obviously, ๐ฐโ๐ ๐๐ = ๐โ๐
๐ฐโ๐ ๐๐ = ๐๐โ ,
and
๐โโ.
(5.18)
โ ( ๐๐โ )๐=1
is an orthonormal basis in the space โโฅ ๐, ๐ . This sysโฅ special orthonormal basis of the space โ๐, ๐ mentioned
Clearly, the system tem will turn out to be the at the beginning of this section. Thus, the matrix representation of the operator ๐ต๐, ๐ : โ๐, ๐ โโ โโฅ ๐, ๐ will be considered with respect to the orthonormal bases ( ๐๐ )โ ๐=1 of the spaces โ๐, ๐ and
โโฅ ๐, ๐ ,
and respectively. Let ) ( โ โ ๐ซ
๐ฌ
(๐๐โ )โ ๐=1
(5.19)
(5.20)
286
V.K. Dubovoy, B. Fritzsche and B. Kirstein
be the matrix representation of the operator ๐ฐโ๐ with respect to the canonical basis (5.9) of the space โ๐ . Then, from (5.18) we infer that the columns ( ) ( ) โ โ and ๐ฌ ๐ซ of the block-matrix (5.20) are the coe๏ฌcients in the series developments of ๐โ๐ and ๐๐โ with respect to the canonical basis (5.9). If โ โ โ๐ then clearly โ= ๐โโฅ ๐, ๐
โ โ
( โ , ๐๐โ ) ๐๐โ .
(5.21)
๐=1
Thus, the matrix representation of the operator ๐โโฅ considered as an opera๐, ๐ tor acting between โ๐ and โโฅ ๐, ๐ equipped with the orthonormal bases (5.9) and โ โ ( ๐๐ )๐=1 has the form ( โโ , ๐ฌ โ ) . From this and the shape (4.5) of the operator ๐ต๐, ๐ , we obtain the following result. o measure. Then the matrix of the operTheorem 5.4. Let ๐ โ โณ1+ ( ๐ ) be a Szegห ator ๐ต๐, ๐ : โ๐, ๐ โโ โโฅ ๐, ๐ โ
โ
with respect to the orthonormal bases ( ๐๐ )๐=1 and ( ๐๐โ )๐=1 of the spaces โ๐, ๐ โ and โโฅ ๐, ๐ , respectively, is given by โ where โ is the block of the matrix given in (5.20). Now Theorem 4.3 can be reformulated in the following way. o measure. Then the Riesz projection ๐+ Corollary 5.5. Let ๐ โ โณ1+ ( ๐ ) be a Szegห 2 โ is bounded in ๐ฟ๐ if and only if โ is boundedly invertible in ๐2 where โ is the block of the matrix given in (5.20). In [3, Corollary 3.7] the matrix โ was expressed in terms of the Schur parameters associated with the measure ๐. In order to write down this matrix we introduce the necessary notions and terminology used in [3]. The matrix โ expressed in terms of the corresponding Schur parameter sequence will the denoted by โ ( ๐พ ). Let { } ฮ๐2 := ๐พ = (๐พ๐ )โ ๐=0 โ ๐2 : ๐พ๐ โ ๐ป, ๐ โ โ0 . Thus, ฮ๐2 is the subset of all ๐พ = (๐พ๐ )โ ๐=0 โ ฮ, for which the product โ โ ( ) 1 โ โฃ๐พ๐ โฃ2 ๐=0
converges. Let us mention the following well-known fact (see, for example, Remark 1.4)
Description of Helson-Szegห o Measures
287
Proposition 5.6. Let ๐ โ โณ1+ ( ๐ ). Then ๐ is a Szegห o measure if and only if ๐พ belongs to ฮ๐2 . For a Schur parameter sequence ๐พ belonging to ฮ๐2 , we note that the sequence (๐ฟ๐ (๐พ))โ ๐=0 introduced in formula (3.12) of [3] via ๐ฟ0 (๐พ) := 1 and, for each positive integer ๐, via ๐ฟ๐ (๐พ) := ๐ โ โ โ โ โ โ โ โ (โ1)๐ ... ๐พ๐1 ๐พ ๐1 +๐ 1 . . . ๐พ๐๐ ๐พ ๐๐ +๐ ๐ ๐=1
๐ 1 +๐ 2 +โ
โ
โ
+๐ ๐ =๐ ๐1 =๐โ๐ 1 ๐2 =๐1 โ๐ 2
๐๐ =๐๐โ1 โ๐ ๐
(5.22) plays a key role. Here the summation runs over all ordered ๐-tuples (๐ 1 , . . . , ๐ ๐ ) of positive integers which satisfy ๐ 1 + โ
โ
โ
+ ๐ ๐ = ๐. For example, โ โ ๐ฟ1 (๐พ) = โ ๐พ๐ ๐พ๐+1 ๐=0
and ๐ฟ2 (๐พ) = โ
โ โ
โ โ
๐พ๐ ๐พ๐+2 +
๐=0
โ โ
๐พ๐1 ๐พ๐1 +1 ๐พ๐2 ๐พ๐2 +1 .
๐1 =1 ๐2 =๐1 โ1
Obviously, if ๐พ โ ฮ๐2 , then the series (5.22) converges absolutely. For each ๐พ = (๐พ๐ )โ ๐=0 โ ฮ๐2 , we set ฮ ๐ :=
โ โ
๐ท ๐พ ๐ , ๐ โ โ0 ,
(5.23)
1 โ โฃ๐พ๐ โฃ2 , ๐ โ โ0 .
(5.24)
๐=๐
where ๐ท๐พ๐ :=
โ
In the space ๐2 we de๏ฌne the coshift mapping ๐ : ๐2 โ ๐2 via โ (๐ง๐ )โ ๐=0 โ (๐ง๐+1 )๐=0 .
(5.25)
The following result is contained in [3, Theorem 3.6, Corollary 3.7]. Theorem 5.7. Let ๐ โ โณ1+ ( ๐ ) be a Szegห o measure and let ๐พ โ ฮ be the Schur parameter sequence associated with ๐. Then ๐พ โ ฮ๐2 and the block โ of the matrix (5.20) has the form โ โ ฮ 1 0 0 ... โ ฮ 2 ๐ฟ1 (๐ ๐พ) ฮ 2 0 . . .โ โ โ 2 โ ฮ 3 ๐ฟ2 (๐ ๐พ) ฮ ๐ฟ (๐ ๐พ) ฮ . . .โ 3 1 3 โ โ .. .. .. .. โ โ(๐พ ) = โ , (5.26) โ .โ . . . โ โ โฮ ๐ ๐ฟ๐โ1 (๐ ๐พ) ฮ ๐ ๐ฟ๐โ2 (๐ 2 ๐พ) ฮ ๐ ๐ฟ๐โ3 (๐ 3 ๐พ) . . . โ โ โ .. .. .. . . . where ฮ ๐ , ๐ฟ๐ ( ๐พ ) and ๐ are given via the formulas (5.23), (5.22), and (5.25), respectively.
288
V.K. Dubovoy, B. Fritzsche and B. Kirstein
Remark 5.8. It follows from Theorems 5.4 and 5.7 that the matrix representation of the operator ๐ต๐, ๐ : โ๐, ๐ โโ โโฅ ๐, ๐ โ โ with respect to the orthonormal bases ( ๐๐ )โ ๐=1 and ( ๐๐ )๐=1 of the spaces โ๐, ๐ โฅ โ and โ๐, ๐ , respectively, is given by the matrix โ ( ๐พ ), where โ ( ๐พ ) has the form (5.26).
6. Characterization of Helson-Szegห o measures in terms of the Schur parameters of the associated Schur function The ๏ฌrst criterion which characterizes Helson-Szegห o measures in the associated Schur parameter sequence was already obtained. It follows by combination of Theorem 1.2, Theorem 4.3, Proposition 5.6, Theorem 5.7, and Remark 5.8. This leads us to the following theorem, which is one of the main results of this paper. Theorem 6.1. Let ๐ โ โณ1+ (๐) and let ๐พ โ ฮ be the sequence of Schur parameters associated with ๐. Then ๐ is a Helson-Szegห o measure if and only if ๐พ โ ฮ๐2 and the operator โโ (๐พ), which is de๏ฌned in ๐2 by the matrix (5.26), is boundedly invertible. Corollary 6.2. Let ๐ โ โณ1+ (๐) and let ๐พ โ ฮ be the sequence of Schur parameters associated with ๐. Then ๐ is a Helson-Szegห o measure if and only if ๐พ โ ฮ๐2 and there exists some positive constant ๐ถ such that for each โ โ ๐2 the inequality โฅโโ (๐พ)โโฅ โฅ ๐ถโฅโโฅ
(6.1)
is satis๏ฌed. Proof. First suppose that ๐พ โ ฮ๐2 and that there exists some positive constant ๐ถ such that for each โ โ ๐2 the inequality (6.1) is satis๏ฌed. From the shape (5.26) of the operator โ(๐พ) it follows immediately that ker โ(๐พ) = {0}. Thus, Ran โโ (๐พ) = ๐2 . From (6.1) it follows that the operator โโ (๐พ) is invertible and ( )โ1 that the corresponding inverse operator โโ (๐พ) is bounded and satis๏ฌes 1( โ )โ1 1 1 โ (๐พ) 1โค 1 ๐ถ โ where ๐ถ is taken from (6.1). Since โ (๐พ) is a bounded linear operator, the operator [โโ (๐พ)]โ1 is closed. Thus Ran โโ (๐พ) = ๐2 and, consequently, the operator โโ (๐พ) is boundedly invertible. Hence, Theorem 6.1 yields that ๐ is a Helson-Szegหo measure. If ๐ is a Helson-Szegหo measure, then Theorem 6.1 yields that โโ (๐พ) is boundedly invertible. Hence, condition (6.1) is trivially satis๏ฌed. โก It should be mentioned that a result similar to Theorem 6.1 was proved earlier using a di๏ฌerent method in [7, De๏ฌnition 4.6, Proposition 4.7 and Theorem 4.8]. More speci๏ฌcally, it was shown that a measure ๐ is a Helson-Szegห o measure if and only if some in๏ฌnite matrix โณ (which is de๏ฌned in [7, formulas (4.1) and (4.2)]) generates a bounded operator in โ2 . It was also shown that the boundedness of โณ
Description of Helson-Szegห o Measures
289
is equivalent to the boundedness of another operator matrix โ de๏ฌned in formula (6.4) of [7]. In order to derive criteria in another way we need some statements on the operator โ(๐พ) which were obtained in [3]. The following result which originates from [3, Theorem 3.12 and Corollary 3.13] plays an important role in the study of the matrix โ(๐พ). Namely, it describes the multiplicative structure of โ(๐พ) and indicates connections to the backward shift. Theorem 6.3. It holds that โ(๐พ) = ๐(๐พ) โ
โ(๐ ๐พ) where
โ
โ โ โ ๐(๐พ) := โ โ โ and ๐ท๐พ๐
๐ท๐พ1 โ๐พ1 ๐พ 2 โ๐พ1 ๐ท๐พ2 ๐พ 3
0 ๐ท๐พ2 โ๐พ2 ๐พ 3
..
. โ๐โ1
โ๐พ1 (
..
. โ๐โ1
0 0 ๐ท๐พ3
โ
โ
โ
โ
โ
โ
...
..
. โ๐โ1
๐ท๐พ๐ )๐พ ๐ โ๐พ3 ( ๐=4 ๐ท๐พ๐ )๐พ ๐ .. .. . . โ := 1 โ โฃ๐พ๐ โฃ2 , ๐ โ โ0 . The matrix ๐(๐พ) satis๏ฌes ๐=2
.. .
๐ท๐พ๐ )๐พ ๐ โ๐พ2 (
(6.2)
๐=3
๐ผ โ ๐(๐พ)๐โ (๐พ) = ๐(๐พ)๐ โ (๐พ) where
โ ๐(๐พ) := col โ๐พ1 , ๐พ2 ๐ท๐พ1 , . . . , ๐พ๐
๐โ1 โ
0 0 0
.. .
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
๐ท๐พ๐ โ
โ
โ
.. .
โ โ โ โ โ (6.3) โ โ
(6.4) โ
๐ท ๐พ๐ , . . . โ
(6.5)
๐=1
The multiplicative structure of โ(๐พ) obtained in Theorem 6.3 gives us some hope that the boundedness of the operator โโ (๐พ) can be reduced to a constructive condition on the Schur parameters via convergence of some in๏ฌnite products (series). This is a promising direction for future work on this problem. Let ๐พ โ ฮ๐2 . For each ๐ โ โ we set (see formula (5.3) in [3]) โ โ ฮ 1 0 0 ... 0 โ ฮ 2 ๐ฟ1 (๐ ๐พ) ฮ 2 0 ... 0 โ โ โ 2 โ ฮ 3 ๐ฟ2 (๐ ๐พ) ฮ ๐ฟ (๐ ๐พ) ฮ ... 0 โ 3 1 3 ๐๐ (๐พ) := โ โ . (6.6) โ .. .. .. .. โ โ . . . . โ 2 3 ฮ ๐ ๐ฟ๐โ1 (๐ ๐พ) ฮ ๐ ๐ฟ๐โ2 (๐ ๐พ) ฮ ๐ ๐ฟ๐โ3 (๐ ๐พ) . . . ฮ ๐ The matrices introduced in (6.6) will play an important role in our investigations. Now we turn our attention to some properties of the matrices ๐๐ (๐พ), ๐ โ โ, which will later be of use. From Corollary 5.2 in [3] we get the following result. Lemma 6.4. Let ๐พ = (๐พ๐ )โ ๐=0 โ ฮ๐2 and let ๐ โ โ. Then the matrix ๐๐ (๐พ) de๏ฌned by (6.6) is contractive. We continue with some asymptotical considerations.
290
V.K. Dubovoy, B. Fritzsche and B. Kirstein
Lemma 6.5. Let ๐พ = (๐พ๐ )โ ๐=0 โ ฮ๐2 . Then: (a) lim๐โโ ฮ ๐ = 1. (b) For each ๐ โ โ, lim๐โโ ๐ฟ๐ (๐ ๐ ๐พ) = 0. (c) For each ๐ โ โ, lim๐โโ ๐๐ (๐ ๐ ๐พ) = ๐ผ๐ .
โโ Proof. The choice of ๐พ implies the convergence of the in๏ฌnite product ๐=0 ๐ท๐พ๐ . This yields (a). Assertion (b) is an immediate consequence of the de๏ฌnition of the sequence (๐ฟ๐ (๐ ๐ ๐พ))โ ๐=1 (see (5.22) and (5.25)). By inspection of the sequence (๐๐ (๐ ๐ ๐พ))โ ๐=1 one can immediately see that the combination of (a) and (b) yields the assertion of (c). โก The following result is given in [3, Lemma 5.3]. Lemma 6.6. Let ๐พ = (๐พ๐ )โ ๐=0 โ ฮ๐2 and let ๐ โ โ. Then ๐๐ (๐พ) = ๐๐ (๐พ) โ
๐๐ (๐ ๐พ),
(6.7)
where ๐๐ (๐พ) := โ
๐ท ๐พ1 โ โ๐พ 1๐พ 2 โ โ โ๐พ ๐ท 1 ๐พ2 ๐พ 3 โ โ .. โ โ (โ . ) ๐โ1 โ๐พ1 ๐ท ๐พ๐ ๐พ ๐ ๐=2
0 ๐ท ๐พ2 โ๐พ2 ๐พ 3 .. (โ . ) ๐โ1 โ๐พ2 ๐ท ๐พ๐ ๐พ ๐ ๐=3
โ๐พ3
(โ
0 0 ๐ท๐พ3 .. .
๐โ1 ๐=4
... ... ... ) ๐ท ๐พ๐ ๐พ ๐
โ
0 0 0 .. .
. . . ๐ท ๐พ๐
โ โ โ โ. โ โ โ
(6.8) Moreover, ๐๐ (๐พ) is a nonsingular matrix which ful๏ฌlls ๐ผ๐ โ ๐๐ (๐พ)๐โ๐ (๐พ) = ๐๐ (๐พ)๐๐โ (๐พ), where
โ ๐๐ (๐พ) := โ๐พ1 , ๐พ2 ๐ท๐พ1 , . . . , ๐พ๐
( ๐โ1 โ
(6.9)
)โ ๐ ๐ท ๐พ๐ โ .
(6.10)
๐=1
Corollary 6.7. Let ๐พ = (๐พ๐ )โ ๐=0 โ ฮ๐2 and let ๐ โ โ. Then the multiplicative decomposition ๐๐ (๐พ) = holds true.
โโ โ โ
๐๐ (๐ ๐ ๐พ)
(6.11)
๐=0
Proof. Combine part (c) of Lemma 6.5 and (6.7). Now we state the next main result of this paper. For โ = (๐ง๐ )โ ๐=1 โ ๐2 and ๐ โ โ we set โ๐ := (๐ง1 , . . . , ๐ง๐ )โค โ โ๐ .
โก
Description of Helson-Szegห o Measures
291
Theorem 6.8. Let ๐ โ โณ1+ (๐) and let ๐พ โ ฮ be the sequence of Schur parameters associated with ๐. Then ๐ is a Helson-Szegห o measure if and only if ๐พ โ ฮ๐2 and there exists some positive constant ๐ถ such that for all โ โ ๐2 the inequality 1( โโ ) 1 1 ๐ 1 1 โ โ 1 ๐ 1 lim lim ๐๐ (๐ ๐พ) โ๐ 1 (6.12) 1 โฅ ๐ถโฅโโฅ ๐โโ ๐โโ 1 1 ๐=0 1 is satis๏ฌed. Proof. In view of (6.11) and condition (c) in Lemma 6.5 the condition (6.12) is equivalent to the fact that for all โ โ ๐2 the inequality lim โฅโโ๐ (๐พ)โ๐ โฅ โฅ ๐ถโฅโโฅ
(6.13)
๐โโ
is satis๏ฌed. This inequality is equivalent to the inequality (6.1).
โก
Theorem 6.8 leads to an alternate proof of an interesting su๏ฌcient condition for a Szegหo measure to be a Helson-Szegหo measure (see Theorem 6.12). To prove this result we will still need some preparations. Lemma 6.9. Let ๐ โ โ. Furthermore, let the nonsingular complex ๐ ร ๐ matrix ๐ and the vector ๐ โ โ๐ be chosen such that ๐ผ๐ โ ๐๐โ = ๐๐โ holds. Then 1 โ
satis๏ฌes
โฅ๐โฅ2โ๐
(6.14)
> 0 and the vector 1 ๐ห := โ ๐โ ๐ 2 1 โ โฅ๐โฅโ๐
(6.15)
๐ผ๐ โ ๐โ ๐ = ๐ห๐หโ .
(6.16)
Proof. The case ๐ = 0๐ร1 is trivial. Now suppose that ๐ โ โ๐ โ{0๐ร1}. From (6.14) we get (๐ผ๐ โ ๐๐โ )๐ = ๐๐ โ ๐ = โฅ๐โฅ2โ๐ โ
๐ (6.17) and consequently 2 ๐๐โ ๐ = (1 โ โฅ๐โฅโ๐ ) โ
๐. (6.18) 2
Hence 1 โ โฅ๐โฅโ๐ is an eigenvalue of ๐๐โ with corresponding eigenvector ๐. Since ๐ is nonsingular, the matrix ๐๐โ is positive Hermitian. Thus, we have 1 โ 2 โฅ๐โฅโ๐ > 0. Using (6.17) we infer 2
(๐ผ๐ โ ๐โ ๐)๐โ ๐ = ๐โ (๐ผ๐ โ ๐๐โ )๐ = โฅ๐โฅโ๐ โ
๐โ ๐. Taking into account (6.18) we can conclude [ ] 2 2 2 2 โฅ๐โ ๐โฅโ๐ = ๐ โ ๐๐โ ๐ = ๐ โ (1 โ โฅ๐โฅโ๐ ) โ
๐ = (1 โ โฅ๐โฅโ๐ ) โ
โฅ๐โฅโ๐
(6.19) (6.20)
and therefore from (6.15) we have โฅห ๐โฅโ๐ = โฅ๐โฅโ๐ > 0.
(6.21)
292
V.K. Dubovoy, B. Fritzsche and B. Kirstein 2
Formulas (6.19), (6.15) and (6.21) show that โฅห ๐ โฅโ๐ is an eigenvalue of ๐ผ๐ โ ๐โ ๐ with corresponding eigenvector ๐ห. From (6.14) and ๐ โ= 0๐ร1 we get rank (๐ผ๐ โ ๐โ ๐) = rank (๐ผ๐ โ ๐๐โ ) = 1. So for each vector โ we can conclude ( (๐ผ๐ โ ๐โ ๐)โ = (๐ผ๐ โ ๐โ ๐) โ,
๐ ห โฅห ๐ โฅ โ๐
)
๐ ห โฅห ๐ โฅโ๐
โ๐
= (โ, ๐ห)โ๐ ๐ห = ๐ห๐หโ โ
โ.
โก
Corollary 6.10. Let the assumptions of Lemma 6.9 be satis๏ฌed. Then for each โ โ โ๐ the inequalities 1
โฅ๐โโฅ โฅ (1 โ โฅ๐โฅ2 ) 2 โฅโโฅ
(6.22)
and 1
โฅ๐โ โโฅ โฅ (1 โ โฅ๐โฅ2 ) 2 โฅโโฅ
(6.23)
are satis๏ฌed. Proof. Applying (6.16) and (6.21) we get for โ โ โ๐ the relation ( ) โฅโโฅ2 โ โฅ๐โโฅ2 = (๐ผ โ ๐โ ๐)โ, โ = โฃ(โ, ๐ห)โฃ2 โค โฅห ๐ โฅ2 โฅโโฅ2 = โฅ๐โฅ2 โฅโโฅ2 . This implies (6.22). Analogously, (6.23) can be veri๏ฌed.
โก
Corollary 6.11. Let ๐พ โ ฮ๐2 , and let the matrix ๐๐ (๐พ) be de๏ฌned via (6.8). Then for all โ โ โ๐ the inequalities ( ๐ ) โ โฅ๐๐(๐พ)โโฅ โฅ ๐ท๐พ๐ โฅโโฅ (6.24) ๐=1
and
( โฅ๐โ๐(๐พ)โโฅ
โฅ
๐ โ
) ๐ท๐พ๐ โฅโโฅ
(6.25)
๐=1
are satis๏ฌed. Proof. The matrix ๐๐ ( ๐พ ) satis๏ฌes the conditions of Lemma 6.9. Here the vector ๐ has the form (6.10). It remains only to mention that in this case we have โก โค ๐โ1 ( ) ) โ( 2 2 2 2 2โฃ 2 โฆ 1 โ โฅ๐โฅ = 1 โ โฃ๐พ1 โฃ โ โฃ๐พ2 โฃ 1 โ โฃ๐พ1 โฃ โ โ
โ
โ
โ โฃ๐พ๐ โฃ 1 โ โฃ๐พ๐ โฃ (6.26) ๐=1
=
๐ โ
(
1 โ โฃ๐พ๐ โฃ2
)
.
โก
๐=1
The above consideration lead us to an alternate proof for a nice su๏ฌcient criterion for the Helson-Szegห o property of a measure ๐ โ โณ1+ ( ๐ ) which is expressed in terms of the modules of the associated Schur parameter sequence.
Description of Helson-Szegห o Measures
293
Regarding the history of Theorem 6.12, it should be mentioned that, in view of a theorem by B.L. Golinskii and I.A. Ibragimov [6], the convergence of the in๏ฌnite product in (6.27) is equivalent to the property that ๐ is absolutely continuous with respect to the Lebesgue measure. The corresponding density is then of the form exp ๐, where ๐ is a real Besov-class function. A Theorem of V.V. Pellerโs [13] states that every function of this form is a density of a Helson-Szegห o measure. This topic was also discussed in detail in [7]. Theorem 6.12. Let ๐ โ โณ1+ ( ๐ ) and let ๐พ โ ฮ be the Schur parameter sequence associated with ๐. If ๐พ โ ฮ๐2 and the in๏ฌnite product โ ( โ โ ) โ 2 1 โ โฃ๐พ๐ โฃ (6.27) ๐=1 ๐=๐
converges, then ๐ is a Helson-Szegห o measure. Proof. Applying successively the estimate (6.25), we get for all ๐, ๐ โ โ and all vectors โ โ โ๐ the chain of inequalities 1โก โโ โค 1 1 โก โโโ โค 1 1 1 1 1 ๐ ๐โ1 โ 1 1 โ 1 1 โ โ( ๐ ) ) ( ๐ โ ๐ 1 1 1โฃ โฆ โฃ โฆ ๐๐ ๐ ๐พ ๐๐ ๐ ๐พ โ1 = 1๐๐ ( ๐ ๐พ ) โ1 1 1 1 1 1 1 ๐=0 ๐=0 1โก โโโ โค 1 1 ๐โ1 1 ๐+๐ โ 1 โ โ( ๐ ) 1 1 โฃ โฆ โฅ ๐ท ๐พ๐ 1 ๐๐ ๐ ๐พ โ1 1 1 ๐=0 1 ๐=๐+1 โฅ
โ
โฅโ โ โฅโ โ =โ โ โฅโ
โ
โ
โ
๐+๐ โ
โ โ ๐ท ๐พ๐ โ โ
โ
๐=๐+1 โ โ
โ โ ๐ท ๐พ๐ โ โ
โ
๐=๐+1 ๐+1 โ โ โ ๐=1 ๐=๐ โ โ โ โ
โ
๐+๐โ1 โ ๐=๐ โ โ
โ
โ ๐ท ๐พ๐ โ โ
โ
โ
โ
โ
โ โ
โ
๐ท ๐พ๐ โ โ
โ
โ
โ
๐=๐
โ
โ
๐ โ ๐=1
โ โ
โ ๐ท๐พ๐ โ โฅโโฅ โ
๐ท๐พ๐ โ โฅโโฅ
๐=1
๐ท๐พ๐ โ โฅโโฅ โ
๐ท๐พ๐ โ โฅโโฅ
(6.28)
๐=1 ๐=๐
From this inequality it follows (6.12) where ๐ถ=
โ โ โ โ
๐ท ๐พ๐ .
๐=1 ๐=๐
Thus, the proof is complete.
โก
294
V.K. Dubovoy, B. Fritzsche and B. Kirstein
Taking into account that the convergence of the in๏ฌnite product (6.27) is equivalent to the strong Szegหo condition โ โ
2
๐ โ
โฃ๐พ๐ โฃ < โ,
๐=1
Theorem 6.12 is an immediate consequence of [7, Theorem 5.3]. The proof of [7, Theorem 5.3] is completely di๏ฌerent from the above proof of Theorem 6.12. It is based on a scattering formalism using CMV matrices. (For a comprehensive exposition on CMV matrices, we refer the reader to Chapter 4 in the monograph Simon [16].) The aim of our next considerations is to characterize the Helson-Szegห o property of a measure ๐ โ โณ1+ ( ๐ ) in terms of some in๏ฌnite series formed from its Schur parameter sequence. The following result provides the key information for the desired characterization. Theorem 6.13. Let ๐พ = ( ๐พ๐ )โ ๐=0 โ ฮโ2 and let ๐ ( ๐พ ) := ๐ผ โ โ ( ๐พ ) โโ ( ๐พ )
(6.29)
where โ ( ๐พ ) is given by (5.26). Then ๐ ( ๐พ ) satis๏ฌes the inequalities 0 โค ๐(๐พ ) โค ๐ผ
(6.30)
and admits the strong convergent series decomposition ๐(๐พ ) =
โ โ
๐๐ ( ๐พ ) ๐๐โ ( ๐พ )
(6.31)
๐=0
where ๐0 ( ๐พ ) := ๐ ( ๐พ ) ,
โก โโ โค ๐โ1 โ ( ) ) ( ๐๐ ( ๐พ ) := โฃ ๐ ๐ ๐ ๐พ โฆ ๐ ๐ ๐ ๐พ , ๐ โ โ,
(6.32)
๐=0
and ๐ ( ๐พ ), ๐ ( ๐พ ) and ๐ are given by (6.3), (6.5) and (5.25), respectively. Proof. Since the matrix โ ( ๐พ ) is a block of the unitary operator matrix given by (5.20) we have โฅโ ( ๐พ )โฅ โค 1. This implies the inequalities (6.30). Using (6.3) and (6.4), we obtain ๐ ( ๐พ ) = ๐ผ โ โ ( ๐พ ) โโ ( ๐พ ) = ๐ผ โ ๐ ( ๐พ ) โ ( ๐ ๐พ ) โโ ( ๐ ๐พ ) ๐โ ( ๐พ ) = ๐ผ โ ๐ ( ๐พ ) ๐โ ( ๐พ ) + ๐ ( ๐พ ) ๐ ( ๐ ๐พ ) ๐โ ( ๐พ ) = ๐ ( ๐พ ) ๐ โ ( ๐พ ) + ๐ ( ๐พ ) ๐ ( ๐ ๐พ ) ๐โ ( ๐พ ) .
Description of Helson-Szegห o Measures
295
Repeating this procedure ๐ โ 1 times, we get ๐ ( ๐พ ) = ๐ ( ๐พ ) ๐ โ ( ๐พ ) + ๐ ( ๐พ ) ๐ ( ๐ ๐พ ) ๐ โ ( ๐ ๐พ ) ๐โ ( ๐พ ) โค โก โโโ โค โก โโโ ๐โ1 ๐โ1 โ ( โ ) ) ( ๐ ๐ ๐ ๐พ โฆ ๐ ( ๐ ๐๐พ ) ๐โ ( ๐ ๐ ๐พ ) โฃ ๐โ ๐ ๐ ๐พ โฆ + โ
โ
โ
+ โฃ ๐=0
๐=0
โค โก โโโ โค โก โโโ ๐โ1 ๐โ1 โ ( โ ) ) ) ( ( ๐ ๐ ๐ ๐พ โฆ ๐ ๐ ๐+1 ๐พ โฃ ๐โ ๐ ๐ ๐พ โฆ +โฃ ๐=0
๐=0
โก โโโ โค โก โโโ โค ๐โ1 ๐โ1 ๐โ1 โ โ ( โ ( ) ) ) ( = ๐๐ ( ๐พ ) ๐๐โ ( ๐พ ) + โฃ ๐ ๐ ๐ ๐พ โฆ ๐ ๐ ๐+1 ๐พ โฃ ๐โ ๐ ๐ ๐พ โฆ. ๐=0
๐=0
๐=0
In view of part (c) of Lemma 6.4 and the shape (6.3) of the matrix ๐ ( ๐พ ) for ๏ฌnite vectors โ โ โ2 (i.e., โ has the form โ = col ( ๐ง1 , ๐ง2 , . . . , ๐ง๐ , 0, 0, . . . ) for some ๐ โ โ) we obtain โค โก โโโ โค โก โโโ ๐โ1 ๐โ1 โ ( โ ) ) ) ( ( ๐ ๐ ๐ ๐พ โฆ ๐ ๐ ๐+1 ๐พ โฃ ๐โ ๐ ๐ ๐พ โฆโ = 0. lim โฃ ๐โโโ
๐=0
๐=0
This implies that the series given by the right-hand side of the formula (6.32) weakly converges to ๐ ( ๐พ ). From the concrete form of this series, its strong convergence follows. Thus, the proof is complete. โก The last main result of this paper is the following statement, which is an immediate consequence of Theorem 6.1 and Theorem 6.13. Theorem 6.14. Let ๐ โ โณ1+ ( ๐ ) and let ๐พ โ ฮ be the sequence of Schur parameters associated with ๐. Then ๐ is a Helson-Szegห o measure if and only if ๐พ โ ฮโ2 and there exists some positive constant ๐ โ ( 0, 1 ) such that the inequality โ โ ๐๐ ( ๐พ ) ๐๐โ ( ๐พ ) โค ( 1 โ ๐ ) ๐ผ (6.33) ๐=0
is satis๏ฌed, where the vectors ๐๐ ( ๐พ ) , ๐ โ โ0 , are given by (6.32). We note that the inequality (6.33) can be considered as a rewriting of condition (6.12) in an additive form. Remark 6.15. Finally, we would like to add that many important properties of Schur functions can be characterized in terms of the matrix, โ ( ๐พ ), given by (5.26). It was shown in [3, Section 5] that the pseudocontinuability of a Schur function is determined by the properties of the matrix โ ( ๐พ ). In [4, Section 2], it was proved that the ๐-recurrence property of Schur parameter sequences of non-inner rational Schur functions is also expressed with the aid of the matrix โ ( ๐พ ). Furthermore, the structure of the matrix โ ( ๐พ ) allows one to determine whether a non-inner Schur function is rational or not(see ([3, Section 5]).
296
V.K. Dubovoy, B. Fritzsche and B. Kirstein
References [1] M.J. Bertin, A. Guilloux, J.P. Schreiber: Pisot and Salem Numbers, Birkhยจ auser, BaselโBostonโBerlin, 1992. [2] M.S. Brodskii: Unitary operator colligations and their characteristic functions (in Russian), Uspek Mat. Nauk 33 (1978), Issue 4, 141โ168. English transl. in: Russian Math. Surveys 33 (1978), Issue 4, 159โ191. [3] V.K. Dubovoy: Shift operators contained in contractions, Schur parameters and pseudocontinuable Schur functions, in: Interpolation, Schur Functions and Moment Problems (eds.: D. Alpay, I. Gohberg), Oper. Theory Adv. Appl., Vol. 165, Birkhยจ auser, Basel, 2006, pp. 175โ250. [4] V.K. Dubovoy, B. Fritzsche, B. Kirstein: The ๐ฎ-recurrence of Schur parameters of non-inner rational Schur functions, in: Topics in Operator Theory, Volume 1: Operators, Matrices and Analytic Functions (eds.: J.A. Ball, V. Bolotnikov, J.W. Helton, L. Rodman, I.M. Spitkovsky), Oper. Theory Adv. Appl., Vol. 202, Birkhยจ auser, Basel, 2010, pp. 151โ194. [5] V.K. Dubovoy, B.F. Fritzsche, B. Kirstein: Shift operators contained in contractions, pseudocontinuable Schur functions and orthogonal systems on the unit circle, Complex Analysis and Operator Theory 5 (2011), 579โ610. [6] B.L. Golinskii, I.A. Ibragimov: On Szegห oโs limit theorem (in Russian), Izv. Akad. Nauk. SSSR, Ser. Mat. 35(1971), 408โ429. English transl. in Math. USSR Izv. 5(1971), 421-444. [7] L.B. Golinskii, A.Ya. Kheifets, F. Peherstorfer, P.M. Yuditskii FaddeevMarchenko scattering for CMV matrices and the strong Szegห o theorem, arXiv: 0807.4017v1 [math.SP] 25 July 2008. ห : A problem in prediction theory, Annali di Mat. Pura ed [8] H. Helson, G. Szego Applicata 4 (1960), 51, 107โ138. [9] A.Ya. Kheifets, F. Peherstorfer, P.M. Yuditskii On scattering for CMV matrices, arXiv: 0706.2970v1 [math.SP] 20 June 2007. [10] P. Koosis: Introduction to ๐ป ๐ Spaces, Cambridge Univ. Press, Cambridge etc. 1998. [11] N.K. Nikolski: Operators, Functions and Systems: An Easy Reading, Math. Surveys and Monographs, V. 92, Contents: V. 1, Hardy, Hankel and Toeplitz (2002). [12] F. Peherstorfer, A.L. Volberg, P.M. Yuditskii CMV matrices with asymptotically constant coe๏ฌcients. Szegห o-Blaschke class, scattering theory, Journal of Functional Analysis 256 (2009), 2157โ2210. [13] V.V. Peller: Hankel operators of class ๐๐ and their applications (rational approximation, Gaussian processes, the problem of majorization of operators) (in Russian), Mat. Sb. 113(1980), 538โ581. English transl. in: Math. USSR Sbornik 41(1982), 443โ479. [14] M. Rosenblum, J. Rovnyak: Topics in Hardy Classes and Univalent Functions, Birkhยจ auser, Basel 1994. ยจ [15] I. Schur: Uber Potenzreihen, die im Inneren des Einheitskreises beschrยจ ankt sind, J. reine u. angew. Math., Part I: 147 (1917), 205โ232, Part II: 148 (1918), 122โ145. [16] B. Simon: Orthogonal Polynomials on the Unit Circle. Part 1: Classical Theory, Amer. Math. Soc. Colloq. Publ., Providence, RI, v. 54 (2004).
Description of Helson-Szegห o Measures
297
[17] T. Tao, C. Thiele: Nonlinear Fourier Analysis, IAS Lectures at Park City, Mathematics Series 2003. Vladimir K. Dubovoy Department of Mathematics and Mechanics Kharkov State University Svobody Square 4 UA-61077 Kharkov, Ukraine e-mail:
[email protected] Bernd Fritzsche, and Bernd Kirstein Mathematisches Institut Universitยจ at Leipzig Augustusplatz 10/11 D-04109 Leipzig, Germany e-mail:
[email protected] [email protected]
Operator Theory: Advances and Applications, Vol. 218, 299โ328 c 2012 Springer Basel AG โ
Divide and Conquer Method for Eigenstructure of Quasiseparable Matrices Using Zeroes of Rational Matrix Functions Y. Eidelman and I. Haimovici Dedicated to the memory of Israel Gohberg, our friend and teacher
Abstract. We study divide and conquer method to compute eigenstructure of matrices with quasiseparable representation. In order to ๏ฌnd the eigenstructure of a large matrix ๐ด we divide the problem into two problems for smaller sized matrices ๐ต and ๐ถ by using the quasiseparable representation of ๐ด. In the conquer step we show that to reconstruct the eigenstructure of ๐ด from those of ๐ต and ๐ถ amounts to the study of the eigenstructure of a rational matrix function. For a Hermitian matrix ๐ด which is order one quasiseparable we completely solve the eigenproblem. Mathematics Subject Classi๏ฌcation (2000). Primary 15A18; Secondary 26C15. Keywords. Quasiseparable, divide and conquer, rational matrix function, Hermitian matrix.
1. Introduction In order to solve the eigenproblem for a large matrix ๐ด which is in quasiseparable representation we represent ๐ด in the form ) ( ๐ต 0 + ๐บ๐ป ๐ด= 0 ๐ถ with smaller sized matrices ๐ต and ๐ถ and a perturbation matrix ๐บ๐ป of small rank that depends on the order of quasiseparability. The matrices ๐ต and ๐ถ have in turn at most the same order of quasiseparability and can therefore be divided further in the same way, until small enough matrices for which the eigenproblem can be solved conveniently. In most cases the two smaller matrices obtained by using an appropriate quasiseparable representation also belong both of them to that class. After the division step of the algorithm is completed and the eigenstructure of the smallest matrices has been found, we perform the conquer step in which the
300
Y. Eidelman and I. Haimovici
division tree is climbed back and we obtain the eigenstructure of a larger matrix ๐ด upon knowing the eigenstructure of two smaller matrices ๐ต and ๐ถ. To do this we should compute the eigenstructure of a small sized matrix function with size equal to the order of perturbation. We study in detail the eigenstructure of such matrix functions. Therefore the paper restates the de๏ฌnition of eigenvalues and Jordan chains for rational matrix functions. We ๏ฌnd in exact arithmetic a correspondence which is one-to-one and onto between the eigenvalues and Jordan chains of the matrix ๐ด and those of a rational matrix function which is built using only the spectral data of the smaller matrices ๐ต and ๐ถ and the perturbation matrix ๐บ๐ป. Although this correspondence is of theoretical importance, in practice, when only approximations of the eigenvalues are determined, we could not choose to compute the Jordan canonical form of the matrices. As the eigenvalue multiplicities are not continuous functions of the matrix entries, computation of the Jordan canonical form is an ill-posed problem. While performing the conquer step we impose more and more restrictive conditions in order to obtain more results. The complete algorithm is obtained for Hermitian matrices with quasiseparable of order one representations. While in theory most of our results apply to general matrices, which can be always represented as quasiseparable of a certain order, in practice the case of the non-Hermitian matrices, or of the matrices which are not order one quasiseparable raises numerous di๏ฌculties. Among the obstacles in the non-symmetric case, which are analyzed in detail in [11], we will mention that the complex roots of the equation occur in conjugate pairs, but after ๏ฌnding one such pair we can remain to work further with a complex matrix and that the roots do not interlace with the poles as in the symmetric case, but can scatter anywhere in the complex plane, as [4] puts it. Also, if the rational matrix function is not in fact a scalar one, as it is the case for (order one) quasiseparable matrix ๐ด, the position of the roots is again quite at random. The present algorithm has complexity ๐(๐ 2 ) in contrast to ๐(๐ 3 ) operations which are required to compute eigenvalues of a non-structured matrix. The detailed analysis of complexity of this algorithms will be done in [5], the computer experiments are planned to be preformed elsewhere. This paper is a continuation of the results presented in [1]. Our results on divide and conquer method generalize the corresponding results for tridiagonal matrices and for diagonal plus semiseparable matrices, concerning algorithms for tridiagonal matrices see [3, 8, 15, 7] and the literature cited therein, concerning diagonal plus semiseparable matrices see [12]. An algorithm for unitary Hessenberg matrices, which also have quasiseparable order one, di๏ฌerent from ones presented in this paper, was developed in [9]. For an important, interesting, complete and up-to-date state of art in the ๏ฌeld of divide and conquer algorithms for eigendecomposition see [15]. Following the exposition there will show that our approach covers all the cases in a uni๏ฌed manner, but there are still other alternative algorithms that solve the problem, for instance the use of arrowhead matrices, which seems to be close to our method.
Divide and Conquer for Quasiseparable Matrices
301
2. Notation and de๏ฌnitions For an ๐ ร๐ matrix ๐ด we denote by ๐ด๐๐ or by ๐ด(๐, ๐) its element on row 1 โค ๐ โค ๐ and on column 1 โค ๐ โค ๐ and by ๐ด(๐ : ๐, ๐ : ๐) the submatrix containing rows 1 โค ๐ โค ๐ โค ๐ inclusively between columns 1 โค ๐ โค ๐ โค ๐ inclusively. In particular, if ๐ = ๐ then we denote ๐ด(๐, ๐ : ๐) and if ๐ < ๐, ๐ = ๐ we denote ๐ด(๐ : ๐, ๐). Let ๐ด = {๐ด๐๐ }๐ ๐,๐=1 be a matrix with block entries ๐ด๐๐ of sizes ๐๐ ร ๐๐ . Assume that the entries of this matrix are represented in the form โง > ๏ฃด ๏ฃด ๐(๐)๐๐๐ ๐(๐), 1 โค ๐ < ๐ โค ๐, โจ ๐(๐), 1 โค ๐ = ๐ โค ๐, ๐ด๐๐ = (2.1) ๏ฃด ๏ฃด โฉ ๐(๐)๐< โ(๐), 1 โค ๐ < ๐ โค ๐. ๐๐
Here ๐(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐(๐) (๐ = 2, . . . , ๐ โ 1) are ๐ฟ ๐ฟ matrices of sizes ๐๐ ร ๐๐โ1 , ๐๐๐ฟ ร ๐๐ , ๐๐๐ฟ ร ๐๐โ1 respectively, ๐(๐) (๐ = 1, . . . , ๐ โ 1), โ(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 2, . . . , ๐ โ 1) are matrices of sizes ๐๐ ร ๐ ๐ ๐๐๐ , ๐๐โ1 ร ๐๐ , ๐๐โ1 ร ๐๐๐ respectively, ๐(๐) (๐ = 1, . . . , ๐ ) are ๐๐ ร ๐๐ matrices. < Also, the operations ๐> ๐๐ and ๐๐๐ are de๏ฌned for positive integers ๐, ๐, ๐ > ๐ as > < ๐> ๐๐ = ๐(๐ โ 1) โ
โ
โ
โ
โ
๐(๐ + 1) for ๐ > ๐ + 1, ๐๐+1,๐ = ๐ผ๐๐ and ๐๐๐ = ๐(๐ + 1) โ
โ
โ
โ
โ
๐(๐ โ 1) < for ๐ > ๐ + 1, ๐๐,๐+1 = ๐ผ๐๐ . The representation of a matrix ๐ด in the form (2.1) is called a quasiseparable representation. The elements ๐(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐(๐) (๐ = 2, . . . , ๐ โ 1); ๐(๐) (๐ = 1, . . . , ๐ โ 1), โ(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 2, . . . , ๐ โ1); ๐(๐) (๐ = 1, . . . , ๐ ) are called quasiseparable generators of the matrix ๐ด. The numbers ๐๐๐ฟ , ๐๐๐ (๐ = 1, . . . , ๐ โ 1) are called the orders of these generators. The elements ๐(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐(๐) (๐ = 2, . . . , ๐ โ 1) and ๐(๐) (๐ = 1, . . . , ๐ โ 1), โ(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 2, . . . , ๐ โ 1) are called also lower quasiseparable generators and upper quasiseparable generators of the matrix ๐ด. For matrices with scalar entries the elements ๐(๐) are numbers and the generators ๐(๐), ๐(๐) and ๐(๐), โ(๐) are rows and columns of the corresponding sizes. We can suppose that for an ๐ ร ๐ matrix the orders of the lower and of the upper quasiseparable generators are the same, ๐๐๐ฟ = ๐๐๐ (๐ = 1, . . . , ๐ โ 1), since otherwise one can pad the smaller ones with zeroes. It follows that we can ask this as a condition for Theorem 3.1 below, without loss of generality. Denote โ โ ๐(๐ + 1) โ โ ๐(๐ + 2)๐(๐ + 1) โ โ โ โ > ๐ ๐(๐ + 3)๐(๐ + 2)๐(๐ + 1) ๐๐+1 = col(๐(๐)๐๐๐ )๐=๐+1 = โ โ , (2.2) โ โ .. โ โ . ๐(๐ )๐(๐ โ 1) โ
โ
โ
๐(๐ + 2)๐(๐ + 1)
302
Y. Eidelman and I. Haimovici ๐ ๐๐ = row(๐> ๐+1,๐ ๐(๐))๐=1 ( = ๐(๐) โ
โ
โ
๐(3)๐(2)๐(1) โฃ
โ
โ
โ
๐บ๐
๐(๐) โ
โ
โ
๐(3)๐(2) ) โฃ ๐(๐) ,
โฃ โ
โ
โ
โฃ ๐(๐)๐(๐ โ 1) โ ๐(1)๐(2)๐(3) โ
โ
โ
๐(๐ โ 1)๐(๐) โ ๐(2)๐(3) โ
โ
โ
๐(๐ โ 1)๐(๐) โ โ .. โ < ๐ . = col(๐(๐)๐๐,๐+1 )๐=1 = โ โ ๐(๐ โ 2)๐(๐ โ 1)๐(๐) โ โ ๐(๐ โ 1)๐(๐) ๐(๐)
(2.3)
โ โ โ โ โ โ, โ โ โ
(2.4)
๐ ๐ป๐+1 = row(๐< ๐๐ โ(๐))๐=๐+1 ( = โ(๐ + 1) โฃ ๐(๐ + 1)โ(๐ + 2) โฃ ๐(๐ + 1)๐(๐ + 2)โ(๐ + 3) โฃ ) โ
โ
โ
โฃ ๐(๐ + 1) โ
โ
โ
๐(๐ โ 1)โ(๐ ) . (2.5) A direct computation shows that
๐ด(๐ + 1 : ๐, 1 : ๐) = ๐๐+1 ๐๐ ,
๐ = 1, . . . , ๐ โ 1.
(2.6)
We point out that the number of columns of ๐๐+1 as well as the number of rows of ๐๐ is ๐๐ , so that one can multiply these matrices and obtain a matrix whose rank is at most ๐๐ . Another direct computation shows that ๐ด(1 : ๐, ๐ + 1 : ๐ ) = ๐บ๐ ๐ป๐+1 ,
๐ = 1, . . . , ๐ โ 1.
(2.7)
We point out also that the number of columns of ๐บ๐ as well as the number of rows of ๐ป๐+1 is ๐๐ , so that one can multiply these matrices and obtain a matrix whose rank is at most ๐๐ . Next we will note down some de๏ฌnitions which could be found in [14] and the references therein. The complex number ๐0 is called a zero (or an eigenvalue) of the rational matrix function ๐น (๐) if det ๐น (๐0 ) = 0 and ๐ โ= 0 is called an eigenvector of ๐น (๐) corresponding to ๐0 if ๐น (๐0 )๐ = 0. If ๐0 is an eigenvector for the zero (eigenvalue) ๐0 and ๐ โ 1 (๐) ๐น (๐0 )๐๐โ๐ = 0, ๐! ๐=0
๐ = 0, 1, . . . , ๐,
then ๐0 , ๐1 , . . . , ๐๐ is called a Jordan chain corresponding to ๐0 . A system ๐10 , ๐11 , . . . , ๐1๐1 , ๐20 , ๐21 , . . . , ๐2๐2 , . . . , ๐๐0 , ๐๐1 , . . . , ๐๐๐๐ of Jordan chains corresponding to ๐0 is a canonical system of Jordan chains if all the Jordan chains are of maximal length among those Jordan chains corresponding to ๐0 which start with an eigenvector which is independent of all the eigenvectors which have been already chosen in the system. In particular, the ๏ฌrst chain in the system is of maximal length among all the Jordan chains corresponding to ๐0 .
Divide and Conquer for Quasiseparable Matrices
303
The numbers ๐1 โฅ ๐2 โฅ โ
โ
โ
โฅ ๐๐ are independent of the particular Jordan chains chosen and they are called the partial multiplicities of ๐0 and ๐1 + ๐2 + โ
โ
โ
+ ๐๐ (the sum of the lengths o๏ฌ all the independent Jordan chains) is called the multiplicity of ๐0 as a zero of ๐น (๐). The Jordan chains chosen for di๏ฌerent eigenvalues can contain the same vectors. For instance for the particular case when the matrix rational function is in fact an 1ร1 (a scalar) function, all the Jordan chains for di๏ฌerent zeroes start with the same eigenvector ๐ = 1. However, if the same Jordan chain corresponds to different eigenvalues ๐1 and ๐2 and it has been chosen in both the canonical systems of Jordan chains, then its length ๐ is counted twice: as a partial multiplicity of ๐1 and as a partial multiplicity of ๐2 as well, when determining the total multiplicity of these eigenvalues. For instance, for the function (๐ โ 1)(๐ โ 2)(๐ โ 3) the same Jordan chain ๐ = 1 gives for each of the three eigenvalues their total multiplicity of 1 each.
3. Divide step 3.1. The main theorem The divide step consists in splitting a single problem into two smaller independent problems with size roughly half the size of the original problem. This is done recursively, until the obtained problems are of a convenient size which is small enough so that they can be solved by standard techniques. In order to assure the next recursion step in the same initial conditions as for the current step, one must show that the two smaller matrices which are obtained in the divide step have quasiseparable representations of at most the same order as the larger initial matrix and that they possibly belong to the same class. Theorem 3.1. Let ๐, ๐ be two positive integers such that ๐ < ๐ and ๐ด = {๐ด๐๐ }๐ ๐,๐=1 be a block matrix with entries of sizes ๐๐ ร๐๐ with lower quasiseparable generators ๐(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐(๐) (๐ = 2, . . . , ๐ โ 1) of orders ๐๐ (๐ = 1, . . . , ๐ โ 1), upper quasiseparable generators ๐(๐) (๐ = 1, . . . , ๐ โ 1), โ(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 2, . . . , ๐ โ 1) of the same orders ๐๐ (๐ = 1, . . . , ๐ โ 1) and diagonal entries ๐(๐) (๐ = 1, . . . , ๐ ). Then the matrix ๐ด is a perturbation of rank ๐๐ at most of a 2 ร 2 block diagonal matrix ) ( ๐ต 0 (3.1) 0 ๐ถ with submatrices ๐ต of size ๐ ร ๐ and ๐ถ of size (๐ โ ๐) ร (๐ โ ๐) which have quasiseparable generators of orders ๐๐ , ๐ = 1, . . . , ๐ โ 1 and of orders ๐๐ , ๐ = ๐ + 1, . . . , ๐ โ 1 respectively.
304
form
Y. Eidelman and I. Haimovici In fact, using the notations (2.2)โ(2.5) one can represent the matrix ๐ด in the ( ๐ด=
where
( ๐1 =
while
๐บ๐ ๐๐+1
)
๐ต 0
0 ๐ถ
,
๐2 =
)
+ ๐1 ๐2 , (
๐๐
(3.2) ๐ป๐+1
)
,
๐ต = ๐ต๐ = ๐ด(1 : ๐, 1 : ๐) โ ๐บ๐ ๐๐ , ๐ถ = ๐ถ๐ = ๐ด(๐ + 1 : ๐, ๐ + 1 : ๐ ) โ ๐๐+1 ๐ป๐+1 . Moreover, the matrix ๐ต has quasiseparable generators
(3.3) (3.4)
> ๐๐ต (๐) = ๐(๐) โ ๐(๐)๐< ๐,๐+1 ๐๐+1,๐โ1 (๐ = 2, . . . , ๐),
๐๐ต (๐) = ๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐๐ต (๐) = ๐(๐) (๐ = 2, . . . , ๐ โ 1); โ๐ต (๐) โ
๐๐ต (๐) < > ๐๐โ1,๐+1 ๐๐+1,๐ ๐(๐)
= ๐(๐) (๐ = 1, . . . , ๐ โ 1), (๐ = 2, . . . , ๐), ๐๐ต (๐) = ๐(๐) (๐ = 2, . . . , ๐ โ 1);
> ๐๐ต (๐) = ๐(๐) โ ๐(๐)๐< ๐,๐+1 ๐๐+1,๐ ๐(๐) (๐ = 1, . . . , ๐) of orders ๐๐ , ๐ = 1, . . . , ๐ โ 1 and the matrix ๐ถ has the quasiseparable generators
๐๐ถ (๐ โ ๐) = ๐(๐), (๐ = ๐ + 2, . . . , ๐ ), < ๐๐ถ (๐ โ ๐) = ๐(๐) โ ๐> ๐+1,๐ ๐๐๐ โ(๐), (๐ = ๐ + 1, . . . , ๐ โ 1),
๐๐ถ (๐ โ ๐) = ๐(๐) (๐ = ๐ + 2, . . . , ๐ โ 1); < ๐๐ถ (๐ โ ๐) = ๐(๐) โ ๐(๐)๐> ๐๐ ๐๐,๐+1 , (๐ = ๐ + 1, . . . , ๐ โ 1), โ๐ถ (๐ โ ๐) = โ(๐), (๐ = ๐ + 2, . . . , ๐ ), ๐๐ต (๐ โ ๐) = ๐(๐) (๐ = ๐ + 2, . . . , ๐ โ 1); < ๐๐ต (๐ โ ๐) = ๐(๐) โ ๐(๐)๐> ๐๐ ๐๐๐ โ(๐), (๐ = ๐ + 1, . . . , ๐ ) of orders ๐๐ , ๐ = ๐ + 1, . . . , ๐ โ 1. Proof. It follows from (2.6), (2.7) that the matrix ๐ด may be partitioned in the form ( ) ๐ด(1 : ๐, 1 : ๐) ๐บ๐ ๐ป๐+1 ๐ด= . (3.5) ๐๐+1 ๐๐ ๐ด(๐ + 1 : ๐, ๐ + 1 : ๐ ) Using (3.5) one can represent the matrix ๐ด in the form ) ) ( ( ( ) ๐บ๐ ๐ต 0 ๐๐ ๐ป๐+1 , (3.6) + ๐ด= 0 ๐ถ ๐๐+1 where ๐ต and ๐ถ satisfy (3.4). Thus we have represented the matrix ๐ด as a sum of a block diagonal 2 ร 2 matrix and a matrix of rank ๐๐ at most. It remains to show that the matrix ๐ต has quasiseparable generators of orders ๐๐ , ๐ = 1, . . . , ๐ โ 1 and the matrix ๐ถ has quasiseparable generators of orders ๐๐ , ๐ = ๐ + 1, . . . , ๐ โ 1 and obtain the formulas for these generators. We will proceed ๏ฌrst for the matrix ๐ต.
Divide and Conquer for Quasiseparable Matrices
305
For 1 โค ๐ < ๐ โค ๐ we have ๐ต(๐, ๐) = ๐ด(๐, ๐) โ ๐บ๐ (๐, 1 : ๐๐ )๐๐ (1 : ๐๐ , ๐) < > = ๐(๐)๐> ๐๐ ๐(๐) โ ๐(๐)๐๐,๐+1 ๐๐+1,๐ ๐(๐).
Using the equality > > ๐> ๐+1,๐ = ๐(๐) โ
โ
โ
๐(๐)๐(๐ โ 1) โ
โ
โ
๐(๐ + 1) = ๐๐+1,๐โ1 ๐๐๐
(3.7)
we conclude that for 1 โค ๐ < ๐ โค ๐ we have > > ๐ต(๐, ๐) = (๐(๐) โ ๐(๐)๐< ๐,๐+1 ๐๐+1,๐โ1 )๐๐๐ ๐(๐).
Thus the matrix ๐ต has lower quasiseparable generators > ๐(๐) โ ๐(๐)๐< ๐,๐+1 ๐๐+1,๐โ1 (๐ = 2, . . . , ๐),
๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐(๐) (๐ = 2, . . . , ๐ โ 1) of orders ๐๐ , ๐ = 1, . . . , ๐ โ 1. Similarly we obtain for ๐ โค ๐ the following diagonal entries for ๐ต ๐ต(๐, ๐) = ๐ด(๐, ๐) โ ๐บ๐ (๐, 1 : ๐๐ )๐๐ (1 : ๐๐ , ๐) > = ๐(๐) โ ๐(๐)๐< ๐,๐+1 ๐๐+1,๐ ๐(๐),
๐ = 1, . . . , ๐
and also the following upper quasiseparable generators of orders ๐๐ , ๐ = 1, . . . , ๐โ1 < > ๐ต(๐, ๐) = ๐(๐)๐< ๐๐ (โ(๐) โ ๐๐โ1,๐+1 ๐๐+1,๐ ๐(๐)),
1 โค ๐ < ๐ โค ๐.
Use now formula ๐< ๐๐ = ๐(๐ + 1)๐(๐ + 2) โ
โ
โ
๐(๐ โ 1) < = (๐(๐ + 1) โ
โ
โ
๐(๐ โ 1))(๐(๐) โ
โ
โ
๐(๐ โ 1)) = ๐< ๐๐ ๐๐โ1,๐ .
(3.8)
In the same way we get that the matrix ๐ถ has quasiseparable generators of orders ๐๐ , ๐ = ๐ + 1, . . . , ๐ โ 1 and we obtain the formulas for these generators. For ๐ + 1 โค ๐ < ๐ โค ๐ we have ๐ถ(๐, ๐) = ๐ด(๐, ๐) โ ๐๐+1 (๐, 1 : ๐๐ )๐ป๐+1 (1 : ๐๐ , ๐) > < = ๐(๐)๐> ๐๐ ๐(๐) โ ๐(๐)๐๐๐ ๐๐๐ โ(๐).
Using again the equality (3.7), namely > > ๐> ๐๐ = ๐(๐ โ 1) โ
โ
โ
๐(๐ + 1)๐(๐) โ
โ
โ
๐(๐ + 1) = ๐๐๐ ๐๐+1,๐
we conclude that > < ๐ถ(๐, ๐) = ๐(๐)๐> ๐๐ (๐(๐) โ ๐๐+1,๐ ๐๐๐ โ(๐)),
๐ + 1 โค ๐ < ๐ โค ๐.
Thus the matrix ๐ถ has lower quasiseparable generators ๐(๐), (๐ = ๐ + 2, . . . , ๐ ), < ๐(๐) โ ๐> ๐+1,๐ ๐๐๐ โ(๐), ๐ = ๐ + 1, . . . , ๐ โ 1 and ๐(๐) (๐ = ๐ + 2, . . . , ๐ โ 1) of orders ๐๐ , ๐ = ๐ + 1, . . . , ๐ โ 1. Similarly we obtain for ๐ โฅ ๐ + 1 the following diagonal entries for ๐ถ < ๐ถ(๐, ๐) = ๐ด(๐, ๐) โ ๐๐+1 (๐, 1 : ๐๐ )๐ป๐+1 (1 : ๐๐ , ๐) = ๐(๐) โ ๐(๐)๐> ๐๐ ๐๐๐ โ(๐)
306
Y. Eidelman and I. Haimovici
and the following upper quasiseparable generators of orders ๐๐ , ๐ = ๐ + 1, . . . , ๐ โ1 ๐ถ(๐, ๐) = ๐ด(๐, ๐) โ ๐๐+1 (๐, 1 : ๐๐ )๐ป๐+1 (1 : ๐๐ , ๐) > < = ๐(๐)๐< ๐๐ โ(๐) โ ๐(๐)๐๐๐ ๐๐๐ โ(๐) < < = (๐(๐) โ ๐(๐)๐> ๐๐ ๐๐,๐+1 )๐๐๐ โ(๐),
๐ + 1 โค ๐ < ๐ โค ๐.
Here we used again formula (3.8) to show that ๐< ๐๐ = ๐(๐ + 1)๐(๐ + 2) โ
โ
โ
๐(๐ โ 1) < = (๐(๐ + 1) โ
โ
โ
๐(๐))(๐(๐ + 1) โ
โ
โ
๐(๐ โ 1)) = ๐< ๐,๐+1 ๐๐๐ .
โก
This theorem generalizes a result obtained in [1, Section 7] for Hermitian matrices. This result is presented below as Corollary 3.2. 3.2. Hermitian and/or tridiagonal matrices It is clear that if the matrix ๐ด is Hermitian, then this property is also preserved for the matrices ๐ต and ๐ถ. In this case, only the computation of the lower quasiseparable generators is needed so that the complexity is less. Indeed, for a Hermitian block matrix, using the given lower quasiseparable generators one can build the following upper quasiseparable generators of the same orders ๐(๐) = (๐(๐))โ , ๐ = 1, . . . , ๐ โ 1,
โ(๐) = (๐(๐))โ , ๐ = 2, . . . , ๐,
โ
๐(๐) = (๐(๐)) , ๐ = 2, . . . , ๐ โ 1.
(3.9)
Corollary 3.2. Let ๐, ๐ be two positive integers such that ๐ < ๐ and ๐ด = {๐ด๐๐ }๐ ๐,๐=1 be a block Hermitian matrix with entries of sizes ๐๐ ร ๐๐ with lower quasiseparable generators ๐(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐(๐) (๐ = 2, . . . , ๐ โ 1) of orders ๐๐ (๐ = 1, . . . , ๐ โ 1) and diagonal entries ๐(๐) (๐ = 1, . . . , ๐ ). Then the matrix ๐ด is a perturbation of rank ๐๐ at most of a 2 ร 2 block diagonal matrix ) ( ๐ต 0 (3.10) 0 ๐ถ with Hermitian submatrices ๐ต of size ๐ ร ๐ and ๐ถ of size (๐ โ ๐) ร (๐ โ ๐) which have lower quasiseparable generators of orders ๐๐ , ๐ = 1, . . . , ๐ โ 1 and of orders ๐๐ , ๐ = ๐ + 1, . . . , ๐ โ 1 respectively. In fact, one can represent the matrix ๐ด in the form (3.2), (3.3) with ) ( ) ( ๐โ๐ โ , ๐2 = ๐๐ ๐๐+1 , (3.11) ๐1 = ๐๐+1 while
๐ต = ๐ต๐ = ๐ด(1 : ๐, 1 : ๐) โ ๐โ๐ ๐๐ ,
โ . ๐ถ = ๐ถ๐ = ๐ด(๐ + 1 : ๐, ๐ + 1 : ๐ ) โ ๐๐+1 ๐๐+1
(3.12)
Divide and Conquer for Quasiseparable Matrices
307
Moreover, the matrix ๐ต has (lower) quasiseparable generators โ > ๐๐ต (๐) = ๐(๐) โ (๐(๐))โ (๐> ๐+1,๐โ1 ) ๐๐+1,๐โ1 (๐ = 2, . . . , ๐),
๐๐ต (๐) = ๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐๐ต (๐) = ๐(๐) (๐ = 2, . . . , ๐ โ 1) of orders ๐๐ , ๐ = 1, . . . , ๐ โ 1 and the matrix ๐ถ has (lower) quasiseparable generators (๐ = ๐ + 2, . . . , ๐ ), ๐๐ถ (๐ โ ๐) = ๐(๐), > โ โ ๐๐ถ (๐ โ ๐) = ๐(๐) โ ๐> ๐+1,๐ (๐๐+1,๐ ) (๐(๐)) ,
(๐ = ๐ + 1, . . . , ๐ โ 1),
๐๐ถ (๐ โ ๐) = ๐(๐)
(๐ = ๐ + 2, . . . , ๐ โ 1)
of orders ๐๐ , ๐ = ๐ + 1, . . . , ๐ โ 1 as in Theorem 3.1. The diagonal entries of the matrices ๐ต and ๐ถ become in the Hermitian case โ > ๐๐ต (๐) = ๐(๐) โ (๐(๐))โ (๐> ๐+1,๐ ) ๐๐+1,๐ ๐(๐)
๐๐ถ (๐ โ ๐) = ๐(๐) โ
> โ โ ๐(๐)๐> ๐๐ (๐๐๐ ) (๐(๐)) ,
(๐ = 1, . . . , ๐), (๐ = ๐ + 1, . . . , ๐ ).
In order to show that the present paper covers the case of a tridiagonal matrix which has been treated extensively in the literature ( see [3, 8, 7] and the literature cited therein) we have yet to prove that our quasiseparable approach for dividing a large matrix also preserves the tridiagonal structure. Corollary 3.3. Let ๐, ๐ be two positive integers such that ๐ < ๐ and let ๐ด = {๐ด๐๐ }๐ ๐,๐=1 be a block tridiagonal matrix โ โ ๐พ1 ๐ฝ1 0 โ
โ
โ
0 0 0 โ ๐ผ1 ๐พ2 ๐ฝ2 โ
โ
โ
0 0 0 โ โ โ โ 0 ๐ผ2 ๐พ3 โ
โ
โ
0 0 0 โ โ โ (3.13) โ โ .. .. .. .. .. .. . . โ โ . . . . . . . โ โ โ 0 0 0 โ
โ
โ
๐ผ๐ โ2 ๐พ๐ โ1 ๐ฝ๐ โ1 โ 0 0 0 โ
โ
โ
0 ๐ผ๐ โ1 ๐พ๐ , where ๐พ๐ are ๐๐ ร ๐๐ matrices, ๐ = 1, . . . , ๐ and ๐ผ๐ , ๐ฝ๐ are ๐๐+1 ร ๐๐ and ๐๐ ร ๐๐+1 matrices respectively, ๐ = 1, . . . , ๐ โ 1. Suppose that the matrix ๐ด has the following block quasiseparable generators ๐(๐) = ๐ผ๐โ1 , (๐ = 2, . . . , ๐ ), ๐(๐) = ๐ผ๐๐ , (๐ = 1, . . . , ๐ โ 1), ๐(๐) = 0๐๐ ร๐๐+1 , (๐ = 2, . . . , ๐ โ 1), ๐(๐) = ๐ผ๐๐ , (๐ = 1, . . . , ๐ โ 1), โ(๐) = ๐ฝ๐โ1 , (๐ = 2, . . . , ๐ ),
(3.14)
๐(๐) = 0๐๐ ร๐๐+1 (๐ = 2, . . . , ๐ โ 1) and the diagonal entries ๐(๐) = ๐พ๐ , ๐ = 1, . . . , ๐ . Then the matrix ๐ด is a perturbation of block rank one of a 2ร2 block diagonal matrix ) ( ๐ต 0 (3.15) 0 ๐ถ
308
Y. Eidelman and I. Haimovici
with block tridiagonal submatrices ๐ต of block size ๐ ร ๐ and ๐ถ of block size (๐ โ ๐) ร (๐ โ ๐) which preserve the quasiseparable generators of order one of the matrix ๐ด and di๏ฌer of it only on diagonal entries. Proof. It follows by (2.2) that ๐๐+1 =
๐ col(๐(๐)๐> ๐๐ )๐=๐+1
by (2.3) that ๐ ๐๐ = row(๐> ๐+1,๐ ๐(๐))๐=1 =
( = (
by (2.4) that ๐ ๐บ๐ = col(๐(๐)๐< ๐,๐+1 )๐=1 =
,
0๐๐ ร๐๐
)
๐ผ๐๐
0๐๐ ร๐๐ (
)
๐ผ๐
0๐๐ ร๐๐ ๐ผ๐๐
,
)
and by (2.5) that
( ) ๐ ๐ฝ๐ 0๐๐ ร๐๐ , ๐ป๐+1 = row(๐< ๐๐ โ(๐))๐=๐+1 = โ๐โ1 โ ๐๐ , ๐๐ = ๐ where ๐๐ = ๐=1 ๐=๐+1 ๐๐ . Using (3.4) it follows that the desired ๐ต and ๐ถ satisfy ๐ต = ๐ต๐ = ๐ด(1 : ๐, 1 : ๐) โ ๐บ๐ ๐๐ โ 0 ๐พ1 ๐ฝ1 0 โ
โ
โ
โ ๐ผ1 ๐พ2 ๐ฝ2 โ
โ
โ
0 โ โ 0 ๐ผ2 ๐พ 3 โ
โ
โ
0 โ =โ . .. .. .. . . . โ . . . . . โ โ 0 0 0 โ
โ
โ
๐ผ๐โ2 0 0 0 โ
โ
โ
0
0 0 0 .. .
๐พ๐โ1 ๐ผ๐โ1
0 0 0 .. .
โ
๐ฝ๐โ1 ๐พ๐ โ ๐ผ๐๐ ,
๐ถ = ๐ถ๐ = ๐ด(๐ + 1 : ๐, ๐ + 1 : ๐ ) โ ๐๐+1 ๐ป๐+1 โ 0 โ
โ
โ
0 ๐พ๐+1 โ ๐ผ๐ ๐ฝ๐ ๐ฝ๐+1 โ ๐ผ ๐พ ๐ฝ โ
โ
โ
0 ๐+1 ๐+2 ๐+2 โ โ 0 ๐ผ๐+2 ๐พ๐+3 โ
โ
โ
0 โ =โ .. .. .. .. . .. โ . . . . โ โ 0 0 0 โ
โ
โ
๐ผ๐ โ2 0 0 0 โ
โ
โ
0
โ โ โ โ โ โ โ โ
0 0 0 .. .
๐พ๐ โ1 ๐ผ๐ โ1
0 0 0 .. .
๐ฝ๐ โ1 ๐พ๐ ,
โ โ โ โ โ โ. โ โ โ
Therefore ๐ต di๏ฌers from the submatrix ๐ด(1 : ๐, 1 : ๐) only on the entry ๐(๐), while ๐ถ di๏ฌers from the submatrix ๐ด(๐ + 1 : ๐, ๐ + 1 : ๐ ) only on the entry ๐(๐ + 1) and it follows that the new matrices are tridiagonal again and that they preserve the quasiseparable generators of ๐ด given in (3.14), namely their generators are ๐๐ต (๐) = ๐ผ๐โ1 , (๐ = 2, . . . , ๐), ๐๐ต (๐) = ๐ผ๐๐ , (๐ = 1, . . . , ๐ โ 1), ๐๐ต (๐) = 0๐๐ ร๐๐+1 , (๐ = 2, . . . , ๐ โ 1),
Divide and Conquer for Quasiseparable Matrices
309
๐๐ต (๐) = ๐ผ๐๐ , (๐ = 1, . . . , ๐ โ 1), โ๐ต (๐) = ๐ฝ๐โ1 , (๐ = 2, . . . , ๐), ๐๐ต (๐) = 0๐๐ ร๐๐+1 (๐ = 2, . . . , ๐ โ 1), ๐๐ถ (๐ โ ๐) = ๐ผ๐โ1 , (๐ = ๐ + 2, . . . , ๐ ), ๐๐ถ (๐ โ ๐) = ๐ผ๐๐ , (๐ = ๐ + 1, . . . , ๐ โ 1), ๐๐ถ (๐ โ ๐) = 0๐๐ ร๐๐+1 , (๐ = ๐ + 2, . . . , ๐ โ 1), ๐๐ถ (๐ โ ๐) = ๐ผ๐๐ , (๐ = ๐ + 1, . . . , ๐ โ 1), โ๐ถ (๐ โ ๐) = ๐ฝ๐โ1 , (๐ = ๐ + 2, . . . , ๐ ), ๐๐ถ (๐ โ ๐) = 0๐๐ ร๐๐+1 (๐ = ๐ + 2, . . . , ๐ โ 1) and their diagonal entries are ๐๐ต (๐) = ๐พ๐ , ๐ = 1, . . . , ๐ โ 1, ๐๐ถ (1) = ๐พ๐+1 โ ๐ผ๐ ๐ฝ๐ ,
๐๐ต (๐) = ๐พ๐ โ ๐ผ๐๐ ,
๐๐ถ (๐ โ ๐) = ๐พ๐ , ๐ = ๐ + 2, . . . , ๐.
Moreover the perturbations given in (3.3) are of block rank one โ โ 0๐๐ ร๐๐ ) ( โ ๐ผ๐๐ โ ๐บ๐ โ, ๐1 = =โ โ โ ๐๐+1 ๐ผ๐ 0๐๐ ร๐๐ ( ) ( ) ๐2 = ๐๐ ๐ป๐+1 = 0๐๐ ร๐๐ ๐ผ๐๐ ๐ฝ๐ 0๐๐ ร๐๐ .
โก
It follows from Corollaries 3.2 and 3.3 that in the case of a matrix ๐ด which is both tridiagonal and Hermitian at the same time the obtained matrices ๐ต and ๐ถ also belong to the same class. 3.3. Algorithms to obtain suitable quasiseparable generators for the divided matrices and the entries of the perturbation matrices The following algorithm obtains in an e๏ฌcient manner lower and upper quasiseparable generators for the matrices ๐ต and ๐ถ. Algorithm 3.4. Let ๐ด = {๐ด๐๐ }๐ ๐,๐=1 be a block matrix with entries of sizes ๐๐ ร ๐๐ with lower quasiseparable generators ๐(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐(๐) (๐ = 2, . . . , ๐ โ 1) of orders ๐๐ (๐ = 1, . . . , ๐ โ 1), upper quasiseparable generators ๐(๐) (๐ = 1, . . . , ๐ โ 1), โ(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 2, . . . , ๐ โ 1) of the same orders ๐๐ (๐ = 1, . . . , ๐ โ 1) and diagonal entries ๐(๐) (๐ = 1, . . . , ๐ ). Let the matrices ๐ต and ๐ถ be given by (3.2). Then a set of quasiseparable generators ๐๐ต (๐) (๐ = 2, . . . , ๐), ๐๐ต (๐) (๐ = 1, . . . , ๐ โ 1), ๐๐ต (๐) (๐ = 2, . . . , ๐ โ 1); ๐๐ต (๐) (๐ = 1, . . . , ๐ โ 1), โ๐ต (๐) (๐ = 2, . . . , ๐), ๐๐ต (๐) (๐ = 2, . . . , ๐ โ 1); ๐๐ต (๐) (๐ = 1, . . . , ๐) of the matrix ๐ต and a set of quasiseparable generators ๐๐ถ (๐) (๐ = 2, . . . , ๐ โ ๐), ๐๐ถ (๐) (๐ = 1, . . . , ๐ โ ๐ โ 1), ๐๐ถ (๐) (๐ = 2, . . . , ๐ โ ๐ โ 1); ๐๐ถ (๐) (๐ = 1, . . . , ๐ โ ๐ โ 1), โ๐ถ (๐) (๐ = 2, . . . , ๐ โ๐), ๐๐ถ (๐) (๐ = 2, . . . , ๐ โ๐โ1); ๐๐ถ (๐) (๐ = 1, . . . , ๐ โ๐) of the matrix ๐ถ which have the same orders as the generators of the matrix ๐ด are obtained with the following algorithm.
310
Y. Eidelman and I. Haimovici
1. Find the quasiseparable generators of ๐ต. 1.1.
๐ข = ๐(๐), ๐๐ต (๐) = ๐(๐) โ ๐(๐)๐(๐), ๐๐ต (๐) = ๐(๐) โ ๐(๐)๐ข, (3.16) ๐ฃ = ๐(๐), โ๐ต (๐) = โ(๐) โ ๐ฃ๐(๐), ๐2 (๐) = ๐(๐),
(3.17)
๐1 (๐) = ๐(๐).
1.2. For ๐ = ๐ โ 1, . . . , 2 perform the following. ๐ค = ๐ข๐(๐), ๐ข = ๐ข๐(๐), ๐ง = ๐(๐)๐ฃ, ๐๐ต (๐) = ๐(๐) โ ๐ง๐ค, ๐๐ต (๐) = ๐(๐) โ ๐ง๐ข, (3.18) ๐ฃ = ๐(๐)๐ฃ, ๐๐ต (๐) = ๐(๐),
๐๐ต (๐) = ๐(๐),
โ๐ต (๐) = โ(๐) โ ๐ฃ๐ค, ๐2 (๐) = ๐ค,
1.3.
๐๐ต (๐) = ๐(๐),
๐1 (1) = ๐(1)๐ฃ, ๐๐ต (1) = ๐(1),
๐๐ต (๐) = ๐(๐),
(3.19) (3.20)
๐1 (๐) = ๐ง. ๐2 (1) = ๐ข๐(1),
๐๐ต (1) = ๐(1) โ ๐1 (1)๐2 (1), ๐๐ต (1) = ๐(1).
(3.21) (3.22)
2. Find the quasiseparable generators of ๐ถ. 2.1. ๐ = ๐ + 1, ๐ข = ๐(๐ ), ๐๐ถ (1) = ๐(๐ ) โ ๐(๐ )โ(๐ ), ๐๐ถ (1) = ๐(๐ ) โ ๐ขโ(๐ ), (3.23) ๐1 (๐ ) = ๐(๐ ), ๐ฃ = ๐(๐ ),
๐2 (๐ ) = โ(๐ ),
๐๐ถ (1) = ๐(๐ ) โ ๐(๐ )๐ฃ.
(3.24)
2.2. For ๐ = ๐ + 2, . . . , ๐ โ 1 perform the following. ๐ = ๐ โ ๐,
๐ค = ๐(๐)๐ข,
๐ง = ๐ฃโ(๐), ๐1 (๐) = ๐ค,
๐๐ถ (๐ ) = ๐(๐) โ ๐ค๐ง, ๐ฃ = ๐ฃ๐(๐),
๐ข = ๐(๐)๐ข,
๐๐ถ (๐ ) = ๐(๐) โ ๐ข๐ง,
๐2 (๐) = ๐ง,
๐๐ถ (๐ ) = ๐(๐),
๐๐ถ (๐ ) = ๐(๐)
๐๐ถ (๐ ) = ๐(๐) โ ๐ค๐ฃ,
๐๐ถ (๐ ) = ๐(๐),
๐1 (๐ ) = ๐(๐ )๐ข,
๐2 (๐ ) = ๐ฃโ(๐ ),
2.3. ๐ = ๐ โ ๐,
๐๐ถ (๐ ) = ๐(๐ ),
(3.25)
โ๐ถ (๐ ) = โ(๐).
๐๐ถ (๐ ) = ๐(๐ ) โ ๐1 (๐ )๐2 (๐ ),
โ๐ถ (๐ ) = โ(๐ ).
(3.26) (3.27)
(3.28) (3.29)
The following algorithm computes suitable lower quasiseparable generators for ๐ต and ๐ถ and the entries of the perturbation matrix in the Hermitian case. Algorithm 3.5. Algorithm for Hermitian matrices Let ๐ด = {๐ด๐๐ }๐ ๐,๐=1 be a block matrix with entries of sizes ๐๐ ร ๐๐ with lower quasiseparable generators ๐(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐(๐) (๐ = 2, . . . , ๐ โ 1) of orders ๐๐ (๐ = 1, . . . , ๐ โ 1) and diagonal entries ๐(๐) (๐ = 1, . . . , ๐ ). Let the matrices ๐ต and ๐ถ be given by (3.2).
Divide and Conquer for Quasiseparable Matrices
311
Then it follows that sets of lower quasiseparable generators and diagonal entries ๐๐ต (๐) (๐ = 2, . . . , ๐),๐๐ต (๐) (๐ = 1, . . . , ๐ โ 1), ๐๐ต (๐) (๐ = 2, . . . , ๐ โ 1); ๐๐ต (๐) (๐ = 1, . . . , ๐) for the matrix ๐ต and ๐๐ถ (๐) (๐ = 2, . . . , ๐ โ ๐), ๐๐ถ (๐) (๐ = 1, . . . , ๐ โ ๐ โ 1), ๐๐ถ (๐) (๐ = 2, . . . , ๐ โ ๐ โ 1); ๐๐ถ (๐) (๐ = 1, . . . , ๐ โ ๐) for the matrix ๐ถ which have the same orders as the generators of the matrix ๐ด are obtained with the following algorithm. 1. Find the lower quasiseparable generators and the diagonal entries of the matrix ๐ต. 1.1. ๐ข = ๐(๐), ๐๐ต (๐) = ๐(๐) โ (๐(๐))โ ๐(๐), ๐๐ต (๐) = ๐(๐) โ (๐(๐))โ ๐ข, (3.30) ๐2 (๐) = ๐(๐). 1.2. For ๐ = ๐ โ 1, . . . , 2 perform the following. ๐ค = ๐ข๐(๐), ๐ข = ๐ข๐(๐), ๐๐ต (๐) = ๐(๐) โ ๐คโ ๐ค, ๐๐ต (๐) = ๐(๐) โ ๐คโ ๐ข, ๐๐ต (๐) = ๐(๐),
๐๐ต (๐) = ๐(๐),
(3.31)
๐2 (๐) = ๐ค
(3.32)
๐2 (1) = ๐ค,
1.3. ๐๐ต (1) = ๐(1),
๐ค = ๐ข๐(1),
๐๐ต (1) = ๐(1) โ ๐คโ ๐ค.
(3.33)
2. Find the lower quasiseparable generators and the diagonal entries of the matrix ๐ถ. 2.1.
๐ = ๐ + 1,
๐ข = ๐(๐ ), ๐๐ถ (1) = ๐(๐ ) โ ๐(๐ )(๐(๐ ))โ ,
(3.34)
๐๐ถ (1) = ๐(๐ ) โ ๐ข(๐(๐ ))โ , ๐2 (๐ ) = (๐(๐ ))โ . 2.2. For ๐ = ๐ + 2, . . . , ๐ โ 1 perform the following. ๐ = ๐ โ ๐, ๐ค = ๐(๐)๐ข, ๐ข = ๐(๐)๐ข, ๐๐ถ (๐ ) = ๐(๐) โ ๐ข๐คโ , โ
๐๐ถ (๐ ) = ๐(๐) โ ๐ค๐ค , 2.3.
๐ = ๐ โ ๐,
๐๐ถ (๐ ) = ๐(๐),
๐๐ถ (๐ ) = ๐(๐ ),
๐๐ถ (๐ ) = ๐(๐),
๐ค = ๐(๐ )๐ข,
(3.35) โ
๐2 (๐) = ๐ค . (3.36)
๐๐ถ (๐ ) = ๐(๐ ) โ ๐ค๐คโ , (3.37)
๐2 (๐ ) = ๐คโ .
4. Conquer step and eigenproblem of rational matrix functions 4.1. The link between the eigenproblem of ๐จ and an eigenproblem for a rational matrix function In the conquer step, the solutions of the smaller problems into which a larger sized problem has been torn are successfully combined two by two to solutions of the next larger problem. Suppose that for the smaller divided matrices ๐ต and ๐ถ of sizes ๐ ร ๐ and respectively (๐ โ ๐) ร (๐ โ ๐) we already have their spectral data, i.e., we have ๐ ร ๐ and (๐ โ ๐) ร (๐ โ ๐) invertible matrices ๐๐ต and respectively ๐๐ถ so
312
Y. Eidelman and I. Haimovici
that ๐๐ตโ1 ๐ต๐๐ต = ๐ฝ๐ต and ๐๐ถโ1 ๐ถ๐๐ถ = ๐ฝ๐ถ where the matrices ๐ฝ๐ต and ๐ฝ๐ถ are in canonical Jordan form. We must compute the spectral data of the twice larger matrix ๐ด which satis๏ฌes (3.2) with the known ๐ ร ๐๐ and respectively ๐๐ ร ๐ matrices ๐1 , ๐2 given by (3.3). Denote ( ) ๐๐ต 0 ๐= . 0 ๐๐ถ Then ๐ is invertible and ) (( ) ๐ต 0 + ๐1 ๐2 ๐ = ๐ฝ + ๐ง1 ๐ง2 ๐ โ1 ๐ด๐ = ๐ โ1 0 ๐ถ where
( ๐ฝ=
๐ฝ๐ต 0
0 ๐ฝ๐ถ
) ,
while
๐ง1 = ๐ โ1 ๐1 , ๐ง2 = ๐2 ๐ are small rank ๐ ร ๐๐ and respectively ๐๐ ร ๐ matrices. We must now ๏ฌnd an invertible ๐ which brings the matrix ๐พ = ๐ฝ + ๐ง 1 ๐ง2
(4.1)
(4.2)
to its canonical Jordan form, i.e., such that ๐ โ1 (๐ฝ + ๐ง1 ๐ง2 )๐ = ๐ฝ๐ด where ๐ฝ๐ด is the canonical Jordan form of the original matrix ๐ด. We then set ๐ = ๐ ๐ to obtain ๐ โ1 ๐ด๐ = ๐ฝ๐ด . We have therefore to study the eigensystem of the matrix ๐พ de๏ฌned in (4.2). Consider the ๐๐ ร ๐๐ matrix function ๐น (๐) = ๐ผ๐๐ โ ๐ง2 (๐๐ผ๐ โ ๐ฝ)โ1 ๐ง1 .
(4.3)
We will show that the eigenproblem of the potentially large ๐ ร ๐ matrix ๐พ can be reduced to the eigenproblem of a small sized ๐๐ ร ๐๐ matrix function ๐น (๐). Finding zeroes for det ๐น (๐), eigenvectors for the small sized matrix which is obtained when we substitute a zero value in ๐น (๐) and possible Jordan chains for those eigenvectors is all that we need, as the following theorem which is a speci๏ฌcation of a result ๏ฌrst appeared in [10]. Theorem 4.1. Suppose that ๐ฝ is an ๐ ร ๐ square matrix, ๐ง1 is an ๐ ร ๐๐ and ๐ง2 is an ๐๐ ร ๐ matrix and that the matrices ๐ฝ and ๐พ = ๐ฝ + ๐ง1 ๐ง2 have no common eigenvalues. Then ๐0 is an eigenvalue of the ๐ ร ๐ matrix ๐พ and ๐ฅ0 , ๐ฅ1 , . . . , ๐ฅ๐ is a Jordan chain of ๐พ corresponding to ๐0 if and only if ๐0 is a zero of ๐น (๐) = ๐ผ๐๐ โ ๐ง2 (๐๐ผ๐ โ ๐ฝ)โ1 ๐ง1 .
Divide and Conquer for Quasiseparable Matrices
313
and
๐ง2 ๐ฅ0 , ๐ง2 ๐ฅ1 , ๐ง2 ๐ฅ2 , . . . , ๐ง2 ๐ฅ๐ is a Jordan chain of ๐น (๐) corresponding to its zero ๐0 . Moreover, if ๐0 , ๐1 , . . . , ๐๐ is a Jordan chain of the rational matrix function ๐น (๐) for its eigenvalue ๐0 , then the corresponding Jordan chain of ๐พ is given by ๐ฆ๐ =
๐ โ
(โ1)๐ (๐0 ๐ผ๐ โ ๐ฝ)โ(๐+1) ๐ง1 ๐๐โ๐ ,
๐ = 0, 1, . . . , ๐.
(4.4)
๐=0
In particular
๐ฆ0 = (๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐0 (4.5) is an eigenvector of ๐พ for its eigenvalue ๐0 . The correspondence between the Jordan chains of ๐พ and ๐น (๐) is one-toone and onto. In particular, the algebraic multiplicity of an eigenvalue ๐0 of ๐พ coincides with the multiplicity of ๐0 as an eigenvalue of ๐น (๐). Proof. Let ๐0 be a zero of ๐น (๐) and let ๐0 โ= 0 be an eigenvector corresponding to ๐0 . Then (๐0 ๐ผ๐ โ ๐พ)(๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐0 = ((๐0 ๐ผ๐ โ ๐ฝ) โ ๐ง1 ๐ง2 )(๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐0 = (๐ผ โ ๐ง1 ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ1 )๐ง1 ๐0 = ๐ง1 ๐น (๐0 )๐0 = 0. In order to prove that (๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐0 is an eigenvector of ๐พ and ๐0 is one of its eigenvalues, it remains only to prove that (๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐0 โ= 0. Indeed, since ๐0 โ= 0 it follows that ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐0 = โ๐น (๐0 )๐0 + ๐ผ๐๐ ๐0 = 0 + ๐0 โ= 0 and therefore
(๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐0 โ= 0. So we proved that ๐0 is an eigenvalue of ๐พ and (๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐0 is one of its eigenvectors. Consider now a Jordan chain ๐0 , ๐1 , . . . , ๐๐ of ๐น (๐) corresponding to ๐0 , i.e., ๐ โ 1 (๐) ๐น (๐0 )๐๐โ๐ = 0, ๐! ๐=0
๐ = 0, 1, . . . , ๐.
If we write down separately the term for ๐ = 0 and we also perform the derivation, then it follows that ๐ผ๐๐ ๐๐ โ ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐๐ +
๐ โ
(โ1)๐+1 ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ(๐+1) ๐ง1 ๐๐โ๐ = 0
๐=1
so that
โ ๐๐ = ๐ง2 โ
๐ โ ๐=0
โ (โ1)๐ (๐0 ๐ผ๐ โ ๐ฝ)โ(๐+1) ๐ง1 ๐๐โ๐ โ ,
๐ = 0, 1, . . . , ๐.
314
Y. Eidelman and I. Haimovici Denote ๐ฆ๐ =
๐ โ
(โ1)๐ (๐0 ๐ผ๐ โ ๐ฝ)โ(๐+1) ๐ง1 ๐๐โ๐ .
๐=0
Then ๐ง2 ๐ฆ๐ = ๐๐ , ๐ = 0, 1, . . . , ๐ and in particular ๐ฆ0 = (๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐0 is the eigenvector that we previously found for ๐พ. It remains to prove that ๐ฆ0 , ๐ฆ1 , . . . , ๐ฆ๐ is a Jordan chain for ๐พ. (๐0 ๐ผ๐ โ ๐พ)๐ฆ๐+1 = ((๐0 ๐ผ๐ โ ๐ฝ) โ ๐ง1 ๐ง2 )๐ฆ๐+1 = (๐0 ๐ผ๐ โ ๐ฝ)
๐+1 โ
(โ1)๐ (๐0 ๐ผ๐ โ ๐ฝ)โ(๐+1) ๐ง1 ๐๐+1โ๐ โ ๐ง1 ๐ง2 ๐ฆ๐+1
๐=0
and since ๐ง2 ๐ฆ๐+1 = ๐๐+1 it follows that (๐0 ๐ผ๐ โ ๐พ)๐ฆ๐+1 =
๐+1 โ
(โ1)๐ (๐0 ๐ผ๐ โ ๐ฝ)โ๐ ๐ง1 ๐๐+1โ๐ โ ๐ง1 ๐๐+1 .
๐=0
Now, the term for ๐ = 0 reduces itself with โ๐ง1 ๐๐+1 , so that we have in fact equality with ๐+1 โ (โ1)๐ (๐0 ๐ผ๐ โ ๐ฝ)โ๐ ๐ง1 ๐๐+1โ๐ ๐=1
and if we denote ๐ = ๐ โ 1, then the sum becomes ๐ โ
(โ1)๐+1 (๐0 ๐ผ๐ โ ๐ฝ)โ(๐+1) ๐ง1 ๐๐โ๐ = โ๐ฆ๐
๐=0
for ๐ = 0, . . . , ๐ โ 1. In total, we proved that (๐พ โ ๐0 ๐ผ๐ )๐ฆ๐+1 = ๐ฆ๐ , so that ๐ฆ0 , ๐ฆ1 , . . . , ๐ฆ๐ is a Jordan chain for ๐พ. Conversely, let now ๐0 be an eigenvalue of ๐พ. Since ๐พ and ๐ฝ have no common eigenvalues it follows that ๐0 ๐ผ๐ โ ๐ฝ is invertible. Let ๐ฅ0 be an eigenvector of ๐พ corresponding to its eigenvalue ๐0 . Then (๐พ โ ๐0 ๐ผ๐ )๐ฅ0 = 0 so that (๐ฝ โ ๐0 ๐ผ๐ )๐ฅ0 = โ๐ง1 ๐ง2 ๐ฅ0 (4.6) and it follows that ๐น (๐0 )๐ง2 ๐ฅ0 = ๐ผ๐๐ ๐ง2 ๐ฅ0 โ ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐ง2 ๐ฅ0 and using (4.6) this is equal to ๐ง2 ๐ฅ0 + ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ1 (๐ฝ โ ๐0 ๐ผ๐ )๐ฅ0 = 0. In order to show that ๐0 is an eigenvalue of ๐น (๐) and to ๏ฌnd ๐ง2 ๐ฅ0 as an eigenvector, it is su๏ฌcient to prove that this vector is not zero. Indeed, since ๐0 is not an eigenvalue of ๐ฝ and ๐ฅ0 โ= 0 then (๐0 ๐ผ๐ โ ๐ฝ)๐ฅ0 โ= 0 and it follows from (4.6) that ๐ง1 ๐ง2 ๐ฅ0 โ= 0 which implies ๐ง2 ๐ฅ0 โ= 0.
Divide and Conquer for Quasiseparable Matrices
315
Consider now a Jordan chain ๐ฅ0 , ๐ฅ1 , . . . , ๐ฅ๐ of the matrix ๐พ corresponding to its eigenvalue ๐0 . We will denote ๐ฅโ1 = 0 and then we can write from the de๏ฌnition of Jordan chains for a matrix ๐พ that (๐พ โ ๐0 ๐ผ๐ )๐ฅ๐ = (๐ฝ + ๐ง1 ๐ง2 โ ๐0 ๐ผ๐ )๐ฅ๐ = ๐ฅ๐โ1 ,
๐ = 0, 1, . . . , ๐,
so that
๐ง1 ๐ง2 ๐ฅ๐โ๐ = โ(๐ฝ โ ๐0 ๐ผ๐ )๐ฅ๐โ๐ + ๐ฅ๐โ๐โ1 . It follows that ๐ โ 1 (๐) ๐น (๐0 )๐ง2 ๐ฅ๐โ๐ = ๐ผ๐๐ ๐ง2 ๐ฅ๐ โ ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ๐ง2 ๐ฅ๐ ๐! ๐=0 โ
๐ โ
(4.7)
(โ1)๐ ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ(๐+1) ๐ง1 ๐ง2 ๐ฅ๐โ๐
๐=1
and using (4.7) this is equal to ๐ง2 ๐ฅ๐ +
๐ โ (โ1)๐ ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ(๐+1) ((๐ฝ โ ๐0 ๐ผ๐ )๐ฅ๐โ๐ โ ๐ฅ๐โ๐โ1 ) ๐=0
๐ ๐ โ โ ๐ โ๐ = ๐ง2 ๐ฅ๐ โ (โ1) ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ) ๐ฅ๐โ๐ + (โ1)๐+1 ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ(๐+1) ๐ฅ๐โ๐โ1 . ๐=0
๐=0
In fact, the ๏ฌrst entry for ๐ = 0 in the ๏ฌrst sum reduces itself with ๐ง2 ๐ฅ๐ , while the second sum has only ๐ โ 1 non-zero factors since the term for ๐ = ๐ contains the fake vector ๐ฅโ1 = 0, therefore ๐ ๐ โ โ 1 (๐) ๐น (๐0 )๐ง2 ๐ฅ๐โ๐ = โ (โ1)๐ ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ๐ ๐ฅ๐โ๐ ๐! ๐=0 ๐=1
+
๐โ1 โ
(โ1)๐+1 ๐ง2 (๐0 ๐ผ๐ โ ๐ฝ)โ(๐+1) ๐ฅ๐โ๐โ1 = 0.
๐=0
But this is the mere de๏ฌnition of the fact that ๐ง2 ๐ฅ0 , ๐ง2 ๐ฅ1 , . . . , ๐ง2 ๐ฅ๐ is a Jordan chain for ๐น (๐). We will prove now that the correspondence established between the Jordan chains of ๐พ and the Jordan chains of ๐น (๐) is onto. Indeed, if ๐0 , ๐1 . . . , ๐๐ is a Jordan chain for ๐น (๐), then we will build using (4.4) the Jordan chain ๐ฆ0 , ๐ฆ1 , . . . , ๐ฆ๐ of ๐พ and we already know that ๐๐ = ๐ง2 ๐ฆ๐ , ๐ = 0, 1, . . . , ๐ so that the original chain ๐0 , ๐1 , . . . , ๐๐ of ๐น (๐) is the image of the Jordan chain ๐ฆ0 , ๐ฆ1 , . . . , ๐ฆ๐ of ๐พ, so that the correspondence is onto. It remains to prove only that the correspondence established between the Jordan chains of ๐พ and the Jordan chains of ๐น (๐) is one-to-one. To this end note ๏ฌrst that two Jordan chains of ๐น (๐) which correspond to Jordan chains of ๐พ and their lengths are di๏ฌerent correspond to di๏ฌerent Jordan chains of ๐พ since corresponding chains have the same length as the original chains for ๐พ. Note also
316
Y. Eidelman and I. Haimovici
that Jordan chains which correspond to di๏ฌerent eigenvalues are di๏ฌerent and as such they are counted twice. (For an explanation see at the end of Subsection 2.) It remains therefore to prove for Jordan chains of the same lengths of the same eigenvalue. Let ๐ฅ๐,0 , ๐ฅ๐,1 , . . . , ๐ฅ๐,๐ , ๐ = 1, 2 two Jordan chains of ๐พ for the eigenvalue ๐0 and suppose that ๐ง2 ๐ฅ1,๐ = ๐ง2 ๐ฅ2,๐ ,
๐ = 0, 1, . . . , ๐.
We must prove that ๐ฅ1,๐ = ๐ฅ2,๐ ,
๐ = 0, 1, . . . , ๐.
We will prove this by induction. For ๐ = 0 we have ๐พ๐ฅ1,0 = ๐0 ๐ฅ1,0 and also ๐พ๐ฅ2,0 = ๐0 ๐ฅ2,0 , therefore (๐ฝ โ ๐0 ๐ผ๐ )๐ฅ1,0 = โ๐ง1 ๐ง2 ๐ฅ1,0 = โ๐ง1 ๐ง2 ๐ฅ2,0 = (๐ฝ โ ๐0 ๐ผ๐ )๐ฅ2,0 , and since ๐0 is not an eigenvalue of ๐ฝ it follows that ๐ฅ1,0 = ๐ฅ2,0 . Suppose now that for a certain ๐ < ๐ we know that ๐ฅ1,๐ = ๐ฅ2,๐ . Then (๐ฝ โ ๐0 ๐ผ๐ )๐ฅ1,๐+1 = ๐ฅ1,๐ โ ๐ง1 ๐ง2 ๐ฅ1,๐+1 = ๐ฅ2,๐ โ ๐ง1 ๐ง2 ๐ฅ2,๐+1 = (๐ฝ โ ๐0 ๐ผ๐ )๐ฅ2,๐+1 , therefore ๐ฅ1,๐+1 = ๐ฅ2,๐+1 .
โก
If the matrices ๐ฝ and ๐พ in the above theorem have common eigenvalues, then all the other eigenvalues of ๐พ still correspond to eigenvalues of ๐น (๐) and have the same multiplicity, while the eigenvalues of ๐พ which were not found by solving the eigenproblem for ๐น (๐) are readily found among the eigenvalues of ๐ฝ. 4.2. Order one quasiseparable matrices with scalar entries If ๐ด is an order one quasiseparable matrix, or at least ๐๐ = 1, then the perturbations ๐1 and ๐2 from (3.3) are vectors and then ๐ง1 , ๐ง2 de๏ฌned in (4.1) are vectors too and the rational function ๐น (๐) from (4.3) is a scalar function. If ๐ด is a matrix with scalar entries and with order one quasiseparable representation then its generators are complex numbers. Proposition 4.2. Suppose that ๐ฝ is an ๐ ร ๐ matrix in Jordan canonical form, ๐ง1 is a column vector and ๐ง2 is a row vector of lengths ๐ and that the matrix ๐ฝ has at least an eigenvalue of geometric multiplicity greater than one. Then ๐ฝ and ๐พ = ๐ฝ + ๐ง1 ๐ง2 have common eigenvalues. Proof. We build ๏ฌrst the function ๐น (๐) using (4.3). Since ๐๐ = 1 it follows that in the present case the function ๐น (๐) is a scalar function and then ๐ผ๐๐ is equal to 1, i.e., (4.8) ๐น (๐) = 1 โ ๐ง2 (๐๐ผ๐ โ ๐ฝ)โ1 ๐ง1 . Suppose that the Jordan canonical matrix ๐ฝ has ๐ Jordan chains which start with independent eigenvectors for all the eigenvalues of ๐ฝ in total. Denote by
Divide and Conquer for Quasiseparable Matrices
317
๐1 , ๐2 , . . . , ๐๐ the lengths of these ๐ Jordan chains. Then it follows that ๐1 + ๐2 + โ
โ
โ
+ ๐๐ = ๐
(4.9)
which is the size of the square matrices ๐ฝ and ๐พ. Denote ๐1 , ๐2 , . . . , ๐๐ respectively the eigenvalues which correspond to the ๐ Jordan chains. Then we can write the matrix ๐ฝ which is in Jordan canonical form as a block diagonal matrix with the blocks โ โ 0 โ
โ
โ
0 0 ๐๐ 1 โ 0 ๐๐ 1 โ
โ
โ
0 0 โ โ โ โ 0 0 ๐๐ โ
โ
โ
0 0 โ โ โ โ .. .. .. .. .. โ .. โ โ . . . . . . โ โ โ 0 0 0 โ
โ
โ
๐๐ 1 โ 0 0 0 โ
โ
โ
0 ๐๐ of size ๐๐ ร ๐๐ for ๐ = 1, 2, . . . , ๐. It follows that (๐๐ผ๐ โ๐ฝ)โ1 is a block diagonal Toeplitz matrix with the blocks โ โ (โ1)๐๐ โ1 (ฮ๐ )โ๐๐ (ฮ๐ )โ1 โ(ฮ๐ )โ2 โ
โ
โ
(โ1)๐๐ โ2 (ฮ๐ )โ(๐๐ โ1) โ 0 (ฮ๐ )โ1 โ
โ
โ
(โ1)๐๐ โ3 (ฮ๐ )โ(๐๐ โ2) (โ1)๐๐ โ2 (ฮ๐ )โ(๐๐ โ1) โ โ โ โ 0 0 โ
โ
โ
(โ1)๐๐ โ4 (ฮ๐ )โ(๐๐ โ3) (โ1)๐๐ โ3 (ฮ๐ )โ(๐๐ โ2) โ โ โ โ โ .. .. .. .. .. โ โ . . . . . โ โ โ โ โ(ฮ๐ )โ2 0 0 โ
โ
โ
(ฮ๐ )โ1 0
0
โ
โ
โ
(ฮ๐ )โ1
0
(4.10) of size ๐๐ ร ๐๐ , for each ๐ = 1, 2, . . . , ๐. Here in (4.10) ฮ๐ denotes ๐ โ ๐๐ . Then (๐๐ผ๐ โ๐ฝ)โ1 ๐ง1 which appears in the de๏ฌnition (4.8) of the scalar function ๐น (๐) and which now is a column vector of length ๐ which we denote by ๐ค = (๐๐ผ๐ โ ๐ฝ)โ1 ๐ง1 ( = ๐ค๐0 +1 ๐ค2 โ
โ
โ
๐ค๐1
๐ค๐1 +1
โ
โ
โ
๐ค๐1 +๐2
โ
โ
โ
๐ค๐1 +...+๐๐
(4.11) )๐
where ๐0 = 0, has its entries of indexes ๐0 + ๐1 + ๐2 + โ
โ
โ
+ ๐๐โ1 + 1, ๐0 + ๐1 + ๐2 + โ
โ
โ
+ ๐๐โ1 + 2, . . . , ๐0 + ๐1 + ๐2 + โ
โ
โ
+ ๐๐ for ๐ = 1, 2, . . . , ๐ equal to ( โ ๐๐ ๐+1 (๐ โ ๐๐ )โ๐ ๐๐ ๐=1 (โ1) โ
โ
โ
(๐ โ ๐๐ )
โ1
โฃ
โ๐๐ โ1 ๐=1
(โ1)๐+1 (๐ โ ๐๐ )โ๐ ๐๐+1
๐๐๐ โ1 โ (๐ โ ๐๐ )
โ2
๐๐๐
โฃ
โ1
(๐ โ ๐๐ )
๐๐๐
โ
โ
โ
)
where ๐๐ , ๐ = 1, 2, . . . , ๐๐ denote entries of the column vector ๐ง1 as follows ๐๐ = (๐ง1 )๐1 +๐2 +โ
โ
โ
+๐๐โ1 +๐ .
(4.12)
318
Y. Eidelman and I. Haimovici Now, from (4.11) and (4.12) it follows that ๐ง2 (๐๐ผ๐ โ ๐ฝ)โ1 ๐ง1 = ๐ง2 ๐ค =
๐๐ ๐ โ โ
๐๐ โ
((๐ง2 )๐1 +โ
โ
โ
+๐๐โ1 +๐
๐=1 ๐=1
(โ1)๐+1 (๐ โ ๐๐ )โ๐ (๐ง1 )๐1 +๐2 +โ
โ
โ
+๐๐โ1 +๐ ).
๐=1
By rearranging the order of the terms in the most inner sum we obtain that โ1
๐ง2 (๐๐ผ๐ โ ๐ฝ)
๐ง1 =
๐ ๐๐ โ โ ๐=1 ๐=1
๐๐,๐ , (๐ โ ๐๐ )๐
(4.13)
where ๐๐,๐ (with ๐ = 1, 2, . . . , ๐ and ๐ = 1, 2, . . . , ๐๐ ) denote proper complex numbers. If the corresponding eigenvalues ๐1 , ๐2 , . . . , ๐๐ of the ๐ Jordan chains of ๐ฝ are all distinct, then the common denominator ๐(๐) of all the fractions in (4.13) will be ๐(๐) = (๐ โ ๐1 )๐1 (๐ โ ๐2 )๐2 โ
โ
โ
โ
โ
(๐ โ ๐๐ )๐๐ which by (4.9) is a polynomial of degree ๐1 + ๐2 + โ
โ
โ
+ ๐๐ = ๐ . But if at least two of the eigenvalues, say ๐๐1 and ๐๐2 , ๐1 โ= ๐2 are equal, then in (4.13) at least one of the denominators appears twice, in our case ๐ โ ๐๐1 = ๐ โ ๐๐2 . Therefore the degree of the common denominator will be less than ๐ , namely ๐ โ min{๐๐1 , ๐๐2 }. Hence if one of the eigenvalues corresponds to more than a Jordan chain, then ๐ง2 (๐๐ผ๐ โ ๐ฝ)โ1 ๐ง1 from (4.13) is in fact equal to the ratio of two polynomials ๐ง2 (๐๐ผ๐ โ ๐ฝ)โ1 ๐ง1 =
๐(๐) , ๐(๐)
(4.14)
where ๐(๐) is the common denominator and deg ๐(๐) โค ๐ โ 1 and deg ๐(๐) < deg ๐(๐). By (4.8) and (4.14) it follows that ๐น (๐) = 1 โ
๐(๐) ๐(๐) โ ๐(๐) = ๐(๐) ๐(๐)
with deg(๐(๐) โ ๐(๐)) = max{deg ๐(๐), deg ๐(๐)} โค (๐ โ 1) since deg ๐(๐) is such. The number of the eigenvalues (zeroes) of the function ๐น (๐) including multiplicities is therefore less than ๐ . But on the other hand the total eigenvalues multiplicity of the ๐ ร ๐ matrix ๐พ is ๐ since the characteristic polynomial of ๐พ has degree ๐ , so that it is not equal to the zero multiplicity of ๐น (๐), but it is strictly larger. This is in contradiction with the result in Theorem 4.1 and it means that this theorem cannot be applied to the matrices ๐พ and ๐ฝ and so it follows that ๐ฝ and ๐พ must have common eigenvalues. โก
Divide and Conquer for Quasiseparable Matrices
319
4.3. Order one quasiseparable matrix ๐จ and diagonalizable matrices ๐ฉ and ๐ช with distinct eigenvalues Suppose further that the Jordan matrix ๐ฝ is in fact diagonal and then denote it by ๐ท. Suppose also that the geometric multiplicity of its eigenvalues is one. This condition asks in fact that the smaller matrices ๐ต and ๐ถ are diagonalizable and that all their eigenvalues are distinct. In this case (๐ง2 )1 (๐ง1 )1 (๐ง2 )2 (๐ง1 )2 (๐ง2 )๐ (๐ง1 )๐ ๐น (๐) = 1 + + + โ
โ
โ
+ (4.15) ๐1 โ ๐ ๐2 โ ๐ ๐๐ โ ๐ where (๐ง1 )๐ , (๐ง2 ), ๐ = 1, . . . , ๐ are the components of the vectors ๐ง1 , ๐ง2 and ๐๐ < ๐๐โ1 < ๐๐ โ2 < โ
โ
โ
< ๐2 < ๐1 the distinct diagonal entries of the diagonal matrix ๐ท. The next Lemma 4.3 gives a su๏ฌcient condition in which the Theorem 4.1 takes place. Lemma 4.3. Let ๐ท be a diagonal ๐ ร ๐ complex matrix and ๐ง1 , ๐ง2 be vectors with ๐ complex components each. Suppose that ๐ท has no equal diagonal entries and that ๐ง1 , ๐ง2 have no zero components. Then ๐ท and the matrix ๐พ = ๐ท + ๐ง1 ๐ง2 given by (4.2) have no common eigenvalues. Proof. Suppose on the contrary that ๐ท and ๐พ have a common eigenvalue ๐ and that ๐ฃ is an eigenvector of ๐พ corresponding to this eigenvalue. Then ๐พ๐ฃ = (๐ท + ๐ง1 ๐ง2 )๐ฃ = ๐๐ฃ
(4.16)
and ๐ฃ โ= 0. Since ๐ท is a diagonal matrix, ๐ท = diag(๐1 , ๐2 , . . . , ๐๐ ), it follows that its eigenvalue ๐ is one of the entries ๐๐ , 1 โค ๐ โค ๐ . Then, if ๐๐ is the corresponding vector in the standard basis of โ๐ , we have that ๐โ๐ (๐ท โ ๐๐ )๐ค = 0
(4.17)
0 = ๐พ๐ฃ โ ๐๐ฃ = (๐ท โ ๐)๐ฃ + ๐ง1 (๐ง2 ๐ฃ) = (๐ท โ ๐๐ )๐ฃ + ๐ง1 (๐ง2 ๐ฃ)
(4.18)
for any vector ๐ค. By (4.16) and by (4.17) 0 = ๐โ๐ ((๐ท โ ๐๐ )๐ฃ + ๐ง1 (๐ง2 ๐ฃ)) = 0 + ๐โ๐ ๐ง1 (๐ง2 ๐ฃ) = (๐ง1 )๐ ๐ง2 ๐ฃ,
(4.19)
where (๐ง1 )๐ is the component ๐ of the vector ๐ง1 , which cannot be zero by the assumptions of the lemma. It follows from (4.19) that ๐ง2 ๐ฃ = 0.
(4.20)
But then (4.18) shows that (๐ท โ ๐๐ )๐ฃ = 0, therefore ๐ฃ is also an eigenvector of ๐ท for the same eigenvalue ๐. Hence ๐ฃ = ๐ผ๐๐ for a complex scalar ๐ผ โ= 0. Therefore ๐ง2 ๐ฃ = ๐ผ(๐ง2 )๐ , where (๐ง2 )๐ is the component ๐ of the vector ๐ง2 , which must be non-zero. Therefore ๐ง2 ๐ฃ โ= 0 which is in contradiction with (4.20). โก
320
Y. Eidelman and I. Haimovici
5. Complete algorithm for Hermitian matrices 5.1. Hermitian order one quasiseparable matrix ๐จ If the initial matrix ๐ด is also Hermitian and its quasiseparable generators satisfy (3.9) then by Corollary 3.2 the smaller matrices ๐ต and ๐ถ are Hermitian too. In this case the results proved in [7] and references therein for the special case of a tridiagonal symmetric matrix ๐ด can be generalized for the larger context of order one quasiseparable Hermitian matrices. Suppose that for the divided matrices ๐ต and ๐ถ of sizes ๐ร๐ and respectively (๐ โ ๐) ร (๐ โ ๐) we already have their Schur decompositions, i.e., we have ๐ ร ๐ and (๐ โ ๐) ร (๐ โ ๐) unitary matrices ๐๐ต and respectively ๐๐ถ so that ๐โ๐ต ๐ต๐๐ต = ๐ท๐ต and ๐โ๐ถ ๐ถ๐๐ถ = ๐ท๐ถ where the matrices ๐ท๐ต and ๐ท๐ถ are diagonal matrices. We must compute the spectral data of the twice larger matrix ๐ด which satis๏ฌes (3.2) with the known column vector ๐1 and row vector ๐2 = ๐1โ given by (3.11). If we denote ( ) ๐๐ต 0 ๐= 0 ๐๐ถ then ๐ is unitary and (( ) ) ๐ต 0 โ โ ๐ ๐ด๐ = ๐ + ๐1 ๐2 ๐ = ๐ท + ๐ง1 ๐ง2 0 ๐ถ where
( ๐ท=
๐ท๐ต 0
0 ๐ท๐ถ
) ,
while
๐ง1 = ๐ โ ๐1 , ๐ง2 = ๐ง1โ = ๐2 ๐ are a column vector which we will also denote by ๐ง and respectively a row vector which is in fact ๐ง โ . We must now ๏ฌnd a unitary ๐ which brings the matrix ๐พ from (4.2), which now becomes ๐พ = ๐ท + ๐ง๐ง โ to its diagonal form, i.e., such that ๐ โ (๐ท + ๐ง๐ง โ )๐ = ๐ท๐ด where ๐ท๐ด is the diagonal matrix in the Schur decomposition of the original matrix ๐ด. We then set ๐ = ๐ ๐ to obtain ๐ โ ๐ด๐ = ๐ท๐ด . In the case when the conditions of Lemma 4.3 are ful๏ฌlled and also ๐ด is a Hermitian matrix it follows that the vector ๐ง has no zero components. In this case, the rational scalar function ๐น (๐) becomes โฃ๐ง1 โฃ2 โฃ๐ง2 โฃ2 โฃ๐ง๐ โฃ2 + + โ
โ
โ
+ ๐1 โ ๐ ๐2 โ ๐ ๐๐ โ ๐ where ๐ง๐ are the components of the vector ๐ง and ๐น (๐) = 1 +
๐๐ > ๐๐โ1 > ๐๐ โ2 > โ
โ
โ
> ๐2 > ๐1
(5.1)
Divide and Conquer for Quasiseparable Matrices
321
are the distinct diagonal entries of the diagonal matrix ๐ท. Because ๐ด is a Hermitian matrix and ๐ง โ= 0 the derivative of ๐น (๐) is negative between the poles ๐๐ , ๐ = 1, . . . , ๐ โฃ๐ง1 โฃ2 โฃ๐ง2 โฃ2 โฃ๐ง๐ โฃ2 ๐น โฒ (๐) = โ โ โ โ
โ
โ
โ (5.2) (๐1 โ ๐)2 (๐2 โ ๐)2 (๐๐ โ ๐)2 so that ๐น (๐) is monotone between its poles. Moreover ๐น (๐) takes all the real values between each two poles, including the value zero. It follows that ๐น (๐) has exactly ๐ roots ๐๐ , ๐ = 1, . . . , ๐ and they satisfy ๐๐ + ๐ง โ ๐ง > ๐๐ > ๐๐ > ๐๐โ1 > ๐๐ โ2 > โ
โ
โ
> ๐2 > ๐1 > ๐1 .
(5.3)
Moreover, from (4.5) we have that the eigenvectors corresponding to the eigenvalues ๐๐ , ๐ = 1, . . . , ๐ are โ โ โ ๐ง1 โ โ โ ๐ง1 ๐ง1 โ ๐ฃ1 = โ โ
๐1 โ๐1 ๐ง2 ๐2 โ๐1
...
โ โ, โ
โ ๐ฃ2 = โ โ
๐ง๐ ๐๐ โ๐1
๐1 โ๐2 ๐ง2 ๐2 โ๐2
...
โ โ, โ
โ
โ
โ
, ๐ฃ๐
โ =โ โ
๐ง๐ ๐๐ โ๐2
๐1 โ๐๐ ๐ง2 ๐2 โ๐๐
...
โ โ โ
(5.4)
๐ง๐ ๐๐ โ๐๐
which must be normalized to obtain the desired orthogonal matrix ๐ . 5.2. The rational function approximation method and the convexifying method for ๏ฌnding zeroes in (5.1) In order to ๏ฌnd the zeroes of the function ๐น (๐) which appeared in (5.1) we will now summarize for completeness two known methods: the local approximation in the region of a root by simple rational functions whose zeroes are easy to compute and as a main method the improved Newton method, i.e., the use of convexifying transformations which precede the search for a root. This methods, which are due to Bunch, Nielsen and Sorensen [2] and respectively to Melman [13] have been especially conceived for rational functions of this type. In both methods, for ๏ฌnding the ๐th root of ๐น (๐) = 1 +
โฃ๐ง1 โฃ2 โฃ๐ง2 โฃ2 โฃ๐ง๐ โฃ2 + + โ
โ
โ
+ ๐1 โ ๐ ๐2 โ ๐ ๐๐ โ ๐
where ๐ = 1, 2, . . . , ๐ โ 1 a linear change of variables ๐ = ๐๐ โ ๐
(5.5)
is performed ๏ฌrst. (Note that the case ๐ = ๐ needs a di๏ฌerent treatment as (5.3) suggests.) This change of variables has numerical advantages for the accurate determination of the updated eigenvectors. After (5.5) the problem becomes to ๏ฌnd the zero ๐๐ of the function ๐น๐ (๐) = 1 +
โฃ๐ง1 โฃ2 โฃ๐ง2 โฃ2 โฃ๐ง๐ โฃ2 + + โ
โ
โ
+ ๐ฟ1 โ ๐ ๐ฟ2 โ ๐ ๐ฟ๐ โ ๐
(5.6)
322
Y. Eidelman and I. Haimovici
where ๐ฟ๐ = ๐๐ โ ๐๐ , ๐ = 1, 2, . . . , ๐ and the root we look for must lie in the interval โ โ ๐โ1 โ 0 < ๐๐ < min โ๐ฟ๐+1 , 1 โ ๐๐ โ . (5.7) ๐=1
The method in [2] which is also recommended in [7] is the following. Denote ฮจ(๐ก) = 1 +
๐ โ โฃ๐ง๐ โฃ2 , ๐ฟ๐ โ ๐ก
ฮฆ(๐ก) =
๐=1
๐ โ โฃ๐ง๐ โฃ2 . ๐ฟ๐ โ ๐ก
(5.8)
๐=๐+1
Then (5.6) becomes โฮจ(๐๐ ) = ฮฆ(๐๐ ) + 1
(5.9)
and both sides are convex but the left side is decreasing and the right side is increasing on (5.7). In order to ๏ฌnd the root, suppose that we already have at a certain stage of the approximation ๐ก๐ between 0 and ๐๐ . The problem is to ๏ฌnd a ๐ก๐+1 โ (๐ก๐ , ๐๐ ), i.e., a better approximation. To this end, the two functions in (5.8) ๐ are approximated by interpolating simpler rational functions ๐โ๐ก , ๐ + ๐ฟ๐+1๐ โ๐ก such that ๐ ๐ = ฮจ(๐ก๐ ), ๐ + = ฮฆ(๐ก๐ ), ๐ โ ๐ก๐ ๐ฟ๐+1 โ ๐ก๐ ๐ ๐ = ฮจโฒ (๐ก๐ ), = ฮฆโฒ (๐ก๐ ). (๐ โ ๐ก๐ )2 (๐ฟ๐+1 โ ๐ก๐ )2 It is easy to compute ๐, ๐, ๐, ๐ and then to solve the quadratic equation ๐ ๐ =1+๐+ ๐ โ ๐ก๐+1 ๐ฟ๐+1 โ ๐ก๐+1 which is an approximation of (5.9). In fact, ๐ก๐+1 = ๐ก๐ + where ๐=
2๐ โ , ๐ + ๐2 โ 4๐
ฮ(1 + ฮฆ๐ ) + ฮจ2๐ /ฮจโฒ๐ ฮจ๐ + โฒ, ๐ ฮจ๐
๐ = 1 + ฮฆ๐ โ ฮฮฆโฒ๐ ,
๐ค = 1 + ฮฆ๐ + ฮจ ๐ ,
ฮฆ๐ = ฮฆ(๐ก๐ ), ฮจ๐ = ฮจ(๐ก๐ ),
(5.10)
๐=
ฮ๐คฮจ๐ , ฮจโฒ๐ ๐
ฮ = ๐ฟ๐+1 โ ๐ก๐ ,
ฮฆโฒ๐ = ฮฆโฒ (๐ก๐ ), ฮจโฒ๐ = ฮจโฒ (๐ก๐ ).
The reasons for arranging the calculations in this way are: ๐ค must be computed anyway for a convergence test, cancellation is minimized and ๐ก๐+1 has an unambiguous sign. In [2] it is proved that starting with any 0 < ๐ก0 < ๐๐ the sequence obtained recursively by (5.10) converges increasingly to ๐๐ quadratically (namely โฃ๐ก๐+1 โ ๐๐ โฃ = ๐(โฃ๐ก๐ โ ๐๐ โฃ2 )).
Divide and Conquer for Quasiseparable Matrices
323
Finally, the case when ๐ = ๐ and we look for the last root is treated. In this case equation (5.9) becomes โฮจ(๐ก) = 1 and accordingly the iterations for obtaining ๐ก๐ are simpler: ๐ก๐+1 = ๐ก๐ +
(1 + ฮจ๐ )ฮจ๐ . โฒ ฮจ๐
We will now describe a method proposed by Melman in [13]. This method is faster, which is important since it has to be used numerous times. It ๏ฌrst performs a further transformation of variables, besides (5.5) and then the function becomes one for which both the Newton method and the secant method converge from any suitably chosen initial point and they do it faster. More speci๏ฌcally, a class of transformation of variables is considered which change the function into a convex one. These transformations must be twice continuously di๏ฌerentiable and also proper, i.e., they are one-to-one and their range (possibly including โ) is su๏ฌcient to cover the values of the original variable. Such a transformation is for instance ๐ค(๐พ) = ๐พ ๐ for 0 < ๐ โค 1. 1 It is shown that if ๐คโฒโฒ (๐พ) โค 0 for all ๐พ such that ๐ค(๐พ) > ๐ฟ๐+1 then the 1 function ๐น๐ (๐) from (5.6) becomes a convex function ๐น๐ ( ๐ค(๐พ) ). It is also shown that if ๐น (๐ฅ) is convex and decreasing (respectively increasing) on a closed interval [๐, ๐] and ๐น (๐)๐น (๐) < 0 then Newtonโs method converges monotonically to the unique solution ๐ฅโ of ๐น (๐ฅ) = 0 from any initial point in [๐, ๐ฅโ ] (respectively [๐ฅโ , ๐]). 1 Moreover, denote the unique solution of ๐น๐ ( ๐ค(๐พ) ) = 0 by ๐พ โ and suppose that ๐คโฒ (๐พ) 1 has also a constant sign for each ๐พ such that ๐ค(๐พ) > ๐ฟ๐+1 > 0. Then Newtonโs 1 method applied to the function ๐น๐ ( ๐ค(๐พ) ) in this interval converges monotonically from any point ( ( ) ) 1 ๐0 โ ๐คโ1 , ๐พโ (5.11) ๐ฟ๐+1 1 or in ๐0 โ [๐พ โ , ๐คโ1 ( ๐ฟ๐+1 )) depending on wether ๐ค is increasing or decreasing. Suppose that ๐ค is increasing and that we start from a point ๐0 as in (5.11). Denote ๐ โ โฃ๐ง๐ โฃ2 ๐
๐ (๐พ) = 1 + โ โฃ๐ง๐ โฃ2 ๐ค(๐พ) + ๐ฟ๐ ๐โ=๐=1
๐ โ ๐=1,๐โ=๐,๐+1
(
โฃ๐ง๐ โฃ 2 ) ๐ฟ๐
๐ค(๐พ) โ
1 ๐ฟ๐
(5.12)
1 which is the rest to remain from ๐น๐ ( ๐ค(๐พ) ) after its dominant most troublesome part ๐+1 โฃ 2 ( โฃ๐ง๐ฟ๐+1 ) ๐ท๐ (๐พ) = (5.13) 1 ๐ค(๐พ) โ ๐ฟ๐+1
is deleted. Then a sequence ๐๐ which converges to the root ๐พ โ faster than the Newton method which starts with the same ๐0 satis๏ฌes โฒ
๐
๐ (๐๐ ) + ๐
๐ (๐๐ )(๐๐+1 โ ๐๐ ) + ๐ท๐ (๐๐+1 ),
324
Y. Eidelman and I. Haimovici
where ๐
๐ (๐พ), ๐ท๐ (๐พ) have been de๏ฌned in (5.12), (5.13) and ๐๐+1 stays in the interval de๏ฌned in (5.11). 1 1 For ๐ = ๐ , the function ๐น๐ ( ๐ค(๐พ) ) is almost the same but, since ๐ = ๐ค(๐พ) , it is โ1 โ1 de๏ฌned on [๐ค (0), โ) and ๐ค (0) can be a starting point (in which the function equals 1). Its dominant part is ๐ท๐ (๐พ) =
( โฃ๐ง๐ฟ11 โฃ )2 ๐ค(๐พ) โ
1 ๐ฟ1
instead of the formula in (5.13). 5.3. Repeated diagonal entries and zero components for ๐ In applying Lemma 5.1 below we need to determine up to machine precision when two diagonal entries of ๐ท are distinct and when an entry of ๐ง is not zero. To this end, suppose that ๐ก๐๐ is a small multiple of the machine precision, for instance ๐ก๐๐ = ๐ข(โฃโฃ๐ทโฃโฃ2 + โฃโฃ๐งโฃโฃ2 ). By Lemma 5.1 we can determine an orthogonal matrix ๐1 and an integer 1 โค ๐ โค ๐ such that ๐๐1 ๐ท๐1 = diag(๐1 , . . . , ๐๐ ) (zeroes are up to ๐ก๐๐) and a vector ๐ค = ๐๐1 ๐ง such that ๐๐+1 โ ๐๐ โฅ ๐ก๐๐ for ๐ = 1, . . . , ๐ โ 1, โฃ๐ค๐ โฃ โฅ ๐ก๐๐ for 1 โค ๐ โค ๐ and โฃ๐ค๐ โฃ < ๐ก๐๐ otherwise. The next Lemma 5.1 shows that one can relax the conditions that ๐ง has only non-zero components and that the diagonal elements of ๐ท are all distinct from one another. Lemma 5.1. Let ๐ท = diag(๐1 , ๐2 , . . . , ๐๐ ) be a diagonal real matrix and let ๐ง be a vector with ๐ components. Then there exists a unitary matrix ๐1 such that if ๐โ1 ๐ท๐1 = diag(๐1 , ๐2 , . . . , ๐๐ ) and ๐ค = ๐1 ๐ง then ๐1 < ๐2 < โ
โ
โ
< ๐๐ โค ๐๐+1 โค โ
โ
โ
โค ๐๐ , ๐ค๐ โ= 0, ๐ = 1, . . . , ๐ and ๐ค๐ = 0, ๐ = ๐ + 1, . . . , ๐ . The proof is the same as the proof in [7] p. 463, which is made there for the tridiagonal case. 5.4. Diagonalizing ๐ซ + ๐๐โ In order to ๏ฌnd the orthogonal matrix ๐ in the Schur decomposition of a Hermitian order one quasiseparable matrix ๐ด we compute ๐ = ๐1 ๐2 . The orthogonal matrix ๐1 is given by Lemma 5.1. We then take as a new ๐ท the matrix ๐โ1 ๐ท๐1 and we take ๐ค as a new ๐ง. It follows that the ๏ฌrst ๐ entries of the diagonal matrix ๐ท are in strictly decreasing order and that the ๏ฌrst ๐ entries of ๐ง are non zero. We proceed with ห 2 such that ๏ฌnding the ๐ ร ๐ matrix ๐ ห โ (๐ท(1 : ๐, 1 : ๐) + ๐ง(1 : ๐)๐ง โ (1 : ๐))๐ ห 2 = diag(๐1 , . . . , ๐๐ ). ๐ 2
(5.14)
We can therefore apply Theorem 4.1 to an ๐ ร ๐ problem, so that we must ๏ฌrst determine the ๐ distinct zeroes of the rational function ๐น (๐) in (5.1) but with only
Divide and Conquer for Quasiseparable Matrices
325
ห 2 is found by ๐ poles. If ๐๐ , ๐ = 1, . . . , ๐ are these zeroes then the ๐๐กโ column of ๐ ห 2 , ๐ผ๐โ๐ ). normalizing ๐ฃ๐ , ๐ = 1, . . . , ๐ from (5.4). Finally, we consider ๐2 = diag(๐ Thus we obtain the following algorithm Algorithm 5.2. Let ๐ท = diag(๐1 , ๐2 , . . . , ๐๐ ) be a diagonal real matrix and let ๐ง = (๐ง๐ )๐ ๐=1 be a vector column. Then the unitary matrix ๐ and the real diagonal matrix ฮ such that ๐ท + ๐ง๐ง โ = ๐ ฮ๐ โ are obtained by the following algorithm. 1. Determine the number ๐ of distinct diagonal entries for ๐ท, the matrix ๐1 such that ห = ๐โ ๐ท๐1 = diag(๐1 , ๐2 , . . . , ๐๐ ) ๐ท 1 with ๐1 < ๐2 < โ
โ
โ
< ๐๐ โค ๐๐+1 โค โ
โ
โ
โค ๐๐ , and the vector ๐โ1 ๐ง = ๐ค = (๐ค๐ )๐ ๐=1 with the ๏ฌrst ๐ entries di๏ฌerent from zero and ๐ค๐+1 = โ
โ
โ
= ๐ค๐ = 0 as in Lemma 5.1. ห = diag(๐1 , ๐2 , . . . , ๐๐ ), ๐ค 2.1. Set ๐ท ห = (๐ค๐ )๐๐=1 and using one of the iteration methods which have been described in Subsection 5.2 compute the ๐ eigenvalues ห + ๐ค( ๐1 , . . . , ๐๐ of the matrix ๐ท ห ๐ค) ห โ with ๐ instead of ๐ . 2.2. Find ๐ eigenvectors ๐ฃ1 , ๐ฃ2 , . . . , ๐ฃ๐ with formula (5.4). (0)
(0)
(0)
2.3. Compute the normalized eigenvectors ๐ฃ1 , ๐ฃ2 , . . . , ๐ฃ๐ by dividing ๐ฃ1 , ๐ฃ2 , . . . , ๐ฃ๐ by the result of formula โ๐น โฒ (๐๐ ) = 3. Set ห2 = ๐
[
โฃ๐ง1 โฃ2 โฃ๐ง2 โฃ2 โฃ๐ง๐ โฃ2 + + โ
โ
โ
+ , 2 2 (๐1 โ ๐๐ ) (๐2 โ ๐๐ ) (๐๐ โ ๐๐ )2 (0)
๐ฃ1
(0)
๐ฃ2
(0)
. . . ๐ฃ๐
]
,
๐ = 1, . . . , ๐.
(5.15)
ฮ = diag(๐1 , . . . , ๐๐ , ๐๐+1 , . . . , ๐๐ )
ห 2. and compute ๐ = ๐1 ๐ 5.5. The complete algorithm Now we are in position to present the complete divide and conquer algorithm to compute eigendecomposition of a Hermitian matrix with quasiseparable of order one representation. ๐ Algorithm 5.3. Let ๐ด = {๐ด๐๐ }๐ ๐,๐=1 be an ๐ ร๐ Hermitian matrix where ๐ = 2 , with lower quasiseparable generators ๐(๐) (๐ = 2, . . . , ๐ ), ๐(๐) (๐ = 1, . . . , ๐ โ 1), ๐(๐) (๐ = 2, . . . , ๐ โ 1) of order one and diagonal entries ๐(๐) (๐ = 1, . . . , ๐ ). Then the ๐ eigenvalues ๐๐ < ๐๐โ1 < โ
โ
โ
< ๐1 of ๐ด and a unitary matrix ๐ such that ๐ โ ๐ด๐ = diag (๐๐ , ๐๐ โ1 . . . , ๐1 )
are obtained by the following algorithm.
326
Y. Eidelman and I. Haimovici
1. For performing the divide step. Set ๐(0,1) (๐) ๐ = 2, . . . , ๐,
๐ (0,1) (๐) ๐ = 1, . . . , ๐ โ 1,
๐(0,1) (๐) ๐ = 2, . . . , ๐ โ 1), ๐(0,1) (๐) ๐ = 1, . . . , ๐. For ๐ = 1, . . . , ๐ perform the following. Set ๐ = 2๐โ1 , ๐ = 2๐โ๐ . For ๐ = 1, 2, . . . , 2๐ using lower quasiseparable generators ๐(๐โ1,๐) (๐) (๐ = 2, . . . , 2๐), ๐ (๐โ1,๐) (๐) (๐ = 1, . . . , 2๐ โ 1), ๐(๐โ1,๐) (๐), (๐ = 2, . . . , 2๐ โ 1) and diagonal entries ๐(๐โ1,๐) (๐) (๐ = 1, . . . , 2๐) of the matrix ๐ด(๐โ1,๐) compute via Algorithm 3.5 lower quasiseparable generators ๐(๐,2๐โ1) (๐), ๐(๐,2๐) (๐) (๐ = 2, . . . , ๐), ๐(๐,2๐โ1) (๐), ๐ (๐,2๐) (๐) (๐ = 1, . . . , ๐ โ 1), ๐(๐,2๐โ1) (๐), ๐(๐,2๐) (๐) (๐ = 2, . . . , ๐ โ 1) and diagonal entries ๐(๐,2๐โ1) (๐), ๐(๐,2๐) (๐) (๐ = 1, . . . , ๐) of the matrices ๐ด(๐,2๐โ1) , ๐ด(๐,2๐) and the vectors ๐ฆ (๐โ1,๐) such that ( (๐,2๐โ1) ) ๐ด 0 (๐โ1,๐) ๐ด = + ๐ฆ (๐โ1,๐) (๐ฆ (๐โ1,๐) )โ . 0 ๐ด(๐,2๐) 2. For performing the conquer step. Set ฮ(0,๐ก) = ๐(๐,๐ก) (๐ก), ๐ (0,๐ก) = 1,
๐ก = 1, . . . , ๐.
For ๐ = 1, . . . , ๐ perform the following. For ๐ = 1, 2, . . . , 2๐ perform the following. 2.1. Compute ) ( (๐โ1,2๐โ1) ๐ 0 (๐,๐) ๐ฆ (๐,๐) = ๐ง 0 ๐ (๐โ1,2๐) and set ๐ท(๐,๐) = ฮ(๐โ1,2๐โ1) โ ฮ(๐โ1,2๐) . 2.2. Using Algorithm 5.2 determine the eigendecomposition ๐ท(๐,๐) + ๐ง (๐,๐) (๐ง (๐,๐) )โ = ๐ (๐,๐) ฮ(๐,๐) (๐ (๐,๐) )โ with a unitary matrix ๐ (๐,๐) and a real diagonal matrix ฮ(๐,๐) 2.3. Compute ๐ (๐,๐) =
(
๐ (๐โ1,2๐โ1) 0
3. Set ๐ = ๐ (๐,1) , ฮ = ฮ(๐,1) .
0
๐ (๐โ1,2๐)
)
๐ (๐,๐) .
Divide and Conquer for Quasiseparable Matrices
327
6. Conclusions We studied the divide and conquer method used for solving the eigenproblem of large matrices with quasiseparable representations. We analyzed the divide step and the conquer step for matrices with arbitrary quasiseparable order. In the conquer step we proved that in order to reconstruct the eigendata of a larger matrix from the one of the two smaller matrices we have to solve the eigenproblem for a ๐ ร ๐ rational matrix function, where ๐ is a quasiseparable order of a matrix. We gave the complete algorithm for our method in the case of quasiseparable of order one Hermitian matrices. In a future work we will show that the results known in literature for the eigenproblem of unitary Hessenberg matrices can also be obtained as a particular case of our method and perform numerical tests.
References [1] T. Bella, Y. Eidelman, I. Gohberg and V. Olshevsky, Computations with quasiseparable polynomials and matrices, Theoretical Computer Science 409: 158โ179 (2008). [2] J.R. Bunch, C.P. Nielsen and D.C. Sorensen, Rank-one modi๏ฌcation of the symmetric eigenproblem, Numer. Math. 31: 31โ48 (1978). [3] J. Cuppen, A divide and conquer method for symmetric tridiagonal eigenproblem, Numerische Mathematik 36: 177โ195 (1981). [4] J.J. Dongarra and M. Sidani, A parallel algorithm for the non-symmetric eigenvalue problem, Report CS-91-137, University of Tennessee, Knoxville (1991); SIAM J. Sci. Comput. 14: 542โ569 (1993). [5] Y. Eidelman, I. Gohberg and I. Haimovici, Separable type representations of matrices and fast algorithms, to appear. [6] Y. Eidelman, I. Gohberg and V. Olshevsky, Eigenstructure of Order-One-Quasiseparable Matrices. Three-term and Two-term Recurrence Relations, Linear Algebra and its Applications 405: 1โ40 (2005). [7] G.H. Golub and C.F. Van Loan, Matrix Computations, John Hopkins, Baltimore 1989. [8] M. Gu and S. Eisenstat, A divide-and-conquer algorithm for the symmetric tridiagonal eigenproblem, SIAM Journal on Matrix Analysis and Applications 16: 172โ191 (1995). [9] M. Gu, R. Guzzo, X.-B. Chi and X.-Q. Cao, A divide-and-conquer algorithm for the symmetric tridiagonal eigenproblem, SIAM Journal on Matrix Analysis and Applications 25: 385โ404 (2003). [10] I. Haimovici, Operator Equations and Bezout Operators for Analytic Operator Functions, Ph.D. Thesis, Technion, Haifa, 1991. [11] E.R. Jessup, A case against a divide and conquer approach to the non-symmetric eigenvalue problem, Applied Numerical Mathematics 12: 403โ420 (1993). [12] N. Mastronardi, E. Van Camp and M. Van Barel, Divide and conquer algorithms for computing the eigendecomposition of symmetric diagonal-plus-semiseparable matrices, Numerical Algorithms, 9: 379โ398 (2005).
328
Y. Eidelman and I. Haimovici
[13] A. Melman, Numerical solution of a secular equation, Numer. Math. 69: 483โ493 (1995). [14] L. Rodman and M. Schaps, On the partial multiplicities of a product of two matrix polynomials, Integral Equations and Operator Theory, Volume 2, Number 4, 565โ599 (1979). [15] R. Vandebril, M. Van Barel and N. Mastronardi, Matrix computations and semiseparable matrices: Eigenvalue and singular value methods, The John Hopkins University Press (2008). Y. Eidelman and I. Haimovici School of Mathematical Sciences Raymond and Beverly Sackler Faculty of Exact Sciences Tel-Aviv University Ramat-Aviv 69978, Israel e-mail:
[email protected] [email protected]
Operator Theory: Advances and Applications, Vol. 218, 329โ343 c 2012 Springer Basel AG โ
An Identity Satis๏ฌed by Certain Orthogonal Vector-valued Functions Robert L. Ellis In memory of Israel Gohberg and his mathematical prowess
Abstract. In this paper we ๏ฌrst de๏ฌne a class of scalar products on ๐2๐ , the product of an even number of copies of the Wiener algebra ๐ . Then we obtain a sequence of orthogonal elements of ๐2๐ for such a scalar product and derive an identity that they satisfy. Mathematics Subject Classi๏ฌcation (2000). 47B35, 42C05. Keywords. Orthogonal vector-valued functions, inde๏ฌnite scalar product, in๏ฌnite Toeplitz matrix, Wiener algebra, Nehari problem, four block problem.
Introduction In [1], a class of vector-valued functions was investigated that are orthogonal for a scalar product on ๐2 = ๐ ร ๐ , where ๐ is the Wiener algebra of absolutely convergent Fourier series on the unit circle. In the simplest โโ case, a scalar product is de๏ฌned as follows by a function ๐ in ๐ , i.e., ๐(๐ง) = ๐=โโ ๐๐ ๐ง ๐ for โฃ๐งโฃ = 1, โโ where the ๐๐ are complex numbers with ๐=โโ โฃ๐๐ โฃ < โ. Denote any element ๐ of ๐2 as a vector ( (1) ) ๐ ๐= . ๐(2) Then a possibly inde๏ฌnite scalar product is de๏ฌned on ๐2 by โ โ โซ 2๐ 1 ๐(๐๐๐ ) 1 โ ๐(๐๐๐ ) ๐๐ โจ๐, ๐โฉ = ๐(๐๐๐ )โ โ 2๐ 0 ๐๐ ๐(๐ ) 1
(1)
where * denotes the conjugate transpose of a matrix. This scalar product can be expressed in a di๏ฌerent way. For this, let ๐บ = (๐๐โ๐ )โ ๐,๐ =โโ be the in๏ฌnite Toeplitz
330
R.L. Ellis
matrix de๏ฌned by the Fourier โ โ
โ
โ
โ โ
โ
โ
โ โ โ
โ
โ
โ ๐บ=โ โ โ โ โ
coe๏ฌcients of ๐: โ
โ
โ
โ
โ
โ
๐1 โ
โ
โ
โ
โ
โ
โ
โ
โ
๐0 ๐1 โ
โ
โ
and let ๐ be the 2 ร 2 block matrix
(
๐ =
๐โ1 ๐0 ๐1 โ
โ
โ
๐ผ ๐บโ
โ โ
โ
โ
๐โ1 ๐0 โ
โ
โ
โ
โ
โ
๐บ ๐ผ
โ โ โ โ โ โ โ
โ
โ
โ โ โ
โ
โ
โ โ
โ
โ
โ
โ
โ
๐โ1 โ
โ
โ
โ
โ
โ
(2)
)
viewed as an operator on โ2 (โโ, โ) ร โ2 (โโ, โ), whose elements will also be considered as vectors of the form ) ( ๐ ๐ where ๐ = (. . . , ๐ผโ1 , ๐ผ0 , ๐ผ1 , . . . )๐ and ๐ = (. . . , ๐ฝโ1 , ๐ฝ0 , ๐ฝ1 , . . . )๐ . Here the superscript ๐ denotes the transpose of a matrix. For any ๐ and ๐ in ๐2 let โ โ โ โ ๐ผ๐ ๐ง ๐ , ๐(2) (๐ง) = ๐ฝ๐ ๐ง ๐ ๐(1) (๐ง) = ๐ (1) (๐ง) =
๐=โโ โ โ
๐พ๐ ๐ง ๐ ,
๐ (2) (๐ง) =
๐=โโ
( ๐๐ = where
Then
๐ ๐
๐=โโ โ โ
๐ฟ๐ ๐ง ๐
๐=โโ
)
(
,
๐๐ =
๐ ๐
)
๐ = (. . . , ๐ผโ1 , ๐ผ0 , ๐ผ1 , . . . )๐ ,
๐ = (. . . , ๐ฝโ1 , ๐ฝ0 , ๐ฝ1 , . . . )๐
๐ = (. . . , ๐พโ1 , ๐พ0 , ๐พ1 , . . . )๐ ,
๐ = (. . . , ๐ฟโ1 , ๐ฟ0 , ๐ฟ1 , . . . )๐ .
(3) โจ๐, ๐โฉ = ๐๐โ ๐ ๐๐ . An orthogonal family {๐๐ โฃ๐ = 0, ยฑ1, ยฑ2, . . . } can be obtained as follows, provided the indicated solutions exist. Suppose that for any integer ๐ there are โ1 vectors (๐) (๐) (๐) ๐๐ = (๐ผ๐(๐) , ๐ผ๐+1 , . . . )๐ , ๐๐ = (. . . , ๐ฝโ๐โ1 , ๐ฝโ๐ )๐ such that ) ( ( )( ) ๐1 ๐ผ ๐บ๐ ๐๐ = (4) ๐บโ๐ ๐ผ ๐๐ 0 where โ โ โ
โ
โ
๐๐+1 ๐๐ (5) ๐บ๐ = โ โ
โ
โ
๐๐+2 ๐๐+1 โ โ
โ
โ
โ
โ
โ
โ
โ
โ
An Identity Satis๏ฌed by Certain Functions and
331
๐1 = (1, 0, 0, . . . )๐
and where ๐ผ denotes variously the identity matrix of the appropriate size. For any integer ๐, let ( ) ๐ผ๐ ๐๐ = ๐ฝ๐ where โ โโ โ โ (๐) (๐) ๐ผ๐ (๐ง) = ๐ผ๐ ๐ง ๐ and ๐ฝ๐ (๐ง) = ๐ฝ๐ ๐ง ๐ . ๐=๐
๐=โ๐
Then {๐๐ โฃ๐ = 0, ยฑ1, ยฑ2, . . . } is an orthogonal family of vectors in ๐2 for the scalar product (1). Furthermore ๐ผ๐ and ๐ฝ๐ satisfy the identity โฃ๐ผ๐ (๐ง)โฃ2 โ โฃ๐ฝ๐ (๐ง)โฃ2 = ๐ผ(๐) ๐
for โฃ๐งโฃ = 1.
(6)
Solutions of (4) will exist and hence an orthogonal family will exist, for example, when โฃโฃ๐บ๐ โฃโฃ < 1 for every integer ๐. The functions {๐๐ }โ ๐=โโ appear in a linear fractional description of all solutions of the Nehari problem. See [4, 5]. Identities similar to (6) also appear in [2, 3]. In this paper the preceding results will be generalized. For any given positive integer ๐, a scalar product will be de๏ฌned on ๐2๐ = ๐ ร ๐ ร โ
โ
โ
ร ๐ , the product of 2๐ copies of ๐ , by means of a function โ โ ๐11 (๐ง) ๐12 (๐ง) โ
โ
โ
๐1๐ (๐ง) โ โ
โ
โ
โ โ
โ
โ
โ
โ
โ
โ
โ
โ
โ ๐(๐ง) = โ (7) โ โ
โ
โ
โ โ
โ
โ
โ
โ
โ
โ
โ
โ
๐๐1 (๐ง) ๐๐2 (๐ง) โ
โ
โ
๐๐๐ (๐ง) where ๐11 , ๐12 , . . . , ๐๐๐ are in ๐ . Then an orthogonal system of vectors in ๐2๐ will be found by solving an equation analogous to (4), and an identity analogous to (6) will be proved.
1. A scalar product Let ๐ denote the Wiener algebra of absolutely convergent Fourier series on the unit circle, and let ๐ be a ๏ฌxed positive integer. Denote by ๐2๐ the product ๐ ร ๐ ร โ
โ
โ
ร ๐ of 2๐ copies of ๐ . The elements of ๐2๐ will be represented as column vectors of the form (๐(1) , ๐(2) , . . . , ๐(2๐) )๐ . Let ๐ be a matrix-valued function as in (7), with ๐11 , ๐12 , . . . , ๐๐๐ in ๐ . Then ๐ de๏ฌnes a weight ( ) ๐ผ ๐(๐ง) ฮฉ(๐ง) = ๐(๐ง)โ ๐ผ for the corresponding possibly inde๏ฌnite scalar product on ๐ given by โซ 2๐ 1 โจ๐, ๐โฉ = ๐(๐๐๐ )โ ฮฉ(๐๐๐ ) ๐(๐๐๐ ) ๐๐. 2๐ 0
(8)
332
R.L. Ellis
First we prove that this scalar product can be re-expressed in a manner similar to (3). For 1 โค ๐, ๐ โค ๐ let โ โ ๐๐๐ (๐ง) = ๐๐(๐,๐) ๐ง ๐ ๐=โโ
(๐,๐)
and let ๐บ๐๐ be the corresponding in๏ฌnite Toeplitz matrix ๐บ๐๐ = (๐๐โ๐ )โ ๐,๐ =โโ . (See (2).) Let ( ) ๐ผ ๐บ ๐ = (9) ๐บโ ๐ผ where ๐บ is the ๐ ร ๐ block matrix (๐บ๐๐ )๐ ๐,๐=1 and ๐ผ denotes the appropriate (1) (2) identity matrix. For any ๐ = (๐ , ๐ , . . . , ๐(2๐) ) in ๐2๐ and for 1 โค ๐ โค 2๐, let โ โ ๐ ๐(๐) (๐ง) = ๐ผ(๐) (10) ๐ ๐ง ๐=โโ
and
(๐)
(๐)
(๐)
๐๐(๐) = (. . . , ๐ผโ1 , ๐ผ0 , ๐ผ1 , . . . )๐
and let
๐๐ = (๐๐(1) , ๐๐(2) , . . . , ๐๐(2๐) )๐ .
Proposition 1.1. For any ๐ and ๐ in ๐2๐ , โจ๐, ๐โฉ = ๐๐โ ๐ ๐๐ .
(11) (๐)
Proof. Denote each ๐(๐) as in (10), and denote each ๐ (๐) as in (10) with ๐ฝ๐ (๐) replacing ๐ผ๐ . Then โซ 2๐ ( ) 1 โจ๐, ๐โฉ = ๐ (1) (๐๐๐ ), . . . , ๐ (2๐) (๐๐๐ ) 2๐ 0 ( )( )๐ ๐ผ ๐(๐๐๐ ) (1) ๐๐ (2๐) ๐๐ ร ๐ (๐ ), . . . , ๐ (๐ ) ๐๐ ๐(๐๐๐ )โ ๐ผ โซ 2๐ ( )( )๐ 1 = ๐ (1) (๐๐๐ ), . . . , ๐ (2๐) (๐๐๐ ) ๐1 (๐๐๐ ), . . . , ๐(2๐) (๐๐๐ ) ๐๐ 2๐ 0 where for 1 โค ๐ โค ๐, ๐๐
(๐)
๐๐ (๐ ) = ๐
๐๐
(๐ ) +
๐ โ
๐๐๐ (๐๐๐ )๐(๐+๐) (๐๐๐ )
๐=1
and for ๐ + 1 โค ๐ โค 2๐, ๐๐ (๐๐๐ ) =
๐ โ ๐=1
๐๐,๐โ๐ (๐๐๐ ) ๐(๐) (๐๐๐ ) + ๐(๐) (๐๐๐ ).
An Identity Satis๏ฌed by Certain Functions Therefore 1 โจ๐, ๐โฉ = 2๐ +
โซ
2๐
โฉ
0
2๐ โ ๐ โ
โง ๐ โจโ
๐ (๐) (๐๐๐ ) ๐(๐) (๐๐๐ ) +
๐=1
๐ โ ๐ โ
๐ (๐) (๐๐๐ ) ๐๐๐ (๐๐๐ ) ๐(๐+๐) (๐๐๐ )
๐=1 ๐=1 2๐ โ
๐ (๐) (๐๐๐ ) ๐๐,๐โ๐ (๐๐๐ ) ๐๐ (๐๐๐ ) +
๐=๐+1 ๐=1
333
๐=๐+1
โซ โฌ ๐ (๐) (๐๐๐ ) ๐(๐) (๐๐๐ ) ๐๐. (12) โญ
Combining the ๏ฌrst and last sums on the right side of (12), we ๏ฌnd that โซ 2๐ {โ 2๐ โ โ โ โ 1 (๐) ๐๐ ๐ โจ๐, ๐โฉ = ๐ฝ๐ ๐โ๐๐๐ ๐ผ(๐) ๐ ๐ 2๐ 0 ๐=โโ ๐ =โโ ๐=1
๐ โ
+
โ โ
(๐)
๐,๐=1 ๐,๐ ,๐ก=โโ ๐ โ
+
โ โ
(๐+๐) ๐๐ก๐
๐ฝ๐ ๐โ๐๐๐ ๐๐ (๐,๐) ๐๐๐ ๐ ๐ผ๐ก
๐
} (๐+๐) ๐ฝ๐
(๐,๐) ๐โ๐๐๐ ๐๐ ๐โ๐๐ ๐
๐,๐=1 ๐,๐ ,๐ก=โโ
=
2๐ โ โ โ ๐=1 ๐=โโ
๐ โ
(๐)
๐ฝ๐ ๐ผ(๐) ๐ + +
๐ โ
โ โ
(๐)
๐,๐=1 ๐,๐ก=โโ โ โ
(๐+๐)
๐ฝ๐
๐,๐=1 ๐,๐ก=โโ
We also have ๐๐โ ๐ ๐๐ = (๐๐(1) , . . . , ๐๐(2๐) )โ
(
๐ผ ๐บโ
(๐) ๐ผ๐ก ๐๐๐ก๐
๐๐
(๐,๐) (๐+๐)
๐ฝ๐ ๐๐โ๐ก ๐ผ๐ก (๐,๐)
(๐)
๐๐กโ๐ ๐ผ๐ก .
๐บ ๐ผ
)
(13)
(๐๐(1) , . . . , ๐๐(2๐) )๐
= (๐๐(1) , . . . , ๐๐(2๐) )โ (๐1 , . . . , ๐2๐ )๐ where for 1 โค ๐ โค ๐, ๐๐ = ๐๐(๐) +
๐ โ
๐บ๐๐ ๐๐(๐+๐)
๐=1
and for ๐ + 1 โค ๐ โค 2๐, ๐๐ =
๐ โ
๐บโ๐,๐โ๐ ๐๐(๐) + ๐๐(๐) .
๐=1
Therefore ๐๐โ ๐ ๐๐
=
๐ โ ๐=1
๐๐โ (๐) ๐๐(๐)
+
๐ โ ๐,๐=1
+
๐๐โ (๐) ๐บ๐,๐ ๐๐(๐+๐)
2๐ โ ๐=๐+1
๐๐โ (๐)
๐ โ ๐=1
๐บโ๐,๐โ๐ ๐๐(๐)
+
2๐ โ ๐=๐+1
๐๐โ (๐) ๐๐(๐) .
334
R.L. Ellis
Combining the ๏ฌrst and last sums, we have ๐๐โ
๐ ๐๐ =
2๐ โ โ โ ๐=1 ๐=โโ
(๐) ๐ฝ๐ ๐ผ(๐) ๐
+ +
๐ โ
โ โ
(๐) (๐,๐)
๐,๐=1 ๐,๐ก=โโ ๐ โ โ โ ๐,๐=1 ๐,๐ก=โโ
๐ฝ๐ ๐๐โ๐ก ๐ผ(๐+๐) ๐ (14) (๐+๐) ๐ฝ๐
(๐,๐) (๐) ๐๐กโ๐ ๐ผ๐ก .
From (13) and (14) we conclude that (11) holds.
โก
2. An orthogonal system We will generate an orthogonal system for the scalar product in the preceding section by solving equations that are appropriate analogs of (4). We continue to let ๐บ = (๐บ๐๐ )๐ ๐,๐=1 be the ๐ ร ๐ block matrix in (9). For any integer ๐, let โ [๐] โ [๐] ๐บ11 โ
โ
โ
๐บ1,๐โ1 ๐ป๐ โ โ โ โ โ ๐บ โ โ 21 โ
โ
โ
๐บ2,๐โ1 ๐บ2๐ [๐] โ โ โ ๐บ๐ = โ โ
(15) โ
โ
โ
โ โ โ
โ โ
โ
โ
โ โ โ ๐บ๐1 โ
โ
โ
๐บ๐,๐โ1 ๐บ๐๐ [๐] โ [๐]
[๐]
where ๐บ11 , . . . , ๐บ1,๐โ1 are formed from ๐บ11 , . . . , ๐บ1,๐โ1 by deleting all rows above the ๐th row; ๐บ2๐ [๐], . . . , ๐บ๐๐ [๐] result from ๐บ2๐ , . . . , ๐บ๐๐ by deleting all columns to the right of the ๐th column; and โ (1,๐) (1,๐) โ โ
โ
โ
๐๐+1 ๐๐ โ โ โ โ (1,๐) (1,๐) โ ๐ป๐ = โ โ
โ
โ
๐๐+2 (16) ๐๐+1 โ โ โ โ โ
โ
โ
โ
โ
โ
โ
โ
โ
is the in๏ฌnite Hankel matrix obtained from ๐บ(1,๐) by deleting all rows above the ๐th row and all columns to the right of the ๐th column. We regard ๐บ๐ as a โsectionโ of ๐บ analogous to the matrix in (5). In generating an orthogonal system we will solve equations in the form ( )( ) ( ) ๐ผ ๐บ๐ ๐๐ ๐1 = . ๐บโ๐ ๐ผ ๐๐ 0 Here ๐ผ represents two di๏ฌerent identity matrices of the appropriate sizes, and ๐1 = (1, 0, 0, . . . )๐ ,
(๐)
๐ ๐๐ = (๐1 , . . . , ๐(๐) ๐ ) ,
(๐)
(๐)
๐๐ = (๐๐+1 , . . . , ๐2๐ )๐
An Identity Satis๏ฌed by Certain Functions
335
where (๐)
= (๐ผ(1,๐) , ๐ผ๐+1 , . . . )๐ ๐
(๐)
(2๐,๐)
๐1
(1,๐)
(2๐,๐)
๐2๐ = (. . . , ๐ผโ๐โ1 , ๐ผโ๐
is in โ1 (๐, โ) )
is in โ1 (โโ, โ๐)
and for 2 โค ๐ โค 2๐ โ 1, (๐)
๐๐
(๐,๐)
(๐,๐)
= (. . . , ๐ผโ1 , ๐ผ0
(๐,๐)
, ๐ผ1
,...)
is in โ1 (โโ, โ).
Theorem 2.1. Let ๐๐๐ (1 โค ๐, ๐ โค ๐) be in ๐ . Suppose that for any integer ๐ there are โ1 -vectors ๐๐ and ๐๐ such that )( ) ) ( ( ๐๐ ๐1 ๐ผ ๐บ๐ (17) = ๐บโ๐ ๐ผ ๐๐ 0 (1)
(2๐) ๐
Let ๐๐ = (๐๐ , . . . , ๐๐
) , where
๐(1) ๐ (๐ง) = ๐(๐) ๐ (๐ง) =
โ โ
(1,๐) ๐
๐ผ๐
๐=๐ โ โ
(๐,๐) ๐
๐=โโ โโ โ
๐๐(2๐) (๐ง) =
๐=โ๐
๐ง
๐ผ๐
๐ง
for 2 โค ๐ โค 2๐ โ 1
(2๐,๐) ๐
๐ผ๐
๐ง
where the ๐ผโs are as described before the theorem. Then {๐๐ }โ ๐=โโ is an orthogonal system in ๐2๐ for the scalar product in (8). Proof. For any integer ๐ let (๐)โฒ
๐1 and
(1,๐)
= (. . . , 0, 0, ๐ผ(1,๐) , ๐ผ๐+1 , . . . )๐ ๐
(๐)โฒ
(2๐,๐)
(2๐,๐)
๐2๐ = (. . . , ๐ผโ๐โ1 , ๐ผโ๐ Then
(๐)โฒ
๐๐(1) = ๐1 ๐
and
, 0, 0, . . . )๐ . (๐)โฒ
๐๐(2๐) = ๐2๐ . ๐
For any two integers ๐ and ๐ with ๐ > ๐ , (11) implies that โจ๐๐ , ๐๐ โฉ = ๐๐โ๐ ๐ ๐๐๐ . But because of the leading zeros in ๐๐(1) and ๐๐(1) and the trailing zeros in ๐๐(2๐) ๐ ๐ ๐ and ๐๐(2๐) , it follows that ๐ )( ( ) ๐๐ ๐ผ ๐บ๐ โ โจ๐๐ , ๐๐ โฉ = (๐๐ ๐๐ ) ๐บโ๐ ๐ผ ๐๐
336
R.L. Ellis
where ๐๐ has at least ๐ โ ๐ leading zeros. Thus (17) implies that ) ( ๐1 = 0. โจ๐๐ , ๐๐ โฉ = (๐๐ ๐๐ )โ 0 This proves the theorem.
โก
Just as the functions {๐๐ }โ ๐=โโ in the Introduction are related to the Nehari problem, the functions {๐๐ }โ ๐=โโ in Theorem 2.1 are related to the Four Block problem. See Section II.4 in [4].
3. An identity In this section we will derive an identity similar to (6) associated with the orthogonal functions in Section 2. We will ๏ฌx an integer ๐ and, for simplicity, suppress ๐ in some of the notation. Thus we will write (17) as ) ( )( ) ( ๐1 ๐ ๐ผ ๐บ๐ = (18) ๐ ๐บโ๐ ๐ผ 0 where ๐ = (๐1 , ๐2 , . . . , ๐๐ )๐
๐ = (๐๐+1 , ๐๐+2 , . . . , ๐2๐ )๐
and
with (1)
๐ ๐1 = (๐ผ(1) ๐ , ๐ผ๐+1 , . . . ) (๐)
(๐)
(๐)
๐๐ = (. . . , ๐ผโ1 , ๐ผ0 , ๐ผ1 , . . . )๐ ๐2๐ = (. . . ,
(2๐) ๐ผโ๐โ1 ,
for 2 โค ๐ โค 2๐ โ 1
(2๐) ๐ผโ๐ )๐ .
To emphasize the analogy with (6), we will also use ๐ผ in place of the function ๐๐ obtained in Theorem 2.1. Thus ๐ผ = (๐ผ1 , . . . , ๐ผ2๐ )๐ , where ๐ผ1 (๐ง) = ๐ผ๐ (๐ง) =
โ โ
(1)
๐ผ๐ ๐ง ๐
๐=๐ โ โ ๐=โโ
๐ผ2๐ (๐ง) =
โ๐ โ ๐=โโ
(๐)
๐ผ๐ ๐ง ๐
for 2 โค ๐ โค 2๐ โ 1
(2๐) ๐
๐ผ๐
๐ง .
For any ๐ = (. . . , ๐1 , ๐0 , ๐1 , . . . )๐ in โ1 (โโ, โ), any ๐ = (๐๐ , ๐๐+1 , ๐๐+2, . . . )๐ in โ1 (๐, โ), and any ๐ = (. . . , ๐โ๐โ1 , ๐โ๐ ) in โ1 (โโ, โ๐), we de๏ฌne in๏ฌnite
An Identity Satis๏ฌed by Certain Functions Toeplitz matrices by
โ
โ โ โ โ ๐ (๐) = โ โ โ โ โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
๐1 โ
โ
โ
โ
โ
โ
โ
โ
โ
๐0 ๐1 โ
โ
โ
๐โ1 ๐0 ๐1 โ
โ
โ
โ
๐๐ ๐๐+1 ๐๐+2 โ
โ
โ
โ 0 ๐๐ ๐๐+1 โ
โ
โ
โ ๐ (๐) = โ 0 0 ๐๐ โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ ๐โ๐ ๐โ๐โ1 ๐โ๐โ2 โ 0 ๐โ๐ ๐โ๐โ1 โ ๐ (๐) = โ 0 0 ๐โ๐ โ
โ
โ
โ
โ
โ
โ
โ
โ
and we let โ โ
โ
โ
โ โ
โ
โ
โ ๐
=โ โ โ
โ
โ
โ โ
โ
โ
โ
โ
โ
โ
โ
โ
0 0 1 โ
โ
โ
โ
โ
โ
0 1 0 โ
โ
โ
โ
โ
โ
1 0 0 โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
337
โ
โ
โ
๐โ1 ๐0 โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
๐โ1 โ
โ
โ
โ
โ
โ
โ โ โ โ โ
โ
โ
โ
โ
โ
โ โ โ
โ
โ
โ โ
โ
โ
โ
โ
โ โ โ โ โ
and
โ โ โ โ โ โ โ
โ
โ
โ โ โ
โ
โ
โ โ
โ
โ
โ
โ
โ
โ โ
โ
โ
๐
+ = โ โ โ
โ
โ
โ
โ
โ
0 0 1 โ
โ
โ
0 1 0 โ
โ
โ
โ 1 0 โ โ. 0 โ โ
โ
โ
Left multiplication by either ๐
or ๐
+ reverses the rows of a matrix, provided the multiplication is possible. Theorem 3.1. Let ๐ผ = (๐ผ1 , ๐ผ2 , . . . , ๐ผ2๐ )๐ be the ๐th orthogonal function obtained from (18) as in Theorem 2.1. Then the identity ๐ โ
2๐ โ
โฃ๐ผ๐ (๐ง)โฃ2 โ
๐=1
โฃ๐ผ๐ (๐ง)โฃ2 = ๐ผ(1) ๐
for โฃ๐งโฃ = 1
๐=๐+1
(1)
holds, where ๐ผ๐ denotes the coe๏ฌcient of ๐ง ๐ in ๐ผ1 . Proof. The matrix ๐บ๐ in (18) is given by (15). The ๏ฌrst and last rows in (18) imply that ๐1 + and
๐โ1 โ โ=1
๐ป๐โ ๐1 +
[๐]
๐บ1โ ๐๐+โ + ๐ป๐ ๐2๐ = ๐1
๐ โ ๐=2
๐บ๐๐ [๐]โ ๐๐ + ๐2๐ = 0.
(19)
(20)
338
R.L. Ellis
Observe from (16) that โ ๐ป๐ ๐2๐
โ
โ
โ
(1,๐)
๐๐
โ โ =โ โ โ
โ
โ
โ
๐๐+2
(1,๐)
๐๐+1
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ โ โ โ โ โ =โ โ โ โ โ
(1,๐)
๐๐+1
(2๐)
๐ผโ๐
(1,๐)
โ
โ
โ
โ
โ
โ
โ โโ โโ โ โ (2๐) โโ ๐ผ โ โ โ๐โ1 โ
โ โ โ โ โ โ โ
(2๐)
๐ผโ๐โ1
(2๐)
๐ผโ๐โ2
(2๐)
(2๐)
๐ผโ๐โ1
(2๐)
0
๐ผโ๐
0
0
๐ผโ๐
โ
โ
โ
โ
โ
โ
โ
โ
โ
(2๐)
๐ผโ๐ โ โ (1,๐) ๐๐ โ
โ
โ
โโ โโ (1,๐) โ โ ๐๐+1 โ
โ
โ
โโ โ โโ โ
โโ โโ โ
โ
โ
โ โโ โ
โ โ โ โ
โ
โ
โ
โ โ โ โ โ โ โ. โ โ โ โ โ
Thus ๐ป๐ ๐2๐ = ๐ (๐2๐ )๐พ๐ (1,๐)
where ๐พ๐ = (๐๐
(1,๐)
, ๐๐+1 , . . . )๐ . Also โ
๐ป๐โ
(21)
โ
โ
โ
โ โ (1,๐) ๐1 = โ ๐๐+1 โ (1,๐) ๐๐ โ โ โ โ =โ โ โ โ
โ
โ
โ
0 0 (1)
๐ผ๐
โ
โ
โ
โ
โ
โ
โ
โ
(1)
๐ผ๐
โ
โ โ โ โ ๐ผ(1) โ โ โ ๐+1 โ โ
โ
โ
โโ โ
โ โ โ โ โ โ
โ (1,๐) ๐๐+1 โ
โ
โ
โ
โโ (1,๐) โ
โ
โ
โ
โ
โ
โ
โ
โ
๐๐ โโ (1) โ 0 ๐ผ๐ โ
โ
โ
โ โ โ ๐ (1,๐) โ โ ๐+1 โโ (1) (1) โ
๐ผ๐ ๐ผ๐+1 โ
โ
โ
โ โ โ โ โ
(1) (1) ๐ผ๐+1 ๐ผ๐+2 โ
โ
โ
โ
(1,๐) ๐๐+2
โ โ โ โ โ โ โ โ
so that ๐
+ ๐ป๐โ ๐1 = ๐ (๐1 )๐พ๐ .
(22)
Since upper triangular Toeplitz matrices commute, it follows from (21) and (22) that ๐ (๐1 ) ๐ป๐ ๐2๐ = ๐ (๐2๐ ) ๐
+ ๐ป๐โ ๐1 .
(23)
An Identity Satis๏ฌed by Certain Functions
339
Solving (19) for ๐ป๐ ๐2๐ and (20) for ๐ป๐โ ๐1 , and substituting in (23) leads to ๐โ1 โ
๐ (๐1 )๐1 โ ๐ (๐1 )๐1 โ
โโ1
= โ๐ (๐2๐ )๐
+ ๐2๐ โ
[๐]
๐ (๐1 ) ๐บ1โ ๐๐+โ
๐ โ
๐ (๐2๐ )๐
+ ๐บ๐๐ [๐]โ ๐๐
๐=2
which we rewrite as ๐ (๐1 ) ๐1 +
๐โ1 โ โ=1
โ
๐ โ
[๐]
๐ (๐1 ) ๐บ1โ ๐๐+โ (24)
๐ (๐2๐ )๐
+ ๐บ๐๐ [๐]โ ๐๐ โ ๐ (๐2๐ )๐
+ ๐2๐ = ๐ (๐1 )๐1 .
๐=2
From rows 2 through ๐ in (18) we have ๐๐ +
๐โ1 โ
๐บ๐โ ๐โ+๐ + ๐บ๐๐ [๐] ๐2๐ = 0
for 2 โค ๐ โค ๐
(25)
โ=1
and from rows ๐ + 1 through 2๐ โ 1 in (18) we have [๐] (๐บ1โ )โ
๐1 +
๐ โ
๐บโ๐โ ๐๐ + ๐๐+โ = 0
for 1 โค โ โค ๐ โ 1.
(26)
๐=2
From (25) it follows that ๐โ1 โ
๐บ๐โ ๐โ+๐ = โ๐๐ โ ๐บ๐๐ [๐] ๐2๐
for 2 โค ๐ โค ๐
โ=1
and hence that ๐ ๐โ1 โ โ
๐ (๐๐ ) ๐
๐บ๐โ ๐โ+๐ = โ
๐=2 โ=1
๐ โ ๐=2
๐ (๐๐ ) ๐
๐๐ โ
๐ โ
๐ (๐๐ ) ๐
๐บ๐๐ [๐] ๐2๐ . (27)
๐=2
Similarly it follows from (26) that ๐โ1 ๐ โโ โ=1 ๐=2
๐ (๐๐+โ )โ ๐บโ๐โ ๐๐ = โ
๐โ1 โ โ=1
๐ (๐๐+โ )โ ๐๐+โ โ
๐โ1 โ โ=1
[๐]
๐ (๐๐+โ )โ (๐บ1โ )โ ๐1 . (28)
340
R.L. Ellis
Next we observe that for 2 โค ๐ โค ๐ and 1 โค โ โค ๐ โ 1, โ โ โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ โ (๐) (๐) โ โ
โ
โ
๐ผ(๐) ๐ผโ1 ๐ผโ2 โ
โ
โ
โ 0 โ โ โ โ (๐) (๐) (๐) โ ๐ (๐๐ ) ๐
๐บ๐๐ ๐โ+๐ = โ โ
โ
โ
๐ผ1 ๐ผ0 ๐ผโ1 โ
โ
โ
โ โ โ โ (๐) (๐) (๐) โ โ ๐ผ1 ๐ผ0 โ
โ
โ
โ โ โ
โ
โ
๐ผ2 โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ โ โ
โ
โ
โ โ โ ร โ โ
โ
โ
โ โ โ โ โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
(๐โ) ๐2
(๐โ) ๐1
(๐โ) ๐0
(๐โ)
๐0
(๐โ)
๐1 ๐0
โ
โ
โ
(๐โ)
๐โ1
๐โ1
(๐โ)
๐โ2
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ โ
โ
โ
โ โ โ โ โ
โ
โ
โ โ โ โ โ
โ
โ
โ
(๐โ) (๐โ)
โ
โ
โ
โ
โ
โ
โ
โ โ โ (โ+๐) โ ๐ผโ1 โ โ โ ๐ผ(โ+๐) โ 0 โ โ (โ+๐) โ ๐ผ1 โ โ โ
โ
โ โ โ โ โ โ โ. โ โ โ โ โ โ
Therefore for any integer ๐, and for 2 โค ๐ โค ๐ and 1 โค โ โค ๐ โ 1, the ๐th entry of ๐ (๐๐ ) ๐
๐บ๐๐ ๐โ+๐ =
โ โ
(๐)
๐,๐ =โโ
(๐โ)
(โ+๐)
๐ผ๐โ๐ ๐โ๐โ๐ ๐ผ๐
.
(29)
In the same way we ๏ฌnd that for any integer ๐, and for 2 โค ๐ โค ๐ and 1 โค โ โค ๐ โ 1, the ๐th entry of ๐ (๐๐+โ )โ ๐บโ๐โ ๐๐ =
โ โ ๐,๐ =โโ
(โ+๐)
(๐โ)
๐ผโ๐+๐ ๐โ๐ +๐ ๐ผ(๐) ๐ .
(30)
The two sums in (29) and (30) are easily seen to be equal, so it follows from (27)โ(30) that ๐ โ
๐ (๐๐ ) ๐
๐๐ +
๐=2
=
๐ โ
๐ (๐๐ ) ๐
๐บ๐๐ [๐] ๐2๐
๐=2 ๐โ1 โ
๐โ1 โ
โ
๐ (๐๐+โ ) ๐๐+โ +
โ=1
(31) โ
๐ (๐๐+โ )
โ=1
[๐] (๐บ1โ )โ
๐1 .
We can carry out a similar analysis of the sums in (24) and (31). We ๏ฌnd that for 1 โค โ โค ๐ โ 1, [๐]
the ๐th entry of ๐ (๐1 ) ๐บ1โ ๐๐+โ =
โ โ โ โ
๐ =0 ๐=โโ
(1)
(1,โ)
๐ผ๐+๐ ๐๐+๐ +๐โ๐ ๐ผ(๐+โ) ๐
for ๐ โฅ 0
(32)
An Identity Satis๏ฌed by Certain Functions and
341
[๐]
the ๐th entry of ๐ (๐๐+โ )โ (๐บ1โ )โ ๐1 โ โ โ โ
=
๐=โโ ๐ =0
(โ)
(1)
๐ผ๐+โ ๐๐โ๐โ๐+๐ ๐ผ๐+๐ ๐
for โ โ < ๐ < โ.
(33)
Therefore [๐]
the ๐th entry of ๐ (๐1 ) ๐บ1โ ๐๐+โ [๐]
= the (โ๐)th entry of ๐ (๐๐+โ )โ (๐บ1โ )โ ๐1
for ๐ โฅ 0
and hence [๐]
the ๐th entry of ๐ (๐1 ) ๐บ1โ ๐๐+โ [๐]
= the ๐th entry of ๐
๐ (๐๐+โ )โ (๐บ1โ )โ ๐1 Thus if we let
โ
โ
โ
โ
โ โ
โ
โ
๐ =โ โ โ
โ
โ
โ
โ
โ
0 โ
โ
โ
โ
โ
โ
โ
โ
โ
0 0 โ
โ
โ
โ
โ
โ
1 0 0 โ
โ
โ
0 1 0 โ
โ
โ
0 0 1 โ
โ
โ
โ
โ
โ
0 0 โ
โ
โ
for ๐ โฅ 0. โ
โ
โ
โ
โ
โ
0 โ
โ
โ
โ โ
โ
โ
โ
โ
โ
โ โ โ
โ
โ
โ โ
โ
โ
be the matrix that projects โ1 (โโ, โ) onto โ1 (0, โ), we can conclude that [๐]
[๐]
๐ (๐1 ) ๐บ1โ ๐๐+โ = ๐ ๐
๐ (๐๐+โ )โ (๐บ1โ )โ ๐1 .
(34)
Similarly, for 1 โค ๐ โค ๐ โ 1, the ๐th entry of ๐ (๐2๐ ) ๐
+ (๐บ๐๐ [๐])โ ๐๐ =
โ โ โ โ
(2๐)
๐ =0 ๐=โโ
and
(๐๐)
(๐)
๐ผโ๐โ๐ ๐๐+๐ +๐+๐ ๐ผ๐
for ๐ โฅ 0
(35)
the ๐th entry of ๐ (๐๐ ) ๐
๐บ๐๐ [๐] ๐2๐ =
โ โ โ โ
(๐)
๐ =0 ๐=โโ
=
โ โ โ โ ๐ =0 ๐=โโ
(๐๐)
(2๐)
(๐๐)
(2๐)
๐ผโ๐ ๐๐โ๐โ๐+๐ ๐ผโ๐โ๐ ๐ผ(๐) ๐๐+๐โ๐+๐ ๐ผโ๐โ๐ ๐
for โ โ < ๐ < โ
so that the ๐th entry of ๐
๐ (๐๐ ) ๐
๐บ๐๐ [๐] ๐2๐ =
โ โ โ โ ๐ =0 ๐=โโ
(๐๐)
(2๐)
๐(๐) ๐๐+๐+๐+๐ ๐ผโ๐โ๐ ๐
for โ โ < ๐ < โ.
(36)
For ๐ โฅ 0 the sums in (35) and (36) are complex conjugates of each other, so ๐ (๐2๐ ) ๐
+ ๐บ๐๐ [๐]โ ๐๐ = ๐ ๐
๐ (๐๐ ) ๐
๐บ๐๐ [๐] ๐2๐ .
(37)
342
R.L. Ellis
Multiplying both sides of (31) by ๐ ๐
and substituting from (34) and (37), we have ๐ โ
๐
๐
๐ (๐๐ ) ๐
๐๐ +
๐=2
๐โ1 โ
=๐
๐ โ
๐ (๐2๐ ) ๐
+ ๐บ๐๐ [๐]โ ๐๐
๐=2
๐
๐ (๐๐+โ )โ ๐๐+โ +
๐โ1 โ
โ=1
so that
๐โ1 โ โ=1
โ=1
[๐] ๐ (๐1 ) ๐บ1โ
=๐
๐ โ
๐๐+โ โ
๐ โ
[๐]
๐ (๐1 ) ๐บ1โ ๐๐+โ
๐ (๐2๐ ) ๐
+ ๐บ๐๐ [๐]โ ๐๐
๐=2
๐
๐ (๐๐ ) ๐
๐๐ โ ๐
๐=2
๐โ1 โ
๐
๐ (๐๐+โ )โ ๐๐+โ .
โ=1
Substituting this into (24), we have ๐ (๐1 ) ๐1 + ๐
๐ โ
๐
๐ (๐๐ ) ๐
๐๐ โ ๐
๐=2 (1) ๐ผ๐ .
(38)
For any function ๐ค(๐ง) = Then (๐)
(๐)โฏ
๐
๐ (๐๐+โ )โ ๐๐+โ โ ๐ (๐2๐ ) ๐
+ ๐2๐
โ=1
= ๐ (๐1 )๐1 =
๐ผ1 (๐ง) ๐ผ1
๐โ1 โ
โโ
๐=โโ
๐ค๐ ๐ง ๐ in ๐ , we let ๐คโฏ (๐ง) =
(1)
(1)
โโ
๐=โโ
๐คโ๐ ๐ง ๐ .
(1)
๐ ๐+1 (๐ง) = (๐ผ(1) + โ
โ
โ
) (๐ผ๐ ๐ง โ๐ + ๐ผ๐+1 ๐ง โ๐โ1 + โ
โ
โ
) ๐ ๐ง + ๐ผ๐+1 ๐ง โ โ
=
๐ฝ๐ ๐ง ๐
๐=โโ
where ๐ฝ๐ =
โ โ ๐ =0
(1)
(1)
๐ผ๐+๐ +๐ ๐ผ๐+๐
for ๐ โฅ 0
and
๐ฝ๐ =
โ โ ๐ =0
(1)
(1)
๐ผ๐+๐ ๐ผ๐+๐ โ๐
for ๐ < 0.
In particular ๐ฝโ๐ = ๐ฝ๐ for โโ < ๐ < โ. But for ๐ โฅ 0, the ๐th entry of ๐ (๐1 ) ๐1 =
โ โ ๐ =0
(1)
(1)
๐ผ๐+๐ ๐ผ๐+๐ +๐ = ๐ฝ๐ .
Thus for ๐ โฅ 0, the ๐th entry in ๐ (๐1 ) ๐1 equals the coe๏ฌcient of ๐ง ๐ in ๐ผ1 (๐ง) ๐ผโฏ1 (๐ง). Similar calculations show that for ๐ โฅ 0, the ๐th entry in ๐ (๐2๐ ) ๐
+ ๐2๐ equals the coe๏ฌcient of ๐ง ๐ in ๐ผ2๐ (๐ง) ๐ผโฏ2๐ (๐ง), and for any ๐ and for 2 โค ๐ โค ๐, the ๐th entry in ๐
๐ (๐๐ ) ๐
๐๐ equals the coe๏ฌcient of ๐ง ๐ in ๐ผ๐ (๐ง)๐ผโฏ๐ (๐ง), and for any ๐ and for 1 โค โ โค ๐ โ 1, the ๐th entry in ๐
๐ (๐๐+โ )โ ๐๐+โ equals the coe๏ฌcient of ๐ง ๐ in ๐ผ๐+โ (๐ง)๐ผโฏ๐+โ (๐ง).
An Identity Satis๏ฌed by Certain Functions
343
From these results and (38) we can conclude that all the coe๏ฌcients of the positive powers of ๐ง in ๐ โ ๐=1
๐ผ๐ (๐ง)๐ผโฏ๐ (๐ง) โ
2๐ โ ๐=๐+1
๐ผ๐ (๐ง)๐ผโฏ๐ (๐ง)
(39)
(1)
are zero and the constant term is ๐ผ๐ . Since the coe๏ฌcients of the negative powers of ๐ง in (39) are the complex conjugates of the coe๏ฌcients of the positive powers, it follows that ๐ โ ๐=1
๐ผ๐ (๐ง)๐ผโฏ๐ (๐ง) โ โ1
For ๐ง on the unit circle, ๐ง hence ๐ โ
2๐ โ ๐=๐+1
(1)
๐ผ๐ (๐ง)๐ผโฏ๐ (๐ง) = ๐ผ๐ . (1)
= ๐ง, so it follows that ๐ผโฏ๐ (๐ง) = ๐ผ๐ (๐ง), ๐ผ๐ is real, and
โฃ๐ผ๐ (๐ง)โฃ2 โ
๐=1
This proves Theorem 3.1.
2๐ โ
โฃ๐ผ๐ (๐ง)โฃ2 = ๐ผ(1) ๐ .
๐=๐+1
โก
It is to be expected that the inversion formula in Section 2 of [1] and the inverse problem in Section 4 of [1] can be generalized to the present situation.โ
References [1] R.L. Ellis and I. Gohberg, โOrthogonal systems related to in๏ฌnite Hankel matrices,โ J. Funct. Anal. 109: 155โ198 (1992) [2] R.L. Ellis, I. Gohberg, and D.C. Lay, โIn๏ฌnite analogues of block Toeplitz matrices and related orthogonal functions,โ Integral Equations and Operator Theory 22: 375โ 419 (1995) [3] R.L. Ellis, I. Gohberg, and D.C. Lay, โOn a class of block Toeplitz matrices,โ Linear Algebra Appl. 241: 225โ245 (1996) [4] I. Gohberg, M.A. Kaashoek, and H.J. Woerdeman, โThe band method for positive and contractive extension problems,โ J. Operator Theory 22: 109โ155 (1989) [5] I. Gohberg, M.A. Kaashoek, and H.J. Woerdeman, โThe band method for positive and strictly contractive extension problems: An alternative version and new applications,โ Integral Equations and Operator Theory 12: 343โ382 (1989) Robert L. Ellis Department of Mathematics University of Maryland College Park, Maryland 20742, USA e-mail:
[email protected]
โ The
author would like to thank the reviewer for several useful suggestions.
Operator Theory: Advances and Applications, Vol. 218, 345โ357 c 2012 Springer Basel AG โ
Invertibility of Certain Fredholm Operators Israel Feldman and Nahum Krupnik To the blessed memory of our dear teacher Israel Gohberg
Abstract. Some new classes of algebras in which each Fredholm operator is invertible are described. Mathematics Subject Classi๏ฌcation (2000). Primary 47A53, Secondary 45E10. Keywords. Fredholm operators, spectrum of linear operators, generalized Gelfand transform.
1. Introduction Let ฮฉ be a unital subalgebra of a Banach algebra ๐ฟ(โฌ), where ๐ฟ = ๐ฟ(โฌ) is the algebra of all linear bounded operators on a Banach space โฌ; ๐ฆ(ฮฉ) โ the ideal of all compact operators ๐พ โ ฮฉ; ๐ฆ(โฌ) := ๐ฆ (๐ฟ(โฌ)) (for short); ๐ฆ0 (ฮฉ) โ the ideal of all ๏ฌnite-dimensional operators ๐พ โ ฮฉ, ๐น = ๐น (โฌ) โ the set of all ๐น -operators (Fredholm operators) on โฌ and ๐บ๐ฟ โ the group of all invertible operators in ๐ฟ. Also spec(๐ด) denotes the spectrum of an operator ๐ด in the algebra ๐ฟ(โฌ) and ๐(๐ด)(= โ โ spec(๐ด)) the regular set of operator ๐ด. Recall that algebra ฮฉ is inverse closed in ๐ฟ(โฌ) if ๐ด โ ฮฉ โฉ ๐บ๐ฟ =โ ๐ดโ1 โ ฮฉ. We say that ฮฉ is ๐น -closed if for each operator ๐ด โ ฮฉ โฉ ๐น, there exists an operator ๐
โ ฮฉ such that at least one of the operators ๐
๐ด โ ๐ผ or ๐ด๐
โ ๐ผ is compact. Note that in this case both operators ๐
๐ด โ ๐ผ and ๐ด๐
โ ๐ผ are compact. In the sequel we say that ฮฉ is an ๐น ๐น -algebra (Fredholm free algebra) if ฮฉ does not have Fredholm operators non-invertible in ๐ฟ(โฌ). A following characterization of Fredholm free ๐ถ โ subalgebras is well known: Theorem 1.1. Let ๐ป be a Hilbert space and let ๐ be a ๐ถ โ -subalgebra of ๐ฟ(๐ป). Then the following two statements are equivalent: (i) Algebra ๐ does not contain non-zero compact operators. (ii) Algebra ๐ is an ๐น ๐น -algebra. The research of the second author was partially supported by Retalon Inc., Toronto, ON, Canada.
346
I. Feldman and N. Krupnik
See, for example, [CL, Theorem 3.5], or (more conveniently) see Corollaries 2.5 and 2.7 below. Theorem 1.1 is no longer true if we replace ๐ โ ๐ฟ(๐ป) by an arbitrary subalgebra ฮฉ โ ๐ฟ(โฌ). In Section 2 we study the connections between the invertibility of ๐น -operators and the structure of compact operators in some subalgebras ฮฉ โ ๐ฟ(โฌ). Let ๐ป be a Hilbert space. A subalgebra ฮฉ โ ๐ฟ(๐ป) is called selfadjoint if ๐ด โ ฮฉ โ ๐ดโ โ ฮฉ. The closure of a selfadjoint ๐น ๐น -subalgebra has a following hereditary property: Theorem 1.2. Let ๐ป be a Hilbert space and let ฮฉ be a selfadjoint subalgebra of ๐ฟ(๐ป). If ฮฉ is an ๐น ๐น -algebra, then ๐ := clos(ฮฉ) is an ๐น ๐น -algebra, too. See, for example, [KF, Theorem 1], but it also follows from Theorem 3.3, Statements 1โ and 3โ below. Some classes of non-selfadjoint subalgebras ฮฉ โ ๐ฟ(โฌ) with the hereditary property like in Theorem 1.2 were studied in [KF], [KMF], [MF]. In Section 3 we continue these studies. We obtain some general properties of Banach subalgebras ๐ which have dense ๐น ๐น -subalgebras, and, in particular, obtain some su๏ฌcient conditions under which algebra ๐ with a dense ๐น ๐น -subalgebra ฮฉ is an ๐น ๐น -algebra, too. The following was stated by A. Markus (see [KF, pp. 11โ12]): Proposition 1.3. Let ๐ โ ๐ฟ(โฌ) be a commutative algebra and ฮฉ its dense subalgebra. If ฮฉ is an ๐น ๐น -algebra, then ๐ is an ๐น ๐น -algebra, too. In Section 3 some generalizations of this statement are obtained for the algebras ๐ which admit so-called Generalized Gelfand Transform as well as for algebras ๐ with standard Amitsur-Levitski polynomial identities (of some order ๐ = 2๐): โ ๐ ๐๐(๐)๐๐(1) ๐๐(2) โ
โ
โ
๐๐(๐) = 0, (๐๐ โ ๐), (1.1) ๐โ๐๐
where ๐ runs through the symmetric group ๐๐ . In Section 4 some illustrative examples and open questions are presented. In the sequel, we suppose all Banach spaces โฌ in๏ฌnite-dimensional and all subalgebras of ๐ฟ(โฌ) (except the ideals!) unital. Sometimes we mention this in the text, but sometimes it is not mentioned. It is our pleasure to thank our friend A. Markus for useful remarks and comments.
2. The structure of compact operators in ๐ญ ๐ญ -subalgebras of ๐ณ(ํ) Let โฌ be a Banach space. In this section we denote by ฮฉ an arbitrary (closed or non-closed) unital subalgebra of ๐ฟ(โฌ) and study the connections between the statements (i) and (ii) of Theorem 1.1 for subalgebra ฮฉ. We start with the following two examples:
Invertibility of Certain Fredholm Operators
347
Example 2.1. Let ๐ (โ= 0) be a ๏ฌnite-dimensional operator in a Hilbert space ๐ป (or in any in๏ฌnite-dimensional Banach space โฌ) and let ๐ 2 = 0. Denote ฮฉ = {๐๐ผ +๐๐ }, where ๐, ๐ โ โ. Let ๐ด = ๐๐ผ + ๐๐. If ๐ = 0 then ๐ด is not a Fredholm operator; if ๐ โ= 0 then ๐ดโ1 = ๐1 ๐ผ โ ๐๐2 ๐. Thus ฮฉ is an ๐น ๐น -algebra, but it contains a ๏ฌnite-dimensional operator ๐. Example 2.2. Let ฮฉ denote the algebra of all lower-triangular Toeplitz operators on โ2 (or on any โ๐ , ๐ โ (1, โ)). Algebra ฮฉ does not contain non-zero compact operators1, but it contains non-invertible Fredholm operator ๐ ๐ฅ = (0, ๐ฅ1 , ๐ฅ2 , . . . ), i.e., ฮฉ is not an ๐น ๐น -algebra. Conclusion 2.3. Examples 2.1 and 2.2 show that for the general subalgebra ฮฉ (even in Hilbert spaces), the statements (i) and (ii) from Theorem 1.1 are independent. Thus (in contrast with Theorem 1.1) an ๐น ๐น -algebra ฮฉ may have non-zero compact operators. In continuation of this section we study the structure of the ideals of compact operators in ๐น ๐น -algebras. Recall that a two-sided ideal ๐ฝ of an algebra ฮฉ is called a nil-ideal (a quasinilpotent ideal) if all its elements are nilpotent (quasinilpotent). Proposition 2.4. Let ฮฉ(โ ๐ฟ(โฌ)) be an ๐น ๐น -algebra. Then ๐ฆ(ฮฉ) is a quasinilpotent ideal in ฮฉ. In particular, ๐ฆ0 (ฮฉ) is a nil-ideal in ฮฉ, and it is not necessarily that ๐ฆ(ฮฉ) = {0} or ๐ฆ0 (ฮฉ) = {0}. Proof. It is clear that ๐ฆ(ฮฉ) is a two-sided ideal in ฮฉ. Let ๐พ โ ๐ฆ(ฮฉ) and ๐ด = ๐พ โ๐๐ผ. If ๐ โ= 0, then ๐ด is an ๐น -operator and by the condition of the proposition it is invertible. Thus spec(๐พ) = {0}, i.e., ๐ฆ(ฮฉ) is a quasinilpotent ideal. In addition, Example 2.1 illustrates that this ideal is not necessarily trivial. In the mentioned example ๐ฆ(ฮฉ) = ๐ฆ0 (ฮฉ) = {๐๐ } is a nil-ideal. To complete the proof we give an example of an ๐น ๐น -algebra which contains in๏ฌnite-dimensional compact operators. Let {๐ } ( ) โ ๐ฅ2 ๐ฅ3 ๐ฅ๐+1 ๐ ๐ฅ โ โ2 , ๐ ๐ฅ := , ,..., ,... ฮฉ= ๐๐ ๐ : ๐ โ โ (2.1) 2 3 ๐+1 ๐=0 and ๐ := clos(ฮฉ) โ ๐ฟ(โ2 ). Here ๐ is a in๏ฌnite-dimensional quasinilpotent compact (Hilbert-Schmidt) operator; ฮฉ = {๐๐ผ} โ ๐ฆ(ฮฉ), where ๐ โ โ. It is clear that ฮฉ is an ๐น ๐น -algebra. Note that clos(ฮฉ) โ ๐ฟ(โ2 ) is an ๐น ๐น -algebra (with in๏ฌnite-dimensional compact operators), too. โก Corollary 2.5. Let ฮฉ(โ ๐ฟ(๐ป)) be a selfadjoint ๐น ๐น -algebra. Then ฮฉ does not have non-zero compact operators. Proof. Let ๐พ โ ๐ฆ(ฮฉ), then ๐พ๐พ โ โ ๐ฆ(ฮฉ), too. By Proposition 2.4 ๐พ๐พ โ is quasinilpotent. Thus โฅ๐พโฅ2 = โฅ๐พ๐พ โ โฅ = max{๐ : ๐ โ spec(๐พ๐พ โ )} = 0. 1 See,
โก
for example, the proof of Statement 4โ in Theorem 2.11 and compare with its Statement 2โ .
348
I. Feldman and N. Krupnik
An inverse question. Let ๐ฆ(ฮฉ) be a quasinilpotent ideal in ฮฉ. Is ฮฉ an ๐น ๐น -algebra? The answer is negative even when ๐ฆ(ฮฉ) is a nil-ideal and (moreover) even if ๐ฆ(ฮฉ) = {0}. This can be con๏ฌrmed by Example 2.2, where ๐ฆ(ฮฉ) = {0}, but ฮฉ is not an ๐น ๐น -algebra. Now we are going to restrict the algebra ฮฉ โ ๐ฟ(โฌ) with some conditions so that the implication (i) โ (ii) would hold in ฮฉ. We start with Proposition 2.6. Let ฮฉ be a ๐น -closed subalgebra of ๐ฟ(โฌ). If ฮฉ does not contain non-zero compact operators, then ฮฉ is an ๐น ๐น -algebra. Proof. Let ๐ด โ ๐น โฉ ฮฉ. Since ฮฉ is ๐น -closed there exists an operator ๐ต โ ฮฉ such that ๐ต๐ดโ๐ผ = ๐พ1 and ๐ด๐ต โ๐ผ = ๐พ2 are compact operators in ฮฉ. By the conditions of the proposition ๐พ1 = ๐พ2 = 0, i.e., ๐ด โ ๐บ๐ฟ. โก Corollary 2.7. Let ฮฉ be a ๐ถ โ -subalgebra of ๐ฟ(๐ป). If ฮฉ does not have non-zero compact operators, then ฮฉ is an ๐น ๐น -algebra. This statement follows from Proposition 2.6 and the following Lemma 2.8. Each ๐ถ โ -subalgebra ฮฉ โ ๐ฟ(๐ป) is ๐น -closed. ห := ๐ฟ(๐ป)/๐ฆ(๐ป). It is Proof. Let ๐ be the canonical homomorphism ๐ฟ(๐ป) โ ๐ฟ โ ห well known that ๐ฟ is a ๐ถ -algebra. Denote ห = {๐ ห โ๐ฟ ห such that ๐ โ1 (๐) ห โฉ ฮฉ โ= โ
}. ฮฉ ห is a ๐ถ โ -subalgebra of ๐ฟ. ห Let ๐ด โ ๐น โฉ ฮฉ, then ๐ดห It is not di๏ฌcult to check that ฮฉ ห (because ๐ถ โ -subalgebras are inverse closed). Thus, there exists is invertible in ฮฉ ห โ ๐ผห = ๐ต ห ๐ดห โ ๐ผห = 0 and hence operators ๐ด๐ต โ ๐ผ and ๐ต๐ด โ ๐ผ ๐ต โ ฮฉ such that ๐ดห๐ต are compact. โก To give another condition which provides the implication (i) โ (ii), we need a following de๏ฌnition. Let ๐ be a subset of ๐ฟ(โฌ). We say that ๐ is symmetric if for any ๐ด โ ๐ there exists an operator ๐ด โ ๐ such that spec(๐ด๐ด) โ โ. Theorem 2.9. Let ฮฉ be a Banach subalgebra of ๐ฟ(โฌ). Assume that the set ๐ of all Fredholm operators ๐ด โ ฮฉ is symmetric. If the algebra ฮฉ does not contain nilpotent ๏ฌnite-dimensional operators, then it is an ๐น ๐น -algebra. Proof. Assume that there exists a non-invertible ๐น -operator ๐ด โ ฮฉ. Then there exists ๐ด โ ฮฉ โฉ ๐น such that spec(๐ด๐ด) โ โ. Since ๐ด is not invertible, it follows that at least one of the operators ๐ด๐ด or ๐ด๐ด (we denote it by ๐ต ) is not invertible. Thus ๐ต (โ ฮฉ) is a non-invertible ๐น -operator and spec(๐ต) โ โ. Recall that spec(๐ต) denotes the spectrum of operator ๐ต in algebra ๐ฟ(โฌ). Let โฑ (๐ต) := {๐ : ๐ต โ ๐๐ผ โ ๐น } denote the set of ๐น -points of operator ๐ต, and let โฑ0 denote the unbounded component of โฑ (๐ต). Since spec(๐ต) โ โ and ๐ต is a non-invertible ๐น -operator, it follows that ๐0 = 0 belongs to unbounded component of ๐น -points of operator ๐ต and hence ([GoKre, Theorem 3.6. ]) it is an isolated ๐น -point of spec(๐ต). Let ฮ
Invertibility of Certain Fredholm Operators
349
denote a circle โฃ๐โฃ = ๐, such that ๐ต โ ๐๐ผ (0 < โฃ๐โฃ โค ๐) is invertible. Since the spectrum of the operator ๐ต โ ๐๐ผ has a connected complement in โ, it follows that (๐ต โ ๐๐ผ)โ1 โ ฮฉ (0 < โฃ๐โฃ โค ๐), and since ฮฉ is a closed algebra it follows that the Riesz projection โซ 1 โ1 โ(๐ต, ฮ) := โ (๐ต โ ๐๐ผ) ๐๐ (2.2) 2๐๐ ฮ belongs to ฮฉ and is a non-zero ๏ฌnite-dimensional operator. By Proposition 2.4 this operator is nilpotent and this contradicts the condition of the theorem. โก Remark 2.10. The condition that ฮฉ โฉ ๐น is symmetric in Theorem 2.9 is essential. Namely, the algebra ฮฉ in Example 2.2 satis๏ฌes all conditions of Theorem 2.9 except the mentioned one, but the implication (i)=โ(ii) fails. We conclude this section by considering the following class of subalgebras without compact operators. Let {๐๐ } โ ๐ฟ(โฌ) denote a sequence of isometries which tends weakly to zero, and let ๐ถ (๐๐ , โฌ) โ ๐ฟ(โฌ) be the commutant of the set {๐๐ }. Denote โฃ๐ดโฃ = inf โฅ๐ด + ๐พโฅ. (2.3) ๐พโ๐ฆ(โฌ)
Theorem 2.11. Let ฮฉ be any unital subalgebra of ๐ถ (๐๐ , โฌ) . Then 1โ . 2โ . 3โ . 4โ . 5โ . 6โ . 7โ .
Equality โฅ๐ดโฅ = โฃ๐ดโฃ holds for all ๐ด โ ๐ถ (๐๐ , โฌ) . The algebra ฮฉ does not have non-zero compact operators. If ฮฉ โ ๐ถ (๐๐ , ๐ป) is a ๐ถ โ -algebra, then it is an ๐น ๐น -algebra. In general, algebra ฮฉ is not necessarily an ๐น ๐น -algebra. If the set of Fredholm operators ๐ด โ ฮฉ is symmetric, then ฮฉ is an ๐น ๐น -algebra. If ฮฉ is a ๐น -closed algebra, then it is an ๐น ๐น -algebra. Let ๐ be a subset of ๐ถ (๐๐ , โฌ) and let the set of all invertible operators ๐ด โ ๐ถ (๐๐ , โฌ) be dense in ๐ โฉ ๐น, then each Fredholm operator from ๐ is invertible. The following known lemma will be used in the proof of this theorem
Lemma 2.12. ([๐พ๐น, ๐ฟ๐๐๐๐ 1]) Let ๐ด๐ โ ๐บ๐ฟ(โฌ) and โฅ๐ด๐ โ ๐ดโฅ โ 0, where ๐ด is a non-invertible F-operator, then there exists a subsequence ๐ด๐๐ such that โ1 โ1 โฅ๐ดโ1 ๐ด๐๐ โ ๐, where ๐ is a ๏ฌnite-dimensional operator. ๐๐ โฅ Proof. Statement 1โ (of Theorem 2.11) follows from [K, Theorem 4.3]. Statement 2โ follows from Statement 1โ . Indeed, if ๐ โ ๐ฆ(ฮฉ) then โฅ๐ โฅ = โฃ๐ โฃ = inf ๐พโ๐ฆ(โฌ) โฃโฃ๐พ + ๐ โฃโฃ = 0. Statement 3โ follows from Statement 2โ and Theorem 1.1. To prove Statement 4โ consider in โ๐ , ๐ โ (1, โ) the sequence {๐ ๐ }, ๐ โ โ of isometries, where ๐ ๐ฅ = (0, ๐ฅ1 , ๐ฅ2 , . . . ). It can be easily checked that {๐ ๐ } tends weakly to zero. It is well known (and can be easily checked) that the commutant of operator ๐ (as well as of the set {๐ ๐ }) coincides with the algebra of all lower triangular Toeplitz matrices. This algebra satis๏ฌes the condition of the theorem, but it contains non-invertible Fredholm operators. For example, ๐ด = ๐. This proves
350
I. Feldman and N. Krupnik
Statement 4โ . Statement 5โ follows from Statement 2โ and Theorem 2.9. Statement 6โ follows from Statement 2โ and Proposition 2.6. Let us prove Statement 7โ . Assume that ๐ โ ๐ is a non-invertible Fredholm operator. Then there exists a sequence ๐๐ โ ๐ถ (๐๐ , โฌ) โฉ ๐บ๐ฟ(โฌ) such that โฅ๐ โ ๐๐ โฅ โ 0. The algebra ๐ถ (๐๐ , โฌ) (as a commutant) is inverse closed and hence ๐๐โ1 โ ๐ถ (๐๐ , โฌ) . By Lemma 2.12 there exists a subsequence ๐๐๐ such that โฅ๐๐โ1 โฅโ1 ๐๐โ1 tends to a non-zero ๏ฌnite๐ ๐ dimensional operator ๐พ. Since ๐ถ (๐๐ , โฌ) is closed, it follows that ๐พ โ ๐ถ (๐๐ , โฌ) . This contradicts Statement 2โ . โก
3. The closure of ๐ญ ๐ญ -subalgebras In this section ฮฉ denotes a (generally non-closed) ๐น ๐น -subalgebra of ๐ฟ (โฌ) . We study some properties of the algebra ๐ := clos(ฮฉ) โ ๐ฟ (โฌ) . These properties can be considered as โsome approximationsโ to the answer to a general Question 3.1. Let ฮฉ โ ๐ฟ (โฌ) be an ๐น ๐น -algebra. Is the closure ๐ = clos(ฮฉ) an ๐น ๐น -algebra, too ? Or, to a weaker Question 3.2. Let ฮฉ be an ๐น ๐น -algebra and let ๐ be inverse closed in ๐ฟ (โฌ). Is ๐ an ๐น ๐น -algebra, too? Questions 3.1 and 3.2 were formulated more than 15 years ago in Lecture Notes [KMF]. As far as we know, the answers to these questions are still unknown. We start with Theorem 3.3. Let ฮฉ โ ๐ฟ(โฌ) be an ๐น ๐น -algebra and ๐ := clos(ฮฉ). 1โ . If ๐ฆ(๐) โ= {0} (๐ฆ0 (๐) โ= {0}) , then it is a quasinilpotent ideal (a nil-ideal) in ๐. If, in particular, ฮฉ is a selfadjoint subalgebra of ๐ฟ(๐ป), then ๐ฆ(๐) = {0}. 2โ . If the algebra ๐ is ๐น -closed, then it is an ๐น ๐น -algebra. In addition, algebra ๐ is inverse closed in ๐ฟ(โฌ). 3โ . If the algebra ๐ is inverse closed and ๐ฆ0 (๐) = {0}, then ๐ is an ๐น ๐น -algebra2 4โ . The algebra ๐ does not contain non-invertible ๐น -operators ๐ด with isolated point ๐0 = 0 of the spectrum of operator ๐ด. 5โ . Let the algebra ๐ be a subalgebra of a commutant ๐ถ (๐๐ , โฌ) , de๏ฌned in Section 2, then ๐ is an ๐น ๐น -algebra. The following known statement will be used in the proof of this theorem. Lemma 3.4. Let ๐ด โ ๐ฟ(โฌ) be a non-invertible ๐น -operator and let there exist ๐ > 0 such that {๐ : 0 < โฃ๐โฃ โค ๐} โ ๐(๐ด). Then there exists a number ๐ฟ > 0 such that for each operator ๐ต โ ๐ฟ(โฌ) with โฅ๐ต โ ๐ดโฅ < ๐ฟ the set {๐ : 0 โค ๐ โค ๐} โฉ spec(๐ต) 2 In
fact, statement 3โ was proved in [KF, Theorem 2], but for completeness we give here a short proof of this statement.
Invertibility of Certain Fredholm Operators
351
consists of a ๏ฌnite number of points ๐๐ (โฃ๐๐ โฃ < ๐) such that ๐ตโ๐๐ ๐ผ are ๐น -operators and โ ๐๐๐ (๐ต) , (3.1) ๐0 (๐ด) = where ๐๐ (๐ต) denotes the algebraic multiplicity of the number ๐. This lemma follows from [GoKre, Theorem 4.3.] Now we are ready to prove Theorem 3.3. Proof. In order to prove the ๏ฌrst statement in 1โ it is enough to show that each compact operator from the algebra ๐ is quasinilpotent. Let ๐พ โ ๐ be a compact operator. If it is not quasinilpotent, then for some 0 โ= ๐1 โ spec(๐พ) the point ๐ = 0 is an isolated point of the spectrum of the non-invertible Fredholm operator ๐ต = ๐พ โ ๐1 ๐ผ โ ๐. By Lemma 3.4 there exists ๐ > 0 such that for each operator ๐ โ ๐ฟ(โฌ) with โฅ๐ต โ ๐ โฅ < ๐, there exists ๐0 โ โ such that the operator ๐ โ ๐0 ๐ผ is a non-invertible Fredholm operator. Taking such an operator ๐ from the dense algebra ฮฉ, we come to a contradiction. This proves the ๏ฌrst statement in 1โ . The second statement from 1โ is evident because ๐ถ โ -algebras (and, in particular, algebra ๐) do not contain quasinilpotent or nil-ideals. 2โ . Let the algebra ๐ be ๐น -closed and ๐ด โ ๐น โฉ ๐. Then there exists ๐ต โ ๐ such that ๐ต๐ด = ๐ผ + ๐พ, where ๐พ (โ ๐) is a compact operator. It follows from 1โ that ๐พ is quasinilpotent. Since spec(๐พ) is nowhere dense in โ, it follows that the spectrum spec(๐พ) in algebras ๐ฟ and ๐ coincide and hence (๐ผ + ๐พ)โ1 โ ๐. Thus (๐ผ + ๐พ)โ1 ๐ต๐ด = ๐ผ and the operator ๐ด is left invertible in ๐. Since also ind ๐ด = 0 (because ๐ด is a limit of a sequence of invertible operators) it follows that ๐ด is invertible. This proves the ๏ฌrst statement of 2โ . Moreover, this proves the second statement of 2โ because ๐ดโ1 = (๐ผ + ๐พ)โ1 ๐ต โ ๐. 3โ . Suppose that ๐ is not an ๐น ๐น -algebra. Then there exists a non-invertible ๐น operator ๐ด โ ๐ and we can take a sequence ๐ด๐ โ ฮฉ โฉ ๐น such that โฅ๐ด๐ โ ๐ดโฅ โ 0. Since ฮฉ is an ๐น ๐น -algebra, it follows that ๐ด๐ โ ฮฉ โฉ ๐บ๐ฟ. Algebra ๐ is inverse โ1 โ1 closed and ๐ดโ1 โ ๐. By Lemma 2.12 โฅ๐ดโ1 ๐ด๐ โ ๐ โ ๐พ0 (๐). Moreover, ๐ ๐ โฅ ๐ โ= 0 because โฅ๐โฅ = 1. This contradicts the conditions of Statement 3โ , and this statement is proved. 4โ . Suppose that ๐ contains a non-invertible ๐น -operator ๐ด. If ๐ = 0 is an isolated point of spec(๐ด), then, by Lemma 3.4, there exists ๐ฟ > 0 such that for each operator ๐ต โ ๐ฟ(โฌ) with โฅ๐ต โ ๐ดโฅ < ๐ฟ, there exists ๐0 such that ๐ต โ ๐0 ๐ผ is a non-invertible ๐น -operator. Like in the proof of 1โ we take ๐ต โ ฮฉ, and come to a contradiction, which proves Statement 4โ . 5โ . Algebra ๐ is a subset of ๐ถ (๐๐ , โฌ) and the set ฮฉ โฉ ๐บ๐ฟ (โ ๐ถ (๐๐ , โฌ) โฉ ๐บ๐ฟ) is dense in ๐ โฉ ๐น. Thus, ๐ถ (๐๐ , โฌ)โฉ ๐บ๐ฟ is dense in ๐ โฉ ๐น, and we are in the condition of Statement 7โ of Theorem 2.11 (where the set ๐ is substituted by the algebra ๐). This proves that ๐ is an ๐น ๐น -algebra. โก
352
I. Feldman and N. Krupnik
Remark 3.5. Example 2.1 and the example used in the proof of Proposition 2.4 (see equalities (2.1)) show that the algebra ๐ in Statement 1โ of Theorem 3.3 may contain ๏ฌnite-dimensional as well as in๏ฌnite-dimensional compact operators. Remark 3.6. The su๏ฌcient condition ๐ฆ0 (๐) = {0} in Statement 3โ of Theorem 3.3 is not necessarily for ๐ to be an ๐น ๐น -algebra. This can be con๏ฌrmed by Example 2.1. We conclude this section by considering certain classes of algebras which admit generalized Gelfand transforms. Let ๐๐ (โ) denote the algebra of all ๐ ร ๐ matrices with entries from โ. De๏ฌnition 3.7. We say that the algebra ๐ โ ๐ฟ(โฌ) admits a generalized Gelfand transform of order ๐ in โฌ if there exists a family of continuous homomorphisms ๐๐ : ๐ โ ๐๐ (โ), ๐ โ ๐ฎ, ๐ = ๐(๐ ) โค ๐
(3.2)
such that for each ๐ด โ ๐ the following implication holds: ๐ด โ ๐ โฉ ๐บ๐ฟ(โฌ) โโ det ๐๐ (๐ด) โ= 0 โ๐ โ ๐ฎ.
(3.3)
If this is the case, then we write ๐ โ ๐บ๐บ๐ (โฌ) and say that the system of homomorphisms {๐๐ } generates a GGT of order ๐ for the algebra ๐ in algebra ๐ฟ(โฌ). Example 3.8. Let ฮฉ โ ๐ฟ(โฌ) be a commutative subalgebra, then ๐ := clos(ฮฉ) โ ๐บ๐บ๐ (โฌ). Indeed, if ๐ is inverse closed, then the Gelfand transform on ๐ (which is responsible for the invertibility of the elements from ๐ in algebra ๐) generates also an ๐บ๐บ๐ (โฌ) for for ฮฉ. Assume that ๐ is not inverse closed. ( ๐ and, in particular, ) ห ห Denote by ๐ ๐ โ ๐ โ ๐ฟ(โฌ) some closed inverse closed commutative subalgebra of ๐ฟ(โฌ). For example, we can take the maximal commutative subalgebra of ๐ฟ(โฌ) which contains ๐. The Gelfand transform in algebra ๐ห generates a ๐บ๐บ๐ (โฌ) for the algebra ๐ห and, in particular, for the algebra ๐. Theorem 3.9. Let ๐ := clos(ฮฉ) โ ๐บ๐บ๐ (โฌ). Then 1โ . ๐ is not necessarily an ๐น ๐น -algebra. 2โ . But, if ฮฉ is an ๐น ๐น -algebra then ๐ is an ๐น ๐น -algebra, too. Proof. 1โ . Let ๐ โ ๐ฟ(โฌ) be an arbitrary non-invertible Fredholm operator and ๐ โ ๐ฟ(โฌ) an arbitrary closed commutative algebra which contains operator ๐. Then ๐ is not an ๐น ๐น -algebra, but (as was shown in Example 3.8) ๐ โ ๐บ๐บ๐ (โฌ). / ๐บ๐ฟ(โฌ), then there exists a homomorphism 2โ . Assume that ๐ด โ ๐น (โฌ)โฉ๐ but ๐ด โ ๐ โ {๐๐ } such that det ๐(๐ด) = 0. Let ๐ด๐ โ ๐น (โฌ) โฉ ฮฉ and โฅ๐ด โ ๐ด๐ โฅ โ 0. Then det ๐(๐ด๐ ) โ det ๐(๐ด) = 0, and hence there exist ๐๐ โ spec(๐(๐ด๐ )) such that ๐๐ โ 0. Denote ๐ต๐ := ๐ด๐ โ ๐๐ ๐ผ. It is clear that ๐ต๐ โ ๐ด, ๐ต๐ โ ฮฉ โฉ ๐น (โฌ) and ๐ต๐ โ / ๐บ๐ฟ(โฌ). This is a contradiction and the theorem is proved. โก
Invertibility of Certain Fredholm Operators
353
Theorem 3.10. Let ฮฉ โ ๐ฟ(โฌ) be a subalgebra with Amitsur-Levitski polynomial identity (1.1) of some order ๐ = 2๐, ๐ โ โ, and let ๐ := clos(ฮฉ) be inverse closed in ๐ฟ(โฌ). If ฮฉ is an ๐น ๐น -algebra, then ๐ is an ๐น ๐น -algebra, too. Proof. Since ๐ := clos(ฮฉ) is a Banach algebra with polynomial identity (1.1), it follows from [K, Theorem 21.1] that it admits a GGT for ๐ in ๐, i.e., there exists a set of homomorphisms ๐๐ : ๐ โ ๐๐ (โ), where ๐ = ๐(๐ ) โค ๐ such that for any operator ๐ด โ ๐ the following implication holds: ๐ด โ ๐บ๐ โโ det ๐๐ (๐ด) โ= 0
โ๐ โ โณ.
(3.4)
Here โณ denote the set of all maximal ideals of algebra ๐. Since ๐ is inverse closed it follows that ๐ด โ ๐บ๐ฟ(โฌ) โโ ๐ด โ ๐บ๐. Thus the set {๐๐ : ๐ โ โณ} generates a ๐บ๐บ๐ (โฌ) for algebra ๐. It remains to use Theorem 3.9. โก Theorem 3.11. Let ๐ be a subset of the center of an algebra ฮฉ โ ๐ฟ(โฌ) and let ฮฉ be a ๏ฌnite-dimensional module over ๐. If ฮฉ is an ๐น ๐น -algebra, then ๐ := clos(ฮฉ) is an ๐น ๐น -algebra, too. Proof. If the algebra ๐ has a dense subalgebra ฮฉ which is a ๏ฌnite-dimensional module over its center, then (see [GK, Corollary 1.2]) it admits a GGT for ๐ in ๐ฟ(โฌ), and we can use Theorem 3.9. โก Corollary 3.12. Let ฮฉ โ ๐ฟ(โฌ) be a smallest (generally non-closed) unital subalgebra generated by arbitrary two idempotent operators ๐, ๐
or by 2๐ idempotents ๐1 , ๐2 , . . . , ๐2๐โ1 , ๐
with some special relations.3 If ฮฉ is an ๐น ๐น -algebra, then ๐ = clos(ฮฉ) is an ๐น ๐น -algebra, too. Proof. If an algebra is generated by two idempotents or by 2๐ idempotents with relations (1โ4) from [BGKKRSS, Section 4], then it admits a GGT for ๐ in ๐ฟ(โฌ) (see [GK] for two idempotents and [BGKKRSS] for 2๐ idempotents). Thus, again we can use Theorem 3.9. โก
4. Some illustrative examples and open questions We start with a following illustrative example. Example 4.1. Let โฌ = ๐ฟ๐ (0, โ), ๐ โ (1, โ). Denote by {๐๐ } the sequence of 1 isometries de๏ฌned by equalities ๐๐ ๐ (๐ฅ) = ๐ ๐ ๐ (๐๐ฅ). It is not di๏ฌcult to check that ๐๐ โ 0 weakly. Denote by ๐๐ the commutant of the set {๐๐}. It follows from Theorem 2.11 that algebra ๐๐ does not contain non-zero compact operators, and โฃ๐ดโฃ = โฅ๐ดโฅ for all ๐ด โ ๐๐ . The algebra ๐๐ contains (for example): Singular integral operator ๐ and Ces` aro operators ๐ถ, ๐ถห de๏ฌned by equalities โซ โ โซ โซ โ 1 ๐ (๐ฆ)๐๐ฆ 1 ๐ฅ ๐ (๐ฆ)๐๐ฆ ห (๐ฅ) = ๐๐ (๐ฅ) = ; ๐ถ๐ (๐ฅ) = ๐ (๐ฆ)๐๐ฆ; ๐ถ๐ ; (4.1) ๐๐ 0 ๐ฆ โ ๐ฅ ๐ฅ 0 ๐ฆ ๐ฅ 3 See
the relations (1โ4) in [BGKKRSS, Section 4].
354
I. Feldman and N. Krupnik
integral operators
โซ
๐ ๐ (๐ฅ) = ๐๐ (๐ฅ) +
โ
0
๐(๐ฅ, ๐ฆ)๐ (๐ฆ)๐๐ฆ,
(๐ โ โ),
(4.2)
where ๐(๐ฅ, ๐ฆ) is measurable on [0, โ) ร [0, โ) and satis๏ฌes the following two conditions: โซ โ ๐(๐ฅ, ๐ฆ) ๐(๐ก๐ฅ, ๐ก๐ฆ) = , (๐ก โ (0, โ)) and ๐พ๐ (๐) := โฃ๐(๐ข, 1)โฃ๐ข1/๐โ1 ๐๐ข < โ. (4.3) ๐ก 0 and shift operators ๐ ๐ (๐ฅ) =
๐ โ
๐๐ ๐ (๐๐ ๐ฅ)
where ๐๐ โ โ, ๐๐ > 0.
(4.4)
๐=1
Consider a few subalgebras of algebra ๐๐ , generated by operators (4.1)โ(4.4). Alg1. Denote by ๐ฎ๐ (โ ๐๐ ) the unital Banach algebra generated by operator ๐. This algebra is symmetric for each ๐ โ (1, โ) (see, for example, [K, Theorem 13.6]). For operator ๐ one can take ๐ = [cos ๐๐ ๐ โ ๐ sin ๐๐ ๐ผ] [cos ๐๐ ๐ผ โ ๐ sin ๐๐ ๐]โ1 ,
(4.5)
where ๐๐ = 2๐/๐. If, in particular, ๐ = 2, then ๐ = ๐ โ = ๐. It follows from Statement 5โ of Theorem 2.11, that ๐ฎ๐ is an ๐น ๐น -algebra for all ๐ โ (1, โ). The algebra ๐ฎ๐ is wide enough. It contains, for example, the operators โซ โ โซ โ 1 ๐ (๐ฆ)๐๐ฆ ๐ฆ ๐ (๐ฆ)๐๐ฆ ๐๐ค ๐ (๐ฅ) = (โฃ๐คโฃ = 1) and ๐ ๐ (๐ฅ) = ln , ๐๐ 0 ๐ฆ + ๐ค๐ฅ ๐ฅ ๐ฆโ๐ฅ 0 see [K, Section 13]; the operators ๐
โ1
๐
โ1
1 ๐ (๐ฅ) = ๐๐
and
โซ ๐ (๐ฅ) =
โ 0
โโ
โซ 0
โ
๐ฆ ๐ (๐ฆ)๐๐ฆ ๐ฅ ๐ฆโ๐ฅ
๐ฅ ๐ (๐ฆ)๐๐ฆ ๐ฆ ๐ฆโ๐ฅ
(๐ โ (1, 2))
(๐ โ (2, โ)),
see [GK1, V. II, p. 98]. Alg2. By โณ๐ (๐ โ (1, โ)) we denote the set of all operators (4.2), which satis๏ฌes the conditions (4.3). It is not di๏ฌcult to check that โณ๐ is an algebra. Indeed, let ๐1 , ๐2 , ๐ correspond to integral operators ๐พ1 , ๐พ2 , ๐พ = ๐พ1 ๐พ2 , where โซ โ โซ โ ๐พ๐ ๐ (๐ฅ) := ๐๐ (๐ฅ, ๐ฆ)๐ (๐ฆ)๐๐ฆ (๐ = 1, 2); ๐(๐ฅ, ๐ฆ) = ๐1 (๐ฅ, ๐ง)๐2 (๐ง, ๐ฆ)๐๐ง. 0
0
Then ๐(๐๐ฅ, ๐๐ฆ) = โซ โ โซ 1 โ ( ๐ง ) (๐ง ) (๐ง ) 1 ๐1 (๐๐ฅ, ๐ง)๐2 (๐ง, ๐๐ฆ)๐๐ง = ๐1 ๐ฅ, ๐2 ,๐ฆ ๐ = ๐(๐ฅ, ๐ฆ). ๐ 0 ๐ ๐ ๐ ๐ 0
(4.6)
Invertibility of Certain Fredholm Operators Next we denote (for short) 1/๐ โ 1 = ๐ and check: $โซ โ $ โซ โ โซ โ $ $ ๐ ๐ $ โฃ๐(๐ข, 1)โฃ๐ข ๐๐ข = ๐ข ๐๐ข $ ๐1 (๐ข, ๐ง)๐2 (๐ง, 1)๐๐ง $$ 0 0 โซ0 โ โซ โ$ ( ๐ข )$$ ( ๐ข )๐ ( ๐ข ) $ โค โฃ๐2 (๐ง, 1)โฃ๐ง ๐ ๐๐ง ,1 $ ๐ $๐1 ๐ง ๐ง ๐ง 0 0 โค ๐พ๐ (๐2 )๐พ๐ (๐1 ).
355
(4.7)
Equalities (4.6) and (4.7) show that โณ๐ is an algebra. Theorem 4.2. The algebra clos (โณ๐ ) is an ๐น ๐น -algebra. Proof. It is known (see, for example, [K-G, Theorem 2]) that the spectrum of the operator (4.2) coincides with the curve โซ โ 1 ๐=๐+ ๐(๐๐ก , 1)๐( ๐ +๐๐ฅ)๐ก ๐๐ก (๐ฅ โ โ) (4.8) โโ
and for each point ๐ of this curve, the operator ๐ด โ ๐๐ผ is not an Fredholm operator. It follows from here that โณ๐ is an ๐น ๐น -algebra. Using Statement 5โ from Theorem 3.3 we obtain that clos (โณ๐ ) is an ๐น ๐น -algebra, too. โก ห Alg3. Denote by ๐๐ the unital Banach algebra generated by operators ๐ถ and ๐ถ. ห = ๐ถ + ๐ถ. ห It can be directly This is a commutative algebra because ๐ถ ๐ถห = ๐ถ๐ถ ห belong to the algebra โณ๐ . It follows from Theorem 4.2 that checked that ๐ถ and ๐ถ ๐๐ is an ๐น ๐น -algebra. Alg4. Denote by ๐ฒ๐ the unital Banach algebra generated by operators (4.4). It is well known that each Fredholm operator from ๐ฒ๐ is invertible. See, for example, the book [A], where the absence of non-invertible Fredholm operators is shown for more general classes of algebras. Since the algebra ๐ฒ๐ is commutative it follows from Proposition 1.3 (see also Example 3.8 & Theorem 3.9) that clos (๐ฒ๐ ) is an ๐น ๐น -algebra. Consider another illustrative example: Example 4.3. Let โฌ๐ := ๐ฟ๐ (ฮ), ๐ โ (1, โ), where ฮ is the unit circle, and let ๐ด โ ๐ฟ(โฌ๐ ) be a singular integral operator โซ ๐ (๐ )๐๐ ๐ด๐ (๐ก) = ๐(๐ก)๐ (๐ก) + ๐(๐ก) , ๐ก โ ฮ, (4.9) ฮ ๐ โ๐ก where ๐ and ๐ are piecewiseโconstant functions continuous on ฮ โ {โ1, 1}. Denote ๐ by ฮฉ๐ the set of operators ๐=1 ๐ด๐1 ๐ด๐2 โ
โ
โ
๐ด๐,โ(๐) , where ๐ โ โ and ๐ด๐๐ are the operators of the form (4.9). Theorem 4.4. Let ๐๐ := clos(ฮฉ๐ ). Then 1โ . Algebra ๐๐ is an ๐น ๐น -algebra if and only if ฮฉ๐ is. 2โ . Algebra ฮฉ๐ is an ๐น ๐น -algebra if and only if ๐ = 2.
356
I. Feldman and N. Krupnik
Proof. The algebra ฮฉ๐ is generated by the following two idempotents: analytical projection (โ ) โ ๐ ๐ ๐ ๐ก๐ = ๐๐ ๐ก๐ , ๐ก โ ฮ, ๐๐ โ โ ๐โฅ0
and the operator ๐
of multiplication by the characteristic function of the upper semi-circle. If ฮฉ๐ is an ๐น ๐น -algebra then, by Corollary 3.12, the algebra ๐๐ := clos(ฮฉ๐ ) is an ๐น ๐น -algebra too. This proves Statement 1โ . Consider an operator ๐ต = ๐๐ + ๐ (โ ฮฉ๐ ), where ๐(๐ก) takes only two values: ยฑ1 and ๐ = ๐ผ โ๐. It follows from [GK1, Ch. 9, Theorem 3.1] that ๐ต is a Fredholm operator if and only if ๐ โ= 2. If ๐ > 2 then ind ๐ด = 1, If ๐ < 2 then ind ๐ด = โ1. Thus the only candidate to be an ๐น ๐น -algebra is the algebra ฮฉ2 . And here is one of the ways to con๏ฌrm that algebra ฮฉ2 is an ๐น ๐น -algebra. Consider the following sequence of operators ( ) โ 2 ๐ (๐ + 1)๐ก + ๐ โ 1 ๐๐ ๐ (๐ก) = ๐ . (4.10) ๐ + 1 + ๐ก(๐ โ 1) ๐ + 1 + ๐ก(๐ โ 1) It is not di๏ฌcult to check that {๐๐ } (โ ๐ฟ(โฌ2 )) is a sequence of isometries which tends weakly to zero and that selfadjoint operators ๐, ๐
commute with all ๐๐ (we omit the details). The algebra ๐2 is a ๐ถ โ -subalgebra of the commutant ๐ห of the set {๐๐ }, and it follows from Statement 3โ of Theorem 2.11 that ๐2 is an ๐น ๐น -algebra. โก Remark 4.5. The algebra ๐ in Theorem 3.10 satis๏ฌes the following two conditions. It is inverse closed and with Amitsur-Levitski polynomial identity (1.1). โ If we omit both of these conditions, then we come to the open Question 3.1. โ If we omit only the second condition, then we come to the open Question 3.2. โ Finally, if we omit only the ๏ฌrst condition, then we come to Question 4.6. Let ฮฉ โ ๐ฟ(โฌ) be an ๐น ๐น -subalgebra with Amitsur-Levitski polynomial identity (1.1) of some order 2๐, ๐ โ โ. Is ๐ := clos(ฮฉ) an ๐น ๐น -algebra, too? If ๐ = 1 then the Amitsur-Levitski identity ๐ฅ1 ๐ฅ2 โ ๐ฅ2 ๐ฅ1 = 0 means that ฮฉ is a commutative algebra and it follows from Proposition 1.3 that the answer to Question 4.6 is positive. As far as we know, Question 4.6 for ๐ > 1 is still open.
Invertibility of Certain Fredholm Operators
357
References [A]
A. Antonevich, Linear Functional Equations. Operator Approach, OT. 83 Birkhยจ auser Verlag, 1996. [BGKKRSS] A. Bยจ ottcher, I. Gohberg, Yu. Karlovich, N. Krupnik, S. Roch, B. Silbermann, I. Sptkovsky, Banach algebras generated by idempotents and applications, Operator Theory., V. 90(1996), 19โ54. [CL] L.A. Coburn and A. Lebov, Algebraic Theory of Fredholm Operators, J. Math. Mech., 1996, 15, 577โ584. [GoKre] I. Gohberg and M.G. Krein, The basic propositions on defect numbers, root numbers and indices of linear operators, Uspehi Mat. Nauk 12, no. 2(74) (1957), 43โ118 (Russian). English transl. Amer. Math, Soc. Transl. (2)13 (1960), 185โ 264. [GK] I. Gohberg and N. Krupnik, Extension theorems for Fredholm and invertibility symbols, IEOT 16 (1993), 515โ529. [GK1] I. Gohberg and N. Krupnik, One-Dimensional Linear Singular Integral Equation, Vol. IโII, Birkhยจ auser Verlag, Basel โ Boston, 1992. [K] N. Krupnik, Banach Algebras with Symbols and Singular Integral Operators, Birkhยจ auser Verlag Basel โ Boston, 1987. [KF] N. Krupnik and I. Feldman, On the invertibility of certain Fredholm operators, Izv. Akad. Nauk MSSR, ser. ๏ฌz. i mat. nauk 2, (1982), 8โ14. (Russian) [KMF] N. Krupnik, A. Markus, I. Feldman, Operator algebras in which all Fredholm operators are invertible, Lecture Notes in Mathematics, Linear and Complex Analysis, 1533 (1994), 124โ125. [K-G] M. Kozhokar-Gonchar, The spectrum of Ces` aro operators, Mat. Issled. 7, no. 4(26) (1972), 94โ103. (Russian) [MF] A. Markus and I. Feldman, On the algebras generated by operators with one-side inverses, Research of Dif. Equat., Shtiintsa, Kishinev, (1983) 42โ46. (Russian) Israel Feldman Department of Mathematics Bar-Ilan University Ramat-Gan 52900, Israel e-mail:
[email protected] Nahum Krupnik 208โ7460 Bathurst Str. Vaughan, L4J 7K9 Ontario, Canada e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 359โ376 c 2012 Springer Basel AG โ
Bernstein Widths and Super Strictly Singular Inclusions F.L. Hernยดandez, Y. Raynaud and E.M. Semenov To the memory of Professor Israel Gohberg
Abstract. The super strict singularity of inclusions between rearrangement invariant function spaces on [0, 1] is studied. Estimates of the Bernstein widths ๐พ๐ of the inclusions ๐ฟโ โ ๐ธ are given. It is showed that if the inclusion ๐ธ โ ๐น is strong and the order continuous part of exp ๐ฟ2 is not included in ๐ธ then the inclusion ๐ธ โ ๐น is super strictly singular. Applications to the classes of Lorentz and Orlicz spaces are given. Mathematics Subject Classi๏ฌcation (2000). 41A46, 46E30. Keywords. Strictly singular operator, rearrangement invariant spaces, Rademacher system, widths.
0. Introduction A linear operator ๐ด between two Banach spaces ๐ธ and ๐น is called strictly singular (SS in short) if ๐ด fails to be an isomorphism on any in๏ฌnite-dimensional subspace of ๐ธ. This concept was introduced by Tosio Kato in [K]. A stronger notion is the following. An operator ๐ด from ๐ธ to ๐น is called super strictly singular (SSS in short) if the sequence of Bernstein widths ๐๐ (๐ด) tends to 0 when ๐ โ โ, where ๐๐ (๐ด) =
sup
inf
๐โ๐ธ,dim ๐=๐ ๐ฅโ๐,โฅ๐ฅโฅ=1
โฅ๐ด๐ฅโฅ๐น .
This notion was introduced ๏ฌrstly by B. Mityagin and A. Pelczynski in [MP]. About widths we refer to [PI]. It is clear that ๐พ โ ๐๐๐ โ ๐๐, where ๐พ denotes the class of compact operators. Properties of SSS operators have been given in [M], [P], [CCT], [FHR], [SSTT] and [S]. This operator ideal has been also named in the literature as ๏ฌnite strictly singular operators ([SSTT], [S]). In the context The authors gratefully acknowledge the support of MTM-grant 2008โ02652, RFBR-grant 08โ01โ 00226a and Complutense University grant.
360
F.L. Hernยดandez, Y. Raynaud and E.M. Semenov
of Banach lattices a weaker notion is the following one ([HR]): An operator ๐ด from a Banach lattices ๐ธ to a Banach space ๐น is said to be disjointly strictly singular (DSS in short) if there is no disjoint sequence on non-null vectors (๐ฅ๐ ) in ๐ธ s.t. the restriction of ๐ด to the subspace [(๐ฅ๐ )] spanned by the vectors (๐ฅ๐ ) is an isomorphism. Clearly SS โ DSS. In general these operator classes ๐พ โ SSS โ SS โ DSS are di๏ฌerent. However any SS operator in a ๐๐ -space (1 โฉฝ ๐ < โ) is compact. This was proved by I. Gohberg, A. Markus and I. Feldman in [GMF] (for ๐ = 2 it was done before by J. Calkin [C]). It easily follows from results of Grothendieck that on probability measure spaces the canonical inclusions ๐ฟโ โ ๐ฟ๐ are SS for any ๐ < โ. More generally, the inclusion ๐ฟโ โ ๐ธ is always SS for any rearrangement invariant space ๐ธ โ= ๐ฟโ on [0, 1], (S. Novikov [N]). In fact it turns out that this inclusion is SSS ([FHR]). This paper is devoted to study SSS inclusions between arbitrary rearrangement invariant function spaces. First in Section 1 we generalize Grothendieckโs result by estimating the Bernstein widths of the inclusions ๐ฟโ โ ๐ธ for r.i. function spaces ๐ธ on [0, 1] (this leads to a new proof of the fact that these inclusions are always SSS). Afterwards we study the SSS property for general inclusions ๐ธ โ ๐น of r.i. spaces on [0, 1]. The main results are given in Section 3 (see Theorem 17 and its Corollaries). The notion of strong inclusion studied in Section 2 plays an important role for that. If ๐ธ, ๐น are r.i. and ๐ธ โ ๐น , then this inclusion is called strong if the topology of the norm of ๐น and that of convergence in measure coincide on the unit ball of ๐ธ. Theorem 17 states that if the inclusion ๐ธ โ ๐น is strong and moreover the order-continuous part ๐บ of the Orlicz space exp๐ฟ2 is not included in ๐ธ then the inclusion ๐ธ โ ๐น is SSS. Recall that a Banach space ๐ธ of measurable functions on [0, 1] is said to be rearrangement invariant (r.i.) if the following conditions hold: 1) if ๐ฆ โ ๐ธ and โฃ๐ฅ(๐ก)โฃ โฉฝ โฃ๐ฆ(๐ก)โฃ a.e., then ๐ฅ โ ๐ธ and โฅ๐ฅโฅ๐ธ โฉฝ โฅ๐ฆโฅ๐ธ ; 2) if ๐ฆ โ ๐ธ and ๐ฅ and ๐ฆ are equimeasurable, then ๐ฅ โ ๐ธ and โฅ๐ฅโฅ๐ธ = โฅ๐ฆโฅ๐ธ . As usual (cf. [LT2] and [KPS]) we shall assume that r.i. spaces ๐ธ are separable or maximal (i.e., ๐ธ = ๐ธ โฒโฒ ), where ๐ธ โฒโฒ denotes the space of measurable functions ๐ฅ for which โฅ๐ฅโฅ๐ธ โฒโฒ = lim โฅ min(โฃ๐ฅโฃ, ๐)โฅ๐ธ < โ. ๐โโ
The space ๐ธ โฒ endowed with the norm โฅ๐ฅโฅ๐ธ โฒ = sup
โฅ๐ฆโฅ๐ธ โฉฝ1
โซ 0
1
๐ฅ(๐ก)๐ฆ(๐ก) ๐๐ก
is an r.i. space. Denote by รฆ๐ the characteristic function of a measurable set ๐. The function ๐๐ธ (๐ ) = โฅรฆ๐ โฅ๐ธ , where ๐ โ [0, 1] is any measurable set of measure ๐ , is named the fundamental function of the r.i. space ๐ธ. We will assume, w.l.o.g., that ๐๐ธ is concave and ๐๐ธ (1) = 1. In this case ๐ฟโ โ ๐ธ โ ๐ฟ1 , and โฅ๐ฅโฅ๐ฟ1 โฉฝ โฅ๐ฅโฅ๐ธ โฉฝ ๐ก โฅ๐ฅโฅ๐ฟโ for any ๐ฅ โ ๐ฟโ . It is known that ๐๐ธ โฒ (๐ก) = . Given ๐ฅ, ๐ฆ โ ๐ฟ1 , we ๐๐ธ (๐ก)
Bernstein Widths and Super Strictly Singular Inclusions
361
โซ๐ โซ๐ shall write ๐ฅ โบ ๐ฆ if 0 ๐ฅโ (๐ก) ๐๐ก โฉฝ 0 ๐ฆ โ (๐ก) ๐๐ก for every ๐ โ [0, 1]. It is well known that โฅ๐ฅโฅ๐ธ โฉฝ โฅ๐ฆโฅ๐ธ provided ๐ฅ โบ ๐ฆ ([LT2], 2.a.8). Important examples of r.i. spaces are the Orlicz, Lorentz and Marcinkiewicz spaces. If M is a positive convex function on [0, โ) with ๐ (0) = 0, the Orlicz space ๐ฟ๐ consists of all measurable functions ๐ฅ(๐ก) on [0, 1] for which โง โซ โซ1 โจ โฌ โฃ๐ฅ(๐ก)โฃ โฅ๐ฅโฅ๐ฟ๐ = inf ๐ > 0 : ๐( )๐๐ก โฉฝ 1 < โ. โฉ โญ ๐ 0
๐ข๐
If ๐๐ (๐ข) = ๐ โ 1, 0 < ๐ < โ, then ๐๐ (๐ข) is convex for ๐ โฉพ 1 and is convex up to equivalence for ๐ < 1. The space ๐ฟ๐๐ is denoted by exp ๐ฟ๐ . The Orlicz space ๐ฟ๐2 is not separable and its separable part (i.e., the closure of ๐ฟโ in ๐ฟ๐2 ) is denoted by ๐บ. The space ๐บ plays an important role in the theory of r.i. spaces. Let us denote by ฮฉ the set of all increasing concave functions ๐(๐ก) on [0, 1] with ๐(0) = 0 and ๐(1) = 1. The Lorentz space ฮ(๐) and ๐ฟ๐,๐ consist of all measurable functions on [0, 1] s.t. โซ 1 โฅ๐ฅโฅฮ(๐) = ๐ฅโ (๐ก) ๐๐(๐ก) < โ, 0
resp. โฅ๐ฅโฅ๐ฟ๐,๐
โง ( โซ 1( )๐ ๐๐ก ) 1๐ ๏ฃด ๏ฃด โ 1/๐ โจ ๐ ๐ฅ (๐ก)๐ก , 1 โฉฝ ๐ < โ, ๐ 0 ๐ก = ๏ฃด โ 1/๐ ๏ฃด ๐=โ โฉ sup ๐ฅ (๐ก)๐ก , 0<๐กโฉฝ1
โ
for 1 < ๐ < โ, where ๐ฅ denotes the decreasing rearrangement of โฃ๐ฅ(๐ก)โฃ. The Marcinkiewicz space ๐ (๐) consists of all measurable functions on [0, 1] s. t. โซ ๐ 1 โฅ๐ฅโฅ๐(๐) = sup ๐ฅโ (๐ก) ๐๐ก < โ. 0<๐ โฉฝ1 ๐(๐ ) 0 It is well known that in a r.i. space ๐ธ the Rademacher system ๐๐ (๐ก) = sign sin 2๐ ๐๐ก, ๐ โ โ generates a subspace isomorphic to ๐2 i๏ฌ ๐บ โ ๐ธ (cf. [LT2], 2.b.4). Hence for a couple of r.i. spaces ๐ธ and ๐น with ๐ธ โ ๐น , if the inclusion ๐ธ โ ๐น is SS then ๐บ โโ ๐ธ. In the extreme case where ๐น = ๐ฟ1 it holds that ๐ธ โ ๐ฟ1 is SS i๏ฌ ๐บ โโ ๐ธ ( [HNS]). Let 1 โฉฝ ๐ < โ. An r.i. space ๐ธ is called ๐-concave if there exists a constant ๐ถ > 0 s. t. 1( ) ๐1 1 ( ๐ ) 1๐ 1 โ 1 โ 1 ๐ 1 1 1โฉพ๐ถ โฃ๐ฅ๐ โฃ๐ โฅ๐ฅ๐ โฅ๐ 1 1 1 ๐=1 1 ๐=1 for any ๐ โ โ and ๐ฅ1 , ๐ฅ2 , . . . , ๐ฅ๐ โ ๐ธ ([LT2], 1.d.3). Given functionals ๐, ๐, we 1 shall write ๐ โ ๐ if ๐ (๐ฅ) โฉฝ ๐(๐ฅ) โฉฝ ๐ถ๐ (๐ฅ) for some constant ๐ถ > 0 and every ๐ฅ ๐ถ from domain of de๏ฌnition. Some results of this article were announced in [S] and [RSH].
362
F.L. Hernยดandez, Y. Raynaud and E.M. Semenov
1. Inclusion of ๐ณโ into r.i. spaces: Generalization of Grothendieckโs result Let ๐ธ, ๐น be a pair of r.i. space and ๐ธ โ ๐น . Given ๐ โ โ, denote ๐พ๐ (๐ธ, ๐น ) =
sup
inf
๐โ๐ธ,dim ๐=๐ โฅ๐ฅโฅ๐ธ =1, ๐ฅโ๐
โฅ๐ฅโฅ๐น .
Clearly ๐พ๐ (๐ธ, ๐น ) are the Bernstein widths of the inclusion operator ๐ผ : ๐ธ โ ๐น . The next statement is simple (there are many similar results). Lemma 1. Let ๐ โ โ and ๐ be an ๐-dimensional subspace of ๐ฟโ . There exists an element ๐ง โ ๐ s. t. โฅ๐งโฅ๐ฟโ = 1 and ๐ง 2 โบ รฆ(0, 1 ) . ๐
Proof. Using the Gram-Schmidt method of orthogonalization we can ๏ฌnd an orthonormal system ๐ฅ1 , ๐ฅ2 , . . . , ๐ฅ๐ in ๐. Then 1 ๐ 1 1โ 1 1 1 ๐๐ ๐ฅ๐ 1 1 $ ๐ $ 1 1 $โ $ ๐=1 $ $ ๐ฟโ 1 sup 1 = sup ๐ ๐ฅ (๐ก) $ $ ๐ ๐ 1 ๐ 1 $ $ ๐ โ {๐๐ }โ=0 1โ 1 2 =1,0โฉฝ๐กโฉฝ1 ๐=1 ๐ ๐ ๐๐ ๐ฅ๐ 1 1 ๐=1 1 1 ๐=1 ๐ฟ2 ( ๐ )1/2 1 )1/2 1 1 (โ 1 ๐ โ 1 1 โ 2 2 1 1 = ๐. = sup ๐ฅ๐ (๐ก) โฉพ1 ๐ฅ๐ 1 0โฉฝ๐กโฉฝ1 1 ๐=1 1 ๐=1 ๐ฟ2
Hence ๐ contains an element ๐ง s. t. โฅ๐งโฅ๐ฟโ = 1 and โฅ๐งโฅ๐ฟ2 and
โซ 0
1
๐ง 2 (๐ก) ๐๐ก โฉฝ
1 โฉฝ โ . Then โฅ๐ง 2 โฅ๐ฟโ = 1 ๐
1 . ๐
It is easy to see that these two estimates imply ๐ง 2 โบ รฆ(0, ๐1 ) .
โก
Theorem 2. Let ๐ธ be an r.i. space and ๐ธ โ= ๐ฟโ . The inclusion ๐ฟโ โ ๐ธ is SSS and ๐๐ธ (1/๐) โฉฝ ๐พ๐ (๐ฟโ , ๐ธ) โฉฝ (๐๐ธ (1/๐))1/2 (1) for any ๐ โ โ. Proof. If ๐ฅ๐ (๐ก) = รฆ( ๐โ1 , ๐ ) (๐ก), 1 โฉฝ ๐ โฉฝ ๐ and ๐ = span{๐ฅ๐ , 1 โฉฝ ๐ โฉฝ ๐}, then ๐
๐
inf
๐ฅโ๐,โฅ๐ฅโฅ๐ฟโ =1
โฅ๐ฅโฅ๐ธ = โฅรฆ(0, ๐1 ) โฅ๐ธ = ๐๐ธ (1/๐) ,
and we get the left inequality (1). By Lemma 1 any ๐-dimensional subspace ๐ โ ๐ฟโ contains an element ๐ง โ ๐ s. t. โฅ๐งโฅ๐ฟโ = 1 and ๐ง 2 โบ รฆ(0, ๐1 ) . Applying ([LT2], 2.a.8) it follows that โฅ๐ง 2 โฅ๐ธ โฉฝ โฅรฆ(0, ๐1 ) โฅ๐ธ = ๐๐ธ (1/๐).
(2)
Bernstein Widths and Super Strictly Singular Inclusions
363
1/2
The space ๐ธ(2), endowed with the norm โฅ๐ฅโฅ๐ธ(2) = โฅ๐ฅ2 โฅ๐ธ , is included in ๐ธ and โฅ๐ฅโฅ๐ธ โฉฝ โฅ๐ฅโฅ๐ธ(2) for any ๐ฅ โ ๐ธ(2) ([LT2], 1.d). The space ๐ธ(2) is called the 2convexication of ๐ธ. Hence โฅ๐งโฅ๐ธ โฉฝ โฅ๐งโฅ๐ธ(2) โฉฝ (๐๐ธ (1/๐)) /2 1
and
โฅ๐ฅโฅ๐ธ โฉฝ โฅ๐งโฅ๐ธ โฉฝ (๐๐ธ (1/๐)) /2 . 1
inf
๐งโ๐,โฅ๐ฅโฅ๐ฟโ =1
Hence ๐พ๐ (๐ฟโ , ๐ธ) โฉฝ (๐๐ธ (1/๐)) /2 , which is the right inequality in (1). Now, since lim ๐๐ธ (๐ก) = 0 for any r.i. space ๐ธ โ= ๐ฟโ , we have lim ๐พ๐ (๐ฟโ , ๐ธ) = 0. โก 1
๐โโ
๐กโ0
The SSS property of the inclusion of ๐ฟโ into any r.i. space ๐ธ โ= ๐ฟโ is also proved by another method in [FHR, Prop. 5.7]. Theorem 2 may be strengthened in the class of 2-convex spaces: Theorem 3. Let ๐ธ be a 2-convex r.i. space. Then ๐๐ธ (1/๐) โฉฝ ๐พ๐ (๐ฟโ , ๐ธ) โฉฝ ๐ถ ๐๐ธ (1/๐)
(3)
for some ๐ถ > 0 and any ๐ โ โ. Proof. It is well known that for any 2-convex r.i. space ๐ธ there exists an r.i. space ๐น s. t. ๐ธ and ๐น (2) coincide up to equivalence of norms ([LT2], 1.d). Therefore it is su๏ฌcient to prove our statement for ๐ธ = ๐น (2). Let ๐ be an ๐-dimensional subspace of ๐ฟโ . By Lemma 1 there exists ๐ง โ ๐ s. t. โฅ๐งโฅ๐ฟโ = 1 and 1/2
1/2
1/2
โฅ๐งโฅ๐ธ = โฅ๐ง 2 โฅ๐น โฉฝ โฅรฆ(0, ๐1 ) โฅ๐น = โฅรฆ2(0, 1 ) โฅ๐น = โฅรฆ(0, ๐1 ) โฅ๐ธ = ๐๐ธ (1/๐). ๐
(4)
Now, if โฅ โ
โฅ๐ธ โฉฝ โฅ โ
โฅ๐น (2) โฉฝ ๐ถโฅ โ
โฅ๐ธ then the constant ๐ถ in (3) coincides with the constant in this inequality. The left part of (3) was proved in Theorem 2. โก Note that if the norms ๐ธ and ๐น (2) coincide, then (4) shows that the constant ๐ถ in (3) equals 1, i.e., ๐พ๐ (๐ฟโ , ๐ธ) = ๐๐ธ (1/๐) for any ๐ โ โ. This condition is satis๏ฌed for ๐ธ = ๐ฟ๐ , for 2 โฉฝ ๐ < โ. Hence ๐พ๐ (๐ฟโ , ๐ฟ๐ ) = ๐๐ฟ๐ (1/๐) = (1/๐) /๐ 1
(5)
for any ๐ โ โ and ๐ โ [2, โ). This statement was proved in [PS]. If 1 โฉฝ ๐ < 2, then 1 ๐พ๐ (๐ฟโ , ๐ฟ๐ ) โฉฝ ๐พ๐ (๐ฟโ , ๐ฟ2 ) = โ . ๐ Let ๐ = span{๐๐ , 1 โฉฝ ๐ โฉฝ ๐} where ๐๐ (๐ก) = sign sin 2๐ ๐๐ก are the Rademacher system. By Khintchine inequality ([LT1], 2.b.3) we have 1 ๐ 1 ( ๐ )1/2 1โ 1 โ 1 1 1 1 2 ๐พ๐ (๐ฟโ , ๐ฟ1 ) โฉพ โ inf ๐๐ ๐๐ 1 โฉพ โ ๐ inf ๐๐ = โ . 1 1 2 โ โฃ๐ โฃ=1 2๐ โฅ ๐๐=1 ๐๐ ๐๐ โฅ =1 1 ๐ฟโ
๐=1
๐ฟ1
๐=1
๐
๐=1
364
F.L. Hernยดandez, Y. Raynaud and E.M. Semenov
So, for any ๐ โ [1, 2) and ๐ โ โ we have 1 1 โ โฉฝ ๐พ๐ (๐ฟโ , ๐ฟ๐ ) โฉฝ โ . (6) ๐ 2๐ Inequalities (5) and (6) show that estimates (1) are precise. Now we want to generalize the obtained results for Lorentz ๐ฟ๐, ๐ -spaces. Proposition 4. Let ๐ 1. ๐พ๐ (๐ฟโ , ๐ฟ๐,๐ ) โ 2. ๐พ๐ (๐ฟโ , ๐ฟ๐,๐ ) โ 3. ๐พ๐ (๐ฟโ , ๐ฟ2,๐ ) =
be an integer. 1 (1/๐) /๐ if 2 < ๐ < โ. 1 1 ( /๐) /2 if 1 < ๐ < 2. 1/2 (1/๐) if 2 โฉฝ ๐ โฉฝ โ.
Proof. 1. Since ๐๐ฟ๐,๐ (๐ก) = ๐ก1/๐ then, by Theorem 2, (1/๐)1/๐ โฉฝ ๐พ๐ (๐ฟโ , ๐ฟ๐,๐ ). To obtain an upper estimate we use Lemma 1 once more. Let ๐ be an ๐-dimensional 1 subspace of ๐ฟโ . There exists ๐ง โ ๐ s. t. โฅ๐งโฅ๐ฟโ = 1 and โฅ๐งโฅ๐ฟ2 โฉฝ โ . By ([BL], ๐ 5.3.1, 3.5.3) ( 1 ) ๐2 2 1 1โ 2 โฅ๐งโฅ๐ฟ๐,๐ โฉฝ ๐ถ๐ โฅ๐งโฅ๐ฟโ๐ โฅ๐งโฅ๐ฟ๐ 2 โฉฝ ๐ถ๐ ๐โ 2 = ๐ถ๐ ๐โ ๐ for some constant ๐ถ๐ > 0 and any ๐ โ โ. Hence ๐พ๐ (๐ฟโ , ๐ฟ๐,๐ ) โฉฝ ๐ถ๐ ๐โ /๐ . 2. Since ๐ฟ2 โ ๐ฟ๐,๐ โ ๐ฟ1 for any ๐ โ (1, 2) ([LT2], 2.b.8) then the needed estimate follows from (6). 1 3. The lower estimate ๐โ /2 โฉฝ ๐พ๐ (๐ฟโ , ๐ฟ2,๐ ) follows from Theorem 2 and equality ๐๐ฟ2,๐ (๐ก) = ๐ก1/2 . Since ๐ฟ2 โ ๐ฟ2,๐ and โฅ๐ฅโฅ๐ฟ2,๐ โฉฝ โฅ๐ฅโฅ๐ฟ2 ([LT2], 2.b.9) we have, by (5), that ๐พ๐ (๐ฟโ , ๐ฟ2,๐ ) โฉฝ ๐พ๐ (๐ฟโ , ๐ฟ2 ) = ๐โ1/2 . Therefore ๐พ๐ (๐ฟโ , ๐ฟ2,๐ ) = ๐โ1/2 for any ๐ โ [2, โ] and ๐ โ โ. โก 1
To ๏ฌnd ๐พ๐ (๐ฟโ , ๐ฟ2,๐ ) for ๐ โ [1, 2) is a more delicate problem. It has been partially solved. Lemma 5. Given ๐ โ โ and 1 โฉฝ ๐ โฉฝ ๐. Then ๐ โ ๐ฅ โ๐ max ๐ โ ๐ โฃ๐ฅ๐ โฃโฉฝ1, ๐ฅ2 โฉฝ๐ ๐=1 ๐=1
๐
(
) 1 is obtained on the sequence ๐ฅ๐ = min 1, ๐๐ โ /2 where ๐ is de๏ฌned by the equation ๐ โ ( 2 ) min 1, ๐ /๐ = ๐. ๐=1
๐ ๐ โ 1 = ๐, then ๐ ln โฉฝ 2๐ ln /2 ๐2 ๐ The proof of Lemmas 5 and 6 is simple (so it is omitted).
Lemma 6. Let 0 < ๐, ๐ < 1. If ๐2 ln Theorem 7. Given ๐ โ โ, ๐
โ1/2
( โฉฝ ๐พ๐ (๐ฟโ , ๐ฟ2,1 ) โฉฝ
2 + ln ๐ ๐
)1/2 .
โ๐ . ๐
Bernstein Widths and Super Strictly Singular Inclusions
365
Proof. The left inequality follows from Theorem 2. By Lemma 1 any ๐-dimensional โ subspace ๐ of ๐ฟโ contains an element ๐ง โ ๐ s. t. โฅ๐งโฅ๐ฟโ = 1 and โฅ๐งโฅ๐ฟ2 โฉฝ 1/ ๐. Therefore ๐พ๐ (๐ฟโ , ๐ฟ2,1 ) โฉฝ sup โฅ๐ฅโฅ๐ฟ2,1 โ
โฅ๐ฅโฅ๐ฟโ โฉฝ1,โฅ๐ฅโฅ๐ฟ2 โฉฝ1/
and
๐
๐
๐พ๐ (๐ฟโ , ๐ฟ2,1 ) โฉฝ sup ๐
โ1 1 โ โ ๐ฅ๐ ๐ 2 . ๐
max ๐
โฃ๐ฅ๐ โฃโฉฝ1,
โ
๐=1
๐ฅ2๐ โฉฝ ๐ ๐
๐=1
Now, applying Lemma 5 we get โซ1 ( ) โ1 โ1 1 1 ๐ ๐พ๐ (๐ฟโ , ๐ฟ2,1 ) โฉฝ min 1, ๐๐ก 2 ๐ก 2 ๐๐ก = ๐ + ๐ ln = ๐ ln 2 ๐ ๐ 0
where ๐ is such that
โซ1 ( ( ))2 โ1 ๐ 1 min 1, ๐๐ก 2 ๐๐ก = ๐2 ln 2 = . ๐ ๐ 0
By Lemma 6 we deduce ๐พ๐ (๐ฟโ , ๐ฟ2,1 ) โฉฝ
โ
โ 1 1 2 โ ln 2 ๐ ๐ = ๐
( ( ) ) 12 ( )1 2 1 + 12 ln ๐ 2 + ln ๐ 2 = . ๐ ๐
Now for 1 โฉฝ ๐ โฉฝ 2, using that the function ๐ โ ln โฅ๐ฅโฅ๐ฟ we get the estimate ๐พ๐ (๐ฟโ , ๐ฟ2,๐ ) โฉฝ for every ๐ โ ๐ . The lower estimate ๐ Thus we have
โ1 2
(2 + ln ๐)
2, 1 ๐
โก
is convex on [0, 1]
1/๐ โ1/2
1
๐2 โฉฝ ๐พ๐ (๐ฟโ , ๐ฟ2,๐ ) follows from Theorem 2.
Corollary 8. If 1 โฉฝ ๐ โฉฝ 2 and ๐ โ ๐ , then 1 1
๐2
1
โฉฝ ๐พ๐ (๐ฟโ , ๐ฟ2,๐ ) โฉฝ
(2 + ln ๐) ๐ 1
๐2
โ 12
.
2. Strong inclusions Let ๐ธ, ๐น be a pair of r.i. spaces and ๐ธ โ ๐น . The inclusion ๐ธ โ ๐น is called strong if lim sup โฅ๐ฅโฅ๐น = 0. ๐โ0 โฅ๐ฅโฅ โฉฝ1 ,mes(supp ๐ฅ)โฉฝ๐ ๐ธ
Clearly any strong inclusion is DSS. If an inclusion ๐ธ โ ๐น is strong, then ๐๐น (๐ก) lim = 0. S.V. Astashkin ([A]) proved that the inverse statement is false. ๐กโ0 ๐๐ธ (๐ก)
366
F.L. Hernยดandez, Y. Raynaud and E.M. Semenov
More precisely, he constructed a pair of r.i. spaces ๐ธ and ๐น with ๐ธ โ ๐น , s.t. ๐๐น (๐ก) lim = 0 and the inclusion ๐ธ โ ๐น is not DSS. ๐กโ0 ๐๐ธ (๐ก) Proposition 9. Let ๐ธ be an r.i. space with ๐ธ โ= ๐ฟโ , ๐ฟ1 . Then the inclusions ๐ฟโ โ ๐ธ โ ๐ฟ1 are strong. Proof. Since lim ๐๐ธ (๐ก) = 0 for any r.i. space ๐ธ โ= ๐ฟโ , we have ๐กโ0
lim
sup
๐โ0 โฅ๐ฅโฅ
๐ฟโ โฉฝ1,mes(supp ๐ฅ)โฉฝ๐
โฅ๐ฅโฅ๐ธ โฉฝ lim ๐๐ธ (๐) = 0. ๐โ0
And, since ๐ธ โ= ๐ฟ1 , so lim ๐๐ธ โฒ (๐ก) = 0, we have ๐กโ0
mes(supp ๐ฅ) โฉฝ ๐ โฅ๐ฅโฅ๐ฟ1 โฉฝ lim โฅ๐ฅโฅ๐ธ โฅรฆsupp ๐ฅ โฅ๐ธ โฒ ๐โ0
โฉฝ lim ๐๐ธ โฒ (๐) = lim ๐โ0
๐โ0
๐ = 0. ๐๐ธ (๐)
Proposition 10. Let ๐ธ, ๐น be a pair of r.i. spaces and assume that )โฒ โซ1 ( ๐ก ๐โฒ๐น (๐ก) ๐๐ก < โ. ๐๐ธ (๐ก)
โก
(7)
0
Then ๐ธ โ ๐น and this inclusion is strong. Proof. It is known ([KPS], 2.5.5, 2.5.7) that ๐ธ โ ๐ (๐ยฏ๐ธ ) and ฮ(๐๐น ) โ ๐น where ๐ยฏ๐ธ (๐ก) = ๐ก/๐๐ธ (๐ก). Assumption (7) implies the inclusion ๐ (๐ ยฏ๐ธ ) โ ฮ(๐๐น ) from which follows ๐ธ โ ๐น ([GHSS]). Let us show that (7) implies that this last inclusion is strong. Indeed, if โฅ๐ฅโฅ๐ (๐ยฏ๐ธ ) โฉฝ 1 and mes(supp ๐ฅ) โฉฝ ๐, then, by ([KPS], 2.2.36), โซ1 โฅ๐ฅโฅฮ(๐๐น ) =
๐ฅ
โ
(๐ก)๐โฒ๐น (๐ก) ๐๐ก
0
Clearly, (7) implies
โซ๐ ( โฉฝ 0
๐ก ๐๐ธ (๐ก)
)โฒ
)โฒ ๐ก ๐โฒ๐น (๐ก) ๐๐ก = 0. ๐โ0 0 ๐๐ธ (๐ก) Thus ๐ (๐ยฏ๐ธ ) โ ฮ(๐๐ธ ) is strong, and hence ๐ธ โ ๐น . โซ
lim
๐
๐โฒ๐น (๐ก)๐๐ก.
(
โก
Denote by ๐ the set of all convex increasing functions on [0, โ) s. t. ๐ (0) = 0, lim ๐ (๐ข)/๐ข = โ. Given an r.i. space ๐น and a function ๐ โ ๐, denote by ๐ขโโ
๐น (๐ ) the r.i. space endowed with the norm: 1 ( )1 { } 1 โฃ๐ฅโฃ 1 1 1 โฅ๐ฅโฅ๐น (๐) = inf ๐ > 0 : 1๐ โฉฝ1 . ๐ 1๐น It is clear that ๐น (๐ ) is an r.i. space. Note that ๐ฟ1 (๐ ) coincides with the Orlicz space ๐ฟ๐ . We need some auxiliary results to give a characterization of strong inclusions.
Bernstein Widths and Super Strictly Singular Inclusions
367
Lemma 11. Let ๐ข๐ , ๐ฃ๐ โฉพ 0 for every ๐ โ โ and lim ๐ข๐ = lim ๐ฃ๐ = โ. There ๐โโ
๐โโ
exists a function ๐ โ ๐ s. t. ๐ (๐ข๐ ) โฉฝ ๐ข๐ ๐ฃ๐ for every ๐ โ โ.
Proof. Without loss of generality we may assume that the sequences {๐ข๐ }, {๐ฃ๐ } are strictly monotone and ๐ข1 = ๐ฃ1 = 0. The set ๐ = {(๐ข๐ , ๐ข๐ ๐ฃ๐ ) , ๐ โ โ} uniquely de๏ฌnes a function ๐ on [0, โ) by: ๐ (๐ฅ) = inf{๐ฆ : (๐ฅ, ๐ฆ) โ conv ๐}. Clearly, ๐ (0) = 0 and ๐ is convex. Note that ๐ is a piecewise linear function, and that the angular points of its graph form an in๏ฌnite subset {(๐ข๐๐ , ๐ข๐๐ ๐ฃ๐๐ ) : ๐ โฉพ 1} of ๐. Indeed let us indicate an algorithm de๏ฌning the ๐๐ โs. Assume that 1 = ๐1 < ๐2 < โ
โ
โ
< ๐๐ have been determined. Then since ๐ข๐ ๐ฃ๐ โ ๐ข๐๐ ๐ฃ๐๐ ๐ข๐ ๐ฃ๐ โฉพ ๐ฃ๐ โ ๐ ๐ โ +โ ๐ข๐ โ ๐ข๐๐ ๐ข๐ the in๏ฌmum
{
} ๐ข๐ ๐ฃ๐ โ ๐ข๐๐ ๐ฃ๐๐ inf : ๐ > ๐๐ ๐ข๐ โ ๐ข๐๐ is attained, and the set ๐ด๐ of minimizers is ๏ฌnite. Then ๐๐+1 = max ๐ด๐ . Let us show that lim ๐ (๐ข) /๐ข = โ. Since the function ๐ is convex and ๐ (0) = 0, the ๐ขโโ
function ๐ (๐ฅ)/๐ฅ is nondecreasing on [0, +โ). But ๐ (๐ฅ) ๐ (๐ข๐๐ ) โฉพ sup = sup ๐ฃ๐๐ = +โ. ๐ฅ ๐ข๐๐ ๐ฅ>0 ๐โฉพ1 ๐โฉพ1
sup
โก
Lemma 12. Let ๐ธ be an r.i. space and ๐ โ ๐. The inclusion ๐ธ(๐ ) โ ๐ธ is strong i๏ฌ ๐ธ โ= ๐ฟโ . Proof. It is evident that ๐ฟโ (๐ ) = ๐ฟโ for any ๐ โ ๐. This proves the ๏ฌrst part of our statement. Suppose that ๐ธ โ= ๐ฟโ . Then lim ๐๐ธ (๐ก) = 0. If โฅ๐ฅโฅ๐ธ(๐) < 1, ๐กโ0
then โฅ๐ (โฃ๐ฅโฃ)โฅ๐ธ โฉฝ 1. Given ๐ โฉพ 1, consider the sets ๐ = {๐ก : ๐ก โ supp ๐ฅ, โฃ๐ฅ(๐ก)โฃ โฉฝ ๐} and ๐ = {๐ก : โฃ๐ฅ (๐ก) โฃ > ๐}. Since ๐ (โฃ๐ฅ (๐ก)โฃ) โฉพ ๐ โฃ๐ฅ (๐ก)โฃ รฆ๐ (๐ก) for every ๐ก โ [0, 1], then 1 1 โฅ๐ฅรฆ๐ โฅ๐ธ โฉฝ โฅ๐ (โฃ๐ฅโฃ)โฅ๐ธ โฉฝ . If mes (supp ๐ฅ) โฉฝ ๐, then ๐ ๐ 1 โฅ๐ฅโฅ๐ธ โฉฝ โฅ๐ฅรฆ๐ โฅ๐ธ + โฅ๐ฅรฆ๐ โฅ๐ธ โฉฝ ๐๐๐ธ (๐) + . ๐ 1
If we take ๐ = (๐๐ธ (๐))โ 2 , then 1
1
1
โฅ๐ฅโฅ๐ธ โฉฝ (๐๐ธ (๐)) 2 (๐) + (๐๐ธ (๐)) 2 = 2(๐๐ธ (๐)) 2 . Hence lim
๐โ0
sup
โฅ๐ฅโฅ๐ธ(๐ ) โฉฝ1, mes(supp ๐ฅ)โฉฝ๐
โฅ๐ฅโฅ๐ธ = 0.
โก
Theorem 13. Let ๐ธ, ๐น be a pair of r.i. spaces with ๐ธ โ ๐น . The inclusion ๐ธ โ ๐น is strong i๏ฌ ๐ธ โ ๐น (๐ ) for some ๐ โ ๐.
368
F.L. Hernยดandez, Y. Raynaud and E.M. Semenov
Proof. The su๏ฌciency follows immediately from Lemma 12. Let us prove the necessity. If the inclusion ๐ธ โ ๐น is strong, then there exists a sequence ๐๐ โ 0 s. t. โฅ๐ฅโฅ๐น โฉฝ 2โ๐ โฅ๐ฅโฅ๐ธ for any ๐ฅ โ ๐ธ with mes (supp ๐ฅ) < ๐๐ . We can assume that ๐ธ โ= ( ๐ฟโ so ๐)๐ธ (๐๐ ) โ 0. Now, by Lemma 11 there exists a function ๐ โ ๐ s. t. 1 ๐ ๐ โฉฝ for any ๐ โ โ. Let us show that ๐ธ โ ๐น (๐ ) where ๐ is ๐๐ธ (๐๐ ) ๐๐ธ (๐๐ ) the above constructed function. Let ๐ฅ โ ๐ธ, โฅ๐ฅโฅ๐ธ = 1. Consider the following sequence of functions { ๐ฅโ (๐ก) , ๐๐ < ๐ก โฉฝ ๐๐โ1 ๐ฅ๐ (๐ก) = 0, for other ๐ก โ [0, 1] where ๐0 = 1. We have 1 = โฅ๐ฅโ โฅ๐ธ โฉพ ๐ฅโ (๐๐ ) ๐๐ธ (๐๐ ) for each ๐ โ โ. Therefore โ1 ๐ฅโ (๐ก) โฉฝ ๐ฅโ (๐๐ ) โฉฝ (๐๐ธ (๐๐ )) for ๐ก โ [๐๐ , 1]. The function ๐ (๐ข) /๐ข is a monotone increasing one. Hence ( ) ๐ (๐ฅ๐ (๐ก)) 1 ๐ (๐ฅ๐ (๐ก)) = ๐ฅ๐ (๐ก) โฉฝ ๐ฅ๐ (๐ก) ๐ ๐๐ธ (๐๐ ) ๐ฅ๐ (๐ก) ๐๐ธ (๐๐ ) ( ) 1 for ๐ก โ (๐๐ , ๐๐โ1 ]. By the construction of ๐ , we have ๐ ๐๐ธ (๐๐ ) โฉฝ ๐. ๐๐ธ (๐๐ ) Consequently ๐ (๐ฅ๐ (๐ก)) โฉฝ ๐๐ฅ๐ (๐ก) and โ โ ๐=1
โฅ๐ (๐ฅ๐ )โฅ๐น โฉฝ
Hence the series
โ โ ๐=1
โ โ ๐=1
2โ๐ โฅ๐ (๐ฅ๐ )โฅ๐ธ โฉฝ
โ โ ๐=1
2โ๐ ๐ โฅ๐ฅ๐ โฅ๐ธ โฉฝ
โ โ
2โ๐ ๐ < โ.
๐=1
๐ ๐ฅ๐ converges in ๐น . On the other hand by the monotone
convergence theorem it converges clearly in ๐ฟ1 to ๐ (๐ฅโ ), which has thus to be also its limit in ๐น . Thus (๐ (๐ฅ))โ = ๐ (๐ฅโ ) belongs to ๐น , and so does ๐ ๐ฅ. Thus the inclusion ๐ธ โ ๐น (๐ ) has been proved. โก
3. SSS inclusions The criterion for SS inclusion of an r.i. space into ๐ฟ1 that was mentioned in Introduction may be straightened as follows. Theorem 14. Let ๐ธ be an r.i. space that does not contain an isomorphic copy of ๐0 . If there exist ๐ถ > 0 and a sequence of subspaces ๐๐ โ ๐ธ with dim ๐๐ = ๐, ๐ โ โ such that โฅ๐ฅโฅ๐ธ โฉฝ ๐ถโฅ๐ฅโฅ๐ฟ1 for any ๐ฅ โ ๐๐ , then ๐บ โ ๐ธ. Proof. Since ๐ธ does not contain a copy of ๐0 one can use a smooth ultraproduct argument. Indeed we may assume that ๐ธ โฒ is not ๐ฟโ , so ๐๐ธ โฒ (0+) = 0. If ๐ is a free ultra๏ฌlter on N, then in the ultrapower ๐ธ๐ the band ๐ต generated by ๐ธ consists
Bernstein Widths and Super Strictly Singular Inclusions
369
of elements [๐ฅ๐ ]๐ de๏ฌned by ๐ธ-equi-integrable sequence (๐ฅ๐ ), while the complementary band ๐ธ โฅ consists of elements represented by bounded sequence (๐ฅ๐ ) with mes(supp(๐ฅ๐ )) โ 0. By Hยจ older inequality in the second case the sequence (๐ฅ๐ ) goes to zero in ๐ฟ1 . Call ๐ the natural inclusion ๐ธ โ ๐ฟ1 . Let as usual its ultrapower map ๐๐ : ๐ธ๐ โ (๐ฟ1 )๐ be de๏ฌned by ๐๐ ([๐ฅ๐ ]๐ ) = [๐(๐ฅ๐ )]๐ . Then by the preceding ๐๐ vanishes on the complementary band ๐ต โฅ of ๐ต in ๐ธ๐ . It is clear that ๐๐ maps ๐ต into the band generated by ๐ฟ1 in its ultrapower (indeed ๐ธ-equiintegrable sequence are a fortiori ๐ฟ1 -equiintegrable). This band ๐ต1 can be identi๏ฌed with a ๐ฟ1 of a big probability space (๐, ฮฃ, ๐) and (since ๐ธ is order continuous) ๐ต is identi๏ฌed to ๐ธ(๐, ฮฃ, ๐) (i.e., ๐ โ ๐ธ(๐, ฮฃ, ๐) i๏ฌ ๐ โ โ ๐ธ), and ๐๐ when restricted to ๐ต is simply the inclusion ๐ธ(๐, ฮฃ, ๐) โ ๐ฟ1 (๐, ฮฃ, ๐). In particular if ๐ธ does not contain ๐0 , neither does the band ๐ต, and it results that ๐ต is a projection band (this remark goes back to [W]). If (๐๐ ) is a sequence of subspaces of ๐ธ with dim ๐๐ = ๐ and โฅ๐ฅโฅ๐ธ โฉฝ ๐ถโฅ๐ฅโฅ1 for any ๐ โฉพ 1 and ๐ฅ โ ๐๐ , consider the ultraproduct ๐ := ฮ ๐ ๐๐ which is an in๏ฌnite-dimensional subspace of ๐ธ๐ with โฅ๐ฅโฅ๐ธ๐ โฉฝ ๐ถโฅ๐๐ ๐ฅโฅ(๐ฟ1 )๐ for every ๐ฅ โ ๐. Let ๐ be the band projection from ๐ธ๐ onto ๐ต. Since ๐๐ = ๐๐ ๐ we have โฅ๐ฅโฅ๐ธ๐ โฉฝ ๐ถโฅ๐๐ ๐๐ฅโฅ(๐ฟ1 )๐ โฉฝ ๐ถโฅ๐ฅโฅ๐ธ๐ for every ๐ฅ โ ๐. Hence ๐ restricts to an isomorphism on ๐, and in particular its range ๐(๐) is an in๏ฌnite-dimensional closed space. Moreover on this subspace the norms of ๐ธ(๐, ฮฃ, ๐) and that of ๐ฟ1 (๐, ฮฃ, ๐) are ๐ถ-equivalent since for ๐ฆ = ๐๐ฅ โ ๐(๐) we have โฅ๐ฆโฅ๐ธ๐ โฉฝ โฅ๐ฅโฅ๐ธ๐ โฉฝ ๐ถโฅ๐๐ ๐๐ฅโฅ(๐ฟ1 )๐ = โฅ๐๐ ๐ฆโฅ(๐ฟ1 )๐ โฉฝ ๐ถโฅ๐ฆโฅ๐ธ๐ . Hence ๐บ(๐, ฮฃ, ๐) โ ๐ธ(๐, ฮฃ, ๐) by Theorem 1 in [AHS]. Coming back on earth to the measure space [0, 1] this means that ๐บ โ ๐ธ. โก Lemma 15. Let ๐ธ be an r.i. space and ๐ง โ ๐ฟ1 โ ๐ธ โฒโฒ .There exists an re๏ฌexive r.i. space ๐ธ1 s.t. ๐ธ โ ๐ธ1 and ๐ง โโ ๐ธ1 . Proof. We may suppose that ๐ง = ๐ง โ . We can ๏ฌnd a function ๐ข โ ๐ธ โฒ s.t. ๐ข = ๐ขโ and โซ 1 ๐ง(๐ก)๐ข(๐ก) ๐๐ก = โ. (8) 0
And moreover we can ๏ฌnd a function ๐ฃ โ ๐ธ โฒ s.t. ๐ฃ = ๐ฃ โ , lim ๐ฃ(๐ก)/๐ข(๐ก) = 0 and ๐กโ0
โซ
1 0
Denote
โซ ๐(๐ ) =
๐ 0
๐ง(๐ก)๐ฃ(๐ก) ๐๐ก = โ. โซ
๐ข(๐ก) ๐๐ก,
๐(๐ ) =
0
๐
๐ฃ(๐ก) ๐๐ก.
Then ๐ and ๐ are concave increasing function, lim ๐(๐ )/๐(๐ ) = 0 and ๐ โ0
๐ (๐) โ ๐ (๐) โ ๐ธ โฒ .
(9)
370
F.L. Hernยดandez, Y. Raynaud and E.M. Semenov
Indeed, if ๐ฅ โ ๐ (๐), โฅ๐ฅโฅ๐ (๐) โฉฝ 1, then โซ ๐ โซ โ ๐ฅ (๐ก) ๐๐ก โฉฝ ๐(๐ ) = 0
๐ 0
๐ข(๐ก) ๐๐ก
(10)
for every ๐ โ [0, 1]. By [LT2], 2.a.8 ๐ฅ โ ๐ธ โฒ and โฅ๐ฅโฅ๐ธ โฒ โฉฝ โฅ๐ขโฅ๐ธ โฒ . It follows from (9) and the well-known formula (๐ (๐))โฒ = ฮ(๐) that ๐ธ โ ฮ(๐) โ ฮ(๐).
(11)
Let ๐ฅ โ ฮ(๐) and ๐(supp ๐ฅ) โฉฝ ๐ for some ๐ > 0. Then โซ 1 โซ ๐ ๐ โฒ (๐ก) โฅ๐ฅโฅฮ(๐) = ๐ฅโ (๐ก) ๐๐(๐ก) = ๐ฅโ (๐ก)๐โฒ (๐ก) โฒ ๐๐ก. ๐ (๐ก) 0 0 Therefore
๐ โฒ (๐ก) โฅ๐ฅโฅฮ(๐) . โฒ 0<๐กโฉฝ๐ ๐ (๐ก)
โฅ๐ฅโฅฮ(๐) โฉฝ sup Since
๐ โฒ (๐ก) ๐ฃ(๐ก) = lim =0 โฒ ๐กโ0 ๐ (๐ก) ๐กโ0 ๐ข(๐ก) lim
we have lim
๐โ0
sup
โฅ๐ฅโฅฮ(๐) โฉฝ1 mes(supp ๐ฅ)โฉฝ๐
โฅ๐ฅโฅฮ(๐) = 0
and by [BVL] the embedding ฮ(๐) โ ฮ(๐) is weakly compact. Using [B, Chap. II, ยง3], we conclude that the space of real interpolation method ([BL, 3.1]) ๐ธ1 := (ฮ(๐), ฮ(๐))๐,๐ is re๏ฌexive for 0 < ๐ < 1 < ๐ < โ. By (8), (11) ๐ง โโ ๐ธ1 and ๐ธ โ ๐ธ1 . โก Denote by โ the class of all r.i. spaces ๐ธ on [0, 1] verifying that there exists a sequence ๐๐ โ 0 s.t. for any subspace ๐ โ ๐ธ with dim ๐ = ๐ we can select ๐ฅ โ ๐ s.t. ๐ฅโ (๐๐ ) < ๐๐ and โฅ๐ฅโฅ๐ธ โฉพ 1. Clearly if ๐ธ1 , ๐ธ2 are r.i. spaces with ๐ธ1 โ โ and ๐ธ2 โ ๐ธ1 then ๐ธ2 โ โ. Theorem 16. Let ๐ธ be an r.i. space. The following conditions are equivalent: i) ๐ธ โ โ. ii) ๐บ โ โ ๐ธ. 1 ( )1 1 1 1/2 1 1 iii) lim 1 min ๐, ln = โ. 1 ๐โโ ๐ก 1๐ธ Proof. (i)โ(ii). Suppose that ๐บ โ ๐ธ. By ([LT2], 2.b.4) there exists a constant ๐ถ > 0 s.t. 1 1 ๐ 1โ 1 1 1 1 โ โฅ๐โฅ๐2 โฉฝ 1 ๐๐ ๐๐ 1 โฉฝ ๐ถโฅ๐โฅ๐2 1 1 2 ๐=1 ๐ธ
Bernstein Widths and Super Strictly Singular Inclusions
371
for any ๐ โ โ and ๐ โ ๐
๐ , where {๐๐ } is the Rademacher system. Theorem 7 in [KS, ch. 2] states that $ { $ ๐ } $โ $ 1 1 $ $ mes ๐ก : $ ๐๐ ๐๐ (๐ก)$ โฉพ โฅ๐โฅ๐2 โฉพ . $ $ 2 32 ๐=1
Therefore ๐ธ โโ โ. (ii)โ(iii). If
1 ( )1 1 1 1/2 1 1 1 lim min ๐, ln < โ, ๐โโ 1 ๐ก 1๐ธ 1 then the function ๐ง(๐ก) = ln1/2 belongs to ๐ธ โฒโฒ . It is well known [L] that the Orlicz ๐ก space exp โซ ๐ ๐ฟ2 coincides up to equivalence with the Marcinkiewicz space ๐ (๐) where 1 ๐(๐ ) = ln1/2 ๐๐ก. If ๐ฅ โ ๐ (๐), then ๐ก 0 โซ ๐ โซ ๐ 1 1 โ ๐ฅ (๐ก) ๐๐ก โฉฝ โฅ๐ฅโฅ๐ (๐) ln /2 ๐๐ก ๐ก 0 0 for every ๐ โ [0, 1]. Hence ๐ฅ โ ๐ธ โฒโฒ and 1 1 1/2 โฅ๐ฅโฅ๐ธ โฒโฒ โฉฝ 1 1ln
1 11 1 โฅ๐ฅโฅ๐(๐) . ๐ก 1๐ธ โฒโฒ
Hence exp ๐ฟ2 = ๐ (๐) โ ๐ธ โฒโฒ and ๐บ is contained in the closure of ๐ฟโ in ๐ธ โฒโฒ , i.e., ๐บ โ ๐ธ. 1 (iii)โ(i). By (iii) we have ln1/2 โโ ๐ธ โฒโฒ . If ๐บ โ ๐ธ, then exp ๐ฟ2 = ๐บโฒโฒ โ ๐ธ โฒโฒ ๐ก 1 and ln1/2 โ ๐ธ โฒโฒ . The obtained contradiction shows that ๐บ โโ ๐ธ and ๐บ โโ ๐ธ โฒโฒ . ๐ก 1 By Lemma 15 there exists a re๏ฌexive r.i. space ๐ธ1 s.t. ln1/2 โโ ๐ธ1 and ๐ธ โ ๐ธ1 . ๐ก 1/2 1 โฒโฒ โฒโฒ Then ๐บ โโ ๐ธ1 . Indeed, if ๐บ โ ๐ธ1 , then ln โ ๐บ โ ๐ธ1 = ๐ธ1 . So, ๐บ โโ ๐ธ1 . ๐ก Since ๐ธ1 is re๏ฌexive, ๐ธ1 does not contain a subspace isomorphic to ๐0 . Suppose that ๐ธ1 โโ โ. Then for some ๐ > 0 and any ๐ โ โ there exists a subspace ๐๐ โ ๐ธ1 , dim ๐๐ = ๐ s.t. ๐ฅโ (๐) โฉพ ๐ for any ๐ฅ โ ๐๐ with โฅ๐ฅโฅ๐ธ = 1. For such ๐ฅ โ ๐๐ we have โซ โฅ๐ฅโฅ๐ฟ1 โฉพ This means that
๐
0
๐ฅโ (๐ก) ๐๐ก โฉพ ๐2 .
โฅ๐ฅโฅ๐ธ1 โฉฝ ๐โ2 โฅ๐ฅโฅ๐ฟ1 for any ๐ฅ โ ๐๐ . Since ๐ธ1 does not contain a subspace isomorphic to ๐0 , then we can apply Theorem 14 and state that ๐บ โ ๐ธ1 . The obtained contradiction proves that ๐ธ1 โ โ and a fortiori ๐ธ โ โ. โก Theorem 17. Let ๐ธ, ๐น be a pair of r.i. spaces. If ๐บ โโ ๐ธ and ๐ธ is strongly included into ๐น , then the inclusion ๐ธ โ ๐น is SSS.
372
F.L. Hernยดandez, Y. Raynaud and E.M. Semenov
Proof. Since ๐บ โโ ๐ธ then, by Theorem 16, ๐ธ โ โ, i.e., there exists a sequence ๐๐ โ 0 s.t. for any subspace ๐๐ โ ๐ธ, dim ๐๐ = ๐ there exists ๐ฅ๐ โ ๐ for which ๐ฅโ๐ (๐๐ ) < ๐๐ and โฅ๐ฅ๐ โฅ๐ธ = 1. We have โฅ๐ฅ๐ โฅ๐น = โฅ๐ฅโ๐ โฅ๐น โฉฝ โฅ๐ฅโ๐ ๐(0,๐๐ ) โฅ๐น + โฅ๐ฅโ๐ ๐(๐๐ ,1) โฅ๐ธ โฉฝ sup โฅ๐ฅ๐๐ โฅ๐น + ๐๐ . โฅ๐ฅโฅ๐ธ =1 mesโฉฝ๐๐
Since ๐ธ is strongly included into ๐น then the ๏ฌrst term in above tends to 0. Hence lim โฅ๐ฅ๐ โฅ๐น = 0.
๐โโ
โก
As a direct consequence of Theorem 17 and Proposition 9 we have Corollary 18. Let ๐ธ be an r.i. space. The inclusion ๐ธ โ ๐ฟ1 is SSS i๏ฌ ๐บ โโ ๐ธ. Corollary 19. Let ๐ธ, ๐น be a pair of r.i. spaces, such that ๐บ โโ ๐ธ and the integral condition (7) in Prop. 10 is satis๏ฌed. Then the inclusion ๐ธ โ ๐น is SSS. This statement immediately follows from Theorem 17 and Proposition 10. Now we apply Corollary 19 for exponential Orlicz spaces. Corollary 20. Let 0 < ๐ < ๐ < โ. The following conditions are equivalent: (i) the inclusion exp ๐ฟ๐ โ exp ๐ฟ๐ is SS; (ii) the inclusion exp ๐ฟ๐ โ exp ๐ฟ๐ is SSS; (iii) ๐ > 2. Proof. The equivalence (i)โโ(iii) was proved in [HNS]. The implication (ii)โ(i) is obvious. Therefore we must prove that the inclusion exp ๐ฟ๐ โ exp ๐ฟ๐ is SSS for ๐ > ๐ > 2. It is well known that up to equivalence 1 ๐ ๐exp ๐ฟ๐ (๐ก) = lnโ ๐ ๐ก for any ๐ > 0. We have )โฒ ) โซ 1( โซ 1( ( )โฒ 1 ๐ ๐ก 1 1๐ โ1 ๐ 1 โ 1๐ โ1 ๐ ๐๐ก โ ๐1 ๐ ๐ 0< ln ๐๐ก = ln โ ln ln 1 ๐ก ๐ก ๐ ๐ก ๐ ๐ก ๐ก 0 0 lnโ ๐ ๐๐ก โซ 1 โซ โ 1 1 1 1 1 ๐ ๐๐ก 1 ๐ < ln ๐ โ ๐ โ1 = ๐ ๐ โ ๐ โ1 ๐๐ = < โ. ๐ 0 ๐ก ๐ก ๐ 1 ๐โ๐ By Corollary 19 the inclusion exp ๐ฟ๐ โ exp ๐ฟ๐ is SSS; so is also the inclusion exp ๐ฟ๐ โ exp ๐ฟ๐ by composition with the bounded inclusion exp ๐ฟ๐ โ exp ๐ฟ๐ . โก Corollary 21. Let ฮ(๐), ฮ(๐) be Lorentz spaces, ๐ โฉฝ ๐ and ๐บ โโ ฮ(๐). The following conditions are equivalent. 1) The inclusion ฮ(๐) โ ฮ(๐) is DSS; 2) the inclusion ฮ(๐) โ ฮ(๐) is SS; 3) the inclusion ฮ(๐) โ ฮ(๐) is SSS; 4) the inclusion ฮ(๐) โ ฮ(๐) is strong; ๐(๐ก) 5) lim = 0. ๐กโ0 ๐(๐ก)
Bernstein Widths and Super Strictly Singular Inclusions
373
Proof. (5)โ(4). By Lemma 2.5.2 in [KPS] sup
โฅ๐ฅโฅฮ(๐) โฉฝ1,mes(supp ๐ฅ)โฉฝ๐
โฅ๐ฅโฅฮ(๐) =
โฅรฆ๐ โฅฮ(๐) ๐(๐ก) = sup . 0<mes ๐โฉฝ๐ โฅรฆ๐ โฅฮ(๐) 0<๐กโฉฝ๐ ๐(๐ก) sup
Hence lim
๐โ0 โฅ๐ฅโฅ
sup
ฮ(๐) โฉฝ1,mes(supp ๐ฅ)โฉฝ๐
โฅ๐ฅโฅฮ(๐) = lim sup
๐โ0 0<๐กโฉฝ๐
๐(๐ก) = 0. ๐(๐ก)
The implication (4)โ(3) follows from Theorem 17. The implications (3)โ(2)โ(1) are obvious. The equivalence (1)โ(5) was proved in [A]. โก Corollary 21 cannot be extended to the class of Orlicz spaces. Given 1 < ๐ < โ, there exists an Orlicz space ๐ฟ๐ such that the inclusion ๐ฟ๐ โ ๐ฟ๐ is DSS but ๐๐ฟ๐ (๐ก) lim sup 1/๐ > 0 (cf. [GHSS]). ๐ก ๐กโ0 Applying Theorem 13 and 17 we get Corollary 22. Let ๐ธ, ๐น be a pair of r.i. spaces with ๐ธ โ ๐น . If ๐บ โโ ๐ธ and ๐ธ โ ๐น (๐ ) for some ๐ โ ๐, then the inclusion ๐ธ โ ๐น is SSS. Using Corollary 22 we get another proof of Corollary 20. Now we present a simple necessary condition for SSS inclusions. Proposition 23. Let ๐ธ, ๐น be a pair of r.i. spaces with ๐ธ โ ๐น and ๐๐ธ = ๐๐น . Then the inclusion ๐ธ โ ๐น is not SSS. Proof. Consider the following two cases: ๐๐ธ (2๐ก) ๐๐ธ (2๐ก) 1) lim inf > 1, 2) lim inf = 1. ๐กโ0 ๐กโ0 ๐๐ธ (๐ก) ๐๐ธ (๐ก) ๐๐ธ (2๐ก) > 2๐พ for some ๐พ > 0. Then ๐๐ธ (๐ก) โฉฝ ๐ถ๐ก๐พ for some ๐ถ > 0 ๐๐ธ (๐ก) and su๏ฌciently small ๐ก > 0. By Proposition 10, ๐ธ โ ๐ฟ๐ for ๐ โ (1/๐พ , โ). Khintchine inequality implies that the norms ๐ธ and ๐น are equivalent on [(๐๐ )] where {๐๐ } is the Rademacher system. This means that the inclusion ๐ธ โ ๐น is not SS ( so neither SSS). ๐๐ธ (2๐ก) ๐๐ธ (๐๐ก) If lim inf = 1, then lim inf = 1 for any ๐ โ โ ([KPS], 1.1.3). ๐กโ0 ๐กโ0 ๐๐ธ (๐ก) ๐๐ธ (๐ก) Therefore there exists a sequence ๐ก๐ โ 0 s.t. ๐๐ธ (๐๐ก๐ ) โฉฝ 2๐๐ธ (๐ก๐ ) for any ๐ โ โ. Let ๐ฅ๐ (๐ก) = รฆ( ๐โ1 ๐ก๐ , ๐ ๐ก๐ ) (๐ก), 1 โฉฝ ๐ โฉฝ ๐. Then ๐ ๐ 1 ๐ 1 1 ๐ 1 1โ 1 1โ 1 1 1 1 1 max โฃ๐๐ โฃ๐๐ธ (๐ก๐ ) โฉฝ 1 ๐๐ ๐ฅ๐ 1 โฉฝ max โฃ๐๐ โฃ 1 ๐ฅ๐ 1 1โฉฝ๐โฉฝ๐ 1 1 1โฉฝ๐โฉฝ๐ 1 1 Let lim inf ๐กโ0
๐=1
๐ธ
๐=1
๐ธ
= max โฃ๐๐ โฃ ๐๐ธ (๐๐ก๐ ) โฉฝ 2 max โฃ๐๐ โฃ ๐๐ธ (๐ก๐ ) 1โฉฝ๐โฉฝ๐
1โฉฝ๐โฉฝ๐
374
F.L. Hernยดandez, Y. Raynaud and E.M. Semenov
and analogously
1 1 ๐ 1โ 1 1 1 max โฃ๐๐ โฃ ๐๐น (๐ก๐ ) โฉฝ 1 ๐๐ ๐ฅ๐ 1 โฉฝ 2 max โฃ๐๐ โฃ ๐๐น (๐ก๐ ). 1โฉฝ๐โฉฝ๐ 1 1 1โฉฝ๐โฉฝ๐ ๐=1
๐น
Since ๐๐ธ = ๐๐น we have ๐พ๐ (๐ธ, ๐น ) โฉพ
max โฃ๐๐ โฃ๐๐น (๐ก๐ )
1โฉฝ๐โฉฝ๐
2 max โฃ๐๐ โฃ๐๐ธ (๐ก๐ )
=
1โฉฝ๐โฉฝ๐
for any ๐ โ โ. This means that the inclusion is not SSS.
1 2 โก
In particular the canonical inclusion ฮ(๐) โ ๐ (๐) is not SSS for any ๐ โ ฮฉ ๐ก where ๐(๐ก) = . And, by Theorem 11 in [AHS], the inclusion ฮ(๐) โ ๐ (๐) is ๐(๐ก) SS provided ๐บ โโ ฮ(๐) and ๐(+0) = 0. So we have: โซ 1 1 Corollary 24. Let ๐ โ ฮฉ with ๐(+0) = 0 and ln /2 1/๐ก ๐๐(๐ก) = โ. Then the inclusion ฮ(๐) โ ๐ (๐) is SS but not SSS. For example, the functions ๐(๐ก) = ln๐ผ
0
๐ satisfy the conditions of Corollary 24 ๐ก
if ๐ผ < โ1/2. Concerning Theorem 17 it is clear, as we mentioned in Introduction, that the assumption ๐บ โโ ๐ธ is necessary for the validity of it. For the class of Lorentz spaces Corollary 21 shows that to be a strong inclusion is also a necessary condition. We do not know what happens in general.
References [A]
Astashkin S.V., Disjointly strictly singular inclusions of symmetric spaces. Mat. Notes 65(1) (1999), 3โ12. [AHS] Astashkin S.V., Hernandez F.L. and Semenov E.M., Strictly singular inclusions of rearrangement invariant spaces and Rademacher spaces. Studia Math. 193(3) (2009), 269โ283. [B] Beauzamy B., Espaces dโinterpolation rยดeels: topologie et gยดeomยดetrie. LNM. 666, Springer Verlag, 1978. [BL] Bergh J., Lยจ ofstrยจ om J., Interpolation spaces, an introduction. Springer Verlag, 1976. [BVL] Bukhvalov A.V., Veksler A.I., Lozanovsky G.Ya., Banach lattices โ some Banach aspects of their theory. Russian Math. Surveys 34 (1979), 159โ213. [C] Calkin J.W., Abstract symmetric boundary conditions. Trans. Amer. Math. Soc., 45(3).(1939), 369โ442. [CCT] Castejยด on A., Corbacho E. and Tarieladze V., AMD-numbers, compactness, strict singularity and the essential spectrum of operators. Georgian Math. J. 9(2) (2002), 227โ270.
Bernstein Widths and Super Strictly Singular Inclusions
375
[FHR] Flores J., Hernยด andez F.L., Raynaud Y., Super strictly singular and cosingular operators and related classes. J. Operator Theory 67 (2012) (to appear). [GHSS] Garcยดฤฑa del Amo A., Hernยด andez F.L., Sยด anchez V.M., Semenov E.M., Disjointly strictly-singular inclusions between rearrangement invariant spaces. J. London Math. Soc., 62 (2000), 239โ252. [GMF] Gohberg I.C., Markus A.S., Feldman I.A., On normally solved operators and ideals related with them. Amer. Math. Soc. Transl. 61 (2) (1967), 63โ84. [HNS]
Hernยด andez F.L., Novikov S.Y., Semenov E.M., Strictly singular embeddings between rearrangement invariant spaces. Positivity 7 (2003), 119โ124.
[HR]
Hernยด andez F.L. and RodrยดฤฑguezโSalinas B., On ๐๐ -complemented copies in Orlicz spaces II. Israel J. of Math. 66 (1989), 27โ55.
[K]
Kato T., Perturbation theory for nullity, de๏ฌciency and other quantities of linear operators. J. Analyse Math. 6 (1958), 261โ322.
[KPS]
Krein S.G., Petunin Yu.I., Semenov E.M., Interpolation of linear operators. AMS, RI, 1982.
[KS]
Kashin B.S., Saakyan A.A., Orthogonal series.Translations Mathematical Monog., 75, American Mathematical Society, Providence, RI, 1989.
[L]
Lorentz G.G., Relations between function spaces. Proc. AMS, 12 (1961), 127โ 132.
[LT1]
Lindenstrauss J., Tzafriri L., Classical Banach Spaces. I. Springer Verlag, 1977.
[LT2]
Lindenstrauss J., Tzafriri L., Classical Banach Spaces. II. Springer Verlag, 1979.
[M]
Milman V.D., Operators of class ๐ถ0 and ๐ถ0โ . Theory of Functions, Functional Analysis and Appl. 10, Kharkov (1970), 15โ26 (Russian).
[MP]
Mityagin B.S. and Peฬทlczyยด nski A., Nuclear operators and approximate dimension. Proc. Inter. Congr. Math. Moscow (1966), 366โ372.
[N]
Novikov S.Ya., Boundary spaces for inclusion maps between rearrangement invariant spaces. Collect. Math. 44(1997), 211โ215.
[P]
Plichko A., Super strictly singular and super strictly cosingular operators in Functional analysis and its Applications, North-Holland math. St. 197. Elsevier. Amsterdam, 2004, 239โ255.
[PI]
Pinkus A., n-Widths in Approximation Theory. Springer Verlag, Berlin, 1985.
[PS]
Parfenov O.G. and Slupko M.V., Bernstein widths of embeddings of Lebesgue spaces. J. of Math. Sciences. 101, 2(2000), 3146โ3148.
[RSH]
Raynaud Y., Semenov E.M. and Hernยด andez F.L., Super strictly singular inclusions between rearrangement invariant spaces. Doklady Mathematics 83 (2011), 216โ218.
[S]
Semenov E.M., Finitely strictly singular embeddings. Doklady Mathematics 81 (2010), 383โ385.
[SSTT] Sari B., Schlumprecht T., Tomczak-Jagerman N. and Troitsky V., On norm closed ideals in ๐ฟ(๐๐ โ ๐๐ ). Studia Math. 179 (2007), 239โ262. [W]
Weis L., Banach lattices with the subsequence splitting property Proc. AMS 105, (1) (1989), 87โ96.
376
F.L. Hernยดandez, Y. Raynaud and E.M. Semenov
F.L. Hernยด andez Departamento de Anยด alisis Matemยด atico Universidad Complutense de Madrid, E-28040 Madrid, Spain e-mail:
[email protected] Y. Raynaud Institut de Mathยดematiques de Jussieu Site Jussieu (Case 247) UPMC-Univ. Paris06 and CNRS F-75252 PARIS cedex 05, France e-mail:
[email protected] E.M. Semenov Department of Mathematics Voronezh State University Voronezh 394693, Russia e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 377โ386 c 2012 Springer Basel AG โ
On Inversion of Certain Structured Linear Transformations Related to Block Toeplitz Matrices M.A. Kaashoek and F. van Schagen Dedicated to the memory of Israel Gohberg. We remember him as an outstanding mathematician, an inspiring teacher and a wonderful friend.
Abstract. This paper presents an explicit inversion formula for certain structured linear transformations that are closely related to ๏ฌnite block Toeplitz matrices. The conditions of invertibility are illustrated by an example. State space techniques from mathematical system theory play an important role. Mathematics Subject Classi๏ฌcation (2000). Primary 47B35; secondary 15A09, 93B99. Keywords. Structured operators, inversion, state space realization, ๏ฌnite block Toeplitz matrices, Gohberg-Heinig inversion formula.
1. Introduction This paper is an addition to Section 2 of [4], where the Gohberg-Heinig formula (see [2]) for the inverse of a ๏ฌnite block Toeplitz matrix is derived using state space techniques from mathematical systems theory. The starting point in Section 2 of [4] is the fact that any ๏ฌnite (๐ + 1)ร(๐ + 1) block Toeplitz matrix ๐ can be represented as โก โค ๐ผ โ ๐ถ๐ด๐ ๐ต โ๐ถ๐ด๐โ1 ๐ต โ
โ
โ
โ๐ถ๐ต โข โฅ โ๐ถ๐ด๐ต โฅ โขโ๐ถ๐ด๐+1 ๐ต ๐ผ โ ๐ถ๐ด๐ ๐ต โ
โ
โ
โข โฅ ๐ =โข (1.1) โฅ, .. .. .. .. โข โฅ . . . . โฃ โฆ โ๐ถ๐ด2๐ ๐ต โ๐ถ๐ด2๐โ1 ๐ต โ
โ
โ
๐ผ โ ๐ถ๐ด๐ ๐ต where ๐ด : ๐ณ โ ๐ณ , ๐ต : ๐ฐ โ ๐ณ , and ๐ถ : ๐ณ โ ๐ฐ are operators (linear transformations) acting between complex linear spaces and ๐ผ is the identity operator
378
M.A. Kaashoek and F. van Schagen
on ๐ฐ. The representation (1.1) allows one to study inversion of ๐ in terms of the operator ๐ โ ๐ =๐ธโ ๐ด๐โ๐ ๐ต๐ถ๐ด๐ : ๐ณ โ ๐ณ , (1.2) ๐=0
where ๐ธ is the identity operator on ๐ณ . To see this note that ๐ = ๐ท โ ๐น ๐บ and ๐ = ๐ธ โ ๐บ๐น , where ๐ท is the (๐ + 1)ร(๐ + 1) block diagonal matrix with ๐ผ as diagonal entries and โค โก ๐ถ โข ๐ถ๐ด โฅ ] [ โฅ โข (1.3) ๐น = โข . โฅ , ๐บ = ๐ด๐ ๐ต ๐ด๐โ1 ๐ต โ
โ
โ
๐ด๐ต ๐ต . . โฃ . โฆ ๐ถ๐ด๐
Assuming ๐ to be invertible, this connection between ๐ and ๐ is used in [4] to give a new proof of the Gohberg-Heinig formula for the inverse of ๐ . In the present paper we present necessary and su๏ฌcient conditions for ๐ to be invertible and we derive a formula for the inverse of ๐ (which was not done in [4]). To do this the four equations in the Gohberg-Heinig theorem are replaced by the equations ๐ ๐พ = ๐ด๐ ๐ต,
๐ ๐ฟ = ๐ต,
๐
๐ = ๐ถ๐ด๐ ,
๐๐ = ๐ถ,
(1.4)
where the operators ๐พ and ๐ฟ from ๐ฐ into ๐ณ and the operators ๐
and ๐ from ๐ณ into ๐ฐ are the unknowns. The following theorem is our main result. Theorem 1.1. Assume there exist operators ๐พ and ๐ฟ from ๐ฐ into ๐ณ and operators ๐
and ๐ from ๐ณ into ๐ฐ satisfying the equations in (1.4). If, in addition, one of the following conditions is satis๏ฌed 1. ๐ผ + ๐ถ๐พ is invertible, 2. ๐ผ + ๐
๐ต is invertible, 3. ๐ผ + ๐ถ๐พ is surjective and ๐ผ + ๐
๐ต is injective, 4. ๐ผ + ๐
๐ต is surjective and ๐ผ + ๐ถ๐พ is injective, then ๐ is invertible and both ๐ผ + ๐ถ๐พ and ๐ผ + ๐
๐ต are invertible. Moreover, in that case ๐ โ ๐ โ1 = ๐ธ + ๐ด๐โ๐ ๐ต(๐๐ด๐ + ๐ถ๐ป๐ ), (1.5) ๐=0
where the linear transformations ๐ป๐ are de๏ฌned recursively by ๐ป0 = 0,
๐ป1 = ๐ด๐พ(๐ผ + ๐ถ๐พ)โ1 ๐ โ ๐ฟ(๐ผ + ๐
๐ต)โ1 ๐
๐ด,
๐ป๐ = ๐ด๐ป๐โ1 + (๐ป1 ๐ด๐โ2 )๐ด
(๐ = 2, . . . , ๐).
We shall give a self-contained proof of the above theorem, not using the connection between ๐ and ๐ . For other recent developments related tot the Gohberg-Heing inversion formula we refer to the extended introduction of [6] and the references given therein.
On Inversion of Certain Structured Linear Transformations
379
The paper consists of two sections not counting the present introduction. The proof of Theorem 1.1 is given in Section 2. When ๐ฐ or ๐ณ is ๏ฌnite dimensional, then injectivity of ๐ implies surjectivity of ๐ and vice versa. As one may expect this property does not hold when both ๐ฐ and ๐ณ are in๏ฌnite dimensional, not even when the four equations in (1.4) are solvable. In Section 3 we present an example to illustrate this fact. In this ๏ฌnal section we also present a corollary to Theorem 1.1 and discuss a few special cases.
2. Proof of the main result It will be convenient ๏ฌrst to state and prove a lemma that covers part of Theorem 1.1. Lemma 2.1. Assume there exist operators ๐พ and ๐ฟ from ๐ฐ into ๐ณ and operators ๐
and ๐ from ๐ณ into ๐ฐ satisfying the equations in (1.4). Then the following two statements hold true: 1. if ๐ผ + ๐
๐ต or ๐ผ + ๐ถ๐พ is injective, then ๐ is injective, 2. if ๐ผ + ๐
๐ต or ๐ผ + ๐ถ๐พ is surjective, then ๐ is surjective. Moreover, if ๐ผ + ๐ถ๐พ or ๐ผ + ๐
๐ต is invertible, then ๐ผ + ๐ถ๐พ, ๐ผ + ๐
๐ต and ๐ are invertible. Proof. The proof of the lemma will be divided into four parts. In the ๏ฌrst two parts we prove the ๏ฌrst statement. The second statement is proved in the third part. The proof of the ๏ฌnal statement is given in the last part. Throughout ฮฉ is โ๐โ1 the operator on ๐ณ de๏ฌned by ฮฉ = ๐=0 ๐ด๐โ1โ๐ ๐ต๐ถ๐ด๐ . Note that ๐ + ๐ต๐ถ๐ด๐ = ๐ธ โ ๐ดฮฉ,
๐ + ๐ด๐ ๐ต๐ถ = ๐ธ โ ฮฉ๐ด.
Hence the following intertwining relations hold true: (๐ + ๐ต๐ถ๐ด๐ )๐ด = ๐ด(๐ + ๐ด๐ ๐ต๐ถ),
ฮฉ(๐ + ๐ต๐ถ๐ด๐ ) = (๐ + ๐ด๐ ๐ต๐ถ)ฮฉ. ๐
(2.1) ๐
Furthermore, we shall use that ๐ + ๐ต๐ถ๐ด is invertible if and only if ๐ + ๐ด ๐ต๐ถ is invertible. Part 1. We assume ๐ผ + ๐
๐ต is injective and prove that ๐ is injective. Note that ๐ผ + ๐
๐ต is injective if and only if ๐ธ + ๐ต๐
is injective. Take ๐ฅ โ Ker ๐ , that is, ๐ ๐ฅ = 0. Then ๐ถ๐ฅ = ๐๐ ๐ฅ = 0, and we see that (๐ธ โ ฮฉ๐ด)๐ฅ = (๐ + ๐ด๐ ๐ต๐ถ)๐ฅ = 0. So ๐ฅ = ฮฉ๐ด๐ฅ and (๐ธ โ ๐ดฮฉ)๐ด๐ฅ = ๐ด(๐ธ โ ฮฉ๐ด)๐ฅ = 0. Since ๐ + ๐ต๐ถ๐ด๐ = (๐ธ + ๐ต๐
)๐ , we have ๐ธ โ ๐ดฮฉ = (๐ธ + ๐ต๐
)๐ . Thus (๐ธ + ๐ต๐
)๐ ๐ด๐ฅ = (๐ธ โ ๐ดฮฉ)๐ด๐ฅ = 0. Now use that ๐ธ + ๐ต๐
is injective. It follows that ๐ ๐ด๐ฅ = 0. We conclude that ๐ ๐ฅ = 0 implies that ๐ ๐ด๐ฅ = 0. By induction we obtain that ๐ ๐ด๐ ๐ฅ = 0 for ๐ = 0, 1, 2, . . .. In particular, using that the fourth equation in (1.4)โ is solvable, we get ๐ถ๐ด๐ ๐ฅ = 0 for ๐ = 0, 1, 2, . . .. Since ๐ฅ = ฮฉ๐ด๐ฅ, ๐ we get that ๐ฅ = ๐=1 ๐ด๐โ๐ ๐ต๐ถ๐ด๐ ๐ฅ = 0, and hence that ๐ is injective. Part 2. Next we assume ๐ผ + ๐ถ๐พ is injective, and we prove that ๐ is injective. Note that ๐ผ + ๐ถ๐พ = ๐ผ + ๐๐ ๐พ = ๐ผ + ๐๐ด๐ ๐ต is injective if and only if ๐ธ + ๐ด๐ ๐ต๐ is injective. As in the previous part, we assume that ๐ฅ โ Ker ๐ . Then we have
380
M.A. Kaashoek and F. van Schagen
that ๐ถ๐ด๐ ๐ฅ = ๐
๐ ๐ฅ = 0. Hence, (๐ธ โ ๐ดฮฉ)๐ฅ = (๐ + ๐ต๐ถ๐ด๐ )๐ฅ = 0. We see that ๐ฅ = ๐ดฮฉ๐ฅ. Next we show that ฮฉ๐ฅ โ Ker ๐ . From ๐ + ๐ด๐ ๐ต๐ถ = (๐ธ + ๐ด๐ ๐ต๐)๐ we see (๐ธ + ๐ด๐ ๐ต๐)๐ ฮฉ๐ฅ = (๐ธ โ ฮฉ๐ด)ฮฉ๐ฅ = ฮฉ(๐ธ โ ๐ดฮฉ)๐ฅ = 0. Since ๐ธ + ๐ด๐ ๐ต๐ is injective, we indeed have ๐ ฮฉ๐ฅ = 0. It follows that ๐ถ๐ด๐โ1 ๐ฅ = ๐ถ๐ด๐โ1 ๐ดฮฉ๐ฅ = ๐ถ๐ด๐ ฮฉ๐ฅ = ๐
๐ ฮฉ๐ฅ = 0. Replacing ๐ฅ by ฮฉ๐ฅ, we conclude that ๐ถ๐ด๐โ1 ฮฉ๐ฅ = 0. Again use that ๐ฅ = ๐ดฮฉ๐ฅ to conclude that ๐ถ๐ด๐โ2 ๐ฅ = 0. Proceeding in this way we get ๐ถ๐ด๐ ๐ฅ = 0 for ๐ = 0, 1, 2, . . .. As we have seen in the previous part, this yields ๐ฅ = 0. Thus ๐ is injective. Part 3. To prove the second statement, we assume ๐ผ + ๐ถ๐พ or ๐ผ + ๐
๐ต is surjective, and we prove that ๐ is surjective. To do this we apply the results of the ๏ฌrst statement to the algebraic dual ๐ # of ๐ . From (1.2) we see that ๐
#
=๐ธโ
๐ โ
# ๐
#
#
# ๐โ๐
(๐ด ) ๐ถ ๐ต (๐ด )
๐=0
=๐ธโ
๐ โ
(๐ด# )๐โ๐ ๐ถ # ๐ต # (๐ด# )๐ .
๐=0
Furthermore, the equations in (1.4) yield ๐พ # ๐ # = ๐ต # (๐ด# )๐ ,
๐ฟ# ๐ # = ๐ต # ,
๐ # ๐
# = (๐ด# )๐ ๐ถ # ,
๐ # ๐# = ๐ถ # .
Our hypotheses imply that ๐ผ + ๐พ # ๐ถ # or ๐ผ + ๐ต # ๐
# is injective. But then we can apply the ๏ฌrst statement of this lemma with ๐ # in place of ๐ , with ๐ด# in place of ๐ด, with ๐ต # in place of ๐ถ, with ๐พ # in place of ๐
, and with ๐ฟ# in place of ๐. It follows that ๐ # is injective, which is equivalent to ๐ being surjective. Part 4. Assume ๐ผ + ๐ถ๐พ or ๐ผ + ๐
๐ต is invertible. Then we know from the ๏ฌrst and second statement that ๐ is invertible. The identity ๐ + ๐ด๐ ๐ต๐ถ = ๐ (๐ธ + ๐พ๐ถ) shows that the invertibility of ๐ and ๐ผ + ๐ถ๐พ yield that ๐ + ๐ด๐ ๐ต๐ถ is invertible. Similarly, using ๐ + ๐ต๐ถ๐ด๐ = (๐ธ + ๐ต๐
)๐ , we see that if ๐ and ๐ผ + ๐
๐ต are invertible, then also ๐ + ๐ต๐ถ๐ด๐ is invertible. Here we use that ๐ผ + ๐ถ๐พ (or ๐ผ + ๐
๐ต) is invertible if and only if ๐ธ +๐พ๐ถ (or ๐ธ +๐ต๐
) is invertible. Recall that ๐ +๐ด๐ ๐ต๐ถ is invertible if and only if ๐ + ๐ต๐ถ๐ด๐ is invertible. Thus our hypotheses imply that ๐ , ๐ + ๐ด๐ ๐ต๐ถ and ๐ + ๐ต๐ถ๐ด๐ are all invertible. But then we see from ๐ + ๐ต๐ถ๐ด๐ = (๐ธ + ๐ต๐
)๐ and ๐ + ๐ด๐ ๐ต๐ถ = ๐ (๐ธ + ๐พ๐ถ) that both ๐ธ + ๐ต๐
and ๐ธ + ๐พ๐ถ are invertible. The latter is equivalent to ๐ผ + ๐ถ๐พ and ๐ผ + ๐
๐ต being invertible. โก Completing the proof of Theorem 1.1. Given Lemma 2.1 it remains to prove the ๏ฌnal statement of the theorem, that is, assuming ๐ , ๐ผ + ๐ถ๐พ and ๐ผ + ๐
๐ต are invertible, we have to derive the formula for ๐ โ1 in (1.5). From (1.2) it is clear that ๐ โ ๐ โ1 = ๐ธ + ๐ด๐โ๐ ๐ต๐ถ๐ด๐ ๐ โ1 . (2.2) ๐=0
In this formula we want to replace ๐ถ๐ด๐ ๐ โ1 for ๐ = 1, . . . , ๐.
On Inversion of Certain Structured Linear Transformations
381
From the ๏ฌrst and third identity in (1.4) we see that ๐ + ๐ด๐ ๐ต๐ถ = ๐ (๐ธ + ๐พ๐ถ), ๐ + ๐ต๐ถ๐ด๐ = (๐ธ + ๐ต๐
)๐. Using the ๏ฌrst identity in (2.1) the two previous formulas yield (๐ธ + ๐ต๐
)๐ ๐ด = (๐ + ๐ต๐ถ๐ด๐ )๐ด = ๐ด(๐ + ๐ด๐ ๐ต๐ถ)
(2.3)
= ๐ด๐ (๐ธ + ๐พ๐ถ). Since ๐ผ + ๐ถ๐พ and ๐ผ + ๐
๐ต are invertible, the same holds true ๐ธ + ๐พ๐ถ and ๐ธ + ๐
๐ต. Thus we can multiply (2.3) from the left by ๐ โ1 (๐ธ + ๐ต๐
)โ1 and from the right by (๐ธ + ๐พ๐ถ)โ1 ๐ โ1 . We obtain ๐ด(๐ธ + ๐พ๐ถ)โ1 ๐ โ1 = ๐ โ1 (๐ธ + ๐ต๐
)โ1 ๐ด. Thus 0 = โ๐ด(๐ธ + ๐พ๐ถ)โ1 ๐ โ1 + ๐ โ1(๐ธ + ๐ต๐
)โ1 ๐ด. By adding ๐ด๐ โ1 โ ๐ โ1 ๐ด to both sides of this equality we get ๐ด๐ โ1 โ ๐ โ1 ๐ด = ๐ด(๐ธ โ (๐ธ + ๐พ๐ถ)โ1 )๐ โ1 โ ๐ โ1 (๐ธ โ (๐ธ + ๐ต๐
)โ1 )๐ด, and therefore ๐ด๐ โ1 โ ๐ โ1 ๐ด = ๐ด๐พ(๐ผ + ๐ถ๐พ)โ1 ๐ถ๐ โ1 โ ๐ โ1 ๐ต(๐ผ + ๐
๐ต)โ1 ๐
๐ด. Now use the de๏ฌnitions of ๐ฟ, ๐ and ๐ป1 to get the identity ๐ด๐ โ1 โ ๐ โ1 ๐ด = ๐ด๐พ(๐ผ + ๐ถ๐พ)โ1 ๐ โ ๐ฟ(๐ผ + ๐
๐ต)โ1 ๐
๐ด = ๐ป1 . We will generalize this by induction to ๐ด๐ ๐ โ1 โ ๐ โ1 ๐ด๐ = ๐ป๐ , ๐ = 1, . . . , ๐, as follows: ๐ด๐ ๐ โ1 โ ๐ โ1 ๐ด๐ = ๐ด(๐ด๐โ1 ๐ โ1 ) โ ๐ โ1 ๐ด๐ = ๐ด(๐ โ1 ๐ด๐โ1 + ๐ป๐โ1 ) โ ๐ โ1 ๐ด๐ = (๐ด๐ โ1 )๐ด๐โ1 + ๐ด๐ป๐โ1 โ ๐ โ1 ๐ด๐ = (๐ โ1 ๐ด + ๐ป1 )๐ด๐โ1 + ๐ด๐ป๐โ1 โ ๐ โ1 ๐ด๐ = ๐ป๐ . Since ๐ถ๐ โ1 = ๐, we proved that ๐ถ๐ด๐ ๐ โ1 = ๐๐ด๐ + ๐ถ๐ป๐ . Inserting this in (2.2) completes the proof of Theorem 1.1. โก
3. Comments and an example Theorem 1.1 has the following corollary. Corollary 3.1. Assume there exist operators ๐พ and ๐ฟ from ๐ฐ into ๐ณ and operators ๐
and ๐ from ๐ณ into ๐ฐ satisfying the equations in (1.4), and let ๐ผ +๐ถ๐พ or ๐ผ +๐
๐ต
382
M.A. Kaashoek and F. van Schagen
be invertible. Then ๐ , ๐ผ + ๐ถ๐พ and ๐ผ + ๐
๐ต are invertible, and ๐ โ1 is given by ๐ โ ๐ด๐โ๐ ๐ต๐๐ด๐ + ๐ โ1 = ๐ธ + ๐=0
+ โ
๐ โ
๐ด๐โ๐ ๐ต
( ๐โ1 โ
๐=1
๐=0
๐ โ
( ๐โ1 โ
๐ด๐โ๐ ๐ต
๐ถ๐ด๐โ๐ ๐พ(๐ผ + ๐ถ๐พ)โ1 ๐๐ด๐
)
) ๐ถ๐ด๐โ1โ๐ ๐ฟ(๐ผ + ๐
๐ต)โ1 ๐
๐ด๐+1 .
๐=0
๐=1
โ๐โ1 Proof. By induction one shows that ๐ป๐ = ๐=0 ๐ด๐โ1โ๐ ๐ป1 ๐ด๐ for ๐ = 1, . . . , ๐, where ๐ป1 = ๐ด๐พ(๐ผ + ๐ถ๐พ)โ1 ๐ โ ๐ฟ(๐ผ + ๐
๐ต)โ1 ๐
๐ด. Using this in (1.5) yields the desired formula for ๐ โ1 . โก By applying the above corollary to the algebraic dual of ๐ one sees that the inverse of ๐ is also given by ๐ โ ๐ โ1 = ๐ธ + ๐ด๐ ๐ฟ๐ถ๐ด๐โ๐ + ๐=0
+
๐ ( ๐โ1 โ โ ๐=1
โ
๐=1
๐โ1 โ ( ๐โ1 โ ๐=1
) ๐ด๐ ๐ฟ(๐ผ + ๐
๐ต)โ1 ๐
๐ด๐โ๐ ๐ต ๐ถ๐ด๐โ๐ ) ๐ด๐+1 ๐พ(๐ผ + ๐ถ๐พ)โ1 ๐๐ด๐โ๐ ๐ต ๐ถ๐ด๐โ๐ .
๐=0
For ๐ = 1, solvability of the four equations in (1.4) directly implies that ๐ is invertible without any further conditions on ๐ผ + ๐ถ๐พ or ๐ผ + ๐
๐ต. Indeed, when ๐ = 1 we have ๐ = ๐ธ โ ๐ด๐ต๐ถ โ ๐ต๐ถ๐ด = ๐ธ โ ๐ ๐พ๐๐ โ ๐ ๐ฟ๐
๐. Hence ๐ (๐ธ + ๐พ๐๐ + ๐ฟ๐
๐ ) = ๐ธ and (๐ธ + ๐ ๐พ๐ + ๐ ๐ฟ๐
)๐ = ๐ธ, which proves that ๐ is invertible. In the proof of Theorem 1.1 injectivity and surjectivity of ๐ are established separately. If ๐ณ or ๐ฐ is ๏ฌnite dimensional, then the operator ๐ is a ๏ฌnite rank perturbation of the identity operator on ๐ณ . For such an operator ๐ one has that dim Ker ๐ = codim Im ๐ , and hence ๐ is injective if and only if ๐ is surjective. However, in general, even when the four equations in (1.4) are solvable, injectivity of ๐ is not equivalent to surjectivity of ๐ . In fact, this already happens for ๐ = 2, as the following example shows. Note that the case ๐ = 1 has to be excluded because of the result mentioned in the preceding paragraph. Example. Take ๐ = 2, and put ๐ฐ = โ2+ and ๐ณ = โ2 โ โ2+ . As before the identity operators on ๐ฐ and ๐ณ are denoted by ๐ผ and ๐ธ, respectively. Thus ๐ผ denotes the identity on โ2+ and ๐ธ stands for the identity on โ2 โ โ2+ . The identity operator
On Inversion of Certain Structured Linear Transformations
383
on โ2 will be denoted by ๐ผ2 . In general, 0 denotes a zero operator. The set {๐1 , ๐2 } denotes the standard basis of โ2 , and {๐1 , ๐2 , . . .} is the standard basis of โ2+ . The forward shift on โ2+ is denoted by ๐; thus ๐๐๐ = ๐๐+1 for ๐ = 1, 2, . . .. Note that the adjoint operator ๐ โ of ๐ is the backward shift on โ2+ , that is, ๐ โ ๐1 = 0 and ๐ โ ๐๐+1 = ๐๐ for ๐ = 1, 2, . . .. We de๏ฌne operators ๐ด, ๐ต, and ๐ถ as follows: ] [ ๐ด11 0 : โ2 โ โ2+ โ โ2 โ โ2+ , ๐ด= 0 ๐ผ [ ] ๐ต1 ๐ต= : โ2+ โ โ2 โ โ2+ , ๐โ ] [ ๐ถ = ๐ถ1 ๐ 2 : โ2 โ โ2+ โ โ2+ . Here ๐ด11 is the operator on โ2 de๏ฌned by ๐ด11 ๐1 = ๐2 , and ๐ด11 ๐2 = ๐1 , and ๐ต1 is the operator from โ2+ to โ2 given by ๐ต1 ๐1 = ๐2 and ๐ต1 ๐๐ = 0 for ๐ = 2, 3, . . .. Furthermore, ๐ถ1 is the operator from โ2 to โ2+ de๏ฌned by ๐ถ1 ๐1 = ๐2 and ๐ถ1 ๐2 = ๐1 . Finally, we set ๐ = ๐ธ โ ๐ด2 ๐ต๐ถ โ ๐ด๐ต๐ถ๐ด โ ๐ต๐ถ๐ด2 . (3.1) Since ๐ด211 = ๐ผ2 , we have ๐ด2 = ๐ธ, and hence ๐ = ๐ธ โ 2๐ต๐ถ โ ๐ด๐ต๐ถ๐ด. Next we write ๐ as a 2 ร 2 operator matrix relative to the direct sum decomposition โ2 โ โ2+ : [ ] ๐11 ๐12 ๐= ๐21 ๐22 ] [ ๐ผ2 โ 2๐ต1 ๐ถ1 โ ๐ด11 ๐ต1 ๐ถ1 ๐ด11 โ2๐ต1 ๐ 2 โ ๐ด11 ๐ต1 ๐ 2 . = โ2๐ โ ๐ถ1 โ ๐ โ ๐ถ1 ๐ด11 ๐ผ โ 2๐ โ ๐ 2 โ ๐ โ ๐ 2 One computes that ๐11 ๐1 = 0 and ๐11 ๐2 = โ๐2 . Since ๐ต1 ๐ 2 = 0, we have ๐12 = 0. The action of ๐21 is given by ๐21 ๐1 = โ2๐1 and ๐21 ๐2 = โ๐1 . Finally, since ๐ โ ๐ is the identity on โ2+ , we see that ๐22 = ๐ผ โ 3๐ โ ๐ 2 = ๐ผ โ 3๐. Now remark that Im ๐ โ span {๐2 } โ โ2+ . Thus ๐1 โโ Im ๐ , and hence ๐ is not surjective. We shall show that ๐ is injective and that the four equations (1.4) do have solutions. Note that the vectors ๐1 , ๐2 , ๐1 , ๐2 , ๐3 , . . . form an orthogonal basis of the Hilbert space โ2 โ โ2+ . We de๏ฌne ๐ to be the forward shift operator on โ2 โ โ2+ with respect to this basis. Thus the action of ๐ is given by ๐ ๐ 1 = ๐2 ,
๐ ๐2 = ๐ 1 ,
๐ ๐๐ = ๐๐+1
(๐ = 1, 2, . . .).
โ2+ ,
Note that Im ๐ is equal to span {๐2 } โ and hence Im ๐ is contained in Im ๐ . The adjoint of ๐ is the backward shift on โ2 โ โ2+ relative to the basis ๐1 , ๐2 , ๐1 , ๐2 , ๐3 , . . .. Thus ๐ โ ๐1 = 0,
๐ โ ๐2 = ๐ 1 ,
๐ โ ๐1 = ๐ 2 ,
๐ โ ๐๐ = ๐๐โ1
(๐ = 2, 3, . . .).
384
M.A. Kaashoek and F. van Schagen
Put ๐ = ๐ โ ๐ . We claim that ๐ is invertible. To see this we ๏ฌrst note that ๐ โ ๐ ๐1 = ๐ โ ๐11 ๐1 + ๐ โ ๐21 ๐1 = ๐ โ ๐21 ๐1 = โ2๐ โ ๐1 = โ2๐2 , ๐ โ ๐ ๐2 = ๐ โ ๐11 ๐2 + ๐ โ ๐21 ๐2 = โ๐ โ ๐2 โ ๐ โ ๐1 = โ๐1 โ ๐2 , ๐ โ ๐ ๐1 = ๐ โ ๐12 ๐1 + ๐ โ ๐22 ๐1 = ๐ โ ๐22 ๐1 = ๐ โ (๐1 โ 3๐2 ) = ๐2 โ 3๐1 , ๐ โ ๐ ๐๐ = ๐ โ ๐22 ๐๐ = ๐ โ (๐๐ โ 3๐๐+1 ) = ๐๐โ1 โ 3๐๐ ,
(๐ = 2, 3, . . .).
Summarizing we have ๐ ๐1 = โ 2๐2 , ๐ ๐1 = ๐2 โ 3๐1 ,
๐ ๐2 = โ๐1 โ ๐2 ,
(3.2)
๐ ๐๐+1 = ๐๐ โ 3๐๐+1
(๐ = 1, 2, 3, . . . ).
(3.3)
Now consider the 2 ร 2 operator matrix representation of ๐ relative to the direct sum decomposition โ2 โ โ2+ : [ ] ๐11 ๐12 ๐= : โ2 โ โ2+ โ โ2 โ โ2+ . ๐21 ๐22 From (3.2) we see that ๐ maps โ2 โ {0} in a one-to-one way onto โ2 โ {0}. Hence ๐11 is invertible and ๐21 = 0. The equalities in (3.3) show that ๐22 = ๐ โ โ 3๐ผ. As ๐ โ is a contraction, it follows that ๐22 is also invertible. Thus ๐ is block upper triangular and its diagonal blocks are invertible. So ๐ is invertible. Since ๐ = ๐ โ ๐ is invertible, ๐ is injective. It remains to prove that for our ๐ the four equations in (1.4) are solvable. Note that ๐ถ๐ โ1 ๐ โ ๐ = ๐ถ and hence ๐ = ๐
= ๐ถ๐ โ1 ๐ โ gives that ๐๐ = ๐ถ and ๐
๐ = ๐ถ๐ด2 , where for the last equality we used the fact that ๐ด2 = ๐ธ. From the de๏ฌnition of ๐ we see that ๐ โ ๐ is the identity operator on โ2 โ โ2+ , and ๐ ๐ โ is the orthogonal projection of โ2 โ โ2+ onto span {๐2 } โ โ2+ . Note that Im ๐ต is contained in span {๐2 } โ โ2+ . We already know that the same holds true for Im ๐ . Thus ๐ ๐ โ ๐ต = ๐ต and ๐ ๐ โ ๐ = ๐ . Now put ๐พ = ๐ฟ = ๐ โ1๐ โ ๐ต. Then ๐ ๐พ = ๐ ๐ โ ๐ ๐พ = ๐ ๐ ๐พ = ๐ ๐ ๐ โ1 ๐ โ ๐ต = ๐ ๐ โ ๐ต = ๐ต,
๐ ๐ฟ = ๐ต = ๐ด2 ๐ต.
For the ๏ฌnal equality we again use that ๐ด2 = ๐ธ. Thus the four equations in (1.4) have solutions. Summarizing we see that ๐ is injective, that the four equations (1.4) have solutions, but that ๐ is not surjective. โก The block Toeplitz matrix ๐ associated to the operator ๐ de๏ฌned by (3.1) is the 3ร3 block operator matrix given by โก โค ๐ผ โ๐ โ๐ โ๐ โข โฅ ๐ผ โ๐ โ๐ โฆ . ๐ = โฃ โ๐ โ๐
โ๐
๐ผ โ๐
On Inversion of Certain Structured Linear Transformations
385
Here, as before, ๐ is the forward shift on โ2+ and ๐ is the operator on โ2+ given by โค โก 1 0 0 0 โ
โ
โ
โข0 0 0 0 โ
โ
โ
โฅ โฅ โข โฅ โข ๐ = โข0 1 0 0 โ
โ
โ
โฅ โฅ โข0 0 1 0 โฆ โฃ .. .. .. .. . . . . Note that Im (๐ผ โ ๐ ) and Im ๐ are contained in span {๐2 , ๐3 , . . .}. Hence ๐ is not surjective, as one expects because ๐ is surjective if and only if ๐ is. Assume that ๐ณ or ๐ฐ is ๏ฌnite dimensional, and let the four equations in (1.4) be solvable. Then ๐ , ๐ผ + ๐
๐ต, ๐ผ + ๐ถ๐พ, ๐ธ + ๐ต๐
and ๐ธ + ๐พ๐ถ are all the sum of an identity operator and an operator of ๏ฌnite rank. For such an operator there exists a well-de๏ฌned determinant that has the usual properties (cf. [3], Sections VII.1 and VII.3). We claim that det(๐ผ + ๐ถ๐พ) = det(๐ผ + ๐
๐ต).
(3.4)
To see this we ๏ฌrst note that det(๐ผ + ๐ถ๐พ) = det(๐ธ + ๐พ๐ถ),
det(๐ผ + ๐
๐ต) = det(๐ธ + ๐ต๐
).
(3.5)
Next, observe that det(๐ธ + ๐ต๐
) det ๐ = det(๐ + ๐ต๐
๐ ) = det(๐ + ๐ต๐ถ๐ด๐ ) = det(๐ธ โ ๐ดฮฉ) = det(๐ธ โ ฮฉ๐ด) = det(๐ + ๐ด๐ ๐ต๐ถ) = det(๐ + ๐ ๐พ๐ถ) = det ๐ det(๐ธ + ๐พ๐ถ). If det ๐ โ= 0, then the above calculation shows that det(๐ธ + ๐พ๐ถ) = det(๐ธ + ๐ต๐
), and hence, by (3.5), the identity (3.4) holds. On the other hand, if det ๐ = 0, then ๐ is not invertible, and we know from Theorem 1.1 that neither ๐ผ + ๐ถ๐พ nor ๐ผ + ๐
๐ต is invertible. In other words, both det(๐ผ + ๐ถ๐พ) and det(๐ผ + ๐
๐ต) are zero, and (3.4) is trivially satis๏ฌed. In the case when dim ๐ฐ = 1, the identity (3.4) recovers the fact that the left upper element and the right lower element of the inverse of a scalar Toeplitz matrix are equal (cf. [5] or Section III.6 in [1]).
References [1] I.C. Gohberg, I.A. Felโdman, Convolution equations and projection methods for their solution, Transl. Math. Monographs Vol. 41, Amer. Math. Soc., Providence, R.I., 1974. [2] I. Gohberg, G. Heinig, The inversion of ๏ฌnite Toeplitz matrices consisting of elements of a non-commutative algebra, Rev. Roum. Math. Pures et Appl. 20 (1974), 623โ 663 (in Russian); English transl. in: Convolution Equations and Singular Integral
386
[3] [4] [5] [6]
M.A. Kaashoek and F. van Schagen Operators, (eds. L. Lerer, V. Olshevsky, I.M. Spitkovsky), OT 206, Birkhยจ auser Verlag, Basel, 2010, pp. 7โ46. I. Gohberg, S. Goldberg, M.A. Kaashoek, Classes of Linear Operators, Volume I, Birkhยจ auser Verlag, Basel, 1990. I. Gohberg, M.A. Kaashoek, F. van Schagen, On inversion of Toeplitz matrices with elements in an algebraic ring, Lin. Alg. Appl. 385 (2004), 381โ389. I. Gohberg, A.A. Semencul, On the invertibility of ๏ฌnite Toeplitz matrices and their continuous analogues, Matem. Issled 7(2), Kishinev (1972), (in Russian). L. Lerer, V. Olshevsky, I.M. Spitkovsky (Eds), Convolution Equations and Singular Integral Operators, OT 206, Birkhยจ auser Verlag, Basel, 2010.
M.A. Kaashoek and F. van Schagen Afdeling Wiskunde, Faculteit der Exacte Wetenschappen VU Universiteit Amsterdam De Boelelaan 1081a, NL-1081 HV Amsterdam, The Netherlands e-mail:
[email protected] [email protected]
Operator Theory: Advances and Applications, Vol. 218, 387โ401 c 2012 Springer Basel AG โ
The Inverse of a Two-level Positive De๏ฌnite Toeplitz Operator Matrix Selcuk Koyuncu and Hugo J. Woerdeman To the memory of Israel Gohberg, an excellent mathematician and an inspiring teacher
Abstract. The Gohberg-Semencul formula allows one to express the entries of the inverse of a Toeplitz matrix using only a few entries (the ๏ฌrst row and the ๏ฌrst column) of the matrix, under some nonsingularity condition. In this paper we will provide a two variable generalization of the GohbergSemencul formula in the case of a positive de๏ฌnite two-level Toeplitz matrix with a symbol of the form โฃ๐โฃ1 2 where ๐ is a stable polynomial of two variables. We also consider the case of operator-valued two-level Toeplitz matrices. In addition, we propose an approximation of the inverse of a multilevel Toeplitz matrix with a positive symbol, and use it as the initial value for a Hotelling iteration to compute the inverse. Numerical results are included. Mathematics Subject Classi๏ฌcation (2000). 15A09 (47B35, 65F30). Keywords. Two-level Toeplitz matrices, stable polynomial, inverse formula, Gohberg-Semencul expressions, Discrete Algebraic Riccati Equation.
1. Introduction Important in the development of computational and theoretical results involving Toeplitz matrices was the Gohberg-Semencul formula which expresses the inverse of Toeplitz ๐ in terms of the ๏ฌrst column and row of ๐ โ1 . The impact of this formula on the ๏ฌeld of structured matrices and numerical algorithms was systematically presented in a book by G. Heinig and K. Rost [4]. Nontrivial generalization to block Toeplitz matrices is the Gohberg-Heinig formula [2]. For the classical onevariable positive de๏ฌnite case the Gohberg-Semencul formula [3] is the following: This research is supported by NSF grant DMS-0901628.
388
S. Koyuncu and H.J. Woerdeman
the inverse of (๐ก๐โ๐ )๐โ1 ๐,๐=0 equals โก โคโก โค โก ๐0 ๐0 . . . ๐๐โ1 ๐๐ โข .. โฅ โข โฅ โข . . . .. .. .. โฆ โ โฃ ... โฃ . โฆโฃ ๐๐โ1 . . . ๐0 ๐0 ๐1 where ๐(๐ง) =
๐ โ
..
. . . . ๐๐
โคโก ๐๐ โฅโข โฆโฃ
... .. .
โค ๐1 .. โฅ , .โฆ
๐๐
๐๐ ๐ง ๐
๐=0
(๐ก๐โ๐ )๐๐,๐=0 (๐๐ )๐๐=0
1 ๐ยฏ0 ๐1 ,
satis๏ฌes = where ๐1 = (1, 0, 0, . . . , 0)๐ . In this paper we consider two-level Toeplitz matrices, which in special cases are block Toeplitz matrices with Toeplitz blocks. We will provide a two variable generalization of the Gohberg-Semencul formula in the case of positive de๏ฌnite two-level Toeplitz matrix with a symbol of the form ๐ (๐ง1 , ๐ง2 ) = โฃ๐ (๐ง11,๐ง2 )โฃ2 where โ๐1 โ๐2 ๐ ๐ ๐ (๐ง1 , ๐ง2 ) = ๐=0 ๐=0 ๐๐๐ ๐ง1 ๐ง2 is a stable polynomial of two variables, i.e., ๐ (๐ง1 , ๐ง2 ) โ= 0 for โฃ๐ง1 โฃ โค 1,โฃ๐ง2 โฃ โค 1. We de๏ฌne a two-level Toeplitz matrix to be a matrix of the form ๐ = (๐กkโl )k,lโฮ where ฮ is a ๏ฌnite subset of โ20 . For instance, when ฮ = {0, 1} ร {0, 1} which we will order lexicographically, ฮ = {(0, 0), (0, 1), (1, 0), (1, 1)}, we get
โก ๐ก0,0 โข๐ก0,1 ๐ =โข โฃ๐ก1,0 ๐ก1,1
๐ก0,โ1 ๐ก0,0 ๐ก1,โ1 ๐ก1,0
๐กโ1,0 ๐กโ1,1 ๐ก0,0 ๐ก0,1
โค ๐กโ1,โ1 ๐กโ1,0 โฅ โฅ ๐ก0,โ1 โฆ ๐ก0,0
(1.1)
In this paper we obtain the following two-variable generalization of the classical Gohberg-Semencul formula. We ๏ฌrst need to introduce some notation. For ๐ = (๐1 , ๐2 ) and ๐ง = (๐ง1 , ๐ง2 ) we let ๐ง ๐ = ๐ง1๐1 ๐ง2๐2 . If ๐ = (๐1 , ๐2 ), we let ๐ denote the set ๐ = ๐1 ร ๐2 , where ๐๐ = {0, . . . , ๐๐ }. Note that ๐ = (๐ก๐โ๐ )๐,๐โ๐ is a block Toeplitz matrix where each of the blocks are Toeplitz; as for instance in (1.1). Finally, we denote ๐ = {๐ง โ โ : โฃ๐งโฃ = 1} and ๐ป = {๐ง โ โ : โฃ๐งโฃ < 1}. Recall that the Loewner order on Hermitian matrices is de๏ฌned via ๐ โค ๐ โโ ๐ โ ๐ โฅ 0, i.e., ๐ โ ๐ is positive semide๏ฌnite. Theorem 1.1. Let ๐ (๐ง1 , ๐ง2 ) =
๐1 โ ๐2 โ
๐๐๐ ๐ง1 ๐ ๐ง2 ๐
and
๐
(๐ง1 , ๐ง2 ) =
๐=0 ๐=0
๐1 โ ๐2 โ ๐=0 ๐=0
be stable operator-valued polynomials, and suppose that โ
โ
๐ (๐ง1 , ๐ง2 )๐ (๐ง1 , ๐ง2 ) = ๐
(๐ง1 , ๐ง2 ) ๐
(๐ง1 , ๐ง2 ).
๐
๐๐ ๐ง1 ๐ ๐ง2 ๐
The Inverse of a Toeplitz Operator Matrix
389
Put ๐ (๐ง1 , ๐ง2 ) = ๐ (๐ง1 , ๐ง2 )โ
โ1
๐ (๐ง1 , ๐ง2 )โ1
= ๐
(๐ง1 , ๐ง2 )โ1 ๐
(๐ง1 , ๐ง2 )โ
โ1
for ๐ง1 , ๐ง2 โ ๐. Put ฮ = ๐ โ {๐}, where ๐ = (๐1 , ๐2 ) and write the Fourier coe๏ฌห ๐), (๐, ๐) โ โค2 . Consider cients of ๐ (๐ง1 , ๐ง2 ) as ๐(๐, ๐ = (๐ห๐1 โ๐2 ,๐1 โ๐2 )(๐1 ,๐1 ),(๐2 ,๐2 )โฮ . Then
๐ โ1 = ๐ด๐ดโ โ ๐ต โ ๐ต โ ๐ถ1โ ๐ท1 โ1 ๐ถ1 โ ๐ถ2โ ๐ท2 โ1 ๐ถ2 ,
(1.2)
where ๐ด = (๐๐โ๐ )๐,๐โฮ ,
๐ต = (๐
๐โ๐ )๐โ๐+ฮ ,
(1.3)
๐โฮ
and ๐ถ1 ,๐ท1 ,๐ถ2 and ๐ท2 are de๏ฌned via (๐ถ1 )๐๐ =
๐1 โ
min{๐2 ,๐2 }
๐1 =๐1 โ๐1
๐2 =0
โ
โ ๐๐โ๐ ๐๐โ๐ โ
๐1โ +๐1 min{๐2 +๐ 2 ,๐2 +๐2 } โ ๐1 =๐1
โ ๐
๐โ๐ ๐
๐โ๐ ,
(1.4)
๐2 =๐2
where ๐ โ ฮ1 = {๐1 + 1, ๐1 + 2, . . .} ร {0, 1, . . . , ๐2 โ 1}, ๐ โ ๐1 ร ๐2 โ {(๐1 , ๐2 )}, (๐ถ2 )๐๐ =
min{๐1 ,๐1 }
โ
๐2 โ
๐1 =0
๐2 =๐2 โ๐2
โ ๐๐โ๐ ๐๐โ๐ โ
min{๐1 +๐1 ,๐1 +๐1 } ๐2 +๐2
โ
โ
๐1 =๐1
๐2 =๐2
โ ๐
๐โ๐ ๐
๐โ๐ ,
(1.5)
where ๐ โ ฮ2 = {0, 1, . . . , ๐1 โ 1} ร {๐2 + 1, ๐2 + 2, . . .} and ๐ โ ๐1 ร ๐2 โ {(๐1 , ๐2 )}, (๐ท1 )๐,๐ห =
min{๐1 ,๐ห1 }
โ
min{๐2 ,๐ห2 }
๐1 =max{๐1 ,๐ห1 }โ๐1
๐2 =0
โ
โ ๐๐โ๐ ๐๐โ๐ ห
min{๐1 ,๐ห1 }+๐1 min{๐2 ,๐ห2 }+๐2
โ
โ
๐ 1 =max{๐1 ,๐ห1 }
๐ 2 =๐2
โ
โ ๐
๐ โ๐ ๐
๐ โ๐ห ,
(1.6)
where ๐, ๐ห โ ฮ1 = {๐1 + 1, ๐1 + 2, . . .} ร {0, 1, . . . , ๐2 โ 1}, and min{๐1 ,๐ห1 }
โ
min{๐2 ,๐ห2 }
๐1 =0
๐2 =max{๐2 ,๐ห2 }โ๐2
(๐ท2 )๐,๐ห =
โ
โ ๐๐โ๐ ๐๐โ๐ ห
min{๐1 ,๐ห1 }+๐1 min{๐2 ,๐ห2 }+๐2
โ
โ
โ
๐ 1 =๐1
๐ 2 =max{๐2 ,๐ห2 }
โ ๐
๐ โ๐ ๐
๐ โ๐ห
(1.7)
where ๐, ๐ห โ ฮ2 = {0, 1, . . . , ๐1 โ 1} ร {๐2 + 1, ๐2 + 2, . . .} and ๐๐ = ๐
๐ = 0 whenever ๐ โโ ๐.
390
S. Koyuncu and H.J. Woerdeman
Thus to compute ๐ โ1 , we have reduced it to computing the inverses of ๐ท1 and ๐ท2 where ๐ท1 and ๐ท2 are traditional matrices. Typically, we โ๐one-level โ๐2 Toeplitz 1 ๐ ๐ would like to use it when ๐ (๐ง1 , ๐ง2 ) = ๐=0 ๐=0 ๐๐๐ ๐ง1 ๐ง2 is in fact a polynomial of degree (๐1 , ๐2 ) where ๐1 โช ๐1 and ๐2 โช ๐2 . In that case, ๐ด, ๐ต, ๐ถ1 ,๐ถ2 ,๐ท1 and ๐ท2 are sparse. Let us start illustrating Theorem 1.1 by giving the following example. Example. Let ๐1 = ๐2 = 2. Given ๐ (๐ง1 , ๐ง2 ) = ๐
(๐ง1 , ๐ง2 ) = ๐00 + ๐01 ๐ง2 + ๐10 ๐ง1 + ๐02 ๐ง22 + ๐20 ๐ง12 where ๐00 = 32 , ๐01 = 13 , ๐02 = 12 , ๐20 = 12 ,๐10 = 13 and ฮ = {0, 1, 2} ร {0, 1, 2} โ {(2, 2)}. In this case the matrices ๐ด, ๐ต, ๐ถ1 , ๐ถ2 , ๐ท1 and ๐ท2 are the following: โก3 2 โข1 โข 31 โข โข 21 โข โข3
0 3 2 1 3
0
0 0 3 2
1 3
0 0
1 2
0 0
0 0 0 0 0 .. .
0 0 0 0 0 .. .
๐ด=โข โข0 โข0 โข โฃ1 2 0
0 0
โก 0 โข0 โข โข0 โข ๐ถ1 = โข0 โข โข0 โฃ. ..
1 3
0 0 0 3 2 1 3 1 2 1 3
0 3 4
0 0 0 0 .. .
0 0 0 0 3 2 1 3
0 1 3
1 6 3 4
0 0 0 .. .
0 0 0 0 0 3 2
0 0 0 0 0 0 0 .. .
0 0 0 0 0 0 3 2 1 3
2 3
0 3 4
0 0 .. .
โค 0 0โฅ โฅ 0โฅ โฅ 0โฅ โฅ, 0โฅ โฅ 0โฅ โฅ 0โฆ
โก 0 โข0 โข โข0 โข โข0 ๐ต=โข โข0 โข โข0 โข โฃ0 0
โค
โก 0 โข0 โข โข0 โข ๐ถ2 = โข0 โข โข0 โฃ. ..
3 2
1 9 2โฅ 3โฅ 1โฅ 6โฅ 3โฅ, 4โฅ
0โฅ .. โฆ .
1 2
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0
3 4
2 3
0 0 0 0 .. .
0 3 4
0 0 .. .
0 0 0 0 0 0 0 0 0 0 0 0 0 .. .
0 0 0 0 0 0 0 0 1 6 3 4
0 0 0 .. .
1 3
0 0 0 0 0 0 0 1 9 2 3 1 6 3 4
0 .. .
1 2
0 0 0 0 0 0 0 0 0 0 0 0 .. .
1โค 3 1โฅ 2โฅ
0โฅ โฅ 0โฅ โฅ, 0โฅ โฅ 0โฅ โฅ 0โฆ 0 โค 0 0โฅ โฅ 0โฅ โฅ , 0โฅ โฅ 0โฅ .. โฆ .
and the top left 8 ร 8 block of the in๏ฌnite block Toeplitz matrices ๐ท1 and ๐ท2 equals to โก 13 โค 1 2 3 2 36 0 0 0 0 3 3 4 1 2 1 3 โข 1 2 13 0 0 โฅ 36 9 3 6 4 โข 32 โฅ 1 13 1 2 3 โข 2 36 0 0 โฅ 9 3 3 4 โข 3 โฅ 2 1 13 1 2 1 3 โข 0 โฅ 2 36 3 3 9 3 6 4 โฅ. โข 3 1 2 1 13 1 2 โข โฅ 2 0 6 3 9 36 3 3 โข 4 โฅ 3 2 1 13 1 2 โฅ โข 0 0 2 36 โข 4 3 3 9 3 โฅ 3 1 2 1 1 โฆ โฃ 0 0 2 13 4 6 3 9 36 3 3 2 1 13 0 0 0 0 2 4 3 3 36
The two-level Toeplitz matrix ๐ of size 8 ร 8 is following. โก
0.6453 โขโ0.1158 โข โขโ0.2241 โข โขโ0.1158 ๐ =โข โข 0.0490 โข โข 0.0674 โข โฃโ0.2241 0.0674
โค โ0.1158 โ0.2241 โ0.1158 0.0490 0.0674 โ0.2241 0.0674 0.6453 โ0.1158 0.0037 โ0.1158 0.0490 0.0304 โ0.2241โฅ โฅ โ0.1158 0.6453 0.0304 0.0037 โ0.1158 0.0839 0.0304 โฅ โฅ 0.0037 0.0304 0.6453 โ0.1158 โ0.2241 โ0.1158 0.0490 โฅ โฅ. โ0.1158 0.0037 โ0.1158 0.6453 โ0.1158 0.0037 โ0.1158โฅ โฅ 0.0490 โ0.1158 โ0.2241 โ0.1158 0.6453 0.0304 0.0037 โฅ โฅ 0.0304 0.0839 โ0.1158 0.0037 0.0304 0.6453 โ0.1158โฆ โ0.2241 0.0304 0.0490 โ0.1158 0.0037 โ0.1158 0.6453
The Inverse of a Toeplitz Operator Matrix
391
2. Proof of the main result To prove Theorem 1.1 we ๏ฌrst recall the following auxiliary results from [8]. Lemma 2.1. Assume that the operator matrix (๐ด๐๐ )2 ๐,๐=1 : ๐ป1 โ ๐ป2 โ ๐ป1 โ ๐ป2 and the operator ๐ด22 are invertible. Then ๐ = ๐ด11 โ ๐ด12 ๐ดโ1 22 ๐ด21 is invertible and ] ]โ1 [ โ1 [ ๐ โ ๐ด11 ๐ด12 = . (2.1) โ โ ๐ด21 ๐ด22 Proof. Follows directly from the factorization ][ [ ] [ ๐ด11 0 ๐ด11 โ ๐ด12 ๐ดโ1 ๐ผ โ๐ด12 ๐ดโ1 22 ๐ด21 22 = ๐ด21 0 ๐ด22 0 ๐ผ
๐ด12 ๐ด22
][
๐ผ
โ๐ดโ1 22 ๐ด21
] 0 . (2.2) ๐ผ โก
Lemma 2.2. Let lower/upper and upper/lower factorization of the inverse of a block matrix be given,as follows: [ ]โ1 [ ][ ] ๐ต11 ๐ต12 0 ๐11 ๐11 ๐12 = (2.3) ๐21 ๐22 ๐ต21 ๐ต22 0 ๐22 [ ][ ] ๐
11 ๐
12 ๐11 0 = , (2.4) 0 ๐
22 ๐21 ๐22 and suppose that ๐
22 and ๐22 are invertible. Then โ1 ๐ต11 = ๐11 ๐11 โ ๐
12 ๐21 .
(2.5)
Proof. Apply Lemma 2.1 with ๐ด11 = ๐11 ๐11 , ๐ด12 = ๐
12 ๐22 , ๐ด21 = ๐
22 ๐21 , ๐ด22 = ๐
22 ๐22 to equality ]โ1 [ ] [ ๐11 ๐11 ๐
12 ๐22 ๐ต11 ๐ต12 . โก = ๐ต21 ๐ต22 ๐
22 ๐21 ๐
22 ๐22 Corollary 2.3. Consider a positive de๏ฌnite operator matrix (๐ต๐๐ )3 ๐,๐=1 of which the lower/upper and upper/lower block Cholesky factorization of its inverse are given,as follows: โคโก โ โค โก โ โ ๐11 ๐21 ๐11 0 0 ๐31 โ โ โฆ 0 โฆ โฃ 0 ๐22 ๐32 (2.6) [(๐ต๐๐ )3 ๐,๐=1 ]โ1 = โฃ๐21 ๐22 โ 0 0 ๐33 ๐31 ๐32 ๐33 โก โ โคโก โค โ โ ๐
11 ๐
21 ๐
31 0 0 ๐
11 โ โ โฆโฃ ๐
21 ๐
22 ๐
22 ๐
32 0 โฆ, =โฃ 0 (2.7) โ ๐
31 ๐
32 ๐
33 0 0 ๐
33 with ๐
22 ,๐22 ,๐33 and ๐
33 invertible. Then โ1 โ โ โ = ๐11 ๐11 โ ๐
21 ๐
21 โ ๐
31 ๐
31 ๐ต11
(2.8)
392
S. Koyuncu and H.J. Woerdeman
Proof. By Lemma 2.2 we have that โ1 ๐ต11
=
โ ๐11 ๐11
[
โ ๐
21
โ
โ ๐
31
[ ] ] ๐
21 ๐
31
which gives (2.8).
โก
Before we prove the Theorem 1.1, we need to introduce some notation. Let โ be a Hilbert space and let โฌ(โ) denote the Banach space of bounded linear operators on โ. We let ๐ฟโ = ๐ฟโ (๐2 ; โฌ(โ)) denote the Lebesgue space of essentially bounded โฌ(โ)-valued measurable functions on ๐2 , and we let ๐ฟ2 = ๐ฟ2 (๐2 ; โ) and ๐ป2 = ๐ป2 (๐2 ; โ) denote the Lebesgue and Hardy space of square integrable 2 โ-valued functions โ on ๐ , ๐respectively. As usual we view ๐ป2 as a subspace of ๐ฟ2 . For ๐ฟ(๐ง) = ๐โโค2 ๐ฟ๐ ๐ง โ ๐ฟโ we will consider its multiplication operator ๐๐ฟ : ๐ฟ2 โ ๐ฟ2 given by (๐๐ฟ (๐ ))(๐ง) = ๐ฟ(๐ง)๐ (๐ง). The Toeplitz operator ๐๐ฟ : ๐ป2 โ ๐ป2 is de๏ฌned as the compressionโof ๐๐ฟ to ๐ป2 . ๐ For ฮ โ โค2 we let ๐ฮ denote the subspace {๐น โ ๐ฟ2 : ๐น (๐ง) = ๐โฮ ๐น๐ ๐ง } of ๐ฟ2 consisting of those functions with Fourier support in ฮ. In addition,we let ๐ฮ denote the orthogonal projection onto ๐ฮ . So, for instance, ๐โ20 is the orthogonal projection onto ๐ป2 and ๐๐ฟ = ๐โ20 ๐๐ฟ ๐โโ2 . 0
Proof of Theorem 1.1. Clearly we have that ๐๐ โ1 = ๐๐ ๐๐ โ = ๐๐
โ ๐๐
. With respect to the decomposition ๐ฟ2 = ๐ป2 โฅ โ ๐ป2 we get that [ ] [ ] [ ] โ โ โ 0 โ 0 ๐๐ = , ๐๐ = , ๐๐ โ1 = , (2.9) โ ๐๐ โ ๐๐ โ ๐๐ โ1 [ ] [ ] โ 0 โ 0 ๐๐
= , ๐๐
โ1 = , (2.10) โ ๐๐
โ ๐๐
โ1 where we used that ๐๐ ยฑ1 [๐ป2 ] โ ๐ป2 and ๐๐
ยฑ1 [๐ป2 ] โ ๐ป2 which follows as ๐ ยฑ1 โ1 and ๐
ยฑ1 are analytic in ๐ป2 . It now follows that ๐๐ = (๐๐ )โ (๐๐ )โ1 and thus ๐๐ โ1 = ๐๐ ๐๐ โ .
(2.11)
Next, decompose ๐ป2 = ๐ฮ โ ๐ฮ โ ๐๐+โ20 , where ฮ = ๐1 ร ๐2 โ {(๐1 , ๐2 )} and ฮ = โ20 โ(ฮ โช (๐ + โ20 )), and write ๐๐ and ๐๐
with respect to this decomposition: โก โค โก โค ๐11 ๐
11 โฆ , ๐๐
= โฃ๐
21 ๐
22 โฆ. ๐๐ = โฃ๐21 ๐22 (2.12) ๐31 ๐32 ๐33 ๐
31 ๐
32 ๐
33 As the Fourier support of ๐ and ๐
lies in ๐, and as ๐ (๐ง)๐ (๐ง)โ = ๐
(๐ง)โ ๐
(๐ง) on ๐2 , it is not hard to show that โ โ ๐๐ ๐๐ โ ๐๐+โ 2 = ๐๐
โ ๐๐
๐๐+โ2 , 0
0
(2.13)
The Inverse of a Toeplitz Operator Matrix
393
which yields that โ โ โ โ ๐31 ๐31 + ๐32 ๐32 + ๐33 ๐33 = ๐
33 ๐
33 ,
โ โ โ โ โ ๐21 ๐31 + ๐22 ๐32 = ๐
32 ๐
33 , ๐11 ๐31 = ๐
31 ๐
11 .
Thus we can factor ๐๐ ๐๐ โ as
๐๐ ๐๐ โ
โก โ ห ๐
11 โฃ =
หโ ๐
21 โ ห 22 ๐
โคโก โ ห 11 ๐
31 ๐
โ โฆโฃ ห ๐
32 ๐
21 โ ๐
33 ๐
31
โค ห 22 ๐
๐
32
๐
33
โฆ,
(2.14)
ห 11 , ๐
ห 21 and ๐
ห 22 . Combining now (2.14) and two factorization of ๐๐ ๐๐ โ for some ๐
โ โ โ หโ ๐
ห หโ ห given via (2.11), we get ๐
22 21 = [ ๐]21 ๐11 โ ๐
32 ๐
31 and ๐
22 ๐
22 = ๐21 ๐21 + ๐ถ1 โ โ หโ ๐
ห หโ ห ๐22 ๐22 โ ๐
32 ๐
32 . Now, we write = ๐
22 21 where ๐ถ1 = ๐ฮ1 ๐
22 ๐
21 and ๐ถ2 หโ ๐
ห ๐ถ2 = ๐ฮ2 ๐
22 21 . We will start only proving (1.4). The proof of (1.5) is similar. To prove (1.4), let ๐ โ ฮ1 = {๐1 + 1, ๐1 + 2, . . .} ร {0, 1, . . . , ๐2 โ 1}, ๐ โ ฮ. Since ๐๐ = ๐
๐ = 0 โ โ when ๐ โโ ๐ = ๐1 ร ๐2 , we get from ๐ถ1 = ๐ฮ1 (๐21 ๐11 โ ๐
32 ๐
31 ) that โ โ โ โ (๐ถ1 )๐๐ = ๐๐โ๐ ๐๐โ๐ โ ๐
๐โ๐ ๐
๐โ๐ . ๐โฮ ๐โ๐โ๐1 ร๐2 ๐โ๐โ๐1 ร๐2
๐โ๐+โ20 ๐โ๐โ๐1 ร๐2 ๐โ๐โ๐1 ร๐2
Note that ๐โ๐ โ ๐1 ร๐2 and ๐โ๐ โ ๐1 ร๐2 imply 0 โค ๐1 โ๐1 โค ๐1 , 0 โค ๐1 โ๐1 โค ๐1 , 0 โค ๐2 โ ๐2 โค ๐2 and 0 โค ๐2 โ ๐2 โค ๐2 . Combining these inequalities we get ๐1 โ ๐1 โค ๐1 โค ๐1 and 0 โค ๐2 โค min{๐2 , ๐2 }. Similarly, since ๐ โ ๐ โ ๐1 ร ๐2 and ๐ โ ๐ โ ๐1 ร ๐2 we get ๐1 โค ๐1 โค ๐1 + ๐1 and ๐2 โค ๐2 โค min{๐2 + ๐2 , ๐2 + ๐2 }. Thus the ๐, ๐th entry of ๐ถ1 equals (๐ถ1 )๐๐ =
๐1 โ
min{๐2 ,๐2 }
๐1 =๐1 โ๐1
๐2 =0
โ
โ ๐๐โ๐ ๐๐โ๐ โ
๐1 +๐1 min{๐2 +๐2 ,๐2 +๐2 }
โ
โ
๐1 =๐1
๐2 =๐2
โ ๐
๐โ๐ ๐
๐โ๐ .
This proves (1.4). หโ ๐
ห = ๐21 ๐ โ + ๐22 ๐ โ โ ๐
โ ๐
32 . Next, we need to compute ๐
21 22 32 [ ] 22 22 ๐ท ๐ธ 1 โ โ ห ๐
ห ห ๐
ห Write ๐
, where ๐ท๐ = ๐ฮ๐ ๐
22 22 = ๐ธ โ 22 22 ๐ฮ๐ , ๐ = 1, 2, and ๐ธ = ๐ท2 โ ห ห 22 ๐ฮ1 ๐
๐
22 ๐ฮ2 . We ๏ฌrst show that ๐ธ = 0. Let ๐ โ ฮ1 = {๐1 + 1, . . .} ร {0, . . . , ๐2 โ 1}, ๐ โ ฮ2 = {0, . . . , ๐1 โ 1} ร {๐2 + 1, . . .}. Note that โ โ โ โ (๐21 ๐21 + ๐22 ๐22 )๐๐ = ๐๐โ๐ ๐๐โ๐ . (2.15) ๐โฮโชฮ ๐โ๐โ๐ ๐โ๐โ๐
394
S. Koyuncu and H.J. Woerdeman
As ๐ โ (๐ โ ๐) โฉ (๐ โ ๐) โฉ (ฮ โช ฮ) is equivalent to ๐1 โ ๐1 โค ๐1 โค ๐1 and ๐2 โ ๐2 โค ๐2 โค ๐2 , we obtain ๐1 โ
โ โ (๐21 ๐21 + ๐22 ๐22 )๐๐ =
๐2 โ
โ ๐๐โ๐ ๐๐โ๐ .
๐1 =๐1 โ๐1 ๐2 =๐2 โ๐2
Next,
โ
โ (๐
32 ๐
32 )๐๐ =
โ ๐
๐โ๐ ๐
๐โ๐ .
(2.16)
๐โ๐+โ20 ๐โ๐โ๐ ๐โ๐โ๐
As ๐ โ (๐ + ๐) โฉ (๐ + ๐) โฉ (๐ + โ20 ) is equivalent to ๐1 โค ๐1 โค ๐1 + ๐1 and ๐2 โค ๐2 โค ๐2 + ๐2 , we obtain โ (๐
32 ๐
32 )๐๐ =
๐1 +๐1 ๐2 +๐2 โ โ
โ ๐
๐โ๐ ๐
๐โ๐ .
๐1 =๐1 ๐2 =๐2 โ โ โ Finally, we need to show that (๐21 ๐21 + ๐22 ๐22 )๐๐ = (๐
32 ๐
32 )๐๐ . It is clear that if ๐1 โ ๐1 > ๐1 or ๐2 โ ๐2 > ๐2 then equality holds as both sides equal 0. Now let us consider the case when ๐1 โ ๐1 โค ๐1 and ๐2 โ ๐2 โค ๐2 . Let ๐ = (๐1 + ๐, ๐ ) โ ฮ1 and ๐ = (ห ๐ , ๐2 + ๐ ห) โ ฮ2 where ๐, ๐ ห โฅ 1, ๐ โ {0, . . . , ๐2 โ 1} and ๐ห โ {0, . . . , ๐1 โ 1}. โ โ Using the fact that ๐ (๐ง)๐ (๐ง) = ๐
(๐ง) ๐
(๐ง) we have ๐ห โ ๐ โ
๐๐1 +๐โ๐1 ,๐ โ๐2 ๐๐หโโ๐1 ,๐2 +ห๐ โ๐2
๐1 =๐ ๐2 =ห ๐ ๐ห โ ๐ โ
=
๐
๐โหโ๐1 ,๐2 +ห๐ โ๐2 ๐
๐1 +๐โ๐1 ,๐ โ๐2 .
(2.17)
๐1 =๐ ๐2 =ห ๐
Substituting ๐ = ๐1 โ ๐1 , ๐ห = ๐1 , ๐ ห = ๐2 โ ๐2 and ๐ = ๐2 into (2.17) we obtain ๐1 โ
๐2 โ
๐๐1 โ๐1 ,๐2 โ๐2 ๐๐โ1 โ๐1 ,๐2 โ๐2
๐1 =๐1 โ๐1 ๐2 =๐2 โ๐2 ๐1 โ
=
๐2 โ
๐
๐โ1 โ๐1 ,๐2 โ๐2 ๐
๐1 โ๐1 ,๐2 โ๐2 .
(2.18)
๐1 =๐1 โ๐1 ๐2 =๐2 โ๐2
Replacing ๐1 + ๐1 by ๐1 and ๐2 + ๐2 by ๐2 in the right hand of (2.18) we obtain ๐1 +๐1 ๐2 +๐2 โ โ
๐
๐โ1 โ๐1 +๐1 ,๐2 โ๐2 +๐2 ๐
๐1 โ๐1 +๐1 ,๐2 โ๐2 +๐2 .
(2.19)
๐1 =๐1 ๐2 =๐2
Replacing ๐1 + ๐1 โ ๐1 + ๐1 by ๐ห1 and ๐2 โ ๐2 + ๐2 + ๐2 by ๐ห2 in (2.19) we obtain ๐1โ +๐1 ๐2โ +๐2 ๐ห1 =๐1 ๐ห2 =๐2
๐
๐โห โ๐ 1
ห
1 ,๐2 โ๐2
๐
๐ห1 โ๐1 ,๐ห2 โ๐2 .
(2.20)
The Inverse of a Toeplitz Operator Matrix
395
Thus (2.18) and (2.20), yield that ๐1 โ
๐2 โ
โ ๐๐โ๐ ๐๐โ๐ =
๐1โ +๐1 ๐2โ +๐2 ๐ห1 =๐1 ๐ห2 =๐2
๐1 =๐1 โ๐1 ๐2 =๐2 โ๐2
โ ๐
ห๐โ๐ ๐
ห๐โ๐ .
This proves that ๐ธ = 0. Now let us prove (1.6). The proof of (1.7) is similar and will be omitted. Let ๐, ๐ห โ ฮ1 . Since ๐๐ = ๐
๐ = 0 when ๐ โโ ๐ = ๐1 ร ๐2 , we get from ๐ท1 = โ ห ห 22 ๐ฮ1 ๐
๐
22 ๐ฮ1 that โ โ โ โ (๐ท1 )๐,๐ห = ๐๐โ๐ ๐๐โ๐ โ ๐
๐ โ๐ ๐
๐ โ๐ห . ห ๐โฮโชฮ1 ๐โ๐โ๐1 ร๐2 ห ๐โ๐โ๐ 1 ร๐2
๐ โ๐+โ20 ๐ โ๐โ๐1 ร๐2 ห ๐ โ๐โ๐ 1 ร๐2
Note that ๐ โ ๐ โ ๐1 ร ๐2 and ๐ห โ ๐ โ ๐1 ร ๐2 implies ๐1 โ ๐1 โค ๐1 โค ๐1 , ๐ห1 โ ๐1 โค ๐1 โค ๐ห1 , 0 โค ๐2 โค ๐2 and 0 โค ๐2 โค ๐ห2 . Combining these inequalities we get max{๐1 , ๐ห1 } โ ๐1 โค ๐1 โค min{๐1 , ๐ห1 } and 0 โค ๐2 โค min{๐2 , ๐ห2 }. Similarly, ๐ โ๐ โ ๐1 ร๐2 and ๐ โ ๐ห โ ๐1 ร๐2 implies that ๐1 โค ๐ 1 โค ๐1 +๐1 , ๐ห1 โค ๐ 1 โค ๐ห1 +๐1 , ห entry of ๐ท1 is given by ๐2 โค ๐ 2 โค ๐2 + ๐2 and ๐2 โค ๐ 2 โค ๐ห2 + ๐2 . Thus ๐, ๐th (1.6). โก
3. Implementation of the formula in Matlab Suppose we are given a two variable scalar-valued stable polynomial ๐ (๐ง1 , ๐ง2 ) =
๐1 โ ๐2 โ
๐๐๐ ๐ง1 ๐ ๐ง2 ๐
๐=0 ๐=0 1 โฃ๐ โฃ2 .
with the symbol of ๐ is of the form We can build the matrices ๐ด,๐ต,๐ถ1 ,๐ท1 , ๐ถ2 ,๐ท2 according to Theorem 1.1. The matrices ๐ท1 and ๐ท2 are generated by matrixvalued symbols of one variable. One way to compute ๐ถ1โ ๐ท1โ1 ๐ถ1 is to factorize ๐ท1โ1 = ๐น ๐น โ with ๐น upper triangular. As ๐ถ1 is typically sparse with entries in the upper part, ๐น ๐ถ1 will also be sparse. The factorization of ๐ท1 (and ๐ท2 ) can be obtained by a direct LU factorization, but also via the so-called Discrete Algebraic Riccati Equation (DARE) in Matlab. We will illustrate the latter method. Suppose ๐ (๐ง) = ๐โ๐ ๐ง โ๐ + โ
โ
โ
+ ๐๐ ๐ง ๐ โฅ 0, โฃ๐งโฃ = 1. We want โ
๐ (๐ง) = ๐(๐ง) ๐(๐ง)
(3.1)
๐
where ๐(๐ง) = ๐0 + โ
โ
โ
+ ๐๐ ๐ง is the outer factor. Note that (3.1) is equivalent to โก โโค โก โ ๐0 ๐0 ๐0 โ
โ
โ
[ ] โข .. โฅ โข .. .. ๐ โ
โ
โ
๐ = โฃ.โฆ 0 โฃ . ๐ . ๐โ๐ ๐โ๐ ๐0 โ
โ
โ
โค ๐โ0 ๐๐ .. โฅ . โฆ
๐โ๐ ๐๐
396
S. Koyuncu and H.J. Woerdeman
having property that ๐ โ
๐โ๐ ๐๐ = ๐0 ,
๐=๐
๐โ1 โ
๐๐ ๐โ๐+1 = ๐1 ,
๐โ2 โ
๐=0
๐๐ ๐โ๐+2 = ๐2 , . . . , ๐0 ๐โ๐โ1 = ๐๐
(3.2)
๐=0
with ๐1โ = ๐โ1 , ๐2โ = ๐โ2 , . . . , ๐๐โ = ๐โ๐ . Therefore, we consider โก โค ๐00 โ
โ
โ
๐0๐ โข .. โฅ โฅ 0. ๐ = โฃ ... . โฆ ๐๐0
โ
โ
โ
(3.3)
๐๐๐
with the property ๐ โ
๐๐๐ = ๐0 ,
๐=0
If we write ๐ = where
๐
= ๐0 ,
โค
๐๐,๐+1 = ๐1 , . . . , ๐0๐ = ๐๐
(3.4)
๐=0
[
โก
๐โ1 โ
๐โ๐ โข .. โฅ ๐ = โฃ . โฆ, ๐โ1
โ๐
๐
๐โ
๐
]
[ +
โก 0 ๐ผ โข .. โข. ๐ด=โข โข. โฃ .. 0 0
๐ดโ ๐๐ด
๐ดโ ๐๐ต
๐ต โ ๐๐ด ๐ต โ ๐๐ต โค โฅ โฅ โฅ, โฅ ๐ผโฆ ... 0 ..
.
] ,
โก โค 0 โข .. โฅ โข โฅ ๐ต = โข.โฅ , โฃ0 โฆ ๐ผ
and ๐ = ๐ โ
then ๐ has the property (3.4). In fact, every ๐ satis๏ฌying (3.4) can be written in this form. Now suppose that ๐ is so that ๐
+ ๐ต โ ๐๐ต > 0 then ๐ โฅ 0 if and only if ๐ โ ๐ดโ ๐๐ด + (๐ + ๐ดโ ๐๐ต)(๐
+ ๐ต โ ๐๐ต)โ1 (๐ โ + ๐ต โ ๐๐ด) โ ๐ โฅ 0.
(3.5)
At the optimal choices of ๐ one can get ๐ โ ๐ดโ ๐๐ด + (๐ + ๐ดโ ๐๐ต)(๐
+ ๐ต โ ๐๐ต)โ1 (๐ โ + ๐ต โ ๐๐ด) โ ๐ = 0
(3.6)
and for one of the optimal ones, we have that ๐ด โ ๐ต(๐
+ ๐ต โ ๐๐ต)โ1 (๐ โ + ๐ต โ ๐๐ด) โ
โ1
(3.7) โ2
โ
โ
has of its eigenvalues in ๐ป. If we let (๐
+ ๐ต ๐๐ต) = ๐0 , ๐ + ๐ต ๐๐ด = [ all ] ๐0 ๐โ๐ โ
โ
โ
๐0 ๐โ1 , then (3.5) becomes the companion matrix of ๐(๐ง) = ๐ง ๐ ๐ผ + โ ๐โ1 โ ๐โ1 + โ
โ
โ
+ ๐โ1 0 ๐1 ๐ง 0 ๐๐ and thus ๐ has all its eigenvalues in ๐ป. A detailed description of this method can be found in Section 15 of [5]. We now give the following example to illustrate how DARE can be used to factorize ๐ท1 and ๐ท2 . Example. Let ๐ (๐ง1 , ๐ง2 ) = 5 + 2๐ง1 + 3๐ง2 + ๐ง1 ๐ง2 + ๐ง12 + ๐ง22 . Since ๐ (๐ง1 , ๐ง2 ) = ๐ (๐ง1 , ๐ง2 )๐ (1/๐ง1 , 1/๐ง2 ) = 41 + 15(๐ง1 + 1/๐ง1 ) + 20(๐ง2 + 1/๐ง2 ) + 5(๐ง12 + 1/๐ง12 ) + 5(๐ง22 + 1/๐ง22 ) + 5(๐ง1 ๐ง2 + 1/๐ง1 ๐ง2 ) + 8(๐ง1 /๐ง2 + ๐ง2 /๐ง1 ) + (๐ง12 /๐ง22 + ๐ง22 /๐ง12 ) + 3(๐ง12 /๐ง2 + ๐ง2 /๐ง12 ) + 2(๐ง1 /๐ง22 + ๐ง22 /๐ง1 ), then the symbol of ๐ is positive. Letting ๐1 = ๐2 = 3,
The Inverse of a Toeplitz Operator Matrix we assemble a bi-in๏ฌnite block as shown below. โก 30 17 5 โข17 39 17 โข โข 5 17 30 โข โข12 7 2 โข โข 5 15 7 โข โข 0 5 12 โข โข5 3 1 โข โข0 5 3 โข โฃ0 0 5 0 0 0
397
Toeplitz matrix ๐ท1 whose top left 10 ร 10 block is 12 7 2 30 17 5 12 5 0 5
5 15 7 17 39 17 7 15 5 3
0 5 12 5 17 30 2 7 12 1
โค 5 0 0 0 3 5 0 0โฅ โฅ 1 3 5 0โฅ โฅ 12 5 0 5 โฅ โฅ 7 15 5 3 โฅ โฅ. 2 7 12 1 โฅ โฅ 30 17 5 12โฅ โฅ 17 39 17 7 โฅ โฅ 5 17 30 2 โฆ 12 7 2 30
We now write the matrix-valued one variable symbol associated with ๐ท1 : ๐ (๐ง) = ๐โ2 where
โก โค 30 17 5 ๐0 = โฃ17 39 17โฆ , 5 17 30
1 1 + ๐โ1 + ๐0 + ๐1 ๐ง + ๐2 ๐ง 2 , ๐ง2 ๐ง
๐โ1
โก 12 =โฃ5 0
โค 7 2 15 7 โฆ , 5 12
๐โ2
โก 5 = โฃ0 0
โค 3 1 5 3โฆ 0 5
โ โ with ๐โ1 = ๐1 and ๐โ2 = ๐2 . โ Suppose ๐ (๐ง) = ๐(๐ง) ๐(๐ง) where ๐(๐ง) = ๐0 + ๐1 ๐ง + ๐2 ๐ง 2 . We write ๐0 = โ โ โ โ ๐0 ๐0 + ๐1 ๐1 +๐2 ๐2 , ๐1 = ๐0 ๐1 +๐โ1 ๐2 , ๐โ1 = ๐โ1 ๐0 +๐โ2 ๐1 , ๐2 = ๐โ0 ๐2 and ๐โ2 = ๐โ2 ๐0 . Using DARE in MATLAB, we can factorize ๐ (๐ง) in the following way: Let โก โค โก โค 0 0 0 1 0 0 0 0 0 โข0 0 0 0 1 0โฅ โข0 0 0โฅ โข โฅ โข โฅ [ ] โข โฅ โข0 0 0โฅ ๐ 0 0 0 0 0 1โฅ โข โฅ ๐
= ๐0 , ๐ = โ2 , ๐ด = โข , ๐ต = โข0 0 0 0 0 0โฅ โข1 0 0โฅ , ๐โ1 โข โฅ โข โฅ โฃ0 0 0 0 0 0โฆ โฃ0 1 0โฆ 0 0 0 0 0 0 0 0 1
๐ธ = ๐ผ6
and ๐ = ๐6
then using [๐, ๐ฟ, ๐บ] = dare (๐ด, ๐ต, ๐, ๐
, ๐, ๐ธ) in MATLAB, we get โก โค โก โค 5.0000 0 0 0.6545 โ0.5445 0.1176 0 โฆ , ๐1 = โฃ0.4961 0.6684 โ0.5944โฆ ๐0 = โฃ3.0000 4.8780 1.0000 2.4258 4.1835 0.2390 0.7171 1.1952 and
โก โค 1.3090 โ0.4548 โ0.3899 ๐2 = โฃ0.9923 1.9690 โ0.1384โฆ . 0.4781 1.3673 2.3647
398
S. Koyuncu and H.J. Woerdeman
Thus we have ๐(๐ง) = ๐0 + ๐1 ๐ง + ๐2 ๐ง 2 . Next we assemble a bi-in๏ฌnite block Toeplitz matrix ๐ท2 whose top left 10 ร 10 block is as shown below. โก โค 35 13 5 18 5 0 5 0 0 0 โข13 39 13 7 20 5 2 5 0 0โฅ โข โฅ โข 5 13 35 3 โฅ 7 18 1 2 5 0 โข โฅ โข18 7 3 35 13 5 18 5 0 5 โฅ โข โฅ โข 5 20 7 13 39 13 7 20 5 2 โฅ โข โฅ โข 0 5 18 5 13 35 3 7 18 1 โฅ . โข โฅ โข 5 2 1 18 7 3 35 13 5 18โฅ โข โฅ โข0 5 2 5 20 7 13 39 13 7 โฅ โข โฅ โฃ0 0 5 0 5 18 5 13 35 3 โฆ 0 0 0 5 2 1 18 7 3 35 Then the matrix-valued one variable symbol associated with ๐ท2 is the following: ๐ (๐ง) = ๐โ2 where
โก โค 35 13 5 ๐0 = โฃ13 39 13โฆ , 5 13 35
โ โ with ๐โ1 = ๐1 and ๐โ2 = ๐2 . We now let
๐
= ๐0 ,
[ ] ๐โ2 ๐= , ๐โ1
1 1 + ๐โ1 + ๐0 + ๐1 ๐ง + ๐2 ๐ง 2 2 ๐ง ๐ง
๐โ1
โก 18 =โฃ5 0
โก
0 โข0 โข โข0 ๐ด=โข โข0 โข โฃ0 0 ๐ธ = ๐ผ6
0 0 0 0 0 0
โค 7 3 20 7 โฆ , 5 18
0 0 0 0 0 0
1 0 0 0 0 0
0 1 0 0 0 0
๐โ2
โค 0 0โฅ โฅ 1โฅ โฅ, 0โฅ โฅ 0โฆ 0
โก 5 = โฃ0 0
โก 0 โข0 โข โข0 ๐ต=โข โข1 โข โฃ0 0
โค 2 1 5 2โฆ 0 5
0 0 0 0 1 0
and ๐ = ๐6
then using [๐, ๐ฟ, ๐บ] = dare (๐ด, ๐ต, ๐, ๐
, ๐, ๐ธ) in MATLAB, we get โก โค โก โค 5.001 2 0.999 2.99 1 โ0.01 4.855 1.625โฆ , ๐1 = โฃโ0.01 3.05 1.04 โฆ ๐0 = โฃ 0 0 0 4.554 0.01 โ0.05 2.88 and
โก 1 ๐2 = โฃ0 0
Thus we have ๐(๐ง) = ๐0 + ๐1 ๐ง + ๐2 ๐ง 2 .
โค 0 0 1.02 0 โฆ. 0.07 1.09
โค 0 0โฅ โฅ 0โฅ โฅ, 0โฅ โฅ 0โฆ 1
The Inverse of a Toeplitz Operator Matrix
399
We now present numerical results for implementation of Theorem 1.1. ๐1
๐2
โฅ ๐ โ1 โ ๐ด๐ดโ โ ๐ต โ ๐ต โ ๐ถ1โ ๐ท1 โ1 ๐ถ1 โ ๐ถ2โ ๐ท2 โ1 ๐ถ2 โฅ
4
4
1.6308๐ โ 013
8
8
3.7907๐ โ 013
16
16
1.0216๐ โ 012
32
32
3.7828๐ โ 012
In the next section we provide an algorithm to approximate ๐ โ1 in case the 1 symbol is not of the form โฃ๐(๐ง)โฃ 2 , and give numerical results.
4. Inversion algorithm and numerical results We now consider the case when the symbol of ๐ is not necessarily of the form โฃ๐โฃ1 2 . It may still be worthwhile to use the results in the previous section for approximating ๐ โ1 . Note that the expression ๐ด๐ดโ โ ๐ต โ ๐ต is easily computable when the polynomial ๐ is known, even when ๐ is a polynomial of more than two variables. Therefore, we may try to approximate the symbol of a multilevel Toeplitz by a symbol of the form โฃ๐โฃ1 2 . In this section we explore this idea. In order to use the above idea, one needs to have a way to go from a positive de๏ฌnite multilevel Toeplitz maโ 1 trix ๐ = (๐ก๐โ๐ )๐,๐โฮ to a stable polynomial ๐ so that ๐ก(๐ง) = ๐โโค๐ ๐ก๐ ๐ง ๐ = โฃ๐(๐ง)โฃ 2. This is a nontrivial step, and in fact in the multivariable case such a polynomial may not exist; see Theorem 1.1.3 in [1] for a necessary and su๏ฌcient condition when such a polynomial exist in the case of two variables. In that case we will use the following idea introduced in > 0, ๐ง โ ๐๐ , we write โ log (๐ก(๐ง)) as โ [7]. For ๐ก(๐ง) ๐ a Fourier series โ log (๐ก(๐ง)) = ๐โโค๐ ๐๐ ๐ง . Let now ๐ป be the half-space ๐ป = {(๐1 , . . . , ๐๐ ) : ๐1 = โ
โ
โ
= ๐๐โ1 = 0, ๐๐ โ= 0 โ ๐๐ > 0}. Then ๐ป โช (โ๐ป) โช {0} = โค๐ and ๐ป โฉ (โ๐ป) = โ
. We now introduce ๐+ (๐ง) = โ 1 ๐ ๐ ๐โ๐ป ๐๐ ๐ง Then โ log(๐ก(๐ง)) = ๐+ (๐ง) + ๐+ (๐ง), ๐ง โ ๐ . Next we compute 2 ๐00 + โ 1 ๐+ (๐ง) ๐ ๐ = ๐โ๐ปโช{0} ๐๐ ๐ง . Note that ๐ก(๐ง) = ๐+ (๐ง) 2 . We now use a ๏ฌnite set of the โฃ๐
โฃ
Fourier coe๏ฌcients ๐๐ , ๐ โ โ๐0 ,of ๐๐+ (๐ง) as the Fourier coe๏ฌcients of the polynomial ๐. With this choice for ๐, the matrices ๐ด and ๐ต are built as in Theorem 1.1. We let ๐1 =๐ด๐ดโ โ ๐ต โ ๐ต and it should be noted that while the symbols ๐ก(๐ง) 1 are not of the form โฃ๐(๐ง)โฃ 2 , where ๐ is stable, with the choice below of Fourier coe๏ฌcients supported in {0, . . . , 4}๐ , the approximations are quite good. Let us mention that in [6] an approximation algorithm is proposed that the inverse of two-level Toeplitz matrices for various typical symbols possess low-tensor rank approximations with Kronecker factor of low displacement rank, and they state
400
S. Koyuncu and H.J. Woerdeman
initial โencouragingโ results. The algorithm in [6] is iterative, namely based on the Hotelling algorithm [9]: ๐๐+1 = 2๐๐ โ ๐๐ ๐ ๐๐ , ๐ = 1, 2, . . . ,
(4.1)
โ1
2
where ๐1 is some initial approximation to ๐ . Since ๐ผ โ ๐ ๐๐ = (๐ผ โ ๐ ๐๐โ1 ) , the iterations (4.1) converge quadratically, provided that โฅ ๐ผ โ ๐ ๐1 โฅ< 1. Using the approximation ๐1 , and performing the Hotelling algorithm we obtain the following results. The number ๐โ indicates the number of iterations the Hotelling algorithm is performed, and ๐โ indicates the corresponding iterate. Example. ๐ก(๐ง1 , ๐ง2 ) = 2.1 โ 12 (๐ง1 2 + ๐ง11 2 ) โ 12 (๐ง2 2 + ๐ง12 2 ). Note that ๐ก is nonsingular on ๐2 . In building ๐ด and ๐ต we only use the Fourier coe๏ฌcient of ๐๐+ with index ห = {0, . . . , 4} ร {0, . . . , 4}. The results are as follows. ๐โ๐พ Table 1 ๐1
๐2
size(๐ )
โฅ ๐ โ1 โ ๐1 โฅ
๐โ
โฅ ๐ โ1 โ ๐โ โฅ
16
16
288 ร 288
0.002311245385348
7
3.3238๐ โ 015
32
32
1088 ร 1088 0.002329157836868
7
3.5367๐ โ 013
48
48
2400 ร 2400 0.002311239597903
8
7.8772๐ โ 015
64
64
4224 ร 4224 0.002311239524231
8
9.1165๐ โ 015
Example. With ๐ก(๐ง1 , ๐ง2 ) = 12 + 1 ) + 19 ( ๐ง๐ง12 ๐ง22 2
๐ง2
๐ง2
2
1
11 6 (๐ง1
+
1 ๐ง1 )
+ ๐ง๐ง21 ) + 14 ( ๐ง12 + ๐ง22 ) + 16 ( ๐ง๐ง12 + ๐ง๐ง22 ) + 2
1
11 1 5 2 6 (๐ง2 + ๐ง2 ) + 2 (๐ง1 2 ๐ง2 1 ๐ง1 ( + ๐ง21 ). Note that 6 ๐ง2
+
+
1 ) ๐ง12
+ 52 (๐ง22 +
๐ก is nonsingular
on ๐ . In building ๐ด and ๐ต we only use the Fourier coe๏ฌcient of ๐๐+ with index ห = {0, . . . , 4} ร {0, . . . , 4}. We obtain the following results. ๐โ๐พ Table 2 size(๐ )
โฅ ๐ โ1 โ ๐1 โฅ ๐โ
โฅ ๐ โ1 โ ๐โ โฅ
๐1
๐2
16
16
288 ร 288 0.047558938791824
5 1.9513๐ โ 014
32
32 1088 ร 1088 0.094929745730251
5 1.4426๐ โ 013
48
48 2400 ร 2400 0.084552586200363
5 2.2439๐ โ 013
64
64 4224 ร 4224 0.086403129147974
5 2.6601๐ โ 013
Below is an experiment in three variables (a case not covered in [6]). Example. ๐ก(๐ง1 , ๐ง2 , ๐ง3 ) = 3.5 โ 12 (๐ง1 + ๐ง11 ) โ 12 (๐ง2 + ๐ง12 ) โ 12 (๐ง3 + ๐ง13 ). In building ๐ด ห = {0, . . . , 4} ร and ๐ต we only use the Fourier coe๏ฌcient of ๐๐+ with index ๐ โ ๐พ {0, . . . , 4}. The results are as follows.
The Inverse of a Toeplitz Operator Matrix
401
Table 3 โฅ ๐ โ1 โ ๐1 โฅ ๐โ
โฅ ๐ โ1 โ ๐โ โฅ
๐1
๐2
๐3
size(๐ )
6
6
6
343 ร 343
0.528074
6
1.5103๐ โ 015
8
8
8
729 ร 729
0.664157
6
1.2905๐ โ 015
10
10
10
1331 ร 1331
0.754442
6
1.9590๐ โ 015
12
12
12
2197 ร 2197
0.815762
6
2.0447๐ โ 015
16
16
16
4913 ร 4913
0.8896
6
2.7554๐ โ 015
References [1] Je๏ฌrey S. Geronimo and Hugo J. Woerdeman. Two variable orthogonal polynomials on the bicircle and structured matrices. SIAM J. Matrix Anal. Appl., 29(3):796โ825 (electronic), 2007. [2] I.C. Gohberg and G. Heinig. Inversion of ๏ฌnite Toeplitz matrices consisting of elements of a noncommutative algebra. Rev. Roumaine Math. Pures Appl. (in Russian), 19:623โ663,1974. [3] I.C. Gohberg and A.A. Semencul. The inversion of ๏ฌnite Toeplitz matrices and their continual analogues. Mat. Issled., 7(2(24)):201โ223, 290, 1972. [4] Georg Heinig and Karla Rost. Algebraic methods for Toeplitz-like matrices and operators. Akademie-Verlag, Berlin, 1984. [5] Peter Lancaster and Leiba Rodman. Algebraic Riccati equations. Oxford Science Publications. The Clarendon Press Oxford University Press, New York, 1995. [6] Vadim Olshevsky, Ivan Oseledets, and Eugene Tyrtyshnikov. Tensor properties of multilevel Toeplitz and related matrices. Linear Algebra Appl., 412(1):1โ21, 2006. [7] Cornelis V.M. van der Mee, Sebastiano Seatzu, and Giuseppe Rodriguez. Spectral factorization of bi-in๏ฌnite multi-index block Toeplitz matrices. Linear Algebra Apply., 343/344:355โ380, 2002. Special issue on structured and in๏ฌnite systems of linear equations. [8] Hugo J. Woerdeman. Estimates of inverses of multivariable Toeplitz matrices. Oper. Matrices, 2(4):507โ515, 2008 [9] Harold Hotelling. Some new methods in matrix calculation. Ann. Math. Statistics, 14:1โ34, 1943 Selcuk Koyuncu and Hugo J. Woerdeman Department of Mathematics Drexel University Philadelphia, PA 19104, USA e-mail:
[email protected] [email protected]
Operator Theory: Advances and Applications, Vol. 218, 403โ424 c 2012 Springer Basel AG โ
Parametrizing Structure Preserving Transformations of Matrix Polynomials Peter Lancaster and Ion Zaballa Dedicated to the memory of Israel Gohberg, good friend and scholar
Abstract. The spectral properties of ๐ ร ๐ matrix polynomials are studied in terms of their (isospectral) linearizations. The main results in this paper concern the parametrization of strict equivalence and congruence transformations of the linearizations. The โcentralizerโ of the appropriate Jordan canonical form plays a major role in these parametrizations. The transformations involved are strict equivalence or congruence according as the polynomials in question have no symmetry, or are Hermitian, respectively. Jordan structures over either the complex numbers or the real numbers are used, as appropriate. Mathematics Subject Classi๏ฌcation (2000). 15A21, 15A54, 47B15. Keywords. Matrix polynomials, structure preserving, transformations.
1. Introduction The objects of study in this paper are ๐ ร ๐ matrix polynomials of the form โโ ๐ ๐ร๐ ๐ฟ(๐) = (or ๐ด๐ โ โ๐ร๐ ) for each ๐ and ๐ดโ is ๐=0 ๐ด๐ ๐ where ๐ด๐ โ โ nonsingular. Two matrix polynomials with nonsingular leading coe๏ฌcients will be said to be isospectral if they have the same elementary divisors or, equivalently, the same underlying Jordan canonical form. (The Jordan form will be over the complex or real ๏ฌelds as the context requires.) It is well known (see [7], [8], [11]) that such a polynomial has an isospectral linearization ๐๐ด โ ๐ต where โก โค โก โค ๐ด1 ๐ด2 โ
โ
โ
๐ด๐ โ๐ด0 0 โ
โ
โ
0 0 โข ๐ด2 โ
โ
โ
๐ด๐ 0 โฅ โข 0 ๐ด2 โ
โ
โ
๐ด๐ โฅ โข โฅ โข โฅ ๐ด=โข . , ๐ต=โข . (1) โฅ โฅ, . . .. โฆ .. โฃ .. โฃ .. 0 โฆ ๐ด๐ 0 โ
โ
โ
0 0 ๐ด๐ 0 โ
โ
โ
0 This work was supported by grants from the EPSRC (United Kingdom), NSERC (Canada), and DGICYT, GV (Spain).
404
P. Lancaster and I. Zaballa
Note also that, when ๐ฟ(๐) is hermitian, so is this linearization and, since ๐ดโ is invertible, ๐ด is also invertible. ห Given isospectral matrix polynomials ๐ฟ(๐) and ๐ฟ(๐), the ๏ฌrst objective is to parametrize all strict equivalence transformations connecting their linearizations. In other words, we are to parametrize all pairs of nonsingular complex matrices ๐ and ๐ for which (with the above de๏ฌnitions) ห โ ๐ต)๐ ห ; ๐ (๐๐ด โ ๐ต) = (๐๐ด
(2)
they determine a strict equivalence transformation. Pairs of matrices (๐, ๐ ) satisfying this property will be called block-symmetric structure preserving transformations (SPT), since they preserve the block-symmetric structure of ๐ด and ๐ต (see [2, 13]). It was shown in [13, Thms. 7, 8] that the structure preserving transformations of two given isospectral matrix polynomials are closely related to their standard triples as de๏ฌned in [7, 8]. As a ๏ฌrst step, it will be shown in this paper that a parametrization of all possible block-symmetric SPTs for two given ๐ ร ๐ isospectral matrix polynomials can be obtained in terms of the centralizer of their common Jordan form, namely, ๐(๐ฝ) := {ฮ โ โโ๐รโ๐ : ฮ๐ฝ = ๐ฝฮ}.
(3)
We ๏ฌrst consider general (non-symmetric) polynomials (see Theorem 2.2), then those with real coe๏ฌcients (Theorem 3.1), and in Section 4 (Theorem 4.3) those with hermitian coe๏ฌcients. Finally, in Section 5, we consider those with real symmetric coe๏ฌcients. In the case of hermitian matrix polynomials the strict equivalence transformations de๏ฌned by (2) are replaced by congruence transformations: ห โ ๐ต, ห ๐ โ (๐๐ด โ ๐ต)๐ = ๐๐ด
(4)
with ๐ nonsingular. These transformations, preserving the symmetries and block structure of ๐ด and ๐ต, will be called structure preserving congruences (SPC, for short). As in the nonsymmetric case, SPCโs and selfadjoint standard triples for a given hermitian matrix polynomial will be shown to be closely related. The definition of selfadjoint standard triples given in [11, p. 244] will be used, and a one-to-one correspondence between SPC matrices and selfadjoint standard triples will be exhibited. To complete this work, it has been found necessary to carefully review canonical structures associated with matrix polynomials, and this has been done in the accompanying paper [15]. A characterization of the set of all SPCs will be obtained in terms of the (suitably modi๏ฌed) centralizer. The invariants known as the sign characteristics associated with real eigenvalues (and sub-sumed in a primitive matrix ๐ ) are to be preserved as well as the complete Jordan structure โ and this motivates the notion of strictly isospectral hermitian matrix polynomials. It will be seen that the
Parametrizing Structure Preserving Transformations
405
role of matrices in the centralizer of ๐ฝ must be restricted to admit a ๐ -unitary property. A matrix polynomial ๐ฟ(๐) is said to be diagonalizable if there is an isospectral diagonal matrix polynomial of the same size and degree. Algorithms have been proposed for the reduction of diagonalizable quadratic polynomials (โ = 2, which we call systems) (see also [2], [5]) and they are the subject of the recent paper [14]. In view of their importance, and for the purpose of illustration, we focus on this quadratic case in Section 2.1 and a detailed example is included. Here, in the terminology of [14], we are concerned with systems which are ๐ท๐ธโ (diagonalizable by strict equivalence over โ applied to a linearization). Sections 3 and 3.1 are analogues of 2 and 2.1, but are devoted to the special case of real matrix polynomials (without symmetries). In Section 3.1 the systems are said to be ๐ท๐ธโ (diagonalizable by strict equivalence over โ applied to a linearization). Section 4 is devoted to the case of hermitian matrix polynomials and includes systems which are ๐ท๐ถโ (diagonalizable to real form by complex congruence). Another natural and important topic concerns the real symmetric matrix polynomials (which are, of course, both real and hermitian). They are considered in Section 5, where the techniques of Sections 3 and 4 are utilised. Here, the systems are also ๐ท๐ถโ but are now diagonalizable to real form by real congruence. Analysis of this case requires some extension of existing theory, and is developed in the accompanying paper [15].
2. General complex matrix polynomials ห Let ๐ฟ(๐) and ๐ฟ(๐) be two โ-degree ๐ ร ๐ matrix polynomials with nonsingular หโ๐ต ห be their leading coe๏ฌcients, as in the introduction. Let ๐๐ด โ ๐ต and ๐๐ด ห linearizations as de๏ฌned in (1). If ๐ฟ(๐) and ๐ฟ(๐) are isospectral then ๐๐ด โ ๐ต and หโ๐ต ห are, as pencils, strictly equivalent; i.e., ๐๐ด ห โ ๐ต)๐ ห ๐ (๐๐ด โ ๐ต) = (๐๐ด
(5)
for some nonsingular ๐ and ๐ . We aim to characterize and parametrize the non ห singular block-symmetric SPTs for ๐ฟ(๐) and ๐ฟ(๐); i.e., all pairs of matrices (๐, ๐ ) for which (5) holds. As shown in [13, Th. 7] SPTs and standard triples are closely related. (The notions of โstandard pairs and triplesโ for a matrix polynomial are carefully developed in [15].) We recall here that, if ๐ถ๐
is the right companion matrix of ๐ฟ(๐), i.e., โก โค 0 ๐ผ๐ โ
โ
โ
0 โข โฅ .. .. .. .. โข โฅ . . . . ๐ถ๐
= โข (6) โฅ โฃ โฆ 0 0 โ
โ
โ
๐ผ๐ โ1 โ๐ดโ1 โ
โ
โ
โ๐ดโ1 โ ๐ด0 โ๐ดโ ๐ด1 โ ๐ดโโ1
406
P. Lancaster and I. Zaballa
and
โก
0 0 .. .
โค
โข โฅ โข โฅ ๐0 = โข (7) โฅ, โฃ โฆ ๐ดโ1 โ then (๐0 , ๐ถ๐
, ๐0 ) is a standard triple of ๐ฟ(๐) and any other standard triple of this matrix polynomial is similar to (๐0 , ๐ถ๐
, ๐0 ). It is also important to realize that [ ๐0 = ๐ผ๐
0
โ
โ
โ
] 0 ,
๐ด๐ถ๐
= ๐ถ๐ฟ ๐ด = ๐ต,
(8)
where ๐ถ๐ฟ is the left companion matrix of ๐ฟ(๐): โก โค 0 โ
โ
โ
0 โ๐ด0 ๐ดโ1 โ โข๐ผ๐ โ
โ
โ
0 โฅ โ๐ด1 ๐ดโ1 โ โข โฅ ๐ถ๐ฟ = โข . . โฅ. . . . . .. .. โฃ .. โฆ โ1 0 โ
โ
โ
๐ผ๐ โ๐ดโโ1 ๐ดโ Since ๐ด is invertible (๐0 ๐ดโ1 , ๐ถ๐ฟ , ๐ด๐0 ) is also a standard triple of ๐ฟ(๐). The block-symmetric SPTs of two matrix polynomials can be characterized by using standard triples as follows: ห Theorem 2.1. Let ๐ฟ(๐) and ๐ฟ(๐) be isospectral matrix polynomials of the same ห size. Then (๐, ๐ ) is a block-symmetric SPT for ๐ฟ(๐) and ๐ฟ(๐) if and only if one (and then both) of the following equivalent conditions holds: (a) โค โก ๐ โข ๐๐ถ๐
โฅ ] [ โฅ โข โโ1 (9) ๐ = โข .. โฅ and ๐ โ1 = ๐ด ๐ ๐ถ๐
๐ โ
โ
โ
๐ถ๐
๐ โฃ . โฆ โโ1 ๐๐ถ๐
ห for a standard triple (๐, ๐ถ๐
, ๐ ) of ๐ฟ(๐). (b) โก ห โค ๐ โข ๐ ห๐ถ ห๐ฟ โฅ ] [ โข โฅ ห หโโ1 ๐ห ห๐ฟ ๐ห โ
โ
โ
๐ถ (10) and ๐ = ๐ห ๐ถ ๐ โ1 = โข . โฅ ๐ด ๐ฟ โฃ .. โฆ ห๐ถ ห โโ1 ๐ ๐ฟ
ห ๐ถ ห๐ฟ , ๐ห ) of ๐ฟ(๐). for a standard triple (๐, The proof follows from the proofs of Theorems 7 and 8 in [13]. ห Notice that, given isospectral matrix polynomials ๐ฟ(๐) and ๐ฟ(๐), the stanห dard triples of ๐ฟ(๐) of the form (๐, ๐ถ๐
, ๐ ) are completely determined by ๐; and ห ๐ถ ห๐ฟ , ๐ห ) are completely determined by the standard triples of ๐ฟ(๐) of the form (๐, ห ๐. It follows that Theorem 2.1 can be used to de๏ฌne a bijective correspondence ห ห for which between block-symmetric SPTs for ๐ฟ(๐) and ๐ฟ(๐) and matrices ๐ (๐)
Parametrizing Structure Preserving Transformations
407
ห ๐ถ ห๐ฟ , ๐ห )) is a standard triple of ๐ฟ(๐) ห (๐, ๐ถ๐
, ๐ ) ((๐, (๐ฟ(๐), respectively). In this section we aim to provide a more concise parametrizing set. Notice, for example, that if no invertibility is required of ๐ and ๐ , then the set of matrix pairs (๐, ๐ ) such that ห โ ๐ต)๐ ห ๐ (๐๐ด โ ๐ต) = (๐๐ด is a linear space. The goal is to obtain a parametrizing space for the blocksymmetric SPTs of two isospectral matrix polynomials which re๏ฌects their linearity and whose dimension can be easily computed. Let ๐ฝ be the Jordan form (over โ) of a matrix polynomial ๐ฟ(๐) โ as above โ and recall the de๏ฌnition (3) of the centralizer of ๐ฝ. If ๐1 , . . . , ๐๐ are the distinct eigenvalues of ๐ฝ, it is known (see for example [1, p. 222]) that ๐(๐ฝ) is a linear space of dimension ๐ ๐ ๐ โ โ (2๐ โ 1)๐๐๐ , (11) ๐= ๐=1 ๐=1
ห where, for eigenvalue ๐๐ of ๐ฟ(๐) (and of ๐ฟ(๐)), ๐ = 1, . . . , ๐, ๐ ๐ is the geometric multiplicity of ๐๐ , and (๐๐1 , . . . , ๐๐๐ ๐ ) is the Segre characteristic. ห Let ฮ denote the set of all block-symmetric SPTs of ๐ฟ(๐) and ๐ฟ(๐): ห โ ๐ต)๐ ห }. ฮ = {(๐, ๐ ) โ โโ๐รโ๐ : ๐ (๐๐ด โ ๐ต) = (๐๐ด As already noted, ฮ is a linear space. The main result in this section is the following theorem โ whose proof is quite straightforward. ห Theorem 2.2. Let ๐ฟ(๐), ๐ฟ(๐) be ๐ร๐ isospectral matrix polynomials with det ๐ดโ โ= ห 0 and det ๐ดโ โ= 0, let ๐ฝ be their common Jordan form, and de๏ฌne ฮ as above. Let ห๐
๐ห = ๐ฝ. Then, the ๐ and ๐ห be invertible matrices such that ๐ โ1 ๐ถ๐
๐ = ๐หโ1 ๐ถ mapping ๐ : ๐(๐ฝ) โโ ฮ de๏ฌned by ห๐หฮ๐ โ1๐ดโ1 , ๐หฮ๐ โ1) ๐(ฮ) = (๐ด is an isomorphism of linear spaces. Proof. It is clear that, provided that ๐ is well de๏ฌned, it is a linear mapping. So the goal is to prove that ๐ is well de๏ฌned and bijective. ห๐หฮ๐ โ1 ๐ดโ1 and ๐ = ๐หฮ๐ โ1 then ๐ = ๐ด๐ ห ๐ดโ1 and so ๐ ๐ด = ๐ด๐ ห . If ๐ = ๐ด Also, bearing in mind (8), ๐๐ต
= = = = = =
ห๐หฮ๐ โ1 ๐ดโ1 ๐ต, ๐ด ห๐หฮ๐ โ1 ๐ถ๐
, ๐ด ห๐หฮ๐ฝ๐ โ1 , ๐ด ห๐ห๐ฝฮ๐ โ1 , ๐ด ห ห๐
๐หฮ๐ โ1 , ๐ด๐ถ ห ๐ต๐.
(๐ด๐ถ๐
= ๐ต) (๐ โ1๐ถ๐
๐ = ๐ฝ) (ฮ๐ฝ = ๐ฝฮ) ห๐
๐ห = ๐ฝ) (๐หโ1 ๐ถ ห ห ห (๐ด๐ถ๐
= ๐ต)
408
P. Lancaster and I. Zaballa
Therefore,
ห๐หฮ๐ โ1 ๐ดโ1 )(๐๐ด โ ๐ต) = (๐๐ด ห โ ๐ต)( ห ๐หฮ๐ โ1 ), (๐ด ห๐หฮ๐ โ1 ๐ดโ1 , ๐หฮ๐ โ1 ) โ ฮ, as required. and (๐ด ห and ๐ห are invertible maThe injectivity of ๐ is immediate because ๐ด, ๐ , ๐ด trices. Let us prove that ๐ is surjective. ห ๐ต)๐ ห . Since ๐ด๐ถ๐
= ๐ต and ๐ด ห๐ถ ห๐
= ๐ต, ห Let (๐, ๐ ) โ ฮ, i.e., ๐ (๐๐ดโ๐ต) = (๐๐ดโ ห โ๐ โ ๐ถ ห๐
)๐ . Thus ๐ ๐ด(๐๐ผโ๐ โ ๐ถ๐
) = ๐ด(๐๐ผ ห๐
)๐ หโ1 ๐ ๐ด(๐๐ผโ๐ โ ๐ถ๐
) = (๐๐ผโ๐ โ ๐ถ ๐ด and the following relations are obtained: ห๐
๐, and ๐ ๐ ๐ฝ๐ โ1 = ๐ห๐ฝ ๐หโ1๐. หโ1 ๐ ๐ด = ๐, ๐ ๐ถ๐
= ๐ถ ๐ด ห๐
๐ห = ๐ฝ. Thus ๐หโ1๐ ๐ ๐ฝ = The last statement is a consequence of ๐ โ1๐ถ๐
๐ = ๐หโ1 ๐ถ โ1 ๐ฝ ๐ห ๐ ๐ . ห ๐ดโ1 = If we put ฮ = ๐หโ1 ๐ ๐ , then ฮ โ ๐(๐ฝ), ๐ = ๐หฮ๐ โ1 and ๐ = ๐ด๐ ห๐หฮ๐ โ1 ๐ดโ1 as desired. ๐ด โก According to this result, ฮ is a linear space of dimension ๐ (see (11)) and ห ๐ต)๐ ห are parameterized nonsingular matrices ๐ and ๐ for which ๐ (๐๐ดโ๐ต) = (๐๐ดโ through nonsingular matrices ๐ in the centralizer of ๐ฝ; a Zariski open set of the linear space ๐(๐ฝ) and a subgroup of the general linear group Glโ๐ (โ). ห๐
๐ห = Notice that nonsingular matrices ๐ and ๐ห for which ๐ถ๐
๐ = ๐ ๐ฝ and ๐ถ ห ๐ ๐ฝ (as used in this construction) necessarily have the partitioned form โก โค โก ห โค ๐ ๐ โข ๐๐ฝ โฅ โข ๐๐ฝ ห โฅ โข โฅ โฅ โข ๐ = โข . โฅ , ๐ห = โข . โฅ , โฃ .. โฆ โฃ .. โฆ ห โโ1 ๐๐ฝ โโ1 ๐๐ฝ ห are full-rank ๐ ร โ๐ matrices. Therefore (๐, ๐ฝ) and (๐, ห ๐ฝ) are where ๐ and ๐ ห Jordan pairs of ๐ฟ(๐) and ๐ฟ(๐), respectively. Notice also that, in the important special case in which ๐ฟ(๐) has all eigenvalues distinct, ๐ = โ๐ and the matrices ฮ parametrizing the block-symmetric SPTs (๐, ๐ ) are nonsingular diagonal matrices. 2.1. Diagonalizable quadratic systems This section concerns an application of Theorem 2.2 to a class of matrix polynomials for which numerical algorithms have been proposed ([2], [3], [5], for example), namely, โdiagonalizableโ systems. By de๏ฌnition, they are polynomials ๐ฟ(๐) of deห gree two for which there exists an isospectral diagonal quadratic system ๐ฟ(๐). Since all semisimple systems are included, the diagonalizable systems are often seen as being widely useful. A complete description of admissible Jordan forms ๐ฝ appears in [14], and we ห ห๐
๐ห = ๐ห๐ฝ. use that information here to parametrize all matrices ๐ห for which ๐ถ
Parametrizing Structure Preserving Transformations
409
This, in turn, determines a parametrization of the pairs (๐, ๐ ) โ ฮ. The theory is illustrated with a detailed example. ห ห๐2 + ๐ท๐ ห +๐พ ห is Let ๐1 , . . . , ๐๐ก be the distinct eigenvalues of ๐ฟ(๐). If ๐ฟ(๐) =๐ 2 ห a diagonal isospectral system then the element in position (๐, ๐) is ๐ ห ๐ ๐ + ๐๐ ๐ + ห ๐๐ . For each ๐ = 1, . . . , ๐ there are two possible cases: Either (i) ๐ ห ๐ ๐2 + ๐ห๐ ๐ + ห ๐๐ = ๐ ห ๐ (๐ โ ๐๐๐ )2 , or (ii) ๐ ห ๐ ๐2 + ๐ห๐ ๐ + ห ๐๐ = ๐ ห ๐ (๐ โ ๐๐๐ )(๐ โ ๐๐๐ ) with ๐๐๐ โ= ๐๐๐ . [ ] ๐๐๐ 1 In the ๏ฌrst case, de๏ฌne ๐ฝ๐ = ; and in the second case de๏ฌne ๐ฝ๐ = 0 ๐๐๐ [ ] โ๐ ๐๐๐ 0 ห . Let ๐ฝ = ๐=1 ๐ฝ๐ . This is a Jordan form of ๐ฟ(๐). 0 ๐๐๐ [ ] [ ] ๐ 1 1 0 Next, when ๐ฝ๐ = ๐๐ put ๐๐ = and notice that 0 ๐๐๐ โ๐๐๐ 1 [ ] [ ] 0 1 0 1 โ1 ๐๐ ๐ฝ๐ ๐๐ = = . โ๐2๐๐ 2๐๐๐ โห ๐๐ /๐ ห ๐ โ๐ห๐ /๐ ห๐ [ ] [ ] ๐๐๐ 1 1 โ ๐๐ โ๐ ๐๐๐ 0 ๐๐๐ โ๐๐๐ , observe that ๐๐ ๐ When ๐ฝ๐ = put ๐๐ = 0 ๐๐๐ โ๐๐๐ 1 [ ] 1 1 โ ๐๐ โ๐ ๐๐ ๐ ๐๐โ1 = ๐๐๐ ๐๐๐ 1 โ ๐๐ โ๐ ๐ ๐
and ๐๐โ1 ๐ฝ๐ ๐๐ =
[
0
๐
] [ 0 1 = ๐๐๐ + ๐๐๐ โห ๐๐ /๐ ห๐
โ๐๐๐ ๐๐๐ โ๐ก Thus, if we de๏ฌne ๐ = ๐=1 ๐๐ , then ๐ก [ โ 0 ๐ โ1 ๐ฝ๐ = โห ๐๐ /๐ ห๐ ๐=1
1 โ๐ห๐ /๐ ห๐
] 1 . โ๐ห๐ /๐ ห๐
]
and there is a permutation matrix ๐ (always the same) such that [ ] 0 ๐ผ๐ ๐ โ1 ๐ ๐ ๐ฝ๐ ๐ = หโ1 ๐พ ห โ๐ หโ1 ๐ท ห . โ๐ ห๐
๐ห = ๐ห๐ฝ. Thus, if ๐ห = (๐ ๐ )โ1 then ๐ห is invertible and ๐ถ Let us apply this construction to a simple example. Example 2.3. Consider the diagonalizable system [ ] [ ] [ ] โ1 โ3 1 2 2 0 1 ๐ฟ(๐) = ๐ +๐ + . 1 3 โ3 โ7 2 4
410
P. Lancaster and I. Zaballa
By ๏ฌrst examining the centralizer of the Jordan form, we will construct a complete parametrization of the pairs ๐, ๐ in ฮ. The eigenvalues of ๐ฟ(๐) are: +1 with algebraic multiplicity 3 and geometric multiplicity 2, and the simple eigenvalue 0. The fact that the eigenvalue +1 has Segre characteristic (2, 1) ensures that ๐ฟ(๐) is diagonalizable (the Segre characteristics (1, 1, 1) and (3) for this eigenvalue are not admissible, see [14]). A diagonal strictly isospectral system is ] [ ] [ 2 ] [ [ ] ][ 2 0 ๐ โ 2๐ + 1 2 0 โ4 0 2 0 0 ห ๐2 + ๐ฟ(๐) = = ๐+ . 0 โ5 0 0 0 5 0 ๐2 โ ๐ 0 โ5 All computations are made with the help of MATLAB and its Symbolic Toolbox. Matrices ๐ฝ1 and ๐ฝ2 are ] ] [ [ 1 1 1 0 , ๐ฝ2 = , ๐ฝ1 = 0 1 0 0 so that
โก 1 โข0 ๐ฝ =โข โฃ0 0
1 1 0 0
0 0 1 0
โค 0 0โฅ โฅ 0โฆ 0
is a Jordan form for ๐ฟ(๐) and the Segre characteristic is ((2, 1), (1)). Thus, according to (11), the dimension of ๐(๐ฝ) is 6. Now, with ] ] [ [ โ 1 0 0 1 ๐2 , , ๐2 = , ๐ = ๐1 ๐1 = โ1 1 โ1 1 and, de๏ฌning the permutation matrix โก 1 โข0 ๐ =โข โฃ0 0 we have ๐ห := (๐ ๐ )โ1
0 0 1 0
โก 1 โข0 =โข โฃ1 0
โค 0 0โฅ โฅ, 0โฆ 1
0 1 0 0 0 0 1 0
0 1 0 1
โค 0 โ1โฅ โฅ. 0โฆ 0
Now Jordan chains of ๐ฟ(๐) for the eigenvalues ๐1 = 1 and ๐2 = 0 are computed following [8, p. 25]. In particular, the Jordan chains of ๐ฟ(๐) for ๐1 = 1 have the form: [ ] [ ] [ ] ๐ ๐ ๐ ๐ฅ01 = , ๐ฅ11 = , ๐ฅ02 = โ๐ ๐ ๐,
Parametrizing Structure Preserving Transformations
411
] 2๐ . In particular, and the eigenvectors of ๐ฟ(๐) for ๐2 = 0 have the form ๐ฅ03 = โ๐ a matrix of Jordan chains of ๐ฟ(๐) is (taking ๐ = ๐ = ๐ = ๐ = 1 and ๐ = ๐ = 0): [ ] 1 1 1 2 ๐= . โ1 0 0 โ1 [
Thus, a matrix ๐ such that ๐ โ1 ๐ถ๐
๐ = ๐ฝ is โก 1 1 [ ] โขโ1 0 ๐ ๐ = =โข โฃ1 2 ๐๐ฝ โ1 โ1 Finally, the matrices ฮ โ ๐(๐ฝ) have the form โก ๐ ๐ ๐ โข0 ๐ 0 ฮ=โข โฃ0 ๐ ๐ 0 0 0
โค 1 2 0 โ1โฅ โฅ. 1 0โฆ 0 0
(see [1, 12]): โค 0 0โฅ โฅ. 0โฆ ๐
ห๐หฮ๐ โ1 ๐ดโ1 : Then MATLAB produces the following answers for ๐ = ๐ด [ 6*a-2*b+6*c, -4*a+2*b-4*c, 4*a-2*b+6*c, 2*b-2*a-4*c ] [ -10*f, 5*f, 0, 0 ] [ -4*a+2*b-6*c, 2*a-2*b+4*c, -2*a+2*b-6*c, -2*b+4*c ] [ -5*d+15*e+10*f, 5*d-10*e-5*f, -5*d+15*e, 5*d-10*e ] and for ๐ = ๐หฮ๐ โ1 : [ -a+b-c, -2*a+2*b-2*c, a-b+2*c, a-2*b+3*c ] [ d-e-f, 2*d-2*e-f, -d+2*e+f, -2*d+3*e+f ] [ b-c, 2*b-2*c, -b+2*c, -a-2*b+3*c ] [ d-e, 2*d-2*e, -d+2*e, -2*d+3*e ] When they are nonsingular (i.e., when ฮ is nonsingular) these matrices ๐ and ๐ de๏ฌne a block-symmetric SPT for the given systems: this can be veri๏ฌed directly from equation (2). Furthermore, according to Theorem 2.2, these are all ห possible structure preserving transformations for ๐ฟ(๐) and ๐ฟ(๐). โก
3. Real matrix polynomials ห If ๐ฟ(๐) and ๐ฟ(๐) are real isospectral matrix polynomials, it may be possible to design algorithms using only real arithmetic so that the matrices ๐ and ๐ for which (2) holds are real. With this in mind, we consider corresponding real Jordan forms (see [15]). The description of the centralizer of a matrix in real Jordan form may be less familiar than its complex counterpart. A simple computation shows, however, that if ๐พ is a matrix in real Jordan form, the real matrices ๐ โ ๐(๐พ) (the centralizer for the real Jordan form) can be described as follows:
412
P. Lancaster and I. Zaballa
ยฏ ๐+1 ,. . . , Let ๐1 , . . . , ๐๐ be the real eigenvalues of ๐พ and ๐๐+1 ,. . . , ๐๐+๐ , ๐ ยฏ ๐๐+๐ be the non-real eigenvalues (in conjugate pairs). Let ๐๐ = (๐๐1 , . . . , ๐๐๐ก๐ ) be the Segre characteristic of ๐พ associated with ๐๐ , ๐๐1 โฅ ๐๐2 โฅ โ
โ
โ
โฅ ๐๐๐ก๐ ,
๐ = 1, . . . , ๐ + ๐ . ยฏ๐ coincide for each ๐ = ๐ + And recall that the Segre characteristics of ๐๐ and ๐ 1, . . . , ๐ + ๐ . Then [it can] be veri๏ฌed that ๐ โ ๐(๐พ) if and only if ๐ = Diag(๐1 , . . . , ๐๐+๐ ) ๐ ๐ and the matrices ๐๐๐ have triangular Toeplitz structure with ๐๐ = ๐๐๐ 1โค๐,๐โค๐ก๐ as follows: โ For the real eigenvalues, ๐ = 1, . . . , ๐, they have the same form as in the complex case, โค โก 1 ๐ ๐๐๐๐๐ ๐๐๐ ๐2๐๐ โ
โ
โ
๐ โ1 โข 0 ๐1 โ
โ
โ
๐ ๐๐ โฅ ๐๐ ๐๐ โฅ โข ๐ โ โ๐๐๐ ร๐๐๐ , ๐๐๐ =โข . . .. โฅ . .. .. โฆ โฃ .. . 0 0 โ
โ
โ
๐1๐๐
๐ ๐๐๐
โก 0 โข0 โข = โข. โฃ .. 0
๐ ๐๐๐
โ
โ
โ
โ
โ
โ
0 0 .. . 0
โ
โ
โ
โ
โ
โ
โก 1 ๐๐๐ โข 0 โข โข . โข .. โข =โข โข 0 โข 0 โข โข . โฃ ..
๐1๐๐ 0 .. . 0 ๐2๐๐ ๐1๐๐ .. . 0 0 .. .
๐2๐๐ ๐1๐๐ .. . 0 โ
โ
โ
โ
โ
โ
.. . โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
โ
.. . โ
โ
โ
โค ๐ ๐๐๐๐๐ ๐ โ1 ๐๐๐๐๐ โฅ โฅ โ โ๐๐๐ ร๐๐๐ , ๐ > ๐, .. โฅ . โฆ ๐1๐๐ โค
๐๐๐๐๐๐ ๐๐๐๐๐๐ โ1 โฅ โฅ .. โฅ . โฅ โฅ ๐๐๐ ร๐๐๐ 1 , ๐ < ๐. ๐๐๐ โฅ โฅโโ โฅ 0 โฅ .. โฅ . โฆ
โ
โ
โ
0 0 โ
โ
โ
0 โ For the non-real conjugate pairs, ๐ = ๐ + 1, . . . , ๐ + ๐ , โก 1 โค ๐ ๐ด๐๐๐๐ ๐ด๐๐ ๐ด2๐๐ โ
โ
โ
๐ โ1 โข 0 ๐ด1๐๐ โ
โ
โ
๐ด๐๐๐๐ โฅ โข โฅ ๐ =โข . ๐๐๐ , .. .. โฅ .. โฃ .. . . . โฆ 0 0 โ
โ
โ
๐ด1๐๐ โก โค ๐ 0 โ
โ
โ
0 ๐ด1๐๐ ๐ด2๐๐ โ
โ
โ
๐ด๐๐๐๐ ๐ โ1 โข0 โ
โ
โ
0 0 ๐ด1๐๐ โ
โ
โ
๐ด๐๐๐๐ โฅ โข โฅ ๐ = โข. , ๐ > ๐, ๐๐๐ .. .. .. โฅ .. โฃ .. โ
โ
โ
... . . . . โฆ 0 โ
โ
โ
0 0 0 โ
โ
โ
๐ด1๐๐
Parametrizing Structure Preserving Transformations
๐ ๐๐๐
โก 1 ๐ด๐๐ โข 0 โข โข . โข .. โข =โข โข 0 โข 0 โข โข . โฃ .. 0
where ๐ดโ๐๐
โค ๐ด๐๐๐๐๐ ๐๐๐ โ1 โฅ ๐ด๐๐ โฅ .. โฅ . โฅ โฅ ๐ด1๐๐ โฅ โฅ , ๐ < ๐, 0 โฅ โฅ .. โฅ . โฆ
๐ด2๐๐ ๐ด1๐๐ .. . 0 0 .. .
โ
โ
โ
โ
โ
โ
.. . โ
โ
โ
โ
โ
โ
[
] โ๐โ๐๐ โ โ2ร2 . ๐โ๐๐
0
๐โ = โ๐๐ ๐๐๐
โ
โ
โ
โ
โ
โ
413
0
Now ๐(๐พ) is a real linear space of dimension ๐ก๐ ๐ก๐ ๐ โ ๐+๐ โ โ โ (2๐ โ 1)๐๐๐ + 2 (2๐ โ 1)๐๐๐ . ๐๐
= ๐=1 ๐=1
(12)
๐=๐+1 ๐=1
Actually, the dimension of the centralizer of a matrix does not depend on the ๏ฌeld but on the degrees of its invariant polynomials (see, for example, [1, p. 222]) an these are the same computed whether on โ or โ. Now if ๐ and ๐ห are real nonsingular matrices satisfying ห๐
๐ห = ๐ห๐พ ๐ถ๐
๐ = ๐ ๐พ and ๐ถ ห๐
the right companion matrices of ๐ฟ(๐) and ๐ฟ(๐), ห with ๐ถ๐
and ๐ถ respectively, and ห โ ๐ต)๐ ห } ฮ๐
= {(๐, ๐ ) โ โโ๐รโ๐ ร โโ๐รโ๐ : ๐ (๐๐ด โ ๐ต) = (๐๐ด then the proof of the following theorem is the same as that of Theorem 2.2. ห Theorem 3.1. Let ๐ฟ(๐) and ๐ฟ(๐) be isospectral real matrix polynomials with ห det ๐ดโ โ= 0 and det ๐ดโ โ= 0. Let ๐พ be a common real Jordan form for ๐ฟ(๐) and ห ๐ฟ(๐). Then, with ๐๐
of (12), the map ๐ : ๐(๐พ) โ ฮ๐
de๏ฌned by ห๐หฮ๐ โ1 ๐ดโ1 , ๐หฮ๐ โ1 ) ๐(ฮ) = (๐ด is an isomorphism of ๐๐
-dimensional real vector spaces. One may also ask what form Theorem 2.1 takes when con๏ฌned to real matrix polynomials. However, using real standard triples as described in [15], the theorem also holds for real matrix polynomials and does not require a separate statement. 3.1. Diagonalizable real quadratic systems As with the theory over โ we now illustrate Theorem 3.1 in the case of diagห onalizable quadratic systems. If ๐ฟ(๐) is a real diagonal system the matrix ๐ห of that theorem can be constructed as for a complex system, but an additional step ห is required. Assume that ๐ฟ(๐) is a real diagonal system with real and complex eigenvalues: ๐1 ,. . . , ๐๐ก . It is shown in [14] that the non-real complex eigenvalues must be semisimple and appear in conjugate pairs.
414
P. Lancaster and I. Zaballa
ห Let ๐ ห ๐ ๐2 + ๐ห๐ ๐ + ห ๐๐ be the polynomial in position (๐, ๐) of ๐ฟ(๐). Then there are three possibilities: ๐๐ = ๐ ห๐ (๐ โ ๐๐๐ )2 with ๐๐๐ โ โ, (i) ๐ ห ๐ ๐2 + ๐ห๐ ๐ + ห 2 ห ห (ii) ๐ ห ๐ ๐ + ๐๐ ๐ + ๐๐ = ๐ ห๐ (๐ โ ๐๐๐ )(๐ โ ๐๐๐ ) with ๐๐๐ โ= ๐๐๐ and real, ห๐ = ๐ ยฏ ๐ ) with ๐๐ โ (iii) ๐ ห ๐ ๐2 + ๐ห๐ ๐ + ๐ ห๐ (๐ โ ๐๐ )(๐ โ ๐ / โ. ๐
๐
๐
In the ๏ฌrst two cases, [ de๏ฌne ๐พ ] ๐ = ๐ฝ๐ as in the complex case of Section 2.1. In the โ๐ก ๐๐ ๐ ๐ third case, let ๐พ๐ = , where ๐๐๐ = ๐๐ + ๐๐๐ . Finally, de๏ฌne ๐พ = ๐=1 ๐ฝ๐ . โ๐๐ ๐๐ [ [ [ ] ] ] ๐๐๐ ๐๐๐ 1 0 1 0 , as in the If ๐พ๐ = or ๐พ = , de๏ฌne ๐๐ = โ๐๐ 1 0 ๐๐๐ [ ๐ ] 0 ๐๐๐ ๐ ๐ ๐๐ , (๐๐ โ= 0), de๏ฌne previous section. If ๐พ๐ = โ๐๐ ๐๐ [ ] 1 0 ๐๐ = โ๐๐ /๐๐ 1/๐๐ [ ] 1 0 , and observe that ๐๐โ1 = ๐๐ ๐๐ ] [ ] [ 0 1 0 1 = . ๐๐โ1 ๐พ๐ ๐๐ = โ(๐2๐ + ๐2๐ ) 2๐๐ โห ๐๐ / ๐ ห๐ โ๐ห๐ /๐ ห๐ โ๐ก If we de๏ฌne ๐ = ๐=1 ๐๐ , then ] ๐ก [ โ 0 1 โ1 ๐ ๐พ๐ = . ห๐ /๐ ห ๐ โ๐ห๐ /๐ ห๐ โ๐ ๐=1
Finally, de๏ฌne the permutation matrix ๐ as in the complex case and set ๐ห = ห๐
๐ห = ๐พ. (๐ ๐ )โ1 . Then ๐ห is real and ๐หโ1 ๐ถ Example 3.2. Let
] ] ] [ [ 3/2 โ1/2 2 11/2 โ9/2 โ3 5 ๐ + . ๐ฟ(๐) = ๐+ โ1/2 3/2 โ9/2 11/2 5 โ3 [
The eigenvalues are ๐1 = โ1 with algebraic multiplicity 2 and geometric multiplicity 1, together with the conjugate pair ๐2 = 2 + ๐ and ๐3 = 2 โ ๐. A diagonal isospectral system is ] [ 2 ] ] [ ] [ [ ๐ + 2๐ + 1 0 1 0 2 0 2 1 0 ห = . + +๐ ๐ฟ(๐) = ๐ 0 ๐2 โ 4๐ + 5 0 5 0 โ4 0 1 Thus [ ๐พ1 =
]
โ1 1 , 0 โ1
[ ๐พ2 =
]
2 1 , โ1 2
โก
โ1 1 0 โข 0 โ1 0 ๐พ =โข โฃ0 0 2 0 0 โ1
โค 0 0โฅ โฅ. 1โฆ 2
Parametrizing Structure Preserving Transformations
415
The Segre characteristic is ((2), (1), (1)), the dimension of ๐(๐พ) is 4 and ฮ โ ๐(๐พ) if and only if โค โก ๐ ๐ 0 0 โข0 ๐ 0 0 โฅ โฅ (13) ฮ=โข โฃ0 0 ๐ โ๐โฆ . 0 0 ๐ ๐ Now,
[ ๐1 =
and
1 1
] 0 , 1
[ ๐2 =
] 1 0 , โ2 1
โก 1 โข0 ๐ =โข โฃ0 0
With these matrices
0 0 1 0 โก
1 โข 0 ๐ห = (๐ ๐ )โ1 โข โฃโ1 0
๐ = ๐1
โ
๐2 ,
โค 0 0โฅ โฅ. 0โฆ 1
0 1 0 0 0 0 1 0
0 1 0 2
โค 0 0โฅ โฅ. 0โฆ 1
Now we compute Jordan chains of ๐ฟ(๐). We proceed as in the complex case and ๏ฌnd that [ ] [ ] ๐ ๐ , ๐ฅ11 = , ๐ฅ01 = ๐ ๐ are the Jordan chains of ๐ฟ(๐) for the eigenvalue ๐1 = โ1. Also, [ ] [ ] ๐ 0 , ๐ฅ03 = , ๐ฅ02 = โ๐ 0 are the real Jordan chains of ๐ฟ(๐) for ๐2 = 2 + ๐ and ๐ = 2 โ ๐. Recall that ๐ and ๐ are the real and imaginary parts of any pair of conjugate complex eigenvectors corresponding to the conjugate complex eigenvalues (see Section 2.1). In this example there are real eigenvectors associated with the complex eigenvalues. Provided that ๐, ๐ and ๐ take real values, the matrix [ ] [ ] ๐ ๐ = with ๐ = ๐ฅ01 ๐ฅ11 ๐ฅ02 ๐ฅ03 ๐๐พ satis๏ฌes ๐ โ1 ๐ถ๐
๐ = ๐พ. In particular, โก 1 โข1 ๐ =โข โฃโ1 โ1
if ๐ = ๐ = 1 and ๐ = 0, then โค 0 1 0 0 โ1 0 โฅ โฅ. 1 2 1โฆ 1 โ2 โ1
ห๐หฮ๐ โ1๐ดโ1 and ๐ = ๐หฮ๐ โ1 : Finally, using (13), we compute ๐ = ๐ด
(14)
416
P. Lancaster and I. Zaballa
U = [ 1/2*b+1/2*a, 1/2*b+1/2*a, -1/2*b, -1/2*b [ 1/2*d+1/4*c, -1/2*d-1/4*c, 5/4*d, -5/4*d [ 1/2*b, 1/2*b, 1/2*a-1/2*b, 1/2*a-1/2*b [ -1/4*d, 1/4*d, 1/4*c-1/2*d, -1/4*c+1/2*d
] ] ] ]
V = [ 1/2*b+1/2*a, 1/2*b+1/2*a, 1/2*b, 1/2*b ] [ d+1/2*c, -d-1/2*c, -1/2*d, 1/2*d ] [ -1/2*b, -1/2*b, 1/2*a-1/2*b, 1/2*a-1/2*b ] [ 5/2*d, -5/2*d, 1/2*c-d, -1/2*c+d ] and check that they are real structure preserving transformations, i.e., that (2) holds. โก
4. Hermitian matrix polynomials When a matrix polynomial ๐ฟ(๐) has hermitian coe๏ฌcients the linearization ๐๐ดโ๐ต (as used above) is also hermitian, and this admits reduction of the linearization by congruence transformations โ see (4). Thus, our ๏ฌrst goal is as follows: for two ๐ร๐ ห hermitian matrix polynomials ๐ฟ(๐) and ๐ฟ(๐) of degree โ with nonsingular leading coe๏ฌcients and congruent linearizations, parametrize all matrices ๐ โ Glโ๐ (โ) หโ๐ต ห such that ๐ โ (๐๐ด โ ๐ต)๐ = ๐๐ด
We will ๏ฌrst prove an analogue of Theorem 2.1 and then analogues of Theorems 2.2 and 3.1. However, this problem is more involved because of the presence of the sign characteristic in the canonical form (see [15]). Hermitian matrix polynomials having the same Jordan form and sign characteristic are said to be strictly isospectral. We use the same notation for the set of matrices to be parameterized: ห โ ๐ต}. ห ฮ = {๐ โ โโ๐รโ๐ : ๐ โ (๐๐ด โ ๐ต)๐ = ๐๐ด
(15)
ห are invertible matrices, so are all matrices in ฮ. Notice that since ๐ด and ๐ด In order to prove the analogue of Theorem 2.1 and introduce the set that will play a role similar to the โcentralizerโ, ๐(๐ฝ) of (3), let us recall some results on selfadjoint standard and Jordan triples of hermitian matrix polynomials. If ๐ฟ(๐) has hermitian coe๏ฌcients ๐ด0 , . . . , ๐ดโ , a standard triple (๐, ๐, ๐ ) of ๐ฟ(๐) is said to be selfadjoint if there is an invertible hermitian matrix ๐ โ โโ๐รโ๐ such that ๐ โ = ๐๐ โ1
and ๐ โ = ๐ ๐ ๐ โ1
(16)
Notice that if such a matrix ๐ exists then ๐ โ = ๐ ๐ . It is also noteworthy that if (๐, ๐, ๐ ) is a selfadjoint triple for ๐ฟ(๐) there is one and only one invertible hermitian matrix ๐ satisfying (16) (see [15]). The second property in (16) can be rewritten as ๐ ๐ = ๐ โ ๐ . This means that ๐ is ๐ -selfadjoint; i.e., selfadjoint with respect to the inde๏ฌnite inner product ห๐
are selfadjoint [๐ฅ, ๐ฆ] = (๐ฅ, ๐ ๐ฆ) = ๐ฆ โ ๐ ๐ฅ (see [9, 11]). In particular, ๐ถ๐
and ๐ถ
Parametrizing Structure Preserving Transformations
417
ห respectively (see (8) and notice that for hermitian matrix with respect to ๐ด and ๐ด, โ polynomials ๐ถ๐ฟ = ๐ถ๐
). Now, the analogue of Theorem 2.1 is: ห Theorem 4.1. Let ๐ฟ(๐) and ๐ฟ(๐) be strictly isospectral hermitian matrix polynomials with nonsingular leading coe๏ฌcients. Then ๐ โ ฮ (of (15)) if and only if ห ๐ถ ห๐
, ๐ด หโ1 ๐ ห โ ) is a selfadjoint ห โ โ๐รโ๐ such that (๐, there is a full rank matrix ๐ triple of ๐ฟ(๐) and โก ห โค ๐ โข ๐ ห๐ถ ห๐
โฅ โฅ โข (17) ๐ = โข . โฅ. โฃ .. โฆ ห๐ถ ห โโ1 ๐ ๐
ห (Since the roles of ๐ฟ(๐) and ๐ฟ(๐) can be interchanged in this statement, a ห similar characterization of ๐ can be given in terms of selfadjoint triples of ๐ฟ(๐) โ as in Theorem 2.1.) ห โ ๐ต. ห Then ๐ โ (๐๐ด โ ๐ต) = (๐๐ดห โ ๐ต)๐ ห โ1 . Proof. Assume that ๐ โ (๐๐ด โ ๐ต)๐ = ๐๐ด โ โ1 ห According This means that (๐ , ๐ ) is a block-symmetric SPT of ๐ฟ(๐) and ๐ฟ(๐). ห ห ห to Theorem 2.1(b), there is a standard triple (๐, ๐ถ๐ฟ , ๐ ) of ๐ฟ(๐) such that โก ห โค ๐ โข ๐ ห๐ถ ห๐ฟ โฅ [ ] โข โฅ ห ห๐ฟ ๐ห โ
โ
โ
๐ถ หโโ1 ๐ห . ๐ = โข . โฅ ๐ด, (18) ๐ โ = ๐ห ๐ถ ๐ฟ โฃ .. โฆ ห๐ถ หโโ1 ๐ ๐ฟ
หโ = ๐ด ห๐ถ ห๐
๐ด ห ๐ด, ห ๐ถ ห๐
, ๐ด หโ1 ๐ห ) is a หโ1 , it follows that (๐ ห๐ฟ = ๐ถ Bearing in mind that ๐ถ ๐
standard triple of ๐ฟ(๐) and โค โก ห โค โก ห โค โก ห๐ด ห ๐ ๐ ๐ โข ๐ โข ๐ ห๐ถ ห๐
โฅ ห๐ถ ห๐ฟ โฅ ห๐ฟ ๐ด หโฅ โข ๐ ห๐ด ห๐ด หโ1 ๐ถ โฅ โข โข โฅ ห โข โฅ ๐ = โข . โฅ๐ด =โข โฅ = โข . โฅ, . .. โฆ โฃ .. โฆ โฃ .. โฆ โฃ ห โโ1 ๐ด ห ห๐ถ หโโ1 ห๐ถ ห โโ1 ห๐ด ห๐ด หโ1 ๐ถ ๐ ๐ ๐ ๐ฟ
๐ฟ
๐
ห=๐ ห ๐ด. ห where ๐ หโ1 ๐ห ) is a selfadjoint triple for ๐ฟ(๐); i.e., ห ๐ถ ห๐
, ๐ด We are to prove that (๐, ห โ1 = ๐ ห ๐ด๐ ห โ1 for some invertible hermitian matrix ๐ . Taking หโ1 = ๐๐ ๐ห โ ๐ด ห ห๐ ห โ . But, using (18), ๐ = ๐ด we only have to show that ๐ห = ๐ด โก โค โก โค ๐ผ๐ ๐ผ๐ โข0โฅ ]โข 0 โฅ [ โข โฅ โฅ ห ๐ ห๐ ห โ, ห๐
๐ หโ โ
โ
โ
๐ถ หโ โข หโ ๐ถ หโโ1 ๐ ๐ห = ๐ โ โข . โฅ = ๐ด โข .. โฅ = ๐ด ๐
โฃ .. โฆ โฃ.โฆ 0
as desired.
0
418
P. Lancaster and I. Zaballa
ห โ) ห such that (๐, ห ๐ถ ห๐
, ๐ด หโ1 ๐ Conversely, assume that there is a full row rank matrix ๐ โ1 ห ห ห is a selfadjoint triple of ๐ฟ(๐) and ๐ is given by (17). Put ๐ = ๐ ๐ด . Then หโ1 ๐ ห๐
= ๐ด หโ1 ๐ถ ห โ ) is a standard triple of ๐ฟ(๐) and, because ๐ถ ห๐ฟ ๐ด, ห so is ห๐ด, ห๐ถ ห๐
, ๐ด (๐ ห โ ). But ห ๐ถ ห๐ฟ , ๐ (๐, โก หห โค โก ห โค ๐ ๐๐ด โข ๐ ห๐ถ ห๐ฟ โฅ ห๐ด ห๐ถ ห๐
โฅ โข ๐ โฅ โข โฅ ห โข = ๐ด ๐ =โข โฅ โข .. .. โฅ โฆ โฃ โฃ . . โฆ ห๐ด ห๐ถ หโโ1 ห โโ1 ๐ ๐ห๐ถ ๐
and
[
หโ ๐โ = ๐
ห โ๐ถ ห๐ฟ ๐
๐ฟ
โ
โ
โ
] ห โ๐ถ ห โโ1 . ๐ ๐ฟ
ห ๐ถ ห๐ฟ , ๐ ห โ ) of ๐ฟ(๐) such that Therefore there is a standard triple (๐, โก ห โค ๐ โข ๐ ห๐ถ ห๐ฟ โฅ [ ] โฅ ห โข หโ ๐ ห โ๐ถ ห โ๐ถ ห๐ฟ โ
โ
โ
๐ ห โโ1 . and ๐ โ = ๐ ๐ = โข . โฅ๐ด ๐ฟ โฃ .. โฆ ห๐ถ หโโ1 ๐ ๐ฟ
ห In By Theorem 2.1(b), (๐ โ , ๐ โ1 ) is a block-symmetric SPT for ๐ฟ(๐) and ๐ฟ(๐). other words ห โ1 ๐ โ (๐๐ด โ ๐ต) = (๐๐ดห โ ๐ต)๐ and the theorem is proved. โก 4.1. The set ฮ in terms of canonical structures Now we prove the analogue of Theorem 2.2 concerning hermitian polynomials. If ๐ฟ(๐) is hermitian, ๐ฝ is a Jordan form for ๐ฟ(๐), and ๐ is the corresponding canonical matrix determined by ๐ฝ and the sign characteristic of ๐ฟ(๐) associated with its real eigenvalues (see [15]) then ๐ โ1 = ๐ and ๐ ๐ฝ = ๐ฝ โ ๐ . Now a Jordan triple (๐, ๐ฝ, ๐ ) of ๐ฟ(๐) is selfadjoint if ๐ โ = ๐๐ . The following result is Theorem 1.10 of [7]. It provides some motivation for the introduction of the set that will play the role of the centralizer ๐(๐ฝ) in the hermitian case (cf. [8, Th. 1.25]). Theorem 4.2. Let (๐, ๐ฝ, ๐ ) be a selfadjoint Jordan triple for the hermitian matrix ห ๐ฝ, ๐ห ) is a selfadjoint Jordan triple for ๐ฟ(๐) if and only polynomial ๐ฟ(๐). Then (๐, if there exists a matrix ๐ โ โโ๐รโ๐ such that ๐ โ ๐ ๐ = ๐ and ห = ๐๐, ๐ฝ = ๐ โ1 ๐ฝ๐, ๐ห = ๐ โ1 ๐ ๐ โ . ๐ A matrix ๐ for which ๐ โ ๐ ๐ = ๐ is said to be ๐ -unitary. We de๏ฌne ๐(๐ฝ, ๐ ) = {๐ โ โโ๐รโ๐ : ๐ โ ๐ ๐ = ๐, and ๐๐ฝ = ๐ฝ๐}.
(19)
Thus, members of ๐(๐ฝ, ๐ ) are the ๐ -unitary matrices that commute with ๐ฝ. This is no longer an open set of a linear space (actually it is closed) but it is still a subgroup of Glโ๐ (โ).
Parametrizing Structure Preserving Transformations
419
ห๐
are selfadjoint with respect to ๐ด and ๐ด, ห respecRecall now that ๐ถ๐
and ๐ถ tively, and then ([7, Th. 1.4] or [11, Th. 5.1.1]) there are nonsingular matrices ๐ and ๐ห such that ๐ฝ = ๐ โ1 ๐ถ๐
๐, ๐ = ๐ โ ๐ด๐, (20) โ ห๐ห, ห๐
๐ห. ๐ = ๐ห ๐ด ๐ฝ = ๐หโ1๐ถ Then, with the de๏ฌnition (15) of ฮ: ห Theorem 4.3. Let ๐ฟ(๐) and ๐ฟ(๐) be strictly isospectral ๐ ร ๐ hermitian matrix หโ nonsingular. Let ๐ฝ and ๐ be a pair of polynomials of degree โ with ๐ดโ and ๐ด ห If ๐ and ๐ห are invertible canonical matrices common to both ๐ฟ(๐) and ๐ฟ(๐). matrices satisfying (20) then the map ๐ : ๐(๐ฝ, ๐ ) โ ฮ given by ๐(๐) = ๐ ๐ ๐ห โ1 is well de๏ฌned and bijective. Proof: Since ๐ and ๐ห are invertible matrices we only have to prove that ๐ is well de๏ฌned with a well-de๏ฌned inverse. Let ๐ โ ๐(๐ฝ, ๐ ) and ๐ = ๐ ๐ ๐หโ1. Put ๐ = ๐ โ1 = ๐ห๐ โ1 ๐ โ1 and ๐ = ห ๐ด๐ ๐ดโ1 . Since ๐(๐ฝ, ๐ ) is a group, ๐ โ1 โ ๐(๐ฝ, ๐ ). In particular, ๐ โ1 โ ๐(๐ฝ) and ห ห ๐ต)๐ ห . i.e., ๐ (๐๐ดโ๐ต) = (๐๐ดโ by Theorem 2.2, (๐, ๐ ) is a SPT for ๐ฟ(๐) and ๐ฟ(๐); Let us show that ๐ = ๐ โ . As ๐ = ๐ โ1 this would imply that ๐ โ ฮ. In fact ห ๐ดโ1 ๐ = ๐ด๐
ห๐ห = ๐ โ ๐ด๐ = ๐ ) ห๐ห๐ โ1๐ โ1 ๐ดโ1 , = ๐ด (๐หโ ๐ด = ๐หโโ ๐ ๐ โ1 ๐ โ1 ๐ โ , (๐ โ ๐ ๐ = ๐ ) = ๐หโโ ๐ ๐ โ1 ๐ โ ๐ โ , = ๐หโโ ๐ โ ๐ โ , = (๐ ๐ ๐ห โ1)โ = ๐ โ .
ห and ๐ โ ๐ต๐ = ๐ต. ห This means that Conversely, let ๐ โ ฮ: ๐ โ ๐ด๐ = ๐ด โ1 ห Thus, by Theorem 2.2, (๐ , ๐ ) is a block-symmetric SPT for ๐ฟ(๐) and ๐ฟ(๐). โ1 โ1 โ โ1 โ1 ห ห ห ๐ = ๐ ๐๐ and ๐ = ๐ด๐ ๐๐ ๐ด for some invertible matrix ๐ โ ๐(๐ฝ). But if ๐ โ1 = ๐ห๐๐ โ1 , then ๐ = ๐ ๐ โ1๐หโ1 . Let us show that ๐ โ ๐(๐ฝ, ๐ ). This will conclude the proof because, since ๐(๐ฝ, ๐ ) is a group, ๐ โ1 will also be in ๐(๐ฝ, ๐ ). ห๐ห๐๐ โ1๐ดโ1 . On the one hand, ๐ โ = ๐หโโ ๐ โโ ๐ โ and on the other hand ๐ โ = ๐ด Thus ห๐ห๐๐ โ1๐ดโ1 ๐ โโ . ๐ โโ = ๐หโ ๐ด ห๐ห = ๐ โ ๐ด๐ = ๐ , so that ๐ โโ = ๐ ๐๐ โ1 , and ๐ โ ๐ ๐ = However, we also have ๐หโ ๐ด ๐ as desired. โก โ
There is still the problem of the geometry of ๐(๐ฝ, ๐ ) (we know that ๐(๐ฝ) is a linear space). We will see in the next section that, in the real case, ๐(๐ฝ, ๐ ) may contain a ๏ฌnite number of matrices.
420
P. Lancaster and I. Zaballa
5. Real symmetric matrix polynomials The next two sections concern a special class of hermitian systems, namely, those that have real symmetric coe๏ฌcients โ and, in particular, have linearizations which are diagonalizable by a real congruence transformation. This problem class includes prototypical models of vibration in viscously damped systems. The analysis of this case requires the notion of a real selfadjoint Jordan triple, (๐๐ , ๐พ, ๐ ๐๐๐ ) associated with a real symmetric matrix polynomial ๐ฟ(๐). In particular, the matrix ๐พ is, of course, a real Jordan canonical form (see Theorem 3.4 of [15]). The set of matrices (de๏ฌning real congruence transformations) to be parameterized is now: ห โ ๐ต}. ห ฮ๐ = {๐ โ โโ๐รโ๐ : ๐ ๐ (๐๐ด โ ๐ต)๐ = ๐๐ด (21) ห are invertible, so Notice that ๐ด and ๐ต are real and symmetric and, since ๐ด and ๐ด are all matrices in ฮ๐ . Given a real selfadjoint Jordan triple (๐๐ , ๐พ, ๐ ๐๐๐ ) associated with a real symmetric matrix polynomial ๐ฟ(๐) then (cf. (19)) de๏ฌne ๐(๐พ, ๐ ) = {ฮ โ โโ๐รโ๐ : ฮ๐ ๐ ฮ = ๐, and ฮ๐พ = ๐พฮ}. Thus, members of ๐(๐พ, ๐ ) are the ๐ -orthogonal matrices that commute with ห are real, ๐พ. As before, this is a subgroup of Glโ๐ (โ). Then, if ๐ฟ(๐) and ๐ฟ(๐) symmetric, and strictly isospectral (have the same canonical matrices ๐พ and ๐ ), then (see Section 4 of [15]) there are real nonsingular matrices ๐ and ๐ห such that ๐ ๐ ๐ด๐ = ๐, ห๐ห, = ๐ ๐ห๐ ๐ด
๐พ = ๐ โ1๐ถ๐
๐, ห๐
๐ห. ๐พ = ๐หโ1 ๐ถ
(22)
The results for real symmetric matrix polynomials are direct analogues of those for hermitian matrix polynomials: ห Theorem 5.1. Let ๐ฟ(๐) and ๐ฟ(๐) be strictly isospectral real symmetric matrix polyห โ โ๐รโ๐ such nomials. Then ๐ โ ฮ๐ if and only if there is a full rank matrix ๐ โ1 ๐ ห ๐ ห ) is a real selfadjoint triple of ๐ฟ(๐) and ห ๐ถ ห๐
, ๐ด that (๐, โก ห โค ๐ โข ๐ ห๐ถ ห๐
โฅ โข โฅ ๐ =โข . โฅ โฃ .. โฆ ห๐ถ หโโ1 ๐ ๐
ห Theorem 5.2. Let ๐ฟ(๐) and ๐ฟ(๐) be strictly isospectral ๐ร๐ real symmetric matrix หโ nonsingular. Let ๐พ and ๐ be the common polynomials of degree โ with ๐ดโ and ๐ด canonical forms for these matrix polynomials (as above). If ๐ and ๐ห are invertible real matrices satisfying (22) then the map ๐ : ๐(๐พ, ๐ ) โ ฮ๐ given by ๐(ฮ) = ๐ ฮ๐หโ1 is well de๏ฌned and bijective.
Parametrizing Structure Preserving Transformations
421
Given the existence of real selfadjoint triples, the proofs are essentially the same as those of Theorems 4.1 and 4.3, respectively. It is only necessary to con๏ฌne the argument to real matrices. Example 5.1. We apply the theory above to the systems of Example 3.2: ] ] [ ] [ [ 3/2 โ1/2 2 11/2 โ9/2 โ3 5 ๐ + , ๐ฟ(๐) = ๐+ โ1/2 3/2 โ9/2 11/2 5 โ3 and
[ 1 ห ๐ฟ(๐) = ๐2 0
] [ 2 ] ] [ ] [ ๐ + 2๐ + 1 0 1 0 2 0 0 = . + +๐ 0 ๐2 โ 4๐ + 5 0 5 0 โ4 1
We already know that they are isospectral. But they are also strictly isospectral systems. In fact, the only real eigenvalue is โ1 and its Segre characteristic is (2). In order to compute the sign characteristic of the elementary divisor (๐ + 1)2 we can use Theorem 3.7 of [7]. It turns out that the sign characteristic of (๐ + 1)2 in both matrices is +1. Thus, the common real Jordan form and sip matrix for these systems are โค โค โก โก 0 1 0 0 โ1 1 0 0 โข1 0 0 0โฅ โข 0 โ1 0 0โฅ โฅ โฅ and ๐ =โข ๐พ =โข โฆ โฃ0 0 0 1โฆ . โฃ0 0 2 1 0 0 1 0 0 0 โ1 2 The matrices in ๐(๐พ) have the form given in (13). Hence ฮ โ ๐(๐พ, ๐ ) if and only if ฮ has the form in (13) and ฮ๐ ๐ ฮ = ๐ . A simple computation shows that, in this case, โก โค 0 ๐2 0 0 โข๐2 2๐๐ 0 0 โฅ โฅ. ฮ๐ ๐ ฮ = โข 2 โฃ0 0 2๐๐ ๐ โ ๐2 โฆ 0 0 ๐2 โ ๐2 โ2๐๐ Thus ฮ๐ ๐ ฮ = ๐ if and only if ๐2 = 1, ๐2 = 1 and ๐ = ๐ = 0. This reveals, for example, that there are only 4 distinct matrices in ๐(๐พ, ๐ ). Next we have to ๏ฌnd matrices ๐ and ๐ห satisfying (22). We already have a matrix ๐ (cf. (14)) for which ๐ โ1 ๐ถ๐
๐ = ๐พ. It turns out that any other matrix satisfying this condition must be of the form ๐๐ป with ๐ป โ ๐(๐พ) and invertible. Recalling (13), we compute โก โค 0 2๐2 0 0 โข2๐2 4๐๐ โฅ 0 0 โฅ. (๐๐ป)๐ ๐ด(๐๐ป) = โข โฃ 0 0 8๐๐ 4๐2 โ 4๐2 โฆ 0 0 4๐2 โ 4๐2 โ8๐๐
422
P. Lancaster and I. Zaballa
Thus (๐๐ป)๐ ๐ด(๐๐ป) = ๐ if and only if ๐2 = 1/2, ๐ = 0, ๐2 = 1/4 and ๐ = 0. โ Taking ๐ = 1/ 2 and ๐ = 1/2, for example, we have โก โ โค 1/ 2 0โ 0 0 โข 0 1/ 2 0 0 โฅ โฅ โ ๐(๐พ), ๐ป=โข โฃ 0 0 1/2 0 โฆ 0 0 0 1/2 and a matrix ๐ such that ๐ ๐ ๐ด๐ = ๐ and ๐ โ1 ๐ถ๐
๐ = ๐พ is ๐ = ๐๐ป. With the above ๐ป โกโ โค 0 1/2 0 โ2/2 โข 2/2 โ1/2 0 โฅ โฅ. โ โ0 ๐ =โข โฃโ 2/2 โฆ 2/2 1 1/2 โ โ โ 2/2 2/2 โ1 โ1/2 ห The same procedure applied to ๐ฟ(๐) shows that, for example, โก โค 1 0 0 0 โข 0 0 1 0โฅ โฅ ๐ห = โข โฃโ1 1 0 0โฆ 0 0 2 1 ห๐ห = ๐ and ๐หโ1 ๐ถ ห๐
๐ห = ๐พ. Notice that this matrix was also obtained satis๏ฌes ๐ห๐ ๐ด by the standard procedure detailed in Section 3.1. ห Thus, all SPCs for ๐ฟ(๐) and ๐ฟ(๐) have the form โกโ โค 2 1 ๐ ๐ 0 0 2 2 โข โ2 โฅ 0 โฅ โข 2 ๐ โ 12 ๐ โ0 โ1 ห ๐ = ๐ ฮ๐ = โข (23) โฅ 2 โฃ 0 0 ๐ 12 ๐ โฆ โ2 2 1 0 0 2 ๐ โ2๐ with ๐2 = 1 and ๐2 = 1. One can easily check that โก 2 โค 2๐ โ 2 0 ๐2 โ 1 0 โข 4 โ 4๐2 0 ๐2 โ 1โฅ ห = โข 20 โฅ, ๐ ๐ ๐ด๐ โ ๐ด โฃ๐ โ1 0 0 0 โฆ 0 ๐2 โ 1 0 0 and
โก 1 โ ๐2 โข ห=โข 0 ๐ ๐ ๐ต๐ โ ๐ต โฃ 0 0
0 5 โ 5๐2 0 0
โค 0 0 0 0 โฅ โฅ, ๐2 โ 1 0 โฆ 0 ๐2 โ 1
which reduce to zero when ๐2 = 1 and ๐2 = 1. Again, only four real SPCs reduce ห ๐ฟ(๐) to the diagonal strictly isospectral system ๐ฟ(๐). It is worth noting that the system ๐ฟ(๐) of this example satis๏ฌes the conditions of Theorem 2 in [14] for being decoupled (diagonalized) by congruence. For, if we write ๐ฟ(๐) in the form ๐ฟ(๐) = ๐ ๐2 +๐ท๐+๐พ then the eigenvalues of ๐๐ +๐พ are 1
Parametrizing Structure Preserving Transformations
423
and 5 so that it is of de๏ฌnite type, and one can check that ๐ท๐ โ1 ๐พ = ๐พ๐ โ1 ๐ท. Hence there is a nonsingular matrix ๐ such that ห ๐ โ ๐ฟ(๐)๐ = ๐ฟ(๐).
(24)
It turns out that our procedure to construct the SPCs for these two systems produces some matrices ๐ satisfying (24). In fact, one can check (see [4]) that the relation ห โ1 ๐ ๐ (๐๐ด โ ๐ต) = (๐๐ดห โ ๐ต)๐ implies ๐ ๐ ห (๐12 ๐ + ๐11 )๐ฟ(๐) = ๐ฟ(๐)(๐ 12 ๐ + ๐11 ) โ1 where ๐ = [๐๐๐ ]๐,๐=1,2 and ๐ = [๐๐๐ ]๐,๐=1,2 . But from (23) we have ๐12 = 0, โ1 ๐12 = 0 and ๐11 = ๐11 . Therefore ๐ ห ๐11 ๐ฟ(๐)๐11 = ๐ฟ(๐),
with
[โ ๐11 =
2 ๐ โ2 2 2 ๐
]
1 2๐ โ 21 ๐
,
๐2 = ๐2 = 1,
de๏ฌnes a strict real congruence between the two symmetric systems. Whether this is a viable procedure for constructing all possible strict real congruences between two given systems remains an open question. โก Acknowledgement The authors are grateful for support and encouragement received from Seamus D. Garvey, Uwe Prells, and Atanas Popov โ partners in the project supported by the EPSRC (UK) under Grant EP/E046290.
References [1] Gantmacher F.R., The Theory of Matrices, vol 1. AMS Chelsea, Providence, Rhode Island, 1998. [2] Garvey S.G., Friswell M.I., and Prells U., Co-ordinate transformations for secondorder systems, Part 1: General transformations, J. Sound and Vibration, 285, 2002, 885โ909. [3] Garvey S.G., Friswell M.I., Prells U., and Chen Z., General isospectral ๏ฌows for linear dynamic systems, Lin. Alg. and its Applications, 385, 2004, 335โ368. [4] Garvey S.G., Lancaster P., Popov A., Prells U., Zaballa I., Filters Connecting Isospectral Quadratic Systems. Preprint. [5] Chu M., and Del Buono N., Total decoupling of a general quadratic pencil, Part 1, J. Sound and Vibration, 309, 2008, 96โ111. [6] Chu M., and Xu S.F., Spectral decomposition of real symmetric quadratic ๐-matrices and its applications, Math. Comp., 78, 2009, 293โ313. [7] Gohberg I., Lancaster P., and Rodman L., Spectral analysis of selfadjoint matrix polynomials, Ann. of Math., 112, 1980, 33โ71.
424
P. Lancaster and I. Zaballa
[8] Gohberg I., Lancaster P., and Rodman L., Matrix Polynomials Academic Press, New York, 1982, and SIAM, Philadelphia, 2009. [9] Gohberg I., Lancaster P., and Rodman L., Matrices and Inde๏ฌnite Scalar Products, Birkhยจ auser, Basel, 1983. [10] Gohberg I., Lancaster P., and Rodman L., Invariant Subspaces of Matrices with Applications, Wiley, New York, 1986 and SIAM, Philadelphia, 2006. [11] Gohberg I., Lancaster P., and Rodman L., Inde๏ฌnite Linear Algebra and Applications, Birkhยจ auser, Basel, 2005. [12] Lancaster P., and Tismenetsky M., The Theory of Matrices, Academic Press, New York, 1985. [13] Lancaster P., and Prells U., Isospectral families of high-order systems, Z. Angew. Math. Mech, 87, 2007, 219โ234. [14] Lancaster P., and Zaballa I., Diagonalizable quadratic eigenvalue problems, Mechanical Systems and Signal Processing, 23, 2009, 1134โ1144. [15] Lancaster P., and Zaballa I., A review of canonical forms for selfadjoint matrix polynomials. Submitted. Peter Lancaster Dept. of Mathematics and Statistics University of Calgary Calgary, AB T2N 1N4, Canada e-mail:
[email protected] Ion Zaballa Departamento de Matematica Aplicada y EIO Universidad del Pais Vasco Apdo 644 E-48080 Bilbao, Spain e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 425โ443 c 2012 Springer Basel AG โ
A Review of Canonical Forms for Selfadjoint Matrix Polynomials Peter Lancaster and Ion Zaballa Dedicated to the memory of Israel Gohberg, good friend and scholar
Abstract. In the theory of ๐ ร ๐ matrix polynomials, the notions of โstandard pairs and triplesโ, and the special cases of โJordan pairs and triplesโ play an important role. There are interesting di๏ฌerences in these constructions according as the analysis is carried out over the complex ๏ฌeld โ, or the real ๏ฌeld โ. A careful review is made of these ideas with special reference to complex hermitian systems, and to real symmetric systems with nonsingular leading coe๏ฌcient. New results are obtained concerning real Jordan structures for real symmetric matrix polynomials. Mathematics Subject Classi๏ฌcation (2000). 15A21, 15A54, 47B15. Keywords. Matrix polynomials, canonical forms.
1. Introduction Standard and Jordan triples for matrix polynomials and their โselfadjointโ forms, were introduced and developed by Gohberg, Lancaster, and Rodman (GLR) in several publications including [3, 4, 8]. In the ๏ฌrst two of these works (of thirty years ago) selfadjoint structures are de๏ฌned for polynomials de๏ฌned over โ and are formulated in terms of canonical forms over โ. Following a lead given more recently in [8], we separate the โselfadjointโ and โcanonicalโ notions and provide careful distinction between systems de๏ฌned over either โ or โ. Also, we take advantage of the comprehensive discussion of canonical forms provided in [11]. Less comprehensive discussions can be found in the GLR works but they are scattered and incomplete. It is our objective in this paper to give a concise and largely self-contained overview of these ideas. For convenience, some of the This work was supported by grants from the EPSRC (United Kingdom), NSERC (Canada), and DGICYT, GV (Spain).
426
P. Lancaster and I. Zaballa
necessary arguments are repeated, but some are new. Our main objective is the development of arguments leading to Theorem 4.3 and (especially) Theorem 4.4 concerning Jordan triples. In Section 5 constructions for chains of generalized eigenvectors are presented in the light of preceding results. โ๐=โ We consider ๐ ร ๐ matrix polynomials of degree โ: say ๐ฟ(๐) = ๐=0 ๐ด๐ ๐๐ with ๐ดโ nonsingular, and all coe๏ฌcient matrices ๐ด๐ โ โ๐ร๐ , or all in โ๐ร๐ . The ๏ฌrst of these is the setting for the greater part of the GLR theory. Where the distinction is not important, we use the symbol ๐ฝ to denote either the ๏ฌeld of real or the ๏ฌeld of complex numbers. In particular, this paper seems to provide the ๏ฌrst comprehensive account of real canonical structures (for real symmetric systems) โ with no hypotheses on the degrees of elementary divisors, and making no positive de๏ฌnite hypotheses on any of the coe๏ฌcients ๐ด๐ . We note that Chu et al. in [1], [2] and [13] have recently used partial canonical structures in studying inverse problems and model updating for some real quadratic systems.
2. Standard pairs and triples Let ๐ฟ(๐) be a an ๐ ร ๐ matrix polynomial over ๐ฝ with nonsingular leading coe๏ฌcient ๐ดโ . Let ๐ถ๐
be the right companion matrix of ๐ฟ(๐), namely, โค โก 0 โ
โ
โ
0 0 ๐ผ๐ โฅ โข 0 0 ๐ผ๐ โ
โ
โ
0 โฅ โข โฅ โข .. .. (1) ๐ถ๐
= โข โฅ, . . โฅ โข โฆ โฃ 0 ๐ผ๐ โ๐ดโ1 โ๐ดโ1 โ๐ดโ1 โ ๐ด0 โ ๐ด1 . . . โ ๐ดโโ1 and de๏ฌne the โblock symmetricโ matrix ๐ด by โก โค ๐ด1 ๐ด2 โ
โ
โ
๐ดโ โข ๐ด2 โ
โ
โ
๐ดโ 0 โฅ โข โฅ ๐ด=โข . .. โฅ . โฃ .. . โฆ ๐ดโ 0 โ
โ
โ
0
(2)
Thus, both ๐ถ๐
and ๐ด are in ๐ฝโ๐รโ๐. Also, ๐ด is nonsingular and if the coe๏ฌcient matrices of ๐ฟ(๐) are real and symmetric, or hermitian, then ๐ดโ = ๐ด so that ๐ด is real and symmetric, or hermitian, according as ๐ฝ = โ or โ. The product ๐ด๐ถ๐
is also block-symmetric, and if ๐ฟ(๐) is hermitian or real symmetric, ๐ด๐ถ๐
is also hermitian or real symmetric, respectively. For this reason ๐ด is sometimes known as the โsymmetrizerโ for ๐ถ๐
(it de๏ฌnes an inde๏ฌnite inner product on ๐ฝโ๐รโ๐ in โ which ๐ถ๐
is selfadjoint; i.e., ๐ด๐ถ๐
= ๐ถ๐
๐ด).
Canonical Forms De๏ฌnition 2.1. (a) A pair of matrices ๐ โ ๐ฝ๐รโ๐ and ๐ if โก ๐ โข ๐๐ โข โข .. โฃ .
427
โ ๐ฝโ๐รโ๐ form a standard pair over ๐ฝ โค โฅ โฅ โฅ โฆ
๐๐ โโ1
is nonsingular. (b) A standard pair (๐, ๐ ) is a standard pair for ๐ฟ(๐) if [ ] ๐ = ๐ผ 0 โ
โ
โ
0 ๐, and ๐ = ๐ โ1 ๐ถ๐
๐ for some nonsingular ๐ โ ๐ฝโ๐รโ๐ . Theorem 2.2. A standard pair (๐, ๐ ) is a standard pair for ๐ฟ(๐) if and only if ๐ฟ(๐, ๐ ) := ๐ดโ ๐๐ โ + โ
โ
โ
+ ๐ด1 ๐๐ + ๐ด0 ๐ = 0. This is Proposition 12.1 of [8]. De๏ฌnition 2.3. (a) Given a standard pair (๐, ๐ ) over ๐ฝ, if โคโ1 โก โก ๐ โข ๐๐ โฅ โข โฅ โข โข ๐ =โข โฅ โข .. โฆ โฃ โฃ . โโ1 ๐๐
โค 0 .. โฅ . โฅ โฅ, 0 โฆ ๐
(3)
for some nonsingular matrix ๐ โ ๐ฝ๐ร๐ then (๐, ๐, ๐ ) is said to be a standard triple (over ๐ฝ). (b) If (๐, ๐ ) is a standard pair for ๐ฟ(๐) and ๐ is de๏ฌned as in (a) with ๐ = ๐ดโ1 โ then (๐, ๐, ๐ ) is said to be a standard triple for ๐ฟ(๐) (over ๐ฝ). Throughout this paper, when saying that (๐, ๐, ๐ ) is a standard triple it is to be understood that, unless speci๏ฌed otherwise, ๐ โ ๐ฝ๐รโ๐ , ๐ โ ๐ฝโ๐รโ๐ and ๐ โ ๐ฝโ๐ร๐ . If (๐, ๐, ๐ ) is a standard triple for ๐ฟ(๐) then (see [4, Prop. 2.1]): [ ] (i) ๐ ๐ ๐ โ
โ
โ
๐ โโ1 ๐ is nonsingular. (ii) ๐ โ[๐ ๐ดโ + ๐ โโ1 ๐ ๐ดโโ1 + โ
โ
]โ
+ [๐ ๐ ๐ด1 + ๐ ๐ด0 = 0, ] (iii) ๐ ๐ ๐ ๐ โ
โ
โ
๐ โโ1 ๐ = 0 โ
โ
โ
0 ๐ดโ1 . โ
This implies that (๐ ๐ , ๐ ๐ , ๐ ๐ ) is a standard triple of ๐ฟ(๐)๐ . The pair (๐, ๐ ) is called a left standard pair of ๐ฟ(๐). The prime example of a standard triple for polynomial ๐ฟ(๐) is โก โค 0 โข .. โฅ [ ] โข โฅ ๐0 = ๐ผ 0 โ
โ
โ
0 , ๐ = ๐ถ๐
, ๐0 = โข . โฅ . (4) โฃ 0 โฆ ๐ดโ1 โ
428
P. Lancaster and I. Zaballa
This is a standard triple for the matrix polynomial ๐ฟ(๐) whose leading coe๏ฌcient is ๐ดโ and the remaining coe๏ฌcients are ๐ดโ times those appearing in the last block row of ๐ถ๐
; i.e., ๐ฟ(๐) = ๐ดโ ๐โ + ๐ดโโ1 ๐โโ1 + โ
โ
โ
+ ๐ด1 ๐ + ๐ด0 . In other words, all matrix polynomials with nonsingular leading coe๏ฌcient admit standard triples. The converse is also true. The proof is based on the fact that if (๐, ๐, ๐ ) is a standard triple for ๐ฟ(๐) then the resolvent form holds: ๐ฟ(๐)โ1 = ๐(๐ผโ๐ ๐ โ ๐ )โ1 ๐,
๐โ / ๐(๐ฟ);
(5)
๐(๐ฟ) being the spectrum (i.e., the set of eigenvalues) of ๐ฟ(๐) (Theorem 14.2 of [12]). Theorem 2.4. If (๐, ๐, ๐ ) is a standard triple with ๐ โ ๐ฝ๐รโ๐ , ๐ โ ๐ฝโ๐รโ๐ and ๐ โ ๐ฝโ๐ร๐ then there is a unique matrix polynomial ๐ฟ(๐) for which (๐, ๐, ๐ ) is a standard triple. Proof. By De๏ฌnition 2.3(a), if (๐, ๐, ๐ ) is a standard triple then there is an invertible matrix ๐ such that { 0 for ๐ = 0, 1, . . . , โ โ 2 ๐ ๐๐ ๐ = ๐ for ๐ = โ โ 1 Thus
โก
๐๐ ๐๐ ๐ .. .
๐๐ ๐ ๐๐ 2 ๐ .. .
โ
โ
โ
โ
โ
โ
.. .
๐๐ โโ1 ๐ ๐๐ โ๐ .. .
โค
โข โฅ โข โฅ rank โข โฅ โฃ โฆ โโ1 โโ2 2โโ2 ๐๐ ๐ ๐๐ ๐ โ
โ
โ
๐๐ โก โค๐ 0 โ
โ
โ
0 ๐ โข0 โ
โ
โ
๐ ๐๐ โ ๐ โฅ โข โฅ = rank โข . โฅ = โ๐ . .. .. โฃ .. โฆ . ๐ โ
โ
โ
๐๐ 2โโ3 ๐ ๐๐ 2(โโ1) ๐ By Theorem 2.8 of [4]1 there is a matrix polynomial ๐ฟ(๐) such that ๐ฟ(๐)โ1 = ๐(๐๐ผโ๐ โ ๐ )โ1 ๐ and by Theorem 14.2.4 of [12] (๐, ๐, ๐ ) is a standard triple for ๐ฟ(๐). This is the only matrix polynomial for which (๐, ๐, ๐ ) is a standard triple ห ห โ1 = ๐(๐๐ผโ๐ โ because, if this triple were a standard triple for ๐ฟ(๐), then ๐ฟ(๐) ๐ )โ1 ๐ = ๐ฟ(๐)โ1 . โก It should be noted that the coe๏ฌcients of ๐ฟ(๐) can be expressed2 in terms of a standard triple for ๐ฟ(๐) (this is Theorem 14.7.1 of [12] and Theorem 2.4 of [4]) and so if (๐, ๐, ๐ ) is real, ๐ฟ(๐) is real too. Although a standard triple de๏ฌnes a matrix polynomial uniquely, a matrix polynomial generally admits many standard triples. The relationship between two standard triples for the same matrix polynomial is clari๏ฌed in the following theorem: 1 After
straightforward generalization to admit nonsingular ๐ดโ possibly di๏ฌerent from ๐ผ. plays an important part in strategies for solving inverse problems in which the coe๏ฌcients are expressed in terms of spectral data. See [9] and [13], for example. 2 This
Canonical Forms
429
Theorem 2.5. If (๐, ๐, ๐ ) is a standard triple for ๐ฟ(๐) over ๐ฝ, and if a triple of matrices (๐1 , ๐1 , ๐1 ) is similar to (๐, ๐, ๐ ) in the sense that ๐1 = ๐๐ โ1,
๐1 = ๐๐ ๐ โ1,
๐1 = ๐๐
(6)
for some invertible ๐ โ ๐ฝโ๐รโ๐ , then (๐1 , ๐1 , ๐1 ) is also a standard triple for ๐ฟ(๐) over ๐ฝ. Conversely, any two standard triples for ๐ฟ(๐) over ๐ฝ are similar. This is Proposition 12.1.3 of [8]. Using the standard triple (4), we note that (๐, ๐, ๐ ) is a standard triple for ๐ฟ(๐) if and only if there is a nonsingular matrix ๐ such that โก โค 0 โข .. โฅ ] [ โข โฅ ๐ = ๐ผ๐ 0 โ
โ
โ
0 ๐, ๐ = ๐ โ1 ๐ถ๐
๐, ๐ = ๐ โ1 โข . โฅ . โฃ 0 โฆ ๐ดโ1 โ It turns out that the matrix ๐ satisfying (6) is uniquely determined either by (๐, ๐ ) and (๐1 , ๐1 ), or by (๐, ๐ ) and (๐1 , ๐1 ). It is given by (see [4, Th. 1.25]): ๐ = ๐ถ(๐1 , ๐1 )โ1 ๐ถ(๐, ๐ ) or ๐ = ๐
(๐1 , ๐1 )๐
(๐, ๐ )โ1 where
โก
๐ ๐๐ .. .
(7)
โค
โฅ โข โฅ โข ๐ถ(๐, ๐ ) = โข โฅ โฆ โฃ โโ1 ๐๐
[ and ๐
(๐, ๐ ) = ๐
๐๐
โ
โ
โ
] ๐ โโ1 ๐ .
(8)
It follows from (7) that ๐ถ(๐, ๐ )๐
(๐, ๐ ) = ๐ถ(๐1 , ๐1 )๐
(๐1 , ๐1 ). Furthermore, if ๐ด is the block-symmetric matrix of (2), then for any standard triple (๐, ๐, ๐ ) of ๐ฟ(๐) we have (see [12, Th. 14.2.5]) ๐ดโ1 = ๐ถ(๐, ๐ )๐
(๐, ๐ ) = ฮ, where
โก โข โข ฮ=โข โฃ
0 0 .. .
ฮโโ1
โ
โ
โ
โ
โ
โ
โ
โ
โ
0
ฮโโ1 .. .
ฮ2โโ3
ฮโโ1 ฮโ .. .
(9) โค โฅ โฅ โฅ โฆ
(10)
ฮ2(โโ1)
is (by de๏ฌnition) the matrix of the moments of ๐ฟ(๐); i.e., the 2โโ2 ๏ฌrst coe๏ฌcients of the resolvent expansion for ๐ฟ(๐)โ1 .
430
P. Lancaster and I. Zaballa
3. Standard triples and hermitian systems Now we consider the notions of standard pairs and triples in the context of real symmetric or complex hermitian matrix polynomials. Using the resolvent form, the following result is easily proved (see Corollary 14.2.1 of [12]): Theorem 3.1. (a) A real matrix polynomial ๐ฟ(๐) has symmetric coe๏ฌcients if and only if, for any real standard triple (๐, ๐, ๐ ) for ๐ฟ(๐), (๐ ๐ , ๐ ๐ , ๐ ๐ ) is also a standard triple for ๐ฟ(๐). (b) A complex matrix polynomial ๐ฟ(๐) has hermitian coe๏ฌcients if and only if, for any complex standard triple (๐, ๐, ๐ ) for ๐ฟ(๐), (๐ โ , ๐ โ , ๐ โ ) is also a standard triple for ๐ฟ(๐). The following de๏ฌnitions are critical โ and may be unfamiliar. Observe that the statements make no reference to matrix polynomials. De๏ฌnition 3.2. (a) A real standard triple (๐, ๐, ๐ ) is said to be real selfadjoint if there is a real nonsingular symmetric matrix ๐ป for which ๐ ๐ = ๐๐ป โ1 ,
๐ ๐ = ๐ป๐ ๐ป โ1,
๐ ๐ = ๐ป๐.
(11)
(b) A complex standard triple (๐, ๐, ๐ ) is said to be selfadjoint if there is a nonsingular hermitian matrix ๐ป for which ๐ โ = ๐๐ป โ1,
๐ โ = ๐ป๐ ๐ป โ1 ,
๐ โ = ๐ป๐.
(12)
Note that, because of the symmetry imposed on ๐ป, the ๏ฌrst and third of the relations in (11) and (12) are equivalent. Note also that this is not the same use of โselfadjoint tripleโ as that of [4, p. 261] but an elementary adaptation of the de๏ฌnition given in [8, p. 244] for hermitian polynomial matrices. The following example shows that โreal selfadjoint standard triplesโ may not be recognizable by inspection. Example 3.3. Let ๐ฟ(๐) be a real matrix polynomial and consider a standard triple of the form (4). It is, of course, a real standard triple if ๐ฟ(๐) has real coe๏ฌcients. Furthermore, if ๐ฟ(๐) is real and symmetric and we de๏ฌne ๐ป = ๐ด (the blocksymmetric matrix of (2)), then it can be veri๏ฌed that (11) holds, i.e., this standard triple is real selfadjoint. Let us take a closer look at the de๏ฌnition of selfadjoint triples. We will focus on the real case but using the translation: โreal โ complexโ, โsymmetric โ hermitianโ and โ๐ โ โ โ the same results and proofs hold for complex matrices. First we show that it is not necessary to require that ๐ป be symmetric in De๏ฌnition 3.2 of a selfadjoint triple. Theorem 3.4. A real standard triple (๐, ๐, ๐ ) is selfadjoint if and only if it is similar to (๐ ๐ , ๐ ๐ , ๐ ๐ ).
Canonical Forms
431
Proof. It is clear from (11) that if (๐, ๐, ๐ ) is selfadjoint then it is similar to (๐ ๐ , ๐ ๐ , ๐ ๐ ). For the converse, assume that there is an invertible matrix ๐ป such that (11) holds. We are to prove that ๐ป is symmetric. As ๐ ๐ = ๐ป๐ ๐ป โ1 and ๐ ๐ = ๐ป๐ it follows that (with ๐
as in (8)) ๐
(๐ ๐ , ๐ ๐ ) = ๐ป๐
(๐, ๐ ) and so
๐ป = ๐
(๐ ๐ , ๐ ๐ )๐
(๐, ๐ )โ1 = ๐ถ(๐, ๐ )๐ ๐
(๐, ๐ )โ1 .
But also ๐ ๐ = ๐๐ป โ1 . Then ๐ถ(๐ ๐ , ๐ ๐ ) = ๐ถ(๐, ๐ )๐ป โ1 . That is to say, ๐ป = ๐ถ(๐ ๐ , ๐ ๐ )โ1 ๐ถ(๐, ๐ ) = ๐
(๐, ๐ )โ๐ ๐ถ(๐, ๐ ). These two expressions for ๐ป yield ๐ป = ๐ป ๐ .
โก
It follows from Theorems 3.4 and 3.1 that all real standard triples of a real symmetric matrix polynomial are selfadjoint (compare with [8, Th. 12.2.2]): Theorem 3.5. Let ๐ฟ(๐) have real coe๏ฌcients with ๐ดโ nonsingular. Then: (a) If ๐ฟ(๐) admits a real selfadjoint standard triple then it is real and symmetric. (b) If ๐ฟ(๐) is real and symmetric then all its real standard triples are selfadjoint. Proof. (a) If (๐, ๐, ๐ ) is a real selfadjoint standard triple for matrix polynomial ๐ฟ(๐) then, by Theorem 3.4, (๐, ๐, ๐ ) and (๐ ๐ , ๐ ๐ , ๐ ๐ ) are similar. Hence, by Theorem 2.5, (๐ ๐ , ๐ ๐ , ๐ ๐ ) is also a standard triple of ๐ฟ(๐). Therefore, by Theorem 3.1, the coe๏ฌcients of ๐ฟ(๐) are symmetric. (b) If ๐ฟ(๐) is a real selfadjoint matrix polynomial and (๐, ๐, ๐ ) is a real standard triple for ๐ฟ(๐) then (๐ ๐ , ๐ ๐ , ๐ ๐ ) is also a standard triple and, according to Theorem 2.5, (๐, ๐, ๐ ) and (๐ ๐ , ๐ ๐ , ๐ ๐ ) are similar. Then Theorem 3.4 implies that (๐, ๐, ๐ ) is selfadjoint. โก Recall that the similarity matrix for two similar standard triples (๐, ๐, ๐ ) and (๐1 , ๐1 , ๐1 ) is given by (7). In the selfadjoint case (๐ ๐ , ๐ ๐ , ๐ ๐ ) plays the role of (๐1 , ๐1 , ๐1 ). Thus: Proposition 3.6. If (๐, ๐, ๐ ) is a real selfadjoint triple then there is one and only one nonsingular real symmetric matrix ๐ป such that ๐ ๐ = ๐๐ป โ1 ,
๐ ๐ = ๐ป๐ ๐ป โ1,
๐๐๐
๐ ๐ = ๐ป๐.
This matrix is given by any of the four equivalent expressions: (i) (ii) (iii) (iv)
๐ป ๐ป ๐ป ๐ป
= ๐
(๐, ๐ )โ๐ ๐ถ(๐, ๐ ), = ๐ถ(๐, ๐ )๐ ๐
(๐, ๐ )โ1 , = ๐ถ(๐, ๐ )๐ ๐ด๐ถ(๐, ๐ ), = ๐
(๐, ๐ )โ๐ ฮ๐
(๐, ๐ )โ1 ,
where ๐ด is the matrix (2) of the coe๏ฌcients of the unique matrix ๐ฟ(๐) for which (๐, ๐, ๐ ) is a selfadjoint triple and ฮ is the matrix (10) of its moments.
432
P. Lancaster and I. Zaballa
Proof. Everything is known but the expressions for ๐ป of items (iii) and (iv). Recall that (cf. (9)) ๐ดโ1 = ๐ถ(๐, ๐ )๐
(๐, ๐ ) = ฮ. Thus
๐ป = ๐ถ(๐, ๐ )๐ ๐
(๐, ๐ )โ1 = ๐ถ(๐, ๐ )๐ ๐ด๐ถ(๐, ๐ )
and also ๐ป = ๐
(๐, ๐ )โ๐ ๐ถ(๐, ๐ ) = ๐
(๐, ๐ )โ๐ ฮ๐
(๐, ๐ )โ1 .
โก
The expressions for ๐ป in items (iii) and (iv) reveal the symmetric structure of ๐ป more clearly, because both ๐ด and ฮ are symmetric provided that ๐ฟ(๐) is symmetric. The following result is an easy consequence of Proposition 3.6: Corollary 3.7. (a) Let ๐ฟ(๐) be a real matrix polynomial with nonsingular leading coe๏ฌcient, and let (๐, ๐, ๐ ) be a real standard triple for ๐ฟ(๐). Then ๐ฟ(๐) is symmetric if and only if ๐ป = ๐ถ(๐, ๐ )๐ ๐
(๐, ๐ )โ1 is symmetric. (b) A standard triple (๐, ๐, ๐ ) is selfadjoint if and only if ๐ป is symmetric. Proof. Note ๏ฌrst that, from (9), we have ๐
(๐, ๐ )โ1 = ๐ด๐ถ(๐, ๐ ). Now we systematically use the fact that ๐ถ(๐, ๐ )๐ ๐
(๐, ๐ )โ1 = ๐ถ(๐, ๐ )๐ ๐ด๐ถ(๐, ๐ ). (a) If ๐ฟ(๐) is symmetric then, since ๐ด is symmetric, ๐ป = ๐ถ(๐, ๐ )๐ ๐
(๐, ๐ )โ1 = ๐ถ(๐, ๐ )๐ ๐ด๐ถ(๐, ๐ ) is symmetric too. Conversely, if ๐ป = ๐ถ(๐, ๐ )๐ ๐
(๐, ๐ )โ1 = ๐ถ(๐, ๐ )๐ ๐ด๐ถ(๐, ๐ ) is symmetric then ๐ด is symmetric and it is plain that ๐ด is symmetric if and only if ๐ด๐ is symmetric for ๐ = 0, 1, . . . , โ. (b) It was proved in Theorem 3.4 that if (๐, ๐, ๐ ) is real selfadjoint then ๐ป is symmetric. Conversely, let (๐, ๐, ๐ ) be a standard triple and ๐ฟ(๐) be the matrix polynomial for which (๐, ๐, ๐ ) is a standard triple. If ๐ป is symmetric, by part (a), ๐ฟ(๐) is selfadjoint. Then, by Theorem 3.1 (๐, ๐, ๐ ) and (๐ ๐ , ๐ ๐ , ๐ ๐ ) are standard triples for ๐ฟ(๐) and, consequently, they are similar. By Theorem 3.4 (๐, ๐, ๐ ) is selfadjoint. โก We aim to show now that the selfadjoint triples of a selfadjoint matrix polynomial are not only similar but unitarily similar. Let us recall this concept (see [8]). Let ๐ป be an ๐ ร ๐ invertible hermitian or symmetric matrix according as ๐ฝ = โ or ๐ฝ = โ. A matrix ๐ โ ๐ฝ๐ร๐ is said to be ๐ป-selfadjoint if ๐ฝ = โ and ๐ โ ๐ป = ๐ป๐ and real ๐ป-selfadjoint if ๐ฝ = โ and ๐ ๐ ๐ป = ๐ป๐ (see [8, p. 48]). De๏ฌnition 3.8. [8, Sec. 4.5, 6.1] (a) Let ๐ป1 , ๐ป2 โ โ๐ร๐ be hermitian invertible matrices and let ๐1 , ๐2 โ โ๐ร๐ be ๐ป1 -selfadjoint and ๐ป2 -selfadjoint, respectively. Then the pairs (๐1 , ๐ป1 ) and (๐2 , ๐ป2 ) are said to be unitarily similar if there exists an invertible matrix
Canonical Forms
433
๐ โ โ๐ร๐ such that ๐2 = ๐โ1 ๐1 ๐ and ๐ป2 = ๐โ ๐ป1 ๐ (๐ is (๐ป1 , ๐ป2 )unitary). (b) Let ๐ป1 , ๐ป2 โ โ๐ร๐ be symmetric invertible matrices and let ๐1 , ๐2 โ โ๐ร๐ be real ๐ป1 -selfadjoint and ๐ป2 -selfdajoint, respectively. Then the pairs (๐1 , ๐ป1 ) and (๐2 , ๐ป2 ) are said to be real unitarily similar if there exists an invertible matrix ๐ โ โ๐ร๐ such that ๐2 = ๐โ1๐1 ๐ and ๐ป2 = ๐๐ ๐ป1 ๐ (๐ is (๐ป1 , ๐ป2 )orthogonal). If (๐, ๐, ๐ ) is a (real) selfadjoint triple then there exists an invertible (symmetric) hermitian matrix ๐ป such that (๐ ๐ = ๐ป๐ ๐ป โ1 ) ๐ โ = ๐ป๐ ๐ป โ1 ; i.e., the โmainโ matrix ๐ is (real) ๐ป-selfadjoint. Recall that such a matrix ๐ป is unique and is given by any of the four expressions in Proposition 3.6. Proposition 3.9. Let (๐1 , ๐1 , ๐1 ) and (๐2 , ๐2 , ๐2 ) be (real) selfadjoint triples and, for ๐ = 1, 2, let ๐ป๐ be the (symmetric) hermitian matrix such that (๐๐๐ = ๐ป๐ ๐๐ ๐ป๐โ1) ๐๐โ = ๐ป๐ ๐๐ ๐ป๐โ1 and (๐๐ = ๐ป๐โ1 ๐๐๐ ) ๐๐ = ๐ป๐โ1 ๐๐โ . If (๐1 , ๐1 , ๐1 ) and (๐2 , ๐2 , ๐2 ) are similar then (๐1 , ๐ป1 ) and (๐2 , ๐ป2 ) are (real) unitarily similar. Proof. The proof will be given for the real case; the complex case is proved similarly. Assume that (๐1 , ๐1 , ๐1 ) and (๐2 , ๐2 , ๐2 ) are similar and let ๐ be the unique nonsingular matrix such that (cf. (7)) ๐2 = ๐1 ๐ โ1 ,
๐2 = ๐๐1 ๐ โ1 ,
Such a matrix has the form ๐ = ๐ถ(๐2 , ๐2 )
โ1
๐2 = ๐๐1 .
๐ถ(๐1 , ๐1 ). Then ๐2 = ๐๐1 ๐ โ1 and
๐ป2 = ๐ถ(๐2 , ๐2 )๐ ๐
(๐2 , ๐2 )โ1 = ๐ โ๐ ๐ถ(๐1 , ๐1 )๐ ๐
(๐1 , ๐1 )โ1 ๐ โ1 = ๐ โ๐ ๐ป1 ๐ โ1 . If ๐ = ๐ โ1 we have ๐2 = ๐โ1 ๐1 ๐ and ๐ป2 = ๐๐ ๐ป1 ๐ so that (๐1 , ๐ป1 ) and (๐2 , ๐ป2 ) are real unitarily similar. โก It will be important for us to note that the converse of this proposition is not true in general. That is to say, given two selfadjoint triples (๐1 , ๐1 , ๐1 ) and (๐2 , ๐2 , ๐2 ), the fact that (๐1 , ๐ป1 ) and (๐2 , ๐ป2 ) are unitarily similar does not guarantee that (๐1 , ๐1 , ๐1 ) and (๐2 , ๐2 , ๐2 ) are similar. A proof of this will need the use of the sign characteristic, a concept that will be introduced in the following section. We defer that proof until the necessary concepts have been discussed (see Section 4.2). However, for a given (real) selfadjoint matrix polynomial, all its (real) selfadjoint triples can be obtained from the โcompanionโ triple de๏ฌned in (4). It has already been shown that, provided that ๐ฟ(๐) is selfadjoint, this primitive triple is also selfadjoint with symmetric (hermitian in the complex case) matrix ๐ด of (2). It turns out that all selfadjoint standard triples for ๐ฟ(๐) can be obtained by applying unitary similarity to (๐ถ๐
, ๐ด): Proposition 3.10. Let ๐ฟ(๐) be a selfadjoint matrix polynomial with nonsingular leading coe๏ฌcient and let its primitive selfadjoint triple be (๐0 , ๐ถ๐
, ๐0 ) as given in (4). Let ๐ป be an โ๐ ร โ๐ symmetric (hermitian if ๐ฝ = โ) invertible matrix and let ๐ โ ๐ฝโ๐รโ๐ be ๐ป-selfadjoint. If (๐, ๐ป) and (๐ถ๐
, ๐ด) are unitarily similar with
434
P. Lancaster and I. Zaballa
๐ = ๐โ1 ๐ถ๐
๐ and ๐ป = ๐๐ ๐ด๐ (๐ป = ๐โ ๐ด๐ if ๐ฝ = โ) then (๐0 ๐, ๐, ๐ป โ1 ๐๐ ๐0๐ ) ((๐0 ๐, ๐, ๐ป โ1 ๐โ ๐0โ ) if ๐ฝ = โ) is a selfadjoint triple of ๐ฟ(๐). Proof. Again, the proof is given in the real case. The proof in the complex case is similar. De๏ฌne ๐ = ๐0 ๐ and ๐ = ๐ป โ1 ๐๐ ๐0๐ . We have to prove ๏ฌrst that (๐, ๐, ๐ ) is similar to (๐0 , ๐ถ๐
, ๐0 ). In fact, it follows from (2) and (4) that ๐0 = ๐ดโ1 ๐0๐ and so ๐ = ๐ป โ1 ๐๐ ๐0๐ = ๐ป โ1 ๐๐ ๐ด๐0 = ๐โ1 ๐ดโ1 ๐โ๐ ๐๐ ๐ด๐0 = ๐โ1 ๐0 . Thus (๐, ๐, ๐ ) โ (๐0 ๐, ๐โ1 ๐ถ๐
๐, ๐โ1 ๐0 ) is the required similarity, and (๐, ๐, ๐ ) is a standard triple of ๐ฟ(๐). We are to prove next that ๐ = ๐ป โ1 ๐ ๐ and ๐ ๐ = ๐ป๐ ๐ป โ1. The ๏ฌrst follows ๐ from the de๏ฌnition of ๐ . Now, bearing in mind that ๐ถ๐
is ๐ด-selfadjoint (๐ถ๐
๐ด= ๐ด๐ถ๐
), we have ๐๐
= =
๐ โ๐ ๐๐ ๐ถ๐
๐ = ๐๐ ๐ด๐ถ๐
๐ดโ1 ๐โ๐ = ๐๐ ๐ด๐๐โ1 ๐ถ๐
๐๐โ1 ๐ดโ1 ๐โ๐ โ1 ๐ป๐ ๐ป ,
and the proposition follows.
โก
This argument provides a construction of (real) selfadjoint triples of a (real) selfadjoint matrix polynomial using unitary similarity, and we have used the primitive triple (๐0 , ๐ถ๐
, ๐0 ) in this construction. However, this role could be played by any selfadjoint triple of ๐ฟ(๐) and the proposition still holds โ with the same proof. In other words, let (๐1 , ๐1 , ๐1 ) be a real selfadjoint triple of ๐ฟ(๐) and ๐ป1 the symmetric invertible matrix satisfying ๐1๐ ๐ป1 = ๐ป1 ๐1 , ๐1 = ๐ป1โ1๐1๐ . If ๐ = ๐โ1 ๐1 ๐ and ๐ป = ๐๐ ๐ป1 ๐ then (๐1 ๐โ1 , ๐, ๐ป โ1๐โ๐ ๐1๐ ) is a real selfadjoint standard triple of ๐ฟ(๐). A similar result holds in the complex case.
4. Canonical structures for hermitian polynomials In this section we review some canonical structures for hermitian (and especially real symmetric) matrix polynomials. Results in the complex hermitian case are well known, and will serve to set-the-scene before discussing the less-well-known case of real symmetric matrix polynomials. First consider the standard triple (๐0 , ๐ถ๐
, ๐0 ) of (4). It follows from Theโ orem 3.1 that (๐0โ , ๐ถ๐
, ๐0โ ) is also a standard triple. Furthermore, with ๐ด given by (2) โ โ ๐ถ๐
= ๐ด๐ถ๐
๐ดโ1 or ๐ด๐ถ๐
= ๐ถ๐
๐ด. (13) โ โ Thus, we have a hermitian pair: ๐ด = ๐ด and (๐ด๐ถ๐
) = ๐ด๐ถ๐
with ๐ด nonsingular. And, of course, this becomes a real symmetric pair when ๐ฟ(๐) has real symmetric coe๏ฌcients. Canonical structures for ๐ฟ(๐) are now determined by congruence transformations applied simultaneously to ๐ด and ๐ด๐ถ๐
over ๐ฝ. For a congruence ๐โ ๐ด๐ = ๐ and ๐โ (๐ด๐ถ๐
)๐ = ๐ ๐ฝ, it follows that ๐ถ๐
= ๐๐ฝ๐โ1 , and this shows that, for ๐,
Canonical Forms
435
we can use a similarity transformation of ๐ถ๐
to Jordan canonical form โ over โ or โ, as the case may be. Simultaneous congruence transformations of this kind have been reviewed recently in [11]. Furthermore, the invertibility of ๐ดโ (and hence ๐ด) assumed here removes troublesome singular structures appearing in the general case. (The hermitian case appears as Theorem 5.1.1 of [8].) De๏ฌnition 4.1. A standard triple of the form (๐, ๐ฝ, ๐ ) (where ๐ฝ is a matrix in Jordan form) is said to be a Jordan triple. If, in addition, ๐ฝ is a Jordan form of the companion matrix ๐ถ๐
for a matrix polynomial ๐ฟ(๐), then (๐, ๐ฝ, ๐ ) is said to be a Jordan triple for ๐ฟ(๐). Then, recalling De๏ฌnition 3.2 of selfadjoint standard triples: De๏ฌnition 4.2. A (real) Jordan triple (๐, ๐ฝ, ๐ ) will be called a (real) selfadjoint Jordan triple if (๐, ๐ฝ, ๐ ) is a (real) selfadjoint standard triple. Of course, in these de๏ฌnitions, the Jordan form is that over โ, or over โ, as appropriate. As in the classical case, ๐ฟ(๐) = ๐ผ๐ ๐ โ ๐ด, the Jordan form displays both the elementary divisor structure of the eigenvalues, but also encodes complete information on eigenvector chains. However, the details are di๏ฌerent in the complex and real cases. 4.1. The complex hermitian case The structure of Jordan triples over โ is familiar from the works of Gohberg et al., and will be summarized here. The structure of Jordan triples over โ may be less familiar, and is the topic of the next section. To help in the description of canonical forms we introduce the primitive ๐ร๐ matrices โค โก โค โก 0 โ
โ
โ
0 1 0 0 0 โ
โ
โ
0 1 โฅ โข .. โข 0 โ
โ
โ
โข . 1 0 โฅ 0 0 โฅ โฅ โฅ โข โข โฅ โข โข .. .. โฅ . .. , ๐บ = (14) ๐น๐ = โข โฅ โข ๐ . 1 . โฅ โฅ โฅ โข โข . โฅ โข โฆ โฃ 0 1 .. 0 โฃ 1 0 . โฆ 1 0 โ
โ
โ
0 0 0 0 โ
โ
โ
0 Note also that ๐น1 = 1, ๐บ1 = 0. In the following formulae, ๐ will always be the degree of an elementary divisor of the pencil ๐๐ผ โ ๐ถ๐
or, what is equivalent, the hermitian pencil ๐๐ด โ ๐ด๐ถ๐
. Suppose that ๐ฟ(๐) is hermitian and has exactly ๐ real elementary divisors with associated real eigenvalues ๐ผ1 , . . . , ๐ผ๐ (not necessarily distinct), and let the degrees of these elementary divisors be ๐1 , . . . , ๐๐ , respectively. Also, let there be exactly ๐ pairs of non-real conjugate eigenvalues, (๐ฝ1 , ๐ฝยฏ1 ), . . . , (๐ฝ๐ , ๐ฝยฏ๐ ) with associated elementary divisors of degrees ๐1 , . . . , ๐๐ , respectively. Then if ๐ฟ(๐) is ๐ ร ๐ with degree โ (and det๐ดโ โ= 0) we will have โ๐ =
๐ โ ๐=1
๐๐ + 2
๐ โ ๐=1
๐๐ .
436
P. Lancaster and I. Zaballa
โ๐ Now it will be convenient to introduce the notation ๐=1 ๐๐ to denote a direct (block diagonal) sum of matrices ๐1 , . . . , ๐๐ . There is a complex congruence transformation which, when applied to ๐๐ด โ ๐ด๐ถ๐
, produces a hermitian pencil ๐๐ โ ๐ ๐ฝ, where ๐ =
๐ โ
๐๐ ๐น๐๐
๐=1
and ๐๐ฝ =
๐ โ
๐๐ (๐ผ๐ ๐น๐๐ + ๐บ๐๐ )
๐ [ โโ
๐=1
๐=1
๐ โโ
๐น2๐๐ ,
(15)
๐=1
0 ๐ฝยฏ๐ ๐น๐๐ + ๐บ๐๐
๐ฝ๐ ๐น๐๐ + ๐บ๐๐ 0
] .
(16)
The numbers ๐1 , . . . , ๐๐ are each equal to either +1 or -1 and, together, they are known as the โsign characteristicโ of the system (either ๐ฟ(๐) or ๐๐ด โ ๐ด๐ถ๐
). This is an important concept which plays an important role in perturbation theory for matrix polynomials, as well as more general matrix functions (see [6], for example). The reduced forms (15) and (16) are obtained from Theorem 6.1 of [11], where existence and uniqueness arguments can be found. To emphasize the dependence of the structure of ๐ on the sign characteristic, ๐ := {๐1 , . . . , ๐๐ }, and on the more primitive Jordan matrix, ๐ฝ, we write ๐ = ๐๐,๐ฝ . It is important to note that ๐๐,๐ฝ โ1 2 is involutory: ๐๐,๐ฝ = ๐ผ, so that ๐๐,๐ฝ = ๐๐,๐ฝ . From (15) and (16) we deduce that the corresponding Jordan canonical form is ] ๐ ๐ [ โ โโ 0 ๐ฝยฏ๐ ๐ผ๐๐ + ๐น๐๐ ๐บ๐๐ (๐ผ๐ ๐ผ๐๐ + ๐น๐๐ ๐บ๐๐ ) , (17) ๐ฝ= 0 ๐ฝ๐ ๐ผ๐๐ + ๐น๐๐ ๐บ๐๐ ๐=1
๐=1
โ
and ๐ฝ ๐๐,๐ฝ = ๐๐,๐ฝ ๐ฝ so that ๐ฝ is ๐๐,๐ฝ -selfadjoint. If all elementary divisors of ๐ฟ(๐) are linear then ๐ฝ will be diagonal, but ๐๐,๐ฝ will be diagonal only if all eigenvalues are real; otherwise ๐๐,๐ฝ is tridiagonal. Since ๐๐ด โ ๐ด๐ถ๐
and ๐๐๐,๐ฝ โ ๐๐,๐ฝ ๐ฝ are congruent pencils, there is an invertible complex matrix ๐ such that ๐๐,๐ฝ = ๐โ ๐ด๐ and ๐๐,๐ฝ ๐ฝ = ๐โ ๐ด๐ถ๐
๐. Then ๐๐,๐ฝ ๐ฝ = ๐โ ๐ด๐๐โ1 ๐ถ๐
๐ = ๐๐,๐ฝ ๐โ1 ๐ถ๐
๐. But ๐๐,๐ฝ is invertible. Therefore (๐ถ๐
, ๐ด) and (๐ฝ, ๐๐,๐ฝ ) are unitarily similar. If we put ๐ = ๐0 ๐ where (๐0 , ๐ถ๐
, ๐0 ) is the primitive selfadjoint triple of ๐ฟ(๐) given in (4), by Proposition 3.10 and bearing โ1 = ๐๐,๐ฝ , (๐, ๐ฝ, ๐๐,๐ฝ ๐ โ ) is a selfadjoint triple of ๐ฟ(๐). We have in mind that ๐๐,๐ฝ just proved: Theorem 4.3. If ๐ฟ(๐) is hermitian and ๐ดโ is nonsingular, then there exists a selfadjoint Jordan triple of the form (๐, ๐ฝ, ๐๐,๐ฝ ๐ โ ) with ๐ฝ and ๐๐,๐ฝ as in (17) and (15). The set of numbers ๐ is determined uniquely by ๐ฟ(๐), up to permutation of signs in the blocks of ๐๐,๐ฝ corresponding to the Jordan blocks of ๐ฝ with the same real eigenvalue and the same size. We emphasize that the de๏ฌnition of a real selfadjoint Jordan triple given in De๏ฌnition 4.2 is more general than that of [3] and [4]. In fact, in [3] and [4]
Canonical Forms
437
a standard triple (๐, ๐, ๐ ) is a selfadjoint Jordan triple if ๐ = ๐ฝ (matrix in Jordan form) and ๐ = ๐๐,๐ฝ ๐ โ . This is more restrictive than De๏ฌnition 4.2 where (๐, ๐, ๐ ) quali๏ฌes as a selfadjoint Jordan triple provided that it is a standard triple, ๐ = ๐ฝ, and ๐ = ๐ป โ1 ๐ โ where ๐ป is any nonsingular hermitian matrix such that ๐ฝ โ ๐ป = ๐ป๐ฝ. Thus, if (๐, ๐ฝ, ๐๐,๐ฝ ๐ โ ) is a standard triple, it is a selfadjoint Jordan triple in both cases, but if ๐ is an invertible matrix such that ๐ฝ = ๐โ1 ๐ฝ๐ and we de๏ฌne ๐ป = ๐โ ๐๐,๐ฝ ๐ then ๐ป is hermitian and nonsingular, ๐ฝ โ ๐ป = ๐ป๐ฝ and (๐, ๐ฝ, ๐ป โ1 ๐ โ ) is a selfadjoint Jordan triple for ๐ฟ(๐) in the sense of De๏ฌnition 4.2, but it is not (unless ๐ป = ๐๐,๐ฝ ) in the sense of [3] and [4]. To stress the di๏ฌerence between the two de๏ฌnitions, compare Theorem 1.10 of [3] with the fact that, according to the complex version of Theorem 3.5 (or [8, Th. 12.2.2]), all Jordan triples of an hermitian matrix polynomial are selfadjoint. 4.2. The real symmetric case When the coe๏ฌcients of ๐ฟ(๐) are real and symmetric, then matrices ๐ด and ๐ด๐ถ๐
(of (13)) are real and symmetric. Now the simultaneous reduction of these two matrices by congruence can be completed over the real ๏ฌeld. Again, a complete discussion can be found in [11]. The relevant result of that paper is now Theorem 9.2. We use the same notations and conventions as Section 4.1 concerning the real eigenvalues, non-real conjugate pairs of eigenvalues and the degrees of their elementary divisors. To handle the case of nonlinear elementary divisors for nonreal eigenvalues it is convenient to introduce another primitive symmetric matrix with even size, say 2๐ ร 2๐: โค โก 0 0 โ
โ
โ
1 0 โข 0 0 0 โ1 โฅ โฅ โข โฅ โข .. โข . 1 0 0 0 โฅ โฅ โข 0 โ1 0 0 โฅ ๐ธ2๐ = โข (18) โฅ. โข โข .. โฅ โข . โฅ โฅ โข โฃ 1 0 0 0 โฆ 0 โ1 โ
โ
โ
0 0 To avoid confusion with the hermitian case, we will denote a real Jordan form for a real symmetric polynomial ๐ฟ(๐) by ๐พ. Also, we write the non-real eigenvalues in real and imaginary parts: ๐ฝ๐ = ๐๐ + ๐๐๐ , for ๐ = 1, . . . , ๐ . Now there is a real congruence transformation which, when applied to ๐๐ด โ ๐ด๐ถ๐
, produces a real symmetric pencil ๐๐๐,๐พ โ ๐๐,๐พ ๐พ, where ๐๐,๐พ =
๐ โ ๐=1
๐๐ ๐น๐๐
๐ โโ ๐=1
๐น2๐๐
(19)
438
P. Lancaster and I. Zaballa
and ๐๐,๐พ ๐พ =
๐ โ
๐๐ (๐ผ๐ ๐น๐๐ + ๐บ๐๐ )
๐=1
[ ๐ ( โโ ๐น2๐๐ โ2 ๐๐ ๐น2๐๐ + ๐๐ ๐ธ2๐๐ + 0 ๐=1
0 02
]) .
(20) The numbers ๐1 , . . . , ๐๐ are each equal to either +1 or โ1 and, together, they are known as the โsign characteristicโ of the system (either ๐ฟ(๐) or ๐๐ด โ ๐ด๐ถ๐
). Once 2 again, the โPโ matrix is involutory: ๐๐,๐พ = ๐ผ. We deduce from these two equations that the real Jordan form is: ๐พ=
๐+๐ โ
๐พ๐ ,
(21)
๐=1
where, for ๐ = 1, . . . , ๐,
๐พ๐ = ๐ผ๐ ๐ผ๐๐ + ๐น๐๐ ๐บ๐๐ ,
and for ๐ = ๐ + 1, . . . , ๐ + ๐ ,
[
๐พ๐ = ๐๐ ๐ผ2๐๐ + ๐๐ ๐น2๐๐ ๐ธ2๐๐ + ๐น2๐๐ [ If ๐๐ =
๐๐ ๐๐
โ๐๐ ๐๐
]
(22) ๐น2๐๐ โ2 0
0 02
] .
for ๐ = ๐ + 1, . . . , ๐ + ๐ , then ๐พ๐ is the 2๐๐ ร 2๐๐ real matrix โก โข โข โข ๐พ๐ = โข โข โข โฃ
๐๐
0
๐ผ2 0 .. .
๐๐ ๐ผ2
0
โ
โ
โ
โ
โ
โ
๐๐ .. .
..
. ๐ผ2
โค 0 .. โฅ . โฅ โฅ โฅ. โฅ โฅ โฆ ๐๐
(23)
Notice also that the matrices of (22) have a familiar โJordanโ structure. Thus, when ๐๐ = 3, for example, โก โค ๐ผ๐ 0 0 ๐ผ๐ ๐ผ๐๐ + ๐น๐๐ ๐บ๐๐ = โฃ 1 ๐ผ๐ 0 โฆ . 0 1 ๐ผ๐ For semisimple real systems, ๐พ and ๐๐,๐พ are diagonal if all eigenvalues are real and, if there is at least one non-real eigenvalue pair, then both are tridiagonal. As in the complex case, the real congruence relating ๐๐ด โ ๐ด๐ถ๐
and ๐๐๐,๐พ โ ๐๐,๐พ ๐พ yields the existence of a real invertible matrix ๐ such that ๐ โ1 ๐ถ๐
๐ = ๐พ
and ๐ ๐ ๐ด๐ = ๐๐,๐พ .
(24)
๐ Now, the fundamental symmetry ๐ด๐ถ๐
= ๐ถ๐
๐ด is equivalent to
๐๐,๐พ ๐พ = ๐พ ๐ ๐๐,๐พ ;
(25)
i.e., ๐พ is real ๐๐,๐พ -selfadjoint and so (๐ถ๐
, ๐ด) and (๐พ, ๐๐,๐พ ) are real unitarily similar. If ๐๐ = ๐0 ๐ where (๐0 , ๐ถ๐
, ๐0 ) is the primitive real selfadjoint triple of
Canonical Forms
439
โ1 ๐ฟ(๐) given in (4), by Proposition 3.10 and bearing in mind that ๐๐,๐พ = ๐๐,๐พ , (๐๐ , ๐พ, ๐๐,๐พ ๐๐๐ ) is a real selfadjoint triple of ๐ฟ(๐). This proves the existence of real Jordan selfadjoint triples for real selfadjoint matrix polynomials:
Theorem 4.4. If ๐ฟ(๐) is real and symmetric and ๐ดโ is nonsingular, then there exists a real selfadjoint Jordan triple of the form (๐๐ , ๐พ, ๐๐,๐พ ๐๐๐ ) with ๐พ and ๐๐,๐พ as in (21) and (19). The set of numbers ๐ is determined uniquely by ๐ฟ(๐), up to permutation of signs in the blocks of ๐๐,๐พ corresponding to the Jordan blocks of ๐พ with the same real eigenvalue and the same size. In both the real and complex cases, we have two independent systems of invariants associated with a selfadjoint matrix polynomial: the elementary divisors and the sign characteristic. The ๏ฌrst system of invariants de๏ฌnes the structure of the real or complex Jordan form and the ๏ฌrst and second together determine the structure of the canonical form ๐๐,๐พ . Following [10] we say that two selfadjoint matrix polynomials are strictly isospectral if and only if they have the same elementary divisors and the same sign characteristic. We are now in position to prove that the converse of Proposition 3.9 does not hold in general. The proof is the same for real or complex matrix polynomials and we are going to focus on the real case. Assume that ๐ฟ1 (๐) and ๐ฟ2 (๐) are strictly isospectral, ๐ฟ1 (๐) โ= ๐ฟ2 (๐) and, for ๐ = 1, 2, let ๐ถ๐
๐ and ๐ด๐ be the right companion form and block-symmetric matrix given by (2) associated with ๐ฟ๐ (๐). Then, since ๐ฟ1 (๐) and ๐ฟ2 (๐) share the same canonical forms ๐พ and ๐๐,๐พ , we conclude from (24) that (๐ถ๐
1 , ๐ด1 ) and (๐ถ๐
,2 , ๐ด2 ) are real unitarily similar. Now, for ๐ = 1, 2, (๐0๐ , ๐ถ๐
๐ , ๐0๐ ) given by (4) is a real selfadjoint triple of ๐ฟ๐ (๐) and (๐01 , ๐ถ๐
1 , ๐01 ) and (๐02 , ๐ถ๐
2 , ๐02 ) are not similar provided that ๐ฟ1 (๐) โ= ๐ฟ2 (๐). For if they were similar there would be a nonsingular matrix ๐ such that (๐02 , ๐ถ๐
2 , ๐02 ) = (๐01 ๐, ๐ โ1๐ถ๐
1 ๐, ๐ โ1 ๐01 ). Then ๐ถ๐
1 ๐ = ๐ ๐ถ๐
2 and ๐ must have the form โค ๐ โข ๐๐ถ๐
2 โฅ โฅ โข ๐ = โข .. โฅ โฃ . โฆ โก
โโ1 ๐๐ถ๐
2
[ ] for some full row rank matrix ๐. But ๐01 = ๐02 = ๐ผ๐ 0 โ
โ
โ
0 . The condition ๐02 = ๐01 ๐ implies ๐ = ๐01 = ๐02 . Then ๐ = ๐ผโ๐ and ๐ถ๐
1 = ๐ถ๐
2 which is a contradiction because ๐ฟ1 (๐) โ= ๐ฟ2 (๐).
440
P. Lancaster and I. Zaballa
5. Chains of generalized eigenvectors The canonical structures ensured by Theorems 4.3 and 4.4 lead to the idea of โchains of generalized eigenvectorsโ which are important in perturbation theory and in many applications. In this section we show how these ideas ๏ฌt into the constructions of this paper. 5.1. Real Jordan triples and real eigenvector chains Given a real selfadjoint triple for a real symmetric system, as in Theorem 4.4, the real matrices (๐๐ , ๐พ) form a standard pair and (see Theorem 2.2) ๐ฟ(๐๐ , ๐พ) =
โ โ
๐ด๐ ๐๐ ๐พ ๐ = 0.
(26)
๐=0
Recall the structure of ๐พ from (21)โ(23) and partition ๐๐ accordingly: [ ] ๐๐ = ๐1 โ
โ
โ
๐๐ ๐๐+1 โ
โ
โ
๐๐+๐ ,
(27)
where the number of columns in ๐๐ and ๐พ๐ agree for each ๐ = 1, . . . , ๐ + ๐ . Then for each ๐ [ ] ๐ ๐๐ ๐พ ๐ = ๐1 ๐พ1๐ โ
โ
โ
๐๐+๐ ๐พ๐+๐ , and it follows from (26) that โ โ
๐ด๐ ๐๐ ๐พ๐๐ = 0 for
๐ = 1, 2, . . . , ๐ + ๐ .
(28)
๐=0
Now, for a real eigenvalue we have 1 โค ๐ โค ๐ and write โก ๐๐ โข [ ] โข 1 ๐๐ ๐๐ = ๐ฅ๐1 โ
โ
โ
๐ฅ๐๐๐ and ๐พ๐ = โข .. .. โฃ . . 1
โค โฅ โฅ โฅ โ โ๐๐ ร๐๐ . โฆ ๐๐
Bearing in mind that ๐ถ๐
๐ = ๐๐พ and ๐๐ = ๐0 ๐ (i.e., ๐๐ is the submatrix of ๐ formed by its ๐ ๏ฌrst rows), the following relations are easily obtained: ๐ฟ(๐๐ )๐ฅ๐๐๐ = 0, ๐ฟ(๐๐ )๐ฅ๐๐๐ โ1 + ๐ฟ ๐ฟ(๐๐ )๐ฅ๐1 + ๐ฟ(1) (๐๐ )๐ฅ๐2 + โ
โ
โ
+
(1)
(๐๐ )๐ฅ๐๐๐ = 0, .. .
1 ๐ฟ(๐๐ โ1) (๐๐ )๐ฅ๐๐๐ = 0, (๐๐ โ 1)!
(29)
where ๐ฟ(๐) (๐๐ ) is the ๐th derivative of ๐ฟ(๐) at ๐๐ . This means that ๐ฅ๐๐๐ , ๐ฅ๐,๐๐ โ1 ,. . . , ๐ฅ๐1 is a real Jordan chain of ๐ฟ(๐) associated with the real eigenvalue ๐๐ (see Section 1.4 of [4], for example). In computation, the ๏ฌrst relation in (29) is used to ๏ฌnd the eigenvector ๐ฅ๐๐๐ , the second to ๏ฌnd ๐ฅ๐,๐๐ โ1 , and so on.
Canonical Forms
441
5.2. Real chains for non-real eigenvalues Real Jordan structures associated with non-real eigenvalues are more troublesome. However, using similar ideas, we can obtain real analogues of (29) for non-real eigenvalues. First, we de๏ฌne matrix functions with arguments ๐ โ โ๐ร๐ and ๐ โ โ๐ร๐ , ๐ being any ๏ฌxed positive integer: ๐ฟ(๐, ๐ ) =
โ โ
๐ด๐ ๐ ๐ ๐ ,
๐ฟ(1) (๐, ๐ ) =
๐=0
โ โ
๐๐ด๐ ๐ ๐ ๐โ1 ,
๐=1
and, for ๐ = 2, . . . , โ, ๐ฟ(๐) (๐, ๐ ) =
โ โ
๐(๐ โ 1) . . . (๐ โ ๐ + 1)๐ด๐ ๐ ๐ ๐โ๐ .
๐=๐
Second, we recall that, by hypothesis, ๐ฟ(๐) has ๐ pairs of non-real conjugate eigenvalues (๐ฝ๐ , ๐ฝยฏ๐ ), ๐ฝ๐ = ๐๐ + ๐๐๐ , with associated elementary divisors of degree ๐๐ . Third, for 1 โค ๐ โค ๐ we write โก โค ๐๐ 0 โ
โ
โ
0 โข .. โฅ โข ๐ผ2 ๐๐ . โฅ โข โฅ [ ] โข โฅ, ๐ผ2 ๐๐ ๐๐+๐ = ๐๐1 โ
โ
โ
๐๐๐๐ and ๐พ๐+๐ = โข 0 โฅ โข . โฅ .. .. โฃ .. โฆ . . 0 โ
โ
โ
๐ผ2 ๐๐ [ ] ๐๐ โ๐๐ with ๐๐๐ โ โ๐โร2 and ๐๐ = . ๐๐ ๐๐ Now, if ๐๐+๐ is the submatrix of ๐ whose columns correspond to those of ๐๐+๐ (of (27)), then the relation ๐ถ๐
๐ = ๐๐พ implies ๐ถ๐
๐๐+๐ = ๐๐+๐ ๐พ๐+๐ ,
1 โค ๐ โค ๐ .
Writing down this equation explicitly it is found that, for each ๐, ๐ฟ(๐๐๐๐ , ๐๐ ) = 0, ๐ฟ(๐๐๐๐ โ1 , ๐๐ ) + ๐ฟ ๐ฟ(๐๐1 , ๐๐ ) + ๐ฟ(1) (๐๐2 , ๐๐ ) + โ
โ
โ
+
(1)
(๐๐๐๐ , ๐๐ ) = 0, .. .
1 ๐ฟ(๐๐ โ1) (๐๐๐๐ , ๐๐ ) = 0. (๐๐ โ 1)!
(30)
We may now de๏ฌne ๐๐๐๐ , ๐๐๐๐ โ1 ,. . . , ๐๐1 to be a real Jordan chain of ๐ฟ(๐) with respect to the pair of non-real eigenvalues (๐ฝ๐ , ๐ฝยฏ๐ ).
442
P. Lancaster and I. Zaballa
5.3. Real chains from complex chains Returning to the GLR theory over โ, observe that Theorem 10.7 of [4] ensures the existence of complex selfadjoint Jordan triples (๐, ๐ฝ, ๐๐,๐ฝ ๐ โ ) (as in Theorem 4.3) for real symmetric systems ๐ฟ(๐) with the special form: [ ] ยฏ 2 ๐2 , ๐ฝ = Diag(๐ฝ1 , ๐ฝยฏ2 , ๐ฝ2 ), ๐ = ๐1 ๐ (31) where the spectrum of ๐ฝ1 is real, the spectrum of ๐ฝ2 contains no real numbers and no conjugate complex pairs, the matrix ๐1 is real and contains real Jordan chains of ๐ฟ(๐), and ๐2 contains complex Jordan chains that are not conjugate in pairs. Starting with the structure of (31), one can apply the procedure of [12, Sec. 6.7] (see also [13]) to produce a real selfadjoint Jordan triple of the form (๐๐ , ๐พ, ๐ป โ1 ๐๐ ) (with ๐พ as in Theorem 4.4), but the symmetric matrix ๐ป may not have the canonical form, ๐๐,๐พ of (19). This procedure can be applied with any unitary matrix ๐ for which ๐ โ ๐ฝ๐ = ๐พ. One can carefully select a unitary matrix ๐ such that ๐ โ ๐ฝ๐ = ๐พ, ๐๐ = ๐๐ is real and ๐ โ ๐๐,๐ฝ ๐ = ๐๐,๐พ , but there are unitary matrices for which the ๏ฌrst two conditions are satis๏ฌed but not the third. To illustrate this situation consider the simplest real selfadjoint quadratic matrix polynomial ๐ฟ(๐) = ๐2 + ๐๐ + ๐ where ๐, ๐ โ โ, and assume that it has two nonreal complex conjugate roots ๐1,2 = ๐ ยฑ ๐๐. Then a complex selfadjoint Jordan with the form (31) is (๐, ๐ฝ, ๐๐,๐ฝ ๐ โ ), where [ ] [ ] ยฏ [ ] ๐ 0 0 1 ยฏ ๐ฅ ๐ฝ= 1 , ๐๐,๐ฝ = , ๐๐ = ๐ฅ 0 ๐1 1 0 and ๐ฅ = ยฑ 2โ1 ๐ (1 โ ๐). In order to obtain real selfadjoint Jordan triples we can use unitary matrices ๐ such that ๐ โ ๐ฝ๐ = ๐พ, ๐๐ ๐ = ๐๐ is real and ๐ โ ๐๐,๐ฝ ๐ is symmetric. Two such matrices are [ ] [ ] 1 1+๐ 1โ๐ 1 ๐ 1 โ ๐ = and ๐ = . 2 1โ๐ 1+๐ 2 โ๐ 1 In fact,
] โ๐ , ๐ [ ๐2 = ๐๐ ๐ = โ โ12๐
๐ โ ๐ฝ๐ = ๐ โ ๐ฝ๐ = ๐พ = [ ๐1 = ๐ ๐ ๐ = 0
] ยฑ โ1๐ ,
and ๐ โ ๐๐,๐ฝ ๐ = ๐๐,๐ฝ = ๐๐,๐พ , but
[
๐ ๐
] ยฑ โ12๐ ,
] โ1 0 ๐ป = ๐ ๐๐,๐ฝ ๐ = . 0 1 โ
[
Both (๐1 , ๐พ, ๐๐,๐พ ๐1๐ ) and (๐2 , ๐พ, ๐ป๐2๐ ) are real selfadjoint Jordan triples of ๐ฟ(๐). It is worth-noticing that the elements of ๐1 are the sum and di๏ฌerence of the realโand imaginary parts of ๐๐ while the elements of ๐2 are, up to multiplication by 2, its imaginary and real parts.
Canonical Forms
443
References [1] Chu M.T., and Xu S., Spectral decomposition of real symmetric quadratic ๐-matrices and its applications, Math. of Comp., 78, 2009, 293โ313. [2] Chu D., Chu M.T., and Lin W.-W., Quadratic model updating with symmetry, positive de๏ฌniteness and no spill-over, SIAM J.Matrix Anal.Appl., 31, 2009, 546โ564. [3] Gohberg I., Lancaster P., and Rodman L., Spectral analysis of selfadjoint matrix polynomials, Ann. of Math., 112, 1980, 33โ71. [4] Gohberg I., Lancaster P., and Rodman L., Matrix Polynomials Academic Press, New York, 1982, and SIAM, Philadelphia, 2009. [5] Gohberg I., Lancaster P., and Rodman L., Matrices and Inde๏ฌnite Scalar Products Birkhยจ auser, Basel, 1983. [6] Gohberg I., Lancaster P., and Rodman L., A sign characteristic for selfadjoint meromorphic matrix functions Applicable Analysis, 16, 1983, 165โ185. [7] Gohberg I., Lancaster P., and Rodman L., Invariant Subspaces of Matrices with Applications, Wiley, New York, 1986 and SIAM, Philadelphia, 2006. [8] Gohberg I., Lancaster P., and Rodman L., Inde๏ฌnite Linear Algebra and Applications, Birkhยจ auser, Basel, 2005. [9] Lancaster P., Inverse spectral problems for semisimple damped vibrating systems, SIAM J. Matrix Anal. Appl., 29, 2007, 279โ301. [10] Lancaster P., and Prells U., Isospectral families of high-order systems, Z. Angew. Math. Mech, 87, 2007, 219โ234. [11] Lancaster P., and Rodman L., Canonical forms for hermitian matrix pairs under strict equivalence and congruence, SIAM Review, 47, 2005, 407โ443 [12] Lancaster P., and Tismenetsky M., The Theory of Matrices, Academic Press, New York, 1985. [13] Lin M.M., Dong B., and Chu M.T., Inverse problems for real symmetric quadratic pencils, IMA Journal of Numerical Analysis (to appear). Peter Lancaster Dept. of Mathematics and Statistics University of Calgary Calgary, AB T2N 1N4, Canada e-mail:
[email protected] Ion Zaballa Departamento de Matematica Aplicada y EIO Universidad del Pais Vasco Apdo 644 E-48080 Bilbao, Spain e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 445โ463 c 2012 Springer Basel AG โ
Linearization, Factorization, and the Spectral Compression of a Self-adjoint Analytic Operator Function Under the Condition (VM) H. Langer, A. Markus and V. Matsaev To the memory of our teacher, colleague and dear friend Izrael Gohberg
Abstract. In this paper we continue the study of spectral properties of a selfadjoint analytic operator function ๐ด(๐ง) under the Virozub-Matsaev condition. As in [6], [7], main tools are the linearization and the factorization of ๐ด(๐ง). We use an abstract de๏ฌnition of a so-called Hilbert space linearization and show its uniqueness, and we prove a generalization of the well-known factorization theorem from [10]. The main results concern properties of the compression ๐ดฮ (๐ง) of ๐ด(๐ง) to its spectral subspace, called spectral compression of ๐ด(๐ง). Close connections between the linearization, the inner linearization, and the local spectral function of ๐ด(๐ง) and of its spectral compression ๐ดฮ (๐ง) are established. Mathematics Subject Classi๏ฌcation (2000). 47A56, 47A68, 47A10. Keywords. Self-adjoint analytic operator function, linearization, spectral function, spectrum of de๏ฌnite type, factorization.
1. Introduction This note is a continuation of [7]. We consider an analytic operator function ๐ด(๐ง) which is de๏ฌned and self-adjoint on a simply connected symmetric open set ๐ โ โ and with values in โ(โ) for some Hilbert space โ; here self-adjoint means that ๐ด(๐ง โ ) = ๐ด(๐ง)โ ,
๐ง โ ๐,
in particular, the operators ๐ด(๐) for ๐ โ ๐โฉโ are self-adjoint. The spectrum ๐(๐ด), the point spectrum ๐๐ (๐ด), and the resolvent set ๐(๐ด) of the operator function ๐ด(๐ง) are de๏ฌned in the usual way (see [6], [8]). It is generally assumed that ๐ contains The authors thank the referee for valuable suggestions.
446
H. Langer, A. Markus and V. Matsaev
the real interval ฮ0 = [๐ผ0 , ๐ฝ0 ], that ๐ด(๐ง) satis๏ฌes on ฮ0 the Virozub-Matsaev condition (VM), and that ๐ผ0 , ๐ฝ0 โ ๐(๐ด). The condition (VM) is formulated at the beginning of Section 3. It means, roughly, that if for some ๐ฅ โ โ, ๐ฅ โ= 0, a curve ฮ0 โ ๐ โโ (๐ด(๐)๐ฅ, ๐ฅ),
๐ โ ฮ0 ,
comes close to the real axis, it must cross the axis with a positive ascent. Under the condition (VM) on ฮ0 , for a neighbourhood ๐ฐ of ฮ0 the set ๐ฐ โ ฮ0 belongs to ๐(๐ด) (see, e.g., [6, Proposition 2.1]). On the other hand, if we suppose from the beginning that (๐ฐ โ ฮ0 ) โ ๐(๐ด), then, as proved in [2], [5], there exists a self-adjoint operator ฮ (the linearization of ๐ด(๐ง)) in some Krein space โฑ, such that the relation ๐ด(๐ง)โ1 = โ๐ โ (ฮ โ ๐ง)โ1 ๐ + ๐ต(๐ง),
๐ง โ ๐ฐ โ ฮ0 ,
(1.1)
holds; here ๐ โ โ(โ, โฑ) and ๐ต(๐ง) is a self-adjoint analytic function in ๐ฐ with values in โ(โ), and the spectrum of the operator function ๐ด(๐ง) in ๐ฐ coincides with ๐(ฮ). If the condition (VM) (or at least the more general condition (๐+ ), see below) is satis๏ฌed, then โฑ is a Hilbert space and ฮ is a self-adjoint operator in this Hilbert space, see [6], [7]. By [7, Theorem 2.4] the operator ๐ maps โ onto โฑ with ker ๐ = (ran ๐ โ )โฅ , ran ๐ โ being closed. Let ๐ธ denote the spectral function of the self-adjoint operator ฮ in โฑ . Then the โ(โ)-valued function ๐(ฮ) = ๐ โ ๐ธ(ฮ)๐,
ฮ โ โ,
(1.2)
where โ is the ring generated by all intervals of โ, is called the local spectral function of the operator function ๐ด(๐ง) on ฮ0 . For an interval ฮ = [๐ผ, ๐ฝ] โ ฮ0 , such that ๐ผ and ๐ฝ are not eigenvalues of the operator function ๐ด(๐ง), the relations (1.1) and (1.2) imply that โซ โฒ 1 ๐(ฮ) = โ ๐ด(๐ง)โ1 ๐๐ง, 2๐i ๐พ(ฮ) where ๐พ(ฮ) is a smooth contour in ๐ฐ which surrounds ฮ and crosses the real axis in ๐ผ and ๐ฝ orthogonally, and the โฒ at the integral denotes the Cauchy principal value at ๐ผ and ๐ฝ. The function ๐ is additive on โ, its values ๐(ฮ), ฮ โ โ, are nonnegative operators in โ with closed range (see [7, Theorem 3.1]). Moreover, ran ๐(ฮ) โ ran ๐(ฮ0 ), and with the notation โ(ฮ) := ran ๐(ฮ) we obtain โ(ฮ) โ โ(ฮ0 ),
ฮ โ โ.
For an interval ฮ โ ฮ0 the subspace โ(ฮ) is called the spectral subspace of the operator function ๐ด(๐ง) for ฮ. Since all these spectral subspaces are contained in โ(ฮ0 ), we call โ(ฮ0 ) sometimes the main spectral subspace of ๐ด(๐ง) for ฮ0 . Observe that unlike the spectral subspaces of a self-adjoint operator, these spectral subspaces of a selfadjoint operator function are not invariant under the values of the operator function. It is one aim of this note to continue the study of the local spectral function and the spectral subspaces of ๐ด(๐ง), which was started in [6], [7].
Self-adjoint Analytic Operator Functions For a self-adjoint operator ๐ด with spectral Function ๐ธ it holds โซ (๐ด โ ๐ก)๐๐ธ(๐ก) = 0 for all intervals ฮ, ฮ
447
(1.3)
and, together with the fact that the values of ๐ธ are orthogonal projections, this relation determines ๐ธ uniquely. For a self-adjoint analytic operator function under the condition (๐+ ) on ฮ0 we have instead (see [6, Theorem 3.4]) โซ ๐ด(๐ก)๐๐(๐ก) = 0 for all intervals ฮ โ ฮ0 , (1.4) ฮ
where ๐ is the local spectral function of the operator function ๐ด(๐ง). The values ๐(ฮ) are nonnegative operators, but in general not projections. For a self-adjoint operator ๐ด with (1.3) we also have โซ ๐๐ธ(๐ก) (๐ด โ ๐ง)โ1 = , ๐ง โ โ โ ฮ0 , (1.5) ฮ0 ๐ก โ ๐ง whereas for a self-adjoint analytic operator function we have instead from (1.1) and (1.2) โซ ๐๐(๐ก) ๐ด(๐ง)โ1 = โ + ๐ต(๐ง), ๐ง โ ๐ฐ(ฮ0 ), (1.6) ฮ0 ๐ก โ ๐ง with a self-adjoint operator function ๐ต(๐ง) which is analytic in a neighbourhood of ฮ0 . In [7] we have introduced the inner linearization ๐ of the operator function ๐ด(๐ง) in the main spectral subspace โ(ฮ0 ), given by โซ ๐ := ๐0โ ฮ(๐0โ )โ1 = ๐ ๐๐(๐)๐(ฮ0 )โ1 . (1.7) ฮ0
It is an operator in โ(ฮ0 ), which is selfadjoint with respect to a new Hilbert inner product, and it has in ฮ0 the same spectrum, eigenvalues and corresponding eigenvectors as the operator function ๐ด(๐ง). In the papers [5], [6], [7] besides (VM) also the weaker condition (๐+ ) was considered. By de๏ฌnition, the condition (๐+ ) holds on ฮ0 , if there exist positive numbers ๐, ๐ฟ, such that for all ๐ โ ฮ0 and ๐ โ โ, โฅ๐ โฅ = 1, we have ( ) โฅ๐ด(๐)๐ โฅ < ๐ =โ ๐ดโฒ (๐)๐, ๐ > ๐ฟ. Under this condition, the operator function ๐ด(๐ง) has still a local spectral function on ฮ0 , however, many of the results in the present note fail. This is shown, e.g., by the example in [6, Remark 7.7]. In this example, ๐ด(๐ง) and hence also ฮ have two eigenvalues, hence dim โฑ โฅ 2. On the other hand dim (ran ๐(ฮ0 )) = 1. Since ๐ โ : โฑ โ โ we have ker ๐ โ โ= {0}, and ๐ โ โ= 0 implies dim (ran ๐ โ ) = 1, hence dim (ran ๐ ) = 1 and, since ๐ : โ โ โฑ , we ๏ฌnd ran ๐ โ= โฑ . Thus, in this example the claims of [7, Theorem 2.4], and also of [7, Remark 3.3] do not hold, although the condition (๐+ ) is satis๏ฌed (comp. also with the ๏ฌrst paragraph on [7, p. 536]). In the present paper we use the relation (1.1) as an abstract de๏ฌnition of the linearization ฮ of the operator function ๐ด(๐ง). Therefore we call this relation
448
H. Langer, A. Markus and V. Matsaev
also the basic relation for the linearization ฮ. Under the condition (๐+ ) a minimal linearization, which is a selfadjoint operator in a Hilbert space, exists, see [5], [6]; in Section 2 we show the uniqueness of this linearization, up to unitary equivalence. As was shown in [5], the linearization constructed there is equivalent to the linearization from [2], where also the more general situation of Banach spaces was considered. This holds also with respect to the linearization in the paper [3]. In the present note, however, we consider only Hilbert spaces and use a formally simpler de๏ฌnition of a linearization (see De๏ฌnition 2.1 below). The proof for the uniqueness of the linearization up to similarity from [3, Theorem 2.1] can be adapted to a proof for uniqueness up to unitary equivalence in our situation. However, for the convenience of the reader we prove this fact here directly. In [7] a factorization result, which goes back to [10], was proved for the case that the main spectral subspace โ(ฮ0 ) coincides with the original space โ. In Section 3 this factorization is extended to the situation where โ(ฮ0 ) can be a proper subspace of โ. We have mentioned already that the spectral subspaces โ(ฮ) of the operator function ๐ด(๐ง) are not invariant under the operators ๐ด(๐ง). However, if ๐ฮ denotes the orthogonal projection onto โ(ฮ) in โ, the compressed operator function ๐ดฮ (๐ง): ๐ดฮ (๐ง)๐ = ๐ฮ ๐ด(๐ง)๐, ๐ โ โ(ฮ), with values in โ(โ(ฮ)) is again a self-adjoint analytic operator function on ๐ which satis๏ฌes the condition (VM) on ฮ0 . We call ๐ดฮ (๐ง) the spectral compression of ๐ด(๐ง) for ฮ. Since it satis๏ฌes the condition (VM) on ฮ it has a local spectral function on ฮ, an inner linearization ๐ ฮ and a Hilbert space linearization ฮฮ . In Section 4 it is shown that ๐ ฮ coincides with the restriction ๐ฮ of ๐ to its (invariant) spectral subspace corresponding to ฮ (Theorem 4.1), and that also ฮฮ is the restriction ฮฮ of ฮ to its (invariant) spectral subspace corresponding to ฮ (Theorem 4.3). Although the spectral subspaces โ(ฮ) of ๐ด(๐ง) are not invariant under the operators ๐ด(๐ง), they have a weaker property, which we call pseudoinvariance. This is shown in Section 5. Finally, in Section 6 we derive an explicit expression for the value ๐({๐0 }) of the local spectral function at a real eigenvalue ๐0 of ๐ด(๐ง). Moreover, a second proof of Theorem 4.3 is given, that is based on a block operator representation of ๐ด(๐ง)โ1 with respect to the spectral subspace and its orthogonal complement. This proof allows us to state in Corollary 6.3 that the local spectral function of the spectral compression ๐ดฮ (๐ง) is the restriction of the local spectral function of ๐ด(๐ง).
2. The linearization and its uniqueness 1. Let ๐ด(๐ง) be a self-adjoint analytic operator function which is de๏ฌned on a symmetric open set ๐ โ โ and with values in โ(โ) for some Hilbert space โ. De๏ฌnition 2.1. Suppose that ๐ contains the closed interval ฮ = [๐ผ, ๐ฝ] and that for some simply connected neighbourhood ๐ฐ of ฮ we have ๐ฐ โ ฮ โ ๐(๐ด). The ห on ฮ, if there exist operator function ๐ด(๐ง) admits a Hilbert space linearization ฮ
Self-adjoint Analytic Operator Functions
449
ห in โฑห and an operator ๐ห โ โ(โ, โฑห), a Hilbert space โฑห, a self-adjoint operator ฮ such that the following holds: ห โ ฮ. (a) ๐(ฮ) โ ห ห (b) ๐ (ฮ โ ๐ง)โ1 ๐ห = โ๐ด(๐ง)โ1 + ๐ต(๐ง), ๐ง โ ๐ฐ โ ฮ, where ๐ต(๐ง) is an operator function which is analytic in ๐ฐ; this relation is called the basic relation for ห the linearization ฮ. If, additionally, { } ห ๐ ๐ห โ : ๐ = 0, 1, 2, . . . , (c) โฑห = c.l.s. ฮ ห is called minimal. then the Hilbert space linearization ฮ It is easy to check that the minimality condition (c) is equivalent to the condition { } ห โ ๐ง)โ1 ๐หโ : ๐ง โ ๐ต , where ๐ต is a subset of ๐(ฮ) ห with at (cโ) โฑห = c.l.s. (ฮ ห least one accumulation point in ๐(ฮ). Remark 2.2. In [5] the existence of a minimal Hilbert space linearization was shown if A(z) satis๏ฌes on ฮ the condition (๐+ ). Under this condition the Hilbert space โฑ and the operator ฮ as constructed in [5] have all the properties (a)โ(c). Remark 2.3. If the stronger condition (VM) is satis๏ฌed on ฮ0 , also the inner linearization ๐ of ๐ด(๐ง) is a Hilbert space linearization of ๐ด(๐ง) on ฮ0 . To see this we ๏ฌrst observe that ๐(๐) = ๐(ฮ) โ ฮ0 , and that the relations (1.1) and (1.7) imply ๐ด(๐ง)โ1 = โ๐ฝ(๐ โ ๐ง)โ1 ๐(ฮ0 ) + ๐ต(๐ง),
(2.1)
where ๐ฝ denotes the embedding from โ(ฮ0 ) into โ. Now we choose in De๏ฌnition 2.1 as โฑห the linear space โ(ฮ0 ) equipped with the inner product ) ( (2.2) โจ๐ฅ, ๐ฆโฉ := ๐(ฮ0 )โ1 ๐ฅ, ๐ฆ , ๐ฅ, ๐ฆ โ โ, and as ๐ห โ โ(โ, โฑห) the mapping ๐ห๐ฅ := ๐(ฮ0 ) ๐ฅ โ โ. Since ๐(ฮ0 ) = ๐ โ ๐ , it is easy to see that then ๐ห โ โ โ(โฑห, โ) is the embedding ๐ฝ of โ(ฮ0 ) into โ. Hence (2.1) is the basic relation for the linearization ๐. It is trivial that also the minimality property (c) holds since the set ๐ ๐ ๐ห โ for ๐ = 0 is equal to ๐หโ = ๐(ฮ0 )โ = โ(ฮ0 ). The following theorem states that a minimal Hilbert space linearization is unique up to unitary equivalence. Theorem 2.4. Let the operator function ๐ด(๐ง) be as at the beginning of this section. If ๐ด(๐ง) admits on ฮ two minimal Hilbert space linearizations ฮ1 and ฮ2 , then ฮ1 and ฮ2 are unitarily equivalent.
450
H. Langer, A. Markus and V. Matsaev
Proof. We start from the basic relations for ฮ1 and ฮ2 : ๐๐โ (ฮ๐ โ ๐ง)โ1 ๐๐ = โ๐ด(๐ง)โ1 + ๐ต๐ (๐ง),
๐ง โ ๐ฐ โ (๐ผ, ๐ฝ), ๐ = 1, 2.
(2.3)
Choose a su๏ฌciently smooth simple positive oriented curve ๐พ (โ ๐ฐ) which surrounds ฮ and passes through the points ๐ผ โ ๐ก, ๐ฝ + ๐ก, where ๐ก > 0 is such that the intervals [๐ผ โ ๐ก, ๐ผ) and (๐ฝ, ๐ฝ + ๐ก] belong to ๐(ฮ๐ ), ๐ = 1, 2. The equality (2.3) implies โฎ 1 โ ๐+โ ๐๐ ฮ๐ ๐๐ = โ ๐ง ๐+โ ๐ด(๐ง)โ1 ๐๐ง, ๐, โ = 0, 1, . . . , ๐ = 1, 2. 2๐i ๐พ Hence for any ๐ โ โ and vectors ๐ฅ๐ โ โ, ๐ = 0, 1, . . . , ๐, we ๏ฌnd for ๐ = 1, 2 (โฎ ) ๐ ๐ โ ( โ ๐+โ ) 1 โ ๐+โ โ1 ๐๐ ฮ๐ ๐๐ ๐ฅ๐ , ๐ฅโ = โ ๐ง ๐ด(๐ง) ๐๐ง ๐ฅ๐ , ๐ฅโ , 2๐i ๐พ ๐,โ=0
๐,โ=0
and therefore ๐ ๐ โ โ ( โ ๐+โ ) ( โ ๐+โ ) ๐1 ฮ1 ๐1 ๐ฅ๐ , ๐ฅโ = ๐2 ฮ2 ๐2 ๐ฅ๐ , ๐ฅโ . ๐,โ=0
๐,โ=0
This relation can be written as โฉ ๐ โช ๐ โ โ ฮ๐1 ๐1 ๐ฅ๐ , ฮโ1 ๐1 ๐ฅโ ๐=0
or
โ=0
=
1 1 ๐ 1โ 1 1 1 ๐ ฮ 1 ๐1 ๐ฅ ๐ 1 1 1 1
โฑ1
๐ โ
๐ โ
ฮ๐2 ๐2 ๐ฅ๐ ,
๐=0
โฑ1
๐=0
Denote
โฉ
ฮโ2 ๐2 ๐ฅโ
โ=0
1 1 ๐ 1โ 1 1 1 ๐ =1 ฮ2 ๐2 ๐ฅ๐ 1 1 1 ๐=0
โช
.
, โฑ2
(2.4)
โฑ2
{ } ๐๐ := span ฮ๐๐ ๐๐ โ : ๐ = 0, 1, . . . ,
๐ = 1, 2.
Because of the minimality property (c), ๐๐ is a dense subset of the Hilbert space โฑ๐ , ๐ = 1, 2. Consider the correspondence ๐ โ
ฮ๐1 ๐1 ๐ฅ๐ โโ
๐=0
๐ โ
ฮ๐2 ๐2 ๐ฅ๐ .
(2.5)
๐=0
It determines a correctly de๏ฌned mapping from ๐1 onto ๐2 , that is, any equality ๐ โ
ฮ๐1 ๐1 ๐ฅ๐ =
๐=0
implies
๐ โ ๐=0
๐ โ
ฮ๐1 ๐1 ๐ฆ๐
(2.6)
ฮ๐2 ๐2 ๐ฆ๐ .
(2.7)
๐=0
ฮ๐2 ๐2 ๐ฅ๐ =
๐ โ ๐=0
Self-adjoint Analytic Operator Functions
451
To see this, we add, if necessary, some ๐ฅ๐ or ๐ฆ๐ equal to 0, such that ๐ = ๐. Then the relation (2.4) yields 1 1 ๐ 1 1 ๐ 1 1โ 1 1โ 1 1 1 1 ๐ ๐ ฮ1 ๐1 (๐ฅ๐ โ ๐ฆ๐ )1 = 1 ฮ2 ๐2 (๐ฅ๐ โ ๐ฆ๐ )1 , 1 1 1 1 1 ๐=0
๐=0
โฑ1
โฑ2
and hence (2.6) and (2.7) are equivalent. By (2.4), the mapping ๐
from ๐1 onto ๐2 , given by (2.5): ) ( ๐ ๐ โ โ ๐ ๐
ฮ1 ๐1 ๐ฅ๐ = ฮ๐2 ๐2 ๐ฅ๐ ๐=0
๐=0
extends by continuity to a unitary mapping from โฑ1 onto โฑ2 , which we also denote by ๐
. The relation ( ๐ ) ( ๐ ) โ โ ๐+1 ๐ ฮ1 ๐1 ๐ฅ๐ = ๐
ฮ1 ๐1 ๐ฅ๐ ๐
ฮ1 ๐=0
๐=0
=
๐ โ
ฮ๐+1 ๐2 ๐ฅ ๐ 2
= ฮ2
๐=0
๐ โ
( ฮ๐2 ๐2 ๐ฅ๐
= ฮ2 ๐
๐=0
๐ โ
) ฮ๐1 ๐1 ๐ฅ๐
๐=0
implies ๐
ฮ1 = ฮ2 ๐
, hence ฮ1 and ฮ2 are unitarily equivalent.
โก
From Theorem 2.4 and Remark 2.2 we obtain: Corollary 2.5. Suppose that the condition (VM) is satis๏ฌed for ๐ด(๐ง) on ฮ0 , and that the endpoints of ฮ0 are regular points for the operator function ๐ด(๐ง). If ฮ is a minimal Hilbert space linearization of ๐ด(๐ง) for ฮ0 , and ๐ is the inner linearization of ๐ด(๐ง) for ฮ0 in โ(ฮ0 ), equipped with the inner product as in (2.2), then ฮ and ๐ are unitarily equivalent. Remark 2.6. Note that also the following inverse of Theorem 2.4 holds. If the operator ฮ1 is a minimal Hilbert space linearization of ๐ด(๐ง) for ฮ0 and if the operator ฮ2 is unitarily equivalent to ฮ1 : ฮ2 = ๐ ฮ1 ๐ โ1, then ฮ2 is also a minimal Hilbert space linearization of ๐ด(๐ง) for ฮ0 . This is clear if we de๏ฌne the corresponding operator ๐2 by ๐2 = ๐ ๐1 . 2. Similar as in De๏ฌnition 2.1, a Krein space linearization can be de๏ฌned if in De๏ฌnition 2.1 the words โHilbert spaceโ are replaced everywhere by โKrein spaceโ. The existence of a Krein space linearization was shown in [5] for any self-adjoint analytic operator function ๐ด(๐ง), de๏ฌned on a domain ๐ = ๐ โ and with compact spectrum in ๐, without assuming the condition (๐+ ). In the particular case of a monic self-adjoint operator polynomial ๐ด(๐ง) = ๐ง ๐ ๐ผ + ๐ง ๐โ1 ๐ต๐โ1 + โ
โ
โ
+ ๐ง๐ต1 + ๐ต0
452
H. Langer, A. Markus and V. Matsaev
in a Hilbert space โ also the linearization given by the companion operator โ โ 0 ๐ผ 0 โ
โ
โ
0 โ 0 0 ๐ผ โ
โ
โ
0 โ โ โ ฮ=โ . . . .. โ .. .. โ .. . โ โ๐ต0 โ๐ต1 โ๐ต2 โ
โ
โ
โ๐ต๐โ1 is a Krein space linearization in this sense. Indeed, in this case we can choose โฑ = โ1 โ โ2 โ โ
โ
โ
โ โ๐ ,
โ1 = โ2 = โ
โ
โ
= โ๐ = โ,
with inner product โจโ
, โ
โฉโฑ = (๐บโ
, โ
), de๏ฌned by the Gram operator โ โ ๐ต1 ๐ต2 โ
โ
โ
๐ต๐โ1 ๐ผ โ ๐ต2 ๐ต3 โ
โ
โ
๐ผ 0โ โ โ โ .. . . .. โ .. .. ๐บ=โ . .โ โ โ โ๐ต๐โ1 ๐ผ โ
โ
โ
0 0โ ๐ผ 0 โ
โ
โ
0 0 and the embedding ๐ which maps โ identically onto โ1 , the ๏ฌrst component of โฑ . It is easy to check that in this case ๐ด(๐ง)โ1 = โ๐ โ (ฮ โ ๐ง)โ1 ๐,
๐ง โ ๐(๐ด).
What concerns Theorem 2.4 in the Krein space situation, the inde๏ฌnite isometry between the sets ๐1 and ๐2 follows as above, but this isometry is in general not continuous, and hence it does not extend to a unique isometry between the spaces โฑ1 and โฑ2 . In other words, two minimal Krein space linearizations are in general only weakly isomorphic, see [1]. However, this isometry between the sets ๐1 and ๐2 extends to a unique isometry between the whole spaces, if, e.g., one of the Krein spaces โฑ1 or โฑ2 (and then also the other) is a Pontryagin space.
3. Factorization of ๐จ(๐) Let ๐ด(๐ง) be a self-adjoint analytic operator function which is de๏ฌned and selfadjoint on a symmetric open set ๐ โ โ. Further we shall always suppose that ๐ contains the real interval ฮ0 = [๐ผ0 , ๐ฝ0 ], that ๐ด(๐ผ0 ) and ๐ด(๐ฝ0 ) are boundedly invertible and that ๐ด(๐ง) satis๏ฌes on ฮ0 the Virozub-Matsaev condition (VM): (VM)
โ๐, ๐ฟ > 0 : ๐ โ ฮ0 , ๐ โ โ, โฅ๐ โฅ = 1, โฃ(๐ด(๐)๐, ๐ )โฃ < ๐ =โ (๐ดโฒ (๐)๐, ๐ ) > ๐ฟ.
Then, there exists a simply connected neighbourhood of ฮ0 which does not contain spectrum of ๐ด outside ฮ0 ; such a neighbourhood is denoted by ๐ฐ, hence ๐ฐ โ ฮ0 โ ๐(๐ด). If โ(ฮ) is a spectral subspace of ๐ด(๐ง), we consider the operator function with values in โ(โ(ฮ), โ), which, for ๐ง โ ๐, maps โ(ฮ) โ ๐ โโ ๐ด(๐ง)๐ โ โ.
Self-adjoint Analytic Operator Functions
453
We call this operator function the restriction of ๐ด(๐ง) to the subspace โ(ฮ). Observe that here the term โrestrictionโ does not mean that โ(ฮ) is an invariant subspace of ๐ด(๐ง) (in that usual sense we shall use this term in Sections 4 and 6). For the orthogonal projection ๐โ(ฮ) in โ onto โ(ฮ) we write for short ๐ฮ . Theorem 3.1. Under the assumptions at the beginning of this section, the restriction of ๐ด(๐ง) to โ(ฮ0 ) admits the following factorization: ๐ด(๐ง)๐ = ๐ (๐ง)(๐ โ ๐ง)๐,
๐ โ โ(ฮ0 ), ๐ง โ ๐ฐ,
(3.1)
where ๐ is the inner linearization of ๐ด(๐ง) ( in โ(ฮ)0 ) and ๐ (๐ง), ๐ง โ ๐ฐ, is an analytic operator function with values in โ โ(ฮ0 ), โ . For each ๐ง in a neighborhood of ฮ0 the operator ๐ (๐ง) is injective and its range (depending on ๐ง) is a closed subspace of โ. Proof. We start from the basic relation (1.1) ๐ด(๐ง)โ1 โ ๐ต(๐ง) = โ๐ โ (ฮ โ ๐ง)โ1 ๐,
๐ง โ ๐ฐ โ ๐(๐ด). (3.2) ( ) Both sides of this relation can be considered as elements of โ โ, โ(ฮ0 ) and we can rewrite (3.2) in the form ๐ด(๐ง)โ1 โ ๐ต(๐ง) = โ๐ฝ๐0โ (ฮ โ ๐ง)โ1 ๐,
๐ง โ ๐ฐ โ ๐(๐ด),
(3.3)
where ๐ฝ is the embedding from โ(ฮ) into โ and ๐0โ was de๏ฌned in [7, p. 542]. By the de๏ฌnition (1.7), ๐ โ ๐ง = ๐0โ (ฮ โ ๐ง)(๐0โ )โ1 , ๐ง โ ๐ฐ โ ๐(๐), and hence (๐ โ ๐ง)โ1 = ๐0โ (ฮ โ ๐ง)โ1 (๐0โ )
โ1
,
๐ง โ ๐ฐ โ ๐(๐).
(๐ โ ๐ง)โ1๐(ฮ0 ) = ๐0โ (ฮ โ ๐ง)โ1 ๐,
๐ง โ ๐ฐ โ ๐(๐),
(3.4)
We shall prove that (3.5) ( ) where the operators on both sides are considered as elements of โ โ, โ(ฮ0 ) . If ๐ โ โ โ โ(ฮ0 ) then (๐ โ ๐ง)โ1 ๐(ฮ0 )๐ = 0 = ๐0โ (ฮ โ ๐ง)โ1 ๐ ๐,
๐ง โ ๐ฐ โ ๐(๐),
(3.6)
if ๐ โ โ(ฮ0 ) then, again for ๐ง โ ๐ฐ โ ๐(๐), (๐ โ ๐ง)โ1 ๐(ฮ0 )๐
= (๐ โ ๐ง)โ1 ๐ โ ๐ ๐ = ๐0โ (ฮ โ ๐ง)โ1 (๐0โ )โ1 ๐ โ ๐ ๐ = ๐0โ (ฮ โ ๐ง)โ1 ๐ ๐,
(3.7)
where for the second equality sign we have used (3.4). Obviously, (3.6) and (3.7) imply (3.5). From (2.1) we have ๐ด(๐ง)โ1 โ ๐ต(๐ง) = โ๐ฝ(๐ โ ๐ง)โ1 ๐(ฮ0 ),
๐ง โ ๐ฐ โ ๐(๐ด).
(3.8)
Multiplying (3.8) by ๐ด(๐ง) from the left we get ๐ผ โ ๐ด(๐ง)๐ต(๐ง) = โ๐ด(๐ง)๐ฝ(๐ โ ๐ง)โ1 ๐(ฮ0 ),
๐ง โ ๐ฐ โ ๐(๐ด),
(3.9)
454
H. Langer, A. Markus and V. Matsaev
where both sides act in โ, and multiplying (3.8) by ๐ด(๐ง) from the right we get (๐ผ โ ๐ต(๐ง)๐ด(๐ง))๐ = โ(๐ โ ๐ง)โ1 ๐(ฮ0 )๐ด(๐ง)๐,
๐ง โ ๐ฐ โ ๐(๐ด), ๐ โ โ. (3.10) ( ) Now let ๐ โ โ(ฮ0 ) and apply both sides of (3.9) to ๐(ฮ0 )โ1 ๐ โ โ(ฮ0 ) : (๐ด(๐ง)๐ต(๐ง) โ ๐ผ)๐(ฮ0 )โ1 ๐ = ๐ด(๐ง)(๐ โ ๐ง)โ1๐,
๐ง โ ๐ฐ โ ๐(๐ด).
(3.11)
Set ๐ (๐ง) := ๐ด(๐ง)(๐ โ ๐ง)โ1 . The relation (3.11) shows that ๐ (๐ง) is an operator function with values in โ(โ(ฮ0 ), โ), which is analytic in a neighborhood of ฮ0 . By the de๏ฌnition of ๐ (๐ง), ๐ด(๐ง) = ๐ (๐ง)(๐ โ ๐ง),
๐ง โ ๐ฐ,
(3.12)
and it remains to prove that for each ๐ง โ ๐ฐ the operator ๐ (๐ง) is injective and that its range is closed. This will follow if we show that for ๐ง โ ๐ฐ and a sequence (๐๐ ) โ โ(ฮ0 ) the relation ๐ (๐ง)๐๐ โ 0,
๐ โ โ,
(3.13)
imply ๐๐ โ 0, ๐ โ โ. To this end we multiply (3.10) from the left by ๐ โ ๐ง to obtain ( ) (๐ โ ๐ง) ๐ต(๐ง)๐ด(๐ง) โ ๐ผ = ๐(ฮ0 )๐ด(๐ง), and apply this relation to the elements ๐ฝ(๐ โ ๐ง)โ1 ๐๐ , ๐ง โ ๐(๐). This gives ( ) ๐(ฮ0 )๐ (๐ง)๐๐ = (๐ โ ๐ง) ๐ต(๐ง)๐ด(๐ง) โ ๐ผ ๐ฝ(๐ โ ๐ง)โ1 ๐๐ =
(๐ โ ๐ง)๐ต(๐ง)๐ด(๐ง)๐ฝ(๐ โ ๐ง)โ1 ๐๐ โ ๐๐ ,
where for the last equality sign we can remove the parentheses since the operator ๐ต(๐ง)๐ด(๐ง) maps โ(ฮ0 ) into โ(ฮ0 ), see (3.10). Hence ๐(ฮ0 )๐ (๐ง)๐๐ = (๐ โ ๐ง)๐ต(๐ง)๐ (๐ง)๐๐ โ ๐๐ ,
(3.14)
for ๐ง โ ๐(๐). Since both sides of (3.14) are continuous functions of ๐ง we can choose โก ๐ง โ ฮ0 . Now (3.13) and (3.14) imply ๐๐ โ 0. Remark 3.2. In the case ๐ด(๐ผ0 ) โช 0,
๐ด(๐ฝ0 ) โซ 0,
(3.15)
Theorem 3.1 becomes the Virozub-Matsaev factorization theorem [10] (see also [7, Theorem 4.4]). Indeed, since the conditions (3.15) are equivalent to the equality โ(ฮ0 ) = โ (see [6, Corollary 7.4]), we have only to check that ran ๐ (๐ง) = โ,
๐ง โ ๐ฐ.
(3.16)
It follows from (3.1) and (3.15) that ran ๐ (๐ผ0 ) = ran ๐ (๐ฝ0 ) = โ. But ran ๐ (๐ง) depends continuously on ๐ง in the gap topology, and hence (3.16) holds.
Self-adjoint Analytic Operator Functions
455
Corollary 3.3. Under the assumptions at the beginning of this section the following statements hold: (a) ๐(๐) = ๐(๐ด) โฉ ฮ0 . (b) ๐๐ (๐) = ๐๐ (๐ด) โฉ ฮ0 , and if ๐0 โ ๐๐ (๐) then ker (๐ โ ๐0 ) = ker ๐ด(๐0 ). (c) The eigenvectors of the operator function ๐ด(๐ง), corresponding to di๏ฌerent eigenvalues in ฮ0 , are linearly independent. If there is an in๏ฌnite number of such eigenvalues, then the corresponding eigenvectors form a Riesz basis in their closed linear span. We mention that the second part of (c) follows from (b) and from the fact that ๐ is similar to a self-adjoint operator (see [7, Theorem 4.1]). Remark 3.4. It was shown in [6, Remark 7.7] that statement (c) of Corollary 3.3 fails if we replace the condition (VM) by the condition (๐+ ).
4. The spectral compression of ๐จ(๐), I 1. Let ๐ด(๐ง) be as at the beginning of Section 3. For an interval ฮ = [๐ผ, ๐ฝ] โ ฮ0 , โ(ฮ) is the corresponding spectral subspace and ๐ฮ is the orthogonal projection ( ) in โ onto โ(ฮ). Consider the operator function ๐ดฮ (๐ง) with values in โ โ(ฮ) , which is de๏ฌned as follows: ๐ดฮ (๐ง)๐ := ๐ฮ ๐ด(๐ง)๐,
๐ โ โ(ฮ).
(4.1)
We call ๐ดฮ (๐ง) the spectral compression of ๐ด(๐ง) for ฮ. It is easy to check, that ๐ดฮ (๐ง) is a self-adjoint analytic operator function, de๏ฌned for ๐ง โ ๐ฐ, which satis๏ฌes the condition (VM) on ฮ0 . Moreover, [7, Lemma 2.2 and Corollary 2.5] imply that ๐ดฮ (๐ผ) โค 0,
๐ดฮ (๐ฝ) โฅ 0.
From [6, Lemma 4.1 (e)] it follows that ๐ดฮ (๐ผโฒ ) โช 0,
๐ดฮ (๐ฝ โฒ ) โซ 0
for all ๐ผโฒ โ [๐ผ0 , ๐ผ), ๐ฝ โฒ โ (๐ฝ, ๐ฝ0 ].
Hence according to [7, Theorem 4.4] ๐ดฮ (๐ง) admits the Virozub-Matsaev factorization (4.2) ๐ดฮ (๐ง) = ๐ ฮ (๐ง)(๐ ฮ โ ๐ง), ๐ง โ ๐ฐ, ฮ where ๐ is the inner linearization of ๐ดฮ on ฮ, ๐ ฮ (๐ง) is an analytic operator function with values in โ(โ(ฮ)) which are invertible operators in โ(ฮ), and ๐ฐ is a neighbourhood of ฮ. On the other hand, if we multiply the factorization (3.1) above from the left by ๐ฮ and apply it only to elements of โ(ฮ) we obtain ( ) (4.3) ๐ดฮ (๐ง)๐ = ๐ฮ ๐ (๐ง) ๐ โ ๐ง ๐, ๐ โ โ(ฮ) The subspace โ(ฮ) is an invariant subspace (even a spectral subspace) of ๐. Therefore this relation can be written as ) ( ๐ดฮ (๐ง) = ๐ฮ ๐ (๐ง)๐ฮ ๐ฮ โ ๐ง , (4.4) where here ๐ฮ denotes the restriction of ๐ to its invariant subspace โ(ฮ).
456
H. Langer, A. Markus and V. Matsaev
Theorem 4.1. The operators ๐ ฮ in (4.2) and ๐ฮ in (4.4) coincide, that is, the inner linearization of the spectral compression ๐ดฮ (๐ง) coincides with the restriction ๐ฮ of the inner linearization ๐ of ๐ด(๐ง) to its invariant subspace โ(ฮ). Proof. The relations (4.2) and (4.4) imply
( ) ๐ ฮ (๐ง)(๐ ฮ โ ๐ง) = ๐ฮ ๐ (๐ง)๐ฮ ๐ฮ โ ๐ง ,
๐ง โ ๐ฐ.
It follows that
( )โ1 = ๐ ฮ (๐ง)โ1 ๐ฮ ๐ (๐ง)๐ฮ , (๐ ฮ โ ๐ง) ๐ฮ โ ๐ง
๐ง โ ๐ฐ โ ๐(๐ฮ ].
The operator function on the right-hand side is bounded and analytic in a neighbourhood of ฮ, the function on the left-hand side is analytic outside ฮ including โ. According to Liouvilleโs theorem, both sides are constant, and letting ๐ง โ โ on the left-hand side it follows that this constant is ๐ผ. โก 2. In this subsection we show that the restriction ฮฮ of a minimal Hilbert space linearization ฮ to its spectral subspace โฑฮ is a minimal Hilbert space linearization of the spectral compression ๐ดฮ (๐ง). We start with the following evident statement. Lemma 4.2. Let ๐บโฒ , ๐บโฒโฒ be self-adjoint operators and ๐ธ โฒ , ๐ธ โฒโฒ , respectively, be their spectral functions. If ฮ is a real interval, by ๐บโฒฮ , ๐บโฒโฒฮ we denote the restrictions of these operators to their invariant subspaces ran ๐ธ โฒ (ฮ), ran ๐ธ โฒโฒ (ฮ), respectively. If ๐บโฒ , ๐บโฒโฒ are unitarily equivalent, then ๐บโฒฮ and ๐บโฒโฒฮ are also unitarily equivalent. Theorem 4.3. Let ฮ in โฑ be a minimal Hilbert space linearization for ฮ0 of the self-adjoint operator function ๐ด(๐ง) in โ as above, and let ฮ be a subinterval of ฮ0 . Then the restriction ฮฮ of ฮ to its invariant subspace โฑฮ is a minimal Hilbert space linearization for ฮ of the compressed operator function ๐ดฮ (๐ง) in โ(ฮ). Proof. Consider the inner linearization ๐ of ๐ด(๐ง). By Corollary 2.5 the operators ฮ and ๐ are unitarily equivalent. Lemma 4.2 implies that also the operators ฮฮ and ๐ฮ are unitarily equivalent. By Theorem 4.1, ๐ฮ is the inner linearization of the operator function ๐ดฮ (๐ง) for ฮ, and hence ๐ฮ is a minimal Hilbert space linearization of ๐ดฮ (๐ง) for ฮ, see Remark 2.2. Now Remark 2.6 implies that ฮฮ is a minimal Hilbert space linearization of ๐ดฮ (๐ง) for ฮ. โก Corollary 4.4. The linearization ฮ of the operator function ๐ด(๐ง) is also a linearization of the compression ๐ดฮ0 (๐ง) to its main spectral subspace โ(ฮ0 ).
5. Pseudoinvariance One of the main properties of a spectral subspace of an operator is the invariance of this subspace under the operator. In our situation, however, the spectral subspace โ(ฮ) of the self-adjoint operator function ๐ด(๐ง) can be an invariant subspace of all operators ๐ด(๐ง) only in some trivial cases. However we shall show that the subspace โ(ฮ) has a property which can be considered as a weak analogue of invariance.
Self-adjoint Analytic Operator Functions
457
If a subspace โ โ โ is not invariant under an operator ๐บ โ โ(โ) then for at least one vector ๐ โ โ we have dist (๐บ๐, โ) > 0. This can happen in the present situation, but we show that no non-zero vector ๐ด(๐ง)๐ with ๐ โ โ(ฮ) can be orthogonal to โ(ฮ). More exactly, we prove: Theorem 5.1. There exists a number ๐ < 1 such that dist (๐ด(๐ง)๐, โ(ฮ)) โค ๐โฅ๐ด(๐ง)๐ โฅ
(5.1)
for all subintervals ฮ โ ฮ0 , all ๐ง โ ฮ0 , and all ๐ โ โ(ฮ). Proof. The relations (4.2), (4.4) and ๐ ฮ = ๐ฮ (see Theorem 4.1) imply that ๐ฮ ๐ (๐ง)๐ฮ (๐ ฮ โ ๐ง) = ๐ ฮ (๐ง)(๐ ฮ โ ๐ง), ฮ
๐ง โ ๐ฐ.
The operator ๐ has only real spectrum, and hence ran (๐ ๐ง โ ๐ฐ โ โ. Therefore (5.2) implies ๐ฮ ๐ (๐ง)๐ = ๐ ฮ (๐ง)๐,
ฮ
(5.2)
โ ๐ง) = โ(ฮ) for
๐ โ โ(ฮ), ๐ง โ ๐ฐ โ โ.
(5.3)
By continuity, (5.3) holds even for all ๐ง โ ๐ฐ. Since ๐ ฮ (๐ง), ๐ง โ ฮ0 , is invertible, we obtain for all ๐ง โ ฮ0 and ๐ โ โ(ฮ) โฅ๐ฮ ๐ (๐ง)๐ โฅ = โฅ๐ ฮ (๐ง)๐ โฅ โฅ ๐พ1 โฅ๐ โฅ, ( )โ1 . Using Theorem 3.1, we have where ๐พ1 = max๐งโฮ0 โฅ๐ ฮ (๐ง)โ1 โฅ โฅ๐ฮ ๐ด(๐ง)๐ โฅ = โฅ๐ฮ ๐ (๐ง)(๐ โ ๐ง)๐ โฅ โฅ ๐พ1 โฅ(๐ โ ๐ง)๐ โฅ.
(5.4)
On the other hand, โฅ๐ด(๐ง)๐ โฅ = โฅ๐ (๐ง)(๐ โ ๐ง)๐ โฅ โค ๐พ2 โฅ(๐ โ ๐ง)๐ โฅ,
(5.5)
where ๐พ2 = max๐งโฮ0 โฅ๐ (๐ง)โฅ. Since 2
(dist (๐ด(๐ง)๐, โ(ฮ))) = โฅ๐ด(๐ง)๐ โฅ2 โ โฅ๐ฮ ๐ด(๐ง)๐ โฅ2 , )1/2 ( . the inequalities (5.4) and (5.5) imply (5.1) with ๐ = 1 โ ๐พ12 ๐พ2โ2
โก
Remark 5.2. The relations (5.4) and (5.5) imply also that, with ๐พ := ๐พ1 ๐พ2โ1 , โฅ๐ฮ ๐ด(๐ง)๐ โฅ โฅ ๐พโฅ๐ด(๐ง)๐ โฅ,
๐ โ โ(ฮ), ๐ง โ ฮ.
(5.6)
6. The spectral compression of ๐จ(๐), II 1. In this section ๐ด(๐ง) is again a self-adjoint analytic operator function as at the beginning of Section 3, and ฮ is a closed subinterval of ฮ0 . Recall that ๐ธ denotes the spectral function of the self-adjoint linearization ฮ, โฑฮ = ran ๐ธ(ฮ), and ๐ is the local spectral function of ๐ด(๐ง).
458
H. Langer, A. Markus and V. Matsaev
First we give an explicit formula for ๐({๐0 }) =: ๐0 if ๐0 โ ฮ0 is an eigenvalue of the operator function ๐ด(๐ง). Then ker ๐ด(๐0 ) = ran ๐0 (see [7, (3.3)]), and the condition (VM) at the beginning of Section 3 implies that ( โฒ ) ๐ด (๐0 )๐, ๐ โฅ ๐ฟโฅ๐ โฅ2 , ๐ โ ker ๐ด(๐0 ). (6.1) Therefore the operator ๐0 ๐ดโฒ (๐0 )๐0 is uniformly positive and hence boundedly invertible on ker ๐ด(๐0 ) = ran ๐0 ; here ๐0 denotes the orthogonal projection onto ker ๐ด(๐0 ). Lemma 6.1. For ๐ โ ker ๐ด(๐0 ) we have ๐ = ๐0 ๐ดโฒ (๐0 )๐,
(6.2)
ห be the measure which is obtained from ๐ by subtracting the possible Proof. Let ๐ point measure at ๐0 : { ๐(ฮ) if ๐0 โ / ฮ, ห ๐(ฮ) := ๐(ฮ) โ ๐({๐0 }) if ๐0 โ ฮ, for intervals ฮ. The relation (1.5) implies ๐ด(๐ง)โ1 = โ
๐0 โ ๐0 โ ๐ง
โซ ฮ0
ห ๐๐(๐ก) + ๐ต(๐ง), ๐กโ๐ง
(6.3)
and hence
โซ ห ๐0 ๐๐(๐ก) ๐ด(๐ง) โ ๐ด(๐ง) + ๐ต(๐ง)๐ด(๐ง). ๐0 โ ๐ง ฮ0 ๐ก โ ๐ง For ๐ โ ker ๐ด(๐0 ) we get โซ ห ๐0 ๐๐(๐ก) ๐ =โ (๐ด(๐ง) โ ๐ด(๐0 ))๐ โ (๐ด(๐ง) โ ๐ด(๐0 ))๐ + ๐ต(๐ง)๐ด(๐ง)๐. (6.4) ๐0 โ ๐ง ฮ0 ๐ก โ ๐ง ๐ผ =โ
The second term on the right-hand side can be written as โซ ห ๐๐(๐ก) ๐ด(๐ง) โ ๐ด(๐0 ) (๐ง โ ๐0 ) ๐. ๐ก โ ๐ง ๐ง โ ๐0 ฮ0 Since lim๐งโ๐0
๐ด(๐ง) โ ๐ด(๐0 ) = ๐ดโฒ (๐0 ) in operator norm, we ๏ฌnd ๐ง โ ๐0 โซ ห ๐๐(๐ก) lim (๐ด(๐ง) โ ๐ด(๐0 ))๐ = 0, ๐งโ๐0 ฮ ๐กโ๐ง 0
and (6.4) implies (6.2).
โก
The formula (6.2) implies the desired description of ๐0 : Proposition 6.2. If ๐0 โ ฮ0 is an eigenvalue of the self-adjoint operator function ๐ด(๐ง) and ๐0 denotes the orthogonal projection onto ker ๐ด(๐0 ), then ๐({๐0 }) = (๐0 ๐ดโฒ (๐0 )๐0 )โ1 ๐0 .
(6.5)
Self-adjoint Analytic Operator Functions
459
If ๐0 is isolated and of ๏ฌnite multiplicity, a corresponding formula was given in [9, Lemma 2.1]. The relation (6.5) implies $ (6.6) ๐ฮ ({๐0 }) = ๐({๐0 })$โ(ฮ) , comp. Corollary 6.3. In the following we need some auxiliary statements. (i) If ๐บ is a self-adjoint operator, ๐ธ its spectral function, and ๐0 โ โ, then for each vector โ it holds lim ๐๐ฆ(๐บ โ ๐0 โ ๐๐ฆ)โ1โ = โ๐ธ({๐0 })โ.
๐ฆโ0
This fact is known and can also easily be checked. In a similar way it follows from (1.6): (ii) If ๐ด(๐ง) is as above and ๐0 โ ฮ0 , then lim ๐๐ฆ ๐ด(๐0 + ๐๐ฆ)โ1 ๐ = ๐({๐0 })๐,
๐ฆโ0
๐ โ โ.
This relation and (6.6) imply (iii) lim๐ฆโ0 ๐๐ฆ ๐ดฮ (๐0 + ๐๐ฆ)โ1 ๐ = ๐({๐0 })๐, ๐ โ โ. 2. In this subsection we give another proof of Theorem 4.3, using the following Schur factorization of a 2 ร 2 block operator matrix. Let โ = โ1 โ โ2 ,
(6.7)
and let ๐บ โ โ(โ) have the corresponding matrix representation ( ) ๐บ11 ๐บ12 ๐บ= . ๐บ21 ๐บ22 If the operators ๐บ22 and ๐บ11 โ ๐บ12 ๐บโ1 22 ๐บ21 are invertible, then also ๐บ is invertible and ( )( )(( ) )โ1 ๐ผ 0 ๐ผ ๐บ12 ๐บโ1 0 ๐บ11 โ ๐บ12 ๐บโ1 22 22 ๐บ21 ๐บโ1 = (6.8) โ๐บโ1 ๐ผ 0 ๐ผ 0 ๐บโ1 22 ๐บ21 22 To prove Theorem 4.3, the decomposition (6.7) is chosen as โ = โ(ฮ) โ โ(ฮ)โฅ .
(6.9) ๐0โ โฑฮ
= โ(ฮ), and hence It follows from [7, Theorem 2.4 and Theorem 4.1] that โฅ ๐ โ(ฮ)โฅ โ โฑฮ . Therefore the basic relation and the fact that (ฮ โ ๐ง)โ1 ๐ for โฅ ๐ โ โฑฮ is analytic on ฮi , the interior of ฮ, imply that ๐ด(๐ง)โ1 ๐, ๐ โ โ(ฮ)โฅ , is analytic on ฮi . In the matrix representation of ๐ด(๐ง)โ1 with respect to the decomposition (6.9): ( ) ๐11 (๐ง) ๐12 (๐ง) โ1 , (6.10) ๐ด(๐ง) =: ๐21 (๐ง) ๐22 (๐ง) with (1.1) we obtain ๐11 (๐ง) = ๐ฮ ๐ด(๐ง)โ1 ๐ฮ = โ๐ฮ ๐0โ (ฮ โ ๐ง)โ1 ๐ ๐ฮ + ๐ฮ ๐ต(๐ง)๐ฮ .
(6.11)
460
H. Langer, A. Markus and V. Matsaev
Since ๐ด(๐ง)โ1 ๐ is analytic in ฮi for ๐ โ โ(ฮ)โฅ , the operator functions ๐12 (๐ง) and ๐22 (๐ง) are analytic on ฮi , and because of the self-adjointness of ๐ด(๐ง) this holds also for ๐21 (๐ง). We show that ๐22 (๐ง) is boundedly invertible on ฮi . Assume that for some ๐ง0 โ ฮi there exists a sequence (๐๐ ) โ โ(ฮ)โฅ , โฅ๐๐ โฅ = 1, such that ๐22 (๐ง0 )๐๐ โ 0 if ๐ โ โ. Then the โ-valued functions ( ) ( ) 0 ๐12 (๐ง)๐๐ ๐ฆ๐ (๐ง) := ๐ด(๐ง)โ1 = , ๐ = 1, 2, . . . , ๐22 (๐ง)๐๐ ๐๐ are analytic on ฮi (this means that they have analytic continuations from the set of non-real points) since the expressions on the right-hand side are analytic on ฮi . Denote ) ( ( ) 0 ๐12 (๐ง0 )๐๐ , ๐ฃ๐ = ๐ข๐ = , ๐ = 1, 2, . . . . ๐22 (๐ง0 )๐๐ 0 Since ๐ฃ๐ โ โ(ฮ), we obtain from (5.6) โฅ๐ฮ ๐ด(๐ง0 )๐ฃ๐ โฅ โฅ ๐พโฅ๐ด(๐ง0 )๐ฃ๐ โฅ, Further,
(
๐ด(๐ง0 )๐ฃ๐ =
0 ๐๐
๐ = 1, 2, . . . .
(6.12)
) โ ๐ด(๐ง0 )๐ข๐ ,
Since ๐ข๐ โ 0 it follows that
๐ฮ ๐ด(๐ง0 )๐ฃ๐ = โ๐ฮ ๐ด(๐ง0 )๐ข๐ , (
0 ๐ด(๐ง0 )๐ฃ๐ โ ๐๐
๐ = 1, 2, . . . .
) โ 0,
๐ โ โ,
(6.13)
and ๐ฮ ๐ด(๐ง0 )๐ฃ๐ โ 0,
๐ โ โ.
(6.14)
Now, if ๐ โ โ, (6.12) and (6.14) imply ๐ด(๐ง0 )๐ฃ๐ โ 0, and from (6.13) it follows that ๐๐ โ 0, a contradiction. The relation (6.10) yields ( )โ1 ๐11 (๐ง) ๐12 (๐ง) ๐ด(๐ง) = . ๐21 (๐ง) ๐22 (๐ง) Now we apply the Schur factorization (6.8) to ๐บ = ๐ด(๐ง)โ1 . Then the left upper block in the matrix for ๐ด(๐ง) equals )โ1 ( ๐11 (๐ง) โ ๐12 (๐ง)๐22 (๐ง)โ1 ๐21 (๐ง) that is
$ ๐ดฮ (๐ง) = ๐ฮ ๐ด(๐ง)$โ(ฮ) = (๐11 (๐ง) โ ๐12 (๐ง)๐22 (๐ง)โ1 ๐21 (๐ง))โ1 .
This relation and (6.11) imply ๐ดฮ (๐ง)โ1 = ๐11 (๐ง) โ ๐12 (๐ง)๐22 (๐ง)โ1 ๐21 (๐ง) = โ๐ฮ ๐0โ (ฮโ๐ง)โ1๐ ๐ฮ +๐ฮ ๐ต(๐ง)๐ฮ โ๐12 (๐ง)๐22 (๐ง)โ1 ๐21 (๐ง).
(6.15)
Self-adjoint Analytic Operator Functions
461
With the spectral subspace โฑฮ0 โฮ of ฮ, corresponding to the set ฮ0 โ ฮ, and the restriction ฮฮ0 โฮ of ฮ to this spectral subspace, the ๏ฌrst term on the right-hand side can be written as ๐ฮ ๐0โ ๐ธ(ฮ)(ฮฮ โ ๐ง)โ1 ๐ธ(ฮ)๐ ๐ฮ + ๐ฮ ๐0โ ๐ธ(ฮ0 โ ฮ)(ฮฮ0 โฮ โ ๐ง)โ1 ๐ธ(ฮ0 โ ฮ)๐ ๐ฮ , and (6.15) becomes ๐ดฮ (๐ง)โ1 + ๐ฮ ๐0โ ๐ธ(ฮ)(ฮฮ โ ๐ง)โ1 ๐ธ(ฮ)๐ ๐ฮ = โ ๐ฮ ๐0โ ๐ธ(ฮ0 โ ฮ)(ฮฮ0 โฮ โ ๐ง)โ1 ๐ธ(ฮ0 โ ฮ)๐ ๐ฮ + ๐ฮ ๐ต(๐ง)๐ฮ โ ๐12 (๐ง)๐22 (๐ง)โ1 ๐21 (๐ง). The operator function on the right-hand side is analytic on ฮi . The operator function on the left-hand side is analytic on a set ๐ฐ โ ฮ, where ๐ฐ is a complex neighbourhood of ฮ: for ๐ดฮ (๐ง)โ1 this follows from [7, Theorem 3.1 (9)], for ฮฮ it is clear from its de๏ฌnition. Therefore the only possible singularities of the expressions on the two sides of this equality are the endpoints of ฮ = [๐ผ, ๐ฝ]. Consider, e.g., the left endpoint ๐ผ. For ๐ โ โฮ , ๐ โ= 0, the function (( ) ) ๐(๐ง) := ๐ดฮ (๐ง)โ1 + ๐ฮ ๐0โ ๐ธ(ฮ)(ฮฮ โ ๐ง)โ1 ๐ธ(ฮ)๐ ๐ฮ ๐, ๐ is analytic in ๐ฐ๐ (๐ผ) := {๐ง : 0 < โฃ๐ง โ ๐ผโฃ < ๐}, for some ๐ > 0, and we have ๐ถ โฃ๐(๐ง)โฃ โค for ๐ง โ ๐ฐ๐ (๐ผ) โ โ. For the term from the second summand in โฃ Im ๐งโฃ the sum on the right-hand side this estimate is obvious, for the ๏ฌrst summand it follows from [6, Proposition 2.1] or (1.6). According to [8, Lemma 33.4], ๐(๐ง) has a simple pole in ๐ผ or is analytic there. The ๏ฌrst case cannot hold since the residue of ๐(๐ง) at its simple pole ๐ผ is zero. To show this it is enough to check that lim ๐๐ฆ๐(๐ผ + ๐๐ฆ) = 0.
๐ฆโ0
Statement (iii) implies that the contribution of the ๏ฌrst summand in the sum on the right-hand side for ๐(๐ง) equals (๐({๐ผ})๐, ๐ ), whereas (i) and the relations (1.2), (6.6) show that the contribution of the second summand equals โ(๐({๐ผ})๐, ๐ ). It remains to prove the minimality of ฮฮ , that is, that an arbitrary ๐ โ โฑฮ can be approximated by ๏ฌnite sums of the form ๐ โ
(ฮฮ โ ๐ง๐,๐ )โ1 ๐ธ(ฮ)๐ ๐ฮ ๐ฅฮ ๐,๐ ,
(6.16)
๐=1
with ๐ง๐,๐ โ ๐ช, a nonempty open subset of ๐(ฮ) โฉ ๐(ฮฮ ), and ๐ฅฮ ๐,๐ โ โ(ฮ), ๐ = 1, 2, . . . , ๐, ๐ = 1, 2, . . . . โ Since the linearization ฮ is minimal, ๐ โ โฑ can be ap๐ proximated by elements ๐=1 (ฮ โ ๐ง๐,๐ )โ1 ๐ ๐ฅ๐,๐ with ๐ฅ๐,๐ โ โ. Because of ๐ โ โฑฮ โ๐ we can also use ๐=1 (ฮฮ โ๐ง๐,๐ )โ1 ๐ธ(ฮ)๐ ๐ฅ๐,๐ , and if we decompose ๐ฅ๐,๐ according โฅ to (6.9) as ๐ฅ๐,๐ = ๐ฅฮ + ๐ฅโฒ๐,๐ with ๐ฅโฒ๐,๐ โ โ(ฮ)โฅ and observe that ๐ โ(ฮ)โฅ โ โฑฮ , โ๐ ๐,๐ โ1 ฮ it follows that ๐=1 (ฮฮ โ ๐ง๐,๐ ) ๐ธ(ฮ)๐ ๐ฅ๐,๐ is an approximating sequence. Since ๐ฅฮ ๐,๐ โ โ(ฮ), this sequence coincides with (6.16), and the proof is complete.
462
H. Langer, A. Markus and V. Matsaev
Corollary 6.3. The local spectral function ๐ฮ of ๐ดฮ (๐ง) is the restriction of the local spectral function ๐ of ๐ด(๐ง) to its invariant subspace โ(ฮ): $ (6.17) ๐ฮ (ฮ) = ๐(ฮ)$โ(ฮ) , where ฮ is any subinterval of ฮ. To see this we observe that for the linearization of the operator function ๐ดฮ (๐ง) the operator ๐ธ(ฮ)๐ ๐ฮ plays the role of ๐ . Hence ๐ฮ (ฮ) = ๐ฮ ๐ โ ๐ธ(ฮ)๐ธ(ฮ)๐ธ(ฮ)๐ ๐ฮ = ๐ฮ ๐ โ ๐ธ(ฮ)๐ ๐ฮ = ๐ฮ ๐(ฮ)๐ฮ , which implies (6.17).
References ห [1] B. Curgus, A. Dijksma, H. Langer, H.S.V. de Snoo: Characteristic functions of unitary colligations and of bounded operators in Krein spaces. Operator Theory: Adv. Appl. 41 (1989), 125โ152. [2] I. Gohberg, M.A. Kaashoek, D.C. Lay: Equivalence, linearization and decomposition of holomorphic operator functions. J. Funct. Anal. 28 (1978), 102โ144. [3] M.A. Kaashoek, C.V.M. van der Mee, L. Rodman: Analytic operator functions with compact spectrum. I. Spectral nodes, linearization and equivalence. Integral Equations Operator Theory 4 (1981), 504โ547. [4] H. Langer, A. Markus, V. Matsaev: Locally de๏ฌnite operators in inde๏ฌnite inner product spaces. Math. Ann. 308 (1997), 405โ424. [5] H. Langer, A. Markus, V. Matsaev: Linearization and compact perturbation of selfadjoint analytic operator functions. Operator Theory: Adv. Appl. 118 (2000), 255โ 285. [6] H. Langer, A. Markus, V. Matsaev: Self-adjoint analytic operator functions and their local spectral function. J. Funct. Anal. 235 (2006), 193โ225. [7] H. Langer, A. Markus, V. Matsaev: Self-adjoint Analytic Operator Functions: Local Spectral Function and Inner Linearization. Integral Equations Operator Theory 63 (2009), 533โ545. [8] A.S. Markus: Introduction to the Spectral Theory of Polynomial Operator Pencils. AMS Translations of Mathematical Monographs, vol. 71, 1988. [9] A. Markus, V. Matsaev: On the basis property for a certain part of the eigenvectors and associated vectors of a self-adjoint operator pencil. Math. USSR Sbornik 61 (1988), 289โ307. [10] A.I. Virozub, V.I. Matsaev: The spectral properties of a certain class of selfadjoint operator-valued functions. Funct. Anal. Appl. 8 (1974), 1โ9.
Self-adjoint Analytic Operator Functions H. Langer Institute for Analysis and Scienti๏ฌc Computing Vienna University of Technology Wiedner Hauptstrasse 8โ10 A-1040 Vienna, Austria e-mail:
[email protected] A. Markus Department of Mathematics Ben-Gurion University of the Negev P.O. Box 653 84105 Beer-Sheva, Israel e-mail:
[email protected] V. Matsaev Department of Mathematics School of Mathematical Sciences Tel Aviv University 69978 Ramat Aviv, Israel e-mail:
[email protected]
463
Operator Theory: Advances and Applications, Vol. 218, 465โ494 c 2012 Springer Basel AG โ
An Estimate for the Splitting of Holomorphic Cocycles. One Variable Jยจ urgen Leiterer Dedicated to the memory of my teacher Israel Gohberg
Abstract. It is well known that every holomorphic cocycle over a domain in the complex plane and with values in the group of invertible elements of a Banach algebra, which is su๏ฌciently close to the unit cocycle, splits holomorphically. We prove this result with certain uniform estimates. Mathematics Subject Classi๏ฌcation (2000). 47A56 32L99. Keywords. Holomorphic cocycle, splitting of cocycles, uniform estimates.
1. Introduction Let ๐ท be an open set in the complex plane, let ๐ฐ = {๐๐ }๐โ๐ผ be an open covering of ๐ท, let ๐ด be a Banach ( algebra ) with unit, 1, and let ๐บ๐ด be the group of invertible of all elements of ๐ด. Let ๐ถ 0 ๐ฐ, ๐ช๐บ๐ด be the ( set ๐บ๐ด ) families ๐ = {๐๐ }๐โ๐ผ of holomorphic 1 be the set of families ๐ = {๐๐๐ }๐,๐โ๐ผ functions ๐๐ : ๐๐ โ ๐บ๐ด, and let ๐ ๐ฐ, ๐ช of holomorphic functions ๐๐๐ : ๐๐ โฉ ๐๐ โ ๐บ๐ด satisfying the cocycle condition ๐๐๐ ๐๐๐ = ๐๐๐
on ๐๐ โฉ ๐๐ ,
Set dist(๐, 1) =
sup
๐โ๐ผ, ๐โ๐๐
โฅ๐๐ (๐) โ 1โฅ
and dist(๐, 1) =
sup
๐,๐โ๐ผ, ๐โ๐๐ โฉ๐๐
โฅ๐๐๐ (๐) โ 1โฅ
๐, ๐ โ ๐ผ. ( ) for ๐ โ ๐ถ 0 ๐ฐ, ๐ช๐บ๐ด ( ) for ๐ โ ๐ 1 ๐ฐ, ๐ช๐บ๐ด .
From the theory of Grauert [G] and Bungart [B] (for one variable, see ( ) also Theorem 5.6.3 in [GL]) it is well known that, for each ๐ โ ๐ 1 ๐ฐ, ๐ช๐บ๐ด with
466
J. Leiterer
( ) dist(๐, 1) < 1, there exists ๐ข โ ๐ถ 0 ๐ฐ, ๐ช๐บ๐ด such that ๐๐๐ = ๐ข๐ ๐ขโ1 ๐
on ๐๐ โฉ ๐๐ ,
๐, ๐ โ ๐ผ.
1
(1.1)
In this paper, we prove the following theorem (see Theorem 4.2, for a slightly more precise version). 1.1. Theorem. Suppose ๐ท is bounded. Let ๐ be the diameter of ๐ท and let ๐ > 0. Assume, for each ๐ โ ๐ท, there exists ๐ โ ๐ผ such that $ { } ๐ท โฉ ๐ โ โ $ โฃ๐ โ ๐โฃ < ๐ โ ๐๐ . 2 (1.2) ( ) Then, for each ๐ โ ๐ 1 ๐ฐ, ๐ช๐บ๐ด satisfying the estimate ๐ dist(๐, 1) โค 26 , (1.3) 2 ๐ ( ) there exists ๐ข โ ๐ถ 0 ๐ฐ, ๐ช๐บ๐ด which solves the Cousin problem (1.1) and satis๏ฌes 225 ๐ dist(๐, 1). (1.4) ๐ Of course, the constant 225 is not optimal. We present it just to show that there is a constant at this place which is independent of ๐ท, ๐, and the Banach algebra ๐ด. The same is true for similar constants during the paper. After linearization, the Cousin problem (1.1) leads to the inhomogeneous Cauchy-Riemann equation. Since, on bounded domains, this equation admits a solution with uniform estimates (cf. Section 3 below), an appropriate version of the implicit function theorem quickly leads to the following result: Let the hypotheses of Theorem 1.1 be ful๏ฌlled and)let ๐ > 0. Then there exists a constant ๐ฟ(> 0 such) ( that, for each ๐ โ ๐ 1 ๐ฐ, ๐ช๐บ๐ด with dist(๐, 1) < ๐ฟ, there exists ๐ข โ ๐ถ 0 ๐ฐ, ๐ช๐บ๐ด which solves the Cousin problem (1.1) and satis๏ฌes the estimate dist(๐ข, 1) < ๐. So it is natural to analyze the proof of the implicit function theorem in order to get a proof of Theorem 1.1. However, the author did not succeed in this way (only a weaker estimate was obtained). Therefore, we go another way. First we study the equation โ๐ ๐ โ1 = ๐, (1.5) โ๐ง where ๐ : ๐ โ ๐ด is a given continuous function and ๐ is searched as a continuous function from ๐ท to ๐บ๐ด. To the knowledge of the author, this equation appears for the ๏ฌrst time in the work of Cornalba and Gri๏ฌths [CG], where, using the Newlander-Nierenberg theorem, local solvability is obtained (for the case ๐ด = ๐ฟ(๐, โ), the algebra of complex ๐ ร ๐ matrices). Then, Gennadi Henkin found a another proof for the local solvability, using uniform estimates for the inhomogeneous Cauchy-Riemann equation.3 Henkinโs proof has the advantage that it gives local solutions with uniform estimates. Analyzing the proof of Henkin, in Section 5, dist(๐ข, 1) โค
1 If
๐บ is connected, this is true also without the condition dist(๐, 1) < 1. (Def. 2.2) we call such coverings ๐-separated. 3 To the knowledge of the author, this proof is published only in the form of an exercise in the book [HL] (Exercise 10 at the end of Chapter 2). 2 Below
An Estimate for the Splitting
467
we obtain a global solution of (1.5) with appropriate uniform estimates, provided ๐ is su๏ฌciently small (Theorem 5.1). Then we prove a version of Theorem 1.1 for the class of continuous functions with continuous Cauchy-Riemann derivative (Theorem 8.3). In the last section, we deduce Theorem 1.1 from Theorems 8.3 and 5.1. The author has two motivations for the present paper. One motivation is to provide the Weierstrass product theorems obtained in [GR1, GR2, GL] for operator functions with some estimates. The second motivation is to provide the Oka-Grauert principle with certain estimates. The latter is also the motivation for another paper of the author [L], where the case of several variables is studied (for ๐ด = ๐ฟ(๐, โ)). Finally let us compare Theorem 1.1 with the following result of B. Berndtsson and J.-P. Rosay [BR]: Let ๐ท = ๐ป be the unit disc in the complex plane, and let ๐บ = ๐บ๐ฟ(๐, โ), the group of invertible complex ๐ ร ๐ matrices. Assume ( ) condition 1 ๐บ๐ฟ(๐,โ) (1.2) is satis๏ฌed for some ๐ > 0. Then, for each ๐ โ ๐ ๐ฐ, ๐ช satisfying the condition โฅ๐ โฅ := sup โฅ๐๐๐ (๐)โฅ < โ, (1.6) 0
(
there exists ๐ข โ ๐ถ ๐ฐ, ๐ช ๏ฌes both โฅ๐ขโฅ :=
sup
๐โ๐ผ, ๐โ๐๐
๐,๐โ๐ผ, ๐โ๐๐ โฉ๐๐
๐บ๐ฟ(๐,โ)
)
โฅ๐ข๐(๐)โฅ < โ
which solves the Cousin problem (1.1) and satis๐๐๐ โฅ๐ขโ1 โฅ :=
sup
๐โ๐ผ, ๐โ๐๐
โฅ๐ขโ1 ๐ (๐)โฅ < โ.
(1.7)
Of course, our condition (1.3) is much stronger than condition (1.6). However, it seems to the author that the method of [BR], under the stronger condition (1.3) (also in the case of matrices), does not give estimate (1.4), although some weaker estimate (not explicitly stated in [BR]) can be obtained analyzing the proof of [BR].
2. Notation Throughout this paper the following notations are used. โ โ is the set of natural numbers, zero included. โโ = โ โ {0}. โค is the set of integers. โ is the complex plane. โ is the real line. โ Banach spaces and Banach algebras are always complex. โ The Lebesgue measure on โ will be denoted by ๐๐. โ Let ๐ท โ โ be an open set, let ๐ธ be a Banach space, and let ๐ : ๐ท โ ๐ธ be continuous. If ๐ is of class ๐ 1 , then we denote by โ๐ the function (and not a di๏ฌerential form) de๏ฌned by ( ) 1 โ๐ โ๐ โ๐ = +๐ 2 โ๐ฅ โ๐ฆ where ๐ฅ, ๐ฆ are the canonical real coordinate functions on โ. If ๐ is only continuous (and possibly not di๏ฌerentiable), then we say that โ๐ is continuous
468
J. Leiterer if there is a continuous function ๐ฃ : ๐ท โ ๐ธ such that โซ โซ ๐๐ฃ ๐๐ = โ (โ๐)๐ข ๐๐ ๐ท
๐ท
(2.1)
for all ๐ถ โ -functions ๐ : ๐ท โ โ with compact support. This function ๐ฃ (which then is uniquely determined) will be denoted by โ๐ . โ If ๐ธ is a Banach space with the norm โฅ โ
โฅ, ๐ is a subset of โ, and ๐ is an ๐ธ-valued function de๏ฌned on ๐, then we set โฅ๐ โฅ๐ = sup โฅ๐ (๐ง)โฅ. ๐งโ๐
(2.2)
โ If ๐ is a subset of โ, then we denote by ๐ the closure of ๐ in โ, and by int ๐ we denote the interior of ๐ with respect to โ. โ If ๐ โ โ, ๐ธ is a Banach space, and ๐ is an ๐ธ-valued function with the domain of de๏ฌnition ๐, then the support of ๐ , supp ๐ , is the maximal relatively closed subset of ๐ such that ๐ โก 0 outside of it. 2.1. In order to give our results also for holomorphic functions which admit a continuous extension to the boundary, or to some part of the boundary of their domain of de๏ฌnition, we will consider sets ๐ โ โ with the property that ๐ โ int ๐.
(2.3)
By a ๐ โ -function on such a set ๐ we mean a function which comes from a ๐ โ function de๏ฌned in some open (with respect to โ) neighborhood of ๐. As a consequence of (2.3), the derivatives of such functions are well de๏ฌned on ๐ by their values on int ๐. The following de๏ฌnition will be used throughout the paper. 2.2. De๏ฌnition. Let ๐ โ โ, let ๐ฐ = {๐๐ }๐โ๐ผ be a covering of ๐ by relatively open sets4 , and let ๐ > 0. Then ๐ฐ will be called if for} each point ๐ โ ๐, $ { ๐-separated there exists an index ๐ โ ๐ผ such that ๐ โฉ ๐ โ โ $ โฃ๐ โ ๐โฃ < ๐ โ ๐๐ .
3. An estimate for the Pompeiju integral 3.1. Let ๐ธ be a Banach space, let ๐ท โ โ be a bounded open set, and let ๐ : ๐ท โ ๐ธ be continuous and bounded. Then it is well known (see, e.g., Theorem 2.1.9 in [GL]) that the function ๐ข : ๐ท โ ๐ธ de๏ฌned by the Pompeiju integral [P] โซ 1 ๐ (๐) ๐ข(๐ง) = โ ๐๐(๐), ๐ง โ ๐ท, (3.1) ๐ ๐ท ๐โ๐ง is continuous on ๐ท and solves the equation โ๐ข = ๐ 4 i.e.,
on ๐ท.
a covering which comes from an open covering of an open (in โ) neighborhood of ๐.
An Estimate for the Splitting
469
Moreover, if ๐ is the diameter of ๐ท, then it is easy to see that โ โฅ๐ขโฅ๐ท โค ๐ 2โฅ๐ โฅ๐ท . โ The constant ๐ 2 is not optimal. But without additional geometric conditions on ๐ท it cannot be improved so much (the case of a square shows that it is > ๐). However, if ๐ท is contained in a โlong and thinโ rectangle, the constant can be improved essentially. To make this precise, we give a de๏ฌnition. 3.2. De๏ฌnition. Let ๐ be a bounded subset of โ such that int ๐ โ= โ
. Denote by ๐๐ the set of pairs (๐, ๐) โ โ2 with 0 < ๐ โค ๐ such that ๐ is contained in a closed rectangle with side lengths ๐ and ๐. As int ๐ โ= โ
, ๐๐ is a closed and bounded in โ2 . Therefore ( ) โ 2 ๐ ๐ถ๐ := min ๐ 2 + log ๐ ๐ (๐,๐)โ๐๐ exists. ๐ถ๐ will be called the rectangle constant of ๐. 3.3. Proposition. Let ๐ธ be a Banach space, let ๐ท โ โ be a bounded open set with the rectangle constant ๐ถ๐ท , and let ๐ : ๐ท โ ๐ธ be a bounded continuous function. Then the solution ๐ข of โ๐ข = ๐ de๏ฌned on ๐ท by the Pompeiju integral (3.1) admits the estimate โฅ๐ขโฅ๐ท โค ๐ถ๐ท โฅ๐ โฅ๐ท . (3.2) Proof. By de๏ฌnition of ๐ถ๐ท , ๐ท is contained in a rectangle with side lengths ๐ and ๐, where ๐ โค ๐, and ( ) โ 2 ๐ ๐ถ๐ท = ๐ 2 + log . ๐ ๐ After a shift and a rotation of ๐ท, we may assume that $ { } $ ๐ท โ ๐ง = ๐ฅ + ๐๐ฆ โ โ $ 0 โค ๐ฅ โค ๐, 0 โค ๐ฆ โค ๐ . Set
{ } $ ๐โ๐ ๐+๐ $ ๐ง = ๐ฅ + ๐๐ฆ โ โ $ โค๐ฅโค , 0โค๐ฆโค๐ , 2 2 { } $ ๐ โ ๐ $ ๐
1 = ๐ง = ๐ฅ + ๐๐ฆ โ โ $ 0 โค ๐ฅ โค , 0โค๐ฆโค๐ , 2 { } $ ๐+๐ $ ๐
2 = ๐ง = ๐ฅ + ๐๐ฆ โ โ $ โค ๐ฅ โค ๐, 0 โค ๐ฆ โค ๐ , 2 ๐ ๐ ๐ง0 = + ๐ . 2 2 Then, for all ๐ง โ ๐ท, โซ โซ โฅ๐ โฅ๐ท ๐๐(๐) โฅ๐ โฅ๐ท ๐๐(๐) โฅ๐ข(๐ง)โฅ โค โค . ๐ โฃ๐ โ ๐งโฃ ๐ โฃ๐ โ ๐ง0 โฃ ๐
0 =
๐
1 โช๐
0 โช๐
2
๐
1 โช๐
0 โช๐
2
(3.3)
470
J. Leiterer
Since
โซ ๐
0
and
โซ ๐
1 โช๐
2
๐๐(๐) < โฃ๐ โ ๐ง0 โฃ
โซ โ โฃ๐โ๐ง0 โฃ<๐/ 2
๐๐(๐) =2 โฃ๐ โ ๐ง0 โฃ
โซ
๐
1
๐๐(๐) = โฃ๐ โ ๐ง0 โฃ
๐๐(๐) <2 โฃ๐ โ ๐ง0 โฃ
โซ
โ ๐/ โซ 2
0
โซ
๐โ๐ 2
0
0
โ 2๐๐ ๐๐ = ๐๐ 2 ๐
๐ ๐ 2
๐๐ฆ ๐ ๐๐ฅ = 2๐ log , ๐ โ๐ฅ
this implies (3.2).
โก
4. The main result In this section, ๐ด is a Banach algebra with unit, 1, and ๐บ๐ด is the group of invertible elements of ๐ด. 4.1. Multiplicative cocycles. Let ๐ be a subset of โ such that ๐ โ int ๐. Then we denote by โฌ๐ช๐ด (๐) the algebra of all bounded continuous functions ๐ : ๐ โ ๐ด which are holomorphic in int ๐, and by โฌ๐ช๐บ๐ด (๐) we denote the group of all ๐ โ โฌ๐ช๐ด (๐) such that ๐ (๐) โ ๐บ๐ด for all ๐ โ ๐ and, moreover, the function ๐ โ1 is also bounded on ๐. The unit element of this group, i.e., the constant function with value 1, will also be denoted by 1. Now let ๐ฐ = {๐๐ }๐โ๐ผ a covering of ๐ by relatively open subsets of ๐. Then we use the following notations: ( ) ๐ถ 0 ๐ฐ, โฌ๐ช๐บ๐ด is the set of families {๐๐ }๐โ๐ผ of functions5 ๐๐ โ โฌ๐ช๐บ๐ด (๐๐ ), ( ) ๐ถ 1 ๐ฐ, โฌ๐ช๐บ๐ด is the set of families {๐๐๐ }๐,๐โ๐ผ of functions6 ๐๐๐ โ โฌ๐ช๐บ๐ด (๐๐ โฉ ๐๐ ), ( ) ( ) and ๐ 1 ๐ฐ, โฌ๐ช๐บ๐ด is the set of all ๐ โ ๐ถ 1 ๐ฐ, โฌ๐ช๐บ๐ด satisfying the multiplicative cocycle condition ๐๐๐ ๐๐๐ = ๐๐๐ on ๐๐ โฉ ๐๐ โฉ ๐๐ , ๐, ๐, ๐ โ ๐ผ. 7 (4.1) ( ) The elements of ๐ 1 ๐ฐ, โฌ๐ช๐บ๐ด are called multiplicative cocycles. Note that the cocycle condition (4.1) implies that โ1 ๐๐๐ = ๐๐๐
and 5 By
๐๐๐ = 1
on ๐๐ โฉ ๐๐ , on ๐๐ ,
๐, ๐ โ ๐ผ,
๐ โ ๐ผ.
(4.2) (4.3)
a family of functions {๐๐ }๐โ๐ผ we mean a map which is de๏ฌned only for the indices ๐ โ ๐ผ with ๐๐ โ= โ
. 6 By a family of functions {๐ } ๐๐ ๐,๐โ๐ผ we mean a map which is de๏ฌned for the ordered pairs (๐, ๐) โ ๐ผ ร ๐ผ with ๐๐ โฉ ๐๐ โ= โ
. 7 More precisely, the cocycle condition means the following: If (๐, ๐, ๐) โ ๐ผ ร ๐ผ ร ๐ผ is an ordered triplet such that ๐๐ โฉ ๐๐ โฉ ๐๐ โ= โ
, then ๐๐๐ (๐)๐๐๐ (๐) = ๐๐๐ (๐) for all ๐ โ ๐๐ โฉ ๐๐ โฉ ๐๐ . In particular, if ๐๐ โฉ ๐๐ โฉ ๐๐ = โ
for all pairwise di๏ฌerent triplets (๐, ๐, ๐) โ ๐ผ ร ๐ผ ร ๐ผ, then the cocycle codition just means that (4.2) and (hence) (4.3) are satis๏ฌed. If moreover ๐๐ โฉ ๐๐ = โ
whenever ๐ โ= ๐, then the cocycle condition reduces to condition (4.3).
An Estimate for the Splitting
471
( ) The element ๐ โ ๐ถ 0 ๐ฐ, โฌ๐ช๐บ๐ด de๏ฌned by ๐๐ โก 1 on ๐๐ , ๐ โ ๐ผ, will be denoted by ( ) 1, and the cocycle ๐ โ ๐ 1 ๐ฐ, โฌ๐ช๐บ๐ด de๏ฌned by ๐๐๐ โก 1 on ๐๐ โฉ ๐๐ , ๐, ๐ โ ๐ผ, will be called the unit cocycle and also denoted by 1. Moreover, we de๏ฌne ( ) โฅ๐ โ 1โฅ = sup โฅ๐๐ โ 1โฅ๐๐ if ๐ โ ๐ถ 0 ๐ฐ, โฌ๐ช๐บ๐ด , (4.4) ๐โ๐ผ ( ) โฅ๐ โ 1โฅ = sup โฅ๐๐๐ โ 1โฅ๐๐ โฉ๐๐ if ๐ โ ๐ถ 1 ๐ฐ, โฌ๐ช๐บ๐ด , (4.5) ๐,๐โ๐ผ
where โฅ๐๐ โ 1โฅ๐๐ and โฅ๐๐๐ โ 1โฅ๐๐ โฉ๐๐ are de๏ฌned by (2.2). The โnumbersโ de๏ฌned by (4.4) and (4.5) can be in๏ฌnite if ๐ผ is in๏ฌnite. However, in this paper, we meet only those ๐ for which โฅ๐ โ 1โฅ < 1. Now the main result of this paper can be stated as follows. 4.2. Theorem. Let ๐ be a bounded subset of โ such that ๐ โ int ๐, and let ๐ถ๐ be the rectangle constant of ๐ (Section 3.2). Let ๐ > 0, let ๐ฐ = {๐๐ }๐โ๐ผ be an ๐-separated ( ) covering of ๐ by relatively open sets (De๏ฌnition 2.2), and let ๐ โ ๐ 1 ๐ฐ, โฌ๐ช๐บ๐ด be a multiplicative cocycle satisfying ๐ 1 and โฅ๐ โ 1โฅ โค . 224 ๐ถ๐ 64 ( ) Then there exists โ โ ๐ถ 0 ๐ฐ, โฌ๐ช๐บ๐ด such that โฅ๐ โ 1โฅ โค
๐๐๐ = โ๐ โโ1 ๐ and
on (
โฅโ โ 1โฅ โค
๐๐ โฉ ๐ ๐ ,
8
๐, ๐ โ ๐ผ,
) 223 ๐ถ๐ 2+ โฅ๐ โ 1โฅ. ๐
(4.6)
(4.7) (4.8)
Of course, the constants 224 and 223 in this theorem are not optimal. The interesting point is that they do not depend on ๐ and ๐.
5. An estimate for the equation ๐ผ โ1 ๐๐ผ = ๐ฝ In this section, ๐ด is a Banach algebra with unit, 1, ๐บ๐ด is the group of invertible elements of ๐ด, ๐ is a bounded subset of โ such that ๐ โ int ๐, and ๐ถ๐ is the rectangle constant of ๐ (Section 3.2). The aim of this section is to prove the following theorem. 5.1. Theorem. Let ๐ : ๐ โ ๐ด be a continuous function such that 1 โฅ๐ โฅ๐ โค . 8๐ถ๐ 8 The
(5.1)
rectangle constant ๐ถ๐ can be small compared to ๐2โ24 although the diameter of ๐ is big compared to ๐. (The case when the diameter of ๐ is small compared to ๐ is not interesting, because then the trivial covering {๐} is a re๏ฌnement of any ๐-separated covering of ๐.) Therefore, in general, the second estimate does not follow from the ๏ฌrst one.
472
J. Leiterer
Then there exists a continuous function ๐ : ๐ โ ๐บ๐ด such that โ๐ is also continuous on ๐ 9 , ๐ โ1 โ๐ = ๐ on ๐, (5.2) and โฅ๐ โ 1โฅ๐ โค 2๐ถ๐ โฅ๐ โฅ๐ .
(5.3)
The following lemma is the main step of the proof of this theorem. 5.2. Lemma. Let the hypotheses of Theorem 5.1 be ful๏ฌlled, and let โฌ๐ ๐ด(๐) be the Banach space of ๐ด-valued bounded continuous functions on ๐ endowed with the sup-norm (2.2). Then there exists a bounded linear operator ๐
: โฌ๐ ๐ด(๐) โ โฌ๐ ๐ด (๐) such that, for all ๐ โ โฌ๐ ๐ด(๐), โ๐
๐ is also bounded and continuous on ๐, โ๐
๐ โ (๐
๐ )๐ = ๐ and
(5.4)
โฅ๐
โฅ โค 2๐ถ๐ ,
(5.5) ๐ด
where โฅ๐
โฅ is the operator norm of ๐
as operator acting in โฌ๐ (๐). Proof. As noticed in Section 3, setting โซ 1 ๐ (๐) (๐ ๐ )(๐ง) = โ ๐๐(๐), ๐ ๐ โ๐ง
๐ง โ ๐,
int ๐
๐ด
for ๐ โ โฌ๐ (๐), we obtain a bounded linear operator ๐ : โฌ๐ ๐ด(๐) โ โฌ๐ ๐ด (๐) such that โ๐ ๐ = ๐ for all ๐ โ โฌ๐ ๐ด(๐) (5.6) and Set
โฅ๐ โฅ โค ๐ถ๐ . ๐ ๐ := โ(๐ ๐ )๐
(5.7)
for ๐ โ โฌ๐ ๐ด (๐).
By (5.1) and (5.7), this de๏ฌnes a bounded linear operator ๐ : โฌ๐ ๐ด (๐) โ โฌ๐ ๐ด (๐) with 1 โฅ๐ โฅ โค . (5.8) 4 By (5.8), id +๐ is invertible, where (id +๐ )โ1 =
โ โ
(โ๐ )๐
(5.9)
๐=0
and
โ 1 1 โ 1 4 1(id +๐ )โ1 1 โค = . ๐ 4 3
(5.10)
๐=0
9 By
this we mean that โ๐ is continuous in int ๐ (in the sense de๏ฌned in Section 2) and admits a continuous extension to ๐.
An Estimate for the Splitting This implies that
473
๐
:= ๐ (id +๐ )โ1
is a well-de๏ฌned bounded linear endomorphism of โฌ๐ ๐ด(๐), where, by (5.7) and (5.10), 4 โฅ๐
โฅ โค โฅ๐ โฅโฅ(id +๐ )โ1 โฅ โค ๐ถ๐ . 3 such that estimate (5.5) is satis๏ฌed. To prove that ๐
has also the other required properties, let ๐ โ โฌ๐ ๐ด (๐) be given. Then we see from (5.6) that โ๐
๐ is continuous on ๐ and โ๐
๐ = (id +๐ )โ1 ๐. Moreover, by de๏ฌnition of ๐ and ๐
, ( ) ๐ (id +๐ )โ1 ๐ = โ ๐ (id +๐ )โ1 ๐ ๐ = โ(๐
๐ )๐. Together this implies โ๐
๐ โ (๐
๐ )๐ = (id +๐ )โ1 ๐ + ๐ (id +๐ )โ1 ๐ = (id +๐ )(id +๐ )โ1 ๐ = ๐. So, also (5.4) is proved.
โก
5.3. Proof of Theorem 5.1. Let ๐
be the operator from Lemma 5.2. Set ๐ = 1+๐
๐ . As โ๐
๐ is continuous on ๐, then also โ๐ is continuous on ๐. (5.3) follows from (5.5). In view of (5.1), this further implies that โฅ๐ โ 1โฅ๐ โค 1/4. In particular, the values of ๐ are invertible. Moreover, from (5.4) we see that ( ) ( ) ( ) ( ) ๐ โ1 โ๐ = ๐ โ1 โ(1 + ๐
๐ ) = ๐ โ1 โ(๐
๐ ) = ๐ โ1 ๐ + (๐
๐ )๐ ( ) = ๐ โ1 (1 + ๐
๐ )๐ = ๐ โ1 ๐ ๐ = ๐. โก
6. Partitions of unity with estimates In this section we use the following notations. For ๐ โ โ and ๐ > 0, we set $ $ { } { } $ $ ๐ต(๐, ๐) = ๐ โ โ $ โฃ๐ โ ๐โฃ < ๐ and ๐ต(๐, ๐) = ๐ โ โ $ โฃ๐ โ ๐โฃ โค ๐ . For ๐ > 0 and ๐ = (๐1 , ๐2 ) โ โค2 , we set ๐1 ๐2 ๐๐๐ = โ ๐ + ๐ โ ๐. 2 2 It is easy to see that
( ) ๐ ๐ต ๐๐๐ , = โ. 2 2
โช ๐โโค
(6.1)
(6.2)
474
J. Leiterer
6.1. Lemma. Let ๐ฝ โ โค2 such that โฏ๐ฝ โฅ 22.10 Then โฉ ( ) ๐ต ๐๐๐ , ๐ = โ
for all ๐ > 0.
(6.3)
๐โ๐ฝ
Proof. For ๐ โ โค2 we denote by ๐ฝ(๐) the set of all indices ๐ โ โค2 such that โฃ๐๐๐ โ ๐๐๐ โฃ < 2๐. By (6.1), this can be written in the form $๐ ๐1 $$2 $$ ๐2 ๐2 $$2 $ 1 $ โ ๐ โ โ ๐$ + $ โ ๐ โ โ ๐$ < 4๐2 , 2 2 2 2 or, equivalently, โฃ๐1 โ ๐1 โฃ2 + โฃ๐2 โ ๐2 โฃ2 < 8. (6.4) Since, for ๏ฌxed ๐, the number of indices ๐ โ โค2 satisfying (6.4) is equal to 21, we get โฏ๐ฝ(๐) = 21 for all ๐ โ โค2 . (6.5) 2 Now let ๐ฝ โ โค with โฏ๐ฝ โฅ 22 be given. Then, by (6.5), for each ๐ โ ๐ฝ, there exist at least one index ๐ โ ๐ฝ such that ๐ โโ ๐ฝ(๐), i.e., โฃ๐๐๐ โ ๐๐๐ โฃ โฅ 2๐ and, hence, ๐ต(๐๐๐ , ๐) โฉ ๐ต(๐๐๐ , ๐) = โ
. In particular, then we have (6.3).
โก
6.2. Lemma. For each ๐ > 0, there exists a ๐ โ -partition of unity {๐๐ }๐โโค2 subordinate to the open covering { ( )} ๐ต ๐๐๐ , ๐ (6.6) 2 ๐โโค
of โ (by (6.2) this is indeed a covering of โ) such that $ $ $ $ $ โ๐๐ $ $ โ๐๐ $ 176 $ $, $ $ for all ๐ โ โค2 . $ โ๐ฅ $ $ โ๐ฆ $ โค ๐
(6.7)
Proof. Denote by โ on of the derivatives โ/โ๐ฅ and โ/โ๐ฆ. Take a ๐ โ -function ๐ : [0, โ[โ [0, 1] such that ๐ โก 1 in a neighborhood of [0, 1], ๐ โก 0 in a neighborhood of [2, โ[, and $ โฒ$ $๐ $ โค 2 everywhere on [0, โ[. (6.8) Set
( ) โฃ๐ โ ๐๐๐ โฃ2 ๐ ห๐ (๐) = ๐ 4 ๐2 2 Then, for all ๐ โ โค ,
for all ๐ โ โ and ๐ โ โค2 .
( ๐) ๐ ห๐ โก 1 in a neighborhood of ๐ต ๐๐๐ , , 2 ( ) ๐ supp ๐ ห๐ โ ๐ตโ ๐๐๐ , โ . 2
10 By
โฏ๐ฝ we denote the number of ๐ฝ.
(6.9) (6.10)
An Estimate for the Splitting Moreover, we set ๐=
โ
475
๐ ห๐ .
๐โโค2
By Lemma 6.1, the sum in the de๏ฌnition of ๐ is locally ๏ฌnite. Therefore, ๐ is a ๐ โ -function on โ and โ โ๐ = โ๐ ห๐ . (6.11) ๐โโค2
From (6.9) and (6.2) we see that ๐โฅ1
everywhere on โ.
(6.12)
Therefore, setting
/ ๐๐ = ๐ ห๐ ๐, ๐ โ โค2 . we obtain a ๐ โ partition of unity {๐๐ }๐โโค2 on โ. By (6.10) this partition of unity is subordinate to the covering (6.6). It remains to prove estimate (6.7). We have ( ) โฃ๐ โ ๐๐๐ โฃ2 4 โฒ (โ ๐ ห๐ )(๐) = 2 ๐ ห 4 (๐ โ ๐๐๐ ). ๐ ๐ ๐2 Taking into account (6.8) and the fact that, by (6.10), ( ) 4โฃ๐ โ ๐๐๐ โฃ2 โฃ๐ โ ๐๐ โฃ < ๐ if ๐ หโฒ๐ โ= 0, ๐2 this implies that
8 on โ. (6.13) ๐ Since, by (6.10) and Lemma 6.1, locally, the sum in (6.11) contains not more than 21 non-zero terms, this further implies that 168 โฃโ๐โฃ โค on โ. (6.14) ๐ As โ๐ ห๐ ๐ ห๐ โ๐๐ = โ 2 โ๐, ๐ โฅ 1, and ๐ ห๐ โค 1, ๐ ๐ from (6.13) and (6.14) we see that 8 168 176 โฃโ๐๐ โฃ โค โฃโ ๐ ห๐ โฃ + โฃโ๐โฃ โค + = . โก ๐ ๐ ๐ โฃโ ๐ ห๐ โฃ โค
6.3. Let ๐ โ โ such that ๐ โ int ๐, and let ๐ฐ = {๐๐ }๐โ๐ผ be a covering of ๐ by relatively open subsets of ๐. We say that {๐๐ }๐โ๐ผ is a ๐ โ partition of unity subordinate to ๐ฐ if (i) for each ๐ โ ๐ผ, ๐๐ is a non-negative real ๐ โ -function on ๐ (in the sense explained in Section 2.1) such that supp ๐๐ is compact and contained in ๐๐ , (ii) for each ๐ โ ๐, there exists a relative open neighborhood ๐ (๐) โ ๐ of ๐ such that ๐๐ โก 0 on ๐ (๐) for all ๐ โ ๐ผ except for a ๏ฌnite number; โ (iii) ๐โ๐ผ ๐๐ โก 1 on ๐.
476
J. Leiterer
If, for some ๐ โ โโ , the number of the set ๐ฝ(๐) in condition (ii) can be always chosen โค ๐, then ๐ฐ will be called of order โค ๐. Note that, by Lemma 6.1, each ๐ โ partition of unity, which is subordinate to the covering { } ๐ โฉ ๐ต(๐๐๐ , ๐) ๐โโค2 (6.15) is of order โค 21. We now combine Lemmas 6.1 and 6.2. 6.4. Lemma. Let ๐ โ โ, let ๐ฐ = {๐๐ }๐โ๐ผ be an ๐-separated covering of ๐ by relatively open subsets of ๐, ๐ > 0 (Def. 2.2). Then there exists a ๐ โ -partition of unity {๐๐ }๐โ๐ผ subordinate to ๐ฐ, which is of order โค 21 and such that $ $ $ $ $ โ๐๐ $ $ โ๐๐ $ 212 $ $, $ $ on ๐, ๐ โ ๐ผ. (6.16) $ โ๐ฅ $ $ โ๐ฆ $ โค ๐ Proof. Denote by โ on of the derivatives โ/โ๐ฅ and โ/โ๐ฆ. Since ๐ฐ is ๐-separated, the covering 6.15 is a re๏ฌnement of ๐ฐ, i.e., there is a map ๐ : โค2 โ ๐ผ such that ( ) ๐ โฉ ๐ต ๐๐๐ , ๐ โ ๐๐ (๐) , ๐ โ โค2 . (6.17) โ By Lemma 6.2,{ there ห๐ }๐โโค2 subordinate to the ( ๐ exists )} a ๐ partition of unity {๐ open covering ๐ต ๐๐ , ๐ ๐โโค2 of โ which satis๏ฌes
โฃโ ๐ ห๐ โฃ โค
176 , ๐
๐ โ โค2 .
Now, for ๐ โ ๐ผ, we de๏ฌne on ๐: ๐๐ = 0 if ๐ โโ ๐ (โค2 ), and โ ๐๐ = ๐ ห๐ if ๐ โ ๐ (โค2 ).
(6.18)
(6.19)
๐โ๐ โ1 (๐)
As the sets ๐ โ1 (๐), ๐ โ ๐ผ, are pairwise disjunct and ๐ผ is the union of these sets, it is clear that โ ๐๐ โก 1 on ๐, and from (6.17) we see that supp ๐๐ โ ๐๐ . Hence, {๐๐ }๐โ๐ผ is a ๐ โ partition of unity subordinate to ๐ฐ. By Lemma 6.1, the partition of unity {๐ ห๐ }๐โโค2 is of order โค 21. Since the sets ๐ โ1 (๐), ๐ โ ๐ผ, are pairwise disjunct, this implies, by (6.19), that also the partition {๐๐ }๐โ๐ผ is of order โค 21. The fact that the partition of unity {๐ ห๐ }๐โโค2 is of order โค 21, moreover implies that, in the sum (6.19), locally, not more than 21 terms are di๏ฌerent from zero. Together with (6.18) this yields the required estimate: โฃโ๐๐ โฃ โค 21
176 212 < . ๐ ๐
โก
An Estimate for the Splitting
477
7. Continuous functions with continuous Cauchy-Riemann derivative. The additive case In this section, ๐ธ is a Banach space, and ๐ is a bounded subset of โ such that ๐ โ int ๐. 7.1. We denote by โฌ๐ ๐ธ (๐) the Banach space of ๐ธ-valued bounded continuous functions on ๐ endowed with the sup-norm โฅโ
โฅ๐ de๏ฌned by (2.2), and by โฌ๐โ๐ธ (๐) we denote the subspace of all ๐ โ โฌ๐ ๐ธ (๐) such that also โ๐ โ โฌ๐ ๐ธ (๐) (the domain of de๏ฌnition of the di๏ฌerential operator โ as an operator in โฌ๐ ๐ธ (๐)). Notice that โฌ๐โ๐ธ (๐) becomes a Banach space if we introduce the norm โฅ โ
โฅโ de๏ฌned by โฅ๐ โฅโ = โฅ๐ โฅ๐ + โฅโ๐ โฅ๐ .
(7.1)
Below we use the following simple fact (see, e.g., Proposition 2.1.2 in [GL]): If ๐ โ โฌ๐โ๐ธ (๐) and ๐ : ๐ โ โ is a bounded continuous function such that โ๐ is also continuous and bounded on ๐, then ๐๐ belongs to โฌ๐โ๐ธ (๐) and โ(๐๐ ) = (โ๐)๐ + ๐โ๐.
(7.2)
Now let ๐ฐ = {๐๐ }๐โ๐ผ be a covering of ๐ by relatively open subsets of ๐. Then we use the following notations:11 ( ) โ ๐ถ 0 ๐ฐ, โฌ๐โ๐ธ is the space of all families {๐๐ }๐โ๐ผ of functions ๐๐ โ โฌ๐โ๐ธ (๐๐ ). ( ) โ ๐ถ 1 ๐ฐ, โฌ๐โ๐ธ is the space of all families {๐๐๐ }๐,๐โ๐ผ of functions ๐๐๐ โ โฌ๐โ๐ธ (๐๐ โฉ ๐๐ ). ( ) ( ) โ ๐ 1 ๐ฐ, โฌ๐โ๐ธ is the subspace of all ๐ โ ๐ถ 1 ๐ฐ, โฌ๐โ๐ธ satisfying the following (additive) cocycle condition: ๐๐๐ + ๐๐๐ = ๐๐๐
on ๐๐ โฉ ๐๐ โฉ ๐๐ ,
๐, ๐, ๐ โ ๐ผ.
(7.3)
The( elements ) of this subspace will be called additive 1-cocycles. โ ๐ถ 2 ๐ฐ, โฌ๐โ๐ธ is the space of all families {๐๐๐๐ }๐,๐,๐โ๐ผ of functions ๐๐๐๐ โ โฌ๐โ๐ธ (๐๐ โฉ ๐๐ โฉ ๐๐ ). ( ) ( ) โ ๐ 2 ๐ฐ, โฌ๐โ๐ธ is the subspace of all ๐ โ ๐ถ 2 ๐ฐ, โฌ๐โ๐ธ (๐) satisfying the following condition (also called cocycle condition): โ๐๐๐๐ + ๐๐๐๐ โ ๐๐๐๐ + ๐๐๐๐ = 0
on ๐๐ โฉ ๐๐ โฉ ๐๐ โฉ ๐๐ ,
๐, ๐, ๐, ๐ โ ๐ผ.
(7.4)
The elements of this subspace will be called additive 2-cocycles. ห are notations from the theory of Cech cohomology with coe๏ฌcients in sheaves, but, in this paper, we will not use this theory, except for some very simple facts, which will be explained. Note that the map ๐ โ โฌ๐ ๐ธ (๐)(๐ ) applied to the relatively open subsets of ๐ is only a presheaf, 11 These
but not a sheaf.
โ
478
J. Leiterer
( ) โ For ๐ โ ๐ถ ๐ ๐ฐ, โฌ๐โ๐ธ (๐) , ๐ = 0, 1, 2, we de๏ฌne โฅ๐ โฅ = sup โฅ๐๐ โฅ๐๐ , โฅโ๐ โฅ = sup โฅโ๐๐ โฅ๐๐ ๐โ๐ผ
๐โ๐ผ
if ๐ = 0,
โฅ๐ โฅ = sup โฅ๐๐๐ โฅ๐๐ โฉ๐๐ , โฅโ๐ โฅ = sup โฅโ๐๐๐ โฅ๐๐ โฉ๐๐ ๐,๐โ๐ผ
๐,๐โ๐ผ
if ๐ = 1,
โฅ๐ โฅ = sup โฅ๐๐๐๐ โฅ๐๐ โฉ๐๐ โฉ๐๐ , โฅโ๐ โฅ = sup โฅโ๐๐๐๐ โฅ๐๐ โฉ๐๐ โฉ๐๐ ๐,๐,๐โ๐ผ
๐,๐,๐โ๐ผ
(7.5)
if ๐ = 2,
and
โฅ๐ โฅโ = โฅ๐ โฅ + โฅโ๐ โฅ. (7.6) Note that can be in๏ฌnite if ๐ผ is in๏ฌnite. The space of all ( these โnumbersโ ) ๐ โ ๐ถ ๐ ๐ฐ, โฌ๐โ๐ธ (๐) with โฅ๐ โฅโ < โ is a Banach space. โ We de๏ฌne linear operators ( ) ( ) ๐ฟ : ๐ถ ๐ ๐ฐ, โฌ๐โ๐ธ (๐) โ ๐ถ ๐+1 ๐ฐ, โฌ๐โ๐ธ (๐) , ๐ = 0, 1, ( ) setting, for ๐ โ ๐ถ 0 ๐ฐ, โฌ๐โ๐ธ (๐) , (๐ฟ๐ )๐๐ = ๐๐ โ ๐๐ ( ) and, for ๐ โ ๐ถ ๐ฐ, โฌ๐โ๐ธ (๐) ,
on ๐๐ โฉ ๐๐ ,
๐, ๐ โ ๐ผ,
(7.7)
1
(๐ฟ๐ )๐๐๐ = โ๐๐๐ + ๐๐๐ โ ๐๐๐ on ๐๐ โฉ ๐๐ โฉ ๐๐ , ๐, ๐, ๐ โ ๐ผ. (7.8) ( ) ๐ ๐ธ โ The element ๐ โ ๐ถ ๐ฐ, โฌ๐โ (๐) de๏ฌned by ๐๐ โก 0 if ๐ = 1, ๐๐๐ โก 0 if ๐ = 1, and ๐๐๐๐ โก 0 if ๐ = 2 will be denoted by 0. It is easy to check that ( ) ( ) ๐ฟ๐ถ ๐ ๐ฐ, โฌ๐โ๐ธ (๐) โ ๐ ๐+1 ๐ฐ, โฌ๐โ๐ธ (๐) , ๐ = 0, 1. (7.9) ( ) Note also that the de๏ฌnition of ๐ฟ is chosen so that an element ๐ โ ๐ถ 1 ๐ฐ, โฌ๐โ๐ธ (๐) is an additive 1-cocycle if and only if ๐ฟ๐ = 0.12 Moreover, ( it is well )known from ห the general theory of Cech cohomology that each ๐ โ ๐ ๐ ๐ฐ, โฌ๐โ๐ธ (๐) , ๐ = 1, 2, is of the form ๐ = ๐ฟ๐ข, ( ) ๐โ1 ๐ธ where ๐ข โ ๐ถ ๐ฐ, โฌ๐โ (๐) . We need a version with estimates of the latter fact, which is stated by the following lemma (the proof of this lemma is a modi๏ฌcation of the corresponding arguments from the general theory). 7.2. Lemma. Let ๐ > 0, and let ๐ฐ be an ๐-separated covering ( ) of ๐ by relatively open sets (De๏ฌnition 2.2). Then, for each ๐ โ ๐ ๐ ๐ฐ, โฌ๐โ๐ธ (๐) , ๐ = 1, 2, such that ( ) โฅ๐ โฅโ < โ, there exists ๐ข โ ๐ถ ๐โ1 ๐ฐ, โฌ๐โ๐ธ (๐) such that ๐ฟ๐ข = ๐,
(7.10)
โฅ๐ขโฅ โค โฅ๐ โฅ
(7.11)
( ) ห the general theory of Cech cohomology, such an operator ๐ฟ is de๏ฌned also on ๐ถ 2 ๐ฐ , โฌ๐ ๐ธ (๐) , โ ( ) and its kernel is ๐ 2 ๐ฐ , โฌ๐ ๐ธ (๐) . Here we do not need this. 12 In
โ
An Estimate for the Splitting and โฅโ๐ขโฅ โค โฅโ๐ โฅ +
479
217 โฅ๐ โฅ. ๐
(7.12)
( ) Proof. Let ๐ฐ = {๐๐ }๐โ๐ผ , and let ๐ = {๐๐๐ }๐,๐โ๐ผ โ ๐ ๐ ๐ฐ, โฌ๐โ๐ธ (๐) with โฅ๐ โฅ + โฅโ๐ โฅ < โ be given. Then, by Lemma 6.4, there exists a ๐ โ partition of unity {๐๐ }๐โ๐ผ subordinated to ๐ฐ, which is of order โค 21, such that $ $ 212 โฃ(โ๐๐ )(๐)$$ โค , ๐ โ โ, ๐ โ ๐ผ. (7.13) ๐ ( ) First let ๐ = 1. Then we de๏ฌne a ๐ข = {๐ข๐}๐โ๐ผ โ ๐ถ 0 ๐ฐ, โฌ๐โ๐ธ (๐) by โ ๐ข๐ = โ ๐๐ ๐๐๐ . (7.14) ๐โ๐ผ
โ
As ๐ is an additive 1-cocycle and ๐๐ โก 1, then ) โ โ ( ๐ข๐ โ ๐ข๐ = ๐๐ โ ๐๐๐ + ๐๐๐ = ๐๐ ๐๐๐ = ๐๐๐ , ๐โ๐ผ
๐โ๐ผ
i.e., we โ have relation (7.10). Estimate (7.11) is clear, since all ๐๐ are non-negative and ๐๐ โก 1. Further, by (7.14) and (7.2), ) โ( โ๐ข๐ = โ ๐๐ โ๐๐๐ + (โ๐๐ )๐๐๐ , ๐ โ ๐ผ. ๐โ๐ผ
Hence โฅโ๐ขโฅ โค โฅโ๐ โฅ + โฅ๐ โฅ
sup
1โค๐โค๐ , ๐โโ
โ
โฃโ๐๐ (๐)โฃ.
๐โ๐ผ
Since {๐๐ } is of order โค 21, now estimate (7.12) ( follows from ) (7.13). Now let ๐ = 2. Then we de๏ฌne a ๐ข โ ๐ถ 1 ๐ฐ, โฌ๐โ๐ธ (๐) setting โ ๐ข๐๐ = โ ๐๐ ๐๐๐๐ on ๐๐ โฉ ๐๐ , ๐, ๐ โ ๐ผ.
(7.15)
๐โ๐ผ
Then, for all ๐, ๐, ๐ โ ๐ผ, (๐ฟ๐ข)๐๐๐ = โ๐ข๐๐ + ๐ข๐๐ โ ๐ข๐๐ =
โ
( ) ๐๐ ๐๐๐๐ โ ๐๐๐๐ + ๐๐๐๐ on ๐๐ โฉ ๐๐ โฉ ๐๐ .
๐โ๐ผ
Moreover, as ๐ is an additive 2-cocycle, for all ๐, ๐, ๐, ๐ โ ๐ผ, 0 = (๐ฟ๐ )๐๐๐๐ = โ๐๐๐๐ + ๐๐๐๐ โ ๐๐๐๐ + ๐๐๐๐ , i.e., ๐๐๐๐ โ ๐๐๐๐ + ๐๐๐๐ = ๐๐๐๐ Hence (๐ฟ๐ข)๐๐๐ =
โ ๐โ๐ผ
๐๐ ๐๐๐๐ = ๐๐๐๐
on ๐๐ โฉ ๐๐ โฉ ๐๐ โฉ ๐๐ . on ๐๐ โฉ ๐๐ โฉ ๐๐ ,
๐, ๐, ๐ โ ๐ผ,
480
J. Leiterer
i.e., โ we have (7.10). Estimate (7.11) is clear, since all ๐๐ are non-negative and ๐๐ โก 1. Further, by (7.15) and (7.2), ) โ( โ๐ข๐๐ = โ ๐๐ โ๐๐๐๐ + (โ๐๐ )๐๐๐๐ , ๐, ๐ โ ๐ผ. ๐โ๐ผ
As all ๐๐ are non-negative and
โ
๐๐ โก 1, this implies that โ$ $ $โ๐๐ (๐)$. โฅโ๐ขโฅ โค โฅโ๐ โฅ + โฅ๐ โฅ sup 1โค๐โค๐ , ๐โโ
๐โ๐ผ
Since {๐๐ } is of order โค 21, now estimate (7.12) again follows from (7.13).
โก
8. Continuous functions with continuous Cauchy-Riemann derivative. The multiplicative case We will use the following proposition. 8.1. Proposition. Let ๐ be a bounded subset of โ such that ๐ โ int ๐, let ๐ด be a Banach algebra with unit, and let ๐บ๐ด is the group of invertible elements of ๐ด. Then (i) If ๐, ๐ : ๐ โ ๐ด are continuous such that also โ๐ and โ๐ are continuous on ๐, then โ(๐ ๐) is continuous on ๐ and โ(๐ ๐) = (โ๐ )๐ + ๐ โ๐. (ii) If ๐ : ๐ โ ๐บ๐ด is continuous such that also โ๐ is continuous on ๐, then ๐ โ1 is continuous on ๐ and โ๐ โ1 = โ๐ โ1 (โ๐ )๐ โ1 . Proof. This is clear when ๐ is open and the functions ๐ and ๐ are of class ๐ โ . The general case follows from this and the fact that, for each continuous function ๐ : int ๐ โ ๐ด such that โ๐ is also continuous on int ๐, there exists a sequence (๐๐ ) of ๐ โ functions ๐๐ : int ๐ โ ๐ด such that, uniformly on the compact subsets of int ๐, both lim ๐๐ = ๐ and lim โ๐๐ = ๐ (see, e.g., Lemma 2.1.3 in [GL]) โก 8.2. Let ๐ be a bounded subset of โ such that ๐ โ int ๐, let ๐ด be a Banach algebra with unit, 1, and let ๐บ๐ด be the group of invertible elements of ๐ด. Since ๐ด is a Banach algebra, the Banach space โฌ๐ ๐ด (๐) introduced in Section 7.1 now is a Banach algebra. Moreover we see from Proposition 8.1 (i) that the subspace โฌ๐โ๐ด (๐) (also introduced in Section 7.1) is a subalgebra of โฌ๐ ๐ด (๐), which becomes a Banach algebra if we introduce the norm (7.1). We denote by โฌ๐โ๐บ๐ด (๐) the set of all ๐ โ โฌ๐โ๐ด (๐) such that ๐ (๐) โ ๐บ๐ฟ(๐, โ) for all ๐ โ ๐ and sup โฅ๐ โ1 (๐)โฅ < โ. (8.1) ๐โ๐
An Estimate for the Splitting
481
It follows from Proposition 8.1 (ii) that, for each ๐ โ โฌ๐โ๐บ๐ด (๐), the function ๐ โ1 again belongs to โฌ๐โ๐บ๐ด(๐), i.e., โฌ๐โ๐บ๐ด (๐) is the group of invertible elements of the algebra โฌ๐โ๐บ๐ด(๐). Notice that the algebra โฌ๐ช๐ด (๐) (Section 4.1) is a subalgebra of โฌ๐โ๐ด (๐), and the group โฌ๐ช๐บ๐ด (๐) is a subgroup of โฌ๐โ๐บ๐ด (๐). Now let ๐ฐ = {๐๐ }๐โ๐ผ be a covering of ๐ by relatively open subsets of ๐. Then, additional to the notations introduced in Section 7.1, here we need also the following notations: ( ) ( ) โ ๐ถ 0 ๐ฐ, โฌ๐โ๐บ๐ด is the group of all ๐ โ ๐ถ 0 ๐ฐ, โฌ๐โ๐บ๐ด such that ๐๐ โ โฌ๐โ๐บ๐ด (๐๐ ) for(all ๐ โ ๐ผ. ) ( ) โ ๐ถ 1 ๐ฐ, โฌ๐โ๐บ๐ด is the set13 of all ๐ โ ๐ถ 1 ๐ฐ, โฌ๐โ๐บ๐ด such that ๐๐๐ โ โฌ๐โ๐บ๐ด(๐๐ โฉ ๐๐ )( for all ๐,)๐ โ ๐ผ. ( ) โ ๐ 1 ๐ฐ, โฌ๐โ๐บ๐ด is the subset of all ๐ โ ๐ถ 1 ๐ฐ, โฌ๐โ๐บ๐ด satisfying the multiplicative cocycle condition ๐๐๐ ๐๐๐ = ๐๐๐
on ๐๐ โฉ ๐๐ โฉ ๐๐ ,
๐, ๐, ๐ โ ๐ผ.
(8.2)
The elements cocycles. ( of this )subset will be(called multiplicative ) โ For ๐ โ ๐ถ 0 ๐ฐ, โฌ๐โ๐บ๐ด and ๐ โ ๐ถ 1 ๐ฐ, โฌ๐โ๐บ๐ด , we de๏ฌne an element ๐ โ ๐ โ ( ) ๐ถ 1 ๐ฐ, โฌ๐โ๐บ๐ด by (๐ โ ๐ )๐๐ = ๐๐โ1 ๐๐๐ ๐๐
on ๐๐ โฉ ๐๐ ,
๐, ๐ โ ๐ผ.
(8.3)
Note that( ๐ โ 1 is )always a multiplicative ( ) cocycle. Notice also that, for all ๐, โ โ ๐ถ 0 ๐ฐ, โฌ๐โ๐บ๐ด and ๐ โ ๐ถ 1 ๐ฐ, โฌ๐โ๐บ๐ด , (๐โ) โ ๐ = โ โ (๐ โ ๐ ).
(8.4)
The aim of this section is to prove the following theorem. 8.3. Theorem. Let ๐ โ โ such that ๐ โ int ๐, and let ๐ฐ = {๐๐ }๐โ๐ผ be an ๐-separated covering of ๐ )by relatively open sets (De๏ฌnition 2.2). ( Let ๐ โ ๐ 1 ๐ฐ, โฌ๐โ๐บ๐ด such that 1 โฅ๐ โ 1โฅ โค . 64 ( ) Then there exists ๐ โ ๐ถ 0 ๐ฐ, โฌ๐โ๐บ๐ด such that
and
๐ โ ๐ = 1,
(8.6)
โฅ๐ โ 1โฅ โค 2โฅ๐ โ 1โฅ,
(8.7)
220 โฅ๐ โ 1โฅ + 2โฅโ๐ โฅ. ๐ We ๏ฌrst prove the following lemma. โฅโ๐โฅ โค
13 This
(8.5)
is also a group, but here we will not use the group structure.
(8.8)
482
J. Leiterer
( ) 8.4. Lemma. Under the hypotheses of Theorem 8.3 there exists ๐ โ ๐ถ 0 ๐ฐ, ๐โ๐บ๐ด such that 1 โฅ๐ โ ๐ โ 1โฅ โค โฅ๐ โ 1โฅ, (8.9) 8 22 2 โฅโ(๐ โ ๐ )โฅ โค โฅ๐ โ 1โฅ2 + 16โฅโ๐ โฅโฅ๐ โ 1โฅ, (8.10) ๐ 65 โฅ๐ โ 1โฅ โค โฅ๐ โ 1โฅ, (8.11) 64 218 33 โฅโ๐โฅ โค โฅ๐ โ 1โฅ + โฅโ๐ โฅ. (8.12) ๐ 32 Proof. We set ๐ = ๐ โ 1. In general, ๐ is not an additive 1-cocycle, i.e., ๐ฟ๐ โ= 0 (Section 7.1). But, as ๐๐๐ = ๐๐๐ ๐๐๐ , we have 1 + ๐๐๐ = (1 + ๐๐๐ )(1 + ๐๐๐ ) = 1 + ๐๐๐ + ๐๐๐ + ๐๐๐ ๐๐๐ and therefore (๐ฟ๐)๐๐๐ = โ๐๐๐ + ๐๐๐ โ ๐๐๐ = ๐๐๐ ๐๐๐ Hence
on ๐๐ โฉ ๐๐ โฉ ๐๐ ,
๐, ๐, ๐ โ ๐ผ.
โฅ๐ฟ๐โฅ โค โฅ๐โฅ2
(8.13)
โฅโ๐ฟ๐โฅ โค 2โฅ๐โฅโฅโ๐โฅ.
(8.14)
and As ๐ฟ๐ is an additive ( )2-cocycle (see (7.9)), it follows from Lemma 7.2 that there exists ๐ข โ ๐ถ 1 ๐ฐ, โฌ๐โ๐ด such that ๐ฟ๐ข = ๐ฟ๐,
(8.15)
โฅ๐ขโฅ โค โฅ๐ฟ๐โฅ, and
217 โฅ๐ฟ๐โฅ. ๐ By (8.13) and (8.14), the last two estimates further imply โฅโ๐ขโฅ โค โฅโ๐ฟ๐โฅ +
โฅ๐ขโฅ โค โฅ๐โฅ2
(8.16)
and
217 โฅ๐โฅ2 . (8.17) ๐ By (8.15), ( ๐ โ ๐ข) is an additive 1-cocycle. Therefore, again from Lemma 7.2 we get ๐ฃ โ ๐ถ 0 ๐ฐ, โฌ๐โ๐ด such that ๐ฟ๐ฃ = ๐ โ ๐ข, (8.18) โฅโ๐ขโฅ โค 2โฅ๐โฅโฅโ๐โฅ +
โฅ๐ฃโฅ โค โฅ๐ โ ๐ขโฅ, and โฅโ๐ฃโฅ โค โฅโ๐ โ โ๐ขโฅ +
217 โฅ๐ โ ๐ขโฅ. ๐
(8.19) (8.20)
An Estimate for the Splitting
483
By (8.19) and (8.16), โฅ๐ฃโฅ โค โฅ๐โฅ + โฅ๐โฅ2 . By (8.5) this implies that 65 โฅ๐ฃโฅ โค โฅ๐โฅ (8.21) 64 and 65 โฅ๐ฃโฅ < 12 . (8.22) 2 In particular โฅ๐ฃโฅ < 1, which implies (by the arguments given)at the beginning of ( Section 8.2) that ๐ := 1 + ๐ฃ belongs to the group ๐ถ 0 ๐ฐ, โฌ๐โ๐บ๐ด . We will show that this ๐ has the required properties. Estimate (8.11) is clear by (8.21). To prove the remaining properties we set ๐ = ๐ โ1 โ 1 + ๐ฃ. (8.23) ) ( ) โ1 ๐ด 0 ๐ด As ๐ and ๐ฃ belong to ๐ถ ๐ฐ, โฌ๐โ , then also ๐ โ ๐ถ ๐ฐ, โฌ๐โ . Moreover, since โฅ๐ฃโฅ < 1, for ๐ โ1 we have the representation โ โ ๐ โ1 = (โ๐ฃ)๐ , ( 0
๐=0
where the convergence is absolute with respect to โฅ โ
โฅ. (Below we shall prove that the convergence is even absolute with respect to the Banach space norm โฅ โ
โฅโ de๏ฌned by (7.6).) Therefore, with the same kind of convergence, โ โ ๐= (โ๐ฃ)๐ . (8.24) ๐=2
More precisely, we see from (8.21) that โ โ โ โ 652 โฅ๐โฅ โค โฅ๐ฃโฅ2 โฅ๐ฃโฅ๐ โค 12 โฅ๐โฅ2 โฅ๐ฃโฅ๐. 2 ๐=0 ๐=0 Since, by (8.22), โ โ
โฅ๐ฃโฅ๐ โค
๐=0
)๐ โ ( โ 65 ๐=0
this implies โฅ๐โฅ โค From (8.23) we see
212
=
1 212 , 65 = 12 2 โ 65 1 โ 212
652 โฅ๐โฅ2 โค 2โฅ๐โฅ2 . โ 65
212
(8.25)
(๐ โ ๐ )๐๐ = (1 โ ๐ฃ๐ + ๐๐ )(1 + ๐๐๐ )(1 + ๐ฃ๐ ) = 1 โ ๐ฃ๐ + ๐๐๐ + ๐ฃ๐ โ ๐ฃ๐ ๐๐๐ + ๐๐๐ ๐ฃ๐ โ ๐ฃ๐ ๐ฃ๐ โ ๐ฃ๐ ๐๐๐ ๐ฃ๐ + ๐๐ (1 + ๐๐๐ )(1 + ๐ฃ๐ ) Since, by (8.18), โ๐ฃ๐ + ๐๐๐ + ๐ฃ๐ = ๐๐๐ โ (๐ฟ๐ฃ)๐๐ = ๐ข๐๐ , this implies that (๐ โ ๐ )๐๐ โ 1 = ๐ข๐๐ โ ๐ฃ๐ ๐๐๐ + ๐๐๐ ๐ฃ๐ โ ๐ฃ๐ ๐ฃ๐ โ ๐ฃ๐ ๐๐๐ ๐ฃ๐ + ๐๐ (1 + ๐๐๐ )(1 + ๐ฃ๐ ). (8.26) Hence โฅ๐ โ ๐ โ 1โฅ โค โฅ๐ขโฅ + 2โฅ๐ฃโฅโฅ๐โฅ + โฅ๐ฃโฅ2 + โฅ๐ฃโฅ2 โฅ๐โฅ + โฅ๐โฅ(1 + โฅ๐โฅ)(1 + โฅ๐ฃโฅ).
484
J. Leiterer
In view of (8.16), (8.21), (8.5), (8.25), and (8.22), this implies ( ) 65 652 652 65 212 + 65 โฅ๐ โ ๐ โ 1โฅ โค 1 + + 12 + 18 + 2 โ
โ
โฅ๐โฅ2 โค 8โฅ๐โฅ2. 32 2 2 64 212
(8.27)
Taking again into account that โฅ๐โฅ โค 1/64, this implies (8.9). From (8.20), (8.16), and (8.17) we see that ) 217 ( โฅโ๐ฃโฅ โค โฅโ๐โฅ + โฅโ๐ขโฅ + โฅ๐โฅ + โฅ๐ขโฅ ๐ ) 217 217 ( โค โฅโ๐โฅ + 2โฅ๐โฅโฅโ๐โฅ + โฅ๐โฅ2 + โฅ๐โฅ + โฅ๐โฅ2 ๐ ๐ ) 217 ( = โฅโ๐โฅ + 2โฅ๐โฅโฅโ๐โฅ + โฅ๐โฅ + 2โฅ๐โฅ2 . ๐ As โฅ๐โฅ โค 1/64, this further implies that โฅโ๐ฃโฅ โค
33 218 โฅโ๐โฅ + โฅ๐โฅ. 32 ๐
(8.28)
Since โ๐ = โ๐ฃ, this proves (8.12). Next we estimate โฅโ๐โฅ. From the product rule (Proposition 8.1 (i)) it follows that โฅโ(โ๐ฃ)๐ โฅ โค ๐โฅโ๐ฃโฅโฅ๐ฃโฅ๐โ1 if ๐ โฅ 1. By (8.21) this implies that โฅโ(โ๐ฃ)๐ โฅ โค ๐โฅโ๐ฃโฅ
65 โฅ๐โฅโฅ๐ฃโฅ๐โ2 64
if ๐ โฅ 2,
and further, by (8.22),
( )๐โ2 65 65 โฅโ(โ๐ฃ) โฅ โค ๐โฅโ๐ฃโฅ โฅ๐โฅ 12 64 2 ๐
if ๐ โฅ 2.
Moreover,
( )๐โ2 ( )โ )๐โ2 )๐โ2 โ โ ( โ ( โ โ 65 ๐ 65 65 ๐ 12 โค sup ๐โ2 =2 2 211 211 ๐โฅ2 2 ๐=2 ๐=2 ๐=2 )๐โ2 โ ( โ 1 32 <2 = . 16 15 ๐=2
Together this implies that โ โ ๐=2
โฅโ(โ๐ฃ)๐ โฅ โค
65 32 7 โ
โฅโ๐ฃโฅโฅ๐โฅ โค โฅโ๐ฃโฅโฅ๐โฅ. 64 15 3
Hence, by (8.24), โฅโ๐โฅ โค
7 โฅโ๐ฃโฅโฅ๐โฅ, 3
An Estimate for the Splitting
485
and further, by (8.28), ( ) 7 33 218 220 5 โฅโ๐โฅ โค โฅโ๐โฅ + โฅ๐โฅ โฅ๐โฅ โค โฅ๐โฅ2 + โฅโ๐โฅโฅ๐โฅ. 3 32 ๐ ๐ 2
(8.29)
From (8.26) and the product rule (Proposition 8.1 (i)) we see โฅโ(๐ โ ๐ )โฅ โค โฅโ๐ขโฅ + 2โฅโ๐ฃโฅโฅ๐โฅ + 2โฅโ๐โฅโฅ๐ฃโฅ + 2โฅโ๐ฃโฅโฅ๐ฃโฅ + 2โฅ๐ฃโฅโฅ๐โฅโฅโ๐ฃโฅ + โฅ๐ฃโฅ2 โฅโ๐โฅ + โฅโ๐โฅ(1 + โฅ๐โฅ)(1 + โฅ๐ฃโฅ) ( ) + โฅ๐โฅ โฅโ๐โฅ + โฅโ๐ฃโฅ + โฅ๐ฃโฅโฅโ๐โฅ + โฅ๐โฅโฅโ๐ฃโฅ . Taking into account that โฅ๐ฃโฅ โค
65 โฅ๐โฅ, 64
โฅ๐ฃโฅ <
65 , 212
โฅ๐โฅ <
1 , 64
and โฅ๐โฅ โค 2โฅ๐โฅ2 โค
โฅ๐โฅ 32
(see (8.21), (8.22), (8.5), and (8.25)), this implies that 65 65 65 โฅโ๐โฅโฅ๐โฅ + โฅโ๐ฃโฅโฅ๐โฅ + 11 โฅ๐โฅโฅโ๐ฃโฅ 32 32 2 ( ) 652 โฅ๐โฅ 65 1 + 18 โฅ๐โฅโฅโ๐โฅ + 2โฅโ๐โฅ + โฅโ๐โฅ + โฅโ๐ฃโฅ + 12 โฅโ๐โฅ + โฅโ๐ฃโฅ 2 32 2 64 ( ) 2 65 65 1 65 โค โฅโ๐ขโฅ + + 18 + + โฅโ๐โฅโฅ๐โฅ 32 2 32 217 ( ) 65 65 1 1 + 2+ + 11 + + 11 โฅโ๐ฃโฅโฅ๐โฅ + 2โฅโ๐โฅ 32 2 32 2
โฅโ(๐ โ ๐ )โฅ โค โฅโ๐ขโฅ + 2โฅโ๐ฃโฅโฅ๐โฅ +
โค โฅโ๐ขโฅ + 3โฅโ๐โฅโฅ๐โฅ + 5โฅโ๐ฃโฅโฅ๐โฅ + 2โฅโ๐โฅ. Together with (8.17), (8.28), and (8.29) this proves (8.10): 217 โฅโ(๐ โ ๐ )โฅ โค 2โฅ๐โฅโฅโ๐โฅ + โฅ๐โฅ2 + 3โฅโ๐โฅโฅ๐โฅ ๐ ( ) ( 20 ) 33 218 2 5 2 +5 โฅโ๐โฅ + โฅ๐โฅ โฅ๐โฅ + 2 โฅ๐โฅ + โฅโ๐โฅโฅ๐โฅ 32 ๐ ๐ 2 ( 17 ) 18 20 2 2 2 โค +5 +2 โฅ๐โฅ2 ๐ ๐ ๐ ( ) 33 222 + 2 + 3 + 5 + 5 โฅโ๐โฅโฅ๐โฅ โค โฅ๐โฅ2 + 16โฅโ๐โฅโฅ๐โฅ. 32 ๐
โก
8.5. Proof of Theorem 8.3. We write for short ๐ถ=
222 . ๐
(8.30)
486
J. Leiterer
( ) Then, from Lemma 8.4 we get sequences (๐๐ )๐โโโ of elements ๐๐ โ ๐ถ 0 ๐ฐ, โฌ๐โ๐บ๐ด ( ) and (๐๐ )๐โโ of cocycles ๐๐ โ ๐ 1 ๐ฐ, โฌ๐โ๐บ๐ด such that ๐0 = ๐ and, for all ๐ โ โ, ๐๐+1 = ๐๐+1 โ ๐๐ , (8.31) 1 โฅ๐๐+1 โ 1โฅ โค โฅ๐๐ โ 1โฅ, (8.32) 8 โฅโ๐๐+1 โฅ โค ๐ถโฅ๐๐ โ 1โฅ2 + 16โฅโ๐๐ โฅโฅ๐๐ โ 1โฅ, (8.33) 65 โฅ๐๐+1 โ 1โฅ โค โฅ๐ โ 1โฅ, (8.34) 64 ๐ ๐ถ 33 โฅโ๐๐+1 โฅ โค โฅ๐๐ โ 1โฅ + โฅโ๐๐ โฅ. (8.35) 16 32 Since ๐0 = ๐ , it follows from (8.32) that 1 1 โฅ๐๐ โ 1โฅ โค ๐ โฅ๐ โ 1โฅ = 3๐ โฅ๐ โ 1โฅ, ๐ โ โ. (8.36) 8 2 Together with (8.34) this yields 65 1 โฅ๐๐ โ 1โฅ โค โฅ๐ โ 1โฅ, ๐ โ โโ . (8.37) 64 23๐โ3 Since โฅ๐ โ 1โฅ โค 1/64, this in particular implies that, for all ๐ โ โโ , โฅ๐๐ โ 1โฅ < 17 and, therefore, โ โ โ โ ( ) โฅ๐๐ โ 1โฅ๐ 1 7 log 1 + โฅ๐๐ โ 1โฅ = โ (โ1)๐ โค โฅ๐๐ โ 1โฅ = โฅ๐๐ โ 1โฅ. ๐ ๐ 7 6 ๐=1 ๐=0 Together with (8.37) this further implies that ( ) 7 1 log 1 + โฅ๐๐ โ 1โฅ โค โฅ๐ โ 1โฅ, ๐ โ โโ . 6 23๐โ3 Hence, for each ๐ โ โ, โ โ โ ( ) 7 โ 1 4 1 log 1 + โฅ๐๐ โ 1โฅ โค โฅ๐ โ 1โฅ = โฅ๐ โ 1โฅ. 6 23๐โ3 3 23๐ ๐=๐ +1
(8.39)
๐=๐ +1
As โฅ๐ โ 1โฅ โค 1/64, this in particular implies that โ ( ) โ 1 log 1 + โฅ๐๐ โ 1โฅ โค . 48 ๐=1 Moreover, for ๐ โ โ and ๐ โ โโ , we have ๐โ +๐ ( ๐ ) โ โ 1 + โฅ๐๐ โ 1โฅ = 1 + ๐
=1 ๐ +1โค๐1 <โ
โ
โ
<๐๐
โค๐ +๐
๐=๐ +1
(8.38)
(8.40)
โฅ๐๐1 โ 1โฅ โ
โ
โ
โฅ๐๐๐
โ 1โฅ.
and, similarly, ๐โ +๐ ๐=๐ +1
๐ โ ( ) 1 + (๐๐ โ 1) = 1 +
โ
๐
=1 ๐ +1โค๐1 <โ
โ
โ
<๐๐
โค๐ +๐
(
) ( ) ๐๐1 โ 1 . . . ๐๐๐
โ 1 .
An Estimate for the Splitting
487
Hence, for ๐ โ โ and ๐ โ โโ , 1 ๐ +๐ 1 1 ๐ +๐ 1 1 โ 1 1 โ ( 1 ) 1 1 1 1 ๐ โ 1 = 1 + (๐ โ 1) โ 1 ๐ ๐ 1 1 1 1 ๐=๐ +1
๐=๐ +1
1 ๐ 1โ =1 1
โ
๐
=1 ๐ +1โค๐1 <โ
โ
โ
<๐๐
โค๐ +๐
โค
๐ โ
โ
๐
=1 ๐ +1โค๐1 <โ
โ
โ
<๐๐
โค๐ +๐
=
๐โ +๐
1 1 (๐๐1 โ 1) . . . (๐๐๐
โ 1)1 1
โฅ๐๐1 โ 1โฅ . . . โฅ๐๐๐
โ 1โฅ
๐โ +๐ ( ) ( ) 1 + โฅ๐๐ โ 1โฅ โ 1 = exp log 1 + โฅ๐๐ โ 1โฅ โ 1.
๐=๐ +1
๐=๐+1
Taking into account (8.40) and the fact that ๐๐ฅ โ 1 โค ๐1/48 ๐ฅ โค
3 ๐ฅ 2
for 0 โค ๐ฅ โค
this implies that 1 ๐ +๐ 1 +๐ ( ) 1 โ 1 3 ๐โ 1 1โค ๐ โ 1 log 1 + โฅ๐ โ 1โฅ , ๐ ๐ 1 1 2 ๐=๐ +1
1 , 48
๐ โ โ, ๐ โ โโ
๐=๐ +1
and, therefore, by (8.39), 1 ๐โ 1 1 +๐ 1 2 1 ๐ ๐ โ 11 1 1 โค 23๐ โฅ๐ โ 1โฅ,
๐ โ โ, ๐ โ โโ .
(8.41)
๐=๐ +1
In particular, as โฅ๐ โ 1โฅ โค 2โ6 , 1 ๐+๐ 1 1 โ 1 1 1 1 ๐๐ โ 11 โค 5+3๐ , 1 1 1 2
๐ โ โ, ๐ โ โโ ,
(8.42)
๐=๐ +1
and, hence,
1 ๐โ 1 1 +๐ 1 1 ๐๐ 1 1 1 < 2,
๐ โ โ, ๐ โ โโ .
(8.43)
๐=๐ +1
From the last two inequalities we see that, for all ๐, ๐ โ โโ , 1 ๐ +๐ 1 1 ๐ 11 ๐ +๐ 1 ๐ โ 1 โ 1 1 โ 11 โ 1 1 1 1 1 1 1 ๐๐ โ ๐๐ 1 โค 1 ๐๐ 11 ๐ ๐ โ 11 1 1 โค 24+3๐ . ๐=1
๐=1
๐=1
๐=๐ +1
(8.44)
( ) Denote by ๐ถ 0 ๐ฐ, โฌ๐ ๐ด the Banach algebra of families ๐ = {๐๐ }๐โ๐ผ of continuous functions ๐๐ : ๐๐ โ ๐ฟ(๐, โ) such that โฅ๐โฅ := sup โฅ๐๐ โฅ๐๐ < โ, ๐โ๐ผ
488
J. Leiterer
( ) and let ๐ถ 0 ๐ฐ , โฌ๐ ๐บ๐ด be the ( group) of invertible elements of this Banach algebra, i.e., the group of all ๐ โ ๐ถ 0 ๐ฐ, โฌ๐ ๐ด such that ๐๐ (๐) โ ๐บ๐ฟ(๐, โ) for all ๐ โ ๐ผ, ๐ โ ๐๐ and sup โฅ๐โ1 ๐ โฅ๐๐ < โ. ๐โ๐ผ โโ Then we see from (8.44) and (8.41) that the in๏ฌnite product ๐ := ๐=1 ๐๐ converges ( ) in ๐ถ 0 ๐ฐ, โฌ๐ ๐บ๐ด , where โฅ๐ โ 1โฅ โค 2โฅ๐ โ 1โฅ (8.45) and, hence (as โฃ๐ โ 1โฅ โค 1/64), 1 โฅ๐ โ 1โฅ โค . (8.46) 32 Moreover, from (8.31) and (8.4) we see that (๐1 . . . ๐๐ ) โ ๐ = ๐๐ ,
๐ โ โโ .
By (8.36) this implies that 1 1 1(๐1 . . . ๐๐ ) โ ๐ โ 11 โค โฅ๐ โ 1โฅ , ๐ โ โโ . 23๐ Hence ๐ โ ๐ = lim (๐1 . . . ๐๐ ) โ ๐ = 1. ๐โโ ( ) Therefore it remains to prove that ๐ belongs to ๐ถ 0 ๐ฐ, โฌ๐โ๐บ๐ด and satis๏ฌes (8.8) ((8.7) is identical (8.45)). As we already ( know) that โฅ๐ โ 1โฅ < 1, for this it is su๏ฌcient to prove that ๐ belongs to ๐ถ 0 ๐ฐ, โฌ๐โ๐ด and satis๏ฌes (8.8). From (8.33) and (8.36) we see that ๐ถ 16 โฅโ๐๐+1 โฅ โค 6๐ โฅ๐ โ 1โฅ2 + 3๐ โฅโ๐๐ โฅโฅ๐ โ 1โฅ, ๐ โ โ. 2 2 Since โฅ๐ โ 1โฅ โค 1/64, this implies that ๐ถ 1 โฅโ๐๐+1 โฅ โค 6๐+6 โฅ๐ โ 1โฅ + 3๐+2 โฅโ๐๐ โฅ, ๐ โ โ. (8.47) 2 2 Setting ๏ฌrst ๐ = 0 and then ๐ = 1, this gives (recall that ๐0 = ๐ ) ๐ถ 1 โฅโ๐1 โฅ โค 6 โฅ๐ โ 1โฅ + 2 โฅโ๐ โฅ (8.48) 2 2 and ( ) ๐ถ 1 ๐ถ 1 ๐ถ 1 โฅโ๐2 โฅ โค 12 โฅ๐ โ 1โฅ + 5 โฅโ๐1 โฅ โค 12 โฅ๐ โ 1โฅ + 5 โฅ๐ โ 1โฅ + 2 โฅโ๐ โฅ 2 2 2 2 26 2 ( ) 1 1 1 ๐ถ 1 = ๐ถ 12 + 11 โฅ๐ โ 1โฅ + 7 โฅโ๐ โฅ โค 10 โฅ๐ โ 1โฅ + 7 โฅโ๐ โฅ. (8.49) 2 2 2 2 2 Next we prove by induction that, for all ๐ โ โโ , ๐ถ 1 โฅโ๐๐+1 โฅ โค ๐+9 โฅ๐ โ 1โฅ + ๐+6 โฅโ๐ โฅ. 2 2
(8.50)
An Estimate for the Splitting
489
For ๐ = 1 this holds true by (8.49). Now let ๐ โ โโ such that (8.50) is already proved. Then, by (8.47) and (8.50), โฅโ๐(๐+1)+1 โฅ โค
๐ถ
โฅ๐ โ 1โฅ +
1
โฅโ๐๐+1 โฅ ( ) ๐ถ 1 ๐ถ 1 โค 6(๐+1)+6 โฅ๐ โ 1โฅ + 3(๐+1)+2 โฅ๐ โ 1โฅ + โฅโ๐ โฅ 2๐+9 2๐+6 2 2 ( ) 1 1 1 = ๐ถ 6๐+12 + 4๐+14 โฅ๐ โ 1โฅ + 4๐+11 โฅโ๐ โฅ 2 2 2 ๐ถ 1 โค (๐+1)+9 โฅ๐ โ 1โฅ + (๐+1)+6 โฅโ๐ โฅ. 2 2 So (8.50) is proved for all ๐ โ โโ . As ๐0 = ๐ , we see from (8.35) that 26(๐+1)+6
23(๐+1)+2
๐ถ 33 โฅ๐ โ 1โฅ + โฅโ๐ โฅ. 16 32 Moreover, from (8.35), (8.36), and (8.48) we see โฅโ๐1 โฅ โค
(8.51)
๐ถ 33 โฅ๐1 โ 1โฅ + โฅโ๐1 โฅ 16 ( 32 ) ๐ถ 33 ๐ถ 1 ๐ถ 33 โค 7 โฅ๐ โ 1โฅ + โฅ๐ โ 1โฅ + โฅโ๐ โฅ โค 5 โฅ๐ โ 1โฅ + 7 โฅโ๐ โฅ. 2 32 26 22 2 2
โฅโ๐2 โฅ โค
(8.52)
If ๐ โฅ 3, then we see from (8.35), (8.36), and (8.50) that ๐ถ 33 โฅ๐ โ 1โฅ + 5 โฅโ๐๐โ1 โฅ 16 ๐โ1 2 ( ) ๐ถ 33 ๐ถ 1 โค 3๐+1 โฅ๐ โ 1โฅ + 5 โฅ๐ โ 1โฅ + ๐+4 โฅโ๐ โฅ 2 2 2๐+7 2 ( ) ๐ถ 1 33 33 = ๐+5 + 7 โฅ๐ โ 1โฅ + ๐+9 โฅโ๐ โฅ 2๐โ4 2 2 2 2
โฅโ๐๐ โฅ โค
(8.53)
๐ถ 33 โฅ๐ โ 1โฅ + ๐+9 โฅโ๐ โฅ. 2๐+5 2 For ๐ โฅ 3, from the product rule we get the estimate 1 ๐ 1 1 ๐ 1 1 ๐ โ1 1 ๐ โ1 1 ๐โ1 1 1 ๐ 1 1 โ 1 1โ 1 1โ 1 โ 1โ 1 1 โ 1 1โ 1 1 1 1 1 1 1 1 ๐๐ 1 โค โฅโ๐1 โฅ1 ๐๐ 1 + โฅโ๐๐ โฅ1 ๐๐ 1 + โฅโ๐๐ โฅ1 ๐๐ 1 โ
1 ๐๐ 1 1 1, โค
๐=1
where, by (8.42),
๐=2
๐=1
๐=2
1โ 1 1โ 1 1 ๐ 1 1 ๐ 1 1 1 1 1โค1+ 1 , ๐ โค 1 + ๐ โ 1 ๐1 ๐ 1 1 1 28 ๐=2 ๐=2
1 ๐โ 1 1 ๐โ 1 1 โ1 1 1 โ1 1 1 1 1 1 ๐๐ 1 โค 1 + 1 ๐ ๐ โ 11 1 1 โค 1 + 25 , ๐=1 ๐=1
๐=1
๐=๐+1
490
J. Leiterer
and 1 ๐โ1 1 1 ๐ 1 ( )( ) 1โ 1 1 โ 1 1 1 1 1 1 1 1 ๐๐ 1 โ
1 ๐๐ 1 โค 1 + 5 1 + 5+3๐ โค 1 + 6 1 2 2 2 ๐=1 ๐=๐+1
if ๐ โฅ 2.
Hence 1 ๐ 1 ( ) ( ) ( ) ๐โ1 1 โ 1 1 1 1 โ 1โ 1 ๐ โค 1 + โฅโ๐ โฅ + 1 + โฅโ๐ โฅ + 1 + โฅโ๐๐ โฅ ๐1 1 ๐ 1 28 26 25 ๐=1 ๐=2 ( ) ( ) ( ) โ 1 1 1 โ โค 1 + 8 โฅโ๐1 โฅ + 1 + 5 โฅโ๐2 โฅ + 1 + 5 โฅโ๐๐ โฅ for ๐ โฅ 3. 2 2 2 ๐=3 Together with (8.51), (8.52), and (8.53) this implies that 1 ๐ 1 ( )( ) 1 โ 1 1 ๐ถ 33 1โ 1 ๐ โค 1 + โฅ๐ โ 1โฅ + โฅโ๐ โฅ ๐1 1 28 16 25 1 ( )( ) 1 ๐ถ 33 + 1+ 5 โฅ๐ โ 1โฅ + 7 โฅโ๐ โฅ 2 25 2 ( )โ ) โ ( 1 ๐ถ 33 + 1+ 5 โฅ๐ โ 1โฅ + โฅโ๐ โฅ 2 2๐+5 2๐+9
for ๐ โฅ 3,
๐=3
and further, taking into account that ( ) ( ) ( ) โ 1 ๐ถ 1 ๐ถ 1 โ ๐ถ ๐ถ ๐ถ ๐ถ ๐ถ 1+ 8 + 1+ 5 + 1 + = + + < 5 5 ๐+5 2 16 2 2 2 2 8 16 64 4 ๐=3 and
( ) ( ) ( ) โ 1 33 1 33 1 โ 33 1+ 8 + 1 + + 1 + 2 25 25 2 7 25 ๐=3 2๐+9 ( ) ( ) 33 1 1 1 1 1 33 1 99 = 5 1 + 2 + 6 + 7 + 8 + 11 < 5 1 + = < 2, 2 2 2 2 2 2 2 2 64
we obtain that 1 ๐ 1 1 โ 1 ๐ถ 1โ ๐๐ 1 1 1 โค 4 โฅ๐ โ 1โฅ + 2โฅโ๐ โฅ for ๐ โฅ 3. 1
(8.54)
In particular, as โฅ๐ โ 1โฅ โค 1/64, 1 ๐ 1 1 โ 1 ๐ถ 1โ ๐๐ 1 1 1 โค 28 + 2โฅโ๐ โฅ for ๐ โฅ 3. 1
(8.55)
An Estimate for the Splitting
491
Moreover, from the product rule and (8.43) we see that 1 ๐ +๐ โ1 1 1 ๐โ 1 1 ๐โ 1 +๐ 1 โ 1 1 1 1 +๐ 1 1 1 1โ 1 โค โฅโ๐๐ +1 โฅ1 1 ๐ ๐ + โฅโ๐ โฅ ๐ 1 ๐ ๐1 ๐ +๐ ๐1 1 1 1 1 1 ๐=๐ +1
๐=๐ +2
+ โค4
๐ +๐ โโ1 ๐=๐ +2 โ โ
๐=๐ +1
1 ๐โ1 1 1 ๐ +๐ 1 1 โ 1 1 โ 1 1 1 โฅโ๐๐ โฅ1 ๐ ๐๐ 1 ๐1 โ
1 1 1 ๐=๐ +1
โฅโ๐๐ โฅ
๐=๐+1
if ๐, ๐ โ โ and ๐ โฅ 3.
๐=๐+1
By (8.53) and taking into account that โฅ๐ โ 1โฅ โค 2โ6 , this implies that 1 ๐โ 1 ( ) ( ) +๐ โ โ โ โ 1 1 ๐ถโฅ๐ โ 1โฅ 33โฅโ๐ โฅ ๐ถ โฅโ๐ โฅ 1โ 1 ๐๐ 1 โค + ๐+7 โค + ๐+1 1 2๐+3 2 2๐+9 2 ๐=๐ +1
๐=๐ +1
๐=๐ +1
(8.56)
๐ถ
โฅโ๐ โฅ ๐ถ + โฅโ๐ โฅ = ๐+9 + ๐+1 < , ๐, ๐ โ โ, ๐ โฅ 3. 2 2 2๐ +1 Again by the product rule, 1 ๐โ 1 1 (โ ( ๐โ ))1 ๐ ๐ +๐ โ 1 +๐ 1 1 1 1โ 1 = 1โ 1 ๐ โ โ ๐ ๐ ๐ โ 1 ๐ ๐1 ๐ ๐ 1 1 1 ๐=1
๐=1
๐=1
๐=๐+1
1 โ 1 1 +๐ 1 1โ 1 1 1 +๐ 1 ๐ 1 1 ๐โ 1 1 ๐ 1 1 ๐โ 1 1 1 1 1 1 1 1 โค 1โ ๐๐ 1 โ
1 ๐๐ โ 1 1 + 1 ๐๐ 1 โ
1โ ๐๐ 1 1 ๐=1
๐=๐+1
๐=1
๐=๐ +1
for all ๐ โ โ and ๐ โ โโ . If ๐, ๐ โฅ 3, then, by (8.55), (8.42), (8.43), and (8.56), this implies that 1 ๐โ 1 ( ) ๐ โ 1 +๐ 1 ๐ถ 1 ๐ถ + โฅโ๐ โฅ ๐ถ + โฅโ๐ โฅ 1โ 1 ๐๐ โ โ ๐๐ 1 โค + 2โฅโ๐ โฅ 5+3๐ + โค . 1 8 ๐ 2 2 2 2๐โ1 ๐=1 ๐=1 โโ Together with (8.44) this implies that the in๏ฌnite product ๐ = ๐=1 ๐๐ converges ( ) even in the Banach space ๐ถ 0 ๐ฐ, โฌ๐โ๐ด endowed with the norm (7.6), where, by (8.54), ๐ถ โฅโ๐โฅ โค โฅ๐ โ 1โฅ + 2โฅโ๐ โฅ. 4 By de๏ฌnition (8.30) of ๐ถ, this completes the proof of Theorem 8.3.
9. Proof of Theorem 4.2
( ) The cocycle ๐ belongs, in particular, to ๐ 1 ๐ฐ, โฌ๐โ๐บ๐ด (Section 8.2). Therefore, in ( ) view of the second estimate in (4.6), from Theorem 8.3 we get ๐ โ ๐ถ 0 ๐ฐ, โฌ๐โ๐บ๐ด such that ๐ โ ๐ = 1, (9.1)
492
J. Leiterer โฅ๐ โ 1โฅ โค 2โฅ๐ โ 1โฅ,
(9.2)
220 โฅ๐ โ 1โฅ. ๐
(9.3)
and โฅโ๐โฅ โค By (9.1) As ๐๐๐
๐๐๐ = ๐๐ ๐๐โ1 on ๐๐ โฉ ๐๐ , ๐, ๐ โ ๐ผ. is holomorphic in ๐๐ โฉ ๐๐ โฉ int ๐, this implies that ) ( ) 0 = (โ๐๐ ๐๐โ1 โ ๐๐ ๐๐โ1 โ๐๐ ๐๐โ1 on ๐๐ โฉ ๐๐ , ๐, ๐ โ ๐ผ,
(9.4)
i.e.,
) ( ) ๐๐โ1 (โ๐๐ = ๐๐โ1 โ๐๐ on ๐๐ โฉ ๐๐ , ๐, ๐ โ ๐ผ. Therefore, we have a well-de๏ฌned continuous function ๐ : ๐ โ ๐ด such that ๐ = ๐๐โ1 โ๐๐
on ๐๐ ,
๐ โ ๐ผ.
(9.5)
Since โฅ๐ โ 1โฅ โค 1/64, we see from (9.2) that โฅ๐ โ 1โฅ โค 1/32 and therefore โฅ๐ โ1 โฅ โค 2.
(9.6)
Together with (9.5) and (9.3), this implies that 221 โฅ๐ โ 1โฅ (9.7) ๐ and further, by the ๏ฌrst estimate in (4.6), 1 โฅ๐ โฅ๐ โค . (9.8) 8๐ถ๐ Therefore, we can apply Theorem 5.1 and obtain a continuous function ๐ : ๐ โ ๐บ๐ด such that โ๐ is also continuous on ๐, โฅ๐ โฅ๐ โค
๐ โ1 โ๐ = ๐
on ๐,
(9.9)
โฅ๐ โ 1โฅ๐ โค 2๐ถ๐ โฅ๐ โฅ๐ . From (9.10) and (9.8) it follows that 1 โฅ๐ โ 1โฅ๐ โค . 4
(9.10) (9.11)
Hence, ๐ belongs to the group (โฌ๐ โ )๐บ๐ด (๐). Therefore, setting โ๐ = ๐๐ ๐ โ1
on ๐๐ , ๐ โ ๐ผ, ( ) ๐บ๐ด in ๐ถ ๐ฐ, โฌ๐โ . By Proposition 8.1, from (9.9) 0
we obtain an element โ := {โ๐ }๐โ๐ผ and (9.5) we see that ( ) ( ) ( ) โโ๐ = โ๐๐ ๐ โ1 โ ๐๐ ๐ โ1 โ๐ ๐ โ1 = โ๐๐ ๐ โ1 โ ๐๐ ๐ ๐ โ1 ( ) ( ) = โ๐๐ ๐ โ1 โ ๐๐ ๐๐โ1 โ๐๐ ๐ โ1 = 0 ( ) on ๐๐ โฉ int ๐, i.e., โ even belongs to ๐ถ 0 ๐ฐ, โฌ๐ช๐บ๐(๐,โ) . From (9.4) we see that โ1 โ๐ โโ1 ๐ ๐๐โ1 = ๐๐ ๐๐โ1 = ๐๐๐ ๐ = ๐๐ ๐
on ๐๐ โฉ ๐๐ , i.e., we have (4.7).
An Estimate for the Splitting
and
It remains to prove (4.8). From (9.11) it follows that ๐ โ1 = โฅ๐ โ1 โ 1โฅ๐ โค
โ โ ๐=1
โฅ1 โ ๐ โฅ๐๐ โค โฅ1 โ ๐ โฅ๐
โ โ ๐=1
493 โโ
๐=0 (1
โ ๐ )๐
1 4 = โฅ1 โ ๐ โฅ๐ . 4๐โ1 3
By (9.10) and (9.7), this further implies that โฅ๐ โ1 โ 1โฅ๐ โค
8 224 ๐ถ๐ โฅ๐ โฅ๐ โค ๐ถ โฅ๐ โ 1โฅ. 3 3๐ ๐
(9.12)
Moreover, by de๏ฌnition of โ๐ , โ๐ โ 1 = ๐๐ ๐ โ1 โ 1 = (๐๐ โ 1)(๐ โ1 โ 1) + (๐๐ โ 1) + (๐ โ1 โ 1). Therefore โฅโ โ 1โฅ โค โฅ๐ โ 1โฅโฅ๐ โ1 โ 1โฅ๐ + โฅ๐ โ 1โฅ + โฅ๐ โ1 โ 1โฅ๐ . Together with (9.2) and (9.12) this implies that 225 224 ๐ถ๐ โฅ๐ โ 1โฅ2 + 2โฅ๐ โ 1โฅ + ๐ถ โฅ๐ โ 1โฅ, 3๐ 3๐ ๐ and further, as โฅ๐ โ 1โฅ โค 1/64, ( 25 ) ( ) 2 224 223 ๐ถ๐ โฅโ โ 1โฅ โค ๐ถ +2+ ๐ถ โฅ๐ โ 1โฅ โค 2 + โฅ๐ โ 1โฅ. 3 โ
64๐ ๐ 3๐ ๐ ๐ โฅโ โ 1โฅ โค
This completes the proof of Theorem 4.2.
References [CG] [BR]
[B] [GL]
Cornalba, M., Gri๏ฌths, Ph., Analytic cycles and vector bundles on non-compact algebraic varieties, Inventiones math. 28 (1975), 1โ106. Berndtsson, B., Rosay, J.-P., Quasi-isometric vector bundles and bounded factorization of holomorphic matrices, Ann. Inst. Fourier, Grenoble 51, 3 (2001), 885โ901. Bungart, L., On analytic ๏ฌber bundles. I. Holomorphic ๏ฌber bundles with in๏ฌnitedimensional ๏ฌbers, Topology 7 (1967), 55โ68. Gohberg, I., Leiterer, J., Holomorphic operator functions of one variable and applications, Birkhยจ auser 2009.
[GR1] Gohberg, I., Rodman, L., Analytic operator-valued functions with prescribed local data, J. Analyse Math. 40 (1981), 90โ128. [GR2] Gohberg, I., Rodman, L., Analytic matrix functions with prescribed local data, Acta Sci. Math. (Szeged) 45 (1983), no. 1-4, 189โ199. [G]
Grauert, H., Analytische Faserungen u ยจber holomorph-vollstยจ andigen Rยจ aumen, Math. Ann. 135 (1958), 263โ273.
[HL]
Henkin, G., Leiterer, J., Theory of functions on complex manifolds, Birkhยจ auser 1984.
494 [L]
[P]
J. Leiterer Leiterer, J., An estimate for the splitting of holomorphic cocycles. Several variables, Proceedings of the Workshop on Geometric Analysis of Several Complex Variables and Related Topics. Edited by: S. Berhanu, A. Meziani, N. Mir, R. Meziani, Y. Barkatou. Contemporary Mathematics, Volume 550, 2011. Pompeiju, D., Sur les singularitยดes des fonctions analytiques uniformes, C. R. Acad. Sci. Paris 139 (1904), 914โ915.
Jยจ urgen Leiterer Institut fยจ ur Mathematik Humboldt-Universitยจ at zu Berlin Rudower Chaussee 25 D-12489 Berlin, Germany e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 495โ512 c 2012 Springer Basel AG โ
The Discrete Algebraic Riccati Equation and Hermitian Block Toeplitz Matrices Leonid Lerer and Andrยดe C.M. Ran Dedicated to the memory of our dear friend and mentor Israel Gohberg
Abstract. This paper discusses the representation of the full set of solutions of the discrete algebraic Riccati equation in terms of two solutions, the di๏ฌerence of which is invertible. It turns out that if two such solutions exist, then all solutions can be described in terms of one of these solutions and the solution of a Stein equation. A complete parametrization is available in that case. Under some additional hypotheses we show that there are two solutions which di๏ฌer by an invertible matrix. In the ๏ฌnal sections we discuss a special case connected to invertible Hermitian block Toeplitz matrices. Mathematics Subject Classi๏ฌcation (2000). Primary 15A24, 47B35 Secondary 47A56 . Keywords. Riccati equations, Hermitian block Toeplitz matrices.
1. Introduction In this paper we discuss the discrete algebraic Riccati equation ๐ = ๐ด๐๐ดโ + ๐ โ ๐ด๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 ๐ต โ ๐๐ดโ .
(1.1)
We do not make any assumption on de๏ฌniteness of ๐
and ๐, and neither on de๏ฌniteness of the associated Popov function, but we shall assume that ๐
is an invertible Hermitian matrix, ๐ is Hermitian, and in addition, that ๐ด is invertible. A more general version of the discrete algebraic Riccati equation plays a role in several problems in the theory of discrete time systems. It appears in the theory of linear quadratic optimal control [23], where one is interested in one particular solution in case of ๏ฌxed end-point control, or in a whole class of The research of the ๏ฌrst author is supported by the Israel Science Foundation, ISF (Grant number 121/09).
496
L. Lerer and A.C.M. Ran
solutions in case of linear or free end-point control [21]. The Riccati equation also appears in realization theory for stochastic linear systems, in ๐ป โ optimal control, in factorization problems and in several other problems, [16, 13, 8, 12]. The structure of the set of all Hermitian solutions has been a subject of study since the late ninety-seventies. There are parametrizations of the set of all solutions in terms of invariant subspaces [16, 22], in terms of symmetric factorizations of certain rational matrix functions or matrix polynomials, see, e.g., [1, 13, 14, 19], and in terms of two solutions, e.g., extremal solutions [23] or unmixed solutions [4, 5, 7, 25, 26, 27]. For the unsymmetric continuous time analogue a parametrization of the set of all solutions in terms of two solutions is given in [6]. The paper discusses two main themes. The ๏ฌrst topic is concerned with the general theory, much in the spirit of the parametrization of all solutions in terms of two particular solutions. In particular, we show that if ๐ = 0, and ๐ด and (๐ดโ )โ1 have no common eigenvalues, then there is a one-to-one correspondence between all solutions and a class of ๐ดโ -invariant subspaces. The second topic concerns the case where the matrices ๐ด and ๐ต have a speci๏ฌc companion-like structure. It is clear that under the assumptions ๐ = 0 and ๐ด is invertible the study of equation (1.1) is closely related to the well-known Stein equation. In [11] a profound connection has been found between Stein equations, block Toeplitz matrices, and inverse and direct problems for related matrix orthogonal polynomials (in particular the relation (4.6) below was discovered there). See also [15], where the connection between orthogonal polynomials and Toeplitz matrices originates. In our view it is of great interest to explore the connections between the results of [11] and the study of Riccati equations (1.1). The speci๏ฌc structure of ๐ด and ๐ต in companion form allows us to describe the set of all solutions of (1.1) in terms of two matrices: one solution, and the inverse of a certain block Toeplitz Hermitian matrix. We start the paper with some preliminary results on the discrete algebraic Riccati equation. In particular, in the next section we show how invertible solutions for the case ๐ = 0 may be found by solving a Stein equation. In the third section we describe all solutions in terms of two solutions whose di๏ฌerence is invertible. This part of the paper is close in spirit to results describing all solutions in terms of two unmixed solutions, see, e.g., [25, 27], compare also [6]. In the fourth section we introduce the special case we wish to consider, describe the connection with block Toeplitz Hermitian matrices, and describe all solutions for this speci๏ฌc case. Because of the speci๏ฌc nature of the coe๏ฌcients, we can derive the description of all solutions under conditions that are weaker than those of the third section. To be more precise, the coe๏ฌcient ๐ is zero in this case. We show that under speci๏ฌc conditions on ๐ด and ๐ต there is a unique invertible solution, which is the inverse of a Hermitian block Toeplitz matrix. All other solutions are given as a product of the invertible solution and certain projections. Surprisingly enough, all solutions are completely determined by their ๏ฌrst block column. In the ๏ฌfth section we present some examples of the special case. In the ๏ฌnal section we discuss the relation of the special case to polynomial factorization, which is closely related to the factorization results for the Popov function.
Discrete Riccati Equations and Block Toeplitz Matrices
497
2. Preliminaries on the discrete algebraic Riccati equation In this section we discuss some properties of the algebraic Riccati equation. Let ๐ and ๐
be ๐ ร ๐ and ๐ ร ๐ Hermitian matrices, respectively, and assume that ๐
is invertible. Let ๐ด be an ๐ ร ๐ matrix. We assume throughout that ๐ด is invertible as well. Let ๐ต be an ๐ ร ๐ matrix. Consider the discrete algebraic Riccati equation ๐ = ๐ด๐๐ดโ + ๐ โ ๐ด๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 ๐ต โ ๐๐ดโ .
(2.1)
Our ๏ฌrst observation is the following: let ๐ be a solution of the equation (2.1). Let ๐ be invertible, and denote ๐ด1 = ๐๐ด๐ โ1 , ๐ต1 = (๐ โ )โ1 ๐ต, ๐1 = ๐๐๐ โ . Then ๐1 = ๐๐๐ โ satis๏ฌes the equation ๐1 = ๐ด1 ๐1 ๐ดโ1 + ๐1 โ ๐ด1 ๐1 ๐ต1 (๐
+ ๐ต1โ ๐1 ๐ต1 )โ1 ๐ต1โ ๐1 ๐ดโ1 and conversely, every solution of this equation gives rise to a solution of the original equation in this way. Our second observation is the following (compare [21, Lemma 3.1]). Proposition 2.1. Let ๐0 be a solution to (2.1), and let ๐ be any other solution. Denote by ๐ด0 = ๐ด โ ๐ด๐0 ๐ต(๐
+ ๐ต โ ๐0 ๐ต)โ1 ๐ต โ , and ๐
0 = ๐
+ ๐ต โ ๐0 ๐ต. The latter matrix is invertible by assumption, since ๐0 is a solution of (2.1). Then ๐ = ๐ โ ๐0 satis๏ฌes the equation ๐ = ๐ด0 ๐ ๐ดโ0 โ ๐ด0 ๐ ๐ต(๐
0 + ๐ต โ ๐ ๐ต)โ1 ๐ต โ ๐ ๐ดโ0 . We shall be interested in a very speci๏ฌc pair of solutions, namely two solutions ๐ and ๐0 such that their di๏ฌerence ๐โ๐0 is invertible. For this reason we consider invertible solutions of the special homogeneous algebraic Riccati equation when ๐ = 0: (2.2) ๐ = ๐ด๐๐ดโ โ ๐ด๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 ๐ต โ ๐๐ดโ . We assume that ๐ด is invertible. Observe that 0 is obviously a solution, but we are especially interested in invertible solutions of (2.2). The following is a well-known result, see, e.g., [25, 26]. Proposition 2.2. Assume that ๐ด is invertible, that ๐
is Hermitian and invertible, and let ๐ be an invertible solution of (2.2). Then ๐ = ๐ โ1 solves the Stein equation (2.3) ๐ โ ๐ดโ ๐ ๐ด = โ๐ต๐
โ1๐ต โ โ1 and conversely, if ๐ is an invertible solution of (2.3), then ๐ = ๐ solves the discrete algebraic Riccati equation (2.2). Proof. The proof is a straightforward computation using the invertibility of ๐ด and ๐ . โก As a result of the above considerations we arrive at the conclusion: assume that ๐ด is invertible and we are given one solution ๐0 of (2.1), then in order to study solutions for which ๐ โ ๐0 is invertible, it is enough to study the invertible solutions of the Stein equation (2.3).
498
L. Lerer and A.C.M. Ran
3. Non-invertible solutions In this section we consider other solutions ๐ of the algebraic Riccati equation ๐ = ๐ด๐๐ดโ โ ๐ด๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 ๐ต โ ๐๐ดโ .
(3.1)
We assume throughout that ๐+ is an invertible solution, and recall our standing assumptions that ๐
is Hermitian and invertible, and that ๐ด is invertible as well. We ๏ฌrst prove the following result, which is well known in the case of positive de๏ฌnite ๐
(compare, e.g., [22, 23, 25]) Theorem 3.1. Let ๐ด be invertible, let ๐
be Hermitian and invertible. Let ๐+ be an invertible solution of (3.1). Let ๐ be an ๐ดโ -invariant subspace which is ๐+ ห . Let ๐๐ be the projection onto (๐+ ๐ )โฅ nondegenerate, that is, โ๐ = (๐+ ๐ )โฅ +๐ along ๐ , and put ๐๐ = ๐+ ๐๐ . Then ๐๐ is a Hermitian solution of (3.1) for which Ker ๐๐ = ๐ . Proof. With respect to the orthogonal decomposition โ๐ = ๐ โ ๐ โฅ we write ( ) ( ) ( ) ๐11 ๐12 0 ๐ด11 ๐ต1 ๐+ = , ๐ด = , ๐ต = , โ ๐ด21 ๐ด22 ๐12 ๐22 ๐ต2 where the (1,2)-entry of ๐ด is zero because ๐ is invariant under ๐ดโ . Also ๐11 is invertible because of the assumption that ๐ is ๐+ -nondegenerate, and so ๐+ ๐ โฉ ๐ โฅ = (0). Then ( ( ( ))โฅ ( ))โฅ ๐ผ ๐11 = Im (๐+ ๐ )โฅ = Im โ1 โ โ ๐12 ๐11 ๐12 ) ( โ1 ( ) โ๐11 ๐12 โ1 = Ker ๐ผ ๐11 . ๐12 = Im ๐ผ One then easily sees that ๐๐ , being the projection onto (๐+ ๐ )โฅ along ๐ , is given by ( ) โ1 ๐12 0 โ๐11 ๐๐ = , 0 ๐ผ and hence ๐๐ = ๐+ ๐๐ is given by ( )( ) ) ( โ1 0 0 ๐11 ๐12 0 โ๐11 ๐12 ๐๐ = . (3.2) = โ1 โ โ ๐12 ๐22 0 ๐22 โ ๐12 0 ๐ผ ๐11 ๐12 โ1 โ It follows that ๐๐ is Hermitian. Denote ๐22 = ๐22 โ ๐12 ๐11 ๐12 . Now ๐22 is the Schur complement of ๐11 in the invertible matrix ๐+ . That is, we have )( )( ) ( โ1 ๐ผ 0 ๐11 0 ๐ผ ๐11 ๐12 . (3.3) ๐+ = โ1 โ 0 ๐22 ๐11 ๐ผ 0 ๐ผ ๐12
It then follows from the invertibility of ๐+ that also ๐22 is invertible. Hence Ker ๐๐ = ๐ . It remains to prove that ๐๐ solves (3.1). For this we use the fact that ๐ = โ1 satis๏ฌes ๐+ (3.4) ๐ โ ๐ดโ ๐ ๐ด = โ๐ต๐
โ1 ๐ต โ .
Discrete Riccati Equations and Block Toeplitz Matrices By (3.3) we can write ( ๐ผ โ1 ๐ = ๐+ = 0
โ1 ๐12 โ๐11 ๐ผ
) ( โ1 ๐11 0
0 โ1 ๐22
)(
๐ผ
โ1 โ โ๐12 ๐11
) 0 . ๐ผ
499
(3.5)
In particular, concentrating on the (2,2)-block in equation (3.4), we see that โ1 โ1 โ ๐ดโ22 ๐22 ๐ด22 = โ๐ต2 ๐
โ1 ๐ต2โ . ๐22
(3.6)
Hence, by Proposition 2.2, ๐22 satis๏ฌes the Riccati equation ๐22 = ๐ด22 ๐22 ๐ดโ22 โ ๐ด22 ๐22 ๐ต2 (๐
+ ๐ต2โ ๐22 ๐ต2 )โ1 ๐ต2โ ๐22 ๐ดโ22 .
(3.7)
Now consider our claim that ๐๐ = ๐ด๐๐ ๐ดโ โ ๐ด๐๐ ๐ต(๐
+ ๐ต โ ๐๐ ๐ต)โ1 ๐ต โ ๐๐ ๐ด.
(3.8)
To show that this is indeed the case, one uses the facts that ( ( ) ) 0 0 0 ๐ต = , ๐ด๐ , ๐ด๐๐ ๐ดโ = ๐ 0 ๐ด22 ๐22 ๐ดโ22 ๐ด22 ๐22 ๐ต2 and
๐
+ ๐ต โ ๐๐ ๐ต = ๐
+ ๐ต2โ ๐22 ๐ต2 .
Then (3.8) is seen to be equivalent to (3.7).
โก
To prove a result in the converse direction we need to make an assumption on ๐ด, but ๏ฌrst we present a result in the converse direction that holds under the assumptions we have used so far. Proposition 3.2. Let ๐ด be invertible and let ๐
be Hermitian and invertible. Let ๐ be a solution of (3.1). Then ๐ = Ker ๐ is invariant under ๐ดโ . Proof. This is an easy consequence of the fact that ๐ด is invertible: if ๐ solves (3.1) then ) ( ๐(๐ดโ )โ1 = ๐ด โ ๐ด๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 ๐ต โ ๐. So, if ๐ฅ โ Ker ๐ then ๐(๐ดโ )โ1 ๐ฅ = 0. It follows that Ker ๐ is (๐ดโ )โ1 -invariant, โก and hence also is ๐ดโ -invariant. We would like to be able to conclude that for any solution ๐ its kernel is ๐+ -nondegenerate. This holds true under an extra assumption on ๐ด, as can be seen in the following theorem. Theorem 3.3. Let ๐ด be invertible, let ๐
be Hermitian and invertible. Assume in addition that ๐ด and (๐ดโ )โ1 have no common eigenvalues. Then, if ๐ solves (3.1) Ker ๐ is ๐+ -nondegenerate. There is a one-to-one correspondence between solutions of (3.1) and subspaces ๐ that are ๐ดโ -invariant and ๐+ -nondegenerate, which is given by ๐ = ๐+ ๐๐ , where ๐๐ is the projection onto (๐+ ๐ )โฅ along ๐ . In particular, there can be only one invertible solution of (3.1).
500
L. Lerer and A.C.M. Ran
Proof. Let ๐ be a solution of (3.1), let ๐ be its kernel, and let us write โ๐ = ๐ โ ๐ โฅ . With respect to this decomposition we write ) ) ( ) ( ( 0 0 ๐11 ๐12 ๐11 ๐12 โ1 , ๐ = ๐ , ๐ = . = ๐+ = โ โ + 0 ๐22 ๐12 ๐22 ๐12 ๐22 In order to prove that ๐ is ๐+ -nondegenerate we need to prove that ๐11 is invertible, or equivalently (see [2]), that ๐22 is invertible. Note that ๐22 is invertible. Using the fact that ๐ is ๐ดโ -invariant (see the previous proposition), we conโ1 โ1 sider the (2,2)-block in the equation ๐+ โ ๐ดโ ๐+ ๐ด = โ๐ต๐
โ1 ๐ต โ , to see that ๐22 satis๏ฌes the equation ๐22 โ ๐ดโ22 ๐22 ๐ด22 = โ๐ต2 ๐
โ1 ๐ต2โ .
(3.9) โ1 ๐22
satis๏ฌes this However, as we have seen in the proof of Theorem 3.1 also equation. If we could show that (3.9) has a unique solution we would be done, โ1 since in that case ๐22 = ๐22 is invertible. Now, since ๐ด and (๐ดโ )โ1 have no common eigenvalues, neither have ๐ดโ22 and ๐ดโ1 22 . This shows that indeed, (3.9) has a unique solution (see Chapter 13 in [17]), and so we have shown that ๐ is ๐+ -nondegenerate. Combining the results in Theorem 3.1 with what we have proved just now, we see that there is indeed a one-to-one correspondence between the set of solutions of (3.1) and the set of subspaces ๐ that are ๐ดโ -invariant and ๐+ -nondegenerate. โก Observe that the uniqueness of the solution of the Stein equation is in fact equivalent to ๐ดโ and ๐ดโ1 having no common eigenvalues. We ๏ฌnally arrive at the main result, which describes all noninvertible solutions of (3.1) in terms of the solution of the Stein equation (2.3). For an analogue for the non-symmetric continuous time algebraic Riccati equation compare [6], Theorem 3.1. Note however, that the parametrization given here is quite di๏ฌerent in nature. Theorem 3.4. Let ๐ด be invertible, let ๐
be Hermitian and invertible. Assume that ๐ด and (๐ดโ )โ1 have no common eigenvalues, and let ๐ be the unique solution of (2.3). Then there is a one-to-one correspondence between solutions of (3.1) and subspaces ๐ that are ๐ดโ -invariant and ๐ -nondegenerate which is given by ๐ = ๐ โ1 ๐๐ , where ๐๐ is the projection onto ๐ ๐ โฅ along ๐ . Proof. The theorem is a direct consequence of the previous one, once we realize that a subspace ๐ is ๐+ -nondegenerate if and only if it is ๐ -nondegenerate, and that (๐+ ๐ )โฅ = ๐ ๐ โฅ . โก Note that if we would assume that ๐
is positive de๏ฌnite and that the pair (๐ด, ๐ต) is controllable, then we get additional information. Indeed, this condition reduces to the controllability of (๐ด22 , ๐ต2 ) being controllable, and then if (3.9) has a solution the inertia of ๐+ is the same as the inertia of ๐ด, as is well known. (This is the discrete-time version of the Chen-Wimmer theorem [3, 24]; the precise formulation may be found in Chapter 13, Theorem 13.2.4 in [17].) This proves the following theorem.
Discrete Riccati Equations and Block Toeplitz Matrices
501
Theorem 3.5. Let ๐ด be invertible, let ๐
be positive de๏ฌnite, and let the pair (๐ด, ๐ต) be controllable. Assume that ๐ด has all its eigenvalues inside the open unit disc, or has all its eigenvalues outside the closed unit disc. Then there is a one-to-one correspondence between all solutions of (3.1) and the set of ๐ดโ -invariant subspaces. This correspondence is given as follows: let ๐๐ be the projection onto (๐+ ๐ )โฅ along ๐ , then the solution corresponding to ๐ is given by ๐๐ = ๐+ ๐๐ . Proof. In case ๐ด has all its eigenvalues inside the unit circle, ๐+ is positive de๏ฌnite, in case ๐ด has all its eigenvalues outside the unit circle ๐+ is negative de๏ฌnite. In these cases any ๐ดโ -invariant subspace is automatically ๐+ -nondegenerate. โก Finally, we comment on the connection between the continuous algebraic Riccati equation and the discrete algebraic Riccati equation (1.1). It is possible, under the assumption in place in this section, to reduce the discrete algebraic Riccati equation to a continuous one, in the sense that the set of solutions is the same. This is done by a Cayley transform type argument, see, e.g., the proof of Theorem 12.2.3 in [16], where this is outlined in detail. However, the coe๏ฌcients of the resulting continuous algebraic Riccati equation are complicated expressions in the original coe๏ฌcients. Moreover, our main new result, being Theorem 3.4 where the one-to-one correspondence of solutions of (3.1) and ๐ดโ -invariant ๐ nondegenerate subspaces is new, cannot be recovered in this way from existing results.
4. Riccati equations with coe๏ฌcients of companion type In this section we discuss a special case of the algebraic Riccati equation. A pair of matrices (๐ด, ๐ต) is called a monic pair if ๐ต is of size ๐๐ ร ๐ for some ๐ and( ๐, ๐ต is of full column rank ) ๐, while ๐ด is of size ๐๐ ร ๐๐ for some ๐, and rank ๐ต ๐ดโ ๐ต . . . (๐ดโ )๐โ1 ๐ต = ๐๐. Note that if (๐ด, ๐ต) is a monic pair, then so is (๐ด + ๐น ๐ต โ , ๐ต) for every ๐น of size ๐๐ ร ๐. When (๐ด, ๐ต) is a monic pair, it follows (see, e.g., [9]) that there exists an invertible matrix ๐ and ๐ ร ๐ matrices ๐พ1 , . . . , ๐พ๐ such that โ โ โ ๐พ1 ๐พ2โ โ
โ
โ
๐พ๐โ โ โ ๐ผ โ โ ๐ผ 0 โ
โ
โ
0 โ โ โ โ โ .. โ .. โ 0โ โ1 โ, . . 0 ๐ผ ๐ โ1 ๐ดโ ๐ = โ ๐ต = (4.1) ๐ โ. โ . โ โ โ .. โ โ . .. โ . . . โ . . . โ 0 0 โ
โ
โ
๐ผ 0 Recall that if ๐0 is a solution of the algebraic Riccati equation, then we denote ๐ด0 = ๐ด + ๐น ๐ต โ , where ๐น = โ๐ด๐0 ๐ต(๐
+ ๐ต โ ๐0 ๐ต)โ1 . Hence, if (๐ด, ๐ต) is a monic pair, then so is (๐ด0 , ๐ต). Theorem 4.1. Under the above conditions on the coe๏ฌcients ๐ด and ๐ต, the invertible solution ๐ of the Riccati equation (2.2) is congruent to the inverse of a block Toeplitz matrix.
502
L. Lerer and A.C.M. Ran
Proof. We ๏ฌrst observe that if ๐ is a solution of (2.2) with the coe๏ฌcients ๐ด, ๐ต and ๐
, then ๐ โ ๐๐ is a solution of (2.2) with the coe๏ฌcients ๐ โ1 ๐ด๐, ๐ โ1 ๐ต, and ๐
. We shall use the matrix ๐ such that (4.1) holds. Let ๐พ denote the matrix โ โ ๐พ1 ๐ผ 0 โ
โ
โ
0 .โ โ .. โ ๐พ2 0 ๐ผ . .. โ โ โ โ .. โ . .. .. (4.2) ๐พ = โ ... . . .โ โ โ โ โ . .. โ .. . ๐ผโ ๐พ๐ 0 โ
โ
โ
โ
โ
โ
0 Let ๐ต be as above. (๐) (๐) De๏ฌne ๐00 = โ๐
, and de๏ฌne ๐๐0 = โ๐พ๐ ๐
. We assume that ๐พ is invertible. The Stein equation (2.3) now becomes ) ( (๐) โ1 (4.3) ๐ โ ๐พ โ ๐ ๐พ = diag โ(๐00 ) 0 โ
โ
โ
0 . The corresponding algebraic Riccati equation is (๐)
๐ = ๐พ๐๐พ โ โ ๐พ๐๐ต(๐00 + ๐ต โ ๐๐ต)โ1 ๐ต โ ๐๐พ โ .
(4.4)
โ1
Now using [11] we see that ๐ = ๐ = ๐๐โ1, where ๐๐ is the block Toeplitz matrix given by ( (๐) ) 0 ห ห โ ๐00 ๐พ, ๐๐ = ๐พ 0 ๐ โ1 ห is given by where ๐พ
โ
(๐)
(๐00 )โ1 โ ๐พ1 โ โ โ ๐พ2 ห =โ ๐พ โ .. โ . โ โ .. โ . ๐พ๐
0 ๐ผ
โ
โ
โ
0
0
๐ผ .. . ..
0
. โ
โ
โ
โ
โ
โ
.. .. ..
โ
โ
โ
. .
. โ
โ
โ
๐ผ 0
โ 0 0โ โ .. โ .โ โ . .. โ .โ โ โ 0โ
๐ผ
We conclude that the invertible solution of the algebraic Riccati equation is the inverse of a block Toeplitz matrix. โก Now assume ๐พ โ and ๐พ โ1 have no common eigenvalues. The latter condition can be stated in terms of a certain matrix polynomial, see [11]. We solve the Stein โ1 and the solution is block-Toeplitz. Then we need to be more equation for ๐+ explicit about the solution ๐. We can take two points of view. The ๏ฌrst is that we are given two solutions (๐) of the algebraic Riccati equation, ๐0 and ๐1 , with ๐1 โ ๐0 invertible. Put ๐๐0
Discrete Riccati Equations and Block Toeplitz Matrices as above, and put ๐(๐) =
๐ โ ๐=0
503
(๐)
๐๐โ๐ ๐๐0 .
โ1 We assume that ๐(๐) does not have a zero ๐0 for which ๐ยฏ0 is also a zero. Let ๐๐ be the (unique, because of all our assumptions) block Toeplitz matrix for which โ (๐) โ โ โ ๐00 ๐ผ โ (๐) โ โ โ โ๐10 โ โ0โ โ ๐๐ โ (4.5) .โ . โ .. โ = โ โ . โ โ .. โ (๐) 0 ๐ ๐0
โ1 ๐+
Then = ๐๐โ1 , and now we can describe all solutions of the algebraic Riccati equation. The second point of view: we can assume that only one solution is given, ๐0 . Now assume in addition that the unique solution of (4.3) is invertible, then that gives us the ๐1 . Equivalently, that assumption is that there is an invertible block โ1 Toeplitz satisfying (4.5). Then again ๐+ = ๐๐โ1 , and we can continue as above. If ๐
> 0 the invertibility of the solution is automatic, see [11, Theorem 8.1]. Now we arrive at the main point of this section. In the previous section we had to assume that ๐ด and (๐ดโ )โ1 do not have common eigenvalues, to guarantee existence of a (unique) invertible solution to the Stein equation (2.3). However, for the particular case under consideration here, we may use Theorem 8.1 of [11] to deduce existence of an invertible solution to the Stein equation even when ๐พ and (๐พ โ )โ1 have a common eigenvalue. In this case an extra condition on the eigenvectors and generalized eigenvectors should be satis๏ฌed. To be precise, let ๐
= โ๐00 be positive de๏ฌnite, and introduce the orthogonal matrix polynomial associated to ๐พ by ๐ (๐) = โ๐(๐)๐
โ1/2 (see [11]). Then Theorem 8.1 in [11] tells us that the Stein equation (2.3) has an invertible solution if and only if for โ1 every symmetric pair of eigenvalues ๐0 , ๐ยฏ0 of ๐ (๐) the following holds true: for any right Jordan chains ๐ฅ1 , . . . , ๐ฅ๐ and ๐ฆ1 , . . . , ๐ฆ๐ of ๐ (๐) corresponding to ๐0 and โ1 ๐0 , respectively, there holds ๐ โ
โจ๐ฅ๐โ๐ , ๐ฆ๐ โฉ = 0, where ๐ = max ๐, ๐.
(4.6)
๐=1
The condition stated above in terms of eigenvalues of the polynomial ๐ (๐) can be reformulated in terms of the eigenvalues of the matrix ๐พ. This proves the following theorem. Theorem 4.2. Let ๐พ and ๐ต be as above, let ๐
be positive de๏ฌnite. If condition (4.6) โ1 holds whenever ๐0 , ๐0 are both eigenvalues of ๐พ, then there is an invertible block Toeplitz Hermitian matrix ๐ = ๐๐โ1 solving (4.3). The solutions to the Riccati equation (4.4) are in one-to-one correspondence to ๐พ-invariant subspaces ๐ which
504
L. Lerer and A.C.M. Ran
are ๐ -nondegenerate. The correspondence is given by ๐๐ = ๐๐โ ๐ โ1 ๐๐ , where ๐๐ is the projection onto ๐ ๐ โฅ along ๐ . If in the theorem above, we replace the condition ๐
> 0 by invertibility of the Hermitian matrix ๐
, then Theorem 7.2 in [11] guarantees the existence of a Hermitian block Toeplitz matrix ๐๐โ1 solving (4.3). However, invertibility of ๐๐โ1 is not guaranteed. We ๏ฌnish this section with some remarks on the inertia of solutions. For the (๐) โ1 case where ๐
= ๐00 is positive de๏ฌnite, the inertia of the solution ๐+ = ๐๐โ1 is given as follows: the number of negative eigenvalues of ๐+ is the number of zeros of the orthogonal matrix polynomial ๐ (๐) inside the unit disc, the number of positive eigenvalues of ๐+ is the number of zeros of ๐ (๐) outside the unit disc, see [11]. In case ๐
is no longer positive de๏ฌnite, the inertia of the solution ๐+ may still be described in terms of zeros of a certain matrix polynomial, see [20]. The inertia of the solution ๐๐ obviously is determined by ๐+ and the projection ๐๐ . In any case: #negative eigenvalues of ๐๐ โค #negative eigenvalues of ๐๐โ1 , #positive eigenvalues of ๐๐ โค #positive eigenvalues of ๐๐โ1 , #zero eigenvalues of ๐๐ = dim ๐. For some more information on the inertia of the solutions in case ๐
> 0, see [18].
5. Computations and examples We start this section with some remarks concerning computation of the solutions. Let ๐ be a ๐-dimensional ๐พ-invariant subspace, and take any basis ๐ค1 , . . . , ๐ค๐ for ) ( ๐ . Put ๐ = ๐ค1 โ
โ
โ
๐ค๐ , and let ๐ = ๐ ๐ท๐ โ be the singular ( value decompo) sition of ๐ . Denote the ๐th column of ๐ by ๐ข๐ , and put ๐1 = ๐ข๐+1 โ
โ
โ
๐ข๐ . Then: ๐1 ๐1โ : โ๐๐ โ โ๐๐ is the orthogonal projection along ๐ onto ๐ โฅ , ๐1 : โ๐๐โ๐ โ โ๐๐ is an isometry onto ๐ โฅ , and ๐1โ : โ๐๐ โ โ๐๐โ๐ is the orthoprojection along ๐ . For the actual computation of ๐๐ we can now use the proof of Theorem 3.1. In fact, we know from (3.2) and (3.5) that with respect to the decomposition โ๐๐ = ๐ โ ๐ โฅ the solution ๐๐ and the block Toeplitz ๐ are given by ( ) ( ) โ โ 0 0 , ๐ = ๐๐ = โ1 . 0 ๐22 โ ๐22 โ1 = ๐1โ ๐ ๐1 , and so ๐22 = (๐1โ ๐ ๐1 )โ1 . Thus Hence, ๐22
๐๐ = ๐1 (๐1โ ๐ ๐1 )โ1 ๐1โ . The computational method expanded here will be used in the following examples. Example 1. As a ๏ฌrst example, consider the scalar case, that is, the case where ๐ = 1. Then there is a ๏ฌnite number of ๐พ-invariant subspaces, since ๐พ is nonderogatory.
Discrete Riccati Equations and Block Toeplitz Matrices
505
Also, such subspaces can be completely described (see, e.g., [10]). We take ๐
= (๐) โ๐00 = โ1. Take for ๐ a one-dimensional subspace, such a subspace is spanned by the vector ( )๐ 1 1 ๐ฅ๐ = 1 ๐๐ โ
โ
โ
๐๐โ1 , ๐
which satis๏ฌes ๐พ๐ฅ๐ = ๐๐ ๐ฅ๐ . The number ๐๐ is a root of the polynomial ๐(๐) = 1 โ ๐1 ๐ โ ๐2 ๐2 โ โ
โ
โ
โ ๐๐ ๐๐ . To see whether ๐ is ๐ -nondegenerate, we have to compute ๐ฅโ๐ ๐ ๐ฅ๐ . Recalling that ๐ satis๏ฌes the Stein equation (2.3), with ๐ด = ๐พ โ , we see that ๐ฅโ๐ (๐ โ ๐พ โ ๐ ๐พ)๐ฅ๐ ๐ฅโ๐ = ๐ฅโ๐ ๐ ๐ฅ๐ (1 โ โฃ๐๐ โฃ2 ) = ๐ฅโ๐ ๐ต๐
โ1 ๐ต โ ๐ฅ๐ = โ1. So, ๐ฅโ๐ ๐ ๐ฅ๐ โ= 0, and ๐ is ๐ -nondegenerate. Example 2. As a second example, we take Consider the matrix โ 5 0 โ0 5/6 ๐พ =โ โโ6 0 0 โ1/6
a case where Theorem 4.2 applies. โ 1 0 0 1โ โ, 0 0โ 0 0
and take ๐
= ๐ผ2 . The eigenvalues of ๐พ are the numbers 2, 1/2, 3, 1/3, and so indeed, there are pairs of eigenvalues symmetrically placed with respect to the unit circle. The corresponding eigenvectors are, respectively, โ โ โ โ โ โ โ โ 1 0 1 0 โ0โ โโ3โ โ0โ โโ2โ โ โ โ โ โ โ โ ๐ฅ1 = โ โโ3โ , ๐ฅ2 = โ 0 โ , ๐ฅ3 = โโ2โ , ๐ฅ4 = โ 0 โ . 0 1 0 1 One sees immediately that the condition (4.6) is satis๏ฌed. To solve the corresponding Stein equation (4.3) one cannot just use Matlabโs dlyap function, as it will complain about multiple solutions. However, a block Toeplitz solution is easily found by solving the equation by hand, and is equal to โ โ 7/120 0 5/120 0 โ 0 โ21/10 0 โ3/2 โ โ ๐ =โ โ5/120 0 7/120 0 โ 0 โ3/2 0 โ21/10 The matrix ๐พ has four eigenvectors, leading to a total of 14 nontrivial invariant subspaces: four one-dimensional subspaces, six two-dimensional ones, and again four three-dimensional ones. One can check directly that all ๐พ-invariant non-trivial subspaces are ๐ nondegenerate, and hence there is a one-to-one correspondence between ๐พ-invariant subspaces and solutions to the algebraic Riccati equation. Thus there are 16
506
L. Lerer and A.C.M. Ran
solutions to the corresponding algebraic Riccati equation: the zero solution, ๐ โ1 and fourteen solutions corresponding to the invariant subspaces given above. To compute the sixteen solutions, one sees that because of the special form of ๐พ, all solutions will have a structure like that of ๐ . The computations can then be done by hand.
6. Connection with the Popov function
( Let (๐พ, ๐ต) be a monic pair, that is ๐พ is as in (4.2), and ๐ต โ = ๐ผ With this pair introduce a matrix polynomial
0 โ
โ
โ
) 0 .
๐ (๐ง) = ๐ผ โ ๐ง๐พ1 โ ๐ง 2 ๐พ2 โ โ
โ
โ
โ ๐ง ๐ ๐พ๐ . (Note that this matrix polynomial is slightly di๏ฌerent from the orthogonal matrix polynomial ๐ introduced earlier.) Let ๐ be a solution to the algebraic Riccati equation (6.1) ๐ = ๐พ๐๐พ โ โ ๐พ๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 ๐ต โ ๐๐พ โ , and denote ๐ = (๐๐๐ )๐๐,๐=1 , where each block ๐๐๐ is of size ๐ ร ๐. We see that ๐
+ ๐ต โ ๐๐ต = ๐
+ ๐11 , and ๐พ๐๐ต = ๐พcol (๐๐1 )๐๐=1 = col (๐พ๐ ๐11 + ๐๐+1 1 ), = 0. Introduce also the matrix polynomial ) ( ๐ฟ(๐ง) = ๐ (๐ง) + ๐ง๐ผ ๐ง 2 ๐ผ โ
โ
โ
๐ง ๐ ๐ผ ๐พ๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 .
where ๐๐+1
1
Using the observations above, the polynomial ๐ฟ becomes ๐ฟ(๐ง) = ๐ (๐ง) +
๐ โ
๐ง ๐ (๐พ๐ ๐11 + ๐๐+1 1 )(๐
+ ๐11 )โ1 .
๐=1
Observe that ๐ฟ is a comonic polynomial. Let us write ๐ฟ(๐ง) = ๐ (๐ง) + ๐ฟ1 (๐ง), and put ๐ฟ2 (๐ง) =
๐ โ
๐ง ๐ (๐พ๐ ๐11 + ๐๐+1 1 ) = ๐ฟ1 (๐ง)(๐
+ ๐11 ).
๐=1
( ) Introduce the function ๐ (๐ง) = ๐ง๐ผ ๐ง 2 ๐ผ โ
โ
โ
๐ง ๐ ๐ผ . Then we have the following expressions for the polynomials ๐, ๐ฟ and ๐ฟ2 ๐ (๐ง) = ๐ผ โ ๐ (๐ง)๐พ๐ต,
(6.2)
๐ฟ2 (๐ง) = ๐ (๐ง)๐พ๐๐ต,
(6.3) โ
โ1
๐ฟ(๐ง) = ๐ (๐ง) + ๐ (๐ง)๐พ๐๐ต(๐
+ ๐ต ๐๐ต)
.
(6.4)
With these notations, the following theorem describes a one-to-one correspondence between solutions and polynomial factorizations.
Discrete Riccati Equations and Block Toeplitz Matrices
507
Theorem 6.1. Let (๐พ โ , ๐ต) be a monic pair. 1. If ๐ is a solution to the algebraic Riccati equation ๐ = ๐พ๐๐พ โ โ ๐พ๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 ๐ต โ ๐๐พ โ then, with ๐ (๐ง) given by (6.2) and with ๐ฟ(๐ง) the comonic polynomial given by (6.4), we have ๐ (๐ง)๐
๐ (ยฏ ๐ง โ1)โ = ๐ฟ(๐ง)(๐
+ ๐11 )๐ฟ(ยฏ ๐ง โ1 )โ .
(6.5)
2. Conversely, if we have a factorization (6.5) of ๐ (๐ง)๐
๐ (ยฏ ๐ง โ1 )โ , where ๐ฟ is a comonic polynomial, then there is a solution ๐ of the algebraic Riccati equation such that ๐ฟ(๐ง) is given by (6.4). Proof. We ๏ฌrst prove part 1. Suppose that ๐ solves the algebraic Riccati equation, then we have to prove that (6.5) holds. We ๏ฌrst rewrite this as follows: ๐ (๐ง)๐
๐ (ยฏ ๐ง โ1 )โ = (๐ (๐ง) + ๐ฟ2 (๐ง)(๐
+ ๐11 )โ1 )(๐
+ ๐11 )(๐ (ยฏ ๐ง โ1 )โ + (๐
+ ๐11 )โ1 ๐ฟ2 (ยฏ ๐ง โ1 )โ ). Expanding on the right-hand side, and cancelling the terms ๐ (๐ง)๐
๐ (ยฏ ๐ง โ1 )โ which appear on both sides, we see that (6.5) is equivalent to ๐ (๐ง)๐11 ๐ (ยฏ ๐ง โ1)โ + ๐ฟ2 (๐ง)๐ (ยฏ ๐ง โ1)โ + ๐ (๐ง)๐ฟ2 (ยฏ ๐ง โ1 )โ + ๐ฟ2 (๐ง)(๐
+ ๐11 )โ1 ๐ฟ2 (ยฏ ๐ง โ1 )โ = 0. Now to show that his holds as a consequence of ๐ being a solution of the algebraic Riccati equation, consider ๐ง โ1 )โ = 0. ๐ (๐ง)(๐ โ ๐พ๐๐พ โ + ๐พ๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 ๐ต โ ๐๐พ โ )๐ (ยฏ Using (6.3) we see that ๐ง โ1 )โ = ๐ (๐ง)๐พ๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 ๐ต โ ๐๐พ โ ๐ (ยฏ ๐ง โ1 )โ . ๐ฟ2 (๐ง)(๐
+ ๐11 )โ1 ๐ฟ2 (ยฏ So, it remains to prove that ๐ (๐ง)(๐ โ ๐พ๐๐พ โ )๐ (ยฏ ๐ง โ1 )โ โ1 โ
๐ง = ๐ (๐ง)๐11 ๐ (ยฏ
(6.6) โ1 โ
) + ๐ฟ2 (๐ง)๐ (ยฏ ๐ง
โ1 โ
) + ๐ (๐ง)๐ฟ2 (ยฏ ๐ง
) .
(6.7)
We compute ๐ (๐ง)๐11 ๐ (ยฏ ๐ง โ1 )โ = ๐11 โ ๐ (๐ง)๐พ๐ต๐11 โ ๐11 ๐ต โ ๐พ โ ๐ (ยฏ ๐ง โ1 )โ โ ๐ (๐ง)๐พ๐ต๐11 ๐ต โ ๐พ โ ๐ (ยฏ ๐ง โ1 )โ , and ๐ง โ1)โ = ๐ (๐ง)๐พ๐๐ต โ ๐ (๐ง)๐พ๐๐ต๐ต โ ๐พ โ ๐ (ยฏ ๐ง โ1 )โ . ๐ฟ2 (๐ง)๐ (ยฏ
508
L. Lerer and A.C.M. Ran
The expression (6.7) becomes ๐11 โ ๐ (๐ง)๐พ๐ต๐11 โ ๐11 ๐ต โ ๐พ โ ๐ (ยฏ ๐ง โ1 )โ โ ๐ (๐ง)๐พ๐ต๐11 ๐ต โ ๐พ โ ๐ (ยฏ ๐ง โ1 )โ + ๐ (๐ง)๐พ๐๐ต โ ๐ (๐ง)๐พ๐๐ต๐ต โ ๐พ โ ๐ (ยฏ ๐ง โ1 )โ + ๐ต โ ๐๐พ โ ๐ (ยฏ ๐ง โ1 )โ โ ๐ (๐ง)๐พ๐ต๐ต โ ๐๐พ โ ๐ (ยฏ ๐ง โ1 )โ Now we use the fact that ๐11 = ๐ต โ ๐๐ต, โ 0 ๐ผ โ .. โ. 0 โ โ .. ๐ = โ. โ โ. โ .. 0
...
(6.8)
as well as ๐พ = ๐ + ๐พ๐ต๐ต โ , where โ 0 ... 0 .. โ .. . .โ โ โ .. .. . . 0โ โ โ 0 ๐ผโ ...
...
0
Combining terms in (6.8) we then have ๐11 + ๐ (๐ง)(๐พ๐๐ต โ ๐พ๐ต๐ต โ ๐๐ต) + (๐ต โ ๐๐พ โ โ ๐ต โ ๐๐ต๐ต โ ๐พ โ )๐ (ยฏ ๐ง โ1 )โ + ๐ (๐ง)(๐พ๐ต๐ต โ ๐๐ต๐ต โ ๐พ โ โ ๐พ๐๐ต๐ต โ ๐พ โ โ ๐พ๐ต๐ต โ ๐๐พ โ )๐ (ยฏ ๐ง โ1 )โ = ๐11 + ๐ (๐ง)๐๐๐ต + ๐ต โ ๐๐ โ ๐ (ยฏ ๐ง โ1 )โ + ๐ (๐ง) ((๐พ๐ต๐ต โ โ ๐พ)๐(๐ต๐ต โ ๐พ โ โ ๐พ โ ) โ ๐พ๐๐พ โ )) ๐ (ยฏ ๐ง โ1 )โ = ๐11 ๐ (๐ง)๐๐๐ต + ๐ต โ ๐๐ โ ๐ (ยฏ ๐ง โ1 )โ + ๐ (๐ง)(โ๐)๐(โ๐ โ )๐ (ยฏ ๐ง โ1 )โ โ ๐ (๐ง)๐พ๐๐พ โ๐ (ยฏ ๐ง โ1 )โ = ๐11 + (๐ (๐ง)๐ + ๐ต โ )๐(๐ต + ๐ โ ๐ (ยฏ ๐ง โ1 )โ ) โ ๐ต โ ๐๐ต โ ๐ (๐ง)๐พ๐๐พ โ๐ (ยฏ ๐ง โ1 )โ = (๐ (๐ง)๐ + ๐ต โ )๐(๐ต + ๐ โ ๐ (ยฏ ๐ง โ1 )โ ) โ ๐ (๐ง)๐พ๐๐พ โ ๐ (ยฏ ๐ง โ1 )โ . Now we note that
( ๐ (๐ง)๐ + ๐ต โ = 0
๐ง๐ผ
โ
โ
โ
) ( ๐ง ๐โ1 ๐ผ + ๐ผ
0 โ
โ
โ
(6.9)
) 1 0 = ๐ (๐ง). ๐ง
So, (๐ (๐ง)๐ + ๐ต โ )๐(๐ต + ๐ โ ๐ (ยฏ ๐ง โ1 )โ ) =
1 ๐ (๐ง)๐๐ง๐ (ยฏ ๐ง โ1 )โ = ๐ (๐ง)๐๐ (ยฏ ๐ง โ1 )โ . ๐ง
Thus, we see that (6.9) equals ๐ (๐ง)(๐ โ ๐พ๐๐พ โ)๐ (ยฏ ๐ง โ1 )โ as desired. We now prove part 2. Suppose that (6.5) holds for some comonic polynomial ๐ฟ and some Hermitian ๐11 for which ๐
+ ๐11 is invertible. Write ๐ฟ(๐ง) = ๐ผ +
Discrete Riccati Equations and Block Toeplitz Matrices โ๐
๐=1
๐ง ๐ ๐ฟ๐ . Then de๏ฌne for ๐ = 1, . . . , ๐ โ 1 the matrices ๐๐+1
1
509
by solving
๐ฟ๐ = โ๐พ๐ + (๐พ๐ ๐11 + ๐๐+1 1 )(๐
+ ๐11 )โ1 , that is, ๐๐+1
1
= (๐ฟ๐ + ๐พ๐ )(๐
+ ๐11 ) โ ๐พ๐ ๐11 = ๐ฟ๐ (๐
+ ๐11 ) + ๐พ๐ ๐
.
โ This determines the ๏ฌrst column of ๐. Put ๐1๐ = ๐๐1 . Next, de๏ฌne ๐๐๐ for ๐, ๐ > 1 recursively from the following equation, where we take ๐๐ ๐+1 = 0 for all ๐ and ๐๐+1 ๐ = 0 for all ๐,
๐๐๐ โ ๐๐+1
๐+1
= ๐พ ๐ ๐1
๐+1
+ ๐๐+1 1 ๐พ๐โ + ๐พ๐ ๐11 ๐พ๐โ
+ (๐พ๐ ๐11 + ๐1
๐+1 )(๐
โ1
+ ๐11 )
(๐11 ๐พ๐โ
(6.10) + ๐๐+1 1 ).
To show that it follows that ๐ satis๏ฌes the algebraic Riccati equation, use again that ๐พ = ๐ + ๐พ๐ต๐ต โ , and rewrite the algebraic Riccati equation as follows: ๐ โ ๐๐๐ โ = ๐พ๐ต๐ต โ ๐๐ โ + ๐๐๐ต๐ต โ ๐พ โ + ๐พ๐ต๐11 ๐ต โ ๐พ โ โ ๐พ๐๐ต(๐
+ ๐11 )โ1 ๐ต โ ๐๐พ โ . The (๐, ๐)-entry of the left-hand side is given by ๐๐๐ โ๐๐+1 ๐+1 , with the agreement that all entries with a column or row index ๐ + 1 are zero. One checks directly that the right-hand side only depends on the entries in the ๏ฌrst column and row of ๐, and moreover, the (๐, ๐)-entry of the right-hand side is equal to the (๐, ๐)-entry of the right-hand side of (6.10). In other words, (6.10) is equivalent to the algebraic Riccati equation. โก Because of the proof of part 2 of the previous theorem, any solution ๐ of the Riccati equation (6.1) with coe๏ฌcients as discussed in the beginning of this section is completely determined by its ๏ฌrst block column. Recalling that the invertible solution is the inverse of a Hermitian block Toeplitz matrix, that is no surprise for the invertible solution, because of the Gohberg-Heinig-Semencul formulas, see, e.g., [11]. Surprisingly enough, this holds for the non-invertible solutions as well. The theorem above can be viewed as a corollary of more general results connecting the solutions of a more general algebraic Riccati equation with factorizations of the so-called Popov function. Below we shall explain how that connection may be used to derive the theorem above. However, here we preferred to give an independent proof, which seems to us more transparent for this case, and also gives additional information as seen from the previous paragraph. Intimately connected to the discrete algebraic Riccati equation ๐ = ๐ด๐๐ดโ + ๐ โ ๐ด๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 ๐ต โ ๐๐ดโ is the so-called Popov function, given by ๐บ(๐ง) = ๐
+ ๐ต โ (๐ง โ1 โ ๐ด)โ1 ๐(๐ง โ ๐ดโ )โ1 ๐ต.
510
L. Lerer and A.C.M. Ran
See, e.g., [14], [13], [16]. In particular, there is a one-to-one correspondence between solutions ๐ and certain symmetric factorizations of ๐บ, which is described as follows: let ๐ be a solution to the discrete algebraic Riccati equation, and put ๐
(๐ง) = ๐ผ + ๐ต โ (๐ง โ1 โ ๐ด)โ1 ๐ด๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 , then
๐บ(๐ง) = ๐
(๐ง)(๐
+ ๐ต โ ๐๐ต)๐
(ยฏ ๐ง โ1 )โ . Note that when ๐ = 0 the Popov function reduces to a constant matrix function. For the particular case where the pair (๐ด, ๐ต) is a monic pair, we may take ( ) ๐ด = ๐พ, ๐ตโ = ๐ผ 0 โ
โ
โ
0 , where ๐พ is given as by (4.2). Computing ๐ต โ (๐ง โ1 โ ๐ด)โ1 = ๐ง๐ต โ (๐ผ โ ๐ง๐ด)โ1 , we see that this is equal to
( ๐ (๐ง)โ1 ๐ง๐ผ
๐ง 2๐ผ
โ
โ
โ
) ๐ง ๐๐ผ ,
where
๐ (๐ง) = ๐ผ โ ๐ง๐พ1 โ ๐ง 2 ๐พ2 โ โ
โ
โ
โ ๐ง ๐ ๐พ๐ . The Popov function then may be replaced by the polynomial โ1 โ ๐ผ ๐ง โ 12 ๐ผ โ ( ) โ๐ง โ ๐ (๐ง)๐บ(๐ง)๐ (ยฏ ๐ง โ1 )โ = ๐ง๐ผ ๐ง 2 ๐ผ โ
โ
โ
๐ง ๐๐ผ ๐ โ . โ + ๐ (๐ง)๐
๐ (ยฏ ๐ง โ1 )โ . โ .. โ ๐ง๐๐ผ
Now for the particular case where ๐ = 0, the latter function still is a nontrivial polynomial. The factors corresponding to solutions ๐ are given by ๐ฟ(๐ง) = ๐ (๐ง)๐
(๐ง), where ๐
(๐ง) is as above, so by ( ) ๐ฟ(๐ง) = ๐ (๐ง) + ๐ง๐ผ ๐ง 2 ๐ผ โ
โ
โ
๐ง ๐ ๐ผ ๐พ๐๐ต(๐
+ ๐ต โ ๐๐ต)โ1 . Thus, the polynomial ๐ฟ is exactly the comonic polynomial from Theorem 6.1.
References [1] F.A. Aliev, B.A. Bordyug and V.B. Larin. Discrete generalized Riccati equations and polynomial matrix factorization. Systems and Control Letters 18 (1992), 49โ59. [2] H. Bart, I. Gohberg, M.A. Kaashoek, The coupling method for solving integral equations. In: Topics in operator theory and networks, the Rehovot workshop, OT 12, Birkhยจ auser Verlag, Basel, 1984, pp. 39โ74. [3] C.T. Chen. A generalization of the inertia theorem. SIAM J. Appl. Math. 25 (1973), 158โ161. [4] D.J. Clements, H.K. Wimmer. Existence and uniqueness of unmixed solutions of the discrete-time algebraic Riccati equation. Systems and Control Letters 50 (2003), 343โ346.
Discrete Riccati Equations and Block Toeplitz Matrices
511
[5] A. Ferrante. On the structure of the solutions of discrete-time algebraic Riccati equation with singular closed-loop matrix. IEEE Trans. Automat. Control 49 (2004), 2049โ2054. [6] A. Ferrante, M. Pavon and S. Pinzoni. Asymmetric algebraic Riccati equation: a homeomorphic parametrization of the set of solutions. Linear Algebra and Appl. 329 (2001), 137โ156. [7] A. Ferrante and H.K. Wimmer. Order reduction of discrete-time algebraic Riccati equations with with singular closed-loop matrix. Oper. Matrices 1 (2007), 61โ70. [8] A.E. Frazho, M.A. Kaashoek, A.C.M. Ran. The non-symmetric discrete algebraic Riccati equation and canonical factorization of rational matrix functions on the unit circle. Integral Equations and Operator Theory. 66 (2010), 215โ229. [9] I. Gohberg, P. Lancaster, L. Rodman. Matrix Polynomials, Academic Press, 1982. [10] I. Gohberg, P. Lancaster, L. Rodman. Invariant subspaces of matrices with Applications, John Wiley & Sons, New York, 1986. [11] I. Gohberg and L. Lerer. Matrix generalizations of M.G. Krein theorems on orthogonal polynomials. In Orthogonal Matrix-valued Polynomials and Applications (ed. I. Gohberg) OT 34, Birkhยจ auser Verlag, 1988, 137โ202. [12] C. Heij, A.C.M. Ran, F. van Schagen, Introduction to Mathematical Systems Theory. Birkhยจ auser Verlag, Basel, 2007. [13] V. Ionescu, C. Oara and M. Weiss. Generalized Riccati Theory and robust control. A Popov function approach. John Wiley, Chichester, 1999. [14] I. Karelin, L. Lerer and A.C.M. Ran. ๐ฝ-symmetric factorizations and algebraic Riccati equations. In Proceedings of the IWOTA 1998 (eds. A. Dijksma, M.A. Kaashoek, A.C.M. Ran), OT 124 Birkhยจ auser Verlag, 2001, 319โ360. [15] M.G. Krein. Distribution of roots of polynomials orthogonal on the unit circle with respect to a sign alternating weight. Theor. Funkcii Funkcional Anal. i. Prilozen. 2 (1966), 131โ137 (Russian). [16] P. Lancaster and L. Rodman. Algebraic Riccati Equations Oxford, UK: Clarendon Press, 1995. [17] P. Lancaster and M. Tismenetsky. The Theory of Matrices. 2nd Edition. Academic Press, San Diego etc. 1985. [18] H. Langer, A.C.M. Ran, D. Temme. Inertia of Hermitian solutions of the algebraic Riccati equation. In: Proceedings of the European Control Conference, 1997. [19] V.B. Larin. The generalized Riccati equations, orthogonal projectors and factorization of matrix polynomials. Soviet J. Automat. Inform. Sci. 22 (1989), 72-77. Translated from Automatika 1989, no. 6, 70โ74, 96. [20] L. Lerer, A.C.M. Ran. A new inertia theorem for Stein equations, inertia of invertible hermitian block Toeplitz matrices and matrix orthogonal polynomials. Integral Equations and Operator Theory 47 (2003), 339โ360. [21] A.C.M. Ran and H.L. Trentelman. Linear quadratic problems with inde๏ฌnite cost for discrete time systems. SIAM J. Matrix Anal. Appl. 14 (1993), 776โ797. [22] M.A. Shayman. Geometry of the algebraic Riccati equations. I, II. SIAM J. Control 21 (1983), 375โ394 and 395โ409.
512
L. Lerer and A.C.M. Ran
[23] J.C. Willems. Least squares stationary optimal control and the algebraic Riccati equation. IEEE Trans. Automatic Control AC-16 (1971), 621โ634. [24] H.K. Wimmer. Inertia theorems for matrices, controllability, and linear vibrations. Linear Algebra and Appl.8 (1974), 337โ343. [25] H.K. Wimmer. Unmixed solutions of the discrete-time algebraic Riccati equation. SIAM J. Control Optim. 30 (1992), 867โ878. [26] H.K. Wimmer. Hermitian solutions of the discrete-time algebraic Riccati equation. Internat. J. Control 63 (1996), 921โ936. [27] H.K. Wimmer. A parametrization of solutions of the discrete-time algebraic Riccati equation based on pairs of opposite unmixed solutions. SIAM J. Control Optim. 44 (2006), 1992โ2005. Leonid Lerer Department of Mathematics Technion-Israel Institute of Technology 3200 Haifa, Israel e-mail:
[email protected] Andrยดe C.M. Ran Department of Mathematics Faculteit of Exact Sciences Vrije Universiteit Amsterdam De Boelelaan 1081a NL-1081 HV Amsterdam, The Netherlands e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 513โ539 c 2012 Springer Basel AG โ
On Cyclic and Nearly Cyclic Multiagent Interactions in the Plane Frยดedยดerique Oggier and Alfred Bruckstein This paper is dedicated to the memory of Professor Israel Gohberg, who was a great mathematician truly interested in engineering applications, and a wonderful person.
Abstract. Cyclic pursuit and local averaging interactions have been extensively analyzed in the context of multiagent gathering, in the ๏ฌeld of distributed robotics. This paper reviews some results on cyclically structured dynamical systems, and discusses their application to nearly cyclic interactions among ๐ point-agents in the plane, leading to formations of interesting limiting geometric con๏ฌgurations. In particular, we consider evolutions that can be modeled by a Toeplitz operator, and explain how they can be decoupled into independent evolving modes, focusing on nearly cyclic interactions that break symmetry leading to factor circulants rather than circulant interaction matrices. Mathematics Subject Classi๏ฌcation (2000). 15A18, 47B35, 68T40. Keywords. Circulant matrices, ๐-circulant matrices, cyclic pursuit, multiagent interaction.
1. Introduction Consider a โswarmโ or โpackโ of ๐ robots in the plane, denoted by ๐ซ0 , ๐ซ1 , . . . , ๐ซ๐ โ1 which can all see each other and are aware of the other robotโs identities (i.e., can distinguish them). We shall de๏ฌne the rules of interaction specifying how each robot ๐ซ๐ moves in response to the (evolution in time of the) con๏ฌguration of the entire swarm. Therefore denoting ๐ซ๐ โs location at time ๐ก to be ๐ซ๐ (๐ก) = ๐ฅ๐ (๐ก)+๐๐ฆ๐ (๐ก) (a complex number), we assume that we can write the swarm evolution equations as follows: ๐๐ซ๐ (๐ก) (๐ถ) = ฮฆ๐ {๐ซ๐ (๐)โฃ๐ =0,1,...,๐ โ1 ; ๐ โค ๐ก} ๐๐ก (๐ท) or ๐ซ๐ (๐ก + 1) = ฮฆ๐ {๐ซ๐ (๐)โฃ๐ =0,1,...,๐ โ1 ; ๐ โค ๐ก} (1) depending on whether the temporal evolution is continuous (๐ถ) or discrete (๐ท). So far the ฮฆ-operators are not speci๏ฌed, and in fact they could be quite involved in
514
F. Oggier and A. Bruckstein (๐ถ)
general. The operator ฮฆ๐ {โ
} provides an instantaneous velocity vector for agent (๐ท) ๐ซ๐ in response to the locations of the other agents in the swarm, while ฮฆ๐ {โ
} will yield the next location for ๐ซ๐ in a synchronous discrete timed evolution. Both the discrete and the continuous operators should produce the same evolutions if we decide to look at the agents in a di๏ฌerent frame of reference, i.e., re-encode their locations using arbitrarily rotated and possibly uniformly scaled coordinates, hence the resulting evolution equations must be similarity invariant. This requirement clearly imposes restrictions on the ฮฆ operators, and these will be discussed in the sequel (see (6)). Linear memoryless operators are an important class of (discrete or continuous) operators which have the form ฮฆ๐ {๐ซ0 , ๐ซ1 , . . . , ๐ซ๐โ1 } =
๐ โ1 โ
๐๐๐ (๐ก)๐ซ๐ (๐ก)
๐=0
๐๐๐ (๐ก)
where are some (complex) numbers, varying perhaps in time, but which do not depend on previous swarm con๏ฌgurations. In this case, Equation (1) describes a linear (generally time varying) systemโs state evolution, and there is a wealth of theory dealing with such systems in the control and signal processing literature. Here we shall mainly be concerned with a special class of (constant) linear Toeplitz operators of the form ฮฆ๐ {๐ซ0 , ๐ซ1 , . . . , ๐ซ๐ โ1 } =
๐ โ1 โ
๐Ind[(๐โ๐)<0] ๐(๐โ๐)mod
๐ ๐ซ๐ (๐ก)
(2)
๐=0
where ๐ is some complex number, and { { ๐โ1 โก ๐๐ โ1 mod ๐ 1 ๐๐ and Ind[๐ < 0] = ๐โ๐ โก ๐๐ โ๐ mod ๐ 0 ๐๐
๐<0 . ๐โฅ0
Writing out explicitly ฮฆ๐ {๐ซ0 , . . . , ๐ซ๐ โ1 } for ๐ = 0, . . . , ๐ โ 1 in matrix form and denoting โก โค ๐ซ0 (๐ก) โข โฅ .. P(๐ก) = โฃ โฆ, . ๐ซ๐โ1 (๐ก)
the swarmโs evolution dynamics becomes ( ) ๐ P(๐ก) or P(๐ก + 1) = ฮฆP(๐ก) ๐๐ก โก ๐0 ๐1 โข ๐๐๐โ1 ๐0 โข โข = โข ๐๐๐โ2 ๐๐๐ โ1 โข .. .. โฃ . . ๐๐1
๐๐2
(3) ๐2 ๐1 ๐0 ... ...
... ... ... .. . ๐๐๐ โ1
๐๐ โ1 ๐๐ โ2 ... ... ๐0
โค โฅ โฅ โฅ โฅ P(๐ก). โฅ โฆ
On Cyclic and Nearly Cyclic Multiagent Interactions
515
Note that if ๐ = 1, the matrix is a special Toeplitz-circulant matrix, since by โ1 de๏ฌnition, a Toeplitz-circulant (or simply circulant) ๐ ร๐ matrix M = [๐๐,๐ ]๐ ๐,๐=0 is obtained by cyclic shift of its ๏ฌrst row (or equivalently, of its ๏ฌrst column): ๐๐,๐ = ๐๐โ1mod๐,๐โ1mod๐ . Otherwise it is a generalization of a circulant called a ๐-factor, or ๐-circulant matrix, where the ๐ ร ๐ matrix M is similarly de๏ฌned by cyclic shifts of its ๏ฌrst row (or column), up to a factor ๐: { ๐๐โ1,๐โ1 0 โค ๐ โค ๐ โ 1, ๐ โค ๐ โค ๐ โ 1 ๐๐,๐ = ๐๐๐โ1mod๐,๐โ1mod๐ else. Such matrices arise in several applications, such as linear systems theory [8, 10], linear algebra [1], geometry [5, 13, 15], and in connection with inverses of Toeplitz matrices [7, 9, 4], coding theory [6] and linear systems of di๏ฌerential equations [18]. In case of ๐ = 1, i.e., when the operator ฮฆ is Toeplitz-circulant, we have that all the robotic agents perform โcyclicallyโ the same operation, i.e., agent ๐ซ๐ will determine its next location (or its velocity) according to the same weighted average performed on ๐ซ๐ , ๐ซ๐+1 , . . . , ๐ซ(๐+๐ )mod๐ (in this order), i.e., โก โค ๐ซ๐ (๐ก) { } โข ๐ซ๐+1 (๐ก) โฅ ๐ซ๐ (๐ก + 1) โข โฅ = [๐ , ๐ , . . . , ๐ ] (4) โข โฅ . 0 1 ๐ โ1 ๐ or ๐๐ก ๐ซ๐ (๐ก) โฃ .. โฆ ๐ซ(๐+๐)mod
which can be rewritten as โก
0
0
0
1
โข โข โข โข { } โข ๐ซ๐ (๐ก + 1) โข = ๐ ยฏ โข ๐ or ๐๐ก ๐ซ๐ (๐ก) โข 1 โข โข 1 โข โฃ 1 0 ...
1
(๐th place)
1
...
... 1
... 0
0
๐ (๐ก)
โค
โฅ โฅโก โฅ โฅ โข .. โฅ โข . โฅ โฅโข โฅโฃ โฅ โฅ โฅ 0 โฆ 0
๐ซ0 (๐ก) ๐ซ1 (๐ก) .. .
โก โข โข โข Zโโข โข โข โฃ
0 .. . .. . 0 1
1 0
0
1 0
0 โ
โ
โ
โฅ โฅ โฅ โฆ
๐ซ๐ โ1 (๐ก)
= ๐Z ยฏ ๐โ1 โ
P(๐ก) where
โค
(5) โค
โฅ โฅ โฅ โฅ and ๐ ยฏ = [๐0 , ๐1 , . . . , ๐๐ โ1 ]. โฅ 1 โฅ 0 1 โฆ 0
516
F. Oggier and A. Bruckstein
This special case, with a circulant matrix ฮฆ, was extensively analyzed before in the context of polygon smoothing evolutions and cyclic pursuits for robotic gathering and formation control, see, e.g., [5, 13, 15, 3, 2, 12, 11, 7]. Note that invariance requirements impose some conditions on the linear evolution operators, as we now discuss. If P(๐ก) is described by the evolution equations ๐ P(๐ก) = ฮฆ(๐ถ) P(๐ก) ๐๐ก or P(๐ก + 1) = ฮฆ(๐ท) P(๐ก) from some initial location P(0) = P(๐ก = 0), and if we re-encode the agentsโ positions via a general similarity transformation of the form Pโฒ (๐ก) โ ๐P(๐ก) + ๐ 1
(6)
where ๐ and ๐ are some complex numbers and 1 = [1, . . . , 1]๐ , we shall have for Pโฒ (๐ก): โ in the continuous case ๐ โฒ ๐ P (๐ก) โ (๐P(๐ก) + ๐ 1) ๐๐ก ๐๐ก ๐ = ๐ P(๐ก) ๐๐ก = ๐ฮฆ(๐ถ) P(๐ก) which is equal to ฮฆ(๐ถ) (๐P(๐ก) + ๐ 1) only if ฮฆ(๐ถ) 1 = 0. โ in the discrete case Pโฒ (๐ก + 1) โ ๐P(๐ก + 1) + ๐ 1 = ๐ฮฆ(๐ท) P(๐ก) + ๐ 1 which is equal to ฮฆ(๐ท) (๐P(๐ก) + ๐ 1) only if ฮฆ(๐ท) 1 = 1. Hence the ฮฆ-matrices that describe linear, time-invariant evolutions need to obey the conditions ฮฆ(๐ถ) 1 = 0 or ฮฆ(๐ท) 1 = 1 in order to have Euclidean or similarity invariant evolutions. In some of our examples, these conditions cannot be satis๏ฌed. However, note that any ๐ ร ๐ matrix ฮฆ may be embedded in an (๐ + 1) ร (๐ + 1) matrix ฮฆ as follows โก โค [ ] 1 [ ] ฮฆ s โข . โฅ ฮฆ1 + s . = โฃ . โฆ 0 ๐ง ๐ง 1 and selecting either ๐ง = 0 and s = โฮฆ1 or ๐ง = 1 and s = โฮฆ1 + 1, we obtain a ฮฆ matrix that describes an invariant evolution of a multi-agent system with an addi๐ tional agent ๐ซ๐ต whose position is stationary ( ๐๐ก ๐ซ๐ต = 0 or ๐ซ๐ต (๐ก+1) = ๐ซ๐ต (๐ก)). This additional agent will act as a โbeaconโ or a set reference point, for the description of the swarm of agents. In this case, setting ๐ซ๐ต = (0, 0), the evolution of the rest of
On Cyclic and Nearly Cyclic Multiagent Interactions
517
the agents will be described by the original matrix ฮฆ. Note that the spatial location of the ๏ฌxed ๐ซ๐ต in the plane may be determined according to the initial location of the agents of the swarm. A good example is the geometric and a๏ฌne invariant decision that can be made by each agent independently to set ๐ซ๐ต , and hence the origin of its Cartesian coordinate system, at the centroid of the agent location constellation at ๐ก = 0. This will make the swarm evolution entirely autonomous. However, an external setting of the location of ๐ซ๐ต might be useful in controlling the swarm and steering it toward a desired place in the environment. One might even desire to move ๐ซ๐ต in time and make the swarm move accordingly, by tracking the beacon point in addition to its own internal dynamics controlled by ฮฆ.
2. Analyzing swarm evolution via mode decoupling Circulant, and ๐-factor circulant matrices have very special structures and this allows us to diagonalize them, essentially by Fourier transform methods. Let us see, in general, how diagonalization yields a way to analyze the evolution of the constellation of robots by decoupling it into independently evolving modes. Indeed assume that the time-invariant matrix ฮฆ can be diagonalized (for example when ฮฆ has distinct eigenvalues, or ฮฆ is normal, i.e., ฮฆโ ฮฆ = ฮฆฮฆโ , hence having a full set of orthonormal eigenvectors), as follows ฮฆ = ๐ โ1 ๐ท๐ where ๐ท = Diag[๐0 , ๐1 . . . ๐๐ โ1 ] displays the eigenvalues of ฮฆ and the columns of ๐ โ1 are the (right) eigenvectors. Now we have that โซ P(๐ก + 1) โฌ or = ๐ โ1 ๐ท๐ P(๐ก) โญ ๐ P(๐ก) ๐๐ก and hence
} ๐ P(๐ก + 1) = ๐ท(๐ P(๐ก)). ๐ (๐ P(๐ก)) ๐๐ก ห In terms of the transformed vector P(๐ก) โ ๐ P(๐ก), the evolution is a decoupled evolution controlled explicitly by the (constant) eigenvalues [10]. Indeed, we have โก ๐ก โค ๐๐ โข โฅ ๐๐ก1 0 โข โฅห ห P(๐ก) =โข โฅ P(0) .. โฃ โฆ . ๐๐ก๐ โ1
0 or
โก โข โข ห P(๐ก) =โข โฃ
๐๐0 ๐ก
(discrete case)
โค ๐๐1 ๐ก 0
0 .. .
โฅ โฅห โฅ P(0) โฆ ๐๐๐ ๐ก
(continous case)
.
518
F. Oggier and A. Bruckstein
Therefore diagonalization enables the explicit solution of the swarm evolution, in the case the ฮฆ matrix is time invariant and has a full set of orthonormal eigenvectors. As we shall see below, ๐-factor circulants are a family of matrices that enable both a nice physical interpretation in terms of cyclic and symmetric interactions among similar agents and an explicit diagonalization via discrete Fourier transform matrices.
3. Diagonalization of factor circulants Factor circulant matrices are very special in that they provide explicit formulae for the diagonalizing transforms and for their eigenvalues. This enables us to analyze in detail the behavior of multiagent interactions when these are cyclic or โnearlyโ cyclic, and fully describe the limiting behaviors of the swarm. For circulants, we have the following results. Consider the unitary Fourier transform matrix โก 0 โค ๐ค ๐ค0 . . . ๐ค0 0 1 โฅ . . . ๐ค๐ โ1 1 โข โข ๐ค ๐ค โฅ [FT] โ โ โข . โฅ . . .. .. โฆ ๐ โฃ .. ๐ค0 ๐ค๐ โ1 . . . ๐ค(๐ โ1)(๐ โ1) [ ] 1 = โ ๐ค(๐โ1)(๐โ1) ๐,๐=1,...,๐ ๐
2๐
where ๐ค = ๐โ๐ ๐ is an ๐ th root of unity. Then C is a Toeplitz-circulant matrix if and only if โก โค ๐๐ โข โฅ ๐1 0 โข โฅ C[FT] = [FT] โข โฅ .. โฃ โฆ . 0 ๐๐ โ1 where ๐0 , ๐1 , . . . , ๐๐โ1 are the eigenvalues of C and are given by ๐๐ =
๐ โ1 โ
2๐
๐๐ ๐โ๐ ๐ ๐๐ .
๐=0
Hence and
[FT]โ C[FT] = Diag[๐0 , ๐1 , . . . , ๐๐ โ1 ] C = [FT]Diag[๐0 , ๐1 , . . . , ๐๐ โ1 ][FT]โ .
To summarize the remarkable properties of circulants, we can state that they are (1) diagonalized by the discrete Fourier Transform, (2) they all commute, (3) their products are circulants, (4) their sums are circulants too, and (5) their inverses/pseudoinverses are circulants, and are readily found [9]. In fact, many of the wonders of modern signal processing algorithms, and linear, time invariant systems theory stem from the above properties.
On Cyclic and Nearly Cyclic Multiagent Interactions
519
The corresponding, and equally remarkable properties of ๐-circulants are, however, much less known and applied. Suppose we consider the following operation on a circulant C = C[๐0 ,๐1 ,...,๐๐ โ1 ] : โค โก โค โก ๐๐ ๐๐ โฅ โข โฅ โข ๐1 0 ๐1 0 โฅ โข โฅ โข C W=โข โฅ โข โฅ, [๐ ,๐ ,...,๐ ] .. .. 0 1 ๐ โ1 โฆ โฃ โฆ โฃ . . 0
๐๐ โ1
0
๐๐ โ1
i.e., W is obtained by pre- and post multiplying C by two diagonal matrices. It is easy to see that we have โก โค ๐ 0 ๐1 . . . ๐0 ๐๐ โ1 ๐ 0 ๐0 โข ๐ 1 ๐0 โฅ ๐ 1 ๐1 . . . ๐1 ๐๐ โ1 โข โฅ W = C[๐0 ,๐1 ,...,๐๐ โ1] โ โข . โฅ= C โ M . . .. .. โฃ .. โฆ ๐๐ โ1 ๐0 ๐๐ โ1 ๐1 . . . ๐๐ โ1 ๐๐ โ1 where โ stands for the Schur Hadamard multiplication [14] (or a โmaskingโ operation) which multiplies matrices element-wise, and M โ [๐๐ ๐๐ ]๐,๐=0,...,๐โ1 . Matrices of the type W inherit interesting diagonalization properties from the original circulant C. The matrix W is a circulant matrix that is modi๏ฌed by a highly structured masking matrix M and W = Diag[๐0 , . . . , ๐๐ โ1 ][FT]Diag[๐0 , . . . , ๐๐ โ1 ][FT]โ Diag[๐0 , . . . , ๐๐ โ1 ]. However, since the masking matrix is neither circulant nor Toeplitz, we shall have to consider some special cases for the {๐0 , ๐1 , . . . , ๐๐ โ1 } and {๐0 , ๐1 , . . . , ๐๐ โ1 } sequences. First of all, note that the factorization above will be of the form โก โค ๐0 โข โฅ โ1 .. W = Uโฃ โฆU . ๐๐ โ1 if and only if (Diag[๐0 , ๐1 , . . . , ๐๐ โ1 ][FT])โ1 = [FT]โ Diag[๐0 , ๐1 , . . . , ๐๐ โ1 ] โ1 โ1 โ โโ [FT]โ Diag[๐โ1 0 , ๐1 , . . . , ๐๐ โ1 ] = [FT] Diag[๐0 , ๐1 , . . . , ๐๐ โ1 ] โ ๐๐ผ๐ or ๐๐ = ๐โ1 ๐ , and U will further be unitary if also ๐๐ = ๐๐ , implying that ๐๐ = ๐ โ๐๐ผ๐ โ and ๐๐ = ๐ = ๐๐ . In this case the masking-matrix multiplying C will be [๐๐๐ผ๐ ๐โ๐๐ผ๐ ] = [๐๐(๐ผ๐ โ๐ผ๐ ) ]๐,๐=0,...,๐ โ1 . The most interesting particular cases of {๐0 , ๐1 , . . . , ๐๐ โ1 } and {๐0 , ๐1 , . . . , ๐๐ โ1 } arise when we have ๐๐ = ๐พ ๐ and ๐๐ = ๐พ โ๐ , ๐ = 0, 1 . . . , ๐ โ 1, for some real
520
F. Oggier and A. Bruckstein
or imaginary ๐พ. In this case โก โข โข โข M=โข โข โฃ
1 ๐พ ๐พ2 .. .
๐พ ๐ โ1
๐พ โ1 1 ๐พ .. .
๐พ ๐ โ2
๐พ โ2 ๐พ โ1 1 .. . ...
... ... ... .. . โก
โข โข โข = Circ[1,๐พ โ1 ,...,๐พ โ(๐ โ1) ] โ โข โข โฃ
๐พ
1 ๐พ๐ ๐พ๐ .. .
๐พ๐
๐พ โ(๐ โ1) ๐พ โ(๐ โ1)+1 ๐พ โ(๐ โ1)+2 .. .
โค โฅ โฅ โฅ โฅ โฅ โฆ
1
1 1 ๐พ๐ ๐พ๐
1 1 1
1 1 1 .. .
1 1 1
. . . ๐พ๐
โค โฅ โฅ โฅ โฅ โฅ โฆ
1
where Circ[1,๐พ โ1 ,...,๐พ โ(๐ โ1) ] is given by โก โข โข โข โข โข โฃ
1
๐พ โ(๐ โ1) ๐พ โ(๐ โ1)+1 .. .
๐พ โ(๐ โ1)+(๐ โ2)
๐พ โ1 1
๐พ โ(๐ โ1) .. .
๐พ โ2 ๐พ โ1 1 .. . ...
... ... ... 1 ๐พ โ(๐ โ1)
โค ๐พ โ(๐ โ1) ๐พ โ(๐ โ1)+1 โฅ โฅ โฅ โฅ. โฅ .. โฆ . 1
Hence the matrix W = C โ M becomes โก โข โข โข W = C[๐0 ,...,๐๐ โ1 ] โ Circ[1,๐พ โ1 ,...,๐พ โ(๐ โ1) ] โ โข โข โฃ
1 ๐พ๐ ๐พ๐ .. .
๐พ๐
1 1 ๐พ๐ ๐พ๐
1 1 1
1 1 1 .. .
. . . ๐พ๐
1 1 1
โค โฅ โฅ โฅ โฅ โฅ โฆ
1
which clearly is a ๐(= ๐พ ๐ )-circulant matrix. To summarize, we have the following result: A ๐-circulant matrix W, denoted by โก โค ๐1 ๐2 ... ๐๐ โ1 ๐0 โข ๐๐๐ โ1 ๐0 ๐1 ... ๐๐ โ2 โฅ โข โฅ โข ๐๐๐ โ2 ๐๐๐ โ1 ๐0 ... ... โฅ W=โข (7) โฅ โข โฅ .. .. .. โฃ . . . ... ... โฆ ๐๐1 ๐๐2 . . . ๐๐๐ โ1 ๐0 can be rewritten as W = Circ[๐0 ,๐1 ๐พ,๐2 ๐พ 2 ,...,๐๐ โ1 ๐พ ๐ โ1 ] โ Circ[1,๐พ โ1 ,...,๐พ โ(๐ โ1) ] โ ฮ
On Cyclic and Nearly Cyclic Multiagent Interactions with
โก โข โข โข ฮ=โข โข โฃ
1 1 ๐ 1 ๐ ๐ .. .. . . ๐ ๐
โค ... 1 ... 1 โฅ โฅ ... 1 โฅ โฅ and ๐พ ๐ = ๐ .. โฅ ๐ 1 . โฆ ... ๐ 1 1 1 1
and hence can be factorized as โก โค โก 1 ๐0 โข ๐พ โฅ โข ๐1 0 โข โฅ โข โข โฅ โข ๐พ2 ๐2 W=โข โฅ [FT] โข โข โฅ โข . .. โฃ โฆ โฃ 0 ... ๐พ ๐ โ1
521
โค
โก
โค
1
โฅ โข ๐พ โ1 โฅ โข โฅ โข ๐พ โ2 โฅ [FT]โ โข โฅ โข .. โฆ โฃ . ๐๐ โ1
โฅ โฅ โฅ โฅ โฅ โฆ ๐พ โ(๐ โ1)
where [๐0 , ๐1 , . . . , ๐๐ โ1 ] are the eigenvalues of Circ[๐0 ,๐1 ๐พ,...,๐๐ โ1 ๐พ ๐ โ1 ] โ Circ[๐0 ,๐1 ,...,๐๐ โ1 ] given by ๐๐ =
๐ โ1 โ
2๐
1
๐๐ โ
๐พ ๐ โ
๐โ๐ ๐ ๐๐ (๐พ โ ๐ ๐ ).
๐=0
Therefore W is readily diagonalized as follows โก โค โก โค 1 1 โก โค โข ๐พ โ1 โฅ โข ๐พ โฅ ๐0 โข โฅ โข โฅ โ2 2 โข .. โฅ โข โฅ โข โฅ โ ๐พ ๐พ โฅWโข โฅ [FT] โฃ โฆ = [FT] โข . โข โฅ โข โฅ .. .. โฃ โฆ โฃ โฆ . . ๐๐ โ1 ๐ โ1 โ(๐ โ1) ๐พ ๐พ = Tโ1 WT, the matrices T and Tโ1 being โก โค โก 1 1 โข โฅ โข ๐พ โข โฅ โข T=โข โฅ [FT] and Tโ1 = [FT]โ โข .. โฃ โฆ โฃ . ๐ โ1 ๐พ
โค ๐พ โ1
..
.
โฅ โฅ โฅ. โฆ ๐พ ๐ โ1
Note that T is not, in general a unitary transformation. In all developments above, we assumed ๐พ to be arbitrary. If ๐พ โ= 0 is a real number, T will be an invertible matrix, as seen before. If however ๐พ is purely imaginary, i.e., ๐พ = ๐๐๐ , then clearly ๐พ โ = ๐โ๐๐ = ๐พ โ1 and the matrix T becomes a unitary transformation, obeying TTโ = Tโ T = ๐ผ. In this case the matrix W will be ๐-factor circulant with ๐ = ๐๐๐๐ .
522
F. Oggier and A. Bruckstein
Remark. Note that (7) can be alternatively written1 as W = ๐0 ๐ผ + ๐1 ฮ + โ
โ
โ
+ ๐๐ โ1 ฮ๐ โ1 with
โก โข โข โข ฮ=โข โข โฃ
0
๐
1 0
โค 1 .. .
..
.
โฅ โ1 โฅ ๐โ โฅ e๐ e๐๐+1 + ๐e๐ e๐1 โฅ= โฅ ๐=1 1 โฆ 0
where e๐ , ๐ = 1, . . . , ๐ are column vectors that form the canonical basis of โ๐ . Thus ๐ โ1 โ W = ๐ (ฮ) โ ๐๐ ฮ ๐ , ๐=0
and by the spectral mapping theorem, the spectrum ๐(W) = ๐ (๐(ฮ)), where ๐(ฮ) is found by computing det(๐๐ผ๐ โ ฮ) = ๐๐ โ ๐. It is also easily seen that W is normal โโ ฮ is normal โโ โฃ๐โฃ = 1, telling in particular that W has a full set of orthonormal vectors as already seen above.
4. Dynamics of a cyclically interacting swarm Returning to the problem of analyzing the dynamics and the long-term behavior of a swarm of robots ๐ซ0 , ๐ซ1 , . . . , ๐ซ๐โ1 interacting according to โก โค ๐0 ๐1 ๐2 ... ๐๐ โ1 ๐0 ๐1 ... ๐๐ โ2 โฅ } โข โข ๐๐๐โ1 โฅ P(๐ก + 1) โข ๐๐๐โ2 ๐๐๐ โ1 ๐0 โฅ . . . ... = โข โฅ P(๐ก) = ฮฆP(๐ก), ๐ or ๐๐ก P(๐ก) โข โฅ .. .. . . โฃ โฆ . . . ... ... ๐๐1 ๐๐2 . . . ๐๐๐ โ1 ๐0 we have that the interaction matrix ฮฆ is ๐-circulant hence it is diagonalizable as follows: โก โค โก โค โก โค 1 1 ๐0 โ1 โข ๐พ โฅ โข ๐พ โฅ โข .. โฅ โข โฅ โข โฅ โข โฅ โ2 2 . 0 โข โฅ โข โฅ โ ๐พ ๐พ โฅ [FT] โข ฮฆ=โข โฅ [FT] โข โฅ โข โฅ . โข โฅ โข โฅ . . . โฃ 0 โฆ .. .. . โฃ โฆ โฃ โฆ ๐ โ1 โ(๐ โ1) ๐ ๐ โ1 ๐พ ๐พ 1 We
would like to thank one of our anonymous reviewers who suggested this.
On Cyclic and Nearly Cyclic Multiagent Interactions
523
1
where ๐พ = ๐ ๐ and ๐๐ =
๐โ1 โ
๐
2๐
๐๐ ๐ ๐ ๐โ๐ ๐ ๐๐ .
๐=0
Therefore de๏ฌning
โก
โข โข ห P(๐ก) โ [FT]โ โข โฃ
1
โค โฅ โฅ โฅ P(๐ก) โฆ
1 โ๐
๐
..
. ๐โ
๐ โ1 ๐
we have decoupled dynamics for the transformed location vector, given by โก โค ๐0 โซ ๐ ห P(๐ก) โฌ โข โฅ ๐1 0 ๐๐ก โข โฅห or =โข โฅ P(๐ก) . . โญ โฃ โฆ . 0 ห + 1) P(๐ก ๐๐โ1 and the evolution of the swarm is controlled by the eigenvalues ๐0 , ๐1 , . . . , ๐๐โ1 . Let us concentrate next on some speci๏ฌc cases of ๐ = [๐0 , . . . , ๐๐โ1 ] and ๐. A โ๐- cyclicโ interaction involves agents that are reacting di๏ฌerently with the agents that follow them to the agents that precede them in the ordering ๐ซ0 , . . . , ๐ซ๐ โ1 . 4.1. Darbouxโs polygon evolution and extensions As a ๏ฌrst example, suppose that we have a generalization of Darbouxโs polygon evolution process [5], which is also a nice model for cyclic pursuit: โก 1 โค 1 0 0 ... 2 2 1 โข 0 โฅ 2 โข โฅ โข โฅ . .. โข 0 โฅ โฅ P(๐ก). P(๐ก + 1) = โข โข โฅ .. โข 0 โฅ . โข โฅ 1 1 โฆ โฃ 0 2 2 ๐ 12 0 0 0 0 12 In this case, we have a ๐-factor circulant with 2๐ 1 2๐ 1 1 1 1 ๐๐ = + ๐ ๐ ๐โ๐ ๐ โ
๐ = (1 + ๐ ๐ ๐โ๐ ๐ โ
๐ ), ๐ = 0, 1, . . . , ๐ โ 1. 2 2 2 Here, the evolution of the polygon vertices (or the agents in cyclic pursuit) is described by โก ๐ก โค ๐0 โข โฅ ๐๐ก1 0 โฅห ห + 1) = โข P(๐ก โข โฅ P(0) . .. โฃ โฆ 0
๐๐ก๐ โ1
524
F. Oggier and A. Bruckstein
where we de๏ฌned
โก
โข โข ห P(๐ก) = [FT]โ โข โฃ
1
โค ๐โ1/๐
0 .. .
0 From this we have โก 1 โข ๐1/๐ โข P(๐ก) = โข โฃ โก โข โข =โข โฃ
๐1/๐
๐โ(๐ โ1)/๐
โค ..
โฅ โฅ ห โฅ [FT]P(๐ก) โฆ
. ๐
1
โฅ โฅ โฅ P(๐ก). โฆ
..
๐ โ1 ๐
โค
โก
โฅ โข โฅ โข โฅ [FT] โข โฆ โฃ
. ๐
๐ โ1 ๐
๐๐ก0
โค ๐๐ก1 0
0 .. .
โฅ โฅห โฅ P(0). โฆ ๐๐ก๐ โ1
The evolution of the polygon vertices (the swarm of robots) when we let the time grow, thus asymptotically depends on the dominant eigenvalues among ๐0 , . . . , ๐๐ โ1 . If ๐ = 1 (which means a circulant cyclic pursuit), we have 2๐ 1 ๐๐ = (1 + ๐โ๐ ๐ โ
๐ ), ๐ = 0, 1, . . . , ๐ โ 1, 2 and ๐0 = 1. Then โก ๐ก โค ๐0 โข โฅ ๐๐ก1 0 โข โฅห lim P(๐ก) = lim [FT] โข โฅP(0) . . ๐กโโ ๐กโโ โฃ โฆ . 0 ๐ก ๐๐ โ1 โก โค 1 โข โฅ ๐๐ก1 0 โข โฅ โข โฅ 0 โข โฅ = lim [FT] โข โฅ[FT]โ P(0). .. ๐กโโ โข โฅ . 0 โข โฅ โฃ โฆ 0 ๐ก ๐๐โ1 Since the dominant eigenvalue ๐0 = 1 and all others have modulus less than one, we have that the limiting behavior is โก โค 1 โฅ 1 โข โข 1 โฅ lim P(๐ก) = โข .. โฅ [1, 1, . . . , 1]P(0). ๐กโโ ๐โฃ . โฆ 1
On Cyclic and Nearly Cyclic Multiagent Interactions
525
Hence the point constellation converges to the centroid of the initial locations. The way this convergence occurs is controlled by the next dominant eigenvalues, which are in this case 1 2๐ (1 + ๐โ๐ ๐ ) 2 2๐(๐ โ1) 1 = (1 + ๐โ๐ ๐ ). 2
๐1 = ๐๐โ1 Indeed, writing
โก P๐ (๐ก) = P(๐ก) โ
1 ๐
โข โข โข โฃ
1 1 .. .
โค โฅ โฅ โฅ [1, 1, . . . , 1]P(0), โฆ
1 we have
โก โข โข P๐ (๐ก) = [FT] โข โฃ
0
โค ๐๐ก1 0
0 .. .
โฅ โฅ โฅ [FT]โ P(0) โฆ ๐๐ก๐ โ1
and, disregarding the faster decaying terms ๐๐ก๐ , ๐ = 2, . . . , ๐ โ 2, we further get โก โค 1 โฅ 1 โข โข ๐ค โฅ lim P๐ (๐ก) = โข โฅ [1, ๐ค, . . . , ๐ค๐ โ1 ]P(0)๐๐ก1 . .. ๐กโโ ๐โฃ โฆ ๐ โ1 ๐ค โก โค 1 โฅ ๐ค๐ โ1 1 โข โข โฅ + โข โฅ [1, ๐ค๐โ1 , . . . , ๐ค(๐ โ1)(๐โ1) ]P(0)๐๐ก๐ โ1 . .. ๐ โฃ โฆ . ๐ค(๐โ1)(๐ โ1)
Hence
โก lim P๐ (๐ก) =
๐กโโ
1 ๐
โข โข โข โฃ
1 ๐ค .. .
๐ค๐ โ1
โค
โก
โฅ 1 โข โฅ โข โฅ ๐ด(๐ก)๐๐ก1 + โข ๐โฃ โฆ
1
๐ค๐ โ1 .. .
โค โฅ โฅ โฅ ๐ต(๐ก)๐๐ก๐ โ1 โฆ
๐ค(๐โ1)(๐ โ1)
where ๐ด(๐ก)๐๐ก1 and ๐ต(๐ก)๐๐ โ1 are some complex numbers, and P๐ (๐ก) will be, in the limit ๐ก โ โ, an a๏ฌne transformation of a regular polygon, i.e., a discrete ellipse (see Figure 1).
526
F. Oggier and A. Bruckstein 1 1
0.9 0.8
0.8 0.7 0.6
0.6
0.5 0.4
0.4 0.3
0.2 0.2 0
0
0.2
0.4
0.6
0.8
1
0.1 0.1
1.2
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.4709
Hkiwtg 30 Vjg e{enke rwtuwkv ecug ) ? 3* ykvj c tcpfqo kpkvkcn rqn{iqp ykvj O ? 9 rqkpvu- vjg tuv iwtg rtgugpvu vjg kpkvkcn eqp iwtcvkqp )vjg qwvukfg qpg* cpf vjg tuv kvgtcvkqp )kpukfg*vjg ugeqpf ujqyu vjg gpvktg gxqnw. vkqp hqt 322 kvgtcvkqpu- vjg ncuv i. wtg fkurnc{u vjg uecngf wr eqp i. wtcvkqp hqt vjg ncuv hgy kvgtcvkqpu0
0.4709
0.4709
0.4708
0.4708
0.4708 0.6204
0.6204
0.6204
0.6204
0.6204
0.6204
0.6204
For the general case where ๐ is some real or complex number, we have that โก โข โข lim P(๐ก) = lim โข ๐กโโ ๐กโโ โฃ โก โข โข โข = lim โข ๐กโโ โข โข โฃ
1
โค ๐1/๐
..
โฅ โข โฅ โข โฅ [FT] โข โฆ โฃ
. ๐
1
๐ โ1 ๐
๐1/๐ ..
โก
. ๐
๐ โ1 ๐
๐๐ก0
โค ๐๐ก1 0
โข โข ห P(0) = [FT]โ โข โฃ
1
โก ๐ก ๐0 โฅ โฅ โข โฅ โฅ [FT] โข โข โฅ โฃ โฅ โฆ
โค 0 0
0 .. .
โค ๐โ1/๐
๐๐ก๐ โ1
โค
where ๐0 = 12 (1 + ๐1/๐ ) is the dominant eigenvalue. Since โก
โฅ โฅห โฅ P(0) โฆ
0 .. .
..
โฅ โฅ โฅ P(0), โฆ
. ๐
โ(๐ โ1) ๐
โฅ โฅห โฅ P(0), โฆ 0
On Cyclic and Nearly Cyclic Multiagent Interactions we then have that
โก
โข โข lim P(๐ก) = ๐๐ก0 โข ๐กโโ โฃ
1
527
โค ๐1/๐
โก โข โข ร ([FT]๐,1 )โ โข โฃ
..
โฅ โฅ โฅ [FT]๐,1 โฆ
. ๐
1
๐ โ1 ๐
โค โฅ โฅ โฅ P(0) โฆ
โ1/๐
๐
..
. ๐
โ(๐ โ1) ๐
and since the ๏ฌrst column of the Fourier transform is a vector of all ones, this further simpli๏ฌes to โค โก 1 1/๐ โฅ โ(๐ โ1) 1 โข ๐. โฅ [1, ๐โ1/๐ , . . . , ๐ ๐ ]P(0). lim P(๐ก) = ๐๐ก0 โข . โฆ โฃ . ๐กโโ ๐ ๐ โ1 ๐ ๐ Therefore, we see that the limiting behavior is dominated by
โก 1 ]๐ก 1/๐ โข โ(๐ โ1) ๐ 1 1 .. (1 + ๐1/๐ ) [1, ๐โ1/๐ , . . . , ๐ ๐ ]P(0) โข lim P(๐ก) = . ๐กโโ โฃ 2 ๐ ๐ โ1 a (complex) scalar ๐ ๐ [
โค โฅ โฅ. โฆ
We can distinguish di๏ฌerent behaviors depending on ๐. 1. If ๐ is real and โฃ๐โฃ < 1, P(๐ก) tends to zero, but the limit behavior will be a linear constellation of points โก โค โก โค 1 1 โข ๐1/๐ โฅ โข 1/๐ โฅ โฅ + ๐(๐ผ๐ก )๐ฆ โข ๐ .. โฅ. .. (๐ผ๐ก )๐ฅ โข โฃ โฆ โฃ โฆ . . ๐
๐ โ1 ๐
๐
๐ โ1 ๐
If โฃ๐โฃ > 1, the constellation of agent locations will diverge in a similar formation. 2. If ๐ is a complex number ๐(๐) ๐๐๐(๐) , the convergence/divergence will depend on the angle of rotation induced by ๐(๐) and on the magnitude ๐(๐) . As seen in the examples provided in Figures 2, 3, 4, 5, 6, 7, in the limit, agents are marching in elliptic or circular arcs, spiralling towards their point of convergence (and in case of divergence, spiralling out to in๏ฌnity). As in Figure 1, the left ๏ฌgure presents the initial con๏ฌguration (in red) and the ๏ฌrst iteration (in blue), the second shows the entire evolution for 100 iterations (unless stated otherwise), the last ๏ฌgure displays the scaled up con๏ฌguration for the last few iterations.
528
F. Oggier and A. Bruckstein 1 1
0.9 0.8
0.8
0.7 0.6
0.6 0.5 0.4 0.4 0.3 0.2
0.2
0.1 0
0
0.2
0.4
0.6
0.8
1
1.2
0
0
0.2
0.4
0.6
0.8
1
โ7
6
x 10
5
4
3
2
1
Hkiwtg 40 ? 203
โ7
0
1
x 10 0
1
2
3
4
5
6
1 0.8
0.8
0.6 0.6 0.4 0.4
0.2 0
0.2
โ0.2 0 โ0.4 โ0.2
โ0.6
โ0.4 โ0.4
โ0.2
0
0.2
0.4
0.6
0.8
1
โ0.8 โ0.8
โ0.6
โ0.4
โ0.2
0
0.2
0.4
0.6
0.8
1
โ0.08
โ0.06
โ0.04
โ0.02
0
0.02
0.04
0.06
0.08
0.08 0.06 0.04 0.02 0 โ0.02 โ0.04 โ0.06 โ0.08
Hkiwtg 50 ? 3
โ0.1 โ0.1
On Cyclic and Nearly Cyclic Multiagent Interactions 1
529
1 0.8
0.8
0.6 0.6 0.4 0.4
0.2 0
0.2
โ0.2 0 โ0.4 โ0.2
โ0.6
โ0.4 โ0.4
โ0.2
0
0.2
0.4
0.6
0.8
1
โ0.8 โ0.8
โ0.6
โ0.4
โ0.2
0
0.2
0.4
0.6
0.8
1
0.4 0.3 0.2 0.1 0 โ0.1 โ0.2 โ0.3
Hkiwtg 60 ? j
โ0.4 โ0.4
1
1
0.9
0.8
0.8
โ0.3
โ0.2
โ0.1
0
0.1
0.2
0.3
0.4
0.6
0.7
0.4
0.6 0.2 0.5 0 0.4 โ0.2
0.3
โ0.4
0.2
โ0.6
0.1 0
0
0.2
0.4
0.6
0.8
1
โ0.8 โ0.8
โ0.6
โ0.4
โ0.2
0
0.2
0.4
0.6
0.8
1
0.5 0.4 0.3 0.2 0.1 0 โ0.1 โ0.2 โ0.3 โ0.4
Hkiwtg 70 ? j
โ0.5 โ0.5
0
0.5
530
F. Oggier and A. Bruckstein 1.2
1 0.9
1 0.8
0.8
0.7 0.6
0.6
0.5
0.4
0.4 0.3
0.2
0.2
0 0.1 0
0
0.2
0.4
0.6
0.8
1
โ0.2 โ0.2
0
0.2
0.4
0.6
0.8
1
1.2
โ3
7
x 10
6 5 4 3 2 1 0 โ1 โ2
Hkiwtg 80 ? gzr)j16 *14
x 10
โ3 โ9
โ8
โ7
โ6
โ5
โ4
โ3
โ2
โ1
0
โ3
1
1.2
1
1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0
0.2
โ0.2
0
โ0.4 โ0.2 โ0.2
0
0.2
0.4
0.6
0.8
1
โ0.4
1.2
โ0.2
0
0.2
0.4
0.6
0.8
1
1.2
โ3
x 10 3
2
1
0
โ1
โ2
Hkiwtg 90 ? gzr)j1 6 , j*14
โ3
x 10 โ2
โ1
0
1
2
3
On Cyclic and Nearly Cyclic Multiagent Interactions
531
4.2. Centroid gathering evolution and extensions As a second example, suppose that agent ๐ซ๐ is moving according to the following linear combination of its own position, the positions of agents higher in the hierarchy, i.e., {๐ซ๐+1 , . . . , ๐ซ๐ โ1 }, and the positions of those lower than itself {๐ซ0 , ๐ซ1 , . . . , ๐ซ๐โ1 }: ๐ โ1 โ
๐ซ๐ (๐ก + 1) = ๐ผ๐ซ๐ (๐ก) + ๐ฝ๐น
๐ซ๐ (๐ก) + ๐ฝ๐ต
๐=๐+1
or
โก
๐ผ โข ๐ฝ๐ต โข โข P(๐ก + 1) = โข โข โฃ ๐ฝ๐ต
๐โ1 โ
๐ซ๐ (๐ก)
๐=0
๐ฝ๐น ๐ผ
๐ฝ๐น ๐ฝ๐น .. .
... ...
๐ฝ๐น ๐ฝ๐น .. .
...
...
๐ฝ๐ต
๐ผ
โค โฅ โฅ โฅ โฅ P(๐ก). โฅ โฆ
Note that if ๐ฝ๐น = ๐ฝ๐ต = (1 โ ๐ผ)/(๐ โ 1), we will have ๐ 1โ๐ผ โ ๐ซ๐ (๐ก) ๐ โ1 ๐=0,๐โ=๐ ( ) ๐๐ผ โ 1 ๐๐ผ โ 1 = ๐ซ๐ (๐ก) + 1 โ ๐ซ๐๐๐๐ก๐๐๐๐ ๐ โ1 ๐ผโ1
๐ซ๐ (๐ก + 1) = ๐ผ๐ซ๐ (๐ก) +
hence all agents move towards the time-invariant centroid on straight lines. For general ๐ฝ๐น and ๐ฝ๐ต , the above matrix is ๐ฝ๐ต /๐ฝ๐น -factor circulant and is diagonalized by โก โค 1 โข โฅ (๐ฝ๐ต /๐ฝ๐น ) 0 โข โฅ ห P(๐ก) = [FT]โ โข โฅ P(๐ก), . .. โฃ โฆ (๐ฝ๐ต /๐ฝ๐น )๐ โ1
0
the modes or eigenvalues being given by ๐๐ = ๐ผ +
๐ โ1 โ
( ๐ฝ๐น
๐=1
๐ฝ๐ต ๐ฝ๐น
) ๐๐
2๐
๐โ๐ ๐ ๐๐ , ๐ = 0, . . . , ๐ โ 1.
Let us consider ๏ฌrst the case of perfectly cyclic interaction, i.e., when ๐ฝ๐ต = ๐ฝ๐น . In this case, the interaction matrix is circulant, and we have ๐๐ = ๐ผ +
๐ โ1 โ ๐=1
2๐
๐ฝ๐น ๐โ๐ ๐ ๐๐ , ๐ = 0, . . . , ๐ โ 1
532
F. Oggier and A. Bruckstein
and ๐0 = ๐ผ + (๐ โ 1)๐ฝ๐น ๐๐ = ๐ผ โ ๐ฝ ๐น +
๐ โ1 โ
2๐
๐ฝ๐น ๐โ๐ ๐ ๐๐ = ๐ผ โ ๐ฝ๐น .
๐=0
For normalization, we shall take ๐ฝ๐น = (1 โ ๐ผ)/(๐ โ 1) and then ๐0 = 1 ๐๐ = (๐ ๐ผ โ 1)/(๐ โ 1), for all ๐. We now have that ห P(๐ก) = [FT]โ P(๐ก) evolves according to โก โข โข โข ห lim P(๐ก) = โข ๐กโโ โข โฃ
1
(
๐ ๐ผโ1 ๐ โ1
โค
)๐ก ..
.
(
๐ ๐ผโ1 ๐โ1
โก โฅ โฅ โข โฅห โข โฅ P(0) = โข โฅ โฃ )๐ก โฆ
1 0 .. .
โค โฅ โฅ ห โฅ [1, 0, . . . , 0]P(0). โฆ
0
Hence โก โข โข lim P(๐ก) = [FT] โข ๐กโโ โฃ
1 0 .. .
โค
โก
โฅ 1 โข โฅ โข โฅ [1, 0, . . . , 0][FT]โ P(0) = โข โฆ ๐โฃ
0
1 1 .. .
โค โฅ โฅ โฅ [1, 1, . . . , 1]P(0), โฆ
1
i.e., as we have already seen, all points converge towards the centroid of the initial constellation. The convergence will be as follows: โก โข ห โโข lim P๐ (๐ก) = P(๐ก) โข ๐กโโ โฃ
1 0 .. .
โค โฅ โฅ ห โฅ [1, 0, . . . , 0]P(0) โฆ
0
( =
โก 0 )๐ก โข ๐๐ผ โ 1 โข โข ๐ โ1 โฃ 0
โค 1
..
โฅ โฅห โฅ P(0). โฆ
. 1
On Cyclic and Nearly Cyclic Multiagent Interactions
0.8
0.8
0.7
0.7
0.6
0.6
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.1
0.2
0.3
0.4
0.5
0.6
0.7
533
0.8
0.9
0.5044 0.5044 0.5044 0.5044 0.5044 0.5044 0.5044 0.5044 0.5044
Hkiwtg :0 ? 3- ? 203322 kvgtcvkqpu
0.5044 0.5044 0.4786
0.4786
0.4786
0.4786
0.4786
0.4787
0.4787
Therefore โ
โก
โข โ โข โ lim P๐ (๐ก) = [FT] โ๐ผ โ โข ๐กโโ โฃ โ
1 0 .. .
โ โฅ โ โฅ ห โฅ [1, 0, . . . , 0]โ [FT]โ P(0) โ โฆ
0 โ ( =
โ
โค
โก
)๐ก โข ๐๐ผ โ 1 โ โข โ โP(0) โ โข โฃ โ ๐ โ1
1 1 .. .
โค
โ
โ โฅ โ โฅ โฅ [1, 1, . . . , 1]P(0)โ . โ โฆ
1 Consequently, all agents will gather towards the centroid by moving on a line from โ๐ โ1 ๐ซ๐ (0) to (1/๐ ) ๐=1 ๐ซ๐ (0) (see Figure 8). Next suppose we have ๐ฝ๐ต โ= ๐ฝ๐น . Then we have a ๐ = ๐ฝ๐ต /๐ฝ๐น factor circulant ห and the modes of the P(๐ก) evolution is controlled by ๐๐ = ๐ผ +
๐ โ1 โ ๐=1
( ๐ฝ๐น
๐ฝ๐ต ๐ฝ๐น
)๐/๐
2๐
๐โ๐ ๐ ๐๐ , ๐ = 0, . . . , ๐ โ 1.
534
F. Oggier and A. Bruckstein
Here ๐0 = ๐ผ โ ๐ฝ๐น + ๐ฝ๐น
๐ โ1 ( โ ๐=0
๐ฝ๐ต ๐ฝ๐น
)๐/๐
๐ฝ๐ต /๐ฝ๐น โ 1 (๐ฝ๐ต /๐ฝ๐น )1/๐ โ 1 ) ( )( 1โ๐ผ 1โ๐ผ ๐โ1 =๐ผโ . + ๐ โ1 ๐ โ1 ๐1/๐ โ 1 Similarly we have that ( )๐/๐ ๐ โ1 โ 2๐ ๐ฝ๐ต ๐ฝ๐น ๐โ๐ ๐ ๐๐ ๐๐ = ๐ผ + ๐ฝ๐น ๐=1 ( )( ) 1โ๐ผ ๐๐โ๐๐๐ โ 1 1โ๐ผ + =๐ผโ . ๐ โ1 ๐ โ1 ๐1/๐ ๐๐๐๐/๐ โ 1 In this example too, as before, we have โก ๐ก โก โค 1 ๐0 1/๐ โข โข โฅ ๐ ๐๐ก1 โข โข โฅ [FT] lim P(๐ก) = lim โข โข โฅ . .. .. ๐กโโ ๐กโโ โฃ โฃ โฆ . = ๐ผ โ ๐ฝ๐น + ๐ฝ๐น
0
โก
โข โข ร [FT]โ โข โฃ
๐ 1
๐ โ1 ๐
๐โ1/๐
0 and if ๐0 is the dominant eigenvalue, we โก 1 1/๐ โข 1 โข ๐ lim P(๐ก) = lim ๐๐ก0 โข .. ๐กโโ ๐กโโ ๐ โฃ .
..
0
โค
โค โฅ โฅ โฅ โฆ ๐๐ก๐ โ1
โฅ โฅ โฅ P(0) โฆ
. ๐
โ(๐ โ1) ๐
shall have โค โฅ โฅ โฅ [1, ๐โ1/๐ , . . . , ๐โ(๐ โ1)/๐ ]P(0). โฆ
๐๐ โ1/๐
Depending on the values selected for ๐, we can get a wealth of interesting behaviors while the solutions converge or diverge to in๏ฌnity, displaying spiralling or in line marching. See Figures 9, 10, 11, 12, 13, 14 where we present a few interesting cases.
On Cyclic and Nearly Cyclic Multiagent Interactions
535
0.8 0.8 0.7 0.7 0.6 0.6 0.5 0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0
0.2
0.4
0.6
0.8
1
โ13
x 10 3.4 3.2 3 2.8 2.6 2.4 2.2 2 1.8
Hkiwtg ;0 ? 207- ? 2033222 kvgtcvkqpu
1.6 x 10 1
1.5
2
2.5
โ13
3
0.8 0.8 0.7 0.7
0.6
0.6
0.5
0.5
0.4 0.3
0.4
0.2 0.3 0.1 0.2 0 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0
0.2
0.4
0.6
0.8
1
โ5
x 10 8
6
4
2
0
โ2
Hkiwtg 320 ? 3- ? 203322 kvgtcvkqpu
โ4 x 10 โ6
โ4
โ2
0
2
4
6
8
10
โ5
536
F. Oggier and A. Bruckstein 0.8 0.8
0.7 0.6
0.7
0.5 0.6 0.4 0.5
0.3 0.2
0.4
0.1 0.3
0 โ0.1
0.2
โ0.2 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
โ0.2
0
0.2
0.4
0.6
0.8
1
0.025
0.02
0.015
0.01
0.005
0
Hkiwtg 330 ? j- ? 203322 kvgtcvkqpu
โ0.005
0
0.005
0.01
0.015
0.02
0.025
0.03
0.8 0.8 0.7 0.7
0.6
0.6
0.5 0.4
0.5 0.3 0.4
0.2
0.3
0.1 0
0.2 โ0.1 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
โ0.2
0
0.2
0.4
0.6
0.8
0.035
0.03
0.025
0.02
0.015
0.01
Hkiwtg 340 ? j- ? 203322 kvgtcvkqpu
0.005
0 โ0.005
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
On Cyclic and Nearly Cyclic Multiagent Interactions
537
0.8 0.8 0.7 0.7 0.6 0.6
0.5
0.5
0.4
0.4
0.3 0.2
0.3
0.1 0.2 0 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0
0.2
0.4
0.6
0.8
1
โ0.015
โ0.02
โ0.025
โ0.03
Hkiwtg 350 ? gzr)j16 *14 - ? 203322 kvgtcvkqpu
โ0.035
โ3
x 10 โ5
0
5
10
15
20
0.8 0.8 0.7 0.7 0.6 0.6
0.5
0.5
0.4
0.4
0.3
0.3
0.2 0.1
0.2
0 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0
0.2
0.4
0.6
0.8
1
โ4
x 10 4 2 0 โ2 โ4 โ6 โ8
Hkiwtg 360 ? gzr)j1 6 , j*14 ? 203- 322 kvgtcvkqpu
โ10 โ12 โ5
โ4
x 10 0
5
10
15
538
F. Oggier and A. Bruckstein
5. Concluding remarks We discussed in this paper a special type of cyclic multiagent interaction modeled by ๐-factor cyclic matrices. Such matrices allow explicit closed form diagonalizations via generalized Fourier transforms hence enable the analysis of the evolution of the swarm via a nice, geometric, modal decomposition process. It is expected that a wealth of further similar, structured and nearly cyclic interactions will also yield explicit closed form solutions for their asymptotic behavior. In fact, we may use evolutions that ๏ฌx one, two [17] or several agents in the swarm and use circulant or ๐-circulant interactions for the rest of them leading to further highly structured matrices that can be diagonalized, and correspondingly leading to interesting and explicitly predictable and designable swarm dynamics. In closing, we note that Turingโs morphogenesis may be regarded as a further example of such dynamics for points in the plane where the ๐ฅ and the ๐ฆ coordinates are subjected to di๏ฌerent linear circulant transformations also readily generalizable to ๐-circulant maps [16]. An analysis of such swarm interaction for multiagent system is forthcoming. Acknowledgment The research of F. Oggier is supported in part by the Singapore National Research Foundation under Research Grant NRF-RF2009-07 and NRF-CRP2-2007-03, and in part by the Nanyang Technological University under Research Grant M58110049 and M58110070. The work of Alfred Bruckstein is supported in part by a Nanyang Technological University visiting professorship, at the SPMS and IMI center.
References [1] E.C. Boman. The Moore Penrose pseudoinverse of an arbitrary, square, k-circulant matrix. Linear and Multilinear Algebra, 50:175โ179, 2002. [2] A.M. Bruckstein, N. Cohen, and A. Efrat. Ants, crickets and frogs in cyclic pursuit. CIS9105 technical report, Computer Science Dept., Technion, 1991. [3] A.M. Bruckstein, G. Sapiro, and D. Shaked. Evolutions of planar polygons. International Journal of Pattern Recognition and Arti๏ฌcial Intelligence, 9(6):991โ1014, 1995. [4] R.E. Cline, R.J. Plemmons, and G. Worm. Generalized inverses of certain Toeplitz matrices. Linear Algebra and Its Applications, 8:25โ33, 1974. [5] M.G. Darboux. Sur un probl`eme de gยดeomยดetrie ยดelยดementaire. Bull. Sci. Math., 2:298โ 304, 1878. [6] P. Elia, F. Oggier, and P. Vijay Kumar. Asymptotically optimal cooperative wireless networks without constellation expansion. IEEE Journal on Selected Areas in Communications on Cooperative Communications and Networking, 25, 2007. [7] P. Feinsilver. Circulants, inversion of circulants, and some related matrix algebras. Linear. Algebra and Appl., 56:29โ43, 1984. [8] I. Gohberg and V. Olshevsky. Circulants, displacements and decompositions of matrices. Integral Equations and Operator Theory, 15:730โ743, 1992.
On Cyclic and Nearly Cyclic Multiagent Interactions
539
[9] R.M. Gray. Toeplitz and circulant matrices: A review. Intelligent systems lab technical memo, Stanford University, 1971โ2006. [10] F. Hirsh and S. Smale. Di๏ฌerential Equations, Dynamical Systems and Linear Algebra. Academic Press, 1974. [11] J.A. Marshall and M.E. Broucke. Symmetry invariance of multiagent formations in self-pursuit. IEEE Transactions on Automatic Control, 53(9):2022โ2032, 2008. [12] J.A. Marshall, M.E. Broucke, and B.A. Francis. Formations of vehicles in cyclic pursuit. IEEE Transactions on Automatic Control, 49(11):1963โ1974, 2004. [13] I.J. Schoenberg. The ๏ฌnite Fourier series and elementary geometry. Amer. Math. Monthly, 57(6):390โ404, 1950. [14] I. Schur. Bemerkungen zur Theorie der beschrยจ ankten Bilinearformen mit unendlich vielen Verยจ anderlichen. J. reine angew. Math., 140:1โ28, 1911. [15] D.B. Shapiro. A periodicity problem in plane geometry. The American Math. Monthly, 91:97โ108, 1984. [16] A.M. Turing. The chemical basis of morphogenesis. Philosophical Transactions of the Royal Society of London, series B, Biological Sciences, 237(641), 1952. [17] I. Wagner and A.M. Bruckstein. Row straightening by local interactions. Circuits, Systems and Signal Processing, 16(3):287โ305, 1997. [18] A.C. Wilde. Di๏ฌerential equations involving circulant matrices. Rocky Mount. J. Math., 13(1):1โ13, 1983. Frยดedยดerique Oggier and Alfred Bruckstein2 Division of Mathematical Sciences School of Physical and Mathematical Sciences Nanyang Technological University, Singapore e-mail:
[email protected] [email protected]
2 Alfred
Bruckstein is visiting Professor from The Technion โ IIT, Haifa, Israel.
Operator Theory: Advances and Applications, Vol. 218, 541โ570 c 2012 Springer Basel AG โ
A Trace Formula for Di๏ฌerential Operators of Arbitrary Order ยจ J. Ostensson and D.R. Yafaev To the memory of Israel Cudicovich Gohberg
Abstract. An operator ๐ป = ๐ป0 + ๐ where ๐ป0 = ๐โ๐ โ ๐ (๐ is arbitrary) and ๐ is a di๏ฌerential operator of order ๐ โ1 with coe๏ฌcients decaying su๏ฌciently rapidly at in๏ฌnity is considered in the space ๐ฟ2 (โ). The goal of the paper is to ๏ฌnd an expression for the trace of the di๏ฌerence of the resolvents (๐ป โ ๐ง)โ1 and (๐ป0 โ ๐ง)โ1 in terms of the Wronskian of appropriate solutions to the di๏ฌerential equation ๐ป๐ข = ๐ง๐ข. This also leads to a representation for the perturbation determinant of the pair ๐ป0 , ๐ป. Mathematics Subject Classi๏ฌcation (2000). 34B25, 35P25, 47A40. Keywords. One-dimensional di๏ฌerential operators, arbitrary order, resolvents, perturbation determinant, trace formula.
1. Introduction 1.1. In the framework of the general operator theory in an abstract Hilbert space, the spectral theory of di๏ฌerential operators ๐ป = ๐โ๐ โ ๐ + ๐ฃ๐ (๐ฅ)โ ๐ โ1 + โ
โ
โ
+ ๐ฃ2 (๐ฅ)โ + ๐ฃ1 (๐ฅ),
โ = ๐/๐๐ฅ,
(1.1)
is the same for all values of ๐ . However, from the point of view of di๏ฌerential equations the problems are essentially di๏ฌerent for ๐ = 2 (for ๐ = 1 it is trivial) and for larger values of ๐ . Suppose that the coe๏ฌcients ๐ฃ๐ (๐ฅ), ๐ = 1, . . . , ๐ , decay su๏ฌciently rapidly as โฃ๐ฅโฃ โ โ, and set ๐ป0 = ๐โ๐ โ ๐ . Let ๐
0 (๐ง) = (๐ป0 โ ๐ง)โ1 and ๐
(๐ง) = (๐ป โ ๐ง)โ1 be the resolvents of the operators ๐ป0 and ๐ป acting in the space ๐ฟ2 (โ). The selfadjointness of the operator ๐ป is inessential for us, and we do not assume it. The ๏ฌrst author is grateful to Ari Laptev for useful and stimulating discussions. The second author was partially supported by the project NONAa, ANR-08-BLANC-0228.
ยจ J. Ostensson and D.R. Yafaev
542
The main goal of the present paper is to ๏ฌnd an expression for the trace ( ) Tr ๐
(๐ง) โ ๐
0 (๐ง) (1.2) in terms of solutions to the di๏ฌerential equation ๐ป๐ข = ๐ง๐ข. In the case ๐ = 2 such an expression was found by V.S. Buslaev and L.D. Faddeev in paper [5]. They considered the problem on the half-line, and the problem on the whole line was discussed by L.D. Faddeev in [7]. 1.2. Let us introduce the notation
โ
โ โ {๐ข1 , . . . , ๐ข๐ } = โ โ
๐ข1 ๐ขโฒ1 .. .
(๐ โ1)
๐ข1
... ... .. .
๐ข๐ ๐ขโฒ๐ .. .
(๐ โ1)
โ โ โ โ โ
(1.3)
. . . ๐ข๐
for the Wronskian matrix of solutions ๐ข1 = ๐ข1 (๐ฅ, ๐ง), . . . , ๐ข๐ = ๐ข๐ (๐ฅ, ๐ง) of the di๏ฌerential equation ๐โ๐ ๐ข(๐ ) (๐ฅ) + ๐ฃ๐ (๐ฅ)๐ข(๐ โ1) (๐ฅ) + โ
โ
โ
+ ๐ฃ2 (๐ฅ)๐ขโฒ (๐ฅ) + ๐ฃ1 (๐ฅ)๐ข(๐ฅ) = ๐ง๐ข(๐ฅ).
(1.4)
We always assume that ๐ง โ โ โ [0, โ) if ๐ is even and that Im ๐ง โ= 0 if ๐ is odd. Let ๐๐ be solutions of the equation ๐ ๐ = ๐๐ ๐ง. We suppose that Re ๐๐ > 0
for ๐ = 1, . . . , ๐ and
Re ๐๐ < 0 for
๐ = ๐ + 1, . . . , ๐.
(1.5)
Here ๐ = ๐/2 if ๐ is even and ๐ = (๐ โ 1)/2 for Im ๐ง > 0 and ๐ = (๐ + 1)/2 for Im ๐ง < 0 if ๐ is odd. We ๏ฌrst explain our result for the case of functions ๐ฃ๐ (๐ฅ) with compact supports. We write ๐ฅ << 0 if ๐ฅ lies to the left of the supports of all ๐ฃ๐ (๐ฅ) and ๐ฅ >> 0 if ๐ฅ lies to the right of this set. Let ๐ข๐ (๐ฅ, ๐ง) be solutions of equation (1.4) such that ๐ข๐ (๐ฅ, ๐ง) = ๐๐๐ ๐ฅ for ๐ฅ << 0 if ๐ = 1, . . . , ๐ Let
and for ๐ฅ >> 0 if ๐ = ๐ + 1, . . . , ๐. (1.6)
W(๐ฅ, ๐ง) = det{๐ข1 (๐ฅ, ๐ง), . . . , ๐ข๐(๐ฅ, ๐ง), ๐ข๐+1 (๐ฅ, ๐ง), . . . , ๐ข๐ (๐ฅ, ๐ง)} be the determinant of matrix (1.3), and let W0 (๐ง) = det{๐๐1 ๐ฅ , . . . , ๐๐๐ ๐ฅ , ๐๐๐+1 ๐ฅ , . . . , ๐๐๐ ๐ฅ }
(1.7) (1.8)
be the corresponding Wronskian for the โfreeโ case where ๐ฃ๐ = 0 for all ๐ = 1, . . . , ๐ . Of course, ( ) โซ ๐ฅ2 ๐ฃ๐ (๐ฆ)๐๐ฆ W(๐ฅ1 , ๐ง) (1.9) W(๐ฅ2 , ๐ง) = exp โ๐๐ ๐ฅ1
for arbitrary points ๐ฅ1 and ๐ฅ2 . We emphasize that the Wronskians W(๐ฅ, ๐ง) and W0 (๐ง) depend on the order of numeration of the numbers ๐๐ , but the normalized Wronskian ฮ(๐ฅ, ๐ง) = W(๐ฅ, ๐ง)/W0 (๐ง) (1.10) does not depend on it.
Trace Formula
543
Our main result is that the normalized Wronskian satis๏ฌes (for all ๐ฅ and all regular points ๐ง of the operator ๐ป) the equation ( ) Tr ๐
(๐ง) โ ๐
0 (๐ง) = โฮ(๐ฅ, ๐ง)โ1 ๐ฮ(๐ฅ, ๐ง)/๐๐ง, (1.11) which we call the trace formula in this paper. Thus the trace of the di๏ฌerence of the resolvents admits an explicit expression in terms of properly chosen solutions of equation (1.4). Then we extend representation (1.11) to general short-range coe๏ฌcients ๐ฃ๐ (๐ฅ) satisfying the assumption โซ โ โฃ๐ฃ๐ (๐ฅ)โฃ2 (1 + ๐ฅ2 )๐ผ ๐๐ฅ < โ, ๐ผ > 1/2, ๐ = 1, . . . , ๐, (1.12) โโ
only. In this case the functions ๐ข๐ (๐ฅ, ๐ง) in de๏ฌnition (1.7) are the solutions of equation (1.4) such that ๐ข๐ (๐ฅ, ๐ง) = ๐๐๐ ๐ฅ (1 + ๐(1))
(1.13)
as ๐ฅ โ โโ if ๐ = 1, . . . , ๐ and as ๐ฅ โ +โ if ๐ = ๐ + 1, . . . , ๐ . Here and below all asymptotic relations for solutions of equation (1.4) are supposed to be ๐ โ 1 times di๏ฌerentiable in ๐ฅ. We emphasize that for ๐ > 2 asymptotics (1.13) DO NOT determine the solutions of equation (1.4) uniquely. However, the Wronskian (1.7) does not depend on speci๏ฌc choice of the solutions satisfying (1.13). Thus we do not need the construction of the book [2] by R. Beals, P. Deift and C. Tomei devoted to the inverse scattering problem. In [2] solutions of equation (1.4) were distinguished uniquely (away from some exceptional set of values of ๐ง) by conditions at both in๏ฌnities. Our construction of solutions of equation (1.4) with asymptotics (1.13) relies on integral equations which are Volterra equations for ๐ = 2 but are only Fredholm equations in the general case. Nevertheless for the construction of solutions with asymptotics (1.13) as ๐ฅ โ +โ (as ๐ฅ โ โโ) we impose conditions on the coe๏ฌcients ๐ฃ๐ (๐ฅ) also as ๐ฅ โ +โ (as ๐ฅ โ โโ) only. Suppose that ๐ฃ๐ = 0. Then W(๐ฅ, ๐ง) = W(๐ง) and hence ฮ(๐ฅ, ๐ง) = ฮ(๐ง) do not depend on ๐ฅ. In this case we identify ฮ(๐ง) with the perturbation determinant for the pair of operators ๐ป0 , ๐ป. We refer to the book [10] by I.C. Gohberg and M.G. Kreหฤฑn for a comprehensive discussion of di๏ฌerent properties of perturbation determinants. Set ๐ = ๐ป โ ๐ป0 = ๐ฃ๐ (๐ฅ)โ ๐ โ1 + โ
โ
โ
+ ๐ฃ2 (๐ฅ)โ + ๐ฃ1 (๐ฅ).
(1.14)
If ๐ฃ๐ = 0, then the operator ๐ ๐
0 (๐ง) for Im ๐ง โ= 0 belongs to the trace class ๐1 , and hence the perturbation determinant ) ( (1.15) ๐ท(๐ง) = Det ๐ผ + ๐ ๐
0 (๐ง) is well de๏ฌned. Of particular importance is the abstract trace formula ( ) Tr ๐
(๐ง) โ ๐
0 (๐ง) = โ๐ท(๐ง)โ1๐๐ท(๐ง)/๐๐ง,
(1.16)
544
ยจ J. Ostensson and D.R. Yafaev
which for de๏ฌnition (1.15) is a direct consequence of the formula for the derivative of a determinant. Comparing equations (1.11) and (1.16) and using that ฮ(๐ง) โ 1 as โฃ Im ๐งโฃ โ โ, we show that ( ) Det ๐ผ + ๐ ๐
0 (๐ง) = ฮ(๐ง). (1.17) Thus the perturbation determinant admits an explicit expression in terms of solutions of equation (1.4). If ๐ฃ๐ โ= 0, then under assumption (1.12) it is still true that (for all regular points ๐ง) (1.18) ๐
(๐ง) โ ๐
0 (๐ง) โ ๐1 , although ๐ ๐
0 (๐ง) โโ ๐1 . Without the condition ๐ฃ๐ = 0, equation (1.16) is satis๏ฌed ห for so-called generalized perturbation determinants ๐ท(๐ง) which are de๏ฌned up to constant factors (see subs. 6.2). According to equation (1.11) in the general case for every ๏ฌxed ๐ฅ โ โ, the function ฮ(๐ฅ, ๐ง) di๏ฌers from each generalized perturbation determinant by a constant (not depending on ๐ง) factor. 1.3. A preliminary step in the proof of the trace formula (1.11) is to ๏ฌnd a convenient representation for the resolvent ๐
(๐ง) of the operator ๐ป. This construction goes probably back to the beginning of the twentieth century. We refer to relatively recent books [1, 2, 12] where its di๏ฌerent versions can be found. We start, however, with writing down necessary formulas in a form convenient for us. A di๏ฌerential equation of order ๐ can, of course, be rewritten as a special system of ๐ di๏ฌerential equations of the ๏ฌrst order. A consideration of ๏ฌrst-order systems without special assumptions on their coe๏ฌcients gives more general and transparent results. A large part of the paper is written in terms of solutions of ๏ฌrstorder systems which implies the results about solutions of di๏ฌerential equations of an arbitrary order as their special cases. Let us brie๏ฌy discuss the structure of the paper. In Sections 2 and 3 we collect necessary formulas for solutions of ๏ฌrst-order systems. They are used in Section 4 for the construction of the integral kernel ๐
(๐ฅ, ๐ฆ, ๐ง) of ๐
(๐ง). In particular, we obtain a new representation for the integral โซ ๐ฅ2 ๐
(๐ฆ, ๐ฆ, ๐ง)๐๐ฆ (1.19) ๐ฅ1
where the points ๐ฅ1 , ๐ฅ2 โ โ are arbitrary. Then passing to the limit ๐ฅ1 โ โโ, ๐ฅ2 โ +โ, we prove the trace formula (1.11) for the coe๏ฌcients ๐ฃ๐ , ๐ = 1, . . . , ๐ , with compact supports. A construction of solutions of equation (1.4) with asymptotics (1.13) is given in Section 5. Here we again ๏ฌrst consider a general system of ๐ di๏ฌerential equations of the ๏ฌrst order. Finally, in Section 6 we give the definition of the normalized Wronskian for operators ๐ป with arbitrary short-range coe๏ฌcients and extend the trace formula to the general case. At the end we prove that the normalized Wronskian coincides with the perturbation determinant. 1.4. We note that there exists a somewhat di๏ฌerent approach to proofs of formulas of type (1.17). It consists of a direct calculation of determinant (1.15) whereas we
Trace Formula
545
proceed from a calculation of trace (1.2). In this way formula (1.17) was proven in [11] for the Schrยจ odinger operator on the half-line. In [11] the Fredholm expansion of determinants was used. ( ) A general approach to a calculation of determinants Det ๐ผ +๐พ was proposed in the book [9] by I.C. Gohberg, S. Goldberg and N. Krupnik. In this book integral operators ๐พ with so-called semi-separable kernels were considered. It is important that operators ๐พ = ๐ ๐
0 (๐ง) ๏ฌt into this class. This approach was applied to the Schrยจ odinger operator in paper [8]. The authors thank F. Gesztesy for pointing out references [11, 9, 8].
2. Resolvent kernel In this section we consider an auxiliary vector problem. 2.1. Suppose that the eigenvalues ๐๐ , ๐ = 1, . . . , ๐ , of an ๐ ร ๐ matrix L0 are distinct. We denote by p๐ = (๐1,๐ , ๐2,๐ , . . . , ๐๐,๐ )๐ก (this notation means that the vector p๐ is considered as a column) eigenvectors of L0 corresponding to its eigenvalues ๐๐ and by pโ๐ eigenvectors of Lโ0 corresponding to its eigenvalues ๐ยฏ๐ . Recall that โจp๐ , pโ๐ โฉ = 0 if ๐ โ= ๐ (here โจโ
, โ
โฉ is the scalar product in โ๐ ). Normalizations of p๐ and pโ๐ are inessential, but we suppose that โจp๐ , pโ๐ โฉ = 1. Then the bases p๐ and pโ๐ , ๐ = 1, . . . , ๐ , are dual to each other. Assume that an ๐ ร ๐ matrix V(๐ฅ) where ๐ฅ โ โ belongs locally to ๐ฟ1 and has compact support. We write ๐ฅ << 0 if ๐ฅ lies to the left of the support of V(๐ฅ) and ๐ฅ >> 0 if ๐ฅ lies to the right of this set. We put L(๐ฅ) = L0 + V(๐ฅ).
(2.1)
Consider the homogeneous equation uโฒ (๐ฅ) = L(๐ฅ)u(๐ฅ)
(2.2) ๐ก
for the vector-valued function u(๐ฅ) = (๐ข1 (๐ฅ), . . . , ๐ข๐ (๐ฅ)) . For arbitrary linearly independent solutions u๐ (๐ฅ) = (๐ข1,๐ (๐ฅ), . . . , ๐ข๐,๐ (๐ฅ))๐ก of this equation, we denote by โ โ ๐ข1,1 (๐ฅ) ๐ข1,2 (๐ฅ) . . . ๐ข1,๐ (๐ฅ) โ ๐ข2,1 (๐ฅ) ๐ข2,2 (๐ฅ) . . . ๐ข2,๐ (๐ฅ) โ โ โ U(๐ฅ) = โ . โ =: {u1 (๐ฅ), u2 (๐ฅ), . . . , u๐ (๐ฅ)} .. .. .. โ โ .. . . . ๐ข๐,1(๐ฅ)
๐ข๐,2 (๐ฅ)
. . . ๐ข๐,๐ (๐ฅ)
the corresponding fundamental matrix. It satis๏ฌes the matrix equation Uโฒ (๐ฅ) = L(๐ฅ)U(๐ฅ). It follows that
(2.3) (2.4)
) ( ๐ det U(๐ฅ)/๐๐ฅ = det U(๐ฅ) tr Uโฒ (๐ฅ)Uโ1 (๐ฅ) = det U(๐ฅ) tr L(๐ฅ)
(2.5)
ยจ J. Ostensson and D.R. Yafaev
546 and hence
det U(๐ฅ2 ) = exp
(
โซ
๐ฅ2 ๐ฅ1
) tr L(๐ฆ)๐๐ฆ det U(๐ฅ1 )
(2.6)
for arbitrary points ๐ฅ1 and ๐ฅ2 . Of course det U(๐ฅ) โ= 0 for all ๐ฅ โ โ. We always suppose that ๐
๐ := Re ๐๐ โ= 0 for all ๐ = 1, . . . , ๐ . Let ๐ and ๐ โ ๐ be the numbers of eigenvalues ๐๐ of the matrix L0 lying in the right and left half-planes, respectively. The cases ๐ = 0 or ๐ = ๐ where all ๐๐ lie in one of the half-planes are not excluded. Let u๐ (๐ฅ) be solutions of equation (2.2) distinguished by the condition u๐ (๐ฅ) = ๐๐๐ ๐ฅ p๐ for ๐ฅ << 0 if ๐
๐ > 0 and for ๐ฅ >> 0 if ๐
๐ < 0.
(2.7)
We denote by K+ and Kโ the linear spaces spanned by all solutions u๐ (๐ฅ) such that ๐
๐ > 0 and such that ๐
๐ < 0, respectively. Clearly, dim K+ = ๐ and dim Kโ = ๐ โ ๐. We assume that (2.8) K+ โฉ Kโ = {0}. Then all nontrivial solutions of equation (2.2) exponentially grow either as ๐ฅ โ +โ or as ๐ฅ โ โโ. In particular, equation (2.2) does not have nontrivial solutions u โ ๐ฟ2 (โ; โ๐ ). If u1 (๐ฅ), . . . , u๐ (๐ฅ) and u๐+1 (๐ฅ), . . . , u๐ (๐ฅ) are arbitrary linear independent solutions from K+ and Kโ respectively, then in view of condition (2.8) all these solutions are linearly independent. It is now convenient to accept the following De๏ฌnition 2.1. Suppose that ๐ columns of matrix (2.3) form a basis in the linear space K+ and other ๐ โ ๐ columns form a basis in Kโ . Then the fundamental matrix U(๐ฅ) is called admissible. Observe that for the โfreeโ case where V(๐ฅ) = 0, we can set U0 (๐ฅ) = {p1 ๐๐1 ๐ฅ , . . . , p๐ ๐๐๐ ๐ฅ }
(2.9)
and
( ) W0 (๐ฅ) = det U0 (๐ฅ) = det{p1 , . . . , p๐ } exp tr L0 ๐ฅ (2.10) โ๐ because tr L0 = ๐=1 ๐๐ . Note that det{p1 , . . . , p๐ } โ= 0 since all eigenvalues of the matrix L0 are distinct. The inverse matrix G0 (๐ฅ) = Uโ1 0 (๐ฅ) satis๏ฌes the relation ยฏ ยฏ (2.11) Gโ0 (๐ฅ) = {pโ1 ๐โ๐1 ๐ฅ , . . . , pโ๐ ๐โ๐๐ ๐ฅ }. 2.2. Next we consider the nonhomogeneous equation ๐ (๐ฅ) + f (๐ฅ), ๐ โฒ (๐ฅ) = L(๐ฅ)๐
f (๐ฅ) = (๐1 (๐ฅ), . . . , ๐๐ (๐ฅ))๐ก ,
(2.12)
where the vector-valued function f (๐ฅ) has compact support. Let us use the standard method of variation of arbitrary constants and set ๐ (๐ฅ) = U(๐ฅ)q(๐ฅ),
q(๐ฅ) = (๐1 (๐ฅ), . . . , ๐๐ (๐ฅ))๐ก ,
Trace Formula so that ๐ (๐ฅ) =
๐ โ
547
๐๐ (๐ฅ)u๐ (๐ฅ).
(2.13)
๐=1
Here U(๐ฅ) is an arbitrary admissible fundamental matrix (2.3). Then it follows from equation (2.4) that qโฒ (๐ฅ) = g(๐ฅ)
where g(๐ฅ) = G(๐ฅ)f (๐ฅ) and G(๐ฅ) = Uโ1 (๐ฅ).
(2.14)
We are looking for a solution of equation (2.12) decaying (exponentially) as โฃ๐ฅโฃ โ โ. It is convenient to accept convention (1.5) on the eigenvalues ๐๐ of the matrix L0 . Set ๐+ = min Re ๐๐ , ๐=1,...,๐
๐โ =
min
๐=๐+1,...,๐
โฃ Re ๐๐ โฃ
(2.15)
and observe that estimates u๐ (๐ฅ) = ๐(๐โ๐ยฑ โฃ๐ฅโฃ ),
๐ฅ โ โโ,
hold for ๐ = 1, . . . , ๐ and the upper sign as well as for ๐ = ๐ + 1, . . . , ๐ and the lower sign. Taking into account (2.13), we see that we have to solve equation (2.14) for di๏ฌerent components ๐๐ (๐ฅ) of q(๐ฅ) by di๏ฌerent formulas. Namely, we set โซ โ ๐๐ (๐ฆ)๐๐ฆ, ๐ = 1, . . . , ๐, ๐๐ (๐ฅ) = โ ๐ฅ โซ ๐ฅ ๐๐ (๐ฅ) = ๐๐ (๐ฆ)๐๐ฆ, ๐ = ๐ + 1, . . . , ๐, โโ
where ๐๐ (๐ฅ) are components of g(๐ฅ). This leads to the following result. Proposition 2.2. Let assumption (2.8) hold, and let (2.3) be an arbitrary admissible fundamental matrix. Then the function โซ โ โซ ๐ฅ ๐ ๐ โ โ ๐ (๐ฅ) = โ u๐ (๐ฅ) (G(๐ฆ)f (๐ฆ))๐ ๐๐ฆ + u๐ (๐ฅ) (G(๐ฆ)f (๐ฆ))๐ ๐๐ฆ (2.16) ๐ฅ
๐=1
โโ
๐=๐+1
satis๏ฌes equation (2.12) and ๐ (๐ฅ) = ๐(๐โ๐ยฑ โฃ๐ฅโฃ ) as ๐ฅ โ โโ. Formula (2.16) can be rewritten as โซ โ ๐ (๐ฅ) = R(๐ฅ, ๐ฆ)f (๐ฆ)๐๐ฆ โโ
(2.17)
where the matrix-valued resolvent kernel (or the Green function) R(๐ฅ, ๐ฆ) = {๐
๐,๐ (๐ฅ, ๐ฆ)} is de๏ฌned by the equality ๐
๐,๐ (๐ฅ, ๐ฆ) = โ
๐ โ ๐=1
๐ข๐,๐ (๐ฅ)๐๐,๐ (๐ฆ)๐(๐ฆ โ ๐ฅ) +
๐ โ ๐=๐+1
๐ข๐,๐ (๐ฅ)๐๐,๐ (๐ฆ)๐(๐ฅ โ ๐ฆ).
(2.18)
ยจ J. Ostensson and D.R. Yafaev
548
Here ๐ is the Heaviside function, i.e., ๐(๐ฅ) = 1 for ๐ฅ โฅ 0 and ๐(๐ฅ) = 0 for ๐ฅ < 0, and ๐๐,๐ are elements of the matrix G. In the matrix notation formula (2.18) means that R(๐ฅ, ๐ฆ) = โU(๐ฅ)P+ Uโ1 (๐ฆ)๐(๐ฆ โ ๐ฅ) + U(๐ฅ)Pโ Uโ1 (๐ฆ)๐(๐ฅ โ ๐ฆ),
(2.19)
where the projections Pยฑ are de๏ฌned in the representation โ๐ = โ๐ โ โ๐ โ๐ by the block matrices ) ) ( ( ๐ผ๐ 0 0 0 , Pโ = P+ = . 0 0 0 ๐ผ๐ โ๐ Expressions (2.18) or (2.19) do not of course depend on the choice of bases ห 1 (๐ฅ), . . . , u ห ๐ (๐ฅ) in the spaces K+ and Kโ . Indeed, if we choose other bases u ห ๐+1 (๐ฅ), . . . , u ห ๐ (๐ฅ), then the corresponding admissible fundamental matriand u ห ห ces U(๐ฅ) and U(๐ฅ) are related by the formula U(๐ฅ) = U(๐ฅ)F where the operator ๐ ๐ ห ห โ1 (๐ฆ) = F : โ โ โ commutes with the projections Pยฑ . It follows that U(๐ฅ)P ยฑU โ1 U(๐ฅ)Pยฑ U (๐ฆ). Evidently, the resolvent kernel (2.19) is a continuous function of ๐ฅ and ๐ฆ away from the diagonal ๐ฅ = ๐ฆ and R(๐ฅ, ๐ฅ + 0, ๐ง) = โU(๐ฅ)P+ Uโ1 (๐ฅ), R(๐ฅ, ๐ฅ โ 0, ๐ง) = U(๐ฅ)Pโ Uโ1 (๐ฅ). It follows that R(๐ฅ, ๐ฅ โ 0, ๐ง) โ R(๐ฅ, ๐ฅ + 0, ๐ง) = I,
(2.20)
where I is the ๐ ร ๐ identity matrix. 2.3. The results of the previous subsection admit a simple operator interpretation. Consider the space ๐ฟ2 (โ; โ๐ ) and de๏ฌne the operator H0 on the Sobolev class H1 (โ; โ๐ ) by the formula H0 = โI โ L0 ,
โ = ๐/๐๐ฅ.
If u(๐ฅ) = ๐ข(๐ฅ)p๐ where ๐ข โ H1 (โ), then (H0 u)(๐ฅ) = (๐ขโฒ (๐ฅ) โ ๐๐ ๐ข(๐ฅ))p๐ , and hence the operator H0 is linearly equivalent to a direct sum of the operators of multiplication by ๐๐ โ ๐๐ , ๐ โ โ, ๐ = 1, . . . , ๐ , acting in the space ๐ฟ2 (โ). It follows that the spectrum of the operator H0 consists of straight lines passing through all points โ๐๐ and parallel to the imaginary axis. In particular, the inverse operator Hโ1 0 exists and is bounded. To de๏ฌne the operator H = โI โ L0 โ V(๐ฅ), we need the following well-known assertion (see paper [3] by M.Sh. Birman). Lemma 2.3. Let ๐ : ๐ฟ2 (โ; ๐๐ฅ) โ ๐ฟ2 (โ; ๐๐) be an integral operator with kernel ๐ก(๐, ๐ฅ) = ๐(๐)๐โ๐๐ฅ๐ ๐ฃ(๐ฅ).
(2.21)
Trace Formula If ๐(๐) = (๐ 2 + 1)โ1/2 and
โซ lim
โฃ๐ฅโฃโโ
๐ฅ+1
549
โฃ๐ฃ(๐ฆ)โฃ2 ๐๐ฆ = 0,
๐ฅ
(2.22)
then the operator ๐ is compact. If the coe๏ฌcients of the matrix V(๐ฅ) satisfy condition (2.22), then according to Lemma 2.3 the operator VHโ1 0 is compact. Hence the operator H is closed on H1 (โ; โ๐ ) and by virtue of the Weyl theorem essential spectra of the operators H and H0 coincide. Condition (2.8) implies that 0 is not an eigenvalue of H so that the inverse operator Hโ1 exists and is bounded. If the matrix-valued function V(๐ฅ) has compact support, then according to Proposition 2.2 the integral kernel of the operator Hโ1 is given by formula (2.19). 2.4. Let the solutions u๐ (๐ฅ) of equation (2.2) be distinguished by conditions (2.7). Let us give expressions for the Wronskian W(๐ฅ) := det U(๐ฅ) in terms of transition matrices Tยฑ de๏ฌned as follows. For ๐ = 1, . . . , ๐ and ๐ฅ >> 0 or ๐ = ๐ + 1, . . . , ๐ and ๐ฅ << 0, we have ๐ โ ๐ก๐,๐ p๐ ๐๐๐ ๐ฅ (2.23) u๐ (๐ฅ) = with some coe๏ฌcients ๐ก๐,๐ . Set
๐=1
โ
๐ก1,1 โ ๐ก2,1 โ T+ = โ . โ ..
๐ก๐,1
and
๐ก1,2 ๐ก2,2 .. .
๐ก๐,2
... ... .. .
โ ๐ก1,๐ ๐ก2,๐ โ โ .. โ . โ
(2.24)
. . . ๐ก๐,๐
โ . . . ๐ก๐+1,๐ . . . ๐ก๐+2,๐ โ โ (2.25) .. โ . .. . . โ ๐ก๐,๐+2 . . . ๐ก๐,๐ ๐ก๐,๐+1 Consider, for example, T+ . Using expressions (2.23) for ๐ = 1, . . . , ๐, we see that for ๐ฅ >> 0 matrix (2.3) equals } {๐ ๐ โ โ ๐๐ ๐ฅ ๐๐ ๐ฅ ๐๐+1 ๐ฅ ๐๐ ๐ฅ ๐ก1,๐ p๐ ๐ , . . . , ๐ก๐,๐ p๐ ๐ , p๐+1 ๐ , . . . , p๐ ๐ . (2.26) U(๐ฅ) = โ
๐ก๐+1,๐+1 โ๐ก๐+2,๐+1 โ Tโ = โ .. โ .
๐=1
๐ก๐+1,๐+2 ๐ก๐+2,๐+2 .. .
๐=1
Below, by the calculation of determinants of matrices, we systematically use that one can add to each column another column multiplied by an arbitrary number. In particular, we have } { ๐ ๐ โ โ ๐๐ ๐ฅ ๐๐ ๐ฅ ๐๐+1 ๐ฅ ๐๐ ๐ฅ W(๐ฅ) = det ๐ก1,๐ p๐ ๐ , . . . , ๐ก๐,๐ p๐ ๐ , p๐+1 ๐ , . . . , p๐ ๐ ๐=1
= det T+ W0 (๐ฅ),
๐=1
๐ฅ >> 0,
(2.27)
ยจ J. Ostensson and D.R. Yafaev
550
where the free Wronskian W0 (๐ฅ) is given by formula (2.10). In view of relation (2.6), it follows that for all ๐ฅ โ โ โซ โ ( ) W(๐ฅ) = exp tr L0 ๐ฅ โ tr V(๐ฆ)๐๐ฆ det T+ det{p1 , . . . , p๐ }. (2.28) ๐ฅ
Quite similarly, using expressions (2.23) for ๐ = ๐ + 1, . . . , ๐ and ๐ฅ << 0, we obtain that W(๐ฅ) = det Tโ W0 (๐ฅ), ๐ฅ << 0, (2.29) and โซ ๐ฅ ( ) W(๐ฅ) = exp tr L0 ๐ฅ + tr V(๐ฆ)๐๐ฆ det Tโ det{p1 , . . . , p๐ }, โ๐ฅ โ โ. (2.30) โโ
This leads to the following result. Proposition 2.4. Let U(๐ฅ) be the fundamental matrix (2.3) where u๐ (๐ฅ) are the solutions of equation (2.2) satisfying conditions (2.7). Let the transition matrices Tยฑ be de๏ฌned by formulas (2.23)โ(2.25). Then the Wronskian W(๐ฅ) = det U(๐ฅ) admits representations (2.28) and (2.30). Putting together equalities (2.28) and (2.30), we see that (โซ โ ) det T+ = exp tr V(๐ฆ)๐๐ฆ det Tโ . โโ
Assumptions (2.8) and det Tยฑ โ= 0 are of course equivalent.
3. Dual problem Some properties of admissible fundamental matrices become more transparent if one considers the dual problem corresponding to the matrix-valued function ห L(๐ฅ) = โLโ (๐ฅ). 3.1. It follows from equation (2.4) for U(๐ฅ) that the inverse operator G(๐ฅ) = Uโ1 (๐ฅ) satis๏ฌes the equation Gโฒ (๐ฅ) = โG(๐ฅ)Uโฒ (๐ฅ)G(๐ฅ) = โG(๐ฅ)L(๐ฅ)
(3.1)
which yields the equation
ห โฒ (๐ฅ) = L(๐ฅ) ห U(๐ฅ) ห U (3.2) โ โ1 ห ห for the matrix-valued function U(๐ฅ) := U (๐ฅ) . Clearly, det U(๐ฅ) โ= 0 so that ห U(๐ฅ) is a fundamental matrix for this equation. Proposition 3.1 below shows that ห it is admissible. Set U(๐ฅ) = {ห ๐ข๐,๐ (๐ฅ)}, G(๐ฅ) = {๐๐,๐ (๐ฅ)}. We use below that ๐ข ห๐,๐ (๐ฅ) = ๐๐,๐ (๐ฅ) = (โ1)๐+๐ ๐๐,๐ (๐ฅ)/W(๐ฅ),
(3.3)
where ๐๐,๐ (๐ฅ) is the minor of the matrix U(๐ฅ) which is the determinant of the matrix cut down from U(๐ฅ) by removing the row with index ๐ and the column with index ๐.
Trace Formula
551
Proposition 3.1. Let u๐ (๐ฅ) be arbitrary linear independent solutions of equation (2.2) from K+ for ๐ = 1, . . . , ๐ and from Kโ for ๐ = ๐ + 1, . . . , ๐ , and let U(๐ฅ) be the corresponding admissible fundamental matrix (2.3). Then for all ๐ = 1, . . . , ๐ we have ๐๐,๐ (๐ฅ) = ๐(๐๐โ ๐ฅ ), ๐๐,๐ (๐ฅ) = ๐(๐
โ๐+ ๐ฅ
),
๐ฅ โ โโ, ๐ฅ โ +โ,
๐ = ๐ + 1, . . . , ๐,
(3.4)
๐ = 1, . . . , ๐,
with positive numbers ๐ยฑ de๏ฌned in (2.15). Proof. Let us prove, for example, the ๏ฌrst of relations (3.4). Changing if necessary the numeration, we can suppose that ๐ = ๐ which is notationally convenient. For ๐ฅ << 0 and some numbers ๐๐,๐ , we have { ๐ } ๐ ๐ ๐ โ โ โ โ U(๐ฅ) = ๐1,๐ p๐ ๐๐๐ ๐ฅ , . . . , ๐๐,๐ p๐ ๐๐๐ ๐ฅ , ๐๐+1,๐ p๐ ๐๐๐ ๐ฅ , . . . , ๐๐,๐ p๐ ๐๐๐ ๐ฅ . ๐=1
๐=1
It follows that
{
๐๐,๐ (๐ฅ) = det
๐ โ ๐=1
(๐)
๐1,๐ p๐ ๐๐๐ ๐ฅ , . . . ,
๐ โ ๐=1
where
๐=1
๐=1
๐ โ ๐=1
(๐) ๐๐+1,๐ p๐ ๐๐๐ ๐ฅ , . . . ,
(๐)
๐๐,๐ p๐ ๐๐๐ ๐ฅ ,
๐ โ ๐=1
} (๐) ๐๐ โ1,๐ p๐ ๐๐๐ ๐ฅ
(๐)
p๐ = (๐1,๐ , . . . , ๐๐โ1,๐ , ๐๐+1,๐ , . . . , ๐๐,๐ )๐ก .
(3.5)
(3.6)
Thus the element with index ๐ is removed from p๐ = (๐1,๐ , . . . , ๐๐,๐ )๐ก so that in (3.5) we take the determinant of the (๐ โ 1) ร (๐ โ 1) matrix. Neglecting columns which are repeated in determinant (3.5), we see that it consists of terms } { (๐) (๐) (๐) ๐๐ ๐ฅ (3.7) det p1 ๐๐1 ๐ฅ , . . . , p(๐) , p๐๐+1 ๐๐๐๐+1 ๐ฅ , . . . , p๐๐ โ1 ๐๐๐๐ โ1 ๐ฅ ๐ ๐ multiplied by some coe๏ฌcients which do not depend on ๐ฅ. Here the indices ๐๐+1 , . . . , ๐๐ โ1 take the values ๐ + 1, . . . , ๐ and ๐๐ โ= ๐๐ if ๐ โ= ๐. Evidently, expression (3.7) equals ๐ โ ( ) ๐๐ , (3.8) ๐๐ exp tr L0 ๐ฅ โ ๐๐ ๐ฅ , tr L0 = ๐=1
for some ๐ = ๐ + 1, . . . , ๐ and a number ๐๐ which does not depend on ๐ฅ. In view of formulas (2.10) and (2.29) after division by W(๐ฅ) this expression is ๐(โฃ๐โ๐๐ ๐ฅ โฃ) as ๐ฅ โ โโ. Hence (3.4) for ๐ = ๐ follows from (3.3). โก ห ห ๐ (๐ฅ) be columns of the matrix U(๐ฅ), that is, Let u ห ห ๐ (๐ฅ)}. U(๐ฅ) = {ห u1 (๐ฅ), . . . , u
ยจ J. Ostensson and D.R. Yafaev
552
Then relations (3.4) can equivalently be rewritten as ห ๐ (๐ฅ) = ๐(๐๐โ ๐ฅ ), u ห ๐ (๐ฅ) = ๐(๐ u
โ๐+ ๐ฅ
),
๐ฅ โ โโ,
๐ = ๐ + 1, . . . , ๐,
๐ฅ โ +โ,
๐ = 1, . . . , ๐.
(3.9)
ห Let us de๏ฌne the resolvent kernel R(๐ฅ, ๐ฆ) in the same way as in subs. 2.2 ห with L(๐ฅ) replaced by L(๐ฅ). According to relations (3.9) the ๏ฌrst ๐ columns of the ห matrix U(๐ฅ) exponentially decay as ๐ฅ โ +โ and its last ๐ โ ๐ columns expoห nentially decay as ๐ฅ โ โโ. Therefore the fundamental matrix U(๐ฅ) is admissible for equation (3.2), and it follows from formula (2.19) applied to the dual problem that หโ U ห+ U ห ห โ1 (๐ฆ)๐(๐ฆ โ ๐ฅ) + U(๐ฅ) ห โ1 (๐ฆ)๐(๐ฅ โ ๐ฆ) ห ห P P R(๐ฅ, ๐ฆ) = โU(๐ฅ)
(3.10)
where ห U(๐ฅ) = Uโ (๐ฅ)โ1 ,
ห ยฑ = Pโ . P
(3.11)
In particular, we see that ห R(๐ฅ, ๐ฆ) = โRโ (๐ฆ, ๐ฅ). This relation implicitly follows also from the results of subs. 2.3. (0)
3.2. Let G0 (๐ฅ) = {๐๐,๐ (๐ฅ)} be the matrix inverse to the free matrix (2.9). The next result supplements Proposition 3.1 and plays an important role in our proof of the trace formula (1.11). Proposition 3.2. Let solutions u๐ (๐ฅ) of equation (2.2) be distinguished by conditions (2.7), and let U(๐ฅ) be the corresponding admissible fundamental matrix (2.3). Then for all ๐ = 1, . . . , ๐ elements ๐๐,๐ (๐ฅ) of the matrix G(๐ฅ) = Uโ1 (๐ฅ) satisfy the relations (0)
๐ฅ โ +โ,
๐ = ๐ + 1, . . . , ๐,
(0)
๐ฅ โ โโ,
๐ = 1, . . . , ๐,
๐๐,๐ (๐ฅ) โ ๐๐,๐ (๐ฅ) = ๐(๐โ๐+ ๐ฅ ), ๐๐,๐ (๐ฅ) โ ๐๐,๐ (๐ฅ) = ๐(๐๐โ ๐ฅ ),
(3.12)
with positive numbers ๐ยฑ de๏ฌned in (2.15). Proof. Let us use the notation introduced in the proof of Proposition 3.1. We shall again prove the ๏ฌrst of relations (3.12) for ๐ = ๐ . According to (2.26) for ๐ฅ >> 0, we have {๐ } ๐ โ โ (๐) ๐๐ ๐ฅ (๐) ๐๐ ๐ฅ (๐) (๐) ๐๐+1 ๐ฅ ๐๐ โ1 ๐ฅ ๐ก1,๐ p๐ ๐ , . . . , ๐ก๐,๐ p๐ ๐ , p๐+1 ๐ , . . . , p๐ โ1 ๐ ๐๐,๐ (๐ฅ) = det ๐=1
๐=1
(3.13) (๐) where p๐ is vector (3.6) obtained from p๐ by removing the component with index ๐. Neglecting columns which are repeated in (๐ โ 1) ร (๐ โ 1) matrix (3.13), we
Trace Formula see that
{
๐๐,๐ (๐ฅ) = det
๐ โ ๐=1
๐ โ ๐=1
(๐)
553
(๐)
๐ก1,๐ p๐ ๐๐๐ ๐ฅ + ๐ก1,๐ p๐ ๐๐๐ ๐ฅ , . . . , }
(๐) ๐ก๐,๐ p๐ ๐๐๐ ๐ฅ
+
(๐) (๐) (๐) ๐ก๐,๐ p๐ ๐๐๐ ๐ฅ , p๐+1 ๐๐๐+1 ๐ฅ , . . . , p๐ โ1 ๐๐๐ โ1 ๐ฅ
.
This determinant is the sum of the term { ๐ } ๐ โ โ (๐) ๐๐ ๐ฅ (๐) ๐๐ ๐ฅ (๐) (๐) ๐๐+1 ๐ฅ ๐๐ โ1 ๐ฅ det ๐ก1,๐ p๐ ๐ , . . . , ๐ก๐,๐ p๐ ๐ , p๐+1 ๐ , . . . , p๐ โ1 ๐ (3.14) ๐=1
๐=1
and of the ๐ terms { ๐ ๐ โ โ (๐) (๐) (๐) (๐) ๐ก2,๐ p๐ ๐๐๐ ๐ฅ , . . . , ๐ก๐,๐ p๐ ๐๐๐ ๐ฅ , p๐+1 ๐๐๐+1 ๐ฅ , . . . det ๐ก1,๐ p๐ ๐๐๐ ๐ฅ , } ๐=1 ๐=1 (๐) ๐๐ โ1 ๐ฅ . . . , p๐ โ1 ๐ ,
det
{โ ๐ ๐=1
โ
โ
โ
(๐)
๐ก1,๐ p๐ ๐๐๐ ๐ฅ , . . . ,
๐ โ ๐=1
(๐)
(3.15) (๐)
(๐)
๐ก๐โ1,๐ p๐ ๐๐๐ ๐ฅ , ๐ก๐,๐ p๐ ๐๐๐ ๐ฅ , p๐+1 ๐๐๐+1 ๐ฅ , . . . (๐) . . . , p๐ โ1 ๐๐๐ โ1 ๐ฅ
For the free case where V(๐ฅ) = 0, we have the exact equality { (๐) } (0) (๐) (๐) ๐๐ ๐ฅ ๐๐,๐ (๐ฅ) = det p1 ๐๐1 ๐ฅ , . . . , p(๐) , p๐+1 ๐๐๐+1 ๐ฅ , . . . , p๐ โ1 ๐๐๐ โ1 ๐ฅ . ๐ ๐ Similarly to (2.27), we ๏ฌnd that determinant (3.14) equals } { (๐) (๐) (๐) ๐๐ ๐ฅ , p๐+1 ๐๐๐+1 ๐ฅ , . . . , p๐ โ1 ๐๐๐ โ1 ๐ฅ . det T+ det p1 ๐๐1 ๐ฅ , . . . , p(๐) ๐ ๐
} .
(3.16)
(3.17)
By virtue of (2.27) expression (3.17) divided by W(๐ฅ) equals expression (3.16) divided by W0 (๐ฅ). Thus for the proof of asymptotics (3.12) it remains to estimate determinants (3.15) by ๐ถ๐โ๐+ ๐ฅ . This is similar to the proof of Proposition 3.1. It su๏ฌces to consider terms corresponding to di๏ฌerent values of index ๐ in di๏ฌerent sums. Therefore, up to some factors not depending on ๐ฅ, determinants (3.15) consist of the terms { (๐) } (๐) (๐) (๐) (๐) det p๐ ๐๐๐ ๐ฅ , p๐1 ๐๐๐1 ๐ฅ , . . . , p๐๐โ1 ๐๐๐๐โ1 ๐ฅ , p๐+1 ๐๐๐+1 ๐ฅ , . . . , p๐ โ1 ๐๐๐ โ1 ๐ฅ (3.18) where the indices ๐1 , . . . , ๐๐โ1 take the values 1, . . . , ๐ and ๐๐ โ= ๐๐ if ๐ โ= ๐. Evidently, (3.18) equals expression (3.8) for some ๐ = 1, . . . , ๐ and a number ๐๐ which does not depend on ๐ฅ. In view of formulas (2.10) and (2.27) after division by W(๐ฅ) this expression is ๐(๐โ๐+ ๐ฅ ) as ๐ฅ โ +โ. This concludes the proof of asymptotics (3.12). โก
ยจ J. Ostensson and D.R. Yafaev
554
ห ห ๐ (๐ฅ) of the matrix U(๐ฅ) In view of equality (2.11) in terms of columns u relations (3.12) can equivalently be rewritten as (cf. (3.9)) ยฏ
ห ๐ (๐ฅ) โ pโ๐ ๐โ๐๐ ๐ฅ = ๐(๐โ๐+ ๐ฅ ), u ห ๐ (๐ฅ) โ u
ยฏ pโ๐ ๐โ๐๐ ๐ฅ
= ๐(๐๐โ ๐ฅ ),
๐ฅ โ +โ,
๐ = ๐ + 1, . . . , ๐,
๐ฅ โ โโ,
๐ = 1, . . . , ๐.
(3.19)
ยฏ
We emphasize that the functions pโ๐ ๐โ๐๐ ๐ฅ exponentially grow at in๏ฌnity while the remainders in formulas (3.19) exponentially decay. 3.3. Let us, ๏ฌnally, verify that the resolvent kernel de๏ฌned by equality (2.19) satis๏ฌes a natural estimate โฃR(๐ฅ, ๐ฆ)โฃ โค ๐ถ๐โ๐โฃ๐ฅโ๐ฆโฃ ,
๐ = min{๐+ , ๐โ },
(3.20)
1
with a constant ๐ถ not depending on ๐ฅ and ๐ฆ. It su๏ฌces to check that โฃU(๐ฅ)Pยฑ Uโ1 (๐ฆ)โฃ โค ๐ถ๐โ๐ยฑ โฃ๐ฅโ๐ฆโฃ ,
ยฑ(๐ฆ โ ๐ฅ) โฅ 0.
(3.21)
We can suppose that U(๐ฅ) is de๏ฌned by formula (2.3) where the solutions u๐ (๐ฅ) of equation (2.2) satisfy condition (2.7). Note that โฃU(๐ฅ)Pยฑ โฃ โค ๐ถ๐โ๐ยฑ โฃ๐ฅโฃ , โ๐ฅ โฅ 0. (3.22) Passing here to the dual problem and using that ๐หยฑ = ๐โ , we see that ห โ โฃ โค ๐ถ๐โ๐ยฑ โฃ๐ฆโฃ , ห P โฃPยฑ Uโ1 (๐ฆ)โฃ = โฃUโ (๐ฆ)โ1 Pยฑ โฃ = โฃU(๐ฆ)
ยฑ๐ฆ โฅ 0.
(3.23)
Let us prove estimate (3.21), for example, for the upper sign. If ๐ฅ โค 0 and ๐ฆ โฅ 0, then (3.21) directly follows from (3.22) and (3.23). Suppose next that ๐ฅ โค 0 and ๐ฆ โค 0. According to formula (2.18) it su๏ฌces to check that โฃ๐ข๐,๐ (๐ฅ)๐๐,๐ (๐ฆ)โฃ โค ๐ถโฃ๐๐๐ (๐ฅโ๐ฆ) โฃ (3.24) for all ๐, ๐ = 1, . . . , ๐ and ๐ = 1, . . . , ๐. Recall that โฃ๐ข๐,๐ (๐ฅ)โฃ โค ๐ถโฃ๐๐๐ ๐ฅ โฃ and that โฃ๐๐,๐ (๐ฆ)โฃ โค ๐ถโฃ๐โ๐๐ ๐ฆ โฃ by Proposition 3.2. This yields (3.24) and hence (3.21) for ๐ฅ โค 0, ๐ฆ โค 0 and the upper sign. Similarly, we obtain that estimate (3.21) is true for ๐ฅ โฅ 0, ๐ฆ โฅ 0 and the lower sign. Using this estimate for the dual problem we see that หโU ห ห โ1 (๐ฆ)โฃ โค ๐ถ๐โห๐โ โฃ๐ฅโ๐ฆโฃ , โฃU(๐ฅ) P
๐ฅ โฅ 0, ๐ฆ โฅ 0, ๐ฅ โฅ ๐ฆ.
Passing here to adjoint matrices and taking into account relations (3.11), we ๏ฌnd that โฃU(๐ฆ)P+ Uโ1 (๐ฅ)โฃ โค ๐ถ๐โ๐+ โฃ๐ฅโ๐ฆโฃ , ๐ฅ โฅ 0, ๐ฆ โฅ 0, ๐ฅ โฅ ๐ฆ. Interchanging ๐ฅ and ๐ฆ, we get (3.21) for ๐ฅ โฅ 0, ๐ฆ โฅ 0 and the upper sign. Thus we have proven the following result. Proposition 3.3. Let assumption (2.8) hold. Then the resolvent kernel (2.19) satis๏ฌes estimate (3.20). 1 Here
and below ๐ถ are di๏ฌerent positive constants whose precise values are of no importance.
Trace Formula
555
Estimate (3.20) shows that formula (2.17) obtained for functions f with compact support extends to all f โ ๐ฟ2 (โ; โ๐ ).
4. Trace formula 4.1. Let us consider the di๏ฌerential operator (1.1) acting in the space ๐ฟ2 (โ). Recall that the operator ๐ป0 = ๐โ๐ โ ๐ is self-adjoint in the space ๐ฟ2 (โ) on domain ๐(๐ป0 ) which is the Sobolev class H๐ (โ). The spectrum ๐(๐ป0 ) of ๐ป0 coincides with [0, โ) for ๐ even and with โ for ๐ odd. If the coe๏ฌcients ๐ฃ๐ , ๐ = 1, . . . , ๐ , of operator (1.14) belong locally to ๐ฟ2 , then the operator ๐ป = ๐ป0 + ๐ is well de๏ฌned by formula (1.1) at least on the class ๐ถ0โ (โ). If all functions ๐ฃ๐ satisfy assumption (2.22), then according to Lemma 2.3 the operator ๐ ๐
0 (๐ง), ๐ง โโ ๐(๐ป0 ), is compact. Therefore the operator ๐ป is closed on domain ๐(๐ป) = ๐(๐ป0 ), and by virtue of the Weyl theorem its essential spectrum ๐ess (๐ป) = ๐ess (๐ป0 ). The spectrum ๐(๐ป) of the operator ๐ป in โ โ ๐ess (๐ป0 ) consists of eigenvalues (not necessarily real) of ๏ฌnite multiplicity which might accumulate to ๐ess (๐ป0 ) only. For the construction of the integral kernel of the resolvent ๐
(๐ง) = (๐ป โ ๐ง)โ1 of the operator ๐ป, we have to solve the equation ๐โ๐ ๐(๐ ) (๐ฅ) + (๐ ๐)(๐ฅ) = ๐ง๐(๐ฅ) + ๐ (๐ฅ),
๐ง โโ ๐(๐ป),
๐ โ ๐ฟ2 (โ).
(4.1)
Let us rewrite it as a system of di๏ฌerential equations (2.12). We introduce vectors ๐ = (๐1 , . . . , ๐๐ )๐ก with components ๐๐ = ๐(๐โ1) , f = ๐๐ (0, 0, . . . , 0, ๐ )๐ก and set โ โ 0 1 ... 0 0 โ 0 0 . . . 0 0โ โ โ โ .. .. .. โ , .. . . L0 (๐ง) = โ . (4.2) . . .โ . โ โ โ โ 0 0 ... 0 1 ๐๐ ๐ง 0 . . . 0 0 โ โ 0 0 ... 0 0 โ 0 0 ... 0 0 โ โ โ โ .. .. .. โ . .. (4.3) V(๐ฅ) = โ๐๐ โ ... . . . . โ โ โ โ โ 0 0 ... 0 0 ๐ฃ1 (๐ฅ) ๐ฃ2 (๐ฅ) . . . ๐ฃ๐ โ1 (๐ฅ) ๐ฃ๐ (๐ฅ) Then equation (4.1) is equivalent to vector equation (2.12) with the operator L(๐ฅ, ๐ง) de๏ฌned by equality (2.1). We emphasize that it now depends on the spectral parameter ๐ง. Matrix (4.2) has eigenvalues ๐๐ such that ๐๐๐ = ๐๐ ๐ง with the corresponding eigenvectors p๐ (๐ง) = (1, ๐๐ , . . . , ๐๐๐ โ1 )๐ก .
(4.4)
It is easy to see that ๐ = ๐/2 if ๐ is even and ๐ = (๐ โ 1)/2 for Im ๐ง > 0 and ๐ = (๐ + 1)/2 for Im ๐ง < 0 if ๐ is odd.
ยจ J. Ostensson and D.R. Yafaev
556
If the coe๏ฌcients ๐ฃ๐ (๐ฅ) have compact supports, then all results of Sections 2 and 3 apply. Now we have u๐ (๐ฅ, ๐ง) = (๐ข1,๐ (๐ฅ, ๐ง), . . . , ๐ข๐,๐ (๐ฅ, ๐ง))๐ก where the func(๐โ1) (๐ฅ, ๐ง) = tions ๐ข๐ (๐ฅ, ๐ง) := ๐ข1,๐ (๐ฅ, ๐ง) satisfy homogeneous equation (1.4) and ๐ข๐ ๐ข๐,๐ (๐ฅ, ๐ง) for ๐ = 2, . . . , ๐ . Hence fundamental matrix (2.3) takes form (1.3). This matrix U(๐ฅ, ๐ง) is admissible if the functions ๐ข1 (๐ฅ, ๐ง), . . . , ๐ข๐ (๐ฅ, ๐ง) belong to ๐ฟ2 (โโ ) and the functions ๐ข๐+1 (๐ฅ, ๐ง), . . . , ๐ข๐ (๐ฅ, ๐ง) belong to ๐ฟ2 (โ+ ). The Wronskian W(๐ฅ, ๐ง) = det U(๐ฅ, ๐ง) satis๏ฌes equations (2.5) and (2.6) where tr L(๐ฅ, ๐ง) = โ๐๐ ๐ฃ๐ (๐ฅ). This yields (1.9). Formula (2.17) means that โซ โ ๐(๐โ1) (๐ฅ) = ๐๐ ๐
๐,๐ (๐ฅ, ๐ฆ, ๐ง)๐ (๐ฆ)๐๐ฆ. โโ
In particular, we have ๐
๐,๐ (๐ฅ, ๐ฆ, ๐ง) = โ๐ฅ๐โ1 ๐
1,๐ (๐ฅ, ๐ฆ, ๐ง),
๐ = 2, . . . , ๐.
(4.5)
Condition (2.8) is equivalent to the assumption that ๐ง is not an eigenvalue of the operator ๐ป. Thus, Proposition 2.2 entails the following result. Proposition 4.1. Suppose that ๐ง โโ ๐(๐ป). Let the matrix R(๐ฅ, ๐ฆ, ๐ง) = {๐
๐,๐ (๐ฅ, ๐ฆ, ๐ง)} be de๏ฌned by formula (2.19) where U(๐ฅ, ๐ง) is an admissible fundamental matrix. Then the resolvent ๐
(๐ง) = (๐ป โ ๐ง)โ1 of the operator ๐ป is the integral operator with kernel ๐
(๐ฅ, ๐ฆ, ๐ง) = ๐๐ ๐
1,๐ (๐ฅ, ๐ฆ, ๐ง). According to formula (2.18) Proposition 4.1 implies that ๐
(๐ฅ, ๐ฆ, ๐ง) = โ ๐๐
๐ โ
๐ข1,๐ (๐ฅ, ๐ง)๐๐,๐ (๐ฆ, ๐ง),
๐ฅ < ๐ฆ,
๐=1
๐
(๐ฅ, ๐ฆ, ๐ง) =๐
๐
๐ โ
(4.6) ๐ข1,๐ (๐ฅ, ๐ง)๐๐,๐ (๐ฆ, ๐ง),
๐ฅ > ๐ฆ,
๐=๐+1
It follows from relations (2.20) and (4.5) that, for ๐ โฅ 2, the function (๐) ๐
(๐ฅ, ๐ฆ, ๐ง) as well as its derivatives ๐
๐ฅ (๐ฅ, ๐ฆ, ๐ง), ๐ = 1, . . . , ๐ โ 2, are continuous functions of ๐ฅ and ๐ฆ while ๐
๐ฅ(๐ โ1) (๐ฅ, ๐ฅ โ 0, ๐ง) โ ๐
๐ฅ(๐ โ1) (๐ฅ, ๐ฅ + 0, ๐ง) = ๐๐ . The case ๐ = 1 is trivial. Although in this case the kernel ๐
(๐ฅ, ๐ฆ, ๐ง) is not a continuous function, the di๏ฌerence ๐
(๐ฅ, ๐ฆ, ๐ง) โ ๐
0 (๐ฅ, ๐ฆ, ๐ง) is continuous and its diagonal values equal zero. 4.2. Let us ๏ฌnd a convenient expression for integral (1.19). Since V does not depend on ๐ง, di๏ฌerentiating equation (2.4) in ๐ง, we have ห ห โฒ (๐ฅ, ๐ง) = L(๐ฅ, ๐ง)U(๐ฅ, ๐ง) + Lห 0 (๐ง)U(๐ฅ, ๐ง). U
Trace Formula
557
Here and below the dot stands for the derivative in ๐ง. Using formula (3.1), we now see that ( ) ห ห ห โฒ (๐ฅ, ๐ง) ๐ G(๐ฅ, ๐ง)U(๐ฅ, ๐ง) /๐๐ฅ = Gโฒ (๐ฅ, ๐ง)U(๐ฅ, ๐ง) + G(๐ฅ, ๐ง)U = G(๐ฅ, ๐ง)Lห 0 (๐ง)U(๐ฅ, ๐ง) =: A(๐ฅ, ๐ง). (4.7) Next, we calculate the matrix A(๐ฅ, ๐ง). It โ 0 0 โ .. .. โ Lห 0 (๐ง) = ๐๐ โ . . โ0 0 1 0 and hence
follows from formula (4.2) that โ ... 0 .โ .. . .. โ โ . . . 0โ ... 0
โ
0 โ .. โ B(๐ฅ, ๐ง) := Lห 0 (๐ง)U(๐ฅ, ๐ง) = ๐๐ โ . โ 0 ๐ข1,1
0 .. .
0 ๐ข1,2
โ ... 0 .. โ .. . . โ โ ... 0 โ . . . ๐ข1,๐
where (cf. (1.3) and (2.3)) ๐ข1,๐ = ๐ข๐ . Since A = GB, this yields the relation ๐๐,๐ =
๐ โ
๐๐,๐ ๐๐,๐ = ๐๐,๐ ๐๐,๐ = ๐๐ ๐๐,๐ ๐ข1,๐
(4.8)
๐=1
for the matrix elements ๐๐,๐ = ๐๐,๐ (๐ฅ, ๐ง) and ๐๐,๐ = ๐๐,๐ (๐ฅ, ๐ง) of the matrices A and B. Putting together equalities (4.7) and (4.8), we obtain the relation ๐๐ ๐ข1,๐ (๐ฅ, ๐ง)๐๐,๐ (๐ฅ, ๐ง) =
๐
๐ โ ๐๐,๐ (๐ฅ, ๐ง)๐ขห ๐,๐ (๐ฅ, ๐ง) ๐๐ฅ ๐=1
and, in particular, ๐๐ ๐ข1,๐ (๐ฅ, ๐ง)๐๐,๐ (๐ฅ, ๐ง) =
๐
๐ โ ๐๐,๐ (๐ฅ, ๐ง)๐ขห ๐,๐ (๐ฅ, ๐ง). ๐๐ฅ
(4.9)
๐=1
Comparing formulas (4.6) and (4.9), we get two representations for diagonal values of the resolvent kernel: ๐
๐
(๐ฅ, ๐ฅ, ๐ง) = โ
๐
๐ โโ ๐๐,๐ (๐ฅ, ๐ง)๐ขห ๐,๐ (๐ฅ, ๐ง), ๐๐ฅ ๐=1 ๐=1
๐
(๐ฅ, ๐ฅ, ๐ง) =
๐ ๐ ๐ โ โ ๐๐,๐ (๐ฅ, ๐ง)๐ขห ๐,๐ (๐ฅ, ๐ง). ๐๐ฅ ๐=๐+1 ๐=1
Integrating the ๏ฌrst of these representations over an interval (๐ฅ1 , ๐ฅ) and the second over an interval (๐ฅ, ๐ฅ2 ), we arrive at the following intermediary result.
ยจ J. Ostensson and D.R. Yafaev
558
Proposition 4.2. Under the assumptions of Proposition 4.1 for all ๐ฅ1 , ๐ฅ2 , ๐ฅ โ โ, the representation holds: โซ ๐ฅ2 ๐ ๐ โ โ ๐
(๐ฆ, ๐ฆ, ๐ง)๐๐ฆ = โ ๐๐,๐ (๐ฅ, ๐ง)๐ขห ๐,๐ (๐ฅ, ๐ง) ๐ฅ1
๐=1 ๐=1
+
๐ ๐ โ โ
๐๐,๐ (๐ฅ2 , ๐ง)๐ขห ๐,๐ (๐ฅ2 , ๐ง) +
๐=๐+1 ๐=1
๐ โ ๐ โ
(4.10) ๐๐,๐ (๐ฅ1 , ๐ง)๐ขห ๐,๐ (๐ฅ1 , ๐ง).
๐=1 ๐=1
Let us consider the ๏ฌrst term in the right-hand side of (4.10). Since ) ( ห ๐ง) , ๐ det U(๐ฅ, ๐ง)/๐๐ง = det U(๐ฅ, ๐ง) tr U(๐ฅ, ๐ง)โ1 U(๐ฅ, we see that ๐ ๐ โ โ
( ) ห ห ๐๐,๐ (๐ฅ, ๐ง)๐ขห ๐,๐ (๐ฅ, ๐ง) = tr G(๐ฅ, ๐ง)U(๐ฅ, ๐ง) = W(๐ฅ, ๐ง)โ1 W(๐ฅ, ๐ง)
(4.11)
๐=1 ๐=1
where the function W(๐ฅ, ๐ง) = det U(๐ฅ, ๐ง). Observe that according to (1.9) this expression does not depend on ๐ฅ which is consistent with formula (4.10). For ๐ = 2, identity (4.10) reduces to formula (1.26) of paper [7]. For ๐ > 2, it is probably new. Identity (4.10) allows us to considerably simplify the calculation ( ) of Tr ๐
(๐ง) โ ๐
0 (๐ง) compared to the presentation of book [13] for ๐ = 2. This is essential for an arbitrary ๐ . 4.3. Now we suppose that the solutions u๐ (๐ฅ, ๐ง) of equation (2.2) are distinguished by condition (2.7) which yields condition (1.6) on the solutions ๐ข๐ (๐ฅ, ๐ง) of equation (1.4); W(๐ฅ, ๐ง) is the Wronskian (1.7). Di๏ฌerent objects corresponding to the โfreeโ operator ๐ป0 = ๐โ๐ โ ๐ will be endowed with index 0 (upper or lower). For the (0) free case, we put ๐ข๐ (๐ฅ) = ๐๐๐ ๐ฅ , ๐ = 1, . . . , ๐ . Let U0 (๐ฅ, ๐ง) be the corresponding fundamental matrix (2.9) with the eigenvectors p๐ (๐ง) de๏ฌned by (4.4), and let W0 (๐ง) be its determinant (1.8). The normalized Wronskian ฮ(๐ฅ, ๐ง) is de๏ฌned (0) by formula (1.10). We denote by ๐๐,๐ (๐ฅ, ๐ง) and ๐๐,๐ (๐ฅ, ๐ง) matrix elements of the matrices G(๐ฅ, ๐ง) = Uโ1 (๐ฅ, ๐ง) and G0 (๐ฅ, ๐ง) = Uโ1 0 (๐ฅ, ๐ง), respectively. For the proof of the trace formula, we combine Propositions 3.2 and 4.2. Indeed, let us subtract from equality (4.10) the same equality for the resolvent ๐
0 (๐ง) = (๐ป0 โ ๐ง)โ1 and consider โซ ๐ฅ2 ) ( (4.12) ๐
(๐ฆ, ๐ฆ, ๐ง) โ ๐
0 (๐ฆ, ๐ฆ, ๐ง) ๐๐ฆ. ๐ฅ1
Since ๐๐๐ = ๐๐ ๐ง, for all ๐ = 1, . . . , ๐ , we have (0)
๐ขห ๐,๐ (๐ฅ, ๐ง) = ๐ขห ๐,๐ (๐ฅ, ๐ง) = ๐(๐๐๐โ1 ๐๐๐ ๐ฅ )/๐๐ง = ๐๐ ๐ โ1 ๐๐โ๐ +๐โ1 (๐ โ 1 + ๐ฅ๐๐ )๐๐๐ ๐ฅ if either ๐ = 1, . . . , ๐ and ๐ฅ << 0 or ๐ = ๐ + 1, . . . , ๐ and ๐ฅ >> 0.
Trace Formula
559
According to condition (1.5), it now directly follows from relations (3.12) that, for all ๐ = 1, . . . , ๐ , (0)
(0)
๐๐,๐ (๐ฅ, ๐ง)๐ขห ๐,๐ (๐ฅ, ๐ง) โ ๐๐,๐ (๐ฅ, ๐ง)๐ขห ๐,๐ (๐ฅ, ๐ง) = ๐(๐ฅ๐โ(๐+ +๐โ )โฃ๐ฅโฃ ) if either ๐ = 1, . . . , ๐ and ๐ฅ โ โโ or ๐ = ๐ + 1, . . . , ๐ and ๐ฅ โ +โ. Therefore using equality (4.10) we see that the contribution to (4.12) of the terms depending on ๐ฅ1 disappear in the limit ๐ฅ1 โ โโ and the contribution of the terms depending on ๐ฅ2 disappear in the limit ๐ฅ2 โ +โ. Thus taking into account equality (4.11), we obtain the following result. Theorem 4.3. Suppose that the coe๏ฌcients ๐ฃ1 , . . . , ๐ฃ๐ of the operator ๐ป have compact supports. Then for ๐ง โโ ๐(๐ป), the limit in the left-hand side exists and โซ ๐ฅ2 ) ( ๐
(๐ฆ, ๐ฆ, ๐ง) โ ๐
0 (๐ฆ, ๐ฆ, ๐ง) ๐๐ฆ = โฮ(๐ฅ, ๐ง)โ1 ๐ฮ(๐ฅ, ๐ง)/๐๐ง (4.13) lim ๐ฅ1 โโโ,๐ฅ2 โ+โ
๐ฅ1
where ๐ฅ โ โ is arbitrary. 4.4. The left-hand side of (4.13) can of course be identi๏ฌed with the trace of the di๏ฌerence ๐
(๐ง) โ ๐
0 (๐ง). To that end, we ๏ฌrst verify inclusion (1.18). We proceed from the following well-known result (see, e.g., survey [4] by M.Sh. Birman and M.Z. Solomyak). Proposition 4.4. Suppose that โซ โ โฌ 2 := (1 + ๐ 2 )๐ผ โฃ๐(๐)โฃ2 ๐๐ < โ, โโ
๐ฑ 2 :=
โซ
โ โโ
(1 + ๐ฅ2 )๐ผ โฃ๐ฃ(๐ฅ)โฃ2 ๐๐ฅ < โ (4.14)
for some ๐ผ > 1/2. Then the integral operator ๐ : ๐ฟ2 (โ; ๐๐ฅ) โ ๐ฟ2 (โ; ๐๐) with kernel (2.21) belongs to the trace class ๐1 and its trace norm satis๏ฌes the estimate โฅ๐ โฅ๐1 โค ๐ถโฌ ๐ฑ where the constant ๐ถ depends on ๐ผ > 1/2 only. Now it is easy to prove the following result. Lemma 4.5. Under assumption (1.12) inclusion (1.18) holds. Proof. Let us proceed from the resolvent identity ๐
(๐ง) โ ๐
0 (๐ง) = โ๐
0 (๐ง)๐ ๐
(๐ง) = โ
๐ โ
๐
0 (๐ง)๐ฃ๐ โ ๐โ1 ๐
(๐ง),
๐ง โโ ๐(๐ป),
(4.15)
๐=1
where ๐ฃ๐ is the operator of multiplication by the function ๐ฃ๐ (๐ฅ). The operators โ ๐ ๐
(๐ง) are bounded because the operator ๐ป0 ๐
(๐ง) is bounded. Proposition 4.4 implies that ๐
0 (๐ง)๐ฃ๐ โ ๐1 if ๐ > 1. Indeed, let ฮฆ : ๐ฟ2 (โ; ๐๐ฅ) โ ๐ฟ2 (โ; ๐๐) be the Fourier transform. Then the operator ฮฆ๐
0 (๐ง)๐ฃ๐ has integral kernel (2.21) where ๐(๐) = (2๐)โ1/2 (๐ ๐ โ ๐ง)โ1 and ๐ฃ(๐ฅ) = ๐ฃ๐ (๐ฅ). By virtue of (1.12) condition (4.14) is satis๏ฌed in this case. If ๐ = 1, then we can use that the operator โฃ๐ฃ1 โฃ1/2 ๐
0 (๐ง) belongs to the Hilbert-Schmidt class. Thus all terms in the right-hand side of (4.15) belong to the trace class. โก
ยจ J. Ostensson and D.R. Yafaev
560
Since the kernel ๐
(๐ฅ, ๐ฆ, ๐ง) (and ๐
0 (๐ฅ, ๐ฆ, ๐ง)) is a continuous function of ๐ฅ, ๐ฆ and the limit in the left-hand side of (4.13) exists, we see (see, e.g., [13], Proposition 3.1.6) that โซ โ ( ) ) ( ๐
(๐ฆ, ๐ฆ, ๐ง) โ ๐
0 (๐ฆ, ๐ฆ, ๐ง) ๐๐ฆ. (4.16) Tr ๐
(๐ง) โ ๐
0 (๐ง) = โโ
Putting together formulas (4.13) and (4.16), we get the trace formula (1.11) for the coe๏ฌcients ๐ฃ1 , . . . , ๐ฃ๐ with compact supports.
5. Integral equations Here we consider di๏ฌerential equation (1.4) with arbitrary short-range coe๏ฌcients. Actually, we follow the scheme of Section 2 and ๏ฌrst consider a more general equation (2.2). 5.1. As usual, we suppose that the eigenvalues ๐๐ , ๐ = 1, . . . , ๐ , of an ๐ ร ๐ matrix L0 are distinct and do not lie on the imaginary axis. We set P๐ = โจโ
, pโ๐ โฉp๐ ,
๐ = 1, . . . , ๐,
(5.1)
where p๐ are eigenvectors of L0 and the vectors pโ๐ form the dual basis. We have P2๐ = P๐ , P๐ P๐ = 0 if ๐ โ= ๐, and L0 P๐ = ๐๐ P๐ ,
๐ โ
P๐ = I.
(5.2)
๐=1
Let a matrix L(๐ฅ) be given by formula (2.1) where we now assume that V โ ๐ฟ1 (โยฑ ).
(5.3) (ยฑ)
We shall show that, for all ๐ = 1, . . . , ๐ , equation (2.2) has solutions u๐ (๐ฅ) such that (ยฑ) (5.4) u๐ (๐ฅ) = ๐๐๐ ๐ฅ (p๐ + ๐(1)), ๐ฅ โ ยฑโ. Thus we construct solutions of (2.2) both (exponentially) decaying and growing at (+) (โ) in๏ฌnity. We emphasize that our construction of the functions u๐ (๐ฅ) (of u๐ (๐ฅ)) (ยฑ)
requires condition (5.3) for ๐ฅ โ โ+ (for ๐ฅ โ โโ ) only. Functions u๐ (๐ฅ) will be de๏ฌned as solutions of integral equations which we borrow from the book [6] (see Problem 29 of Chapter 3). For de๏ฌniteness, we consider the case ๐ฅ โ โโ and put (โ) u ๐ = u๐ . Set โ โ K๐ (๐ฅ) = P๐ ๐๐๐ ๐ฅ ๐(โ๐ฅ) โ P๐ ๐๐๐ ๐ฅ ๐(๐ฅ), (5.5) ๐:๐
๐ >๐
๐
๐:๐
๐ โค๐
๐
where ๐
๐ = Re ๐๐ . It follows from relations (5.2) that Kโฒ๐ (๐ฅ) = L0 K๐ (๐ฅ) โ ๐ฟ(๐ฅ)I,
(5.6)
Trace Formula
561
where ๐ฟ(๐ฅ) is the Dirac function. We also use the estimate ) ( โ ๐๐
๐ ๐ฅ ๐(โ๐ฅ) + ๐๐
๐ ๐ฅ ๐(๐ฅ) , โฃK๐ (๐ฅ)โฃ โค ๐ถ๐
(5.7)
๐:๐
๐ >๐
๐
which is a direct consequence of de๏ฌnition (5.5). In particular, we see that โฃK๐ (๐ฅ)โฃ โค ๐ถ๐ ๐๐
๐ ๐ฅ ,
โ๐ฅ โ โ.
(5.8)
Let ๐๐ be the characteristic function of an interval ๐. Consider the integral equation โซ โ u๐ (๐ฅ) = ๐๐๐ ๐ฅ p๐ โ K๐ (๐ฅ โ ๐ฆ)V(๐ฆ)๐(โโ,๐) (๐ฆ)u๐ (๐ฆ)๐๐ฆ, ๐ฅ < ๐, (5.9) โโ
for a function u๐ (๐ฅ) = u๐ (๐ฅ; ๐) depending on the parameter ๐ which will be chosen later. If ๐
๐ = max๐ ๐
๐ , then the ๏ฌrst sum in (5.5) is absent. In this case we can omit ๐(โโ,๐) (๐ฆ) in (5.9) so that (5.9) becomes a Volterra integral equation. However (5.9) is only a Fredholm equation for other values of ๐. Suppose that a function u๐ (๐ฅ) satis๏ฌes the estimate u๐ (๐ฅ) = ๐(๐๐
๐ ๐ฅ ) as ๐ฅ โ โโ and equation (5.9). Di๏ฌerentiating (5.9) and using (5.6) we see that uโฒ๐ (๐ฅ) = ๐๐ ๐๐๐ ๐ฅ p๐ + V(๐ฅ)๐(โโ,๐) (๐ฅ)u๐ (๐ฅ) โซ โ โ L0 K๐ (๐ฅ โ ๐ฆ)V(๐ฆ)๐(โโ,๐) (๐ฆ)u๐ (๐ฆ)๐๐ฆ.
(5.10)
โโ
Putting together equations (5.9) and (5.10) we ๏ฌnd that a solution u๐ (๐ฅ) of integral equation (5.9) satis๏ฌes also the di๏ฌerential equation uโฒ๐ (๐ฅ) = L0 u๐ (๐ฅ) + V(๐ฅ)๐(โโ,๐) (๐ฅ)u๐ (๐ฅ), which reduces to equation (2.2) for ๐ฅ < ๐. Let us set u๐ (๐ฅ; ๐) = ๐๐๐ ๐ฅ w๐ (๐ฅ; ๐), and rewrite equation (5.9) as โซ w๐ (๐ฅ; ๐) = p๐ โ
๐
โโ
๐ฅ < ๐,
K๐ (๐ฅ โ ๐ฆ)๐โ๐๐ (๐ฅโ๐ฆ) V(๐ฆ)w๐ (๐ฆ; ๐)๐๐ฆ.
(5.11)
(5.12)
By virtue of assumption (5.3) and estimate (5.8) we can choose the parameter ๐ such that โซ ๐ โซ ๐ $ $ $ $ $K๐ (๐ฅ โ ๐ฆ)๐โ๐๐ (๐ฅโ๐ฆ) V(๐ฆ)$๐๐ฆ โค ๐ถ $V(๐ฆ)$๐๐ฆ < 1, โ๐ฅ โ โ. (5.13) โโ
โโ
Then equation (5.12) can be solved in the space ๐ฟโ ((โโ, ๐); โ๐ ) by the method of successive approximations. This result can also be reformulated in the following way. Let โ๐ (๐) be the integral operator with kernel Q๐ (๐ฅ, ๐ฆ) = K๐ (๐ฅ โ ๐ฆ)๐โ๐๐ (๐ฅโ๐ฆ) V(๐ฆ)
(5.14)
ยจ J. Ostensson and D.R. Yafaev
562
acting in the space ๐ฟโ ((โโ, ๐); โ๐ ). Then ( w๐ (๐) = ๐ผ โ โ๐ (๐))โ1 p๐
(5.15)
where the inverse operator exists because โฅโ๐ (๐)โฅ < 1. Clearly, the function u๐ (๐ฅ; ๐) de๏ฌned by formula (5.11) satis๏ฌes di๏ฌerential equation (2.2) for ๐ฅ < ๐. Since a solution of a di๏ฌerential equation of ๏ฌrst order is determined uniquely by its value at one point, it su๏ฌces to require equality (5.11) only for one ๐ฅ < ๐. Then the corresponding solution can be extended to all ๐ฅ โ โ. Now we are in a position to give the precise de๏ฌnition. (โ)
De๏ฌnition 5.1. Let w๐ (โ
; ๐โ ) โ ๐ฟโ ((โโ, ๐โ ); โ๐ ), ๐ = 1, . . . , ๐ , be the solution of equation (5.12) where ๐ = ๐โ is a su๏ฌciently large negative number. (โ) We de๏ฌne u๐ (๐ฅ; ๐โ ) as the solution of di๏ฌerential equation (2.2) which satis๏ฌes (+)
condition (5.11) for some (and then for all) ๐ฅ < ๐. The solutions u๐ (๐ฅ; ๐+ ), ๐ = 1, . . . , ๐ , are de๏ฌned quite similarly if (โโ, ๐โ ) is replaced by (๐+ , โ) where ๐+ is a su๏ฌciently large positive number. It remains to verify asymptotics (5.4) for the function u๐ (๐ฅ). According to (5.11) and (5.12) it su๏ฌces to check that the integral in the right-hand side of (5.12) tends to zero as ๐ฅ โ โโ. Using estimate (5.7) and the inclusion w๐ โ ๐ฟโ ((โโ, ๐โ ); โ๐ ), we see that this integral is bounded by โซ ๐ฅ โ โซ ๐ $ $ $ $ $V(๐ฆ)$๐๐ฆ + ๐ถ ๐(๐
๐ โ๐
๐ )(๐ฅโ๐ฆ) $V(๐ฆ)$๐๐ฆ. (5.16) ๐ถ โโ
๐:๐
๐ >๐
๐
๐ฅ
The ๏ฌrst integral here tends to zero as ๐ฅ โ โโ by virtue of condition (5.3). Each of the integrals over (๐ฅ, ๐) can be estimated by โซ ๐ฅ/2 โซ ๐ $ $ $ $ (๐
๐ โ๐
๐ )๐ฅ/2 $ $ $V(๐ฆ)$๐๐ฆ. V(๐ฆ) ๐๐ฆ + ๐ ๐ฅ/2
๐ฅ
Since ๐
๐ > ๐
๐ , this expression tends to zero as ๐ฅ โ โโ by virtue again of condition (5.3). Thus we arrive at the following result. Proposition 5.2. Let assumption (5.3) hold, and let ๐+ (๐โ ) be a su๏ฌciently large (ยฑ) positive (negative) number. Then, for all ๐ = 1, . . . , ๐ , the functions u๐ (๐ฅ; ๐ยฑ ) (see De๏ฌnition 5.1) satisfy equation (2.2) and have asymptotics (5.4). Solutions of equation (2.2) are of course not determined uniquely by asymp(ยฑ) totics (5.4). In particular, the solutions u๐ (๐ฅ; ๐ยฑ ) generically depend on the choice of the parameter ๐ยฑ . (ยฑ) Let u๐ (๐ฅ; ๐, ๐) = u๐ (๐ฅ; ๐ยฑ , ๐) be the function constructed above for the cut-o๏ฌ coe๏ฌcient V๐ (๐ฅ) = ๐(โ๐,๐) (๐ฅ)V(๐ฅ); thus function (5.15) is now replaced by ( w๐ (๐, ๐) = ๐ผ โ โ๐ (๐)๐(โ๐,๐) )โ1 p๐ . (5.17)
Trace Formula Since
563
( ) lim โฅโ๐ (๐) 1 โ ๐(โ๐,๐) โฅ๐ฟโ (โโ,๐) = 0,
๐โโ
we see that u๐ (๐ฅ; ๐, ๐) โ u๐ (๐ฅ, ๐) as ๐ โ โ for all ๏ฌxed ๐ฅ < ๐. This relation extends to all ๐ฅ โ โ because solutions of di๏ฌerential equations depend continuously on initial data. Therefore Proposition 5.2 can be supplemented by the following result. Lemma 5.3. Under the assumptions of Proposition 5.2, let (ยฑ)
u๐ (๐ฅ; ๐ยฑ )
and
(ยฑ)
u๐ (๐ฅ; ๐ยฑ , ๐)
be the solutions of equations (2.2) with V(๐ฅ) and V๐ (๐ฅ), respectively, speci๏ฌed in De๏ฌnition 5.1. Then for all ๐ = 1, . . . , ๐ the relation (ยฑ)
(ยฑ)
lim u๐ (๐ฅ; ๐ยฑ , ๐) = u๐ (๐ฅ; ๐ยฑ )
(5.18)
๐โโ
holds uniformly in ๐ฅ on compact intervals of โ. (ยฑ)
5.2. If a function u๐ (๐ฅ) satis๏ฌes equation (2.2) and has asymptotics (5.4), then (ยฑ)
adding to u๐ (๐ฅ) a solution with a more rapid decay (or less rapid growth) as ๐ฅ โ ยฑโ we obtain again a solution of equation (2.2) with the same asymptotics (5.4). It is natural to expect that this procedure exhausts the arbitrariness in the (ยฑ) de๏ฌnition of u๐ (๐ฅ). The precise result will be formulated in Lemma 5.5. The following assertion is almost obvious. (ยฑ)
(ยฑ)
Lemma 5.4. Suppose that solutions u1 , . . . , u๐ of the di๏ฌerential equation (2.2) have asymptotics (5.4) as ๐ฅ โ ยฑโ. Then for each of the signs โ ยฑ โ the functions (ยฑ) (ยฑ) u1 , . . . , u๐ are linearly independent. Proof. It follows from (5.4) that (ยฑ)
(ยฑ)
det{u1 (๐ฅ), . . . , u๐ (๐ฅ)} = det{p1 , . . . , p๐ } exp
๐ (โ
)( ) ๐๐ ๐ฅ 1 + ๐(1)
๐=1
as ๐ฅ โ ยฑโ. Since det{p1 , . . . , p๐ } โ= 0, this expression is not zero for su๏ฌciently large ยฑ๐ฅ. โก (ยฑ)
(ยฑ)
Lemma 5.5. Suppose that solutions u1 , . . . , u๐ of the di๏ฌerential equation (2.2) (ยฑ) ห ๐ be an arbitrary solution of (2.2) with have asymptotics (5.4) as ๐ฅ โ ยฑโ. Let u asymptotics (5.4) as ๐ฅ โ ยฑโ. Then necessarily โ (ยฑ) (ยฑ) (ยฑ) (ยฑ) ห ๐ (๐ฅ) = u๐ (๐ฅ) + u ๐๐,๐ u๐ (๐ฅ) (5.19) ยฑ(๐
๐ โ๐
๐ )<0 (ยฑ)
for some numbers ๐๐,๐ .
ยจ J. Ostensson and D.R. Yafaev
564
(โ)
Proof. As before, we set u๐ (๐ฅ) = u๐ (๐ฅ). According to Lemma 5.4 we have ห ๐ (๐ฅ) = u
๐ โ
๐๐,๐ u๐ (๐ฅ)
(5.20)
๐=1
with some numbers ๐๐,๐ . Therefore it follows from asymptotic relations (5.4) that โ ๐๐๐ ๐ฅ (p๐ + ๐(1)) = ๐๐,๐ ๐๐๐ ๐ฅ (p๐ + ๐(1)) + ๐(๐๐
๐ ๐ฅ ) ๐
๐ โค๐
๐
as ๐ฅ โ โโ. Since the vectors p1 , . . . , p๐ are linearly independent, this relation implies that ๐๐,๐ = 0 if ๐ โ= ๐ and ๐๐,๐ = 1. Thus equality (5.20) leads to (5.19). โก 5.3. Let us return to di๏ฌerential equation (1.4) with coe๏ฌcients satisfying the assumption ๐ฃ๐ โ ๐ฟ1 (โยฑ ), ๐ = 1, . . . , ๐, (5.21) only. De๏ฌne as usual the matrices L0 (๐ง) and V(๐ฅ) by formulas (4.2) and (4.3). Now ๐๐๐ = ๐๐ ๐ง and the vectors p๐ (๐ง), ๐ = 1, . . . , ๐ , are given by formula (4.4). It is easy to control the dependence on ๐ง of matrices (5.1). (๐)
Lemma 5.6. Elements ๐๐,๐ (๐ง) of the matrices P๐ (๐ง), ๐ = 1, . . . , ๐ , obey the relation (๐)
๐๐,๐ (๐ง) = ๐(โฃ๐โฃ๐โ๐ ),
โฃ๐โฃ๐ = โฃ๐งโฃ โ โ.
(5.22)
Proof. Obviously, the basis dual to p๐ (๐ง) consists of the vectors pโ๐ (๐ง) = (๐๐,1 , ๐๐,2 ๐๐โ1 , . . . , ๐๐,๐ ๐๐โ๐ +1 ) where the numbers ๐๐,๐ do not depend on โฃ๐งโฃ. It now follows from equality (4.4) (๐) โก that ๐๐,๐ (๐ง) = ๐ยฏ๐,๐ ๐๐๐โ1 ๐ยฏ๐โ๐+1 , which yields (5.22). The next step is to control the dependence on ๐ง of matrices (5.14). (๐)
Lemma 5.7. Elements ๐๐,๐ (๐ฅ, ๐ฆ, ๐ง) of the matrices Q๐ (๐ฅ, ๐ฆ, ๐ง), ๐ = 1, . . . , ๐ , admit the estimate (๐) (5.23) โฃ๐๐,๐ (๐ฅ, ๐ฆ, ๐ง)โฃ โค ๐ถโฃ๐โฃ๐โ๐ โฃ๐ฃ๐ (๐ฆ)โฃ with a constant ๐ถ not depending on ๐ฅ, ๐ฆ and ๐ง. (๐)
Proof. It follows from (5.5) and (5.22) that elements ๐ ๐,๐ (๐ฅ, ๐ง) of the matrix K๐ (๐ฅ)๐โ๐๐ ๐ฅ satisfy the estimate (cf. (5.8)) (๐)
โฃ๐ ๐,๐ (๐ฅ, ๐ง)โฃ โค ๐ถโฃ๐โฃ๐โ๐ . This directly implies (5.23) because, by de๏ฌnition (4.3), (๐)
(๐)
๐๐,๐ (๐ฅ, ๐ฆ, ๐ง) = โ๐๐ ๐ ๐,๐ (๐ฅ โ ๐ฆ, ๐ง)๐ฃ๐ (๐ฆ).
โก
Trace Formula (ยฑ)
(ยฑ)
(ยฑ)
565 (ยฑ)
Let u๐ (๐ฅ, ๐ง; ๐ยฑ ) (recall that u๐ = (๐ข1,๐ , . . . , ๐ข๐,๐ )๐ก ) be the solution of equation (2.2) speci๏ฌed in De๏ฌnition 5.1. Then the function (ยฑ)
(ยฑ)
๐ข๐ (๐ฅ, ๐ง; ๐ยฑ ) := ๐ข1,๐ (๐ฅ, ๐ง; ๐ยฑ )
(5.24)
satis๏ฌes also equation (1.4) and according to (4.4) asymptotics (5.4) imply asymptotics (1.13). Therefore Proposition 5.2 and Lemma 5.3 entail the following result. Proposition 5.8. Let assumption (5.21) hold, let โฃ๐งโฃ โฅ ๐ > 0 and let ๐+ = ๐+ (๐) (๐โ = ๐โ (๐)) be a su๏ฌciently large positive (negative) number. Then for every (ยฑ) ๐ = 1, . . . , ๐ the function ๐ข๐ (๐ฅ, ๐ง; ๐ยฑ ) determined by De๏ฌnition 5.1 and equality (5.24) satis๏ฌes equation (1.4) and has asymptotics (1.13) as ๐ฅ โ ยฑโ. Moreover, (ยฑ) the corresponding solutions ๐ข๐ (๐ฅ, ๐ง; ๐ยฑ , ๐) of equation (1.4) with cut-o๏ฌ coe๏ฌcients ๐(โ๐,๐) (๐ฅ)๐ฃ๐ (๐ฅ), ๐ = 1, . . . , ๐ , satisfy the relation (ยฑ)
(ยฑ)
lim ๐ข๐ (๐ฅ, ๐ง; ๐ยฑ , ๐) = ๐ข๐ (๐ฅ, ๐ง; ๐ยฑ )
๐โโ
(5.25)
uniformly in ๐ฅ on compact intervals of โ. This relation remains true for ๐ โ 1 (ยฑ) derivatives of the functions ๐ข๐ . By de๏ฌnition (5.5), the kernels K๐ (๐ฅ, ๐ง) depend analytically on ๐ง except on the rays where Re ๐๐ = Re ๐๐ for some root ๐๐ โ= ๐๐ of the equation ๐ ๐ = ๐๐ ๐ง. In addition to the condition ๐ง โโ ๐(๐ป0 ), this excludes also the half-line ๐ง < 0 for even ๐ and the line Re ๐ง = 0 for odd ๐ . Hence the same is true for the functions (ยฑ) (ยฑ) ๐ข๐ (๐ฅ, ๐ง; ๐ยฑ ) if โฃ๐งโฃ > ๐ > 0. Thus, for ๏ฌxed ๐ฅ and ๐ยฑ , the functions ๐ข๐ (๐ฅ, ๐ง; ๐ยฑ ) are analytic functions of ๐ง if โฃ๐งโฃ > ๐ > 0, Im ๐ง โ= 0 for ๐ even and if Im ๐ง โ= 0, (ยฑ) Re ๐ง โ= 0 for ๐ odd. On the rays where Re ๐๐ = Re ๐๐ , the limits of ๐ข๐ (๐ฅ, ๐ง; ๐ยฑ ) from both sides exist but di๏ฌer, in general, from each other by a term which decays faster (or grows less rapidly) than ๐๐๐ ๐ฅ as ๐ฅ โ +โ or ๐ฅ โ โโ. 5.4. Integral equations (5.9) turn out also to be useful (even for functions ๐ฃ๐ (๐ฅ) (ยฑ) with compact supports) for a study of asymptotics of the solutions ๐ข๐ (๐ฅ, ๐ง; ๐ยฑ ) of di๏ฌerential equation (1.4) as โฃ๐งโฃ โ โ. We choose the sign โ โ โ, ๏ฌx the parameter ๐ = ๐โ and index ๐ and drop them out of notation. Consider system (5.12) of ๐ equations for components ๐ค๐ (๐ฅ, ๐ง), ๐ = 1, . . . , ๐ , of a vector-valued function w(๐ฅ, ๐ง). Set ๐ค๐ (๐ฅ, ๐ง) = ๐ ๐โ1 ๐ค ห๐ (๐ฅ, ๐ง) and take into account equality (4.4). Then we obtain for ๐ค ห๐ (๐ฅ, ๐ง) a system โซ ๐ ๐ โ ๐ค ห๐ (๐ฅ, ๐ง) = 1 โ ๐ ๐โ๐ ๐๐,๐ (๐ฅ, ๐ฆ, ๐ง)๐ค ห๐ (๐ฆ, ๐ง)๐๐ฆ, ๐ฅ < ๐, (5.26) ๐=1
โโ
where the elements ๐๐,๐ of the matrix Q satisfy inequality (5.23). In particular, the operator in the right-hand side of (5.26) is uniformly bounded as โฃ๐งโฃ โ โ. Assume additionally that ๐ฃ๐ (๐ฅ) = 0. Then according to (5.23) the norm of the operator in the right-hand side of (5.26) is ๐(โฃ๐โฃโ1 ) as โฃ๐งโฃ โ โ. Therefore for su๏ฌciently large โฃ๐โฃ system (5.26) can be solved in the space ๐ฟโ((โโ, ๐); โ๐ )
ยจ J. Ostensson and D.R. Yafaev
566
by the method of successive approximations and ๐ค ห๐ (๐ฅ, ๐ง) = 1 + ๐(โฃ๐โฃโ1 ), ๐ = 1, . . . , ๐ . As we have already seen in the proof of Proposition 5.2, the solution of system (5.26) necessarily has the asymptotics ๐ค ห๐ (๐ฅ, ๐ง) = 1 + ๐(1), ๐ = 1, . . . , ๐ , as ๐ฅ โ โโ. De๏ฌne as usual ๐ข(๐ฅ, ๐ง) as a solution of equation (1.4) such that ๐ข(๐ฅ, ๐ง) = ๐๐๐ฅ ๐ค ห1 (๐ฅ, ๐ง) for ๐ฅ < ๐. Thus we obtain the following result. Proposition 5.9. Let assumption (5.21) hold, and let ๐ฃ๐ = 0. Fix arbitrary ๐ยฑ . Then for all ๐ = 1, . . . , ๐ and all su๏ฌciently large โฃ๐งโฃ equation (1.4) has solutions (ยฑ) ๐ข๐ (๐ฅ, ๐ง; ๐ยฑ ) with asymptotics (1.13) as ๐ฅ โ ยฑโ and such that ( ) (ยฑ) ๐ข๐ (๐ฅ, ๐ง; ๐ยฑ ) = ๐๐๐ ๐ฅ 1 + ๐(โฃ๐งโฃโ1/๐ ) , โฃ๐งโฃ โ โ, (5.27) for all ยฑ(๐ฅ โ ๐ยฑ ) > 0. Remark 5.10. If assumption (5.21) is true for both signs, then we can set ๐ = +โ in equation (5.26). For su๏ฌciently large โฃ๐งโฃ, such an equation can again be solved by the method of successive approximations.
6. The Wronskian and the perturbation determinant 6.1. Let us de๏ฌne the Wronskian W(๐ฅ) for di๏ฌerential equation (2.2) where the matrix-valued function L(๐ฅ) is given by formula (2.1) and V(๐ฅ) satis๏ฌes assumption (5.3) (for both signs) only. To justify the de๏ฌnition below, we start with the following observation. ห ๐ of the di๏ฌerential equaLemma 6.1. Suppose that both sets of solutions u๐ and u tion (2.2) have asymptotics (5.4) as ๐ฅ โ โโ for ๐ = 1, . . . , ๐ and as ๐ฅ โ +โ for ๐ = ๐ + 1, . . . , ๐ . Then for all ๐ฅ ห ๐ (๐ฅ)}. det{u1 (๐ฅ), . . . , u๐ (๐ฅ)} = det{ห u1 (๐ฅ), . . . , u
(6.1)
Proof. Let us proceed from Lemma 5.5. To simplify notation, we suppose that ๐
1 โฅ โ
โ
โ
โฅ ๐
๐ > ๐
๐+1 โฅ โ
โ
โ
โฅ ๐
๐ .
(6.2)
First, we check that for all ๐ = 1, . . . , ๐, ห ๐ (๐ฅ)} = det{u1 (๐ฅ), . . . , u๐ (๐ฅ), u ห ๐+1 (๐ฅ), . . . , u ห ๐ (๐ฅ)}. det{ห u1 (๐ฅ), . . . , u
(6.3)
ห 1 (๐ฅ) = u1 (๐ฅ). Suppose that For ๐ = 1 this equality is obvious because necessarily u (6.3) is true for some ๐. Then using (5.19) we see that the left-hand side of (6.3) equals โ ห ๐+2 (๐ฅ), . . . , u ห ๐ (๐ฅ)}. ๐๐+1,๐ u๐ (๐ฅ), u det{u1 (๐ฅ), . . . , u๐ (๐ฅ), u๐+1 (๐ฅ) + ๐
๐ >๐
๐+1
Since according to (6.2) the contribution of the sum over ๐
๐ > ๐
๐+1 equals zero, this yields relation (6.3) for ๐ + 1 and hence for all ๐ = 1, . . . , ๐.
Trace Formula
567
Quite similarly, we can verify that for all ๐ = ๐, . . . , ๐ + 1 ห ๐+1 (๐ฅ), . . . , u ห ๐ (๐ฅ)} det{u1 (๐ฅ), . . . , u๐ (๐ฅ), u ห ๐+1 (๐ฅ), . . . , u ห ๐ (๐ฅ), u๐+1 (๐ฅ), . . . , u๐ (๐ฅ)}. = det{u1 (๐ฅ), . . . , u๐ (๐ฅ), u Putting together equalities (6.3) and (6.4) for ๐ = ๐, we get (6.1).
(6.4) โก
Now we are in a position to de๏ฌne the Wronskian W(๐ฅ). De๏ฌnition 6.2. Let u๐ (๐ฅ), ๐ = 1, . . . , ๐ , be arbitrary solutions of equation (2.2) with asymptotics (5.4) as ๐ฅ โ โโ if ๐ = 1, . . . , ๐ and as ๐ฅ โ โ if ๐ = ๐+1, . . . , ๐ . We set (6.5) W(๐ฅ) = det{u1 (๐ฅ), . . . , u๐ (๐ฅ)}. Recall that solutions of equation (2.2) with asymptotics (5.4) exist according to Proposition 5.2. Although they are not unique, according to Lemma 6.1 the Wronskian W(๐ฅ) does not depend (up to a numeration of eigenvalues ๐๐ ) on a speci๏ฌc choice of such solutions. In particular, we have (โ)
(+)
(+)
W(๐ฅ) = det{u1 (๐ฅ; ๐โ ), . . . , u(โ) ๐ (๐ฅ; ๐โ ), u๐+1 (๐ฅ; ๐+ ), . . . , u๐ (๐ฅ; ๐+ )}
(6.6)
where ๐+ (๐โ ) are su๏ฌciently large positive (negative) numbers and the solutions (ยฑ) u1 (๐ฅ; ๐ยฑ ) are constructed in Proposition 5.2. Of course, De๏ฌnition 6.2 applies if the matrices L0 (๐ง) and V(๐ฅ) are given by formulas (4.2) and (4.3), respectively. In this case the Wronskian W(๐ฅ, ๐ง) depends analytically on the parameter ๐ง โโ ๐(๐ป0 ). Indeed, if additionally Im ๐ง โ= 0 for ๐ even and Re ๐ง โ= 0 for ๐ odd, then this fact directly follows from the analyticity (ยฑ) of the solutions u๐ (๐ฅ, ๐ง; ๐ยฑ ), ๐ = 1, . . . , ๐ (see subs. 5.3). Moreover, according to Lemma 6.1 the Wronskian W(๐ฅ, ๐ง) is continuous (in contrast to the solutions (ยฑ) u๐ (๐ฅ, ๐ง; ๐ยฑ )) on the critical rays where Re ๐๐ = Re ๐๐ for some ๐๐ โ= ๐๐ . Therefore its analyticity in a required region follows from Moreraโs theorem. Evidently, W(๐ฅ, ๐ง) = 0 if and only if ๐ง is an eigenvalue of the operator ๐ป. 6.2. Let us return to the trace formula (1.11) established so far for coe๏ฌcients ๐ฃ๐ (๐ฅ), ๐ = 1, . . . , ๐ , with compact supports. Suppose that assumption (1.12) holds. Then condition (5.21) is satis๏ฌed for both signs. Let us approximate ๐ฃ๐ (๐ฅ) by the cut(โ) o๏ฌ functions ๐(โ๐,๐)(๐ฅ)๐ฃ๐ (๐ฅ). We denote by ๐ข๐ (๐ฅ, ๐ง; ๐โ , ๐), ๐ = 1, . . . , ๐, and (+)
by ๐ข๐ (๐ฅ, ๐ง; ๐+ , ๐), ๐ = ๐ + 1, . . . , ๐ , the solutions of equation (1.4) with the coe๏ฌcients ๐(โ๐,๐) ๐ฃ๐ determined by De๏ฌnition 5.1 and equality (5.24). Let us use formula (6.6) for the Wronskian W๐ (๐ฅ, ๐ง) for equation (1.4) with cut-o๏ฌ coe๏ฌcients ๐(โ๐,๐) (๐ฅ)๐ฃ๐ (๐ฅ). Then it follows from relation (5.25) that lim W๐ (๐ฅ, ๐ง) = W(๐ฅ, ๐ง).
๐โโ
(6.7)
In view of analyticity in ๐ง of these functions we also have ห ห ๐ (๐ฅ, ๐ง) = W(๐ฅ, ๐ง). lim W
๐โโ
(6.8)
ยจ J. Ostensson and D.R. Yafaev
568
Set ๐ป๐ = ๐ป0 + ๐๐ where the operator ๐๐ is de๏ฌned by formula (1.14) with the coe๏ฌcients ๐(โ๐,๐) ๐ฃ๐ . Let us write down formula (4.13) for the operator ๐ป๐ and pass to the limit ๐ โ โ. By virtue of (6.7) and (6.8) the right-hand side of (4.13) converges to the corresponding expression for the operator ๐ป. It is possible to verify that the same is true for the left-hand side of (4.13). We shall not however dwell upon it and establish the trace formula in form (1.11). Using the resolvent identity ๐
(๐ง) โ ๐
๐ (๐ง) = โ
๐ โ
๐
๐ (๐ง)(๐ โ ๐๐ )๐
(๐ง),
๐
๐ (๐ง) = (๐ป๐ โ ๐ง)โ1 ,
๐=1
we see that for ๐ง โโ ๐(๐ป) โฅ๐
(๐ง) โ ๐
๐ (๐ง)โฅ๐1 โค ๐ถโฅ๐
0 (๐ง)(๐ โ ๐๐ )๐
0 (๐ง)โฅ๐1 โค ๐ถ1
๐ โ
โฅ๐
0 (๐ง)๐ฃ๐ (1 โ ๐๐ )โฅ๐1 .
๐=1
According to Proposition 4.4 there is (for ๐ โฅ 2) the estimate โซ 2 โฅ๐
0 (๐ง)๐ฃ๐ (1 โ ๐๐ )โฅ๐1 โค ๐ถ โฃ๐ฃ๐ (๐ฅ)โฃ2 (1 + ๐ฅ2 )๐ผ ๐๐ฅ, ๐ผ > 1/2, โฃ๐ฅโฃโฅ๐
whence
lim โฅ๐
๐ (๐ง) โ ๐
(๐ง)โฅ๐1 = 0.
๐โโ
Thus, using trace formula (1.11) for cut-o๏ฌ perturbations ๐๐ and passing to the limit ๐ โ โ, we deduce it for ๐ . This leads to the following result. Recall that the normalized Wronskian ฮ(๐ฅ, ๐ง) is de๏ฌned by formula (1.10). Theorem 6.3. Under assumption (1.12) the trace formula (1.11) holds for all ๐ง โโ ๐(๐ป). If inclusion (1.18) holds, then equation (1.16) is satis๏ฌed for a generalized perturbation determinant ( ) ห ๐ง0 (๐ง) = Det ๐ผ + (๐ง โ ๐ง0 )๐
(๐ง0 )๐ ๐
0 (๐ง) (6.9) ๐ท ห ๐ง0 (๐ง) is the usual perturbation deterwhere ๐ง0 โโ ๐(๐ป). It is easy to see that ๐ท minant for the pair ๐
0 (๐ง0 ), ๐
(๐ง0 ) at the point (๐ง โ ๐ง0 )โ1 . Of course, equation ห (1.16) for a function ๐ท(๐ง) ๏ฌxes it up to a constant factor only. We note that for di๏ฌerent โreference pointsโ, generalized perturbation determinants are conห ๐ง1 (๐ง) = ๐ท ห ๐ง0 (๐ง1 )โ1 ๐ท ห ๐ง0 (๐ง). Moreover, if ๐ ๐
0 (๐ง) โ ๐1 , nected by the formula ๐ท โ1 ห ๐ง0 (๐ง) = ๐ท(๐ง0 ) ๐ท(๐ง) where ๐ท(๐ง) is the perturbation determinant (see then ๐ท formula (1.15)) for the pair ๐ป0 , ๐ป. Comparing equations (1.11) and (1.16) we see that for all ๐ฅ โ โ and all ๐ง0 โโ ๐(๐ป) โซ ๐ฅ ( ) ๐ ห ๐ง0 (๐ง) ฮ(๐ฅ, ๐ง) = ๐ถ(๐ฅ0 , ๐ง0 ) exp โ ๐ ๐ฃ๐ (๐ฆ)๐๐ฆ ๐ท ๐ฅ0
where the constant ๐ถ(๐ฅ0 , ๐ง0 ) โ= 0 does not depend on ๐ฅ and ๐ง.
Trace Formula
569
6.3. Suppose now that ๐ฃ๐ = 0. Then the Wronskian W(๐ฅ, ๐ง) =: W(๐ง) does not depend on ๐ฅ, and it is easy to deduce from Proposition 5.9 that ( ) (6.10) W(๐ง) = W0 (๐ง) 1 + ๐(โฃ๐งโฃโ1/๐ ) , โฃ๐งโฃ โ โ. For the proof, it su๏ฌces to choose ๐+ < 0, ๐โ > 0 and use asymptotics (5.27) at ๐ฅ = 0. As a side remark, we note that according to (6.10) the set of complex eigenvalues of the operator ๐ป is bounded. It follows from (6.10) that the normalized Wronskian (1.10) satis๏ฌes the relation ฮ(๐ง) = 1 + ๐(โฃ๐งโฃโ1/๐ ), โฃ๐งโฃ โ โ. (6.11) Since, by Proposition 4.4, ๐ ๐
0 (๐ง) โ ๐1 , the perturbation determinant is correctly de๏ฌned by formula (1.15) and (see book [10]) ( ) lim Det ๐ผ + ๐ ๐
0 (๐ง) = 1. (6.12) โฃ Im ๐งโฃโโ
Comparing equations (1.11) and (1.16), we obtain that ) ( ฮ(๐ง) = ๐ถ Det ๐ผ + ๐ ๐
0 (๐ง)
(6.13)
for some constant ๐ถ. Moreover, taking into account relations (6.11) and (6.12), we see that ๐ถ = 1 in (6.13). Let us formulate the result obtained. Theorem 6.4. Suppose that ๐ฃ๐ = 0 and that the coe๏ฌcients ๐ฃ๐ , ๐ = 1, . . . , ๐ โ 1, satisfy assumption (1.12). Then ฮ(๐ฅ, ๐ง) =: ฮ(๐ง) does not depend on ๐ฅ and relation (1.17) is true for all ๐ง โโ ๐(๐ป). 6.4. Finally, we note that for a derivation of the trace formula (1.11) the approximation of ๐ฃ๐ by cut-o๏ฌ functions ๐(โ๐,๐) ๐ฃ๐ is not really necessary. From the very beginning, we could work with functions ๐ฃ๐ satisfying assumption (1.12) only. Then the text of Sections 2, 3 and 4 remains unchanged if, for all ๐ = 1, . . . , ๐ , the (โ) functions ๐๐๐ ๐ฅ p๐ are replaced for ๐ฅ << 0 by u๐ (๐ฅ; ๐โ ) where ๐โ is a su๏ฌciently (+)
big negative number and they are replaced for ๐ฅ >> 0 by u๐ (๐ฅ; ๐+ ) where ๐+ is a su๏ฌciently big positive number. In particular, the de๏ฌnition of the transition (ยฑ) matrices in subs. 2.4 can be given in terms of the solutions u๐ (๐ฅ; ๐ยฑ ). However, a preliminary consideration of coe๏ฌcients ๐ฃ๐ with compact supports seems to be intuitively more clear.
References [1] N.I. Akhieser and I.M. Glasman, The theory of linear operators in Hilbert space, vols. I, II, Ungar, New York, 1961. [2] R. Beals, P. Deift and C. Tomei, Direct and inverse scattering on the line, Math. surveys and monographs, N 28, Amer. Math. Soc., Providence, R. I., 1988.
570
ยจ J. Ostensson and D.R. Yafaev
[3] M.Sh. Birman, On the spectrum of singular boundary-value problems, Matem. Sb. 55, no. 2 (1961), 125โ174 (Russian); English transl.: Eleven Papers on Analysis, Amer. Math. Soc. Transl. (2), vol. 53, Amer. Math. Soc., Providence, Rhode Island, 1966, 23โ60. [4] M.Sh. Birman and M.Z. Solomyak, Estimates for the singular numbers of integral operators, Russian Math. Surveys 32 (1977), 15โ89. [5] V.S. Buslaev and L.D. Faddeev, Formulas for traces for a singular Sturm-Liouville di๏ฌerential operator, Soviet Math. Dokl. 1 (1960), 451โ454. [6] E.A. Coddington and N. Levinson, Theory of ordinary di๏ฌerential equations, McGraw-Hill, New York, 1955. [7] L.D. Faddeev, Inverse problem of quantum scattering theory. II, J. Soviet. Math. 5, 1976, 334โ396. [8] F. Gesztesy and K.A. Makarov, (Modi๏ฌed) Fredholm determinants for operators with matrix-valued semi-separable integral kernels revisited, Integral Eqs. Operator Theory 47 (2003), 457โ497; Erratum 48 (2004), 425โ426. [9] I.C. Gohberg, S. Goldberg and N. Krupnik, Traces and determinants for linear operators, Operator Theory: Advances and Applications 116, Birkhยจ auser, Basel, 2000. [10] I.C. Gohberg and M.G. Kreหฤฑn, Introduction to the theory of linear nonselfadjoint operators, Nauka, Moscow, 1965; Engl. transl.: Amer. Math. Soc. Providence, R. I., 1969. [11] R. Jost and A. Pais, On the scattering of a particle by a static potential, Phys. Rev. 82 (1951), 840โ851. [12] M.A. Naimark, Linear di๏ฌerential operators, Ungar, New York, 1967. [13] D.R. Yafaev, Mathematical scattering theory. Analytic theory, Amer. Math. Soc., Providence, Rhode Island, 2010. ยจ J. Ostensson Department of Mathematics Uppsala University Box 480 SE-751 06 Uppsala, Sweden e-mail:
[email protected] D.R. Yafaev IRMAR, Universitยดe de Rennes I Campus de Beaulieu F-35042 Rennes Cedex, France e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 571โ582 c 2012 Springer Basel AG โ
Jordan Structures and Lattices of Invariant Subspaces of Real Matrices Leiba Rodman Dedicated to the memory of Israel Gohberg
Abstract. Real matrices having the same Jordan structure are characterized in terms of isomorphisms and linear isomorphisms of lattices of their invariant subspaces. Mathematics Subject Classi๏ฌcation (2000). 15A21, 47A15. Keywords. Invariant subspace, Jordan structure, real matrix.
1. Introduction Let F = C, the complex ๏ฌeld, or F = R, the real ๏ฌeld. Two matrices ๐ด, ๐ต โ C๐ร๐ are said to have the same C-Jordan structure if the number ๐ of distinct eigenvalues ๐1 (๐ด), . . . , ๐๐ (๐ด) of ๐ด and ๐1 (๐ต), . . . , ๐๐ (๐ต) of ๐ต is the same, and there exists a permutation ๐ : {1, 2, . . . , ๐ } โ {1, 2, . . . , ๐ } such that the partial multiplicities of ๐๐ (๐ด) as an eigenvalue of ๐ด are identical with those of ๐๐(๐) (๐ต) as an eigenvalue of ๐ต, for ๐ = 1, 2, . . . , ๐ . The partial multiplicities corresponding to ๐ โ ๐(๐ด), ๐ด โ F๐ร๐ , are the sizes of the Jordan blocks (in the Jordan form of ๐ด over C) having the eigenvalue ๐; the partial multiplicity ๐ is repeated the number of times equal to the number of Jordan blocks of size ๐ having the eigenvalue ๐. โIdenticalโ in the above de๏ฌnition includes having the same number of repetitions. For example, if ๐ด โ C14ร14 is nilpotent with the partial multiplicities 3, 3, 3, 3, 2 corresponding to the eigenvalue zero, and ๐ต โ C14ร14 is a nilpotent matrix with partial multiplicities 3, 3, 2, 2, 2, 2, then ๐ต and ๐ด do not have the same C-Jordan structure. This notion was studied and used in [8, 6, 5, 12, 1] and other sources, mainly in connection with various aspects of matrix perturbation theory. In particular, it was proved in [8] that two matrices have the same C-Jordan structure if and only if the lattices of invariant subspaces of ๐ด and ๐ต are isomorphic. Moreover, in this case an isomorphism of these lattices can be given by means of a linear transformation (if this happens, we say that the lattices are linearly isomorphic).
572
L. Rodman
In this paper, we study matrices having the same Jordan structure and the relationship of this property with isomorphisms of the lattices of invariant subspaces, in the context of real matrices and invariant subspaces in the corresponding real vector space. As it turns out, there are essential di๏ฌerences with the complex case. For example, isomorphic lattices of invariant subspaces for real matrices need not be linearly isomorphic. Our main results are Theorems 2.1 and 3.2. In the former, characterizations of having the same Jordan structure (in the context of real matrices) are given in terms of isomorphisms of lattices of invariant subspaces. In the latter, these characterizations are specialized to the case of close (in norm) matrices. In that case, isomorphic lattices of (real) invariant subspaces are necessarily linearly isomorphic. We use the following notation throughout: The spectrum of a matrix (=the set of eigenvalues, including nonreal eigenvalues of real matrices) ๐ด will be denoted ๐(๐ด). Ker ๐ด := {๐ฅ โ F๐ : ๐ด๐ฅ = 0} is the kernel of ๐ด โ F๐ร๐ , and Im ๐ด := {๐ด๐ฅ โ F๐ : ๐ฅ โ F๐ } is the image (or range) of ๐ด. โ๐ (๐ด) := Ker(๐ด โ ๐๐ผ)๐ โ F๐ is the root subspace of a matrix ๐ด โ F๐ร๐ corresponding to the eigenvalue ๐ โ F, and โ๐ยฑ๐๐ (๐ด) := Ker(๐ด2 โ 2๐๐ด + (๐2 + ๐ 2 )๐ผ)๐ โ R๐ is the real root subspace of ๐ด โ R๐ร๐ corresponding to a pair of nonreal complex conjugate eigenvalues ๐ ยฑ ๐๐ of ๐ด. Span {๐ฅ1 , . . . , ๐ฅ๐ } is the subspace spanned by the vectors ๐ฅ1 , . . . , ๐ฅ๐ . The operator matrix norm (=the largest โซsingular value) 1 โฅ๐ดโฅ is used throughout, for ๐ด โ C๐ร๐ . We denote by ๐ฮฉ (๐ด) = 2๐๐ (๐ง๐ผ โ ๐ด)โ1 ๐๐ง ฮ ๐ร๐ the spectral projection of ๐ด โ C associated with eigenvalues included in a set ฮฉ โ C; here ฮ is a suitable (simple, closed, recti๏ฌable) contour such that ฮฉ โฉ ๐(๐ด) is inside ฮ and ๐(๐ด) โ ฮฉ is outside ฮ. Finally, for real numbers, ๐ and ๐ > 0, we let โก โค ๐ ๐ 1 0 โ
โ
โ
0 0 โข โ๐ ๐ 0 1 โ
โ
โ
0 0 โฅ โข โฅ โข 0 0 ๐ ๐ โ
โ
โ
0 0 โฅ โข โฅ โข .. .. โฅ โข 0 0 โ๐ ๐ โ
โ
โ
โฅ . . โข โฅ ๐ฝ2๐ (๐ ยฑ ๐๐) = โข . (1.1) โฅ โ R2๐ร2๐ . . . . .. .. .. โข .. โฅ 1 0 โข โฅ โข . โฅ .. .. .. โข .. . . . 0 1 โฅ โข โฅ โฃ 0 0 0 0 ๐ ๐ โฆ 0
0
0
0
โ๐
๐
2. Matrices with the same Jordan structure For the real case the de๏ฌnition of matrices having the same Jordan structure is modi๏ฌed (comparing with the complex case), and actually we need two versions: Two matrices ๐ด โ R๐ร๐ , ๐ต โ R๐ร๐ are said to have the same weak R-Jordan structure if the following properties hold:
Jordan Structures and Lattices of Invariant Subspaces
573
(1) the number ๐ of distinct eigenvalues with nonnegative imaginary parts ๐1 (๐ด), . . . , ๐๐ (๐ด) of ๐ด and of distinct eigenvalues with nonnegative imaginary parts ๐1 (๐ต), . . . , ๐๐ (๐ต) of ๐ต is the same; (2) there exists a permutation ๐ : {1, 2, . . . , ๐ } โ {1, 2, . . . , ๐ } such that the partial multiplicities of ๐๐ (๐ด) as an eigenvalue of ๐ด are identical with those of ๐๐(๐) (๐ต) as an eigenvalue of ๐ต, for ๐ = 1, 2, . . . , ๐ . Two matrices ๐ด, ๐ต โ R๐ร๐ are said to have the same strong R-Jordan structure if the following properties hold: (3) the number ๐ of distinct real eigenvalues ๐1 (๐ด), . . . , ๐๐ (๐ด) of ๐ด and of distinct real eigenvalues ๐1 (๐ต), . . . , ๐๐ (๐ต) of ๐ต is the same; (4) there exist permutation ๐ : {1, 2, . . . , ๐ } โ {1, 2, . . . , ๐ } such that the partial multiplicities of ๐๐ (๐ด) as an eigenvalue of ๐ด are identical with those of ๐๐(๐) (๐ต) as an eigenvalue of ๐ต, for ๐ = 1, 2, . . . , ๐ ; (5) the number ๐ก of distinct eigenvalues with positive imaginary parts (๐1 + ๐๐1 )(๐ด), . . . , (๐๐ก + ๐๐๐ก )(๐ด) of ๐ด and of distinct eigenvalues with positive imaginary parts (๐1 + ๐๐1 )(๐ต), . . . , (๐๐ก + ๐๐๐ก )(๐ต) of ๐ต, is the same; (6) there exist permutation ๐ : {1, 2, . . . , ๐ก} โ {1, 2, . . . , ๐ก} such that the partial multiplicities of (๐๐ + ๐๐๐ )(๐ด) as an eigenvalue of ๐ด are identical with those of (๐๐(๐) + ๐๐๐(๐) )(๐ต) as an eigenvalue of ๐ต, for ๐ = 1, 2, . . . , ๐ก. Thus, if ๐ด and ๐ต both have either only nonreal spectra or only real spectra, the notions of the weak and strong R-Jordan structure are identical. Part 1 in the following theorem is included for completeness; it was proved in [8]. Theorem 2.1. Part 1. The following statements are equivalent for ๐ด โ C๐ร๐ , ๐ต โ C๐ร๐ : (1a) ๐ = ๐ and ๐ด and ๐ต have the same C-Jordan structure; (1b) The lattices LatC (๐ด) of ๐ด-invariant subspaces in C๐ and LatC (๐ต) of ๐ตinvariant subspaces in C๐ are isomorphic, i.e., there exists a bijective map ๐ : LatC (๐ด) โโ LatC (๐ต) such that ๐(โณ1 โฉ โณ2 ) = ๐(โณ1 ) โฉ ๐(โณ2 ) and ๐(โณ1 + โณ2 ) = ๐(โณ1 ) + ๐(โณ2 ) for every โณ1 , โณ2 โ LatC (๐ด); (1c) ๐ = ๐ and the lattices LatC (๐ด) and LatC (๐ต) are linearly isomorphic, i.e., there exists an invertible matrix ๐ โ C๐ร๐ such that ๐ โณ โ LatC (๐ต) if and only if โณ โ LatC (๐ด). Part 2. The following statements are equivalent for ๐ด, ๐ต โ R๐ร๐ : (2a) ๐ด and ๐ต have the same strong R-Jordan structure; (2b) The lattices LatR (๐ด) of ๐ด-invariant subspaces in R๐ and LatR (๐ต) of ๐ต-invariant subspaces in R๐ are isomorphic with an isomorphism ๐ : LatR (๐ด) โ LatR (๐ต)
574
L. Rodman such that
dim ๐(โณ) = dim โณ (2.1) for all root subspaces โณ = โ๐ (๐ด), ๐ โ ๐(๐ด) โฉ R, and โณ = โ๐ยฑ๐๐ (๐ด), ๐ ยฑ ๐๐ โ ๐(๐ด) โ R; (2c) The lattices LatR (๐ด) and LatR (๐ต) are linearly isomorphic, i.e., there exists an invertible matrix ๐ โ R๐ร๐ such that ๐ โณ โ LatR (๐ต) if and only if โณ โ LatR (๐ด). Part 3. The following statements are equivalent for ๐ด โ R๐ร๐ , ๐ต โ R๐ร๐ : (3a) ๐ด and ๐ต have the same weak R-Jordan structure; (3b) The lattices LatR (๐ด) of ๐ด-invariant subspaces in R๐ and LatR (๐ต) of ๐ตinvariant subspaces in R๐ are isomorphic. The following two examples will clarify the di๏ฌerences between the complex and real cases in Theorem 2.1. Example 2.2. Let ๐ด1 = 0 โ R
1ร1
[ ,
๐ด2 =
0 โ1
1 0
]
โ R2ร2 .
The lattices LatR (๐ด1 ) and LatR (๐ด2 ) are isomorphic. Example 2.3. Let
] 0 1 , โ1 0 โค โก 0 1 1 0 ] [ โข โ1 0 0 1 โฅ 0 0 6ร6 โฅ โโข ๐ต= โฃ 0 0 0 1 โฆโR . 0 0 0 0 โ1 0 Then ๐ด and ๐ต have same weak R-Jordan structure but not the same strong RJordan structure, and the lattices LatR (๐ด) and LatR (๐ต) are isomorphic but not linearly isomorphic. Moreover, ๐ด and ๐ต do not have the same Jordan structure as complex matrices, so LatC (๐ด) and LatC (๐ต) are not isomorphic. [
๐ด=
0 1 0 0
]
[
โ
0 1 โ1 0
]
[
โ
We need two lemmas for the proof of Theorem 2.1. [ ๐Let ] ๐ be the alge๐ bra (isomorphic to C) of 2 ร 2 real matrices of the form โ๐ ๐ , and denote by โณ๐ร๐ (๐) โ R2๐ร2๐ the set of ๐ ร ๐ matrices with entries in ๐. Lemma 2.4. Let ๐ โ โณ๐ร๐ (๐), ๐ โ โณ๐ร๐ (๐) be matrices with no real eigenvalues. Then all solutions ๐ โ R2๐ร2๐ of ๐ ๐ = ๐๐ belong to โณ๐ร๐ (๐). Proof. Replacing ๐ by ๐ โ1 ๐ ๐, ๐ by ๐ โ1๐ ๐ , and ๐ by ๐ โ1 ๐๐ , where ๐ โ โณ๐ร๐ (๐), ๐ โ โณ๐ร๐ (๐) are suitable invertible matrices, we may assume without loss of generality that ๐ and ๐ are real almost upper triangular Jordan forms, i.e., direct sums of real Jordan blocks as in (1.1). Furthermore, using induction on ๐ and on ๐, we can assume that actually ๐ = ๐ฝ2๐ (๐ ยฑ ๐๐), ๐ = ๐ฝ2๐ (๐โฒ ยฑ ๐๐ โฒ ), where
Jordan Structures and Lattices of Invariant Subspaces
575
๐, ๐โฒ โ R, ๐, ๐ โฒ > 0. If ๐ + ๐๐ โ= ๐โฒ + ๐๐ โฒ , the only solution is ๐ = 0, so we may assume ๐ = ๐โฒ and ๐ = ๐ โฒ . We can also take ๐ = 0. Now the result follows by elementary calculations using the easily veri๏ฌable fact that an equation ] ] [ [ 0 ๐ 0 ๐ + ๐, ๐ โ R2ร2 , ๐ โ ๐, ๐=๐ โ๐ 0 โ๐ 0 holds if and only if ๐ โ ๐ and ๐ = 0.
โก
Lemma 2.5. (a) Assume that ๐ด โ R๐ร๐ , ๐ต โ R๐ร๐ are either nilpotent, or ๐(๐ด) = ๐(๐ต) = {๐, โ๐}. Then LatR (๐ด) and LatR (๐ต) are isomorphic if and only if ๐ = ๐ and ๐ด and ๐ต have the same R-Jordan structure. (b) Assume that ๐ด โ R๐ร๐ , ๐ต โ R๐ร๐ are such that ๐ด is nilpotent and ๐(๐ต) = {๐, โ๐}. Then Then LatR (๐ด) and LatR (๐ต) are isomorphic if and only if ๐ด and ๐ต have the same weak R-Jordan structure. Proof. Let ๐ : ๐ โ C be the standard algebra isomorphism de๏ฌned by [ ] ๐ ๐ ๐ = ๐ + ๐๐, โ๐ ๐ and extend it entrywise to matrices; thus, for ๐ = [๐๐,๐ ]๐,๐ ๐=1,๐=1 โ โณ๐ร๐ , let ๐ร๐ โ C . ๐(๐ ) = [๐(๐๐,๐ )]๐,๐ ๐=1,๐=1 First, we prove the claim that if ๐ โ โณ๐ร๐ has no real eigenvalues, then LatR (๐ ) is isomorphic to LatC (๐(๐ )). It su๏ฌces to consider the case when ๐ has a real Jordan form and ๐(๐ ) = {ยฑ๐}. Every 2๐-dimensional subspace ๐ฉ โ LatR (๐ ) is the column space of a real matrix ๐ with linearly independent columns that satis๏ฌes equation of the form ๐ ๐ = ๐๐, where ๐ is a real Jordan form, in particular ๐ โ โณ๐ร๐ (๐). By Lemma 2.4, ๐ โ โณ๐ร๐ (๐). The column space ๐ฉ โฒ of the matrix ๐(๐) is obviously ๐(๐ )-invariant. We let ๐(๐ฉ ) = ๐ฉ โฒ . It turns out that ๐ is well de๏ฌned (i.e., ๐(๐ฉ ) depends only on the column space of ๐, and does not depend on the choice of ๐ itself), and is actually a lattice isomorphism between LatR (๐ ) and LatC (๐(๐ )). Indeed, assume that ๐, ๐ โฒ โ โณ๐ร๐ (๐) with linearly independent columns give rise to the same subspace ๐ฉ , i.e., ๐ = ๐ โฒ ๐ for some invertible real matrix ๐ . Since the kernel of ๐ โฒ is zero, we have that ๐(๐ โฒ ) has also zero kernel, and there exists ๐ โฒ โ C๐ร๐ such that ๐ โฒ ๐(๐ โฒ ) = ๐ผ, or ๐โ1 (๐ โฒ )๐ โฒ = ๐ผ. Now ๐ = ๐โ1 (๐ โฒ )๐ obviously belongs to โณ๐ร๐ (๐), and ๐(๐), ๐(๐ โฒ ) have the same column space, as claimed. If ๐ฉ1 โ ๐ฉ2 , ๐ฉ1 , ๐ฉ2 โ LatR (๐ ), then for the corresponding matrices ๐1 and ๐2 we have ๐1 = ๐2 ๐ for some real matrix ๐ , and as before we show that ๐ โ โณ๐โฒ ร๐โฒ (๐) for appropriate ๐โฒ , ๐ โฒ , hence also ๐(๐ฉ1 ) โ ๐(๐ฉ2 ). All other parts or our claim are easily veri๏ฌed. Proof of Part (a). We may assume that both ๐ด and ๐ต are in the real Jordan form. Then the โifโ part is obvious because we may take ๐ด and ๐ต equal. The โonly ifโ part in case ๐ด and ๐ต are nilpotent, is a particular case of [8, Theorem 2.1]. The โonly ifโ part in case ๐(๐ด) = ๐(๐ต) = {๐, โ๐} follows by using the claim stated and proved in the preceding paragraph.
576
L. Rodman
Part (b) is proved similarly using the isomorphism of LatR (๐ต) and LatC (๐(๐ต)), and [8, Theorem 2.1]. โก We also need the following known fact: Proposition 2.6. Let ๐ด โ F๐ร๐ . Then the maximal number of distinct elements in an increasing (by inclusion) chain of ๐ด-invariant subspaces in F๐ is ๐ + 1 in the complex case, and is โ โ 1 1+ (algebraic multiplicity of ๐)+ (algebraic multiplicity of ๐) 2 ๐โ๐(๐ด)โฉR
๐โ๐(๐ด)โฉ(CโR)
in the real case. Proof. The complex case is obvious by using the Jordan form. In the real case, use the fact that every ๐ด-invariant subspace is the direct sum of its intersections with the root subspaces of ๐ด, thereby reducing the proof to the cases when ๐(๐ด) = ๐, ๐ โ R, or ๐(๐ด) = {๐ ยฑ ๐๐}, ๐ โ R, ๐ > 0. The former case is obvious again by using the real Jordan form, and the latter case follows from Lemma 2.5(b). โก Proof. We prove Theorem 2.1. Part 1 was proved in [8], see also [7]. Note that in [8, 7] it was assumed from the beginning that ๐ = ๐; however, (1b) easily implies that ๐ = ๐: Indeed, for ๐ด โ C๐ร๐ , a maximal increasing (by inclusion) chain of ๐ด-invariant subspaces has exactly ๐ + 1 elements (Proposition 2.6). We prove Part 2. The implication (2a) =โ (2c) follows as in the complex case (see [8, 7]), while (2c) =โ (2b) is trivial. We provide details for the proof of (2b) =โ (2a), following for the large part a line of argument analogous to that of [8]. Suppose that ๐ : LatR (๐ด) โโ LatR (๐ต) is a lattice isomorphism with the property (2.1). Let ๐1 , . . . , ๐๐ be all the distinct real eigenvalues of ๐ด, and let ๐๐+1 ยฑ ๐๐๐+1 , . . . , ๐๐ ยฑ ๐๐๐ be all the distinct pairs of complex conjugate nonreal eigenvalues of ๐ด. Let ๐ฉ๐ = ๐(โ๐๐ (๐ด)) for ๐ = 1, 2, . . . , ๐, and ๐ฉ๐ = ๐(โ๐๐ ยฑ๐๐๐ (๐ด)) for ๐ = ๐ + 1, ๐ + 2, . . . , ๐. Then R๐ is a direct sum of ๐ต-invariant subspaces ๐ฉ1 , . . . , ๐ฉ๐ . We claim that ๐(๐ตโฃ๐ฉ๐ ) โฉ ๐(๐ตโฃ๐ฉ๐ ) = โ
, for ๐ โ= ๐ (๐, ๐ = 1, 2, . . . , ๐). Indeed, assume the contrary, i.e., ๐0 โ ๐(๐ตโฃ๐ฉ๐ ) โฉ ๐(๐ตโฃ๐ฉ๐ ) for some ๐ โ= ๐. Consider ๏ฌrst the case when ๐0 is real. Let ๐ฉ = Span (๐ฆ1 + ๐ฆ2 ), where ๐ฆ1 , resp. ๐ฆ2 , are eigenvectors of ๐ตโฃ๐ฉ๐ , resp. ๐ตโฃ๐ฉ๐ , corresponding to the eigenvalue ๐0 . Clearly, ๐ฉ is ๐ต-invariant. Let โณ := ๐ โ1 (๐ฉ ) โ LatR (๐ด). Since โณ must contain a two-dimensional ๐ด-invariant subspace (in the case ๐(๐ดโฃโณ ) is nonreal) or a onedimensional ๐ด-invariant subspace (in the case ๐ดโฃโณ has a real eigenvalue), and since ๐ is a lattice isomorphism, it follows that โณ has dimension two (in the case ๐(๐ดโฃโณ ) is nonreal) or dimension one (in the case ๐ดโฃโณ has a real eigenvalue). Therefore, โณ โ โ๐๐ (๐ด) or โณ โ โ๐๐ ยฑ๐๐๐ (๐ด) for some ๐. This implies ๐ฉ โ ๐ฉ๐ ห โ
โ
โ
+๐ฉ ห ๐ is a for some ๐, ๐ = 1, 2, . . . , ๐, a contradiction with the fact that ๐ฉ1 + direct sum.
Jordan Structures and Lattices of Invariant Subspaces
577
Now consider the case when ๐0 = ๐0 + ๐๐0 , ๐0 โ R, ๐0 โ R โ {0}, is nonreal. Then there are linearly independent vectors ๐ฆ1 , ๐ฆ1โฒ โ ๐ฉ๐ and ๐ฆ2 , ๐ฆ2โฒ โ ๐ฉ๐ such that ๐ต๐ฆ๐ = ๐0 ๐ฆ๐ โ ๐0 ๐ฆ๐โฒ , ๐ฆ2 , ๐ฆ1โฒ
๐ต๐ฆ๐โฒ = ๐0 ๐ฆ๐ + ๐0 ๐ฆ๐โฒ ,
for ๐ = 1, 2.
๐ฆ2โฒ }.
Let ๐ฉ = Span {๐ฆ1 + + Then ๐ฉ is ๐ต-invariant, but does not properly contain any nonzero ๐ต-invariant subspace. So for โณ := ๐ โ1 (๐ฉ ) we obtain as in the preceding paragraph that either โณ โ โ๐๐ (๐ด) or โณ โ โ๐๐ ยฑ๐๐๐ (๐ด) for some ๐. It follows that ๐ฉ โ ๐ฉ๐ for some ๐, ๐ = 1, 2, . . . , ๐, and we obtain a contradiction as before. Next we show that each ๐ฉ๐ is actually a real root subspace for ๐ต. Indeed, assuming the contrary, for some ๐ we have ห โ
โ
โ
+โ ห ๐โฒ (๐ต)+โ ห ๐โฒ ยฑ๐๐ โฒ (๐ต)+ ห โ
โ
โ
+โ ห ๐โฒ ยฑ๐๐ โฒ (๐ต), ๐ฉ๐ = โ๐โฒ1 (๐ต)+ ๐ ๐+1 ๐+1 โ โ โฒ , . . . , ๐โฒโ ยฑ ๐๐โโฒ where ๐โฒ1 , . . . , ๐โฒ๐ are some distinct real eigenvalues of ๐ต, ๐โฒ๐+1 ยฑ ๐๐๐+1 are some distinct pairs of nonreal complex conjugate eigenvalues of ๐ต, and โ > 1. Letting โณ๐ = ๐ โ1 (โ๐โฒ๐ (๐ต)) for ๐ = 1, 2, . . . , ๐, and โณ๐ = ๐ โ1 (โ๐โฒ๐ ยฑ๐๐๐โฒ (๐ต)) for ๐ = ๐ + 1, ๐ + 2, . . . , โ, we have
ห โ
โ
โ
+โณ ห ๐ +โณ ห ๐+1 + ห โ
โ
โ
+โณ ห โ, โ๐๐ (๐ด) = โณ1 + if ๐ โ {1, 2, . . . , ๐} and ห โ
โ
โ
+โณ ห ๐ +โณ ห ๐+1 + ห โ
โ
โ
+โณ ห โ, โ๐๐ ยฑ๐๐๐ (๐ด) = โณ1 + if ๐ โ {๐ + 1, ๐ + 2, . . . , ๐}. Assume ๐ โ {๐ + 1, ๐ + 2, . . . , ๐}, and let ๐ฆ1 , ๐ฆ1โฒ โ โณ1 , ๐ฆ2 , ๐ฆ2โฒ โ โณ2 be linearly independent vectors such that ๐ด๐ฆ๐ = ๐๐ ๐ฆ๐ โ ๐๐ ๐ฆ๐โฒ , ๐ฆ2 , ๐ฆ1โฒ
๐ด๐ฆ๐โฒ = ๐๐ ๐ฆ๐ + ๐๐ ๐ฆ๐โฒ ,
for ๐ = 1, 2,
๐ฆ2โฒ ).
and let โณ = Span (๐ฆ1 + + Then ๐(โณ) is ๐ต-invariant, is contained in ๐ฉ๐ , but is not contained in any of โ๐โฒ๐ (๐ต) or โ๐โฒ๐ ยฑ๐๐๐โฒ (๐ต). This is impossible, because โณ does not properly contain any nonzero ๐ด-invariant subspace, and therefore ๐(โณ) does not properly contain any nonzero ๐ต-invariant subspace. If ๐ โ {1, 2, . . . , ๐}, then we obtain a contradiction in a similar way, by considering the ๐ด-invariant subspace Span(๐ฅ1 +๐ฅ2 ), where ๐ฅ1 and ๐ฅ2 are eigenvectors of ๐ดโฃโณ1 and ๐ดโฃโณ2 , respectively (cf. the proof of [8, Theorem 2.1]). Thus, we must have โ = 1. We have proved that every ๐ฉ๐ is a root subspace of ๐ต corresponding either to a real eigenvalue, or to a pair of nonreal complex conjugate eigenvalues. We also notice that LatR (๐ดโฃโ๐๐ (๐ด) ) is isomorphic to LatR (๐ตโฃ๐ฉ๐ ), for ๐ = 1, 2, . . . , ๐, and LatR (๐ดโฃโ๐๐ ยฑ๐๐๐ (๐ด) ) is isomorphic to LatR (๐ตโฃ๐ฉ๐ ) for ๐ = ๐ + 1, ๐ + 2, . . . , ๐. Using Lemma 2.5 and the condition (2.1), we easily see that ๐ด and ๐ต have the same strong R-Jordan structure. Proof of Part 3. (3a) =โ (3b) follows from Lemma 2.5, by considering root subspaces of ๐ด and ๐ต associated with eigenvalues that correspond under the permutation ๐ of (2). Assume now (3b) holds, and let ๐ : LatR (๐ด) โโ LatR (๐ต) be a lattice isomorphism. As in the proof of Part 2, assuming that ๐1 , . . . , ๐๐ be all the dis-
578
L. Rodman
tinct real eigenvalues of ๐ด, and ๐๐+1 ยฑ ๐๐๐+1 , . . . , ๐๐ ยฑ ๐๐๐ be all the distinct pairs of complex conjugate nonreal eigenvalues of ๐ด, we obtain that the images ๐ฉ๐ := ๐(โ๐๐ (๐ด)) (๐ = 1, 2, . . . , ๐) and ๐ฉ๐ := ๐(๐
๐๐ ยฑ๐๐๐ (๐ด)) (๐ = ๐ = 1, ๐ + 2, . . . , ๐) are root subspaces of ๐ต. It follows also that LatR (๐ดโฃโ๐๐ (๐ด) ) is isomorphic to LatR (๐ตโฃ๐ฉ๐ ), for ๐ = 1, 2, . . . , ๐, and LatR (๐ดโฃโ๐๐ ยฑ๐๐๐ (๐ด) ) is isomorphic to LatR (๐ตโฃ๐ฉ๐ ) for ๐ = ๐ + 1, ๐ + 2, . . . , ๐. Now Lemma 2.5 shows that ๐ด and ๐ต have the same weak R-Jordan structure. โก
3. Structure preserving neighborhoods Let ๐ด โ F๐ร๐ , and let ฮฉ be a non-empty set of distinct eigenvalues of ๐ด; ฮฉ = {๐1 , . . . , ๐๐ }. In this section, it will be always assumed that, in the case F = R, ฮฉ is closed under complex conjugation. For a ๏ฌxed positive ๐ฟ, the {ฮฉ; ๐ฟ}F -structure preserving neighborhood of ๐ด is de๏ฌned to consist of all matrices ๐ต โ F๐ร๐ that satisfy the following properties: 1. โฅ๐ต โ ๐ดโฅ < ๐ฟ; 2. for every ๐ = 1, 2, . . . , ๐, there exists exactly one eigenvalue, call it ๐๐ (๐ต), of ๐ต in the open disc ๐ท(๐๐ ; ๐ฟ) := {๐ค โ C : โฃ๐ค โ ๐๐ โฃ < ๐ฟ}, perhaps of high multiplicity, and the partial multiplicities of the eigenvalue ๐๐ (๐ต) of ๐ต are identical with those of the eigenvalue ๐๐ of ๐ด. In the above de๏ฌnition, one should think of ๐ฟ as small โ smaller than a ๏ฌxed number which depends only on ๐ด. If ๐ด โ R๐ร๐ , and ฮฉ consists of (not necessarily all) distinct real eigenvalues, then the eigenvalues contained in the discs ๐ท(๐๐ ; ๐ฟ), ๐๐ โ ฮฉ, of any ๐ต โ R๐ร๐ that belongs to the {ฮฉ; ๐ฟ}R -structure preserving neighborhood of ๐ด, are necessarily real (assuming ๐ฟ is su๏ฌciently small). Proposition 3.1. Let ๐ด โ F๐ร๐ and let ๐1 , . . . , ๐๐ be the distinct eigenvalues of ๐ด with algebraic multiplicities ๐ผ1 , . . . , ๐ผ๐ , respectively. Let 1 0 < ๐ฟโฒ < min โฃ๐ โ ๐โฃ. 2 ๐,๐โ๐(๐ด), ๐โ=๐ Then for every ๐ฟ > 0 such that ๐ฟ โค ๐ฟโฒ
and
3.46๐(2โฅ๐ดโฅ + ๐ฟ)๐โ1 ๐ฟ โค (๐ฟ โฒ )๐ , ๐ร๐
(3.1) โฒ
we have the property that if ๐ต โ F , โฅ๐ต โ ๐ดโฅ < ๐ฟ, then the disk ๐ท(๐๐ ; ๐ฟ ) contains exactly ๐ผ๐ eigenvalues of ๐ต (counted with multiplicities), for ๐ = 1, 2, . . . , ๐. Proposition 3.1 is a consequence of the main result of [9]; the constant 3.46, which is an improvement on results obtained earlier in [2, 11], is taken from there. We do not aim at the best possible constant in this proposition. Theorem 3.2. Let ๐ด โ F๐ร๐ . Then for every ๐ฟ > 0 su๏ฌciently small, and for every nonempty set ฮฉ of distinct eigenvalues of ๐ด, the following statements are equivalent:
Jordan Structures and Lattices of Invariant Subspaces
579
(๐ผ) ๐ต belongs in the {ฮฉ; ๐ฟ}F-structure preserving neighborhood of ๐ด; (๐ฝ) โฅ๐ต โ ๐ดโฅ < ๐ฟ, and the lattices of invariant subspaces of ๐ดโฃIm ๐ฮฉ (๐ด) and of ๐ตโฃIm ๐โช ๐ท(๐,๐ฟโฒ ) (๐ต) are isomorphic; ๐โฮฉ (๐พ) โฅ๐ต โ ๐ดโฅ < ๐ฟ, and the lattices of invariant subspaces of ๐ดโฃIm ๐ฮฉ (๐ด) and of ๐ตโฃIm ๐โช ๐ท(๐,๐ฟโฒ ) (๐ต) are linearly isomorphic. ๐โฮฉ
Here, ๐ฟ โฒ is taken from Proposition 3.1. Proof. In the complex case, the result follows from Theorem 2.1, Part 1, combined with Proposition 3.1. Consider now the real case. Suppose (๐ผ) holds. Let ๐ฟ โฒ and ๐ฟ > 0 be as in Proposition 3.1. Then, we see in view of Proposition 3.1 that ๐ดโฃIm ๐๐ (๐ด) and ๐ตโฃIm ๐๐ท(๐,๐ฟโฒ ) (๐ต) have the same strong R-Jordan structure for every real ๐ โ ฮฉ, and ๐ดโฃIm ๐๐ยฑ๐๐ (๐ด) and ๐ตโฃIm ๐๐ท(๐+๐๐,๐ฟโฒ )โช๐ท(๐โ๐๐,๐ฟโฒ ) (๐ต) have the same strong R-Jordan structure for every pair ๐ ยฑ ๐๐ โ ฮฉ, ๐ โ R, ๐ > 0. Theorem 2.1, Part 2 now yields (๐พ). Since (๐พ) โ (๐ฝ) is trivial, it remains to prove that (๐ฝ) โ (๐ผ). Thus, assume (๐ฝ) holds. By Theorem 2.1, Part 3, ๐ดโฃIm ๐ฮฉ (๐ด)
and ๐ตโฃIm ๐โช
โฒ (๐ต) ๐โฮฉ ๐ท(๐,๐ฟ )
(3.2)
have the same weak R-Jordan structure. It will be convenient to write ฮฉ = {๐1 , . . . , ๐๐ , ๐๐+1 ยฑ ๐๐๐+1 , . . . , ๐๐ ยฑ ๐๐๐ },
(3.3)
where ๐1 , . . . , ๐๐ are distinct real numbers, and ๐๐+1 ยฑ ๐๐๐+1 , . . . , ๐๐ ยฑ ๐๐๐ are distinct pairs of nonreal complex conjugate numbers. Then (โช๐โฮฉ ๐ท(๐, ๐ฟ โฒ )) โฉ ๐(๐ต) = {๐โฒ1 , . . . , ๐โฒ๐โฒ , ๐โฒ๐โฒ +1 ยฑ ๐๐๐โฒ โฒ +1 , . . . , ๐โฒ๐ ยฑ ๐๐๐โฒ }
(3.4)
where ๐โฒ1 , . . . , ๐โฒ๐โฒ are distinct real numbers, and ๐โฒ๐โฒ +1 ยฑ ๐๐๐โฒ โฒ +1 , . . . , ๐โฒ๐ ยฑ ๐๐๐โฒ are distinct pairs of nonreal complex conjugate numbers. Since (3.2) have the same weak R-Jordan structure, the number ๐ is the same in (3.3) and (3.4). Using Proposition 3.1, it is easy to see that in every disc ๐ท(๐๐ + ๐๐๐ ; ๐ฟ โฒ ), ๐ = ๐ + 1, . . . , ๐, there is only one, necessarily nonreal, eigenvalue of ๐ต (of algebraic multiplicity equal to that of ๐๐ +๐๐๐ as an eigenvalue of ๐ด); otherwise, we obtain a contradiction with the number ๐ being the same in (3.3) and (3.4). On the other hand, there may be either exactly one real eigenvalue or exactly one pair of nonreal complex conjugate eigenvalues of ๐ต in every disc ๐ท(๐๐ ; ๐ฟ โฒ ), for ๐ = 1, 2, . . . , ๐. Thus, ๐โฒ โค ๐, and we may arrange (3.3) and (3.4) so that ๐โฒ๐ โ ๐ท(๐๐ ; ๐ฟ โฒ ) for ๐ = 1, 2, . . . , ๐โฒ ; ๐โฒ๐ + ๐๐๐โฒ โ ๐ท(๐๐ + ๐๐๐ ; ๐ฟ โฒ ) for ๐ = ๐ + 1, ๐ + 2, . . . , ๐; and ๐โฒ๐ + ๐๐๐โฒ โ ๐ท(๐๐ ; ๐ฟ โฒ ) for ๐ = ๐โฒ + 1, . . . , ๐. In fact, ๐โฒ = ๐. Indeed, by Proposition 2.6, the maximal number of elements in an increasing chain of ๐ดโฃIm ๐ฮฉ (๐ด) -invariant subspaces is 1+
๐ โ ๐=1
(algebraic multiplicity of ๐๐ ) +
๐ โ ๐=๐+1
(algebraic multiplicity of ๐๐ + ๐๐๐ ), (3.5)
580
L. Rodman
whereas that number for ๐ตโฃIm ๐โช
โฒ (๐ต) ๐โฮฉ ๐ท(๐,๐ฟ )
โฒ
1+
๐ โ
(algebraic multiplicity of ๐โฒ๐ ) +
๐=1
is
๐ โ
(algebraic multiplicity of ๐โฒ๐ + ๐๐๐โฒ ).
๐=๐โฒ +1
(3.6) The numbers (3.5) and (3.6) cannot be equal unless ๐โฒ = ๐, on the other hand, (3.5) and (3.6) must be the same in view of the assumption (๐ฝ). Thus, ๐โฒ = ๐. It will be convenient to change notation, and let ๐1 , . . . , ๐๐ , resp. ๐1โฒ , . . . , ๐๐โฒ be all eigenvalues of ๐ดโฃIm ๐ฮฉ (๐ด) , resp. of ๐ตโฃIm ๐โช ๐ท(๐,๐ฟโฒ ) (๐ต) , with nonnegative imag๐โฮฉ inary parts arranged so that ๐๐โฒ โ ๐ท(๐๐ , ๐ฟ โฒ ), for ๐ = 1, . . . , ๐. Denote by ๐ผ๐ = (๐ผ๐,1 โฅ ๐ผ๐,2 โฅ โ
โ
โ
โฅ ๐ผ๐,๐ โฅ โ
โ
โ
),
๐ = 1, 2, . . . , ๐,
the sequence of partial multiplicities of the eigenvalue ๐๐ of ๐ด, arranged in the nondecreasing order and extended inde๏ฌnitely by zeros, and similarly ๐ผโฒ๐ = (๐ผโฒ๐,1 โฅ ๐ผโฒ๐,2 โฅ โ
โ
โ
โฅ ๐ผโฒ๐,๐ โฅ โ
โ
โ
),
๐ = 1, 2, . . . , ๐,
for the eigenvalue ๐๐โฒ of ๐ต. At this point we recall the well-known majorization relation between nonincreasing sequences of nonnegative integers having ๏ฌnite sum. Let ๐ผ = (๐ผ1 โฅ ๐ผ2 โฅ โ
โ
โ
โฅ ๐ผ๐ โฅ โ
โ
โ
), ๐ฝ = (๐ฝ1 โฅ ๐ฝ2 โฅ โ
โ
โ
โฅ ๐ฝ๐ โฅ โ
โ
โ
) be two such sequences. We say that ๐ฝ majorizes ๐ผ, notation: ๐ฝ เชฐ ๐ผ if ๐ โ ๐=1
๐ฝ๐ โฅ
๐ โ
๐ผ๐ ,
๐ = 1, 2, . . . ,
๐=1
and
โ โ
๐ฝ๐ =
๐=1
โ โ
๐ผ๐ .
๐=1
A particular case of the main result of [10, 3] shows that ๐ผโฒ๐ เชฐ ๐ผ๐ ,
๐ = 1, 2, . . . , ๐,
(3.7)
if ๐ฟ is su๏ฌciently small. (We use here the facts that ๐๐โฒ is the only eigenvalue of ๐ต in the disc ๐ท๐๐ ,๐ฟโฒ and that ๐ = ๐โฒ .) Now let ๐ : {1, 2, . . . , ๐} be the permutation that exists by the de๏ฌnition of ๐ดโฃIm ๐ฮฉ (๐ด) and ๐ตโฃIm ๐โช ๐ท(๐,๐ฟโฒ ) (๐ต) having the same ๐โฮฉ weak R-Jordan structure, and let (๐1 , . . . , ๐๐ฃ ) be a cycle in ๐. Then using (3.7) we have ๐ผ๐1 = ๐ผโฒ๐2 เชฐ ๐ผ๐2 = ๐ผโฒ๐3 เชฐ ๐ผ๐3 = โ
โ
โ
เชฐ ๐ผ๐๐ฃ = ๐ผโฒ๐1 เชฐ ๐ผ๐1 . Thus, the equality holds throughout. Repeating this argument for every cycle of ๐, we see that we can take ๐ to be the identity. This proves (๐ผ). โก The proof of Theorem 3.2 show that in the complex case the theorem holds for every ๐ฟ > 0 satisfying (3.1), and in the real case the theorem holds for every ๐ฟ > 0 satisfying (3.1) and for which (3.7) is valid.
Jordan Structures and Lattices of Invariant Subspaces
581
4. Concluding remarks Theorems 2.1 (Part 3) and 3.2 allow us to extend the main result of [8] to real matrices, with essentially the same proof. We only formulate the result, omitting details of proof. For a given ๐ด โ R๐ร๐ , let ฮฅ(๐ด, ๐ต) = inf{โฅ๐ผ โ ๐โฅ}, where the in๏ฌmum is taken over all invertible matrices ๐ โ R๐ร๐ such that โณ โ LatR (๐ด)
โโ
๐(โณ) โ LatR (๐ต).
(4.1)
Remark 4.1. In view of Theorems 2.1 (Part 3) and 3.2, there exists ๐ฟ > 0 (depending on ๐ด only) such that the set of invertible ๐ โ R๐ร๐ with the property (4.1) is nonempty as soon as โฅ๐ต โ ๐ดโฅ < ๐ฟ and ๐ต and ๐ด have the same weak R-Jordan structure. Let dist (LatR (๐ด), LatR (๐ต)) = max
{
โณโLatR (๐ด)
sup
๐ฉ โLatR (๐ต)
inf
โฅ๐๐ฉ โ ๐โณ โฅ,
sup
inf
} โฅ๐๐ฉ โ ๐โณ โฅ
โณโLatR (๐ต)
๐ฉ โLatR (๐ด)
be the distance between the lattice of invariant subspaces of ๐ด โ R๐ร๐ and that of ๐ต โ R๐ร๐ ; here ๐๐ฉ is the orthogonal projection on the subspace ๐ฉ . Note that dist (LatR (๐ด), LatR (๐ต)) โค 1
(4.2)
for all ๐ด, ๐ต โ R๐ร๐ , as it follows, for example, from [4, Theorem S4.5]. Theorem 4.2. Given ๐ด โ R๐ร๐ , there exists ๐ฟ > 0 such that sup
ฮฅ(๐ด, ๐ต) < โ, โฅ๐ต โ ๐ดโฅ
(4.3)
where the supremum is taken over all ๐ต โ R๐ร๐ that satisfy โฅ๐ต โ ๐ดโฅ < ๐ฟ and have the same weak R-Jordan structure as ๐ด does. Moreover, sup
dist (LatR (๐ด), LatR (๐ต)) < โ, โฅ๐ต โ ๐ดโฅ
(4.4)
where the supremum is taken over all ๐ต โ R๐ร๐ which have the same weak RJordan structure as ๐ด does. Using Remark 4.1, one proves as in [8] that (4.3) holds for su๏ฌciently small ๐ฟ > 0. In view of [8, Theorem 3.1] (which is valid in the real case as well), we have that (4.4) holds provided { ( )โ1 } 1 ฮฅ(๐ด, ๐ต) โฅ๐ต โ ๐ดโฅ < min ๐ฟ, sup (4.5) 2 โฅ๐ต โ ๐ดโฅ
582
L. Rodman
and ๐ด and ๐ต have the same weak R-Jordan structure; here ๐ฟ > 0 is such that (4.3) holds. Using (4.2), we obtain the following inequality for every ๐ต having the same weak R-Jordan structure as ๐ด: dist (LatR (๐ด), LatR (๐ต)) โค min{1/๐ฟ0 , ๐ } โฅ๐ต โ ๐ดโฅ, where ๐ is the supremum in (4.3), and ๐ฟ0 is the right-hand side of (4.5).
References [1] T. Bella, V. Olshevsky, and U. Prasad, Lipschitz stability of canonical Jordan bases of ๐ป-selfadjoint matrices under structure-preserving perturbations. Linear Algebra Appl. 428 (2008), 2130โ2176. [2] R. Bhatia, L. Elsner, and G. Krause, Bounds for the variation of the roots of a polynomial and the eigenvalues of a matrix. Linear Algebra Appl. 142 (1990), 195โ 209. [3] H. den Boer and G.Ph.A. Thijsse, Semistability of sums of partial multiplicities under additive perturbation. Integral Equations Operator Theory 3 (1980), 23โ42. [4] I. Gohberg, P. Lancaster, and L. Rodman, Matrix Polynomials. Academic Press, 1982; republication, SIAM 2009. [5] I. Gohberg, P. Lancaster, and L. Rodman, Matrices and Inde๏ฌnite Scalar Products. Birkhยจ auser Verlag, Basel, 1983. [6] I. Gohberg, P. Lancaster, and L. Rodman. Inde๏ฌnite Linear Algebra and Applications. Birkhยจ auser Verlag, 2005. [7] I. Gohberg, P. Lancaster, and L. Rodman. Invariant Subspaces of Matrices with Applications, J. Wiley, New York, 1986; republication, SIAM, 2006. [8] I. Gohberg and L. Rodman, On the distance between lattices of invariant subspaces of matrices. Linear Algebra Appl. 76 (1986), 85โ120. [9] G.M. Krause, Bounds for the variation of matrix eigenvalues and polynomial roots. Linear Algebra Appl. 208/209 (1994), 73โ82. [10] A.S. Markus and E.E. Parilis, The change of the Jordan structure of a matrix under small perturbations. Linear Algebra Appl. 54 (1983), 139โ152. [11] D. Phillips, Improving spectral-variation bounds with Chebyshev polynomials. Linear Algebra Appl. 133 (1990), 165โ173. [12] L. Rodman, Similarity vs unitary similarity and perturbation analysis of sign characteristic: Complex and real inde๏ฌnite inner products. Linear Algebra Appl. 416 (2006), 945โ1009. Leiba Rodman Department of Mathematics College of William and Mary Williamsburg, VA 23187-8795, USA e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 583โ612 c 2012 Springer Basel AG โ
Pseudospectral Functions for Canonical Di๏ฌerential Systems. II J. Rovnyak and L.A. Sakhnovich To the memory of Israel Gohberg
Abstract. A spectral theory is constructed for canonical di๏ฌerential systems whose Hamiltonians have selfadjoint matrix values. In contrast with the case of nonnegative Hamiltonians, eigenvalues in general can be complex, and root functions as well as eigenfunctions come into play. Eigentransforms are de๏ฌned and turn out to be isometric on the span of root functions with respect to a suitably de๏ฌned inde๏ฌnite inner product on entire functions. Mathematics Subject Classi๏ฌcation (2000). Primary 34L10; Secondary 47B50, 47E05, 46C20, 34B09. Keywords. Canonical di๏ฌerential equation, root function, pseudospectral function, spectral function, inde๏ฌnite inner product, eigentransform.
1. Introduction We are concerned with the spectral theory of canonical di๏ฌerential systems ๐๐ = ๐๐ง๐ฝ๐ป(๐ฅ)๐, 0 โค ๐ฅ โค โ, ๐๐ฅ [ (1.1) ] ๐ผ๐ 0 ๐ (0, ๐ง) = 0. In (1.1), we assume that [ 0 ๐ฝ= ๐ผ๐
] ๐ผ๐ , 0
[ ] ๐1 (๐ฅ, ๐ง) ๐ (๐ฅ, ๐ง) = , ๐2 (๐ฅ, ๐ง)
(1.2)
where ๐1 (๐ฅ, ๐ง) and ๐2 (๐ฅ, ๐ง) are ๐-dimensional vector-valued functions, and ๐ง is a complex parameter. As in [3], the Hamiltonian ๐ป(๐ฅ) is assumed to be a measurable 2๐ ร 2๐ matrix-valued function such that โซ โ โ ๐ป(๐ฅ) = ๐ป(๐ฅ) a.e. and โฅ๐ป(๐ฅ)โฅ ๐๐ฅ < โ. (1.3) 0
584
J. Rovnyak and L.A. Sakhnovich
For technical reasons, we also assume throughout that [ ] 0 ๐ป(๐ฅ) = 0 a.e. on [0, โ] =โ ๐ = 0. ๐
(1.4)
For any such system we de๏ฌne ๐ฟ2 (๐ป๐๐ฅ) as a Kreหฤฑn space of (equivalence classes of) 2๐-dimensional vector-valued functions on [0, โ]. Write ๐ป(๐ฅ) = ๐ป+ (๐ฅ) โ ๐ปโ (๐ฅ) where ๐ปยฑ (๐ฅ) are measurable, ๐ปยฑ (๐ฅ) โฅ 0, and ๐ป+ (๐ฅ)๐ปโ (๐ฅ) = 0 a.e. As a linear space, ๐ฟ2 (๐ป๐๐ฅ) is the set of measurable 2๐-dimensional vector-valued functions ๐ on [0, โ] such that โซ โ ๐ (๐ก)โ [๐ป+ (๐ก) + ๐ปโ (๐ก)]๐ (๐ก) ๐๐ก < โ. 0
Two functions ๐1 and ๐2 in ๐ฟ2 (๐ป๐๐ฅ) are identi๏ฌed if ๐ป(๐ฅ)[๐1 (๐ฅ) โ ๐1 (๐ฅ)] = 0 a.e. Taken with the inner product โซ โ โจ๐1 , ๐2 โฉ๐ป = ๐2 (๐ฅ)โ ๐ป(๐ฅ)๐1 (๐ฅ) ๐๐ฅ, ๐1 , ๐2 โ ๐ฟ2 (๐ป๐๐ฅ), 0
2
๐ฟ (๐ป๐๐ฅ) is a Kreหฤฑn space. In a natural way we can view ๐ฟ2 (๐ปยฑ ๐๐ฅ) as closed subspaces, and then ๐ฟ2 (๐ป๐๐ฅ) = ๐ฟ2 (๐ป+ ๐๐ฅ) โ ๐ฟ2 (๐ปโ ๐๐ฅ) is a fundamental decomposition. De๏ฌne an eigentransform ๐น = ๐ ๐ for any ๐ in ๐ฟ2 (๐ป๐๐ฅ) by โซ โ [ ] 0 ๐ผ๐ ๐ (๐ฅ, ๐งยฏ)โ ๐ป(๐ฅ)๐ (๐ฅ) ๐๐ฅ, ๐น (๐ง) = 0
(1.5)
where ๐ (๐ฅ, ๐ง) is the unique 2๐ ร 2๐ matrix-valued function such that ๐๐ = ๐๐ง๐ฝ๐ป(๐ฅ)๐, ๐๐ฅ ๐ (0, ๐ง) = ๐ผ2๐ ,
0 โค ๐ฅ โค โ, ๐ง โ โ.
(1.6)
The function ๐ (๐ฅ, ๐ง) is continuous on [0, โ]รโ and entire in ๐ง for each ๏ฌxed ๐ฅ. For each ๐ in ๐ฟ2 (๐ป๐๐ฅ), ๐น = ๐ ๐ is an ๐-dimensional vector-valued entire function. Throughout we write [ ] ๐(๐ง) ๐(๐ง) โ ๐ (โ, ๐งยฏ) = . (1.7) ๐(๐ง) ๐(๐ง) Here ๐(๐ง), ๐(๐ง), ๐(๐ง), ๐(๐ง) are ๐ ร ๐ matrix-valued entire functions. Consider ๏ฌrst the de๏ฌnite case, that is, ๐ป(๐ฅ) โฅ 0 a.e. Then ๐ฟ2 (๐ป๐๐ฅ) is a Hilbert space. In this case, by a spectral function for (1.1) is meant a nondecreasing ๐ ร ๐ matrix-valued function ๐ (๐ฅ) of real ๐ฅ such that the eigentransform ๐ acts as an isometry from ๐ฟ2 (๐ป๐๐ฅ) into ๐ฟ2 (๐๐). We call ๐ (๐ฅ) a pseudospectral function for (1.1) if ๐ is a partial isometry from ๐ฟ2 (๐ป๐๐ฅ) into ๐ฟ2 (๐๐ ). Pseudospectral
Pseudospectral Functions
585
functions can be constructed using a boundary condition at the right endpoint of the interval [0, โ]. The boundary condition has the form [ โ ] ๐
๐โ ๐ (โ, ๐ง) = 0, where ๐
and ๐ are ๐ ร ๐ matrices such that ๐
โ ๐ + ๐โ ๐
= 0, and such that the entire function ๐(๐ง)๐
+ ๐(๐ง)๐ has invertible values except at isolated points. Then ๐ฃ(๐ง) = ๐[๐(๐ง)๐
+ ๐(๐ง)๐][๐(๐ง)๐
+ ๐(๐ง)๐]โ1 is meromorphic in the complex plane, ๐ฃ(๐ง) = ๐ฃ(ยฏ ๐ง )โ at all points of analyticity, and ๐ฃ(๐ง) has nonnegative imaginary part in the upper half-plane. In particular, ๐ฃ(๐ง) has only real and simple poles and a representation ] โซ โ[ 1 ๐ก ๐ฃ(๐ง) = ๐ผ + ๐ฝ๐ง + โ ๐๐ (๐ก), (1.8) 1 + ๐ก2 โโ ๐ก โ ๐ง where ๐ (๐ฅ) is a nondecreasing ๐ ร ๐ matrix-valued step function with jumps at the poles of ๐ฃ(๐ง), and ๐ผ = ๐ผโ and ๐ฝ โฅ 0 are constant ๐ร๐ matrices. The function ๐ (๐ฅ) is a pseudospectral function. The isometric set for the eigentransform ๐ is the closed span of eigenfunctions. See [4, Chapter 4] and Theorems 4.2.2, 4.2.4, and 4.2.5 in [3]. In this paper we generalize the preceding constructions to Hamiltonians such that ๐ป(๐ฅ) = ๐ป(๐ฅ)โ a.e. We introduce a meromorphic function ๐ฃ(๐ง) in the same way as before. Now, however, ๐ฃ(๐ง) can have nonreal and nonsimple poles, and in general there is no representation of ๐ฃ(๐ง) in the form (1.8). In place of eigenfunctions, we have to deal now with eigenchains of root functions. The role of a pseudospectral function is replaced by a notion of pseudospectral data, which consists of the collection of poles and principal parts of the meromorphic function ๐ฃ(๐ง). The poles and principal parts of ๐ฃ(๐ง) are used to construct an inner product โจโ
, โ
โฉ on vectorvalued entire functions. According to our main result, Theorem 4.7, the identity โซ โ ๐2 (๐ก)โ ๐ป(๐ก)๐1 (๐ก) ๐๐ก = โจ๐น1 , ๐น2 โฉ 0
holds whenever ๐1 and ๐2 are ๏ฌnite linear combinations of root functions and ๐น1 and ๐น2 are their eigentransforms. This agrees with Theorem 4.1.11 of [3] for the special case when ๐ฃ(๐ง) has only simple poles. The general case turns out to be quite a bit more involved. In Section 2 of the paper, we expand the function ๐ (๐ฅ, ๐ง) in a Taylor series about a point ๐ง = ๐ค. The higher-order coe๏ฌcients in this expansion do not arise in the de๏ฌnite theory, but they are important in the general case considered here. In Section 3 we derive explicit formulas for the root functions and their eigentransforms. These formulas are needed for the main results of the paper, which appear in Section 4. Remark. We thank the referee for the comment that the construction of a related linear operator and its resolvent might yield insights into our main results. We leave this as an open question. Concerning such related linear operators, see the remark
586
J. Rovnyak and L.A. Sakhnovich
preceding Proposition 3.1. See also Section 3 of [3], where resolvent operators for canonical di๏ฌerential systems are investigated.
2. Taylor expansions and their coe๏ฌcients Assume given a system (1.1)โ(1.4). De๏ฌne ๐ (๐ฅ, ๐ง) and ๐(๐ง), ๐(๐ง), ๐(๐ง), ๐(๐ง) as in (1.6) and (1.7). By (1.6), ๐ ๐ (๐ก, ๐งยฏ)โ ๐ฝ๐ (๐ก, ๐ค) = ๐(๐ค โ ๐ง)๐ (๐ก, ๐งยฏ)โ ๐ป(๐ก)๐ (๐ก, ๐ค) ๐๐ก a.e. on [0, โ] for all complex ๐ค and ๐ง. We deduce that ๐ (๐ฅ, ๐งยฏ)โ ๐ฝ๐ (๐ฅ, ๐ง) = ๐ (๐ฅ, ๐ง)๐ฝ๐ (๐ฅ, ๐งยฏ)โ = ๐ฝ, and
Set
โซ 0
โ
๐ (๐ก, ๐งยฏ)โ ๐ป(๐ก)๐ (๐ก, ๐ค) ๐๐ก =
๐ (๐ฅ, ๐ง) =
โ โ
(2.1)
๐ (โ, ๐งยฏ)โ ๐ฝ๐ (โ, ๐ค) โ ๐ฝ . ๐(๐ค โ ๐ง)
๐๐ (๐ฅ, ๐ค)(๐ง โ ๐ค)๐ ,
(2.2)
(2.3)
๐=0
[
] โ โ [ ๐(๐ง) ๐(๐ง) ๐๐ (๐ค) = ๐(๐ง) ๐(๐ง) ๐๐ (๐ค) ๐=0
] ๐๐ (๐ค) (๐ง โ ๐ค)๐ , ๐๐ (๐ค)
(2.4)
for all ๐ฅ in [0, โ] and ๐ค in โ. Using the values ๐ฅ = 0 and ๐ฅ = โ, we get ๐0 (0, ๐ค) = ๐ผ2๐ , ๐๐ (0, ๐ค) = 0, ๐ โฅ 1, [ ] ๐ (๐ค) ยฏ โ ๐๐ (๐ค) ยฏ โ ๐๐ (โ, ๐ค) = ๐ , ๐ โฅ 0. ๐๐ (๐ค) ยฏ โ ๐๐ (๐ค) ยฏ โ
(2.5)
For each ๐ โฅ 0, ๐๐ (๐ฅ, ๐ค) is continuous on [0, โ] and entire in ๐ค for ๏ฌxed ๐ฅ. To prove this, represent the coe๏ฌcients as Cauchy integrals as in (2.8) below and use the corresponding properties for ๐ (๐ฅ, ๐ง). Proposition 2.1. For every ๐ค โ โ, ๐ ๐0 (๐ฅ, ๐ค) = ๐๐ค๐ฝ๐ป(๐ฅ)๐0 (๐ฅ, ๐ค), ๐๐ฅ ๐ ๐๐ (๐ฅ, ๐ค) = ๐๐ค๐ฝ๐ป(๐ฅ)๐๐ (๐ฅ, ๐ค) + ๐๐ฝ๐ป(๐ฅ)๐๐โ1 (๐ฅ, ๐ค), ๐๐ฅ a.e. on [0, โ].
(2.6) ๐ โฅ 1,
Proof. The ๏ฌrst equation in (2.6) holds by (1.6) since ๐0 (๐ฅ, ๐ง) = ๐ (๐ฅ, ๐ง). Since ๐๐ (0, ๐ค) = 0 for ๐ โฅ 1, the second equation in (2.6) is equivalent to โซ ๐ฅ โซ ๐ฅ ๐๐ (๐ฅ, ๐ค) = ๐๐ค๐ฝ ๐ป(๐ก)๐๐ (๐ก, ๐ค) ๐๐ก + ๐๐ฝ ๐ป(๐ก)๐๐โ1 (๐ก, ๐ค) ๐๐ก, (2.7) 0
0
Pseudospectral Functions
587
0 โค ๐ฅ โค โ. Let ฮ be a circular path around ๐ค in the counterclockwise direction. For each ๐ฅ in [0, โ] and ๐ โฅ 0, โซ 1 ๐ (๐ฅ, ๐) ๐๐ (๐ฅ, ๐ค) = ๐๐. (2.8) 2๐๐ ฮ (๐ โ ๐ค)๐+1 To prove (2.7), ๏ฌrst write (1.6) in the form โซ ๐ (๐ฅ, ๐) = ๐ผ2๐ + ๐๐๐ฝ
๐ฅ
0
๐ป(๐ก)๐ (๐ก, ๐) ๐๐ก.
โซ Since we assume ๐ โฅ 1, ฮ ๐๐/(๐ โ ๐ค)๐+1 = 0. Thus ] โซ โซ [ โซ ๐ฅ 1 ๐ (๐ฅ, ๐) 1 ๐๐ ๐๐ = ๐๐๐ฝ ๐ป(๐ก)๐ (๐ก, ๐) ๐๐ก 2๐๐ ฮ (๐ โ ๐ค)๐+1 2๐๐ ฮ (๐ โ ๐ค)๐+1 0 [ โซ โซ ๐ฅ 1 = ๐๐ค๐ฝ ๐ป(๐ก)๐ (๐ก, ๐) ๐๐ก 2๐๐ ฮ 0 ] โซ ๐ฅ ๐๐ + ๐(๐ โ ๐ค)๐ฝ ๐ป(๐ก)๐ (๐ก, ๐) ๐๐ก (๐ โ ๐ค)๐+1 0 โซ ๐ฅ โซ 1 ๐ (๐ก, ๐) = ๐๐ค๐ฝ ๐ป(๐ก) ๐๐ ๐๐ก 2๐๐ ฮ (๐ โ ๐ค)๐+1 0 โซ ๐ฅ โซ 1 ๐ (๐ก, ๐) + ๐๐ฝ ๐ป(๐ก) ๐๐ ๐๐ก. 2๐๐ ฮ (๐ โ ๐ค)๐ 0 By (2.8), this is the same as (2.7). The interchange in order of integration is justi๏ฌed because โฅ๐ป(๐ก)โฅโฅ๐ (๐ก, ๐)โฅ is integrable over [0, โ] ร ฮ. โก Proposition 2.2. For all ๐ค โ โ, ๐ฅ โ [0, โ], and ๐ โฅ 0, โ โ ๐๐ (๐ฅ, ๐ค) ยฏ โ ๐ฝ๐๐ (๐ฅ, ๐ค) = ๐๐ (๐ฅ, ๐ค)๐ฝ๐๐ (๐ฅ, ๐ค) ยฏ โ ๐+๐=๐
๐+๐=๐
{ =
๐ฝ,
๐ = 0,
0,
๐ โฅ 1.
(2.9)
Proof. By (2.1) and (2.3), โ โ
โ
๐
๐๐ (๐ฅ, ๐ค) ยฏ (๐ง โ ๐ค) ๐ฝ
๐=0
=
โ โ ๐=0 โ โ ๐=0
๐๐ (๐ฅ, ๐ค)(๐ง โ ๐ค)๐ ๐๐ (๐ฅ, ๐ค)(๐ง โ ๐ค)๐ ๐ฝ
โ โ
๐๐ (๐ฅ, ๐ค) ยฏ โ (๐ง โ ๐ค)๐ = ๐ฝ.
๐=0
The relations (2.9) follow on expanding the products and collecting powers of ๐งโ๐ค: the constant terms equal ๐ฝ, and all other coe๏ฌcients are zero. โก
588
J. Rovnyak and L.A. Sakhnovich
Corollary 2.3. For every ๐ค โ โ, ๐0 (๐ค)๐0 (๐ค) ยฏ โ + ๐0 (๐ค)๐0 (๐ค) ยฏ โ = 0,
๐0 (๐ค) ยฏ โ ๐0 (๐ค) + ๐0 (๐ค) ยฏ โ ๐0 (๐ค) = 0,
๐0 (๐ค)๐0 (๐ค) ยฏ โ + ๐0 (๐ค)๐0 (๐ค) ยฏ โ = ๐ผ๐ ,
๐0 (๐ค) ยฏ โ ๐0 (๐ค) + ๐0 (๐ค) ยฏ โ ๐0 (๐ค) = ๐ผ๐ , (2.10) ๐0 (๐ค) ยฏ โ ๐0 (๐ค) + ๐0 (๐ค) ยฏ โ ๐0 (๐ค) = 0.
๐0 (๐ค)๐0 (๐ค) ยฏ โ + ๐0 (๐ค)๐0 (๐ค) ยฏ โ = 0, For all ๐ โฅ 1,
โ
[๐๐ (๐ค)๐๐ (๐ค) ยฏ โ + ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ ] = 0,
๐+๐=๐
โ
[๐๐ (๐ค)๐๐ (๐ค) ยฏ โ + ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ ] = 0,
๐+๐=๐
โ
(2.11)
[๐๐ (๐ค)๐๐ (๐ค) ยฏ โ + ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ ] = 0,
๐+๐=๐
and
โ
[๐๐ (๐ค) ยฏ โ ๐๐ (๐ค) + ๐๐ (๐ค) ยฏ โ ๐๐ (๐ค)] = 0,
๐+๐=๐
โ
[๐๐ (๐ค) ยฏ โ ๐๐ (๐ค) + ๐๐ (๐ค) ยฏ โ ๐๐ (๐ค)] = 0,
๐+๐=๐
โ
(2.12)
[๐๐ (๐ค) ยฏ โ ๐๐ (๐ค) + ๐๐ (๐ค) ยฏ โ ๐๐ (๐ค)] = 0.
๐+๐=๐
Proof. These identities follow on choosing ๐ฅ = โ in (2.9) and expanding using (1.7). The relations (2.10) follow from the case ๐ = 0 and coincide with the formulas (2.1.5) of [3]. Suppose ๐ โฅ 1. Then by (2.9), [ ] โ 0 0 = ๐๐ (โ, ๐ค) ยฏ โ ๐ฝ๐๐ (โ, ๐ค) 0 0 ๐+๐=๐ [ ][ ] โ ๐๐ (๐ค) ๐๐ (๐ค) ๐๐ (๐ค) ยฏ โ ๐๐ (๐ค) ยฏ โ = ๐๐ (๐ค) ๐๐ (๐ค) ๐๐ (๐ค) ยฏ โ ๐๐ (๐ค) ยฏ โ ๐+๐=๐ [ ] โ ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ + ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ + ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ = , ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ + ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ + ๐๐ (๐ค)๐๐ (๐ค) ยฏ โ ๐+๐=๐ yielding (2.11). We prove (2.12) in a similar way using (2.9). Proposition 2.4. For all ๐ค, ๐ง โ โ and ๐ โฅ 0, [ ] โซ โ [ ] 0 0 ๐ผ๐ ๐ (๐ก, ๐งยฏ)โ ๐ป(๐ก)๐๐ (๐ก, ๐ค) ๐๐ก ๐ผ๐ 0 โ โ ๐(๐ง)๐๐ (๐ค) โ ยฏ โ + ๐(๐ง)๐๐ (๐ค) ยฏ โ =๐ = ฮ๐๐ (๐ค)(๐ง โ ๐ค)๐ . ๐+1 (๐ง โ ๐ค) ๐+๐=๐ ๐=0
โก
(2.13)
Pseudospectral Functions In (2.13), for all ๐, ๐ โฅ 0, โ ฮ๐๐ (๐ค) = ๐ [๐๐+๐+1 (๐ค)๐๐ (๐ค) ยฏ โ + ๐๐+๐+1 (๐ค)๐๐ (๐ค) ยฏ โ ],
589
(2.14)
๐+๐=๐
and the middle expression is interpreted by continuity for ๐ง = ๐ค. Moreover, [ ] โซ โ [ ] 0 โ 0 ๐ผ๐ ๐๐ (๐ก, ๐ค) ฮ๐๐ (๐ค) = ยฏ ๐ป(๐ก)๐๐ (๐ก, ๐ค) ๐๐ก (2.15) ๐ผ ๐ 0 and
ฮ๐๐ (๐ค) ยฏ โ = ฮ๐๐ (๐ค).
(2.16)
Proof. By (2.2) and (1.7), [ ] โซ โ [ ] 0 0 ๐ผ๐ ๐ (๐ก, ๐งยฏ)โ ๐ป(๐ก)๐ (๐ก, ๐) ๐๐ก (2.17) ๐ผ๐ 0 [ ] ยฏ โ + ๐(๐ง)๐(๐) ยฏ โ [ ] ๐ (โ, ๐งยฏ)โ ๐ฝ๐ (โ, ๐) โ ๐ฝ 0 ๐(๐ง)๐(๐) = 0 ๐ผ๐ =๐ . ๐ผ๐ ๐(๐ โ ๐ง) ๐งโ๐ Using the expansions ๐ (๐ก, ๐) = [ ยฏ โ] ๐(๐) ยฏ โ = ๐(๐)
โ โ
๐๐ (๐ก, ๐ค)(๐ โ ๐ค)๐ ,
๐=0 โ [ โ ๐=0
] ๐๐ (๐ค) ยฏ โ (๐ โ ๐ค)๐ , ๐๐ (๐ค) ยฏ โ
๐ ๐ = ๐งโ๐ ๐งโ๐ค we obtain โซ โ 0
[
โ โ 1 (๐ โ ๐ค)๐ =๐ , ๐โ๐ค (๐ง โ ๐ค)๐+1 ๐=0 1โ ๐งโ๐ค
(2.18) (2.19) (2.20)
[ ] โ โ ] 0 โ 0 ๐ผ๐ ๐ (๐ก, ๐งยฏ) ๐ป(๐ก) ๐๐ (๐ก, ๐ค) (๐ โ ๐ค)๐ ๐๐ก ๐ผ๐ ๐=0 [ ] ยฏ โ [ ] ๐(๐) ๐ = ๐(๐ง) ๐(๐ง) ยฏ โ ๐งโ๐ ๐(๐) [ ] โ โ โ โ [ ] ๐๐ (๐ค) ยฏ โ (๐ โ ๐ค)๐ ๐ ๐(๐ง) ๐(๐ง) = (๐ โ ๐ค) ๐ (๐ง โ ๐ค)๐+1 ๐๐ (๐ค) ยฏ โ ๐=0 ๐=0 =
โ โ ๐=0
(๐ โ ๐ค)๐ ๐
โ ๐(๐ง)๐๐ (๐ค) ยฏ โ + ๐(๐ง)๐๐ (๐ค) ยฏ โ . ๐+1 (๐ง โ ๐ค) ๐+๐=๐
(2.21)
In fact, in (2.21) the ๏ฌrst equality is identical to (2.17) by the Taylor expansion for ๐ (๐ก, ๐) in (2.18); the second equality substitutes the two Taylor expansions in (2.19) and (2.20); the third equality collects powers of ๐ โ ๐ค. The ๏ฌrst equality in (2.13) follows from (2.21) on interchanging the order of integration and summation on the left and comparing coe๏ฌcients.
590
J. Rovnyak and L.A. Sakhnovich
To prove the second equality in (2.13), expand ๐(๐ง) and ๐(๐ง) in Taylor series, and write โ ๐(๐ง)๐๐ (๐ค) ยฏ โ + ๐(๐ง)๐๐ (๐ค) ยฏ โ ๐ (๐ง โ ๐ค)๐+1 ๐+๐=๐ [ ] โ โ โ [ ] ๐๐ (๐ค) ยฏ โ 1 ๐๐ (๐ค) ๐๐ (๐ค) (๐ง โ ๐ค)๐ =๐ โ (๐ง โ ๐ค)๐+1 ๐ ( ๐ค) ยฏ ๐ ๐+๐=๐ ๐=0 [ ] โ โ โ [ ] ๐๐ (๐ค) ยฏ โ ๐ ๐๐+๐+1 (๐ค) ๐๐+๐+1 (๐ค) =๐ โ (๐ง โ ๐ค) ๐ ( ๐ค) ยฏ ๐ ๐+๐=๐ ๐=โ๐โ1 [ ] โ โ โ [ ] ๐๐ (๐ค) ยฏ โ ๐ ๐๐+๐+1 (๐ค) ๐๐+๐+1 (๐ค) =๐ โ (๐ง โ ๐ค) ๐ ( ๐ค) ยฏ ๐ ๐+๐=๐ ๐=0 [ ] โ1 โ โ [ ] ๐๐ (๐ค) ยฏ โ ๐ ๐๐+๐+1 (๐ค) ๐๐+๐+1 (๐ค) +๐ โ (๐ง โ ๐ค) ๐ ( ๐ค) ยฏ ๐ ๐+๐=๐ ๐=โ๐โ1
=
โ โ
ฮ๐๐ (๐ค)(๐ง โ ๐ค)๐ + Term 2.
(2.22)
๐=0
Here Term 2 = 0, since by the ๏ฌrst equality in (2.13), proved above, the left side of (2.22) is entire. Thus (2.22) yields the second equality in (2.13) with ฮ๐๐ (๐ค) de๏ฌned by (2.14). The identity (2.15) follows from (2.13) on expanding ๐ (๐ก, ๐งยฏ)โ in a Taylor series about ๐ง = ๐ค and comparing coe๏ฌcients. Then (2.16) follows from (2.15). โก
3. Root spaces and eigenchains We now add a boundary condition at the right endpoint of the interval [0, โ]. Thus we consider a system [
๐ผ๐
๐๐ = ๐๐ง๐ฝ๐ป(๐ฅ)๐, 0 โค ๐ฅ โค โ, ]๐๐ฅ [ โ ] 0 ๐ (0, ๐ง) = 0, ๐
๐โ ๐ (โ, ๐ง) = 0,
(3.1)
subject to the conditions (1.2)โ(1.4). De๏ฌne ๐(๐ง), ๐(๐ง), ๐(๐ง), ๐(๐ง) by (1.7) as before. We assume two conditions: (1โ ) ๐
and ๐ are ๐ ร ๐ matrices such that ๐
โ ๐ + ๐โ ๐
= 0; (2โ ) the values of ๐(๐ง)๐
+ ๐(๐ง)๐ are invertible except at isolated points. There are many choices of matrices meeting these conditions because ๐(0) = 0 and ๐(0) = ๐ผ๐ . The operator ๐
โ ๐
+ ๐โ ๐ is invertible, since otherwise ๐(๐ง)๐
+ ๐(๐ง)๐ has no invertible value, in violation of (2โ ).
Pseudospectral Functions
591
Notice that (2โ ) assures that the function ๐ฃ(๐ง) = ๐[๐(๐ง)๐
+ ๐(๐ง)๐][๐(๐ง)๐
+ ๐(๐ง)๐]โ1
(3.2)
is de๏ฌned except at isolated points. This function is meromorphic on โ, and it satis๏ฌes ๐ฃ(๐ง) = ๐ฃ(ยฏ ๐ง )โ by [3, Proposition 2.3.1]. The poles and principal parts of ๐ฃ(๐ง) contain important information for the spectral theory of the system (3.1). (0) For each ๐ โ โ, let ๐๐ be the set of all solutions ๐ = ๐ (๐ฅ) of (3.1) with (0) (๐) (๐+1) ๐ง = ๐. If ๐๐ , . . . , ๐๐ have been de๏ฌned, let ๐๐ be the set of all ๐ = ๐ (๐ฅ) such that ๐๐ = ๐๐๐ฝ๐ป(๐ฅ)๐ + ๐ฝ๐ป(๐ฅ)๐ (๐) , ๐๐ฅ (3.3) [ ] [ โ ] ๐ผ๐ 0 ๐ (0) = 0, ๐
๐โ ๐ (โ) = 0, (๐)
(0)
(1)
for some ๐ (๐) โ ๐๐ . We call ๐๐ , ๐๐ , . . . root spaces. Elements of these spaces are root functions. Root spaces are linear spaces which we view as subspaces of ๐ฟ2 (๐ป๐๐ฅ). By Proposition 3.2 below there is a largest root space ๐๐ =
โ โช ๐=0
(๐)
(๐)
๐๐ = ๐๐ .
(3.4)
We say that ๐ is an eigenvalue for (3.1) if ๐๐ โ= {0} as a subspace of ๐ฟ2 (๐ป๐๐ฅ). Remark. Following [2, 4], we work directly with canonical di๏ฌerential systems and make no use of underlying operators on ๐ฟ2 (๐ป๐๐ฅ). Nevertheless, it may be noted that our de๏ฌnitions of eigenvalue and root space are equivalent to standard operator de๏ฌnitions. The root subspaces ๐0 , ๐1 , . . . for a bounded linear operator ๐ and eigenvalue ๐ are de๏ฌned recursively by ๐0 = ker (๐ โ ๐๐ผ) and ๐๐+1 = {๐ : (๐ โ ๐๐ผ)๐ โ ๐๐ } for all ๐ = 0, 1, . . . . With due attention to domains, the same de๏ฌnition is used for an unbounded operator. If ๐ป(๐ฅ) has invertible values, we can take ๐ = โ๐๐ป(๐ฅ)โ1 ๐ฝ ๐/๐๐ฅ with domain speci๏ฌed by boundary values [4, p. 49]. The two notions of eigenvalue and root space then coincide. In principle, one can reduce to the case of invertible Hamiltonian with a transformation given in [4, ห (๐ฅ, ๐ง) = ๐โ๐๐ง๐พ๐ฅ ๐ (๐ฅ, ๐ง) for some ๐พ > 0. This p.143] that replaces ๐ (๐ฅ, ๐ง) with ๐ ห yields a new system with selfadjoint Hamiltonian ๐ป(๐ฅ) = ๐ป(๐ฅ) โ ๐พ๐ฝ. If ๐ป(๐ฅ) is ห bounded and ๐พ is su๏ฌciently large, ๐ป(๐ฅ) has invertible values. The transformation is well behaved with respect to eigenvalues and root spaces. We do not use these constructions and therefore omit details. Proposition 3.1. For any complex number ๐, the following are equivalent: (i) ๐ is an eigenvalue of (3.1); (ii) ๐(๐)๐
+ ๐(๐)๐ is not invertible; (iii) ๐ is a pole of ๐ฃ(๐ง). The eigenvalues of (3.1) are isolated points in the complex plane and occur in ยฏ conjugate pairs ๐, ๐.
592
J. Rovnyak and L.A. Sakhnovich
As a preliminary to the proof, consider an ๐ ร ๐ matrix-valued analytic function ๐น (๐ง) on a region ฮฉ which has invertible values except at isolated points. If ๐ โ ฮฉ and ๐น (๐) is not invertible, there is an ๐ โฅ 1 such that ๐น (๐ง) = ๐น๐ (๐ง)๐ (๐ง),
(3.5)
where ๐น๐ (๐ง) is analytic on ฮฉ, ๐น๐ (๐) is invertible, and ๐ (๐ง) is a polynomial of the form [ ] [ ][ ] ๐ (๐ง) = ๐ผ โ ๐๐ + ๐๐ (๐ง โ ๐) โ
โ
โ
๐ผ โ ๐2 + ๐2 (๐ง โ ๐) ๐ผ โ ๐1 + ๐1 (๐ง โ ๐) (3.6) for some rank-one projections ๐1 , ๐2 , . . . , ๐๐ . To see this, let Let ๐ be the order of ๐ as a zero of det ๐น (๐ง). Since ๐น (๐) is not invertible, there is a ๐1 โ= 0 in โ๐ such that ๐น (๐)๐1 = 0. Let ๐1 be the projection on the span of ๐1 , and set [ ] ๐1 ๐น1 (๐ง) = ๐น (๐ง) ๐ผ โ ๐1 + . ๐งโ๐ Since ๐น (๐)๐1 = 0, we can de๏ฌne ๐น1 (๐) so that ๐น1 (๐ง) is analytic on ฮฉ. We have [ ] ๐น (๐ง) = ๐น1 (๐ง) ๐ผ โ ๐1 + ๐1 (๐ง โ ๐) , det ๐น (๐ง) . ๐งโ๐ If ๐ = 1, then det ๐น1 (๐) โ= 0 because then ๐ is a zero of det ๐น (๐ง) of order 1. The assertion follows in the case ๐ = 1. In general, we proceed in the same way but repeat the procedure ๐ times. det ๐น1 (๐ง) =
Proof of Proposition 3.1. Everything here is in [3, Proposition 4.1.8] except for the equivalence of (ii) and (iii). Clearly (iii) implies (ii), so what remains is to show that (ii) implies (iii). We argue by contradiction, assuming that (ii) holds but (iii) fails. Write ๐ฃ(๐ง) = ๐๐ข1 (๐ง)๐ข2 (๐ง)โ1 , where ๐ข1 (๐ง) = ๐(๐)๐
+๐(๐)๐ and ๐ข2 (๐ง) = ๐(๐)๐
+๐(๐)๐. Here ๐ข2 (๐ง) has invertible values except at isolated points, ๐ข2 (๐) is not invertible, and ๐ is a removable singularity of ๐ฃ(๐ง). Applying (3.5) to ๐น (๐ง) = ๐ข2 (๐ง), we obtain ๐ข2 (๐ง) = ๐ขห2 (๐ง)๐ (๐ง), where ๐ข ห2 (๐ง) is entire, ๐ข ห2 (๐) is invertible, and ๐ (๐ง) has the form (3.6). Set ๐ข ห1 (๐ง) = ๐ข1 (๐ง)๐ (๐ง)โ1 ,
๐ง โ= ๐.
Then for all ๐ง โ= ๐, ๐ฃ(๐ง) = ๐๐ข1 (๐ง)๐ (๐ง)โ1 ๐ข ห2 (๐ง)โ1 = ๐ห ๐ข1 (๐ง)ห ๐ข2 (๐ง)โ1 . Since ๐ is a removable singularity of ๐ฃ(๐ง) and ๐ข ห2 (๐) is invertible, ๐ is a removable singularity of ๐ข ห1 (๐ง). Therefore we can de๏ฌne ๐ข ห1 (๐) so that ๐ข ห1 (๐ง) is entire. By (1.7), [ ] [ ] [ ] ๐ข ห1 (๐ง)๐ (๐ง) ๐ข (๐ง) ๐
= 1 = ๐ (โ, ๐งยฏ)โ . ๐ข ห2 (๐ง)๐ (๐ง) ๐ข2 (๐ง) ๐
Pseudospectral Functions
593
We can choose ๐ โ= 0 in โ๐ such that ๐ (๐)๐ = 0, and then we get [ ] [ ] 0 ยฏ โ ๐
๐. = ๐ (โ, ๐) 0 ๐ โ ยฏ Since ๐ (โ, ๐) is invertible, ๐
๐ = ๐๐ = 0. The desired contradiction follows because ๐
โ ๐
+ ๐โ ๐ is invertible under our assumptions. โก Proposition 3.2. The root spaces for (3.1) are ๏ฌnite dimensional. Moreover, for every eigenvalue ๐ of (3.1), there is a ๐ โฅ 0 such that (0)
(1)
(๐)
๐๐ โ ๐๐ โ โ
โ
โ
โ ๐๐
(๐+1)
= ๐๐
= โ
โ
โ
.
(3.7)
Proof. By Proposition 4.1.4(ii) of [3] the root spaces for (3.1) coincide with the root spaces for a nonzero eigenvalue of a compact operator. The assertions thus follow from well-known properties of compact operators (see, e.g., [1, Chapter I]). โก We call ๐ (0) (๐ฅ), ๐ (1) (๐ฅ), . . . , ๐ (๐) (๐ฅ) an eigenchain for the system (3.1) for an eigenvalue ๐ if [ ๐ผ๐
๐๐ (0) = ๐๐๐ฝ๐ป(๐ฅ)๐ (0) , ๐๐ฅ ] [ โ ] 0 ๐ (0) (0) = 0, ๐
๐โ ๐ (0) (โ) = 0,
and for each ๐ = 1, . . . , ๐, [ ๐ผ๐
๐๐ (๐) = ๐๐๐ฝ๐ป(๐ฅ)๐ (๐) + ๐ฝ๐ป(๐ฅ)๐ (๐โ1) , ๐๐ฅ ] [ โ ] 0 ๐ (๐) (0) = 0, ๐
๐โ ๐ (๐) (โ) = 0.
Every root function ๐ (๐ฅ) is the last member ๐ (๐ฅ) = ๐ (๐) (๐ฅ) of some eigenchain. We use this fact to prove the following orthogonality relation, which generalizes Proposition 4.1.1 of [3]. Proposition 3.3. For any complex ๐1 and ๐2 , if ๐ โ ๐๐1 and ๐ โ ๐๐2 , then โซ โ ๐(๐1 โ ๐ยฏ2 ) ๐(๐ก)โ ๐ป(๐ก)๐ (๐ก) ๐๐ก = 0. (3.8) 0
Hence if ๐ is a nonreal eigenvalue for (3.1), the root space ๐๐ is a neutral subspace of ๐ฟ2 (๐ป๐๐ฅ). A subspace of an inde๏ฌnite inner product space is called neutral if the inner product of any two of its elements is zero. [ ] ๐
Lemma 3.4. Let ๐ = ran . ๐ (1) If โ, ๐ โ โ๐ , the following are equivalent: [ ] [ ] ๐ โ (i) ๐
โ โ + ๐โ ๐ = 0; (ii) โ ๐; (iii) โ ๐ โฅ. โ ๐ (2) If ๐ด and ๐ต are ๐ร๐ matrices such that ๐ด๐
+๐ต๐ = 0, then [๐ด ๐ต]๐ = {0}.
594
J. Rovnyak and L.A. Sakhnovich
Proof of Lemma 3.4. Since ๐
โ ๐
+ ๐โ ๐ is invertible, ๐ is the range of a one-toone operator from โ๐ into โ2๐ and hence dim ๐ = ๐. Since ๐
โ ๐ + ๐โ ๐
= 0, ๐ฝ๐ โ ๐ โฅ . By a dimension argument ๐ฝ๐ = ๐ โฅ , and so ๐ = ๐ฝ๐ โฅ . The assertions in (1) follow. To prove (2), consider any ๐ โ ๐ and ๐ข โ โ๐ . Since ๐ด๐
+ ๐ต๐ = 0 by assumption, ๐
โ ๐ดโ ๐ข + ๐โ ๐ต โ ๐ข = 0. By part (1), [ โ ] ๐ด ๐ข โ ๐ โฅ. ๐ต โ๐ข [ โ ]โ [ ] ๐ด ๐ข Therefore ๐ = 0. By the arbitrariness of ๐ข, ๐ด ๐ต ๐ = 0. โก ๐ตโ๐ข Proof of Proposition 3.3. The assertion is trivial if ๐1 = ๐ยฏ2 , so assume that ๐1 = โ ๐ยฏ2 . We must show that in this case, โซ โ ๐(๐ก)โ ๐ป(๐ก)๐ (๐ก) ๐๐ก = 0. (3.9) 0
(0)
(1)
(๐1 )
Let ๐ (๐ฅ), ๐ (๐ฅ), . . . , ๐ (๐ฅ) and ๐ (0) (๐ฅ), ๐ (1) (๐ฅ), . . . , ๐ (๐2 ) (๐ฅ) be eigenchains (๐1 ) (๐2 ) with ๐ (๐ฅ) = ๐ (๐ฅ) and ๐ (๐ฅ) = ๐(๐ฅ). Set ๐ (โ1) (๐ฅ) = ๐ (โ1) (๐ฅ) = 0. Then ๐๐ (๐+1) = ๐๐1 ๐ฝ๐ป(๐ฅ)๐ (๐+1) + ๐ฝ๐ป(๐ฅ)๐ (๐) , ๐๐ฅ ๐๐ (๐+1) = ๐๐2 ๐ฝ๐ป(๐ฅ)๐ (๐+1) + ๐ฝ๐ป(๐ฅ)๐ (๐) , ๐๐ฅ and
[ [
๐ผ๐
๐ผ๐
] 0 ๐ (๐) (0) = 0, ] 0 ๐ (๐) (0) = 0,
[ โ ๐
[ โ ๐
] ๐โ ๐ (๐) (โ) = 0, ] ๐โ ๐ (๐) (โ) = 0,
for all ๐ = โ1, 0, . . . , ๐1 and ๐ = โ1, 0, . . . , ๐2 . By the boundary conditions and Lemma 3.4(1), ๐ (๐) (0)โ ๐ฝ๐ (๐) (0) = ๐ (๐) (โ)โ ๐ฝ๐ (๐) (โ) = 0 for all ๐ = โ1, 0, . . . , ๐1 and ๐ = โ1, 0, . . . , ๐2 . Hence for the same values of ๐, ๐, โซ โ[ $โ ] ๐๐ (๐) ๐๐ (๐) โ $ ๐ (๐) (๐ก)โ ๐ฝ + ๐ฝ๐ (๐) (๐ก) ๐๐ก = ๐ (๐) (๐ก)โ ๐ฝ๐ (๐ก)$ = 0. (3.10) ๐๐ก ๐๐ก 0 0 We show that
โฉ โช ๐(๐1 โ ๐ยฏ2 ) ๐ (๐) , ๐ (๐)
๐ป
โฉ โช โฉ โช = โ ๐ (๐โ1) , ๐ (๐) โ ๐ (๐) , ๐ (๐โ1) ,
๐ = 0, 1, . . . , ๐1 ,
๐ป
๐ = 0, 1, . . . , ๐2 ,
๐ป
(3.11)
Pseudospectral Functions
595
where โจโ
, โ
โฉ๐ป is the inner product of ๐ฟ2 (๐ป๐๐ฅ). In fact, โซ โ ๐(๐1 โ ๐ยฏ2 ) ๐ (๐) (๐ก)โ ๐ป(๐ก)๐ (๐) (๐ก) ๐๐ก 0
โซ =
โ 0
[ ] ๐ (๐) (๐ก)โ ๐ฝ ๐๐1 ๐ฝ๐ป(๐ก)๐ (๐) (๐ก) ๐๐ก โซ +
โซ =
By (3.10), ๐(๐1 โ ๐ยฏ2 )
โซ
โ 0
๐
(๐)
โ
โ 0
(๐ก) ๐ป(๐ก)๐
โ 0
[
]โ ๐๐2 ๐ฝ๐ป(๐ก)๐ (๐) (๐ก) ๐ฝ๐ (๐) (๐ก) ๐๐ก
] ๐๐ (๐) (๐โ1) ๐ (๐ก) ๐ฝ โ ๐ฝ๐ป(๐ก)๐ ๐๐ก ๐๐ก ]โ โซ โ [ (๐) ๐๐ + โ ๐ฝ๐ป(๐ก)๐ (๐โ1) ๐ฝ๐ (๐) (๐ก) ๐๐ก. ๐๐ก 0 (๐)
(๐)
โ
[
โซ (๐ก) ๐๐ก = โ
0
โซ โ
0
โ
โ
๐ (๐) (๐ก)โ ๐ป(๐ก)๐ (๐โ1) (๐ก) ๐๐ก ๐ (๐โ1) (๐ก)โ ๐ป(๐ก)๐ (๐) (๐ก) ๐๐ก,
which proves (3.11). The proof is completed by repeated application of (3.11). Start by choosing ๐ = ๐1 and ๐ = ๐2 in (3.11). For each term on the right, multiply by ๐1 โ ๐ยฏ2 , and repeat. Eventually we reach ๐ = 0 or ๐ = 0 for each term, and then ๐ (๐โ1) (๐ฅ) = ๐ (โ1) (๐ฅ) = 0 or ๐ (๐โ1) (๐ฅ) = ๐ (โ1) (๐ฅ) = 0 accordingly. In the end, we arrive at (3.9), as was to be shown. โก We shall need explicit formulas for eigenchains. Such formulas can be derived from the Taylor expansions (2.3) and (2.4). Set ๐พ(๐ง) = ๐(๐ง)๐
+ ๐(๐ง)๐
and ๐พ๐ (๐ง) = ๐๐ (๐ง)๐
+ ๐๐ (๐ง)๐,
๐ โฅ 0.
(3.12)
Proposition 3.5. The general form of an eigenchain ๐ (0) (๐ฅ), ๐ (1) (๐ฅ), . . . , ๐ (๐) (๐ฅ) for (3.1) for an eigenvalue ๐ is [ ] 0 ๐ (0) (๐ฅ) = ๐0 (๐ฅ, ๐) , ๐0 [ ] [ ] 0 0 (1) ๐ (๐ฅ) = (โ๐)๐1 (๐ฅ, ๐) + ๐0 (๐ฅ, ๐) , ๐0 ๐1 [ ] [ ] [ ] 0 0 0 (2) 2 ๐ (๐ฅ) = (โ๐) ๐2 (๐ฅ, ๐) + (โ๐)๐1 (๐ฅ, ๐) + ๐0 (๐ฅ, ๐) , (3.13) ๐0 ๐1 ๐2 โ
โ
โ
596
J. Rovnyak and L.A. Sakhnovich ๐
(๐)
[ ] [ ] 0 0 ๐โ1 (๐ฅ) = (โ๐) ๐๐ (๐ฅ, ๐) + (โ๐) ๐๐โ1 (๐ฅ, ๐) ๐0 ๐1 [ ] 0 + โ
โ
โ
+ ๐0 (๐ฅ, ๐) , ๐๐ ๐
where ๐0 , ๐1 , . . . , ๐๐ are vectors in โ๐ satisfying ยฏ โ ๐0 = 0 , ๐พ0 (๐) ยฏ โ ๐0 + ๐พ0 (๐) ยฏ โ ๐1 = 0 , (โ๐)๐พ1 (๐) ยฏ โ ๐0 + (โ๐)๐พ1 (๐) ยฏ โ ๐1 + ๐พ0 (๐) ยฏ โ ๐2 = 0 , (โ๐)2 ๐พ2 (๐)
(3.14)
โ
โ
โ
โ ๐โ1 โ โ ยฏ ยฏ ยฏ (โ๐) ๐พ๐ (๐) ๐0 + (โ๐) ๐พ๐โ1 (๐) ๐1 + โ
โ
โ
+ ๐พ0 (๐) ๐๐ = 0 . ๐
Proof. The case ๐ = 0 follows from [3, Proposition 3.1.2]. We proceed by induction for the general case. Assume that the assertion is known up to the ๐th stage for some ๐ โฅ 0. Consider an eigenchain ๐ (0) (๐ฅ), . . . , ๐ (๐) (๐ฅ), ๐ (๐+1) (๐ฅ). In particular,
[
๐ผ๐
๐๐ (๐+1) = ๐๐๐ฝ๐ป(๐ฅ)๐ (๐+1) + ๐ฝ๐ป(๐ฅ)๐ (๐) , ๐๐ฅ ] [ โ ] 0 ๐ (๐+1) (0) = 0, ๐
๐โ ๐ (๐+1) (โ) = 0.
(3.15)
By the inductive assumption, we can represent ๐ (0) (๐ฅ), . . . , ๐ (๐) (๐ฅ) in the form (3.13)โ(3.14) with ๐ = ๐. Set { [ ] [ ] 0 0 ๐ห (๐+1) (๐ฅ) = โ๐ (โ๐)๐ ๐๐+1 (๐ฅ, ๐) + (โ๐)๐โ1 ๐๐ (๐ฅ, ๐) ๐0 ๐1 [ ] [ ]} 0 0 + โ
โ
โ
+ (โ๐)๐2 (๐ฅ, ๐) + ๐1 (๐ฅ, ๐) . ๐๐โ1 ๐๐ By (2.6),
{ ( [ ] [ ]) ๐๐ห (๐+1) 0 0 ๐ = โ๐ (โ๐) ๐๐๐ฝ๐ป(๐ฅ)๐๐+1 (๐ฅ, ๐) + ๐๐ฝ๐ป(๐ฅ)๐๐ (๐ฅ, ๐) ๐ ๐ ๐๐ฅ 0 0 ( [ ] [ ]) 0 0 ๐โ1 + (โ๐) ๐๐๐ฝ๐ป(๐ฅ)๐๐ (๐ฅ, ๐) + ๐๐ฝ๐ป(๐ฅ)๐๐โ1 (๐ฅ, ๐) ๐1 ๐1 + โ
โ
โ
( [ ] [ ]) 0 0 + (โ๐) ๐๐๐ฝ๐ป(๐ฅ)๐2 (๐ฅ, ๐) + ๐๐ฝ๐ป(๐ฅ)๐1 (๐ฅ, ๐) ๐๐โ1 ๐๐โ1 } ( [ ] [ ]) 0 0 + ๐๐๐ฝ๐ป(๐ฅ)๐1 (๐ฅ, ๐) + ๐๐ฝ๐ป(๐ฅ)๐0 (๐ฅ, ๐) ๐๐ ๐๐
Pseudospectral Functions
597
{
[ ] [ ] 0 0 = ๐๐๐ฝ๐ป(๐ฅ)(โ๐) (โ๐)๐ ๐๐+1 (๐ฅ, ๐) + (โ๐)๐โ1 ๐๐ (๐ฅ, ๐) ๐0 ๐1 [ ] [ ]} 0 0 + โ
โ
โ
+ (โ๐)๐2 (๐ฅ, ๐) + ๐1 (๐ฅ, ๐) ๐๐โ1 ๐๐ { [ ] [ ] 0 0 + ๐ฝ๐ป(๐ฅ) (โ๐)๐ ๐๐ (๐ฅ, ๐) + (โ๐)๐โ1 ๐๐โ1 (๐ฅ, ๐) ๐0 ๐1 [ ] [ ]} 0 0 + โ
โ
โ
+ (โ๐)๐1 (๐ฅ, ๐) + ๐0 (๐ฅ, ๐) . ๐๐โ1 ๐๐ Thus
๐๐ห (๐+1) = ๐๐๐ฝ๐ป(๐ฅ)๐ห (๐+1) + ๐ฝ๐ป(๐ฅ)๐ (๐) . ๐๐ฅ In view of (3.15), it follows that
[ By (2.5), ๐ผ๐
๐ (๐ (๐+1) โ ๐ห (๐+1) ) = ๐๐๐ฝ๐ป(๐ฅ)(๐ (๐+1) โ ๐ห (๐+1) ). ๐๐ฅ ] 0 (๐ (๐+1) (0) โ ๐ห (๐+1) (0)) = 0 โ 0 = 0. Therefore [ ] 0 (๐+1) (๐+1) ห ๐ (๐ฅ) โ ๐ (๐ฅ) = ๐0 (๐ฅ, ๐) ๐๐+1
for some ๐๐+1 โ โ๐ . By the de๏ฌnition of ๐ห (๐+1) (๐ฅ), [ ] [ ] 0 0 ๐ (๐+1) (๐ฅ) = (โ๐)๐+1 ๐๐+1 (๐ฅ, ๐) + (โ๐)๐ ๐๐ (๐ฅ, ๐) ๐0 ๐1 [ ] [ ] 0 0 + โ
โ
โ
+ (โ๐)๐1 (๐ฅ, ๐) + ๐0 (๐ฅ, ๐) . ๐๐ ๐๐+1 [ ] The boundary condition ๐ผ๐ 0 ๐ (๐+1) (0) = 0 imposes no condition on ๐๐+1 . [ ] A restriction on ๐๐+1 is imposed by the condition ๐
โ ๐โ ๐ (๐+1) (โ) = 0. By ยฏ โ + ๐โ ๐๐ (๐) ยฏ โ = ๐พ๐ (๐) ยฏ โ , the the second equation in (2.5) and the identity ๐
โ ๐๐ (๐) restriction on ๐๐+1 is that ยฏ โ ๐0 + (โ๐)๐ ๐พ๐ (๐) ยฏ โ ๐1 + โ
โ
โ
+ (โ๐)๐พ1 (๐) ยฏ โ ๐๐ + ๐พ0 (๐) ยฏ โ ๐๐+1 = 0. (โ๐)๐+1 ๐พ๐+1 (๐) Thus the eigenchain ๐ (0) (๐ฅ), . . . , ๐ (๐) (๐ฅ), ๐ (๐+1) (๐ฅ) has the required form. The steps are reversible, and the inductive step follows. โก We also need formulas for the eigentransforms (1.5) of an eigenchain. These are given in the next result in both explicit and recursive forms. Proposition 3.6. Let ๐ (0) (๐ฅ), ๐ (1) (๐ฅ), . . . , ๐ (๐) (๐ฅ) be an eigenchain for (3.1) of the form (3.13), and let ๐น (0) (๐ง), ๐น (1) (๐ง), . . . , ๐น (๐) (๐ง) be the corresponding eigentransforms.
598
J. Rovnyak and L.A. Sakhnovich
(1) For each ๐ = 0, . . . , ๐, ๐ โ
๐น (๐) (๐ง) =
(โ๐)๐ ๐
๐=0
=
โ (โ ๐ โ
โ ๐(๐ง)๐๐ (๐) ยฏ โ + ๐(๐ง)๐๐ (๐) ยฏโ ๐๐โ๐ (๐ง โ ๐)๐+1 ๐+๐=๐
) (โ๐)๐ ฮ๐๐ (๐) ๐๐โ๐ (๐ง โ ๐)๐ ,
(3.16)
๐=0
๐=0
where the coe๏ฌcients in the last expression are as in Proposition 2.4. (2) The functions ๐น (0) (๐ง), ๐น (1) (๐ง), . . . , ๐น (๐) (๐ง) in (1) are given recursively ยฏ โ + ๐(๐ง)๐0 (๐) ยฏโ ๐(๐ง)๐0 (๐) ๐น (0) (๐ง) = ๐ ๐0 , ๐งโ๐ and [ ] ๐น (๐โ1) (๐ง) โ ๐(๐ง) ๐(๐ง) ๐ฝ๐ (๐) (โ) (๐) ๐น (๐ง) = , ๐ = 1, . . . , ๐. ๐(๐ง โ ๐) Moreover, for all ๐ = 0, . . . , ๐, { [ ] ๐ฝ๐ (๐) (โ) ๐ฝ๐ (๐โ1) (โ) ๐น (๐) (๐ง) = โ ๐(๐ง) ๐(๐ง) + 2 ๐(๐ง โ ๐) ๐ (๐ง โ ๐)2 } ๐ฝ๐ (0) (โ) + โ
โ
โ
+ ๐+1 . ๐ (๐ง โ ๐)๐+1
by (3.17)
(3.18)
(3.19)
Notice that if we set ๐น (โ1) (๐ง) โก 0, then (3.18) agrees with (3.17) when ๐ = 0. Proof. (1) By (3.13), ๐
(๐)
(๐ฅ) =
๐ โ
[
๐
(โ๐) ๐๐ (๐ฅ, ๐)
๐=0
Hence by (2.13), ๐น (๐) (๐ง) = = = = =
โซ
โ
0
๐ โ ๐=0 ๐ โ ๐=0 ๐ โ
[
0
๐๐โ๐
.
] 0 ๐ผ๐ ๐ (๐ก, ๐งยฏ)โ ๐ป(๐ก)๐ (๐) (๐ก) ๐๐ก
(โ๐)๐
โซ 0
(โ๐)๐ ๐ (โ๐)๐
โ
[
[ ] ] 0 0 ๐ผ๐ ๐ (๐ก, ๐งยฏ)โ ๐ป(๐ก)๐๐ (๐ก, ๐) ๐๐ก ๐๐โ๐ ๐ผ๐
โ ๐(๐ง)๐๐ (๐) ยฏ โ + ๐(๐ง)๐๐ (๐) ยฏโ ๐๐โ๐ (๐ง โ ๐)๐+1 ๐+๐=๐
โ โ
๐=0 ๐=0 โ (โ ๐ โ
ฮ๐๐ (๐)๐๐โ๐ (๐ง โ ๐)๐ ๐
)
(โ๐) ฮ๐๐ (๐) ๐๐โ๐ (๐ง โ ๐)๐ .
๐=0
]
๐=0
The two equalities in (3.16) follow.
Pseudospectral Functions
599
(2) Consider an eigenchain ๐ (0) (๐ฅ), ๐ (1) (๐ฅ), . . . , ๐ (๐) (๐ฅ) of the form (3.13), and let ๐น (0) (๐ง), ๐น (1) (๐ง), . . . , ๐น (๐) (๐ง) be the corresponding eigentransforms. The identity (3.17) is a special case of (3.16). Suppose ๐ = 1, . . . , ๐. Then ๐๐ (๐) = ๐๐๐ฝ๐ป(๐ฅ)๐ (๐) + ๐ฝ๐ป(๐ฅ)๐ (๐โ1) , ๐๐ฅ [ ] [ โ ] ๐ผ๐ 0 ๐ (๐) (0) = 0, ๐
๐โ ๐ (๐) (โ) = 0. Thus โซ โ 0
[ 0
๐ผ๐
]
๐๐ (๐) ๐ (๐ก, ๐งยฏ) ๐ฝ ๐๐ก = ๐๐ก โ
โซ
โ 0
[ ๐๐ 0 โซ
+ = ๐๐๐น Integration by parts yields [ ๐๐๐น (๐) (๐ง) + ๐น (๐โ1) (๐ง) = 0
โ = 0
[
0
] 0 ๐ผ๐ ๐ (๐ก, ๐งยฏ)โ ๐ป(๐ก)๐ (๐โ1) (๐ก) ๐๐ก
(๐ง) + ๐น (๐โ1) (๐ง).
$โ $ ] ๐ผ๐ ๐ (๐ก, ๐งยฏ)โ ๐ฝ๐ (๐) (๐ก)$$ โซ
[
(๐)
โ[
] ๐ผ๐ ๐ (๐ก, ๐งยฏ)โ ๐ป(๐ก)๐ (๐) (๐ก) ๐๐ก
โ[ 0
]
๐ก=0
) ]( ๐ 0 ๐ผ๐ ๐ (๐ก, ๐งยฏ)โ ๐ฝ๐ (๐) (๐ก) ๐๐ก ๐๐ก
๐ผ๐ ๐ (โ, ๐งยฏ)โ ๐ฝ๐ (๐) (โ) โซ โ ) [ ]( 0 ๐ผ๐ โ ๐๐ง๐ (๐ก, ๐งยฏ)โ ๐ป(๐ก)๐ฝ ๐ฝ๐ (๐) (๐ก) ๐๐ก โ 0
]
= 0 ๐ผ๐ ๐ (โ, ๐งยฏ)โ ๐ฝ๐ (๐) (โ) + ๐๐ง๐น (๐) (๐ง). [ ] [ ] Since 0 ๐ผ๐ ๐ (โ, ๐งยฏ)โ = ๐(๐ง) ๐(๐ง) by (1.7), we obtain (3.18). We prove (3.19) by iterating (3.17) and (3.18). By (3.17), [ ] ยฏโ [ ] ๐ (๐) [ ] ๐ฝ๐ (0) (โ) 1 ๐น (0) (๐ง) = โ ๐(๐ง) ๐(๐ง) ๐ฝ 0 ยฏ โ = โ ๐(๐ง) ๐(๐ง) , ๐0 (๐) ๐(๐ง โ ๐) ๐(๐ง โ ๐) which is the case ๐ = 0 of (3.19). By (3.18) with ๐ = 1, { } [ ] 1 ๐น (1) (๐ง) = ๐น (0) (๐ง) โ ๐(๐ง) ๐(๐ง) ๐ฝ๐ (1) (โ) ๐(๐ง โ ๐) { } [ ] ๐ฝ๐ (0) (โ) [ ] 1 = โ ๐(๐ง) ๐(๐ง) โ ๐(๐ง) ๐(๐ง) ๐ฝ๐ (1) (โ) ๐(๐ง โ ๐) ๐(๐ง โ ๐) { } [ ] ๐ฝ๐ (1) (โ) ๐ฝ๐ (0) (โ) = โ ๐(๐ง) ๐(๐ง) + 2 , ๐(๐ง โ ๐) ๐ (๐ง โ ๐)2 proving (3.19) for ๐ = 1. The general case follows by a straightforward induction. โก
600
J. Rovnyak and L.A. Sakhnovich
4. Main results We assume given a system (3.1) satisfying (1.2)โ(1.4), with operators ๐
and ๐ satisfying (1โ ) and (2โ ). Let ๐ (๐ฅ, ๐ง) be the unique solution of (1.6). As before, we set ๐ฃ(๐ง) = ๐[๐(๐ง)๐
+ ๐(๐ง)๐][๐(๐ง)๐
+ ๐(๐ง)๐]โ1 , (4.1) where ๐ (โ, ๐งยฏ)โ =
[ ] ๐(๐ง) ๐(๐ง) . ๐(๐ง) ๐(๐ง)
(4.2)
Recall that ๐ฃ(๐ง) = ๐ฃ(ยฏ ๐ง )โ , and the only singularities of ๐ฃ(๐ง) are poles, which occur at the points where ๐(๐ง)๐
+ ๐(๐ง)๐ is not invertible (see Proposition 3.1). By Proposition 3.1, the eigenvalues of (3.1) coincide with the poles of ๐ฃ(๐ง). In De๏ฌnition 4.2 we use the poles of ๐ฃ(๐ง) to introduce an inner product space โ0 (๐ฃ) whose elements are ๐-dimensional vector-valued entire functions. Our main result, Theorem 4.7, asserts that the eigentransform (1.5) acts an isometry on the span of root functions in ๐ฟ2 (๐ป๐๐ฅ) to โ0 (๐ฃ). For each ๐ค โ โ, write ๐ฃ(๐ง) = โ
๐พฯฐ (๐ค) ๐พ1 (๐ค) โ โ
โ
โ
โ + ๐ฃห(๐ง), ฯฐ (๐ง โ ๐ค) ๐งโ๐ค
(4.3)
where ๐ฃห(๐ง) is analytic at ๐ง = ๐ค. Here ฯฐ = ฯฐ๐ค โฅ 1 is chosen large enough that such a representation exists. The value of ฯฐ is not important, and zero coe๏ฌcients can be added at will. Such a representation is nontrivial only for poles, but it is notationally convenient to also allow ๐ค to be a point of analyticity for ๐ฃ(๐ง), in which case all coe๏ฌcients are zero. Since ๐ฃ(๐ง) = ๐ฃ(ยฏ ๐ง )โ , ๐ฃ(๐ง) = โ
๐พฯฐ (๐ค)โ ๐พ1 (๐ค)โ โ โ
โ
โ
โ + ๐ฃห(ยฏ ๐ง )โ , ฯฐ (๐ง โ ๐ค) ยฏ ๐งโ๐ค ยฏ
(4.4)
where ๐ฃห(ยฏ ๐ง )โ is analytic at ๐ง = ๐ค. ยฏ Hence ๐พ๐ (๐ค) ยฏ = ๐พ๐ (๐ค)โ , ๐ = 1, . . . , ฯฐ. Proposition 4.1. Let ๐ be an eigenvalue for (3.1), and write ๐ฃ(๐ง) as in (4.3) for ๐ค = ๐. Let ๐ข โ โ๐ be any vector, and de๏ฌne ๐ (0) (๐ฅ), ๐ (1) (๐ฅ), . . . , ๐ (ฯฐโ1) (๐ฅ) by (3.13) with ๐ = ฯฐ โ 1 and ๐๐ = (โ๐)๐ ๐พฯฐโ๐ (๐)๐ข,
๐ = 0, . . . , ฯฐ โ 1.
(4.5)
Then ๐ (0) (๐ฅ), ๐ (1) (๐ฅ), . . . , ๐ (ฯฐโ1) (๐ฅ) is an eigenchain for (3.1). Proof. We must show that ๐0 , . . . , ๐ฯฐโ1 satisfy (3.14). By (4.1) and (3.12), ๐[๐(๐ง)๐
+ ๐(๐ง)๐] = ๐ฃ(๐ง)๐พ(๐ง) = ๐ฃ(๐ง)
โ โ
๐พ๐ (๐ค)(๐ง โ ๐ค)๐ .
๐=0
Hence by (4.3),
[
] โ ๐พฯฐ (๐ค) ๐พ1 (๐ค) โ ๐[๐(๐ง)๐
+ ๐(๐ง)๐] = โ + โ
โ
โ
+ ๐พ๐ (๐ค)(๐ง โ ๐ค)๐ + ๐ฃห(๐ง)๐พ(๐ง). (๐ง โ ๐ค)ฯฐ ๐ง โ ๐ค ๐=0
Pseudospectral Functions
601
Since the left side is analytic at ๐ค, we deduce ฯฐ relations by expanding the ๏ฌrst term on the right side and equating coe๏ฌcients of negative powers of ๐ง โ ๐ค to zero: ๐พฯฐ (๐ค)๐พ0 (๐ค) = 0, ๐พฯฐ (๐ค)๐พ1 (๐ค) + ๐พฯฐโ1 (๐ค)๐พ0 (๐ค) = 0, ๐พฯฐ (๐ค)๐พ2 (๐ค) + ๐พฯฐโ1 (๐ค)๐พ1 (๐ค) + ๐พฯฐโ2 (๐ค)๐พ0 (๐ค) = 0, โ
โ
โ
(4.6)
๐พฯฐ (๐ค)๐พฯฐโ1 (๐ค) + ๐พฯฐโ1 (๐ค)๐พฯฐโ2 (๐ค) + โ
โ
โ
+ ๐พ1 (๐ค)๐พ0 (๐ค) = 0. On replacing ๐ค by ๐ค ยฏ and taking adjoints, we deduce (3.14).
โก
We introduce an inner product space that will be used in Theorem 4.7 to describe the action of the eigentransform (1.5) on root functions. De๏ฌnition 4.2. Let โ0 (๐ฃ) be the set of entire functions ๐น (๐ง) with values in โ๐ such that ๐ฃ(๐ง)๐น (๐ง) has ๏ฌnitely many poles. For ๐น, ๐บ โ โ0 (๐ฃ) and ๐ค โ โ, set โซ 1 โจ๐น, ๐บโฉ๐ค = โ ๐บ(ยฏ ๐ง )โ ๐ฃ(๐ง)๐น (๐ง) ๐๐ง, (4.7) 2๐๐ ฮ๐ค where ฮ๐ค is a counterclockwise circle about ๐ค, chosen small enough that ๐ฃ(๐ง) is analytic on ฮ๐ค and its interior except perhaps at ๐ง = ๐ค. Set โ โจ๐น, ๐บโฉ = โจ๐น, ๐บโฉ๐ค . (4.8) ๐คโโ
We identify entire functions ๐น and ๐บ in โ0 (๐ฃ) such that ๐ฃ(๐ง)[๐น (๐ง)โ๐บ(๐ง)] is entire (or, more precisely, has only removable singularities). The integral in (4.7) is independent of the choice of ฮ๐ค . All but ๏ฌnitely many terms of the sum in (4.8) are zero, and hence โจ๐น, ๐บโฉ is well de๏ฌned. Lemma 4.3. Let ๐น, ๐บ โ โ0 (๐ฃ), ๐ค โ โ, and let ๐น (๐ง) =
โ โ
๐น๐ (๐ค)(๐ง โ ๐ค)๐
and
๐บ(๐ง) =
๐=0
โ โ
๐บ๐ (๐ค)(๐ง ยฏ โ ๐ค) ยฏ ๐
(4.9)
๐=0
be Taylor expansions about ๐ค and ๐ค, ยฏ respectively. If ๐ฃ(๐ง) is given by (4.3), then โจ๐น, ๐บโฉ๐ค = or, equivalently, โก ๐บ0 (๐ค) ยฏ โข ๐บ1 (๐ค) ยฏ โข โจ๐น, ๐บโฉ๐ค = โข .. โฃ .
๐บฯฐโ1 (๐ค) ยฏ
โคโ โก โฅ โฅ โฅ โฆ
ฯฐ โ
โ
๐บ๐ (๐ค) ยฏ โ ๐พ๐ (๐ค)๐น๐ (๐ค),
(4.10)
๐=1 ๐+๐=๐โ1
๐พ1 (๐ค)
โข ๐พ (๐ค) โข 2 โข โฃ ๐พฯฐ (๐ค)
๐พ2 (๐ค)
โ
โ
โ
๐พฯฐโ1 (๐ค)
๐พฯฐ (๐ค)
๐พ3 (๐ค)
โ
โ
โ
โ
โ
โ
๐พฯฐ (๐ค)
0
0
โ
โ
โ
0
0
โคโก โฅโข โฅโข โฅโข โฆโฃ
๐น0 (๐ค) ๐น1 (๐ค) .. .
๐นฯฐโ1 (๐ค)
โค โฅ โฅ โฅ. โฆ
602
J. Rovnyak and L.A. Sakhnovich
Proof. By (4.3) and (4.9), โ๐บ(ยฏ ๐ง )โ ๐ฃ(๐ง)๐น (๐ง) =
โ โ
๐บ๐ (๐ค) ยฏ โ (๐ง โ ๐ค)๐
๐=0
=
ฯฐ โ โ โ โ โ
ฯฐ โ โ ๐พ๐ (๐ค) โ ๐น๐ (๐ค)(๐ง โ ๐ค)๐ + ๐(๐ง) ๐ (๐ง โ ๐ค) ๐=0 ๐=1
๐บ๐ (๐ค) ยฏ โ ๐พ๐ (๐ค)๐น๐ (๐ค)(๐ง โ ๐ค)๐+๐โ๐ + ๐(๐ง),
๐=1 ๐=0 ๐=0
where ๐(๐ง) is analytic at ๐ง = ๐ค. Only the terms with ๐ + ๐ โ ๐ = โ1 make a contribution to the integral (4.7). Therefore โซ ฯฐ โ โ 1 โ โจ๐น, ๐บโฉ๐ค = โ ๐บ(ยฏ ๐ง ) ๐ฃ(๐ง)๐น (๐ง) ๐๐ง = ๐บ๐ (๐ค) ยฏ โ ๐พ๐ (๐ค)๐น๐ (๐ค), 2๐๐ ฮ๐ค ๐=1 ๐+๐โ๐=โ1 which is equivalent to (4.10).
โก
Proposition 4.4. The inner product (4.8) is linear and symmetric. Proof. Linearity in the ๏ฌrst variable is clear from (4.10). Fix ๐น, ๐บ โ โ0 (๐ฃ). For any ๐ค โ โ, ๐พ๐ (๐ค)โ = ๐พ๐ (๐ค), ยฏ ๐ = 1, . . . , ฯฐ. Hence by (4.9) and (4.10), โจ๐บ, ๐น โฉ๐ค = Therefore โจ๐บ, ๐น โฉ =
โ ๐คโโ
ฯฐ โ
โ
๐=1 ๐+๐=๐โ1
โจ๐บ, ๐น โฉ๐ค =
๐น๐ (๐ค)โ ๐พ๐ (๐ค)๐บ ยฏ ๐ (๐ค) ยฏ = โจ๐น, ๐บโฉ๐ค .
โ ๐คโโ
This proves symmetry.
โจ๐บ, ๐น โฉ๐ค =
โ ๐คโโ
โจ๐น, ๐บโฉ๐ค = โจ๐น, ๐บโฉ. โก
We come now to a critical property of eigentransforms of root functions. In the case of simple poles, Lemma 4.1.10 in [3] provides what is needed. The next result is a generalization to arbitrary poles, which is stated in di๏ฌerent language but essentially accomplishes the same thing. (๐โ1)
Proposition 4.5. (1) If ๐ (๐ฅ) โ ๐๐ for some ๐ โฅ 1 and ๐น = ๐ ๐ , then ๐ฃ(๐ง)๐น (๐ง) is analytic in the complex plane except perhaps for a pole at ๐ of order at most ๐. (2) If ๐น = ๐ ๐ where ๐ is a ๏ฌnite linear combination of root functions of (3.1), then ๐น โ โ0 (๐ฃ). Proof. (1) By (3.19),
{ } [ ] ๐ฝ๐ (๐โ1) (โ) ๐ฝ๐ (๐โ2) (โ) ๐ฝ๐ (0) (โ) ๐น (๐ง) = โ ๐(๐ง) ๐(๐ง) + 2 +โ
โ
โ
+ ๐ . ๐(๐ง โ ๐) ๐ (๐ง โ ๐)2 ๐ (๐ง โ ๐)๐ [ โ ] Here the boundary conditions ๐
๐โ ๐ (๐) (โ) = 0 together with Lemma 3.4 imply that [ ] ๐
(๐) ๐ฝ๐ (โ) โ ๐ = ran , ๐ = 0, . . . , ๐ โ 1. ๐
Pseudospectral Functions It follows that
603
[ ] ] ๐
๐น (๐ง) = ๐(๐ง) ๐(๐ง) ๐(๐ง), ๐ [
(4.11)
where
๐1 ๐2 ๐๐ + +โ
โ
โ
+ ๐งโ๐ (๐ง โ ๐)2 (๐ง โ ๐)๐ for some vectors ๐1 , ๐2 , . . . , ๐๐ in โ๐ . By (4.1), ๐ฃ(๐ง)[๐(๐ง)๐
+ ๐(๐ง)๐] = ๐[๐(๐ง)๐
+ ๐(๐ง)๐]. Hence [ ] [ ] ๐
๐ฃ(๐ง) ๐(๐ง) ๐(๐ง) = ๐[๐(๐ง)๐
+ ๐(๐ง)๐]. ๐ ๐(๐ง) =
(4.12)
(4.13)
By (4.11) and (4.13),
[ ] [ ] ๐
๐ฃ(๐ง)๐น (๐ง) = ๐ฃ(๐ง) ๐(๐ง) ๐(๐ง) ๐(๐ง) = ๐[๐(๐ง)๐
+ ๐(๐ง)๐]๐(๐ง). ๐
(4.14)
Since ๐(๐ง) and ๐(๐ง) are entire functions, (4.14) and (4.12) show that ๐ฃ(๐ง)๐น (๐ง) is analytic in โ except perhaps for a pole at ๐ of order at most ๐. (2) This is immediate from (1). โก Corollary 4.6. Suppose ๐ (๐ฅ) โ ๐๐ and โซ โ [ ] 0 ๐ผ๐ ๐ (๐ก, ๐งยฏ)โ ๐ป(๐ก)๐ (๐ก) ๐๐ก. ๐น (๐ง) = 0
Then for any ๐บ(๐ง) in โ0 (๐ฃ), โจ๐น, ๐บโฉ = โจ๐น, ๐บโฉ๐ . Proof. By (4.8),โthe problem is to show that โจ๐น, ๐บโฉ๐ค = 0 for all ๐ค = โ ๐. Fix ๐ค โ= ๐. โ Write ๐น (๐ง) = ๐=0 ๐น๐ (๐ค)(๐ง โ ๐ค)๐ . Let ๐ฃ(๐ง) be given by (4.3). By Proposition 4.5(1), ๐ฃ(๐ง)๐น (๐ง) is analytic at ๐ง = ๐ค, and so ๐พฯฐ (๐ค)๐น0 (๐ค) = 0 ๐พฯฐโ1 (๐ค)๐น0 (๐ค) + ๐พฯฐ (๐ค)๐น1 (๐ค) = 0 โ
โ
โ
๐พ1 (๐ค)๐น0 (๐ค) + ๐พ2 (๐ค)๐น1 (๐ค) + โ
โ
โ
+ ๐พฯฐ (๐ค)๐นฯฐโ1 (๐ค) = 0. These relations say that โก ๐พ1 (๐ค) ๐พ2 (๐ค) โข ๐พ (๐ค) ๐พ (๐ค) 3 โข 2 โข โฃ ๐พฯฐ (๐ค)
0
โ
โ
โ
โ
โ
โ
๐พฯฐโ1 (๐ค) ๐พฯฐ (๐ค)
โ
โ
โ
โ
โ
โ
0
and hence โจ๐น, ๐บโฉ๐ค = 0 by Lemma 4.3.
โคโก โค โก โค ๐พฯฐ (๐ค) ๐น0 (๐ค) 0 โข ๐น1 (๐ค) โฅ โข0โฅ 0 โฅ โฅโข โฅ โข โฅ โฅโข โฅ = โข .. โฅ, .. โฆโฃ โฆ โฃ.โฆ . ๐นฯฐโ1 (๐ค) 0 0 โก
We are now ready to state and prove our main result, which generalizes Theorem 4.1.11 of [3].
604
J. Rovnyak and L.A. Sakhnovich
Theorem 4.7. (1) Let ๐ (๐ฅ) and ๐(๐ฅ) be ๏ฌnite linear combinations of root functions for the system (3.1), and let ๐น (๐ง), ๐บ(๐ง) be their eigentransforms. Then โซ โ ๐(๐ก)โ ๐ป(๐ก)๐ (๐ก) ๐๐ก = โจ๐น, ๐บโฉ . (4.15) 0
(2) Suppose ๐ โ ๐ฟ2 (๐ป๐๐ฅ), and let ๐น be its eigentransform. If ๐ is orthogonal to every root function of (3.1), then ๐น = 0 as an element of โ0 (๐ฃ). The de๏ฌnite case is treated in [3], and in this case more can be said. In the de๏ฌnite case, ๐ฃ(๐ง) is a Nevanlinna function which is meromorphic on the complex plane. Its poles are real and simple and coincide with the eigenvalues ๐1 , ๐2 , . . . of (3.1). In the Nevanlinna representation ] โซ โ[ 1 ๐ก ๐ฃ(๐ง) = ๐ผ + ๐ฝ๐ง + โ ๐๐ (๐ก), 1 + ๐ก2 โโ ๐ก โ ๐ง the nondecreasing function ๐ (๐ก) is constant on the intervals between poles, and the jump in ๐ (๐ก) at a pole ๐๐ is ๐๐ = โ Res ๐ฃ(๐ง) = ๐พ1 (๐๐ ). ๐ง=๐๐
In this case, the inner product (4.8) on โ0 (๐ฃ) is the inner product of ๐ฟ2 (๐๐ ), and Theorem 4.7 is subsumed in the more precise Theorem 4.2.2 of [3]. In the terminology of De๏ฌnition 4.2.3 of [3], ๐ (๐ก) is a pseudospectral function for (1.1). In general, the inner product โจ๐น, ๐บโฉ in (4.15) depends on the collection of poles ๐ค of ๐ฃ(๐ง) and coe๏ฌcients ๐พ1 (๐ค), ๐พ2 (๐ค), . . . in (4.3). Because of (4.15), we call this collection pseudospectral data for (1.1). Proof of Theorem 4.7, Part (1). By linearity and symmetry, it is su๏ฌcient to prove (4.15) when ๐ (๐ฅ) and ๐(๐ฅ) are root functions, say ๐ (๐ฅ) โ ๐๐1 and ๐(๐ฅ) โ ๐๐2 . Case 1: ๐1 โ= ๐ยฏ2 In this case, the left side of (4.15) is zero by Proposition 3.3. By Corollary 4.6, โจ๐น, ๐บโฉ = โจ๐น, ๐บโฉ๐1 . Since ๐ยฏ1 โ= ๐2 , ๐ฃ(๐ง)๐บ(๐ง) is analytic at ๐ง = ๐ยฏ1 by Proposition 4.5(1). Therefore ๐พฯฐ (๐ยฏ1 )๐บ0 (๐ยฏ1 ) = 0 ๐พฯฐโ1 (๐ยฏ1 )๐บ0 (๐ยฏ1 ) + ๐พฯฐ (๐ยฏ1 )๐บ1 (๐ยฏ1 ) = 0 โ
โ
โ
ยฏ ยฏ ยฏ ยฏ ยฏ ยฏ ๐พ1 (๐1 )๐บ0 (๐1 ) + ๐พ2 (๐1 )๐บ1 (๐1 ) + โ
โ
โ
+ ๐พฯฐ (๐1 )๐บฯฐโ1 (๐1 ) = 0.
Pseudospectral Functions Since ๐พ๐ (๐1 ) = ๐พ๐ (๐ยฏ1 )โ , ๐ = 1, . . . , ฯฐ, โก โคโ โก ๐พ1 (๐1 ) ๐พ2 (๐1 ) ๐บ0 (๐ยฏ1 ) โข ๐บ1 (๐ยฏ1 ) โฅ โข ๐พ (๐ ) ๐พ (๐ ) 3 1 โข โฅ โข 2 1 โข โฅ โข .. โฃ โฆ โฃ . ๐บฯฐโ1 (๐ยฏ1 ) ๐พฯฐ (๐1 ) 0
โ
โ
โ
605
๐พฯฐโ1 (๐1 ) ๐พฯฐ (๐1 )
โ
โ
โ
โ
โ
โ
๐พฯฐ (๐1 )
0
โ
โ
โ
0
0
โค
โก โคโ 0 โฅ โข0โฅ โฅ โข โฅ โฅ = โข .. โฅ . โฆ โฃ.โฆ 0
Hence โจ๐น, ๐บโฉ๐1 = 0 by Lemma 4.3. Case 2: ๐1 = ๐ยฏ2 ยฏ As a ๏ฌrst step we derive the formula (4.18) for the Put ๐1 = ๐ and ๐2 = ๐. left side of (4.15). Suppose ๐ฃ(๐ง) = โ
๐พฯฐ (๐ค) ๐พ1 (๐ค) โ โ
โ
โ
โ + ๐ฃห(๐ง), ฯฐ (๐ง โ ๐ค) ๐งโ๐ค
(4.16)
By adding zero terms in (4.16), we can choose ฯฐ as large as we wish. Hence in view of the inclusions (3.7), we can assume without loss of generality that ๐ (๐ฅ) and ๐(๐ฅ) are root functions of the same order ฯฐ, that is, they belong to eigenchains ๐ (0) (๐ฅ), ๐ (1) (๐ฅ), . . . , ๐ (ฯฐโ1) (๐ฅ) = ๐ (๐ฅ), ๐ (0) (๐ฅ), ๐ (1) (๐ฅ), . . . , ๐ (ฯฐโ1) (๐ฅ) = ๐(๐ฅ). Thus
[ ] [ ] 0 0 ๐ (๐ฅ) = (โ๐)ฯฐโ1 ๐ฯฐโ1 (๐ฅ, ๐) + (โ๐)ฯฐโ2 ๐ฯฐโ2 (๐ฅ, ๐) ๐0 ๐1 [ ] [ ] 0 0 + โ
โ
โ
+ (โ๐)๐1 (๐ฅ, ๐) + ๐0 (๐ฅ, ๐) ๐ฯฐโ2 ๐ฯฐโ1
and ๐(๐ฅ) = (โ๐)
ฯฐโ1
[ ] [ ] 0 0 ฯฐโ2 ยฏ ยฏ ๐ฯฐโ1 (๐ฅ, ๐) + (โ๐) ๐ฯฐโ2 (๐ฅ, ๐) โ0 โ1 [ ] [ ] 0 0 ยฏ ยฏ + โ
โ
โ
+ (โ๐)๐1 (๐ฅ, ๐) + ๐0 (๐ฅ, ๐) , โฯฐโ2 โฯฐโ1
where the conditions on ๐0 , . . . , ๐ฯฐโ1 and โ0 , . . . , โฯฐโ1 in Proposition 3.5 are met. In the former case, by (3.14) these conditions can be written: [ ] ยฏโ [ โ ] ๐ (๐) ๐
๐โ 0 ยฏ โ ๐0 = 0, ๐0 (๐) { [ ] [ ] } ยฏโ ยฏโ [ โ ] ๐ (๐) ๐ (๐) ๐
๐โ (โ๐) 1 ยฏ โ ๐0 + 0 ยฏ โ ๐1 = 0, ๐1 (๐) ๐0 (๐) { [ ] [ ] [ ] } ยฏโ ยฏโ ยฏโ [ โ ] ๐ (๐) ๐ (๐) ๐ (๐) ๐
๐โ (โ๐)2 2 ยฏ โ ๐0 + (โ๐) 1 ยฏ โ ๐1 + 0 ยฏ โ ๐2 = 0, ๐2 (๐) ๐1 (๐) ๐0 (๐) โ
โ
โ
606
J. Rovnyak and L.A. Sakhnovich [ โ ๐
[ ] [ ] { ยฏโ ยฏโ ฯฐโ1 ๐ฯฐโ1 (๐) ฯฐโ2 ๐ฯฐโ2 (๐) ๐ (โ๐) ยฏ โ ๐0 + (โ๐) ยฏ โ ๐1 + โ
โ
โ
๐ฯฐโ1 (๐) ๐ฯฐโ2 (๐) [ [ ] ] } ยฏโ ยฏโ ๐1 (๐) ๐0 (๐) +(โ๐) ยฏ โ ๐ฯฐโ2 + ๐0 (๐) ยฏ โ ๐ฯฐโ1 = 0. ๐1 (๐) (4.17) โ
By (2.15), โซ
โ
0
]
๐(๐ก)โ ๐ป(๐ก)๐ (๐ก) ๐๐ก =
โซ
โ ฯฐโ1 โ 0
[ ๐๐ โโฯฐโ1โ๐ 0
๐=0
โ
ฯฐโ1 โ
[ ] 0 (โ๐) ๐๐ (๐ก, ๐) ๐๐ก ๐ ๐ผ๐ ฯฐโ1โ๐
ฯฐโ1 โ ๐=0
=
] ยฏ โ ๐ป(๐ก)โ
๐ผ๐ ๐๐ (๐ก, ๐)
๐
๐๐ (โ๐)๐ โโฯฐโ1โ๐ ฮ๐๐ (๐)๐ฯฐโ1โ๐ .
(4.18)
๐,๐=0
We next derive the formula (4.26) for the right side of (4.15). By Corollary 4.6 and Lemma 4.3, โจ๐น, ๐บโฉ = โจ๐น, ๐บโฉ๐ =
ฯฐ โ
โ
ยฏ โ ๐พ๐ (๐)๐น๐ (๐) ๐บ๐ (๐)
๐=1 ๐+๐=๐โ1
ยฏ โ ๐พ1 (๐)๐น0 (๐) = ๐บ0 (๐) ยฏ โ ๐พ2 (๐)๐น0 (๐) + ๐บ0 (๐) ยฏ โ ๐พ2 (๐)๐น1 (๐) + ๐บ1 (๐) + โ
โ
โ
ยฏ โ ๐พฯฐ (๐)๐น0 (๐) + ๐บฯฐโ2 (๐) ยฏ โ ๐พฯฐ (๐)๐น1 (๐) + ๐บฯฐโ1 (๐) ยฏ โ ๐พฯฐ (๐)๐นฯฐโ1 (๐) + โ
โ
โ
+ ๐บ0 (๐) ] [ ยฏ โ ๐พ1 (๐)๐น0 (๐) + ๐พ2 (๐)๐น1 (๐) + โ
โ
โ
+ ๐พฯฐ (๐)๐นฯฐโ1 (๐) = ๐บ0 (๐) [ ] ยฏ โ ๐พ2 (๐)๐น0 (๐) + ๐พ3 (๐)๐น1 (๐) + โ
โ
โ
+ ๐พฯฐ (๐)๐นฯฐโ2 (๐) + ๐บ1 (๐) + โ
โ
โ
] [ ยฏ โ ๐พฯฐโ1 (๐)๐น0 (๐) + ๐พฯฐ (๐)๐น1 (๐) + ๐บฯฐโ2 (๐) ยฏ โ ๐พฯฐ (๐)๐น0 (๐) , + ๐บฯฐโ1 (๐)
(4.19)
where ๐น (๐ง) =
โ โ ๐=0
๐น๐ (๐ค)(๐ง โ ๐ค)๐
and ๐บ(๐ง) =
โ โ ๐=0
๐บ๐ (๐ค)(๐ง ยฏ โ ๐ค) ยฏ ๐.
Pseudospectral Functions
607
By (3.16), ๐น (๐ง) =
โ ( ฯฐโ1 โ โ ๐=0
๐บ(๐ง) =
๐=0
โ ( ฯฐโ1 โ โ ๐=0
๐
)
(โ๐) ฮ๐๐ (๐) ๐ฯฐโ1โ๐ (๐ง โ ๐)๐ , ) ยฏ ยฏ๐, (โ๐) ฮ๐๐ (๐) ๐ฯฐโ1โ๐ (๐ง โ ๐) ๐
๐=0
and so ๐น๐ (๐) =
ฯฐโ1 โ
(โ๐)๐ ฮ๐๐ (๐) ๐ฯฐโ1โ๐ ,
(4.20)
ยฏ โฯฐโ1โ๐ , (โ๐)๐ ฮ๐๐ (๐)
(4.21)
๐=0
ยฏ = ๐บ๐ (๐)
ฯฐโ1 โ ๐=0
for all ๐ = 0, 1, . . . , ฯฐ โ 1. [ ] ๐
Claim: For every ๐ โ ran , ๐ [ ] [ ] ๐ฃ(๐ง) ๐(๐ง) ๐(๐ง) ๐ = ๐ ๐(๐ง) ๐(๐ง) ๐.
(4.22)
The claim follows on writing (4.1) in the form [ ] [ ] [ ] ๐
[ ] ๐
๐ฃ(๐ง) ๐(๐ง) ๐(๐ง) = ๐ ๐(๐ง) ๐(๐ง) . ๐บ ๐บ Now by (3.19),
{ [ ] ๐ฝ๐ (ฯฐโ1) (โ) ๐ฝ๐ (ฯฐโ2) (โ) ๐น (๐ง) = โ ๐(๐ง) ๐(๐ง) (โ๐) + (โ๐)2 ๐งโ๐ (๐ง โ ๐)2 + โ
โ
โ
+ (โ๐)ฯฐ
๐ฝ๐ (0) (โ) (๐ง โ ๐)ฯฐ
} .
(4.23)
By (4.17) and Lemma 3.4(1), [ ] [ ] ยฏโ ๐ (๐) ๐
๐ฝ๐ (0) (โ) = 0 ยฏ โ ๐0 โ ran , ๐0 (๐) ๐ [ ] [ ] [ ] ยฏโ ยฏโ ๐1 (๐) ๐0 (๐) ๐
๐ฝ๐ (โ) = (โ๐) ยฏ โ ๐0 + ๐0 (๐) ยฏ โ ๐1 โ ran ๐ , ๐1 (๐) [ ] [ ] [ ] [ ] ยฏโ ยฏโ ยฏโ ๐1 (๐) ๐0 (๐) ๐
(2) 2 ๐2 (๐) ๐ฝ๐ (โ) = (โ๐) ยฏ โ ๐0 + (โ๐) ๐1 (๐) ยฏ โ ๐1 + ๐0 (๐) ยฏ โ ๐2 โ ran ๐ , ๐2 (๐) (1)
โ
โ
โ
608
J. Rovnyak and L.A. Sakhnovich ๐ฝ๐
(ฯฐโ1)
ฯฐโ1
(โ) = (โ๐)
[ ] [ ] ยฏโ ยฏโ ๐ฯฐโ1 (๐) ฯฐโ2 ๐ฯฐโ2 (๐) ยฏ โ ๐0 + (โ๐) ยฏ โ ๐1 + โ
โ
โ
๐ฯฐโ1 (๐) ๐ฯฐโ2 (๐) [ ] [ ] [ ] ยฏโ ยฏโ ๐1 (๐) ๐0 (๐) ๐
+ (โ๐) ยฏ โ ๐ฯฐโ2 + ๐0 (๐) ยฏ โ ๐ฯฐโ1 โ ran ๐ . ๐1 (๐)
Therefore by (4.23) and the claim,
{ [ ] ๐ฝ๐ (ฯฐโ1) (โ) ๐ฝ๐ (ฯฐโ2) (โ) ๐ฃ(๐ง)๐น (๐ง) = โ๐ฃ(๐ง) ๐(๐ง) ๐(๐ง) (โ๐) + (โ๐)2 ๐งโ๐ (๐ง โ ๐)2 } ๐ฝ๐ (0) (โ) + โ
โ
โ
+ (โ๐)ฯฐ (๐ง โ ๐)ฯฐ { (ฯฐโ1) [ ] ๐ฝ๐ (โ) ๐ฝ๐ (ฯฐโ2) (โ) = โ๐ ๐(๐ง) ๐(๐ง) (โ๐) + (โ๐)2 ๐งโ๐ (๐ง โ ๐)2 } ๐ฝ๐ (0) (โ) + โ
โ
โ
+ (โ๐)ฯฐ . (4.24) (๐ง โ ๐)ฯฐ
For the left side of (4.24), the series expansions of ๐ฃ(๐ง) and ๐น (๐ง) yield ๐พฯฐ (๐)๐น0 (๐) ๐พฯฐโ1 (๐)๐น0 (๐) + ๐พฯฐ (๐)๐น1 (๐) โ (๐ง โ ๐)ฯฐ (๐ง โ ๐)ฯฐโ1 ๐พ1 (๐)๐น0 (๐) + ๐พ2 (๐)๐น1 (๐) + โ
โ
โ
+ ๐พฯฐ (๐)๐นฯฐโ1 (๐) โ โ
โ
โ
โ ๐งโ๐ + holomorphic part . (4.25)
๐ฃ(๐ง)๐น (๐ง) = โ
The numerators here are key to calculating (4.19). We next show that these numerators are very simple expressions. In fact, by (4.24) and (4.25), โ
๐พฯฐ (๐)๐น0 (๐) ๐พฯฐโ1 (๐)๐น0 (๐) + ๐พฯฐ (๐)๐น1 (๐) โ (๐ง โ ๐)ฯฐ (๐ง โ ๐)ฯฐโ1 ๐พ1 (๐)๐น0 (๐) + ๐พ2 (๐)๐น1 (๐) + โ
โ
โ
+ ๐พฯฐ (๐)๐นฯฐโ1 (๐) โ โ
โ
โ
โ ๐งโ๐ + holomorphic part { [ ] ๐ฝ๐ (ฯฐโ1) (โ) ๐ฝ๐ (ฯฐโ2) (โ) = โ๐ ๐(๐ง) ๐(๐ง) (โ๐) + (โ๐)2 ๐งโ๐ (๐ง โ ๐)2 + โ
โ
โ
+ (โ๐)ฯฐ =โ
{ [
๐0 (๐)
] [ ๐0 (๐) + ๐1 (๐)
โ
}
] ๐1 (๐) (๐ง โ ๐) [ + ๐2 (๐)
{
๐ฝ๐ (0) (โ) (๐ง โ ๐)ฯฐ
} ] ๐2 (๐) (๐ง โ ๐)2 + โ
โ
โ
โ
(0) ๐ฝ๐ (ฯฐโ1) (โ) ๐ฝ๐ (ฯฐโ2) (โ) (โ) ฯฐโ1 ๐ฝ๐ + (โ๐) + โ
โ
โ
+ (โ๐) ๐งโ๐ (๐ง โ ๐)2 (๐ง โ ๐)ฯฐ
} .
Pseudospectral Functions Therefore
609
[ ๐พฯฐ (๐)๐น0 (๐) = ๐0 (๐)
] ๐0 (๐) (โ๐)ฯฐโ1 ๐ฝ๐ (0) (โ) [ ] ยฏโ ] ๐ (๐) [ = (โ๐)ฯฐโ1 ๐0 (๐) ๐0 (๐) 0 ยฏ โ ๐0 ๐0 (๐) = (โ๐)ฯฐโ1 ๐0 ,
the last equality holding by (2.10). By (2.10) and (2.11), ๐พฯฐโ1 (๐)๐น0 (๐) + ๐พฯฐ (๐)๐น1 (๐) [ [ ] ] = ๐1 (๐) ๐1 (๐) (โ๐)ฯฐโ1 ๐ฝ๐ (0) (โ) + ๐0 (๐) ๐0 (๐) (โ๐)ฯฐโ2 ๐ฝ๐ (1) (โ) [ ] ยฏโ ] ๐ (๐) [ = (โ๐)ฯฐโ1 ๐1 (๐) ๐1 (๐) 0 ยฏ โ ๐0 ๐0 (๐) ] ] } [ { [ ยฏโ ยฏโ [ ] ๐ (๐) ๐ (๐) + (โ๐)ฯฐโ2 ๐0 (๐) ๐0 (๐) (โ๐) 1 ยฏ โ ๐0 + 0 ยฏ โ ๐1 ๐1 (๐) ๐0 (๐) = (โ๐)ฯฐโ2 ๐1 . We continue in this way, obtaining at the last stage ๐พ1 (๐)๐น0 (๐) + ๐พ2 (๐)๐น1 (๐) + โ
โ
โ
+ ๐พฯฐ (๐)๐นฯฐโ1 (๐) [ ] = ๐ฯฐโ1 (๐) ๐ฯฐโ1 (๐) (โ๐)ฯฐโ1 ๐ฝ๐ (0) (โ) ] [ + ๐ฯฐโ2 (๐) ๐ฯฐโ2 (๐) (โ๐)ฯฐโ2 ๐ฝ๐ (1) (โ) + โ
โ
โ
[ ] + ๐0 (๐) ๐0 (๐) ๐ฝ๐ (ฯฐโ1) (โ) [ ] ยฏโ ] ๐ (๐) [ = (โ๐)ฯฐโ1 ๐ฯฐโ1 (๐) ๐ฯฐโ1 (๐) 0 ยฏ โ ๐0 ๐0 (๐) [ { [ ] ] } ยฏโ ยฏโ ] [ ๐ (๐) ๐ (๐) + (โ๐)ฯฐโ2 ๐ฯฐโ2 (๐) ๐ฯฐโ2 (๐) (โ๐) 1 ยฏ โ ๐0 + 0 ยฏ โ ๐1 ๐1 (๐) ๐0 (๐) + โ
โ
โ
{ [ ] [ ] ยฏโ ยฏโ ] [ ฯฐโ1 ๐ฯฐโ1 (๐) ฯฐโ2 ๐ฯฐโ2 (๐) + ๐0 (๐) ๐0 (๐) (โ๐) ยฏ โ ๐0 + (โ๐) ยฏ โ ๐1 ๐ฯฐโ1 (๐) ๐ฯฐโ2 (๐) [ ] } ยฏโ ๐0 (๐) + โ
โ
โ
+ ยฏ โ ๐ฯฐโ1 ๐0 (๐) = ๐ฯฐโ1 . Thus (4.19) yields ยฏ โ ๐ฯฐโ1 + ๐บ1 (๐) ยฏ โ (โ๐)๐ฯฐโ2 + โ
โ
โ
โจ๐น, ๐บโฉ = ๐บ0 (๐) ยฏ โ (โ๐)ฯฐโ2 ๐1 + ๐บฯฐโ1 (๐) ยฏ โ (โ๐)ฯฐโ1 ๐0 + ๐บฯฐโ2 (๐) =
ฯฐโ1 โ ๐=0
ยฏ โ (โ๐)๐ ๐ฯฐโ1โ๐ ๐บ๐ (๐)
(4.26)
610
J. Rovnyak and L.A. Sakhnovich The ๏ฌnal step is to compare (4.18) and (4.26). By (4.21), ยฏ = ๐บ๐ (๐)
ฯฐโ1 โ
ยฏ ฯฐโ1โ๐ (โ๐)๐ ฮ๐๐ (๐)โ
๐=0
and so by (2.16), ยฏโ= ๐บ๐ (๐)
ฯฐโ1 โ
ยฏโ= ๐๐ โโฯฐโ1โ๐ ฮ๐๐ (๐)
๐=0
ฯฐโ1 โ
ยฏ . ๐๐ โโฯฐโ1โ๐ ฮ๐๐ (๐)
๐=0
Therefore (4.26) yields โจ๐น, ๐บโฉ =
ฯฐโ1 โ ( ฯฐโ1 โ ๐=0
) (โ๐)๐ ๐ฯฐโ1โ๐
ยฏ ๐๐ โโฯฐโ1โ๐ ฮ๐๐ (๐)
๐=0
=
ฯฐโ1 โ
๐๐ (โ๐)๐ โโฯฐโ1โ๐ ฮ๐๐ (๐)๐ฯฐโ1โ๐ =
๐,๐=0
โซ
โ 0
๐(๐ก)โ ๐ป(๐ก)๐ (๐ก) ๐๐ก,
where the last equality is by (4.18). We have veri๏ฌed (4.15), and this completes the proof. โก Proof of Theorem 4.7, Part (2). According to De๏ฌnition 4.2, to show that ๐น = 0 as an element of โ0 (๐ฃ), we must show that ๐ฃ(๐ง)๐น (๐ง) is entire, that is, it is analytic at every pole of ๐ฃ(๐ง). ยฏ Let ๐ be a pole of ๐ฃ(๐ง), and represent ๐ฃ(๐ง) as in (4.3) for ๐ค = ๐ and ๐ค = ๐. The coe๏ฌcients in these representations satisfy ยฏ โ = ๐พ๐ (๐), ๐พ๐ (๐) Write ๐น (๐ง) =
โโ
๐=0
๐ = 1, . . . , ฯฐ.
(4.27)
๐น๐ (๐)(๐ง โ ๐)๐ . Since
{ [ ] } ๐พฯฐ (๐) ๐พฯฐโ1 (๐) ๐พ1 (๐) ๐ฃ(๐ง)๐น (๐ง) = โ + + โ
โ
โ
+ + ๐ช(1) โ
(๐ง โ ๐)ฯฐ (๐ง โ ๐)ฯฐโ1 ๐งโ๐ { } ฯฐโ1 โ
๐น0 (๐) + ๐น1 (๐)(๐ง โ ๐) + โ
โ
โ
+ ๐นฯฐโ1 (๐)(๐ง โ ๐) +โ
โ
โ
, the problem is to show that ๐พฯฐ (๐)๐น0 (๐) = 0, ๐พฯฐ (๐)๐น1 (๐) + ๐พฯฐโ1 (๐)๐น0 (๐) = 0, โ
โ
โ
๐พฯฐ (๐)๐นฯฐโ1 (๐) + ๐พฯฐโ1 (๐)๐นฯฐโ2 (๐) + โ
โ
โ
+ ๐พ1 (๐)๐น0 (๐) = 0.
(4.28)
Pseudospectral Functions By (1.5) and (2.3),
โซ
๐น (๐ง) = =
โ
0
โ โ
[ 0
611
] ๐ผ๐ ๐ (๐ฅ, ๐งยฏ)โ ๐ป(๐ฅ)๐ (๐ฅ) ๐๐ฅ
(๐ง โ ๐)๐
โซ
๐=0
and so for all ๐ = 0, 1, 2, . . . , โซ ๐น๐ (๐) =
โ 0
[ 0
โ 0
[
] ยฏ โ ๐ป(๐ฅ)๐ (๐ฅ) ๐๐ฅ, 0 ๐ผ๐ ๐๐ (๐ฅ, ๐)
] ยฏ โ ๐ป(๐ฅ)๐ (๐ฅ) ๐๐ฅ . ๐ผ๐ ๐๐ (๐ฅ, ๐)
(4.29)
By Proposition 3.1, ๐ยฏ is an eigenvalue of (3.1). Since ๐ is orthogonal to all root functions of (3.1), ๐ is orthogonal to the root functions for the eigenvalue ๐ยฏ provided by Proposition 4.1. Denote these functions ๐ (0) (๐ฅ), . . . , ๐ (ฯฐโ1) (๐ฅ).
(4.30)
Explicit formulas for the functions (4.30) are given by (3.13) and (4.5) with ๐ ยฏ Thus for each ๐ = 0, 1, . . . , ฯฐ โ 1, replaced by ๐. [ ] [ ] 0 0 (๐) ๐ ๐โ1 ยฏ ยฏ + (โ๐) ๐๐โ1 (๐ฅ, ๐) ๐ (๐ฅ) = (โ๐) ๐๐ (๐ฅ, ๐) ๐0 ๐1 ] [ [ ] 0 0 ยฏ ยฏ + ๐0 (๐ฅ, ๐) + โ
โ
โ
+ (โ๐)๐1 (๐ฅ, ๐) ๐๐โ1 ๐๐ [ [ ] ] 0 0 ๐ ๐โ1 ยฏ ยฏ = (โ๐) ๐๐ (๐ฅ, ๐) + (โ๐) ๐๐โ1 (๐ฅ, ๐) ยฏ ยฏ ๐พฯฐ (๐)๐ข (โ๐)๐พฯฐโ1 (๐)๐ข ] [ 0 ยฏ + โ
โ
โ
+ (โ๐)๐1 (๐ฅ, ๐) ยฏ (โ๐)๐โ1 ๐พฯฐโ๐+1 (๐)๐ข [ ] 0 ยฏ + ๐0 (๐ฅ, ๐) , ยฏ (โ๐)๐ ๐พฯฐโ๐ (๐)๐ข where ๐ข is an arbitrary vector in โ๐ . We obtain โซ โ ๐ 0 = (โ๐) ๐ (๐) (๐ก)โ ๐ป(๐ก)๐ (๐ก) ๐๐ก โซ =
0
โ
0
[
] ยฏ โ ๐๐ (๐ก, ๐)๐ป(๐ก)๐ ยฏ 0 ๐ขโ ๐พฯฐ (๐) (๐ก) ๐๐ก โซ
+
โ
[
0
] ยฏ โ ๐๐โ1 (๐ก, ๐)๐ป(๐ก)๐ ยฏ 0 ๐ขโ ๐พฯฐโ1 (๐) (๐ก) ๐๐ก โซ
+ โ
โ
โ
+ โซ +
0
โ
[
0
โ[
0
] ยฏ โ ๐1 (๐ก, ๐)๐ป(๐ก)๐ ยฏ ๐ขโ ๐พฯฐโ๐+1 (๐) (๐ก) ๐๐ก
] ยฏ โ ๐0 (๐ก, ๐)๐ป(๐ก)๐ ยฏ 0 ๐ขโ ๐พฯฐโ๐ (๐) (๐ก) ๐๐ก.
612
J. Rovnyak and L.A. Sakhnovich
In view of (4.27) and (4.29) and the arbitrariness of ๐ข, we conclude that ๐พฯฐ (๐)โ ๐น๐ (๐) + ๐พฯฐโ1 (๐)โ ๐น๐โ1 (๐) + โ
โ
โ
+ ๐พฯฐโ๐+1 (๐)โ ๐น1 (๐) + ๐พฯฐโ๐ (๐)โ ๐น0 (๐) = 0, which is equivalent to the system (4.28). We have shown that ๐ฃ(๐ง)๐น (๐ง) is analytic at every pole ๐ of ๐ฃ(๐ง). Therefore โก ๐น = 0 as an element of โ0 (๐ฃ), and the proof is complete.
References [1] I.C. Gohberg and M.G. Kreหฤฑn, Introduction to the theory of linear nonselfadjoint operators, American Mathematical Society, Providence, R.I., 1969. [2] I.C. Gohberg and M.G. Kreหฤฑn, Theory and applications of Volterra operators in Hilbert space, American Mathematical Society, Providence, R.I., 1970. [3] J. Rovnyak and L.A. Sakhnovich, Pseudospectral functions for canonical di๏ฌerential systems, Oper. Theory Adv. Appl., vol. 191, Birkhยจ auser, Basel, 2009, pp. 187โ219. [4] L.A. Sakhnovich, Spectral theory of canonical di๏ฌerential systems. Method of operator identities, Oper. Theory Adv. Appl., vol. 107, Birkhยจ auser Verlag, Basel, 1999. J. Rovnyak University of Virginia Department of Mathematics P. O. Box 400137 Charlottesville, VA 22904โ4137, USA e-mail:
[email protected] L.A. Sakhnovich 99 Cove Avenue Milford, CT 06461, USA e-mail:
[email protected]
Operator Theory: Advances and Applications, Vol. 218, 613โ638 c 2012 Springer Basel AG โ
Operator Identities for Subnormal Tuples of Operators Daoxing Xia For the memory of Professor I. Gohberg
Abstract. Some formulas for the products of resolvents of subnormal ๐-tuples of operators as well as ๐-tuples of commuting operators are established. Mathematics Subject Classi๏ฌcation (2000). Primary 47B20. Keywords. Subnormal ๐-tuple of operators, commuting ๐-tuple of operators, resolvent.
1. Introduction A ๐-tuple of operators ๐ = (๐1 , . . . , ๐๐ ) on a Hilbert space โ is said to be subnormal if there is a commuting ๐-tuple โ = (๐1 , . . . , ๐๐ ) of normal operators on a Hilbert space โ0 containing โ as a subspace, such that ๐๐ = ๐๐ โฃโ . In this case โ is said to be a normal extension of ๐. A normal extension is said to be minimal if there is no proper subspace of โ0 โ โ which reduces โ. The minimal normal extension (m.n.e.) of a subnormal tuple of operators exists and is essentially unique. There are several papers studying subnormal ๐-tuples of operators such as [1], [3], [4], [5], [6], [8], [11], [12], [15], [20]. โ Let ๐ be the closure of ๐,๐ [๐๐โ , ๐๐ ]โ. Then ๐ is said to be the defect space. In the ๏ฌrst part of this paper, the formulas for calculating the product of resolvents, a kind of Lifschitz-Brodski kernel ๐ ๐ โ โ (1) ๐๐ (๐๐โ๐ โ ๐คยฏ๐ )โ1 (๐๐๐ โ ๐ง๐ )โ1 โฃ๐ , ๐=1
๐=1
will be given, where โ is the m.n.e. of a subnormal ๐-tuple of ๐ = (๐1 , . . . , ๐๐ ), 1 โค ๐๐ , ๐๐ โค ๐, ๐ is the defect space, ๐๐ is the projection from โ0 to ๐ ,
614
D. Xia
๐ง๐ โ ๐(๐๐๐ ) and ๐ค๐ โ ๐(๐๐๐ ). If ๐ง๐ โ ๐(๐๐๐ ) and ๐ค๐ โ ๐(๐๐๐ ) then (1) is equal to ๐๐
๐ โ ๐=1
(๐๐โ๐ โ ๐คยฏ๐ )โ1
๐ โ
(๐๐๐ โ ๐ง๐ )โ1 โฃ๐ .
(2)
๐=1
Notice that if ๐ is pure, i.e., if there is no proper subspace ๐น โ โ reducing ๐ such that ๐โฃ๐น is normal, then โง โซ ๐ โฌ โ โจโ โ = closure of (๐๐ โ ๐ง๐ )โ1 ๐ผ : ๐ง๐ โ ๐(๐๐ ), ๐ผ โ ๐ . (3) โฉ โญ ๐=1
Thus the calculation of (2) provides a way to calculate the inner product of any two vectors in โ. Let us review some of the theory of single subnormal or hyponormal operators related to the subject in this paper. Let ๐ be a subnormal operator on a Hilbert space โ with m.n.e. ๐ on โ0 โ โ. We have [13] proved that the defect space def
๐ = closure of [๐ โ , ๐]โ is invariant with respect to ๐ โ . It is evident that [๐ โ , ๐]๐ โ ๐ . Then we de๏ฌned ๐ฟ(๐ ) operators def
๐ถ = [๐ โ , ๐]โฃ๐
def
and ฮ = (๐ โ โฃ๐ )โ
and proved that {๐ถ, ฮ} is a complete unitary invariant for pure subnormal operator ๐. We [13] also de๏ฌned an idempotent ๐ฟ(๐ )-valued analytic function, the mosaic for ๐, as follows: โซ ๐ขโฮ ๐(๐ง) = ๐(๐๐ข), ๐ง โ ๐(๐ ), ๐(๐) ๐ข โ ๐ง where ๐(โ
) = ๐๐ ๐ธ(โ
)โฃ๐ , ๐ธ(โ
) is the spectral measure of ๐ , and ๐๐ is the projection from โ0 to ๐ . Then ๐(๐ง) = 0 for ๐ง โ ๐(๐). We then de๏ฌned a rational function def
๐
(๐ง) = ๐ถ(๐ง โ ฮ)โ1 + ฮโ , ๐ง โ ๐(ฮ) from which we derived [๐
(๐ง), ๐(๐ง)] = 0,
for ๐ง โ ๐(ฮ)
โฉ
๐(๐ ).
Let def
๐(๐ง, ๐ค) = (๐ค ยฏ โ ฮโ)(๐ง โ ฮ) โ ๐ถ. The Lifschitz-Brodski kernel (1) in this case is def
๐(๐ง, ๐ค) = ๐๐ (๐ โ โ ๐ค) ยฏ โ1 (๐ โ ๐ง)โ1 โฃ๐ ,
๐ง, ๐ค โ ๐(๐ ).
Operator Identities for Subnormal Tuples of Operators
615
We then proved that [13] ๐(๐ง, ๐ค) = (๐ผ โ ๐(๐ค)โ )๐(๐ง, ๐ค)โ1 โ ๐(๐ง, ๐ค)โ1 ๐(๐ง), if ๐ง, ๐ค โ ๐(๐ ) and ๐(๐ง, ๐ค) is invertible. When dim ๐ < โ, {๐ถ, ฮ} is a pair of matrices and is a very useful tool for studying ๐. For example, in this case ๐(๐ ) โ {๐ง : det ๐(๐ง, ๐ง) = 0}, ๐(๐)โ๐(๐ ) is covered by a union of quadrature domains in Riemann surfaces, and there is a ๏ฌnite set of branched covers (๐
๐ , ๐๐ ) that are quadrature domains in Riemann surfaces (see [17]) such that ๐(๐) equals to the closure of the union of the images ๐๐ of Riemann surfaces ๐
๐ . In [21] and [22] Yakubovich proved that when dim ๐ < โ, the algebraic curve attached to a single subnormal operator ๐ should be divided naturally into two halves, an explicit formula for the mosaic ๐(๐ง) is given, that uses these halves, and the corresponding functional models of ๐ on Riemann surfaces are investigated. If ๐(๐)โ๐(๐ ) is a quadrature domain ๐ท โ โ, then ๐
(๐ง)๐(๐ง) = ๐(๐ง)๐
(๐ง) = ๐(๐ง)๐(๐ง), where ๐(โ
) is the Schwartz function of ๐ท. Besides, the mosaic ๐(๐ง) is the parallel projection to the eigenspace of the matrix ๐
(๐ง) corresponding to the eigenvalue ๐(๐ง). For a hyponormal operator ๐ป on a Hilbert space โ, let ๐ = closure of [๐ป โ , ๐ป]โ. Then ๐ป โ ๐ โโ ๐ for some hyponormal operator ๐ป. M. Putinar [9] introduced the subspace โ def ๐ฆ = closure of {๐ป โ๐ ๐ : ๐ = 0, 1, 2, . . .}. Then ๐ป โ ๐ฆ โ ๐ฆ and [๐ป โ , ๐ป]๐ฆ โ ๐ฆ. In that case he introduced def
๐ถ = [๐ป โ , ๐ป]โฃ๐ฆ
def
and ฮ = (๐ป โ โฃ๐ฆ )โ ,
which are in ๐ฟ(๐ฆ). This pair {๐ถ, ฮ} is also a complete unitary invariant for a pure hyponormal operator ๐ป. In the case of dim ๐ = 1, and dim ๐ฆ < โ, Gustafsson and Putinar ([7] and [9]) studied the unique pure hyponormal operator ๐ป satisfying the condition that the interior domain ๐ท of ๐(๐ป) is a quadrature domain. The author also proved that the Schwartz function ๐(๐ง) of ๐ท satis๏ฌes det ๐(๐ง, ๐(๐ง)) = 0, where ๐(๐ง, ๐ค) = (๐ค ยฏ โ ฮโ )(๐ง โ ฮ) โ ๐ถ. There are several very interesting results of their linear analysis of quadrature domains, some of which are related to ๐ถ and ฮ.
616
D. Xia
The author [17], [18] also introduced the mosaic ๐(โ
) related to the hyponormal operator associated with quadrature domain. This ๐(โ
) is also a meromorphic function on ๐ท, satisfying ๐(โ
)2 = ๐(โ
). Similar to (1), let ๐(๐ง, ๐ค) = ๐๐ฆ (๐ป โ โ ๐ค) ยฏ โ1 (๐ป โ ๐ง)โ1 โฃ๐ฆ ,
๐ง, ๐ค โ ๐(๐ป).
In the case dim ๐ = 1, (without the restriction dim ๐พ < โ), J. Pincus, D. Xia, and J. Xia [10] derived the formula { } โซ โซ 1 ๐(๐)๐๐ด(๐) (๐(๐ง, ๐ค)๐, ๐) = 1 โ exp โ , ๐ (๐ โ ๐ง)(๐ยฏ โ ๐ค) ยฏ where ๐ โ ๐ satis๏ฌes โฅ๐โฅ = โฅ[๐ป โ , ๐ป]โฅ, and ๐(โ
) is the Pincus principal function. If ๐ป is also associated with the quadrature domain, then [7], [9], [16] (๐(๐ง, ๐ค)๐, ๐) = 1 โ
det(๐(๐ง, ๐ค)) . det(๐ง โ ฮ) det(๐คยฏ โ ฮโ )
All of the above show that the objects ๐ถ, ฮ, ๐
(โ
), ๐(โ
), ๐(โ
, โ
) are useful tools in the theory of subnormal operators as well as the theory of hyponormal operators. In ยง2, we will introduce a generalization of ๐ถ, ฮ, ๐
(โ
), ๐(โ
), ๐(โ
, โ
) in the case of a subnormal ๐-tuple of operators. Most of them have been studied in [15]. In ยง3, we will give the formula for (1). In ยง4, we will generalize the formula for (2) to the case of a commuting ๐-tuple of operators. Some papers about the application of these formulas are being prepared.
2. Analytic model for a subnormal ๐-tuple of operators Let ๐ = (๐1 , . . . , ๐๐ ) be a pure subnormal ๐-tuple of operators on a Hilbert space โ with m.n.e. โ on โ0 โ โ. Let ๐ be the defect space: โ ๐ = closure of {[๐๐โ , ๐๐ ]โ : ๐, ๐ = 1, 2, . . . , ๐}. (4) Then as shown in [15], ๐ is invariant with respect to ๐๐โ , and [๐๐โ , ๐๐ ] for ๐, ๐ = 1, 2, . . . , ๐. Denote the operators on ๐ by def
def
๐ถ๐๐ = [๐๐โ , ๐๐ ]โฃ๐ and ฮ๐ = (๐๐โ โฃ๐ )โ
(5)
for ๐, ๐ = 1, 2, . . . , ๐. Let ๐ธ(โ
) be the spectral measure of โ on sp(โ). De๏ฌne an ๐ฟ(๐ )-valued positive measure def
๐(โ
) = ๐ โฃ๐ ๐ธ(โ
)โฃ๐
(6)
๐
ห = ๐
2 (๐พ(๐, ๐)) be the Hilbert space on sp(โ). Let ๐พ(๐) = ร ๐(๐๐ ) โ โ๐ and โ completion of
๐=1
โง ๐ โ โจโ โฉ
๐=1
(๐๐ โ ๐ข๐ )โ1 ๐ผ : ๐ผ โ ๐, ๐๐ โ ๐(๐๐ )
โซ โฌ โญ
Operator Identities for Subnormal Tuples of Operators with respect to the inner product def
(๐, ๐) =
617
โซ ๐ ๐(โ)
(๐(๐๐ข)๐ (๐ข), ๐(๐ข)).
(7)
Theorem[15]. Let ๐ be a pure subnormal ๐-tuple of operators on a separable Hilbert space โ with a m.n.e. โ on a Hilbert space โ0 โ โ. Let ๐ be the defect space of ๐. Then there is a unitary operator ๐ from โ0 onto the Hilbert space ๐ฟ2 (๐) of all measurable and square integrable functions on sp(โ) with respect to inner product (7) satisfying the following conditions: ห ๐ โ = โ, ๐ ๐ (๐ )๐ผ = ๐ (โ
)๐ผ,
๐ผ โ ๐,
for all M-valued bounded Borel functions ๐ on sp(โ), (๐ ๐๐ ๐ โ1 ๐ )(๐ข) = ๐ข๐ ๐ (๐ข), (๐ ๐๐โ ๐ โ1 ๐ )(๐ข) = ๐ข ยฏ๐ ๐ (๐ข) + (ฮโ๐ โ ๐ข ยฏ๐ )๐ (ฮ) ห where for ๐ = 1, 2, . . . , ๐ and ๐ โ โ,
โซ
๐ (ฮ) =
๐(๐๐ข)๐ (๐ข),
and ฮ๐ is de๏ฌned as in (5). ห def Let ๐ห๐ = ๐ ๐๐ ๐ โ1 . Then ๐ = (๐ห1 , . . . , ๐ห๐ ) is said to be the analytic model for ๐. From now on we only have to study the analytic model ๐, and simply identify ห ๐ with ๐ ห etc. In our calculation, we have to use several formulas in [15]. โ with โ, For any ๐1 , . . . , ๐๐ โ {1, 2, . . . , ๐}, de๏ฌne an operator ๐ โ def ๐๐๐ ,...,๐๐ (๐ง๐ , . . . , ๐ง๐ ) = ๐๐ (๐๐๐ โ ๐๐๐ ๐โ ) (๐๐๐ โ ๐ง๐ )โ1 โฃ๐ โซ =
๐ ๐(โ)
๐=1
(๐ข๐๐ โ ฮ๐๐ )๐(๐๐ข) ๐ โ (๐ข๐๐ โ ๐ง๐ )
(8)
๐=1
on ๐ , for ๐ง๐ โ ๐(๐๐๐ ), where ๐โ is the projection from ๐ฆ to โ. In [15], ๐๐๐ ,...,๐๐ is denoted by ๐
๐๐ ,...,๐๐ (๐ง๐ , . . . , ๐ง๐ ). Later in ยง3 we will sometimes denote ๐๐๐ ,...,๐๐ by ๐ ห{๐๐ ,...,๐๐ } . Let ๐{๐1 ,...,๐๐ } (๐ง1 , . . . , ๐ง๐ ) be the matrix (๐๐๐ )๐,๐=1,...,๐ , where { 0 if ๐ > ๐, ๐๐๐ = ๐๐๐ ,...,๐๐ if ๐ โค ๐. Thus ๐{๐1 ,...,๐๐ } is an ๐ฟ(๐ ๐ )-valued holomorphic function on def
๐(๐1 , . . . , ๐๐ ) = ๐(๐๐1 ) ร โ
โ
โ
ร ๐(๐๐๐ ). It is called a mosaic. In [15], it is proved that ๐{๐1 ,...,๐๐ } is idempotent, i.e., ๐2{๐1 ,...,๐๐ } = ๐{๐1 ,...,๐๐ } .
(9)
618
D. Xia
Let us de๏ฌne a kind of โconjugateโof ๐{๐1 ,...,๐๐ } as ๐โ {๐1 ,...,๐๐ } which is a matrix (๐๐๐ )๐,๐=1,2,...,๐ where โง if ๐ = ๐, โจ ๐ผ โ ๐๐โ๐+1 (๐ง๐โ๐+1 )โ 0 if ๐ > ๐, ๐๐๐ = โฉ โ๐๐๐โฒ ,๐๐โฒ +1 ,...,๐๐โฒ (๐ง๐ โฒ , ๐ง๐ โฒ +1 , . . . , ๐ง๐โฒ )โ if ๐ > ๐, where ๐โฒ = ๐ + 1 โ ๐ and ๐ โฒ = ๐ + 1 โ ๐. This ๐โ {๐1 ,...,๐๐ } is a little bit di๏ฌerent from the ๐โ {๐1 ,...,๐๐ } in [15], but it is only a kind of rearrangement of entries. The function
๐โ {๐1 ,...,๐๐ } (๐ง1 , . . . , ๐ง๐ )โ is also holomorphic on ๐(๐1 , . . . , ๐๐ ) and it is idempotent, 2
๐โ {๐1 ,...,๐๐ } = ๐โ {๐1 ,...,๐๐ } .
(10)
The author has written a monograph โThe Analytic Theory of Subnormal Operatorsโand submitted it for publication, in which contains all of the results in [15] with notations which coincide with those in this paper. Let us denote the operator ๐๐
๐ โ
(๐๐โ๐ โ ๐ค ยฏ๐ )โ1
๐=1
๐ โ
(๐๐๐ โ ๐ง๐ )โ1 โฃ๐ in ๐ฟ(๐ )
๐=1
by ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ (๐ง1 , . . . , ๐ง๐ ; ๐ค1 , . . . , ๐ค๐ ); then for ๏ฌxed ๐ค1 , . . . , ๐ค๐ , ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ is holomorphic for (๐ง1 , . . . , ๐ง๐ ) โ ๐(๐1 , . . . , ๐๐ ) and the kernel ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ is hermitian, i.e., ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ (๐ง1 , . . . , ๐ง๐; ๐ค1 , . . . , ๐ค๐ )โ = ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ (๐ค1 , . . . , ๐ค๐ ; ๐ง1 , . . . , ๐ง๐ ). It is easy to see that ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ =
โซ ๐ โ ๐=1
(ยฏ ๐ข๐๐
๐(๐๐ข) . ๐ โ โ๐ค ยฏ๐ ) (๐ข๐๐ โ ๐ง๐ )
(11)
๐=1
Let us denote an ordered integer set {๐1 , . . . , ๐๐ } by ๐๐ and {๐1 , . . . , ๐๐ } by ๐๐ . For example ๐๐๐ ,๐๐ means ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ . For any two ๏ฌnite tuples of integers, ๐๐ , ๐ = 1, 2, . . . , ๐ and ๐๐ , ๐ = 1, 2, . . . , ๐ which satisfy 1 โค ๐๐ , ๐๐ โค ๐, de๏ฌne the operator matrix ๐๐๐ ,๐๐ , which means ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ , as โ โ ๐๐๐ ,๐1 ๐๐๐ ,๐2 โ
โ
โ
๐๐๐ ,๐๐ โ๐๐๐โ1 ,๐1 ๐๐๐โ1 ,๐2 โ
โ
โ
๐๐๐โ1 ,๐๐ โ โ โ โ ๐๐๐ ,๐๐ = โ (12) โ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .โ . โ ๐๐2 ,๐1 ๐๐2 ,๐2 โ
โ
โ
๐๐2 ,๐๐ โ ๐๐1 ,๐1 ๐๐1 ,๐2 โ
โ
โ
๐๐1 ,๐๐
Operator Identities for Subnormal Tuples of Operators
619
This matrix in (12) is a little bit di๏ฌerent from the matrix ๐ต de๏ฌned on p. 630 of [15]. We form an (๐ + ๐) ร (๐ + ๐) matrix by block matrices as ( โ ) def ๐๐๐ ๐๐๐ ,๐๐ โ๐๐ ,๐๐ = (13) 0 ๐๐ ๐ where 0 in (13) is an ๐ ร ๐ matrix with all entries zero. Similar to the proof of Theorem 5 in [15], we may prove that โ๐๐ ,๐๐ is idempotent, i.e., โ2๐๐ ,๐๐ = โ๐๐ ,๐๐ .
(14)
Actually here โ๐๐ ,๐๐ is almost the matrix ๐๐๐ in (59) of [15]. From (14), we have ๐๐๐ ,๐๐ = ๐โ ๐๐ ๐๐๐ ,๐๐ + ๐๐๐ ,๐๐ ๐๐๐ .
(15)
๐๐๐ (๐ง, ๐ค) = (ฮโ๐ โ ๐ค)(ฮ ยฏ ๐ โ ๐ง) โ ๐ถ๐๐ .
(16)
De๏ฌne In the case of ๐1 = 1, ๐1 = 1, (15) becomes ๐1;1 (๐ง, ๐ค) = (๐ผ โ ๐1 (๐ค)โ )๐1;1 (๐ง, ๐ค) โ ๐1;1 (๐ง, ๐ค)๐(๐ง).
(17)
Let us review the single subnormal operator ๐ case: ๐1 = ๐ and โซ ๐(๐๐ข) ๐1;1 (๐ง, ๐ค) = (ยฏ ๐ข โ ๐ค)(๐ข ยฏ โ ๐ง) where ๐ข = ๐ข1 . Then from [13] as shown in ยง1, we have ๐1;1 (๐ง, ๐ค) = (๐ผ โ ๐1 (๐ค)โ )๐11 (๐ง, ๐ค)โ1 โ ๐11 (๐ง, ๐ค)โ1 ๐1 (๐ง). Comparing this with (17), it suggests that in the right-hand side of (15), the matrix ๐๐๐ ,๐๐ may be replaced by some rational functions of ฮ๐ , ฮโ๐ and ๐ถ๐๐ . That is the origin of this paper. In [15] and [19], we introduced the rational function def
๐
๐๐ (๐ง) = ๐ถ๐๐ (๐ง โ ฮ๐ )โ1 + ฮโ๐ ,
๐ง โ ๐(ฮ๐ ).
In [19], we have proved that [๐
๐1 ๐ (๐ง), ๐
๐2 ๐ (๐ง)] = 0
(18)
where [๐ด, ๐ต] = ๐ด๐ต โ ๐ต๐ด. In the case of ๐ง๐ โ ๐(๐๐ ), see Theorem 2 of this paper. Besides, in [15], we introduced some ๐ ร ๐ matrices ๐
๐,๐1 ,...,๐๐ (๐ง1 , . . . , ๐ง๐ ) = (๐๐๐ ), which also can be denoted by ๐
๐,๐ฟ๐ , when ๐ฟ๐ stands for the tuple of integers ๐1 , . . . , ๐๐ satisfying 1 โค ๐๐ โค ๐. The matrix ๐ถ๐ฟ๐ in [15] actually is โ๐
๐,๐ฟ here. In the matrix (๐๐๐ ), โง ๐ โ ๏ฃด ๏ฃด โจ โ๐ถ๐๐๐ (ฮ๐๐ โ ๐ง๐ )โ1 if ๐ < ๐, ๐=๐ ๐๐๐ = (19) ๏ฃด ๐
๐๐๐ (๐ง๐ ) if ๐ = ๐, ๏ฃด โฉ 0 if ๐ > ๐.
620
D. Xia
It is easy to see that ๐
๐,๐ฟ๐ โ ๐ค is invertible, i๏ฌ ๐
๐๐๐ โ ๐ค, ๐ = 1, 2, . . . , ๐ are invertible. We also proved in [15] that [๐๐ฟ๐ , ๐
๐,๐ฟ๐ ] = 0.
(20)
In ยง4, we will prove that for ๐ง = (๐ง1 , . . . , ๐ง๐ ), ๐ง๐ โ ๐(๐๐๐ ), [๐
๐,๐ฟ๐ (๐ง1 , . . . , ๐ง๐ ), ๐
๐โฒ ,๐ฟ๐ (๐ง1 , . . . , ๐ง๐ )] = 0
(21)
for any 1 โค ๐, ๐โฒ โค ๐. It is still open whether (21) is true if it is only assumed that ๐ง๐ โ ๐(ฮ๐๐ ) and ๐ > 1.
3. Calculation of ๐พ๐ท๐ ,๐ธ๐ For ๐ค โ ๐(ฮ๐ ), if (๐
๐๐ ๐ (๐ค)โ โ ๐ง๐ ) is invertible, ๐ = 1, 2, . . . , ๐. De๏ฌne an operator โ โ ๐ โ def def ยฏ โ1 ๐๐,๐๐ = ๐๐;๐1 ,...,๐๐ (๐ง1 , . . . , ๐ง๐ ; ๐ค) = โ (๐
๐๐ ๐ (๐ค)โ โ ๐ง๐ )โ1 โ (ฮโ๐ โ ๐ค) ๐=1
(22) on ๐ , where ๐๐ = {๐1 , . . . , ๐๐ }. By (18) the does not depend on the order of product.
๐ โ ๐=1
in the right-hand side of (22)
For ๐๐ = {๐1 , . . . , ๐๐ }, ๐๐ = {๐1 , . . . , ๐๐ }, if ๐ง๐ , ๐ค๐ โ ๐(ฮ๐ ) and ๐๐๐ (๐ง๐ , ๐ค๐ ) is invertible, de๏ฌne operators ๐๐๐ ,๐1 , ๐๐๐ ,๐2 , . . . , ๐๐๐ ,๐๐ in ๐ฟ(๐ ) by the formula ( ) ๐๐๐ ,๐1 ๐๐๐ ,๐2 โ
โ
โ
๐๐๐ ,๐๐ ( ) (23) ยฏ2 )โ1 โ
โ
โ
(๐
๐๐ ,๐๐ โ ๐ค ยฏ๐ )โ1 = ๐๐1 ,๐1 ๐๐1 ,๐2 โ
โ
โ
๐๐1 ,๐๐ (๐
๐2 ,๐๐ โ ๐ค where ๐๐1 ,๐๐ is ๐๐1 ;๐1 ,...,๐๐ de๏ฌned in (22), for ๐ โฅ 2. Let us comment on the product of (23). Suppose โ โ ๐ต11 ๐ต12 ๐ต13 โ
โ
โ
๐ต1๐ โ 0 ๐ต22 ๐ต23 โ
โ
โ
๐ต2๐ โ โ โ 0 0 ๐ต33 โ
โ
โ
๐ต3๐ โ ยฏ2 )โ1 โ
โ
โ
(๐
๐๐ ,๐๐ โ ๐ค ยฏ๐ )โ1 = โ (๐
๐2 ,๐๐ โ ๐ค โ โ. โ. . . . . . . . . . . . . . . . . . . . . . . . . . .โ 0 0 0 โ
โ
โ
๐ต๐๐ Then (23) means ๐๐๐ ,๐๐ =
๐ โ
๐๐1 ,๐๐ ๐ต๐๐ .
๐=1
The ๐๐๐ ,๐๐ stands for ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ (๐ง1 , . . . , ๐ง๐ ; ๐ค1 , . . . , ๐ค๐ ) etc.
Operator Identities for Subnormal Tuples of Operators
621
Let ๐๐๐ ,๐๐ = ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ (๐ง1 , . . . , ๐ง๐ ; ๐ค1 , . . . , ๐ค๐ ) be the matrix โ โ ๐๐๐ ,๐1 ๐๐๐ ,๐2 โ
โ
โ
๐๐๐ ,๐๐ โ๐๐๐โ1 ,๐1 ๐๐๐โ1 ,๐2 โ
โ
โ
๐๐๐โ1 ,๐๐ โ โ โ โ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .โ . โ โ โ ๐๐2 ,๐1 ๐๐2 ,๐2 โ
โ
โ
๐๐2 ,๐๐ โ ๐๐1 ,๐1 ๐๐1 ,๐2 โ
โ
โ
๐๐1 ,๐๐ Theorem 1. Let ๐ = (๐1 , . . . , ๐๐ ) be a pure subnormal ๐-tuple of operators on a separable Hilbert space โ with minimal normal extension โ = (๐1 , . . . , ๐๐ ) on ๐ฆ โ โ. For integers ๐๐ , ๐๐ , ๐ = 1, 2, . . . , ๐, ๐ = 1, 2, . . . , ๐ satisfying 1 โค ๐๐ , ๐๐ โค ๐, if ๐ง๐ โ ๐(ฮ๐๐ ) โฉ ๐(๐๐๐ ), ๐ = 1, 2, . . . , ๐, ๐ค๐ โ ๐(ฮ๐๐ ) โฉ ๐(๐๐๐ ), ๐ = 1, 2, . . . , ๐ satisfy the condition that ๐๐๐ ,๐๐ (๐ง๐ , ๐ค๐ ) are invertible, ๐ = 1, 2, . . . , ๐, ๐ = 1, 2, . . . , ๐, then ๐๐๐ ,๐๐ = ๐โ ๐๐ ๐๐๐ ,๐๐ โ ๐๐๐ ,๐๐ ๐๐๐ ,
(24)
where ๐๐๐ ,๐๐ stands for ๐๐1 ,...,๐๐ ;๐1 ,...,๐๐ (๐ง1 , . . . , ๐ง๐ ; ๐ค1 , . . . , ๐ค๐ ). Proof. We will prove (24) by mathematical induction with respect to ๐ and ๐. First consider the case ๐ = ๐ = 1. If ๐ = ๐ = 1, then (24) is equivalent to the following: Lemma 1. If ๐ง โ ๐(ฮ๐ ) โฉ ๐(๐๐ ), ๐ค โ ๐(ฮ๐ ) โฉ ๐(๐๐ ) and ๐๐๐ (๐ง, ๐ค) is invertible, then ๐๐;๐ (๐ง, ๐ค) = (๐ผ โ ๐๐ (๐ค)โ )๐๐๐ (๐ง, ๐ค)โ1 โ ๐๐๐ (๐ง, ๐ค)โ1 ๐๐ (๐ง). Proof. The proof is similar to that for Lemma 6 in [14]. But in order to make this paper readable, we give the details. By (13) in [15], for any ๐, ๐, 1 โค ๐, ๐ โค ๐, we have ๐๐๐ (๐ข๐ , ๐ข๐ )๐(๐๐ข) = ๐(๐๐ข)๐๐๐ (๐ข๐ , ๐ข๐ ) = 0,
(25)
where ๐ข = (๐ข1 , . . . , ๐ข๐ ). Therefore ๐๐๐ (๐ง, ๐ค)๐(๐๐ข) = ((๐ค ยฏโ๐ข ยฏ๐ )(๐ง โ ฮ๐ ) โ (ยฏ ๐ข๐ โ ฮโ๐ )(๐ข๐ โ ๐ง))๐(๐๐ข). Thus
โซ
๐๐๐ (๐ง, ๐ค)๐(๐๐ข) (ยฏ ๐ข๐ โ ๐ค)(๐ข ยฏ ๐ โ ๐ง) โซ โซ ๐ข ยฏ๐ โ ฮโ๐ (๐ง โ ๐ข๐ + ๐ข๐ โ ฮ๐ ) =โ ๐(๐๐ข) โ ๐(๐๐ข) ๐ข๐ โ ๐ง ๐ข ยฏ๐ โ ๐ค ยฏ โซ ๐ข ยฏ๐ โ ฮโ๐ = ๐ผ โ ๐๐ (๐ง) โ ๐(๐๐ข). ๐ข ยฏ๐ โ ๐ค ยฏ
(26)
๐ข ยฏ๐ โ ฮโ๐ ๐(๐๐ข)๐๐๐ (๐ง, ๐ค) = ๐๐๐ (๐ง, ๐ค)๐โ๐ . ๐ขยฏ๐ โ ๐ค ยฏ
(27)
๐๐๐ (๐ง, ๐ค)๐๐;๐ (๐ง, ๐ค) =
Let us prove that โซ
622
D. Xia
Firstly, we have โซ โซ ๐ข ยฏ๐ โ ฮโ๐ ๐(๐๐ข) โ โ โ ๐(๐๐ข)(๐ค ยฏ โ ฮ๐ ) = (๐คยฏ โ ฮ๐ ) + (๐ค ยฏ โ ฮ๐ ) (๐ค ยฏ โ ฮโ๐ ) ๐ข ยฏ๐ โ ๐ค ยฏ ๐ข ยฏ๐ โ ๐ค ยฏ โซ ๐(๐๐ข) โ โ = (๐ค ยฏ โ ฮ๐ ) + (๐ค ยฏ โ ฮ๐ ) (ยฏ ๐ข๐ โ ฮโ๐ โ (ยฏ ๐ข๐ โ ๐ค)) ยฏ ๐ข ยฏ๐ โ ๐ค ยฏ = (๐ค ยฏ โ ฮโ๐ )๐๐ (๐ค)โ . (28) Next, we have to prove that โซ ๐ข ยฏ๐ โ ฮโ๐ ๐(๐๐ข)((๐ค ยฏ โ ฮโ๐ )ฮ๐ + ๐ถ๐๐ ) = ((๐ค ยฏ โ ฮโ๐ )ฮ๐ + ๐ถ๐๐ )๐๐ (๐ค)โ . ๐ข ยฏ๐ โ ๐ค ยฏ By (25), the left-hand side of (29) is equal to โซ ๐ข ยฏ๐ โ ฮโ๐ ๐(๐๐ข)((๐ค ยฏ โ ฮโ๐ )ฮ๐ + (ยฏ ๐ข๐ โ ฮโ๐ )(๐ข๐ โ ฮ๐ )) ๐ข ยฏ๐ โ ๐ค ยฏ โซ ๐ข ยฏ๐ โ ฮโ๐ = ๐(๐๐ข)((๐ค ยฏโ๐ข ยฏ๐ )ฮ๐ + (ยฏ ๐ข๐ โ ฮโ๐ )๐ข๐ ) ๐ข ยฏ๐ โ ๐ค ยฏ โซ ๐ข ยฏ๐ โ ฮโ๐ = (ยฏ ๐ข๐ โ ฮโ๐ )๐ข๐ ๐(๐๐ข) , ๐ขยฏ๐ โ ๐ค ยฏ โซ since (ยฏ ๐ข๐ โ ฮโ๐ )๐(๐๐ข) = 0. By (25) again, we have
(29)
(30)
(ยฏ ๐ข๐ โ ฮโ๐ )๐ข๐ ๐(๐๐ข) = (๐ถ๐๐ + (ยฏ ๐ข๐ โ ฮโ๐ )ฮ๐ )๐(๐๐ข). Thus the right-hand side of (30) is equal to โซ ๐ถ๐๐ ๐๐ (๐ค)โ + (ยฏ ๐ข๐ โ ฮโ๐ )ฮ๐ ๐(๐๐ข)(ยฏ ๐ข๐ โ ฮโ๐ )(ยฏ ๐ข๐ โ ๐ค) ยฏ โ1 (31) โซ = ๐ถ๐๐ ๐๐ (๐ค)โ + (๐ค ยฏ โ ฮโ๐ )ฮ๐ ๐๐ (๐ค)โ + (ยฏ ๐ข๐ โ ๐ค)ฮ ยฏ ๐ ๐(๐๐ข)(ยฏ ๐ข๐ โ ฮโ๐ )(ยฏ ๐ข๐ โ ๐ค) ยฏ โ1 . However, the third term in the right-hand side of (31) is zero, which proves (29). From (28) and (29), we get (27). From (26) and (27), we get the lemma. โก In the case ๐ = 1, Theorem 1 is equivalent to the following: Lemma 2. If ๐ค โ ๐(ฮ๐ )โฉ๐(๐๐ ), ๐ง๐ โ ๐(ฮ๐๐ )โฉ๐(๐๐๐ ), 1 โค ๐, ๐๐ โค ๐ and ๐๐๐๐ (๐ง๐ , ๐ค), ๐ = 1, 2, . . . , ๐ are invertible, then ๐๐,๐๐ = (๐ผ โ ๐โ๐ )๐๐,๐๐ โ
๐ โ
๐๐,๐๐ ๐๐๐ ,...,๐๐ ,
(32)
๐=1
where ๐๐ = {๐1 , . . . , ๐๐ }. Proof. In (32), ๐๐,๐๐ means ๐{๐},๐๐ or ๐๐;๐1 ,...,๐๐ , and ๐๐,๐๐ means ๐{๐},๐๐ . Let us prove it by the mathematical induction with respect to the number of ๐โs. For
Operator Identities for Subnormal Tuples of Operators
623
the case that there is only one of ๐โs, says, ๐1 , (32) is equivalent to Lemma 1. Assume that (32) holds good for ๐2 , . . . , ๐๐ (there are ๐ โ 1 ๐โs), i.e., ๐๐,๐2 ,...,๐๐ = (๐ผ โ ๐โ๐ )๐๐;๐2 ,...,๐๐ โ
๐ โ
๐๐;๐2 ,...,๐๐ ๐๐๐ ,...,๐๐ .
(33)
๐=2
We have to prove that (32) holds good for ๐๐ = {๐1 , ๐2 , . . . , ๐๐ }. By (25) again, we have ๐๐๐1 ๐(๐๐ข) = ((ฮโ๐ โ ๐ค)(๐ข ยฏ ๐1 โ ๐ง1 ) โ (ยฏ ๐ข๐ โ ๐ค)(๐ข ยฏ ๐1 โ ฮ๐1 ))๐(๐๐ข). Therefore
โซ
๐๐๐1 ๐๐,๐๐ =
๐๐๐1 (๐ง, ๐ค)๐(๐๐ข) = (ฮโ๐ โ ๐ค)๐ ยฏ ๐;๐2 ,...,๐๐ โ ๐๐1 ,...,๐๐ . ๐ โ (ยฏ ๐ข๐ โ ๐ค) ยฏ (๐ข๐๐ โ ๐ง๐ )
(34)
๐=1
From ๐๐๐ (๐ง, ๐ค)โ1 (ฮโ๐ โ ๐ค) ยฏ = (๐
๐๐ (๐ค)โ โ ๐ง)โ1 , [(๐
๐๐ (๐ค)โ โ ๐ง)โ1 , ๐ผ โ ๐๐ (๐ค)โ ] = 0 (see (20)), (33) and (34), it follows that ๐๐,๐๐ = (๐
๐1 ๐ (๐ค)โ โ ๐ง1 )โ1 ๐๐;๐2 ,...,๐๐ โ ๐โ1 ๐๐1 ๐๐1 ,...,๐๐ = (๐ผ โ ๐๐ (๐ค)โ )(๐
๐1 ๐ (๐ค)โ โ ๐ง1 )โ1 ๐๐;๐2 ,...,๐๐ โ
(35)
๐ โ
(๐
๐1 ๐ (๐ค)โ โ ๐ง1 )โ1 ๐๐;๐2 ,...,๐๐ ๐๐๐ ,...,๐๐ โ ๐โ1 ๐๐1 ๐๐1 ,...,๐๐ .
๐=2
By (22), we have ๐๐;๐1 = ๐โ1 ๐๐1 and ๐๐;๐1 ,...,๐๐ = (๐
๐1 ๐ (๐ค)โ โ ๐ง1 )โ1 ๐๐;๐2 ,...,๐๐ . Thus (35) implies (32), which proves the lemma.
โก
In the case of ๐ = 1, Theorem 1 is equivalent to the following: Lemma 3. If ๐ค๐ โ ๐(ฮ๐๐ ) โฉ ๐(๐๐๐ ), ๐ง โ ๐(ฮ๐ ) โฉ ๐(๐๐ ), 1 โค ๐๐ , ๐ โค ๐ and ๐๐๐ ๐ (๐ง, ๐ค๐ ), ๐ = 1, 2, . . . , ๐ is invertible, then ๐๐๐ ;๐ (๐ง; ๐ค1 , . . . , ๐ค๐ ) = โ
๐ โ ๐โ๐๐ ,...,๐๐ ๐๐๐ ;๐ + (๐ผ โ ๐๐ )๐๐๐ ;๐ .
(36)
๐=1
where ๐๐ = {๐1 , . . . , ๐๐ } Proof. The proof of this lemma is similar to the proof of Lemma 2. For the case ๐ = 1, (36) is just Lemma 1. Assume that (36) holds for ๐2 , . . . , ๐๐ , i.e., ๐๐2 ,...,๐๐ ;๐ (๐ง; ๐ค2 , . . . , ๐ค๐ ) = โ
๐ โ ๐=2
๐โ๐๐ ,...,๐๐ ๐๐2 ,...,๐๐ ;๐ + (๐ผ โ ๐๐ )๐๐2 ,...,๐๐ ;๐ . (37)
624
D. Xia
From ๐(๐๐ข)๐๐1 ๐ (๐ข) = ๐(๐๐ข)((ยฏ ๐ข๐1 โ ๐ค ยฏ1 )(ฮ๐ โ ๐ง) โ (ยฏ ๐ข๐1 โ ฮโ๐1 )(๐ข๐ โ ๐ง)), it follows that โซ ๐(๐๐ข)((ยฏ ๐ข๐1 โ ๐ค ยฏ1 )(ฮ๐ โ ๐ง) โ (ยฏ ๐ข๐1 โ ฮโ๐1 )(๐ข๐ โ ๐ง)) ๐๐๐ ,๐ ๐๐1 ๐ = ๐ โ (๐ข๐ โ ๐ง) (ยฏ ๐ข๐๐ โ ๐ค ยฏ๐ ) = ๐๐2 ,...,๐๐ ;๐ (ฮ๐ โ ๐ง) โ
๐=1 โ ๐๐๐ .
From (ฮ๐ โ ๐ง)๐๐1 ๐ (๐ง, ๐ค1 )โ1 = (๐
๐1 ๐ (๐ง) โ ๐ค ยฏ1 )โ1 , (20) and (37), it follows that ๐๐๐ ,๐ = ๐๐2 ,...,๐๐ ;๐ (๐
๐1 ๐ (๐ง) โ ๐ค ยฏ1 )โ1 โ ๐โ๐๐ ๐๐1 ๐ (๐ง, ๐ค1 )โ1 = โ
๐ โ
๐โ๐๐ ,...,๐๐ ๐๐2 ,...,๐๐ ;๐ (๐
๐1 ๐ (๐ง) โ ๐ค ยฏ1 )โ1
(38)
๐=2
โ (๐ผ โ ๐๐ )๐๐2 ,...,๐๐ ;๐ (๐
๐1 ๐ (๐ง) โ ๐ค ยฏ1 )โ1 โ ๐โ๐๐ ๐๐1 ๐ (๐ง, ๐ค1 )โ1 . But from (23) it is easy to see that ๐๐2 ,...,๐๐ ;๐ (๐
๐1 ๐ (๐ง) โ ๐ค ยฏ1 )โ1 = (ฮ๐ โ ๐ง)โ1
๐ โ (๐
๐๐ ๐ (๐ง) โ ๐ค ยฏ๐ )โ1 = ๐๐๐ ,๐ . ๐=1
Therefore (38) is equivalent to (37), which proves the lemma.
โก
Now let us continue to prove Theorem 1. It is easy to see that (24) is equivalent to the following: ๐๐๐ ,๐๐ = (๐ผ โ ๐โ๐๐ )๐๐๐ ,๐๐ โ
๐โ1 โ
๐ โ
๐=1
๐=1
๐โ๐๐๐ ๐๐๐ ,๐๐ โ
๐๐๐ ,๐๐ ๐๐ห ๐๐ ,
(39)
ห ๐๐ means ๐๐ , . . . , ๐๐ , for any natural numbers where ๐๐๐ means ๐๐ , . . . , ๐๐ and ๐ ๐ and ๐. Lemma 2 shows that (39) holds for ๐ = 1, and any ๐. Assume that (39) holds for ๐ = ๐ฃ โ 1 โฅ 1. Let us prove that (39) holds for ๐ = ๐ฃ and any ๐, by mathematical induction with respect to ๐. Lemma 3 shows that (39) is true for ๐ = 1. We assume that (39) is true for ๐ = ๐ฃ โ 1 and that ๐ is replaced by ๐ โ 1. Notice that ๐(๐๐ข)๐๐1 ๐๐ (๐ง๐ , ๐ค1 ) ( ) = ๐(๐๐ข) (ยฏ ๐ข๐1 โ ๐ค ยฏ1 )(ฮ๐๐ โ ๐ง๐ ) โ (ยฏ ๐ข๐1 โ ๐ค ยฏ1 )(๐ข๐๐ โ ๐ง๐ ) + (ฮโ๐1 โ ๐ค ยฏ1 )(๐ข๐๐ โ ๐ง๐ ) . If ๐ > 1, then ๐๐๐ฃ ,๐๐ ๐๐1 ๐๐ ( ) โซ ๐(๐๐ข) (ยฏ ๐ข๐1 โ ๐ค ยฏ1 )(ฮ๐๐ โ ๐ง๐ ) โ (ยฏ ๐ข๐1 โ ๐ค ยฏ1 )(๐ข๐๐ โ ๐ง๐ ) + (ฮโ๐1 โ ๐ค ยฏ1 )(๐ข๐๐ โ ๐ง๐ ) = ๐ฃ ๐ โ โ (ยฏ ๐ข๐๐ โ ๐ค ยฏ๐ ) (๐ข๐๐ โ ๐ง๐ ) ๐=1
= ๐๐ห๐ฃ ,๐๐ (ฮ๐๐ โ ๐ง๐ ) โ ๐๐ห๐ฃ ,๐๐โ1
๐=1 + ๐๐๐ฃ ,๐๐โ1 (ฮโ๐1
โ๐ค ยฏ1 ),
Operator Identities for Subnormal Tuples of Operators
625
where ๐ห๐ฃ = {๐2 , . . . , ๐๐ฃ }. Since there are only ๐ฃ โ 1 natural numbers in ๐ห๐ฃ , we may apply (39) to ๐๐ห๐ฃ ,๐๐ and ๐๐ห๐ฃ ,๐๐โ1 . Besides, by the hypothesis of mathematical induction with respect to ๐, we may also use the formula (39) for ๐๐๐ฃ ,๐๐โ1 . Thus ๐๐๐ฃ ,๐๐ = (๐ผ1 + ๐ผ2 + ๐ผ3 )๐โ1 ๐1 ๐๐ , where โ ๐ผ1 = โ(๐ผ โ ๐โ๐๐ฃ )๐๐ห๐ฃ ,๐๐ โ โ
๐ฃโ1 โ
๐ โ
๐=2
๐=1
๐โ๐๐ ๐ฃ ๐๐ห๐ ,๐๐ โ
๐ผ2 = โ โ(๐ผ โ ๐โ๐๐ฃ )๐๐ห๐ฃ ,๐๐โ1 โ and
โ
๐ผ3 = โ(๐ผ โ ๐โ๐๐ฃ )๐๐๐ฃ ,๐๐โ1 โ
๐ฃโ1 โ ๐=2
(40) โ
๐๐ห๐ฃ ,๐๐ ๐๐ห ๐๐ โ (ฮ๐๐ โ ๐ง๐ ),
๐โ๐๐ ๐ฃ ๐๐ห๐ ,๐๐โ1 โ
๐=1
๐ฃโ1 โ
๐โ1 โ
๐=1
๐=1
๐โ๐๐ ๐ฃ ๐๐๐ ,๐๐โ1 โ
โ
๐โ1 โ
๐๐ห๐ฃ ,๐๐ ๐๐ห ๐(๐โ1) โ ,
โ
๐๐ห๐ฃ ,๐๐ ๐๐ห ๐(๐โ1) โ (ฮโ๐1 โ ๐ค ยฏ1 ).
Let us rearrange the terms in the summation of (40). Then ๐๐๐ฃ ,๐๐ = ๐ฝ1 + ๐ฝ2 ,
(41)
where ห ๐๐ฃ ,๐๐ โ ๐ฝ1 = (๐ผ โ ๐โ๐๐ฃ )๐
๐ฃโ1 โ ๐=2
ห ๐๐ ,๐๐ โ ๐โ ๐ ห ๐โ๐๐ ๐ฃ ๐ ๐1 ๐ฃ ๐1 ,๐๐ ,
(42)
where
( ) โ ห ๐๐ ,๐๐ = ๐ ห ๐ ยฏ1 ) ๐โ1 ๐1 ๐๐ , ๐๐ ,๐๐ (ฮ๐๐ โ ๐ง๐ ) โ ๐๐ห๐ ,๐๐โ1 + ๐๐๐ ,๐๐โ1 (ฮ๐1 โ ๐ค
(43)
ห ๐๐ ,๐1 = ๐ ห ยฏ1 )โ1 , besides, for ๐ > 1 and ๐ ๐๐ ,๐1 (๐
๐1 ๐1 (๐ง๐ ) โ ๐ค ๐ ๐โ1 โ โ โ1 ๐๐ห๐ฃ ,๐๐ ๐๐ห ๐๐ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ยฏ1 ) + ๐๐ห๐ฃ ,๐ห ๐ ๐๐ห ๐(๐โ1) ๐โ1 ๐ฝ2 = โ ๐1 ๐๐ ๐=1
โ
๐โ1 โ ๐=1
๐=1
(44)
๐๐ห๐ฃ ,๐๐ ๐๐ห ๐(๐โ1) (ฮโ๐1 โ ๐ค ยฏ1 )๐โ1 ๐1 ๐๐ ,
since (ฮ๐๐ โ ๐ง๐ )๐โ1 ยฏ1 )โ1 , and ๐๐1 ,๐๐โ1 (ฮโ๐1 โ ๐ค ยฏ1 )๐โ1 ๐1 ๐๐ = (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ๐1 ๐๐ = ๐๐1 ,๐๐ by (22). Now, let us prove that ห ๐๐ ,๐๐ = ๐๐๐ ,๐๐ . ๐
(45)
626
D. Xia
According to (23), we only have to prove that ) ( ห ๐๐ ,๐1 ๐ ห ๐๐ ,๐2 โ
โ
โ
๐ ห ๐๐ ,๐๐ (๐
๐1 ,๐๐ โ ๐ค ยฏ1 ) ๐ ( ) = ๐๐ห๐ ,๐1 ๐๐ห๐ ,๐2 โ
โ
โ
๐๐ห๐ ,๐๐ ,
(46)
by mathematical induction with respect to ๐, where ๐ โฅ 2. From (23), (46) holds ห ๐๐ ;๐1 can be replaced by ๐๐๐ ;๐1 . Assume that (46) holds while ๐ is for ๐ = 1 and ๐ ห ๐๐ ,๐๐ = ๐๐๐ ,๐๐ replaced by ๐ โ 1 โฅ 1. According to the de๏ฌnition of ๐โs of (23), ๐ for ๐ = 1, 2, . . . , ๐ โ 1. In order to prove that (46) holds good for ๐, we only have to prove that ๐ฟ = 0, where def
๐ฟ = โ
๐โ1 โ
๐๐๐ ,๐๐ ๐ถ๐1 ๐๐
๐=1
๐ โ ห ๐๐ ,๐๐ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค (ฮ๐๐ โ ๐ง๐ )โ1 + ๐ ยฏ1 ) โ ๐๐ห๐ ,๐๐ . ๐=๐
By the hypothesis of mathematical induction, โ
๐โ1 โ
๐๐๐ ,๐๐ ๐ถ๐1 ๐๐
๐=1
๐ โ (ฮ๐๐ โ ๐ง๐ )โ1 ๐=๐
=โ
๐โ1 โ
๐โ1 โ
๐=1
๐=๐
(
๐๐๐ ,๐๐ ๐ถ๐1 ๐๐
) (ฮ๐๐ โ ๐ง๐ )โ1 (ฮ๐๐ โ ๐ง๐ )โ1
[ = ๐๐ห๐ ,๐๐โ1 โ ๐๐๐ ,๐๐โ1 (๐
๐1 ๐๐โ1 (๐ง๐โ1 ) โ ๐ค ยฏ1 ) ] โ1 โ ๐๐๐ ,๐๐โ1 ๐ถ๐1 ๐๐โ1 (ฮ๐๐โ1 โ ๐ง๐โ1) โ
(ฮ๐๐ โ ๐ง๐ )โ1 ( ) = ๐๐ห๐ ,๐๐โ1 โ ๐๐๐ ,๐๐โ1 (ฮโ๐1 โ ๐ค ยฏ1 ) (ฮ๐๐ โ ๐ง๐ )โ1 , since ยฏ1 = โ๐ถ๐1 ๐๐โ1 (ฮ๐๐โ1 โ ๐ง๐โ1)โ1 + ฮโ1 โ ๐ค ยฏ1 . ๐
๐1 ๐๐โ1 (๐ง๐โ1 ) โ ๐ค Thus ห ๐๐ ,๐๐ โ ๐ ห ๐ฟ = {๐ ยฏ๐ )โ1 + ๐๐ห๐ ,๐๐โ1 ๐โ1 ๐1 ๐๐ ๐๐ ,๐๐ (๐
๐1 ๐๐ (๐ง) โ ๐ค โ ๐๐๐ ,๐๐โ1 (ฮโ๐1 โ ๐ค ยฏ1 )๐โ1 ยฏ๐ ) ๐1 ๐๐ }(๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค which equals zero by (43). Therefore (46) is proved and so does (45). From (42) and (45), we have ๐ฝ1 = (๐ผ โ ๐โ๐๐ฃ )๐๐๐ฃ ,๐๐ โ
๐ฃโ1 โ ๐=2
๐โ๐๐ ๐ฃ ๐๐๐ ,๐๐ โ ๐โ๐1 ๐ฃ ๐๐1 ,๐๐ .
Next, let us study ๐ฝ2 . From ๐ฟ = 0 and (45), we have ๐๐ห๐ฃ ,๐๐ = โ
๐โ1 โ ๐ =1
๐๐๐ฃ ,๐๐ ๐ถ๐1 ๐๐
๐ โ (ฮ๐๐ โ ๐ง๐ )โ1 + ๐๐๐ฃ ,๐๐ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ยฏ1 ). ๐=๐
(47)
Operator Identities for Subnormal Tuples of Operators
627
Thus ๐ฝ2 = โ
๐โ1 ๐ ๐ โ โ โ [โ ๐๐๐ฃ ,๐๐ ๐ถ๐1 ๐๐ (ฮ๐๐ โ ๐ง๐ )โ1 ๐ =1
๐=1
๐=๐
+ ๐๐๐ฃ ,๐๐ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ยฏ1 )]๐๐ห ๐๐ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐คยฏ1 )โ1 +
๐โ1 โ
(
โ
๐ =1
๐=1
โ
๐โ1 ๐ โ โ ) ๐๐๐ฃ ,๐๐ ๐ถ๐1 ๐๐ (ฮ๐๐ โ ๐ง๐ )โ1 + ๐๐๐ฃ ,๐๐ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ยฏ1 ) ๐๐ห ๐(๐โ1) ๐โ1 ๐1 ๐๐ ๐=๐
๐โ1 โ ๐=1
๐๐ห๐ฃ ,๐๐ ๐๐ห ๐(๐โ1) (ฮโ๐1 โ ๐ค ยฏ1 )๐โ1 ๐1 ๐๐ .
(48)
For ๐ โ 1 โฅ ๐ โฅ 1, let us group all the terms with coe๏ฌcient ๐๐๐ฃ ,๐๐ in the right-hand side of (48). That is ๐๐๐ฃ ,๐๐ (๐พ1 + ๐พ2 + ๐พ3 ), where โ
๐ โ
๐พ1 =โ
โ ๐ โ ๐ถ๐1 ๐๐ (ฮ๐๐ โ ๐ง๐ )โ1 ๐๐ห ๐๐โ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ยฏ1 )๐๐ห ๐ ๐โ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ยฏ1 )โ1 ,
๐=๐ +1
โ
๐พ2 =โโ
(49)
๐=๐
โ ๐ โ ๐ถ๐1 ๐๐ (ฮ๐๐ โ ๐ง๐ )โ1 ๐๐ห ๐(๐โ1)+ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ยฏ1 )โ ๐๐ห ๐ (๐โ1) ๐โ1 ๐1 ๐๐ ,
๐โ1 โ
๐=๐ +1
๐=๐
and ยฏ1 )๐โ1 ๐พ3 = โ๐๐ห ๐ (๐โ1) (ฮโ๐1 โ ๐ค ๐1 ๐๐ . By (20), we have โ โ ๐โ1 ๐ โ โ ๐พ1 =โ ๐๐ห ๐ ๐ ๐ถ๐1 ๐๐ (ฮ๐๐ โ ๐ง๐ )โ1 โ ๐๐ห ๐ ๐ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ยฏ1 )โ (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ยฏ1 )โ1 ๐=๐
and
โ
๐พ2 =โโ
๐=๐
๐โ2 โ ๐=๐
Notice that
๐๐ห ๐ ๐ ๐ถ๐1 ๐๐
๐โ1 ๐1 ๐๐
๐โ1 โ ๐=๐
โ (ฮ๐๐ โ ๐ง๐ )โ1 + ๐๐ห ๐ (๐โ1) (๐
๐1 ๐๐โ1 (๐ง๐โ1 ) โ ๐ค ยฏ1 )โ ๐โ1 ๐1 ๐๐ .
= (ฮ๐๐ โ ๐ง๐ )โ1 (๐
๐1 ๐๐ (๐ง๐ ) โ ๐ค ยฏ1 )โ1 and
ยฏ1 = โ๐ถ๐1 ๐๐โ1 (ฮ๐๐โ1 โ ๐ง๐โ1 )โ1 + (ฮโ๐1 โ ๐ค ยฏ1 ); ๐
๐1 ๐๐โ1 (๐ง๐โ1 ) โ ๐ค we then have ๐พ1 + ๐พ 2 + ๐พ 3 + ๐ ห๐ห ๐ ๐ ( ) =๐ ห๐ห ๐ (๐โ1) ๐ถ๐1 ๐๐โ1 (ฮ๐๐โ1 โ ๐ง๐โ1 )โ1 โ ฮโ๐1 + ๐ค ยฏ1 + ๐
๐1 ๐๐โ1 (๐ง๐โ1 ) โ ๐ค ยฏ1 ๐โ1 ๐1 ๐๐ = 0.
(50)
628
D. Xia
From (48), (49) and (50), it follows that ๐ โ ห๐ห ๐ ๐ . ๐ฝ2 = โ ๐๐๐ฃ ,๐๐ ๐
(51)
๐ =1
From (41), (47) and (51), we get (39), which proves the theorem.
โก
4. Some operator identities for a commuting ๐-tuple of operators Let ๐ = (๐1 , . . . , ๐๐ ) be a commuting ๐-tuple of operators, i.e., [๐๐ , ๐๐ ] = 0, for ๐, ๐ = 1, 2, . . . , ๐, on a Hilbert space โ. Let โ def ๐ = ๐๐ = closure of {[๐๐โ, ๐๐ ]โ : ๐, ๐ = 1, 2, . . . , ๐} be the defect space of ๐. In the general case, it is di๏ฌerent from the subnormal case; ๐ may not be an invariant subspace of ๐๐โ , ๐ = 1, 2, . . . , ๐. De๏ฌne โ (52) ๐ฆ = closure of {๐1โ๐1 ๐2โ๐2 โ
โ
โ
๐๐โ๐๐ ๐๐ : ๐1 , . . . , ๐๐ = 0, 1, 2, . . .}. Then ๐ฆ is invariant with respect to ๐๐โ and [๐๐โ , ๐๐ ], ๐, ๐ = 1, 2, . . . , ๐. Similar to (5), de๏ฌne def
def
๐ถ๐๐ = [๐๐โ , ๐๐ ]โฃ๐ฆ and ฮ๐ = (๐๐โ โฃ๐ฆ )โ .
(53)
If ๐ is subnormal, then ๐ฆ = ๐ and the operators ฮ๐ and ๐ถ๐๐ de๏ฌned in (53) coincide with that in (5) except for changing ๐๐ to ๐๐ . About the study of ๐ฆ and ๐ถ๐๐ , ฮ๐ , see [5], [6], [9], [16], [18], [19], [20]. We use the same de๏ฌnition of ๐๐๐ (๐ง, ๐ค) in (16) for the commuting operator tuple ๐. Let ๐๐ฆ be the projection from โ to ๐ฆ. โฉ โฉ Lemma 4. If ๐ค โ ๐(ฮ๐ ) ๐(๐๐ ) and ๐ง โ ๐(ฮ๐ ) ๐(๐๐ ), then ๐๐๐ (๐ง, ๐ค) is invertible and ๐๐ฆ (๐๐โ โ ๐ค) ยฏ โ1 (๐๐ โ ๐ง)โ1โฃ๐ฆ = ๐๐๐ (๐ง, ๐ค)โ1 .
(54)
In the case of ๐ being subnormal, ๐(ฮ๐ ) โ ๐(๐๐ ) and this lemma is just the one in [15]. def
def
ยฏ๐ )โ1 and ๐ต = (๐๐ โ ๐ง๐ )โ1 , then from [๐๐โ , ๐๐ ] = ๐ถ๐๐ ๐๐ฆ Proof. Let ๐ด = (๐๐โ โ ๐ค and [๐ด, ๐ต] = ๐ด๐ตโฃ๐ฆ ๐ถ๐๐ ๐๐ฆ ๐ต๐ด
(55)
we have (56) ๐๐ฆ ๐ด๐ตโฃ๐ฆ = ๐๐ฆ ๐ด๐ตโฃ๐ฆ ๐ถ๐๐ ๐๐ฆ ๐ต๐ดโฃ๐ฆ + ๐๐ฆ ๐ต๐ดโฃ๐ฆ . โฉ โ Notice that if ๐ง โ ๐(ฮ๐ ) ๐(๐๐ ), then (ฮโ๐ โ ๐งยฏ)๐ฆ = ๐ฆ. Thus (๐๐ โ ๐งยฏ)๐ฆ = ๐ฆ. โ โ For ๐ฆ โ ๐ฆ, let ๐ข โ ๐ฆ satisfying ๐ฆ = (ฮ๐ โ ๐งยฏ)๐ข. Then (๐๐ โ ๐งยฏ)๐ข = ๐ฆ and โ โ ๐งยฏ)โ1 ๐ฆ = (ฮโ๐ โ ๐งยฏ)โ1 ๐ฆ = ๐ข. (๐๐
(57)
Operator Identities for Subnormal Tuples of Operators
629
Therefore, for any operator ๐ โ ๐ฟ(โ) and ๐ฅ, ๐ฆ โ ๐ฆ, โ โ ๐งยฏ)โ1 ๐ฆ) (๐ต๐ ๐ฅ, ๐ฆ) = (๐ ๐ฅ, (๐๐
= (๐ ๐ฅ, (ฮโ๐ โ ๐งยฏ)โ1 ๐ฆ) = ((ฮ๐ โ ๐ง)โ1 ๐๐ฆ ๐ ๐ฅ, ๐ฆ). Thus ๐๐ฆ (๐๐ โ ๐ง)โ1๐ โฃ๐ฆ = (ฮ๐ โ ๐ง)โ1 ๐๐ฆ ๐ โฃ๐ฆ .
(58)
(๐๐โ โ ๐ค) ยฏ โ1 โฃ๐ฆ = (ฮโ๐ โ ๐ค) ยฏ โ1 .
(59)
From (57), we have From (56), (58) and (59), it follows that ยฏ โ1 + (ฮ๐ โ ๐ง)โ1 (ฮโ๐ โ ๐ค) ยฏ โ1 . ๐๐ฆ ๐ด๐ตโฃ๐ฆ = ๐๐ฆ ๐ด๐ตโฃ๐ฆ ๐ถ๐๐ (ฮ๐ โ ๐ง)โ1(ฮโ๐ โ ๐ค) Thus ๐๐ฆ ๐ด๐ตโฃ๐ฆ ๐๐๐ (๐ง, ๐ค) = ๐ผโฃ๐ฆ , where ๐ผโฃ๐ฆ is the identity operator on ๐ฆ. Similarly from the commutator formula [๐ด, ๐ต] = ๐ต๐ดโฃ๐ฆ ๐ถ๐๐ ๐๐ฆ ๐ด๐ต, we have ๐๐๐ ๐๐ฆ ๐ด๐ตโฃ๐ฆ = ๐ผ๐ฆ . Therefore ๐๐๐ (๐ง, ๐ค) is invertible and (54) holds good.
โก
Part of the following lemma has appeared in [20]. โฉ Lemma 5. If ๐ง โ ๐(ฮ๐ ) ๐(๐๐ ), then [๐
๐๐ (๐ง), ๐
๐๐ (๐ง)] = 0, ๐, ๐ = 1, 2, . . . , ๐. โฉ Furthermore, if ๐ค๐ โ ๐(ฮ๐๐ ) ๐(๐๐๐ ), 1 โค ๐๐ โค ๐, then ๐๐ฆ
๐ โ
โ (๐๐ โ๐ค ยฏ๐ )โ1 (๐๐ โ ๐ง)โ1 โฃ๐ฆ = (ฮ๐ โ ๐ง)โ1 ๐
๐=1
๐ โ
(๐
๐๐ ๐ (๐ง) โ ๐ค ยฏ๐ )โ1 ,
(60)
(61)
๐=1
and ๐๐ฆ (๐๐โ โ ๐งยฏ)โ1 ( =
๐ โ
(๐๐๐ โ ๐ค๐ )โ1 โฃ๐ฆ
๐=1 ๐ โ
(๐
๐๐ ๐ (๐ง)โ โ ๐ค๐ )โ1
) (ฮโ๐ โ ๐งยฏ)โ1
๐=1
=
๐ โ
[(ฮ๐๐ โ ๐ค๐ )โ1 (๐
๐๐๐ (๐ค๐ ) โ ๐งยฏ)โ1 ๐ถ๐๐๐ (ฮ๐๐ โ ๐ค๐ )โ1
๐=1,๐โ=๐ก
+ (ฮ๐๐ โ ๐ค๐ )โ1 ](ฮ๐๐ก โ ๐ค๐ก )โ1 (๐
๐๐๐ก (๐ค๐ก ) โ ๐งยฏ)โ1 .
(62)
630
D. Xia
โ Proof. We write ๐ด๐ = (๐๐ โ๐ค ยฏ๐ )โ1 and ๐ต = (๐๐ โ ๐ง)โ1 . By (55), we have ๐ ( ๐ ) ( ๐โ1 ) ( ๐โ1 ) โ โ โ ๐๐ฆ ๐ด๐ ๐ตโฃ๐ฆ = ๐๐ฆ ๐ด๐ ๐ตโฃ๐ฆ ๐ถ๐๐ ๐๐ฆ ๐ต๐ด๐ โฃ๐ฆ + ๐๐ฆ ๐ด๐ ๐ต๐ด๐ โฃ๐ฆ . ๐=1
๐=1
Thus
( ๐๐ฆ
๐ โ
) ๐ด๐
๐=1
(
๐ โ
= ๐๐ฆ
( ) ๐ตโฃ๐ฆ ๐ผ โ ๐ถ๐๐ (ฮ๐ โ ๐ง)โ1 (ฮโ๐๐ โ ๐คยฏ๐ )โ1 ) ๐ด๐
๐=1
or
( ๐๐ฆ
๐ โ
๐=1
๐ตโฃ๐ฆ (ฮโ๐๐ โ ๐คยฏ๐ )โ1
) ๐ด๐
๐=1
๐ตโฃ๐ฆ = ๐๐ฆ
( ๐โ1 โ
) ๐ด๐
๐ตโฃ๐ฆ (๐
๐๐ ๐ (๐ง) โ ๐ค ยฏ๐ )โ1 .
(63)
๐=1
By mathematical induction with respect to ๐, using ยฏ ๐๐๐ (๐ง, ๐ค) = (๐
๐๐ (๐ง) โ ๐ค)(ฮ ๐ โ ๐ง), (54) and (63), we may prove that ( ๐ ) โ ๐ด๐ ๐ตโฃ๐ฆ = (ฮ๐ โ ๐ง๐ )โ1 (๐
๐1 ๐ (๐ง) โ ๐ค ยฏ1 )โ1 โ
โ
โ
(๐
๐๐ ๐ (๐ง) โ ๐ค ยฏ๐ )โ1 . ๐๐ฆ
(64)
(65)
๐=1
In the case of ๐1 = ๐ and ๐2 = ๐ in (65), we have ๐๐ฆ (๐๐โ โ ๐ค ยฏ1 )โ1 (๐๐โ โ ๐ค ยฏ2 )โ1 (๐๐ โ ๐ง)โ1 โฃ๐ฆ = (ฮ๐ โ ๐ง๐ )โ1 (๐
๐๐ (๐ง) โ ๐ค ยฏ1 )โ1 (๐
๐๐ (๐ง) โ ๐ค ยฏ2 )โ1 .
(66)
In (66) exchanging ๐ and ๐, ๐ค1 and ๐ค2 , we have (60) since ยฏ1 )โ1 , (๐๐โ โ ๐ค ยฏ2 )โ1 ] = 0. [(๐๐โ โ ๐ค Therefore (65) implies (61). Taking adjoints of both sides of (61), we have (62).
(67) โก
De๏ฌne ๐
๐,{๐1 ,...,๐๐ } = (๐๐๐ ) related to ๐ถ๐๐ , ฮ๐ which is the same as in ยง2, such as in (19), ๐๐๐ = ๐
๐๐๐ (๐ง๐ ) and ๐๐๐ = 0 for ๐ > ๐. Theorem 2. Let ๐ = (๐1 , . . . , ๐๐ ) be aโฉcommuting ๐-tuple of operators on โ. If 1 โค ๐, ๐, ๐1 , . . . , ๐๐ โค ๐ and ๐ง๐ โ ๐(ฮ๐๐ ) ๐(๐๐๐ ), then [๐
๐,๐ฝ (๐ง), ๐
๐,๐ฝ (๐ง)] = 0 where ๐ง = (๐ง1 , . . . , ๐ง๐ ), ๐ฝ = {๐1 , . . . , ๐๐ }.
(68)
Operator Identities for Subnormal Tuples of Operators
631
Proof. In the case of ๐ = 1, ๐
๐๐1 (๐ง) = ๐
๐๐1 (๐ง1 ). Thus (60) implies (68) for ๐ = 1. Hence we only have to prove (68) for ๐ โฅ 2. For simplicity of notation, write ๐ด๐ = (๐๐๐ โ ๐ง๐ )โ1 , ๐ต๐ = (๐๐โ โ ๐ค ยฏ๐ )โ1 , ๐ต๐ = โ1 ห โ1 โ๐ค ยฏ๐ ) , ๐
๐๐ = (๐
๐๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ ) for ๐ค๐ โ ๐(๐๐ ), ๐ = 1, 2, . . . , ๐. Let ๐๐ = (ฮ๐๐ โ ๐ง๐ )โ1 and (๐๐โ
def
ยฏ๐ )๐ถ๐๐๐ ๐๐ โ
โ
โ
๐๐ + ๐ถ๐๐๐ ๐๐ โ
โ
โ
๐๐ (๐
๐๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ ) ๐๐๐ (๐, ๐) = (๐
๐๐๐ (๐ง๐ ) โ ๐ค ๐โ1 โ
โ
๐ถ๐๐๐ ๐๐ โ
โ
โ
๐๐ ๐ถ๐๐๐ ๐๐ โ
โ
โ
๐๐ ,
๐=๐+1
for ๐ โ ๐ > 1, and def
๐๐๐ (๐, ๐) = (๐
๐๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ )๐ถ๐๐๐ ๐๐ ๐๐ + ๐ถ๐๐๐ ๐๐ ๐๐ (๐
๐๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ ) for ๐ = ๐ + 1. It is easy to see that (68) is equivalent to ๐๐๐ (๐, ๐) = ๐๐๐ (๐, ๐)
for ๐ < ๐,
(69)
and arbitrary ๐๐ , . . . , ๐๐ . Let ห ๐๐ ๐ถ๐๐ ๐ด๐๐ = ๐
๐
๐ โ
๐๐
and
๐ต๐๐ = ๐ถ๐๐๐
๐=๐
๐ โ
ห ๐๐ , ๐๐ ๐
๐=๐
for ๐ โฅ ๐. De๏ฌne ๐ธ1 = ๐ธ1 (๐, ๐) = ๐ผ and ๐ธ๐ = ๐ธ๐ (๐, ๐), ๐ > 1 by the recurrence formula ๐ธ๐ =
๐โ1 โ
(70)
๐ด๐๐ ๐ธ๐ .
๐=1
De๏ฌne ๐น1 = ๐ผ and ๐น๐ , ๐ > 1 by the recurrence formula ห ๐๐ ๐ถ๐๐๐ ๐๐ )๐น๐โ1 ๐น๐ = (๐๐ + ๐๐ ๐
= (๐
๐๐ ๐ (๐ค๐ )โ โ ๐ง๐ )โ1 ๐น๐โ1 .
(71)
We have to prove that ๐โ1 โ ๐โ1 โ
๐๐ ๐ธ๐ = ๐น๐โ1 ๐1 ,
๐ = 2, 3, . . .
(72)
๐=1 ๐ =๐
and ห ๐๐ ๐ถ๐๐ ๐๐ ๐น๐โ1 ๐1 , ๐ธ๐ = ๐
๐
๐ = 2, 3, . . . ,
(73)
It is easy to see that (72) and (73) hold good for ๐ = 2. Suppose that (72) and (73) hold good for ๐ = 2, 3, . . . , ๐; then we have to prove that (72) and (73) hold
632
D. Xia
good for ๐ = ๐ + 1. It is easy to see that ๐ ๐ โ โ ๐๐ ๐ธ๐ = ๐๐ ๐น๐โ1 ๐1 + ๐๐ ๐ธ๐ ๐=1 ๐ =๐
ห ๐๐ ๐ถ๐๐ ๐๐ ๐น๐โ1 ๐1 = ๐๐ ๐น๐โ1 ๐1 + ๐๐ ๐
๐ ห = (๐๐ + ๐๐ ๐
๐๐ ๐ถ๐๐ ๐๐ )๐น๐โ1 ๐1 ๐
= ๐น๐ ๐1 which proves (72) for ๐ = ๐ + 1. Then from (70), we have ( ๐โ1 ๐โ1 ) โโ ห ๐๐ ๐ธ๐ + ๐ด(๐+1)๐ ๐ธ๐ ๐ธ๐+1 = ๐
๐(๐+1) ๐ถ๐๐ ๐๐+1 ๐๐ ๐+1
๐=1 ๐ =๐
ห ๐(๐+1) ๐ถ๐๐ ๐๐+1 ๐๐ ๐น๐โ1 ๐1 + ๐
ห ๐(๐+1) ๐ถ๐๐ ๐๐+1 ๐๐ ๐
ห ๐๐ ๐ถ๐๐ ๐๐ ๐น๐โ1 ๐1 =๐
๐+1 ๐+1 ๐ ห ห = ๐
๐(๐+1) ๐ถ๐๐ ๐๐+1 (๐๐ + ๐๐ ๐
๐๐ ๐ถ๐๐ ๐๐ )๐น๐โ1 ๐1 ๐+1
๐
ห ๐(๐+1) ๐ถ๐๐ ๐๐+1 ๐น๐ ๐1 =๐
๐+1 by (71), which proves (73) for ๐ = ๐ + 1. Hence (72) and (73) hold good. De๏ฌne def ห ห ๐๐๐ = ๐๐๐ (๐, ๐) = ๐
๐๐ ๐๐๐ (๐, ๐)๐
๐๐ .
(74)
๐๐(๐โ1) = ๐ด๐(๐โ1) + ๐ต๐(๐โ1)
(75)
Then
and ๐๐๐ = ๐ด๐๐ + ๐ต๐๐ โ
๐โ1 โ
๐ด๐๐ ๐ต๐๐
for ๐ > ๐ + 1.
(76)
๐=๐+1
De๏ฌne ๐1 = ๐ผ and ๐๐ for ๐ > 1 by the recurrence formula ๐๐ =
๐โ1 โ
๐๐๐ ๐๐ .
(77)
๐=1
We have to prove that ๐๐ = ๐ธ๐ +
๐โ1 โ
๐ต๐๐ ๐๐ ,
๐=1
by mathematical induction.
๐ = 2, 3, . . .
(78)
Operator Identities for Subnormal Tuples of Operators
633
It is obvious that (78) holds good for ๐ = 2. Suppose (78) holds good for ๐ = 2, 3, . . . , ๐ โ 1. Then from (76) and (77), we have ( ) ๐โ1 ๐โ1 ๐โ2 ๐โ1 โ โ โ โ ๐ด๐๐ โ ๐๐ โ ๐ต๐๐ ๐๐ = (๐๐๐ โ ๐ต๐๐ )๐๐ = ๐ด๐(๐โ1) ๐๐โ1 + ๐ด๐๐ ๐ต๐ ๐ ๐๐ ๐=1
๐=1
= ๐ด๐1 +
๐โ1 โ
( ๐ด๐๐
๐๐ โ
๐โ1 โ ๐=2
๐=1
) ๐ต๐๐ ๐๐
๐ =๐+1
,
๐ =1
๐=2
which is equal to ๐ด๐1 +
๐โ1 โ
๐ด๐๐ ๐ธ๐ by the hypothesis of the induction. By (70), it is
equal to ๐ธ๐ , which proves (78) for all ๐ โฅ 2. From (62), (71) and (73), we have ห ๐๐ ๐ถ๐๐๐ ๐๐ ๐๐ฆ ๐ต๐ ๐
๐โ1 โ
ห ๐1 , ๐ด๐ โฃ๐ฆ = ๐ธ๐ (๐, ๐)๐
for ๐ โฅ 2.
(79)
๐=1
De๏ฌne ๐๐ = ๐๐ (๐, ๐) = ๐๐ฆ ๐ต๐ ๐ต๐
๐ โ
๐ด๐ โฃ๐ฆ .
๐=1
Then by (56), (58) and (61), we have ๐๐ = ๐๐ฆ ๐ต๐ ๐ต๐ ๐ด๐ ๐ถ๐๐๐ ๐๐ฆ ๐ด๐ ๐ต๐ ห ๐๐ ๐
ห ๐๐ ๐ถ๐๐๐ ๐๐ ๐๐ฆ ๐ต๐ = ๐๐ ๐
๐โ1 โ
๐ด๐ โฃ๐ฆ + ๐๐ฆ ๐ต๐ ๐ด๐ ๐ต๐
๐=1 ๐โ1 โ
๐โ1 โ
๐ด๐ โฃ๐ฆ
๐=1
๐ด๐ โฃ ๐ฆ
๐=1
+ ๐๐ฆ ๐ต๐ ๐ด๐ โฃ๐ฆ ๐ถ๐๐๐ ๐๐ฆ ๐ด๐ ๐ต๐ ๐ต๐
๐โ1 โ
๐โ1 โ
๐=1
๐=1
๐ด๐ โฃ๐ฆ + ๐๐ฆ ๐ด๐ ๐ต๐ ๐ต๐
๐ด๐ โฃ๐ฆ .
By (58) and (79), we have ห ๐๐ ๐ธ๐ ๐
ห ๐1 + (๐๐ ๐
ห ๐๐ ๐ถ๐๐ ๐๐ + ๐๐ )๐๐โ1 . ๐๐ = ๐๐ ๐
๐
(80)
We have to prove that ห ๐๐ ๐๐ ๐
ห ๐1 + ๐๐ ๐๐โ1 ๐๐ = ๐๐ ๐
(81)
by mathematical induction. It is easy to see that (81) holds good for ๐ = 2, since ห ๐1 ๐
ห ๐1 and ๐1 = ๐1 ๐
ห ๐1 + ๐ถ๐๐ ๐2 ๐1 = (๐ธ2 + ๐ต21 )๐
ห ๐1 . ห ๐1 = ๐2 ๐
๐ธ2 ๐
2 Suppose (81) holds good for ๐ = 2, 3, . . . , ๐ โ 1. Then from (80) ห ๐๐ ๐ธ๐ ๐
ห ๐1 + ๐๐ ๐
ห ๐๐ ๐ถ๐๐ ๐๐ (๐๐โ1 ๐
ห ๐(๐โ1) ๐๐โ1 ๐
ห ๐1 + ๐๐โ1 ๐๐โ2 ) + ๐๐ ๐๐โ1 ๐๐ = ๐๐ ๐
๐ ห ๐๐ (๐ธ๐ + ๐ต๐(๐โ1) ๐๐โ1 )๐
ห ๐๐ ๐ถ๐๐ ๐๐ ๐๐โ1 ๐๐โ2 + ๐๐ ๐๐โ1 . ห ๐1 + ๐๐ ๐
= ๐๐ ๐
๐
634
D. Xia
Continuing this process, we may prove that โ โ ๐โ1 โ ห ๐1 + ๐๐ ๐๐โ1 . ห ๐๐ โ๐ธ๐ + ๐๐ = ๐๐ ๐
๐ต๐๐ ๐๐ โ ๐
(82)
๐=1
From (78) and (82), we may prove that (81) holds good for all ๐ โฅ 2. From the fact that [๐ต๐ , ๐ต๐ ] = 0, we have ๐๐ (๐, ๐) = ๐๐ (๐, ๐). Therefore (81) implies that ห ๐๐ ๐๐ (๐, ๐)๐
ห๐1 = ๐
ห ๐๐ ๐๐ (๐, ๐)๐
ห ๐1 . ๐
(83)
From (74) and (77), we have ๐โ1 โ
ห ๐๐ ๐
ห ๐๐ ๐๐ ๐
ห ๐1 = ๐
ห ๐๐ ๐๐1๐
ห ๐1 ๐
ห ๐1 + ๐
ห ๐๐ ๐
ห ๐๐ ๐๐1 ๐
ห ๐๐ ๐๐ ๐
ห ๐1 . ๐
(84)
๐=2
For ๐ = 2, from ๐21 = ๐2 , (83) and (84), we have (69) for ๐ = 2 and ๐ = 1. But ๐1 , . . . , ๐๐ are arbitrary numbers in {1, 2, . . . , ๐}, therefore (69) holds good for ๐ = ๐ + 1. Assume that (69) holds good for ๐ = ๐ + 1, . . . , ๐ + ๐, ๐ โฅ 1. Then from (83) and (84) in which ๐ = 1, ๐ = ๐ + 2, we may prove that (69) holds good for ๐ = ๐ + 2, ๐ = 1. Thus (69) holds good for ๐ = ๐ + (๐ + 1), which proves (69) for any ๐ > ๐ and hence the theorem. โก
5. Resolvents formula for a commuting ๐-tuple of operators Let ๐ = {๐1 , . . . , ๐๐ } be a commuting ๐-tuple on a Hilbert space โ. We de๏ฌne ๐ฆ, ๐ถ๐๐ , ฮ๐ etc. as in ยง4. Let us adopt the same matrix ๐๐๐ ,๐๐ for ๐๐ = {๐1 , . . . , ๐๐ } and ๐๐ = {๐1 , . . . , ๐๐ }, 1 โค ๐๐ , ๐๐ โค ๐ as in ยง3. Let ๐๐๐ ,๐๐ be the ๐ฟ(๐ฆ)-valued function def
๐๐๐ ,๐๐ (๐ง1 , . . . , ๐ง๐ ; ๐ค1 , . . . , ๐ค๐ ) = ๐๐ฆ
๐ โ
(๐๐โ๐ โ ๐ค ยฏ๐ )โ1
๐=1
๐ โ
(๐๐๐ โ ๐ง๐ )โ1 โฃ๐ฆ ,
๐=1
for ๐ง๐ , ๐ค๐ โ ๐(๐๐ ). De๏ฌne ๐๐๐ ,๐๐ as in (12). Theorem 3. Let ๐ = {๐1 , . . . , ๐๐ } be a commuting ๐-tuple on a Hilbert space โ. Let ๐ง = (๐ง1 , . . . , ๐ง๐ ) and ๐ค = (๐ค1 , . . . , ๐ค๐ ) satisfy the condition that ๐ค๐ โ ๐(๐๐๐ ), ๐ = 1, 2, . . . , ๐ and ๐ง๐ โ ๐(๐๐๐ ), ๐ = 1, 2, . . . , ๐. Then ๐๐๐ ,๐๐ (๐ง, ๐ค) = ๐๐๐ ,๐๐ (๐ง, ๐ค).
(85)
Proof. In Lemma 5, the formulas (61) and (62) are equivalent to (85) in the case of ๐ = 1 and ๐ = 1 respectively. Let us prove (85) by mathematical induction. Suppose (85) holds good for ๐ = ๐ โ 1 โฅ 1. Let us calculate ๐๐๐ ,๐๐ for any ๐. ยฏ๐ )โ1 and ๐ต๐ = (๐๐๐ โ ๐ง๐ )โ1 . Then by the commutator Let ๐ด๐ = (๐๐โ๐ โ ๐ค formula [๐ด๐ , ๐ต๐ ] = ๐ต๐ ๐ด๐ โฃ๐ฆ ๐ถ๐๐ ๐๐ ๐๐ฆ ๐ด๐ ๐ต๐
Operator Identities for Subnormal Tuples of Operators
635
and ๐ด๐ โฃ๐ฆ = (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 we have ๐๐๐ ,๐๐ = ๐๐ฆ ๐ด1 โ
โ
โ
๐ด๐ ๐ต1 โ
โ
โ
๐ต๐ โฃ๐ฆ ๐ โ = ๐๐ฆ ๐ด1 โ
โ
โ
๐ด๐โ1 ๐ต1 โ
โ
โ
๐ต๐ โฃ๐ฆ (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 ๐ถ๐๐ ๐๐ ๐๐ฆ ๐ด๐ ๐ต๐ โ
โ
โ
๐ต๐ โฃ๐ฆ ๐=1
+ ๐๐๐โ1 ,๐๐ (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 . By (62), we have ๐๐๐ ,๐๐ =
๐ โ
๐๐๐โ1 ,๐๐ ๐๐๐
๐=1
where ๐๐๐ = (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 ๐ถ๐๐ ๐๐ ยฏ๐ )โ1 , since ๐๐๐ = (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค
๐ โ ๐ =๐
(๐
๐๐ ๐๐ (๐ค๐ )โ โ ๐ง๐ )โ1 (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 , for ๐ < ๐ and
ยฏ๐ )โ1 (๐ถ๐๐ ๐๐ ๐๐ฆ ๐ด๐ ๐ต๐ โฃ๐ฆ + 1) (ฮโ๐๐ โ ๐ค ( ) = (ฮโ๐๐ โ ๐ค) ยฏ โ1 (โ(๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ ) + (ฮโ๐๐ โ ๐ค ยฏ๐ ))(๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ )โ1 + 1 = (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ )โ1 . Therefore ( ๐๐๐ ,๐1
๐๐๐ ,๐2
โ
โ
โ
) ( ๐๐๐ ,๐๐ = ๐๐๐โ1 ,๐1
๐๐๐โ1 ,๐2
โ
โ
โ
) ๐๐๐โ1 ,๐๐ ๐น๐
where ๐น๐ = (๐๐๐ )๐,๐=1,2,...,๐ and ๐๐๐ = 0 for ๐ > ๐. By the hypothesis of mathematical induction, ) ( ) ( ๐๐๐โ1 ,๐1 โ
โ
โ
๐๐๐โ1 ,๐๐ = ๐๐๐โ1 ,๐1 โ
โ
โ
๐๐๐โ1 ,๐๐ . Therefore from (23), to show that ๐๐๐ ,๐๐ = ๐๐๐ ,๐๐ , we only have to prove that ยฏ๐ )โ1 . ๐น๐ = (๐
๐๐ ,๐๐ โ ๐ค
(86)
To prove (86), we only have to show that, for any pair (๐, ๐), 1 โค ๐, ๐ โค ๐, ๐๐๐ ๐๐๐ = ๐ผ
(87)
and ๐ โ
๐๐๐ ๐๐ ๐ = 0,
๐ < ๐,
๐ =๐
where (๐๐๐ )๐,๐=1,2,...,๐ = ๐
๐๐ ,๐๐ โ ๐ค ยฏ๐ . Thus it is obvious that (87) holds good. To prove (88), notice that ๐ โ ๐๐๐ ๐๐ ๐ = ๐ผ1 + ๐ผ2 + ๐ผ3 , ๐ =๐
(88)
636
D. Xia
for ๐ โ ๐ > 0, where ๐ผ1 = ๐๐๐ ๐๐๐ = (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ )(ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 ๐ถ๐๐ ๐๐ ๐โ1 โ
๐ผ2 =
๐โ1 โ
๐๐๐ ๐๐ ๐ =
๐ =๐+1
๐ โ (๐
๐๐ ๐๐ (๐ค๐ )โ โ ๐ง๐ )โ1 (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 , ๐ =๐
๐ด๐๐ ๐ด๐ ๐ ,
๐ =๐+1
where ๐ด๐๐ = ๐ถ๐๐ ๐๐
๐ โ (ฮ๐๐ก โ ๐ง๐ก )โ1 (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 , ๐ก=๐
for 1 โค ๐ โค ๐ โค ๐, if ๐ > ๐ + 1, and ๐ผ2 = 0, if ๐ = ๐ + 1. Besides, ๐ผ3 = โ๐ถ๐๐ ๐๐
๐ โ (ฮ๐๐ก โ ๐ง๐ก )โ1 (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ )โ1 . ๐ก=๐
However for ๐ > ๐ + 1, ๐ผ2 = โ
๐โ1 โ
๐ถ๐๐ ๐๐
๐ โ
๐ =๐+1
) ( (ฮ๐๐ก โ ๐ง๐ก )โ1 โ (๐
๐๐ ๐๐ (๐ค๐ )โ โ ๐ง๐ ) + (ฮ๐๐ โ ๐ง๐ )
๐ก=๐
๐ โ ยฏ๐ )โ1 โ
(๐
๐๐ก ๐๐ (๐ค๐ )โ โ ๐ง๐ก )โ1 (ฮโ๐๐ โ ๐ค ๐ก=๐
=
๐โ1 โ
๐ถ๐๐ ๐๐
๐ =๐+1
โ
๐โ1 โ
๐ โ
โ1
(ฮ๐๐ก โ ๐ง๐ก )
๐ =๐+1
(89) โ
๐ โ1 โ
(ฮ๐๐ก โ ๐ง๐ก )โ1
โ1
(๐
๐๐ก ๐๐ (๐ค๐ ) โ ๐ง๐ก )
๐ก=๐ +1
๐ก=๐
๐ถ๐๐ ๐๐
๐ โ
(ฮโ๐๐
โ1
โ๐ค ยฏ๐ )
๐ โ (๐
๐๐ก ๐๐ (๐ค๐ )โ โ ๐ง๐ก )โ1 (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 .
๐ก=๐
๐ก=๐
Most of the terms in the two summations of the right-hand side of (89) cancel each other. Thus ๐ผ2 = ๐ฝ1 + ๐ฝ2 , where ๐โ1
๐ฝ1 = ๐ถ๐๐ ๐๐
โ ๐ก=๐
= ๐ถ๐๐ ๐๐ = โ๐ผ3 ,
(ฮ๐๐ก โ ๐ง๐ก )โ1 (๐
๐๐ ๐๐ (๐ค๐ )โ โ ๐ง๐ )โ1 (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1
๐ โ (ฮ๐๐ก โ ๐ง๐ก )โ1 (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ )โ1 ๐ก=๐
since ยฏ๐ )โ1 = (ฮ๐๐ โ ๐ง๐ )โ1 (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ )โ1 . (๐
๐๐ ๐๐ (๐ค๐ )โ โ ๐ง๐ )โ1 (ฮโ๐๐ โ ๐ค
Operator Identities for Subnormal Tuples of Operators
637
The term ๐ฝ2 = โ๐ถ๐๐ ๐๐ (ฮ๐๐ โ ๐ง๐ )
โ1
๐ โ
(๐
๐๐ก ๐๐ (๐ค๐ )โ โ ๐ง๐ก )โ1 (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1
๐ก=๐+1
๐ ( ) โ โ = (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ ) โ (ฮ๐๐ โ ๐ค ยฏ๐ ) (๐
๐๐ก ๐๐ (๐ค๐ )โ โ ๐ง๐ก )โ1 (ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 .
(90)
๐ก=๐+1
But the product of the ๏ฌrst four factors in ๐ผ1 from the left is (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ )(ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 ๐ถ๐๐ ๐๐ (๐
๐๐ ๐๐ (๐ค๐ )โ โ ๐ง๐ )โ1 ( ) = (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ ) โ (๐
๐๐ ๐๐ (๐ค๐ )โ โ ๐ง๐ ) + (ฮ๐๐ โ ๐ง๐ ) (๐
๐๐ ๐๐ (๐ค๐ )โ โ ๐ง๐ )โ1 = โ(๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ ) + (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ )(ฮ๐๐ โ ๐ง๐ )๐๐๐ ๐๐ (๐ง๐ , ๐ค๐ )โ1 (ฮโ๐ โ ๐ค ยฏ๐ )
(91)
ยฏ๐ ) + (ฮโ๐๐ โ ๐ค ยฏ๐ ). = โ(๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค
Thus ๐ผ1 + ๐ฝ2 = 0, which proves that ๐ผ1 + ๐ผ2 + ๐ผ3 = 0, for ๐ > ๐ + 1. If ๐ = ๐ + 1, then ๐ผ2 = 0, ๐ผ1 = (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ )(ฮโ๐๐ โ ๐ค ยฏ๐ )โ1 ๐ถ๐๐ ๐๐ (๐
๐๐ ๐๐ (๐ค๐ )โ โ ๐ง๐ )โ1 ๐๐๐ ๐๐ +1 (๐ง๐+1 , ๐ค๐ )โ1 ( ) = โ (๐
๐๐ ๐๐ (๐ง๐ ) โ ๐ค ยฏ๐ ) + (ฮโ๐๐ โ ๐ค ยฏ๐ ) ๐๐๐ ๐๐+1 (๐ง๐+1 , ๐ค๐ )โ1 by (91) and ๐ผ3 = โ๐ถ๐๐ ๐๐ (ฮ๐๐ โ ๐ง๐ )โ1 ๐๐๐ ๐๐+1 (๐ง๐+1 , ๐ค๐ )โ1 . Thus ๐ผ1 + ๐ผ3 = 0 which proves ๐ผ1 + ๐ผ2 + ๐ผ3 = 0 as well, and hence it proves (88). Therefore (85) is proved โก
References [1] A. Athevale, On joint hyponormality of operators, Proc. Amer. Math. Soc. 103 (1988), 417โ423. [2] J.B. Conway, The Theory of Subnormal Operators, Math. Sur. Mono. V.36, Amer. Math. Soc. 1991, 1โ435. [3] J.B. Conway, Towards a functional calculus for subnormal tuples: the minimal extension, Trans. Amer. Math. Soc. 329 (1991), 543โ577, the minimal extension and approximation in several complex variable, Proc. Symp. Pure Math. 51, Part I, Amer. Math. Soc. Providence (1990). [4] R.E. Curto, Joint hyponormality: a bridge between hyponormality and subnormality, Proc. Symp. Pure Math. 5 (1990), 69โ91. [5] J. Eschmeier and M. Putinar, Some remarks on spherical isometries, in Vol. โSystems, Approximation, Singular Integral Operators and Related Topicsโ(A.A. Borichev and N.K. Nikolskii, eds.), Birkhยจ auser, Basel et al. (2001), 271โ292. [6] J. Gleason, Matrix construction of subnormal tuples of ๏ฌnite type, Journal of Math. Anal. Appl. 284(2) (2003), 593โ602. [7] B. Gustafsson and M. Putinar, Linear analysis of quadrature domains. II, Israel J. Math. 119 (2000), 187โ216.
638
D. Xia
[8] M. Putinar, Spectral inclusion for subnormal n-tuples, Proc. Amer. Math. Soc. 90 (1984), 405โ406. [9] M. Putinar, Linear analysis of quadrature domains. III, J. Math. Anal. Appl. 239(1) (1999), 101โ117. [10] J.D. Pincus, D. Xia and J. Xia, The analytic model of a hyponormal operator with rank one self-commutators, Integr. Equ. Oper. Theory, 7 (1984), 516โ535. Note on this paper, Integr. Equ. Oper. Theory, 7 (1984), 893โ894. [11] J.D. Pincus, D. Xia, A trace formula for subnormal tuples of operators, Integr. Equ. Oper. Theory, 14 (1991), 390โ398. [12] J.D. Pincus, D. Zheng, A remark on the spectral multiplicity of normal extensions of commuting subnormal tuples, Integr. Equ. Oper. Theory, 16 (1993), 145โ153. [13] D. Xia, Analytic model of subnormal operators, Integr. Equ. Oper. Theory, 10 (1987), 255โ289. [14] D. Xia, Analytic theory of subnormal operators, Integr. Equ. Oper. Theory, 10 (1987), 890โ903. [15] D. Xia, Analytic theory of a subnormal n-tuple of operators, in Operator Theory, Operator Algebra and Applications, Proceedings of Symp. on Pure Math. 51(1) (1990), 617โ640. [16] D. Xia, Hyponormal operators with ๏ฌnite rank self-commutators and quadrature domains, J. Math. Anal. Appl. 203 (1996), 540โ559. [17] D. Xia, Trace formulas for some operators related to quadrature domains in Riemann surfaces, Integr. Equ. Oper. Theory, 47 (2003), 123โ130. [18] D. Xia, Hyponormal operators with rank one self-commutator and quadrature domains, Integr. Equ. Oper. Theory, 48 (2004), 115โ135. [19] D. Xia, On a class of operators of ๏ฌnite type, Integr. Equ. Oper. Theory, 54 (2006), 131โ150. [20] D. Xia, Right spectrum and trace formula of subnormal tuples of operators of ๏ฌnite type, Integr. Equ. Oper. Theory, 55 (2006), 439โ452. [21] D.V. Yakubovich, Subnormal operators of ๏ฌnite type I, Xiaโs model and real algebraic curves, Revista Matem. Iber. 14 (1998), 95โ115. [22] D.V. Yakubovich, Subnormal operators of ๏ฌnite type II, Structure theorems, Revista Matem. Iber. 14 (1998), 623โ689. [23] D.V. Yakubovich, Real separated algebraic curves, quadrature domains, Ahlfors type functions and operator theory, Jour. Func. Analysis, 236 (2006), 25โ58. Daoxing Xia Department of Mathematics Vanderbilt University Nashville, TN 37240, USA e-mail:
[email protected] [email protected]