Fractional Differentiation Inequalities

Fractional Differentiation Inequalities George A. Anastassiou Fractional Differentiation Inequalities 123 George ...

Author: George A. Anastassiou

81 downloads 1160 Views 14MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Fractional Differentiation Inequalities

George A. Anastassiou

Fractional Differentiation Inequalities

123

George Anastassiou University of Memphis Department of Mathematical Sciences Memphis, TN 38152 [email protected]

ISBN 978-0-387-98127-7 e-ISBN 978-0-387-98128-4 DOI 10.1007/978-0-387-98128-4 Springer Dordrecht Heidelberg London New York Library of Congress Control Number: 2009928453 Mathematics Subject Classification (2000): 26A33, 26D10, 26D15 c Springer Science+Business Media, LLC 2009 All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Cover illustration: Created by Sven Geier of the California Institute of Technology. The chaos image is entitled “Sufficiently Advanced Technology.” The fractal image is entitled “Unbalanced.” Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com)

To the unknown mathematician struggling for the truth, and my wife Koula and daughters Angela and Peggy

Contents

Preface

xiii

1 Introduction

1

2 Opial-Type Inequalities for Functions and Their Ordinary and Canavati Fractional Derivatives 7 2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 3 Canavati Fractional Opial-Type Inequalities and Diﬀerential Equations 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . 3.2 Preliminaries . . . . . . . . . . . . . . . . . . . . 3.3 Main Results . . . . . . . . . . . . . . . . . . . . 3.4 Applications . . . . . . . . . . . . . . . . . . . . . 3.5 Other Fractional Diﬀerential Equations . . . . .

Fractional

4 Riemann–Liouville Opial-Type Inequalities for Derivatives 4.1 Introduction and Preliminaries . . . . . . . . . . 4.2 Main Results . . . . . . . . . . . . . . . . . . . . 4.3 Applications . . . . . . . . . . . . . . . . . . . . .

Fractional

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . . . . . . . . . . . . . . .

23 23 24 26 34 38

41 41 44 48 vii

viii

Contents

5 Opial-Type Lp -Inequalities for Riemann–Liouville Fractional Derivatives 5.1 Introduction and Preliminaries . . . . . . . . . . . . . . . . 5.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . .

53 53 56

6 Opial-Type Inequalities Involving Derivatives of Two Functions and 6.1 Introduction . . . . . . . . . . . . 6.2 Preliminaries . . . . . . . . . . . 6.3 Main Results . . . . . . . . . . . 6.4 Applications . . . . . . . . . . . .

67 67 69 72 96

Canavati Fractional Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . .

. . . .

7 Opial-Type Inequalities for Riemann–Liouville Fractional Derivatives of Two Functions with Applications 107 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 7.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 7.3 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . 109 7.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 8 Canavati Fractional Opial-Type Inequalities for Several Functions and Applications 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . 8.3 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . 8.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . .

149 149 150 152 169

9 Riemann–Liouville Fractional Opial-Type for Several Functions and Applications 9.1 Introduction . . . . . . . . . . . . . . . . . 9.2 Background . . . . . . . . . . . . . . . . . 9.3 Main Results . . . . . . . . . . . . . . . . 9.4 Applications . . . . . . . . . . . . . . . . .

Inequalities . . . .

. . . .

. . . .

179 179 180 181 195

10 Converse Canavati Fractional Opial-Type Inequalities for Several Functions 10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 10.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . 10.3 Main Results . . . . . . . . . . . . . . . . . . . . . . . . 10.3.1 Results Involving Two Functions . . . . . . . . . 10.3.2 Results Involving Several Functions . . . . . . .

. . . . .

. . . . .

205 205 206 209 209 220

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

11 Converse Riemann–Liouville Fractional Opial-Type Inequalities for Several Functions 229 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 229 11.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 230

Contents

11.3 Main Results . . . . . . . . . . . . . . . . . . . . . . 11.3.1 Results Involving Two Functions . . . . . . . 11.3.2 Results Involving Several Functions . . . . . 11.3.3 Results with Respect to Generalized Riemann Liouville Fractional Derivative . . . . . . . .

. . . . . . – . .

ix

. . 231 . . 231 . . 244 . . 251

12 Multivariate Canavati Fractional Taylor Formula 257 12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 12.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258 13 Multivariate Caputo Fractional Taylor Formula 269 13.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 269 13.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 14 Canavati Fractional Multivariate Opial-Type Inequalities on Spherical Shells 279 14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 279 14.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280 15 Riemann–Liouville Fractional Multivariate Opial-Type Inequalities over a Spherical Shell 15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 15.2 Background—I . . . . . . . . . . . . . . . . . . . . . . . . 15.3 Background—II . . . . . . . . . . . . . . . . . . . . . . . . 15.4 Background—III . . . . . . . . . . . . . . . . . . . . . . . 15.5 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . 15.5.1 Riemann–Liouville Fractional Opial-Type Inequalities Involving One Function . . . . . . . . 15.5.2 Riemann–Liouville Fractional Opial-Type Inequalities Involving Two Functions . . . . . . . . 15.5.3 Riemann–Liouville Fractional Opial-Type Inequalities Involving Several Functions . . . . . .

. 369

16 Caputo Fractional Multivariate Opial-Type over a Spherical Shell 16.1 Introduction . . . . . . . . . . . . . . . . . . 16.2 Background—I . . . . . . . . . . . . . . . . 16.3 Main Results . . . . . . . . . . . . . . . . . 16.3.1 Results Involving One Function . . . 16.3.2 Results Involving Two Functions . . 16.3.3 Results Involving Several Functions 16.4 Background—II . . . . . . . . . . . . . . . . 16.5 Main Results on a Spherical Shell . . . . . . 16.5.1 Results Involving One Function . . . 16.5.2 Results Involving Two Functions . .

. . . . . . . . . .

. . . . .

319 319 320 323 329 334

. 334 . 350

Inequalities . . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

391 391 392 397 397 402 411 419 424 424 427

x

Contents

16.5.3 Results Involving Several Functions . . . . . . . . . 431 16.6 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 436 17 Poincar´ e-Type Fractional Inequalities 17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 17.2 Fractional Poincar´e Inequalities Results . . . . . . . . 17.3 Applications of Fractional Poincar´e Inequalities . . . . 17.4 Fractional Mean Poincar´e Inequalities . . . . . . . . . 17.5 Applications of Fractional Mean Poincar´e Inequalities

. . . . .

. . . . .

. . . . .

445 445 446 457 473 479

18 Various Sobolev-Type Fractional Inequalities 483 18.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 483 18.2 Various Univariate Sobolev-Type Fractional Inequalities . . 484 18.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 503 19 General Hilbert–Pachpatte-Type Integral Inequalities 505 19.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 505 19.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . 506 20 General Multivariate Hilbert–Pachpatte-Type Integral Inequalities 523 20.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 523 20.2 Symbols and Basics . . . . . . . . . . . . . . . . . . . . . . . 524 20.3 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . 527 21 Other Hilbert–Pachpatte-Type Fractional Inequalities 21.1 Background . . . . . . . . . . . . . . . . . 21.2 Univariate Results . . . . . . . . . . . . . 21.3 Multivariate Results . . . . . . . . . . . .

Integral 545 . . . . . . . . . . 545 . . . . . . . . . . 550 . . . . . . . . . . 553

22 Canavati Fractional and Other Approximation of Csiszar’s f -Divergence 563 22.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . 563 22.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . 568 23 Caputo and Riemann–Liouville Fractional Approximation of Csiszar’s f -Divergence 577 23.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . 577 23.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581 24 Canavati Fractional Ostrowski-Type Inequalities 589 24.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 589 24.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 591

Contents

xi

25 Multivariate Canavati Fractional Ostrowski-Type Inequalities 595 25.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 595 25.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 598 26 Caputo Fractional Ostrowski-Type Inequalities 26.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 26.2 Univariate Results . . . . . . . . . . . . . . . . . . . . . . . 26.3 Multivariate Results . . . . . . . . . . . . . . . . . . . . . .

615 615 618 622

27 Appendix 635 27.1 Conversion Formulae for Diﬀerent Kinds of Fractional Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . 635 27.2 Some Basic Fractional Derivatives . . . . . . . . . . . . . . 638 References

641

List of Symbols

671

Index

673

Preface

In 1991, although I had graduated in 1984, my great Ph.D. thesis advisor J. H. B. Kemperman recommended that I read the very interesting paper of J. Canavati [101], having to do with fractional derivatives and containing a fractional Taylor formula. The hope was to use this in the study of positive linear operators in approximation theory, so I could extend my Ph.D. thesis results to arbitrary order derivatives. At that time I was not successful regarding that goal. So I put this important paper aside for a few years, but it was always in my mind to use it somewhere else. This fortunately happened in 1996 when I saw the proof of the basic Opial-type inequality (see Theorem 1.2). I was able to extend it, via the use of [101], to the fractional level. That led to my paper [17], for which I received an award in 2001 from the Academy of Athens, Greece, as the best article of the year for mathematical analysis, the “Pallas Award and Prize”, an international competition for Greeks. All the above encouraged to me to do more research in the area of fractional inequalities (see [15–66]). So ﬁnally a natural consequence was this research monograph, the ﬁrst of its kind in the area of fractional inequalities. Fractional diﬀerentiation inequalities are by themselves an important and great mathematical topic for research. Furthermore these have many applications; the most important ones are in establishing the uniqueness of the solution in fractional diﬀerential equations and systems and in fractional partial diﬀerential equations. Also they provide upper bounds to the solution of the above equations. In this monograph we give many such applications. Each chapter is self-contained and can be read independently of the others and several graduate courses and seminars can be designed xiii

xiv

Preface

based on this monograph. The list of fractional diﬀerentiation inequalities we study includes Opial, Poincar´e, Sobolev, Hilbert, information theory, and Ostrowski. We give results for the above using three diﬀerent types of fractional derivatives: Canavati, Riemann – Liouville, and Caputo. We examine the univariate and multivariate cases. The ﬁnal preparation of this book took place in Memphis, USA, and in Athens, Greece, during 2008. Fractional calculus has become very useful over the last forty years due to its many applications in almost all the applied sciences. We now see applications in acoustic wave propagation in inhomogeneous porous material, diﬀusive transport, ﬂuid ﬂow, dynamical processes in self-similar structures, dynamics of earthquakes, optics, geology, viscoelastic materials, biosciences, bioengineering, medicine, economics, probability and statistics, astrophysics, chemical engineering, physics, splines, tomography, ﬂuid mechanics, electromagnetic waves, nonlinear control, signal processing, control of electronic power, converters, chaotic dynamics, polymer science, proteins, polymer physics, electrochemistry, statistical physics, rheology, thermodynamics, and neural networks, among others. By now almost all ﬁelds of research in science and engineering use fractional calculus to better describe them. So as expected, this monograph, as part of fractional calculus, is useful for researchers and graduate students in research, seminars, and advanced graduate courses, in pure and applied mathematics, engineering, and all other applied sciences. I would like to thank my family for their support and for their tolerance in accepting my continuous mathematical habit. Also I am greatly indebted and thankful to Raluca Pop and Razvan Mezei for the heroic and fantastic technical preparation of the manuscript in a very short time. George A. Anastassiou November 1, 2008 Memphis, TN, USA

1 Introduction

This monograph is about fractional diﬀerentiation inequalities. These inequalities have applications in fractional diﬀerential equations in establishing the uniqueness of the solution of initial value problems, giving upper bounds to their solutions, and these inequalities by themselves are of great interest and deserve to be studied thoroughly. This monograph is the ﬁrst of its kind, and is a natural outgrowth of the author’s research work (see [13–66]) over the last ﬁfteen years. In Chapters 2–11 and 14–16 we study Opial fractional inequalities, in Chapters 12 and 13 we study multivariate fractional Taylor formulae needed for our multivariate results. In Chapter 17 we present Poincar´e fractional inequalities, in Chapter 18 we present Sobolev fractional inequalities, and in Chapters 19–21 we study Hilbert–Pachpatte fractional inequalities. In Chapters 22–23 we derive results using fractional derivatives in the study of Csiszar’s f -divergence in information theory and we ﬁnish with Chapters 24–26 where we present results for Ostrowski fractional inequalities. The writing style in the monograph was chosen to help the reader. Each chapter can be read independently of any other chapter and each contains a complete story in itself. So all tools needed per chapter to describe the main results and applications can be found therein. Our results are presented for all three main types of fractional derivatives: Riemann–Liouville, Caputo, and Canavati types for the univariate and multivariate cases. At the end we have an appendix where we present conversion formulae from one type of fractional derivative to another, we justify the treatment G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 1,

1

2

1. Introduction

of fractional inequalities per type separately, and we give a table of basic functions’ fractional derivatives. The monograph concludes with a very rich list of 412 references. At this point we would like to mention the basic inequalities involving ordinary derivatives. They are presented here at the fractional level and expanded in all possible directions. Theorem 1.1. (1960, Z. Opial [315]) Let f ∈ C 1 ([0, h]) such that f (0) = f (h) = 0, f (x) > 0 on (0, h) . Then

h

|f (x) f (x)| dx ≤

0

h 4

h

f 2 (x) dx,

(1.1)

0

where h/4 is the best constant. Opial’s inequality (1.1) has an important modiﬁcation which is used in many applications to diﬀerential equations (see D. Willett [406]). We emulate the next inequality at the fractional level, regarding our study of Opial fractional inequalities. Theorem 1.2. ([4, p. 8]) Let x (t) be absolutely continuous in [0, a] , and x (0) = 0. Then a a a 2 |x (t) x (t)| dt ≤ (x (t)) dt, (1.2) 2 0 0 the last inequality holds as equality, iﬀ x (t) = ct, where c is a constant. We continue with the Poincar´e inequality. Theorem 1.3. [2] Let Ω ⊂ Rn , n ∈ N; it holds that uLp (Ω) ≤ C ∇uLp (Ω) ,

(1.3)

for functions u with vanishing mean value over Ω, 1 ≤ p ≤ ∞, for general Ω, where ∇uLp (Ω) is deﬁned as the Lp -norm of the Euclidean norm of ∇u. Next we give the Sobolev inequality. Theorem 1.4. ([159, p. 263], the Gagliardo–Nirenberg–Sobolev inequality) Let 1 ≤ p < n. Then there exists a constant C, depending only on p and n, such that (1.4) uLp∗ (Rn ) ≤ C DuLp (Rn ) ,

1. Introduction

3

for all u ∈ Cc1 (Rn ) . Here p∗ := np/n − p, p∗ > p, and Du is the gradient of u. We also mention Hilbert’s inequality. Theorem 1.5. ([191], Theorem 316) If p > 1, q = p/ (p − 1) and

∞

f (x) dx ≤ F, p

0

then

0

∞

0

∞

g q (y) dy ≤ G,

0 ∞

π f (x) g (y) dxdy < F 1/p G1/q , x+y sin (π/p)

(1.5)

where f, g are nonnegative measurable functions, unless f ≡ 0 or g ≡ 0. The constant π cos ec (π/p) is the best possible in (1.5) . See also the ordinary Hilbert–Pachpatte inequality, Theorem 19.2 here, and in B. Pachpatte [320, Theorem 1]. For the Csiszar’s f -divergence description please see Preliminaries 22.1. We ﬁnish with the famous Ostrowski inequality.

Theorem 1.6. [318] Let f ∈ C ([a, b]) , x ∈ [a, b] . Then 2 1 b x − a+b 1 2 + f (t) dt − f (x) ≤ (b − a) f ∞ . 2 b − a a 4 (b − a)

(1.6)

Inequality (1.6) is sharp. For the last also see Theorem 22.1, p. 498 of [19]; there we prove that the optimal function for the sharpness of (1.6) is f ∗ (t) = |t − x| (b − a) , α > 1. α

The author was the ﬁrst to study Opial fractional derivative inequalities; in 1998, see [15, 17] and [18], which involve fractional derivatives of the Canavati type. For the last papers the author’s motivation was Theorem 1.2. Many years later he discovered the following very important theorem, giving one more reason for writing this monograph. Theorem 1.7. ([261, Theorem 2, 1985], Opial inequality for fractional integrals, by E. Love) If p > 0, q > 0, p + q = r ≥ 1, 0 ≤ a < b ≤ ∞, γ < r, ω (x) is decreasing and positive in (a, b) , f (x) is measurable and

4

1. Introduction

nonnegative on (a, b) , I α is the Riemann–Liouville operator of fractional integration deﬁned by x α−1 (x − t) f (t) dt for α > 0, (I α f ) (x) = Γ (α) a I 0 f (x) = f (x) ,

∞ and I β f is deﬁned similarly for β ≥ 0. Here Γ (α) = 0 e−t tα−1 dt, the gamma function. Then b

q p β I f (x) xγ−αp−βq−1 ω (x) dx [(I α f ) (x)] a

≤C

b

r

[f (x)] xγ−1 ω (x) dx,

(1.7)

a

where C=

Γ (1 − (γ/r)) Γ (α + 1 − (γ/r))

p

Γ (1 − (γ/r)) Γ (β + 1 − (γ/r))

q .

In a 1695 letter to L’Hospital, Leibniz asked, “Can we generalize ordinary derivatives to ones of arbitrary order?” L’Hospital replied to Leibniz with another question, “What if the order is 1/2?” Then Leibniz, in a letter dated September 30, 1695 replied, “It will lead to a paradox, from which one day many useful consequences will be drawn.” That was the beginning of fractional calculus. The subject has been ongoing for more than 300 years now. Many famous mathematicians contributed to the topic over the years such as Liouville, Euler, Laplace, Lagrange, Riemann, Weyl, Fourier, Abel, Lacroix, Grunwald, and Letnikov. For more details on the history of the subject see [311] and [295]. The ﬁrst attempt to give a rigorous deﬁnition of fractional derivatives and study the subject in depth was by Liouville during 1832–1855. The concept grew out of his earlier work in electromagnetism. Fractional derivatives describe solutions of fractional integral equations, many times arising from physics, as done by Abel in 1823 to solve the brachistochrone problem. Fractional calculus has a long history but, from the applicative point of view, it fell into oblivion for hundreds of years because of lack of applications to other sciences such as physics and engineering. This oblivion was also due to its complexity and lack of physical and geometric connection. Only in 2002 (see [335]) did Igor Podlubny show a serious and convincing geometric and physical interpretation of fractional integration and fractional diﬀerentiation. However, fractional calculus has emerged as very useful over the last forty years due to its many applications in almost all applied sciences. We now see applications in acoustic wave propagation in inhomogeneous porous material, diﬀusive transport, ﬂuid ﬂow,

1. Introduction

5

dynamical processes in self-similar structures, dynamics of earthquakes, optics, geology, viscoelastic materials, biosciences, bioengineering, medicine, economics, probability and statistics, astrophysics, chemical engineering, physics, splines, tomography, ﬂuid mechanics, electromagnetic waves, nonlinear control, signal processing, control of electronic power, converters, chaotic dynamics, polymer science, proteins, polymer physics, electrochemistry, statistical physics, rheology, thermodynamics, neural networks, and many others. By now almost all ﬁelds of research in science and engineering use fractional calculus to better describe them (see also [197]). The current mathematical theory of fractional calculus seems to lag behind the needs for mathematical modeling of all the above applications. Therefore there exists a lot of room for mathematical theoretical expansion on the subject. Most of the above applications lead to the study of fractional diﬀerential equations; an excellent source on the topic is [333]. The recent development of fractal theory has shown great connections to fractional calculus. In [248], Y. Liang and W. Su have shown the linear relationship between the Hausdorﬀ dimension of the graph of a generalized Weierstrass fractal function and the order of its fractional calculus. And there exists a large set of literature on the last topic. In general, fractional calculus currently provides the best descriptions for fractals and chaotic situations. We must not overlook the great fractional calculus encyclopedic monograph by S. Samko, A. Kilbas, and O. Marichev [366], and supports which strongly all the above recent scientiﬁc activities; see also [134] by K. Diethelm. Fractional derivatives might exist more easily than ordinary ones (see [352]), and thus potentially be more attractive. Fractional calculus has developed a lot in recent years and especially since 1974 when the ﬁrst international conference in the ﬁeld took place, organized by B. Ross, in New Haven, Connecticut, USA. Now we frequently see such conferences. Overall we observed for fractional calculus a very long but essential infancy of about 250 years, a great rich adolescence of about 50 years, and we foresee a magniﬁcent mature future for many years to come providing the best descriptions of the complexity of our modern science.

2 Opial-Type Inequalities for Functions and Their Ordinary and Canavati Fractional Derivatives

Several Lp -form Opial-type inequalities [315] are presented involving functions and their ordinary and generalized fractional derivatives. The above follow a generalization of Taylor’s formula for generalized fractional derivatives. The chapter ends with the application of the derived inequalities in proving the uniqueness of solution/upper bound to the solution of some known very general fractional diﬀerential equations. This treatment is based on [18].

2.1 Preliminaries In the sequel we follow [101]. Let g ∈ C([0, 1]), n := [ν], [·] the integral part, ν > 0, and α := ν − n (0 < α < 1) . Deﬁne x 1 (Jν g)(x) := (x − t)ν−1 g(t)dt, 0 ≤ x ≤ 1, (2.1) Γ(ν) 0 the integral, where Γ is the gamma function; Γ (ν) = ∞ Riemann–Liouville −t ν−1 e t dt. We deﬁne the subspace C ν ([0, 1]) of C n ([0, 1]): 0 C ν ([0, 1]) := {g ∈ C n ([0, 1]) : J1−α Dn g ∈ C 1 ([0, 1])}, where D := d/dx. C ν ([0, 1]) is a Banach space with norm

[[f ]]ν = max f ∞ , f ∞ , . . . , f (n−1) , Dν f ∞ , if ν ≥ 1, ∞

G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 2,

7

8

2. Opial-Type Inequalities

and [[f ]]ν = Dν f ∞ , if 0 < ν < 1, for more see [101]. So let g ∈ C ν ([0, 1]); we deﬁne the Canavati ν-fractional derivative of g as (2.2) Dν g := DJ1−α Dn g. When ν ≥ 1 we have the Taylor’s formula g(t) = g(0)+g (0)t+g (0)

t2 tn−1 +· · ·+g (n−1) (0) +(Jν Dν g)(t), ∀t ∈ [0, 1]. 2! (n − 1)! (2.3)

When 0 < ν < 1 we get ∀t ∈ [0, 1].

g(t) = (Jν Dν g)(t),

(2.4)

Next we carry the above notions over to arbitrary [a, b] ⊆ R. Let x, x0 ∈ [a, b] such that x ≥ x0 , x0 is ﬁxed. Let f ∈ C([a, b]) and deﬁne x 1 (x − t)ν−1 f (t)dt, x0 ≤ x ≤ b, (2.5) (Jνx0 f )(x) := Γ(ν) x0 the generalized Riemann–Liouville integral. We deﬁne the subspace Cxν0 ([a, b]) of C n ([a, b]): x0 Dn f ∈ C 1 ([x0 , b])}. Cxν0 ([a, b]) := {f ∈ C n ([a, b]) : J1−α

So let f ∈ Cxν0 ([a, b]); we deﬁne the generalized ν-fractional derivative of f over [x0 , b] as x0 f (n) (f (n) := Dn f ). (2.6) Dxν0 f := DJ1−α Notice that x0 (J1−α f (n) )(x) =

1 Γ(1 − α)

x

(x − t)−α f (n) (t)dt

x0

exists for f ∈ Cxν0 ([a, b]). We present the following generalization of Taylor’s formula, the fractional Taylor formula (see also [101]). Theorem 2.1. Let f ∈ Cxν0 ([a, b]), x0 ∈ [a, b] ﬁxed. (i) If ν ≥ 1 then f (x) = f (x0 )+f (x0 )(x−x0 )+f (x0 )

(x − x0 )2 (x − x0 )n−1 +· · ·+f (n−1) (x0 ) 2 (n − 1)!

+ (Jνx0 Dxν0 f )(x), all x ∈ [a, b] : x ≥ x0 .

(2.7)

2.1 Preliminaries

9

(ii) If 0 < ν < 1 we get f (x) = (Jνx0 Dxν0 f )(x),

all x ∈ [a, b] : x ≥ x0 .

(2.8)

Proof. For x = x0 (2.7) holds trivially. Otherwise assume that x = x0 and deﬁne w := ϕ(t) := x0 + t(x − x0 ), 0 ≤ t ≤ 1. Thus ϕ(0) = x0 and ϕ(1) = x. Let g(t) := f (x0 + t(x − x0 )); that is, g(0) = f (x0 ) and g(1) = f (x). Here we see that (ν ≥ 1), g (i) (t) = f (i) (x0 + t(x − x0 ))(x − x0 )i ,

i = 0, 1, . . . , n − 1.

In particular g (i) (0) = f (i) (x0 )(x − x0 )i . Furthermore g ∈ C ν ([0, 1]). Consequently from (2.3) we have f (x)

g (n−1) (0) g (0) + ··· + + (Jν Dν g)(1) 2 (n − 1)! f (x0 ) = f (x0 ) + f (x0 )(x − x0 ) + (x − x0 )2 2 f (n−1) (x0 ) (x − x0 )n−1 + (Jν Dν g)(1), ν ≥ 1. +··· + (n − 1)! = g(1) = g(0) + g (0) +

It remains to translate (Jν Dν g)(1) for any ν > 0. Notice that w − x0 ; t = 0 iﬀ w = x0 , t= x − x0 and dw . dt = x − x0 Denote z := x0 + y(x − x0 ) = ϕ(y), y ∈ [0, 1], thus

z−w z − x0 , and y − t = ; x − x0 x − x0 in particular t = y iﬀ w = z. We observe that y 1 (n) (y − t)(1−α)−1 · (J1−α g )(y) = Γ(1 − α) 0 y=

f (n) (x0 + t(x − x0 ))(x − x0 )n dt

=

(x − x0 )n Γ(1 − α)

=

(J x0 f (n) )(z) (x − x0 )n 1−α (1−α) . (x − x0 )

y

(y − t)(1−α)−1 f (n) (x0 + t(x − x0 ))dt 0

10

2. Opial-Type Inequalities

That is, x0 f (n) )(z) (J1−α g (n) )(y) = (x − x0 )n−(1−α) · (J1−α

or x0 (J1−α g (n) )(y) = (x − x0 )ν−1 (J1−α f (n) )(x0 + y(x − x0 )).

Then (D(J1−α g (n) ))(y)

= =

x0 (x − x0 )ν D(J1−α f (n) )(x0 + y(x − x0 )) (x − x0 )ν (Dxν0 f )(z).

That is, (Dν g)(y) = (x − x0 )ν (Dxν0 f )(z).

(2.9)

In particular (Dν g)(1) = (x − x0 )ν (Dxν0 f )(x). Next by using (2.9) we examine (Jν Dν g)(1)

1

=

1 Γ(ν)

=

(x − x0 )ν Γ(ν)

= = =

(1 − t)ν−1 (Dν g)(t)dt 0

0

1

(1 − t)ν−1 (Dxν0 f )(w)dt

(x − x0 )ν 1 (x − w)ν−1 dw (Dxν0 f )(w) ν−1 Γ(ν) (x − x ) (x − x0 ) 0 x 0 1 (x − w)ν−1 (Dxν0 f )(w)dw Γ(ν) x0 (Jνx0 Dxν0 f )(x).

That is, (Jν Dν g)(1) = (Jνx0 Dxν0 f )(x),

(2.10)

so that (2.7) becomes clear. Also (2.8) is obvious by (2.4) and (2.10).

Note. (1) (Dxn0 f ) = f (n) , n ∈ N. (2) Let f ∈ Cxν0 ([a, b]), ν ≥ 1 and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Then by (2.7), f (x) = (Jνx0 Dxν0 f )(x). That is, 1 f (x) = Γ(ν)

x

x0

(x − t)ν−1 (Dxν0 f )(t)dt,

all x ∈ [a, b]: x ≥ x0 . Notice that (2.11) is true also when 0 < ν < 1.

(2.11)

2.2 Main Results

11

2.2 Main Results We give our ﬁrst result on Opial-type inequalities for fractional derivatives. Theorem 2.2. Let f ∈ Cxν0 ([a, b]), ν ≥ 1 and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b]: x ≥ x0 . Let p, q > 1 such that 1/p + 1/q = 1. Then

x

x0

|f (w)| |(Dxν0 f )(w)|dw ≤

2−1/q (x − x0 )(pν−p+2)/p Γ(ν)((pν − p + 1)(pν − p + 2))1/p 2/q

x

· x0

|(Dxν0 f )(w)|q dw

.

(2.12)

Proof. From (2.11) and H¨ older’s inequality we get (x ≥ x0 ) |f (x)|

1 Γ(ν)

≤

1 Γ(ν)

≤

x

x0

(x − t)ν−1 |(Dxν0 f )(t)|dt 1/p

x

((x − t)

) dt

x0

(x − x0 ) · Γ(ν)(pν − p + 1)1/p (pν−p+1)/p

= Set

x

ν−1 p

x0

(|Dxν0 f |(t))q dt

x

x0

1/q (2.13)

1/q

(|Dxν0 f |(t))q dt

.

x

z(x) := x0

(|Dxν0 f |(t))q dt,

(z(x0 ) = 0).

Then z (x) = (|Dxν0 f |(x))q and |Dxν0 f |(x) = (z (x))1/q ,

all x0 ≤ x ≤ b.

Therefore from (2.13) we have |f (w)| |Dxν0 f |(w) ≤

w

· x0

|Dxν0 f |(t))q dt

(w − x0 )(pν−p+1)/p Γ(ν)(pν − p + 1)1/p

1/q · z (w) , all x0 ≤ w ≤ x.

(2.14)

12

2. Opial-Type Inequalities

Next we integrate (2.14) over [x0 , x] x |f (w)| |Dxν0 f |(w)dw x0 x 1 (w − x0 )pν−p+1/p · (z(w)z (w))1/q dw ≤ Γ(ν)(pν − p + 1)1/p x0 1/p x 1 pν−p+1 ≤ (w − x ) dw 0 Γ(ν)(pν − p + 1)1/p x0 x 1/q · z(w)z (w)dw x0

(x − x0 )(pν−p+2)/p (z(x))2/q · Γ(ν)(pν − p + 1)1/p (pν − p + 2)1/p 21/q 2/q x 2−1/q (x − x0 )(pν−p+2)/p ν q = · |D f (w)| dw , x0 Γ(ν)((pν − p + 1)(pν − p + 2))1/p x0 =

Thus establishing (2.12).

The counterpart of the previous result follows. Theorem 2.3. Let f ∈ Cxν0 ([a, b]), ν > 0, and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν] when ν ≥ 1. Here x, x0 ∈ [a, b]: x > x0 . Assume that (Dxν0 f ) = 0 over [x0 , b] and is of ﬁxed sign. Let 0 < p < 1 and q : 1/p + 1/q = 1. Then x 2−1/q (x − x0 )(pν−p+2)/p |f (w)| |(Dxν0 f )(w)|dw ≥ Γ(ν)((pν − p + 1)(pν − p + 2))1/p x0 x 2/q ν q · |(Dx0 f )(w)| dw . (2.15) x0

Proof. From (2.7), (2.8), and (2.11) we have w 1 |f (w)| = (w − t)ν−1 |(Dxν0 f )(t)|dt, all x0 ≤ w ≤ x. Γ(ν) x0 Here 0 < p < 1 and q = p/p − 1 < 0. By H¨ older’s inequality we get 1/p w 1/q w 1 |f (w)| ≥ (w − t)p(ν−1) dt · |(Dxν0 f )(t)|q dt Γ(ν) x0 x0 1/q w 1 (w − x0 )pν−p+1/p ν q = |(Dx0 f )(t)| dt , (2.16) Γ(ν) (pν − p + 1)1/p x0 all w > x0 . Set w z(w) := (|Dxν0 f |(t))q dt, (z(x0 ) = 0), x0

2.2 Main Results

13

so that z (w) = (|Dxν0 f |(w))q and |(Dxν0 f )(w)| = (z (w))1/q ,

all x0 ≤ w ≤ x.

Then by (2.16) we obtain |f (w)| |(Dxν0 f )(w)| ≥

1 (w − x0 )(pν−p+1)/p · (z(w)z (w))1/q , Γ(ν) (pν − p + 1)1/p

all x0 < w ≤ x. Let x0 < θ ≤ w ≤ x and θ ↓ x0 ; then by integration of the last inequality we ﬁnd x x |f (w)| |(Dxν0 f )(w)|dw = lim |f (w)| |(Dxν0 f )(w)|dw θ↓x 0 x0 θ x 1 · lim (w − x0 )(pν−p+1)/p (z(w)z (w))1/q dw ≥ Γ(ν)(pν − p + 1)1/p θ↓x0 θ (here z(w)z (w) > 0) 1/p x 1 pν−p+1 · lim (w − x0 ) dw ≥ Γ(ν)(pν − p + 1)1/p θ↓x0 θ 1/q x · lim z(w)z (w)dw θ↓x0

θ

=

(z 2 (x) − z 2 (θ))1/q (x − x0 )pν−p+2/p · lim Γ(ν)(pν − p + 1)1/p (pν − p + 2)1/p θ↓x0 21/q

=

2−1/q (x − x0 )pν−p+2/p (z(x))2/q . Γ(ν)((pν − p + 1)(pν − p + 2))1/p

That establishes (2.15).

An extreme case follows. Proposition 2.4. Here p = 1, q = ∞. Let f ∈ Cxν0 ([a, b]), ν > 0, and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν] when ν ≥ 1. Here x, x0 ∈ [a, b]: x ≥ x0 . Then x (x − x0 )ν+1 Dxν0 f 2∞ . (2.17) |f (w)| |Dxν0 f |(w)dw ≤ Γ(ν + 2) x0 Proof. From (2.7), (2.8), and (2.11) we have for all x0 ≤ w ≤ x, w 1 |f (w)| ≤ (w − t)ν−1 |(Dxν0 f )(t)|dt Γ(ν) x0 w 1 ν−1 ≤ (w − t) dt Dxν0 f ∞,[x0 ,w] Γ(ν) x0 Dxν0 f ∞,[x0 ,w] (w − x0 )ν · . = Γ(ν) ν

14

2. Opial-Type Inequalities

That is, |f (w)| ≤ that is, |f (w)| ≤

Dxν0 f ∞,[x0 ,w] (w − x0 )ν ; Γ(ν + 1)

Dxν0 f ∞ (w − x0 )ν , Γ(ν + 1)

Thus |f (w)| |(Dxν0 f )(w)| ≤ and

Dxν0 f ∞ (w − x0 )ν |(Dxν0 f )(w)|, Γ(ν + 1) Dνx0 f ∞ Γ(ν + 1)

x

x0

|f (w)| |(Dxν0 f )(w)|dw

≤

x

x0

x

(w − x0 )ν |(Dxν0 f )(w)|dw

x0 2 x

(Dxν0 f ∞ ) Γ(ν + 1)

≤ Therefore

all x0 ≤ w ≤ x.

|f (w)| |(Dxν0 f )(w)|dw ≤

(w − x0 ) dw . ν

x0

(x − x0 )ν+1 (Dxν0 f ∞ )2 . ν+1 Γ(ν + 1)

Remark 2.5. Let g ∈ C([x0 , b]); then one can easily see that x0 (Jνx0 g)() (x) = (Jν− g)(x), all x ∈ [x0 , b],

∈ {1, . . . , n − 1}, n := [ν], ν ≥ 1. For f ∈ Cxν0 ([a, b]), we therefore have x0 Dxν0 f )(x), (Jνx0 Dxν0 f )() (x) = (Jν−

all x ∈ [x0 , b], ∈ {1, . . . , n − 1}. Assuming that f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, by (2.7) we obtain f (x) = (Jνx0 Dxν0 f )(x),

x ≥ x0 ,

and we get x0 Dxν0 f )(x), f () (x) = (Jν−

x ≥ x0

and ∈ {1, . . . , n − 1}. That is, f

()

1 (x) = Γ(ν − )

x

x0

(x − t)ν−−1 (Dxν0 f )(t)dt,

all x ∈ [x0 , b], ∈ {1, . . . , n − 1}, n := [ν], ν ≥ 1.

(2.18)

2.2 Main Results

15

We present Theorem 2.6. Let f ∈ Cxν0 ([a, b]), ν ≥ 1 and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b]: x ≥ x0 , and = 1, . . . , n − 1. Let p, q > 1 such that 1/p + 1/q = 1. Then

x

|f () (w)| |(Dxν0 f )(w)|dw ≤

x0

2−1/q (x − x0 )(νp−p−p+2)/p Γ(ν − ) · ((νp − p − p + 1)(νp − p − p + 2))1/p

2/q

x

· x0

|(Dxν0 f )(w)|q dw

.

(2.19)

Proof. From (2.18) and H¨ older’s inequality we get (x ≥ x0 ) |f () (x)| ≤

≤

1 Γ(ν − )

1 Γ(ν − )

x

x0

(x − t)ν−−1 |(Dxν0 f )(t)|dt

x

((x − t)ν−−1 )p dt

1/p ·

x0

=

1/q

x

x0

|(Dxν0 f )(t)|q dt

(x − x0 )(νp−p−p+1)/p · (z(x))1/q , Γ(ν − )(νp − p − p + 1)1/p

where

(2.20)

x

z(x) := x0

|(Dxν0 f )(t)|q dt,

(z(x0 ) = 0).

Here z (x) = |(Dxν0 f )(x)|q , and |(Dxν0 f )(x)| = (z (x))1/q ,

x ∈ [x0 , b].

Hence from (2.20) we have |f () (w)| |(Dxν0 f )(w)| ≤ all x0 ≤ w ≤ x.

(w − x0 )(νp−p−p+1)/p · (z(w)z (w))1/q , Γ(ν − )(νp − p − p + 1)1/p

16

2. Opial-Type Inequalities

Integrating the last inequality over [x0 , x] we get

x

x0

|f () (w)| |(Dxν0 f )(w)|dw 1 Γ(ν − )(νp − p − p + 1)1/p

≤

x

(w − x0 )(νp−p−p+1)/p x0

1 Γ(ν − )(νp − p − p + 1)1/p 1/p x 1/q x (w − x0 )νp−p−p+1 dw z(w)z (w)dw ·(z(w)z (w))1/q dw ≤

x0

x0

(x − x0 )(νp−p−p+2)/p (z(x))2/q = · , Γ(ν − )(νp − p − p + 1)1/p (νp − p − p + 2)1/p 21/q proving (2.19).

Next we present the counterpart of the last result. Theorem 2.7. Let f ∈ Cxν0 ([a, b]), ν ≥ 1 and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b]: x > x0 , and = 1, . . . , n − 1. Assume that (Dxν0 f ) = 0 over [x0 , b] and of ﬁxed sign. Let 0 < p < 1 and q : 1/p + 1/q = 1. Then

x

|f () (w)| |(Dxν0 f )(w)|dw ≥

x0

2−1/q (x − x0 )(νp−p−p+2)/p Γ(ν − )((νp − p − p + 1)(νp − p − p + 2))1/p

2/q

x

· x0

|Dxν0 f (w)|q dw

.

(2.21)

Proof. From (2.18) and the assumption of the theorem we have |f () (x)| =

1 Γ(ν − )

x

x0

(x − t)ν−−1 |Dxν0 f (t)|dt.

Here q = p/p − 1 < 0. By H¨ older’s inequality (0 x0 .

2.2 Main Results

Call

x

z(x) := x0

then

17

(|Dxν0 f |(t))q dt,

z(x0 ) = 0;

z (x) = |(Dxν0 f )(x)|q

with

|(Dxν0 f )(x)| = (z (x))1/q ,

all x ≥ x0 .

Therefore |f () (w)| |(Dxν0 f )(w)| ≥

(w − x0 )(νp−p−p+1)/p · (z(w)z (w))1/q , Γ(ν − )(νp − p − p + 1)1/p

all x0 < w ≤ x. Consequently x |f () (w)| |(Dxν0 f )(w)|dw = lim θ↓x0

x0

θ

x

|f () (w)| |(Dxν0 f )(w)|dw

1 ≥ · lim Γ(ν − )(νp − p − p + 1)1/p θ↓x0 1/q x · lim z(w)z (w)dw θ↓x0

=

1/p

x

(w − x0 )

νp−p−p+1

dw

θ

θ

(x − x0 )(νp−p−p+2)/p (z(x))2/q · . Γ(ν − )(νp − p − p + 1)1/p (νp − p − p + 2)1/p 21/q

That proves (2.21).

An extreme case follows. Proposition 2.8. Here p = 1, q = ∞. Let f ∈ Cxν0 ([a, b]) and f (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν], ν ≥ 1. Here x, x0 ∈ [a, b]: x ≥ x0 and = 1, . . . , n − 1. Then (i)

x

x0

|f () (w)| |(Dxν0 f )(w)|dw ≤

(x − x0 )ν−+1 (Dxν0 f ∞ )2 . Γ(ν − + 2)

Proof. From (2.18) we get |f

()

(w)|

≤ =

1 Γ(ν − )

w

(w − t)

ν−−1

x0

dt Dxν0 f ∞

(w − x0 )ν− 1 Dxν0 f ∞ . Γ(ν − ) ν−

(2.22)

18

2. Opial-Type Inequalities

That is, |f () (w)| ≤

(w − x0 )ν− Dν f ∞ , Γ(ν − + 1) x0

all x0 ≤ w ≤ x. Furthermore we see that |f () (w)| |(Dxν0 f )(w)| ≤ Finally it holds x |f () (w)| |(Dxν0 f )(w)|dw

x

≤

x0

(w − x0 )

ν−

dw

x0

= proving (2.22).

(w − x0 )ν− (Dxν0 f ∞ )2 . Γ(ν − + 1)

(Dxν0 f ∞ )2 Γ(ν − + 1)

(x − x0 )ν−+1 (Dxν0 f ∞ )2 , (ν − + 1) Γ(ν − + 1)

2.3 Applications (i) Uniqueness of solution to (see also [3]) (Daν f )(t) = F (t, f, f , . . . , f (n−1) ) : f (i) (a) = ai , 0 ≤ i ≤ n − 1, n := [ν], ν ≥ 1.

(2.23)

Here f ∈ Caν ([a, b]); F is a continuous function on [a, b] × Rn , t ∈ [a, b]. Also F fulﬁlls the Lipschitz condition: |F (t, z0 , z1 , . . . , zn−1 ) − F (t, z0 , z1 , . . . , zn−1 )| ≤

n−1

qi (t)|zi − zi |,

i=0

where qi (t) ≥ 0, 0 ≤ i ≤ n − 1 are continuous functions on [a, b]. Call n−1 (b − a)ν−i . qi ∞ φ(b) := 2Γ(ν − i) (ν − i)(2ν − 2i − 1) i=0 Assume that φ(b) < 1.

(2.24)

Next we prove uniqueness to the solution of (2.23). Let f1 , f2 ∈ Caν ([a, b]) fulﬁlling (2.23); that is, (Daν fj )(t) = F (t, fj , fj , . . . , fj

(n−1)

(i)

) : fj (a) = ai , 0 ≤ i ≤ n − 1; j = 1, 2.

2.3 Applications

Call g := f1 − f2 . Then (n−1) (n−1) ) − F (t, f2 , f2 , . . . , f2 ), (Daν g)(t) = F (t, f1 , f1 , . . . , f1 (i) such that g (a) = 0, 0 ≤ i ≤ n − 1.

19

(2.25)

Here |F (t, f1 , f1 , . . . , f1

(n−1)

≤

n−1

) − F (t, f2 , f2 , . . . , f2

(n−1)

(i)

(i)

qi (t)|f1 (t) − f2 (t)| =

i=0

n−1

)|

qi (t)|g (i) (t)|.

i=0

From (Daν g(t))2 = (Daν g(t)){F (t, f1 , f1 , . . . , f1

(n−1)

we obtain (Daν g(t))2

≤

|Daν g(t)|

) − F (t, f2 , f2 , . . . , f2

n−1

(n−1)

)},

(i)

qi (t)|g (t)|

i=0

=

n−1

qi (t)(|Daν g(t)| |g (i) (t)|).

i=0

Integrating the last inequality we get

b

(Daν g(t))2 dt ≤ a

n−1 b a

i=0

≤

n−1 i=0

qi ∞

qi (t)|g (i) (t)| |Daν g(t)|dt

b

|g (i) (t)| |Daν g(t)|dt.

(2.26)

a

From inequalities (2.12), (2.19), when p = q = 2 and x0 = a, we ﬁnd that b |g (i) (t)| |Daν g(t)|dt a

(b − a)ν−i ≤ 2Γ(ν − i) (2ν − 2i − 1)(ν − i)

b

((Daν g)(t))2 dt,

for i = 0, 1, . . . , n − 1. Combining (2.26) and (2.27) we derive b b ν 2 ((Da g)(t)) dt ≤ φ(b) ((Daν g(t))2 dt. a

(2.27)

a

(2.28)

a

b If a (Daν g(t))2 dt = 0, then from (2.28) we get φ(b) ≥ 1, a contradiction by (2.24).

20

2. Opial-Type Inequalities

Thus

b

(Daν g(t))2 dt = 0, a

and (Daν g(t))2 = 0, a.e. in t ∈ [a, b]; that is, Daν g(t) = 0, a.e. in t ∈ [a, b]. Furthermore (Jνa Daν g)(t) ≡ 0, over [a, b]. But g (i) (a) = 0, 0 ≤ i ≤ n − 1 by (2.25). Finally from (2.7), Theorem 2.1 (fractional Taylor’s formula) we ﬁnd that g(t) ≡ 0 on [a, b]. That is, f1 = f2 on [a, b], proving the uniqueness to the solution of initial value problem (2.23). (ii) Upper bounds on Daν f , solution f , and so on (see also [3]). Let the initial value problem ⎧ (Daν f ) (t) = F (t, f, f , f , . . . , f (n−1) , (Daν f )), a ≤ t ≤ b ⎪ ⎪ ⎨ a≤t≤b (2.29) f (i) (a) = 0, 0 ≤ i ≤ n − 1, n := [ν], ν ≥ 1 ⎪ ⎪ ⎩ ν and Da f (a) = A ∈ R. Here F is continuous on [a, b] × Rn+1 and |F (t, x0 , x1 , . . . , xn )| ≤

n−1

qi (t)|xi |,

i=0

qi (t) ≥ 0, 0 ≤ i ≤ n − 1 continuous on [a, b], also f ∈ Caν ([a, b]). See that (Daν f )(t)(Daν f ) (t) = (Daν f )(t)F (t, f, f , . . . , f (n−1) , (Daν f )), and (a ≤ x ≤ b) x

(Daν f )(t)(Daν f ) (t)dt a x = (Daν f )(t)F (t, f, f , . . . , f (n−1) , (Daν f ))dt a x |(Daν f )(t)| |F (t, f, f , . . . , f (n−1) , (Daν f ))|dt ≤ a n−1 x ≤ |(Daν f )(t)| qi (t)|f (i) (t)| dt a

=

i=0 x n−1

a

≤

n−1 i=0

qi (t)|f (i) (t)| |(Daν f )(t)|dt

i=0

qi ∞

x

|f (i) (t)| |(Daν f )(t)|dt. a

2.3 Applications

21

That is, x n−1 x ((Daν f )(t))2 qi ∞ · |f (i) (t)| |(Daν f )(t)|dt. ≤ 2 a a i=0 That is, ((Daν f )(x))2

≤A +2 2

n−1

x

qi ∞

|f (i) (t)| |(Daν f )(t)|dt.

(2.30)

a

i=0

From inequalities (2.12), (2.19), when p = q = 2 and x0 = a, we get that x (x − a)ν−i |f (i) (t)| |(Daν f )(t)|dt ≤ 2Γ(ν − i) (ν − i)(2ν − 2i − 1) a x ((Daν f )(t))2 dt, (2.31) · a

for all i = 0, 1, . . . , n − 1. Call

and Q(x) :=

θ(x) := ((Daν f )(x))2 ,

(2.32)

ρ := A2 ,

(2.33)

n−1

(x − a)ν−i qi ∞ Γ(ν − i) (ν − i)(2ν − 2i − 1) i=0

,

all a ≤ x ≤ b. From (2.30) and using (2.31) through (2.34) we obtain x θ(x) ≤ ρ + Q(x) θ(t)dt,

(2.34)

(2.35)

a

all a ≤ x ≤ b. Here ρ, Q(x), θ(x) ≥ 0 and Q(a) = 0. Set x θ(t)dt, w(a) = 0; w(x) :=

(2.36)

a

that is, w (x) = θ(x). Therefore (2.35) becomes w (x) ≤ ρ + Q(x)w(x), all a ≤ x ≤ b; that is, w (t) ≤ ρ + Q(t)w(t),

all a ≤ t ≤ x.

(2.37)

Next we see that (see also [76]) w (t)e− and

t a

Q(s)ds

≤ ρe−

w(t)e−

t a

t a

Q(s)ds

Q(s)ds

+ e−

≤ ρe−

t a

t a

Q(s)ds

Q(s)ds

,

Q(t)w(t), (2.38)

22

2. Opial-Type Inequalities

all a ≤ t ≤ x. Integrating (2.38) over [a, x] we ﬁnd x w(x)e− a Q(s)ds ≤ ρ

x

e−

t a

Q(s)ds

dt,

a

and

x

w(x) ≤ ρ e

a

Q(s)ds

x

−

e

t a

Q(s)ds

dt ,

(2.39)

a

all a ≤ x ≤ b. Using (2.39) into (2.35) we obtain

x

θ(x) ≤ ρ + Q(x)ρ e

a

Q(s)ds

x

−

e

t a

Q(s)ds

dt .

a

So we have proved that x |(Daν f )(x)| ≤ |A| 1 + Q(x) e a Q(s)ds ·

x

−

e

t a

Q(s)ds

1/2

dt

=: γ(x),

(2.40)

a

all a ≤ x ≤ b. Given f (i) (a) = 0, i = 0, 1, . . . , n − 1; ν ≥ 1 from (2.11) and (2.18) we have that x 1 (x − t)ν−i−1 ((Daν f )(t))dt. (2.41) f (i) (x) = Γ(ν − i) a Using (2.40) into (2.41) we derive the upper bounds x 1 (i) |f (x)| ≤ (x − t)ν−i−1 γ(t)dt, Γ(ν − i) a all i = 0, 1, . . . , n − 1, for all a ≤ x ≤ b.

(2.42)

3 Canavati Fractional Opial-Type Inequalities and Fractional Diﬀerential Equations

A series of very general Opial-type inequalities [315] is presented involving fractional derivatives of diﬀerent orders. These are based on Taylor’s formula for fractional derivatives. The results are applied in proving uniqueness to the solutions of very general fractional initial value problems of fractional ordinary diﬀerential equations. This treatment is based on [61].

3.1 Introduction The Canavati fractional Opial inequalities that we deal with have the general form b |Dν 1 f (w)|α |Dν 2 f (w)|β g(w) dw a

ζ

b

≤ K

|D f (w)| |D f (w)| p(w) dw ν1

δ

ν2

ε

a

for certain continuous functions f on [a, b]. Here q, p are weight functions and Dγ f denotes the Canavati fractional derivative of f of order γ. These results are presented in Section 3.3, after Section 3.2 (Preliminaries). In Section 3.4, these results are applied to obtain uniqueness criteria for initial value problems for fractional diﬀerential equations of the form Dν f +

k

qj Dγ j f + qk+1 f = h.

j=1

G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 23 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 3,

24

3. Canavati Fractional Opial Inequalities

Other sorts of fractional diﬀerential equations are studied in the literature. Section 3.5 contains remarks on this and attempts to put the results of Section 3.4 in some perspective.

3.2 Preliminaries In the sequel we follow [101]. Let g ∈ C([0, 1]). Let ν be a positive number, n := [ν], and α := ν − n (0 < α < 1). Deﬁne x 1 (x − t)ν−1 g(t) dt, 0 ≤ x ≤ 1, (3.1) (Jν g)(x) := Γ(ν) 0 the Riemann–Liouville integral, where Γ is the gamma function. We deﬁne the subspace C ν ([0, 1]) of C n ([0, 1]) as follows. C ν ([0, 1]) := {g ∈ C n ([0, 1]) : J1−α Dn g ∈ C 1 ([0, 1])}, where D := d/dx. So for g ∈ C ν ([0, 1]), we deﬁne the Canavati ν-fractional derivative of g as Dν g := DJ1−α Dn g. (3.2) When ν ≥ 1 we have the fractional Taylor’s formula t2 tn−1 + · · · + g (n−1) (0) 2! (n − 1)! for all t ∈ [0, 1].

g(t) = g(0) + g (0)t + g (0) + (Jν Dν g)(t),

(3.3)

When 0 < ν < 1 we ﬁnd g(t) = (Jν Dν g)(t),

for all t ∈ [0, 1].

(3.4)

Next we carry the above notions over to arbitrary [a, b] ⊆ R (see [17]). Let x, x0 ∈ [a, b] such that x ≥ x0 , where x0 is ﬁxed. Let f ∈ C([a, b]) and deﬁne x 1 x0 (Jν f )(x) := (x − t)ν−1 f (t) dt, x0 ≤ x ≤ b, (3.5) Γ(ν) x0 the generalized Riemann–Liouville integral. We deﬁne the subspace Cxν0 ([a, b]) of C n ([a, b]): x0 Dn f ∈ C 1 ([x0 , b])}. Cxν0 ([a, b]) := {f ∈ C n ([a, b]) : J1−α

For f ∈ Cxν0 ([a, b]), we deﬁne the generalized ν-fractional derivative of f over [x0 , b] as x0 Dxν0 f := DJ1−α f (n) (f (n) := Dn f ). (3.6)

3.2 Preliminaries

25

Notice that x0 (J1−α f (n) )(x)

1 = Γ(1 − α)

x

(x − t)−α f (n) (t) dt

x0

exists for f ∈ Cxν0 ([a, b]). We recall the following fractional generalization of Taylor’s formula (see [17, 101]). Theorem 3.1. Let f ∈ Cxν0 ([a, b]), x0 ∈ [a, b] ﬁxed. (i) If ν ≥ 1 then f (x) = f (x0 ) + f (x0 )(x − x0 ) + f (x0 ) f (n−1) (x0 )

(x − x0 )2 + ···+ 2

(x − x0 )n−1 + (Jνx0 Dxν0 f )(x), for all x ∈ [a, b] : x ≥ x0 . (3.7) (n − 1)!

(ii) If 0 < ν < 1 then f (x) = (Jνx0 Dxν0 f )(x),

for all x ∈ [a, b] : x ≥ x0 .

(3.8)

We make Remark 3.2. (1) (Dxn0 f ) = f (n) ; n ∈ N. (2) Let f ∈ Cxν0 ([a, b]), ν ≥ 1 and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Then by (3.7) f (x) = (Jνx0 Dxν0 f )(x). That is, f (x) =

1 Γ(ν)

x

x0

(x − t)ν−1 (Dxν0 f )(t) dt,

(3.9)

for all x ∈ [a, b] with x ≥ x0 . Notice that (3.9) is true, also when 0 < ν < 1. We need the following lemma from [17]. Lemma 3.3. Let f ∈ C([a, b]), μ, ν > 0. Then x0 Jμx0 (Jνx0 f ) = Jμ+ν (f ).

We also make

(3.10)

26

3. Canavati Fractional Opial Inequalities

Remark 3.4. Let ν ≥ 1, γ ≥ 0, be such that ν − γ ≥ 1, so that γ < ν. Call n := [ν], α := ν − n ; m := [γ], ρ := γ − m. Note that ν − m ≥ 1 and n − m ≥ 1. Let f ∈ Cxν0 ([a, b]) be such that f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Hence by (3.7) f (x) = (Jνx0 Dxν0 f )(x),

for all x ∈ [a, b] : x ≥ x0 .

Therefore by Leibnitz’s formula and Γ(p + 1) = pΓ(p), p > 0, we ﬁnd that for all x ≥ x0 .

x0 f (m) (x) = (Jν−m Dxν0 f )(x),

(3.11)

It follows that f ∈ Cxγ0 ([a, b]) and thus x0 (Dxγ0 f )(x) := (DJ1−ρ f (m) )(x)

exists for all x ≥ x0 .

(3.12)

By the use of (3.11) we have on [x0 , b], x0 x0 x0 x0 x0 (f (m) ) = J1−ρ (Jν−m Dxν0 f ) = (J1−ρ ◦ Jν−m )(Dxν0 f ) J1−ρ x0 x0 ν ν = (Jν−m+1−ρ (Dx0 f ) = Jν−γ+1 (Dx0 f ).

by (3.10). That is, x0 (J1−ρ f (m) )(x)

1 = Γ(ν − γ + 1)

x

x0

(x − t)ν−γ (Dxν0 f )(t) dt.

Therefore x0 f (m) )(x)) = (Dxγ0 f )(x) = D((J1−ρ

1 · Γ(ν − γ)

x

x0

(x−t)(ν−γ)−1 (Dxν0 f )(t) dt; (3.13)

hence x0 (Dxν0 f ))(x) (Dxγ0 f )(x) = (Jν−γ

and is continuous in x on [x0 , b]. (3.14)

3.3 Main Results The next theorem is a specialization of Theorem 2 from [15]. It presents a fractional Opial-type inequality. Theorem 3.5. Let γ 1 , γ 2 ≥ 0, ν ≥ 1 be such that ν − γ 1 , ν − γ 2 ≥ 1 and f ∈ Cxν0 ([a, b]) with f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b] with x ≥ x0 . Let q be a nonnegative continuous function on [a, b]. Denote 1/2 x 2 (q(w)) dw . (3.15) Q := x0

3.3 Main Results

Then

27

x

q(w)|Dxγ01 (f )(w)| |Dxγ02 (f )(w)|dw x0 x ν 2 ≤ K(q, γ 1 , γ 2 , ν, x, x0 ) · (Dx0 f (w)) dw ,

(3.16)

x0

where

K(q, γ 1 , γ 2 , ν, x, x0 ) :=

Q √ 3 6

1 · Γ(ν − γ 1 ) · Γ(ν − γ 2 )

·

(x − x0 )2ν−γ 1 −γ 2 −1/2 (ν − γ 1 − 56 )1/6 · (ν − γ 2 − 56 )1/6 · (4ν − 2γ 1 − 2γ 2 − 73 )1/2

(3.17)

Proof. From (3.13) we have, for x0 ≤ w ≤ x, j = 1, 2, w 1 γ (w − t)ν−γ j −1 (Dxν0 f )(t)dt (Dx0j f )(w) = Γ(ν − γ j ) x0 x 1 ν−γ −1 (w − t)+ j (Dxν0 f )(t)dt. = Γ(ν − γ j ) x0

(3.18)

Hence

x

q(w) · x0

2

⎛ γ |(Dx0j (f ))(w)|dw

≤ Q⎝

x

2

⎞1/2 γ ((Dx0j (f ))(w))2 dw⎠

x0 j=1

j=1

by (3.18), ≤Q

⎧ ⎨ ⎩

x

2

x0 j=1

1 Γ(ν − γ j )

x

≤Q·

x

· x0

|(Dxν0 f )(t)|3/2 (w

(w − x0

2

x0 j=1

−

x

ν−γ −1 t)+ j |(Dxν0 f )(t)|dt

1 · (x − x0 )2/3 (Γ(ν − γ j ))2

(ν−γ −1)3/2 t)+ j dt

4/3

12 dw

⎫1/2 ⎬

2 dw

⎭

= Q · (x − x0 )2/3 ·

⎧ ⎫ 4/3 ⎬1/2 2 x ⎨ x 1 (ν−γ −1)3/2 · |Dxν0 f (t)|3/2 (w − t)+ j dt dw ⎩ x0 ⎭ Γ(ν − γ ) x j 0 j=1 j=1 2

≤ Q · (x − x0 )2/3 ·

2

1 Γ(ν − γj ) j=1

28

·

3. Canavati Fractional Opial Inequalities

⎧ ⎨ ⎩

x

⎫1/2 x 1/3 ⎬ (ν−γ −1)·6 ((Dxν0 f )(t))2 dt · (w − t)+ j dt · dw ⎭ x0 x0

2

x0 j=1

x

⎞ x 1 2/3 ⎝ ν 2 ⎠ · = Q · (x − x0 ) · ((Dx0 f )(t)) dt Γ(ν − γ j ) x0 j=1 ⎛

·

⎧ ⎨ ⎩

x

2

x0 j=1

2

x

(w −

(ν−γ −1)·6 t)+ j dt

1/3 · dw

x0

⎫1/2 ⎬ ⎭

⎞ x 1 2/3 ⎝ ν 2 ⎠ · = Q · (x − x0 ) · ((Dx0 f )(w)) dw Γ(ν − γ j ) x0 j=1 ⎛

·

⎧ ⎨ ⎩

x

x0

2

⎫1/2 2 (w − x0 )((ν−γ j −1)·6+1)/3 ⎬ dw ⎭ ((ν − γ j − 1)6 + 1)1/3 j=1

⎞ x 1 2/3 ⎝ ν 2 ⎠ · = Q · (x − x0 ) · ((Dx0 f )(w)) dw Γ(ν − γ j ) x0 j=1 ⎛

⎛

2

⎞ x 1/2 " 1 ( 2j=1 (ν−γ j −1)2)+2/3 ⎝ ⎠ · · (w − x0 ) · dw ((ν − γ j − 1)6 + 1)1/6 x0 j=1 2

(x − x0 )2ν−γ 1 −γ 2 −1/2 · =Q· Γ(ν − γ 1 )Γ(ν − γ 2 ) ·

(6ν − 6γ 1 −

5)1/6 (6ν

x

x0

((Dxν0 f )(w))2 dw

1 . − 6γ 2 − 5)1/6 (4ν − 2γ 1 − 2γ 2 − 73 )1/2

We have proved that

x

q(w) ·

x0

#

2

Q γ · |(Dx0j (f ))(w)|dw ≤ √ 3 6 j=1

(x − x0 )2ν−γ 1 −γ 2 −1/2 $ Γ(ν − γ 1 ) · Γ(ν − γ 2 ) · (ν − γ 1 − 56 )1/6 (ν − γ 2 − 56 )1/6 · (4ν − 2γ 1 − 2γ 2 − 73 )1/2 ·

x x0

We have established (3.16).

((Dxν0 f )(w))2 dw .

We next give the following very general fractional Opial-type inequality.

3.3 Main Results

29

Theorem 3.6. Let γ ≥ 0, ν ≥ 1, ν − γ ≥ 1, α, β > 0, r > α, r > 1; let p > 0, q ≥ 0 be continuous functions on [a, b]. Let f ∈ Cxν0 ([a, b]) with f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Let x, x0 ∈ [a, b] with x ≥ x0 . Then x q(w)|Dxγ0 f (w)|β |Dxν0 f (w)|α dw ≤ K(p, q, γ, ν, α, β, r, x, x0 ) x0

(α+β/r)

x

· x0

p(w)|Dxν0 f (w)|r dw

.

(3.19)

Here K(p, q, γ, ν, α, β, r, x, x0 ) :=

x

·

−α 1/(r−α)

((q(w)) · (p(w)) r

α α+β

)

α/r ·

1 (Γ(ν − γ))β

r−α/r

· (P1 (w))

(β(r−1)/r−α)

· dw

x0

(3.20) with

w

P1 (w) :=

(p(t))−1/r−1 · (w − t)(ν−γ−1)(r/(r−1)) · dt.

(3.21)

x0

Proof. Again from (3.13) we have w 1 (Dxγ0 f )(w) = (w − t)ν−γ−1 (Dxν0 f )(t)dt, Γ(ν − γ) x0 That is, |Dxγ0 f (w)|

for all x0 ≤ w ≤ x.

w 1 (w − t)ν−γ−1 |Dxν0 f (t)|dt Γ(ν − γ) x0 w 1 p(t)−1/r (w − t)ν−γ−1 p(t)1/r |(Dxν0 f (t)|dt = Γ(ν − γ) x0 w (r−1)/r 1 −1/r−1 (ν−γ−1)r/(r−1) ≤ · p(t) (w − t) dt Γ(ν − γ) x0 w 1/r ν r · p(t)|Dx0 f (t)| dt ≤

x0

1 (P1 (w))r−1/r Γ(ν − γ)

= Call

1/r

w

x0

p(t)|Dxν0 f (t)|r dt

.

w

z(w) := x0

p(t)|Dxν0 f (t)|r dt,

x0 ≤ w ≤ x,

(so z(x0 ) = 0).

30

3. Canavati Fractional Opial Inequalities

Then

z (w) = p(w)|Dxν0 f (w)|r .

For α > 0 we get |Dxν0 f (w)|α = p(w)−α/r · (z (w))α/r . For β > 0 we ﬁnd q(w)|Dxγ0 f (w)|β |Dxν0 f (w)|α 1 ≤ q(w) (P1 (w))β(r−1)/r (z(w))β/r (p(w))−α/r (z (w))α/r . (Γ(ν − γ))β Hence

x x0

q(w)|Dxγ0 f (w)|β |Dxν0 f (w)|α dw

x

q(w) (P1 (w))β(r−1)/r p(w)−α/r (z(w))β/r (z (w))α/r dw (Γ(ν − γ))β x (r−α)/r 1 r −α 1/r−α β(r−1)/(r−α) ≤ · ((q(w)) (p(w)) ) · (P (w)) · dw 1 (Γ(ν − γ))β x0 x α/r · (z(w))β/α z (w)dw = (∗).

≤

x0

x0

Set K0

:

(∗)

=

x (r−α)/r 1 r −α 1/r−α β(r−1)/(r−α) ((q(w)) (p(w)) ) · (P (w)) · dw 1 (Γ(ν − γ))β x0 x α/r K0 (z(w))β/α dz(w)

=

x0

α/r (z(w))(β/α)+1 x K0 · β +1 x0 α α/r (β+α)/α z(x) K0 · β+α

=

=

=

K0 ·

α

α α+β

α/r · (z(x))(β+α)/r .

Clearly,

K(p, q, γ, ν, α, β, r, x, x0 ) = K0

We have proved (3.19).

α α+β

α/r .

Corollary 3.7. Let γ ≥ 0, ν ≥ 1, ν − γ ≥ 1, let q be a nonnegative continuous function on [a, b]. Let f ∈ Cxν0 ([a, b]) with f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Let x, x0 ∈ [a, b] with x ≥ x0 . Then

3.3 Main Results

31

x

q(w)|Dxγ0 f (w)| |Dxν0 f (w)|dw x ν 2 ≤ K(q, γ, ν, x, x0 ) (Dx0 f (w)) dw ,

x0

(3.22)

x0

where K(q, γ, ν, x, x0 ) =

1 · Γ(ν − γ) 2(2ν − 2γ − 1) 1

·

1/2

x

(q(w))2 (w − x0 )2ν−2γ−1 dw

.

(3.23)

x0

Proof. Apply Theorem 3.6 with α = β = 1, r = 2, p = 1.

Next we give another basic fractional Opial-type inequality. Theorem 3.8. Let ν ≥ 1, α, β > 0, r > α, r > 1; let p > 0, q ≥ 0 be continuous functions on [a, b]. Let f ∈ Cxν0 ([a, b]) with f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Let x, x0 ∈ [a, b] with x ≥ x0 . Then x q(w)|f (w)|β |Dxν0 f (x)|α dw ≤ K ∗ (p, q, ν, α, β, r, x, x0 ) x0

((α+β)/r)

x

· x0

p(w)|Dxν0 f (w)|r dw

.

(3.24)

Here

∗

K (p, q, ν, α, β, r, x, x0 ) :=

x

α α+β

α/r ·

1 · (Γ(ν))β

((q(w))r · (p(w))−α )1/(r−α) · (P1∗ (w))β(r−1)/(r−α) · dw

(r−α)/r ,

x0

(3.25) with P1∗ (w) :=

w

(p(t))−1/r−1 · (w − t)(ν−1)(r/r−1) dt.

x0

Proof. From (3.9) we have w 1 f (w) = (w − t)ν−1 (Dxν0 f )(t)dt, Γ(ν) x0

x0 ≤ w ≤ x.

(3.26)

32

3. Canavati Fractional Opial Inequalities

That is, |f (w)|

w 1 (w − t)ν−1 |Dxν0 f (t)|dt Γ(ν) x0 w 1 p(t)−1/r (w − t)ν−1 p(t)1/r |Dxν0 f (t)|dt Γ(ν) x0

≤ =

(applying H¨ older’s inequality with indices r/(r−1),r)

1 Γ(ν)

≤

p(t)−1/r−1 (w − t)(ν−1)r/r−1 dt

(r−1)/r .

x0

1/r

w

p(t)|Dxν0 f (t)|r dt

x0

1 (P ∗ (w))r−1/r Γ(ν) 1

= Call

w

1/r

w

x0

p(t)|Dxν0 f (t)|r dt

.

w

z(w) := x0

Then

p(t)|Dxν0 f (t)|r dt,

x0 ≤ w ≤ x, (z(x0 ) = 0).

z (w) = p(w)|Dxν0 f (w)|r .

For α > 0 we obtain |Dxν0 f (w)|α = (p(w))−α/r (z (w))α/r . For β > 0 we have q(w)|f (w)|β |Dxν0 f (w)|α

≤ q(w) ·

1 · (P1∗ (w))β(r−1)/r (Γ(ν))β

· (z(w))β/r · (p(w))−α/r (z (w))α/r . Thus

x

q(w)|f (w)|β |Dxν0 f (w)|α dw x q(w) ≤ (P1∗ (w))β(r−1)/r (p(w))−α/r (z(w))β/r (z (w))α/r dw β x0 (Γ(ν)) (r−α)/r x 1 r −α 1/(r−α) ∗ β(r−1)/(r−α) ≤ ((q(w)) (p(w)) ) (P1 (w)) dw (Γ(ν))β x0 x α/r β/α · (z(w)) z (w)dw = (∗)

x0

x0

3.3 Main Results

33

call K0∗

:

1 = (Γ(ν))β

x

= K0∗

x

1/(r−α)

(q(w)) (p(w)) x0

(P1∗ (w))β(r−1)/r−α dw (∗)

−α

r

r−α/r

α/r (z(w))β/α dz(w)

x0

x α/r (z(w))(β/α)+1 = β x0 α +1 α/r z(x)β+α/α ∗ ∗ = K0 = K0 β+α K0∗

α

α α+β

Clearly K ∗ (p, q, ν, α, β, r, x, x0 ) = K0∗ We have established (3.24).

α/r

α α+β

(z(x))(α+β)/r .

α/r .

Corollary 3.9. Let ν ≥ 1, and let q ≥ 0 be a continuous function on [a, b]. Let f ∈ Cxν0 ([a, b]) with f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Let x, x0 ∈ [a, b] : x ≥ x0 . Then

x

x0

q(w)|f (w)| |Dxν0 f (w)|dw x ≤ K ∗ (q, ν, x, x0 ) · (Dxν0 f (w))2 dw ,

(3.27)

x0

where K ∗ (q, ν, x, x0 ) =

1 Γ(ν) 2(2ν − 1) x 1/2 2 2ν−1 · (q(w)) (w − x0 ) dw .

1

·

(3.28)

x0

Proof. Apply Theorem 3.8 for α = β = 1, r = 2, p = 1. Remark 3.10. Notice that K(q, γ 1 , γ 2 , ν, x, x0 ) (see (3.17)), K(q, γ, ν, x, x0 ) (see (3.23)), and K ∗ (q, ν, x, x0 ) (see (3.28)), all converge to zero as x → x0 .

34

3. Canavati Fractional Opial Inequalities

Remark 3.11. Let ν ≥ 1 and γ := ν − 1. Clearly: γ ≥ 0 and ν − γ = 1. Let f ∈ Cxν0 ([a, b]) such that f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Then by (3.13) we ﬁnd x f )(x) = (Dxν0 f )(t)dt, ((Dxν−1 f )(x0 ) = 0); (Dxν−1 0 0 x0

that is,

(Dxν−1 f ) (x) = (Dxν0 f )(x), 0

all x ≥ x0 .

3.4 Applications (i) We prove locally the uniqueness of the solution of the following fractional initial value problem (IVP). Theorem 3.12. Let ν ≥ 1, γ j ≥ 0; j = 1, 2, . . . , k ∈ N be such that all ν − γ j ≥ 1 for all j. Also we assume the ordering ν > γ1 > γ2 > γ3 > · · · > γk. We consider continuous functions q1 , q2 , . . . , qk , qk+1 , h from [a, b] into R. Let x, x0 ∈ [a, b] such that x ≥ x0 . Consider the following fractional diﬀerential equation (Dxν0 f )(w) +

k

γ

qj (w) · (Dx0j f )(w) + qk+1 (w) · f (w) = h(w),

(3.29)

j=1

for all x0 ≤ w ≤ x, such that f (i) (x0 ) = ai ∈ R, i = 0, 1, . . . , n − 1, where n := [ν]. Then there is at most one solution f to this problem. Proof. Let f1 , f2 ∈ Cxν0 ([x0 , x]) fulﬁll the initial value problem (3.29). Call g := f1 − f2 ∈ Cxν0 ([x0 , x]). Then by (3.22) we obtain (Dxν0 g)(w) +

k

γ

qj (w) · (Dx0j g)(w) + qk+1 (w) · g(w) = 0,

j=1

true for all x0 ≤ w ≤ x, such that g (i) (x0 ) = 0, Thus (Dxν0 g)(w) = −

k j=1

all i = 0, 1, . . . , n − 1. γ

qj (w)(Dx0j g)(w) − qk+1 (w) · g(w),

3.4 Applications

35

and k

((Dxν 0 g)(w))2 = −

γ

qj (w) · (Dx0j g)(w) · (Dxν 0 g)(w) − qk+1 (w) · g(w) · (Dxν 0 g)(w).

j=1

That is, ≤

((Dxν0 g)(w))2

k

γ

|qj (w)| · |Dx0j g(w)| · |Dxν0 g(w)|

j=1

+ |qk+1 (w)| · |g(w)| · |Dxν0 g(w)|, which is true for all x0 ≤ w ≤ x. Consequently we have x δ := ((Dxν0 g)(w))2 · dw x0

≤

k j=1

x

x0

γ

|qj (w)| · |Dx0j g(w)| · |Dxν0 g(w)| · dw

x

+ x0

|qk+1 (w)| · |g(w)| · |Dxν0 g(w)| · dw =: (∗).

From (3.22) and (3.27) we ﬁnd (∗)

≤

k

j=1

x

K(|qj |, γ j , ν, x, x0 ) · x0

(Dxν0 g(w))2 dw

+ K ∗ (|qk+1 |, ν, x, x0 ) · (Dxν0 g(w))2 dw x0 ⎧⎛ ⎫ ⎞ k ⎨ ⎬ ⎝ = K(|qj |, γ j , ν, x, x0 )⎠ + K ∗ (|qk+1 |, ν, x, x0 ) · δ. ⎩ ⎭ x

j=1

Set K :=

k

K(|qj |, γ j , ν, x, x0 ) + K ∗ (|qk+1 |, ν, x, x0 ).

j=1

By Remark 3.10, K → 0 as x → x0 . So we get δ ≤ K · δ; that is, δ(1 − K) ≤ 0;

36

3. Canavati Fractional Opial Inequalities

also notice that δ ≥ 0. So for ε > 0 suﬃciently small such that |x − x0 | ≤ ε and K ≤ 1, we obtain δ ≤ 0. That is, δ = 0 for x : |x − x0 | ≤ ε. That is, x ((Dxν0 g)(w))2 dw = 0 for x : |x − x0 | ≤ ε. x0

That means = 0 for w : |w − x0 | ≤ ε. Hence by (3.9) we obtain Dxν0 g(w)

for w : |w − x0 | ≤ ε.

g(w) = 0

We have proved the local uniqueness for the solution of IVP (3.29); more precisely we have for all w with |w − x0 | ≤ ε.

f1 (w) = f2 (w)

(ii) Let ν ≥ 2, γ 1 := ν − 1, γ j ≥ 0; j = 2, 3, . . . , k ∈ N such that ν − γ j ≥ 2; j = 2, 3, . . . , k. Also we assume the order ν > γ1 > γ2 > γ3 > · · · > γk. Let n := [ν]; then n − 1 = [ν − 1]. We consider again IVP (3.29). Now we prove the uniqueness of the solution to (3.29) over the whole interval [x0 , b]. Again let f1 , f2 ∈ Cxν0 ([x0 , b]) fulﬁll the IVP (3.29). Call g := f1 − f2 ∈ ([x0 , b]). Then again by (3.29) we obtain Cxν0 ([x0 , b]). Clearly g ∈ Cxν−1 0 (Dxν0 g)(w) +

k

γ

qj (w) · (Dx0j g)(w) + qk+1 (w) · g(w) = 0,

j=1

all x0 ≤ w ≤ b, g (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Hence (Dxν0 g)(w) · (Dxν−1 g)(w) 0

k

+

γ

qj (w) · (Dx0j g)(w) · (Dxν−1 g)(w) 0

j=1

+ qk+1 (w) · g(w) · (Dxν−1 g)(w) = 0. 0

Next we integrate over [x0 , x ˜] ⊆ [x0 , b], k x x ˜ ˜ γ ν ν−1 (Dx0 g)(w) · (Dx0 g)(w) · dw + qj (w) · (Dx0j g)(w) · (Dxν−1 g)(w)dw 0

x0

j=1

x0

x ˜

+ x0

qk+1 (w) · g(w) · (Dxν−1 g)(w)dw = 0. 0

By Remark 3.11 we get that ((Dxν−1 g)(˜ x))2 0

= −2

k j=1

x ˜

x0

γ

qj (w) · (Dx0j g)(w) · (Dxν−1 g)(w) · dw 0

x ˜

−2 x0

qk+1 (w) · g(w) · (Dxν−1 g)(w) · dw. 0

3.4 Applications

37

Therefore we have ≤ 2

g)(˜ x))2 ((Dxν−1 0

k j=1

x ˜

x0

γ

|qj (w)| · |(Dx0j g)(w)| · |(Dxν−1 g)(w)| · dw 0

x ˜

+2· x0

|qk+1 (w)| · |g(w)| · |(Dxν−1 g)(w)| · dw =: (∗). 0

Next we use Corollaries 3.7 and 3.9 in the above setting. We have k x ˜ ν−1 2 K(|qj |, γ j , ν − 1, x ˜, x0 ) · (Dx0 g(w)) dw (∗) ≤ 2 · j=1

+ 2 · K ∗ (|qk+1 |, ν − 1, x ˜, x0 ) ·

x0

x0

=

x ˜

(Dxν−1 g(w))2 dw 0

⎧⎛ ⎫ ⎞ k ⎨ ⎬ 2· ⎝ K(|qj |, γ j , ν − 1, x ˜, x0 )⎠ + K ∗ (|qk+1 |, ν − 1, x ˜, x0 ) ⎩ ⎭ j=1 x ˜

· x0

(Dxν−1 g(w))2 dw . 0

Set

⎧⎛ ⎫ ⎞ k ⎨ ⎬ T (˜ x) := 2· ⎝ K(|qj |, γ j , ν − 1, x ˜, x0 )⎠ + K ∗ (|qk+1 |, ν − 1, x ˜, x0 ) ≥ 0. ⎩ ⎭ j=1

We have established that

x ˜

g)(˜ x))2 ((Dxν−1 0

≤ T (˜ x)· x0

(Dxν−1 g(w))2 0

· dw , for all x ˜ : x0 ≤ x ˜ ≤ x ≤ b.

x) ≥ 0, a constant for each x ∈ [x0 , b]. Thus Clearly T (x) = maxx0 ≤˜x≤x T (˜ we have shown that x ˜

g)(˜ x))2 ≤ T (x) · ((Dxν−1 0

x0

((Dxν−1 g)(w))2 · dw , 0

all x0 ≤ x ˜ ≤ x.

By Gronwall’s inequality we ﬁnd ((Dxν−1 g)(˜ x))2 = 0, 0

x0 ≤ x ˜ ≤ x,

and (Dxν−1 g)(˜ x) = 0, 0

for all x ˜ with x0 ≤ x ˜ ≤ x.

38

3. Canavati Fractional Opial Inequalities

By (3.9) we now get that g(˜ x) = 0,

all x ˜ : x0 ≤ x ˜ ≤ x.

That is, f1 = f2 over [x0 , x], proving the uniqueness of the solution to IVP (3.29) over the whole interval [x0 , x]. (iii) Consider again IVP (3.29) as in (ii). Additionally assume that h ≡ 0 and ai = 0, all i = 0, 1, . . . , n − 1. Then the unique solution exists and it is the trivial solution zero.

3.5 Other Fractional Diﬀerential Equations In this section we review the basic theory of fractional diﬀerential equations based on Miller – Ross fractional derivatives. The exposition follows Podlubny [333]. Consider the initial value problem D

σn

u(t) +

n−1

pj (t)Dσn−j u(t) + pn (t)u(t) = h(t),

(3.30)

j=1

Dσk −1 u(t)t=0 = bk ,

k = 1, . . . , n,

(3.31)

where Dσk σ k −1

D

σk

:= := =

Dαk Dαk−1 · · · Dα1 , Dαk −1 Dαk−1 · · · Dα1 , k αj , k = 1, . . . , n, j=1

0

<

αj ≤ 1,

j = 1, . . . , n,

and u ∈ L1 [0, τ ]. Note that we have switched from x to τ , and we have replaced x0 by 0 (and we are calling the unknown function u rather than f ). For an example of a concrete partial diﬀerential equation coming from the applications consider (Dγ )2 u + 2aDγ u = Δu, "N where u = u(t, x) for t ≥ 0 and x ∈ RN , Δ = j=1 ∂ 2 /∂x2j is the spatial Laplacian, a is a positive constant and 0 < γ < 1; and Dγ is the usual fractional derivative of order γ with respect to time t. This fractional telegraph equation was introduced in [150]; see [151] for the experimental background. Let u ˆ(t, ξ) (for t ≥ 0, ξ ∈ RN ) be the spatial Fourier transform of u(t, x). Then u ˆ satisﬁes (Dγ )2 u ˆ + 2aDγ u ˆ = −|ξ|2 u ˆ,

3.5 Other Fractional Diﬀerential Equations

39

which is a family of fractional diﬀerential equations of the form (3.30), one for each ξ ∈ RN . This problem arises in the study of suspensions, coming from the ﬂuid dynamical modeling of certain blood ﬂow phenomena (see [150] and [151]). We return to the initial value problem (3.30), (3.31). According to Theorem 3.2 of [333], if each pj is in C[0, τ ], the problem (3.30), (3.31) has a unique solution in the class L1 [0, τ ]. Thus both existence and uniqueness hold for (3.30), (3.31). The number of required initial conditions is n, whereas the order of the equation is σ n , and σ n can be any number in the half-open interval (0, n]. For instance, we can have σ n = 5/2 and n = 4 or even n = 6. Thus the number of initial conditions required to specify the solution uniquely is not only a function of the order of the equation; it depends on the decomposition (involving Dσk ) which leads to the rigorous interpretation of the equation. In the case of our uniqueness criteria (see Section 3.4), ν is the order of the equation and n = [ν] is the number of speciﬁed initial conditions. (Thus if ν = 5/2, necessarily n = 2.)

4 Riemann–Liouville Opial-Type Inequalities for Fractional Derivatives

This chapter provides Opial-type inequalities for generalized Riemann– Liouville fractional derivatives. The inequalities are given for integrable functions with a minimal restriction on the order of the derivatives. This treatment relies on [64].

4.1 Introduction and Preliminaries The original Opial inequality [315] (see also [297, p. 114]) states the following. Theorem 4.1. If f ∈ C 1 [0, a] with f (0) = f (a) = 0 and f (x) > 0 on (0, a), then a a a |f (x)f (x)| dx ≤ (f (x))2 dx. 4 0 0 The constant a/4 is the best possible. This result for classical derivatives has been generalized in several directions (see, for instance, [3, 4]). This chapter is analogous to [17] by the author. Using the diﬀerent Riemann–Liouville deﬁnition of the fractional derivative, we obtain inequalities for integrable rather than continuous functions, while being able to relax the conditions on the order of fractional derivatives. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 41 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 4,

42

4. Riemann–Liouville Opial-Type Inequalities

We give a brief survey of some facts about Riemann–Liouville fractional derivatives needed in the sequel; for more details see the monograph [366, Chapter 1]. Let x > 0. By C m [0, x] we denote the space of all functions on [0, x] that have continuous derivatives up to order m, and AC[0, x] is the space of all absolutely continuous functions on [0, x]. By AC m [0, x] we denote the space of all functions g ∈ C m [0, x] with g (m−1) ∈ AC[0, x] . For any ν ∈ R we denote by [ν] the integral part of ν (the integer k satisfying k ≤ ν < k + 1). By L1 (0, x) we denote the space of all functions integrable on the interval (0, x), and by L∞ (0, x) the set of all functions measurable and essentially bounded on (0, x). Clearly, L∞ (0, x) ⊂ L1 (0, x). Let ν > 0. For any f ∈ L1 (0, x) the Riemann–Liouville fractional integral of f of order ν is deﬁned by s 1 I νf (s) = (s − t)ν−1 f (t) dt, s ∈ [0, x], (4.1) Γ(ν) 0 and the Riemann–Liouville fractional derivative of f of order ν by m m s d d 1 Dνf (s) = I m−νf (s) = (s − t)m−ν−1 f (t) dt, ds Γ(m − ν) ds 0 (4.2) where m = [ν] + 1. In addition, we make the conventions D−νf := I νf if 0 < ν ≤ 1. (4.3) If ν is a positive integer, then Dνf = (d/ds)ν f . Let us remark that a somewhat more general deﬁnition of the Riemann–Liouville fractional derivative is used in the literature with an anchor point a other than 0: Let a ∈ R be ﬁxed, s ≥ a, and let fa (t) = f (a + t) be a translation of f . Then set D0f := f =: I 0f,

I −νf := Dνf if ν > 0,

Daν f (s) := Dνfa (s − a). All our results stated for the fractional derivative deﬁned by (4.2) have an interpretation for the fractional derivative with a general anchor point. Let ν > 0 and m = [ν] + 1. We deﬁne the space I ν(L1 (0, x)) as the set of all functions f on [0, x] of the form f = I ν ϕ for some ϕ ∈ L1 (0, x) (see Deﬁnition 1.2.3 in [366, p. 43]). According to Theorem 1.2.3 in [366, p. 43], the latter characterization is equivalent to the condition I m−νf ∈ AC m [0, x],

d j I m−νf (0) = 0, ds

j = 0, 1, . . . , m − 1.

(4.4) (4.5)

A function f ∈ L1 (0, x) satisfying (4.4) is said to have an integrable fractional derivative Dνf (see Deﬁnition 1.2.4 in [366, p. 44]). We express these conditions in terms of fractional derivatives.

4.1 Introduction and Preliminaries

43

Lemma 4.2. Let ν > 0 and m = [ν] + 1. A function f ∈ L1 (0, x) has an integrable fractional derivative Dνf if and only if Dν−kf ∈ C[0, x], k = 1, . . . , m,

and

Dν−1f ∈ AC[0, x].

(4.6)

Furthermore, f ∈ I ν(L1 (0, x)) if and only if f has an integrable fractional derivative Dνf and satisﬁes the conditions Dν−kf (0) = 0 for k = 1, . . . , m.

(4.7)

Proof. Note that k k d d I m−νf = I k−(ν−m+k)f = Dν−m+kf ds ds in view of the deﬁnition of the fractional derivative and the equation [ν − m + k]+1 = k. Then (4.6) is equivalent to (4.4) and (4.7) is equivalent to (4.5). (For k = m we use the convention Dν−mf = I m−νf in (4.6).) We need the following result on the law of indices for fractional integration and diﬀerentiation using the uniﬁed notation (4.3). Lemma 4.3. (Theorem 1.2.5 [366, p. 46]) The law of indices I u I vf = I u+vf

(4.8)

is valid in the following cases. (i) v > 0, u + v > 0, and f ∈ L1 (0, x). (ii) v < 0, u > 0, and f ∈ I −v (L1 (0, x)). (iii) u < 0, u + v < 0, and f ∈ I −u−v (L1 (0, x)) . Finally we give an integral representation of the fractional derivative Dγf (see also [187]). Theorem 4.4. Let ν > γ ≥ 0, let f ∈ L1 (0, x) have an integrable fractional derivative Dνf ∈ L∞ (0, x), and let Dν−kf (0) = 0 for k = 1, . . . , [ν] + 1. Then s 1 Dγf (s) = (s − t)ν−γ−1 Dνf (t) dt, s ∈ [0, x]. (4.9) Γ(ν − γ) 0 Here Dγf ∈ AC [0, x] for ν − γ ≥ 1, and Dγf ∈ C [0, x] for ν − γ ∈ (0, 1) . Proof. Set u = ν − γ > 0 and v = −ν < 0. According to Lemma 4.2, f ∈ I −v (L1 (0, x)) . Then case (ii) of Lemma 4.3 guarantees that the law of indices holds for this choice of u, v; namely I ν−γDνf = I uI vf = I u+vf =I −γf =Dγf ; this is (4.9).

44

4. Riemann–Liouville Opial-Type Inequalities

4.2 Main Results The ﬁrst result here is an Opial-type inequality involving Riemann–Liouville fractional derivatives. We assume that x, ν, γ are real numbers, x > 0, ν, γ ≥ 0, and that f ∈ L1 (0, x). The standard assumption on f is that f ∈ I ν(L1 (0, x)); we prefer to spell this out in the formulation of each theorem by specifying that f has an integrable fractional derivative Dν f satisfying (4.6). Theorem 4.5. Let 1/p+1/q = 1 with p, q > 1, let γ ≥ 0, ν > γ +1−1/p, and let f ∈ L1 (0, x) have an integrable fractional derivative Dν f ∈ L∞ (0, x) such that Dν−j f (0) = 0 for j = 1, . . . , [ν] + 1. Then

x

2/q

x

|Dν f (s) Dν f (s)| ds ≤ Ω(x) · 0

|Dν f (s)|q ds

,

(4.10)

0

where Ω(x) =

x(rp+2)/p , 21/q Γ(r + 1)((rp + 1)(rp + 2))1/p

r = ν − γ − 1.

(4.11)

Proof. We write Φ(t) = |Dν f (t)| and r = ν − γ − 1. Because 1 − 1/p > 0, we have ν > γ, and Theorem 4.4 applies. Furthermore, r > −1, and t → older’s (s − t)r ∈ L1 (0, s) for any s ∈ [0, x]. Let 0 < s ≤ x. Applying H¨ inequality to (4.9), we get |Dν f (s)| ≤ Write z(s) =

s 0

s(rp+1)/p 1 Γ(r + 1) (rp + 1)1/p

1/q

s

Φ(t)q dt

.

(4.12)

0

Φ(t)q dt. Then z (t) = Φ(t)q almost everywhere in (0, s), |Dν f (s)| = (z (s))1/q

and |Dν f (s) Dν f (s)| ≤

a.e. in (0, x),

s(rp+1)/p (z(s)z (s))1/q Γ(r + 1)(rp + 1)1/p

a.e. in (0, x).

The function srp+1 (z(s)z (s))1/q is integrable over (0, x) as rp + 1 ≥ 0 and z(s)z (s) is measurable and essentially bounded on (0, x). Applying H¨older’s inequality, we obtain x 1/p x 1/q x srp+1 (z(s)z (s))1/q ds ≤ srp+1 ds z(s)z (s) ds 0

0

x(rp+2)/p (z(s))2/q ≤ . (rp + 2)1/p 21/q

0

4.2 Main Results

45

The result then follows when we observe that |Dγ f (s) Dν f (s)| is integrable as Dγ f ∈ AC[0, x] for ν − γ ≥ 1, and Dγ f ∈ C ([0, x]) for ν − γ ∈ (0, 1) , and Dν f ∈ L∞ (0, x). The following result deals with the extreme case of the preceding theorem when p = 1 and q = ∞. Theorem 4.6. Let ν > γ ≥ 0, and let f ∈ L1 (0, x) have an integrable fractional derivative Dν f ∈ L∞ (0, x) such that Dν−j f (0) = 0 for j = 1, . . . , [ν] + 1. Then x |Dγ f (s) Dν f (s)| ds ≤ Ω1 (x) · ess sup|Dν f (s)|2 , (4.13) s∈[0,x]

0

where Ω1 (x) =

xr+2 , Γ(r + 3)

r = ν − γ − 1.

Proof. A straightforward application of Theorem 4.4. Theorem 4.5 has the following counterpart for the case 0 γ ≥ 0 , and let f ∈ L1 (0, x) have an integrable fractional derivative Dν f ∈ L∞ (0, x) that is of the same sign a.e. in (0, x) , with (Dν f )−1 ∈ L∞ (0, x) such that Dν−j f (0) = 0 for j = 1, . . . , [ν] + 1. Then

x

0

2/q

x

|Dγ f (s) Dν f (s)| ds ≥ Ω(x) ·

|Dν f (s)|q ds

,

(4.14)

0

where Ω(x) is deﬁned by (4.11). Proof. The proof follows a similar pattern as the proof of Theorem 4.5. Because 0 < p < 1, we need to apply the reverse H¨older’s inequality [297, p. 135]

x

|u(s)v(s)| ds ≥ 0

1/p

x

|u(s)|p 0

1/q

x

|v(s)|q 0

valid for any u ∈ Lp (0, x) and v ∈ Lq (0, x). Secondly, the assumption that (Dν f )−1 ∈ L∞ (0, x) is needed because q < 0 . The details of the proof are omitted. Under slightly strengthened hypotheses of Theorem 4.5 we obtain the following inequality involving fractional derivatives of three orders.

46

4. Riemann–Liouville Opial-Type Inequalities

Theorem 4.8. Let 1/p+1/q = 1 with p, q > 1, let γ ≥ 0, ν ≥ γ +2−1/p, and let f ∈ L1 (0, x) have an integrable fractional derivative Dν f ∈ L∞ (0, x) such that Dν−j f (0) = 0 for j = 1, . . . , [ν] + 1. Then x 2/q x |Dγ f (s) Dγ+1 f (s)| ds ≤ Ω2 (x) · |Dν f (s)|q ds , (4.15) 0

0

where Ω2 (x) =

x2(rp+1)/p , 2(Γ(r + 1))2 (rp + 1)2/p

r = ν − γ − 1.

(4.16)

Proof. Write Φ(t) = |Dν f (t)| and r = ν − γ − 1. From Theorem 4.4 and from the deﬁnition of the fractional integral we obtain |Dγ f (x)| ≤ U (x) := I r+1 Φ(x),

|Dγ+1 f (x)| ≤ I r Φ(x) = U (x) .

Observing that U (x) = (I r+1 Φ(x)) = I r Φ(x) and using H¨ older’s inequality, we get x x |Dγ f (t) Dγ+1 f (t)| dt ≤ U (t)U (t) dt = 12 U 2 (x) 0

0

1 = 2(Γ(r + 1))2 ≤

1 2(Γ(r + 1))2

2

x

(x − t) Φ(t) dt r

0

x

(x − t)rp dt 0

x2(rp+1)/p 1 = 2(Γ(r + 1))2 (rp + 1)2/p

2/p

2/q Φ(t)q dt

x

0 x q

Φ(t) dt

2/q .

0

The following result is concerned with the case when p = 1 and q = ∞ in the preceding theorem. The proof is straightforward, and we skip it. Theorem 4.9. Let γ ≥ 0, ν > γ + 1, and let f ∈ L1 (0, x) have an integrable fractional derivative Dν f ∈ L∞ (0, x) such that Dν−j f (0) = 0 for j = 1, . . . , [ν] + 1. Then x |Dγ f (s) Dγ+1 f (s)| ds ≤ Ω3 (x) · ess sup|Dν f (t)|2 , (4.17) t∈[0,x]

0

where Ω3 (x) =

x2(ν−γ) . 2(Γ(ν − γ + 1))2

(4.18)

Remark 4.10. We show that inequality (4.17) is sharp, attained for the function f (t) = tν . From the known properties of the gamma function, s Γ(u)Γ(v) u+v−1 s (s − t)u−1 tv−1 dt = , u, v > 0. Γ(u + v) 0

4.2 Main Results

47

Let 0 ≤ j ≤ [ν] + 1, m = [ν] − j + 1 and α = ν − [ν]. Then 1 − α > 0, ν + 1 > 0, and m s d 1 ν−j f (s) = (s − t)(1−α)−1 t(ν+1)−1 dt D Γ(1 − α) ds 0 m Γ(1 − α)Γ(ν + 1) d 1 sm+j = Γ(1 − α) Γ(m + j + 1) ds Γ(ν + 1) j s . = j! Then Dν−j f (0) = 0 for j = 1, . . . , [ν] + 1, and Dν f (s) = Γ(ν + 1). Using Theorem 4.4, we obtain Dγ f (s) =

Γ(ν + 1) sν−γ , Γ(ν − γ + 1)

Hence x |D f (s) D γ

0

γ+1

Dγ+1 f (s) =

Γ(ν + 1) ν−γ−1 s . Γ(ν − γ)

x (Γ(ν + 1))2 f (s)| ds = s2(ν−γ)−1 ds Γ(ν − γ + 1)Γ(ν − γ) 0 2 Γ(ν + 1) 1 x2(ν−γ) . = 2 Γ(ν − γ + 1)

On the other hand, with Ω3 (x) given by (4.18) and ess sup|Dν f (s)|2 = (Γ(ν + 1))2 , the left side of (4.17) is equal to the right side of (4.17) for f (t) = tν . We give a counterpart of Theorem 4.8 for the case 0 γ + 1, and let f ∈ L1 (0, x) have an integrable fractional derivative Dν f ∈ L∞ (0, x) that is of the same sign a.e. in (0, x) , with (Dν f )−1 ∈ L∞ (0, x) such that Dν−j f (0) = 0 for j = 1, . . . , [ν] + 1. Then x 2/q x γ γ+1 ν q |D f (s) D f (s)| ds ≥ Ω2 (x) · |D f (s)| ds , (4.19) 0

0

where Ω2 (x) is given by (4.16). We derive yet another useful variant of an Opial-type inequality. Theorem 4.12. Let 1/p+1/q = 1 with p, q > 1, let γ ≥ 0, ν > γ+1−1/p, and let f ∈ L1 (0, x) have an integrable fractional derivative Dν f ∈ L∞ (0, x) such that Dν−j f (0) = 0 for j = 1, . . . , [ν] + 1. Then, for any m > 0, x m/q x γ m ν q |D f (s)| ds ≤ Ω4 (x) · |D f (s)| ds , (4.20) 0

0

48

4. Riemann–Liouville Opial-Type Inequalities

where Ω4 (x) =

x(rm+1+(m/p)) , (Γ(r + 1))m (rm + 1 + (m/p))(rp + 1)m/p

r = ν−γ−1. (4.21)

Proof. Inequality (4.12) holds under the hypotheses of the theorem. Raising both sides of (4.12) to the power of m and integrating from 0 to x, we get the result. The extreme case of Theorem 4.12 with p = 1 and q = ∞ follows. The proof is omitted, as it is again a straightforward application of Theorem 4.4. Theorem 4.13. Let ν > γ ≥ 0, and let f ∈ L1 (0, x) have an integrable fractional derivative Dν f ∈ L∞ (0, x) such that Dν−j f (0) = 0 for j = 1, . . . , [ν] + 1. Then, for any m > 0, x |Dγ f (s)|m ds ≤ Ω5 (x) · ess sup|Dν f (t)|m , (4.22) 0

where Ω5 (x) =

t∈[0,x]

xm(ν−γ)+1 . (m (ν − γ) + 1) (Γ(ν − γ + 1))m

4.3 Applications (i) Uniqueness of solution to fractional initial value problem ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨

Let γ i ≥ 0, ν > γ i + 1/2, i = 1, . . . , r ∈N. Let f ∈ L1 (0, x) have an integrable fractional derivative Dν f ∈ L∞ (0, x) such that Dν−j f (0) = αj ∈R, j = 1, . . . , [ν] + 1. ⎪ ⎪ ⎪ Furthermore, let ⎪ ⎪ ⎪ ⎩ Dν f (t) = F (t, {Dγ i f (t)}r ) for all t ∈ [0, x]. i=1 (4.23) Here F (t, x1 , . . . , xr ) is continuous for (x1 , . . . , xr ) ∈ Rr , bounded for t ∈ [0, x], and fulﬁlls the Lipschitz condition |F (t, z1 , . . . , zr ) − F (t, z1 , . . . , zr )| ≤

r

qi (t) |zi − zi |,

(4.24)

i=1

where qi (t) ≥ 0 are bounded on [0, x], i = 1, . . . , r. For i = 1, . . . , r and 0 ≤ s ≤ x we deﬁne r sν−γ i Δi (s) := , ψ(s) = qi ∞ Δi (s), 2Γ(ν − γ i ) (ν − γ i )(2ν − 2γ i − 1) i=1 (4.25)

4.3 Applications

49

where qi ∞ = supt∈[0,x] |qi (t)|. We assume that ψ(x) :=

r

qi ∞ Δi (x) < 1.

(4.26)

i=1

Let g ∈ L1 (0, x) have an integrable fractional derivative Dν g ∈ L∞ (0, x) such that Dν−j g(0) = 0, j = 1, . . . , [ν] + 1. Then, by Theorem 4.4, we have s 1 γi D g(s) = (s − t)ν−γ i −1 Dν g(t) dt, s ∈ [0, x], i = 1, . . . , r. Γ(ν − γ i ) 0 (4.27) When p = q = 2, from (4.10) we get for i = 1, . . . , r, x x |(Dγ i g)(w)| |(Dν g)(w)| dw ≤ Δi (x) |(Dν g)(w)|2 dw. (4.28) 0

0

Let f1 , f2 solve (4.23); that is, let for k = 1, 2, (Dν fk )(t) = F (t, {(Dγ i fk )(t)}ri=1 ),

t ∈ [0, x],

and Dν−j fk (0) = αj ∈ R,

j = 1, . . . [ν] + 1.

If g := f1 − f2 , then Dν f (t) = F (t, {(Dγ i f1 )(t)}ri=1 ) − F (t, {(Dγ i f2 )(t)}ri=1 ),

(4.29)

and Dν−j g(0) = 0,

j = 1, . . . , [ν] + 1.

By (4.24), |F (t, Dγ 1 f1 (t), . . . , Dγ r f1 (t)) − F (t, Dγ 1 f2 (t), . . . , Dγ r f2 (t))| r ≤ qi (t)|Dγ i f1 (t) − Dγ i f2 (t)| i=1

=

r

qi (t)|Dγ i g(t)|

i=1

≤

r

qi ∞ |Dγ i g(t)|.

(4.30)

i=1

Thus |Dν g(t)|2 = |Dν g(t)| |F (t, {Dγ i f1 (t)}ri=1 ) − F (t, {Dγ i f2 (t)}ri=1 )| r ν ≤ |D g(t)| qi ∞ |Dγ i g(t)| i=1

=

r i=1

qi ∞ |Dγ i g(t)| |Dν g(t)|.

50

4. Riemann–Liouville Opial-Type Inequalities

Consequently,

x

|Dν g(t)|2 dt ≤ 0

r

qi ∞

|Dγ i g(t)| |Dν g(t)| dt 0

i=1 (4.28)

≤

r

qi ∞ Δi (x)

i=1 (4.26)

x

x

|Dν g(t)|2 dt 0

x

≤ ψ(x)

|Dν g(t)|2 dt; 0

that is,

x

x

|D g(t)| dt ≤ ψ(x) ν

|Dν g(t)|2 dt.

2

0

(4.31)

0

x If 0 |Dν g(t)|2 dt = 0, then by (4.31) we obtain ψ(x) ≥ 1, a contradiction x to the assumption ψ(x) < 1. Hence 0 |Dν g(t)|2 dt = 0; that is, Dν g(t) = 0 a.e. in [0, x]. But Dν−j g(0) = 0, j = 1, . . . , [ν] + 1. Then from (4.9) for γ = 0 we ﬁnd g(t) ≡ 0 in [0, x]. This implies f1 = f2 on [0, x], proving the uniqueness of the solution to the initial value problem (4.23). (ii) Upper bounds on Dν f , solution f , and others ⎧ Consider the initial value problem for 0 ≤ t ≤ x: ⎪ ⎪ ⎪ ⎪ ⎪ (Dν f ) (t) = F (t, {Dγ i f (t)}ri=1 , Dν f ). ⎪ ⎪ ⎪ ⎨ γ ≥ 0, ν > γ + 1/2, i = 1, . . . , r ∈ N. i i ⎪ (0, x) has an integrable fractional derivative Here f ∈ L 1 ⎪ ⎪ ⎪ ν ∞ ⎪ ⎪ D f ∈ L (0, x) which is absolutely continuous. ⎪ ⎪ ⎩ We assume that Dν−j f (0) = 0, j = 1, . . . , [ν] + 1 and Dν f (0) = A ∈R. (4.32) Here F is Lebesgue measurable on [0, x] × Rr+1 , and fulﬁlls the condition |F (t, x1 , . . . , xr , xr+1 )| ≤

r

qi (t)|xi |,

(4.33)

i=1

where qi (t) ≥ 0 are bounded on [0, x], i = 1, . . . , r. We see that Dν f (t)(Dν f ) (t) = Dν f (t)F (t, {Dγ i f (t)}ri=1 , Dν f (t)), and for 0 ≤ s ≤ x we have s Dν f (t)(Dν f ) (t) dt = 0

0

s

Dν f (t)F (t, {Dγ i f (t)}ri=1 , Dν f (t)) dt.

4.3 Applications

Hence

s

ν 2 1 2 |D f (t)|

0

51

s

≤

|Dν f (t)||F (t, {Dγ i f (t)}ri=1 , Dν f (t))| dt r s (4.33) ≤ |Dν f (t)| qi ∞ |Dγ i f (t)| dt 0

0

=

r

qi ∞

i=1

i=1

s

|D f (t)| |D f (t)| dt . γi

ν

0

Recall the notation (4.25) for Δi (s) and ψ(s). Then s r ν 2 2 γi ν |D f (s)| ≤ A + 2 qi ∞ |D f (t)| |D f (t)| dt i=1 (4.10)

≤ A + 2

r

0

qi ∞ Δi (s)

i=1 s

s

|Dν f (t)|2 dt 0

|Dν f (t)|2 dt;

= A2 + ψ(s) 0

that is,

s

|Dν f (s)|2 ≤ A2 + ψ(s)

|Dν f (t)|2 dt.

(4.34)

0

Set θ(s) := |Dν f (s)|2 for 0 ≤ s ≤ x and ρ := A2 . Then s θ(s) ≤ ρ + ψ(s) θ(t) dt, 0

where ρ ≥ 0, ψ(s) ≥ 0, ψ(0) = 0, θ(s) ≥ 0 for all 0 ≤ s ≤ x. We can apply the generalized Gronwall lemma [144, Corollary 1.1.2] (with H(x) = x) to obtain t s exp(−Ψ(t)) dt , Ψ(t) = ψ(u) du. θ(s) ≤ ρ 1 + ψ(s) exp(Ψ(s)) 0

0

(4.35) We have shown that |Dν f (s)| ≤ |A| 1 + ψ(s) exp(Ψ(s))

1/2

s

exp(−Ψ(t)) dt

=: K(s)

0

(4.36) for all 0 ≤ s ≤ x. From (4.9) with γ = 0 we get s 1 |f (s)| ≤ (s − t)ν−1 |Dν f (t)| dt Γ(ν) 0 for all 0 ≤ s ≤ x. Applying (4.36), we obtain s 1 |f (s)| ≤ (s − t)ν−1 K(t) dt. Γ(ν) 0

(4.37)

52

4. Riemann–Liouville Opial-Type Inequalities

Also from (4.9) we have |Dγ i f (s)| ≤

1 Γ(ν − γ i )

s

(s − t)ν−γ i −1 |Dν f (t)| dt

0

for all 0 ≤ s ≤ x, i = 1, . . . , r. Finally, by (4.36) we ﬁnd s 1 |Dγ i f (s)| ≤ (s − t)ν−γ i −1 K(t) dt Γ(ν − γ i ) 0 for all 0 ≤ s ≤ x and i = 1, . . . , r.

(4.38)

5 Opial-type Lp-Inequalities for Riemann–Liouville Fractional Derivatives

This chapter presents a class of Lp -type Opial inequalities for generalized Riemann–Liouville fractional derivatives of integrable functions. The novelty of this approach is the use of the index law for fractional derivatives instead of a Taylor’s formula, which enables us to relax restrictions on the orders of fractional derivatives. This treatment relies on [65].

5.1 Introduction and Preliminaries The Opial inequality, which appeared in [315], is of great interest in diﬀerential equations and other areas of mathematics, and has attracted a great deal of attention in the recent literature. For classical derivatives it has been generalized in several directions (see, for instance, [3, 4, 323]). Love gave a generalization for fractional integrals [261]. This chapter is inspired by the author’s paper [15]. Here we consider Lebesgue integrable functions, whereas [15] dealt with continuous functions using a diﬀerent deﬁnition of the fractional derivative. Our brief survey of basic facts about fractional derivatives is based on the monograph [366] by Samko et al.; most of the results needed in the sequel are contained in Chapter 1 of [366]. The crucial result is Theorem 5.4, which replaces Taylor’s formula in the derivation of various estimates. Throughout the chapter, x denotes a ﬁxed positive number. By C m [0, x] we denote the space of all functions on [0, x] that have continuous derivatives up to order m, and AC[0, x] is the space of all absolutely continuous G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 53 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 5,

5. Opial-type Lp -Inequalities

54

functions on [0, x]. By AC m [0, x] we denote the space of all functions g ∈ C m [0, x] with g (m−1) ∈ AC[0, x]. For any α ∈ R we denote by [α] the integral part of α (the integer k satisfying k ≤ α < k + 1). If p ∈ R, p > 0, by Lp (0, x) we denote the space of all Lebesgue measurable functions f for which |f |p is Lebesgue integrable on the interval (0, x), and by L∞ (0, x) the set of all functions measurable and essentially bounded on (0, x). For any f ∈ L∞ (0, x) we write f ∞ = supt∈[0,x] |f (t)|. We observe that L∞ (0, x) ⊂ Lp (0, x) for all p > 0. For any a ∈ R we write a+ = max(a, 0) and a− = (−a)+ . For the sake of completeness we give a proof of the following known result which provides a basis for the existence of fractional integrals and is needed in another context in the chapter. Lemma 5.1. Let f ∈ L1 (0, x) and let α > −1 be a real number. Then s (s − t)α f (t) dt F (s) = 0

exists for almost all s ∈ [0, x], and F ∈ L1 (0, x). α

Proof. Deﬁne k : Ω := [0, x] × [0, x] → R by k (s, t) = (s − t)+ ; that is, % (s − t)α if 0 ≤ t < s ≤ x, k(s, t) = 0 if 0 ≤ s ≤ t ≤ x. Then k is measurable on Ω, and

x 0

t

k(s, t) ds = 0

x

k(s, t) ds +

x

k(s, t) ds = t

(s − t)α ds = (α + 1)−1 (x − t)α+1 .

t

Because the repeated integral x x x dt k(s, t)|f (t)| ds = (α + 1)−1 (x − t)α+1 |f (t)| dt 0

0

0

exists and is ﬁnite, the function (s, t) → k(s, t)f (t) is integrable over Ω by Tonelli’s theorem, and the conclusion follows from Fubini’s theorem. Let α > 0. For any f ∈ L1 (0, x) the Riemann–Liouville fractional integral of f of order α is deﬁned by s 1 (s − t)α−1 f (t) dt, s ∈ [0, x]. (5.1) I αf (s) = Γ(α) 0 By Lemma 5.1 the integral on the right side of (5.1) exists for almost all s ∈ [0, x], and I αf ∈ L1 (0, x). The Riemann–Liouville fractional derivative of f ∈ L1 (0, x) of order α is deﬁned by m m s d d 1 α m−α D f (s) = I f (s) = (s − t)m−α−1 f (t) dt, ds Γ(m − α) ds 0 (5.2)

5.1 Introduction and Preliminaries

55

where m = [α] + 1, provided that the derivative exists. In addition, we set I −αf := Dαf if α > 0,

D−αf := I αf if 0 < α ≤ 1. (5.3) If α is a positive integer, then Dαf = (d/ds)α f . A more general deﬁnition of fractional integrals and derivatives uses an anchor point other than 0: let f ∈ L1 (a, b), where −∞ < a < b < ∞. For any s ∈ [a, b] set D0f := f =: I 0f,

α f (s) := Ia+

1 Γ(α)

s

(s−t)α−1 f (t) dt,

α Ib− f (s) :=

a

1 Γ(α)

b

(s−t)α−1 f (t) dt. s

The two fractional derivatives are then deﬁned by an obvious modiﬁcation of (5.2). All our results stated for the specialized fractional derivative (5.2) have an interpretation for the fractional derivatives with a general anchor point. Let α > 0 and m = [α] + 1. A function f ∈ L1 (0, x) is said to have an integrable fractional derivative Dαf (see Deﬁnition 2.4 in [366, p. 44]) if I m−αf ∈ AC m [0, x].

(5.4)

1

We deﬁne the space I α(L (0, x)) as the set of all functions f on [0, x] of the form f = I α ϕ for some ϕ ∈ L1 (0, x) (see Deﬁnition 2.3 in [366, p. 43]). We express these conditions in terms of fractional derivatives. Lemma 5.2. Let α > 0 and m = [α] + 1. A function f ∈ L1 (0, x) has an integrable fractional derivative Dαf if and only if Dα−kf ∈ C[0, x], k = 1, . . . , m,

and

Dα−1f ∈ AC[0, x].

(5.5)

1

Furthermore, f ∈ I α(L (0, x)) if and only if f has an integrable fractional derivative Dαf and satisﬁes the conditions Dα−kf (0) = 0 for

k = 1, . . . , m.

(5.6)

Proof. Notice that k k d d m−α I f= I k−(α−m+k)f = Dα−m+kf ds ds in view of the deﬁnition of fractional derivative and the equation [α − m + k] +1 = k. Then (5.5) is equivalent to (5.4) and (5.6) is equivalent to condition (2.56) in [366, p. 43]. (For k = m we use the convention Dα−mf = I m−αf in (5.5).)

56

5. Opial-type Lp -Inequalities

We need the following result on the law of indices for fractional integration and diﬀerentiation using the uniﬁed notation (5.3). Lemma 5.3. (Theorem 2.5 in [366, p. 46]) The law of indices I μI νf = I μ+νf

(5.7)

is valid in the following cases.: (i) ν > 0, μ + ν > 0 and f ∈ L1 (0, x). (ii) ν < 0, μ > 0 and f ∈ I −ν(L1 (0, x)). (iii) μ < 0, μ + ν < 0 and f ∈ I −μ−ν(L1 (0, x)). The following theorem is a powerful analogue of Taylor’s formula with vanishing fractional derivatives of lower orders. In this chapter it is used as the main tool for deriving inequalities. Observe that we do not require α ≥ β + 1 but merely α > β. Theorem 5.4. Let α > β ≥ 0, let f ∈ L1 (0, x) have an integrable fractional derivative Dαf , and let Dα−kf (0) = 0 for k = 1, . . . , [α] + 1. Then s 1 Dβf (s) = (s − t)α−β−1 Dαf (t) dt, s ∈ [0, x]. (5.8) Γ(α − β) 0

Proof. Set μ = α − β > 0 and ν = −α < 0. According to Lemma 5.2, 1 f ∈ I −ν(L (0, x)). Then case (ii) of Lemma 5.3 guarantees that the law of indices holds for this choice of μ, ν; namely I α−βDαf = I μI νf = I μ+νf = I −βf = Dβf ; this proves the result. Notice that the existence of the integral on the right in (5.8) is guaranteed by Lemma 5.1.

5.2 Main Results We assume here that x, ν are positive real numbers, and that f ∈ L1 (0, x). The standard assumption on f is that f ∈ I ν(L1 (0, x)); this is equivalent to f having an integrable fractional derivative Dνf satisfying (5.5). In addition we require that Dνf be essentially bounded to guarantee that Dνf ∈ Lp (0, x) for p > 0. We tabulate the notation used in this section. The inequalities between ν and μi are assumed throughout.

5.2 Main Results

l x, ν, ri r μi αi α β ω1 , ω2 ω sk , sk

57

A positive integer. Positive real numbers, i = 1, . . . , l. "l r = i=1 ri . Real numbers satisfying 0 ≤ μi < ν, i = 1, . . . , l. αi = ν − μi − 1, i = 1, . . . , l. α = max {(αi )− : i = 1, . . . , l}. β = max {(αi )+ : i = 1, . . . , l}. Continuous positive weight functions on [0, x]. Continuous nonnegative weight function on [0, x]. sk > 0 and 1/sk + 1/sk = 1, k = 1, 2.

For brevity we write μ = (μ1 , . . . , μl ) for a selection of the orders μi of fractional derivatives, and r = (r1 , . . . , rl ) for a selection of the constants ri . We derive a very general Opial-type inequality involving Riemann– Liouville fractional derivatives of an integrable function f , which is analogous to [323, Theorem 1.3] for ordinary derivatives and to [15, Theorem 2] for fractional derivatives. Theorem 5.5. Let f ∈ L1 (0, x) have an integrable fractional derivative D f ∈ L∞ (0, x) such that Dν−jf (0) = 0 for j = 1, . . . , [ν] + 1. For k = 1, 2, let sk > 1, let p ∈ R satisfy ν

αs2 < 1,

p>

s2 , 1 − αs2

(5.9)

and let σ = 1/s2 − 1/p. Finally let

x

Q1 =

s1

1/s1

ω 1 (τ ) dτ

x

and Q2 =

0

−s2 /p

ω 2 (τ )

r/s2 dτ

.

0

(5.10) Then

x

ω 1 (τ ) 0

l

ω 2 (τ )|Dνf (τ )|p dτ

r/p ,

0

i=1

where ρ :=

x

|Dμif (τ )|ri dτ ≤ Q1 Q2 C1 xρ+1/s1

(5.11)

"l

i=1 αi ri + σr and

C1 = C1 (ν, μ, r, p, s1 , s2 ) := &l i=1

σ rσ

. Γ(ν − μi )ri (αi + σ)ri σ (ρs1 + 1)1/s1 (5.12)

Proof. First we show that the conditions on s2 and p guarantee that, for i = 1, . . . l, (a) p > s2 > 1,

(b) αi s2 > −1,

(c) αi + σ > 0.

(5.13)

5. Opial-type Lp -Inequalities

58

This is clear if α = 0. If α > 0, then 0 < 1 − αs2 < 1 and p > s2 /(1 − αs2 ) > s2 > 1. For each i ∈ {1, . . . , l}, αi ≥ −α, and αi s2 ≥ −αs2 > −1; furthermore,

αi + σ = αi +

1 + αi s2 1 − αs2 1 1 1 1 − = − ≥ − > 0. s2 p s2 p s2 p

For brevity we write ν i ki (τ , t) = (τ − t)α + , i = 1, . . . , l, Φ(t) = |D f (t)|, 0 ≤ τ , t ≤ x.

From (5.13) it follows that ki (τ , ·) ∈ Ls2 (0, x) and ki (τ , ·) ∈ L1/σ (0, x).

(5.14)

Let i ∈ {1, . . . , l} and τ ∈ [0, x]. We then apply H¨ older’s inequality twice (with the conjugate indices s2 , s2 and p/s2 , p/(p − s2 )) taking into account (5.14) and the fact that ω −1 2 , ω 2 , and Φ are (essentially) bounded:

x

x

ki (τ , t)Φ(t) dt = 0

ω 2 (t)−1/p ω 2 (t)1/p Φ(t)ki (τ , t)dt

0

x

−s2 /p

≤

ω 2 (t)

dt

0

1/s2

x s2 /p

ω 2 (t) 0

≤

1/s2

1/r Q2

s2

s2

Φ(t) ki (τ , t) dt 1/p

x p

1/σ

ω 2 (t)Φ(t) dt 0

1/r

ki (τ , t) 1/p

x

ω 2 (t)Φ(t)p dt

= Q2

σ

x

0

dt

0

σ σ τ αi +σ . (αi + σ)σ

By Theorem 5.4, Γ(ν − μi )|Dμif (τ )| ≤

τ

(τ − t)αi Φ(t) dt = 0

x

ki (τ , t)Φ(t) dt. 0

(5.15)

5.2 Main Results

59

Therefore

x

ω 1 (τ ) 0

l

|Dμif (τ )|ri dτ

i=1

x

≤ 0

l

1 ω 1 (τ ) Γ(ν − μi )ri i=1

x

≤

ω 1 (τ ) 0

ri

x

ki (τ , t)Φ(t) dt

l

1 r /r Q2i ri Γ(ν − μ ) i i=1

0

σ rσ

Q2 Γ(ν − μi )ri (αi + σ)ri σ l x (αi +σ)ri · ω 1 (τ ) τ dτ

i=1

r/p

x

ω 2 (t)Φ(t) dt 0

x

p

= ΔQ2

ω 1 (τ )τ ρ dτ

ω 2 (t)Φ(t) dt 0

≤ ΔQ2

r/p

x p

i=1

0

ri /p

x

ω 2 (t)Φ(t)p dt

σ ri σ · τ (αi +σ)ri dτ (αi + σ)ri σ = &l

0

r/p

x

ω 2 (t)Φ(t)p dt 0

Δ = Q2 (ρs1 + 1)1/s1

dτ

0

x

ω 1 (τ )s1 dτ 0

x

1/s1

τ ρs1 dτ

0

r/p

ω 2 (t)Φ(t)p dt

1/s1

x

Q1 xρ+1/s1

0

&l where Δ := σ rσ /( i=1 Γ(ν − μi )ri (αi + σ)ri σ ). This completes the proof. Next we consider the extreme case p = ∞ in analogy with [15, Proposition 1]. Theorem 5.6. Let f ∈ L1 (0, x) have an integrable fractional derivative D f ∈ L∞ (0, x) such that Dν−jf (0) = 0 for j = 1, . . . , [ν] + 1. Then ν

x

ω(τ ) 0

where ρ =

l

|Dμif (τ )|ri dτ ≤

i=1

"l

i=1 (ν

ρ

ω∞ xρ

&l

ri i=1 Γ(ν − μi + 1)

Dνf r∞ ,

− μi )ri + 1.

Proof. By Theorem 5.4, |Dμif (τ )| ≤

1 Γ(ν − μi )

τ

(τ − t)αi |Dνf (t)| dt, 0

(5.16)

5. Opial-type Lp -Inequalities

60

which implies |Dμif (τ )| ≤

Dνf ∞ τ ν−μi Dνf ∞ τ ν−μi . = Γ(ν − μi ) ν − μi Γ(ν − μi + 1)

(5.17)

The result then follows when we raise (5.17) to the power of ri , take the product from 1 to l, multiply by ω(τ ), and integrate with respect to τ from 0 to x. We have the following counterpart of Theorem 5.5 with s1 , s2 ∈ (0, 1) and p negative. Theorem 5.7. Let f ∈ L1 (0, x) have an integrable fractional derivative Dνf ∈ L∞ (0, x) that is of the same sign a.e. in (0, x) and satisﬁes Dν−jf (0) = 0, j = 1, . . . , [ν] + 1. For k = 1, 2, let 0 < sk < 1, let p < 0, and let σ = 1/s2 − 1/p. Then

x

ω 1 (τ ) 0

l

|Dμif (τ )|ri dτ ≥ Q1 Q2 C1 xρ+1/s1

r/p

x

ω 2 (τ )|Dνf (τ )|p dτ 0

i=1

(5.18) "l α r + σr, Q and Q are deﬁned by (5.10), and C1 is where ρ = i i 1 2 i=1 deﬁned by (5.12). Proof. Combining Theorem 5.4 with the hypotheses on Dνf , we have τ x μi αi Γ(ν − μi )|D f (τ )| = (τ − t) Φ(t) dt = ki (τ , t)Φ(t) dt, (5.19) 0

0

where Φ(t) = D f or Φ(t) = −D f (depending on the sign of Dνf in (0, x)). Because αi > −1 and 0 < s2 < 1, we have αi s2 > −1. Furthermore, i σ = 1/s2 − 1/p > 0. Writing ki (τ , t) = (τ − t)α + (i = 1, . . . , l), we have ν

ki (τ , ·) ∈ Ls2 (0, x)

ν

and ki (τ , ·) ∈ L1/σ (0, x).

(5.20)

We can now retrace the proof of Theorem 5.5, relying on (5.20) and using the reverse H¨older’s inequality in place of H¨ older’s inequality proper (as 0 < sk < 1 for k = 1, 2 and p < 0). A possible choice of p in this theorem is p = (s1 s22 )/(s1 s2 − 1). This results in an inequality similar to the one obtained earlier by the author [15, Theorem 3]. We obtain yet another counterpart of Theorem 5.5 if we assume that s1 , s2 , and p lie in the interval (0, 1). In this case the hypotheses on s1 , s2 , and p are of necessity more restrictive. Theorem 5.8. Let f ∈ L1 (0, x) have an integrable fractional derivative Dνf ∈ L∞ (0, x) that is of the same sign a.e. in (0, x) and satisﬁes

,

5.2 Main Results

61

Dν−jf (0) = 0, j = 1, . . . , [ν] + 1. For k = 1, 2, let 0 < sk < 1, let rs1 ≤ 1, p ∈ R, s2 s2
(b) − 1 < αi + σ < 0.

(5.22)

Because 1 − αs2 + s2 > 0 and 1 + βs2 ≥ 1, inequality (5.22) (a) follows directly from (5.21). Furthermore, we have αi + σ = (1 + αi s2 )/s2 − 1/p, and 1 + αi s2 1 + βs2 1 1 1 1 − αs2 − ≤ − < − < 0. −1 < s2 p s2 p s2 p This proves (5.22) (b). Because αi > −1 and 0 < s2 < 1, we have αi s2 > −1. Furthermore, i σ < 0, and αi /σ > −1. Writing ki (τ , t) = (τ − t)α + (i = 1, . . . , l), we have ki (τ , ·) ∈ Ls2 (0, x)

and ki (τ , ·) ∈ L1/σ (0, x).

As in the proof of the preceding theorem we have τ x μi αi Γ(ν − μi )|D f (τ )| = (τ − t) Φ(t) dt = ki (τ , t)Φ(t) dt, 0

(5.23)

(5.24)

0

where Φ(t) = Dνf or Φ(t) = −Dνf (depending on the sign of Dνf in (0, x)). We can now retrace the proof of Theorem 5.5, relying on (5.23) and using the reverse H¨older’s inequality in place of H¨ older’s inequality proper o (as 0 < sk < 1 for k = 1, 2 and 0 −rs1 ≥ −1

i=1

taking into account the assumption rs1 ≤ 1.

We present a version of Opial’s inequality with l = 2 motivated by Pang and Agarwal’s extension [323, Theorem 1.1] of an inequality due to Fink [160] for classical derivatives. This was further extended in [15, Theorem 4] to fractional derivatives. Our proof is similar to the one given in [323]. In view of the auxiliary inequalities used, in particular of (5.27), the theorem does not extend easily to l > 2.

5. Opial-type Lp -Inequalities

62

Theorem 5.9. Let f ∈ L1 (0, x) have an integrable fractional derivative Dνf ∈ L∞ (0, x) such that Dν−jf (0) = 0 for j = 1, . . . , [ν] + 1. Let ν > μ2 ≥ μ1 + 1 ≥ 1. If p, q > 1 are such that 1/p + 1/q = 1, then

x

2ν−μ1 −μ2 −1+2/q

|D f (τ )||D f (τ )| dτ ≤ C2 x μ1

μ2

0

2/p

x

|D f (τ )| dτ ν

p

,

0

(5.25) where C2 = C2 (ν, μ1 , μ2 , p) is given by C2 := (1/2)1/p . Γ(ν − μ1 )Γ(ν − μ2 + 1)((ν − μ1 )q + 1)1/q ((2ν − μ1 − μ2 − 1)q + 2)1/q (5.26) Proof. First an auxiliary inequality: write αi = ν − μi − 1 for i = 1, 2; in view of the hypothesis μ2 ≥ μ1 + 1 we have α1 − α2 − 1 ≥ 0. Let 0 ≤ t ≤ s ≤ x. Then x

α2 α1 α2 1 (τ − t)α dτ + (τ − s)+ + (τ − s)+ (τ − t)+ 0

≤

1 (x − t)α1 (x − s)α2 +1 . (ν − μ2 )

(5.27)

This is veriﬁed by estimating the integrand in (5.27) (with τ ≥ s ≥ t): (τ − t)α1 (τ − s)α2 + (τ − s)α1 (τ − t)α2 = (τ − t)α1 −α2 −1 (τ − t)α2 +1 (τ − s)α2 + (τ − s)α1 −α2 −1 (τ − s)α2 +1 (τ − t)α2

≤ (x − t)α1 −α2 −1 (τ − t)α2 +1 (τ − s)α2 + (τ − s)α2 +1 (τ − t)α2 (where the last inequality requires α1 − α2 − 1 ≥ 0); (5.27) follows from x

α2 +1 α2 +1 2 2 (τ − t)+ (τ − s)α (τ − t)α + + (τ − s)+ + dτ = 0

1 α +1 [(x − t)(x − s)] 2 . α2 + 1 In the following calculation we abbreviate c1 := (Γ(ν − μ2 )Γ(ν − μ1 ))−1 ,

c2 := (Γ(ν − μ2 + 1)Γ(ν − μ1 ))−1 ,

c3 := (ν − μ2 )q + 1,

:= 2ν − μ1 − μ2 − 1 + 1/q.

By Theorem 5.4, Dμif (τ ) =

1 Γ(ν − μi )

x ν i (τ − t)α + D f (t) dt,

0

i = 1, 2.

5.2 Main Results

63

Using this representation, the auxiliary inequality (5.27), and H¨ older’s inequality, we obtain x |Dμ1f (τ )||Dμ2f (τ )| dτ 0 x x x α1 α2 ν ν ≤ c1 |D f (t)|(τ − t)+ dt |D f (s)|(τ − s)+ ds dτ 0 0 0 x x |Dνf (t)| |Dνf (s)| = c1 0 t x

α1 α2 α1 α2 (τ − t)+ (τ − s)+ + (τ − s)+ (τ − t)+ dτ ds dt · 0 x x ν ν α1 α2 +1 ≤ c2 |D f (t)| |D f (s)|(x − t) (x − s) ds dt 0 t x x ν α1 ν α2 +1 |D f (t)|(x − t) |D f (s)|(x − s) ds dt = c2 0

≤ c2

t

x

0 −1/q

t

|Dνf (t)|(x − t)q

x

−1/q c2 c3 (q

x

|Dνf (s)|p ds

|D f (s)| ds ν

−1/q (q+1)/q

This implies (5.25).

q(α2 +1)

x

p

1 2

ds

t 1/p

x

p t

+ 1)

(x − s)

p

t

|D f (t)| ν

0

≤

x

0

≤

ν

1/q

x

|D f (s)| ds

α1

= c2 c3

−1/q c2 c3

1/p

x

|D f (t)|(x − t) ν

dt

1/p

1/q

x

(x − t) dt q

dt 0

2 1/p

x

|D f (t)| dt ν

p

.

0

In the following theorem we address the case when the function |Dνf | is monotone. Theorem 5.10. Let f ∈ L1 (0, x) have an integrable fractional derivative D f ∈ L∞ (0, x) such that Dν−jf (0) = 0 for j = 1, . . . , [ν]+1 and that |Dνf | is decreasing on [0, x]. Let l ≥ 2. If p, q > 1 are such that 1/p + 1/q = 1 "l and i=1 αi p > −1, then ν

0

x

l

|D f (τ )| dτ ≤ C3 x μi

ν

lq

,

(5.28)

.

(5.29)

0

i=1

where γ :=

1/q

x

|D f (t)| dt

(γp+lp+1)/p

"l i=1

αi and

C3 = C3 (ν, μ, p) :=

p (γp +

1)1/p (γp

+ p + 1)

&l i=1

Γ(ν − μi )

dt

5. Opial-type Lp -Inequalities

64

Proof. By Theorem 5.4, 1 |D f (τ )| ≤ Γ(ν − μi )

x ν i (τ − t)α + |D f (t)| dt.

μi

0

ν i The integrand t → (τ − t)α + |D f (t)| is decreasing (and integrable) on [0, x] for all τ ∈ [0, x]. By Chebyshev’s inequality for the product of integrals [183, p. 1099], l

xl−1

|Dμif (τ )| ≤ &l

i=1 Γ(ν − μi )

i=1

xl−1

≤ &l

i=1 Γ(ν − μi )

xl−1

≤ &l i=1

Γ(ν − μi ) xl−1

≤ &l i=1

Γ(ν − μi )

x

0

l

ν i (τ − t)α + |D f (t)| dt

i=1 x

(τ − t)γ+ |Dνf (t)|l dt

0

1/p

τ

0

τ γp+1 γp + 1

1/q

x

(τ − t)γp dt

|Dνf (t)|lq dt 0

1/p

1/q

x

|D f (t)| dt ν

xl−1 τ (γp+1)/p = &l (γp + 1)1/p i=1 Γ(ν − μi )

lq

0

1/q

x

|D f (t)| dt ν

lq

.

0

Integrating with respect to τ from 0 to x, we get the result. Condition x "l α p > −1 was needed in order to apply H¨ o lder’s inequality to 0 (τ − i i=1 t)γ+ |Dνf (t)|l dt.

The following extreme case of the theorem resembles [15, Proposition 4].

Theorem 5.11. Let the hypotheses of Theorem 5.10 be satisﬁed, but let p = 1 and q = ∞. Then 0

where γ :=

"l i=1

x

l

|Dμif (τ )| dτ ≤ C4 xγ+l+1 Dνf l∞ ,

(5.30)

i=1

αi and

C4 = C4 (ν, μ) :=

1 (γ + 1)(γ + l + 1)

&l i=1

Γ(ν − μi )

.

(5.31)

5.2 Main Results

Proof. As in the proof of Theorem 5.10 we have l i=1

1

|Dμif (τ )| ≤ &l

l

i=1 Γ(ν − μi ) i=1

≤ &l

τ

l−1

i=1 Γ(ν − μi )

τ

|Dνf (t)|(τ − t)αi dt

0

Dνf l∞

τ

(τ − t)γ dt 0

τ γ+l Dνf l∞ . ≤ &l (γ + 1) i=1 Γ(ν − μi ) Integrating over [0, x] with respect to τ we obtain (5.30).

65

6 Opial-Type Inequalities Involving Canavati Fractional Derivatives of Two Functions and Applications

A wide variety of very general but basic Lp (1 ≤ p ≤ ∞)-form Opial-type inequalities [315] is established involving generalized Canavati fractional derivatives [17, 101] of two functions in diﬀerent orders and powers. The above rely on a generalization of Taylor’s formula for generalized Canavati fractional derivatives [17]. Several other concrete results of special interest are derived from the developed results. The sharpness of inequalities is established there. Finally applications of some of these special inequalities are given in establishing the uniqueness of solution and in giving upper bounds to solutions of initial value problems involving a very general system of two fractional diﬀerential equations. Also upper bounds to various fractional derivatives of the solutions that are involved in the above systems are presented. This treatment relies on [26].

6.1 Introduction Opial inequalities appeared for the ﬁrst time in [315] and since then many authors have dealt with them in diﬀerent directions and for various cases. For a complete recent account of activity in this ﬁeld see [4], and there still remains a very active area of research. One of the main attractions to these inequalities is their applications, especially to establishing uniqueness G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 67 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 6,

68

6. Canavati Fractional Derivatives of Two Functions

and upper bounds of the solution of initial value problems in diﬀerential equations. The author was the ﬁrst to present Opial inequalities involving fractional derivatives of functions [15, 17] with applications to fractional diﬀerential equations. Fractional derivatives come up naturally in a number of ﬁelds, especially in physics; see the recent book [197]. Some topics include fractional kinetics of Hamiltonian chaotic systems, polymer physics and rheology, regular variation in thermodynamics, biophysics, fractional time evolution, and fractal time series among others. One also deals there with stochastic fractionaldiﬀerence equations and fractional diﬀusion equations. Great applications of these can be found in the study of DNA sequences. Other fractional diﬀerential equations arise in the study of suspensions, coming from the ﬂuid dynamical modeling of certain blood ﬂow phenomena. An excellent account of the study of fractional diﬀerential equations can be found in the recent book [333]. One can also have applications of fractional calculus to viscoelasticity, Bode’s analysis of feedback ampliﬁers, capacitor theory, electrical circuits, electronanalytical chemistry, biology, control theory, ﬁtting of experimental data, and fractional-order physics. The study of fractional diﬀerential equations ranges from the very theoretical topics of existence and uniqueness of solution to ﬁnding numerical solutions. The study of fractional calculus started in 1695 with L’Hospital and Leibniz, and was continued later by Fourier in 1822 and Abel in 1823, and continues to our day in an increased fashion due to its many applications and the necessity of dealing with fractional phenomena and structures. So this ﬁeld is keeping a lot of people active and interested. In this chapter the author is greatly motivated and inspired by the very important article [3]. Of course there the authors are dealing with other kinds of derivative. So here the author continues his study of fractional Opial inequalities now involving two diﬀerent functions and produces a wide variety of corresponding results with important applications to systems of two fractional diﬀerential equations. Dealing with two functions makes the study more complicated and involved. We start in Section 6.2 with preliminaries, we continue in Section 6.3 with the main results and we ﬁnish in Section 6.4 with the applications. To give a ﬂavor to the reader of the kind of inequalities we are dealing with, we brieﬂy mention

b

q(w)[|(Daγ 1 f1 )(w)|λα |(Daν f1 )(w)|λν + |(Daγ 1 f2 )(w)|λα |(Daν f2 )(w)|λν ]dw a

≤ C(a, b, q(w), γ 1 , ν, λα , λν , p(w), p) ' (((λα +λν )/p) b ν p ν p · p(w)[|(Da f1 )(w)| + |(Da f2 )(w)| ]dw , a

6.2 Preliminaries

69

for certain continuous functions f1 , f2 , p(w), q(w) on [a, b] , all exponents and orders are fractional, and so on. Furthermore, one system of fractional diﬀerential equations we are working on brieﬂy looks like (Daν fj )(t) = Fj (t, {(Daγ i f1 )(t)}ri=1 , {(Daγ i f2 )(t)}ri=1 ), all t ∈ [a, b], (i)

for j = 1, 2, and with fj (a) = aij ∈ R, i = 0, 1, . . . , n − 1, where n := [ν], ν ≥ 2, and so on. In the literature there are many diﬀerent deﬁnitions of fractional derivatives, some of them being equivalent; see [197, 333]. In this chapter we use one of the most recent due to J. Canavati [101], generalized in [15] and [17] by the author. One of the advantages of the Canavati fractional derivative is that in applications to fractional initial value problems we need only n initial conditions, as with the ordinary derivative case, whereas with other deﬁnitions of fractional derivatives we need n + 1 or more conditions; see [333].

6.2 Preliminaries In the sequel we follow [101]. Let g ∈ C([0, 1]). Let ν be a positive number, n := [ν], and α := ν − n (0 < α < 1). Deﬁne x 1 (Jν g)(x) := (x − t)ν−1 g(t)dt, 0 ≤ x ≤ 1, (6.1) Γ(ν) 0 the Riemann–Liouville integral, where Γ is the gamma function. We deﬁne the subspace C ν ([0, 1]) of C n ([0, 1]) as follows. C ν ([0, 1]) := {g ∈ C n ([0, 1]) : J1−α Dn g ∈ C 1 ([0, 1])}, where D := d/dx. So for g ∈ C ν ([0, 1]), we deﬁne the Canavati ν-fractional derivative of g as (6.2) Dν g := DJ1−α Dn g. When ν ≥ 1 we have the fractional Taylor’s formula t2 tn−1 + · · · + g (n−1) (0) 2! (n − 1)! for all t ∈ [0, 1].

g(t) = g(0) + g (0)t + g (0) + (Jν Dν g)(t),

(6.3)

When 0 < ν < 1 we ﬁnd g(t) = (Jν Dν g)(t),

for all t ∈ [0, 1].

(6.4)

70

6. Canavati Fractional Derivatives of Two Functions

Next we carry the above notions over to arbitrary [a, b] ⊆ R (see [17]). Let x, x0 ∈ [a, b] such that x ≥ x0 , where x0 is ﬁxed. Let f ∈ C([a, b]) and deﬁne x 1 x0 (x − t)ν−1 f (t)dt, x0 ≤ x ≤ b, (6.5) (Jν f )(x) := Γ(ν) x0 the generalized Riemann–Liouville integral. We deﬁne the subspace Cxν0 ([a, b]) of C n ([a, b]): x0 Cxν0 ([a, b]) := {f ∈ C n ([a, b]) : J1−α Dn f ∈ C 1 ([x0 , b])}.

For f ∈ Cxν0 ([a, b]), we deﬁne the generalized ν-fractional derivative of f over [x0 , b] as x0 Dxν0 f := DJ1−α f (n) (f (n) := Dn f ). (6.6) Notice that x0 (J1−α f (n) )(x)

1 = Γ(1 − α)

x

(x − t)−α f (n) (t)dt

x0

exists for f ∈ Cxν0 ([a, b]). We recall the following fractional generalization of Taylor’s formula (see [17, 101]). Theorem 6.1. Let f ∈ Cxν0 ([a, b]), x0 ∈ [a, b] ﬁxed. (i) If ν ≥ 1 then f (x) = f (x0 ) + f (x0 )(x − x0 ) + f (x0 ) f (n−1) (x0 )

(x − x0 )2 + ···+ 2

(x − x0 )n−1 + (Jνx0 Dxν0 f )(x), for all x ∈ [a, b] : x ≥ x0 . (6.7) (n − 1)!

(ii) If 0 < ν < 1 then f (x) = (Jνx0 Dxν0 f )(x), for all x ∈ [a, b] : x ≥ x0 .

(6.8)

We make Remark 6.2. (1) (Dxn0 f ) = f (n) , n ∈ N. (2) Let f ∈ Cxν0 ([a, b]), ν ≥ 1, and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Then by (6.7) f (x) = (Jνx0 Dxν0 f )(x).

6.2 Preliminaries

That is, 1 f (x) = Γ(ν)

71

x

x0

(x − t)ν−1 (Dxν0 f )(t)dt,

(6.9)

for all x ∈ [a, b] with x ≥ x0 . Notice that (6.9) is true also when 0 < ν < 1. We need from [17] Lemma 6.3. Let f ∈ C([a, b]), μ, ν > 0. Then x0 Jμx0 (Jνx0 f ) = Jμ+ν (f ).

(6.10)

We also make Remark 6.4. Let ν ≥ 1, γ ≥ 0, be such that ν − γ ≥ 1, so that γ < ν. Call n := [ν], α := ν − n ; m := [γ], ρ := γ − m. Note that ν − m ≥ 1 and n − m ≥ 1. Let f ∈ Cxν0 ([a, b]) be such that f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Hence by (6.7) f (x) = (Jνx0 Dxν0 f )(x), for all x ∈ [a, b] : x ≥ x0 . Therefore by Leibnitz’s formula and Γ(p + 1) = pΓ(p), p > 0, we get that x0 Dxν0 f )(x), for all x ≥ x0 . f (m) (x) = (Jν−m

(6.11)

It follows that f ∈ Cxγ0 ([a, b]) and thus x0 f (m) )(x) exists for all x ≥ x0 . (Dxγ0 f )(x) := (DJ1−ρ

(6.12)

By the use of (6.11) we have on [x0 , b], x0 x0 x0 x0 x0 (f (m) ) = J1−ρ (Jν−m Dxν0 f ) = (J1−ρ ◦ Jν−m )(Dxν0 f ) J1−ρ x0 x0 = Jν−m+1−ρ (Dxν0 f ) = Jν−γ+1 (Dxν0 f ),

by (6.10). That is, x0 (J1−ρ f (m) )(x)

1 = Γ(ν − γ + 1)

x

x0

(x − t)ν−γ (Dxν0 f )(t)dt.

Therefore (Dxγ0 f )(x)

=

x0 D((J1−ρ f (m) )(x))

1 · = Γ(ν − γ)

x

x0

(x − t)(ν−γ)−1 (Dxν0 f )(t)dt;

(6.13) x0 (Dxν0 f ))(x) and is continuous in x on [x0 , b]. hence (Dxγ0 f )(x) = (Jν−γ In particular when ν ≥ 2 we have x f )(x) = (Dxν0 f )(t)dt, x ≥ x0 . (6.14) (Dxν−1 0 x0

That is,

f ) = Dxν0 f, (Dxν−1 0

(Dxν−1 f )(x0 ) = 0. 0

72

6. Canavati Fractional Derivatives of Two Functions

6.3 Main Results We present our ﬁrst main result here: Theorem 6.5. Let ν ≥ 1, γ 1 , γ 2 ≥ 0 be such that ν − γ 1 ≥ 1, ν − γ 2 ≥ 1, (i) (i) and f1 , f2 ∈ Cxν0 ([a, b]) with f1 (x0 ) = f2 (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b] : x ≥ x0 . Consider also p(t) > 0, and q(t) ≥ 0 continuous functions on [x0 , b]. Let λν > 0 and λα , λβ ≥ 0 such that λν < p, where p > 1. Set

w

Pk (w) :=

(w − t)(ν−γ k −1)p/p−1 (p(t))−1/p−1 dt, k = 1, 2, x0 ≤ x ≤ b;

x0

(6.15) A(w) :=

−λν /p

· (P2 (w)) (p(w)) q(w) · (P1 (w)) (Γ(ν − γ 1 ))λα · (Γ(ν − γ 2 ))λβ λα (p−1/p)

λβ (p−1/p)

(6.16)

(p−λν )/p

x

A(w)p/p−λν dw

A0 (x) :=

,

,

(6.17)

x0

and

⎧ ⎨ 21−((λα +λν )/p) , δ 1 :=

⎩

if λα + λν ≤ p, (6.18) if λα + λν ≥ p.

1,

If λβ = 0, we obtain that

x

x0

q(w) |(Dxγ01 f1 )(w)|λα · |(Dxν0 f1 )(w)|λν +

+|(Dxγ01 f2 )(w)|λα · |(Dxν0 f2 )(w)|λν dw λν /p λν · δ1 ≤ A0 (x)|λβ =0 · λα + λν *((λα +λν )/p) ) x

· p(w) |(Dxν0 f1 )(w)|p + |(Dxν0 f2 )(w)|p dw (6.19) . x0

Proof. From (6.13) we have (Dxγ0k fj )(w)

1 = Γ(ν − γ k )

w

x0

(w − t)ν−γ k −1 (Dxν0 fj )(t)dt,

for k = 1, 2, j = 1, 2, and for all x0 ≤ w ≤ b.

6.3 Main Results

73

Next applying H¨ older’s inequality with indices p, p/(p−1) we get γ

|(Dx0k fj )(w)| ≤ ≤

1 Γ(ν − γ k )

w x0

(w − t)ν−γ k −1 (p(t))−1/p (p(t))1/p |(Dxν 0 fj )(t)|dt

w p/(p−1) (p−1)/p 1 (w − t)ν−γ k −1 (p(t))−1/p dt Γ(ν − γ k ) x0 1/p w p(t)|(Dxν 0 fj )(t)|p dt x0

=

1 (P (w))p−1/p Γ(ν − γ k ) k

w x0

1/p p(t)|(Dxν 0 fj )(t)|p dt .

That is, it holds |(Dxγ0k fj )(w)|

1 (Pk (w))p−1/p ≤ Γ(ν − γ k )

1/p

w

x0

p(t)|(Dxν0 fj )(t)|p dt

. (6.20)

Put

w

p(t)|(Dxν0 fj )(t)|p dt,

zj (w) := x0

thus

zj (w) = p(w)|(Dxν0 fj )(w)|p ,

zj (x0 ) = 0,

j = 1, 2.

Hence we have |(Dxγ0k fj )(w)| ≤ and

1 (Pk (w))(p−1)/p (zj (w))1/p , Γ(ν − γ k )

|(Dxν0 fj )(w)|λν = p(w)−λν /p (zj (w))λν /p ,

j = 1, 2.

Therefore we obtain q(w)|(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν 1 1 ≤ q(w) (P1 (w))λα (p−1/p) (z1 (w))λα /p (Γ(ν − γ 1 ))λα (Γ(ν − γ 2 ))λβ (P2 (w))λβ (p−1/p) (z2 (w))λβ /p (p(w))−λν /p (z1 (w))λν /p = A(w)(z1 (w))λα /p (z2 (w))λβ /p (z1 (w))λν /p . Consequently, by another H¨ older’s inequality application, we ﬁnd (by p/λν > 1) x q(w)|(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν dw x0

)

x

≤ A0 (x)

λα /λν

(z1 (w)) x0

(z2 (w))λβ /λν z1 (w)dw

*λν /p . (6.21)

74

6. Canavati Fractional Derivatives of Two Functions

Similarly one ﬁnds x q(w)|(Dxγ02 f1 )(w)|λβ |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν dw x0

)

x

≤ A0 (x)

λβ /λν

(z1 (w))

(z2 (w))λα /λν z2 (w)dw

*λν /p . (6.22)

x0

Taking λβ = 0 and adding (6.21) and (6.22) we obtain

, + γ γ q(w) |(Dx01 f1 )(w)|λα |(Dxν 0 f1 )(w)|λν + |(Dx01 f2 )(w)|λα |(Dxν 0 f2 )(w)|λν dw

x x0

≤ A0 (x)|λβ =0

%)

x

(z1 (w))λα /λν z1 (w)dw

x0

)

x

+

(z2 (w))λα /λν z2 (w)dw

*λν /p

*λν /p -

x0

λν /p λν λα + λν (λα +λν )/p λν /p % x p(t)|(Dxν0 f1 )(t)|p dt

= A0 (x)|λβ =0 (z1 (x))(λα +λν )/p + (z2 (x))(λα +λν )/p

= A0 (x)|λβ =0

λν λα + λν

x0

x

+ x0

(λα +λν )/p -

p(t)|(Dxν0 f2 )(t)|p dt

=: (∗).

In this chapter we frequently use the following basic inequalities. ≤ (a + b)r ≤ ar + br , a, b ≥ 0, 0 ≤ r ≤ 1, (6.23) r r−1 r r ≤ (a + b) ≤ 2 (a + b ), a, b ≥ 0, r ≥ 1. (6.24)

2r−1 (ar + br ) ar + br

Finally using (6.23), (6.24), and (6.18) we get λν /p λν (∗) ≤ A0 (x)|λβ =0 · · δ1 λa + λν (λα +λν )/p x

ν p ν p · p(t) |(Dx0 f1 )(t)| + |(Dx0 f2 )(t)| dt . x0

Inequality (6.19) has been established.

It follows the counterpart of the last theorem. Theorem 6.6. All here are as in Theorem 6.5. Denote λ /λ 2 β ν − 1, if λβ ≥ λν , δ 3 := 1, if λβ ≤ λν .

(6.25)

6.3 Main Results

75

If λα = 0, then it holds

, + γ γ q(w) |(Dx02 f2 )(w)|λβ · |(Dxν 0 f1 )(w)|λν + |(Dx02 f1 )(w)|λβ |(Dxν 0 f2 )(w)|λν dw

x x0

λν /p λν λ /p δ3 ν (6.26) λβ + λν ((λν +λβ )/p) x

ν p ν p · p(w) |(Dx0 f1 )(w)| + |(Dx0 f2 )(w)| dw , all x0 ≤ x ≤ b. ≤ (A0 (x)|λα =0 ) 2(p−λν )/p

x0

Proof. When λα = 0 from (6.21) and (6.22) we obtain x q(w)|(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν dw x0

)

x

≤ (A0 (x)|λα =0 )

(z2 (w))λβ /λν z1 (w)dw

*λν /p ,

(6.27)

,

(6.28)

x0

and

x

x0

q(w)|(Dxγ02 f1 )(w)|λβ |(Dxν0 f2 )(w)|λν dw )

x

≤ (A0 (x)|λα =0 )

(z1 (w))λβ /λν z2 (w)dw

*λν /p

x0

all x0 ≤ x ≤ b. Adding (6.27) and (6.28) we get x q(w) |(Dxγ02 f2 )(w)|λβ · |(Dxν0 f1 )(w)|λν + |(Dxγ02 f1 )(w)|λβ · x0

|(Dxν0 f2 )(w)|λν dw ≤ (A0 (x)|λα =0 ) %) *λν /p ) x *λν /p x (z2 (w))λβ /λν z1 (w)dw + (z1 (w))λβ /λν z2 (w)dw x0

x0

≤ (A0 (x)|λα =0 ) · 2p−λν /p · (M (x))λν /p =: (∗), by 0 < λν /p < 1 and (6.23), where x (z2 (w))λβ /λν z1 (w) + (z1 (w))λβ /λν z2 (w))dw. M (x) := x0

Call δ 2 :=

1, 21−(λβ /λν ) ,

if λβ ≥ λν , if λβ ≤ λν .

(6.29)

(6.30)

76

6. Canavati Fractional Derivatives of Two Functions

Next we work on M (x). We have that x M (x) = (z1 (w))λβ /λν + (z2 (w))λβ /λν (z1 (w) + z2 (w))dw x x 0+ , (z1 (w))λβ /λν z1 (w) + (z2 (w))λβ /λν z2 (w) dw − x0

(by (6.23), (6.24))

x

δ2 (z1 (w) + z2 (w))λβ /λν (z1 (w) + z2 (w)) dw x0 + , λν (z1 (x))((λν +λβ )/λν ) + (z2 (w))(λν +λβ /λν ) − λβ + λν λ λν ν ((λν +λβ )/λν ) = δ 2 (z1 (x) + z2 (x)) − λν + λβ λβ + λν + , ((λν +λβ )/λν ) ((λν +λβ )/λν ) (z1 (x)) + (z2 (w)) ) * λν δ 2 (z1 (x) + z2 (x))((λν +λβ )/λν ) − = (z1 (x))((λν +λβ )/λν ) + (z2 (x))((λν +λβ )/λν ) λβ + λν + (6.24) λν δ 2 2((λν +λβ )/λν )−1 (z1 (x))((λν +λβ )/λν ) ≤ λβ + λν , + (z2 (x))((λν +λβ )/λν ) − (z1 (x))((λν +λβ )/λν ) + (z2 (x))((λν +λβ )/λν ) λν = (δ 2 2λβ /λν − 1) λβ + λν , + (z1 (x))((λν +λβ )/λν ) + (z2 (x))((λν +λβ )/λν ) notice δ 3 = δ 2 2λβ /λν − 1 (6.24) λν ≤ δ 3 (z1 (x) + z2 (x))((λν +λβ )/λν ) . λβ + λν ≤

That is, we present that λν M (x) ≤ δ 3 (z1 (x) + z2 (x))((λν +λβ )/λν ) . λβ + λν

(6.31)

Consequently by (6.29) and (6.31) we get (∗) ≤ (A0 (x)|λα =0 ) 2p−λν /p = (A0 (x)|λα =0 ) 2p−λν /p

x

·

p(t) x0

λν λβ + λν λν λβ + λν

|(Dxν0 f1 )(t)|p

We have established (6.26).

+

λν /p

λ /p

δ 3 ν (z1 (x) + z2 (x))((λν +λβ )/p) λν /p

λ /p

δ3 ν

|(Dxν0 f2 )(t)|p

((λν +λβ )/p) dt

.

6.3 Main Results

77

The full case when λα , λβ = 0 follows. Theorem 6.7. All here are as in Theorem 6.5. Denote ((λ +λ )/λ ) 2 α β ν − 1, if λα + λβ ≥ λν , γ˜ 1 := 1, if λα + λβ ≤ λν , and

γ˜ 2 :=

(6.32)

if λα + λβ + λν ≥ p, if λα + λβ + λν ≤ p.

1, 21−(λα +λβ +λν /p)

(6.33)

Then x q(w) |(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν x0

+ |(Dxγ02 f1 )(w)|λβ |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν dw λν /p + , λν λλαν /p γ˜ 2 + 2p−λν /p (˜ γ 1 λβ )λν /p ≤ A0 (x) (λα + λβ )(λα + λβ + λν ) x ((λα +λβ +λν )/p) · p(w)(|(Dxν0 f1 )(w)|p + |(Dxν0 f2 )(w)|p )dw , (6.34) x0

all x0 ≤ x ≤ b. Proof. Here we use the basic inequality ab ≤

bq ap + , p q

(6.35)

where a, b ≥ 0 and p, q > 1 such that 1/p + 1/q = 1. From (6.21) we obtain x q(w)|(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν dw x0

)

x

≤ A0 (x) x0

λα λα + λβ

(z1 (w))((λα +λβ )/λν ) +

*λν /p λβ ((λα +λβ )/λν ) (z2 (w)) z1 (w)dw λα + λβ ' λν /p (6.23) λα λν (z1 (x))((λα +λβ +λν )/λν ) ≤ A0 (x) λα + λβ (λα + λβ + λν )

x

+ x0

λβ λα + λβ

(z2 (w))((λα +λβ )/λν ) z1 (w)dw

λν /p ( .

78

6. Canavati Fractional Derivatives of Two Functions

Therefore

x

q(w)|(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν dw % λν /p λν λα ≤ A0 (x) · (z1 (x))((λα +λβ +λν )/p) (λα + λβ )(λα + λβ + λν ) λν /p λν /p x λβ ((λα +λβ )/λν ) .(6.36) + (z2 (w)) z1 (w)dw λα + λβ x0 x0

Similarly using (6.22) we ﬁnd x q(w)|(Dxγ02 f1 )(w)|λβ |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν dw x0 % λν /p λν λα ≤ A0 (x) (z2 (x))((λα +λβ +λν )/p) (λα + λβ )(λα + λβ + λν ) λν /p λν /p x λβ ((λα +λβ )/λν ) + (z1 (w)) z2 (w)dw . (6.37) λα + λβ x0 Next adding (6.36) and (6.37) we observe x Ω := q(w) |(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν x0

+ |(Dxγ02 f1 )(w)|λβ |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν dw % λν /p λν λα ≤ A0 (x) (λα + λβ )(λα + λβ + λν ) + , (z1 (x))((λα +λβ +λν )/p) + (z2 (x))((λα +λβ +λν )/p) λν /p λν /p ' x λβ (λα +λβ /λν ) + (z1 (w)) z2 (w)dw λα + λβ x0 ( x

+

(z2 (w))((λα +λβ )/λν ) z1 (w)dw

x0 (6.23)

≤ A0 (x)

+

%

λν /p

λν λα (λα + λβ )(λα + λβ + λν )

λν /p

(z1 (x))((λα +λβ +λν )/p) + (z2 (x))((λα +λβ +λν )/p) λν /p λβ p−λν /p λν /p + ·2 · (Γ(x)) , λα + λβ

,

6.3 Main Results

79

where

x

Γ(x) :=

(z1 (w))((λα +λβ )/λν ) z2 (w) + (z2 (w))((λα +λβ )/λν ) z1 (w) dw.

x0

Again, see (6.30) and (6.31), we get γ˜ 1 λν Γ(x) ≤ (z1 (x) + z2 (x))((λα +λβ +λν )/λν ) . λα + λβ + λν Hence by (6.38) we obtain % Ω ≤ A0 (x) + +

λβ λα + λβ

λν /p

2p−λν /p

≤ A0 (x) +2

λβ λα + λβ

γ˜ 1 λν λα + λβ + λν

)

x

·

p(t)

x0

λν /p

λν /p

γ˜ 1 λν λα + λβ + λν

-

|(Dxν0 f1 )(t)|p

+

λν /p -

λν /p +

,

(z1 (x) + z2 (x))((λα +λβ +λν )/p)

λν λα (λα + λβ )(λα + λβ + λν )

λν (λα + λβ )(λα + λβ + λν )

= A0 (x)

λν /p

(z1 (x))((λα +λβ +λν )/p) + (z2 (x))((λα +λβ +λν )/p)

%

p−λν /p

λν λα (λα + λβ )(λα + λβ + λν )

(6.38)

λν /p γ˜ 2

(z1 (x)+z2 (x))((λα +λβ +λν )/p)

λλαν /p γ˜ 2 + 2p−λν /p (˜ γ 1 λβ )λν /p

|(Dxν0 f2 )(t)|p

,

*((λα +λβ +λν )/p) dt

.

We have proved Ω ≤ A0 (x) ) ·

x x0

λν (λα + λβ )(λα + λβ + λν )

λν /p +

λ /p

λαν

p(w) |(Dxν 0 f1 )(w)|p + |(Dxν 0 f2 )(w)|p dw

We have established (6.34).

γ˜ 2 + 2(p−λν )/p (˜ γ 1 λβ )λν /p

,

*((λα +λβ +λν )/p) .

A special important case follows. Theorem 6.8. Let ν ≥ 2 and γ 1 ≥ 0 such that ν − γ 1 ≥ 2. Let f1 , f2 ∈ (i) (i) Cxν0 ([a, b]) with f1 (x0 ) = f2 (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here

80

6. Canavati Fractional Derivatives of Two Functions

x, x0 ∈ [a, b] : x ≥ x0 . Consider also p(t) > 0, and q(t) ≥ 0 continuous functions on [x0 , b]. Let λα ≥ 0, 0 < λα+1 < 1, and p > 1. Denote λ /λ 2 α α+1 − 1, if λα ≥ λα+1 , θ3 := , (6.39) 1, if λα ≤ λα+1 (1−λα+1 ) λα+1 x θ3 λα+1 (1/1−λα+1 ) (q(w)) dw , L(x) := 2 λα + λα+1 x0 and

x

P1 (x) :=

(x − t)(ν−γ 1 −1)p/(p−1) (p(t))−1/(p−1) dt,

x0

T (x) :=

L(x) ·

P1 (x)(p−1/p) Γ(ν − γ 1 )

(6.40)

(6.41)

(λα +λα+1 ) ,

(6.42)

and

Then

x

x0

ω 1 := 2((p−1)/p)(λα +λα+1 ) ,

(6.43)

Φ(x) := T (x)ω 1 .

(6.44)

q(w) |(Dxγ01 f1 )(w)|λα |(Dxγ01 +1 f2 )(w)|λα+1

(6.45)

+|(Dxγ01 f2 )(w)|λα |(Dxγ01 +1 f1 )(w)|λα+1 dw ) x *((λα +λα+1 )/p) ≤ Φ(x) p(w) |(Dxν0 f1 )(w)|p + |(Dxν0 f2 )(w)|p dw , x0

all x0 ≤ x ≤ b. Proof. For convenience we set γ 2 := γ 1 + 1. From (6.13) we obtain w 1 |(Dxγ0k fj )(w)| ≤ (w − t)ν−γ k −1 |(Dxν0 fj )(t)|dt =: gj,γ k (w), Γ(ν − γ k ) x0 (6.46) where j = 1, 2, k = 1, 2, all x0 ≤ w ≤ b. We observe that γ (6.47) (Dx01 fj (x) = (Dxγ01 +1 fj )(x) = (Dxγ02 fj )(x), all x0 ≤ x ≤ b. And also (gj,γ 1 (w)) = gj,γ 2 (w); Notice that if ν − γ 2 = 1, then gj,γ 2 (w) =

gj,γ k (x0 ) = 0.

w

x0

|(Dxν0 fj )(t)|dt.

(6.48)

6.3 Main Results

81

Next we apply H¨ older’s inequality with indices 1/λα+1 , 1/(1 − λα+1 ); we obtain x

q(w)|(Dxγ01 f1 )(w)|λα |(Dxγ01 +1 f2 (w)|λα+1 dw

x0

x

q(w)(g1,γ 1 (w))λα ((g2,γ 1 (w)) )λα+1 dw

≤ x0

(1−λα+1 )

x

≤

(6.49)

(q(w))(1/1−λα+1 ) dw x0

x

x0

(g1,γ 1 (w))λα /λα+1 (g2,γ 1 (w)) dw

λα+1 .

Similarly we get

x

x0

q(w)|(Dxγ01 f2 )(w)|λα |(Dxγ01 +1 f1 (w)|λα+1 dw

(1−λα+1 )

x

≤

(q(w))(1/1−λα+1 ) dw

(6.50)

x0 x

x0

(g2,γ 1 (w))λα /λα+1 (g1,γ 1 (w)) dw

λα+1 .

Adding (6.49) and (6.50) we observe x q(w) |(Dxγ01 f1 )(w)|λα |(Dxγ01 +1 f2 )(w)|λα+1 x0

+ |(Dxγ01 f2 )(w)|λα |(Dxγ01 +1 f1 )(w)|λα+1 dw

(1−λα+1 )

x

≤

(q(w))(1/1−λα+1 ) dw x0

'

x λα /λα+1

x0

(g1,γ 1 (w))

x λα /λα+1

+ x0 (6.23)

≤

(g2,γ 1 (w))

λα+1

(g2,γ 1 (w)) dw

λα+1 (

(g1,γ 1 (w)) dw

x (1−λα+1 ) 2 (q(w))(1/1−λα+1 ) dw x0

) ·

x x0

* *λα+1 [(g1,γ 1 (w))λα /λα+1 (g2,γ 1 (w)) + (g2,γ 1 (w))λα /λα+1 (g1,γ 1 (w)) dw

82

6. Canavati Fractional Derivatives of Two Functions

(note (6.30) and the proof of (6.31); accordingly here we have) ≤

x (1−λα+1 ) λα+1 λα+1 θ3 2 (q(w))(1/1−λα+1 ) dw · λα + λα+1 x0 (λα +λα+1 ) · g1,γ 1 (x) + g2,γ 1 (x)

(λα +λα+1 ) = L(x) g1,γ 1 (x) + g2,γ 1 (x) x L(x) = (x − t)ν−γ 1 −1 (p(t))−1/p (p(t))1/p |(Dxν0 f1 )(t)| (λ +λ ) (Γ(ν − γ 1 )) α α+1 x0

$(λα +λα+1 ) + |(Dxν0 f2 )(t)| dt (applying H¨ older’s inequality with indices p/p−1 and p we ﬁnd) L(x) (Γ(ν − γ 1 ))(λα +λα+1 ) x (p−1/p)(λα +λα+1 ) · (x − t)(ν−γ 1 −1)p/p−1 (p(t))−1/p−1 dt

≤

x0

x

· x0

p p(t) Dxν0 f1 (t) + |(Dxν0 f2 )(t)| dt )

*(λα +λα+1 /p)

x

= T (x) · )

(λα +λα+1 /p)

x0

p(t)(|(Dxν0 f1 )(t)| + |(Dxν0 f2 )(t)|)p dt *(λα +λα+1 /p)

x

≤ Φ(x) · x0

p(t)(|(Dxν0 f1 )(t)|p

We have proved (6.45).

+

|(Dxν0 f2 )(t)|p )dt

.

Next we treat the case of exponents λβ = λα + λν . Theorem 6.9. All here are as in Theorem 6.5. Consider the special case λβ = λα + λν . Denote T˜(x) := A0 (x)

λν λα + λν

λν /p

2(p−2λα −3λν )/p .

Then

x

x0

q(w) |(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λα +λν |(Dxν0 f1 )(w)|λν

+|(Dxγ02 f1 )(w)|λα +λν |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν dw

(6.51)

6.3 Main Results

2((λα +λν )/p)

x

≤ T˜(x) x0

83

p(w)(|(Dxν0 f1 )(w)|p + |(Dxν0 f2 )(w)|p )dw

, (6.52)

all x0 ≤ x ≤ b. Proof. We apply (6.21) and (6.22) for λβ = λα + λν and add to ﬁnd x q(w) |(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λα +λν |(Dxν0 f1 )(w)|λν x0

+|(Dxγ02 f1 )(w)|λα +λν |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν dw %) *λν /p x ≤ A0 (x) (z1 (w))λα /λν (z2 (w))(λα /λν )+1 z1 (w)dw x0

)

x

+

(z1 (w))(λα /λν )+1 (z2 (w))λα /λν z2 (w)dw

*λν /p -

x0

(6.23)

x

≤ A0 (x)21−(λν /p)

+

(z1 (w))λα /λν (z2 (w))(λα /λν )+1 z1 (w)+

x0

, λν /p (z1 (w))λα /λν +1 (z2 (w))λα /λν z2 (w) dw

x

1−λν /p

λα /λν

= A0 (x)2

(z1 (w)z2 (w)) x0

1−λν /p

= A0 (x)2

[z2 (w)z1 (w)

+

z1 (w)z2 (w)]dw

x

·

λα /λν

(z1 (w)z2 (w)) x0

λν /p

λν /p

(z1 (w)z2 (w)) dw

λν /p (z1 (x)z2 (x))(λα /λν )+1 = A0 (x)2 λα λν + 1 λν /p λν = A0 (x)2p−λν /p (z1 (x)z2 (x))(λα +λν )/p λα + λν λν /p (λα +λν /p) λν (z1 (x) + z2 (x))2 p−λν /p ≤ A0 (x)2 λα + λν 4 1−λν /p

= T˜(x)(z1 (x) + z2 (x))2(λα +λν )/p

x

= T˜(x)

p(w) x0

|(Dxν0 f1 )(w)|p

We have established (6.52).

+

|(Dxν0 f2 )(w)|p

Next follow special cases of the above theorems.

2((λα +λν )/p) dw

.

84

6. Canavati Fractional Derivatives of Two Functions

Corollary 6.10. (to Theorem 6.5; λβ = 0, p(t) = q(t) = 1) It holds x

|(Dxγ01 f1 )(w)|λα |(Dxν0 f1 )(w)|λν + |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν dw x0

((λα +λν )/p)

x

≤ C1 (x) · x0

(|(Dxν0 f1 )(w)|p

+

|(Dxν0 f2 )(w)|p )dw

(6.53)

all x0 ≤ x ≤ b, where C1 (x) := A0 (x)|λβ =0 · δ 1 :=

λν λα + λν

λν /p · δ1 ,

if λα + λν ≤ p, if λα + λν ≥ p

21−(λα +λν /p) , 1,

(6.54)

.

We ﬁnd that ·

A0 (x)|λβ =0 =

(p − 1)((λα p−λα )/p) (Γ(ν − γ 1 ))λα (νp − γ 1 p − 1)((λα p−λα )/p)

(p − λν )(p−λν /p) (λα νp − λα γ 1 p − λα + p − λν )(p−λν /p)

·(x−x0 )((λα νp−λα γ 1 p−λα +p−λν )/p). (6.55)

Proof. Here we need to calculate A0 (x)|λβ =0 . From (6.17) we have

x

A0 (x)|λβ =0 =

p/p−λν A(w)|λβ =0 dw

(p−λν )/p ,

x0

where from (6.16) we ﬁnd A(w)|λβ =0 = and here from (6.15) it is

w

P1 (w) =

(P1 (w))λα ((p−1)/p) , (Γ(ν − γ 1 ))λα

(w − t)((ν−γ 1 −1)/(p−1))p dt.

x0

Therefore we obtain

P1 (w) =

p−1 νp − γ 1 p − 1

(w − x0 )(νp−γ 1 p−1)/(p−1) ,

and A(w)|λβ =0

=

p−1 νp − γ 1 p − 1

(λα p−λα /p) ·

·(w − x0 )(λα νp−λα γ 1 p−λα )/p .

1 (Γ(ν − γ 1 ))λα

6.3 Main Results

85

That is, A0 (x)|λβ =0

(p − 1)(λα p−λα /p) (νp − γ 1 p − 1)(λα p−λα /p) (Γ(ν − γ 1 ))λα ((p−λν )/p) x (λα νp−λα γ 1 p−λα /p−λν ) (w − x0 ) dw ·

=

x0

(p − 1)(λα p−λα /p) (Γ(ν − γ 1 ))λα (νp − γ 1 p − 1)((λα p−λα )/p) (p − λν )(p−λν /p) · (λα νp − λα γ 1 p − λα + p − λν )((p−λν )/p)

=

·(x − x0 )(λα νp−λα γ 1 p−λα +p−λν /p) . Corollary 6.11. (to Theorem 6.5; λβ = 0, p(t) = q(t) = 1 , λα = λν = 1, p = 2). In detail: Let ν ≥ 1, γ 1 , γ 2 ≥ 0, such that ν − γ 1 ≥ 1, ν − γ 2 ≥ 1 and f1 , f2 ∈ (i) (i) Cxν0 ([a, b]) with f1 (x0 ) = f2 (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b] : x ≥ x0 . Then x

|(Dxγ01 f1 )(w)| |(Dxν0 f1 )(w)| + |(Dxγ01 f2 )(w)| |(Dxν0 f2 )(w)| dw x0

≤

x

x0

(x − x0 )(ν−γ 1 ) √ √ 2Γ(ν − γ 1 ) ν − γ 1 2ν − 2γ 1 − 1

[((Dxν0 f1 )(w))2 + ((Dxν0 f2 )(w))2 ]dw ,

all x0 ≤ x ≤ b. Proof. We apply Corollary 6.10. Here δ 1 = 1 and λν /p λν 1 =√ . λα + λν 2 Furthermore we have A0 (x)|λβ =0

=

1 √ Γ(ν − γ 1 ) 2ν − 2γ 1 − 1 1 · (x − x0 )(ν−γ 1 ) . · √ √ 2 ν − γ1

Finally we get C1 (x) =

(x − x0 )(ν−γ 1 ) √ . √ 2Γ(ν − γ 1 ) ν − γ 1 2ν − 2γ 1 − 1

(6.56)

86

6. Canavati Fractional Derivatives of Two Functions

Corollary 6.12. (to Theorem 6.6; λα = 0, p(t) = q(t) = 1). It holds x

|(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν + |(Dxγ02 f1 )(w)|λβ |(Dxν0 f2 )(w)|λν dw x0

x

≤ C2 (x) x0

|(Dxν0 f1 )(w)|p + |(Dxν0 f2 )(w)|p dw

((λν +λβ )/p) ,

(6.57)

all x0 ≤ x ≤ b, where C2 (x) := (A0 (x)|λα =0 ) 2(p−λν )/p

2λβ /λν − 1, 1,

δ 3 :=

λν λβ + λν

λν /p

λ /p

δ3 ν ,

(6.58)

if λβ ≥ λν , if λβ ≤ λν

.

We ﬁnd that (A0 (x)|λα =0 )

=

(p − 1)((λβ p−λβ )/p) (Γ(ν − γ 2 ))λβ (νp − γ 2 p − 1)((λβ p−λβ )/p) (p − λν )(p−λν /p) · (λβ νp − λβ γ 2 p − λβ + p − λν )((p−λν )/p)

·(x − x0 )(λβ νp−λβ γ 2 p−λβ +p−λν /p) . Proof. Here we need to calculate A0 (x)|λα =0 . From (6.17) we have

(p−λν )/p

x

A0 (x)|λα =0 =

(A(w)|λα =0 )

p/p−λν

dw

x0

where from (6.16) we ﬁnd A(w)|λα =0 =

(P2 (w))λβ ((p−1)/p) , (Γ(ν − γ 2 ))λβ

and here from (6.15) it is

w

P2 (w) =

(w − t)(ν−γ 2 −1)p/(p−1) dt.

x0

Therefore we obtain P2 (w) =

p−1 νp − γ 2 p − 1

(w − x0 )(νp−γ 2 p−1)/(p−1) ,

(6.59)

6.3 Main Results

87

and (p − 1)(λβ p−λβ /p) (w − x0 )(λβ νp−λβ γ 2 p−λβ /p) . (νp − γ 2 p − 1)(λβ p−λβ /p) (Γ(ν − γ 2 ))λβ

A(w)|λα =0 = That is,

A0 (x)|λα =0

=

(p − 1)(λβ p−λβ /p) (νp − γ 2 p − 1)(λβ p−λβ /p) (Γ(ν − γ 2 ))λβ ((p−λν )/p) x (λβ νp−λβ γ 2 p−λβ /p−λν ) (w − x0 ) dw ·

= ·

x0

(p − 1)(λβ p−λβ /p) (νp − γ 2 p − 1)(λβ p−λβ /p) (Γ(ν − γ 2 ))λβ

(p − λν )(p−λν /p) (λβ νp − λβ γ 2 p − λβ + p − λν )((p−λν )/p)

·(x − x0 )(λβ νp−λβ γ 2 p−λβ +p−λν /p) . Corollary 6.13. (to Theorem 6.6; λα = 0, p(t) = q(t) = 1, λβ = λν = 1, p = 2). In detail: Let ν ≥ 1, γ 1 , γ 2 ≥ 0, such that ν − γ 1 ≥ 1, ν − γ 2 ≥ 1, and f1 , (i) (i) f2 ∈ Cxν0 ([a, b]) with f1 (x0 ) = f2 (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b] : x ≥ x0 . Then x

|(Dxγ02 f2 )(w)| |(Dxν0 f1 )(w)| + |(Dxγ02 f1 )(w)| |(Dxν0 f2 )(w)| dw x0 (x − x0 )(ν−γ 2 ) ≤ √ √ √ 2Γ(ν − γ 2 ) ν − γ 2 2ν − 2γ 2 − 1 x ν 2 ν 2 ((Dx0 f1 )(w)) + ((Dx0 f2 (w)) dw , (6.60) · x0

all x0 ≤ x ≤ b. λ /p

Proof. We apply Corollary 6.12. Here δ 3 ν

Furthermore we have (A0 (x)|λα =0 ) =

λν λβ + λν

λν /p

= 1, 2p−λν /p =

√

2,

1 =√ . 2

(x − x0 )(ν−γ 2 ) √ √ √ 2 Γ(ν − γ 2 ) ν − γ 2 2ν − 2γ 2 − 1

.

88

6. Canavati Fractional Derivatives of Two Functions

Finally we get C2 (x) =

(x − x0 )(ν−γ 2 ) √ √ √ 2 Γ(ν − γ 2 ) ν − γ 2 2ν − 2γ 2 − 1

.

Corollary 6.14. (to Theorem 6.7; λα = λβ = λν = 1, p = 3, p(t) = q(t) = 1). It holds x [[|(Dxγ01 f1 )(w)||(Dxγ02 f2 )(w)||(Dxν0 f1 )(w)| x0

+|(Dxγ02 f1 )(w)||(Dxγ01 f2 )(w)||(Dxν0 f2 )(w)|]dw ≤ A0 (x) x √ 1 3 ν 3 ν 3 2+ √ (|(Dx0 f1 )(w)| + |(Dx0 f2 )(w)| )dw , · 3 6 x0

(6.61)

all x0 ≤ x ≤ b. Here A0 (x) = 4(x − x0 )(2ν−γ 1 −γ 2 ) . Γ(ν − γ 1 )Γ(ν − γ 2 )[3(3ν − 3γ 1 − 1)(3ν − 3γ 2 − 1)(2ν − γ 1 − γ 2 )]2/3 (6.62) Proof. We apply inequality (6.34). Here γ˜ 1 = 3, γ˜ 2 = 1, and

Furthermore

λν (λα + λβ )(λα + λβ + λν )

λν /p

1 . = √ 3 6

√ 3 λλαν /p γ˜ 2 + 2(p−λν )/p (˜ γ 1 λβ )λν /p = 1 + 12.

√ √ The product of the last two quantities is ( 3 2 + 1/ 3 6). It still remains to ﬁnd A0 (x) for this case. Here by (6.17),

2/3

x

A(w)3/2 dw

A0 (x) =

,

x0

and by (6.16), A(w) =

(P1 (w))2/3 (P2 (w))2/3 . Γ(ν − γ 1 )Γ(ν − γ 2 )

Also by (6.15) we have for k = 1, 2, w Pk (w) = (w − t)((3ν−3γ k −3)/2) dt. x0

6.3 Main Results

That is, Pk (w) =

2(w − x0 )3ν−3γ k −1/2 , (3ν − 3γ k − 1)

89

k = 1, 2,

and A(w) =

24/3 (w − x0 )(6ν−3γ 1 −3γ 2 −2/3) . Γ(ν − γ 1 )Γ(ν − γ 2 )(3ν − 3γ 1 − 1)2/3 (3ν − 3γ 2 − 1)2/3

Finally we ﬁnd A0 (x) =

24/3 Γ(ν − γ 1 )Γ(ν − γ 2 )(3ν − 3γ 1 − 1)2/3 (3ν − 3γ 2 − 1)2/3 2/3 x (w − x0 )((6ν−3γ 1 −3γ 2 −2)/2) dw ·

x0

=

4 · (x − x0 )(2ν−γ 1 −γ 2 ) Γ(ν − γ 1 )Γ(ν − γ 2 ) 1 . [3(3ν − 3γ 1 − 1)(3ν − 3γ 2 − 1)(2ν − γ 1 − γ 2 )]2/3

Corollary 6.15. (to Theorem 6.8; λα = 1, λα+1 = 1/2, p = 3/2, p(t) = q(t) = 1). In detail: Let ν ≥ 2 and γ 1 ≥ 0 such that ν − γ 1 ≥ 2. Let f1 , f2 ∈ Cxν0 ([a, b]) with (i) (i) f1 (x0 ) = f2 (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b] : x ≥ x0 . Then x) . γ +1 |(Dxγ01 f1 )(w)| |(Dx01 f2 )(w)|+ x0 * . γ 1 +1 γ1 |(Dx0 f2 )(w)| |(Dx0 f1 )(w)| dw * ) x ν 3/2 ν 3/2 (|(Dx0 f1 )(w)| + |(Dx0 f2 )(w)| )dw , (6.63) ≤ Φ(x) · x0

all x0 ≤ x ≤ b. Here we ﬁnd 2 (x − x0 )((3ν−3γ 1 −1)/2) Φ(x) = √ . 3ν − 3γ 1 − 2 (Γ(ν − γ 1 ))3/2

(6.64)

√ √ Proof. We apply inequality (6.45). Here ω 1 = 2; that is, Φ(x) = 2 · T (x), and θ3 = 3. Furthermore we ﬁnd L(x) = 2(x − x0 ), and P1 (x) =

(x − x0 )(3ν−3γ 1 −2) . (3ν − 3γ 1 − 2)

90

6. Canavati Fractional Derivatives of Two Functions

Thus T (x)

= =

3

2(x − x0 ) ·

(x − x0 )(3ν−3γ 1 −2/3)

3/2

(3ν − 3γ 1 − 2)Γ(ν − γ 1 ) √ 2 · (x − x0 )(3ν−3γ 1 −1/2) √ . (Γ(ν − γ 1 ))3/2 · 3ν − 3γ 1 − 2

Corollary 6.16. (to Theorem 6.9; p = 2(λα + λν ) > 1, p(t) = q(t) = 1). It holds x [|(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λα +λν |(Dxν0 f1 )(w)|λν x0

+|(Dxγ02 f1 )(w)|λα +λν |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν ]dw x + , (|(Dxν0 f1 )(w)|2(λα +λν ) + (|(Dxν0 f2 )(w)|)2(λα +λν ) dw , ≤ T˜(x) x0

(6.65) all x0 ≤ x ≤ b. Here T˜(x) = A0 (x) and

λν 2(λα + λν )

λν /2(λα +λν )

A0 (x) = σσ ∗ (x − x0 )θ ,

where σ :=

,

(6.66)

(6.67)

1 (Γ(ν − γ 1

))λα (Γ(ν

− γ 2 ))λα +λν (2λ2α +2λα λν −λα )/(2λα +2λν )

2λα + 2λν − 1 2λα ν + 2λν ν − 2λα γ 1 − 2λν γ 1 − 1 ((2λα +2λν −1)/2) 2λα + 2λν − 1 · (6.68) 2λα ν + 2λν ν − 2λα γ 2 − 2λν γ 2 − 1 σ ∗ := (2λα + λν ) 4λ2α ν + 6λα λν ν − 2λ2α γ 1 − −1 ,((2λα +λν )/2(λα +λν )) −2λα λν γ 1 − 2λ2α γ 2 − 4λα λν γ 2 + 2λ2ν ν − 2λ2ν γ 2 , ·

(6.69) and

θ :=

4λ2α ν + 6λα λν ν − 2λ2α γ 1 − 2λα λν γ 1 − 2λ2α γ 2 − 4λα λν γ 2 + 2λ2ν ν − 2λ2ν γ 2 . 2λα + 2λν (6.70)

6.3 Main Results

91

Proof. We apply inequality (6.52). The constant T˜(x) comes from (6.51) and the assumption on p here. Still we need to determine A0 (x). Here from (6.17) we have

(2λα +λν )/2(λα +λν )

x 2(λα +λν )/2λα +λν

A0 (x) =

(A(w))

dw

,

x0

where from (6.16) we ﬁnd A(w) =

(P1 (w))λα (2λα +2λν −1/2λα +2λν ) (P2 (w))((2λα +2λν −1)/2) , (Γ(ν − γ 1 ))λα (Γ(ν − γ 2 ))λα +λν

and from (6.15) we ﬁnd for k = 1, 2, w Pk (w) = (w − t)(ν−γ k −1)((2λα +2λν )/2λα +2λν −1) dt. x0

Hence for k = 1, 2, Pk (w) =

(2λα + 2λν − 1)(w − x0 )((2λα ν+2λν ν−2λα γ k −2λν γ k −1)/(2λα +2λν −1)) , (2λα ν + 2λν ν − 2λα γ k − 2λν γ k − 1)

and ⎛ ⎝

A(w) = σ(w − x0 )

4λ2α ν + 6λα λν ν − 2λ2α γ 1 − 2λ2α γ 2 − 2λα λν γ 1 − 4λα λν γ 2 − 2λ2ν γ 2 + 2λ2ν ν − 2λα − λν

⎞ ⎠ /(2λα +2λν )

.

Finally we obtain A0 (x) = σ· ⎛ ⎜ ⎜ x ⎜ ⎜ (w − x0 ) ⎜ ⎜ x0 ⎝

4λ2α ν + 6λα λν ν − 2λ2α γ 1 − 2λ2α γ 2 − 2λα λν γ 1 − 4λα λν γ 2 − 2λ2ν γ 2 + 2λ2ν ν − 2λα − λν (2λα + λν )

⎞ (2λα +λν ) 2(λα +λν )

⎟ ⎟ ⎟ dw⎟ ⎟ ⎟ ⎠

= σσ ∗ (x − x0 )θ . Corollary 6.17. (to Theorem 6.9; p = 4, λα = λν = 1, p(t) = q(t) = 1). It holds x |(Dxγ01 f1 )(w)|((Dxγ02 f2 )(w))2 |(Dxν0 f1 )(w)|+ x0

((Dxγ02 f1 )(w))2 |(Dxγ01 f2 )(w)| |(Dxν0 f2 )(w)| dw

92

6. Canavati Fractional Derivatives of Two Functions

∗

x

≤ T (x) x0

(((Dxν0 f1 )(w))4

all x0 ≤ x ≤ b. Here T ∗ (x) = and

+

((Dxν0 f2 )(w))4 )dw

,

A˜0 (x) √ , 2

(6.72)

˜ A˜0 (x) = σ ˜σ ˜ ∗ (x − x0 )θ ,

where 1 · σ ˜ := Γ(ν − γ 1 )(Γ(ν − γ 2 ))2 σ ˜ ∗ :=

3 4ν − 4γ 1 − 1

(6.71)

(6.73)

3/4 ·

3 12ν − 4γ 1 − 8γ 2

3 4ν − 4γ 2 − 1

3/2 , (6.74)

3/4 ,

(6.75)

and ˜θ := 3ν − γ − 2γ . 1 2 Proof. By Corollary 6.16.

(6.76)

We continue with related results regarding the sup-norm · ∞ . Theorem 6.18. Let ν ≥ 1, γ 1 , γ 2 ≥ 0, such that ν − γ 1 ≥ 1, ν − γ 2 ≥ 1, (i) (i) and f1 , f2 ∈ Cxν0 ([a, b]) with f1 (x0 ) = f2 (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b] : x ≥ x0 . Consider p(x) ≥ 0 a continuous function on [x0 , b]. Let λα , λβ , λν ≥ 0. Set ρ(x) :=

Then

(x − x0 )(νλα −γ 1 λα +νλβ −γ 2 λβ +1) p(x)∞ . (νλα − γ 1 λα + νλβ − γ 2 λβ + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λβ (6.77)

x

x0

p(w) |(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν

+|(Dxγ02 f1 )(w)|λβ |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν dw

≤

, ρ(x) + ν 2λ 2λ 2(λα +λν ) 2(λα +λν ) Dx0 f1 ∞ , + Dxν0 f1 ∞ β + Dxν0 f2 ∞ β + Dxν0 f2 ∞ 2 (6.78)

all x0 ≤ x ≤ b. Proof. From (6.13) we get for j = 1, 2; k = 1, 2, that w 1 γk ν−γ k −1 |(Dx0 fj )(w)| ≤ (w − t) dt Dxν0 fj ∞ Γ(ν − γ k ) x0 =

(w − x0 )ν−γ k 1 Dxν0 fj ∞ . Γ(ν − γ k ) (ν − γ k )

6.3 Main Results

93

That is, |(Dxγ0k fj )(w)| ≤

(w − x0 )(ν−γ k ) Dxν0 fj ∞ , Γ(ν − γ k + 1)

(6.79)

all x0 ≤ w ≤ b. Then we have |(Dxγ01 f1 )(w)|λα

≤

(w − x0 )(ν−γ 1 )λα Dxν0 f1 λ∞α , (Γ(ν − γ 1 + 1))λα

(6.80)

|(Dxγ02 f2 )(w)|λβ

≤

(w − x0 )(ν−γ 2 )λβ λ Dxν0 f2 ∞β , (Γ(ν − γ 2 + 1))λβ

(6.81)

|(Dxν0 f1 )(w)|λν

≤

Dxν0 f1 λ∞ν .

(6.82)

Multiplying (6.80) through (6.82) we obtain |(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν ≤

(w − x0 )(νλα −γ 1 λα +νλβ −γ 2 λβ ) λ Dxν0 f1 λ∞α +λν Dxν0 f2 ∞β . (6.83) (Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λβ

Similarly we observe |(Dxγ02 f1 )(w)|λβ

≤

(w − x0 )(ν−γ 2 )λβ λ Dxν0 f1 ∞β , (Γ(ν − γ 2 + 1))λβ

(6.84)

|(Dxγ01 f2 )(w)|λα

≤

(w − x0 )(ν−γ 1 )λα Dxν0 f2 λ∞α , (Γ(ν − γ 1 + 1))λα

(6.85)

|(Dxν0 f2 )(w)|λν

≤

Dxν0 f2 λ∞ν .

(6.86)

Multiplying (6.84) through (6.86) we ﬁnd |(Dxγ02 f1 )(w)|λβ |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν ≤

(w − x0 )(νλβ −γ 2 λβ +νλα −γ 1 λα ) λ Dxν0 f1 ∞β Dxν0 f2 λ∞α +λν . (6.87) λ λ β α (Γ(ν − γ 2 + 1)) (Γ(ν − γ 1 + 1))

Adding (6.83) and (6.87) we have |(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν +|(Dxγ02 f1 )(w)|λβ |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν ≤

(w − x0 )(νλα −γ 1 λα +νλβ −γ 2 λβ ) (Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λβ

+ , λ λ · Dxν0 f1 λ∞α +λν Dxν0 f2 ∞β + Dxν0 f1 ∞β Dxν0 f2 λ∞α +λν .

(6.88)

94

6. Canavati Fractional Derivatives of Two Functions

It follows that x p(w)[|(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λν x0

+|(Dxγ02 f1 )(w)|λβ |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν ]dw x p(x)∞ x0 (w − x0 )(νλα −γ 1 λα +νλβ −γ 2 λβ ) dw ≤ (Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λβ + , λ λ · Dxν0 f1 λ∞α +λν Dxν0 f2 ∞β + (Dxν0 f1 ∞β Dxν0 f2 λ∞α +λν , ρ(x) + ν 2λβ 2λβ ν ν ν 2(λα +λν ) α +λν ) . Dx0 f1 2(λ ≤ + D f + D f + D f ∞ ∞ 1 2 2 ∞ x0 x0 x0 ∞ 2 We have established (6.78).

Some special cases to the last theorem follow. Theorem 6.19. (as in Theorem 6.18; λβ = 0). It holds x p(w) |(Dxγ01 f1 )(w)|λα |(Dxν0 f1 )(w)|λν + x0

|(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν dw (x − x0 )(νλα −γ 1 λα +1) p(x)∞ ≤ (νλα − γ 1 λα + 1)(Γ(ν − γ 1 + 1))λα · [Dxν0 f1 λ∞α +λν + Dxν0 f2 λ∞α +λν ],

(6.89)

all x0 ≤ x ≤ b. Proof. As in Theorem 6.18, similarly.

Theorem 6.20. (as in Theorem 6.18; λβ = λα + λν ). It holds x p(w)[|(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λα +λν |(Dxν0 f1 )(w)|λν x0

+|(Dxγ02 f1 )(w)|λα +λν |(Dxγ01 f2 )(w)|λα |(Dxν0 f2 )(w)|λν ]dw (x − x0 )(2νλα −γ 1 λα +νλν −γ 2 λα −γ 2 λν +1 ≤ (2νλα − γ 1 λα + νλν − γ 2 λα − γ 2 λν + 1) p(x)∞ · (Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λα +λν α +λν ) α +λν ) + (Dxν0 f2 )2(λ · Dxν0 f1 2(λ , all x0 ≤ x ≤ b. ∞ ∞

(6.90)

6.3 Main Results

Proof. As in Theorem 6.18, similarly.

95

Theorem 6.21. (as in Theorem 6.18; λν = 0, λα = λβ ). It holds x p(w) |(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λα + x0

|(Dxγ02 f1 )(w)|λα |(Dxγ01 f2 )(w)|λα dw

ν 2λα α ≤ ρ∗ (x)[Dxν0 f1 2λ ∞ + Dx0 f2 ∞ ], all x0 ≤ x ≤ b.

(6.91)

Here ∗

ρ (x) :=

(x − x0 )(2νλα −γ 1 λα −γ 2 λα +1) · p(x)∞ . (2νλα − γ 1 λα − γ 2 λα + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λα (6.92)

Proof. As in Theorem 6.18, when λν = 0, we follow the proof and we obtain x p(w) |(Dxγ01 f1 )(w)|λα |(Dxγ02 f2 )(w)|λβ + x0

|(Dxγ02 f1 )(w)|λβ |(Dxγ01 f2 )(w)|λα dw λ

λ

≤ ρ(x)[Dxν0 f1 λ∞α Dxν0 f2 ∞β + Dxν0 f1 ∞β Dxν0 f2 λ∞α ],

(6.93)

all x0 ≤ x ≤ b, which inequality is by itself of interest. Then setting λα = λβ into (6.93) and (6.77) we derive (6.91). Theorem 6.22. (as in Theorem 6.18; λα = 0, λβ = λν ). It holds x p(w) |(Dxγ02 f2 )(w)|λβ |(Dxν0 f1 )(w)|λβ + x0

≤

(Dxγ02 f1 )(w)|λβ |(Dxν0 f2 )(w)|λβ dw (x − x0 )(νλβ −γ 2 λβ +1) · p(x)∞ (νλβ − γ 2 λβ + 1)(Γ(ν − γ 2 + 1))λβ 2λ

·

2λ

[(Dxν0 f1 )∞ β + (Dxν0 f2 )∞ β ],

(6.94)

all x0 ≤ x ≤ b. Proof. Again we follow the proof of Theorem 6.18. Corollary 6.23. (to Theorem 6.21; all are as in Theorem 6.18; λν = 0, λα = λβ , γ 2 = γ 1 + 1). It holds x p(w) |(Dxγ01 f1 )(w)|λα |(Dxγ01 +1 f2 )(w)|λα + x0

96

6. Canavati Fractional Derivatives of Two Functions

|(Dxγ01 +1 f1 )(w)|λα |(Dxγ01 f2 )(w)|λα dw ≤

(x − x0 )(2νλα −2γ 1 λα −λα +1) p(x)∞ (2νλα − 2γ 1 λα − λα + 1)(ν − γ 1 )λα (Γ(ν − γ 1 ))2λα ν 2λα α · [Dxν0 f1 2λ ∞ + Dx0 f2 ∞ ], all x0 ≤ x ≤ b.

(6.95)

Proof. Obvious. Corollary 6.24. (to Corollary 6.23). In detail : Let ν ≥ 2, γ 1 ≥ 0 such that ν − γ 1 ≥ 2 and f1 , f2 ∈ Cxν0 ([a, b]) with (i) (i) f1 (x0 ) = f2 (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b] : x ≥ x0 . Then x |(Dxγ01 f1 )(w)||(Dxγ01 +1 f2 )(w)|+ x0

|(Dxγ01 +1 f1 )(w)| |(Dxγ01 f2 )(w)| dw (x − x0 )2(ν−γ 1 ) ≤ 2(ν − γ 1 )2 (Γ(ν − γ 1 ))2 [Dxν0 f1 2∞ + Dxν0 f2 2∞ ],

all x0 ≤ x ≤ b.

(6.96)

Proof. Obvious. Proposition 6.25. Inequality (6.96) is sharp, in fact it is attained when f1 = f2 . Proof. Clearly (6.96) collapses to

x

|(Dxγ01 f1 )(w)| |(Dxγ01 +1 f1 )(w)|dw ≤

x0

(x − x0 )2(ν−γ 1 ) 2(ν − γ 1 )2 (Γ(ν − γ 1 ))2

Dxν0 f1 2∞ , (6.97)

all x0 ≤ x ≤ b. In [17], see Propositions 9 and 10; there we proved that (6.97) is sharp. In fact it is attained.

6.4 Applications We present a uniqueness of solution result for a system of fractional diﬀerential equations.

6.4 Applications

97

Theorem 6.26. Let ν ≥ 1, γ i ≥ 0, ν − γ i ≥ 1, i = 1, . . . , r ∈ N, n := [ν], (i) fj ∈ Caν ([a, b]), j = 1, 2; fj (a) = aij ∈ R, i = 0, 1, . . . , n − 1. Furthermore we have for j = 1, 2 that (Daν fj )(t) = Fj (t, {(Daγ i f1 )(t)}ri=1 , {(Daγ i f2 )(t)}ri=1 ) , all t ∈ [a, b]. (6.98) Here Fj are continuous functions on [a, b] × Rr × Rr , and satisfy the Lipschitz condition |Fj (t, z1 , . . . , zr , y1 , . . . , yr ) − Fj (t, z1 , . . . , zr , y1 , . . . , yr )| r ≤ [q1,i,j (t)|zi − zi | + q2,i,j (t)|yi − yi |], j = 1, 2, (6.99) i=1

where q1,i,j (t), q2,i,j (t) ≥ 0, 1 ≤ i ≤ r are continuous functions over [a, b]. Call M1,i := max(q1,i,1 ∞ , q2,i,2 ∞ ) and M2,i := max(q2,i,1 ∞ , q1,i,2 ∞ ). (6.100) Assume here that r (b − a)ν−γ i M1,i M2,i √ + √ · φ∗ (b) := < 1. √ 2 Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1 2 i=1 (6.101) Then if the system (6.98) has two pairs of solutions (f1 , f2 ) and (f1∗ , f2∗ ) we prove that fj = fj∗ , j = 1, 2; that is, we have uniqueness of solution. Proof. Assume there are two pairs of solutions (f1 , f2 ), (f1∗ , f2∗ ) to sys(i) (i) ∗(i) tem (6.98). Set gj := fj − fj∗ , j = 1, 2. Then gj = fj − fj and (i)

gj (a) = 0, i = 0, 1, . . . , n − 1; j = 1, 2. It holds (Daν gj )(t)

= Fj (t, {(Daγ i f1 )(t)}ri=1 , {(Daγ i f2 )(t)}ri=1 ) −Fj (t, {(Daγ i f1∗ )(t)}ri=1 , {(Daγ i f2∗ )(t)}ri=1 ).

Hence by (6.99) we have |(Daν gj )(t)| ≤

r

[q1,i,j (t)|(Daγ i g1 )(t)| + q2,i,j (t)|(Daγ i g2 )(t)|].

i=1

Thus |(Daν gj )(t)| ≤

r [q1,i,j ∞ |(Daγ i g1 )(t)| + q2,i,j ∞ |(Daγ i g2 )(t)|]. i=1

The last implies that |(Daν gj )(t)|2

≤

r

[q1,i,j ∞ |(Daγ i g1 )(t)||(Daν gj )(t)|

i=1

+q2,i,j ∞ |(Daγ i g2 )(t)| |(Daν gj )(t)|.

98

6. Canavati Fractional Derivatives of Two Functions

Consequently we observe

b

(((Daν g1 )(t))2 + ((Daν g2 )(t))2 )dt

I := a

≤

r

'

b

q1,i,1 ∞

|(Daγ i g1 )(t)| |(Daν g1 )(t)|dt+ a

i=1

(

b

q2,i,1 ∞

|(Daγ i g2 )(t)| |(Daν g1 )(t)|dt a

+

r

'

b

q1,i,2 ∞

|(Daγ i g1 )(t)| |(Daν g2 )(t)|dt+ a

i=1

q2,i,2 ∞

(

b

|(Daγ i g2 )(t)| |(Daν g2 )(t)|dt a

≤

r

a

i=1

+

r i=1

b

[|(Daγ i g1 )(t)| |(Daν g1 )(t)| + |(Daγ i g2 )(t)| |(Daν g2 )(t)|]dt

M1,i

b

[|(Daγ i g2 )(t)| |(Daν g1 )(t)| + |(Daγ i g1 )(t)| |(Daν g2 )(t)|]dt =: (∗).

M2,i a

However by Corollary 6.11 we obtain

b

[|(Daγ i g1 )(t)| |(Daν g1 )(t)| + |(Daγ i g2 )(t)| |(Daν g2 )(t)|]dt (b − a)ν−γ i √ ≤ · I. (6.102) √ 2Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1 a

Also by Corollary 6.13 we ﬁnd

b

[|(Daγ i g2 )(t)| |(Daν g1 )(t)| + |(Daγ i g1 )(t)| |(Daν g2 )(t)|]dt (b − a)ν−γ i ≤ √ · I. (6.103) √ √ 2Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1 a

Therefore by (6.102) and (6.103) we ﬁnd (∗)

r ) ≤ M1,i i=1

+M2,i

(b − a)ν−γ i √ √ 2Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1

(b − a)ν−γ i √ √ √ 2Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1

·I

( · I = φ∗ (b)I.

6.4 Applications

99

We have established that I ≤ φ∗ (b)I.

(6.104)

If I = 0 then φ∗ (b) ≥ 1, a contradiction by the assumption (6.101) that φ∗ (b) < 1. Therefore I = 0, implying that ((Daν g1 )(t))2 + ((Daν g2 )(t))2 = 0, a.e. [a, b]. That is, ((Daν g1 )(t))2 = 0,

((Daν g2 )(t))2 = 0, a.e. [a, b].

That is, (Daν g1 )(t) = 0,

(Daν g2 )(t) = 0, a.e. [a, b].

But for j = 1, 2 we get that (i)

gj (a) = 0,

0 ≤ i ≤ n − 1.

Hence from fractional Taylor’s Theorem 6.1 (see (6.7)) we ﬁnd that gj (t) = 0 on [a, b]. That is fj ≡ fj∗ , j = 1, 2, proving the uniqueness claim of the theorem. Next we give upper bounds on Daν fj , solutions fj , and so on, all involved in a system of fractional diﬀerential equations. Theorem 6.27. Let ν ≥ 1, γ i ≥ 0, ν − γ i ≥ 1, i = 1, . . . , r ∈ N, n := [ν], (i) fj ∈ Caν ([a, b]), j = 1, 2; fj (a) = 0, i = 0, 1, . . . , n − 1, and (Daν fj )(a) = Aj ∈ R. Furthermore for a ≤ t ≤ b we have holding the system of fractional diﬀerential equations (Daν fj ) (t) = Fj (t, {(Daγ i f1 )(t)}ri=1 , (Daν f1 )(t), {(Daγ i f2 )(t)}ri=1 , (Daν f2 )(t)), (6.105) for j = 1, 2. Here Fj are continuous functions on [a, b] × Rr+1 × Rr+1 such that |Fj (t, x1 , x2 , . . . , xr , xr+1 , y1 , . . . , yr , yr+1 ) | r ≤ [q1,i,j (t)|xi | + q2,i,j (t)|yi |],

(6.106)

i=1

where q1,i,j (t), q2,i,j (t) ≥ 0, 1 ≤ i ≤ r, j = 1, 2 , are continuous functions on [a, b]. Call M1,i := max(q1,i,1 ∞ , q2,i,2 ∞ ), and M2,i := max(q2,i,1 ∞ , q1,i,2 ∞ ). (6.107)

100

6. Canavati Fractional Derivatives of Two Functions

Also we set (a ≤ x ≤ b) ((Daν f1 )(x))2 + ((Daν f2 )(x))2 , A21 + A22 , r √ (M1,i + 2M2,i ) ·

θ(x) := ρ := Q(x) :=

(6.108) (6.109)

i=1

(x − a)ν−γ i √ √ Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1

,

(6.110)

and χ(x) :=

√

) x ρ· 1 + Q(x) · e( a Q(s)ds)

x

e−(

t a

Q(s)ds)

*1/2

dt

. (6.111)

a

Then

θ(x) ≤ χ(x),

a ≤ x ≤ b.

(6.112)

Consequently it holds |(Daν fj )(x)| |fj (x)|

≤ χ(x), x 1 ≤ (x − t)ν−1 χ(t)dt, Γ(ν) a

(6.113) (6.114)

for j = 1, 2, a ≤ x ≤ b. Also it holds that |(Daγ i fj )(x)| ≤

1 Γ(ν − γ i )

x

(x − t)ν−γ i −1 χ(t)dt,

(6.115)

a

for j = 1, 2; i = 1, . . . , r, a ≤ x ≤ b. Proof. We see that (Daν fj )(t) · (Daν fj ) (t) = (Daν fj )(t) · Fj (t, {(Daγ i f1 )(t)}ri=1 , (Daν f1 )(t), {(Daγ i f2 )(t)}ri=1 , (Daν f2 )(t)), all a ≤ x ≤ b. We then integrate the last equality

x

(Daν fj )(t) · ((Daν fj ) (t))dt x (Daν fj )(t) · Fj (t, {(Daγ i f1 )(t)}ri=1 , = a

a

(Daν f1 )(t), {(Daγ i f2 )(t)}ri=1 , (Daν f2 )(t))dt.

6.4 Applications

101

Hence we ﬁnd

((Daν fj )(t))2 x |a 2

x

≤

|(Daν fj )(t)| |Fj · · · |dt

a x

|(Daν fj )(t)|

≤

' r

(

a

{q1,i,j (t)|(Daγ i f1 )(t)|

i=1 r

≤

dt

x

q1,i,j ∞

i=1

+

q2,i,j (t)|(Daγ i f2 )(t)|}

|(Daν fj )(t)| |(Daγ i f1 )(t)|dt a

x

+ q2,i,j ∞

|(Daν fj )(t)| |(Daγ i f2 )(t)|dt .

a

Thus we obtain ((Dαν fj )(x))2

≤ A2j + 2

r q1,i,j ∞ · i=1

+ q2,i,j ∞ ·

x

|(Daν fj )(t)| |(Daγ i f1 )(t)dt

a x

|(Daν fj )(t)| |(Daγ i f2 (t)|dt

.

a

Consequently we write θ(x) ≤ ρ + 2

r i=1

x [|(Daν f1 )(t)| |(Daγ i f1 )(t)| + |(Daν f2 )(t)| |(Daγ i f2 )(t)|]dt M1,i a x ν γi ν γi [|(Da f2 )(t)| |(Da f1 )(t)| + |(Da f1 )(t)| |(Da f2 )(t)|]dt + M2,i a (by (6.56), (6.60))

≤

%

ρ+2

r i=1

x (x − a)(ν−γ i ) √ θ(t)dt M1,i · √ 2Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1 a x (x − a)(ν−γ i ) · +M2,i √ θ(t)dt √ √ 2Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1 a x = ρ + Q(x) θ(t)dt.

a

That is, we get

x

θ(x) ≤ ρ + Q(x)

θ(t)dt, a

(6.116)

102

6. Canavati Fractional Derivatives of Two Functions

all a ≤ x ≤ b. Here ρ ≥ 0, Q(x) ≥ 0, Q(a) = 0, θ(x) ≥ 0, all a ≤ x ≤ b. Solving the integral inequality (6.116), exactly as in [17] (see pp. 224–225, Application 3.2 and inequalities (44), (47) there) we ﬁnd that θ(x) ≤ χ(x), a ≤ x ≤ b. Then (6.113) is obvious. Next from (6.9) we have x x 1 1 |fj (x)| ≤ (x − t)ν−1 |Daν fj (t)|dt ≤ (x − t)ν−1 χ(t)dt, Γ(ν) a Γ(ν) a all a ≤ x ≤ b, j = 1, 2, proving (6.114). Finally from (6.13) we ﬁnd x 1 |(Daγ i fj )(x)| ≤ (x − t)ν−γ i −1 |(Daν fj )(t)|dt Γ(ν − γ i ) a x 1 (x − t)ν−γ i −1 χ(t)dt, ≤ Γ(ν − γ i ) a all a ≤ x ≤ b, j = 1, 2, i = 1, . . . , r, proving (6.115). In the ﬁnal application we give similar upper bounds to Theorem 6.27, but under diﬀerent conditions. Theorem 6.28. Let a = b, ν ≥ 2, γ i ≥ 0, ν − γ i ≥ 1, i = 1, . . . , r ∈ N, (i) n := [ν], fj ∈ Caν ([a, b]), j = 1, 2; fj (a) = 0, i = 0, 1, . . . , n − 1, and (Daν fj )(a) = Aj ∈ R. Furthermore for a ≤ t ≤ b we have holding the system of fractional diﬀerential equations (Daν fj ) (t)

= Fj (t, {(Daγ i f1 )(t)}ri=1 , (Daν f1 )(t), {(Daγ i f2 )(t)}ri=1 , (Daν f2 )(t)), for j = 1, 2. (6.117)

For ﬁxed i∗ ∈ {1, . . . , r} we assume that γ i∗ +1 = γ i∗ + 1, and ν − γ i∗ ≥ 2, where γ i∗ , γ i∗ +1 ∈ {γ 1 , . . . , γ r }. Call k := γ i∗ , γ := γ i∗ +1 (i.e., γ = k+1). Here Fj are continuous functions on [a, b] × Rr+1 × Rr+1 such that |Fj (t, x1 , . . . , xr , xr+1 , y1 , . . . , yr , yr+1 )| ≤ q1,j (t)|xi∗ | |yi∗ +1 | + q2,j (t)|yi∗ | |xi∗ +1 |, where q1,j , q2,j ≡ 0, q1,j (t), q2,j (t) ≥ 0

(6.118)

6.4 Applications

103

are continuous functions over [a, b]. Put M := max(q1,1 ∞ , q2,1 ∞ , q1,2 ∞ , q2,2 ∞ ),

(6.119)

θ(x) := |(Daν f1 )(x)| + |(Daν f2 )(x)|,

(6.120)

a ≤ x ≤ b,

ρ := |A1 | + |A2 |.

(6.121)

Also we deﬁne Q(x) :=

4M √ 3ν − 3k − 2

·

a ≤ x ≤ b.

σ := Q(x)∞ , We assume that

3ν−3k−1 (x − a)( 2 ) , (Γ(ν − k))3/2

(6.122) (6.123)

√ (b − a)σ ρ < 2.

(6.124)

(x − a) · (σ − Q(x)) · [σρ2 (x − a) − 4ρ3/2 ] + 4ρ , √ (2 − σ ρ(x − a))2

(6.125)

Call ϕ ˜ (x) :=

all a ≤ x ≤ b. Then

θ(x) ≤ ϕ ˜ (x);

(6.126)

in particular, it holds |(Daν fj )(x)| ≤ ϕ ˜ (x),

j = 1, 2,

(6.127)

all a ≤ x ≤ b. Furthermore we ﬁnd |fj (x)|

≤

|(Daγ i fj )(x)|

≤

x 1 (x − t)ν−1 ϕ ˜ (t)dt, Γ(ν) a x 1 (x − t)ν−γ i −1 ϕ ˜ (t)dt, Γ(ν − γ i ) a

(6.128) (6.129)

j = 1, 2, i = 1, . . . , r, all a ≤ x ≤ b. Proof. For a ≤ x ≤ b we ﬁnd x x ν (Da fj ) (t)dt = Fj (t, {(Daγ i f1 )(t)}ri=1 , a

a

(Daν f1 )(t), {(Daγ i f2 )(t)}ri=1 , (Daν f2 )(t))dt. That is,

(Daν fj )(x) = Aj +

x

Fj (t, . . .)dt. a

104

6. Canavati Fractional Derivatives of Two Functions

Then we observe

|(Daν fj )(x)|

≤ |Aj | +

x

|Fj (t, . . .)|dt a

x

≤ |Aj | +

. [q1,j (t)|(Dak f1 )(t)| |(Dak+1 f2 )(t)|

a

. +q2,j (t)|(Dak f2 )(t)| |(Dak+1 f1 )(t)|]dt x ) . k ≤ |Aj | + M |(Da f1 )(t)| |(Dak+1 f2 )(t)|+ . a

* . |(Dak f2 )(t)| |(Dak+1 f1 )(t)| dt . Hence

) . θ(x) ≤ ρ + 2M |(Dak f1 )(t)| |(Dak+1 f2 )(t)|+ a * . k+1 k |(Da f2 )(t)| |(Da f1 )(t)| dt x

(x − a)((3ν−3k−1)/2) · (Γ(ν − k))3/2 x · (|(Daν f1 )(t)|3/2 + |(Daν f2 )(t)|3/2 )dt

(by (6.63), (6.64))

≤

ρ+

4M √ 3ν − 3k − 2

a

4M (x − a)((3ν−3k−1)/2) ρ+ √ · (Γ(ν − k))3/2 3ν − 3k − 2 x · (|(Daν f1 )(t)| + |(Daν f2 )(t)|)3/2 dt .

(6.24)

≤

a

We have proved that

x

θ(x) ≤ ρ + Q(x) ·

(θ(t))3/2 dt,

(6.130)

a

all a ≤ x ≤ b. Notice that θ(x) ≥ 0, ρ ≥ 0, Q(x) ≥ 0, and Q(a) = 0; also it is σ > 0. Call x (θ(t))3/2 dt, w(a) = 0, all a ≤ x ≤ b. (6.131) 0 ≤ w(x) := a

That is, and

w (x) = (θ(x))3/2 ≥ 0, θ(x) = (w (x))2/3 ,

all a ≤ x ≤ b.

(6.132)

6.4 Applications

105

Hence we rewrite (6.130) as follows, (w (x))2/3 ≤ ρ + Q(x)w(x), So that

a ≤ x ≤ b.

(6.133)

(w (x))2/3 ≤ ρ + σw(x) < ρ + ε + σw(x),

for ε > 0 arbitrarily small. Hence w (x) < (ρ + ε + σw(x))3/2 ,

a ≤ x ≤ b.

Here (ρ + ε + σw(x))3/2 > 0. In particular it holds w (t) < (ρ + ε + σw(t))3/2 , and

a ≤ t ≤ x,

w (t) < 1. (ρ + ε + σw(t))3/2

(6.134)

The last is the same as 2 − (ρ + ε + σw(t))−1/2 < 1. σ Therefore after integration we have x σ − d((ρ + ε + σw(t))−1/2 ) ≤ (x − a), 2 a and

(ρ + ε + σw(t))−1/2 |ax ≤

σ (x − a). 2

That is, (ρ + ε)−1/2 − (ρ + ε + σw(x))−1/2 ≤ and

(ρ + ε)−1/2 −

σ (x − a), 2

σ (x − a) ≤ (ρ + ε + σw(x))−1/2 . 2

That is, 2 − σ(ρ + ε)1/2 (x − a) 1 ≤ . 2(ρ + ε)1/2 (ρ + ε + σw(x))1/2 By assumption (6.124) we get √ (x − a)σ ρ < 2,

all a ≤ x ≤ b.

Then for suﬃciently small ε > 0 we still have (x − a)σ(ρ + ε)1/2 < 2.

(6.135)

106

6. Canavati Fractional Derivatives of Two Functions

That is, 2 − (x − a)σ(ρ + ε)1/2 > 0,

all a ≤ x ≤ b.

(6.136)

From (6.135) and (6.136) we get (ρ + ε + σw(x))1/2 ≤

2(ρ + ε)1/2 , 2 − σ(ρ + ε)1/2 (x − a)

all a ≤ x ≤ b.

Solving (6.137) for w(x) we ﬁnd * ) 4 ρ+ε √ − 1 , w(x) ≤ σ (2 − σ( ρ + ε)(x − a))2 for ε > 0 suﬃciently small and for all a ≤ x ≤ b. Letting ε → 0 we obtain * ρ) 4 −1 , w(x) ≤ √ σ (2 − σ ρ(x − a))2

(6.137)

(6.138)

(6.139)

for all a ≤ x ≤ b. That is, w(x) ≤

4ρ3/2 (x − a) − σρ2 (x − a)2 , √ (2 − σ ρ(x − a))2

(6.140)

for all a ≤ x ≤ b. Then by (6.130) and (6.140) we ﬁnd )

* 4ρ3/2 (x − a) − σρ2 (x − a)2 θ(x) ≤ ρ + Q(x) =ϕ ˜ (x), √ (2 − σ ρ(x − a))2

(6.141)

for all a ≤ x ≤ b. That is, (6.141) implies (6.126). Then by (6.120) and (6.126), inequality (6.127) is obvious. Finally by (6.9), (6.13), and (6.127), the inequalities (6.128) and (6.129) are clear, respectively.

7 Opial-Type Inequalities for Riemann– Liouville Fractional Derivatives of Two Functions with Applications

A wide variety of very general but basic Lp (1 ≤ p ≤ ∞)-form Opialtype inequalities [315] is established involving Riemann–Liouville fractional derivatives [17, 230, 295, 314] of two functions in diﬀerent orders and powers. From the developed results are derived several other concrete results of special interest. The sharpness of inequalities is established there. Finally applications of some of these special inequalities are given in establishing the uniqueness of solution and in giving upper bounds to solutions of initial value fractional problems involving a very general system of two fractional diﬀerential equations. Also upper bounds to various Riemann– Liouville fractional derivatives of the solutions that are involved in the above systems are presented. This treatment is based on [48].

7.1 Introduction We start in Section 7.2 with background, we continue in Section 7.3 with the main results, and we ﬁnish in Section 7.4 with the applications. To give the reader a ﬂavor of the kind of inequalities we are dealing with, we brieﬂy mention x q(w)[|(Dγ 1 f1 )(w)|λα |(Dν f1 )(w)|λν + |(Dγ 1 f2 )(w)λα |(Dν f2 )(w)|λν ]dw 0

G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 107 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 7,

108

7. Riemann–Liouville Fractional Derivatives

≤ C(x, q(w), γ 1 , ν, λα , λν , p(w), p) )

*λα +λν /p

x

p(w)[|(Dν f1 )(w)p + |(Dν f2 )(w)|p ]dw

·

, x ≥ 0,

0

for functions f1 , f2 ∈ L1 (0, x); p(w), q(w), 1/p(w) ∈ L∞ (0, x), all exponents and orders are fractional, and so on. Here Dβ f stands for the Riemann– Liouville fractional derivative of order β ≥ 0. Furthermore one system of fractional diﬀerential equations we are working on looks brieﬂy like r

r

(Dν fj ) (t) = Fj (t, {(Dγ i f1 )(t)}i=1 , {(Dγ i f2 )(t)}i=1 ) all t ∈ [0, x], for j = 1, 2, and with Dν−i fj (0) = dij ∈ R, i = 1, ..., n + 1, where n := [ν] is the integral part of ν > 0, and so on.

7.2 Background We need Deﬁnition 7.1. (see [187, 295, 314]). Let α ∈ R+ − {0}. For any f ∈ L1 (0, x); x ∈ R+ − {0}, the Riemann–Liouville fractional integral of f of order α is deﬁned by s 1 (s − t)α−1 f (t)dt, ∀s ∈ [0, x], (7.1) (Jα f )(s) := Γ(α) 0 and the Riemann–Liouville fractional derivative of f of order α by Dα f (s) :=

1 Γ(m − a)

d ds

m

s

(s − t)m−α−1 f (t)dt,

(7.2)

0

where m := [α] + 1, [·] is the integral part. In addition, we set D0 f := f := J0 f, J−α f = Dα f if α > 0, D−α f := Jα f , if 0 < α ≤ 1 . If α ∈ N, then Dα f = f (α) , the ordinary derivative. Deﬁnition 7.2. [87]. We say that f ∈ L1 (0, x) has an L∞ fractional derivative Dα f in [0, x], x ∈ R+ − {0}, iﬀ Dα−k f ∈ C([0, x]), k = 1, . . . , m := [α] + 1; α ∈ R+ − {0}, and Dα−1 f ∈ AC([0, x]) (absolutely continuous functions) and Dα f ∈ L∞ (0, x). We need

7.3 Main Results

109

Lemma 7.3. [187]. Let α ∈ R+ , β > α, let f ∈ L1 (0, x), x ∈ R+ − {0}, have an L∞ fractional derivative Dβ f in [0, x],and let Dβ−k f (0) = 0 for k = 1, . . . , [β] + 1. Then s 1 Dα f (s) = (s − t)β−α−1 Dβ f (t)dt, ∀ s ∈ [0, x]. (7.3) Γ(β − α) 0 Here Dα f ∈ AC([0, x]) for β − α ≥ 1 and Dα f ∈ C([0, x]) for β − α ∈ (0, 1), hence Dα f ∈ L∞ (0, x) and Dα f ∈ L1 (0, x).

7.3 Main Results We present our ﬁrst main result. Theorem 7.4. Let α1 , α2 ∈ R+ , β > α1 , α2 , β − αi > (1/p), p > 1, i = 1, 2 and let f1 , f2 ∈ L1 (0, x), x ∈ R+ − {0} have, respectively, L∞ fractional derivatives Dβ f1 , Dβ f2 in [0,x], and let Dβ−k fi (0) = 0, for k = 1, . . . , [β]+ 1; i = 1, 2. Consider also p(t) > 0 and q(t) ≥ 0, with all p(t), 1/p(t), q(t) ∈ L∞ (0, x). Let λβ > 0 and λα1 , λα2 ≥ 0, such that λβ 1. Put s (s − t)p(β−αi −1)/p−1 (p(t))−1/(p−1) dt, i = 1, 2; 0 ≤ s ≤ x, Pi (s) := 0

(7.4)

q(s)(P1 (s))λα1 (p−1/p) (P2 (s))λα2 ((p−1)/p) (p(s))−λβ /p A(s) := (Γ(β − α1 ))λα1 (Γ(β − α2 ))λα2 x (p−λβ )/p A0 (x) := (A(s))p/(p−λβ ) ds ,

(7.5) (7.6)

0

and δ 1 :=

21−((λα1 +λβ )/p) , if 1, if

λα1 + λβ ≤ p, λα1 + λβ ≥ p.

(7.7)

If λα2 = 0, we obtain that x + λβ λβ , λ λ ds ≤ q(s) |Dα1 f1 (s)| α1 Dβ f1 (s) + |Dα1 f2 (s)| α1 Dβ f2 (s) 0

A0 (x)|λα2 =0 ) δ1 0

x

λβ λα1 + λβ

λβ /p

*((λα1 +λβ )/p) + p β p , β p(s) D f1 (s) + D f2 (s) ds .

(7.8)

110

7. Riemann–Liouville Fractional Derivatives

Proof. From (7.3) we have Dαi fj (s) =

1 Γ(β − αi )

s

(s − t)β−αi −1 Dβ fj (t)dt, ∀ s ∈ [0, x], i = 1, 2; j = 1, 2.

0

(7.9) Next, applying H¨ older’s inequality with indices p, p/(p − 1), we ﬁnd s 1 −1/p 1/p |Dαi fj (s)| ≤ (s − t)β−αi −1 (p(t)) (p(t)) |Dβ fj (t)|dt Γ(β − αi ) 0 (7.10) s p/(p−1) (p−1)/p 1 (s − t)β−αi −1 (p(t))−1/p dt (7.11) ≤ Γ(β − αi ) 0

s

×

1/p p(t)|Dβ fj (t)|p dt =

0

1 (Pi (s))(p−1)/p Γ(β − αi )

s

1/p p(t)|Dβ fj (t)|p dt .

0

(7.12) That is, it holds 1 (Pi (s))p−1/p |D fj (s)| ≤ Γ(β − αi )

αi

Set

1/p

s β

p

p(t)|D fj (t)| dt

(7.13)

s

p(t)|Dβ fj (t)|p dt,

zj (s) :=

.

0

(7.14)

0

thus, zj (s) = p(s)|Dβ fj (s)|p , a.e. in(0, x), zj (0) = 0; j = 1, 2.

(7.15)

Hence, we have |Dαi fj (s)| ≤

1 (Pi (s))p−1/p (zj (s))1/p , Γ(β − αi )

(7.16)

and |Dβ fj (s)|λβ = (p(s))−λβ /p (zj (s))λβ /p , a.e. in(0, x), i = 1, 2; j = 1, 2. (7.17) Therefore we obtain q(s)|(Dα1 f1 )(s)|λα1 |(Dα2 f2 )(s)|λα2 |(Dβ f1 )(s)|λβ ≤ q(s)

1 (P1 (s))(λα1 (p−1)/p) (z1 (s))λα1 /p (Γ(β − α1 ))λα1

(7.18)

1 (P2 (s))λα2 (p−1/p) (z2 (s))λα2 /p (p(s))−λβ /p (z1 (s))λβ /p (Γ(β − α2 ))λα2 (7.19) (7.20) = A(s)(z1 (s))λα1 /p (z2 (s))λα2 /p (z1 (s))λβ /p , a.e.in(0, x).

7.3 Main Results

111

Consequently, by another H¨ older’s inequality application, we ﬁnd (by p/λβ > 1) x λβ λ λ q(s) |(Dα1 f1 ) (s)| α1 |(Dα2 f2 ) (s)| α2 Dβ f1 (s) ds 0

x

≤ A0 (x)

(z1 (s))λα1 /λβ (z2 (s))λα2 /λβ z1 (s)ds

λβ /p .

(7.21)

0

Similarly, one ﬁnds x λβ λ λ q(s) |(Dα1 f2 ) (s)| α1 |(Dα2 f1 )(s)| α2 (Dβ f2 )(s) ds 0

x

≤ A0 (x)

(z1 (s))λα2 /λβ (z2 (s))λα1 /λβ z2 (s)ds

λβ /p .

(7.22)

0

Taking λα2 = 0 and adding (7.21) and (7.22), we get

x 0

)

λβ λβ * α1 λα β λα β α1 1 1 q(s) (D f1 )(s) ds (D f1 )(s) + (D f2 )(s) (D f2 )(s)

≤ A0 (x)|λα2 =0 +

x

'

x

(z1 (s))λα1 /λβ (z1 (s))ds

0

(z2 (s))λα1 /λβ z2 (s)ds

λβ /p

λβ /p ( =

(7.23)

0

A0 (x) |λα

2

+ =0

(z1 (x))((λα1 +λβ )/p) + (z2 (x))((λα1 +λβ )/p)

A0 (x)|λα2 =0

λβ λ α1 + λ β

λβ /p '

x

p(t)|Dβ f1 (t)|p dt

λβ /p λβ = λ α1 + λ β (7.24)

((λα1 +λβ )/p)

0

x

p(t)|Dβ f2 (t)|p dt

+

,

(λα1 +λβ /p) ( =: (∗).

(7.25)

0

In this study, we frequently use the basic inequalities, 2r−1 (ar + br ) ≤ (a + b)r ≤ ar + br , a, b ≥ 0, 0 ≤ r ≤ 1,

(7.26)

ar + br ≤ (a + b)r ≤ 2r−1 (ar + br ), a, b ≥ 0, r ≥ 1.

(7.27)

and At last using (7.7), (7.26), and (7.27), we ﬁnd

(∗) ≤ A0 (x)|λα2 =0

λβ λα1 + λβ

λβ /p δ1

112

7. Riemann–Liouville Fractional Derivatives

)

x

+ p p , p(t) Dβ f1 (t) + Dβ f2 (t) dt

*(λα1 +λβ /p) .

(7.28)

0

Inequality (7.8) has been proved. The counterpart of the last theorem follows. Theorem 7.5. All here are as in Theorem 7.4. Denote λ /λ 2 α2 β − 1, if λα2 ≥ λβ , δ 3 := 1, if λα2 ≤ λβ .

(7.29)

If λα1 = 0, then

) λβ λβ * λ λ q(s) Dα2 f2 (s) α2 (Dβ f1 )(s) + Dα2 f1 (s) α2 Dβ f2 (s) ds

x 0

λβ /p λβ λ /p δ3 β λα2 + λβ (λβ +λα2 )/p x

β p β p · p(s) |D f1 (s)| + |D f2 (s)| ds . ≤ A0 (x) λα1 =0 2(p−λβ )/p

(7.30)

0

Proof. When λα1 = 0 from (7.21) and (7.22) we obtain

x

+ λβ λβ , λ λ ds ≤ q(s) |Dα2 f2 (s)| α2 Dβ f1 (s) + |Dα2 f1 (s)| α2 Dβ f2 (s)

0

A0 (x)|λα1 =0

x

+

'

x λα2 /λβ

(z2 (s))

(z1 (s))ds

λβ /p

0

(z1 (s))λα2 /λβ z2 (s)ds

λβ /p ( ≤

(7.31)

0

A0 (x) λα1 =0 21−(λβ /p) (M (x))λβ /p =: (∗), by 0 < λβ /p < 1 and by (7.26), where x+ , (z2 (s))λα2 /λβ z1 (s) + (z1 (s))λα2 /λβ z2 (s) ds. M (x) :=

(7.32)

(7.33)

0

Call δ 2 :=

1, if 21−(λα2 /λβ ) , if

λα2 ≥ λβ , λα2 ≤ λβ .

(7.34)

7.3 Main Results

113

We observe that x ((z1 (s))λα2 /λβ + (z2 (s))λα2 /λβ )(z1 (s) + z2 (s))ds M (x) = 0

x

−

+

, (z1 (s))λα2 /λβ z1 (s) + (z2 (s))λα2 /λβ z2 (s) ds

0

(by (7.26),(7.27))

≤

−

λβ λα2 + λβ

δ2 +

x

(z1 (s) + z2 (s))λα2 /λβ (z1 (s) + z2 (s)) ds

0

(z1 (x))((λα2 +λβ )/λβ ) + (z2 (x))((λα2 +λβ )/λβ )

= δ 2 (z1 (x) + z2 (x))(λα2 +λβ /λβ ) −

(7.27)

≤

=

(7.35)

λβ λα2 + λβ

,

+ , λβ (z1 (x))((λα2 +λβ )/λβ ) + (z2 (x))((λα2 +λβ )/λβ ) λα2 + λβ + λβ δ 2 (z1 (x) + z2 (x))(λα2 +λβ /λβ ) − = λα2 + λβ + ,, (z1 (x))(λα2 +λβ /λβ ) + (z2 (x))(λα2 +λβ /λβ )

λβ λα2 + λβ

λβ λα2 + λβ

+

(7.36)

(7.37)

(7.38)

δ 2 2λα2 /λβ (z1 (x))(λα2 +λβ /λβ ) + (z2 (x))(λα2 +λβ /λβ )

,, + (7.39) − (z1 (x))(λα2 +λβ /λβ ) + (z2 (x))(λα2 +λβ )/λβ , + (δ 2 2λα2 /λβ − 1) (z1 (x))(λα2 +λβ /λβ ) + (z2 (x))(λα2 +λβ /λβ ) (7.40)

(notice that δ 3 = δ 2 2λα2 /λβ − 1) (7.27) λβ ≤ δ 3 (z1 (x) + z2 (x))((λα2 +λβ )/λβ ) . λα2 + λβ That is, we get that λβ M (x) ≤ δ 3 (z1 (x) + z2 (x))((λα2 +λβ )/λβ ) . λα2 + λβ

(7.41)

(7.42)

Therefore, by (7.32) and (7.42), we ﬁnd (∗) ≤ A0 (x)|λα1 =0 2(p−λβ )/p

λβ λα2 + λβ

λβ /p

λ /p

δ 3 β (z1 (x) + z2 (x))(λα2 +λβ )/p

(7.43)

114

7. Riemann–Liouville Fractional Derivatives

λβ /p ) x + p λβ λ /p δ3 β p(s) Dβ f1 (s) + λα2 + λβ 0 β p , ,((λα2 +λβ )/p) D f2 (s) ds . (7.44)

= A0 (x)|λα1 =0 2(p−λβ /p)

We have established (7.30).

The complete case λα1 , λα2 = 0 follows. Theorem 7.6. All here are as in Theorem 7.4. Denote ((λ +λ )/λ ) 2 α1 α2 β − 1, if λα1 + λα2 ≥ λβ , γ˜1 := 1, if λα1 + λα2 ≤ λβ , and

1, if 21−((λα1 λα2 +λβ )/p) , if

γ˜2 :=

Then

x

λα1 + λα2 + λβ ≥ p, λα1 + λα2 + λβ ≤ p.

|Dα2 f1 (s)|

λα2

|Dα1 f2 (s)|

λβ + λα2 )(λα1 + λα2 + λβ )

(λα1

x

·

(7.46)

+ λβ λ λ q(s) |Dα1 f1 (s)| α1 |Dα2 f2 (s)| α2 Dβ f1 (s) +

0

A0 (x)

(7.45)

λα1

, β D f2 (s)λβ ds ≤

λβ /p + , λ /p · λα1β γ˜2 + 2(p−λβ )/p (γ˜1 λα2 )λβ /p

(λα1 +λα2 +λβ )/p p β p β ds p(s) D f1 (s) + D f2 (s) .

((7.47))

0

Proof. Here we use the known inequality ab ≤

bq ap + , p q

(7.48)

where a, b ≥ 0 and p, q > 1 such that 1/p + 1/q = 1. From (7.21) we obtain x λβ λ λ q(s) |Dα1 f1 (s)| α1 |Dα2 f2 (s)| α2 Dβ f1 (s) ds ≤ 0

)

x

)

A0 (x)

0

λ α2 λα1 + λα2

λα1 (z1 (s))(λα1 +λα2 /λβ ) + λα1 + λα2 *λβ /p * (λα1 +λα2 /λβ ) (z2 (s)) z1 (s)ds

(7.49)

7.3 Main Results

'

(7.26)

≤ A0 (x)

x

+ 0

λα1 λβ λα1 + λα2

λ α2 λα1 + λα2

(z1 (x))((λα1 +λα2 +λβ )/λβ ) (λα1 + λα2 + λβ )

(z2 (s))((λα1 +λα2 )/λβ ) z1 (s)ds

115

λβ /p

λβ /p ( .

(7.50)

Consequently, x q(s)|Dα1 f1 (s)|λα1 |Dα2 f2 (s)|λα2 |Dβ f1 (s)|λβ ds ≤ 0

% A0 (x)

(λα1

+

λα1 λβ + λα2 )(λα1 + λα2 + λβ )

λα2 λα1 + λα2

λβ /p

x

λβ /p (z1 (x))((λα1 +λα2 +λβ )/p)

(z2 (s))(λα1 +λα2 /λβ ) z1 (s)ds

(7.51)

λβ /p .

0

Working similarly, we get x q(s)|Dα1 f2 (s)|λα1 |Dα2 f1 (s)|λα2 |Dβ f2 (s)|λβ ds 0

% ≤ A0 (x)

(λα1

λ α2 λα1 + λα2

λα1 λβ + λα2 )(λα1 + λα2 + λβ )

λβ /p

x

λβ /p (z2 (x))((λα1 +λα2 +λβ )/p) +

(z1 (s))(λα1 +λα2 /λβ ) z2 (s)ds

λβ /p .

(7.52)

0

Next, adding (7.51) and (7.52), we observe x + λβ λ λ Ω := q(s) |Dα1 f1 (s)| α1 |Dα2 f2 (s)| α2 Dβ f1 (s) + 0

|Dα1 f2 (s)| % ≤ A0 (x)

λα1

|Dα2 f1 (s)|

(λα1

λα2

, β D f2 (s)λβ ds

λα1 λβ + λα2 )(λα1 + λα2 + λβ )

λβ /p

(z1 (x))(λα1 +λα2 +λβ /p) + (z2 (x))((λα1 +λα2 +λβ )/p) + λβ /p λβ /p ' x λ α2 ((λα1 +λα2 )/λβ ) (z2 (s)) z1 (s)ds + λα1 + λα2 0 ( x

0

(z1 (s))((λα1 +λα2 )/λβ ) z2 (s)ds

λβ /p

(7.53)

116

7. Riemann–Liouville Fractional Derivatives

%

(7.26)

≤ A0 (x)

(λα1

λ α1 λ β + λα2 )(λα1 + λα2 + λβ )

λβ /p

(z1 (x))((λα1 +λα2 +λβ )/p) + (z2 (x))((λα1 +λα2 +λβ )/p) +

λα2 λα1 + λα2

λβ /p

2 1−(λβ /p) (Γ∗ (x))

λβ /p

,

(7.54)

where ∗

x

Γ (x) :=

(z2 (s))((λα1 +λα2 )/λβ ) z1 (s) + (z1 (s))((λα1 +λα2 )/λβ ) z2 (s) ds.

0

(7.55)

Again see (7.33) and (7.42). Similarly we obtain λβ γ˜1 Γ∗ (x) ≤ (z1 (x) + z2 (x))((λα1 +λα2 +λβ )/λβ ) . λα1 + λα2 + λβ

(7.56)

So by (7.56) we ﬁnd % Ω ≤ A0 (x)

λα2 λα1 + λα2

(λα1

λα1 λβ + λα2 )(λα1 + λα2 + λβ )

λβ /p

(z1 (x))(λα1 +λα2 +λβ /p) + (z2 (x))((λα1 +λα2 +λβ )/p) +

λβ /p

(p−λβ )/p

2

λα1

λβ γ˜1 + λα2 + λβ

λβ /p

-

(z1 (x) + z2 (x))

(λα +λα +λβ /p) 1 2

(7.57) %

by (7.26),(7.27)

≤

A0 (x)

(λα1

λα1 λβ + λα2 )(λα1 + λα2 + λβ )

λβ /p

γ˜2 (z1 (x) + z2 (x))(λα1 +λα2 +λβ /p) +

λα2 λα1 + λα2

λβ /p

2(p−λβ /p)

λα1

λβ γ˜1 + λα2 + λβ

λβ /p

-

(z1 (x) + z2 (x))(λα1 +λα2 +λβ /p)

(7.58) % = A0 (x)

λα2 λα1 + λα2

(λα1

λβ /p

λα1 λβ + λα2 )(λα1 + λα2 + λβ )

2(p−λβ )/p

λα1

λβ γ˜1 + λα2 + λβ

λβ /p γ˜2 +

λβ /p -

(z1 (x) + z2 (x))(λα1 +λα2 +λβ /p)

(7.59)

7.3 Main Results = A0 (x)

(λα1

)

λβ + λα2 )(λα1 + λα2 + λβ )

λβ /p +

x

p(s)[|D f1 (s)| + |D f2 (s)| ]ds β

p

β

λ /p

λαβ1

p

117

γ˜2 + 2(p−λβ )/p (γ˜1 λα2 )λβ /p

,

*(λα1 +λα2 +λβ )/p .

(7.60)

0

We have proved that Ω ≤ A0 (x)

(λα1

)

λβ + λα2 )(λα1 +λα2 +λβ )

λβ /p +

x

p(s)[|D f1 (s)| + |D f2 (s)| ]ds β

p

β

p

λ /p

λαβ1

γ˜2 + 2(p−λβ )/p (γ˜1 λα2 )λβ /p

,

*(λα1 +λα2 +λβ )/p .

(7.61)

0

We have proved (7.47). We need

Theorem 7.7. Let 0 ≤ s ≤ x and f ∈ L∞ ([0, x]), r > 0. Deﬁne s F (s) := (s − t)r f (t)dt.

(7.62)

0

Then there exists

s F (s) = r. (s − t)r−1 f (t)dt, all s ∈ [0, x].

(7.63)

0

Proof. Fix s0 ∈ [0, x] and notice that s0 r F (s0 ) = (s0 − t) f (t)dt = 0

0

x

χ[0,s0 ] (t)(s0 − t)r f (t)dt.

We call g(s, t) := χ[0,s] (t)(s − t)r f (t), which is a Lebesgue integrable function for every s ∈ [0, x]. That is, g(s0 , t) = χ[0,s0 ] (t)(s0 − t)r f (t), all t ∈ [0, x], and F (s0 ) = x g(s0 , t)dt. 0 We would like to study if there exists χ[0,s0 +h] (t)(s0 + h − t)r − χ[0,s0 ] (t)(s0 − t)r ∂g(s0 , t) = f (t) lim . h→0 ∂s h (7.64) We distinguish the following cases. (1) Let x ≥ t > s0 ; then there exist small enough h > 0 such that t ≥ s0 ± h. That is, χ[0,s0 ±h] (t) = χ[0,s0 ] (t) = 0.

118

7. Riemann–Liouville Fractional Derivatives

Hence, there exists ∂g(s0 , t) = 0, ∂s

t : s0 < t ≤ x.

all

(7.65)

(2) Let 0 ≤ t < s0 ; then there exist small enough h > 0 such that t < s0 ± h. That is, χ[0,s0 ±h] (t) = χ[0,s0 ] (t) = 1. In that case ∂g(s0 , t) = f (t) ∂s

(s0 + h − t)r − (s0 − t)r lim h→0 h

= r(s0 − t)r−1 f (t), (7.66)

exists for almost all t : 0 ≤ t < s0 . (3) Let t = s0 . Then we see that ∂g+ (s0 , s0 ) hr r−1 = f (s0 ) lim . = f (s0 ) lim h h→0+ h h→0+ ∂s

(7.67)

The last limit does not exist if 0 < r < 1, equals f (s0 ) if r = 1, and may not exist, and equals 0 if r > 1. Notice also that ∂g− (s0 , s0 ) = f (s0 ) ∂s

lim

h→0−

χ[0,s0 +h] (s0 )hr h

= f (s0 )

lim χ[0,s0 +h] hr−1

h→0−

= 0,

by χ[0,s0 +h] (s0 ) = 0, h < 0. That is, ∂g− (s0 , s0 ) = 0. (7.68) ∂s In general as a conclusion we get that ∂g(s0 , t)/∂s exists for almost all t ∈ [0, x]. Next we deﬁne the diﬀerence quotient at s0 , χ[0,s0 +h] (t)(s0 + h − t)r − χ[0,s0 ] (t)(s0 − t)r Ds0 (h, t) := f (t) , (7.69) h for h = 0, and Ds0 (0, t) := 0. Again we distinguish the following cases. (1) Let x ≥ t > s0 ; then there exist small enough h > 0 such that t > s0 ± h. Clearly then Ds0 (h, t) = 0. (2) Let 0 ≤ t < s0 ; then there exist small enough h > 0 such that t < s0 ± h. Thus (s0 ± h − t)r − (s0 − t)r Ds0 (±h, t) = f (t) . (7.70) ±h

7.3 Main Results

119

Call ρ := s0 − t > 0; clearly ρ ≤ x. Deﬁne ϕ(h) :=

(s0 + h − t)r − (s0 − t)r (ρ + h)r − ρr = h h

(7.71)

for h close to zero, r > 0. That is, Ds0 (h, t) = f (t)ϕ(h). If r = 1, then ϕ(h) = 1 and Ds0 (h, t) = f (t).

(7.72)

We now treat the following subcases. (2(i)) If r > 1, then γ(ρ) := ρr , 0 ≤ ρ ≤ x, is convex and increasing. If h > 0, then by the mean value theorem we get that ϕ(h) < rxr−1 . That is, |Ds0 (h, t)| ≤ rxr−1 |f (t)|.

(7.73)

If h < 0, then, similarly, again we get ϕ(h) =

ρr − (ρ + h)r < rxr−1 . −h

That is, for small |h| we have |Ds0 (h, t)| ≤ rxr−1 |f (t)| ,

r ≥ 1.

(7.74)

(2(ii)) If 0 < r < 1, then γ(ρ) = ρr , 0 ≤ ρ ≤ x is concave and increasing. Let h > 0; then ϕ(h) < ρr−1 = (s0 − t)r−1 and for h < 0, again ϕ(h) ≤ ρr−1 = (s0 − t)r−1 . That is, (7.75) |Ds0 (h, t)| ≤ (s0 − t)r−1 |f (t)|, for small |h|. (3) Case of t = s0 ; then Ds0 (h, s0 ) = f (s0 ) hr−1 ,

for h > 0,

(7.76)

and Ds0 (h, s0 ) = 0,

for h < 0.

(7.77)

So, if r ≥ 1 we obtain |Ds0 (h, s0 )| ≤ |f (s0 )|xr−1 ,

(7.78)

for small |h|. If 0 < r < 1, then for small h > 0 the function Ds0 (h, s0 ) may be unbounded. In conclusion we get:

120

7. Riemann–Liouville Fractional Derivatives

(I) For r ≥ 1, that |Ds0 (h, t)| ≤ rxr−1 ||f (t)||∞ < +∞,

(7.79)

for almost all t ∈ [0, x]. (II) For 0 < r < 1, that |Ds0 (h, t)| ≤ λ(t) for almost all t ∈ [0, x],

where λ(t) :=

(s0 − t)r−1 |f (t)|, 0,

f or

0 ≤ t < s0 , s0 ≤ t ≤ x.

(7.80)

(7.81)

Clearly λ is integrable on [0, x]. Then by Theorem 24.5, pp. 193–194 of [10] we get that ∂g(s0 , ·)/∂s deﬁnes an integrable function, and there exists s0 x ∂g(s0 , t) F (s0 ) = dt = r (s0 − t)r−1 f (t)dt+ ∂s 0 0 x s0 0 dt = r (s0 − t)r−1 f (t)dt. (7.82) 0

s0

That proves the claim.

We proceed with a special important case. Theorem 7.8. Let β > α1 + 1, α1 ∈ R+ , and let f1 , f2 ∈ L1 (0, x), x ∈ R+ − {0} have, respectively, L∞ fractional derivatives Dβ f1 , Dβ f2 in [0, x], and let Dβ−k fi (0) = 0 , for k = 1, . . . , [β] + 1; i = 1, 2. Consider also p(t) > 0 and q(t) ≥ 0, with p(t), 1/p(t), q(t) ∈ L∞ (0, x). Let λα ≥ 0, 0 < λα+1 < 1, and p > 1. Denote λ /(λ ) 2 α α+1 − 1, if λα ≥ λα+1 , (7.83) θ3 := 1, if λα ≤ λα+1 , x (1−λα+1 ) λα+1 θ3 λα+1 (1/(1−λα+1 )) L(x) := 2 (q(s)) ds , (7.84) λα + λα+1 0

and P1 (x) :=

x

(x − s)(β−α1 −1)p/(p−1) (p(s))−1/(p−1) ds,

(7.85)

0

T (x) := L(x)

P1 (x)((p−1)/p) Γ(β − α1 )

(λα +λα+1 ) ,

(7.86)

and ω 1 := 2((p−1)/p)(λα +λα+1 ) ,

(7.87)

Φ(x) := T (x)ω 1 .

(7.88)

7.3 Main Results

121

Then

x 0

)

λα+1 λα+1 * α1 λα α1 +1 λα α1 +1 α1 q(s) D f1 (s) f2 (s) + D f2 (s) f1 (s) ds D D

) ≤ Φ(x)

x

*(λα +λα+1 )/p p β p β p(s) D f1 (s) + D f2 (s) . ds

(7.89)

0

Proof. For convenience, we set α2 := α1 + 1.

(7.90)

By Lemma 7.3, we have Dα1 fi (s) =

1 Γ(β − α1 )

s

(s − t)β−α1 −1 Dβ fi (t)dt,

(7.91)

0

∀ s ∈ [0, x], i = 1, 2, and 1 Γ(β − α1 − 1)

Dα2 fi (s) = Dα1 +1 fi (s) =

s

(s − t)β−α1 −2 Dβ fi (t)dt,

0

(7.92) ∀ s ∈ [0, x], i = 1, 2. By Theorem 7.7 we obtain that

(Dα1 fi (s)) = Dα1 +1 fi (s) = Dα2 fi (s),

(7.93)

∀ s ∈ [0, x], i = 1, 2. So we have 1 |D fj (s)| ≤ Γ(β − αi )

αi

s

(s − t)β−αi −1 |Dβ fj (t)|dt =: gj,αi (s),

(7.94)

0

where j = 1, 2, i = 1, 2, 0 ≤ s ≤ x. We also have by Theorem 7.7 that

(gj,α1 (s)) = gj,α2 (s); gj,αi (0) = 0.

(7.95)

Notice that, if β − α2 = 1, then

s

|Dβ fj (t)|dt.

gj,α2 (s) =

(7.96)

0

Next, we apply H¨ older’s inequality with indices 1/λα+1 , 1/(1 − λα+1 ) to ﬁnd x q(s)|Dα1 f1 (s)|λα |Dα1 +1 f2 (s)|λα+1 ds ≤ 0

122

7. Riemann–Liouville Fractional Derivatives

x

q(s)(g1,α1 (s))λα ((g2,α1 (s)) )λα+1 ds ≤

(7.97)

0

x

(q(s))

(1/1−λα+1 )

(1−λα+1 ) ds

0

x

(g1,α1 (s))

0

λα /λα+1

λα+1 (g2,α1 (s)) ds .

(7.98) Similarly, we get x q(s) |Dα1 f2 (s)|

λα

α +1 D 1 f1 (s)λα+1 ds ≤

0

x

(1−λα+1 ) (q(s)) (1/1−λα+1 ) ds

0

x 0

λα+1 (g2,α1 (s)) λα /λα+1 g1,α1 (s) ds .

(7.99) Adding the last two inequalities (7.98), (7.99), we get

, + q(s) |Dα1 f1 (s)|λα |Dα1 +1 f2 (s)|λα+1 + |Dα1 f2 (s)|λα |Dα1 +1 f1 (s)|λα+1 ds ≤

x 0

x

(1−λα+1 ) ' x λα+1 (q(s)) (1/1−λα+1 ) ds (g1,α1 (s)) λα /λα+1 (g2,α1 (s)) ds +

0

0

x

(g2,α1 (s)) λα /λα+1 (g1,α1 (s)) ds

λα+1 ( (7.100)

0

x (1−λα+1 ) (1/1−λα+1 ) ≤ 2 (q(s)) ds ×

(7.26)

0

)

x

+ , *λα+1 (g1,α1 (s))λα /λα+1 (g2,α1 (s)) + (g2,α1 (s))λα /λα+1 (g1,α1 (s)) ds

0

(7.101) (notice (7.33) and (7.42), so similarly we have) x (1−λα+1 ) λα+1 λα+1 θ3 (1/(1−λα+1 )) ≤ 2 (q(s)) ds λα + λα+1 0 (g1,α1 (x) + g2,α1 (x))(λα +λα+1 )

(7.102)

= L(x)(g1,α1 (x) + g2,α1 (x))(λα +λα+1 )

(7.103)

= )

x

β−α1 −1

(x − t)

L(x) × (Γ(β − α1 ))(λα +λα+1 )

−1/p

(p(t))

1/p

(p(t))

β D f1 (t) + Dβ f2 (t) dt

*(λα +λα+1 )

0

(7.104) (applying H¨ older’s inequality with indices p/(p − 1) and p, we ﬁnd)

7.3 Main Results ≤

L(x) (Γ(β − α1 ))(λα +λα+1 )

)

)

x

123

* (p−1/p)(λα +λα+1 ) (x − t)(β−α1 −1)p/(p−1) (p(t))−1/(p−1) dt

0

*(λα +λα+1 /p)

x

p(t)(|Dβ f1 (t)| + |Dβ f2 (t)|)p dt 0

L(x)(P1 (x))(p−1/p)(λα +λα+1 ) (Γ(β − α1 ))(λα +λα+1 )

)

x

=

(7.105)

p *(λα +λα+1 /p) p(t) Dβ f1 (t) + Dβ f2 (t) dt

0

(7.106) ) = T (x)

x

p p(t) Dβ f1 (t) + Dβ f2 (t) dt

*(λα +λα+1 /p) (7.107)

0

)

x

≤ T (x)2(p−1/p)(λα +λα+1 ) 0

p p *(λα +λα+1 /p) p(t) Dβ f1 (t) + Dβ f2 (t) dt

(7.108) )

x

= Φ(x)

*(λα +λα+1 )/p p β p β dt p(t) D f1 (t) | +| D f2 (t) .

(7.109)

0

We have proved (7.89).

We continue with Theorem 7.9. All here are as in Theorem 7.4. Consider the special case of λα2 = λα1 + λβ . Denote λβ /p λβ ˜ T (x) := A0 (x) 2(p−2λα1 −3λβ )/p . (7.110) λ α1 + λ β Then

x

+ λβ λ λ +λ q(s) |Dα1 f1 (s)| α1 |Dα2 f2 (s)| α1 β Dβ f1 (s) +

0

|Dα2 f1 (s)| T˜(x)

x

λα1 +λβ

|Dα1 f2 (s)|

λα1

, β D f2 (s)λβ ds ≤

p p p(s) Dβ f1 (s) + Dβ f2 (s) ds

2(λα1 +λβ )/p .

(7.111)

0

Proof. We apply (7.21) and (7.22) for λα2 = λα1 + λβ , and by adding we ﬁnd x + λβ λ λ +λ q(s) |Dα1 f1 (s)| α1 |Dα2 f2 (s)| α1 β Dβ f1 (s) + 0

124

7. Riemann–Liouville Fractional Derivatives

|Dα1 f2 (s)| '

λα1

|Dα2 f1 (s)|

x

≤ A0 (x)

λα1 /λβ

(z1 (s))

λα1 +λβ

, β D f2 (s)λβ ds

(z2 (s))λα1 +λβ /λβ z1 (s)ds

0

x (λα1 +λβ )/λβ

+

(z1 (s))

(z2 (s))λα1 /λβ z2 (s)ds

λβ /p

λβ /p ( (7.112)

0

)

(7.26)

≤ A0 (x)2

1−(λβ /p)

x

×

(z1 (s))λα1 /λβ (z2 (s))λα1 /λβ +1 z1 (s)+

0

,λβ /p (z1 (s))(λα1 /λβ )+1 (z2 (s))λα1 /λβ z2 (s) ds = )

x

1−(λβ /p)

λα1 /λβ

A0 (x)2

(z1 (s)z2 (s))

[z2 (s)z1 (s)

+

(7.113)

z1 (s)z2 (s)] ds

*λβ /p =

0

) A0 (x)2(p−λβ )/p

x

(z1 (s)z2 (s))(λα1 /λβ ) (z1 (s)z2 (s)) ds

0

*λβ /p

(7.114) (7.115)

λβ /p λβ = A0 (x)2 (z1 (x)z2 (x)) (7.116) λα1 + λβ λβ /p 2(λα1 +λβ )/p λβ z1 (x) + z2 (x) ≤ A0 (x)2(p−λβ )/p (7.117) λα1 + λβ 2 (p−λβ /p)

(λα1 +λβ /p)

= T˜(x)(z1 (x) + z2 (x))2(λα1 +λβ )/p = *2(λα1 +λβ )/p ) x p β p β ˜ ds T (x) p(s) D f1 (s) + D f2 (s) .

(7.118) (7.119)

0

We have proved (7.111).

Next follow special cases of the above theorems. Corollary 7.10. (to Theorem 7.4) Set λα2 = 0, p(t) = q(t) = 1. Then x+ λβ λβ , λ λ |Dα1 f1 (s)| α1 Dβ f1 (s) + |Dα1 f2 (s)| α1 Dβ f2 (s) ds ≤ 0

) C1 (x)

x

*(λα1 +λβ /p) + , Dβ f1 (s)p + Dβ f2 (s)p ds ,

(7.120)

0

where

C1 (x) := A0 (x)|λα2 =0

λβ λα1 + λβ

λβ /p δ1,

(7.121)

7.3 Main Results

λα1 + λβ ≤ p, λα1 + λβ ≥ p.

21−((λα1 +λβ )/p) , if 1, if

δ 1 :=

125

(7.122)

We ﬁnd that

A0 (x)|λα2 =0 =

(p − 1)((λα1 p−λα1 )/p) (Γ(β − α1 ))λα1 (βp − α1 p − 1)((λα1 p−λα1 )/p)

(p − λβ )((p−λβ )/p) (λα1 βp − λα1 α1 p − λα1 + p − λβ )((p−λβ )/p)

·

×x((λα1 β p− λα1 α1 p−λα1 + p− λβ )/p) .

(7.123)

Proof. Here we need to calculate A0 (x)|λα2 =0 . From (7.6) we have A0 (x)|λα2 =0 =

x

0

p/(p−λβ ) A(s)|λα2 =0 ds

(p−λβ )/p ,

(7.124)

where from (7.5) we get A(s)|λα2 =0 =

(P1 (s))λα1 (p−1)/p , (Γ(β − α1 ))λα1

(7.125)

and here, from (7.4), it is

s

P1 (s) =

(s − t) p(β−α1 −1)/(p−1) dt.

(7.126)

0

Therefore we obtain

P1 (s) =

p−1 β p − α1 p − 1

s((β p−α1 p−1)/(p−1)) ,

(7.127)

and A(s)|λα2 =0 =

p−1 β p − α1 p − 1

(λα

1 p−λα1 /p)

1 (Γ(β − α1 ))λα1

s(λα1 β p−λα1 α1 p−λα1 )/p .

(7.128) That is, A0 (x)|λα2 =0 =

(p − 1)((λα1 p−λα1 )/p) (β p − α1 p − 1)((λα1 p−λα1 )/p) (Γ(β − α1 ))λα1 ((p−λβ )/p)

x ((λα1 β p−λα1 α1 p−λα1 )/(p−λβ ))

s 0

ds

=

126

7. Riemann–Liouville Fractional Derivatives

(p − 1)((λα1 p−λα1 )/p) × (Γ(β − α1 ))λα1 (β p − α1 p − 1)((λα1 p−λα1 )/p) (p − λβ )((p−λβ )/p) (λα1 βp − λα1 α1 p − λα1 + p − λβ )((p−λβ )/p)

(7.129)

×x((λα1 β p− λα1 α1 p−λα1 + p− λβ )/p) . We continue with Corollary 7.11. (to Theorem 7.4: set λα2 = 0, p(t) = q(t) = 1, λα1 = λβ = 1, p = 2.) In detail: let α1 ∈ R+ , β > α1 , β − α1 > (1/2), and let f1 , f2 ∈ L1 (0, x), x ∈ R+ −{0}, have, respectively, L∞ fractional derivatives Dβ f1 , Dβ f2 in [0, x], and let Dβ−k fi (0) = 0 for k = 1, . . . , [β] + 1; i = 1, 2. Then x

α1 |D f1 (s)| Dβ f1 (s) + |Dα1 f2 (s)| Dβ f2 (s) ds ≤

0

x(β−α1 ) √ √ 2Γ(β − α1 ) β − α1 2β − 2α1 − 1

x

)

Dβ f1 (s)

2

2 * ds . + Dβ f2 (s)

0

(7.130)

Proof. We apply Corollary 7.10. Here, δ 1 = 1 and

λβ λ α1 + λ β

λβ /p

1 =√ . 2

(7.131)

Furthermore, we have 1 x(β−α1 ) √ √ √ A0 (x)|λα2 =0 = . (7.132) Γ(β − α1 ) 2β − 2α1 − 1 2 β − α1 Finally, we get C(x) =

x(β−α1 ) √ . 2 Γ(β − α1 ) β − α1 2β − 2α1 − 1 √

(7.133)

We continue with Corollary 7.12. (to Theorem 7.5, λα1 = 0, p(t) = q(t) = 1). It holds x+ λβ λβ , λ λ |Dα2 f2 (s)| α2 Dβ f1 (s) + |Dα2 f1 (s)| α2 Dβ f2 (s) ds ≤ 0

7.3 Main Results

C2 (x)

x

(λβ +λα2 )/p + , Dβ f1 (s)p + Dβ f2 (s)p ds .

127

(7.134)

0

Here C2 (x) := A0 (x)|λα1 =0 2(p−λβ )/p

2λα2 /λβ − 1, 1,

δ 3 :=

λβ λα2 + λβ if if

λβ /p

λ /p

δ3 β ,

λα2 ≥ λβ , λα2 ≤ λβ .

(7.135)

(7.136)

We ﬁnd that

A0 (x)|λα1 =0 =

(p − 1)((λα2 p−λα2 )/p) (Γ(β − α2 ))λα2 (β p − α2 p − 1)((λα2 p−λα2 )/p)

(p − λβ )((p−λβ )/p) (λα2 βp − λα2 α2 p − λα2 + p − λβ )((p−λβ )/p)

·x((λα2 β p− λα2 α2 p− λα2 +p−λβ )/p) .

(7.137)

Proof. Here we need to calculate A0 (x)|λα1 =0 . From (7.6) we have A0 (x)|λα1 =0 =

0

x

p/(p−λβ ) A(s)|λα1 =0 ds

(p−λβ )/p ,

(7.138)

and from (7.5) we get (P2 (s))λα2 (p−1)/p A(s)|λα1 =0 = , (Γ(β − α2 ))λα2 also by (7.4) we have s P2 (s) = (s − t)(β−α2 −1)p/(p−1) dt, 0 ≤ s ≤ x.

(7.139)

(7.140)

0

Therefore, we obtain p−1 P2 (s) = s((β p−α2 p−1)/(p−1)) , β p − α2 p − 1

(7.141)

and (p − 1)((λα2 p−λα2 )/p) s((λα2 β p−λα2 α2 p−λα2 )/p) A(s)|λα1 =0 = . (β p − α2 p − 1)((λα2 p−λα2 )/p) (Γ(β − α2 ))λα2

(7.142)

128

7. Riemann–Liouville Fractional Derivatives

That is,

A0 (x)|λα1 =0 =

(p − 1)((λα2 p−λα2 )/p) (β p − α2 p − 1)((λα2 p−λα2 )/p)(Γ(β−α2 ))

λα 2

·

((p−λβ )/p)

x ((λα2 β p−λα2 α2 p−λα2 )/(p−λβ ))

s

ds

(7.143)

0

=

(p − 1)((λα2 p−λα2 )/p) · (β p − α2 p − 1)((λα2 p−λα2 )/p) (Γ(β − α2 ))λα2 (p − λβ )((p−λβ )/p) (λα2 βp − λα2 α2 p − λα2 + p − λβ )((p−λβ )/p) ×x((λα2 β p− λα2 α2 p−λα2 + p− λβ )/p) .

(7.144)

We give Corollary 7.13. (to Theorem 7.5, λα1 = 0, p(t) = q(t) = 1, λα2 = λβ = 1, p = 2). In detail: let α2 ∈ R+ , β > α2 , β − α2 > (1/2) and let f1 , f2 ∈ L1 (0, x), x ∈ R+ −{0}, have, respectively, L∞ fractional derivatives Dβ f1 , Dβ f2 in [0, x], and let Dβ−k fi (0) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Then x

α2 |D f2 (s)| Dβ f1 (s) + |Dα2 f1 (s)| Dβ f2 (s) ds 0

≤

C2∗ (x)

x

+ , Dβ f1 (s)2 + Dβ f2 (s)2 ds ,

(7.145)

0

where C2∗ (x) := √ 2

x(β−α2 ) . √ √ Γ(β − α2 ) β − α2 2β − 2α2 − 1

λ /p

Proof. We apply Corollary 7.12. Here δ 3 β

Furthermore, we have A0 (x)|λα1 =0 = √

λβ λ α2 + λ β

λβ /p

(7.146)

= 1, 2(p−λβ )/p =

1 =√ . 2

x(β−α2 ) √ √ 2 Γ(β − α2 ) β − α2 2β − 2α2 − 1

√ 2, (7.147)

.

(7.148)

7.3 Main Results

Finally, we ﬁnd

C2∗ (x) = A0 (x)|λα1 =0 ,

129

(7.149)

proving the claim. We continue with Corollary 7.14. (to Theorem 7.6, λα1 = λα2 = λβ = 1, p = 3, p(t) = q(t) = 1). It holds x

α1 |D f1 (s)| |Dα2 f2 (s)| Dβ f1 (s) + |Dα2 f1 (s)| |Dα1 f2 (s)| Dβ f2 (s) ds 0

≤ A0 (x)

√ 3

1 2+ √ 3 6

x

Dβ f1 (s)|3 + Dβ f2 (s)3 ds .

(7.150)

0

Here, A0 (x) =

4x(2β−α1 −α2 ) . Γ(β − α1 )Γ(β − α2 )[3(3β − 3α1 − 1)(3β − 3α2 − 1)(2β − α1 − α2 )]2/3 (7.151)

Proof. We apply inequality (7.47). Here γ˜1 = 3, γ˜2 = 1, and (λα1

λβ + λα2 )(λα1 + λα2 + λβ )

λβ /p

1 . = √ 3 6

(7.152)

Furthermore, +

λλαβ1 /p γ˜2 + 2(p−λβ )/p (γ˜1 λα2 )

The product of the last two quantities is in our case. Here, by (7.6), A0 (x) =

λβ /p

, =1+

√ 3 2+

1 √ 3 6

√ 3 12.

(7.153)

. Next we ﬁnd A0 (x)

2/3

x

(A(s))

3/2

ds

,

(7.154)

0

and, by (7.5), A(s) =

(P1 (s))2/3 (P2 (s))2/3 . Γ(β − α1 )Γ(β − α2 )

(7.155)

Also, by (7.4) we have Pi (s) = 0

s

(s − t)(β−αi −1)3/2 dt, i = 1, 2.

(7.156)

130

7. Riemann–Liouville Fractional Derivatives

That is, Pi (s) =

2 s(3β−3αi −1)/2 , i = 1, 2, (3β − 3αi − 1)

(7.157)

and A(s) =

24/3 s((6β−3α1 −3α2 −2)/3) . (7.158) Γ(β − α1 )Γ(β − α2 )(3β − 3α1 − 1)2/3 (3β − 3α2 − 1)2/3

Finally, we get 24/3 · A0 (x) = Γ(β − α1 )Γ(β − α2 )(3β − 3α1 − 1)2/3 (3β − 3α2 − 1)2/3

x

s((6β−3α1 −3α2 −2)/2) ds

2/3 =

0

4x(2β−α1 −α2 ) . Γ(β − α1 )Γ(β − α2 )[3(3β − 3α1 − 1)(3β − 3α2 − 1)(2β − α1 − α2 )]2/3 (7.159) We give Corollary 7.15. (to Theorem 7.8, here λα = 1, λα+1 = 1/2, p = 3/2, p(t) = q(t) = 1). In detail: let β > α1 + 1, α1 ∈ R+ and let f1 , f2 ∈ L1 (0, x), x ∈ R+ − {0} have, respectively, L∞ fractional derivatives Dβ f1 , Dβ f2 in [0, x], and let Dβ−k fi (0) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Then x+ , |Dα1 f1 (s)| |Dα1 +1 f2 (s)| + |Dα1 f2 (s)| |Dα1 +1 f1 (s)| ds 0

≤ Φ∗ (x)

)

x

* Dβ f1 (s)3/2 + Dβ f2 (s)3/2 ds ,

(7.160)

0

where Φ∗ (x) =

2 x(3β−3α1 −1)/2 √ . (Γ(β − α1 ))3/2 3β − 3α1 − 2

(7.161)

Proof. We apply here Theorem 7.8. We have θ3 = 3, L(x) = P1 (x) = and T (x) =

√

2

√ √ 2 x,

x 3β−3α1 −2 , (3β − 3α1 − 2)

x(3β−3α1 −1)/2 √ (Γ(β − α1 ))3/2 3β − 3α1 − 2

,

(7.162)

7.3 Main Results

131

√ also, ω 1 = 2. Finally, we ﬁnd Φ∗ (x) = which proves the claim.

2 x(3β−3α1 −1)/2 √ , (Γ(β − α1 ))3/2 3β − 3α1 − 2

(7.163)

We continue with Corollary 7.16. (to Theorem 7.9, here p = 2(λα1 + λβ ) > 1, p(t) = q(t) = 1). It holds x+ λβ λ λ +λ |Dα1 f1 (s)| α1 |Dα2 f2 (s)| α1 β Dβ f1 (s) 0

λβ , λ +λ λ ds ≤ + |Dα2 f1 (s)| α1 β |Dα1 f2 (s)| α1 Dβ f2 (s) x 2(λα1 +λβ ) β 2(λα1 +λβ ) β ˜ ˜ T (x) D f1 (s) ds . + D f2 (s)

(7.164)

0

Here, ˜(x) := A˜0 (x) T˜ and

λβ 2(λα1 + λβ )

σ := ·

,

(7.165)

A˜0 (x) := σ σ ∗ xθ ,

where

(λβ /2(λα1 +λβ ))

(Γ(β − α1 ))

λα1

(7.166)

1 (Γ(β − α2 ))λα1 +λβ (2λ2α +2λα1 λβ −λα1 )/(2λα1 +2λβ )

1 2λα1 + 2λβ − 1 2λα1 β + 2λβ β − 2 λα1 α1 − 2λβ α1 − 1 ((2λα1 +2λβ −1)/2) 2λα1 + 2λβ − 1 · , 2λα1 β + 2λβ β − 2λα1 α2 − 2λβ α2 − 1

(7.167)

σ 0 := [(2λα1 + λβ )

2

2

2

2

2

4λα β + 6λα1 λβ β − 2λα α1 − 2λα1 λβ α1 − 2λα α2 − 4λα1 λβ α2 + 2λβ β − 2λβ α2 1

1

1

(2λα1 +λβ )/2(λα1 +λβ )

σ ∗ := σ 0

and θ :=

,

4λ2α1 β + 6λα1 λβ β − 2λ2α1 α1 − 2λα1 λβ α1 + 2λα1 + 2λβ

−1 *

,

(7.168)

132

7. Riemann–Liouville Fractional Derivatives

−2λ2α1 α2 − 4λα1 λβ α2 + 2λ2β β − 2λ2β α2 . 2λα1 + 2λβ

(7.169)

Proof. We apply Theorem 7.9. The constant T˜˜(x) comes from (7.110) in our case. Still, we need to determine A˜0 (x). From (7.6) we get here 10 (x) = A

((2λα1 +λβ )/2(λα1 +λβ ))

x 2(λα1 +λβ )/(2λα1 +λβ )

(A(s))

ds

,

0

(7.170) where, from (7.5) we get also here # $ P1 (s)λα1 ((2λα1 +2λβ −1)/2(λα1 +λβ )) P2 (s)((2λα1 +2λβ −1)/2) A(s) = , (Γ(β − α1 ))λα1 (Γ(β − α2 ))λα1 +λβ (7.171) and, from (7.4), we ﬁnd for i = 1, 2 that s (s − t)[(2λα1 +2λβ )(β−αi −1)/(2λα1 +2λβ −1)] dt, (7.172) Pi (s) = 0

all 0 ≤ s ≤ x. Hence for i = 1, 2 we obtain Pi (s) =

(2λα1 + 2λβ − 1)s((2λα1 β+2λβ β−2λα1 αi −2λβ αi −1)/(2λα1 +2λβ −1)) , (2λα1 β + 2λβ β − 2λα1 αi − 2λβ αi − 1) (7.173)

and A(s) = σs

⎫ ⎧ ⎨ (4λ2 β+6λα λβ β−2λ2 α1 −2λ2 α2 −2λα λβ α1 −4λα λβ α2 −2λ2 α2 +2λ2 β−2λα −λβ ) ⎬ α1 α1 α1 β β 1 1 1 1 (2λα1 +2λβ ) ⎭ ⎩

(7.174) Finally, we ﬁnd ˜ 0) = σ A(x ⎛ ⎜ ⎜ ⎜ ⎜ ⎜ x ⎜ s ⎜ ⎜ 0 ⎜ ⎜ ⎝

2 2 (4λ2 α β + 6λα1 λβ β − 2λα α1 − 2λα α2 − 2λα1 λβ α1 1

−4λα1 λβ α2 −

1 1 2 2λ2 β α2 + 2λβ β − 2λα1 − (2λα +λβ ) 1

λβ )

= σσ ∗ xθ .

⎞ (2λα +λβ )/2(λα +λβ ) 1 1 ⎟ ⎟ ⎟ ⎟ ⎟ ds⎟ ⎟ ⎟ ⎟ ⎟ ⎠

(7.175)

.

7.3 Main Results

133

We present the following interesting special case. Corollary 7.17. (to Theorem 7.9, here p = 4, λα1 = λβ = 1, p(t) = q(t) = 1). It holds x α1

|D f1 (s)|(Dα2 f2 (s))2 |Dβ f1 (s)| + (Dα2 f1 (s))2 |Dα1 f2 (s)||Dβ f2 (s)| ds 0 ∗

≤ T (x)

x

4 β 4 β D f1 (s) + D f2 (s) ds .

(7.176)

0

Here T ∗ (x) = and

A∗0 (x) √ , 2

(7.177)

˜

A∗0 (x) = σ ˜σ ˜ ∗ xθ ,

(7.178)

where σ ˜

1 Γ(β − α1 )(Γ(β − α2 ))2 3/2 3 , 4β − 4α2 − 1

: =

σ ˜ ∗ :=

3 4β − 4α1 − 1

3 12β − 4α1 − 8α2

3/4

(7.179) 3/4 ,

(7.180)

and ˜θ := 3 β − α1 − 2α2 .

Proof. By Corollary 7.16.

(7.181)

We continue related results regarding the L∞ − norm · ∞ . Theorem 7.18. Let α1 , α2 ∈ R+ , β > α1 , α2 , and let f1 , f2 ∈ L1 (0, x), x ∈ R+ −{0}, have, respectively, L∞ fractional derivatives Dβ f1 , Dβ f2 in [0, x], and let Dβ−k fi (0) = 0 for k = 1, . . . , [β] + 1; i = 1, 2. Consider p(s) ≥ 0, p(s) ∈ L∞ (0, x). Let λα1 , λα2 , λβ ≥ 0. Set ρ(x) := %

p(s)∞ x(βλα1 −α1 λα1 +βλα2 −α2 λα2 +1) . (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα2 [βλα1 − α1 λα1 + βλα2 − α2 λα2 + 1] (7.182)

134

7. Riemann–Liouville Fractional Derivatives

Then

x

+ λβ λ λ p(s) |Dα1 f1 (s)| α1 |Dα2 f2 (s)| α2 Dβ f1 (s) +

0

|Dα2 f1 (s)|

λα2

|Dα1 f2 (s)|

λα1

, β D f2 (s)λβ ds ≤

2λα 2λα 2(λα +λβ ) , ρ(x) + β 2(λ +λ ) . D f1 ∞ α1 β + Dβ f1 ∞ 2 + Dβ f2 ∞ 2 + Dβ f2 ∞ 1 2 (7.183) Proof. From (7.9) we get s 1 |D fj (s)| ≤ (s − t)β−αi −1 |Dβ fj (t)|dt Γ(β − αi ) 0 s 1 β−αi −1 (s − t) dt Dβ fj ∞ ≤ Γ(β − αi ) 0 αi

=

sβ−αi Dβ fj ∞ sβ−αi = Dβ fj ∞ . Γ(β − αi ) (β − αi ) Γ(β − αi + 1)

(7.184) (7.185)

That is, |Dαi fj (s)| ≤

sβ−αi Dβ fj ∞ , Γ(β − αi + 1)

(7.186)

∀ s ∈ [0, x], i = 1, 2; j = 1, 2. Then we have |Dα1 f1 (s)|λα1 ≤

s(β−α1 )λα1 λ Dβ f1 ∞α1 , (Γ(β − α1 + 1))λα1

(7.187)

|Dα2 f2 (s)|λα2 ≤

s(β−α2 )λα2 λ Dβ f2 ∞α2 , (Γ(β − α2 + 1))λα2

(7.188)

λ

|Dβ f1 (s)|λβ ≤ Dβ f1 ∞β .

(7.189)

Multiplying the last three inequalities (7.187) through (7.189), we obtain |Dα1 f1 (s)|λα1 |Dα2 f2 (s)|λα2 |Dβ f1 (s)|λβ ≤ s(βλα1 −α1 λα1 +βλα2 −α2 λα2 ) λ +λ λ Dβ f1 ∞α1 β Dβ f2 ∞α2 . (7.190) (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα2 Similarly we see that |Dα2 f1 (s)|λα2 ≤

s(β−α2 )λα2 λ Dβ f1 ∞α2 , (Γ(β − α2 + 1))λα2

(7.191)

|Dα1 f2 (s)|λα1 ≤

s(β−α1 )λα1 λ Dβ f2 ∞α1 , λα1 (Γ(β − α1 + 1))

(7.192)

7.3 Main Results λ

|Dβ f2 (s)|λβ ≤ Dβ f2 ∞β .

135

(7.193)

Multiplying the last three inequalities (7.191) through (7.193), we get |Dα2 f1 (s)|λα2 |Dα1 f2 (s)|λα1 |Dβ f2 (s)|λβ ≤ s(βλα2 −α2 λα2 +βλα1 −α1 λα1 ) λ λ +λ Dβ f1 ∞α2 Dβ f2 ∞α1 β . (7.194) (Γ(β − α2 + 1))λα2 (Γ(β − α1 + 1))λα1 Adding (7.190), (7.194) produces |Dα1 f1 (s)|λα1 |Dα2 f2 (s)|λα2 |Dβ f1 (s)|λβ + |Dα2 f1 (s)|λα2 |Dα1 f2 (s)|λα1 |Dβ f2 (s)|λβ

s(βλα1 −α1 λα1 +βλα2 −α2 λα2 ) · (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα2 + λα λα λα +λβ , λ +λ . Dβ f1 ∞α1 β Dβ f2 ∞ 2 + Dβ f1 ∞ 2 Dβ f2 ∞ 1 ≤

It follows that x

(7.195)

p(s) |Dα1 f1 (s)|λα1 |Dα2 f2 (s)|λα2 |Dβ f1 (s)|λβ +

0

|Dα2 f1 (s)|λα2 |Dα1 f2 (s)|λα1 |Dβ f2 (s)|λβ ds ≤ x p(s)∞ 0 s(βλα1 −α1 λα1 +βλα2 −α2 λα2 ) ds · (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα2 + λα λα λα +λβ , λ +λ Dβ f1 ∞α1 β Dβ f2 ∞ 2 + Dβ f1 ∞ 2 Dβ f2 ∞ 1 =

(7.196)

p(s)∞ x(βλα1 −α1 λα1 +βλα2 −α2 λα2 +1) · (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα2 [βλα1 − α1 λα1 + βλα2 − α2 λα2 + 1]

+

λ

Dβ f1 ∞α1 )

+λβ

β λα2 β λα2 β λα1 +λβ , D f2 + D f1 ∞ D f2 ∞ ∞

(7.197)

* 2λα ρ(x) 2(λα +λβ ) 2 β 2λα2 β 2(λα1 +λβ ) ≤ + Dβ f2 + D f1 + D f2 Dβ f1 ∞ 1 . 2 ∞ ∞ ∞ (7.198)

We have proved (7.183).

We present some special cases of the last theorem. Theorem 7.19. (all are as in Theorem 7.18; λα2 = 0). It holds x

p(s) |Dα1 f1 (s)|λα1 |Dβ f1 (s)|λβ + |Dα1 f2 (s)|λα1 |Dβ f2 (s)|λβ ds ≤ 0

+ β λα1 +λβ , p(s)∞ x(βλα1 −α1 λα1 +1) λα1 +λβ β D f2 D . f + ∞ 1 ∞ (Γ(β − α1 + 1))λα1 [βλα1 − α1 λα1 + 1] (7.199)

136

7. Riemann–Liouville Fractional Derivatives

Proof. As in Theorem 7.18, similarly.

We continue with Theorem 7.20. (all are as in Theorem 7.18; λα2 = λα1 + λβ ). It holds x p(s) |Dα1 f1 (s)|λα1 |Dα2 f2 (s)|λα1 +λβ |Dβ f1 (s)|λβ + 0

|Dα2 f1 (s)|λα1 +λβ |Dα1 f2 (s)|λα1 |Dβ f2 (s)|λβ ds ≤

p(s)∞ (Γ(β − α1 + 1)) (Γ(β − α2 + 1))λα1 +λβ λα1

x(2βλα1 −α1 λα1 +βλβ −α2 λα1 −α2 λβ +1) (2βλα1 − α1 λα1 + βλβ − α2 λα1 − α2 λβ + 1) + 2(λα1 +λβ ) , 2(λ +λ ) . Dβ f1 ∞ α1 β + Dβ f2

∞

Proof. As in Theorem 7.18, similarly.

(7.200)

We give Theorem 7.21. (all are as in Theorem 7.18; λβ = 0, λα1 = λα2 ). It holds x

p(s) |Dα1 f1 (s)|λα1 |Dα2 f2 (s)|λα1 + |Dα2 f1 (s)|λα1 |Dα1 f2 (s)|λα1 ds 0

+ 2λα , 2λ ≤ ρ∗ (x) Dβ f1 ∞ α1 + Dβ f2 ∞ 1 ,

(7.201)

where % ∗

ρ (x) :=

p(s)∞ x(2βλα1 −α1 λα1 −α2 λα1 +1) (Γ(β − α1 + 1)Γ(β − α2 + 1))λα1 (2βλα1 − α1 λα1 − α2 λα1 + 1)

.

(7.202)

Proof. As in Theorem 7.18, when λβ = 0, we follow the proof and we obtain x

p(s) |Dα1 f1 (s)|λα1 |Dα2 f2 (s)|λα2 + |Dα2 f1 (s)|λα2 |Dα1 f2 (s)|λα1 ds 0

+ λα λα , λ λ ≤ ρ(x) Dβ f1 ∞α1 Dβ f2 ∞ 2 + Dβ f1 ∞α2 Dβ f2 ∞ 1 ,

(7.203)

7.3 Main Results

137

which inequality is in itself of interest, and so on. Then setting λα1 = λα2 into (7.203) and (7.182), we ﬁnally derive (7.201) and (7.202). We give Theorem 7.22. (all are as in Theorem 7.18; λα1 = 0, λα2 = λβ ). It holds x

p(s) |Dα2 f2 (s)|λα2 |Dβ f1 (s)|λα2 + |Dα2 f1 (s)|λα2 |Dβ f2 (s)|λα2 ds 0

≤

(β λα2

x(βλα2 −α2 λα2 +1) p(s)∞ − α2 λα2 + 1)(Γ(β − α2 + 1))λα2

)

2λα2

Dβ f1 ∞

2λα * 2 + Dβ f2 . ∞

(7.204)

Proof. As in Theorem 7.18.

We continue with Corollary 7.23. (to Theorem 7.21; all are as in Theorem 7.18; λβ = 0, λα1 = λα2 , α2 = α1 + 1). It holds x

p(s) |Dα1 f1 (s)|λα1 |Dα1 +1 f2 (s)|λα1 + |Dα1 +1 f1 (s)|λα1 |Dα1 f2 (s)|λα1 ds 0

≤

(2β λα1

x(2βλα1 −2α1 λα1 −λα1 +1) p(s)∞ · − 2α1 λα1 − λα1 + 1)(β − α1 )λα1 (Γ(β − α1 ))2λα1 + 2λα , 2λ (7.205) Dβ f1 ∞ α1 + Dβ f2 ∞ 1 .

Proof. By Theorem 7.21.

We give Corollary 7.24. (to Corollary 7.23) In detail: let α1 ∈ R+ , β > α1 + 1 and let f1 , f2 ∈ L1 (0, x), x ∈ R+ − {0}, have, respectively, L∞ fractional derivative Dβ f1 , Dβ f2 in [0, x], and let Dβ−k fi (0) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Then x α1

|D f1 (s)| |Dα1 +1 f2 (s)| + |Dα1 +1 f1 (s)| |Dα1 f2 (s)| ds 0

138

7. Riemann–Liouville Fractional Derivatives

≤

x2(β−α1 ) 2(β − α1 )2 (Γ(β − α1 ))2

+ 2 , Dβ f1 2∞ + Dβ f2 ∞ .

(7.206)

Proof. In (7.205) set λα1 = 1 and p(s) = 1. Proposition 7.25. Inequality (7.206) is sharp; in fact it is attained when f1 = f2 . Proof. Clearly (7.206) when f1 = f2 collapses to x x2(β−α1 ) |Dα1 f1 (s)| |Dα1 +1 f1 (s)|ds ≤ Dβ f1 2∞ . 2(β − α1 )2 (Γ(β − α1 ))2 0 (7.207) In [64] (see Theorem 2.5, Remark 2.6) we proved that (7.207) is sharp; in fact it is attained by f1 (s) = sβ .

7.4 Applications We present a uniqueness of solution result for a system of fractional diﬀerential equations involving Riemann–Liouville fractional derivatives. Theorem 7.26. Let αi ∈ R+ , β > αi , β − αi > (1/2), i = 1, . . . , r ∈ N, and let fj ∈ L1 (0, x), j = 1, 2; x ∈ R+ − {0}, have, respectively, L∞ fractional derivative Dβ fj in [0, x], and let Dβ−k fj (0) = akj ∈ R for k = 1, . . . , [β] + 1; j = 1, 2. Furthermore for j = 1, 2 we have that r

r

Dβ fj (s) = Fj (s, {Dαi f1 (s)}i=1 , {Dαi f2 (s)}i=1 ) ,

(7.208)

all s ∈ [0, x]. Here Fj (s, z1 , . . . , zr , y1 , . . . , yr ) are continuous for (z1 , . . . , zr , y1 , . . . , yr ) ∈ R2r , bounded for s ∈ [0, x], and satisfy the Lipschitz condition |Fj (s, z1 , . . . , zr , y1 , . . . , yr ) − Fj (s, z1 , . . . , zr , y1 , . . . , yr )| ≤ r

[q1,i,j (s)|zi − zi | + q2,i,j (s)|yi − yi |] ,

(7.209)

i=1

j = 1, 2, where q1,i,j (s), q2,i,j (s) ≥ 0 are bounded on [0, x], 1 ≤ i ≤ r. Call M1,i := max(q1,i,1 ∞ , q2,i,2 ∞ ) and M2,i := max(q2,i,1 ∞ , q1,i,2 ∞ ). (7.210)

7.4 Applications

139

Assume here that r M1,i M2,i xβ−αi √ √ φ∗ (x) := + √ < 1. 2 Γ(β − αi ) β − αi 2β − 2αi − 1 2 i=1 (7.211) Then, if the system (7.208) has two pairs of solutions (f1 , f2 ) and (f1∗ , f2∗ ), we prove that fj = fj∗ , j = 1, 2; that is, we have the uniqueness of solution property for the fractional system (7.208). Proof. Assume there are two pairs of solutions (f1 , f2 ), (f1∗ , f2∗ ) to system (7.208). Set gj := fj − fj∗ , j = 1, 2 . Then Dβ−k gj (0) = 0, k = 1, ..., [β] + 1; j = 1, 2. It holds

r

(7.212)

r

Dβ gj (s) = Fj (s, {Dαi f1 (s)}i=1 , {Dαi f2 (s)}i=1 ) −Fj (s, {Dαi f1∗ (s)}i=1 , {Dαi f2∗ (s)}i=1 ) . r

r

(7.213)

Hence by (7.209) we have |Dβ gj (s)| ≤

r

[g1,i,j (s)|Dαi g1 (s)| + g2,i,j (s)|Dαi g2 (s)|].

(7.214)

[g1,i,j ∞ |Dαi g1 (s)| + g2,i,j ∞ |Dαi g2 (s)|].

(7.215)

i=1

Thus r

|Dβ gj (s)| ≤

i=1

The last implies that (Dβ gj (s))2 ≤

r g1,i,j ∞ |Dαi g1 (s)| |Dβ gj (s)|+ i=1

g2,i,j ∞ |Dαi g2 (s)| |Dβ gj (s)| ,

j = 1, 2.

(7.216)

Consequently we see that x β (D g1 (s))2 + (Dβ g2 (s))2 ds ≤ I := 0 r ) i=1

g1,i,1 ∞

x 0

|Dαi g1 (s)| |Dβ g1 (s)|ds + g2,i,1 ∞

+

r ) g1,i,2 ∞ i=1

0

* |Dαi g2 (s)| |Dβ g1 (s)|ds

x 0

x

|Dαi g1 (s)| |Dβ g2 (s)|ds+

140

7. Riemann–Liouville Fractional Derivatives

g2,i,2 ∞ ≤

r

r

|D g2 (s)| |D g2 (s)|ds αi

β

(7.217)

0

αi

|D g1 (s)| |Dβ g1 (s)| + |Dαi g2 (s)| |Dβ g2 (s)| ds

x

M1,i

0

i=1

+

*

x

αi

|D g2 (s)| |Dβ g1 (s)| + |Dαi g1 (s)| |Dβ g2 (s)| ds

x

M2,i

=: (∗).

0

i=1

(7.218) However by Corollary 7.11, (7.130) we have x [|Dαi g1 (s)| |Dβ g1 (s)| + |Dαi g2 (s)| |Dβ g2 (s)|]ds 0

≤

x(β−αi ) √ √ 2Γ(β − αi ) β − αi 2β − 2αi − 1

I.

(7.219)

Also by Corollary 7.13, (7.145) and (7.146) we ﬁnd x [|Dαi g2 (s)| |Dβ g1 (s)| + |Dαi g1 (s)| |Dβ g2 (s)|]ds 0

√

≤

x(β−αi ) √ √ 2Γ(β − αi ) β − αi 2β − 2αi − 1

I.

(7.220)

Therefore by (7.219) and (7.220) we get (∗) ≤

r

M1,i

i=1 r i=1

M2,i

x(β−αi ) I √ √ 2Γ(β − αi ) β − αi 2β − 2αi − 1

x(β−αi ) I √ √ √ 2Γ(β − αi ) β − αi 2β − 2αi − 1

We have established that

+

= φ∗ (x)I.

I ≤ φ∗ (x) I.

(7.221)

(7.222)

(7.223)

∗

If I = 0 then φ (x) ≥ 1, a contradiction by assumption (7.211) that φ∗ (x) < 1. Therefore I = 0, implying that (Dβ g1 (s))2 + (Dβ g2 (s))2 = 0, a.e. in [0, x].

(7.224)

That is, 2 2 β D g1 (s) = 0, Dβ g2 (s) = 0, a.e.

in [0, x].

(7.225)

That is, Dβ g1 (s) = 0, Dβ g2 (s) = 0, a.e. in [0, x].

(7.226)

7.4 Applications

141

By (7.212) and Lemma 7.3, apply (7.3) for α = 0; we ﬁnd g1 (s) ≡ g2 (s) ≡ 0 over [0, x]. The last implies fj = fj∗ , j = 1, 2, over [0, x], thus proving the uniqueness of solution to the initial value problem of this theorem. Next we give upper bounds on Dβ fj , solutions fj , and so on, all involved in a system of fractional diﬀerential equations involving Riemann–Liouville fractional derivatives. Theorem 7.27. Let αi ∈ R+ , β > αi , β − αi > (1/2), i = 1, . . . , r ∈ N, and let fj ∈ L1 (0, x), j = 1, 2; x ∈ R+ − {0}, have, respectively, fractional derivative Dβ fj in [0, x] that is absolutely continuous on [0, x], and let Dβ−k fj (0) = 0 for k = 1, . . . , [β] + 1; j = 1, 2. Furthermore (Dβ fj )(0) = Aj ∈ R. Furthermore for 0 ≤ s ≤ x we have holding the system of fractional diﬀerential equations

Dβ fj

# # $r $r (s) = Fj s, Dαi f1 (s) i=1 , Dβ f1 (s), Dαi f2 (s) i=1 , Dβ f2 (s) , (7.227)

for j = 1, 2. Here Fj is Lebesgue measurable on [0, x] × Rr+1 × Rr+1 such that |Fj (s, x1 x2 , ..., xr , xr+1 , y1 , ..., yr , yr+1 )| ≤

r

q1,i,j (s) · |xi | + q2,i,j (s)|yi | ,

i=1

(7.228)

where q1,i,j (s), q2,i,j (s) ≥ 0, 1 ≤ i ≤ r, j = 1, 2, are bounded on [0,x]. Call M1,i := max(q1,i,1 ∞ , q2,i,2 ∞ ),

(7.229)

M2,i := max(q2,i,1 ∞ , q1,i,2 ∞ ).

(7.230)

and Also we set (0 ≤ s ≤ x) θ(s) := ((Dβ f1 )(s))2 + ((Dβ f2 )(s))2 ,

(7.231)

(7.232) ρ := A21 + A22 , r β−α i √ s √ √ Q(s) := (M1,i + 2 M2,i ) , Γ(β − αi ) β − αi 2β − 2αi − 1 i=1 (7.233) and ) √ ( 0s Q(t)dt) · χ(s) := ρ· 1 + Q(s) · e

s

−(

e

z 0

Q(t)dt)

*1/2

dz

. (7.234)

0

Then

θ(s) ≤ χ(s),

all

0 ≤ s ≤ x.

(7.235)

142

7. Riemann–Liouville Fractional Derivatives

Consequently it holds |Dβ fj (s)| ≤ χ(s), s 1 |fj (s)| ≤ (s − t)β−1 χ(t)dt, Γ(β) 0 for j = 1, 2, 0 ≤ s ≤ x. Also it holds that

1 |D fj (s)| ≤ Γ(β − αi ) αi

s

(7.236) (7.237)

(s − t)β−αi −1 χ(t)dt,

(7.238)

0

for j = 1, 2, i = 1, . . . , r, all 0 ≤ s ≤ x. Proof. We observe that r (Dβ fj )(s)(Dβ fj ) (s) = Dβ fj )(s) · Fj (s, {Dαi f1 (s)}i=1 , β r D f1 (s), {Dαi f2 (s)}i=1 , Dβ f2 (s) ,

(7.239)

j = 1, 2, all 0 ≤ s ≤ x. We then integrate (7.239), y y r (Dβ fj )(s)(Dβ fj ) (s)ds = (Dβ fj )(s)·Fj s, {Dαi f1 (s)}i=1 , Dβ f1 (s), 0

0

r {Dαi f2 (s)}i=1 , Dβ f2 (s) ds, 0 ≤ y ≤ x.

(7.240)

Hence we obtain y y (7.228) ((Dβ fj )(s))2 ≤ |Dβ fj (s)| |Fj · · · |ds ≤ 2 0 0 ' ( y r β αi αi |D fj (s)| [g1,i,j (s)|D f1 (s)| + g2,i,j (s)|D f2 (s)|] ds 0

i=1

≤

r g1,i,j ∞

(7.241)

y

|Dβ fj (s)| |Dαi f1 (s)|ds

0

i=1

y

+ g2,i,j ∞

|D fj (s)| |D f2 (s)|ds β

αi

.

(7.242)

0

Thus we ﬁnd ((Dβ fj (y))2 ≤ A2j + 2

r

g1,i,j ∞

y

|Dβ fj (s)| |Dαi f1 (s)|ds 0

i=1

+ g2,i,j ∞

y

|D fj (s)| |D f2 (s)|ds β

0

αi

.

(7.243)

7.4 Applications

143

Consequently we write r

θ(y) ≤ ρ+2

+ M2,i

y

|Dβ f1 (s)| |Dαi f1 (s)| + |Dβ f2 (s)| |Dαi f2 (s)| ds

0

i=1

y

M1,i

|D f2 (s)| |D f1 (s)| + |D f1 (s)| |D f2 (s)| ds β

αi

0

%

β

αi

(7.244)

y y (β−αi ) 0 θ(s)ds √ √ ≤ ρ+2 M1,i 2Γ(β − αi ) β − αi 2β − 2αi − 1 i=1 y y (β−αi ) 0 θ(s)ds (7.245) +M2,i √ √ √ 2Γ(β − αi ) β − αi 2β − 2αi − 1 y = ρ + Q(y) θ(s)ds. (7.246)

(by (7.130),(7.145))

r

0

That is, we get

y

θ(y) ≤ ρ + Q(y)

θ(s)ds,

(7.247)

0

all 0 ≤ y ≤ x . Here ρ ≥ 0, Q(y) ≥ 0, Q(0) = 0, θ(y) ≥ 0 , all 0 ≤ y ≤ x . Solving the integral inequality (7.247), exactly as in [17] (see pp. 224–225, Application 3.2 and inequalities (44), (47) there) we ﬁnd that θ(y) ≤ χ(y), 0 ≤ y ≤ x, (7.248) proving (7.235). Then (7.236) is obvious. Next from (7.3), the case of α = 0, we obtain s s 1 1 β−1 β |fj (s)| ≤ (s − t) |D fj (t)|dt ≤ (s − t)β−1 χ(t)dt, Γ(β) 0 Γ(β) 0 (7.249) proving (7.237). Finally, again from (7.3), we ﬁnd s 1 (s − t)β−αi −1 |Dβ fj (t)|dt (7.250) |Dαi fj (s)| ≤ Γ(β − αi ) 0 s 1 ≤ (s − t)β−αi −1 χ(t)dt, (7.251) Γ(β − αi ) 0 thus proving (7.238).

In the last application we give similar upper bounds as in Theorem 7.27, but under diﬀerent conditions.

144

7. Riemann–Liouville Fractional Derivatives

Theorem 7.28. Let αi ∈ R+ , β > αi i = 1, . . . , r ∈ N, and fj ∈ L1 (0, x), j = 1, 2; x ∈ R+ − {0}, have, respectively, fractional derivative Dβ fj in [0, x] that is absolutely continuous on [0, x], and let Dβ−μ fj (0) = 0 for μ = 1, . . . , [β] + 1; j = 1, 2. And (Dβ fj )(0) = Aj ∈ R. Furthermore for 0 ≤ s ≤ x we have holding the system of fractional diﬀerential equations β r D fj (s) = Fj s, {Dαi f1 (s)}i=1 , (Dβ f1 )(s), r {Dαi f2 (s)}i=1 , Dβ f2 (s) , for j = 1,2. (7.252) For ﬁxed i∗ ∈ {1, ..., r} we assume that αi∗ +1 = αi∗ +1 , where αi∗ , αi∗ +1 ∈ {α1 , . . . , αr }. Call k := αi∗ , γ := αi∗ + 1; that is, γ = k + 1. Here Fj is Lebesgue measurable on [0, x] × Rr+1 × Rr+1 such that |Fj (s, x1 , ..., xr , xr+1 , y1 , ..., yr , yr+1 )| ≤ q1,j (s)|xi∗ | |yi∗ +1 | + q2,j (s)|yi∗ | |xi∗ +1 |,

(7.253)

where both 0 ≡ q1,j , q2,j ≥ 0, bounded functions over [0, x]. Put M := max(q1,1 ∞ , q2,1 ∞ , q1,2 ∞ , q2,2 ∞ ), θ(s) := |(Dβ f1 )(s)| + |(Dβ f2 )(s)|,

0 ≤ s ≤ x,

(7.254)

ρ := |A1 | + |A2 |. (7.255)

Also we deﬁne

((3β−3k−1)/2) 4M s Q(s) := √ , 3β − 3k − 2 (Γ(β − k))3/2 ((3β−3k−1)/2) 4M x , σ := √ (Γ(β − k))3/2 3β − 3k − 2 all 0 ≤ s ≤ x. We assume that Call

(7.256) (7.257)

√ xσ ρ < 2.

(7.258)

s(σ − Q(s)) σρ2 s − 4ρ3/2 + 4ρ , ϕ ˜ (s) := √ (2 − σ ρ s)2

(7.259)

all 0 ≤ s ≤ x. Then

θ(s) ≤ ϕ ˜ (s);

(7.260)

in particular, it holds |Dβ fj (s)| ≤ ϕ ˜ (s),

j = 1, 2,

for all 0 ≤ s ≤ x. Furthermore we ﬁnd s 1 (s − t)β−1 ϕ ˜ (t)dt, |fj (s)| ≤ Γ(β) 0

(7.261)

(7.262)

7.4 Applications

1 |D fj (s)| ≤ Γ(β − αi )

s

αi

(s − t)β−αi −1 ϕ ˜ (t)dt,

145

(7.263)

0

j = 1, 2; i = 1, . . . , r, all 0 ≤ s ≤ x. Proof. Clearly, here Dβ fj are L∞ fractional derivatives in [0, x]. For 0 ≤ s ≤ y ≤ x, by (7.252) we get that y y β r D fj (s)ds = Fj (s, {Dαi f1 (s)}i=1 , Dβ f1 (s), 0

0

r {D f2 (s)}i=1 , Dβ f2 (s) ds. αi

That is,

(7.264)

y

(Dβ fj )(y) = Aj +

Fj (s, . . .)ds.

(7.265)

0

Then we observe

y

|(Dβ fj )(y)| ≤ |Aj | +

|Fj (s, ...)|ds ≤ |Aj |+ 0

y

+

q1,j ∞ |(Dk f1 )(s)|

|(Dk+1 f2 )(s)| + q2,j ∞ |(Dk f2 )(s)|

, |(Dk+1 f1 )(s)| ds

0

(7.266) , y+ |(Dk f1 )(s)| |(Dk+1 f2 )(s)| + |Dk f2 (s)| |(Dk+1 f1 )(s)| ds .

≤ |Aj | + M 0

(7.267)

Therefore

y

θ(y) ≤ ρ + 2M

+

|Dk f1 (s)|

|Dk+1 f2 (s)| + |Dk f2 (s)|

, |Dk+1 f1 (s)| ds

0

(7.268)

y (3β−3k−1/2) (Γ(β − k))3/2 y β 3/2 β 3/2 · (|D f1 )(s)| + |(D f2 (s)| )ds

((7.160),(7.161))

≤

ρ+

0

(7.27)

≤

ρ+

√

4M 3β − 3k − 2

4M √ 3β − 3k − 2

y (3β−3k−1/2) (Γ(β − k))3/2

y

·

(7.269)

(|Dβ f1 (s)| + |Dβ f2 (s)|)3/2 ds .

0

(7.270)

We have proved that

y

θ(y) ≤ ρ + Q(y)

(θ(s))3/2 ds,

(7.271)

0

all 0 ≤ y ≤ x. Notice that θ(y) ≥ 0, ρ ≥ 0, Q(y) ≥ 0, and Q(0) = 0, also it is σ > 0.

146

7. Riemann–Liouville Fractional Derivatives

Call

0 ≤ w(y) :=

y

(θ(s))3/2 ds, w(0) = 0, all 0 ≤ y ≤ x.

(7.272)

0

That is, and

w (y) = (θ(y))3/2 ≥ 0 θ(y) = (w (y))2/3 ,

over

all

[0, x],

(7.273)

0 ≤ y ≤ x.

(7.274)

Hence we rewrite (7.271) as follows, (w (y))2/3 ≤ ρ + Q(y)w(y), So that

all

0 ≤ y ≤ x.

(w (y))2/3 ≤ ρ + σ w(y) < ρ + ε + σ w(y),

(7.275) (7.276)

for ε > 0 arbitrarily small. Thus w (y) < (ρ + ε + σ w(y))3/2 ,

all

0 ≤ y ≤ x.

(7.277)

Here (ρ + ε + σ w(y))3/2 > 0 . In particular it holds w (s) < (ρ + ε + σ w(y))3/2 , and

all

0 ≤ s ≤ y,

w (s) < 1. (ρ + ε + σ w(s))3/2

(7.279)

The last is the same as 2 − (ρ + ε + σ w(s))−1/2 < 1. σ

(7.280)

Therefore after integration we get y σ − d((ρ + ε + σ w(s))−1/2 ) ≤ y, 2 0 and

0 σ (ρ + ε + σ w(s))−1/2 ≤ y. 2 y

That is,

(7.281)

(7.282) σ y, 2

(7.283)

σ y ≤ (ρ + ε + σ w(y))−1/2 . 2

(7.284)

(2 − σ(ρ + ε)1/2 y) 1 ≤ . 1/2 2(ρ + ε) (ρ + ε + σ w(y))1/2

(7.285)

(ρ + ε)−1/2 − (ρ + ε + σ w(y))−1/2 ≤ and

(7.278)

(ρ + ε)−1/2 −

That is,

7.4 Applications

147

By assumption (7.258) we have √ y σ ρ < 2,

all

0 ≤ y ≤ x.

(7.286)

Then for suﬃciently small ε > 0 we still have y σ(ρ + ε)1/2 < 2.

(7.287)

That is, 2 − y σ(ρ + ε)1/2 > 0,

0 ≤ y ≤ x.

(7.288)

2(ρ + ε)1/2 , (2 − σ(ρ + ε)1/2 y)

(7.289)

all

From (7.285) and (7.288) we obtain (ρ + ε + σ w(y))1/2 ≤

all 0 ≤ y ≤ x. Solving (7.289) for w(y) we ﬁnd * ) 4 ρ+ε √ −1 , w(y) ≤ σ (2 − σ( ρ + ε)y)2 for ε > 0 suﬃciently small and for all 0 ≤ y ≤ x. Letting ε → 0 we obtain ρ 4 −1 , w(y) ≤ √ σ (2 − σ y ρ)2

(7.290)

(7.291)

for all 0 ≤ y ≤ x. That is, w(y) ≤

(4 y ρ3/2 − σ y 2 ρ2 ) , √ (2 − σ y ρ)2

(7.292)

all 0 ≤ y ≤ x. Hence by (7.271) through (7.292) we derive )

* (4 y ρ3/2 − σ y 2 ρ2 ) θ(y) ≤ ρ + Q(y) =ϕ ˜ (y), √ (2 − σ y ρ)2

(7.293)

all 0 ≤ y ≤ x. We have proved (7.260) and (7.261). Next by applying Lemma 7.3, formula (7.3) for α = 0 , and using (7.261), we obtain (7.262). Finally by again applying Lemma 7.3, formula (7.3) for αi , and using (7.261), we get (7.263). The proof of the theorem now is complete.

8 Canavati Fractional Opial-Type Inequalities for Several Functions and Applications

A wide variety of very general Lp (1 ≤ p ≤ ∞)-form Opial-type inequalities [315] is presented involving generalized Canavati fractional derivatives [17, 101] of several functions in diﬀerent orders and powers. The above are based on a generalization of Taylor’s formula for generalized Canavati fractional derivatives [17]. From the established results are derived several other particular results of special interest. Applications of some of these special inequalities are given in proving the uniqueness of solution and in giving upper bounds to solutions of initial value problems involving a very general system of several fractional diﬀerential equations. Upper bounds to various fractional derivatives of the solutions that are involved in the above systems are given too. This treatment is based on [27].

8.1 Introduction Here the author continues his study of Canavati fractional Opial inequalities now involving several diﬀerent functions and produces a wide variety of corresponding results with important applications to systems of several fractional diﬀerential equations. This chapter is a generalization of Chapter 6. We start in Section 8.2 with preliminaries, we continue in Section 8.3 with the main results, and we ﬁnish in Section 8.4 with applications. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 149 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 8,

150

8. Canavati Fractional Opial-Type Inequalities

To give the reader an idea of the kind of inequalities we are dealing with, we brieﬂy mention a simple one ⎛ ⎞ x M ⎝ |(Daγ fj )(w)| |(Daν fj )(w)|⎠ dw a

≤

j=1

(x − a) √ √ 2Γ(ν − γ) ν − γ 2ν − 2γ − 1 ν−γ

⎧ ⎨ ⎩

⎛ x

⎝

a

M

⎞ 2 ((Daν fj )(w)) ⎠ dw

j=1

⎫ ⎬ ⎭

,

all a ≤ x ≤ b, for a certain kind of continuous function fj , j = 1, . . . , M ∈ N; ν ≥ 1, γ ≥ 0, ν − γ ≥ 1, and so on. Furthermore one system of fractional diﬀerential equations we are dealing with is of the form (Daν fj )(t) = Fj (t, {(Daγ i f1 )(t)}ri=1 , {(Daγ i f2 )(t)}ri=1 , . . . , {(Daγ i fM )(t)}ri=1 ) ,

all t ∈ [a, b],

(i) fj (a)

for j = 1, 2, . . . , M ∈ N and with = aij ∈ R, i = 0, 1, . . . , n − 1, where n := [ν], ν ≥ 2, and so on. In the literature there are many diﬀerent deﬁnitions of fractional derivatives, some of them equivalent (see [197, 333]). In this chapter we use one of the most recent due to J. Canavati [101], generalized in [15] and [17] by the author. One of the advantages of Canavati fractional derivatives is that in applications to fractional initial value problems we need only n initial conditions as with the ordinary derivative case, whereas with other deﬁnitions of fractional derivatives we need n + 1 or more initial conditions (see [333]).

8.2 Preliminaries In the sequel we follow [101]. Let g ∈ C([0, 1]). Let ν be a positive number, n := [ν], and α := ν − n (0 < α < 1). Deﬁne x 1 (x − t)ν−1 g(t) dt, 0 ≤ x ≤ 1, (8.1) (Jν g)(x) := Γ(ν) 0 the Riemann–Liouville integral, where Γ is the gamma function. We deﬁne the subspace C ν ([0, 1]) of C n ([0, 1]) as follows, C ν ([0, 1]) := {g ∈ C n ([0, 1]) : J1−α Dn g ∈ C 1 ([0, 1])}, where D := d/dx. So for g ∈ C ν ([0, 1]), we deﬁne the Canavati ν-fractional derivative of g as (8.2) Dν g := DJ1−α Dn g.

8.2 Preliminaries

151

When ν ≥ 1 we have the Taylor’s formula t2 tn−1 + · · · + g (n−1) (0) 2! (n − 1)! for all t ∈ [0, 1].

g(t) =g(0) + g (0)t + g (0) + (Jν Dν g)(t),

(8.1)

When 0 < ν < 1 we ﬁnd g(t) = (Jν Dν g)(t),

for all t ∈ [0, 1].

(8.4)

Next we transfer the above notions over to arbitrary [a, b] ⊆ R (see [17]). Let x, x0 ∈ [a, b] such that x ≥ x0 , where x0 is ﬁxed. Let f ∈ C([a, b]) and deﬁne x 1 (Jνx0 f )(x) := (x − t)ν−1 f (t) dt, x0 ≤ x ≤ b, (8.5) Γ(ν) x0 the generalized Riemann–Liouville integral . We deﬁne the subspace Cxν0 ([a, b]) of C n ([a, b]): x0 Dn f ∈ C 1 ([x0 , b])}. Cxν0 ([a, b]) := {f ∈ C n ([a, b]) : J1−α

For f ∈ Cxν0 ([a, b]), we deﬁne the generalized ν-fractional derivative of f over [x0 , b] as x0 f (n) (f (n) := Dn f ). (8.6) Dxν0 f := DJ1−α Observe that x0 (J1−α f (n) )(x) =

1 Γ(1 − α)

x

(x − t)−α f (n) (t) dt

x0

exists for f ∈ Cxν0 ([a, b]). We mention the following fractional generalization of Taylor’s formula (see [17, 101]); see also Theorem 2.1. Theorem 8.1. Let f ∈ Cxν0 ([a, b]), x0 ∈ [a, b], ﬁxed. (i) If ν ≥ 1 then f (x) =f (x0 ) + f (x0 )(x − x0 ) + f (x0 )

(x − x0 )2 2

(x − x0 )n−1 (n − 1)! x0 ν + (Jν Dx0 f )(x), for all x ∈ [a, b] : x ≥ x0 . + · · · + f (n−1) (x0 )

(8.2)

(ii) If 0 < ν < 1 then f (x) = (Jνx0 Dxν0 f )(x),

for all x ∈ [a, b] : x ≥ x0 .

(8.8)

152

8. Canavati Fractional Opial-Type Inequalities

We make Remark 8.2. (1) (Dxn0 f ) = f (n) , n ∈ N. (2) Let f ∈ Cxν0 ([a, b]), ν ≥ 1, and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1; n := [ν]. Then by (8.7) f (x) = (Jνx0 Dxν0 f )(x). That is, 1 f (x) = Γ(ν)

x

x0

(x − t)ν−1 (Dxν0 f )(t) dt,

(8.9)

for all x ∈ [a, b] with x ≥ x0 . Notice that (8.9) is true, also when 0 < ν < 1. We also make Remark 8.3. Let ν ≥ 1, γ ≥ 0, such that ν − γ ≥ 1, so that γ < ν. Call n := [ν], α := ν − n ; m := [γ], ρ := γ − m. Note that ν − m ≥ 1 and n − m ≥ 1. Let f ∈ Cxν0 ([a, b]) be such that f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Hence by (8.7) f (x) = (Jνx0 Dxν0 f )(x),

for all x ∈ [a, b] : x ≥ x0 .

Therefore by Leibnitz’s formula and Γ(p + 1) = pΓ(p), p > 0, we get that x0 Dxν0 f )(x), f (m) (x) = (Jν−m

for all x ≥ x0 .

x0 f (m) )(x) exists It follows that f ∈ Cxγ0 ([a, b]) and thus (Dxγ0 f )(x) := (DJ1−ρ for all x ≥ x0 . We easily obtain x 1 x0 γ (m) (x − t)(ν−γ)−1 (Dxν0 f )(t) dt, (Dx0 f )(x) = D((J1−ρ f )(x)) = Γ(ν − γ) x0 (8.10) and thus x0 (Dxν0 f ))(x) (Dxγ0 f )(x) = (Jν−γ

and is continuous in x on [x0 , b].

8.3 Main Results Here we often use the following basic inequalities. Let a1 , . . . , an ≥ 0, n ∈ N; then ar1 + · · · + arn ≤ (a1 + · · · + an )r ,

r ≥ 1,

(8.11)

8.3 Main Results

153

and ar1 + · · · + arn ≤ n1−r (a1 + · · · + an )r ,

0 ≤ r ≤ 1.

(8.12)

Our ﬁrst result follows next. Theorem 8.4. Let ν ≥ 1, γ 1 , γ 2 ≥ 0, such that ν − γ 1 ≥ 1, ν − γ 2 ≥ 1, (i) and fj ∈ Cxν0 ([a, b]) with fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν], j = 1, . . . , M ∈ N. Here x, x0 ∈ [a, b] : x ≥ x0 . Also consider p(t) > 0, and q(t) ≥ 0 continuous functions on [x0 , b]. Let λν > 0 and λα , λβ ≥ 0 such that λν < p, where p > 1. Set w Pk (w) := (w − t)(ν−γ k −1)p/p−1 (p(t))−1/p−1 dt, k = 1, 2, (8.13) x0

x0 ≤ w ≤ b; A(w) :=

q(w) · (P1 (w))λα (p−1/p) · (P2 (w))λβ ((p−1)/p) (p(w))−λν /p ; (8.14) (Γ(ν − γ 1 ))λα · (Γ(ν − γ 2 ))λβ x (p−λν )/p A(w)p/p−λν dw . (8.15) A0 (x) := x0

Call

ϕ1 (x) := A0 (x)|λβ =0 · δ ∗1 :=

M 1−(λα +λν /p) , 2(λα +λν /p)−1

λν λα + λν

λν /p ,

(8.16)

if λα + λν ≤ p, if λα + λν ≥ p.

(8.17)

If λβ = 0, we obtain that ⎞ ⎛ x M q(w) ⎝ |(Dxγ01 fj )(w)|λα |(Dxν0 fj )(w)|λν ⎠ dw x0

j=1

⎡ ≤ δ ∗1 · ϕ1 (x) · ⎣

x

x0

⎛ p(w) ⎝

M

⎞

⎤((λα +λν )/p)

|(Dxν0 fj )(w)|p ⎠ dw⎦

,

j=1

(8.18) all x0 ≤ x ≤ b. Proof. From Theorem 2 of [26] (see here Theorem 6.5) we get x q(w) |(Dxγ01 fj )(w)|λα |(Dxν0 fj )(w)|λν x0

(8.19) + |(Dxγ01 fj+1 )(w)|λα |(Dxν0 fj+1 )(w)|λν dw * (λα +λν )/p ) x

, ≤ δ 1 ϕ1 (x) p(w) |(Dxν0 fj )(w)|p + |(Dxν0 fj+1 )(w)|p dw x0

154

8. Canavati Fractional Opial-Type Inequalities

j = 1, 2, . . . , M − 1, where ⎧ ⎨ 21− (λα +λν )/p , δ 1 := ⎩ 1,

if λα + λν ≤ p, if λα + λν ≥ p.

(8.20)

Hence by adding all the above we ﬁnd

x

x0

M −1 |(Dxγ01 fj )(w)|λα |(Dxν0 fj )(w)|λν q(w) j=1

+|(Dxγ01 fj+1 )(w)|λα |(Dxν0 fj+1 )(w)|λν ≤ δ 1 ϕ1 (x) ·

M −1)

x

x0

j=1

dw

p(w)[|(Dxν0 fj )(w)|p *

(λα +λν )/p

+|(Dxν0 fj+1 )(w)|p ] dw

(8.21)

.

Also it holds x q(w) |(Dxγ01 f1 )(w)|λα |(Dxν0 f1 )(w)|λν x0

(8.22) + |(Dxγ01 fM )(w)|λα |(Dxν0 fM )(w)|λν dw * (λα +λν )/p ) x

. ≤ δ 1 ϕ1 (x) p(w) |(Dxν0 f1 )(w)|p + |(Dxν0 fM )(w)|p dw x0

Call

% ε1 =

1, M 1−

if λα + λν ≥ p

λα +λν /p

if λα + λν ≤ p.

,

Adding (8.21) and (8.22), and using (8.11) and (8.12) we have ⎞ ⎛ x M 2 q(w) ⎝ |(Dxγ01 fj )(w)|λα |(Dxν0 fj )(w)|λν ⎠ dw x0

j=1

%M −1) ≤ δ 1 ϕ1 (x) j=1

x

x0

p(w)[|(Dxν0 fj )(w)|p *

+|(Dxν0 fj+1 )(w)|p ]dw

(λα +λν )/p

(8.23)

8.3 Main Results

+

x

x0

155

-

(λα +λν )/p

p(w)[|(Dxν0 f1 )(w)|p + |(Dxν0 fM )(w)|p ]dw

≤ δ 1 ε1 ϕ1 (x)

x

x0

) * (λα +λν )/p M . p(w) 2 |(Dxν0 fj )(w)|p dw j=1

We have proved

⎛ x

q(w) ⎝

x0

M

⎞ |(Dxγ01 fj )(w)|λα |(Dxν0 fj )(w)|λν ⎠ dw

j=1

λα +λν /p −1 ≤ δ1 2 ε1 ϕ1 (x) ⎫ ) * ⎬ M x p(w) |(Dxν0 fj )(w)|p dw · ⎭ ⎩ x0 ⎧ ⎨

(λα +λν )/p

.

(8.24)

j=1

Clearly here we have δ ∗1 = δ 1 2 λα +λν /p −1 ε1 . From (8.24) and (8.25) we derive (8.18).

(8.25)

Next we give Theorem 8.5. All here are as in Theorem 8.4. Denote % 2λβ /λν − 1, if λβ ≥ λν , δ 3 := 1, if λβ ≤ λν , % 1, if λν + λβ ≥ p, ε2 := if λν + λβ ≤ p, M 1− λν +λβ /p ,

(8.26)

(8.27)

and p−λν /p ϕ2 (x) := A0 (x) λα =0 2

λν λβ + λν

λν /p

λ /p

δ3 ν .

If λα = 0, then

x

x0

%M −1 |(Dxγ02 fj+1 )(w)|λβ |(Dxν0 fj )(w)|λν q(w) j=1

(8.28)

156

8. Canavati Fractional Opial-Type Inequalities

|(Dxγ02 fj )(w)|λβ |(Dxν0 fj+1 )(w)|λν

+

+ |(Dxγ02 fM )(w)|λβ |(Dxν0 f1 )(w)|λν + |(Dxγ02 f1 )(w)|λβ |(Dxν0 fM )(w)|λν ≤ 2 (λν +λβ )/p ε2 ϕ2 (x) ·

dw

x

p(w)

x0

·

) M

* ν p |(Dx0 fj )(w)| dw

(λν +λβ )/p

x ≥ x0 .

,

(8.29)

j=1

Proof. From Theorem 3 of [26] (see here Theorem 6.6) we have

x

x0

q(w) |(Dxγ02 fj+1 )(w)|λβ |(Dxν0 fj )(w)|λν

+ |(Dxγ02 fj )(w)|λβ |(Dxν0 fj+1 )(w)|λν dw x ≤ ϕ2 (x) p(w) |(Dxν0 fj )(w)|p + x0 (λν +λβ )/p

|(Dxν0 fj+1 )(w)|p dw ,

(8.30)

for j = 1, . . . , M − 1. Hence by adding all the above we get

x

x0

M −1 |(Dxγ02 fj+1 )(w)|λβ |(Dxν0 fj )(w)|λν q(w) j=1

+

|(Dxγ02 fj )(w)|λβ |(Dxν0 fj+1 )(w)|λν

%M −1 ≤ ϕ2 (x) j=1

dw

x

x0

p(w)[|(Dxν0 fj )(w)|p

+ |(Dxν0 fj+1 )(w)|p dw

-

(λν +λβ )/p

.

(8.31)

8.3 Main Results

157

Similarly it holds

q(w) |(Dxγ02 fM )(w)|λβ |(Dxν0 f1 )(w)|λν x0

+ |(Dxγ02 f1 )(w)|λβ |(Dxν0 fM )(w)|λν dw x ≤ ϕ2 (x) p(w) |(Dxν0 f1 )(w)|p + x0 (λν +λβ )/p

|(Dxν0 fM )(w)|p dw . x

(8.32)

Adding (8.31) and (8.32) and using (8.11), (8.12) we produce (8.29).

The general case follows.

Theorem 8.6. All here are as in Theorem 8.4. Denote

γ˜ 1 :=

% 2 λα +λβ /λν − 1,

if λα + λβ ≥ λν , if λα + λβ ≤ λν ,

1,

(8.33)

and % γ˜ 2 :=

1,

if λα + λβ + λν ≥ p,

1− λα +λβ +λν /p

2

if λα + λβ + λν ≤ p.

,

(8.34)

Set

λν /p λν A0 (x) · (λα + λβ )(λα + λβ + λν ) ) * p−λν /p λν /p λν /p (˜ γ 1 λβ ) · λα γ˜ 2 + 2 ,

ϕ3 (x) :=

(8.35)

and % ε3 :=

1, M 1−

if λα + λβ + λν ≥ p,

(λα +λβ +λν )/p

,

if λα + λβ + λν ≤ p.

(8.36)

158

8. Canavati Fractional Opial-Type Inequalities

Then

x

x0

)M −1 |(Dxγ01 fj )(w)|λα |(Dxγ02 fj+1 )(w)|λβ |(Dxν0 fj )(w)|λν q(w) j=1

+|(Dxγ02 fj )(w)|λβ |(Dxγ01 fj+1 )(w)|λα |(Dxν0 fj+1 )(w)|λν + |(Dxγ01 f1 )(w)|λα |(Dxγ02 fM )(w)|λβ |(Dxν0 f1 )(w)|λν +|(Dxγ02 f1 )(w)|λβ |(Dxγ01 fM )(w)|λα |(Dxν0 fM )(w)|λν ≤ 2 λα +λβ +λν /p ε3 ϕ3 (x) ·

%

*

dw (8.37)

x

p(w) x0

) M

- (λα +λβ +λν )/p

* |(Dxν0 fj )(w)|p dw

,

j=1

all x0 ≤ x ≤ b. Proof. From Theorem 4 of [26] (see here Theorem 6.7) and adding all together we have M −1 x j=1

x0

γ γ q(w) |(Dx01 fj )(w)|λα |(Dx02 fj+1 )(w)|λβ |(Dxν 0 fj )(w)|λν

γ γ +|(Dx02 fj )(w)|λβ |(Dx01 fj+1 )(w)|λα |(Dxν 0 fj+1 )(w)|λν dw (8.38) M −1 x p(w) |(Dxν 0 fj )(w)|p ≤ ϕ3 (x) j=1

x0

+|(Dxν 0 fj+1 )(w)|p dw

(λα +λβ +λν )/p

,

all x0 ≤ x ≤ b. Also it holds

x x0

γ γ q(w) |(Dx01 f1 )(w)|λα |(Dx02 fM )(w)|λβ |(Dxν 0 f1 )(w)|λν

γ γ (8.39) + |(Dx02 f1 )(w)|λβ |(Dx01 fM )(w)|λα |(Dxν 0 fM )(w)|λν dw x (λα +λβ +λν /p) ≤ ϕ3 (x) , p(w)(|(Dxν 0 f1 )(w)|p + |(Dxν 0 fM )(w)|p ) dw x0

all x0 ≤ x ≤ b. Adding (8.38) and (8.39), along with (8.11) and (8.12) we produce (8.37).

8.3 Main Results

159

We continue with

Theorem 8.7. Let ν ≥ 2 and γ 1 ≥ 0 such that ν − γ 1 ≥ 2. Let fj ∈ (i) with fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν], j = 1, . . . , M ∈ N. Here x, x0 ∈ [a, b] : x ≥ x0 . Also consider p(t) > 0, and q(t) ≥ 0 continuous functions on [x0 , b]. Let λα ≥ 0, 0 < λα+1 < 1, and p > 1. Denote Cxν0 ([a, b])

% λα /λα+1 − 1, 2 θ3 := 1,

if λα ≥ λα+1 , if λα ≤ λα+1

(8.40)

x (1−λα+1 ) 2 (q(w)) 1/1−λα+1 dw

L(x) :=

x0

θ3 λα+1 λα + λα+1

λα+1 ,

(8.41)

and

x

P1 (x) := x0

(x − t)(ν−γ 1 −1)p/p−1 (p(t))−1/(p−1) dt,

⎞(λα +λα+1 ) (p−1)/p (x) P 1 ⎠ , T (x) :=L(x) · ⎝ Γ(ν − γ 1 )

(8.42)

⎛

(8.43)

and ω 1 := 2((p−1)/p)(λα +λα+1 ) ,

(8.44)

Φ(x) := T (x)ω 1 .

(8.45)

Also put % ε4 :=

1, M 1−

if λα + λα+1 ≥ p,

(λα +λα+1 )/p

,

if λα + λα+1 ≤ p

.

(8.46)

160

8. Canavati Fractional Opial-Type Inequalities

Then

x

x0

⎧⎧ −1 ⎨⎨M |(Dxγ01 fj )(w)|λα |(Dxγ01 +1 fj+1 )(w)|λα+1 q(w) ⎩⎩ j=1

+|(Dxγ01 fj+1 )(w)|λα |(Dxγ01 +1 fj )(w)|λα+1 + |(Dxγ01 f1 )(w)|λα |(Dxγ01 +1 fM )(w)|λα+1

γ1 λα γ 1 +1 λα+1 dw +|(Dx0 fM )(w)| |(Dx0 f1 )(w)|

) x λα +λα+1 /p ε4 Φ(x) ≤2 p(w) x 0 * (λα +λα+1 )/p M , |(Dxν0 fj )(w)|p dw

(8.47)

j=1

all x0 ≤ x ≤ b. Proof. From Theorem 5 [26] (see here Theorem 6.8) we get

x

q(w) x0

M −1

|(Dxγ01 fj )(w)|λα |(Dxγ01 +1 fj+1 )(w)|λα+1

j=1

+|(Dxγ01 fj+1 )(w)|λα |(Dxγ01 +1 fj )(w)|λα+1 dw M −1) x ≤ Φ(x) p(w) |(Dxν0 fj )(w)|p j=1

x0

+|(Dxν0 fj+1 )(w)|p dw all x0 ≤ x ≤ b. Also it holds

(8.48)

* (λα +λα+1 )/p

q(w) |(Dxγ01 f1 )(w)|λα |(Dxγ01 +1 fM )(w)|λα+1 x0

+|(Dxγ01 fM )(w)|λα |(Dxγ01 +1 f1 )(w)|λα+1 dw ) x ≤ Φ(x) p(w)(|(Dxν0 f1 )(w)|p x0 * (λα +λα+1 )/p

+|(Dxν0 fM )(w)|p dw , x

(8.49)

all x0 ≤ x ≤ b. Adding (8.48) and (8.49), along with (8.11) and (8.12) we derive (8.47).

8.3 Main Results

161

Next comes the following theorem. Theorem 8.8. All here are as in Theorem 8.4. Consider the special case λβ = λα + λν . Denote λν /p λν T˜(x) := A0 (x) 2 (p−2λα −3λν )/p , λα + λν % 1, if 2(λα + λν ) ≥ p, ε5 := . M 1− 2(λα +λν )/p , if 2(λα + λν ) ≤ p

(8.50) (8.51)

Then

x

x0

%M −1 |(Dxγ01 fj )(w)|λα |(Dxγ02 fj+1 )(w)|λα +λν |(Dxν0 fj )(w)|λν q(w) j=1

+|(Dxγ02 fj )(w)|λα +λν |(Dxγ01 fj+1 )(w)|λα |(Dxν0 fj+1 )(w)|λν + |(Dxγ01 f1 )(w)|λα |(Dxγ02 fM )(w)|λα +λν |(Dxν0 f1 )(w)|λν

γ2 λα +λν γ1 λα ν λν dw +|(Dx0 f1 )(w)| |(Dx0 fM )(w)| |(Dx0 fM )(w)|

) x ˜ ε5 T (x) ≤2 p(w) x0 * 2 (λα +λν )/p M , |(Dxν0 fj )(w)|p dw

2 λα +λν /p

(8.52)

j=1

all x0 ≤ x ≤ b. Proof. Based on Theorem 6 [26] (see here Theorem 6.9). The rest is as in the proof of Theorem 8.7. Next we give special cases of the above theorems. Corollary 8.9. (to Theorem 8.4; λβ = 0, p(t) = q(t) = 1). It holds x M x0

|(Dxγ01 fj )(w)|λα |(Dxν0 fj )(w)|λν

dw

j=1

' ) M * ( x ∗ ν p |(Dx0 fj )(w)| dw ≤ δ 1 ϕ1 (x) x0

all x0 ≤ x ≤ b.

j=1

(λα +λν )/p

(8.53)

162

8. Canavati Fractional Opial-Type Inequalities

In (8.53) A0 (x)λ =0 of ϕ1 (x) is given in [26], Corollary 1, by Equation β (55) there; see here (6.55). Corollary 8.10. (to Theorem 8.4; λβ = 0, p(t) = q(t) = 1 , λα = λν = 1, p = 2). In detail : let ν ≥ 1, γ 1 ≥ 0, such that ν − γ 1 ≥ 1, fj ∈ Cxν0 ([a, b]) (i) with fj (x0 ) = 0, i = 1, . . . , n − 1, n := [ν], j = 1, . . . , M ∈ N. Here x, x0 ∈ [a, b] : x ≥ x0 . Then x M x0

|(Dxγ01 fj )(w)| |(Dxν0 fj )(w)|

dw

(8.54)

j=1

(x − x0 )ν−γ 1 √ ≤ √ 2Γ(ν − γ 1 ) ν − γ 1 2ν − 2γ 1 − 1 ⎧ ⎫ ) * ⎬ M ⎨ x · ((Dxν0 fj )(w))2 dw , ⎩ x0 ⎭ j=1

all x0 ≤ x ≤ b. Proof. Based on our Corollary 8.9 and Corollary 1 of [26], especially Equation (55) there; see here (6.55). Corollary 8.11. (to Theorem 8.5, λα = 0, p(t) = q(t) = 1). It holds x %M −1 |(Dxγ02 fj+1 )(w)|λβ |(Dxν0 fj )(w)|λν x0

j=1

+|(Dxγ02 fj )(w)|λβ |(Dxν0 fj+1 )(w)|λν

+ |(Dxγ02 fM )(w)|λβ |(Dxν0 f1 )(w)|λν

γ2 λβ ν λν dw +|(Dx0 f1 )(w)| |(Dx0 fM )(w)| ≤ 2 λν +λβ /p ε2 ϕ2 (x)· ⎧ ⎫ * ⎬ (λν +λβ )/p M ⎨ x ) |(Dxν0 fj )(w)|p dw , ⎩ x0 ⎭

(8.55)

j=1

all x0 ≤ x ≤b. In (8.55), A0 (x)λ =0 of ϕ2 (x) is given in [26], Corollary 3 by Equation α (59); see here (6.59).

8.3 Main Results

163

Corollary 8.12. (to Theorem 8.5, λα = 0, p(t) = q(t) = 1, λβ = λν = 1, p = 2). In detail : let ν ≥ 1, γ 2 ≥ 0 such that ν − γ 2 ≥ 1 and fj ∈ Cxν0 ([a, b]) (i) with fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν], j = 1, . . . , M ∈ N. Here x, x0 ∈ [a, b] : x ≥ x0 . Then x %M −1 |(Dxγ02 fj+1 )(w)| |(Dxν0 fj )(w)| x0

j=1

+|(Dxγ02 fj )(w)| |(Dxν0 fj+1 )(w)

+ |(Dxγ02 fM )(w)| |(Dxν0 f1 )(w)|

+|(Dxγ02 f1 )(w)| |(Dxν0 fM )(w)|

dw

√ 2(x − x0 )(ν−γ 2 ) √ ≤ √ Γ(ν − γ 2 ) ν − γ 2 2ν − 2γ 2 − 1 ⎧ ⎡ ⎤ ⎫ M ⎨ x ⎬ ⎣ ((Dxν fj )(w))2 ⎦ dw , 0 ⎩ x0 ⎭

(8.56)

j=1

all x0 ≤ x ≤ b. Proof. From our Corollary 8.11 and Corollary 3 of [26], especially Equation (59); see here (6.59). Corollary 8.13. (to Theorem 8.6, λα = λβ = λν = 1, p = 3, p(t) = q(t) = 1). It holds x )M −1 |(Dxγ01 fj )(w)| |(Dxγ02 fj+1 )(w)||(Dxν0 fj )(w)| x0

j=1

+|(Dxγ02 fj )(w)| |(Dxγ01 fj+1 )(w)| |(Dxν0 fj+1 )(w)| + |(Dxγ01 f1 )(w)| |(Dxγ02 fM )(w)| |(Dxν0 f1 )(w)| *

γ2 γ1 ν +|(Dx0 f1 )(w)| |(Dx0 fM )(w)| |(Dx0 fM )(w)| dw ⎡ ⎤ * x ) M |(Dxν0 fj )(w)|3 dw ⎦ , ≤ 2ϕ∗3 (x) · ⎣ x0

all x0 ≤ x ≤ b. Here ϕ∗3 (x)

:=

(8.57)

j=1

√ 3

1 2+ √ 3 6

A0 (x),

(8.58)

164

8. Canavati Fractional Opial-Type Inequalities

where in this special case A0 (x) :=

4(x − x0 )(2ν−γ 1 −γ 2 ) . Γ(ν − γ 1 )Γ(ν − γ 2 )[3(3ν − 3γ 1 − 1)(3ν − 3γ 2 − 1)(2ν − γ 1 − γ 2 )]2/3 (8.59)

Proof. From Theorem 8.6 and Equation (62) of [26], which is here Equation (8.59). Corollary 8.14. (to Theorem 8.7, λα = 1, λα+1 = 1/2, p = 3/2, p(t) = q(t) = 1). In detail : Let ν ≥ 2 and γ 1 ≥ 0 such that ν − γ 1 ≥ 2. Let fj ∈ Cxν0 ([a, b]) with (i) fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν], j = 1, . . . , M ∈ N. Here x, x0 ∈ [a, b] : x ≥ x0 . Call 3ν−3γ 1 −1/2 2 (x − x ) 0 , (8.60) Φ∗ (x) := √ 3ν − 3γ 1 − 2 (Γ(ν − γ 1 ))3/2 all x0 ≤ x ≤ b. Then

x %M . −1 γ +1 |(Dxγ01 fj )(w)| |(Dx01 fj+1 )(w)| x0

j=1

.

+|(Dxγ01 fj+1 )(w)|

γ +1 |(Dx01 fj )(w)|

. γ +1 + |(Dxγ01 f1 )(w)| |(Dx01 fM )(w)| .

γ 1 +1 γ1 +|(Dx0 fM )(w)| |(Dx0 f1 )(w)| dw ⎡ ∗ ⎣ ≤ 2Φ (x) ·

⎛ x

x0

⎝

M

⎞

⎤

|(Dxν0 fj )(w)|3/2 ⎠ dw⎦ ,

(8.61)

j=1

all x0 ≤ x ≤ b. Proof. Based on Theorem 8.7 here, and Equation (64) of [26] to establish coeﬃcient Φ∗ (x) in (8.61). Corollary 8.15. (to Theorem 8.8, p = 2(λα + λν ) > 1, p(t) = q(t) = 1). It holds x %M −1 |(Dxγ01 fj )(w)|λα |(Dxγ02 fj+1 )(w)|λα +λν |(Dxν0 fj )(w)|λν x0

j=1

8.3 Main Results

+|(Dxγ02 fj )(w)|λα +λν |(Dxγ01 fj+1 )(w)|λα |(Dxν0 fj+1 )(w)|λν

165

+ |(Dxγ01 f1 )(w)|λα |(Dxγ02 fM )(w)|λα +λν |(Dxν0 f1 )(w)|λν

dw +|(Dxγ02 f1 )(w)|λα +λν |(Dxγ01 fM )(w)|λα |(Dxν0 fM )(w)|λν ⎡ ≤ 2T˜(x) ⎣

⎛ x

⎝

x0

M

⎞

⎤

|(Dxν0 fj )(w)|2(λα +λν ) ⎠ dw⎦ ,

(8.62)

j=1

all x0 ≤ x ≤ b. Here T˜(x) in (8.62) is given precisely by Equations (66)–(70) of [26]; see here (6.66)–(6.70). Corollary 8.16. (to Theorem 8.8, p = 4, λα = λν = 1, p(t) = q(t) = 1). It holds x %M −1 |(Dxγ01 fj )(w)|((Dxγ02 fj+1 )(w))2 |(Dxν0 fj )(w)| x0

j=1

+((Dxγ02 fj )(w))2 |(Dxγ01 fj+1 )(w)| |(Dxν0 fj+1 )(w)|

+ |(Dxγ01 f1 )(w)|((Dxγ02 fM )(w))2 |(Dxν0 f1 )(w)|

γ2 2 γ1 ν +((Dx0 f1 )(w)) |(Dx0 fM )(w)| |(Dx0 fM )(w)| dw ⎡ ≤ 2T˜(x) ⎣

x

x0

⎛ ⎞ ⎤ M ⎝ ((Dxν fj )(w))4 ⎠ dw⎦ , 0

(8.63)

j=1

all x0 ≤ x ≤ b. Here in (8.63) we have that T˜(x) = T ∗ (x) of Corollary 8 in [26] for it (see there Equations (72)–(76), and here (6.72)–(6.76)) . Next we present the supremum case. Theorem 8.17. Let ν ≥ 1, γ 1 , γ 2 ≥ 0, such that ν − γ 1 ≥ 1, ν − γ 2 ≥ 1, (i) and fj ∈ Cxν0 ([a, b]) with fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν], j = 1, . . . , M ∈ N. Here x, x0 ∈ [a, b] : x ≥ x0 . Consider p(x) ≥ 0 a continuous function on [x0 , b]. Let λα , λβ , λν ≥ 0. Set ρ(x) :=

(x − x0 )(νλα −γ 1 λα +νλβ −γ 2 λβ +1) p(x)∞ . (νλα − γ 1 λα + νλβ − γ 2 λβ + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λβ (8.64)

166

8. Canavati Fractional Opial-Type Inequalities

Then

x

x0

%M −1 |(Dxγ01 fj )(w)|λα |(Dxγ02 fj+1 )(w)|λβ |(Dxν0 fj )(w)|λν p(w) j=1

+|(Dxγ02 fj )(w)|λβ |(Dxγ01 fj+1 (w)|λα |(Dxν0 fj+1 )(w)|λν

+ |(Dxγ01 f1 )(w)|λα |(Dxγ02 fM )(w)|λβ |(Dxν0 f1 )(w)|λν

γ2 λβ γ1 λα ν λν dw +|(Dx0 f1 )(w)| |(Dx0 fM )(w)| |(Dx0 fM )(w)| M 2λβ ν 2(λα +λν ) ν ≤ ρ(x) {(Dx0 fj )∞ + (Dx0 fj )∞ } ,

(8.65)

j=1

all x0 ≤ x ≤ b. Proof. Based on Theorem 7 of [26]; see here Theorem 6.18.

Similarly we give Theorem 8.18. (as in Theorem 8.17, λβ = 0). It holds

x

x0

M γ1 λα ν λν p(w) |(Dx0 fj )(w)| |(Dx0 fj )(w)| dw j=1

≤

(x − x0 )(νλα −γ 1 λα +1) p(x)∞ (νλα − γ 1 λα + 1)(Γ(ν − γ 1 + 1))λα ⎞ ⎛ M Dxν0 fj λ∞α +λν ⎠ , ·⎝

(8.66)

j=1

all x0 ≤ x ≤ b. Proof. Based on Theorem 8 of [26]; see here Theorem 6.19.

It follows Theorem 8.19. (as in Theorem 8.17, λβ = λα + λν ). It holds

x

x0

%M −1 |(Dxγ01 fj )(w)|λα |(Dxγ02 fj+1 )(w)|λα +λν |(Dxν0 fj )(w)|λν p(w) j=1

8.3 Main Results

+|(Dxγ02 fj )(w)|λα +λν |(Dxγ01 fj+1 )(w)|λα |(Dxν0 fj+1 )(w)|λν

167

+ |(Dxγ01 f1 )(w)|λα |(Dxγ02 fM )(w)|λα +λν |(Dxν0 f1 )(w)|λν

γ2 λα +λν γ1 λα ν λν dw +|(Dx0 f1 )(w)| |(Dx0 fM )(w)| |(Dx0 fM )(w)| ≤

2(x − x0 )(2νλα −γ 1 λα +νλν −γ 2 λα −γ 2 λν +1) (2νλα − γ 1 λα + νλν − γ 2 λα − γ 2 λν + 1)(Γ(ν − γ 1 + 1))λα ⎞ ⎛ M p(x)∞ α +λν ) ⎠ , (8.67) Dxν0 fj 2(λ ·⎝ ∞ (Γ(ν − γ 2 + 1))(λα +λν ) j=1

all x0 ≤ x ≤ b. Proof. By Theorem 9 of [26]; see here Theorem 6.20.

We continue with Theorem 8.20. (as in Theorem 8.17, λν = 0, λα = λβ ). It holds %M −1 x |(Dxγ01 fj )(w)|λα |(Dxγ02 fj+1 )(w)|λα p(w) x0

j=1

+|(Dxγ02 fj )(w)|λα |(Dxγ01 fj+1 )(w)|λα

+ |(Dxγ01 f1 )(w)|λα |(Dxγ02 fM )(w)|λα

γ2 λα γ1 λα dw +|(Dx0 f1 )(w)| |(Dx0 fM )(w)| ⎡ ≤ 2ρ∗ (x) ⎣

M

⎤ α⎦ , Dxν0 fj 2λ ∞

(8.68)

j=1

all x0 ≤ x ≤ b. Here we have ∗

ρ (x) :=

(x − x0 )(2νλα −γ 1 λα −γ 2 λα +1) p(x)∞ . (2νλα − γ 1 λα − γ 2 λα + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λα (8.69)

Proof. Based on Theorem 10 of [26]; see here Theorem 6.21. Next we give

168

8. Canavati Fractional Opial-Type Inequalities

Theorem 8.21. (as in Theorem 8.17, λα = 0, λβ = λν ). It holds %M −1 x |(Dxγ02 fj+1 )(w)|λβ |(Dxν0 fj )(w)|λβ p(w) x0

j=1

+|(Dxγ02 fj )(w)|λβ |(Dxν0 fj+1 )(w)|λβ

+ |(Dxγ02 fM )(w)|λβ |(Dxν0 f1 )(w)|λβ

γ2 λβ ν λβ dw +|(Dx0 f1 )(w)| |(Dx0 fM )(w)| ≤2·

(x − x0 )(νλβ −γ 2 λβ +1) p(x)∞ (νλβ − γ 2 λβ + 1)(Γ(ν − γ 2 + 1))λβ ⎡ ⎤ M 2λ ⎣ Dxν0 fj ∞ β ⎦ ,

(8.70)

j=1

all x0 ≤ x ≤ b. Proof. Based on Theorem 11 of [26]; see here Theorem 6.22.

Some special cases follow. Corollary 8.22. (to Theorem 8.20, all are as in Theorem 8.17, λν = 0, λα = λβ , γ 2 = γ 1 + 1). It holds %M −1 x |(Dxγ01 fj )(w)|λα |(Dxγ01 +1 fj+1 )(w)|λα p(w) x0

j=1

+|(Dxγ01 +1 fj )(w)|λα |(Dxγ01 fj+1 )(w)|λα

+ |(Dxγ01 f1 )(w)|λα |(Dxγ01 +1 fM )(w)|λα

γ 1 +1 λα γ1 λα dw +|(Dx0 f1 )(w)| |(Dx0 fM )(w)| ≤2·

(x − x0 )(2νλα −2γ 1 λα −λα +1) p(x)∞ (2νλα − 2γ 1 λα − λα + 1)(ν − γ 1 )λα (Γ(ν − γ 1 ))2λα ⎡ ⎤ M α⎦ ·⎣ , Dxν0 fj 2λ ∞ j=1

all x0 ≤ x ≤ b.

(8.71)

8.4 Applications

Proof. Based on Corollary 9 of [26]; see here Corollary 6.23.

169

Corollary 8.23. (to Corollary 8.22). In detail : let ν ≥ 2, γ 1 ≥ 0, such (i) that ν − γ 1 ≥ 2 and fj ∈ Cxν0 ([a, b]) with fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν], j = 1, . . . , M ∈ N. Here x, x0 ∈ [a, b] : x ≥ x0 . Then x %M −1 |(Dxγ01 fj )(w)| |(Dxγ01 +1 fj+1 )(w)| x0

j=1

+|(Dxγ01 +1 fj )(w)| |(Dxγ01 fj+1 )(w)|

+ |(Dxγ01 f1 )(w)| |(Dxγ01 +1 fM )(w)|

+|(Dxγ01 +1 f1 )(w)| |(Dxγ01 fM )(w)| dw ≤

(x − x0 )2(ν−γ 1 ) (ν − γ 1 )2 (Γ(ν − γ 1 ))2

⎛ ⎝

M

⎞ Dxν0 fj 2∞ ⎠ ,

(8.72)

j=1

all x0 ≤ x ≤ b. Proof. Based on Corollary 10 of [26]; see here Corollary 6.24. Corollary 8.24. (to Corollary 8.23). It holds ⎛ ⎞ x M ⎝ |(Dxγ01 fj )(w)| |(Dxγ01 +1 fj )(w)|⎠ dw x0

j=1

≤

(x − x0 )2(ν−γ 1 ) 2(ν − γ 1 )2 (Γ(ν − γ 1 ))2

⎛ ⎝

M

⎞ Dxν0 fj 2∞ ⎠ ,

(8.73)

j=1

all x0 ≤ x ≤ b. Proof. Based on Equation (97) of [26]; see here (6.97).

8.4 Applications We present our ﬁrst application. Theorem 8.25. Let ν ≥ 1, γ i ≥ 0, ν − γ i ≥ 1, i = 1, . . . , r ∈ N, n := [ν], (i) fj ∈ Caν ([a, b]), j = 1, 2, 3, . . . , M , fj (a) = aij ∈ R, i = 0, 1, . . . , n − 1. Furthermore we have for j = 1, 2, . . . , M that # γ $r # γ $r # γ $r (Daν fj )(t) = Fj t, (Da i f1 )(t) i=1 , (Da i f2 )(t) i=1 , . . . , (Da i fM )(t) i=1 , (8.74)

170

8. Canavati Fractional Opial-Type Inequalities

all t ∈ [a, b]. Here Fj are continuous functions on [a, b] × (Rr )M and satisfy the Lipschitz condition Fj (t, x11 , x12 , . . . , x1r , x21 , . . . , x2r , x31 , . . . , x3r , . . . , xM 1 , . . . , xM r ) −Fj (t, x11 , x12 , . . . , x1r , x21 , . . . , x2r , x31 , . . . , x3r , xM 1 , . . . , xM r ) M r ≤ q,i,j (t)|xi − xi | , (8.75) i=1

=1

j = 1, 2, . . . , M , where all q,i,j ≥ 0, 1 ≤ i ≤ r, are continuous functions over [a, b]. Call # $ W := max q,i,j ∞ , , j = 1, 2, . . . , M, i = 1, . . . , r . (8.76) Assume here that r ν−γ i 1 (b − a) M − 1 √ φ∗ (b) := W + √ < 1. √ 2 Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1 2 i=1 (8.77) Then if system (8.74) has two M -tuples of solutions (f1 , f2 , . . . , fM ) and ∗ ) we prove that (f1∗ , f2∗ , . . . , fM fj = fj∗ ,

j = 1, 2, . . . , M ;

that is, we have uniqueness of solution. Proof. Assume that there are two M -tuples of solutions (f1 , f2 , . . . , fM ) ∗ ) satisfying the system (8.74). Set gj := fj − fj∗ , j = and (f1∗ , . . . , fM (i)

(i)

∗(i)

(i)

and gj (a) = 0, i = 0, 1, . . . , n − 1; 1, 2, . . . , M . Then gj = fj − fj j = 1, 2, . . . , M . It holds # $r # $r (Daν gj )(t) = Fj t, (Daγ i f1 )(t) i=1 , . . . , (Daγ i fM )(t) i=1 # $r # $r ∗ − Fj t, (Daγ i f1∗ )(t) i=1 , . . . , (Daγ i fM )(t) i=1 . Therefore by (8.75) we get |(Daν gj )(t)|

r q1,i,j (t)|(Daγ i g1 )(t)| + q2,i,j (t)|(Daγ i g2 )(t)| ≤ i=1

+ · · · + qM,i,j (t)|(Daγ i gM )(t)| . And thus |(Daν gj )(t)| ≤

r q1,i,j ∞ |(Daγ i g1 )(t)| + q2,i,j ∞ |(Daγ i g2 )(t)| i=1

+ · · · + qM,i,j ∞ |(Daγ i gM )(t)| .

8.4 Applications

171

Furthermore we have r

|(Daν gj )(t)| ≤W

|(Daγ i g1 )(t)| + |(Daγ i g2 )(t)|

i=1

+ ··· +

|(Daγ i gM )(t)|

.

(8.78)

Clearly (8.78) implies M M r ν 2 (Da gj )(t) ≤W |(Daγ i g1 )(t)| |(Daν gj )(t)| j=1

i=1 j=1

+ |(Daγ i g2 )(t)| |(Daν gj )(t)|

+ · · · + |(Daγ i gM )(t)| |(Daν gj )(t)| .

(8.79)

Integrating (8.79) we observe ⎛

b

I :=

⎞ M ν 2 ⎝ (Da gj )(t) ⎠ dt

a

≤W

j=1

% r M ) i=1 j=1

b

|(Daγ i g1 )(t)| |(Daν gj )(t)| dt

a

b

|(Daγ i g2 )(t)| |(Daν gj )(t)| dt

+ a

b

+ ··· +

*|(Daγ i gM )(t)| |(Daν gj )(t)| dt .

(8.80)

a

That is,

I

≤ W

% r ) M b i=1

+

a

τ ,m∈{1,...,M } τ =m

+

dt

|(Daγ i gλ )(t)| |(Daν gλ )(t)|

λ=1

b

(|(Daγ i gm )(t)| |(Daν gτ )(t)| a

|(Daγ i gτ )(t)| |(Daν gm )(t)|

* dt .

(8.81)

172

8. Canavati Fractional Opial-Type Inequalities

Using Corollary 2 from here and Corollary 4 of [26] (see here Corollary 6.11 and Corollary 6.13), we obtain % r ) (b − a)ν−γ i √ I ≤W I √ 2Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1 i=1 *(b − a)ν−γ i (M − 1)I . (8.82) + √ √ √ 2Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1 That is, we get that

I ≤ φ∗ (b) · I.

(8.83)

If I = 0 then φ∗ (b) ≥ 1, a contradiction by the assumption that φ (b) < 1; see (8.77). Therefore I = 0, implying that ∗

M

2 (Daν gλ )(t) = 0,

a.e. in [a, b].

λ=1

That is, (Daν gλ )2 (t) = 0,

a.e. in [a, b].

That is, (Daν gλ )(t) = 0,

λ = 1, 2, . . . , M, a.e. in [a, b].

But for λ = 1, 2, . . . , M we get that (i)

gλ (a) = 0,

0 ≤ i ≤ n − 1.

Hence from fractional Taylor’s Theorem 8.1 we get that gλ (t) = 0 on [a, b]. That is, fλ = fλ∗ , λ = 1, 2, . . . , M, proving the uniqueness argument of this theorem. Another related application follows. Theorem 8.26. Let ν ≥ 1, γ i ≥ 0, ν − γ i ≥ 1, i = 1, . . . , r ∈ N, (i) n := [ν], fj ∈ Caν ([a, b]), j = 1, 2, . . . , M ; fj (a) = 0, i = 0, 1, . . . , n − 1, and (Daν fj )(a) = Aj ∈ R. Furthermore for a ≤ t ≤ b we have holding the system of fractional diﬀerential equations (Daν fj ) (x) = Fj t, ({(Daγ i fλ )(t)}ri=1 , (Daν fλ )(t)); λ = 1, 2, . . . , M , j = 1, 2, . . . , M. (8.84) Here Fj are continuous functions on [a, b] × (Rr+1 )M such that |Fj (t, x11 , x12 , . . . , x1r , x1,r+1 ; x21 , x22 , . . . , x2r , x2,r+1 ; x31 , x32 , . . . , x3,r+1 ;

8.4 Applications

xM 1 , xM 2 , . . . , xM,r+1 )| ≤

M r i=1

173

q,i,j (t)|xi | ,

(8.85)

=1

where q,i,j (t) ≥ 0,

1 ≤ i ≤ r; , j = 1, 2, . . . , M,

are continuous functions on [a, b]. Call # $ W := max q,i,j ∞ ; , j = 1, 2, . . . , M, i = 1, . . . , r .

(8.86)

Also we set (a ≤ x ≤ b) θ(x) :=

M

2 (Daν fλ )(x) ,

(8.87)

λ=1

ρ :=

M

A2λ ,

(8.88)

λ=1

r i=1

√ Q(x) := W 1 + 2(M − 1) (x − a)ν−γ i √ √ Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1

(8.89)

and χ(x) :=

√

) x t *1/2 x Q(s)ds − a Q(s)ds a · dt e ρ · 1 + Q(x) · e . (8.90)

a

Then

θ(x) ≤ χ(x),

a ≤ x ≤ b.

(8.91)

Consequently we get |(Daν fj )(x)| ≤ χ(x), x 1 |fj (x)| ≤ (x − t)ν−1 χ(t) dt, Γ(ν) a all a ≤ x ≤ b, j = 1, 2, . . . , M . Also it holds x 1 γi |(Da fj )(x)| ≤ (x − t)ν−γ i −1 χ(t) dt, Γ(ν − γ i ) a

(8.92) (8.93)

(8.94)

all a ≤ x ≤ b, j = 1, 2, . . . , M , i = 1, . . . , r. Proof. We easily get that (a ≤ x ≤ b), x x ν ν (Da fj )(t)(Da fj ) (t)dt = (Daν fj )(t) · Fj t, ({(Daγ i fλ )(t)}ri=1 , a a (8.95) (Daν fλ )(t)); λ = 1, 2, . . . , M dt.

174

8. Canavati Fractional Opial-Type Inequalities

Hence we obtain x ((Daν fj )(t))2 2

x

≤

|(Daν fj )(t)| |Fj · · · |dt a

a

x

≤

) * r M |(Daν fj )(t)| q,i,j (t)|(Daγ i f )(t)| dt

a r M

≤

i=1

≤ W

i=1

q,i,j ∞

=1

x

|(Daν fj )(t)| |(Daγ i f )(t)| dt a

=1

r M

x

|(Daν fj )(t)| |(Daγ i f )(t)| dt

.

a

i=1 =1

Thus we have for j = 1, . . . , M that

2 (Daν fj )(x)

≤ A2j + 2W ·

r M i=1 =1

x

|(Daν fj )(t)| |(Daγ i f )(t)| dt

.

(8.96)

a

Consequently it holds % r M M θ(x) ≤ ρ + 2W % = ρ + 2W +

i=1

j=1 =1

i=1

r x M

τ ,m∈{1,...,M }

a

-

x

|(Daν fj )(t)| |(Daγ i f )(t)| dt

a

|(Daγ i fλ )(t)| |(Daν fλ )(t)| dt

λ=1 x

(|(Daγ i fm )(t)| |(Daν fτ )(t)| a

τ =m

+ |(Daγ i fτ )(t)| |(Daν fm )(t)|) dt .

(8.97)

Using Corollary 2 from here and Corollary 4 of [26] (see here Corollary 6.11 and Corollary 6.13) we obtain % r (x − a)ν−γ i √ θ(x) ≤ ρ + 2W √ 2Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1 i=1 x (x − a)ν−γ i (8.98) θ(t) dt + √ √ √ 2Γ(ν − γ i ) ν − γ i 2ν − 2γ i − 1 a x (M − 1) θ(t) dt . a

8.4 Applications

Hence we have

175

x

θ(t) dt, all a ≤ x ≤ b.

θ(x) ≤ ρ + Q(x)

(8.99)

a

Here ρ ≥ 0, Q(x) ≥ 0, Q(a) = 0, θ(x) ≥ 0, all a ≤ x ≤ b. As in the proof of Theorem 13 of [26] (see also [17]), we get (8.91) and (8.92). Using (8.9) we get (8.93), and using (8.10) we establish (8.94). Finally we give a specialized application. Theorem 8.27. Let a = b, ν ≥ 2, γ i ≥ 0, ν − γ i ≥ 1, i = 1, . . . , r ∈ N, (i) n := [ν], fj ∈ Caν ([a, b]), j = 1, 2, . . . , M ; fj (a) = 0, i = 0, 1, . . . , n − 1, and (8.100) (Daν fj )(a) = Aj ∈ R. Furthermore for a ≤ t ≤ b we have holding the system of fractional diﬀerential equations (Daν fj ) (t) = Fj t, ({(Daγ i f )(t)}ri=1 , (Daν f )(t));

= 1, . . . , M , for j = 1, 2, . . . , M. (8.101) For ﬁxed i∗ ∈ {1, . . . , r} we assume that γ i∗ +1 = γ i∗ + 1, and ν − γ i∗ ≥ 2, where γ i∗ , γ i∗ +1 ∈ {γ 1 , . . . , γ r }. Call k := γ i∗ , γ := γ i∗ + 1; that is, γ = k + 1. Here Fj are continuous functions on [a, b] × (Rr+1 )M such that |Fj (t, x11 , x12 , . . . , x1r , x1,r+1 ; x21 , x22 , . . . , x2r , x2,r+1 ; x31 , x32 , . . . , x3r , x3,r+1 ; . . . ; xM 1 , xM 2 , . . . , xM r , xM,r+1 )| %M −1 . . q,1,j (t)|xi∗ | |x+1,i∗ +1 | + q,2,j (t)|x+1,i∗ | |x,i∗ +1 | ≤ =1

. . , + qM,1,j (t)|x1i∗ | |xM,i∗ +1 | + qM,2,j (t)|xM i∗ | |x1,i∗ +1 |

(8.102)

where all 0 ≤ q,1,j , q,2,j ≡ 0 are continuous functions over [a, b]. Put $M # (8.103) W := max q,1,j ∞ , q,2,j ∞ ,j=1 . Also set θ(x) :=

M

|(Daν fj )(x)|,

a ≤ x ≤ b,

(8.104)

j=1

ρ :=

M

|Aj |,

(8.105)

j=1

Φ∗ (x) :=

2 √ 3ν − 3k − 2

(x − a) (3ν−3k−1)/2 , (Γ(ν − k))3/2

(8.106)

176

8. Canavati Fractional Opial-Type Inequalities

all a ≤ x ≤ b, and Q(x) := 2M W Φ∗ (x),

a ≤ x ≤ b.

σ := Q(x)∞ , We assume that

a ≤ x ≤ b,

√ (b − a)σ ρ < 2.

(8.107) (8.108) (8.109)

Call * 4ρ3/2 (x − a) − σρ2 (x − a)2 , all a ≤ x ≤ b. (8.110) ϕ ˜ (x) := ρ + Q(x) · √ (2 − σ ρ(x − a))2 )

Then θ(x) ≤ ϕ ˜ (x),

all a ≤ x ≤ b;

(8.111)

j = 1, . . . , M, all a ≤ x ≤ b.

(8.112)

in particular we have |(Daν fj )(x)| ≤ ϕ ˜ (x), Furthermore we get |fj (x)| ≤ and |(Daγ i fj )(x)| ≤

1 Γ(ν)

1 Γ(ν − γ i )

x

(x − t)ν−1 ϕ ˜ (t) dt,

(8.113)

a

x

(x − t)ν−γ i −1 ϕ ˜ (t) dt,

(8.114)

a

j = 1, . . . , M ; i = 1, . . . , r; all a ≤ x ≤ b. Proof. Notice that W > 0 and σ > 0. For a ≤ x ≤ b we get x x ν (Da fj ) (t)dt = Fj t, ({Daγ i f )(t)}ri=1 , (Daν f )(t)); a a

= 1, . . . , M dt, j = 1, . . . , M.(8.115) That is,

x

(Daν fj )(x) = Aj +

Fj (t, . . .) dt.

(8.116)

a

Then we observe |(Daν fj )(x)| ≤ |Aj | +

x

|Fj (t, . . .)| dt a

≤ |Aj | +

x %M −1 a

=1

. γ γ +1 q,1,j (t)|(Da i∗ f )(t)| |(Da i∗ f+1 )(t)|

8.4 Applications

.

γ +q,2,j (t)|(Da i∗ f+1 )(t)|

γ +1 |(Da i∗ f )(t)|

. γ i∗ γ +1 + qM,1,j (t)|(Da f1 )(t)| |(Da i∗ fM )(t)| . γ i∗ γ i∗ +1 + qM,2,j (t)|(Da fM )(t)| |(Da f1 )(t)| dt. Thus |(Daν fj )(x)|

177

(8.117)

%M −1 . x |(Dak f )(t)| |(Dak+1 f+1 )(t)| ≤ |Aj | + W a

=1

. k+1 k +|(Da f+1 )(t)| |(Da f )(t)| . + |(Dak f1 )(t)| |(Dak+1 fM )(t)| . + |(Dak fM )(t)| |(Dak+1 f1 )(t))| dt. By Corollary 8.14 we obtain |(Daν fj )(x)|

∗

x

≤ |Aj | + 2W Φ (x) a

M

(8.118)

|(Daν f )(t)|3/2

dt ,

(8.119)

=1

j = 1, 2, . . . , M . Therefore by adding all of the inequalities (8.119) we get M x ∗ ν 3/2 θ(x) ≤ ρ + 2M Φ (x)W |(Da f )(t)| dt a

⎛ (by (8.11))

≤

ρ + 2M Φ∗ (x)W ⎝

θ(x) ≤ ρ + (2M Φ∗ (x)W )

x

3/2

⎞ dt⎠ .

(8.120)

(θ(t))3/2 dt , all a ≤ x ≤ b.

(8.121)

a

That is,

=1

x M

|(Daν f )(t)|

=1

a

More precisely we get that x (θ(t))3/2 dt , θ(x) ≤ ρ + Q(x)

a ≤ x ≤ b.

(8.122)

a

Notice that θ(x) ≥ 0, ρ ≥ 0, Q(x) ≥ 0, and Q(a) = 0 by Φ∗ (a) = 0. Acting here as in the proof of Theorem 14 of [26] (see here Theorem 6.28), we derive (8.111) and (8.112). Using (8.9) we get (8.113), and using (8.10) we establish (8.114).

9 Riemann–Liouville Fractional Opial-Type Inequalities for Several Functions and Applications

A wide variety of very general Lp (1 ≤ p ≤ ∞)-form Opial-type inequalities [315] is presented involving Riemann–Liouville fractional derivatives [17, 230, 295, 314] of several functions in diﬀerent orders and powers. Several other particular results of special interest from the established results are derived. Applications of some of these special inequalities are given in proving the uniqueness of solution and in giving upper bounds to solutions of initial value fractional problems involving a very general system of several fractional diﬀerential equations. Upper bounds to various Riemann–Liouville fractional derivatives of the solutions that are involved in the above systems are given too. This treatment is based on [46].

9.1 Introduction Here the author continues his study of Riemann–Liouville fractional Opialtype inequalities now involving several diﬀerent functions and produces a wide variety of corresponding results with important applications to systems of several fractional diﬀerential equations. This chapter continues Chapter 7. We start in Section 9.2 with background, we continue in Section 9.3 with the main results, and we ﬁnish in Section 9.4 with applications. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 179 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 9,

180

9. Riemann–Liouville Fractional Opial Inequalities

To give the reader an idea of the kind of inequalities we are dealing with, we brieﬂy mention a simple one, ⎛ ⎞ x M ⎝ |(Dγ fj )(w)| |(Dν fj )(w)|⎠ dw ≤ 0

j=1

xν−γ √ √ 2Γ(ν − γ) ν − γ 2ν − 2γ − 1

⎧ ⎨ ⎩

0

⎛ x

⎝

M

⎞ ((Dν fj )(w))2 ⎠ dw

j=1

⎫ ⎬ ⎭

,

x ≥ 0, for functions fj ∈ L1 (0, x), j = 1, . . . , M ∈ N; ν > γ ≥ 0, and so on. Here Dβ f stands for the Riemann–Liouville fractional derivative of f of order β ≥ 0. Furthermore one system of fractional diﬀerential equations we are dealing with brieﬂy is of the form (Dν fj )(t) = Fj (t, {(Dγ i f1 )(t)}ri=1 , {(Dγ i f2 )(t)}ri=1 , . . . , {(Dγ i fM )(t)}ri=1 ), all t ∈ [a, b], j = 1, . . . , M ; Dν−k fj (0) = αkj ∈ R, k = 1, . . . , [ν] + 1. Here [ν] is the integral part of ν.

9.2 Background We need Deﬁnition 9.1. (see [187, 295, 314]). Let α ∈ R+ − {0} . For any f ∈ L1 (0, x); x ∈ R+ − {0}, the Riemann–Liouville fractional integral of f of order α is deﬁned by s 1 (s − t)α−1 f (t)dt, ∀s ∈ [0, x], (9.1) (Jα f )(s) := Γ(α) 0 and the Riemann–Liouville fractional derivative of f of order α by m s d 1 Dα f (s) := (s − t)m−α−1 f (t)dt, Γ(m − a) ds 0

(9.2)

where m := [α] + 1, [·] is the integral part. In addition, we set D0 f := f := J0 f, J−α f = Dα f if α > 0, D−α f := Jα f , if 0 < α ≤ 1. If α ∈ N, then Dα f = f (α) the ordinary derivative. Deﬁnition 9.2. [187]. We say that f ∈ L1 (0, x) has an L∞ fractional derivative Dα f in [0, x], x ∈ R+ − {0}, iﬀ Dα−k f ∈ C([0, x]),

9.3 Main Results

181

k = 1, . . . , m := [α] + 1; α ∈ R+ − {0}, and Dα−1 f ∈ AC([0, x]) (absolutely continuous functions) and Dα f ∈ L∞ (0, x). We need Lemma 9.3. [187]. Let α ∈ R+ , β > α, let f ∈ L1 (0, x), x ∈ R+ − {0}, have an L∞ fractional derivative Dβ f in [0, x], and let Dβ−k f (0) = 0 for k = 1, . . . , [β] + 1. Then s 1 (s − t)β−α−1 Dβ f (t)dt, ∀ s ∈ [0, x]. (9.3) Dα f (s) = Γ(β − α) 0 Clearly here Dα f ∈ AC([0, x]) for β − α ≥ 1 and in C([0, x]) for β − α ∈ (0, 1), hence Dα f ∈ L∞ (0, x) and Dα f ∈ L1 (0, x).

9.3 Main Results Here we often use the following basic inequalities. Let α1 , α2 , . . . , αn ≥ 0, n ∈ N; then ar1 + ... + arn ≤ (a1 + ... + an )r ,

r ≥ 1,

(9.4)

0 ≤ r ≤ 1.

(9.5)

and ar1 + ... + arn ≤ n1−r (a1 + ... + an )r , Our ﬁrst result follows. Theorem 9.4. Let α1 , α2 ∈ R+ , β > α1 , α2 , β − αi > (1/p), p > 1, i = 1, 2, and let fj ∈ L1 (0, x), j = 1, . . . , M ∈ N, x ∈ R+ − {0} have, respectively, L∞ fractional derivatives Dβ fj in [0, x], and let Dβ−k fj (0) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M . Consider also p(t) > 0 and q(t) ≥ 0, with all p(t), 1/p(t), q(t) ∈ L∞ (0, x). Let λβ > 0 and λα1 , λα2 ≥ 0, such that λβ < p. Set s Pi (s) := (s − t)p(β−αi −1)/p−1 (p(t))−1/(p−1) dt, i = 1, 2; 0 ≤ s ≤ x, 0

A(s) :=

(9.6) λα1 (p−1/p)

λα2 ((p−1)/p)

−λβ /p

(P2 (s)) (p(s)) q(s)(P1 (s)) (Γ(β − α1 ))λα1 (Γ(β − α2 ))λα2 x (p−λβ )/p p/(p−λβ ) (A(s)) ds , A0 (x) := 0

,

(9.7)

(9.8)

182

9. Riemann–Liouville Fractional Opial Inequalities

and δ ∗1 :=

M 1−((λα1 +λβ )/p) , 2(λα1 +λβ /p)−1 ,

λα1 + λβ ≤ p, λα1 + λβ ≥ p.

if if

Call

ϕ1 (x) := (A0 (x)|λα2 =0 )

λβ λα1 + λβ

(9.9)

λβ /p .

(9.10)

If λα2 = 0, we obtain that, ⎞ ⎛ x M q(s) ⎝ |Dα1 fj (s)|λα1 |Dβ fj (s)|λβ ⎠ ds ≤ 0

j=1

⎡ δ ∗1 ϕ1 (x) ⎣

⎛

x

p(s) ⎝

0

M

⎞

⎤(λα1 +λβ /p)

|Dβ fj (s)|p ⎠ ds⎦

.

(9.11)

j=1

Proof. By Theorem 4 of [48] and here Theorem 7.4 we obtain

x

+ λβ λβ , λ λ ds ≤ q(s) |Dα1 fj (s)| α1 Dβ fj (s) + |Dα1 fj+1 (s)| α1 Dβ fj+1 (s)

0

(λβ /p) ) x + p λβ δ1 p(s) Dβ fj (s) + λα1 + λβ 0 , , (λ +λ /p) β α1 β D fj+1 (s)p ds , j = 1, 2..., M − 1, (9.12)

A0 (x)|λα2 =0

where

21−((λα1 +λβ )/p) , if 1, if

δ 1 :=

λα1 + λβ ≤ p, λα1 + λβ ≥ p.

(9.13)

Hence by adding all the above we get

x

q(s) 0

%M −1 +

|D

α1

fj (s)|

λα1

β

|D fj (s)|

λβ

+ |D

α1

fj+1 (s)|

λα1

β

|D fj+1 (s)|

λβ

,

ds

j=1

≤ δ 1 ϕ1 (x)

⎧ −1 ) ⎨M ⎩

j=1

x

p(s) |Dβ fj (s)|p + |Dβ fj+1 (s)|p ds

⎫ *(λα1 +λβ )/p ⎬ ⎭

0

.

(9.14) Also it holds x

q(s) |Dα1 f1 (s)|λα1 |Dβ f1 (s)|λβ + |Dα1 fM (s)|λα1 |Dβ fM (s)|λβ ds ≤ 0

)

x

p(s) |D f1 (s)| + |D fM (s)|

δ 1 ϕ1 (x) 0

β

p

β

p

*(λα1 +λβ )/p ds

.

(9.15)

9.3 Main Results

Call

λα1 + λβ ≥ p, λα1 + λβ ≤ p.

1, if M 1−(λα1 +λβ /p) , if

ε1 =

(9.16)

Adding (9.14) and (9.15), and using (9.4) and (9.5) we have ⎞ ⎛ x M 2 q(s) ⎝ |Dα1 fj (s)|λα1 |Dβ fj (s)|λβ ⎠ ds ≤ 0

δ 1 ϕ1 (x)

)

x

p(s) |Dβ fj (s)|p + |Dβ fj+1 (s)|p ds

*(λα1 +λβ /p) +

0

j=1

x

(9.17)

j=1

⎧ −1 ) ⎨M ⎩

183

p(s) |D f1 (s)| + |D fM (s)| β

p

β

p

*((λα1 +λβ )/p) ds

≤

0

⎧ ⎨ δ 1 ε1 ϕ1 (x)

⎩

⎛ x

p(s) ⎝2

0

M

⎞ |Dβ fj (s)|p ⎠ ds

j=1

⎫((λα1 +λβ )/p) ⎬ .

⎭

(9.18)

We have derived

⎛ x

q(s) ⎝

0

M

⎞ |Dα1 fj (s)|λα1 |Dβ fj (s)|λβ ⎠ ds ≤

j=1

δ 1 2(λα1 +λβ /p)−1 ε1 ϕ1 (x)

⎧ ⎨ ⎩

0

⎡ x

p(s) ⎣

M

⎤ |Dβ fj (s)|p ⎦ ds

j=1

⎫((λα1 +λβ )/p) ⎬ ⎭

. (9.19)

Clearly here we have δ ∗1 = δ 1 2(λα1 +λβ /p)−1 ε1 . From (9.19) and (9.20) we derive (9.11).

(9.20)

Next we give Theorem 9.5. All here are as in Theorem 9.4. Denote λ /λ 2 α2 β − 1, if λα2 ≥ λβ , δ 3 := 1, if λα2 ≤ λβ , ε2 :=

1, if M 1−(λβ +λα2 /p) , if

λβ + λα2 ≥ p, λβ + λα2 ≤ p,

(9.21)

(9.22)

184

9. Riemann–Liouville Fractional Opial Inequalities

and ϕ2 (x) := A0 (x) λα1 =0 2(p−λβ /p)

λβ λα2 + λβ

λβ /p

(λβ /p)

δ3

.

(9.23)

If λα1 = 0, then

x

q(s)

%%M −1 +

0

|D

α2

fj+1 (s)|

λα2

β

|D fj (s)|

λβ

+ |D

α2

fj (s)|

λα2

β

|D fj+1 (s)|

λβ

,

-

j=1

$ + |Dα2 fM (s)|λα2 |Dβ f1 (s)|λβ + |Dα2 f1 (s)|λα2 |Dβ fM (s)|λβ ds ≤

2(λβ +λα2 /p) ε2 ϕ2 (x)

⎧ ⎨ ⎩

⎡ x

p(s) ⎣

0

M

⎤ |Dβ fj (s)|p ⎦ ds

j=1

⎫((λβ +λα2 )/p) ⎬ . (9.24)

⎭

Proof. From Theorem 5 of [48] and here Theorem 7.5, we have

x

q(s) |Dα2 fj+1 (s)|λα2 |Dβ fj (s)|λβ + |Dα2 fj (s)|λα2 |Dβ fj+1 (s)|λβ ds ≤

0

x

ϕ2 (x)

p(s) |Dβ fj (s)|p + |Dβ fj+1 (s)|p ds

(λβ +λα2 )/p ,

(9.25)

0

for j = 1, . . . , M − 1. Hence by adding all of the above we ﬁnd

x

q(s) 0

M −1 +

|D

α2

fj+1 (s)|

λα2

β

|D fj (s)|

λβ

+ |D

α2

fj (s)|

λα2

β

|D fj+1 (s)|

λβ

,

ds

j=1

⎛ ≤ ϕ2 (x) ⎝

M −1 x

p(s) |Dβ fj (s)|p + |Dβ fj+1 (s)|p ds

(λβ +λα2 )/p

⎞ ⎠.

0

j=1

(9.26) Similarly it holds

x

q(s) |Dα2 fM (s)|λα2 |Dβ f1 (s)|λβ + |Dα2 f1 (s)|λα2 |Dβ fM (s)|λβ ds ≤

0

ϕ2 (x)

x

p(s) |Dβ f1 (s)|p + |Dβ fM (s)|p ds

(λβ +λα2 )/p .

(9.27)

0

Adding (9.26) and (9.27) and using (9.4), (9.5) we derive (9.24).

9.3 Main Results

185

The general case follows. Theorem 9.6. All here are as in Theorem 9.4. Denote ((λ +λ )/λ ) 2 α1 α2 β − 1, if λα1 + λα2 ≥ λβ , γ˜1 := 1, if λα1 + λα2 ≤ λβ ,

(9.28)

and γ˜2 :=

1, if 21−((λα1 +λα2 +λβ )/p) , if

λα1 + λα2 + λβ ≥ p, λα1 + λα2 + λβ ≤ p.

(9.29)

Set ϕ3 (x)

:

λβ /p λβ = A0 (x) (λα1 + λα2 )(λα1 + λα2 + λβ ) + , λλαβ1 /p γ˜2 + 2(p−λβ )/p (γ˜1 λα2 )λβ /p ,

(9.30)

and ε3 :=

1, M 1−(λα1 +λα2 +λβ /p) ,

if if

λα1 + λα2 + λβ ≥ p, λα1 + λα2 + λβ ≤ p.

(9.31)

Then ⎡

x

q(s) ⎣

0

M −1

|Dα1 fj (s)|λα1 |Dα2 fj+1 (s)|λα2 |Dβ fj (s)|λβ +

j=1

|Dα2 fj (s)|λα2 |Dα1 fj+1 (s)|λα1 |Dβ fj+1 (s)|λβ + α1 |D f1 (s)|λα1 |Dα2 fM (s)|λα2 |Dβ f1 (s)|λβ

+ |Dα2 f1 (s)|λα2 |Dα1 fM (s)|λα1 |Dβ fM (s)|λβ ds ⎧ ⎤ ⎫((λα1 +λα2 +λβ )/p) ⎡ M ⎬ ⎨ x p(s) ⎣ |Dβ fj (s)|p ⎦ ds . ≤ 2(λα1 +λα2 +λβ /p) ε3 ϕ3 (x) ⎭ ⎩ 0 j=1

(9.32)

Proof. From Theorem 6 of [48] and here Theorem 7.6, and by adding we ﬁnd M −1 x j=1

0

q(s) |Dα1 fj (s)|λα1 |Dα2 fj+1 (s)|λα2 |Dβ fj (s)|λβ +

186

9. Riemann–Liouville Fractional Opial Inequalities

|Dα2 fj (s)|λα2 |Dα1 fj+1 (s)|λα1 |Dβ fj+1 (s)|λβ ds ≤ ϕ3 (x)

M −1 x

p(s)(|D fj (s)| + |D fj+1 (s)| )ds β

p

β

p

(λα1 +λα2 +λβ )/p .

0

j=1

(9.33) Also it holds

x

q(s) |Dα1 f1 (s)|λα1 |Dα2 fM (s)|λα2 |Dβ f1 (s)|λβ +

0

|Dα2 f1 (s)|λα2 |Dα1 fM (s)|λα1 |Dβ fM (s)|λβ ds ≤ (λα1 +λα2 +λβ )/p x ϕ3 (x) p(s) |Dβ f1 (s)|p + |Dβ fM (s)|p ds .

(9.34)

0

Adding (9.33) and (9.34), along with (9.4), (9.5) we derive (9.32).

We continue with Theorem 9.7. Let β > α1 + 1, α1 ∈ R+ and let fj ∈ L1 (0, x), j = 1, . . . , M ∈ N, x ∈ R+ − {0} have, respectively, L∞ fractional derivatives Dβ fj , in[0, x], and let Dβ−k fj (0) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M . Consider also p(t) > 0 and q(t) ≥ 0, with p(t), 1/p(t), q(t) ∈ L∞ (0, x). Let λα ≥ 0, 0 < λα+1 < 1, and p > 1. Denote λ /(λ ) 2 α α+1 − 1, if λα ≥ λα+1 , θ3 := (9.35) 1, if λα ≤ λα+1 , x (1−λα+1 ) λα+1 θ3 λα+1 (1/(1−λα+1 )) L(x) := 2 (q(s)) ds , (9.36) λα + λα+1 0

and P1 (x) :=

x

(x − s)(β−α1 −1)p/(p−1) (p(s))−1/(p−1) ds,

(9.37)

0

T (x) := L(x)

P1 (x)(p−1/p) Γ(β − α1 )

(λα +λα+1 ) ,

(9.38)

and ω 1 := 2(p−1/p)(λα +λα+1 ) ,

(9.39)

Φ(x) := T (x) ω 1 .

(9.40)

Also put ε4 :=

1, M 1−(λα +λα+1 /p) ,

if if

λα + λα+1 ≥ p, λα + λα+1 ≤ p.

(9.41)

9.3 Main Results

Then

x

q(s) 0

⎧⎧ −1 ⎨⎨M ⎩⎩

187

|Dα1 fj (s)|λα |Dα1 +1 fj+1 (s)|λα+1 +

j=1

|Dα1 fj+1 (s)|λα |Dα1 +1 fj (s)|λα+1

$

$ + |Dα1 f1 (s)|λα |Dα1 +1 fM (s)|λα+1 + |Dα1 fM (s)|λα |Dα1 +1 f1 (s)|λα+1 ds ≤ ⎞ ⎤((λα +λα+1 )/p) ⎡ ⎛ x M 2(λα +λα+1 /p) ε4 Φ(x) ⎣ p(s) ⎝ |Dβ fj (s)|p ⎠ ds⎦ . 0

j=1

(9.42)

Proof. From Theorem 8 of [48] and here Theorem 7.8, we ﬁnd

x

q(s) 0

M −1

α1 |D fj (s)|λα |Dα1 +1 fj+1 (s)|λα+1 +

j=1

|Dα1 fj+1 (s)|λα |Dα1 +1 fj (s)|λα+1 ds ≤ Φ(x)

M −1 ) x

*((λα +λα+1 )/p) p(s)(|D fj (s)| + |D fj+1 (s)| )ds β

p

β

p

.

0

j=1

(9.43) Similarly it holds x

q(s) |Dα1 f1 (s)|λα |Dα1 +1 fM (s)|λα+1 + |Dα1 fM (s)|λα |Dα1 +1 f1 (s)|λα+1 ds 0

) ≤ Φ(x)

*(λα +λα+1 )/p

x

p(s)(|D f1 (s)| + |D fM (s)| )ds β

p

β

p

.

(9.44)

0

Adding (9.43) and (9.44), along with (9.4), (9.5) we obtain (9.42). Next comes the following theorem. Theorem 9.8. All are as in Theorem 9.4. Consider the special case of λα2 = λα1 + λβ . Denote λβ /p λβ T˜(x) := A0 (x) 2(p−2λα1 −3λβ )/p , (9.45) λα1 + λβ 1, if 2(λα1 + λβ ) ≥ p, ε5 := (9.46) M 1−(2(λα1 +λβ )/p) , if 2(λα1 + λβ ) ≤ p.

188

9. Riemann–Liouville Fractional Opial Inequalities

Then 0

x

⎧⎧ −1 + ⎨⎨M λβ λ λ +λ q(s) |Dα1 fj (s)| α1 |Dα2 fj+1 (s)| α1 β Dβ fj (s) ⎩⎩ j=1

+ |Dα2 fj (s)|

λα1 +λβ

|Dα1 fj+1 (s)|

λα1

, β D fj+1 (s)λβ

+ λβ λ λ +λ + |Dα1 f1 (s)| α1 |Dα2 fM (s)| α1 β Dβ f1 (s) + |Dα2 f1 (s)|

λα1 +λβ

⎡ 22(λα1 +λβ )/p ε5 T˜(x) ⎣

|Dα1 fM (s)|

λα1

, β D fM (s)λβ ds ≤

⎞ ⎤2(λα1 +λβ )/p M p β D fj (s) ⎠ ds⎦ p(s) ⎝ . (9.47) ⎛

x

0

j=1

Proof. Based on Theorem 9 of [48] and here Theorem 7.9. The rest is as in the proof of Theorem 9.7. Next we give a special case of the above theorems. Corollary 9.9. (to Theorem 9.4, λα2 = 0, p(t) = q(t) = 1). It holds ⎛ ⎞ x M λβ λ ⎝ |Dα1 fj (s)| α1 Dβ fj (s) ⎠ ds ≤ 0

j=1

⎡ δ ∗1 ϕ1 (x) ⎣

x

⎡ ⎤ ⎤(λα1 +λβ /p) M p β D fj (s) ⎦ ds⎦ ⎣ .

0

(9.48)

j=1

In (9.48), (A0 (x) |λα =0 ) of ϕ1 (x) is given in [48], Corollary 10, Equation 2 (123) there; here see (7.123). Corollary 9.10. (to Theorem 9.4, λα2 = 0, p(t) = q(t) = 1, λα1 = λβ = 1, p = 2). In detail; let α1 ∈ R+ , β > α1 , β − α1 > (1/2), and let fj ∈ L1 (0, x), j = 1, . . . , M ∈ N, x ∈ R+ − {0}, have, respectively, L∞ fractional derivatives Dβ fj in [0, x], and let Dβ−k fj (0) = 0 for k = 1, . . . , [β] + 1; j = 1, . . . , M . Then ⎞ ⎛ x M ⎝ |Dα1 fj (s)| Dβ fj (s)⎠ ds ≤ 0

j=1

9.3 Main Results

x(β−α1 ) √ √ 2Γ(β − α1 ) β − α1 2β − 2α1 − 1

⎧ ⎨ ⎩

0

x

189

⎡ ⎤ ⎫ M ⎬ β 2 ⎣ D fj (s) ⎦ ds . ⎭ j=1

(9.49)

Proof. Based on our Corollary 9.9 and Corollary 11 of [48] (see inequality (130) there; also see here Corollary 7.11). Corollary 9.11. (to Theorem 9.5, λα1 = 0, p(t) = q(t) = 1). It holds ⎧⎧

−1 ) x ⎨⎨M

0

⎩⎩

D

j=1

α2

⎫ λβ λβ *⎬ λα β λα β α 2 fj+1 (s) 2 D fj (s) + D fj (s) 2 D fj+1 (s) ⎭

+ λβ λβ , λ λ ds ≤ + |Dα2 fM (s)| α2 Dβ f1 (s) + |Dα2 f1 (s)| α2 Dβ fM (s)

2(λβ +λα2 /p) ε2 ϕ2 (x)

⎧ ⎨ ⎩

0

x

⎡ ⎤ ⎫((λβ +λα2 )/p) M ⎬ p β D fj (s) ⎦ ds ⎣ . ⎭

(9.50)

j=1

In (9.50), (A0 (x) |λα =0 ) of ϕ2 (x) is given in [48], Corollary 12, Equation 1 (137) there; also see Corollary 7.12 here. Corollary 9.12. (to Theorem 9.5, λα1 = 0, p(t) = q(t) = 1, λα2 = λβ = 1, p = 2). In detail; let α2 ∈ R+ , β > α2 , β − α2 > (1/2), and let fj ∈ L1 (0, x), j = 1, . . . , M ∈ N, x ∈ R+ − {0}, have, respectively, L∞ fractional derivatives Dβ fj in [0,x], and let Dβ−k fj (0) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M . Then ⎫ ⎧⎧ x ⎨⎨M −1 ⎬ β

α2 |D fj+1 (s)| D fj (s) + |Dα2 fj (s)| Dβ fj+1 (s) + ⎭ 0 ⎩⎩ j=1

$ α2 |D fM (s)| Dβ f1 (s) + |Dα2 f1 (s)| Dβ fM (s) ds ⎤ ⎫ ⎧ ⎡ M √ (β−α ) ⎬ ⎨ x 2 2x ⎣ (Dβ fj (s))2 ⎦ ds . √ √ ≤ ⎭ Γ(β − α2 ) β − α2 2β − 2α2 − 1 ⎩ 0 j=1

(9.51)

Proof. From Corollary 9.11 and Corollary 13 of [48], especially Equation (146) there. Also see Corollary 7.13 here.

190

9. Riemann–Liouville Fractional Opial Inequalities

Corollary 9.13. (to Theorem 9.6, λα1 = λα2 = λβ = 1, p = 3, p(t) = q(t) = 1). It holds ⎡ x M −1 α1 ⎣ |D fj (s)| |Dα2 fj+1 (s)| Dβ fj (s) + 0

j=1

|Dα2 fj (s)| |Dα1 fj+1 (s)| Dβ fj+1 (s) + +

,, |Dα1 f1 (s)| |Dα2 fM (s)| Dβ f1 (s) + |Dα2 f1 (s)| |Dα1 fM (s)| Dβ fM (s) ds ≤ ' ' M (( 3 x β ∗ 2 ϕ3 (x) (9.52) D fj (s) ds . 0

Here ϕ∗3

:=

j=1

√ 1 3 2+ √ 3 6

A0 (x),

(9.53)

where in this special case, A0 (x) = 4x(2β−α1 −α2 ) . Γ(β − α1 )Γ(β − α2 )[3(3β − 3α1 − 1)(3β − 3α2 − 1)(2β − α1 − α2 )]2/3 (9.54)

Proof. From Theorem 9.6 and Corollary 14 of [48]; see there Equation (151) which is here (9.54); see also Corollary 7.14. Corollary 9.14. (to Theorem 9.7, λα = 1, λα+1 = 1/2, p = 3/2, p(t) = q(t) = 1). In detail: let β > α1 + 1, α1 ∈ R+ and let fj ∈ L1 (0, x), j = 1, . . . , M ∈ N, x ∈ R+ − {0} have, respectively, L∞ fractional derivatives Dβ fj in [0, x], and let Dβ−k fj (0) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M . Set ( 3β−3α1 −1/2) 2 x Φ∗ (x) := √ . (9.55) 3β − 3α1 − 2 (Γ(β − α1 ))3/2 Then

x

%%M −1 +

0

|Dα1 fj (s)|

|Dα1 +1 fj+1 (s)| + |Dα1 fj+1 (s)|

|Dα1 +1 fj (s)|

,

+

j=1

+

|Dα1 f1 (s)|

, |Dα1 +1 fM (s)| + |Dα1 fM (s)| |Dα1 +1 f1 (s)| ds ≤ ⎛ ⎞ ⎤ ⎡ x M ⎝ (9.56) |Dβ fj (s)|3/2 ⎠ ds⎦ . 2Φ∗ (x) ⎣ 0

j=1

9.3 Main Results

191

Proof. Based on Theorem 9.7 here, and Corollary 15 of [48] (see there Equation (161) which is here (9.55)); see also Corollary 7.15 here. Corollary 9.15. (to Theorem 9.8, here p = 2(λα1 + λβ ) > 1, p(t) = q(t) = 1). It holds ⎧⎧ x ⎨⎨M −1 + λβ λ λ +λ |Dα1 fj (s)| α1 |Dα2 fj+1 (s)| α1 β Dβ fj (s) + 0 ⎩⎩ j=1

|Dα2 fj (s)| +

λα1 +λβ

λα1

|Dα2 fM (s)|

λα1 +λβ

|Dα1 fM (s)|

|Dα1 f1 (s)|

|Dα2 f1 (s)|

|Dα1 fj+1 (s)|

⎡ 2 T˜(x) ⎣

λα1

, β D fj+1 (s)λβ +

λα1 +λβ

λα1

β D f1 (s)λβ +

, β D fM (s)λβ ds ≤

⎛ x

⎞ ⎤ M 2(λα1 +λβ ) β D fj (s) ⎝ ⎠ ds⎦ .

0

(9.57)

j=1

Here T˜(x) in (9.57) is given by (9.45) and in detail by T˜˜(x) of [48]; see there Corollary 16 and Equations (165) – (169). Also see here Corollary 7.16 and (7.165)–(7.169) . Corollary 9.16. (to Theorem 9.8, p = 4, λα1 = λβ = 1, p(t) = q(t) = 1). It holds ⎧⎧ x ⎨⎨M −1 + 2 |Dα1 fj (s)| (Dα2 fj+1 (s)) Dβ fj (s) + 0 ⎩⎩ j=1

, 2 (Dα2 fj (s)) |Dα1 fj+1 (s)| Dβ fj+1 (s) + +

2 |Dα1 f1 (s)| (Dα2 fM (s)) Dβ f1 (s) +

, 2 (Dα2 f1 (s)) |Dα1 fM (s)| Dβ fM (s) ds ≤ ⎡ 2 T˜(x) ⎣

0

x

⎛ ⎞ ⎤ M β 4 ⎝ D fj (s) ⎠ ds⎦ .

(9.58)

j=1

Here in (9.58) we have that T˜(x) = T ∗ (x) of Corollary 17 of [48]; for that see Equations (177)–(181) there. Also see here Corollary 7.17, Equations (7.177)–(7.181).

192

9. Riemann–Liouville Fractional Opial Inequalities

Next we present the L∞ case. Theorem 9.17. Let α1 , α2 ∈ R+ , β > α1 , α2 , and let fj ∈ L1 (0, x), j = 1, . . . , M ∈ N, x ∈ R+ − {0}, have, respectively, L∞ fractional derivatives Dβ fj in [0, x], and let Dβ−k fj (0) = 0 for k = 1, . . . , [β] + 1; j = 1, . . . , M . Consider p(s) ≥ 0, p(s) ∈ L∞ (0, x) . Let λα1 , λα2 , λβ ≥ 0. Set p(s)∞ ρ(x) := (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα2 x(βλα1 −α1 λα1 +βλα2 −α2 λα2 +1) [βλα1 − α1 λα1 + βλα2 − α2 λα2 + 1] Then

x

p(s) 0

⎧⎧ −1 + ⎨⎨M ⎩⎩

|Dα1 fj (s)|

λα1

|Dα2 fj+1 (s)|

λα2

.

(9.59)

β D fj (s)λβ +

j=1

|Dα2 fj (s)|

λα2

|Dα1 fj+1 (s)|

λα1

, β D fj+1 (s)λβ +

+ λβ λ λ |Dα1 f1 (s)| α1 |Dα2 fM (s)| α2 Dβ f1 (s) λβ , λ λ ds ≤ + |Dα2 f1 (s)| α2 |Dα1 fM (s)| α1 Dβ fM (s) ⎧ ⎫ M ⎨ β 2(λα1 +λβ ) β 2λα2 ⎬ D fj ρ(x) + D fj ∞ . ∞ ⎩ ⎭

(9.60)

j=1

Proof. Based on Theorem 18 of [48]; see also Theorem 7.18 here.

Similarly we give Theorem 9.18. (as in Theorem 9.17, λα2 = 0). It holds ⎞ ⎛ x M p(s) ⎝ |Dα1 fj (s)|λα1 |Dβ fj (s)|λβ ⎠ ds ≤ 0

j=1

x(βλα1 −α1 λα1 +1) p(s)∞ (βλα1 − α1 λα1 + 1)(Γ(β − α1 + 1))λα1

⎞ M β λα1 +λβ D fj ⎠. ⎝ ∞ ⎛

j=1

(9.61) Proof. Based on Theorem 19 of [48]; see Theorem 7.19 here.

9.3 Main Results

193

It follows Theorem 9.19. (as in Theorem 9.17, λα2 = λα1 + λβ ). It holds ⎧⎧ x −1 + ⎨⎨M λβ λ λ +λ |Dα1 fj (s)| α1 |Dα2 fj+1 (s)| α1 β Dβ fj (s) p(s) + ⎩⎩ 0 j=1

|Dα2 fj (s)| +

λα1 +λβ

λα1

|Dα2 fM (s)|

λα1 +λβ

|Dα1 fM (s)|

|Dα1 f1 (s)|

|Dα2 f1 (s)|

|Dα1 fj+1 (s)|

λα1

, β D fj+1 (s)λβ +

λα1 +λβ

λα1

β D f1 (s)λβ +

, β D fM (s)λβ ds ≤

2 x(2βλα1 −α1 λα1 +βλβ −α2 λα1 −α2 λβ +1) (2βλα1 − α1 λα1 + βλβ − α2 λα1 − α2 λβ + 1) p(s)∞ (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))(λα1 +λβ ) ⎞ ⎛ M β 2(λα1 +λβ ) D fj ⎠. ⎝ ∞

(9.62)

j=1

Proof. By Theorem 20 of [48]; see Theorem 7.20 here.

We continue with Theorem 9.20. (as in Theorem 9.17, λβ = 0, λα1 = λα2 ). It holds ⎧⎧ x −1 + ⎨⎨M λ λ |Dα1 fj (s)| α1 |Dα2 fj+1 (s)| α1 + p(s) ⎩⎩ 0 j=1

|Dα2 fj (s)|

λα1

|Dα1 fj+1 (s)|

λα1

,

, + λ λ λ λ + |Dα1 f1 (s)| α1 |Dα2 fM (s)| α1 + |Dα2 f1 (s)| α1 |Dα1 fM (s)| α1 ds ⎤ M Dβ fj 2λα1 ⎦ . ≤ 2 ρ∗ (x) ⎣ ∞ ⎡

j=1

Here we have

(9.63)

194

9. Riemann–Liouville Fractional Opial Inequalities

ρ∗ (x) := (2β λα1 − α1 λα1

x(2βλα1 −α1 λα1 −α2 λα1 +1) p(s)∞ . − α2 λα1 + 1)(Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα1 (9.64)

Proof. Based on Theorem 21 of [48]; see here Theorem 7.21.

Next we give Theorem 9.21. (as in Theorem 9.17, λα1 = 0, λα2 = λβ ). It holds ⎧⎧ x −1 + ⎨⎨M λα λ |Dα2 fj+1 (s)| α2 Dβ fj (s) 2 + p(s) ⎩⎩ 0 j=1

|Dα2 fj (s)|

λα2

, β D fj+1 (s)λα2

, β D f1 (s)λα2 + |Dα2 f1 (s)|λα2 Dβ fM (s)λα2 ds ≤ ⎞ ⎛ M β 2λα2 x(βλα2 −α2 λα2 +1) p(s)∞ D fj ⎠ . (9.65) ⎝ 2 ∞ (βλα2 − α2 λα2 + 1)(Γ(β − α2 + 1))λα2 j=1 +

|Dα2 fM (s)|

λα2

Proof. Based on Theorem 22 of [48]; see Theorem 7.22 here.

Some special cases follow. Corollary 9.22. (to Theorem 9.20, all are as in Theorem 9.17, λβ = 0, λα1 = λα2 , α2 = α1 + 1). It holds ⎧⎧ x −1 + ⎨⎨M λα λ |Dα1 fj (s)| α1 Dα1 +1 fj+1 (s) 1 + p(s) ⎩⎩ 0 j=1

, α +1 D 1 fj (s)λα1 |Dα1 fj+1 (s)|λα1 ) * λα λα λ λ 1 α1 +1 1 α1 D fM (s) α1 + Dα1 f1 (s) α1 Dα1 +1 fM (s) + D f1 (s) ds ≤

2

(2βλα1

x(2βλα1 −2α1 λα1 −λα1 +1) p(s)∞ − 2α1 λα1 − λα1 + 1)(β − α1 )λα1 (Γ(β − α1 ))2λα1 ⎤ ⎡ M β 2λα1 D fj ⎦. ⎣ ∞ j=1

(9.66)

9.4 Applications

195

Proof. Based on Corollary 23 of [48]; see Corollary 7.23 here. Corollary 9.23. (to Corollary 9.22 ). In detail : let α1 ∈ R+ , β > α1 +1, and let fj ∈ L1 (0, x), j = 1, . . . , M ∈ N, x ∈ R+ − {0}, have, respectively, L∞ fractional derivatives Dβ fj in [0, x], and let Dβ−k fj (0) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M . Then ⎧⎧ ⎫ x ⎨⎨M −1 ⎬ α1

|D fj (s)| |Dα1 +1 fj+1 (s)| + |Dα1 +1 fj (s)| |Dα1 fj+1 (s)| + ⎭ 0 ⎩⎩ j=1

$ α1 |D f1 (s)| |Dα1 +1 fM (s)| + |Dα1 +1 f1 (s)| |Dα1 fM (s)| ds ⎛ ⎞ M β 2 x2( β−α1 ) D fj ⎠ . ⎝ ≤ (9.67) ∞ (β − α1 )2 (Γ(β − α1 ))2 j=1 Proof. Based on Corollary 24 of [48]; see Corollary 7.24 here. Corollary 9.24. (to Corollary 29.3). It holds ⎛ ⎞ x M ⎝ |Dα1 fj (s)| |Dα1 +1 fj (s)|⎠ ds ≤ 0

j=1

x2 (β−α1 ) 2(β − α1 )2 (Γ(β − α1 ))2

⎞ M β 2 D fj ⎠ . ⎝ ∞ ⎛

(9.68)

j=1

Proof. Based on inequality (207) of [48]; see also (7.207) here.

9.4 Applications We present our ﬁrst application. Theorem 9.25. Let αi ∈ R+ , β > αi , β − αi > (1/2), i = 1, . . . , r ∈ N, and let fj ∈ L1 (0, x), j = 1, . . . , M ∈ N; x ∈ R+ − {0}, have, respectively, L∞ fractional derivatives Dβ fj in [0, x], and let Dβ−k fj (0) = αkj ∈ R for k = 1, . . . , [β] + 1; j = 1, . . . , M . Furthermore for j = 1, . . . , M , we have that r,M Dβ fj (s) = Fj s, {Dαi fj (s)}i=1,j=1 , (9.69)

196

9. Riemann–Liouville Fractional Opial Inequalities

all s ∈ [0, x]. Here Fj (s, z1 , z2 , . . . , zM ) are continuous for (z1 , z2 , . . . , zM ) ∈ (Rr )M , bounded for s ∈ [0, x], and satisfy the Lipschitz condition |Fj (t; x11 , x12 , ..., x1r ; x21 , x22 , ..., x2r ; x31 , ..., x3r ; ..., xM 1 , ..., xM r )

−Fj (t, x11 , x12 , ..., x1r ; x21 , x22 , ..., x2r ; x31 , ..., x3r ; ..., xM 1 , ..., xM r )| ≤ M r q,i,j (t)|xi − xi | , (9.70) i=1

=1

j = 1, 2, . . . , M , where all q,i,j (s) ≥ 0 are bounded on [0, x], 1 ≤ i ≤ r, = 1, . . . , M . Call W := max {q,i,j ∞ , | , j = 1, 2, ..., M ; i = 1, ..., r} .

(9.71)

Assume here that ∗

φ (x) := W

1 M −1 + √ 2 2

r i=1

xβ−αi √ √ Γ(β − αi ) β − αi 2β − 2αi − 1

< 1.

(9.72)

Then, if the system (9.69) has two pairs of solutions (f1 , f2 , . . . , fM ) and ∗ ), we prove that fj = fj∗ , j = 1, 2, . . . , M ; that is, we have (f1∗ , f2∗ , . . . , fM uniqueness of solution for the fractional system (9.69). Proof. Assume that there are two M -tuples of solutions (f1 , . . . , fM ) and ∗ (f1∗ , . . . , fM ) satisfying the system (9.69). Set gj := fj −fj∗ , j = 1, 2, . . . , M. β−k Then D gj (0) = 0, k = 1, . . . , [β] + 1; j = 1, . . . , M. It holds # $r,M r,M Dβ gj (s) = Fj s, {Dαi fj (s)}i=1,j=1 −Fj s, Dαi fj∗ (s) i=1,j=1 . (9.73) Hence by (9.70) we get |D gj (s)| ≤ β

M r i=1

Thus |D gj (s)| ≤ W β

αi

g,i,j (t)|D g (s)| .

(9.74)

=1

M r i=1

|D g (s)| . αi

(9.75)

=1

The last implies (D gj (s)) ≤ W β

2

M r i=1

=1

|D gj (s)| |D g (s)| , β

αi

(9.76)

9.4 Applications

197

and M

(Dβ gj (s))2 ≤ W

r M M

j=1

|Dβ gj (s)| |Dαi g (s)|.

(9.77)

i=1 j=1 =1

Integrating (9.77) we observe

x

I :=

⎛ ⎞ M ⎝ (Dβ gj (s))2 ⎠ ds ≤

0

W

j=1

⎧ M r M ⎨ ⎩

i=1 j=1

x

|Dβ gj (s)| |Dαi g (s)|ds

⎫ ⎬

0

=1

⎭

.

(9.78)

That is, I≤W

% r ' i=1

τ ,m∈{1,...,M } τ =m

x

0

x

0

M

|D gλ (s)| |D gλ (s)| ds + αi

β

λ=1

⎤⎫ ⎪ ⎬ αi ⎥ |D gm (s)| |Dβ gτ (s)| + |Dαi gτ (s)| |Dβ gm (s)| ds ⎦ . ⎪ ⎭

(9.79) Using Corollary 9.10 from here, and Corollary 13 of [48] (see also here Corollary 7.13) we ﬁnd % r ) x(β−αi ) I √ √ I≤W + 2 Γ(β − αi ) β − αi 2β − 2αi − 1 i=1

x(β−αi ) (M − 1)I √ √ √ 2 Γ(β − αi ) β − αi 2β − 2αi − 1

That is, we get that

* .

(9.80)

I ≤ φ∗ (x) I.

(9.81)

∗

∗

If I = 0 then φ (x) ≥ 1, a contradiction by assumption that φ (x) < 1; see (9.72). Therefore I = 0, implying that M

(Dβ gj (s))2 = 0,

a.e. in [0, x].

(9.82)

j=1

That is, (Dβ gj (s))2 = 0,

a.e. in [0, x],

j = 1, ..., M,

(9.83)

and Dβ gj (s) = 0,

a.e. in [0, x],

j = 1, ..., M.

(9.84)

198

9. Riemann–Liouville Fractional Opial Inequalities

By Dβ−k gj (0) = 0, k = 1, . . . , [β] + 1; j = 1, . . . , M , and Lemma 9.3, apply (9.3) for α = 0, we ﬁnd gj (s) ≡ 0, all s ∈ [0, x], all j = 1, . . . , M . The last implies fj = fj∗ , j = 1, ..., M , over [0, x], thus proving the uniqueness of solution to the initial value problem of this theorem. Another related application follows. Theorem 9.26. Let αi ∈ R+ , β > αi , β − αi > (1/2), i = 1, . . . , r ∈ N, and let fj ∈ L1 (0, x), j = 1, . . . , M ∈ N; x ∈ R+ − {0}, have, respectively, fractional derivatives Dβ fj in [0, x], that are absolutely continuous on [0, x], and let Dβ−k fj (0) = 0 for k = 1, . . . , [β] + 1; j = 1, . . . , M . And (Dβ fj )(0) = Aj ∈ R . Furthermore for 0 ≤ s ≤ x we have holding the system of fractional diﬀerential equations β # $M r,M D fj (s) = Fj s, {Dαi fj (s)}i=1,j=1 , Dβ fj (s) j=1 , (9.85) for j = 1, . . . , M . Here Fj is Lebesgue measurable on [0, x] × (Rr+1 )M such that |Fj (t; x11 , x12 , ..., x1r , x1,r+1 ; x21 , x22 , ..., x2r , x2,r+1 ; ... M r q,i,j (t)|xi | , (9.86) xM 1 , xM 2 , ..., xM,r+1 )| ≤ i=1

=1

where q,i,j ≥ 0, 1 ≤ i ≤ r; , j = 1, . . . , M , are bounded on [0, x]. Set W := max{q,i,j ∞ ; , j = 1, ..., M ; i = 1, ..., r}.

(9.87)

Also we put (0 ≤ s ≤ x) θ(s) :=

M

(Dβ fλ (s))2 ,

(9.88)

λ=1

ρ :=

Q(s) := W (1 +

√

2(M − 1))

M

A2λ ,

(9.89)

λ=1 r i=1

sβ−αi √ √ , Γ(β − αi ) β − αi 2β − 2αi − 1 (9.90)

and χ(s) :=

) s √ ρ 1 + Q(s) · e( 0 Q(t)dt) ·

s

e−(

t 0

Q(y)dy)

*1/2

dt

. (9.91)

0

Then

θ(s) ≤ χ(s),

0 ≤ s ≤ x.

(9.92)

9.4 Applications

199

Consequently we get |Dβ fj (s)| ≤ χ(s), s 1 |fj (s)| ≤ (s − t)β−1 χ(t)dt, Γ(β) 0

(9.93) (9.94)

all 0 ≤ s ≤ x, j = 1, . . . , M . Also it holds s 1 αi |D fj (s)| ≤ (s − t)β−αi −1 χ(t)dt, Γ(β − αi ) 0

(9.95)

all 0 ≤ s ≤ x, j = 1, . . . , M, i = 1, . . . , r. Proof. We see that (Dβ fj )(s)(Dβ fj ) (s) = (Dβ fj )(s) Fj

M $r,M # s, Dαi fj (s) i=1,j=1 , Dβ fj (s) , j=1

(9.96) j = 1, . . . , M , all 0 ≤ s ≤ x. We then integrate (9.96), y (Dβ fj )(s)(Dβ fj ) (s)ds = 0

0 y

# $M r,M (Dβ fj )(s) Fj s, {Dαi fj (s)}i=1,j=1 , Dβ fj (s) j=1 ds,

(9.97)

0 ≤ y ≤ x. Hence we obtain

y y (9.86) ((Dβ fj )(s))2 β ≤ |D f (s)| |F · · · |ds ≤ j j 2 (9.87) 0 0 ' r M ( y β αi W |D fj (s)| |D f (s)| ds = 0

W

i=1

r M i=1

=1

=1

y

|D f (s)| |D fj (s)|ds αi

β

.

(9.98)

0

Thus we have for j = 1, . . . , M that r M β 2 2 (D fj (y)) ≤ Aj + 2W i=1

y

|D f (s)| |D fj (s)|ds αi

β

.

0

=1

(9.99) Consequently it holds ⎛ ⎛ M r M ⎝ θ(y) ≤ ρ + 2W ⎝ i=1

j=1

=1

y

⎞⎞ ⎠⎠ = |Dαi f (s)| |Dβ fj (s)|ds

0

(9.100)

200

9. Riemann–Liouville Fractional Opial Inequalities

% r %

ρ + 2W

i=1

τ ,m∈{1,...,M } τ =m

y

0

y

0

M

|Dαi fλ (t)| |Dβ fλ (t)| dt

+

λ=1

⎫⎫ ⎪ ⎬⎪ ⎬ αi . |D fm (t)| |Dβ fτ (t)| + |Dαi fτ (t)| |Dβ fm (t)| dt ⎪ ⎭⎪ ⎭

(9.101) Using Corollary 9.10 from here, and Corollary 13 of [48] (see here Corollary 7.13) we ﬁnd % θ(y) ≤ ρ + 2 W

r i=1

%

y (β−αi ) √ √ 2 Γ(β − αi ) β − αi 2β − 2αi − 1

y (β−αi ) √ √ √ 2 Γ(β − αi ) β − αi 2β − 2αi − 1

y

θ(t)dt +

0

(M − 1)

y

θ(t)dt

.

0

(9.102) Hence we have

y

θ(y) ≤ ρ + Q(y)

θ(t)dt,

all 0 ≤ y ≤ x.

(9.103)

0

Here ρ ≥ 0, Q(y) ≥ 0, Q(0) = 0, θ(y) ≥ 0, all 0 ≤ y ≤ x. As in the proof of Theorem 27 of [48] (here also see the proof of Theorem 7.27; see also [17]) we get (9.92), (9.93). Using Lemma 9.3 (see (9.3)), for α = 0, and (9.93) we obtain (9.94). Again using (9.3) and (9.93) we get (9.95). Finally we give a specialized application. Theorem 9.27. Let αi ∈ R+ , β > αi , i = 1, . . . , r ∈ N, and fj ∈ L1 (0, x), j = 1, . . . , M ∈ N; x ∈ R+ − {0} , have, respectively, fractional derivatives Dβ fj in [0, x], that are absolutely continuous on [0, x], and let Dβ−μ fj (0) = 0 for μ = 1, . . . , [β] + 1; j = 1, . . . , M . And (Dβ fj )(0) = Aj ∈ R. Furthermore for 0 ≤ s ≤ x we have holding the system of fractional diﬀerential equations β # $M r,M D fj (s) = Fj s, {Dαi fj (s)}i=1,j=1 , Dβ fj (s) j=1 , (9.104) for j = 1, . . . , M . For ﬁxed i∗ ∈ {1, . . . , r} we assume that αi∗ +1 = αi∗ + 1 , where αi∗ , αi∗ +1 ∈ {α1 , . . . , αr }. Call k := αi∗ , γ := αi∗ + 1; that is, γ = k + 1 . Here Fj is Lebesgue measurable on [0, x] × (Rr+1 )M such that |Fj (t, x11 , x12 , ..., x1r , x1,r+1 ; x21 , x22 , ..., x2r , x2,r+1 ; x31 , x32 , ..., x3r , x3,r+1 ; ...; xM 1 , xM 2 , ..., xM r , xM,r+1 )|

9.4 Applications

201

%%M −1 . . ≤ + q,1,j (t)|xi∗ | |x+1,i∗ +1 | + q,2,j (t)|x+1,i∗ | |x,i∗ +1 | =1

. . qM,1,j (t)|x1i∗ | |xM,i∗ +1 | + qM,2,j (t)|xM i∗ | |x1,i∗ +1 | ,

(9.105)

where all 0 ≤ q,1,j , q,2,j ≡ 0, are bounded over [0, x]. Call M

W := max {q,1,j ∞ , q,2,j ∞ },j=1 . Also set θ(s) :=

M

|(Dβ fj )(s)|,

0 ≤ s ≤ x,

(9.106)

(9.107)

j=1 M

ρ :=

|Aj |,

j=1

∗

Φ (s) :=

2 √ 3β − 3k − 2

s((3β−3k−1)/2) , (Γ(β − k))3/2

(9.108)

0 ≤ s ≤ x,

(9.109)

all 0 ≤ s ≤ x, and Q(s) := 2M W Φ∗ (s),

4M W x((3β−3k−1)/2) . σ := Q(s)∞ = √ 3β − 3k − 2 (Γ(β − k))3/2 We assume that Set

√ xσ ρ < 2.

(9.110)

(9.111)

* 4ρ3/2 s − σρ2 s2 , all 0 ≤ s ≤ x. ϕ ˜ (s) := ρ + Q(s) √ (2 − σ ρ s)2 )

(9.112)

Then θ(s) ≤ ϕ ˜ (s),

all

0 ≤ s ≤ x,

(9.113)

in particular we have ˜ (s), |Dβ fj (s)| ≤ ϕ

j = 1, ..., M, for all 0 ≤ s ≤ x.

(9.114)

Furthermore we get |fj (s)| ≤ and |Dαi fj (s)| ≤

1 Γ(β)

1 Γ(β − αi )

s

(s − t)β−1 ϕ ˜ (t)dt,

(9.115)

(s − t)β−αi −1 ϕ ˜ (t)dt,

(9.116)

0

0

s

202

9. Riemann–Liouville Fractional Opial Inequalities

j = 1, . . . , M ; i = 1, . . . , r, all 0 ≤ s ≤ x. Proof. Notice here that W > 0 and σ > 0 . Clearly, here Dβ fj are L∞ fractional derivatives in [0, x]. For 0 ≤ s ≤ y ≤ x, by (9.104) we get that y y β # $M r,M D fj (s)ds = Fj s, {Dαi fj (s)}i=1,j=1 , Dβ fj (s) j=1 ds. 0

0

(9.117)

That is,

y

β

(D fj )(y) = Aj +

Fj (s, ...)ds.

(9.118)

0

Then we observe that

y

|Dβ fj (y)| ≤ |Aj | +

|Fj (s, ...)|ds ≤ |Aj |+ 0

y

W

%%M −1

0

(|D

αi∗

. f (s)|

|Dαi∗ +1 f+1 (s)| + |Dαi∗ f+1 (s)|

.

|Dαi∗ +1 f (s)|)

=1

+(|Dαi∗ f1 (s)| |Dαi∗ +1 fM (s)| + |Dαi∗ fM (s)| |Dαi∗ +1 f1 (s)|) ds. (9.119) That is, %%M −1 . y β |D fj (y)| ≤ |Aj | + W |Dk f (s)| |Dk+1 f+1 (s)|+ 0

=1

. |Dk f+1 (s)| |Dk+1 f (s)| +

. . ds. |Dk f1 (s)| |Dk+1 fM (s)| + |Dk fM (s)| |Dk+1 f1 (s)|

(9.120)

By Corollary 9.14 we obtain

∗

s

|D fj (s)| ≤ |Aj | + 2W φ (s) β

M

0

|D f (t)| β

3/2

dt ,

j = 1, . . . , M , all 0 ≤ s ≤ x. Therefore by adding all of inequalities (9.121) we ﬁnd M s ∗ β 3/2 θ(s) ≤ ρ + 2M W φ (s) dt , |D f (t)| 0

⎛ (9.4)

≤ ρ + 2M W φ∗ (s) ⎝

0

s

(9.121)

=1

(9.122)

=1

M =1

3/2 |Dβ f (t)|

⎞ dt⎠ .

(9.123)

9.4 Applications

That is, ∗

θ(s) ≤ ρ + (2M W φ (s))

203

s 3/2

(θ(t))

dt ,

(9.124)

all 0 ≤ s ≤ x. More precisely we get that s 3/2 (θ(t)) dt , all 0 ≤ s ≤ x. θ(s) ≤ ρ + Q(y)

(9.125)

0

0

Notice that θ(s) ≥ 0, ρ ≥ 0, Q(s) ≥ 0, and Q(0) = 0 by Φ∗ (0) = 0. Acting here as in the proof of Theorem 28 of [48] (also see Theorem 7.28 here) we derive (9.113) and (9.114). Using Lemma 9.3 (see (9.3)), along with (9.114), we obtain (9.115) and (9.116).

10 Converse Canavati Fractional Opial-Type Inequalities for Several Functions

A collection of very general Lp (0 < p < 1)-form converse Opial-type inequalities [315] is presented involving generalized Canavati fractional derivatives [17, 101] of several functions in diﬀerent orders and powers. Other particular results of special interest are derived from the established results. This treatment is based on [51].

10.1 Introduction Opial inequalities appeared for the ﬁrst time in [315] and since then many authors have dealt with them in diﬀerent directions and for various cases. For a complete account of the recent activity in this ﬁeld see [4], and it still remains a very active area of research. One of the main attractions to these inequalities is their applications, especially in proving the uniqueness and upper bounds of the solution of initial value problems in diﬀerential equations. The author was the ﬁrst to present Opial-type inequalities involving fractional derivatives of functions in [15, 17] with applications to fractional diﬀerential equations. Fractional derivatives come up naturally in a number of ﬁelds, especially in physics; see the recent books [197, 333]. Here the author continues his study of fractional Opial-type inequalities now involving several diﬀerent functions and produces a wide variety of converse results. To give the reader an idea of the kind of inequalities we are dealing with, we brieﬂy mention G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 205 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 10,

206

10. Converse Canavati Fractional Inequalities

a speciﬁc one (see Corollary 10.15). ⎛ ⎞ x M λα v λv γ Dx 1 fj (w) Dx fj (w) ⎠ dw ≥ ⎝ 0 0 x0

j=1

⎡ C (x) ⎣

x

x0

⎡ ⎤ ⎤(λα +λv /p) M p v Dx fj (w) ⎦ dw⎦ ⎣ , 0 j=1

all x0 ≤ x ≤ b. In the last inequality C (x) is a constant that depends on x0 , x, and the involved parameters, γ 1 ≥ 0, 1 ≤ v − γ 1 < 1/p, 0 p. Here fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [v] (integral part); j = 1, . . . , M. γ And Dx01 fj , Dxv0 fj are the generalized (of Canavati) type [15, 101] fractional derivatives of fj of orders γ 1 , v, respectively.

10.2 Preliminaries In the sequel we follow [101]. Let g ∈ C ([0, 1]) . Let v be a positive number, n := [v], and α := v − n (0 < α < 1) . Deﬁne x 1 v−1 (x − t) g (t) dt, 0 ≤ x ≤ 1, (10.1) (Jv g) (x) := Γ (v) 0 the Riemann–Liouville integral, where Γ is the gamma function. We deﬁne the subspace C v ([0, 1]) of C n ([0, 1]) as follows. # $ C v ([0, 1]) := g ∈ C n ([0, 1]) : J1−α Dn g ∈ C 1 ([0, 1]) , where D := d/dx. So for g ∈ C v ([0, 1]) , we deﬁne the Canavati v-fractional derivative of g as (10.2) Dv g := DJ1−α Dn g. When v ≥ 1 we have the fractional Taylor’s formula g (t) = g (0) + g (0) t + g” (0) + (Jv Dv g) (t) ,

t2 tn−1 + . . . + g (n−1) (0) 2! (n − 1)! for all t ∈ [0, 1] .

(10.3)

When 0 < v < 1 we ﬁnd g (t) = (Jv Dv g) (t) ,

for all t ∈ [0, 1] .

(10.4)

10.2 Preliminaries

207

Next we carry the above notions over to arbitrary [a, b] ⊆ R (see [17]). Let x, x0 ∈ [a, b] such that x > x0 , where x0 < b is ﬁxed. Let f ∈ C ([a, b]) and deﬁne x 1 v−1 (x − t) f (t) dt, x0 ≤ x ≤ b, (10.5) (Jvx0 f ) (x) := Γ (v) x0 the generalized Riemann–Liouville integral. We deﬁne the subspace Cxv0 ([a, b]) of C n ([a, b]) : # $ x0 Cxv0 ([a, b]) := f ∈ C n ([a, b]) : J1−α Dn f ∈ C 1 ([x0 , b]) ; clearly Cx00 ([a, b]) = C ([a, b]); also Cxn0 ([a, b]) = C n ([a, b]) , n ∈ N. For f ∈ Cxv0 ([a, b]) , we deﬁne the generalized v-fractional derivative of f over [x0 , b] as x0 f (n) := Dn f . Dxv0 f := DJ1−α f (n) (10.6) Notice that

x0 J1−α f (n)

1 (x) = Γ (1 − α)

x

−α

(x − t)

f (n) (t) dt

x0

exists for f ∈ Cxv0 ([a, b]) . We recall the following fractional generalization of Taylor’s formula (see [17, 101]). Theorem 10.1. Let f ∈ Cxv0 ([a, b]) , x0 ∈ [a, b] ﬁxed. (i) If v ≥ 1 then f (x) = f (x0 ) + f (x0 ) (x − x0 ) + f ” (x0 ) n−1

f (n−1) (x0 )

(x − x0 ) (n − 1)!

+ Jvx0 Dxv0 f (x) ,

2

(x − x0 ) + ...+ 2

for all x ∈ [a, b] : x ≥ x0 . (10.7)

(ii) If 0 < v < 1 then f (x) = Jvx0 Dxv0 f (x) ,

for all x ∈ [a, b] : x ≥ x0 .

(10.8)

We make Remark 10.2. (1) Dxn0 f = f (n) , n ∈ N. (2) Let f ∈ Cxv0 ([a, b]) , v ≥ 1, and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [v] . Then by (10.7), f (x) = Jvx0 Dxv0 f (x) .

208

10. Converse Canavati Fractional Inequalities

That is, 1 f (x) = Γ (v)

x

(x − t)

v−1

x0

v Dx0 f (t) dt,

(10.9)

for all x ∈ [a, b] with x ≥ x0 . Notice that (10.9) is true, also when 0 < v < 1. We need the following from [17]. Lemma 10.3. Let f ∈ C ([a, b]) , μ, v > 0. Then x0 Jμx0 (Jvx0 f ) = Jμ+v (f ) .

(10.10)

We also make Remark 10.4. Let v ≥ γ + 1, γ ≥ 0, so that γ < v. Call n := [v] , α := v − n; m := [γ] , ρ := γ − m. Note that v − m ≥ 1 and n − m ≥ 1. Let f ∈ Cxv0 ([a, b]) be such that f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Hence by (10.7), f (x) = Jvx0 Dxv0 f (x) , for all x ∈ [a, b] : x ≥ x0 . Therefore by Leibnitz’s formula and Γ (p + 1) = pΓ (p) , p > 0, we get that x0 Dxv0 f (x) , for all x ≥ x0 . (10.11) f (m) (x) = Jv−m It follows that f ∈ Cxγ0 ([a, b]) and thus γ x0 f (m) (x) exists for all x ≥ x0 . Dx0 f (x) := DJ1−ρ

(10.12)

By the use of (10.11) we have on [x0 , b], x0 v x0 x0 x0 x0 f (m) = J1−ρ Jv−m Dxv0 f = J1−ρ Dx0 f J1−ρ ◦ Jv−m v v x0 x0 Dx0 f = Jv−γ+1 Dx0 f , = Jv−m+1−ρ by (10.10). That is,

x0 J1−ρ f (m) (x) =

1 Γ (v − γ + 1)

x

v−γ

(x − t) x0

Dxv0 f (t) dt.

Therefore

1 x0 · Dxγ0 f (x) = D J1−ρ f (m) (x) = Γ (v − γ) x (v−γ)−1 v Dx0 f (t) dt; (x − t) x0

hence

γ x0 v Dx0 f (x) = Jv−γ Dx0 f (x)

(10.13)

10.3 Main Results

209

and is continuous in x on [x0 , b]. In particular when v ≥ 2 we have x v−1 v Dx0 f (x) = Dx0 f (t) dt, x ≥ x0 . (10.14) x0

That is,

Dxv−1 f 0

v−1 Dx0 f (x0 ) = 0.

= Dxv0 f,

10.3 Main Results 10.3.1 Results Involving Two Functions We present our ﬁrst main result. Theorem 10.5. Let γ j ≥ 0, 1 ≤ v − γ j < 1/p, 0 < p < 1, j = 1, 2, (i)

(i)

and f1 , f2 ∈ Cxv0 ([a, b]) with f1 (x0 ) = f2 (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [v] . Here x, x0 ∈ [a, b] : x ≥ x0 . We assume here that Dxv0 fi is of ﬁxed sign on [x0 , b] , i = 1, 2. Consider also p (t) > 0 and q (t) > 0 continuous functions on [x0 , b]. Let λv > 0 and λα , λβ ≥ 0 such that λv > p. Set w (v−γ k −1)p/p−1 −1/p−1 Pk (w) := (w − t) (p (t)) dt, k = 1, 2, x0 ≤ x ≤ b; x0

(10.15) A (w) :=

q (w) · (P1 (w))

λα (p−1/p)

· (P2 (w)) λα

(Γ (v − γ 1 ))

(Γ (v − γ 2 ))

(p (w))

λβ

p/(p−λv )

(A (w))

dw

−λv /p

, (10.16)

(p−λv )/p

x

A0 (x) :=

(λβ (p−1)/p)

,

(10.17)

x0

and δ 1 := 21−((λα +λv )/p) .

(10.18)

If λβ = 0, we obtain that

x

x0

+ λα λv q (w) Dxγ01 f1 (w) · Dxv0 f1 (w) + , γ Dx 1 f2 (w)λα · Dxv f2 (w)λv dw 0 0

≥ A0 (x) |λβ =0 ·

λv λα + λv

)

λv /p

x

· δ1 · x0

p p (w) Dxv0 f1 (w) +

v Dx f2 (w)p dw ((λα +λv )/p) . 0

(10.19)

210

10. Converse Canavati Fractional Inequalities

Proof. From (10.13) and assumption we get w γ 1 v−γ k −1 v Dx k fj (w) = Dx0 fj (t) dt, (w − t) 0 Γ (v − γ k ) x0 for k = 1, 2, j = 1, 2 and for all x0 ≤ w ≤ b. Next applying H¨ older’s inequality with indices p, p/p − 1 we have w γ 1 v−γ k −1 −1/p 1/p Dx k fj (w) = (w − t) (p (t)) (p (t)) 0 Γ (v − γ k ) x0 v Dx fj (t) dt 0

≥

1 Γ (v − γ k )

w

(w − t)

v−γ k −1

(p (t))

−1/p

p−1/p

p/p−1 dt

x0

w

x0

p p (t) Dxv0 fj (t) dt

1 p−1/p (Pk (w)) = Γ (v − γ k )

w

x0

1/p

p p (t) Dxv0 fj (t) dt

1/p .

That is, it holds γ Dx k fj (w) ≥ 0

1 p−1/p (Pk (w)) Γ (v − γ k )

w

x0

p p (t) Dxv0 fj (t) dt

1/p .

(10.20)

Put

w

zj (w) := x0

thus,

p p (t) Dxv0 fj (t) dt,

p zj (w) = p (w) Dxv0 fj (w) , zj (x0 ) = 0; j = 1, 2.

Hence, we have γ Dx k fj (w) ≥ 0

and

1 p−1/p 1/p (Pk (w)) (zj (w)) , Γ (v − γ k )

λv /p v Dx fj (w)λv = p (w)−λv /p zj (w) , 0

j = 1, 2.

Therefore we obtain λα λ λv q (w) Dxγ01 f1 (w) Dxγ02 f2 (w) β Dxv0 f1 (w) ≥ q (w)

1 (Γ (v − γ 1 ))

λα (p−1/p)

λα

(P1 (w))

λα /p

(z1 (w))

10.3 Main Results

1 λβ

(Γ (v − γ 2 ))

(P2 (w))

λβ (p−1/p)

(z2 (w))

λα /p

(z2 (w))

= A (w) (z1 (w))

λβ /p

λβ /p

(p (w))

−λv /p

z1 (w)

211

λv /p

λv /p z1 (w) .

Consequently, by another H¨ older’s inequality application, we ﬁnd (by p/λv < 1) x λα λ λv q (w) Dxγ01 f1 (w) Dxγ02 f2 (w) β Dxv0 f1 (w) dw x0

)

x

≥ A0 (x)

(z1 (w))

λα /λv

λβ /λv

(z2 (w))

*λv /p

z1 (w) dw

.

(10.21)

x0

Similarly one derives x λ λα λv q (w) Dxγ02 f1 (w) β Dxγ01 f2 (w) Dxv0 f2 (w) dw x0

)

x

≥ A0 (x)

(z1 (w))

λβ /λv

(z2 (w))

λα /λv

*λv /p

z2 (w) dw

.

(10.22)

x0

Taking λβ = 0 and adding (10.21) and (10.22) we obtain x + λα λv q (w) Dxγ01 f1 (w) · Dxv0 f1 (w) + x0

, γ Dx 1 f2 (w)λα · Dxv f2 (w)λv dw 0 0 %) *λv /p x λα /λv ≥ A0 (x) |λβ =0 (z1 (w)) z1 (w) dw + x0

)

x

(z2 (w))

λα /λv

*λv /p -

z2 (w) dw

x0

(λ +λ )/p = A0 (x) |λβ =0 (z1 (x)) α v + (λα +λv )/p

(z2 (x)) = A0 (x) |λβ =0

λv λα + λv

x

+ x0

λv /p %

λv λα + λv

x

x0

p p (t) Dxv0 f2 (t) dt

λv /p

p p (t) Dxv0 f1 (t) dt (λα +λv )/p =: (∗) .

(λα +λv )/p

212

10. Converse Canavati Fractional Inequalities

In this chapter we frequently use the basic inequalities r

2r−1 (ar + br ) ≤ (a + b) ≤ ar + br , r

ar + br ≤ (a + b) ≤ 2r−1 (ar + br ) ,

a, b ≥ 0, 0 ≤ r ≤ 1,

(10.23)

a, b ≥ 0, r ≥ 1.

(10.24)

Finally using (10.23) , (10.24), and (10.18) we ﬁnd

(∗) ≥ A0 (x) |λβ =0 ·

x

x0

λv λα + λv

λv /p · δ1

p p

p (t) Dxv0 f1 (t) + Dxv0 f2 (t) dt

(λα +λv )/p .

Inequality (10.19) has been established. Here we see that (p/p − 1) (v − γ i − 1) + 1 > 0, −1/(p − 1) > 0 and p (t) ∈ C ([x0 , b]) , thus (see (10.15)) Pi (w) ∈ R for every w ∈ [x0 , b] , also Pi (w) is continuous and bounded on [x0 , b] for i = 1, 2. By λv > p > 0, we have 0 < p/λv < 1, p/p − λv < 0. We see that 1 1 λ λ λ /p = (Γ (v − γ 1 )) α (Γ (v − γ 2 )) β (p (w)) v A (w) q (w) λα (1−p/p)

(P1 (w))

(P2 (w))

λβ (1−p/p)

∈ C ([x0 , b]) ,

and 1/A (w) > 0 on (x0 , b] , 1/A (x0 ) = 0. Therefore 0 < A0 (x) < ∞, and all we have done in this proof is valid. The counterpart of the last theorem follows. Theorem 10.6. All here are as in Theorem 10.5. Further assume λβ ≥ λv . Denote (10.25) δ 2 := 21−(λβ /λv ) , δ 3 := (δ 2 − 1) 2−(λβ /λv ) . If λα = 0, then x x0

+ λ λv q (w) Dxγ02 f2 (w) β · Dxv0 f1 (w) + , γ Dx 2 f1 (w)λβ · Dxv f2 (w)λv dw 0 0

≥ (A0 (x) |λα =0 ) 2

(p−λv )/p

λv λβ + λv

λv /p

λ /p

δ3 v

10.3 Main Results

x

· x0

p p

p (w) Dxv0 f1 (w) + Dxv0 f2 (w) dw

213

((λv +λβ )/p) ,

(10.26)

all x0 ≤ x ≤ b. Proof. When λα = 0 from (10.21) and (10.22) we have x λ λv q (w) Dxγ02 f2 (w) β Dxv0 f1 (w) dw x0

)

x

≥ (A0 (x) |λα =0 )

(z2 (w))

λβ /λv

*λv /p

z1 (w) dw

,

(10.27)

x0

and

x

x0

λ λv q (w) Dxγ02 f1 (w) β Dxv0 f2 (w) dw )

x

≥ (A0 (x) |λα =0 )

(z1 (w))

λβ /λv

*λv /p

z2 (w) dw

,

(10.28)

x0

all x0 ≤ x ≤ b. Adding (10.27) and (10.28) we obtain x + λ λv q (w) Dxγ02 f2 (w) β · Dxv0 f1 (w) + x0

, γ Dx 2 f1 (w)λβ · Dxv f2 (w)λv dw 0 0 %) * x

≥ (A0 (x) |λα =0 )

λβ /λv

(z2 (w)) x0

)

x

λβ /λv

(z1 (w))

z1 (w) dw

λv /p

+

*λv /p -

z2 (w) dw

x0

≥ (A0 (x) |λα =0 ) · 2p−λv /p · (M (x))

λv /p

=: (∗) ,

by λv /p > 1 and (10.24) , where x λ /λ λ /λ M (x) := (z2 (w)) β v z1 (w) + (z1 (w)) β v z2 (w) dw.

(10.29)

(10.30)

x0

Next we work on M (x). We have that x λ /λ λ /λ (z1 (w)) β v + (z2 (w)) β v z1 (w) + z2 (w) dw M (x) = x0

x

−

+

λβ /λv

(z1 (w)) x0

λβ /λv

z1 (w) + (z2 (w))

, z2 (w) dw

214

10. Converse Canavati Fractional Inequalities

(by (10.24))

≥

−

x

δ2

(z1 (w) + z2 (w))

λβ /λv

(z1 (w) + z2 (w)) dw

x0

λv λβ + λv

+

((λv +λβ )/λv )

(z1 (x))

(λv +λβ /λv )

= δ 2 (z1 (x) + z2 (x))

((λv +λβ )/λv )

,

+ (z2 (x))

λv λv + λβ

−

λv λβ + λv ,

+ (λ +λ /λ ) (λ +λ /λ ) (z1 (x)) v β v + (z2 (x)) v β v + λv (λ +λ /λ ) = δ 2 (z1 (x) + z2 (x)) v β v − λβ + λv , (λ +λ /λ ) (λ +λ /λ ) (z1 (x)) v β v + (z2 (x)) v β v

+ λv (λ +λ /λ ) (λ +λ /λ ) δ 2 (z1 (x)) v β v + (z2 (x)) v β v λβ + λv , (λ +λ /λ ) (λ +λ /λ ) − (z1 (x)) v β v + (z2 (x)) v β v , + λv (λ +λ /λ ) (λ +λ /λ ) = (δ 2 − 1) (z1 (x)) v β v + (z2 (x)) v β v λβ + λv (10.24) λv (λ +λ /λ ) ≥ δ 3 (z1 (x) + z2 (x)) v β v . λβ + λv

(10.24)

≥

That is, we present that λv (λ +λ /λ ) M (x) ≥ δ 3 (z1 (x) + z2 (x)) v β v . λβ + λv Consequently, by (10.29) and (10.31) we obtain (∗) ≥ (A0 (x) |λα =0 ) 2 p−λv /p λ /p

δ3 v

λv λβ + λv

(λ +λ /p)

λv /p

(z1 (x) + z2 (x)) v β λv /p λv λ /p = (A0 (x) |λα =0 ) 2 p−λv /p δ3 v λβ + λv (λv +λβ /p) x v p v p

p (t) Dx0 f1 (t) + Dx0 f2 (t) dt . x0

We have established (10.26) . A special important case follows.

(10.31)

10.3 Main Results

215

Theorem 10.7. Let v ≥ 2 and γ 1 ≥ 0 such that 2 ≤ v − γ 1 < 1/p, (i) (i) 0 < p < 1. Let f1 , f2 ∈ Cxv0 ([a, b]) with f1 (x0 ) = f2 (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [v] . Here x, x0 ∈ [a, b] : x ≥ x0 . We assume here that Dxv0 fj is of ﬁxed sign on [x0 , b] , j = 1, 2. Consider also p (t) > 0 and q (t) > 0 continuous functions on [x0 , b] . Let λα ≥ λα+1 > 1. Denote (10.32) θ3 := 21−(λα /λα+1 ) − 1 2−λα /λα+1 , L (x) := 2

(1−λα+1 )

x

(1/1−λα+1 )

(q (w))

dw

x0

θ3 λα+1 λα + λα+1

λα+1 , (10.33)

and

x

(x − t)

P1 (x) :=

(v−γ 1 −1)p/p−1

(p (t))

−1/p−1

dt,

(10.34)

x0

T (x) := L (x) ·

(p−1/p)

P1 (x) Γ (v − γ 1 )

(λα +λα+1 ) ,

ω 1 := 2(p−1/p)(λα +λα+1 ) ,

(10.35)

Φ (x) := T (x) ω 1 .

(10.36)

and

Then

x

x0

+ λα λα+1 q (w) Dxγ01 f1 (w) Dxγ01 +1 f2 (w)

λα λα+1 , + Dxγ01 f2 (w) Dxγ01 +1 f1 (w) dw

)

x

≥ Φ (x) x0

p p p (w) Dxv0 f1 (w) + Dxv0 f2 (w) dw

(10.37)

*(λα +λα+1 )/p ,

all x0 ≤ x ≤ b. Proof. For convenience we set γ 2 := γ 1 +1. From (10.13) and assumption we obtain w γ 1 v−γ k −1 v Dx k fj (w) = Dx0 fi (t) dt =: gj,γ k (w), (w − t) 0 Γ (v − γ k ) x0 (10.38) where j = 1, 2, k = 1, 2, all x0 ≤ x ≤ b. We observe that γ (10.39) Dx01 fj (x) = Dxγ01 +1 fj (x) = Dxγ02 fj (x) , all x0 ≤ x ≤ b. And also gj,γ 1 (w) = gj,γ 2 (w) ; gj,γ k (x0 ) = 0.

(10.40)

216

10. Converse Canavati Fractional Inequalities

Notice that if v − γ 2 = 1, then gj,γ 2 (w) =

w

x0

v Dx fj (t) dt. 0

Next we apply H¨ older’s inequality with indices 1/λα+1 < 1, 1/(1 − λα+1 ) < 0; we obtain x λα λα+1 q (w) Dxγ01 f1 (w) Dxγ01 +1 f2 (w) dw x0

x

= x0

λα λα+1 g2,γ 1 (w) q (w) g1,γ 1 (w) dw

(1−λα+1 )

x

≥

(10.41)

(1/1−λα+1 )

(q (w))

dw

x0

λα /λα+1 g1,γ 1 (w) g2,γ 1 (w) dw

x

x0

λα+1 .

Similarly we have x λα λα+1 q (w) Dxγ01 f2 (w) Dxγ01 +1 f1 (w) dw x0

(1−λα+1 )

x

≥

(1/1−λα+1 )

(q (w))

dw

x0

x

x0

λα /λα+1 g2,γ 1 (w) g1,γ 1 (w) dw

λα+1 .

Adding (10.41) and (10.42) we see that x + λα λα+1 q (w) Dxγ01 f1 (w) Dxγ01 +1 f2 (w) x0

λα λα+1 , dw + Dxγ01 f2 (w) Dxγ01 +1 f1 (w)

x

≥

(1−λα+1 ) (1/1−λα+1 )

(q (w))

dw

x0

'

x

x0

x

+ x0

λα /λα+1 g1,γ 1 (w) g2,γ 1 (w) dw λα /λα+1 g2,γ 1 (w) g1,γ 1 (w) dw

λα+1 λα+1 (

(10.42)

10.3 Main Results (10.24)

≥

2

)

217

(1−λα+1 )

x

(q (w))

(1/1−λα+1 )

dw

x0

x

· x0

+ λα /λα+1 g1,γ 1 (w) g2,γ 1 (w)

+ g2,γ 1 (w)

λα /λα+1 , ,λα+1 g1,γ 1 (w) dw

(note (10.30) and the proof of (10.31); accordingly here we have) x (1−λα+1 ) (1/1−λα+1 ) ≥ 2 (q (w)) dw x0

λα+1 θ3 λα + λα+1

λα+1

(λα +λα+1 ) g1,γ 1 (x) + g2,γ 1 (x)

(λα +λα+1 ) = L (x) g1,γ 1 (x) + g2,γ 1 (x) L (x)

=

x

v−γ 1 −1

(x − t)

(λα +λα+1 )

−1/p

(p (t))

1/p

(p (t))

x0 (Γ (v − γ 1 )) v v $(λα +λα+1 ) Dx f1 (t) + Dx f2 (t) dt 0 0 p and p we ﬁnd) (applying H¨ older’s inequality with indices p−1 x L (x) (v−γ 1 −1)p/p−1 ≥ · (x − t) (λα +λα+1 ) x0 (Γ (v − γ 1 )) (p−1/p)(λα +λα+1 )

(p (t))

x

x0

−1/p−1

dt

p p (t) Dxv0 f1 (t) + Dxv0 f2 (t) dt

)

x

= T (x) · x0

)

x

≥ Φ (x) · x0

(λα +λα+1 /p)

p p (t) Dxv0 f1 (t) + Dxv0 f2 (t) dt

*(λα +λα+1 /p)

p p p (t) Dxv0 f1 (t) + Dxv0 f2 (t) dt

We have proved (10.37) .

*(λα +λα+1 /p) .

Next we treat the case of exponents λβ = λα + λv . Theorem 10.8. All here are as in Theorem 10.5. Consider the special case of λβ = λα + λv .

218

10. Converse Canavati Fractional Inequalities

Assume here for j = 1, 2 that x p zj (x) := p (t) Dxv0 fj (t) dt ∈ [H, Ψ] , 0 < H < Ψ, x0

h :=

Ψ > 1, H

Mh (1) :=

(h − 1) h1/h−1 . e ln h

(10.43)

Denote T˜ (x) := A0 (x)

λv /p

λv λα + λv

2p−2λα −3λv /p (Mh (1))

−2(λα +λv )/p

.

(10.44) Then

x

x0

+ λα λα +λv v Dx f1 (w)λv q (w) Dxγ01 f1 (w) Dxγ02 f2 (w) 0

λα +λv γ , Dx 1 f2 (w)λα Dxv f2 (w)λv dw + Dxγ02 f1 (w) 0 0

x

≥ T˜ (x) x0

p p p (w) Dxv0 f1 (w) + Dxv0 f2 (w) dw

2((λα +λv )/p) , (10.45)

all x0 ≤ x ≤ b. Proof. We apply (10.21) and (10.22) for λβ = λα + λv and add to ﬁnd x + λα λα +λv v Dx f1 (w)λv q (w) Dxγ01 f1 (w) Dxγ02 f2 (w) 0 x0

λα +λv γ , Dx 1 f2 (w)λα Dxv f2 (w)λv dw + Dxγ02 f1 (w) 0 0 %) * x

≥ A0 (x)

λα /λv

(z1 (w))

(z2 (w))

λα /λv +1

z1 (w) dw

x0

)

x

+

λα /λv +1

(z1 (w))

(z2 (w))

λα /λv

z2

λv /p

*λv /p (w) dw

x0 (10.24)

≥

x

1−λv /p

+

λα /λv

(z1 (w))

A0 (x) 2

(z2 (w))

λα /λv +1

x0

, λv /p λ /λ (z2 (w)) α v z2 (w) dw x λα /λv = A0 (x) 21−λv /p (z1 (w) z2 (w))

+ (z1 (w))

λα /λv +1

x0

[z2 (w) z1 (w) + z1 (w) z2 (w)] dw}

λv /p

z1 (w)

10.3 Main Results

x

= A0 (x) 21−λv /p

(z1 (w) z2 (w)) x0

λα /λv

219

λv /p

(z1 (w) z2 (w)) dw

λ /λ +1 λv /p (z1 (x) z2 (x)) α v = A0 (x) 2 λα λv + 1 λv /p λv (λ +λ )/p p−λv /p = A0 (x) 2 (z1 (x) z2 (x)) α v λα + λv 1−λv /p

(see [377]) ≥ A0 (x) 2p−λv /p

λv λα + λv

λv /p

z1 (x) + z2 (x) 2Mh (1)

2(λα +λv )/p

2(λ +λ )/p = T˜ (x) (z1 (x) + z2 (x)) α v

x

= T˜ (x) x0

p p p (w) Dxv0 f1 (w) + Dxv0 f2 (w) dw

2(λα +λv /p) .

We have established (10.45) . Special cases of the above theorems follow next. Corollary 10.9. (to Theorem 10.5; λβ = 0, p (t) = q (t) = 1). Then

x + x0

, Dxγ 1 f1 (w)λα Dxv f1 (w)λv + Dxγ 1 f2 (w)λα Dxv f2 (w)λv dw 0 0 0 0

Dxv f1 (w)p + Dxv f2 (w)p dw 0 0

x

≥ C1 (x) ·

x0

(λα +λv /p) ,

(10.46)

all x0 ≤ x ≤ b, where C1 (x) := A0 (x) |λβ =0 ·

λv λα + λv

λv /p · δ1,

(10.47)

δ 1 := 21−(λα +λv /p) We have that A0 (x) |λβ =0 =

%

(p − 1) λα

(Γ (v − γ 1 ))

·

(10.48)

((λα p−λα )/p)

(vp − γ 1 p − 1)

((λα p−λα )/p)

-

(p−λv /p)

(p − λv )

((p−λv )/p)

(λα vp − λα γ 1 p − λα + p − λv ) (x − x0 )

((λα vp−λα γ 1 p−λα +p−λv )/p)

.

·

(10.49)

220

10. Converse Canavati Fractional Inequalities

Proof. By Theorem 10.5. The constant A0 (x) |λβ =0 was calculated in [26]; see here (6.55). Corollary 10.10. (to Theorem 10.6; λα = 0, p (t) = q (t) = 1, λβ ≥ λv ). Then

x + x0

, Dxγ 2 f2 (w)λβ Dxv f1 (w)λv + Dxγ 2 f1 (w)λβ Dxv f2 (w)λv dw 0 0 0 0

Dxv f1 (w)p + Dxv f2 (w)p dw 0 0

x

≥ C2 (x)

x0

(λv +λβ /p) ,

(10.50)

δ3 v .

(10.51)

all x0 ≤ x ≤ b, where C2 (x) := (A0 (x) |λα =0 ) 2

(p−λv )/p

We have that

%

(A0 (x) |λα =0 ) =

(p − 1) λβ

(Γ (v − γ 2 ))

·

λv λβ + λv

λv /p

λ /p

((λβ p−λβ )/p)

(vp − γ 2 p − 1)

((λβ p−λβ )/p)

-

(p−λv /p)

(p − λv )

(λβ vp − λβ γ 2 p − λβ + p − λv )

(10.52)

((p−λv )/p)

((λβ vp−λβ γ 2 p−λβ +p−λv )/p)

(x − x0 )

·

.

Proof. By Theorem 10.6. The constant (A0 (x) |λα =0 ) was calculated in [26]; see here (6.59).

10.3.2 Results Involving Several Functions Here we use the following basic inequality. Let α1 , . . . , αn ≥ 0, n ∈ N; then n r r r r−1 r a1 + . . . + an ≤ (a1 + . . . + an ) ≤ n ai , r ≥ 1. (10.53) i=1

We present Theorem 10.11. Let γ 1 , γ 2 ≥ 0 such that 1 ≤ v − γ i < 1/p, 0 < p < 1, (i) i = 1, 2, and fj ∈ Cxv0 ([a, b]) with fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [v] , j = 1, . . . , M ∈ N. Here x, x0 ∈ [a, b] : x ≥ x0 . We assume that Dxv0 fj is of ﬁxed sign on [x0 , b] , j = 1, . . . , M. Consider also p (t) > 0, and q (t) > 0 continuous functions on [x0 , b] . Let λv > 0 and λα , λβ ≥ 0 such that λv > p.

10.3 Main Results

Set

w

(v−γ k −1)p/p−1

(w − t)

Pk (w) :=

−1/p−1

(p (t))

221

dt, k = 1, 2; x0 ≤ w ≤ b;

x0

(10.54) A (w) :=

q (w) (P1 (w))

(λα (p−1)/p)

(P2 (w)) λα

(Γ (v − γ 1 ))

(Γ (v − γ 2 ))

(p (w))

−λv /p

λβ

(A (w))

p/p−λv

dw

; (10.55)

(p−λv )/p

x

A0 (x) :=

(λβ (p−1)/p)

.

(10.56)

,

(10.57)

x0

Call

ϕ1 (x) := A0 (x) |λβ =0 ·

λv λα + λv

λv /p

δ ∗1 := M 1−(λα +λv /p) .

(10.58)

If λβ = 0, we obtain that ⎞ ⎛ x M λα v λv γ Dx 1 fj (w) Dx fj (w) ⎠ dw q (w) ⎝ 0 0 x0

j=1

⎡ ∗ ⎣ ≥ δ 1 · ϕ1 (x) ·

x

x0

⎞ ⎤((λα +λv )/p) ⎛ M p v Dx fj (w) ⎠ dw⎦ p (w) ⎝ , (10.59) 0 j=1

all x0 ≤ x ≤ b. Proof. By Theorem 10.5 we have x + λα λv q (w) Dxγ01 fj (w) Dxv0 fj (w) x0

λα λv , dw + Dxγ01 fj+1 (w) Dxv0 fj+1 (w) )

x

≥ δ 1 ϕ1 (x) x0

p p

p (w) Dxv0 fj (w) + Dxv0 fj+1 (w) dw

(10.60) *((λα +λv )/p) ,

j = 1, 2, . . . , M − 1. Hence by adding all the above we derive ⎛ x M −1 + γ Dx 1 fj (w)λα Dxv fj (w)λv q (w) ⎝ 0 0 x0

⎞ λα v λv , γ ⎠ dw + Dx01 fj+1 (w) Dx0 fj+1 (w) j=1

(10.61)

222

10. Converse Canavati Fractional Inequalities

⎛ ≥δ 1 ϕ1 (x) · ⎝

M −1 ) x x0

j=1

p p (w) Dxv0 fj (w)

⎞ p (λα +λv /p) v ⎠. + Dx0 fj+1 (w) dw Also it holds x + λα λv q (w) Dxγ01 f1 (w) Dxv0 f1 (w) x0 λα λv , dw (10.62) + Dxγ01 fM (w) Dxv0 fM (w) *((λα +λv )/p) ) x p p

≥ δ 1 ϕ1 (x) p (w) Dxv0 f1 (w) + Dxv0 fM (w) dw . x0

Adding (10.61) and (10.62) , and using (10.53) we have ⎞ ⎛ x M λα v λv γ Dx 1 fj (w) Dx fj (w) ⎠ dw q (w) ⎝ 2 0 0 x0

j=1

≥ δ 1 ϕ1 (x)

⎧⎧ −1 ) ⎨⎨M ⎩⎩

j=1

x

x0

p p (w) Dxv0 fj (w)

(10.63)

p (λα +λv /p) + Dxv0 fj+1 (w) dw

x

+ x0

p p

p (w) Dxv0 f1 (w) + Dxv0 fM (w) dw

M 1−(λα +λv /p) δ 1 ϕ1 (x)

⎧ ⎨ ⎩

((λα +λv )/p) ≥

⎤ ⎫((λα +λv )/p) M ⎬ p v Dx fj (w) ⎦ dw p (w) ⎣2 . 0 ⎭ ⎡

x

x0

j=1

(10.64) We have proved ⎞ ⎛ x M λα v λv γ Dx 1 fj (w) Dx fj (w) ⎠ dw ≥ q (w) ⎝ 0 0 x0

j=1

M 1−(λα +λv /p) δ 1 2(λα +λv /p)−1 ϕ1 (x) ⎧ ⎤ ⎫((λα +λv )/p) ⎡ M ⎨ x ⎬ p v Dx fj (w) ⎦ dw · p (w) ⎣ , 0 ⎩ x0 ⎭ j=1

thus proving (10.59).

(10.65)

10.3 Main Results

223

Next we give Theorem 10.12. All here are as in Theorem 10.11. Assume λβ ≥ λv . Denote λv /p λv λ /p (p−λv )/p ϕ2 (x) := (A0 (x) |λα =0 ) 2 δ3 v . (10.66) λβ + λv If λα = 0, then

x

x0

⎧⎧ −1 + ⎨⎨M γ Dx 2 fj+1 (w)λβ Dxv fj (w)λv q (w) 0 0 ⎩⎩ j=1

λ λv , + Dxγ02 fj (w) β Dxv0 fj+1 (w) + λ λv + Dxγ02 fM (w) β Dxv0 f1 (w) λ λv , dw ≥ + Dxγ02 f1 (w) β Dxv0 fM (w)

x

M 1−((λv +λβ )/p) 2((λv +λβ )/p) ϕ2 (x) ·

p (w) x0

⎡ ⎤ ⎫((λv +λβ )/p) M ⎬ p v Dx fj (w) ⎦ dw ⎣ , 0 ⎭

x ≥ x0 .

(10.67)

j=1

Proof. From Theorem 10.6 we obtain x + λ λv q (w) Dxγ02 fj+1 (w) β Dxv0 fj (w) x0

λ λv , dw + Dxγ02 fj (w) β Dxv0 fj+1 (w)

x

≥ ϕ2 (x) x0

p p

p (w) Dxv0 fj (w) + Dxv0 fj+1 (w) dw

((λv +λβ )/p) , (10.68)

for j = 1, . . . , M − 1. Hence by adding all of the above we ﬁnd ⎛ x M −1 + γ Dx 2 fj+1 (w)λβ Dxv fj (w)λv q (w) ⎝ 0 0 x0

j=1

λ λv , + Dxγ02 fj (w) β Dxv0 fj+1 (w) dw

224

10. Converse Canavati Fractional Inequalities

≥ ϕ2 (x)

⎧ −1 ⎨M ⎩

j=1

x

x0

p p (w) Dxv0 fj (w)

p ((λv +λβ )/p) . + Dxv0 fj+1 (w) dw Similarly it holds x x0

(10.69)

+ λ λv q (w) Dxγ02 fM (w) β Dxv0 f1 (w)

λ λv , + Dxγ02 f1 (w) β Dxv0 fM (w) dw

x

≥ ϕ2 (x) x0

p p

p (w) Dxv0 f1 (w) + Dxv0 fM (w) dw

((λv +λβ )/p) . (10.70)

Adding (10.69), (10.70), and using (10.53) we derive (10.67) .

We continue with Theorem 10.13. Let v ≥ 2 and γ 1 ≥ 0 such that 2 ≤ v − γ 1 < 1/p, (i) 0 < p < 1. Let fj ∈ Cxv0 ([a, b]) with fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [v] , j = 1, . . . , M ∈ N. Here x, x0 ∈ [a, b] : x ≥ x0 . Assume that Dxv0 fj is of ﬁxed sign on [x0 , b] , j = 1, . . . , M. Consider also p (t) > 0, and q (t) > 0 continuous functions on [x0 , b] . Let λα ≥ λα+1 > 1; Φ is as in Theorem 10.7. Then ⎧⎧ x −1 + ⎨⎨M γ Dx 1 fj (w)λα Dxγ 1 +1 fj+1 (w)λα+1 q (w) 0 0 ⎩⎩ x0 j=1

λα λα+1 , + Dxγ01 fj+1 (w) Dxγ01 +1 fj (w) + λα λα+1 + Dxγ01 f1 (w) Dxγ01 +1 fM (w) λα λα+1 , dw ≥ + Dxγ01 fM (w) Dxγ01 +1 f1 (w) M 1−((λα +λα+1 )/p) 2((λα +λα+1 )/p) Φ (x) ⎞ ⎤((λα +λα+1 )/p) ⎛ ⎡ x M p v Dx fj (w) ⎠ dw⎦ ⎣ p (w) ⎝ , x0

all x0 ≤ x ≤ b.

0

j=1

(10.71)

10.3 Main Results

225

Proof. From Theorem 10.7 we obtain

x

q (w) x0

M −1 +

γ Dx 1 fj (w)λα Dxγ 1 +1 fj+1 (w)λα+1 0

0

j=1

λα λα+1 , dw + Dxγ01 fj+1 (w) Dxγ01 +1 fj (w)

≥ Φ (x)

M −1 ) x x0

j=1

p p p (w) Dxv 0 fj (w) + Dxv 0 fj+1 (w) dw

*(λα +λα+1 /p) , (10.72)

all x0 ≤ x ≤ b. Also it holds

x

x0

+ λα λα+1 q (w) Dxγ01 f1 (w) Dxγ01 +1 fM (w)

λα λα+1 , dw + Dxγ01 fM (w) Dxγ01 +1 f1 (w) )

x

≥ Φ (x) x0

p p p (w) Dxv0 f1 (w) + Dxv0 fM (w) dw

*((λα +λα+1 )/p) ,

(10.73) all x0 ≤ x ≤ b. Adding (10.72) and (10.73) , along with (10.53) we derive (10.71). Next comes the following theorem. Theorem 10.14. All here are as in Theorem 10.11. Consider the special case of λβ = λα + λv . Here T˜ (x) is as in (10.44) . Assume here for j = 1, . . . , M that x p p (t) Dxv0 fj (t) dt ∈ [H, Ψ] , 0 < H < Ψ. zj (x) := x0

Then

x

x0

⎧⎧ −1 + ⎨⎨M γ Dx 1 fj (w)λα Dxγ 2 fj+1 (w)λα +λv Dxv fj (w)λv q (w) 0 0 0 ⎩⎩ j=1

λα +λv γ , Dx 1 fj+1 (w)λα Dxv fj+1 (w)λv + Dxγ02 fj (w) 0 0

226

10. Converse Canavati Fractional Inequalities

+ λα λα +λv v Dx f1 (w)λv + Dxγ01 f1 (w) Dxγ02 fM (w) 0

λα +λv γ , Dx 1 fM (w)λα Dxv fM (w)λv dw ≥ + Dxγ02 f1 (w) 0 0 (10.74)

M (1−2(λα +λv )/p) 22(λα +λv /p) T˜ (x) ⎡ ⎞ ⎤(2(λα +λv )/p) ⎛ x M p v Dx fj (w) ⎠ dw⎦ ×⎣ p (w) ⎝ , 0

x0

j=1

all x0 ≤ x ≤ b. Proof. Based on Theorem 10.8. The rest is as in the proof of Theorem 10.13. We continue with Corollary 10.15. (to Theorem 10.11, λβ = 0, p (t) = q (t) = 1). Then ⎛ ⎞ x M λα v λv γ Dx 1 fj (w) Dx fj (w) ⎠ dw ⎝ x0

0

0

j=1

⎡ ≥ δ ∗1 ϕ1 (x) ⎣

x

x0

⎡ ⎤ ⎤((λα +λv )/p) M p v Dx fj (w) ⎦ dw⎦ ⎣ , 0

(10.75)

j=1

all x0 ≤ x ≤ b. In (10.75) , A0 (x) |λβ =0 of ϕ1 (x) is given by (10.49). Proof. Based on Theorem 10.11.

Corollary 10.16. (to Theorem 10.12, λα = 0, p (t) = q (t) = 1). It holds ⎧⎧ x ⎨⎨M −1 + γ Dx 2 fj+1 (w)λβ Dxv fj (w)λv 0 0 ⎩ ⎩ x0 j=1

λ λv , + Dxγ02 fj (w) β Dxv0 fj+1 (w) + λ λv + Dxγ02 fM (w) β Dxv0 f1 (w) λ λv , + Dxγ02 f1 (w) β Dxv0 fM (w) dw ≥

10.3 Main Results M

1−(λv +λβ /p)

((λv +λβ )/p)

2

⎧ ⎨ ϕ2 (x)

⎩

227

⎤ ⎫((λv +λβ )/p) M ⎬ p v Dx fj (w) ⎦ dw ⎣ , 0 ⎭ ⎡

x x0

j=1

(10.76)

all x0 ≤ x ≤ b. In (10.76) , (A0 (x) |λα =0 ) of ϕ2 (x) is given by (10.52) . Proof. Based on Theorem 10.12.

11 Converse Riemann–Liouville Fractional Opial-Type Inequalities for Several Functions

A collection of very general Lp (0 < p < 1)-form converse Opial-type inequalities [315] is presented involving Riemann – Liouville fractional derivatives [17, 230, 295, 314] of several functions in diﬀerent orders and powers. Other particular results of special interest are derived from the established results. This treatment is based on [47].

11.1 Introduction Opial inequalities appeared for the ﬁrst time in [315] and since then many authors dealt with them in diﬀerent directions and for various cases. For a complete account of the recent activity in this ﬁeld see [4], and it still remains a very active area of research. One of the main attractions to these inequalities is their applications, especially in proving the uniqueness and upper bounds of the solution of initial value problems in diﬀerential equations. The author was the ﬁrst to present Opial inequalities involving fractional derivatives of functions in [15, 17] with applications to fractional diﬀerential equations. See also [64, 65]. Fractional derivatives come up naturally in a number of ﬁelds, especially in physics, for example; see the recent books [197, 333]. Here the author continues his study of Riemann – Liouville fractional Opial-type inequalities now involving several diﬀerent functions and produces a wide variety of reverse results. To give the reader an idea of the kind of converse inequalities we are dealing with, we brieﬂy mention a speciﬁc one (see Corollary 11.15). G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 229 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 11,

230

11. Converse Riemann–Liouville Fractional Inequalities ⎛

x 0

⎞ M λβ + λα β α1 D fj (s) 1 D fj (s) ⎠ ds ≥ M 1−((λα1 +λβ )/p) ⎝ j=1

λβ λα1 + λβ

λβ /p

1

(Γ (β − α1 ))λα1

p − λβ λα1 βp − λα1 α1 p − λα1 + p − λβ

(p−λβ /p)

p−1 βp − α1 p − 1

λα

1

(p−1/p)

x((λα1 βp−λα1 α1 p−λα1 +p−λβ )/p)

,

⎤((λα +λβ )/p) 1 M p β ⎣ , D fj (s) ds⎦ ⎡

x 0

j=1

where α1 ≥ 0, 0 < β − α1 < 1/p, 0 p. All fractional derivatives involved here are of Riemann – Liouville type. The L∞ fractional derivatives Dβ fj in [0, x] are each of ﬁxed sign a.e. on [0, x] , and Dβ−k fj (0) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M. Here [·] denotes the integral part of the number; Γ stands for the gamma function.

11.2 Background We need Deﬁnition 11.1. (see [187, 295, 314]). Let α ∈ R+ − {0} . For any f ∈ L1 (0, x) ; x ∈ R+ − {0} , the Riemann – Liouville fractional integral of f of order α is deﬁned by s 1 α−1 (s − t) f (t) dt, ∀s ∈ [0, x] , (11.1) (Jα f ) (s) := Γ (α) 0 and the Riemann – Liouville fractional derivative of f of order α by m s d 1 m−α−1 α D f (s) := (s − t) f (t) dt, (11.2) Γ (m − a) ds 0 where m := [α] + 1, [·] is the integral part. In addition, we set D0 f := f := J0 f, J−α f = Dα f if α > 0, D−α f := Jα f, if 0 < α ≤ 1. If α ∈ N, then Dα f = f (α) the ordinary derivative. Deﬁnition 11.2. [187]. We say that f ∈ L1 (0, x) has an L∞ fractional derivative Dα f in [0, x] , x ∈ R+ − {0} , iﬀ Dα−k f ∈ C ([0, x]) , k = 1, . . . , m := [α] + 1; α ∈ R+ − {0} , and Dα−1 f ∈ AC ([0, x]) (absolutely continuous functions) and Dα f ∈ L∞ (o, x) .

11.3 Main Results

231

We need Lemma 11.3. [187]. Let α ∈ R+ , β > α, let f ∈ L1 (o, x) , x ∈ R+ −{0} , have an L∞ fractional derivative Dβ f in [0, x] , and let Dβ−k f (0) = 0 for k = 1, . . . , [β] + 1. Then s 1 β−α−1 (s − t) Dβ f (t) dt, ∀s ∈ [0, x] . (11.3) Dα f (s) = Γ (β − α) 0 Here Dα f ∈ AC ([0, x]) for β − α ≥ 1 and Dα f ∈ C ([0, x]) for β − α ∈ (0, 1) , hence Dα f ∈ L∞ (o, x) and Dα f ∈ L1 (o, x) .

11.3 Main Results 11.3.1 Results Involving Two Functions We present our ﬁrst main result. Theorem 11.4. Let αi ∈ R+ , 0 < β − αi < 1/p, 0 < p < 1, i = 1, 2, and let f1 , f2 ∈ L1 (0, x) , x ∈ R+ − {0} have, respectively, L∞ fractional derivatives Dβ f1 , Dβ f2 in [0, x] , each of ﬁxed sign a.e. on [0, x] , and let Dβ−k fi (0) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Consider also p (t) > 0 and q (t) ≥ 0, with all p (t) , 1/p (t), q (t) , 1/q (t) ∈ L∞ (0, x) . Let λβ > 0 and λα1 , λα2 ≥ 0, such that λβ > p. Call s p(β−αi −1)/p−1 −1/(p−1) Pi (s) := (s − t) (p (t)) dt, i = 1, 2; 0 ≤ s ≤ x, 0

A (s) :=

(11.4)

q (s) (P1 (s))

λα1 ((p−1)/p)

(Γ (β − α1 )) A0 (x) :=

λα2 ((p−1)/p)

(P2 (s)) λα1

x

(A (s))

(Γ (β − α2 )) p/(p−λβ )

−λβ /p

(p (s))

λα2

(11.5)

(p−λβ )/p ds

(11.6)

δ 1 := 21−((λα1 +λβ )/p) .

(11.7)

0

and

If λα2 = 0, we obtain that 0

x

+ λ β λβ , λ λ ds ≥ q (s) |Dα1 f1 (s)| α1 Dβ f1 (s) + |Dα1 f2 (s)| α1 Dβ f2 (s)

232

11. Converse Riemann–Liouville Fractional Inequalities

A0 (x) |λα2 =0

λβ λα1 + λβ

λβ /p

)

x

δ1

+ p p (s) Dβ f1 (s) +

0

β , ,((λα1 +λβ )/p) D f2 (s)p ds .

(11.8)

Proof. From (11.3) and assumption we have s 1 β−αi −1 β D fj (t) dt, |Dαi fj (s)| = (s − t) Γ (β − αi ) 0

(11.9)

∀s ∈ [0, x] , i = 1, 2; j = 1, 2. Next, applying H¨ older’s inequality with indices p, p/ (p − 1) , we ﬁnd s 1 β−αi −1 |Dαi fj (s)| = (s − t) Γ (β − αi ) 0 −1/p 1/p (11.10) (p (t)) (p (t)) Dβ fj (t) dt

≥

1 Γ (β − αi ) ×

s

s

β−αi −1

(s − t)

(p (t))

−1/p

(p−1)/p

p/p−1 dt

(11.11)

0

p p (t) Dβ fj (t) dt

1/p

1 p−1/p (Pi (s)) Γ (β − αi ) 1/p s β p p (t) D fj (t) dt . (11.12)

0

=

0

That is, it holds 1 p−1/p (Pi (s)) |D fj (s)| ≥ Γ (β − αi )

αi

s

p p (t) Dβ fj (t) dt

1/p .

0

(11.13)

Put zj (s) :=

s

p p (t) Dβ fj (t) dt,

(11.14)

0

thus, p zj (s) = p (s) Dβ fj (s) , a.e. in (0, x) , zj (0) = 0; j = 1, 2.

(11.15)

Hence, we have |Dαi fj (s)| ≥

1 p−1/p 1/p (Pi (s)) (zj (s)) , Γ (β − αi )

(11.16)

11.3 Main Results

and

λβ /p β D fj (s)λβ = (p (s))−λβ /p zj (s) ,

233

(11.17)

a.e. in (o, x) , i = 1, 2; j = 1, 2. Therefore we obtain λα1

q (s) |(Dα1 f1 ) (s)|

≥ q (s)

1 (Γ (β − α1 ))

1

|(Dα2 f2 ) (s)|

(P1 (s))

λα1

λα2 (p−1/p)

(P2 (s))

λα2

(Γ (β − α2 ))

λα2

|(Dβ f1 ) (s)|

λα1 (p−1/p)

λα2 /p

(z2 (s))

λβ

λα1 /p

(z1 (s))

−λβ /p

(11.18)

(p (s))

z1 (s)

λβ /p (11.19)

= A (s) (z1 (s))

λα1 /p

λα2 /p

(z2 (s))

z1 (s)

λβ /p

, a.e. in (0, x) .

(11.20)

Consequently, by another H¨ older’s inequality application, we get (by (p/λβ ) < 1) x λ λ λ q (s) |(Dα1 f1 ) (s)| α1 |(Dα2 f2 ) (s)| α2 |(Dβ f1 ) (s)| β ds 0

≥ A0 (x)

x

(z1 (s))

λα1 /λβ

λα2 /λβ

(z2 (s))

λβ /p

z1 (s) ds

. (11.21)

0

Similarly, one ﬁnds x λ λ λ q (s) |(Dα1 f2 ) (s)| α1 |(Dα2 f1 ) (s)| α2 |(Dβ f2 ) (s)| β ds 0

≥ A0 (x)

x

(z1 (s))

λα2 /λβ

λα1 /λβ

(z2 (s))

λβ /p

z2 (s) ds

.

(11.22)

0

Taking λα2 = 0 and adding (11.21) and (11.22) , we derive

x

, + q (s) |(Dα1 f1 ) (s)|λα1 |(Dβ f1 ) (s)|λβ + |(Dα1 f2 ) (s)|λα1 |(Dβ f2 ) (s)|λβ ds

0

≥ A0 (x) |λα2 =0

'

x

λα1 /λβ

(z1 (s)) 0

λβ /p + z1 (s) ds

234

11. Converse Riemann–Liouville Fractional Inequalities

x

λα1 /λβ

(z2 (s))

λβ /p (

=

z2 (s) ds

(11.23)

0

+ λ +λ /p A0 (x) |λα2 =0 (z1 (x))( α1 β ) + λ +λ /p (z2 (x))( α1 β )

A0 (x) |λα2 =0

x

+

λβ λα1 + λβ

,

λβ /p

λβ λα1 + λβ

⎡ λβ /p ⎢ ⎣

x

=

p p (t) Dβ f1 (t) dt

(11.24)

(λα1 +λβ /p)

0

p p (t) Dβ f2 (t) dt

(λα1 +λβ /p)

0

⎤ ⎥ ⎦ =: (∗) .

(11.25)

In this study, we frequently use the basic inequalities r

2r−1 (ar + br ) ≤ (a + b) ≤ ar + br , a, b ≥ 0, 0 ≤ r ≤ 1,

(11.26)

and r

ar + br ≤ (a + b) ≤ 2r−1 (ar + br ) , a, b ≥ 0, r ≥ 1.

(11.27)

At last using (11.7) , (11.26), and (11.27) , we get (∗) ≥ A0 (x) |λα2 =0 )

x

λβ λα1 + λβ

λβ /p δ1

*(λα1 +λβ /p) + p p , . p (t) Dβ f1 (t) + Dβ f2 (t) dt

(11.28)

0

Inequality (11.8) has been proved. We remark the following. Here we see that (p/p − 1) (β − αi − 1) + 1 > 0, −1/(p − 1) > 0 and p (t) ∈ L∞ (0, x) ; thus (see (11.4)) Pi (s) ∈ R, for every s ∈ [0, x] ; also Pi (s) is continuous and bounded on [0, x] . By λβ > p > 0, we have 0 < p/λβ < 1, p/p − λβ < 0. We observe that 1 1 λ λ λ /p = (Γ (β − α1 )) α1 (Γ (β − α2 )) α2 (p (s)) β A (s) q (s) λα1 (1−p/p)

(P1 (s))

(P2 (s))

λα2 (1−p/p)

∈ L∞ (0, x) ,

11.3 Main Results

235

and 1/A (s) > 0, a.e. on [0, x] . Therefore 0 < A0 (x) < ∞, and all we have done in this proof is valid.

The counterpart of the last theorem follows. Theorem 11.5. All here are as in Theorem 11.4. Further assume λα2 ≥ λβ . Denote (11.29) δ 2 := 21−(λα2 /λβ ) and δ 3 := (δ 2 − 1) 2−λα2 /λβ

(11.30)

If λα1 = 0, then

x

) λβ λβ * q (s) |(Dα2 f2 ) (s)|λα2 Dβ f1 (s) + |(Dα2 f1 ) (s)|λα2 Dβ f2 (s) ds

0

λβ /p λβ λ /p δ3 β λα2 + λβ (λβ +λα2 )/p x + p p , · p (s) Dβ f1 (s) + Dβ f2 (s) ds . ≥ A0 (x) |λα1 =0 2(p−λβ )/p

(11.31)

0

Proof. When λα1 = 0 from (11.21) and (11.22) we ﬁnd

x

+ λβ λβ , λ λ ds ≥ q (s) |Dα2 f2 (s)| α2 Dβ f1 (s) + |Dα2 f1 (s)| α2 Dβ f2 (s)

0

(11.32) A0 (x) |λα1 =0

'

x

λα2 /λβ

(z2 (s))

λβ /p

z1 (s) ds

+

0

x

(z1 (s))

λα2 /λβ

( λβ /p z2 (s) ds ≥

0

λ /p A0 (x) |λα1 =0 21−λβ /p (M (x)) β := (∗) ,

(11.33)

by λβ /p > 1 and by (11.27) , where

x

+ (z2 (s))

M (x) := 0

We observe that

λα2 /λβ

z1 (s) + (z1 (s))

λα2 /λβ

, z2 (s) ds.

(11.34)

236

11. Converse Riemann–Liouville Fractional Inequalities

x

M (x) =

(z1 (s))

λα2 /λβ

+ (z2 (s))

λα2 /λβ

z1 (s) + z2 (s) ds

0

x

−

+ (z1 (s))

λα2 /λβ

λα2 /λβ

z1 (s) + (z2 (s))

, z2 (s) ds

(11.35)

0

(by (11.27))

≥

x

δ2

(z1 (s) + z2 (s))

λα2 /λβ

(z1 (s) + z2 (s)) ds

0

−

λβ λα2 + λβ

+

(z1 (x))(

(λα2 +λβ )/λβ )

(λ +λ )/λ + (z2 (x))( α2 β β )

λα2 +λβ /λβ )

= δ 2 (z1 (x) + z2 (x))( −

λβ λα2 + λβ

+

(z1 (x))(

(λα2 +λβ )/λβ )

= +

λβ λ α2 + λ β

(11.36)

(λ +λ )/λ + (z2 (x))( α2 β β )

λβ λα2 + λβ

,

, (11.37)

+ ,, δ 2 (z1 (x) + z2 (x))(λα2 +λβ /λβ ) − (z1 (x))(λα2 +λβ /λβ ) + (z2 (x))(λα2 +λβ /λβ )

11.38

(11.27)

≥

λβ λα2 + λβ

+

λ +λ /λ λ +λ /λ δ 2 (z1 (x))( α2 β β ) + (z2 (x))( α2 β β )

,, + λ +λ /λ λ +λ /λ − (z1 (x))( α2 β β ) + (z2 (x))( α2 β ) β =

λβ λα2 + λβ

(11.39)

+ , λ +λ /λ λ +λ /λ (δ 2 − 1) (z1 (x))( α2 β β ) + (z2 (x))( α2 β β ) (11.40)

(11.27)

≥

λβ λα2 + λβ

That is, we get that

λ +λ /λ δ 3 (z1 (x) + z2 (x))( α2 β β ) .

(11.41)

11.3 Main Results

M (x) ≥

λβ λα2 + λβ

λ +λ /λ δ 3 (z1 (x) + z2 (x))( α2 β β ) .

237

(11.42)

Therefore, by (11.32) , (11.33), and (11.42) , we derive (∗) ≥ A0 (x) |λα1 =0 2(p−λβ )/p

= A0 (x) |λα1 =0 2(p−λβ )/p x

λβ λα2 + λβ

λβ /p

λ +λ /p (z1 (x) + z2 (x))( α2 β )

λ /p

δ3 β

)

λβ λα2 + λβ

λβ /p

(11.43)

λ /p

δ3 β

*(λα2 +λβ /p) + p p , . p (s) Dβ f1 (s) + Dβ f2 (s) ds

(11.44)

0

We have established (11.31) . We need Theorem 11.6. (see [48]) Let 0 ≤ s ≤ x and f ∈ L∞ ([0, x]) , r > 0. Deﬁne s r F (s) := (s − t) f (t) dt. (11.45) 0

Then there exists

s

r−1

(s − t)

F (s) = r

f (t) dt, all s ∈ [0, x]

(11.46)

0

We proceed with a special important case. Theorem 11.7. Let 1 < β − α1 < 1/p, 0 < p < 1, α1 ∈ R+ and let f1 , f2 ∈ L1 (0, x) , x ∈ R+ − {0} have, respectively, L∞ fractional derivatives Dβ f1 , Dβ f2 in [0, x] , each of ﬁxed sign a.e. in [0, x] , and let Dβ−k fi (0) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Consider also p (t) > 0 and q (t) ≥ 0, with p (t) , 1/p (t), q (t) , 1/q (t) ∈ L∞ (0, x) . Let λα ≥ λα+1 > 1. Denote θ3 := 21−(λα /λα+1 ) − 1 2−λα /λα+1 (11.47) L (x) := 2

1−λα+1

x

(q (s)) 0

(1/(1−λα+1 ))

ds

θ3 λα+1 λα + λα+1

λα+1 . (11.48)

238

11. Converse Riemann–Liouville Fractional Inequalities

and

x

(β−α1 −1)p/(p−1)

(x − s)

P1 (x) :=

(p (s))

−1/(p−1)

ds,

(11.49)

0

T (x) := L (x)

((p−1)/p)

P1 (x) Γ (β − α1 )

(λα +λα+1 ) (11.50)

and ω 1 := 2(p−1/p)(λα +λα+1 ) ,

(11.51)

Φ (x) := T (x) ω 1 .

(11.52)

Then

x

, + λ λ q (s) |Dα1 f1 (s)|λα Dα1 +1 f2 (s) α+1 + |Dα1 f2 (s)|λα Dα1 +1 f1 (s) α+1 ds

0

) ≥ Φ (x)

x

*(λα +λα+1 )/p p p p (s) Dβ f1 (s) + Dβ f2 (s) ds .

(11.53)

0

Proof. For convenience, we set α2 := α1 + 1.

(11.54)

By Lemma 11.3 and assumption we have

|Dα1 fi (s)| =

1 Γ (β − α1 )

s

(s − t)

β−α1 −1

β D fi (t) dt,

(11.55)

0

∀s ∈ [0, x] , i = 1, 2, and |Dα2 fi (s)| = Dα1 +1 fi (s) = s 1 β−α1 −2 β D fi (t) dt, (s − t) Γ (β − α1 − 1) 0

(11.56)

∀s ∈ [0, x] , i = 1, 2. By Theorem 11.6 we obtain that

(Dα1 fi (s)) = Dα1 +1 fi (s) = Dα2 fi (s) , ∀s ∈ [0, x] , i = 1, 2.

(11.57)

11.3 Main Results

239

So we have 1 |D fj (s)| = Γ (β − αi )

s

β−αi −1

(s − t)

αi

β D fj (t) dt =: gj,α (s) , i

0

(11.58) where j = 1, 2, i = 1, 2, 0 ≤ s ≤ x. We also have by Theorem 11.6 that

(gj,α1 (s)) = gj,α2 (s) ; gj,αi (0) = 0.

(11.59)

Notice that, if β − α2 = 1, then

β D fj (t) dt.

s

gj,α2 (s) =

(11.60)

0

Next, we apply H¨ older’s inequality with indices 1/λα+1 < 1, 1/(1 − λα+1 ) < 0 to obtain x λα+1 λ q (s) |Dα1 f1 (s)| α Dα1 +1 f2 (s) ds = (11.61) 0

x

q (s) (g1,α1 (s))

λα

(g2,α1 (s))

λα+1

ds ≥

0

(1−λα+1 )

x

(1/1−λα+1 )

(q (s))

ds

0

x

λα /λα+1

(g1,α1 (s))

λα+1

(g2,α1 (s)) ds

.

(11.62)

0

Similarly, we ﬁnd x q (s) |Dα1 f2 (s)| 0

λα

α +1 D 1 f1 (s)λα+1 ds ≥ (1−λα+1 )

x

(1/1−λα+1 )

(q (s))

ds

0

x

(g2,α1 (s))

λα /λα+1

(g1,α1 (s)) ds

λα+1 .

(11.63)

0

Adding the last two inequalities (11.62), (11.63), we derive

x 0

, + λ λ q (s) |Dα1 f1 (s)|λα Dα1 +1 f2 (s) α+1 + |Dα1 f2 (s)|λα Dα1 +1 f1 (s) α+1 ds

240

11. Converse Riemann–Liouville Fractional Inequalities

x

≥

(q (s))(1/1−λα+1 ) ds

⎡ (1−λα+1 ) ⎣

0

x 0

x

λα /λα+1

(g2,α1 (s))

λα+1

(g1,α1 (s))λα /λα+1 (g2,α1 (s)) ds

+

λα+1 (

(11.64)

(g1,α1 (s)) ds

0 (11.27)

≥

2

(1−λα+1 )

x

(q (s))

(1/1−λα+1 )

×

ds

0

)

x

+ , *λα+1 λα /λα+1 λα /λα+1 (g1,α1 (s)) (g2,α1 (s)) + (g2,α1 (s)) (g1,α1 (s)) ds

0

(11.65) (notice (11.34) and (11.42) , so similarly we have) ≥ 2

(1−λα+1 )

x

(1/1−λα+1 )

(q (s))

ds

0

λα+1 θ3 λα + λα+1

λα+1 (g1,α1 (x) + g2,α1 (x))

(λα +λα+1 )

(11.66)

(λα +λα+1 )

= L (x) (g1,α1 (x) + g2,α1 (x)) = )

x

L (x) (Γ (β − α1 ))

(λα +λα+1 )

(11.67)

×

, *(λα +λα+1 ) + (x − t)β−α1 −1 (p (t))−1/p (p (t))1/p Dβ f1 (t) + Dβ f2 (t) dt

0

(11.68)

(applying H¨ older’s inequality with indices p/(p − 1) and p, we ﬁnd) ≥

L (x) (Γ (β − α1 ))(λα +λα+1 )

)

x

(x − t)

(β−α1 −1)p/(p−1)

(p (t))

−1/(p−1)

* ((p−1)/p)(λα +λα+1 ) dt

0

(11.69) )

x

p p (t) Dβ f1 (t) + Dβ f2 (t) dt

*((λα +λα+1 )/p) =

0

L (x) (P1 (x))(p−1/p)(λα +λα+1 ) (Γ (β − α ))(λα +λα+1 ) 1

)

x 0

p * ((λα +λα+1 )/p) p (t) Dβ f1 (t) + Dβ f2 (t) dt

(11.70)

11.3 Main Results

)

x

= T (x)

p p (t) Dβ f1 (t) + Dβ f2 (t) dt

241

*(λα +λα+1 /p)

0 (11.26)

≥

T (x) 2(p−1/p)(λα +λα+1 )

)

x 0

)

x

= Φ (x)

p p * (λα +λα+1 /p) p (t) Dβ f1 (t) + Dβ f2 (t) dt

(11.71)

*(λα +λα+1 )/p p β p β dt p (t) D f1 (t) + D f2 (t) .

(11.72)

0

We have proved (11.53) . We continue with

Theorem 11.8. All here are as in Theorem 11.4. Consider the special case of λα2 = λα1 + λβ . Assume here for j = 1, 2 that x p zj (x) := p (t) Dβ fj (t) dt ∈ [H, Ψ] , 0 < H < Ψ, (11.73) 0

along with h :=

Ψ > 1, H

Mh (1) =

(h − 1) h1/h−1 . e ln h

(11.74)

Denote T˜ (x) := A0 (x)

λβ λα1 + λβ

λβ /p

−2 λ +λ /p 2(p−2λα1 −3λβ )/p (Mh (1)) ( α1 β ) .

(11.75) Then

x

+ λβ λ λ +λ q (s) |Dα1 f1 (s)| α1 |Dα2 f2 (s)| α1 β Dβ f1 (s) +

0

|Dα2 f1 (s)| T˜ (x)

x

λα1 +λβ

λα1

|Dα1 f2 (s)|

, β D f2 (s)λβ ds ≥

2(λα1 +λβ )/p p β p β ds p (s) D f1 (s) + D f2 (s) .

(11.76)

0

Proof. We apply (11.21) and (11.22) for λα2 = λα1 + λβ , and by adding we obtain

242

11. Converse Riemann–Liouville Fractional Inequalities

+ λβ λ λ +λ q (s) |Dα1 f1 (s)| α1 |Dα2 f2 (s)| α1 β Dβ f1 (s) +

x

0 λα1

|Dα1 f2 (s)| '

|Dα2 f1 (s)|

x

≥ A0 (x)

(z1 (s))

λα1 /λβ

λα1 +λβ

, β D f2 (s)λβ ds

λα1 +λβ /λβ

(z2 (s))

λβ /p

z1 (s) ds

+

0

x

λα /λβ 1 λ +λ /λ (z1 (s))( α1 β ) β (z2 (s)) z2 (s) ds

λβ /p ( (11.77)

0

)

(11.27)

≥

×

1−(λβ /p)

A0 (x) 2

x

(z1 (s))

λα1 /λβ

λα1 /λβ +1

(z2 (s))

z1 (s) + =

0 λ /λ +1 λ /λ (z1 (s))( α1 β ) (z2 (s)) α1 β z2 (s) ds

) A0 (x) 21−(λβ /p)

x

,λβ /p =

(11.78)

λα /λβ 1

(z1 (s) z2 (s)) 0

+ , ,λβ /p z2 (s) z1 (s) + z1 (s) z2 (s) ds = )

x

(p−λβ )/p

A0 (x) 2

(11.79)

(λα1 /λβ ) (z1 (s) z2 (s)) (z1 (s) z2 (s)) ds

*λβ /p

0

(11.80)

λ +λ /p = A0 (x) 2(p−λβ /p) (z1 (x) z2 (x))( α1 β )

λβ λα1 + λβ

λβ /p (11.81)

(see [377]) ≥ A0 (x) 2

(p−λβ /p)

λβ λα1 + λβ

λβ /p

z1 (x) + z2 (x) 2Mh (1)

2(λα1 +λβ )/p (11.82)

2 λ +λ /p = T˜ (x) (z1 (x) + z2 (x)) ( α1 β ) =

(11.83)

11.3 Main Results

)

*2(λα1 +λβ )/p p p p (s) Dβ f1 (s) + Dβ f2 (s) ds .

x

T˜ (x)

243

(11.84)

0

We have proved (11.76) .

Next are special cases of the above theorems. Corollary 11.9. (to Theorem 11.4) Set λα2 = 0, p (t) = q (t) = 1. Then

x

+

λα1

|Dα1 f1 (s)|

, β D f1 (s)λβ + |Dα1 f2 (s)|λα1 Dβ f2 (s)λβ ds ≥

0

)

x

C1 (x)

*(λα1 +λβ /p) + , Dβ f1 (s)p + Dβ f2 (s)p ds ,

0

where C1 (x) := A0 (x) |λα2 =0

λβ λα1 + λβ

λβ /p δ1 ,

(11.85)

with δ 1 = 21−((λα1 +λβ )/p) .

(11.86)

We have that A0 (x) |λα2 =0 =

%

1 λα1

(Γ (β − α1 ))

((λα1 p−λα1 )/p) p−1 βp − α1 p − 1 (11.87)

p − λβ λα1 βp − λα1 α1 p − λα1 + p − λβ

((p−λβ )/p)

×

x((λα1 βp−λα1 α1 p−λα1 +p−λβ )/p) . Proof. By Theorem 11.4. The constant A0 (x) |λα2 =0 was calculated in [48]. We continue with Corollary 11.10. (to Theorem 11.5, λα1 = 0, p (t) = q (t) = 1, λα2 ≥ λβ ). Then

244

11. Converse Riemann–Liouville Fractional Inequalities

x

+

|Dα2 f2 (s)|

λα2

, β D f1 (s)λβ + |Dα2 f1 (s)|λα2 Dβ f2 (s)λβ ds ≥

0

(11.88)

x

C2 (x)

(λβ +λα2 )/p + , Dβ f1 (s)p + Dβ f2 (s)p ds .

0

Here C2 (x) := A0 (x) |λα1 =0 2(p−λβ )/p

λβ λα2 + λβ

λβ /p

λ /p

δ3 β .

(11.89)

We have that

%

A0 (x) |λα1 =0 =

1 (Γ (β − α2 ))

λα2

((λα2 p−λα2 )/p) p−1 βp − α2 p − 1 (11.90)

p − λβ λα2 βp − λα2 α2 p − λα2 + p − λβ

((p−λβ )/p)

·

x((λα2 βp−λα2 α2 p−λα2 +p−λβ )/p) . Proof. By Theorem 11.5. The constant A0 (x) |λα1 =0 was calculated in [48].

11.3.2 Results Involving Several Functions Here we use the following basic inequalities. Let α1 , α2 , . . . , αn ≥ 0, n ∈ N, then

ar1

+ ... +

arn

r

≤ (a1 + . . . + an ) ≤ n

r−1

n

ari

, r ≥ 1,

(11.91)

i=1

and

r

nr−1 (ar1 + . . . + arn ) ≤ (a1 + . . . + an ) ≤

n i=1

ari , 0 ≤ r ≤ 1.

(11.92)

11.3 Main Results

245

We present Theorem 11.11. Let αi ∈ R+ , 0 < β − αi < 1/p, 0 < p < 1, i = 1, 2, and let fj ∈ L1 (0, x) , j = 1, . . . , M ∈ N, x ∈ R+ − {0} have, respectively, L∞ fractional derivatives Dβ fj in [0, x] , each of ﬁxed sign a.e. on [0, x] , and let Dβ−k fj (0) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M. Consider also p (t) > 0 and q (t) ≥ 0, with all p (t) , 1/p (t), q (t) , 1/q (t) ∈ L∞ (0, x) . Let λβ > 0 and λα1 , λα2 ≥ 0, such that λβ > p. Put s p(β−αi −1)/p−1 −1/(p−1) Pi (s) := (s − t) (p (t)) dt, i = 1, 2; 0 ≤ s ≤ x, 0

A (s) :=

(11.93)

q (s) (P1 (s))

λα1 (p−1/p)

(Γ (β − α1 ))

λα2 (p−1/p)

(P2 (s)) λα1

x

A0 (x) :=

(p (s))

−λβ /p

,

λα2

(Γ (β − α2 )) p/(p−λβ )

(A (s))

(11.94)

(p−λβ )/p ds

,

(11.95)

0

and δ ∗1 := M 1−((λα1 +λβ )/p) .

(11.96)

Call ϕ1 (x) := A0 (x) |λα2 =0

λβ λα1 + λβ

λβ /p .

(11.97)

If λα2 = 0, we obtain that ⎞ ⎛ x M λ β λ q (s) ⎝ |Dα1 fj (s)| α1 Dβ fj (s) ⎠ ds ≥ 0

j=1

⎡ ∗ ⎣ δ 1 ϕ1 (x) 0

x

⎞ ⎤((λα1 +λβ )/p) ⎛ M p β D fj (s) ⎠ ds⎦ p (s) ⎝ .

(11.98)

j=1

Proof. By Theorem 11.4 we get

x 0

)

λβ λβ * α1 λα β λα β α1 1 1 q (s) D fj (s) ds ≥ D fj (s) + D fj+1 (s) D fj+1 (s)

246

11. Converse Riemann–Liouville Fractional Inequalities

A0 (x) |λα2 =0

λβ λα1 + λβ

λβ /p

) δ1

x

+ p p (s) Dβ fj (s) +

0

, ,(λα1 +λβ /p) β D fj+1 (s)p ds , j = 1, 2, . . . , M − 1.

(11.99)

Hence by adding all the above we obtain ⎧ x −1 + ⎨M λβ λ |Dα1 fj (s)| α1 Dβ fj (s) + q (s) ⎩ 0 j=1 λβ , λ ds |Dα1 fj+1 (s)| α1 Dβ fj+1 (s) ⎧ −1 ) ⎨M

≥ δ 1 ϕ1 (x)

⎩

x

0

j=1

⎫ , *(λα1 +λβ /p) ⎬ + p p p (s) Dβ fj (s) + Dβ fj+1 (s) ds . ⎭ (11.100)

Also it holds

x

+ λβ λβ , λ λ ds ≥ q (s) |Dα1 f1 (s)| α1 Dβ f1 (s) + |Dα1 fM (s)| α1 Dβ fM (s)

0

)

x

δ 1 ϕ1 (x)

*((λα1 +λβ )/p) + p β p , β p (s) D f1 (s) + D fM (s) ds

(11.101)

0

Adding (11.100) and (11.101) , and using (11.91) we have ⎞ ⎛ x M λβ λ 2 q (s) ⎝ |Dα1 fj (s)| α1 Dβ fj (s) ⎠ ds ≥ 0

δ 1 ϕ1 (x)

j=1

⎧ −1 ) ⎨M ⎩

j=1

)

x

x

*((λα1 +λβ )/p) + p β p , β + p (s) D fj (s) + D fj+1 (s) ds

0

*(λα1 +λβ /p) + p p , p (s) Dβ f1 (s) + Dβ fM (s) ds

(11.102)

≥

0

M 1−(λα1 +λβ /p) δ 1 ϕ1 (x)

⎧ ⎨ ⎩

0

⎞ ⎫((λα1 +λβ )/p) M ⎬ p β D fj (s) ⎠ ds p (s) ⎝2 . ⎭ ⎛

x

j=1

(11.103)

11.3 Main Results

247

We have established

⎛ x

q (s) ⎝

0

M

λα1

|Dα1 fj (s)|

⎞ λβ β D fj (s) ⎠ ds ≥

j=1

M 1−(λα1 +λβ /p) δ 1 2(λα1 +λβ /p)−1 ⎧ ⎨ ϕ1 (x)

⎩

x

0

⎤ ⎫(λα1 +λβ /p) ⎡ M ⎬ p β D fj (s) ⎦ ds p (s) ⎣ , ⎭

(11.104)

j=1

thus proving (11.98) . Next we give Theorem 11.12. All here are as in Theorem 11.11. Further assume λα2 ≥ λβ . Denote ϕ2 (x) := A0 (x) |λα1 =0 2(p−λβ )/p If λα1 = 0, then

x

0

|Dα2 fM (s)|

λβ λ α2 + λ β

λβ /p

λ /p

δ3 β .

(11.105)

⎧ −1 + ⎨M λβ λ |Dα2 fj+1 (s)| α2 Dβ fj (s) + q (s) ⎩ j=1

|Dα2 fj (s)| +

λα2

λα2

, β D fj+1 (s)λβ +

, β D f1 (s)λβ + |Dα2 f1 (s)|λα2 Dβ fM (s)λβ ds ≥ (11.106)

M 1−(λβ +λα2 /p) 2(λβ +λα2 /p) ϕ2 (x)

⎧ ⎨ ⎩

⎤ ⎫((λβ +λα )/p) 2 M ⎬ p ⎦ β ⎣ . p (s) D fj (s) ds ⎭ ⎡

x 0

j=1

Proof. From Theorem 11.5 we have

x 0

) λβ λβ * λ λ q (s) Dα2 fj+1 (s) α2 Dβ fj (s) + Dα2 fj (s) α2 Dβ fj+1 (s) ds ≥

248

11. Converse Riemann–Liouville Fractional Inequalities

x

ϕ2 (x)

(λβ +λα2 /p) + p p , , (11.107) p (s) Dβ fj (s) + Dβ fj+1 (s) ds

0

for j = 1, . . . , M − 1. Hence by adding all of the above we derive ⎛ x M −1 + λβ λ q (s) ⎝ |Dα2 fj+1 (s)| α2 Dβ fj (s) + 0

j=1 λα2

|Dα2 fj (s)| ⎛ ϕ2 (x) ⎝

M −1 x

, β D fj+1 (s)λβ ds ≥

⎞ , (λβ +λα2 /p) + p p ⎠. p (s) Dβ fj (s) + Dβ fj+1 (s) ds

0

j=1

(11.108) Similarly it holds

x

+ λβ λβ , λ λ ds ≥ q (s) |Dα2 fM (s)| α2 Dβ f1 (s) + |Dα2 f1 (s)| α2 Dβ fM (s)

0

ϕ2 (x)

x

(λβ +λα2 /p) + p p , . p (s) Dβ f1 (s) + Dβ fM (s) ds

(11.109)

0

Adding (11.108) and (11.109) and using (11.91) we derive (11.106). We continue with Theorem 11.13. Let 1 < β − α1 < 1/p, 0 < p < 1, α1 ∈ R+ , and let fj ∈ L1 (0, x) , j = 1, . . . , M ∈ N, x ∈ R+ − {0} have, respectively, L∞ fractional derivatives Dβ fj in [0, x] , each of ﬁxed sign a.e. on [0, x] , and let Dβ−k fj (0) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M. Consider also p (t) > 0 and q (t) ≥ 0, with p (t) , 1/p (t), q (t) , 1/q (t) ∈ L∞ (0, x) . Let λα ≥ λα+1 > 1; Φ is as in Theorem 11.7. Then ⎧⎧ x −1 + ⎨⎨M λα+1 λ |Dα1 fj (s)| α Dα1 +1 fj+1 (s) q (s) + ⎩⎩ 0 j=1

|Dα1 fj+1 (s)|

λα

, α +1 D 1 fj (s)λα+1

, + λ λ + |Dα1 f1 (s)|λα Dα1 +1 fM (s) α+1 + |Dα1 fM (s)|λα Dα1 +1 f1 (s) α+1 ds ≥

11.3 Main Results

M

1−(λα +λα+1 /p)

2(λα +λα+1 /p) Φ (x)

'

x 0

M p β p (s) D fj (s)

249

(((λα +λα+1 )/p) ds

.

j=1

(11.110)

Proof. From Theorem 11.7 we have x M −1 + λα+1 λ |Dα1 fj (s)| α Dα1 +1 fj+1 (s) q (s) + 0

j=1 λα

|Dα1 fj+1 (s)|

≥ Φ (x)

M −1 ) x

, α +1 D 1 fj (s)λα+1 ds

*(λα +λα+1 /p) p β p β ds p (s) D fj (s) + D fj+1 (s) .

0

j=1

(11.111) Similarly it holds x

+ λα+1 λ q (s) |Dα1 f1 (s)| α Dα1 +1 fM (s) +

0 λα

|Dα1 fM (s)| )

x

≥ Φ (x)

, α +1 D 1 f1 (s)λα+1 ds

*(λα +λα+1 /p) p β p β ds p (s) D f1 (s) + D fM (s) . (11.112)

0

Adding (11.111) and (11.112) , along with (11.91) we obtain (11.110) .

We present Theorem 11.14. All are as in Theorem 11.11. Consider the special case of λα2 = λα1 + λβ . Here T˜ (x) is as in (11.75) . Assume here for j = 1, . . . , M that x p p (t) Dβ fj (t) dt ∈ [H, Ψ] , 0 < H < Ψ. zj (x) := 0

Then

x

q (s) 0

⎧⎧ −1 + ⎨⎨M ⎩⎩

j=1

|Dα1 fj (s)|

λα1

|Dα2 fj+1 (s)|

λα1 +λβ

β D fj (s)λβ

250

11. Converse Riemann–Liouville Fractional Inequalities

λα1 +λβ

+ |Dα2 fj (s)|

|Dα1 fj+1 (s)|

λα1

, β D fj+1 (s)λβ

+ λβ λ λ +λ + |Dα1 f1 (s)| α1 |Dα2 fM (s)| α1 β Dβ f1 (s)

+ |Dα2 f1 (s)|

λα1 +λβ

λα1

|Dα1 fM (s)| ⎡

M

(1−(2(λα1 +λβ )/p)) 2(λα1 +λβ )/p 2

T˜ (x) ⎣

, β D fM (s)λβ ds ≥

(11.113)

⎛

x 0

⎞ ⎤ 2(λα +λβ )/p 1 M p β p (s) ⎝ . D fj (s) ⎠ ds⎦ j=1

Proof. Based on Theorem 11.8. The rest of the proof is as in the proof of Theorem 11.13. We continue with Corollary 11.15. (to Theorem 11.11, λα2 = 0, p (t) = q (t) = 1). Then ⎛ ⎞ x M λβ λ ⎝ |Dα1 fj (s)| α1 Dβ fj (s) ⎠ ds ≥ 0

j=1

⎡ δ ∗1 ϕ1 (x) ⎣

x

0

⎡ ⎤ ⎤((λα1 +λβ )/p) M p β D fj (s) ⎦ ds⎦ ⎣ .

(11.114)

j=1

Here A0 (x) |λα2 =0 within ϕ1 (x) is given by (11.87) . Proof. Based on Theorem 11.11.

We also give Corollary 11.16. (to Theorem 11.12, λα1 = 0, p (t) = q (t) = 1). It holds ⎧⎧ x ⎨⎨M −1 + λβ λ |Dα2 fj+1 (s)| α2 Dβ fj (s) + 0 ⎩⎩ j=1

|Dα2 fj (s)|

λα2

, β D fj+1 (s)λβ +

+ λβ λβ , λ λ |Dα2 fM (s)| α2 Dβ f1 (s) + |Dα2 f1 (s)| α2 Dβ fM (s) ds ≥

11.3 Main Results M

1−(λβ +λα2 /p)

(λβ +λα2 /p)

2

⎧ ⎨ ϕ2 (x)

⎩

251

⎤ ⎫((λβ +λα )/p) 2 M p ⎬ β ⎣ ⎦ . D fj (s) ds ⎭ ⎡

x 0

j=1

(11.115)

In (11.115) , within ϕ2 (x) , A0 (x) |λα1 =0 is given by (11.90) . Proof. Based on Theorem 11.12.

11.3.3 Results with Respect to Generalized Riemann – Liouville Fractional Derivative We give the notion of a generalized Riemann – Liouville fractional derivative at arbitrary anchor point a ∈ R; see [64]. Deﬁnition 11.17. Let v ≥ 0; deﬁne (Dav f ) (s) := (Dv fa ) (s − a) , s ≥ a,

(11.116)

where fa (t) := f (t + a) is the translate function, for v = 0 both sides equal to f (s) , and for v = n ∈ N we get (Dan f ) (s) = f (n) (s) , the ordinary derivative. Clearly here (Dav f ) (z + a) = (Dv fa ) (z) . For p (s) ,

Dav f

(s) ∈ L∞ (a, x) , x > a, a, x ∈ R we get

w

p (y) (Dav f ) (y) dy = a

(11.117)

w−a

p (z + a) (Dv fa ) (z) dz,

(11.118)

a

all a ≤ w ≤ x. So here we transfer the results of Sections 11.3.1 and 11.3.2, using the above concepts, to arbitrary interval [a, x] . Thus earlier results are applied to fa over [0, x − a] , for f over [a, x] and use the generalized Riemann – Liouville fractional derivative. This method was applied extensively in [45]; see there for all details concerning this transfer. We need Deﬁnition 11.18. [45] We say that f ∈ L1 (a, w) , a < w; a, w ∈ R has an L∞ fractional derivative Daβ f [β > 0] in [a, w] , iﬀ (1) Daβ−k f ∈ C ([a, w]) , k = 1, . . . , m := [β] + 1

252

11. Converse Riemann–Liouville Fractional Inequalities

(2) Daβ−1 f ∈ AC ([a, w]) , and (3) Daβ f ∈ L∞ (a, w) . We need Lemma 11.19. (see [45]) Let β > α ≥ 0 and let f ∈ L1 (a, w) have an L∞ fractional derivative Daβ f in [a, w] , and let Daβ−k f (a) = 0, k = 1, . . . , [β] + 1. Then s 1 β−α−1 Daα f (s) = (s − t) Daβ f (t) dt, (11.119) Γ (β − α) a all a ≤ s ≤ w. Clearly here Daα f ∈ AC ([a, w]) for β − α ≥ 1, and in C ([a, w]) for β − α ∈ (0, 1) , hence Daα f ∈ L∞ (a, w) , and Daα f ∈ L1 (a, w) . Notice that for β > 0 that β−k Da f (a) = Dβ−k fa (0) , all k = 1, . . . , [β] + 1. (11.120) Here we only show two such transfers; for the rest of the results so far we can get similar corresponding transfers, acting as in [45]. We give Theorem 11.20. Let αi ≥ 0, 0 < β − αi < 1/p, 0 < p < 1, i = 1, 2, and let f1 , f2 ∈ L1 (a, x) , a, x ∈ R, a < x, have, respectively, L∞ fractional derivatives Daβ f1 , Daβ f2 in [a, x] , each of ﬁxed sign a.e. on [a, x] , and let Daβ−k fi (a) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Consider also p (t) > 0 and q (t) ≥ 0, with all p (t) ,

1 1 , q (t) , ∈ L∞ (a, x) . p (t) q (t)

Let λβ > 0 and λα1 , λα2 ≥ 0, such that λβ > p.

11.3 Main Results

253

Put

s

Pi (s) :=

(s − t)p(β−αi −1)/p−1 (p (t + a))−1/(p−1) dt, i = 1, 2; 0 ≤ s ≤ x−a,

0

A (s) :=

(11.121)

q (s + a) (P1 (s))

λα1 (p−1/p)

λα2 (p−1/p)

(P2 (s)) λα1

(Γ (β − α1 ))

(Γ (β − α2 ))

−λβ /p

(p (s + a))

,

λα2

(11.122)

(p−λβ )/p

x−a

A0 (x − a) :=

(A (s))

p/(p−λβ )

ds

,

(11.123)

0

and δ 1 := 21−(λα1 +λβ /p) .

(11.124)

If λα2 = 0, then

x

+ λ β λβ , λ λ ds ≥ q (s) |Daα1 f1 (s)| α1 Daβ f1 (s) + |Daα1 f2 (s)| α1 Daβ f2 (s)

a

A0 (x − a) |λα2 =0

λβ λα1 + λβ

λβ /p

) δ1

x

+ p p (s) Daβ f1 (s) +

a

, ,((λα1 +λβ )/p) β Da f2 (s)p ds .

(11.125)

Proof. We apply (11.8) for f1a := f1 (· + a) , f2a := f2 (· + a) , p (· + a) , q (· + a) and the fractional derivatives Dβ f1a , Dβ f2a , Dαi f1a , Dαi f2a , i = 1, 2, all on [0, x − a] . Namely here we have that the functions f1a , f2a , p (· + a) , q (· + a) fulﬁll all the assumptions of Theorem 11.4 on the interval [0, x − a] , so for λα2 = 0 we get by (11.8) that

x−a 0

) λβ λβ * q (s + a) |Dα1 f1a (s)|λα1 Dβ f1a (s) + |Dα1 f2a (s)|λα1 Dβ f2a (s) ds

254

11. Converse Riemann–Liouville Fractional Inequalities

≥ A0 (x − a) |λα2 =0

λβ λα1 + λβ

λβ /p

)

x−a

δ1

+ p p (s + a) Dβ f1a (s) +

0

, ,(λα1 +λβ /p) β D f2a (s)p ds .

(11.126)

Using (11.117) , and from (11.126) we ﬁnd

x−a

+ λβ λ q (s + a) |Daα1 f1 (s + a)| α1 Daβ f1 (s + a) +

0

|Daα1 f2 (s + a)|

A0 (x − a) |λα2 =0

λβ λα1 + λβ

λα1

, β Da f2 (s + a)λβ ds ≥

λβ /p

)

x−a

δ1

+ p p (s + a) Daβ f1 (s + a) +

0

, ,(λα1 +λβ /p) β Da f2 (s + a)p ds .

(11.127)

Let Fa (s) := F (s + a) , Ga (s) := G (s + a) denote the integrands of left and right integrals of (11.127) . Notice easily that Fa , Ga ∈ L∞ (0, x − a) , that is, F, G ∈ L∞ (a, x) , hence F, G ∈ L1 (a, x) . Thus

x−a

0

x

Fa (s) ds =

F (s) ds, a

x−a

Ga (s) ds = 0

x

G (s) ds, a

proving (11.125) . Notice for a = 0, inequality (11.125) collapses to inequality (11.8) . We ﬁnish this chapter with Theorem 11.21. Let αi ≥ 0, 0 < β − αi < 1/p, 0 < p < 1, i = 1, 2, and let fj ∈ L1 (a, x) , j = 1, . . . , M ∈ N, a, x ∈ R, a < x, have, respectively, L∞ fractional derivatives Daβ fj in [a, x] , each of ﬁxed sign a.e. on [a, x] , and let Daβ−k fi (a) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M. Consider also p (t) > 0 and q (t) ≥ 0,

11.3 Main Results

with all

255

1 1 , q (t) , ∈ L∞ (a, x) . p (t) q (t)

p (t) , Let

λβ > 0 and λα1 , λα2 ≥ 0, such that λβ > p. Call

s

Pi (s) :=

(s − t)p(β−αi −1)/p−1 (p (t + a))−1/(p−1) dt, i = 1, 2; 0 ≤ s ≤ x−a,

0

A (s) :=

(11.128)

q (s + a) (P1 (s))

λα1 (p−1/p)

λα2 ((p−1)/p)

(P2 (s)) λα1

(Γ (β − α1 ))

(Γ (β − α2 ))

−λβ /p

(p (s + a))

λα2

, (11.129)

(p−λβ )/p

x−a

A0 (x − a) :=

(A (s))

p/(p−λβ )

ds

,

(11.130)

0

and δ ∗1 := M (1−(λα1 +λβ /p)) .

(11.131)

Call ϕ1 (x − a) := A0 (x − a) |λα2 =0

λβ λ α1 + λ β

λβ /p .

(11.132)

If λα2 = 0, then

⎛ x

q (s) ⎝

a

M

λα1

|Daα1 fj (s)|

⎞ λβ β Da fj (s) ⎠ ds ≥

j=1

⎡ δ ∗1 ϕ1 (x − a) ⎣ a

⎞ ⎤((λα1 +λβ )/p) M p β Da fj (s) ⎠ ds⎦ p (s) ⎝ . ⎛

x

(11.133)

j=1

Proof. Based on Theorem 11.11 (see (11.98)) and similar to Theorem 11.20.

12 Multivariate Canavati Fractional Taylor Formula

We present here is a multivariate fractional Taylor formula using the Canavati deﬁnition of fractional derivative. As related results we present that the order of fractional-ordinary partial diﬀerentiation is immaterial, we discuss fractional integration by parts, and we estimate the remainder of our multivariate fractional Taylor formula. This treatment is based on [40].

12.1 Introduction The main motivation here comes from Canavati [101], Anastassiou [17], and Anastassiou [19], where there is presented a Taylor univariate fractional formula by using an appropriate deﬁnition of fractional derivative introduced ﬁrst in Canavati [101]. So we extend this formula to the multivariate fractional case over a compact and convex subset of Rk , k ≥ 2, for all fractional orders ν > 0. We give an estimate to the remainder of our multivariate fractional Taylor formula. We give under mild and natural assumptions that the order of fractional-ordinary partial diﬀerentiation is immaterial. Also we present some fractional integration by parts results. The main overall ingredient here is the Riemann–Liouville integral.

G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 257 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 12,

258

12. Multivariate Canavati Fractional Taylor Formula

12.2 Results We make Remark 12.1. We follow Anastassiou [19, p. 540] (see also Canavati [101] and Anastassiou [17]). Let [a, b] ⊆ R. Let x, x0 ∈ [a, b] such that x ≥ x0 , x0 is ﬁxed. Let f ∈ C([a, b]) and deﬁne x 1 (x − t)ν−1 f (t)dt, x0 ≤ x ≤ b, (12.1) (Jνx0 f )(x) = Γ(ν) x0 ν > 0, the generalized Riemann–Liouville integral. We consider the subspace Cxν0 ([a, b]) of C n ([a, b]), n := [ν], α := ν − n(0 < α < 1): x0 Cxν0 ([a, b]) := {f ∈ C n ([a, b]) : J1−α f (n) ∈ C 1 ([x0 , b])}.

(12.2)

Hence, let f ∈ Cxν0 ([a, b]) , we deﬁne the generalized ν - fractional derivative of f over [x0 , b] (see also Canavati [101] and Anastassiou [17]) as x0 f (n) ) . Dxν0 f := (J1−α

Notice that

x0 J1−α f (n) (x) =

1 Γ(1 − α)

x

(x − t)−α f (n) (t)dt

(12.3)

(12.4)

x0

exists for f ∈ Cxν0 ([a, b]). Let fx0 (t) := f (x0 + t), 0 ≤ t ≤ b − x0 , x ≥ x0 . By change of variable we obtain (12.5) (D0ν fx0 )(x − x0 ) = (Dxν0 f )(x). When ν ∈ N then the fractional derivative collapses to the usual one. We mention the fractional Taylor formula. See Anastassiou [19, p. 540], Canavati [101], and Anastassiou [17]. Theorem 12.2. Let f ∈ Cxν0 ([a, b]), x0 ∈ [a, b] ﬁxed. (i) If ν ≥ 1, then it holds f (x) = f (x0 ) + f (x0 )(x − x0 ) + f (x0 )

+(Jνx0 Dxν0 f )(x),

(x − x0 )2 (x − x0 )n−1 + · · · + f (n−1) (x0 ) 2 (n − 1)!

x ∈ [a, b] : x ≥ x0 .

all

(12.6)

(ii) If 0 < ν < 1 we have f (x) = (Jνx0 Dxν0 f )(x),

all

x ∈ [a, b] : x ≥ x0 .

(12.7)

12.2 Results

259

We transfer Theorem 12.2 to the multivariate case. We make Remark 12.3. Let Q be a compact and convex subset of Rk , k ≥ 2; z := (z1 , . . . , zk ), x0 := (x01 , . . . , x0k ) ∈ Q . Let f ∈ C n (Q), n ∈ N. Call gz (t) := f (x0 + t(z − x0 )), 0 ≤ t ≤ 1; gz (0) = f (x0 ), gz (1) = f (z). (12.8) Then ⎡ j ⎤ k ∂ gz(j) (t) = ⎣ (zi − x0i ) f ⎦ (x0 + t(z − x0 )), ∂x i i=1 j = 0, 1, 2, . . . , n, and gz(n) (0)

' =

k

∂ (zi − x0i ) ∂xi i=1

(12.9)

n ( f (x0 ).

(12.10)

If all fα (x0 ) := ∂ α f /∂xα (x0 ) = 0, α := (α1 , . . . , αk ), αi ∈ Z+ , i = "k (l) 1, . . . , k ; |α| := i=1 αi =: l, then gz (0) = 0, where l ∈ {0, 1, . . . , n}. We quote that k ∂f gz (t) = (zi − x0i ) (x0 + t(z − x0 )). (12.11) ∂xi i=1 First let 1 ≤ ν < 2 ; then here n := [ν] = 1 and α = ν − 1; 1 − α = 2 − ν. Because 0 ≤ ν − 1 < 1, then n∗ := [ν − 1] = 0, α∗ = ν − 1 − n∗ = ν − 1, and 1 − α∗ = 2 − ν. Set x 1 (Jν gz )(x) := (x − t)ν−1 gz (t)dt, (12.12) Γ(ν) 0 0 ≤ x ≤ 1. Consider C ν ([0, 1]) := {g ∈ C 1 ([0, 1]) : J2−ν g ∈ C 1 ([0, 1])} ⊆ C 1 ([0, 1]), 1 ≤ ν < 2. (12.13) Assume that as a function of t : fxi (x0 + t(z − x0 )) ∈ C ν−1 ([0, 1]), i = (ν) (ν) 1, . . . , k, then there exists the fractional derivative gz , gz = (J2−ν gz ) . The last comes by using (12.11) to have x 1 (J2−ν gz )(x) = (x − t)1−ν gz (t)dt (12.14) Γ(2 − ν) 0

260

12. Multivariate Canavati Fractional Taylor Formula

"k =

− x0i ) Γ(2 − ν)

i=1 (zi

x

(x − t)1−ν 0

0 ≤ x ≤ 1. Hence it holds k

1 Γ(2 − ν)

(zi − x0i )

i=1

k

=

∂f (x0 + t(z − x0 ))dt, ∂xi

(J2−ν gz (x)) = x

(x − t)1−ν 0

∂f (x0 + t(z − x0 ))dt ∂xi

(12.15)

(zi − x0i )(J2−ν (fxi (x0 + t(z − x0 )))) .

(12.16)

(12.17)

i=1

That is, gz(ν) (t) =

k

(zi − x0i )

i=1

(ν−1) ∂f (x0 + t(z − x0 )) , ∂xi

(12.18)

0 ≤ t ≤ 1, 1 ≤ ν < 2. Thus the remainder turns to k

(Jν gz(ν) )(t) =

(zi − x0i )[Jν (fxi (x0 + t(z − x0 )))(ν−1) ](t).

(12.19)

i=1

By (12.6) applied on gz we obtain f (z) = gz (1) = f (x0 ) + (Jν gz(ν) )(1).

(12.20)

That is, it holds f (z) = f (x0 ) +

k

(zi − x0i )[Jν (fxi (x0 + t(z − x0 )))(ν−1) ](1).

(12.21)

i=1

More precisely we ﬁnd f (z) = f (x0 ) +

k i=1

(zi − x0i )

1 Γ(ν)

1

(1 − t)ν−1 (fxi (x0 + t(z − x0 )))(ν−1) dt.

(12.22)

0

From Remark 12.3 we have established the basic multivariate Canavati fractional Taylor formula. Theorem 12.4. Let f ∈ C 1 (Q), Q compact and convex ⊆ Rk , k ≥ 2. For ﬁxed x0 , z ∈ Q, assume that as a function of t : fxi (x0 + t(z − x0 )) ∈ C ν−1 ([0, 1]), 1 ≤ ν < 2, all i = 1, . . . , k. Then

12.2 Results

261

(i) f (z1 , . . . , zk ) = f (x01 , . . . , x0k )+ k (zi − x0i ) Γ(ν)

i=1

1

(1 − t)ν−1 (fxi (x0 + t(z − x0 )))(ν−1) dt.

(12.23)

0

(ii) Given f (x0 ) = 0, then f (z) =

k (zi − x0i ) i=1

Γ(ν)

1

(1 − t)ν−1 (fxi (x0 + t(z − x0 )))(ν−1) dt. (12.24)

0

We make Remark 12.5. Continuing from Remark 12.3. Here f ∈ C 2 (Q), Q ⊆ R2 , we have gz (t) = (z1 − x01 )2

∂2f (x0 + t(z − x0 )) + 2(z1 − x01 )(z2 − x02 ) ∂x21

∂2f ∂2f (x0 + t(z − x0 )) + (z2 − x02 )2 (x0 + t(z − x0 )). ∂x1 ∂x2 ∂x22

(12.25)

Let 2 ≤ ν < 3; then n := [ν] = 2, α := ν − n = ν − 2, 1 − α = 3 − ν. Set ν ∗ := ν −2; then n∗ := [ν −2] = 0, α∗ = (ν −2)−n∗ = ν −2, 1−α∗ = 3−ν. We have (0 ≤ x ≤ 1) x 1 (x − t)2−ν gz (t)dt = (J3−ν gz )(x) = Γ(3 − ν) 0 x 1 2 (x − t)2−ν fx1 x1 (x0 + t(z − x0 ))dt (z1 − x01 ) Γ(3 − ν) 0 x 1 (x − t)2−ν fx1 x2 (x0 + t(z − x0 ))dt +2(z1 − x01 )(z2 − x02 ) Γ(3 − ν) 0 x 1 (x − t)2−ν fx2 x2 (x0 + t(z − x0 ))dt. (12.26) +(z2 − x02 )2 Γ(3 − ν) 0 That is, it holds. (J3−ν gz )(x) = (z1 − x01 )2 (J3−ν (fx1 x1 (x0 + t(z − x0 ))))(x)+ 2(z1 − x01 )(z2 − x02 )(J3−ν (fx1 x2 (x0 + t(z − x0 ))))(x)+ (z2 − x02 )2 (J3−ν (fx2 x2 (x0 + t(z − x0 ))))(x).

(12.27)

262

12. Multivariate Canavati Fractional Taylor Formula

Assuming now that fx1 x1 (x0 + t(z − x0 )), fx1 x2 (x0 + t(z − x0 )), fx2 x2 (x0 + t(z − x0 )), as functions of t belong to C (ν−2) ([0, 1]) we obtain that there exists gz(ν) (t) = (z1 − x01 )2 (fx1 x1 (x0 + t(z − x0 )))(ν−2) + 2(z1 − x01 )(z2 − x02 )(fx1 x2 (x0 + t(z − x0 )))(ν−2) + (z2 − x02 )2 (fx2 x2 (x0 + t(z − x0 )))(ν−2) .

(12.28)

Next we observe that (Jν gz(ν) )(x) = (z1 − x01 )2 (Jν (fx1 x1 (x0 + t(z − x0 )))(ν−2) )(x) +2(z1 − x01 )(z2 − x02 )(Jν (fx1 x2 (x0 + t(z − x0 )))(ν−2) )(x) +(z2 − x02 )2 (Jν (fx2 x2 (x0 + t(z − x0 )))(ν−2) )(x).

(12.29)

We have proved via (12.6) the next Taylor type result. Theorem 12.6. Let f ∈ C 2 (Q), Q compact and convex ⊆ R2 . For ﬁxed x0 , z ∈ Q assume that as functions of t : fx1 x1 (x0 + t(z − x0 )), fx1 x2 (x0 + t(z − x0 )), fx2 x2 (x0 + t(z − x0 )) ∈ C (ν−2) ([0, 1]), where 2 ≤ ν < 3. Then (i) f (z1 , z2 ) = f (x01 , x02 ) + (z1 − x01 ) (z1 − x01 )2

1 Γ(ν)

(z2 − x02 )2

1 Γ(ν)

(1 − t)ν−1 (fx1 x1 (x0 + t(z − x0 )))(ν−2) dt 1 Γ(ν)

1

(1 − t)ν−1 (fx1 x2 (x0 + t(z − x0 )))(ν−2) dt+ 0

1

(1 − t)ν−1 (fx2 x2 (x0 + t(z − x0 )))(ν−2) dt. ∂f ∂x1 (x0 )

f (z1 , z2 ) = (z1 − x01 )2 +2(z1 − x01 )(z2 − x02 ) 1 Γ(ν)

(12.30)

0

(ii) When f (x0 ) =

(z2 − x02 )2

1

0

+2(z1 − x01 )(z2 − x02 )

∂f ∂f (x0 ) + (z2 − x02 ) (x0 )+ ∂x1 ∂x2

1 Γ(ν)

1 Γ(ν)

=

∂f ∂x2 (x0 )

= 0 , then

1

(1 − t)ν−1 (fx1 x1 (x0 + t(z − x0 )))(ν−2) dt 0

1

(1 − t)ν−1 (fx1 x2 (x0 + t(z − x0 )))(ν−2) dt+ 0

1

(1 − t)ν−1 (fx2 x2 (x0 + t(z − x0 )))(ν−2) dt. 0

(12.31)

12.2 Results

263

The following general multivariate Canavati fractional Taylor formula is valid. Theorem 12.7. Let f ∈ C n (Q), Q compact and convex ⊆ Rk , k ≥ 2 ; here ν ≥ 1 such that n = [ν]. For ﬁxed x0 , z ∈ Q assume that as functions of t : fα (x0 + t(z − x0 )) ∈ C (ν−n) ([0, 1]), for all α := (α1 , . . . , αk ), αi ∈ "k Z+ , i = 1, . . . , k; |α| := i=1 αi = n. Then (i) f (z1 , . . . , zk ) = f (x01 , . . . , x0k ) +

k

(zi − x0i )

i=1

) "

k i=1 (zi

−

+

∂f (x01 , . . . , x0k ) ∂xi

2 * f (x01 , . . . , x0k )

∂ x0i ) ∂x i

+ ...

2

) "

k i=1 (zi

−

∂ x0i ) ∂x i

n−1 * f (x01 , . . . , x0k ) +

(n − 1)!

1 Γ(ν)

1

(1 − t)ν−1 0

⎧' k ⎨ ⎩

∂ (zi − x0i ) ∂xi i=1

n ((ν−n) f

⎫ ⎬ (x0 + t(z − x0 )) dt. ⎭ (12.32)

(ii) If all fα (x0 ) = 0, α := (α1 , . . . , αk ), αi ∈ Z+ , i = 1, . . . , k, |α| := "k i=1 αi = l, l = 0, . . . , n − 1, then 1 f (z1 , . . . , zk ) = Γ(ν)

1

(1 − t)ν−1 0

⎫ ⎧' n ((ν−n) k ⎬ ⎨ ∂ (zi − x0i ) f (x0 + t(z − x0 )) dt. ⎭ ⎩ ∂xi i=1

Proof. Use (12.6).

(Note that fractional diﬀerentiation is a linear operation.) We make

(12.33)

264

12. Multivariate Canavati Fractional Taylor Formula

Remark 12.8. Continuing from the previous remarks. Let here 0 < ν < 1. Assume that f (x0 + t(z − x0 )) ∈ C ν ([0, 1]) as a function of t . Then by gz (t) := f (x0 + t(z − x0 )) we have (ν)

(ν)

gz (t) = (f (x0 + t(z − x0 )))(ν) , and (Jν gz )(t) = (Jν (f (x0 + t(z − x0 )))(ν) )(t),

t ∈ [0, 1]. Hence (Jν gz(ν) )(1) =

1 Γ(ν)

1

(1 − t)ν−1 (f (x0 + t(z − x0 )))(ν) dt.

(12.34)

0

We have established the next multivariate Canavati fractional Taylor formula when 0 < ν < 1. Theorem 12.9. Let bounded f : Q → R, where Q convex ⊆ Rk , k ≥ 2, such that as a function of t : f (x0 +t(z−x0 )) ∈ C ν ([0, 1]), 0 < ν < 1, x0 , z ∈ Q being ﬁxed. Then 1 1 (1 − t)ν−1 (f (x0 + t(z − x0 )))(ν) dt. (12.35) f (z1 , . . . , zk ) = Γ(ν) 0 Proof. Use (12.7).

We make Remark 12.10. Next we study the ordinary partial derivatives of fractional derivatives. Let 0 < α < 1, f ∈ C 1 ([0, 1]2 ), x ∈ [0, 1] ﬁxed, and consider x (x − t)−α f (t, z)dt, (12.36) γ(x, z) := 0

∀ z ∈ [0, 1]. We observe that x x |γ(x, z)| ≤ (x − t)−α |f (t, z)|dt ≤ f ∞ (x − t)−α dt = 0

0

f ∞ That is, the function

f ∞ x1−α ≤ < +∞. 1−α 1−α

ρ(t) := (x − t)−α f (t, z)

(12.37)

is Lebesgue integrable in t ∈ [0, x], ∀ z ∈ [0, 1]. Thus one can consider integration in (12.36) over [0, x), ∀ z ∈ [0, 1]. Also the function (12.38) λ(z) := (x − t)−α f (t, z)

12.2 Results

265

is diﬀerentiable in z ∈ [0, 1], ∀ t ∈ [0, x). That is, we have λ (z) = (x − t)−α Moreover,

∂f (t, z) , ∀t ∈ [0, x). ∂z

∂f |λ (z)| ≤ (x − t)−α ∂z , ∞

(12.39)

(12.40)

∀ (t, z) ∈ [0, x) × [0, 1]. The R.H.S (12.40) is integrable in t ∈ [0, x] and nonnegative. Hence by H. Bauer [79, pp. 103–104] we obtain that (x − t)−α ∂f (t, z)/∂z is integrable in t ∈ [0, x) and x ∂f (t, z) ∂γ(x, z) = dt, (12.41) (x − t)−α ∂z ∂z 0 ∀ z ∈ [0, 1]. We have proved Lemma 12.11. Let 0 < α < 1, f ∈ C 1 ([0, 1]2 ), 0 ≤ x ≤ 1. Then x x ∂f (t, z) ∂ −α dt, (12.42) (x − t) f (t, z)dt = (x − t)−α ∂z ∂z 0 0 ∀ z ∈ [0, 1]. We make Remark 12.12. Assume now 0 < α < 1, f ∈ C n+1 ([0, 1]2 ), n ∈ N. Then by Lemma 12.11 we get x x n n+1 ∂ f −α ∂ f −α ∂ (t, z)dt. (12.43) (x − t) (t, z)dt = (x − t) n n ∂z ∂t ∂t ∂z 0 0 Let now ν > 0, n := [ν], α := ν − n. We suppose existence of n ∂ g ∂ ν g(x, z) ∂ g (ν) (x, z) := = (·, z) (x, z). J 1−α ∂xν ∂x ∂tn

(12.44) (ν)

We also assume here that g ∈ C n+1 ([0, 1]2 ), and (g (ν) (x, z))z , gz (x, z) exist and are jointly continuous in (x, z) ∈ [0, 1]2 , [ν] = n ∈ N. Then it holds n ∂ g ∂ ∂ (ν) ∂ (ν) (g (x, z)) = (·, z) (x, z) = (g (x, z))z = J1−α ∂z ∂z ∂x ∂tn

266

12. Multivariate Canavati Fractional Taylor Formula

∂ ∂x

n ∂ g (·, z) (x, z)(by (12.43)) = J1−α ∂tn n+1 ∂ ∂ g(·, z) (x, z) J1−α ∂x ∂z∂tn n ∂ ∂ g (·, z) = gz(ν) (x, z). = J1−α z ∂x ∂tn

∂ ∂z

That is, ∀(x, z) ∈ [0, 1]2 , ν > 0.

(g (ν) (x, z))z = gz(ν) (x, z),

(12.45)

In brief, it holds (g (ν) )z = (gz )(ν) .

(12.46)

Under more similar suitable assumptions one obtains (g (ν) )zz = (gzz )(ν) , (g (ν) )z1 z2 = (gz1 z2 )(ν) , (g (ν) )z1 z2 z3 = (gz1 z2 z3 )(ν) , . . . . (12.47) We have established that the order of fractional-ordinary partial diﬀerentiation is immaterial. Theorem 12.13. Let g ∈ C n+1 ([0, 1]2 ), and ν > 0 such that [ν] = n ∈ N. (ν) Assume the existence of g (ν) (x, z) , and (g (ν) (x, z))z , gz (x, z) both exist 2 and are jointly continuous in (x, z) ∈ [0, 1] . Then (12.48) (g (ν) (x, z))z = gz(ν) (x, z), ∀(x, z) ∈ [0, 1]2 .

We make Remark 12.14. Next comes fractional integration by parts. Let f, g ∈ C ν ([0, 1]), ν > 0, n := [ν], α := ν − n. Here g (ν) =

d(J1−α g (n) ) (ν) d (J1−α f (n) ) ,f . = dx dx

That is, d (J1−α g (n) ) = g (ν) dx, d (J1−α f (n) ) = f (ν) dx. We observe that

1

1

(J1−α f (n) )(x)g (ν) (x)dx = 0

(J1−α f (n) )(x)d (J1−α g (n) )(x) = 0

(J1−α f (n) )(1)(J1−α g (n) )(1) −

1

(J1−α g (n) )(x)d (J1−α f (n) )(x) = 0

12.2 Results

267

1

(J1−α f (n) )(1)(J1−α g (n) )(1) −

(J1−α g (n) )(x)f (ν) (x)dx.

(12.49)

0

Next let us take g ∈ C ν ([0, 1]), 1 ≤ ν < 2, n := [ν] = 1, α := ν − n = ν − 1, 1 − α = 2 − ν, and f ∈ C 1 ([0, 1]). Then g (ν) (x) = d (J2−ν g )(x)/dx; that is, d (J2−ν g )(x) = g (ν) (x)dx. Hence 1

1

f (x)g (ν) (x)dx = 0

f (1)(J2−ν g )(1) −

f (x)d(J2−ν g )(x) =

0

1

(J2−ν g )(x)f (x)dx.

(12.50)

0

We have established the following fractional integration by parts formulae. Theorem 12.15. (i) Let f, g ∈ C ν ([0, 1]), ν > 0, n := [ν], α := ν − n. Then 1 (J1−α f (n) )(x)g (ν) (x)dx = 0

(J1−α f

(n)

)(1)(J1−α g

(n)

1

)(1) −

(J1−α g (n) )(x)f (ν) (x)dx.

(12.51)

0

(ii) Let g ∈ C ν ([0, 1]), 1 ≤ ν < 2, f ∈ C 1 ([0, 1]). Then

1

f (x)g (ν) (x)dx = f (1)(J2−ν g )(1) −

0

1

(J2−ν g )(x)f (x)dx. (12.52)

0

We make the last Remark 12.16. Here we estimate the remainder (12.32). By deﬁnition in this chapter (see (12.2), (12.3)), the fractional derivatives are continuous functions. So the function ⎧' ⎫ n ((ν−n) k ⎨ ⎬ ∂ Gν (t) := (zi − x0i ) f (x0 + t(z − x0 )) , (12.53) ⎩ ⎭ ∂xi i=1

t ∈ [0, 1], that appears in the remainder of (12.32), is continuous in t. We write the remainder (12.32) as Rν :=

1 Γ(ν)

1

(1 − t)ν−1 0

268

12. Multivariate Canavati Fractional Taylor Formula

⎧' k ⎨ ⎩

(zi − x0i )

i=1

1 Γ(ν) We obtain |Rν | ≤

1 Γ(ν)

∂ ∂xi

n ((ν−n) f

⎫ ⎬ (x0 + t(z − x0 )) dt = ⎭

1

(1 − t)ν−1 Gν (t)dt, ν ≥ 1.

(12.54)

0

1

(1 − t)ν−1 |Gν (t)|dt ≤ 0

1 Γ(ν)

1

|Gν (t)|dt = 0

1 Gν L1 ([0,1]) . Γ(ν) Also for p, q > 1 : 1/p + 1/q = 1, we get 1 1 |Rν | ≤ (1 − t)ν−1 |Gν (t)|dt ≤ Γ(ν) 0 1/p 1 1/q 1 1 ν−1 p q ((1 − t) ) dt |Gν (t)| dt = Γ(ν) 0 0 1 1 Gν Lq ([0,1]) . Γ(ν) (p(ν − 1) + 1)1/p In the case p = q = 2 we have |Rν | ≤ Finally we get that 1 |Rν | ≤ Γ(ν)

1 1 √ Gν L2 ([0,1]) . Γ(ν) 2ν − 1

1

(1 − t)ν−1 |Gν (t)|dt ≤ 0

Gν ∞ . Γ(ν + 1)

(12.55)

(12.56)

(12.57)

(12.58)

We have established the following remainder estimate. Theorem 12.17. All here are as in Theorem 12.7. Let Rν be the remainder in (12.32) (see (12.54)), and Gν as in (12.53). Then Gν Lq ([0,1]) Gν L1 ([0,1]) |Rν | ≤ min , , Γ(ν) Γ(ν)(p(ν − 1) + 1)1/p Gν L2 ([0,1]) Gν ∞ √ , , (12.59) Γ(ν) 2ν − 1 Γ(ν + 1) where p, q > 1 : 1/p + 1/q = 1. Comment. The chain rule as in ordinary diﬀerentiation is not possible in Canavati fractional diﬀerentiation. That limits us a lot from using the multivariate Canavati fractional Taylor formula, as we employ the usual one involving only ordinary partial derivatives of functions.

13 Multivariate Caputo Fractional Taylor Formula

This is a continuation of Chapter 12. We establish here a multivariate fractional Taylor formula via the Caputo fractional derivative. The fractional remainder is expressed as a composition of two Riemann–Liouville fractional integrals. We estimate the remainder. This treatment is based on [53].

13.1 Background We start with Deﬁnition 13.1. [134] Let ν ≥ 0; the operator Jaν , deﬁned on L1 (a, b) by x 1 ν−1 Jaν f (x) := (x − t) f (t) dt (13.1) Γ (ν) a for a ≤ x ≤ b, is called the Riemann–Liouville fractional integral operator of order ν. For ν = 0, we set Ja0 := I, the identity operator. Here Γ stands for the gamma function. By Theorem 2.1 of [134, p. 13], Jaν f (x), ν > 0, exists for almost all x ∈ [a, b] and Jaν f ∈ L1 (a, b), where f ∈ L1 (a, b) . Here AC n ([a, b]) is the space of functions with absolutely continuous (n − 1)–st derivative. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 269 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 13,

270

13. Multivariate Caputo Fractional Taylor Formula

We need to mention Deﬁnition 13.2. [58, 134] Let ν ≥ 0; n := ν , · is the ceiling of the number, f ∈ AC n ([a, b]) . We call the Caputo fractional derivative x 1 n−ν−1 (n) ν (x − t) f (t) dt, (13.2) D∗a f (x) := Γ (n − ν) a ∀x ∈ [a, b] . ν f (x) exists almost everywhere for x ∈ [a, b] . The above function D∗a ν 0 f = f. If ν ∈ N, then D∗a f = f (ν) the ordinary derivative; it is also D∗a We need Theorem 13.3. (Taylor expansion for Caputo derivatives, [134, p. 40]) Assume ν ≥ 0, n = ν , and f ∈ AC n ([a, b]) . Then f (x) =

n−1 k=0

f (k) (a) 1 k (x − a) + k! Γ (ν)

x

ν−1

(x − t) a

ν D∗a f (t) dt,

(13.3)

∀x ∈ [a, b] .

13.2 Results We establish analogues of Theorem 13.3 to the multivariate case. We make Remark 13.4. Let Q be a compact and convex subset of Rk , k ≥ 2; z := (z1 , . . . , zk ) , x0 := (x01 , . . . , x0k ) ∈ Q. Let f ∈ C n (Q) , n ∈ N. Set gz (t) := f (x0 + t (z − x0 )) , 0 ≤ t ≤ 1; gz (0) = f (x0 ) , gz (1) = f (z) .

(13.4)

Then ⎡ j ⎤ k ∂ gz(j) (t) = ⎣ (zi − x0i ) f ⎦ (x0 + t (z − x0 )) , ∂x i i=1 j = 0, 1, 2, . . . , n, and gz(n)

(0) =

' k

∂ (zi − x0i ) ∂xi i=1

(13.5)

n ( f (x0 ) .

(13.6)

13.2 Results

271

If all fα (x0 ) := ∂ α f /∂xα (x0 ) = 0, α := (α1 , . . . , αk ) , αi ∈ Z+ , i = "k (l) 1, . . . , k; |α| := i=1 αi =: l, then gz (0) = 0, where l ∈ {0, 1, . . . , n} . We quote that k ∂f (zi − x0i ) (x0 + t (z − x0 )) . (13.7) gz (t) = ∂x i i=1 When f ∈ C 2 (Q) , Q ⊆ R2 , we have gz (t) = (z1 − x01 )

2

2 (z1 − x01 ) (z2 − x02 ) (z2 − x02 )

∂2f (x0 + t (z − x0 )) + ∂x21 ∂2f (x0 + t (z − x0 )) + ∂x1 ∂x2

∂2f (x0 + t (z − x0 )) , ∂x22

2

(13.8)

and so on. Clearly here gz ∈ C n ([0, 1]) , hence gz ∈ AC n ([0, 1]) . Let now ν > 0 with ν = n. By applying (13.3) for gz we ﬁnd gz (1) =

n−1 l=0

(l)

gz (0) 1 + l! Γ (ν)

1

ν−1

(1 − t) 0

ν D∗0 gz (t) dt.

(13.9)

Here we observe by (13.2) that ν D∗0 gz (t) =

1 Γ (n − ν)

t

n−ν−1

(t − s)

gz(n) (s) ds.

(13.10)

0

Let us consider the case of 0 < ν ≤ 1; that is, n = 1. Then t 1 −ν ν D∗0 gz (t) = (t − s) gz (s) ds Γ (1 − ν) 0 (13.7)

=

k

1 Γ (1 − ν) =

k i=1

1 Γ (1 − ν)

t

−ν

(t − s) 0

∂f (zi − x0i ) (x0 + s (z − x0 )) ds = ∂xi i=1

k

t

(zi − x0i )

i=1

(t − s) 0

(zi − x0i ) J01−ν

−ν

∂f (x0 + s (z − x0 )) ds ∂xi

∂f (x0 + t (z − x0 )) . ∂xi

(13.11)

272

13. Multivariate Caputo Fractional Taylor Formula

That is, ν D∗0 gz (t) =

k

(zi − x0i ) J01−ν

i=1

∂f (x0 + t (z − x0 )) , ∂xi

(13.12)

for all t ∈ [0, 1] . Consequently by (13.9) and (13.12) we get 1 1 ν−1 (1 − t) f (z) = f (x0 ) + Γ (ν) 0 k

(zi −

x0i ) J01−ν

i=1

∂f (x0 + t (z − x0 )) ∂xi )

k

1 (zi − x0i ) f (x0 ) + Γ (ν) i=1 J01−ν

1

dt =

ν−1

(1 − t) 0

* ∂f (x0 + t (z − x0 )) dt . ∂xi

(13.13)

Based on the last comments we present the following basic multivariate fractional fundamental theorem. Theorem 13.5. Let Q be a compact and convex subset of Rk , k ≥ 2; z := (z1 , . . . , zk ) , x0 := (x01 , . . . , x0k ) ∈ Q, f ∈ C 1 (Q) , 0 < ν ≤ 1. Then f (z) = f (x0 ) +

k i=1

J01−ν

) (zi − x0i )

1 Γ (ν)

1

ν−1

(1 − t) 0

* ∂f (x0 + t (z − x0 )) dt . ∂xi

We make Remark 13.6. This is a continuation of Remark 13.4. Let now 1 < ν ≤ 2; that is, n = 2, f ∈ C 2 (Q) , Q ⊆ R2 . Then t 1 (13.8) 1−ν ν D∗0 gz (t) = (t − s) gz (s) ds = Γ (2 − ν) 0 1 Γ (2 − ν)

t

1−ν

(t − s) 0

) 2 2 ∂ f (x0 + s (z − x0 )) + (z1 − x01 ) ∂x21

(13.14)

13.2 Results

273

∂2f (x0 + s (z − x0 )) + ∂x1 ∂x2 * 2 2 ∂ f (z2 − x02 ) (x0 + s (z − x0 )) ds = ∂x22

2 (z1 − x01 ) (z2 − x02 )

) (z1 − x01 )

2

1 Γ (2 − ν)

t

(t − s)

1−ν

0

∂2f (x0 + s (z − x0 )) ds ∂x21

*

+2 (z1 − x01 ) (z2 − x02 ) )

1 Γ (2 − ν)

+ (z2 − x02 )

2

t

1−ν

(t − s) 0

1 Γ (2 − ν)

t

(t − s)

J02−ν

+2 (z1 − x01 ) (z2 − x02 )

2

J02−ν

*

∂2f (x0 + s (z − x0 )) ds ∂x22

∂2f (x0 + t (z − x0 )) ∂x21

+ (z2 − x02 )

1−ν

0

2

= (z1 − x01 )

∂2f (x0 + s (z − x0 )) ds ∂x1 ∂x2

(13.15)

∂2f (x0 + t (z − x0 )) ∂x1 ∂x2

2 ∂ f (x + t (z − x )) . J02−ν 0 0 ∂x22

That is, we get

+2 (z1 − x01 ) (z2 − x02 ) 2

(z2 − x02 )

J02−ν

2

J02−ν

ν gz (t) = (z1 − x01 ) D∗0

J02−ν

∂2f (x + t (z − x )) 0 0 ∂x21

∂2f (x0 + t (z − x0 )) ∂x1 ∂x2

∂2f (x0 + t (z − x0 )) ∂x22

(z2 − x02 )

∂f 1 (x0 ) + ∂x2 Γ (ν)

, 0 ≤ t ≤ 1.

∂f (x0 ) + ∂x1

1

ν−1

(1 − t) 0

+

Thus by (13.6), (13.7), (13.9), and (13.16) we obtain f (z) = f (x0 ) + (z1 − x01 )

(13.16)

274

13. Multivariate Caputo Fractional Taylor Formula

) 2 ∂ f 2 2−ν (x0 + t (z − x0 )) + (z1 − x01 ) J0 ∂x21 2 (z1 − x01 ) (z2 − x02 ) + (z2 − x02 )

2

J02−ν

J02−ν

* ∂2f (x0 + t (z − x0 )) dt ∂x22

J02−ν

2

1 Γ (ν)

ν−1

(1 − t) 0

J02−ν

1

∂2f (x0 + t (z − x0 )) ∂x21

+2 (z1 − x01 ) (z2 − x02 )

1 Γ (ν)

J02−ν

2

1 Γ (ν)

dt +

1

ν−1

(1 − t) 0

∂2f (x0 + t (z − x0 )) ∂x1 ∂x2

(z2 − x02 )

(13.17)

∂f ∂f (x0 ) + (z2 − x02 ) (x0 ) + ∂x1 ∂x2

= f (x0 ) + (z1 − x01 ) (z1 − x01 )

∂2f (x0 + t (z − x0 )) ∂x1 ∂x2

1

dt +

ν−1

(1 − t) 0

∂2f (x0 + t (z − x0 )) ∂x22

dt .

(13.18)

We have established the following Caputo fractional bivariate Taylor formula. Theorem 13.7. Let f ∈ C 2 (Q) , Q ⊆ R2 compact and convex, z := (z1 , z2 ), x0 := (x01 , x02 ) ∈ Q, and 1 < ν ≤ 2. Then (1) f (z1 , z2 ) = f (x01 , x02 ) + (z1 − x01 ) (z2 − x02 )

1 Γ (ν)

1

0

∂f 2 (x01 , x02 ) + (z1 − x01 ) ∂x2

ν−1

(1 − t)

∂f (x01 , x02 ) + ∂x1

J02−ν

∂2f (x0 + t (z − x0 )) ∂x21

dt

13.2 Results

+2 (z1 − x01 ) (z2 − x02 )

J02−ν

1 Γ (ν)

1

ν−1

(1 − t) 0

∂2f (x0 + t (z − x0 )) ∂x1 ∂x2

(z2 − x02 )

J02−ν

2

1 Γ (ν)

275

1

dt +

ν−1

(1 − t) 0

∂2f (x0 + t (z − x0 )) ∂x22

dt .

(13.19)

Additionally assume that f (x0 ) =

∂f ∂f (x0 ) = (x0 ) = 0, ∂x1 ∂x2

then (2) f (z1 , z2 ) = (z1 − x01 )

J02−ν

2

J02−ν

1 Γ (ν)

1

0

(z2 − x02 )

J02−ν

1 Γ (ν)

dt

1

ν−1

(1 − t) 0

∂2f (x0 + t (z − x0 )) ∂x1 ∂x2 2

ν−1

(1 − t)

∂2f (x0 + t (z − x0 )) ∂x21

+2 (z1 − x01 ) (z2 − x02 )

1 Γ (ν)

1

dt +

ν−1

(1 − t) 0

∂2f (x0 + t (z − x0 )) ∂x22

dt .

(13.20)

We make Remark 13.8. This is another continuation of Remark 13.4. Let ν > 0, n = ν , f ∈ C n (Q) . By (13.4), (13.6), and (13.9) we get f (z) = f (x0 ) +

k i=1

(zi − x0i )

∂f (x0 ) + ∂xi

276

13. Multivariate Caputo Fractional Taylor Formula

) n−1

"k

(zi − x0i )

i=1

l * f (x0 )

l!

l=2

+

∂ ∂xi

1 Γ (ν)

1

ν−1

(1 − t) 0

ν D∗0 gz (t) dt.

(13.21)

But we have ν D∗0 gz (t) =

1 Γ (n − ν) 1 Γ (n − ν)

' k

(t − s)

= J0n−ν

n−ν−1

gz(n) (s) ds =

0

t

n−ν−1

(t − s) 0

∂ (zi − x0i ) ∂x i i=1

%' k

t

n ( f (x0 + s (z − x0 )) ds

∂ (zi − x0i ) ∂x i i=1

n (

-

f (x0 + t (z − x0 )) .

(13.22)

We have proved the following general multivariate Caputo fractional Taylor formula. Theorem 13.9. Let ν > 0, n = ν , f ∈ C n (Q) , where Q is a compact and convex subset of Rk , k ≥ 2; z := (z1 , . . . , zk ) , x0 := (x01 , . . . , x0k ) ∈ Q. Then k ∂f (x0 ) 1) f (z) = f (x0 ) + (zi − x0i ) + ∂xi i=1 ) n−1

"k

∂ i=1 (zi − x0i ) ∂xi

l!

l=2

+ %' k

l * f (x0 )

1 Γ (ν)

∂ (zi − x0i ) ∂x i i=1

1

ν−1

(1 − t)

n−ν J0

0

n ( f (x0 + t (z − x0 ))

-( dt.

(13.23)

13.2 Results

277

Additionally assume that fα (x0 ) = 0, α := (α1 , . . . , αk ) , αi ∈ Z+ , i = "k 1, . . . , k; |α| := i=1 αi =: r, r = 0, . . . , n − 1; then 2) f (z) = %' k

∂ (zi − x0i ) ∂x i i=1

1 Γ (ν)

1

ν−1

(1 − t)

J0n−ν

0

n (

-(

f (x0 + t (z − x0 ))

dt =: Rν .

(13.24)

We continue with Remark 13.10. Here we estimate the remainder of (13.23) , which is the same as Rν of (13.24). The function %' k n ( ∂ (zi − x0i ) f Gν (t) := J0n−ν ∂xi i=1 (x0 + t (z − x0 ))} , t ∈ [0, 1] ,

(13.25)

which appears in Rν is continuous; see Proposition 114 of [45] and Rν ∈ R. Similarly the remainder of (13.14) exists and the same holds for all the integrals of (13.19); they are all real numbers. So we can write 1 1 ν−1 (1 − t) Gν (t) dt, ν > 0. (13.26) Rν = Γ (ν) 0 When ν ≥ 1 we obtain 1 |Rν | ≤ Γ (ν) 1 Γ (ν)

1

1

|Gν (t)| dt = 0

ν−1

(1 − t)

|Gν (t)| dt ≤

0

1 Gν L1 (0,1) ; Γ (ν)

That is, |Rν | ≤

1 Gν L1 (0,1) . Γ (ν)

(13.27)

Also for p, q > 1 : 1/p + 1/q = 1 and with p (ν − 1) + 1 > 0 for ν > 0, we derive 1 1 ν−1 |Rν | ≤ (1 − t) |Gν (t)| dt ≤ Γ (ν) 0

278

13. Multivariate Caputo Fractional Taylor Formula

1 Γ (ν)

1

ν−1

(1 − t)

1/p

p

q

|Gν (t)| dt

dt

0

1/q

1

=

0

1 1 Gν Lq ([0,1]) . Γ (ν) (p (ν − 1) + 1)1/p That is, |Rν | ≤

1 1 Gν Lq ([0,1]) . Γ (ν) (p (ν − 1) + 1)1/p

(13.28)

In the case of p = q = 2 and ν > 1/2 we have 1 1 √ Gν L2 ([0,1]) . Γ (ν) 2ν − 1

|Rν | ≤

(13.29)

Finally we get that |Rν | ≤

1 Γ (ν)

1

ν−1

(1 − t)

|Gν (t)| dt ≤

0

that is, |Rν | ≤

Gν ∞ ; Γ (ν + 1)

Gν ∞ , ν > 0. Γ (ν + 1)

(13.30)

We have established the following remainder estimate. Theorem 13.11. All here are as in Theorem 13.9. Let Rν be the remainder in (13.23) , and Gν (t) , t ∈ [0, 1] as in (13.25) , ν ≥ 1. Then % Gν Lq ([0,1]) Gν L1 ([0,1]) |Rν | ≤ min , , 1/p Γ (ν) Γ (ν) (p (ν − 1) + 1) Gν L2 ([0,1]) Gν ∞ √ , Γ (ν) 2ν − 1 Γ (ν + 1) where p, q > 1 : 1/p + 1/q = 1.

,

(13.31)

14 Canavati Fractional Multivariate Opial-Type Inequalities on Spherical Shells

Here we introduce the concept of multivariate Canavati fractional diﬀerentiation especially of the fractional radial diﬀerentiation, by extending the univariate deﬁnition of [101]. Then we present Opial-type inequalities over compact and convex subsets of RN , N ≥ 2 , mainly over spherical shells, studying the problem in all possibilities. Our results involve one, two, or more functions. This treatment is based on [44].

14.1 Introduction This chapter is motivated by the articles of Opial [315], Beesack [80], and Anastassiou [15, 17–19, 26, 27, 61]. We would like to mention Theorem 14.1. (Opial [315, 1960]). Let c > 0 , and y(x) be real, continuously diﬀerentiable on [0, c], with y(0) = y(c) = 0 . Then 0

c

c |y(x)y (x)|dx ≤ 4

c

(y (x))2 dx.

(14.1)

0

Equality holds for the function y(x) = x on [0, c/2], and y(x) = c − x on [c/2, c]. The next result implies Theorem 14.1 and is often used in applications. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 279 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 14,

280

14. Canavati Fractional Multivariate Opial Inequalities

Theorem 14.2. (Beesack [80,1962]). Let b > 0. If y(x) is real, continuously diﬀerentiable on [0, b], and y(0) = 0 then

b

|y(x)y (x)|dx ≤

0

b 2

b

(y (x))2 dx.

(14.2)

0

Equality holds only for y = mx where m is a constant. We describe here our speciﬁc multivariate setting. Let the balls B(0, R1 ), B(0, R2 ); 0 < R1 < R2 . Here B(0, R) := {x ∈ RN : |x| < R} ⊆ RN , N ≥ 2, R > 0 , and the sphere S N −1 := {x ∈ RN : |x| = 1}, where | · | is the Euclidean norm. Let dω be the element of surface measure on S N −1 and let ω N = S N −1 dω = 2π N/2 /Γ(N/2). For x ∈ RN − {0} we can write uniquely x = rω , where r = |x| > 0, and ω = x/r ∈ S N −1 , |ω| = 1. Let the spherical shell A := B(0, R2 ) − B(0, R1 ) . We have that V ol(A) = ω N (R2N − R1N )/N . Indeed A¯ = [R1 , R2 ] × S N −1 . ¯ it holds For F ∈ C(A) R2

F (x)dx = A

S N −1

F (rω)rN −1 dr dω;

(14.3)

R1

we often use this formula here. In this chapter we present a series of various fractional multivariate Opial-type inequalities over spherical shells and arbitrary domains. Opialtype inequalities ﬁnd applications in establishing the uniqueness of solution of initial value problems for diﬀerential equations and their systems; see [406].

14.2 Results We make Remark 14.3. We introduce here the partial Canavati type fractional derivatives. Let f : [0, 1]2 → R. Let ν > 0 , n := [ν] , α := ν − n, 0 < α < 1; μ > 0, m := [μ], β := μ − m, 0 < β < 1. Assume ∃ ∂ m+n f (t, s)/∂xn ∂y m ∈ C([0, 1]2 ); then (x − t)−α (y − s)−β ∂ m+n f (t, s)/ ∂xn ∂y m is integrable over [0, x] × [0, y]; x, y ∈ [0, 1]; that is, x y ∂ m+n f (t, s) F (x, y) := (x − t)−α (y − s)−β dt ds (14.4) ∂xn ∂y m 0 0 is real-valued. Thus, by Fubini’s theorem, the order of integration in (14.4) does not matter.

14.2 Results

281

Let now g ∈ C([0, 1]); we deﬁne ∞ the Riemann–Liouville integral ; Γ is the gamma function: Γ(ν) := 0 e−t tν−1 dt, as 1 (Jν g)(x) := Γ(ν)

x

(x − t)ν−1 g(t)dt, 0 ≤ x ≤ 1.

(14.5)

0

We consider here the space C ν ([0, 1]) := {g ∈ C n ([0, 1]) : J1−α g (n) ∈ C 1 ([0, 1])};

(14.6)

then the ν – fractional derivative of g is deﬁned by g (ν) := (J1−α g (n) ) ; see [101]. We assume here f (· , y) ∈ C ν ([0, 1]), ∀ y ∈ [0, 1]; then we deﬁne the ν-partial fractional derivate of f with respect to x : ∂f ν (· , y)/∂xν as ∂ ν f (x, y) ∂ n f (x, y) ∂ := (14.7) , ∀ (x, y) ∈ [0, 1]2 . J1−α ∂xν ∂x ∂xn Also, we assume f (x, ·) ∈ C μ ([0, 1]), ∀ x ∈ [0, 1] , where C μ ([0, 1]) := {g ∈ C m ([0, 1]) : J1−β g (m) ∈ C 1 ([0, 1])}.

(14.8)

Then we deﬁne the μ – partial fractional derivative of f with respect to y: ∂f μ /∂y μ (x, ·) as ∂f μ (x, y) ∂mf ∂ := (14.9) J1−β m (x, y) , ∀ (x, y) ∈ [0, 1]2 . ∂y μ ∂y ∂y Deﬁne the space # C ν+μ ([0, 1]2 ) := f ∈ C n+m ([0, 1]2 ) : J1−α J1−β

∂ n f (·, y) ∂xn

∂ m f (x, ·) ∂xm

∈ C 1 ([0, 1]), ∀ y ∈ [0, 1]; ∈ C 1 ([0, 1]), ∀ x ∈ [0, 1];

$ ∃ Fx , Fy , Fyx ∈ C([0, 1]2 ) .

(14.10)

Deﬁne the mixed Canavati fractional partial derivative: ∂ ν+μ f (x, y) := ∂xν ∂y μ x y 1 ∂2 ∂ n+m f (t, s) (x − t)−α (y − s)−β dt ds. Γ(1 − α)Γ(1 − β) ∂x∂y 0 0 ∂xn ∂y m (14.11)

282

14. Canavati Fractional Multivariate Opial Inequalities

One can have anchor points x0 , y0 = 0; then all the above deﬁnitions go through for x ≥ x0 , y ≥ y0 . Conclusion 1: Clearly then we have Fxy = Fyx , and ∂ ν+μ f ∂ μ+ν f = , ∂xν ∂y μ ∂y μ ∂xν

(14.12)

so the order of fractional diﬀerentiation is immaterial. Here, it is by deﬁnition ∂ μ+ν f (x, y) := ∂y μ ∂xν ∂2 1 Γ(1 − α) Γ(1 − β) ∂y∂x

0

x

y

(x − t)−α (y − s)−β

0

∂ m+n f (t, s) dt ds. ∂y m ∂xn (14.13)

Comment: (1) Let ν = 0; then n = α = 0, and (14.11) becomes

∂2 ∂ μ f (x, y) 1 = μ ∂y Γ(1 − β) ∂x∂y

0

x

0

y

(y − s)−β

∂ m f (t, s) dt ds = ∂y m

x

y ∂2 ∂ m f (t, s) 1 (y − s)−β dt ds = Γ(1 − β) ∂y∂x 0 0 ∂y m x y m 1 ∂ ∂ −β ∂ f (t, s) (y − s) ds dt =: (). Γ(1 − β) ∂y ∂x 0 ∂y m 0 (14.14) Notice for ﬁxed y we have that (y − s)−β ∂ m f (t, s)/∂y m is integrable over [0, y], so the function y ∂ m f (t, s) ϕ(t) := (y − s)−β ds (14.15) ∂y m 0

is real-valued for any t ∈ [0, x]. By continuity of ∂ m f /∂y m we have true that ∀ ε > 0 ∃ δ > 0 : whenever |t1 − t2 | < δ we have |∂ m f (t1 , s)/∂y m − ∂ m f (t2 , s)/∂y m | < ε . We further have m y ∂ f (t1 , s) ∂ m f (t2 , s) (y − s)−β − ϕ(t1 ) − ϕ(t2 ) = ds. ∂y m ∂y m 0 Hence |ϕ(t1 ) − ϕ(t2 )| ≤

y

−β

(y − s) 0

m ∂ f (t1 , s) ∂ m f (t2 , s) ds ≤ − ∂y m ∂y m

14.2 Results

ε 0

y

(y − s)−β ds =

ε y 1−β , 1−β

283

(14.16)

proving ϕ(t) continuous. Consequently y 1 ∂ ∂ m f (x, s) ∂ μ f (x, y) () = (y − s)−β ds =: . m Γ(1 − β) ∂y ∂y ∂y μ 0 (14.17) Conclusion 2: When ν = 0, the fractional mixed partial derivative collapses to the single fractional partial derivative. (2) Let μ = 0; then m = β = 0 , and (14.11) becomes x y n ∂2 1 ∂ ν f (x, y) −α ∂ f (t, s) = (x − t) dt ds = ∂xν Γ(1 − α) ∂x∂y 0 0 ∂xn y x ∂ ∂ ∂ n f (t, s) 1 (x − t)−α dt ds (14.18) Γ(1 − α) ∂x ∂y ∂xn 0 0 x ∂ n f (t, s) (notice (x − t)−α dt is continuous in s ∈ [0, y]) = ∂xn 0 x n 1 ∂ ∂ ν f (x, y) −α ∂ f (t, y) (x − t) dt =: . (14.19) Γ(1 − α) ∂x ∂xn ∂xν 0 Conclusion 3: When μ = 0, the mixed fractional derivative collapses again to the single one. (3) Let now n = ν ∈ N (i.e., α = 0); then x n ∂ f (t, y) ∂ ν f (x, y) ∂ ∂ n f (x, y) = dt = , (14.20) ν n ∂x ∂x ∂x ∂xn 0 the ordinary one. (4) When m = μ ∈ N (i.e., β = 0), then y m ∂ μ f (x, y) ∂ f (x, s) ∂ ∂ m f (x, y) = ds = , ∂y μ ∂y 0 ∂y m ∂y m

(14.21)

the ordinary one. (5) Furthermore, let ﬁnally both ν = n ∈ N and μ = m ∈ N (i.e., α = β = 0). Then x y n+m ∂ f (t, s) ∂2 ∂ ν+μ f (x, y) = dt ds = ∂xν ∂y μ ∂x∂y 0 0 ∂xn ∂y m y x n+m ∂ f (t, s) ∂ ∂ dt ds ∂x ∂y ∂xn ∂y m 0 0

284

14. Canavati Fractional Multivariate Opial Inequalities

x n+m ∂ f (t, y) ∂ ∂ n+m f (x, y) = dt = , (14.22) ∂x ∂xn ∂y m ∂xn ∂y m 0 proving that the fractional mixed partial collapses to the ordinary one. Fractional diﬀerentiation is a linear operation. Conclusion 4: The above deﬁnitions we gave for the fractional partial derivatives are natural extensions of the ordinary positive integer ones. Having introduced the fractional partial derivatives we are ready to develop our Opial-type results. We make Remark 14.4. First we consider a general domain. Let Q be a compact and convex subset of RN , N ≥ 2; z := (z1 , ..., zN ), x0 := (x01 , ...x0N ) ∈ Q be ﬁxed. Let f ∈ C n (Q), n ∈ N. Set gz (t) = f (x0 + t(z − x0 )), 0 ≤ t ≤ 1; gz (0) = f (x0 ), gz (1) = f (z). Then it holds

⎡ j ⎤ N ∂ gz(j) (t) = ⎣ (zi − x0i ) f ⎦ (x0 + t(z − x0 )), ∂x i i=1

(14.23)

j = 0, 1, 2, . . . , n, and in particular gz (t) =

N i=1

(zi − x0i )

∂f (x0 + t(z − x0 )), ∂xi

(14.24)

0 ≤ t ≤ 1. Clearly here gz ∈ C n ([0, 1]). Let ﬁrst 1 ≤ ν < 2; in that case we take n := [ν] = 1. Following [40] and by assuming that as a function of t : (ν) fxi (x0 + t(z − x0 )) ∈ C ν−1 ([0, 1]), i = 1, . . . , N, then there exists gz = (J2−ν gz ) , and it holds (ν−1) N ∂f (ν) gz (t) = (zi − x0i ) (x0 + t(z − x0 )) , (14.25) ∂xi i=1 0 ≤ t ≤ 1. Also here we have "N (zi − x0i ) t (J2−ν gz )(t) = i=1 (t − s)1−ν fxi (x0 + s(z − x0 ))ds, (14.26) Γ(2 − ν) 0 0 ≤ t ≤ 1. Clearly (J2−ν gz )(t) ∈ C 1 ([0, 1]) and (J2−ν gz )(0) = 0. Therefore by (14.2) we get s s s ν 2 |J2−ν gz (t)| |Dν gz (t)| dt ≤ (D gz (t)) dt, ∀ s ∈ [0, 1]. (14.27) 2 0 0

14.2 Results

285

We have established the following Opial-type result. Theorem 14.5. Let Q be a compact and convex subset of RN , N ≥ 2; z, x0 ∈ Q be ﬁxed; 1 ≤ ν < 2. Let f ∈ C 1 (Q). Assume that as a function of t : fxi (x0 + t(z − x0 )) ∈ C ν−1 ([0, 1]), i = 1, . . . , N . Then t s N 1 1−ν (t − s) fxi (x0 + s(z − x0 ))ds (zi − x0i ) Γ(2 − ν) 0 0 i=1

N (ν−1) (zi − x0i )(fxi (x0 + t(z − x0 ))) dt ≤ i=1

2 N s s (ν−1) (zi − x0i )(fxi (x0 + t(z − x0 ))) dt, 2 0 i=1

(14.28)

∀ s ∈ [0, 1]. Remark 14.6. (Continuation) Let here ν ≥ 2 and n := [ν], β := ν −n. We assume that as functions of t : fα (x0 + t(z − x0 )) ∈ C (ν−n) ([0, 1]), for "N all α := (α1 , . . . , αk ), αi ∈ Z+ , i = 1, . . . , N ; |α| := i=1 αi = n. Clearly (ν) (n) then there exists gz = (J1−β gz ) , and it holds gz(ν) (t)

=

' N

∂ (zi − x0i ) ∂xi i=1

n ((ν−n) f

(x0 + t(z − x0 )),

(14.29)

all t ∈ [0, 1]. Of course, it holds (J1−β gz(n) )(t) %' N

∂ (zi − x0i ) ∂xi i=1

(14.23)

=

n (

1 Γ(1 − β)

t

(t − s)−β

0

-

f (x0 + s(z − x0 )) ds.

(14.30)

(n)

Notice (J1−β gz )(0) = 0. Hence again by (14.2) we get s s s ν |J1−β gz(n) (t)| |Dν gz (t)|dt ≤ (D gz (t))2 dt, ∀ s ∈ [0, 1]. (14.31) 2 0 0 We have proved the following general Opial-type result. Theorem 14.7. Let Q be a compact and convex subset of RN , N ≥ 2; z, x0 ∈ Q be ﬁxed ; ν ≥ 2, n := [ν], β := ν − n . Let f ∈ C n (Q).

286

14. Canavati Fractional Multivariate Opial Inequalities

Assume that as a function of t : fα (x0 + t(z − x0 )) ∈ C (ν−n) ([0, 1]), for "N all α := (α1 , ..., αk ), αi ∈ Z+ , i = 1, . . . , N ; |α| := i=1 αi = n . Then 1 Γ(1 − β)

s t

0

0

−β

(t − s)

⎫ ⎧ ⎡ n ⎤ N ⎬ ⎨ ∂ ⎣ (zi − x0i ) f ⎦ (x0 + s(z − x0 )) ds ⎭ ⎩ ∂xi i=1

' n ((ν−n) N ∂ (zi − x0i ) f (x0 + t(z − x0 )) dt ≤ ∂x i i=1 ⎫2 ⎧' n ((ν−n) N ⎬ s s⎨ ∂ (zi − x0i ) f (x0 + t(z − x0 )) dt, ⎭ 2 0 ⎩ ∂xi

(14.32)

i=1

∀ s ∈ [0, 1]. Note. Following the last pattern one can transfer any univariate Opialtype inequality(see [4]) into this fractional multivariate general setting. Inasmuch as no chain rule is valid in the fractional diﬀerentiation, inequalities such as (14.31), (14.32) are not revealing themselves to totally decompose into all of their ingredients. Next, working over spherical shells we obtain a series of various Opial-type fractional multivariate inequalities that look nice and are very clear. We give Deﬁnition 14.8. (see [17; 19, p. 540]) In the following we carry earlier notions introduced in Remark 14.3 over to arbitrary [a, b] ⊆ R. Let x, x0 ∈ [a, b] such that x ≥ x0 , x0 is ﬁxed. Let f ∈ C([a, b]) and deﬁne x 1 (Jνx0 f )(x) := (x − t)ν−1 f (t)dt, x0 ≤ x ≤ b, (14.33) Γ(ν) x0 the generalized Riemann–Liouville integral. We consider the subspace Cxν0 ([a, b]) of C n ([a, b]): x0 Cxν0 ([a, b]) := {f ∈ C n ([a, b]) : J1−α

f (n) ∈ C 1 ([x0 , b])}

(14.34)

Hence, let f ∈ Cxν0 ([a, b]); we deﬁne the generalized ν - fractional derivative of f over [x0 , b] as x0 f (n) . (14.35) Dxν0 f := J1−α Notice that x0 J1−α f (n) (x) =

1 Γ(1 − α)

x

x0

(x − t)−α f (n) (t)dt

(14.36)

14.2 Results

287

exists for f ∈ Cxν0 ([a, b]). Next we use Theorem 14.9. [15; 19, p. 567] Let γ i ≥ 0, ν ≥ 1 such that ν − γ i ≥ 1; i = 1, . . . , l and f ∈ Cxν0 ([a, b]) with f (j) (x0 ) = 0, j = 0, 1, . . . , n − 1, n := [ν]. Here x, x0 ∈ [a, b] : x ≥ x0 . Let q1 , q2 > 0 be continuous "l functions on [a, b] and ri > 0 : i=1 ri = r. Let s1 , s1 > 1 : 1/s1 +1/s1 = 1 and s2 , s2 > 1 : 1/s2 + 1/s2 = 1 and p > s2 . Furthermore suppose that

x

Q1 :=

s1

1/s1

(q1 (ω)) dω

< +∞

(14.37)

x0

and

x

Q2 :=

(q2 (ω))−s2 /p dω

r/s2 < +∞.

(14.38)

x0

Call σ := p − s2 /ps2 . Then

x

q1 (ω) x0

·

l i=1

l

|(Dxγ0i (f ))(ω)|ri dω ≤ Q1 Q2

i=1

σ ri σ ri (Γ(ν − γ i )) (ν − γ i − 1 + σ)ri σ

"l

(x − x0 )( i=1 (ν−γ i −1)ri +σr+1/s1 ) · "l (( i=1 (ν − γ i − 1)ri s1 ) + rs1 σ + 1)1/s1 x r/p ν p · q2 (ω)|(Dx0 f )(ω)| dω .

(14.39)

x0

We next work in the setting of spherical shells introduced in Section 14.1, Introduction. We need ¯ and Deﬁnition 14.10. Let ν > 0, n := [ν], α := ν − n, f ∈ C n (A), ν ν ¯ A is a spherical shell. Assume that there exists ∂R1 f (x)/∂r ∈ C(A), given by r ν n f (x) ∂R ∂ 1 −α ∂ f (tω) 1 := (r − t) dt , (14.40) ∂rν Γ(1 − α) ∂r ∂rn R1 ¯ that is, x = rω, r ∈ [R1 , R2 ], and ω ∈ S N −1 . where x ∈ A; ν We call ∂R1 f /∂rν the radial fractional derivative of f of order ν .

288

14. Canavati Fractional Multivariate Opial Inequalities

We need ¯ Lemma 14.11. Let γ ≥ 0, ν ≥ 1 such that ν − γ ≥ 1. Let f ∈ C n (A) ν ν ¯ x ∈ A, ¯ A a spherical shell. Further f (x)/∂r ∈ C( A), and there exists ∂R 1 assume that ∂ j f (R1 ω)/∂rj = 0, j = 0, 1, . . . , n − 1, n := [ν], ∀ ω ∈ S N −1 . γ ¯ f (x)/∂rγ ∈ C(A). Then there exists ∂R 1 ν Proof. The assumption implies that ∂R f (rω)/∂rν ∈ C([R1 , R2 ]), ∀ω ∈ 1 ν S N −1 ; that is, f (rω) ∈ CR ([R1 , R2 ], ∀ ω ∈ S N −1 . Following [17], and [19, 1 γ f (r ω)/∂rγ , given by pp. 544–545], we get that there exists ∂R 1 γ f (r ω) ∂R 1 1 = ∂rγ Γ(ν − γ)

r

(r − t)ν−γ−1 R1

ν ∂R f (tω) 1 dt; ∂rν

(14.41)

γ indeed f (r ω) ∈ CR ([R1 , R2 ]), ∀ ω ∈ S N −1 . 1 Hence R2 γ ∂R f (r ω) ∂ ν f (t ω) 1 ν−γ−1 R1 1 = X (t)(r − t) dt. (14.42) [R ,r] 1 ∂rγ Γ(ν − γ) R1 ∂rν

Let rn → r, ω n → ω; then X[R1 ,rn ] (t) → X[R1 ,r] (t), a.e.; also (rn − t)ν−γ−1 → (r − t)ν−γ−1 , and ν ν ∂R ∂R f (t ω n ) f (t ω) 1 1 → . ∂rν ∂rν

Furthermore it holds that X[R1 ,rn ] (t)(rn − t)ν−γ−1 X[R1 ,r] (t)(r − t)ν−γ−1

ν ∂R f (t ω) 1 , ∂rν

However we have X[R1 ,rn ] (t)|rn − t|

ν−γ−1

ν ∂R f (t ω n ) 1 −→ ∂rν

a.e. on [R1 , R2 ].

(14.43)

ν ∂R1 f (t ω n ) ≤ ∂rν

ν ∂R1 f (R2 − R1 )ν−γ−1 ∂rν < ∞. ∞

(14.44)

Thus, by the dominated convergence theorem we obtain R2 ∂ ν f (t ω n ) X[R1 ,rn ] (t)(rn − t)ν−γ−1 R1 ν dt → ∂r R1

R2

R1

X[R1 ,r] (t)(r − t)ν−γ−1

ν ∂R f (t ω) 1 dt, ∂rν

(14.45)

14.2 Results

proving the claim.

289

We present the following very general result. Theorem 14.12. Let γ i ≥ 0, ν ≥ 1,such that ν − γ i ≥ 1; i = ¯ and there exists ∂ ν f (x)/∂rν ∈ 1, . . . , l, n := [ν]. Let f ∈ C n (A) R1 ¯ ¯ C(A), x ∈ A; A is a spherical shell: A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2. Furthermore assume that ∂ j f /∂rj , j = 0, 1, . . . , n−1, vanish "l on ∂B(0, R1 ). Let ri > 0 : i=1 ri = p. Let s1 , s1 > 1 : 1/s1 + 1/s1 = 1, and s2 , s2 > 1 : 1/s2 + 1/s2 = 1, and p > s2 . Denote Q1 = and

⎛ Q2 =

(N −1)s1 +1

(N −1)s1 +1

− R1 (N − 1)s1 + 1

R2

(1−N )s2 /p+1

⎝ R2

1/s1 ,

(1−N )s2 /p+1

− R1 s

(1 − N ) p2 + 1

(14.46)

⎞p/s2 ⎠

.

(14.47)

Call σ := p − s2 /p s2 . Also call C := Q1 Q2

l i=1

" Then

(R2 − R1 )( l i=1 (ν

σ ri σ (Γ(ν − γ i ))ri (ν − γ i − 1 + σ)ri σ

"l

p i=1 (ν−γ i −1)ri + s2

+ s1 −1) 1

1/s1 . − γ i − 1)ri s1 + s1 ( sp2 − 1) + 1

ri ν l γi ∂R1 f (x) p ∂R1 f (x) dx ≤ C ∂rν dx. ∂rγ i A

(14.48)

(14.49)

A i=1

ν Proof. The assumption implies that f (r ω) ∈ C n ([R1 , R2 ]), ∂R f (r ω)/ 1 ν N −1 ∂r ∈ C([R1 , R2 ]), ∀ ω ∈ S . By Theorem 14.9 we have ν R2 R2 l γi ri ∂ f (r ω) p ∂R1 f (r ω) N −1 R1 dr, ∀ ω ∈ S N −1 . rN −1 dr ≤ C r ∂rν γi ∂r R1 R1 i=1 (14.50) Therefore it holds ri l γi R2 ∂R1 f (r ω) N −1 r dr d ω ≤ ∂rγ i S N −1 R1 i=1

290

14. Canavati Fractional Multivariate Opial Inequalities

R2

C

r S N −1

N −1

R1

ν ∂R1 f (r ω) p dr d ω. ∂rν

Using Lemma 14.11 and by (14.3) we derive (14.49).

(14.51)

We mention Theorem 14.13. [15; 19, p. 573] Let γ i ≥ 0, ν ≥ 1 such that ν − γ i ≥ 1; i = 1, . . . , l and f ∈ Cxν0 ([a, b]) with f (j) (x0 ) = 0, j = 0, 1..., n − 1, n := [ν]. Here x, x0 ∈ [a, b] : x ≥ x0 . Let q˜(w) ≥ 0 continuous on [a,b] "l and ri > 0 : i=1 ri = r. Then

x

q˜(w) · x0

% ≤

˜ q ∞ (Dxν0 f ∞ )r

&l

i=1 (Γ(ν

− γ i + 1))ri

l

(|Dxν0 f |(w))ri dw

i=1

- % ·

"l

(x − x0 )rν− i=1 ri γ i +1 "l (rν − i=1 ri γ i + 1)

.

(14.52)

We give Theorem 14.14. Let γ i ≥ 0, ν ≥ 1, such that ν − γ i ≥ 1; i = ¯ and there exists ∂ ν f (x)/∂rν ∈ 1, . . . , l, n := [ν]. Let f ∈ C n (A) R1 ¯ ¯ C(A), x ∈ A; A is a spherical shell: A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2. Furthermore assume that ∂ j f /∂rj , j = 0, 1, . . . , n − 1, vanish on "l ∂B(0, R1 ). Let ri > 0 : i=1 ri = r. Call "l

R2N −1 (R2 − R1 )rν−

M := &l

ri γ i +1

"l

> 0.

(14.53)

ri ν r l γi ∂ f 2π N/2 ∂R1 f (x) R1 . dx ≤ M γ i ∂r Γ(N/2) ∂rν ∞,A¯ A i=1

(14.54)

i=1 (Γ(ν

Then

i=1

− γ i + 1))ri (rν −

i=1 ri γ i

+ 1)

Proof. By Theorem 14.13 we get that r l γ ν r R2 ∂Ri f (rω) i ∂R1 f N −1 1 r dr ≤ M ∂rν ¯ . γ i ∂r R1 ∞,A

(14.55)

i=1

Hence it holds R2

r

S N −1

R1

N −1

ri l γi ∂R1 f (r ω) dr d ω ≤ M ω N ∂rγ i

i=1

ν r ∂R1 f ∂rν ¯ . ∞,A (14.56)

14.2 Results

Using (14.3) and Lemma 14.11, we establish the claim.

291

We need Theorem 14.15. [61]. Let γ ≥ 0, ν ≥ 1, ν −γ ≥ 1, α, β > 0, r > α, r > 1; let p > 0, q ≥ 0 be continuous functions on [a, b]. Let f ∈ Cxν0 ([a, b]) with f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Let x, x0 ∈ [a, b] with x ≥ x0 . Then x q(w)|Dxγ0 f (w)|β |Dxν0 f (w)|α dw ≤ K(p, q, γ, ν, α, β, r, x, x0 ) x0

x

· x0

r p(w) Dxν0 f (w) dw

Here

(α+β/r)

α α+β

.

(14.57)

α/r

1 (Γ(ν − γ))β x r−α/r · (q(w))r · (p(w))−α )1/(r−α) · (P1 (w))(β(r−1)/r−α) · dw , K(p, q, γ, ν, α, β, r, x, x0 ) :=

·

x0

(14.58) with

w

P1 (w) :=

(p(t))−1/r−1 · (w − t)(ν−γ−1)(r/r−1) · dt.

(14.59)

x0

We present Theorem 14.16. Let γ ≥ 0, ν ≥ 1, n := [ν], ν − γ ≥ 1, α, β > 0, α + ¯ and there exists ∂ ν f (x)/∂rν ∈ C(A), ¯ x ∈ A; ¯ A is β > 1. Let f ∈ C n (A) R1 a spherical shell: A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2. Furthermore assume that ∂ j f /∂rj = 0, for j = 0, 1, . . . , n − 1, on ∂B(0, R1 ). Then ν α γ ν ∂R1 f (x) β ∂R ∂R1 f (x) α+β f (x) 1 dx ≤ K dx. ∂rγ ∂rν ∂rν A A

(14.60)

Here K=

α α+β

α/(α+β)

1 (Γ(ν − γ))β

R2

β/(α+β) rN −1 (P1 (r))(a+β−1) dr ,

R1

(14.61) with

r

t(1−N )/(a+β−1) (r − t)(ν−γ−1)(α+β)/(α+β−1) dr.

P1 (r) := R1

(14.62)

292

14. Canavati Fractional Multivariate Opial Inequalities

Proof. The assumption implies that f (rω) ∈ C n ([R1 , R2 ]) and ν ∂R f (r ω)/∂rν ∈ C([R1 , R2 ]), ∀ ω ∈ S N −1 . Hence by Theorem 14.15, 1 ∀ ω ∈ S N −1 , we get that γ ν α ν R2 R2 ∂ f (r ω) β ∂R ∂ f (r ω) α+β f (r ω) N −1 R1 N −1 R1 1 r r dr. ∂rγ ∂rν dr ≤ K ∂rν R1 R1 (14.63) Therefore it holds γ β ν α R2 ∂ f (r ω) ∂ f (r ω) R R 1 dr dω rN −1 1 γ ∂r ∂rν S N −1 R1

R2

r

=K S N −1

N −1

R1

ν ∂R1 f (r ω) α+β dr d ω . ∂rν

Using Lemma 14.11 and by (14.3) we derive (14.60).

(14.64)

We need Theorem 14.17. [61]. Let ν ≥ 1, α, β > 0, r > α, r > 1; let p > 0, q ≥ 0 be continuous functions on [a, b]. Let f ∈ Cxν0 ([a, b]) with f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Let x, x0 ∈ [a, b] with x ≥ x0 . Then x α β g(w) |f (w)| Dxν f (x) dw ≤ K ∗ (p, q, ν, α, β, r, x, x0 ) 0

x0

x

· x0

r p(w) Dxν0 f (w) dw

Here

∗

K (p, q, ν, α, β, r, x, x0 ) :=

x

·

(α+β/r)

α α+β

.

(14.65)

α/r · 1/(Γ(ν))β

((q(w))r · (p(w))−α )1/(r−α) · (P1∗ (w))(β(r−1)/r−α) · dw

(r−α)/r ,

x0

(14.66) with P1∗ (w)

w

:=

(p(t))−1/r−1 · (w − t)(ν−1)(r/(r−1)) dt.

(14.67)

x0

Based on Theorem 14.17 we give similarly: Theorem 14.18. Let ν ≥ 1, n := [ν], α, β > 0, α + β > 1. Let ¯ and there exists ∂ ν f (x)/∂rν ∈ C(A), ¯ x ∈ A; ¯ A is a spherical f ∈ C n (A) R1

14.2 Results

293

shell : A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2. Furthermore assume that ∂ j f /∂rj = 0, for j = 0, 1, . . . , n − 1, on ∂B(0, R1 ). Then ν α ν ∂R1 f (x) α+β β ∂R1 f (x) ∗ |f (x)| dx ≤ K dx. (14.68) ∂rν ∂rν A A Here

∗

K :=

α α+β

(α/α+β)

1 (Γ(ν))β

(β/(α+β))

R2

r

N −1

(P1∗ (r))(α+β−1) dr

,

R1

(14.69) with P1∗ (r) :=

r

t(1−N /α+β−1) (r − t)(α+β)(ν−1)/(α+β−1) dt.

(14.70)

R1

Next we present a set of multivariate fractional Opial-type inequalities involving two functions over the shell. We need Theorem 14.19. [26]. Let ν ≥ 1, γ 1 , γ 2 ≥ 0, such that 1, ν − γ 2 ≥ 1, and f1 , f2 ∈ Cxν0 ([a, b]) with (i)

(i)

f1 (x0 ) = f2 (x0 ) = 0,

i = 0, 1, . . . , n − 1,

n := [ν].

ν − γ1 ≥ (14.71)

Here, x, x0 , ∈ [a, b] : x ≥ x0 . Consider also p(t) > 0, and q(t) ≥ 0 continuous functions on [x0 , b]. Let λν > 0 and λα , λβ ≥ 0, such that λν < p, where p > 1. Set

w

Pk (w) :=

(w − t)(ν−γ k −1)p/(p−1) (p(t))−1/(p−1) dt,

k = 1, 2,

x0 ≤ w ≤ b,

x0

(14.72)

A(w) :=

q(w) · (P1 (w))λα ((p−1)/p) · (P2 (w))λβ ((p−1)/p) (p(w))−λν /p , (Γ(ν − γ 1 ))λα · (Γ(ν − γ 2 ))λβ (14.73) x (p−λν )/p A(w)p/(p−λν ) dw , (14.74) A0 (x) := x0

and

δ 1 :=

21−((λα +λν )/p) , 1,

if if

λα + λν ≤ p, λα + λν ≥ p. (14.75)

If λβ = 0, we obtain that x + λν λα q(w) (Dxγ01 f1 )(w) · Dxν0 f1 (w) + x0

294

14. Canavati Fractional Multivariate Opial Inequalities

, λ Dx 1 f2 (w)λα · Dxν f2 (w)λν dw 0 0 ≤ A0 (x) λβ =0 · )

x

· x0

λν λα + λν

λν /p · δ1

p p

p(w) Dxν0 f1 (w) + Dxν0 f2 (w) dw

*((λα +λν )/p) .

(14.76)

Similarly, by (14.76), we derive Theorem 14.20. Let ν ≥ 1, γ 1 , γ 2 ≥ 0, such that ν − γ 1 ≥ 1, ν − γ 2 ≥ ¯ and there exist ∂ ν f1 (x)/∂rν , ∂ ν f2 (x)/ 1, n := [ν], and f1 , f2 ∈ C n (A), R1 R1 ν ¯ A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2. Furthermore ∂r ∈ C(A), assume ∂ j f1 /∂rj = ∂ j f2 /∂rj = 0, for j = 0, 1, . . . , n − 1, on ∂B(0, R1 ). Let λν > 0 and λα > 0; λβ ≥ 0, p := λα + λν > 1. Call

w

Pk (w) :=

(w − t)(ν−γ k −1)p/(p−1) t(1−N )/(p−1) dt,

(14.77)

R1

k = 1, 2, A(w) :=

R1 ≤ w ≤ R 2 , w(N −1)(1−(λν /p)) (P1 (w))λα ((p−1)/p) (P2 (w))λβ ((p−1)/p) , (14.78) (Γ(ν − γ 1 ))λα (Γ(ν − γ 2 ))λβ

λα /p

R2

(A(w))p/λα dw

A0 (R2 ) :=

.

(14.79)

R1

Take the case of λβ = 0. Then ⎡ λα λν ∂γ1 ∂ν f R1 1 R 1 + ⎣ f (x) (x) 1 γ ν ∂r A ∂r 1

⎤ γ λα λν ∂ 1f ∂ν f R1 2 R1 2 (x) (x) ⎦ dx ≤ ∂rγ 1 ∂rν

A0 (R2 )|λβ =0

We need

λν p

p ν p * (λν /p) ) ν ∂ f ∂R1 f1 + R1 2 (x) dx. (x) ∂rν ∂rν A (14.80)

14.2 Results

295

Theorem 14.21. [26]. All here are as in Theorem 14.19. Denote λ /λ 2 β ν − 1, if λβ ≥ λν , δ 3 := 1, if λβ ≤ λν . If λα = 0, then x x0

+ λ λν q(w) Dxγ02 f2 (w) β · Dxν0 f1 (w) + , γ Dx 2 f1 (w)λβ · Dxν f2 (w)λν dw 0 0

≤ (A0 (x)|λα =0 ) 2p−λν /p

x

· x0

λν λβ + λν

λν /p

p p

p(w) Dxν0 f1 (w) + Dxν0 f2 (w) dw

λ /p

δ3 ν

((λν +λβ )/p) ,

(14.81)

all x0 ≤ x ≤ b.

Similarly, by (14.81), we derive Theorem 14.22. All basic assumptions are as in Theorem 14.20. Let λν > 0, λα = 0, λβ > 0, p := λν + λβ > 1, P2 deﬁned by (14.77). Now it is w(N −1)(1−λν /p) (P2 (w))λβ ((p−1)/p) , (14.82) A(w) := (Γ(ν − γ 2 ))λβ λβ /p R2

(A(w))p/λβ dw

A0 (R2 ) :=

.

(14.83)

R1

Denote δ 3 :=

2λβ /λν − 1, 1,

if if

λβ ≥ λν , λβ ≤ λν .

(14.84)

Then

⎡ ⎤ ∂ γ 2 f (x) λβ ∂ ν f (x) λν ∂ γ 2 f (x) λβ ∂ ν f (x) λν 2 1 1 2 R R R R + 1 ⎦ dx ≤ ⎣ 1 1 1 ∂rγ 2 ∂rν ∂rγ 2 ∂rν A

λβ /p

A0 (R2 ) 2

λν p

(λν /p)

λ /p δ3 ν

p ν ν ∂R1 f1 (x) p ∂R f (x) + 1 2 ∂rν dx. ∂rν A (14.85)

296

14. Canavati Fractional Multivariate Opial Inequalities

We need Theorem 14.23. [26]. All here are as in Theorem 14.19 , (λα , λβ = 0). Denote ⎧ ⎨ 2((λα +λβ )/λν )−1, if λα + λβ ≥ λν , (14.86) γ˜1 := 1, if λα + λβ ≤ λν , ⎩ and

1, if λα + λβ + λν ≥ p, 1−((λα +λβ +λν )/p) , if λα + λβ + λν ≤ p. 2

γ˜2 := Then

(14.87)

+ λα λ λν q(w) Dxγ01 f1 (w) · Dxγ02 f2 (w) β · Dxν0 f1 (w)

x

x0

λ λα λν , dw + Dxγ02 f1 (w) β · Dxγ01 f2 (w) · Dxν0 f2 (w)

≤ A0 (x)

x

· x0

λν (λα + λβ )(λα + λβ + λν )

λν /p +

λν /p λα γ˜2 + 2(p−λν )/p ( γ˜1 λβ )λν /p

p p p(w) Dxν0 f1 (w) + Dxν0 f2 (w) dw

,

((λα +λβ +λν )/p) , (14.88)

all x0 ≤ x ≤ b. Similarly, by (14.88), we obtain Theorem 14.24. Let all basics be as in Theorem 14.20. Here, λν , λα , λβ > 0, p := λα + λβ + λν > 1. Also Pk , k = 1, 2 as in (14.77), and A(w) is as in (14.78). Here it is λα +λβ /p R2

(A(w))p/λα +λβ dw

A0 (R2 ) :=

,

(14.89)

R1

γ˜1 := Then

2((λα +λβ )/λν ) − 1, 1,

if λα + λβ ≥ λν , if λα + λβ ≤ λν .

⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λβ ∂ ν f (x) λν 1 2 R1 1 R R 1 1 + ⎣ γ1 γ2 ∂r ∂r ∂rν A

(14.90)

14.2 Results

297

⎤ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν R1 1 R1 2 R1 2 ⎦ dx ≤ ∂rγ 2 ∂rγ 1 ∂rν A0 (R2 )

(λν /p) + , λν λλαν /p + 2(λα +λβ )/p (γ˜1 λβ )λν /p (λα + λβ )p p ν ν ∂R1 f1 (x) p ∂R f (x) + 1 2 (14.91) ∂rν dx. ∂rν A

We need Theorem 14.25. [26]. Let ν ≥ 2 and γ 1 ≥ 0, such that ν − γ 1 ≥ 2. Let f1 , f2 ∈ Cxν0 ([a, b]) with (i)

(i)

f1 (x0 ) = f2 (x0 ) = 0,

i = 0, 1, . . . , n − 1,

n := [ν]. Here x, x0 ∈ [a, b] : x ≥ x0 . Consider also, p(t) > 0, and q(t) ≥ 0 continuous functions on [x0 , b]. Let λα ≥ 0,

0 < λα+1 < 1,

and p > 1. Denote θ3 :=

L(x) :=

2λα /(λα+1 ) − 1, 1,

if if

λα ≥ λα+1 , λα ≤ λα+1 ,

(1−λα+1 ) x λα+1 θ3 λα+1 (q(w))(1/1−(λα+1 )) dw , 2 λα + λα+1 x0 (14.92)

and

x

P1 (x) :=

(x − t)(ν−γ 1 −1)p/(p−1) (p(t))−1/(p−1) dt,

(14.93)

x0

T (x) := L(x) ·

P1 (x)((p−1)/p) Γ(ν − γ 1 )

(λα +λα+1 ) ,

(14.94)

and ω 1 := 2(p−1/p)(λα +λα+1 ) ,

Φ(x) := T (x) ω 1 . Then

x

x0

+ λα λα+1 q(w) Dxγ01 f1 (w) · Dxγ01 +1 f2 (w) +

(14.95)

(14.96)

298

14. Canavati Fractional Multivariate Opial Inequalities

, γ Dx 1 f2 (w)λα · Dxγ 1 +1 f1 (w)λα+1 dw ≤ 0 0 )

x

Φ(x) x0

p p p(w) · Dxν0 f1 (w) + Dxν0 f2 (w) dw

*((λα +λα+1 )/p) , (14.97)

all

x0 ≤ x ≤ b.

Similarly, by (14.97), we obtain Theorem 14.26. Let ν ≥ 2, γ 1 ≥ 0, such that ν − γ 1 ≥ 2, n := ¯ and there exist ∂ ν f1 (x)/∂rν , ∂ ν f2 (x)/∂rν ∈ [ν]. Let f1 , f2 ∈ C n (A) R1 R1 ¯ A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2. Furthermore assume C(A), ∂ j f1 /∂rj = ∂ j f2 /∂rj = 0, j = 0, 1, . . . , n − 1, on ∂B(0, R1 ). Let λα > 0, 0 < λα+1 < 1, such that p := λα + λα+1 > 1. Denote (λ /λ ) 2 α α+1 − 1 if λα ≥ λα+1 (14.98) θ3 := 1, if λα ≤ λα+1 , )

L(R2 ) := λ * (1−λα+1 ) θ λ α+1 (1 − λα+1 ) (N −λα+1 )/(1−λα+1 ) (N −λα+1 )/(1−λα+1 ) 3 α+1 2 − R1 , R2 (N − λα+1 ) p (14.99)

and

R2

P1 (R2 ) :=

(R2 − t) (ν−γ 1 −1)p/(p−1) t (1−N )/(p−1) dt,

R1

Φ(R2 ) := L(R2 )

P1 (R2 )(p−1) (Γ(ν − γ 1 ))p

(14.100)

2p−1 .

(14.101)

Then ⎡ ⎤ λα γ +1 γ 1 ∂ 1 f (x) λα+1 ∂ γ 1 f (x) λα ∂ γ 1 +1 f (x) λα+1 ∂ f (x) 1 2 2 1 R1 R1 ⎣ R1 ⎦ dx + R1 γ ∂r 1 ∂rγ 1 +1 ∂rγ 1 ∂rγ 1 +1 A p ν ν ∂R1 f1 (x) p ∂R f (x) 1 2 ≤ Φ(R2 ) ∂rν + ∂rν dx. A

(14.102)

We need Theorem 14.27. [26]. All here are as in Theorem 14.19. Consider the special case λβ = λα + λν . Denote λν /p λν ˜ T (x) := A0 (x) 2(p−2λα −3λν )/p . (14.103) λα + λν

14.2 Results

Then

x

x0

299

+ λα λα +λν ν Dx f1 (w)λν q(w) Dxγ01 f1 (w) Dxγ02 f2 (w) 0

λα +λν γ , Dx 1 f2 (w)λα Dxν f2 (w)λν dw ≤ + Dxλ02 f1 (w) 0 0

x

T˜(x) x0

p p p(w) Dxν0 f1 (w) + Dxν0 f2 (w) dw

2((λα +λν )/p) , (14.104)

all x0 ≤ x ≤ b. Similarly, by (14.104), we obtain Theorem 14.28. Here all are as in Theorem 14.20. Consider the case λβ = λα + λν ; λα ≥ 0, λν > 0, λβ > 1/2, p := 2λβ . Here Pk , k = 1, 2, as in (14.77) and A(w) is as in (14.78). Call ((2λα +λν )/p) R2

(A(w))p/(2λα +λν )

A0 (R2 ) :=

.

(14.105)

R1

Also set

T˜(R2 ) := A0 (R2 )

λν λβ

λν /p 2(−λν /p) .

(14.106)

⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λβ ∂ ν f (x) λν 1 2 R1 1 R R + ⎣ 1 1 γ γ1 2 ∂r ∂r ∂rν A ⎤ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν 1 2 2 R1 R1 R1 ⎦ dx ≤ ∂rγ 2 ∂rγ 1 ∂rν

Then

T˜(R2 )

p ν ν ∂R1 f1 (x) p ∂R f (x) + 1 2 ∂rν dx. ∂rν A

(14.107)

We need Theorem 14.29. [26]. Let ν ≥ 1, γ 1 , γ 2 ≥ 0, such that ν − γ 1 ≥ (i) (i) 1, ν − γ 2 ≥ 1 and f1 , f2 ∈ Cxν0 ([a, b]) with f1 (x0 ) = f2 (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν]. Here, x, x0 ∈ [a, b] : x ≥ x0 . Consider p(x) ≥ 0 continuous functions on [x0 , b]. Let λα , λβ , λν ≥ 0. Set ρ(x) :=

(x − x0 )(νλα −γ 1 λα +νλβ −γ 2 λβ +1) p(x)∞ . (νλα − γ 1 λα + νλβ − γ 2 λβ + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λβ (14.108)

300

14. Canavati Fractional Multivariate Opial Inequalities

Then

+ λα λ λν q(w ) Dxγ01 f1 (w ) Dxγ02 f2 (w ) β Dxν0 f1 (w ) +

x

x0

, λ Dx 2 f1 (w)λβ Dxγ 1 f2 (w)λα Dxν f2 (w)λν dw 0 0 0 ≤

, ρ(x) + ν 2(λα +λν ) 2(λα +λν ) β β Dx0 f1 ∞ , + Dxν0 f1 2λ + Dxν0 f2 2λ + Dxν0 f2 ∞ ∞ ∞ 2 (14.109)

all x0 ≤ x ≤ b.

Similarly, by (14.109), we get Theorem 14.30. Same basic assumptions as in Theorem 14.20. Let λα , λβ , λν ≥ 0. Call ρ(R2 ) :=

R2N −1 (R2 − R1 )(νλα −γ 1 λα +νλβ −γ 2 λβ +1) . (νλα − γ 1 λα + νλβ − γ 2 λβ + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λβ (14.110)

Then

⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λβ ∂ ν f (x) λν 1 2 R1 1 R + ⎣ R1 1 γ γ1 2 ∂r ∂r ∂rν A

⎤ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν 1 2 2 R1 R1 R1 ⎦ dx ≤ ∂rγ 2 ∂rγ 1 ∂rν ' ν ∂ ν f 2(λα +λν ) ∂R f 2λβ π N/2 R1 1 1 1 ρ(R2 ) + + ∂rν Γ(N/2) ∂rν ∞ ∞ ( ν ν ∂R1 f2 2λβ ∂R f2 2(λα +λν ) 1 + . ∂rν ∂rν ∞ ∞

(14.111)

We need Theorem 14.31. [26]. Assume, as in Theorem 14.29, λβ = 0. It holds

x

x0

+ λα λν p(w) Dxγ01 f1 (w) · Dxν0 f1 (w) +

, γ Dx 1 f2 (w)λα · Dxν f2 (w)λν dw ≤ 0 0

14.2 Results

301

+ ν λα +λν ν λα +λν , (x − x0 )(νλα −γ 1 λα +1) p(x)∞ Dx f1 , + Dx0 f2 ∞ · 0 ∞ (νλα − γ 1 λα + 1)(Γ(ν − γ 1 + 1))λα (14.112) all x0 ≤ x ≤ b. Similarly, by (14.112), we derive Theorem 14.32. All are as in Theorem 14.30. Assume λβ = 0. Then ⎡ ⎤ λα γ 1 ∂ ν f (x) λν ∂ γ 1 f (x) λα ∂ ν f (x) λν ∂ f (x) 1 2 1 2 R1 R1 + R1 ⎦ dx ≤ ⎣ R1 ∂rγ 1 ∂rν ∂rγ 1 ∂rν A ⎛ ⎤ ∂ ν f1 λα +λν ∂ ν f2 λα +λν R2N −1 (R2 − R1 )(νλα −γ 1 λα +1) 2π N/2 R1 R1 ⎝ ⎦. + ∂rν ∂rν Γ(N/2) (νλα − γ 1 λα + 1)(Γ(ν − γ 1 + 1))λα ∞

∞

(14.113)

We need Theorem 14.33. [26]. (In relationship to Theorem 14.29, λβ = λα + λν .) It holds x + λα λα +λν ν Dx f1 (w)λν + p(w) Dxγ01 f1 (w) Dxγ02 f2 (w) 0 x0

, γ Dx 2 f1 (w)λα +λν Dxγ 1 f2 (w)λα Dxν f2 (w)λν dw ≤ 0 0 0

(x − x0 )(2νλα −γ 1 λα +νλν −γ 2 λα −γ 2 λν +1) (2νλα − γ 1 λα + νλν − γ 2 λα − γ 2 λν + 1) p(x)∞ · (Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λα +λν 2(λα +λν ) ν 2(λα +λν ) · Dxν0 f1 ∞ , + Dx0 f2 ∞ all

(14.114)

x0 ≤ x ≤ b.

Similarly, by (14.114), we derive Theorem 14.34. All are as in Theorem 14.30. Assume λβ = λα + λν . Then ⎡ λα γ λα +λν λ γ 1 ν ∂R ∂R1 f1 (x) ∂R21 f2 (x) f (x) ν 1 1 ⎣ ∂rν + ∂rγ 1 ∂rγ 2 A

302

14. Canavati Fractional Multivariate Opial Inequalities

γ ∂ 2 f (x) λα +λν R1 1 ∂rγ 2

⎤ γ ∂ 1 f (x) λα ∂ ν f (x) λν N/2 R1 2 R1 2 ⎦ dx ≤ 2π ∂rγ 1 ∂rν Γ(N/2)

R2N −1 (R2 − R1 )(2νλα −γ 1 λα +νλν −γ 2 λα −γ 2 λν +1) (2νλα − γ 1 λα + νλν − γ 2 λα − γ 2 λν + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λα +λν

2(λα +λν ) ν 2(λα +λν ) ν ∂R ∂ f f 1 2 R 1 1 . + ∂rν ∂rν ∞ ∞

(14.115)

We need Theorem 14.35. [26]. (In relationship to Theorem 14.29, λν = 0, λα = λβ .) It holds x + λα λα p(w) Dxγ01 f1 (w) · Dxγ02 f2 (w) + x0

, γ Dx 2 f1 (w)λα · Dxγ 1 f2 (w)λα dw ≤ 0 0 + 2λα ν 2λα , ρ∗ (x) Dxν f1 + Dx f2 , 0

∞

0

∞

(14.116)

all x0 ≤ x ≤ b. Here (x − x0 )(2νλα −γ 1 λα −γ 2 λα +1) · p(x)∞ ρ∗ (x) := . (2νλα − γ 1 λα − γ 2 λα + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λα (14.117) We get, by (14.116), the result Theorem 14.36. All are as in Theorem 14.30. Assume λν = 0, λα = λβ . Then ⎡ ⎤ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λα ∂ γ 2 f (x) λα ∂ γ 1 f (x) λα R1 2 R1 2 ⎦ R 1 ⎣ R1 1 dx ≤ + 1 γ γ1 γ2 γ1 2 ∂r ∂r ∂r ∂r A

2π N/2 Γ(N/2)

where ρ∗ (R2 ) :=

' ( ν ν ∂R f 2λα ∂R f 2λα 1 1 1 2 , + ρ (R2 ) ∂rν ∞ ∂rν ∞ ∗

(14.118)

R2N −1 (R2 − R1 )(2νλα −γ 1 λα −γ 2 λα +1) . (2νλα − γ 1 λα − γ 2 λα + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λα (14.119)

14.2 Results

303

We need Theorem 14.37. [26]. (In relationship to Theorem 14.29, λα = 0, λβ = λν .) It holds x + λ λ p(w) Dxγ02 f2 (w) β · Dxν0 f1 (w) β + x0

γ , Dx 2 f1 (w)λβ · Dxν f2 (w)λβ dw ≤ 0 0 + ν 2λβ ν 2λβ , (x − x0 )(νλβ −γ 2 λβ +1) · p(x)∞ Dx f1 + Dx0 f2 ∞ , · 0 ∞ (νλβ − γ 2 λβ + 1)(Γ(ν − γ 2 + 1))λβ (14.120) all x0 ≤ x ≤ b.

We get, by (14.120), the next result. Theorem 14.38. All are as in Theorem 14.30. Assume λα = 0 and λβ = λν . Then ⎡ ⎤ λβ λβ λβ γ 2 λβ γ 2 ν ν ∂ f (x) ∂R f1 (x) ∂R f1 (x) ∂R1 f2 (x) ⎦ ⎣ R1 2 dx ≤ 1 ν + 1 γ γ2 ∂r 2 ∂rν ∂r ∂r A

2π N/2 Γ(N/2)

R2N −1 (R2 − R1 )(νλβ −γ 2 λβ +1) (νλβ − γ 2 λβ + 1)(Γ(ν − γ 2 + 1))λβ ( ν ∂R1 f2 2λβ + . ∂rν ∞

' ν ∂R f 2λβ 1 1 ∂rν ∞ (14.121)

We make ¯ j = 1, . . . , M ∈ Assumption 14.39. Let ν ≥ 1, n := [ν], fj ∈ C n (A), ν ν ¯ A := B(0, R2 ) − B(0, R1 ) ⊆ N, and there exist ∂R1 fj /∂r ∈ C(A), N R , N ≥ 2. Furthermore assume that ∂ i fj /∂ri = 0, i = 0, 1, . . . , n − 1, on ∂B(0, R1 ), for all j = 1, . . . , M . Next we present a set of multivariate fractional Opial-type inequalities involving several functions over the shell. We need Theorem 14.40. [27] Let ν ≥ 1, γ 1 , γ 2 ≥ 0, such that ν−γ 1 ≥ 1, ν−γ 2 ≥ (i) 1 and fj ∈ Cxν0 ([a, b]) with fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν], j =

304

14. Canavati Fractional Multivariate Opial Inequalities

1, . . . , M ∈ N. Here, x, x0 ∈ [a, b] : x ≥ x0 . Consider also p(t) > 0, and q(t) ≥ 0 continuous functions on [x0 , b]. Let λν > 0 and λα , λβ ≥ 0 such that λν < p, where p > 1. Call w (w − t)(ν−γ k −1)p/p−1 (p(t))−1/p−1 dt, k = 1, 2, x0 ≤ w ≤ b; Pk (w) := x0

(14.122) A(w) :=

q(w) · (P1 (w))λα ((p−1)/p) · (P2 (w))λβ ((p−1)/p) (p(w))−λν /p ; (14.123) (Γ(ν − γ 1 ))λα · (Γ(ν − γ 2 ))λβ x (p−λν )/p A0 (x) := A(w)p/(p−λν ) dw . (14.124) x0

Set

ϕ1 (x) := A0 (x) λβ =0 · δ ∗1

M 1−(λα +λν /p) , 2(λα +λν /p) − 1,

:=

λν λα + λν

λν /p ,

(14.125)

if λα + λν ≤ p, if λα + λν ≥ p.

(14.126)

If λβ = 0, we obtain that ⎞ ⎛ x M λα ν λν γ Dx 1 fj (w) · Dx fj (w) ⎠ dw ≤ q(w) ⎝ 0 0 x0

j=1

⎡ δ ∗1 · ϕ1 (x) · ⎣

⎞ ⎤((λα +λν )/p) M p ν Dx fj (w) ⎠ dw⎦ p(w) ⎝ , 0 ⎛

x

x0

(14.127)

j=1

all x0 ≤ x ≤ b. Similarly, by (14.127), we derive Theorem 14.41. Let fj , j = 1, . . . , M,as in Assumption 14.39. Let γ 1 , γ 2 ≥ 0, such that ν − γ 1 ≥ 1, ν − γ 2 ≥ 1. Let λν > 0, and λα > 0; λβ ≥ 0, p := λα + λν > 1. Set w Pk (w) := (w − t)(ν−γ k −1)p/(p−1) t(1−N )/(p−1) dt, (14.128) R1

k = 1, 2, A(w) :=

R1 ≤ w ≤ R 2 , w(N −1)(1−(λν /p)) (P1 (w))λα ((p−1)/p) (P2 (w))λβ ((p−1)/p) , (14.129) (Γ(ν − γ 1 ))λα (Γ(ν − γ 2 ))λβ λα /p R2

(A(w))p/λα dw

A0 (R2 ) := R1

.

(14.130)

14.2 Results

305

Take the case of λβ = 0. Then λα λν M γ1 ν ∂R1 fj (x) ∂R1 fj dx ∂rγ 1 ∂rν A j=1

≤ A0 (R2 )|λβ =0

⎡ ⎤ p (λν /p) M ν ∂ f (x) λν R1 j ⎣ ⎦. ∂rν dx p A j=1

(14.131)

We need Theorem 14.42. [27]. All here are as in Theorem 14.40. Denote λ /λ 2 β ν − 1, if λβ ≥ λν , (14.132) δ 3 := 1, if λβ ≤ λν , ε2 :=

1, if λν + λβ ≥ p, M 1−((λν +λβ )/p) , if λν + λβ ≤ p,

(14.133)

and ϕ2 (x) := (A0 (x)|λα =0 )2(p−λν /p)

λν λβ + λν

λν /p

λ /p

δ3 ν .

(14.134)

If λα = 0, then

x

x0

⎧⎧ −1 + ⎨⎨M γ Dx 2 fj+1 (w)λβ Dxν fj (w)λν q(w) 0 0 ⎩⎩ j=1

λ λν , + Dxγ02 fj (w) β Dxν0 fj+1 (w) + λ λν + Dxγ02 fM (w) β Dxν0 f1 (w) λ λν , dw ≤ + Dxγ02 f1 (w) β Dxν0 fM (w) 2((λν +λβ )/p) ε2 ϕ2 (x) ·

⎧ ⎨ ⎩

x

x0

⎤ ⎫((λν +λβ )/p) ⎡ M ⎬ p ν Dx fj (w) ⎦ dw p(w) · ⎣ , 0 ⎭ j=1

(14.135) x ≥ x0 .

Similarly, by (14.135), we obtain

306

14. Canavati Fractional Multivariate Opial Inequalities

Theorem 14.43. All basic assumptions are as in Theorem 14.41. Let λν > 0, λα = 0; λβ > 0, p := λν + λβ > 1, P2 deﬁned by (14.128). Now it is A(w) :=

w(N −1)(1−λν /p) (P2 (w))λβ (p−1/p) , (Γ(ν − γ 2 ))λβ

λβ /p

R2

A0 (R2 ) :=

(14.136)

(A(w))

p/λβ

dw

.

(14.137)

R1

Denote δ 3 :=

2 λβ /λν − 1, 1,

Call

if λβ ≥ λν , if λβ ≤ λν .

ϕ2 (R2 ) := A0 (R2 )2 λβ /p

λν p

λν /p

(14.138)

λ /p

δ3 ν .

(14.139)

⎧⎧ ⎡ λβ λ ⎨⎨M −1 γ 2 ν ∂ f (x) ∂R f (x) ν 1 j + ⎣ R1 j+1 ∂rγ 2 ∂rν A ⎩⎩

Then

j=1

⎤⎫ γ ∂ 2 f (x) λβ ∂ ν f (x) λν ⎬ R1 j R1 j+1 ⎦ + ∂rγ 2 ⎭ ∂rν ⎡ ∂ γ 2 f (x) λβ ∂ ν f (x) λν R1 1 + ⎣ R1 M γ 2 ∂r ∂rν ⎤⎫ γ ∂ 2 f (x) λβ ∂ ν f (x) λν ⎬ R1 1 R1 M ⎦ dx ≤ ∂rγ 2 ∂rν ⎭ ⎡

⎤ M ν ∂R1 fj (x) p ⎦. 2ϕ2 (R2 ) ⎣ ∂rν dx j=1

(14.140)

A

We need Theorem 14.44. [27]. All here are as in Theorem 14.40, (λα , λβ = 0). ((λ +λ )/λ ) 2 α β ν − 1, if λα + λβ ≥ λν , γ˜1 := (14.141) 1, if λα + λβ ≤ λν , and

γ˜2 :=

1, if λα + λβ + λν ≥ p, 21−((λα +λβ +λν )/p) , if λα + λβ + λν ≤ p.

(14.142)

14.2 Results

Set

ϕ3 (x) := A0 (x) ·

λν (λα + λβ )(λα + λβ + λν )

307

λν /p (14.143)

and ε3 :=

if λα + λβ + λν ≥ p, if λα + λβ + λν ≤ p,

1, M 1−(λα +λβ +λν /p) ,

(14.144)

Then

⎡ x

x0

q(w) ⎣

M −1 +

γ Dx 1 fj (w)λα Dxγ 2 fj+1 (w)λβ Dxν fj (w)λν 0 0 0

j=1

λ λα λν , + Dxγ02 fj (w) β Dxγ01 fj+1 (w) Dxν0 fj+1 (w) + λα λ λν + Dxγ01 f1 (w) Dxγ02 fM (w) β Dxν0 f1 (w) λ λα λν ,, dw + Dxγ02 f1 (w) β Dxγ01 fM (w) Dxν0 fM (w) ⎧ ⎤ ⎫ ⎡ M ⎬ ⎨ x p Dxν fj (w) ⎦ dw ((λα +λβ +λν )/p) , ≤ 2(λα +λβ +λν /p) ε3 ϕ3 (x)· p(w)⎣ 0 ⎭ ⎩ x0 j=1

(14.145) all x0 ≤ x ≤ b.

Similarly, by (14.145), we obtain Theorem 14.45. All basic assumptions are as in Theorem 14.41. Here λν , λα , λβ > 0, p := λα + λβ + λν > 1, Pk as in (14.128). A is as in (14.129). Here

(λα +λβ )/p

R2 p/(λα +λβ )

A0 (R2 ) :=

(A(w))

,

,

(14.146)

R1

γ˜1 :=

2(λα +λβ /λν ) − 1, 1,

if λα + λβ ≥ λν , if λα + λβ ≤ λν .

(14.147)

Set ϕ3 (R2 ) := A0 (R2 )

λν (λα + λβ )p

(λν /p)

+

, (λν /p) λα + 2((λα +λβ )/p) (γ˜1 λβ )(λν /p) . (14.148)

308

14. Canavati Fractional Multivariate Opial Inequalities

Then A

⎡

M −1

⎣

j=1

⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λβ ∂ ν f (x) λν j j+1 R1 j R R 1 1 + ⎣ ∂rγ 1 ∂rγ 1 ∂rν

⎤ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν R1 j R1 j+1 R1 j+1 ⎦+ ∂rγ 2 ∂rγ 1 ∂rν ⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λβ ∂ ν f (x) λν 1 M R1 1 R R 1 1 + ⎣ ∂rγ 1 ∂rγ 2 ∂rν ⎤⎤ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν R1 1 R1 M R1 M ⎦⎦ dx ≤ ∂rγ 2 ∂rγ 1 ∂rν ⎡

⎤ M ν ∂R1 fj (x) p ⎦. 2ϕ3 (R2 ) ⎣ ∂rν dx j=1

(14.149)

A

We need Theorem 14.46. [27]. Let ν ≥ 2, and γ 1 ≥ 0, such that ν − γ 1 ≥ 2. (i) Let fj ∈ Cxν0 ([a, b]) with fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν], j = 1, . . . , M ∈ N. Here, x, x0 ∈ [a, b] : x ≥ x0 . Consider also p(t) > 0, , and q(t) ≥ 0 continuous functions on [x0 , b]. Let λα ≥ 0, 0 < λα+1 < 1, and p > 1 . Denote (λ /λ ) 2 α α+1 − 1, if λα ≥ λα+1 , θ3 := (14.150) 1, if λα ≤ λα+1 , L(x) := 2

(1−λα+1 )

x

(q(w))

(1/1−λα+1 )

dw

x0

θ3 λα+1 λα + λα+1

λα+1 , (14.151)

and

x

P1 (x) :=

(x − t)((ν−γ 1 −1)p/(p−1)) (p(t))−1/(p−1) dt,

(14.152)

x0

T (x) := L(x) ·

P1 (x)(p−1/p) Γ(ν − γ 1 )

(λα +λα+1 ) ,

(14.153)

and ω 1 := 2((p−1)/p)(λα +λα+1 ) , Φ(x) := T (x) ω 1 .

(14.154)

14.2 Results

309

Also put ε4 := Then

x

x0

if λα + λα+1 ≥ p, if λα + λα+1 ≤ p

1, M 1−(λα +λα+1 /p) ,

.

(14.155)

⎧⎧ −1 + ⎨⎨M γ Dx 1 fj (w)λα Dxγ 1 +1 fj+1 (w)λα+1 q(w) 0 0 ⎩⎩ j=1

λα λα+1 , + Dxγ01 fj+1 (w) Dxγ01 +1 fj (w) + λα λα+1 + Dxγ01 f1 (w) Dxγ01 +1 fM (w) λα λα+1 , dw + Dxγ01 fM (w) Dxγ01 +1 f1 (w) ⎡ ≤ 2(λα +λα+1 /p) ε4 Φ(x) ⎣

⎞ ⎤((λα +λα+1 )/p) M p ν Dx fj (w) ⎠ dw⎦ p(w) ⎝ , 0 ⎛

x

x0

j=1

(14.156) all x0 ≤ x ≤ b.

Similarly, by (14.156), we get Theorem 14.47. Let all be as in Assumption 14.39. Here ν ≥ 2, γ 1 ≥ 0 such that ν − γ 1 ≥ 2. Let λα > 0, 0 < λα+1 < 1, such that p := λα + λα+1 > 1. Denote (λ /λ ) 2 α α+1 − 1, if λα ≥ λα+1 , (14.157) θ3 := 1, if λα ≤ λα+1 , ) 1 − λα+1 (N −λα+1 )/(1−λα+1 ) R2 L(R2 ) := 2 − N − λα+1

(N −λα+1 )/(1−λα+1 ) R1

λα+1 ,(1−λα+1 ) θ λ 3 α+1 , λα + λα+1

(14.158)

and

R2

P (R2 ) :=

(R2 − t)(ν−γ 1 −1)(p/(p−1)) t (1−N )/(p−1) dt,

(14.159)

R1

Φ(R2 ) := L(R2 )

P1 (R2 )(p−1) (Γ(ν − γ 1 ))p

2(p−1) .

(14.160)

310

14. Canavati Fractional Multivariate Opial Inequalities

⎧⎧ ⎡ λα γ +1 λα+1 ⎨⎨M −1 γ 1 ∂R1 fj (x) ∂R11 fj+1 (x) ⎣ + ∂rγ 1 ∂rγ 1 +1 A ⎩⎩ j=1 ⎤⎫ γ ∂ 1 f (x) λα ∂ γ 1 +1 f (x) λα+1 ⎬ j R1 R1 j+1 ⎦ + ∂rγ 1 +1 ⎭ ∂rγ 1

Then

⎡ ∂ γ 1 f (x) λα ∂ γ 1 +1 f (x) λα+1 1 M R ⎣ R1 + 1 ∂rγ 1 ∂rγ 1 +1 ⎤⎫ γ ∂ 1 f (x) λα ∂ γ 1 +1 f (x) λα+1 ⎬ 1 R1 R1 M ⎦ dx ≤ ∂rγ 1 ∂rγ 1 +1 ⎭ ⎡

⎤ M ν ∂R1 fj (x) p ⎦. 2Φ(R2 ) ⎣ ∂rν dx

(14.161)

A

j=1

We need Theorem 14.48. [27]. All here are as in Theorem 14.40. Consider the special case λβ = λα + λν . Denote T˜(x) := A0 (x) ε5 :=

λν λα + λν

λν /p

2(p−2λα −3λν /p) ,

1, if 2(λα + λν ) ≥ p, 1−(2(λα +λν )/p) , if 2(λα + λν ) ≤ p M

(14.162)

.

(14.163)

Then

x

x0

⎧⎧ −1 + ⎨⎨M γ Dx 1 fj (w)λα Dxγ 2 fj+1 (w)λα +λν Dxν fj (w)λν q(w) 0 0 0 ⎩⎩ j=1

λα +λν γ , Dx 1 fj+1 (w)λα Dxν fj+1 (w)λν + Dxγ02 fj (w) 0 0 + λα λα +λν Dxν f1 (w)λν + Dxγ01 f1 (w) Dxγ02 fM (w) 0 λα +λν γ , Dx 1 fM (w)λα Dxν fM (w)λν dw + Dxγ02 f1 (w) 0 0 ⎡ ≤ 22((λα +λν )/p) ε5 T˜(x) ⎣

⎞ ⎤(2(λα +λν )/p) M p ν Dx fj (w) ⎠ dw⎦ p(w) ⎝ , 0 ⎛

x

x0

j=1

(14.164)

14.2 Results

all

311

x0 ≤ x ≤ b.

Similarly, by (14.164), we have Theorem 14.49. Here all are as in Theorem 14.41. Consider the case λβ = λα + λν ; λα ≥ 0, λν > 0, λβ > 1/2, p := 2λβ . Here Pk , k = 1, 2, as in (14.128) and A is as in (14.129). Set (2λα +λν /p) R2

(A(w))p/(2λα +λν ) dw

A0 (R2 ) :=

.

(14.165)

R1

Also put

T˜(R2 ) := A0 (R2 )

Then

λν λβ

(λν /p)

2−(λν /p) .

⎧⎧ ⎡ λα γ λα +λν ⎨⎨M −1 γ 1 ∂R1 fj (x) ∂R21 fj+1 (x) ⎣ ∂rγ 1 ∂rγ 2 A ⎩⎩ j=1

(14.166)

ν ∂R1 fj (x) λν ∂rν +

⎤⎫ γ ∂ 2 f (x) λα +λν ∂ γ 1 f (x) λα ∂ ν f (x) λν ⎬ R1 j+1 R1 j+1 R1 j ⎦ + ∂rγ 2 ⎭ ∂rγ 1 ∂rν ⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λα +λν ∂ ν f (x) λν 1 M R R 1 1 R1 1 ⎣ ∂rν + ∂rγ 1 ∂rγ 2 ⎤⎫ γ ∂ 2 f (x) λα +λν ∂ γ 1 f (x) λα ∂ ν f (x) λν ⎬ R1 M R1 M R1 1 ⎦ dx ∂rγ 1 ∂rν ∂rγ 2 ⎭ ⎡

⎤ M ν ∂R1 fj (x) p ⎦. ≤ 2 T˜(R2 ) ⎣ ∂rν dx j=1

(14.167)

A

We need Theorem 14.50. [27]. Let ν ≥ 1, γ 1 , γ 2 ≥ 0, such that ν − γ 1 ≥ 1, ν − (i) γ 2 ≥ 1 and fj ∈ Cxν0 ([a, b]) with fj (x0 ) = 0, i = 0, 1, . . . , n − 1, n := [ν], j = 1, . . . , M ∈ N. Here, x, x0 ∈ [a, b] : x ≥ x0 . Consider p(x) ≥ 0 continuous functions on [x0 , b]. Let λα , λβ , λν ≥ 0. Set ρ(x) :=

(x − x0 )(νλα −γ 1 λα +νλβ −γ 2 λβ +1) p(x)∞ . (νλα − γ 1 λα + νλβ − γ 2 λβ + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λβ (14.168)

312

14. Canavati Fractional Multivariate Opial Inequalities

Then

x

x0

⎧⎧ −1 + ⎨⎨M γ Dx 1 fj (w)λα Dxγ 2 fj+1 (w)λβ Dxν fj (w)λν + p(w) 0 0 0 ⎩⎩ j=1

, γ Dx 2 fj (w)λβ Dxγ 1 fj+1 (w)λα Dxν fj+1 (w)λν + 0 0 0 + λ λ Dxγ 1 f1 (w) α Dxγ 2 fM (w) β Dxν f1 (w)λν + 0 0 0 γ , Dx 2 f1 (w)λβ Dxγ 1 fM (w)λα Dxν fM (w)λν dw 0 0 0 ⎧ ⎫ M ⎨ ν 2(λα +λν ) ν 2λβ ⎬ Dx fj ≤ ρ(x) + Dx0 fj ∞ , (14.169) 0 ∞ ⎩ ⎭ j=1

all

x0 ≤ x ≤ b.

Similarly, by (14.169), we have Theorem 14.51. All are as in Assumption 14.39. Let γ 1 , γ 2 ≥ 0, such that ν − γ 1 ≥ 1, ν − γ 2 ≥ 1; λα , λβ , λν ≥ 0. Set ρ(R2 ) :=

Then

R2N −1 (R2 − R1 )(νλα −γ 1 λα +νλβ −γ 2 λβ +1) . (νλα − γ 1 λα + νλβ − γ 2 λβ + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λβ (14.170)

⎧⎧ ⎡ λα γ λβ ⎨⎨M −1 γ 1 ∂R1 fj (x) ∂R21 fj+1 (x) ⎣ ∂rγ 1 ∂rγ 2 A ⎩⎩ j=1

ν ∂R1 fj (x) λν ∂rν +

⎤⎫ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν ⎬ R1 j R1 j+1 R1 j+1 ⎦ + ∂rγ 2 ⎭ ∂rγ 1 ∂rν ⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λβ ∂ ν f (x) λν 1 M R1 1 R + ⎣ R1 1 ∂rγ 1 ∂rγ 2 ∂rν ⎤⎫ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν ⎬ R1 1 R1 M R1 M ⎦ dx ≤ ∂rγ 2 ∂rγ 1 ∂rν ⎭ ⎧ ⎫ % 2(λα +λν ) ν 2λβ -⎬ M ⎨ ν ∂R ∂ f f 2π N/2 j j R 1 1 ρ (R2 ) + . ∂rν ∂rν ⎩ ⎭ Γ(N/2) ∞ ∞ j=1

(14.171)

14.2 Results

313

We need Theorem 14.52. [27]. (As in Theorem 14.50, λβ = 0.) It holds ⎞ ⎛ x M λα ν λν γ Dx 1 fj (w) Dx fj (w) ⎠ dw ≤ p(w) ⎝ 0

x0

0

j=1

(x − x0 )(νλα −γ 1 λα +1) p(x)∞ (νλα − γ 1 λα + 1)(Γ(ν − γ 1 + 1))λα

⎞ M ν λα +λν Dx fj ⎠ , (14.172) ·⎝ 0 ∞ ⎛

j=1

all x0 ≤ x ≤ b. Similarly, by (14.172), we obtain Theorem 14.53. Here all are as in Theorem 14.51. Case of λβ = 0. Then ⎛ ⎞ λα γ1 M ∂ ν f (x) λν ∂ f (x) R1 j dx⎠ ≤ ⎝ R1 j γ1 ∂r ∂rν A j=1

N/2

2π Γ(N/2)

⎞ ⎛ M λα +λν ν ∂ f − R1 ) R1 j ⎠. ⎝ ∂rν (νλα − γ 1 λα + 1)(Γ(ν − γ 1 + 1))λα ∞ j=1 R2N −1 (R2

(νλα −γ 1 λα +1)

(14.173) We need Theorem 14.54. [27]. (As in Theorem 14.50, λβ = λα + λν .) It holds ⎧⎧ x −1 + ⎨⎨M γ Dx 1 fj (w)λα Dxγ 2 fj+1 (w)λα +λν Dxν fj (w)λν + p(w) 0 0 0 ⎩⎩ x0 j=1

, γ Dx 2 fj (w)λα +λν Dxγ 1 fj+1 (w)λα Dxν fj+1 (w)λν + 0 0 0 + Dxγ 1 f1 (w)λα Dxγ 2 fM (w)λα +λν Dxν f1 (w)λν + 0 0 0 , γ Dx 2 f1 (w)λα +λν Dxγ 1 fM (w)λα Dxν fM (w)λν dw ≤ 0 0 0

2(x − x0 )(2νλα −γ 1 λα +νλν −γ 2 λα −γ 2 λν +1) (2νλα − γ 1 λα + νλν − γ 2 λα − γ 2 λν + 1) p(x)∞ (Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))(λα +λν )

314

14. Canavati Fractional Multivariate Opial Inequalities

⎞ M ν 2(λα +λν ) Dx fj ⎠, ·⎝ 0 ∞ ⎛

(14.174)

j=1

all x0 ≤ x ≤ b.

Similarly, by (14.174), we derive Theorem 14.55. Here all are as in Theorem 14.51. Case of λβ = λα + λν . Then ⎧⎧ ⎡ λα γ λα +λν λ ⎨⎨M −1 γ 1 ν ∂R ∂R1 fj (x) ∂R21 fj+1 (x) fj (x) ν 1 ⎣ ∂rν + ∂rγ 1 ∂rγ 2 A ⎩⎩ j=1

γ ∂ 2 f (x) λα +λν R1 j ∂rγ 2

⎤⎫ γ ∂ 1 f (x) λα ∂ ν f (x) λν ⎬ j+1 R1 R1 j+1 ⎦ + ⎭ ∂rγ 1 ∂rν

⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λα +λν 1 M R R 1 1 ⎣ ∂rγ 1 ∂rγ 2 γ ∂ 2 f (x) λα +λν R1 1 ∂rγ 2 4π N/2 · Γ(N/2)

ν ∂R1 f1 (x) λν ∂rν +

⎤⎫ γ ∂ 1 f (x) λα ∂ ν f (x) λν ⎬ M M R1 R1 ⎦ dx ≤ ∂rγ 1 ∂rν ⎭

R2N −1 (2νλα − γ 1 λα + νλν − γ 2 λα − γ 2 λν + 1)

(R2 − R1 )(2νλα −γ 1 λα +νλν −γ 2 λα −γ 2 λν +1) (Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))(λα +λν ) ⎛ ⎞ M ν ∂R1 fj 2(λα +λν ) ⎠. ·⎝ ∂rν j=1

(14.175)

∞

We need Theorem 14.56. [27]. (As in Theorem 14.50, λν = 0, λα = λβ .) It holds ⎧⎧ x −1 + ⎨⎨M γ Dx 1 fj (w)λα Dxγ 2 fj+1 (w)λα p(w) 0 0 ⎩⎩ x0 j=1

14.2 Results

315

λα λα , + Dxγ02 fj (w) Dxγ01 fj+1 (w) + λα λα + Dxγ01 f1 (w) Dxγ02 fM (w) λα λα , dw + Dxγ02 f1 (w) Dxγ01 fM (w) ⎤ M 2λ α Dxν fj ⎦ , ≤ 2 ρ∗ (x) ⎣ 0 ∞ ⎡

(14.176)

j=1

all x0 ≤ x ≤ b. Here we have (x − x0 )(2νλα −γ 1 λα −γ 2 λα +1) p(x)∞ ∗ ρ (x) := . (2νλα − γ 1 λα − γ 2 λα + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λα (14.177)

Similarly, by (14.177), we derive Theorem 14.57. Here all are as in Theorem 14.51. Case of 0, λα = λβ . Then ⎧⎧ ⎡ λα γ λα ⎨⎨M −1 γ 1 ∂R1 fj (x) ∂R21 fj+1 (x) ⎣ ∂rγ 1 ∂rγ 2 A ⎩⎩ j=1 ⎤⎫ γ ∂ 2 f (x) λα ∂ γ 1 f (x) λα ⎬ R1 j+1 ⎦ R1 j + ∂rγ 2 ⎭ ∂rγ 1

λν =

⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λα 1 M R R 1 1 + ⎣ ∂rγ 1 ∂rγ 2 ⎤⎫ γ ∂ 2 f (x) λα ∂ γ 1 f (x) λα ⎬ R1 M ⎦ R1 1 dx + ∂rγ 2 ∂rγ 1 ⎭ ≤

4π N/2 Γ(N/2)

⎤ 2λα M ν ∂ f R1 j ⎦ . ρ∗ (R2 ) ⎣ ∂rν ⎡

j=1

(14.178)

∞

Here we have ρ∗ (R2 ) :=

R2N −1 (R2 − R1 )(2νλα −γ 1 λα −γ 2 λα +1) . (2νλα − γ 1 λα − γ 2 λα + 1)(Γ(ν − γ 1 + 1))λα (Γ(ν − γ 2 + 1))λα (14.179)

316

14. Canavati Fractional Multivariate Opial Inequalities

We need Theorem 14.58. [27]. (As in Theorem 14.50, λα = 0, λβ = λν .) It holds ⎧⎧ x −1 + ⎨⎨M γ Dx 2 fj+1 (w)λβ Dxν fj (w)λβ p(w) 0 0 ⎩⎩ x0 j=1

λ λ , + Dxγ02 fj (w) β Dxν0 fj+1 (w) β + λ λ + Dxγ02 fM (w) β Dxν0 f1 (w) β λ λ , + Dxγ02 f1 (w) β Dxν0 fM (w) β dw ⎤ ⎡ M ν 2λβ (x − x0 )(νλβ −γ 2 λβ +1) p(x)∞ Dx fj ⎦ , (14.180) ⎣ ≤ 2· 0 ∞ (νλβ − γ 2 λβ + 1)(Γ(ν − γ 2 + 1))λβ j=1 all

x0 ≤ x ≤ b.

Similarly, by (14.180), we give the following. Theorem 14.59. Here all are as in Theorem 14.51. Case of 0, λβ = λν . Then ⎧⎧ ⎡ λβ λ ⎨⎨M −1 γ 2 ν ∂R ∂R1 fj+1 (x) f (x) β 1 j ⎣ ∂rν ∂rγ 2 A ⎩⎩ j=1

λα =

⎤⎫ γ ∂ 2 f (x) λβ ∂ ν f (x) λβ ⎬ j R1 j+1 ⎦ + R1 γ ∂r 2 ⎭ ∂rν ⎡ ∂ γ 2 f (x) λβ ∂ ν f (x) λβ M R1 1 R 1 + ⎣ ∂rγ 2 ∂rν ⎤⎫ γ ∂ 2 f (x) λβ ∂ ν f (x) λβ ⎬ 1 R1 M ⎦ dx ≤ + R1 γ ∂r 2 ∂rν ⎭

4π N/2 Γ(N/2)

R2N −1 (R2 − R1 )(νλβ −γ 2 λβ +1) (νλβ − γ 2 λβ + 1)(Γ(ν − γ 2 + 1))λβ

⎤ ⎡ M 2λβ ν ∂R f 1 j ⎦. ⎣ ∂rν ∞ j=1 (14.181)

14.2 Results

317

To extend the above research we give a motivational result regarding fractional integration by parts. Proposition 14.60. Let f ∈ C 1 ([0, 1]) and g ∈ C ν ([0, 1]), ν > 0, n := [ν], α := ν − n. Then

1

f (x)g

(ν)

(n)

(x)dx = f (1)(J1−α g

0

)(1) −

1

(J1−α g (n) )(x)f (x)dx.

0

(14.182)

Proof. Here g (ν) = d (J1−α g (n) )(x)/dx; that is, d (J1−α g (n) )(x) = g (x) dx. Hence, by ordinary integration by parts we have: 1 1 f (x)g (ν) (x) dx = f (x) d (J1−α g (n) )(x) = (ν)

0

0

f (1)(J1−α g (n) )(1) −

1

(J1−α g (n) )(x)f (x) dx,

0

by (J1−α g (n) )(0) = 0. Now we are ready to give Deﬁnition 14.61. Let ν > 0, n := [ν], α := ν − n, g : [0, 1] → R such that there exists g (n) which is measurable. Assume that (J1−α g (n) ) ∈ L1 ([0, 1]). We say that g (ν) ∈ L1 ([0, 1]) is a weak fractional derivative of order ν for g, iﬀ 1 1 u(x) g (ν) (x) dx = − (J1−α g (n) )(x)u (x) dx, (14.183) 0

0

∞

∀ u ∈ C ([0, 1]) : u(1) = 0. Based on the above Deﬁnition 14.61, we can extend the concept of weak fractional diﬀerentiation to anchor points x0 = 0, and to the multivariate case, especially to the radial case. Then we try to generalize the results of this chapter.

15 Riemann–Liouville Fractional Multivariate Opial-Type Inequalities over a Spherical Shell

Here we introduce the concept of the Riemann–Liouville fractional radial derivative for a function deﬁned on a spherical shell. Using polar coordinates we are able to derive multivariate Opial-type inequalities over a spherical shell of RN , N ≥ 2, by studying the topic in all possibilities. Our results involve one, two, or more functions. We also produce several generalized univariate fractional Opial-type inequalities, many of which are used to achieve the main goals. This treatment is based on [45].

15.1 Introduction This chapter is motivated by articles of Opial [315], Bessack [80], Anastassiou, Koliha, and Pecaric [64, 65], and Anastassiou [46, 48]. Opial-type inequalities usually ﬁnd applications in establishing the uniqueness of solution of initial value problems for diﬀerential equations and their systems; see Willett [406]. In this chapter we present a series of various Riemann–Liouville fractional multivariate Opial-type inequalities over spherical shells. To achieve our goal we use polar coordinates, and we introduce and use the Riemann–Liouville fractional radial derivative. We work on the spherical shell, and not on the ball, because a radial derivative cannot be deﬁned at zero. So, we reduce the problem to a univariate one. Consequently we use a large array of univariate Opial-type inequalities involving generalized Riemann–Liouville fractional derivatives; these are Riemann–Liouville fractional derivatives deﬁned at the arbitrary anchor G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 319 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 15,

320

15. Fractional Multivariate Opial Inequalities

point a ∈ R. So we also present a very large set of generalized univariate Riemann–Liouville fractional Opial-type inequalities transferred from earlier ones, proved at anchor point zero for the standard Riemann–Liouville fractional derivative. In our results we involve one, two, or several functions. But ﬁrst we need to develop an extensive background in three parts, and then follow the main results in three subsections.

15.2 Background—I Here we follow [356, pp. 149–150] and [383, pp. 87–88]. Also here RN , N > 1 denotes the N -tuple of reals R, and N denotes the natural numbers. Let us denote by dx ≡ λRN (dx) the Lebesgue measure on RN , N > 1, and S N −1 := {x ∈ RN : |x| = 1} the unit sphere on RN , where | · | stands for the Euclidean norm in RN . Also denote the ball B(0, R) := {x ∈ RN : |x| < R} ⊆ RN , R > 0, and the spherical shell A := B(0, R2 ) − B(0, R1 ), 0 < R1 < R2 . For x ∈ RN − {0} we can write uniquely x = rw, where r = |x| > 0 , and w = x/r ∈ S N −1 , |w| = 1. Clearly here RN − {0} = (0, ∞) × S N −1 , also the map Φ : RN − {0} → S N −1 : Φ(x) =

x |x|

is continuous. In addition, A = [R1 , R2 ] × S N −1 . Let us denote by dw ≡ λS N −1 (w) the surface measure on S N −1 to be deﬁned as the image under Φ of N · λRN restricted to the Borel class of B(0, 1) − {0}. More precisely, the last deﬁnition is as follows. Let A ⊂ S N −1 be a Borel set, and let A˜ := {ru : 0 < r < 1, u ∈ A} ⊂ RN ; we deﬁne ˜ λS N −1 (A) = N · λRN (A). Noting Φ(rx) = Φ(x), all r > 0 and x ∈ RN − {0}, one can conclude that N f ◦ Φ(x)dx = r f ◦ Φ(x)dx B(0,r)−{0}

and thus

rN f ◦ Φ(x)dx = N B(0,r)−{0}

B(0,1)−{0}

S N −1

f (w)λS N −1 (dw),

15.2 Background—I

321

for all f nonnegative and measurable functions on (S N −1 , BS N −1 ); B stands for the Borel class. We denote by 2π N/2 dw = wN ≡ λS N −1 (S N −1 ) = Γ(N/2) S N −1 the surface area of S N −1 and we get the volume |B(0, r)| =

2π N/2 rN wN rN = , N N Γ(N/2)

so that |B(0, 1)| = 2π N/2 /(N Γ(N/2)). Clearly here Vol(A) = |A| = Next, deﬁne

2π N/2 (R2N − R1N ) wN (R2N − R1N ) = . N N Γ(N/2)

ψ : (0, ∞) × S N −1 → RN − {0}

by ψ(r, w) := rw; ψ is a one-to-one and onto function. Thus (r, w) ≡ ψ −1 (x) = (|x|, Φ(x)) are called the polar coordinates of x ∈ RN − {0}. Finally, deﬁne the measure RN on (0, ∞), B(0,∞) by RN (Γ) = rN −1 dr, any Γ ∈ B(0,∞) . Γ

We mention the following very important theorem. Theorem 15.1. (See exercise 6, pp. 149–150 in [356] and Theorem 5.2.2 pp. 87–88 of [383].) We have that λRN = (RN × λS N −1 ) ◦ ψ −1 on BRN −{0} . particular, if f is a nonnegative Borel measurable function on In RN , BRN , then the Lebesgue integral N −1 f (x)dx = r f (rw)λS N −1 (dw) dr RN

(0,∞)

S N −1

=

f (rw)r S N −1

N −1

dr λS N −1 (dw).

(15.1)

(0,∞)

Clearly 15.1 is true for f a Borel integrable function taking values in R.

322

15. Fractional Multivariate Opial Inequalities

Using the facts that: (i) The Lebesgue measure of a Lebesgue measurable set K equals the Lebesgue measure of a Borel set (i.e., there exist M , an Fσ , and T a Gδ sets: M ⊂ K ⊂ T with λRN (K) = λRN (M ) = λRN (T ) ; see [354], p. 62). (ii) For each g Lebesgue measurable function, there exists an f Borel measurable function such that g = f a.e. (see [356, p. 145]), we get valid that (15.1) is true for Lebesgue integrable functions f : RN → R. We give the important Proposition 15.2. Let f : B(0, R) → R, R > 0, be a Lebesgue integrable function. Then R

f (x)dx = S N −1

B(0,R)

f (rw)rN −1 dr dw.

(15.2)

0

⎧ ⎨ f (x), x ∈ B(0, R),

Proof. Call F (x) :=

⎩

0, x ∈ RN − B(0, R).

Then apply (15.1) for F to get (15.2) easily. At last here we give the main tool for this chapter. Proposition 15.3. Let f : A → R be a Lebesgue integrable function, where A := B(0, R2 ) − B(0, R1 ), 0 < R1 < R2 . Then

f (x)dx =

f (rw)r S N −1

A

R2

N −1

R1

Proof. Apply (15.1) for ⎧ ⎨ f (x), x ∈ A, F (x) := then (15.3) is valid.

⎩

0, x ∈ RN − A;

dr dw.

(15.3)

15.3 Background—II

323

We also need the following well-known result. Proposition 15.4. Let f : [a, b] → R, be a Lebesgue integrable function. Then b−a b f (z)dz = f (t + a)dt. (15.4) a

0

So if fa (t) := f (a + t), the translation of f , then b−a b f (z)dz = fa (t)dt. a

(15.5)

0

15.3 Background—II Here we deﬁne the Riemann–Liouville fractional derivative that we use. Deﬁnition 15.5. (See [64, 65, 187].) Let α > 0. For any f ∈ L1 (0, x); x > 0, the Riemann–Liouville fractional integral of f of order α is deﬁned by s 1 (Jα f ) (s) := (s − t)α−1 f (t)dt, (15.6) Γ(α) 0 all s ∈ [0, x], and the Riemann–Liouville fractional derivative of f of order α by m s d 1 Dα f (s) := (s − t)m−α−1 f (t)dt, (15.7) Γ(m − α) ds 0 where m = [α] + 1, [·] is the integer part. In addition, we set D0 f := f := J0 f, J−α f = Dα f, if α > 0, D−α f := Jα f, if 0 < α ≤ 1. If α ∈ N, then Dα f = f (α) the ordinary derivative. Deﬁnition 15.6. (See [187].) We say that f ∈ L1 (0, x) has an L∞ fractional derivative Dα f in [0, x] if x > 0, if and only if Dα−k f ∈ C([0, x]), k = 1, . . . , m := [α] + 1; α > 0, and Dα−1 f ∈ AC([0, x]) (absolutely continuous functions),

324

15. Fractional Multivariate Opial Inequalities

and Dα f ∈ L∞ (0, x). We mention Lemma 15.7. (See [187].) Let β > α ≥ 0. Let f ∈ L1 (0, x), x > 0, have an L∞ fractional derivative Dβ f in [0, x] and let Dβ−k f (0) = 0 for k = 1, . . . , [β] + 1. Then Dα f (s) :=

1 Γ(β − α)

s

(s − t)β−α−1 Dβ f (t)dt,

(15.8)

0

all s ∈ [0, x]. Clearly here Dα f ∈ AC([0, x]) for β − α ≥ 1 and Dα f ∈ C([0, x]), for β − α ∈ (0, 1), hence Dα f ∈ L∞ (0, x), and Dα f ∈ L1 (0, x). Next, we deﬁne the generalized Riemann–Liouville fractional derivative with arbitrary anchor point a ∈ R; see [64]. Deﬁnition 15.8. Let v ≥ 0, deﬁne (Dav f )(s) := (Dv fa )(s − a), s ≥ a,

(15.9)

for v = 0 both sides equal to f (s), and for v = n ∈ N we easily get that (Dan f )(s) = f (n) (s), the ordinary derivative. Clearly here (Dav f )(z + a) = (Dv fa )(z).

(15.10)

We use p(s) and Dav f (s) in L∞ (a, x), x > a, a, x ∈ R. In that case by using (15.5) we obtain w w−a v p(y)(Da f )(y)dy = p(z + a)(Dv fa )(z)dz, (15.11) a

0

for all a ≤ w ≤ x, a, x ∈ R, which identity we use a lot in this chapter. Our initial intention is to transfer Riemann–Liouville fractional Opial inequalities, [46, 48, 64, 65] applied to fa over [0, w − a], for f over [a, w] and use the generalized Riemann–Liouville fractional derivative. For that we observe the following. Lemma 15.9. f ∈ L1 (a, w) if and only if fa ∈ L1 (0, w − a), where w ≥ a, a, w ∈R; fa (t) := f (a + t).

15.3 Background—II

Proof. We see that w

w−a

|f (z)|dz =

|fa (t)|dt.

a

325

0

We need Lemma 15.10. Let F (s) := f (s−a), a ∈ R be ﬁxed. Here f : [0, w−a] → R, where w > a and F : [a, w] → R. Then (i) F ∈ C([a, w]) if and only if f ∈ C([0, w − a]). (ii) F ∈ L∞ (a, w) if and only if f ∈ L∞ (0, w − a). (iii) F ∈ AC([a, w]) if and only if f ∈ AC([0, w − a]). Proof. It is based on the fact that the map g : [a, w] → [0, w − a], such that g(s) := s − a is one to one and onto. (i) (⇒) Let F be continuous, and let zn , z ∈ [0, w − a] : zn → z; that is, zn + a → z + a, here zn + a, z + a ∈ [a, w]. Hence F (zn + a) → F (z + a) (i.e., f (zn ) → f (z)), proving continuity of f . (⇐) Let f be continuous, and let sn → s; sn , s ∈ [a, w] ⇐⇒ sn − a, s − a ∈ [0, w − a], and sn − a → s − a. Hence f (sn − a) → f (s − a); that is, F (sn ) → F (s). That is, F is continuous. (ii) We see that |F (s)| = |f (s − a)| ≤ f ∞,

[0,w−a]

a.e. in s ∈ [a, w]. Hence F ∞,

[a,w]

≤ f ∞,

[0,w−a] .

Also |f (s − a)| = |F (s)| ≤ F ∞,

[a,w]

a.e. in s ∈ [a, w]. So that f ∞,

[0,w−a]

F ∞,

[a,w]

≤ F ∞,

[a,w] ;

that is, proving the claim.

= f ∞,

[0,w−a] ,

326

15. Fractional Multivariate Opial Inequalities

(iii) (⇒) Let F be absolutely continuous; that is, for every > 0 there is a δ > 0 such that whenever (a1 , b1 ), . . . , (an , bn ) are disjoint open subintervals of [a, w], then n

(bi − ai ) < δ ⇒

i=1

n

|F (bi ) − F (ai )| < .

i=1

Here (ai − a, bi − a) ⊂ [0, w − a], i = 1, . . . , n and also disjoint. Rewriting the last statement we have n

((bi − a) − (ai − a)) < δ ⇒

i=1

n

|f (bi − a) − f (ai − a)| < ;

i=1

that is, f is absolutely continuous. Notice that any open subinterval (ai , bi ) ⊂ [0, w − a] has the form (ai − a, bi − a), where (ai , bi ) ⊂ [a, w], for each i = 1, . . . , n; by ai = ai − a, bi = bi − a. (⇐) Assume now f is absolutely continuous; that is, for every > 0 there is a δ > 0: for any (a1 , b1 ), . . . , (an , bn ) that are disjoint subintervals of [0, w − a], then n

(bi − ai ) < δ ⇒

n

i=1

|f (bi ) − f (ai )| < .

i=1

The last statement is rewritten as (∀ > 0) (∃ δ > 0) : (a1 + a, b1 + a), . . . , (an + a, bn + a) ⊂ [a, w] (they are disjoint open subintervals); then n

((bi + a) − (ai + a)) < δ ⇒

i=1

n

|F (bi + a) − F (ai + a)| < ,

i=1

by f (bi ) = F (bi + a), f (ai ) = F (ai + a). Therefore F is absolutely continuous. Notice again here that any open subinterval (ai , bi ) ⊂ [a, w] has the form (ai + a, bi + a), where (ai , bi ) ⊂ [0, w − a] all i = 1, . . . , n; by ai = ai + a, bi = bi + a . We need Lemma 15.11. Here a < w, a, w ∈ R. Then p(s) ∈ L∞ (a, w) if and only if δ(z) := p(a + z) ∈ L∞ (0, w − a). In fact δ∞,

[0,w−a]

= p(s)∞,

[a,w] .

15.3 Background—II

327

Proof. Let L1 stand for the class of Lebesgue measurable sets. Assume p(s) is Lebesgue measurable on [a, w]. Then for any c ∈ R we have L1 ([a, w]) {x ∈ [a, w] : p(x) ≤ c} = {a+(x−a) ∈ [a, w] : p(a+(x−a)) ≤ c} = a + {(x − a) ∈ [0, w − a] : p(a + (x − a)) ≤ c} = a + {u ∈ [0, w − a] : p(a + u) ≤ c}.

Therefore {u ∈ [0, w − a] : δ(u) ≤ c} = {u ∈ [0, w − a] : p(a + u) ≤ c} = −a + {x ∈ [a, w] : p(x) ≤ c} ∈ L1 ([0, w − a]) , for all c ∈ R. Hence δ is Lebesgue measurable on [0, w − a]. Assume now that δ is Lebesgue measurable on [0, w − a]. Then for any c ∈ R we have L1 ([0, w − a]) {z ∈ [0, w − a] : δ(z) ≤ c} = {z ∈ [0, w − a] : p(a + z) ≤ c} = {(a + z) − a ∈ [0, w − a] : p(a + z) ≤ c} = −a + {(a + z) ∈ [a, w] : p(a + z) ≤ c} = −a + {x ∈ [a, w] : p(x) ≤ c}; that is, {x ∈ [a, w] : p(x) ≤ c} = a + {z ∈ [0, w − a] : δ(z) ≤ c} ∈ L1 ([a, w]) , for all c ∈ R. Hence p(s) is Lebesgue measurable on [a, w]. We do have that |δ(z)| = |p(a + z)| ≤ p(s)∞, a.e. z ∈ [0, w − a]. Hence δ∞,

[0,w−a]

[a,w] ,

≤ p(s)∞,

|p(a + z)| = |δ(z)| ≤ δ∞,

[a,w] .

Also we have

[0,w−a] ,

a.e. z ∈ [0, w − a]. Hence p(s)∞,[a,w] ≤ δ∞,[0,w−a] , proving the claim. We continue with Deﬁnition 15.12. We say that f ∈ L1 (a, w), a < w; a, w ∈ R has an L∞ fractional derivative Daβ f (β > 0) in [a, w], if and only if (1) Daβ−k f ∈ C([a, w]), k = 1, . . . , m := [β] + 1; (2)Daβ−1 f ∈ AC([a, w]); and (3)Daβ f ∈ L∞ (a, w). Based on Lemmas 15.9 and 15.10, the last deﬁnition 15.12 is equivalent step by step to

328

15. Fractional Multivariate Opial Inequalities

Deﬁnition 15.13. We say that fa (s) := f (a + s) ∈ L1 (0, w − a) has an L∞ fractional derivative Dβ fa in [0, w − a], β > 0, w > a; a, w ∈ R, if and only if (1) Dβ−k fa ∈ C([0, w − a]), k = 1, . . . , m := [β] + 1; (2) Dβ−1 fa ∈ AC([0, w − a]); and (3) Dβ fa ∈ L∞ (0, w − a). Deﬁnition 15.14. Here we deﬁne for s ≥ a, Daβ−m f (s) = Daβ−([β]+1) f (s) := Dβ−([β]+1) fa (s−a) = J([β]+1−β) fa (s−a) = 1 Γ ([β] + 1 − β)

s−a

(s − (a + t))[β]−β f (a + t)dt,

(15.12)

0

where f ∈ L1 (a, w), a < w; a, w ∈ R. Notice that 0 < [β] + 1 − β ≤ 1. If f ∈ L∞ (a, w), then s 1 (s − t)[β]−β f (t)dt. Daβ−m f (s) = Γ(m − β) a Remark 15.15. Notice that β−k Da f (a) = Dβ−k fa (0), f or k = 1, . . . , [β] + 1.

(15.13)

Based on Lemma 15.7 we get Lemma 15.16. Let β > α ≥ 0 and let f ∈ L1 (a, w) (if and only if fa ∈ L1 (0, w − a), a < w; a, w ∈ R ) have an L∞ fractional derivative Daβ f in [a, w] (if and only if fa have an L∞ fractional derivative Dβ fa in [0, w − a]), and let β−k Da f (a) = 0, k = 1, . . . , [β] + 1 (which is the same as Dβ−k fa (0) = 0, f or k = 1, . . . , [β] + 1 ). Then s 1 α (s − t)β−α−1 Dβ fa (t)dt, (15.14) (i) D fa (s) = Γ(β − α) 0 all s ∈ [0, w − a]. (ii)

Daα f (s)

1 = Γ(β − α)

s

(s − t)β−α−1 Daβ f (t)dt,

(15.15)

a

all a ≤ s ≤ w. Clearly here Daα f ∈ AC([a, w]) for β − α ≥ 1, and Daα f ∈ C([a, w]) for β − α ∈ (0, 1);

15.4 Background—III

329

hence Daα f ∈ L∞ (a, w) and Daα f ∈ L1 (a, w), and likewise for Dα fa on [0, w − a]. Proof. By Lemma 15.7, and by Deﬁnition 15.8 and (15.14), we have (15.14)

(Daα f ) (s) = (Dα fa ) (s−a)

=

1 Γ(β − α) (15.4)

=

proving (15.15).

=

s−a

1 Γ(β − α)

s−a

((s−a)−t)β−α−1 Daβ f (t+a)dt

0

(s − (t + a))β−α−1 Daβ f (t + a)dt

0

1 Γ(β − α)

s

(s − t)β−α−1 Daβ f (t)dt,

(15.16)

a

15.4 Background—III We make Remark 15.17. Let f ∈ L1 (a, w), where a < w; a, w ∈ R. Let β > 0, a ≤ s ≤ w; by Deﬁnition 15.8 we have m s−a β d 1 (s − a − t)m−β−1 f (a + t)dt, (15.17) Da f (s) = Γ(m − β) ds 0 where m := [β] +1. If β = 0, then Daβ f (s) = f (s). Let now F ∈ L1 (A) = L1 [R1 , R2 ] × S N −1 . For a ﬁxed w ∈ S N −1 , deﬁne gw (r) := F (rw) = F (x), where x ∈ A := B(0, R2 ) − B(0, R1 ), x 0 < R1 ≤ r ≤ R2 , r = |x|, w = ∈ S N −1 . r By Fubini’s theorem gw ∈ L1 [R1 , R2 ], B[R1 ,R2 ] , RN , for λS N −1 -almost every w ∈ S N −1 . Call / L1 [R1 , R2 ], B[R1 ,R2 ] , RN } K(F ) := {w ∈ S N −1 : gw ∈ = {w ∈ S N −1 : F (· w) ∈ (15.18) / L1 [R1 , R2 ], B[R1 ,R2 ] , RN }. That is, λS N −1 (K(F )) = 0. Of course, Θ(F ) := [R1 , R2 ] × K(F ) ⊂ A and λRN (Θ(F )) = 0.

330

15. Fractional Multivariate Opial Inequalities

By (15.17) then we have

β DR g 1 w

1 (r) = Γ(m − β)

d dr

m

r−R1

(r − R1 − t)m−β−1 gw (R1 + t)dt,

0

where β > 0, m := [β] + 1, r ∈ [R1 , R2 ]. If β = 0, then

β DR g 1 w

(15.19) (r) =

gw (r). Formula (15.19) is written for all w ∈ S N −1 − K(F ). We set β g DR (r) ≡ 0, for all w ∈ K(F ), r ∈ [R1 , R2 ], any β > 0. w 1

The above leads to the following deﬁnition. Deﬁnition 15.18. Let β > 0, m := [β] + 1, F ∈ L1 (A); A is the spherical shell. We deﬁne ⎧ β ⎨ F (x) ∂R 1 := ⎩ ∂rβ

1 Γ(m−β)

d m r−R1 (r dr 0

0, for w ∈ K(F ),

− R1 − t)m−β−1 F ((R1 + t)w)dt, for w ∈ S N −1 − K(F ),

where x = rw ∈ A, r ∈ [R1 , R2 ], w ∈ S N −1 . If β = 0, deﬁne β F (x) ∂R 1 := F (x). ∂rβ We call β ∂R F (x) 1 ∂rβ the Riemann–Liouville radial fractional derivative of F of order β.

(15.20)

We make Remark 15.19. If f ∈ L∞ (a, w), then (15.17) becomes β Da f (s) =

1 Γ(m − β)

d ds

m

s

(s − t)m−β−1 f (t)dt,

(15.21)

a

β > 0, m := [β] + 1, s ∈ [a, w]. If F is a Lebesgue measurable function from A into R and bounded, that is, there exists M ∗ > 0 : |F (x)| ≤ M ∗ , all x ∈ A, then of course F ∈ L1 (A). Clearly then |gw (r)| ≤ M ∗ , all r ∈ [R1 , R2 ], and all w ∈ S N −1 .

15.4 Background—III

331

Therefore (15.19) becomes

β DR g 1 w

1 (r) = Γ(m − β)

d dr

m

r

(r − t)m−β−1 gw (t)dt,

(15.22)

R1

where m := [β] + 1, β > 0, r ∈ [R1 , R2 ], for all w ∈ S N −1 − K(F ). In this last case, (15.20) becomes ⎧ d m r 1 (r − t)m−β−1 F (tw)dt, ⎪ ⎪ Γ(m−β) dr R1 β ⎨ ∂R1 F (x) := for w ∈ S N −1 − K(F ), ⎪ ∂rβ ⎪ ⎩ 0, for w ∈ K(F ), (15.23) where x = rw ∈ A, r ∈ [R1 , R2 ], w ∈ S N −1 . We need Theorem 15.20. Let β > α > 0 and F ∈ L1 (A). Assume that β ∂R F (x) 1 ∈ L∞ (A). ∂rβ β F (rw) takes real values for almost all r ∈ [R1 , R2 ], Further assume that DR 1 β N −1 , and for these |DR F (rw)| ≤ M1 for some M1 > 0. for each w ∈ S 1 N −1 − K(F ), we assume that F (· w) have an L∞ fractional For each w ∈ S β derivative DR F (· w) in [R1 , R2 ], and that 1 β−k F (R1 w) = 0, k = 1, . . . , [β] + 1. DR 1

Then α F (x) α ∂R 1 1 = DR1 F (rw) = α ∂r Γ(β − α)

r

R1

β (r − t)β−α−1 DR F (tw)dt, 1

(15.24) is true for all x ∈ A, that is, is true for all r ∈ [R1 , R2 ] and for all w ∈ S N −1 . Here α DR1 F (· w) ∈ AC([R1 , R2 ]) for β − α ≥ 1 and

α DR F (· w) ∈ C([R1 , R2 ]) for β − α ∈ (0, 1), 1

for all w ∈ S N −1 . Furthermore α ∂R F (x) 1 ∈ L∞ (A). ∂rα

332

15. Fractional Multivariate Opial Inequalities

In particular, it holds 1 F (x) = F (rw) = Γ(β)

r

R1

β (r − t)β−1 DR F (tw)dt, 1

(15.25)

for all r ∈ [R1 , R2 ] and w ∈ S N −1 − K(F ); x = rw, and F (· w) ∈ AC([R1 , R2 ]) for β ≥ 1 and F (· w) ∈ C([R1 , R2 ]) for β ∈ (0, 1), for all w ∈ S

N −1

− K(F ).

Proof. Here we observe that for each w ∈ S N −1 − K(F ), we have F (· w) ∈ L1 [R1 , R2 ], B[R1 ,R2 ] , RN . By our assumptions and Lemma 15.16, we have r α 1 β DR1 F (rw) = (r − t)β−α−1 DR F (tw)dt, 1 Γ(β − α) R1

(15.26)

for all r ∈ [R1 , R2 ] and w ∈ S N −1 − K(F ), for β > α ≥ 0, thus initially proving (15.25) by setting α = 0 in (15.26). Here α F (· w) ∈ AC([R1 , R2 ]), for all w ∈ S N −1 − K(F ), β − α ≥ 1, DR 1

and α F (· w) ∈ C([R1 , R2 ]), for β − α ∈ (0, 1). DR 1

Formula (15.26) for α > 0, is true for all r ∈ [R1 , R2 ] and w ∈ S N −1 , and α DR1 F (· w) ∈ AC([R1 , R2 ]), for all w ∈ S N −1 , β − α ≥ 1 and α F (· w) ∈ C([R1 , R2 ]), for β − α ∈ (0, 1). DR 1

Thus proving (15.24). Fixing r ∈ [R1 , R2 ], the function β F (tw) δ r (t, w) := (r − t)β−α−1 DR 1

is measurable on

[R1 , r] × S N −1 , B [R1 ,r] × B S N −1 .

Here B [R1 ,r] × B S N −1 stands for the complete σ- algebra generated by B [R1 ,r] × B S N −1 , where B X stands for the completion of BX . Then we get that r r β β−α−1 |δ r (t, w)|dt dw = (r − t) |DR1 F (tw)|dt dw S N −1

R1

S N −1

R1

15.4 Background—III

8 ∂ β F (x) 8 8 8 ≤ 8 R1 β 8 ∂r ∞,

([R1 ,r]×S N −1 )

S N −1

r

333

(r − t)β−α−1 dt dw

R1

(15.27) 8 ∂ β F (x) 8 8 8 = 8 R1 β 8 ∂r ∞, 8 ∂ β F (x) 8 8 R1 8 8 8 ∂rβ ∞,

([R1 ,r]×S N −1 )

A

2π N/2 Γ(N/2)

2π N/2 Γ(N/2)

(r − R1 )β−α ≤ (β − α)

(R2 − R1 )β−α < ∞. (β − α)

(15.28)

Hence δ r (t, w) is integrable on [R1 , r] × S N −1 , B [R1 ,r] × B S N −1 . Consequently, by Fubini’s theorem and (15.24), we obtain that α DR1 F (rw), β > α > 0, is integrable in w over S N −1 , BS N −1 . So we have that α DR1 F (rw), β > α > 0, is continuous in r ∈ [R1 , R2 ] for each w ∈ S N −1 , and measurable in w ∈ S N −1 for each r ∈ [R1 , R2 ]. So, it is a Carath´eodory function. Here [R1 , R2 ] is a separable metric space and S N −1 is a measurable space, and the function takes values in R∗ = R ∪ {±∞}, is a metric space. αwhich F (rw), β > α > 0 is Therefore by Theorem 20.15, p. 156 of [10], DR 1 jointly B[R1 ,R2 ] × BS N −1 -measurable on [R1 , R2 ] × S N −1 = A, that is, Lebesgue measurable on A. Indeed then we have that r ⏐ ⏐ α 1 ⏐ β β−α−1 ⏐ D F (rw)| ≤ (r − t) F (tw) | DR ⏐ ⏐dt R1 1 Γ(β − α) R1 ≤

β DR F (· w)∞, 1

[R1 ,R2 ]

Γ(β − α) ≤

r

(r − t)β−α−1 dt R1

≤

(r − R1 )β−α M1 Γ(β − α) (β − α) (15.29)

M1 (R2 − R1 )β−α := τ < ∞, Γ(β − α + 1)

for all w ∈ S N −1 and all r ∈ [R1 , R2 ]; that is, we proved that α F (rw)| ≤ τ < ∞, | DR 1 for all w ∈ S N −1 and all r ∈ [R1 , R2 ], hence proving that α ∂R F (x) 1 ∈ L∞ (A). ∂rα

(15.30)

334

15. Fractional Multivariate Opial Inequalities

We have ﬁnished our proof. We have built the machinery to do Riemann–Liouville fractional Opialtype inequalities on the spherical shell. Now we are ready to present our main results next.

15.5 Main Results 15.5.1 Riemann–Liouville Fractional Opial-Type Inequalities Involving One Function We mention Theorem 15.21. (See [64].) Let 1/p + 1/q = 1 with p, q > 1, let γ ≥ 0, v > γ + 1 − 1/p, and f ∈ L1 (0, x) have an L∞ fractional derivative Dv f in [0, x], x > 0, such that Dv−j f (0) = 0, for j = 1, . . . , [v] + 1. Then

x

|D f (s)| ds

v

0

2/q

x

|D f (s)| |D f (s)|ds ≤ Ω(x) γ

v

q

(15.31)

0

where Ω(x) :=

x(rp+2)/p 21/q Γ(r + 1) ((rp + 1)(rp + 2))

(15.32)

1/p

and r := v − γ − 1.

(15.33)

We transfer Theorem 15.21 to arbitrary anchor point a ∈ R. We present Theorem 15.22. Let 1/p + 1/q = 1 with p, q > 1, let γ ≥ 0, v > γ + 1 − 1/p, and f ∈ L1 (a, x) have an L∞ fractional derivative Dav f in [a, x], a, x ∈ R, a < x, such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1. Then

x

|Daγ f (s)| |Dav f (s)|ds ≤ Ω(x − a) a

where Ω is as in (15.32).

2/q

x

|Dav f (s)|q ds a

,

(15.34)

15.5 Main Results

335

Proof. By Lemma 15.9, fa ∈ L1 (0, x − a) with L∞ fractional derivative Dv fa in [0, x − a]; see Deﬁnitions 15.12 and 15.13. Furthermore it holds (see (15.13)), Dv−j fa (0) = 0, for j = 1, . . . , [v] + 1. Therefore by (15.31) we have

x−a

x−a

|Dγ fa (s)| |Dv fa (s)|ds ≤ Ω(x−a)

|Dv fa (s)|q ds

0

2/q . (15.35)

0

Using (15.10) we have

x−a

x−a

| (Daγ f ) (s+a)| | (Dav f ) (s+a)|ds ≤ Ω(x−a)

0

2/q | (Dav f ) (s + a)|q ds .

0

By Lemma 15.16, we have that Daγ f ∈ AC([a, x]) for v − γ ≥ 1 and Daγ f ∈ C([a, x]) for v − γ ∈ (0, 1). Clearly then by Proposition 15.4 we get

x−a

| (Daγ f ) (s + a)| | (Dav f ) (s + a)|ds = 0

x

|Daγ f (s)| | (Dav f ) (s)|ds a

and

x−a

| (Dav f ) (s + a)|q ds = 0

(15.36)

x

| (Dav f ) (s)|q ds, a

notice here functions under right-hand side integrations are integrable. That proves (15.34). We mention Theorem 15.23. (See [64].) Let v > γ ≥ 0, and let f ∈ L1 (0, x) have an L∞ fractional derivative Dv f in [0, x], x > 0, such that Dv−j f (0) = 0, for j = 1, . . . , [v] + 1. Then

x

|Dγ f (s)| |Dv f (s)|ds ≤ Ω1 (x) ess sups∈[0,x] |Dv f (s)|2 ,

(15.37)

0

where Ω1 (x) =

xr+2 , r = v − γ − 1. Γ(r + 3)

(15.38)

336

15. Fractional Multivariate Opial Inequalities

We give the general transfer. Theorem 15.24. Let v > γ ≥ 0, and let f ∈ L1 (a, x) have an L∞ fractional derivative Dav f in [a, x], a, x ∈ R, a < x, such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1. Then x |Daγ f (s)| |Dav f (s)|ds ≤ Ω1 (x − a) ess sups∈[a,x] |Dav f (s)|2 ,

(15.39)

a

where Ω1 is as in (15.38). Proof. By Lemma 15.9, fa ∈ L1 (0, x − a) with L∞ fractional derivative Dv fa in [0, x − a]; see Deﬁnitions 15.12 and 15.13. Furthermore it holds (see (15.13)), Dv−j fa (0) = 0, f or j = 1, . . . , [v] + 1. Therefore by (15.37) we have

x−a

|Dγ fa (s)| |Dv fa (s)|ds ≤ Ω1 (x − a) ess sups∈[0,x−a] |Dv fa (s)|2 . 0

(15.40)

Using (15.10) we get

x−a

|Daγ f (s+a)| | (Dav f ) (s+a)|ds ≤ Ω1 (x−a)ess sups∈[0,x−a] |Dav f (s+a)|2 . 0

We have again by Proposition 15.4 that

x−a

|Daγ f (s + a)| |Dav f (s + a)|ds = 0

x

|Daγ f (s)| |Dav f (s)|ds.

(15.41)

a

Also by Lemma 15.11 we obtain ess sups∈[0,x−a] |Dav f (s + a)|2 = ess sups∈[a,x] |Dav f (s)|2 , thus proving the claim.

We give the transfer Theorem 15.25. Let 1/p + 1/q = 1 with 0 γ ≥ 0, and let f ∈ L1 (a, x) have an L∞ fractional derivative Dav f in [a, x], a, x ∈ R, a < x, such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1.

15.5 Main Results

337

Additionally assume (1/Dav f ) ∈ L∞ (a, x) and that Dav f has the same sign a.e. in (a, x). Then

x

|Daγ f (s)|

|Dav f (s)|ds

≥ Ω(x − a)

a

2/q

x

|Dav f (s)|q ds

(15.42)

a

where Ω is deﬁned by (15.32). Proof. Based on Theorem 2.3, of [64], special case of a = 0. It is a similar method of proving as in Theorem 15.22. We give the transfer. Theorem 15.26. Let 1/p + 1/q = 1 with p, q > 1 let γ ≥ 0, v ≥ γ + 2 − 1/p, and f ∈ L1 (a, x) have an L∞ fractional derivative Dav f in [a, x], a, x ∈ R, a < x, such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1. Then x

|Daγ f (s)|

|Daγ+1 f (s)|ds

≤ Ω2 (x − a)

a

2/q

x

|Dav f (s)|q ds

,

(15.43)

a

where Ω2 (t) :=

t2

(rp+1)/p 2

2 (Γ(r + 1)) (rp + 1)

2/p

, r = v − γ − 1, t ≥ 0.

(15.44)

Proof. Based on Theorem 2.4 of [64], special case of a = 0. It is a similar method of proving as in Theorem 15.22. We present the transfer. Theorem 15.27. Let γ ≥ 0, v > γ + 1, and let f ∈ L1 (a, x) have an L∞ fractional derivative Dav f in [a, x], a, x ∈ R, a < x, such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1. Then x |Daγ f (s)| |Daγ+1 f (s)|ds ≤ Ω3 (x − a) ess supt∈(a,x) |Dav f (t)|2 , (15.45) a

where Ω3 (t) :=

t2(v−γ)

2,

2 (Γ(v − γ + 1))

t ≥ 0.

(15.46)

338

15. Fractional Multivariate Opial Inequalities

Proof. Based on Theorem 2.5, of [64], special case of a = 0. It is a similar method of proving as in Theorem 15.24. We further give Proposition 15.28. Inequality (15.45) is sharp; namely it is attained by f∗ (s) := (s − a)v , a ≤ s ≤ x , v > γ + 1 , γ ≥0. Proof. Here we are acting as in Remark 2.6 of [64]. We use the known formula s Γ(u) Γ(v) (s − a)u+v−1 , u, v > 0. (15.47) (s − t)u−1 (t − a)v−1 dt = Γ(u + v) a Let 0 ≤ j ≤ [v] + 1, m := [v] − j + 1, α := v − [v]. We have 1 − α > 0, v + 1 > 0, and by (15.21) we obtain Dav−j f∗ (s) = Dav−j (s−a)v =

1 Γ(1 − α)

d ds

m

s

(s−t)(1−α)−1 (t−a)(v+1)−1 dt

a

(15.48)

=

Γ(1 − α) Γ(v + 1) 1 Γ(1 − α) Γ(m + j + 1)

d ds

m (s − a)m+j =

Γ(v + 1) (s − a)j ; j! (15.49)

that is, Dav−j (s − a)v =

Γ(v + 1) (s − a)j . j!

(15.50)

Hence Dav−j f∗ (a) = 0, for j = 1, . . . , [v] + 1

(15.51)

Dav f∗ (s) = Dav (s − a)v = Γ(v + 1).

(15.52)

and Using Lemma 15.16, in particular applying (15.15), we obtain Daγ (s − a)v =

Γ(v + 1)(s − a)v−γ , Γ(v − γ + 1)

Daγ+1 (s − a)v =

Γ(v + 1) (s − a)v−γ−1 . Γ(v − γ) (15.53)

Therefore

2

x

|Daγ f∗ (s)| |Daγ+1 f∗ (s)|ds =

L.H.S. (15.45) = a

(Γ(v + 1)) Γ(v − γ) Γ(v − γ + 1)

15.5 Main Results

x

(s−a)2(v−γ)−1 ds = a

1 2

Γ(v + 1) Γ(v − γ + 1)

339

2 (x−a)2(v−γ) = R.H.S. (15.45), (15.54)

thus proving the claim.

We present Theorem 15.29. Let 1/p + 1/q = 1 with 0 γ + 1, and let f ∈ L1 (a, x) have an L∞ fractional derivative Dav f in [a, x], a, x ∈ R, a < x, such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1. Also assume (1/Dav f ) ∈ L∞ (a, x) and that Dav f has the same sign a.e. in (a, x). Then

x

|Daγ f (s)|

|Daγ+1 f (s)|ds

≥ Ω2 (x − a)

a

2/q

x

|Dav f (s)|q ds

,

(15.55)

a

where Ω2 is given by (15.44). Proof. We transfer here for arbitrary anchor point a ∈ R, Theorem 2.7 of [64]. We apply the earlier established method; see Theorem 15.22. We present Theorem 15.30. Let 1/p + 1/q = 1 with p, q > 1, let γ ≥ 0, v > γ + 1 − 1/p, and let f ∈ L1 (a, x) have an L∞ fractional derivative Dav f in [a, x], a, x ∈ R, a < x, such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1; let m > 0. Then x |Daγ f (s)|m ds ≤ Ω4 (x − a) a

m/q

x

|Dav f (s)|q ds

,

(15.56)

a

where Ω4 (t) :=

m

(Γ(r + 1))

t(rm+1+(m/p)) , rm + 1 + m (rp + 1)m/p p

r := v − γ − 1, t ≥ 0.

(15.57)

Proof. Based on Theorem 2.8 of [64] and its transfer to arbitrary anchor point a ∈ R.

340

15. Fractional Multivariate Opial Inequalities

We give Theorem 15.31. Let v > γ ≥ 0, and let f ∈ L1 (a, x) have an L∞ fractional derivative Dav f in [a, x], a, x ∈ R, a < x, such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1; let m > 0. Then x |Daγ f (s)|m ds ≤ Ω5 (x − a)esssupt∈[a,x] |Dav f (t)|m ,

(15.58)

a

where Ω5 (t) :=

m

(Γ(v − γ + 1))

t(v−γ)m+1 (v − γ − 1) m + 1 +

m p

((v − γ)m + 1)

,

t ≥ 0. (15.59)

Proof. Based on Theorem 2.9 of [64], and so on. We next give the notation valid for the rest of Section 15.5.1; we follow [65]. Notation 15.32. Here we call • l: a positive integer. • v, ri : positive real numbers, i = 1, . . . , l, r =

"l

i=1 ri .

• μi : real numbers satisfying 0 ≤ μi < v, i = 1, . . . , l. • αi = v − μi − 1, i = 1, . . . , l. • α = max{(αi )− : i = 1, . . . , l}, where (αi )− := (−αi )+ . • β = max{(αi )+ : i = 1, . . . , l}, where (αi )+ := max (αi , 0). • w1 , w2 : continuous positive weight functions on [a, x], a, x ∈ R, a < x. • w: continuous nonnegative weight function on [a, x]. • sk , sk : sk > 0 and 1/sk + 1/sk = 1 k = 1, 2. We write μ = (μ1 , . . . , μl ) for a selection of the orders μi of fractional derivatives, and r = (r1 , . . . , rl ) for a selection of the constants ri .

15.5 Main Results

341

We mention Theorem 15.33. [65] Let f ∈ L1 (0, x) have an L∞ fractional derivative Dv f in [0, x], x > 0, such that Dv−j f (0) = 0, for j = 1, . . . , [v] + 1. Here a = 0. For k = 1, 2, let sk > 1 and p > 0 satisfy αs2 < 1, p >

s2 1 − αs2

(15.60)

and let σ = 1/s2 − 1/p. Finally, let

x

Q1 :=

1/s1

s1

w1 (τ ) dτ

; Q2 :=

0

−s2 /p

w2 (τ )

r/s2 dτ

.

(15.61)

0

Then

x

w1 (τ ) 0

l

|Dμi f (τ )|ri dτ ≤

i=1

r/p

x

w2 (τ ) |D f (τ )| dτ

ρ+(1/s1 )

v

Q1 Q2 C1 x where ρ :=

x

p

,

(15.62)

0

"l i=1

αi ri + σr, and σ rσ

C1 := C1 (v, μ, r, p, s1 , s2 ) := &l i=1

. Γ(v − μi )ri (αi + σ)ri σ (ρs1 + 1)1/s1 (15.63)

We transfer the last theorem to arbitrary anchor point a ∈ R. Theorem 15.34. Here all constant and parameter notation is as in Theorem 15.33. Let f ∈ L1 (a, x), a < x, a, x ∈ R, have an L∞ fractional derivative Dav f in [a, x], such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1. Let Q1 (a) :=

x

s1

1/s1

w1 (τ ) dτ

, Q2 (a) :=

a

x

−s2 /p

w2 (τ )

r/s2 dτ

.

a

(15.64) Then

x

w1 (τ ) a

l i=1

|Daμi f (τ )|ri dτ ≤

342

15. Fractional Multivariate Opial Inequalities

r/p

x

Q1 (a)Q2 (a)C1 (x − a)ρ+(1/s1 )

w2 (τ ) |Dav f (τ )|p dτ

.

(15.65)

a

Proof. By Lemma 15.9, fa ∈ L1 (0, x − a) with an L∞ fractional derivative Dv fa in [0, x − a]; see Deﬁnitions 15.12, and 15.13. Furthermore it holds (see (15.13)) Dv−j fa (0) = 0, for j = 1, . . . , [v] + 1. Notice that

x−a

Q1 (a) :=

w1 (τ + a)s1 dτ

1/s1

x−a

, Q2 (a) :=

w2 (τ + a)−s2 /p dτ

r/s2 .

a

0

(15.66)

Next we apply (15.62) on [0, x − a] to fa with respect to w1 (a + τ ), w2 (a + τ ), τ ∈ [0, x − a]. We have

l

x−a

w1 (a + τ ) 0

|Dμi fa (τ )|ri dτ ≤

i=1

r/p

x−a

Q1 (a)Q2 (a)C1 (x − a)ρ+(1/s1 )

w2 (a + τ ) |Dv fa (τ )|p dτ

.

0

(15.67) Equivalently, via (15.10), we write

x−a

w1 (a + τ ) 0

l

|Daμi f (a + τ )|ri dτ ≤

i=1

Q1 (a)Q2 (a)C1 (x − a)

ρ+(1/s1 )

r/p

x−a

w2 (a + τ )

|Dav f (a

p

+ τ )| dτ

.

0

(15.68) μ By Lemma 15.16, we have that Da i f ∈ AC([a, x]). Hence, by Proposition 15.4, we get (15.65). Next, we apply Theorem 15.34 to the spherical shell A. We give Theorem 15.35. Here all constant and parameter notation is as in Theorem 15.33. Let f ∈ L1 (A) with v ∂R f (x) 1 ∈ L∞ (A), x ∈ A; ∂rv

A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2, 0 < R1 < R2 .

15.5 Main Results

343

v Further assume that DR f (rw) ∈ R for almost all r ∈ [R1 , R2 ], for each 1 N −1 v w ∈S , and for these |DR f (rw)| ≤ M1 for some M1 > 0. For each 1 N −1 w∈S − K(F ), we assume that f (· w) has an L∞ fractional derivative v f (· w) in [R1 , R2 ], and that DR 1 v−j f (R1 w) = 0, j = 1, . . . , [v] + 1. DR 1

"l We take p = i=1 ri , and 0 ≤ μ1 < μ2 ≤ μ3 ≤ . . . ≤ μl < v. If μ1 = 0 we set r1 = 1. Denote Q1 (R1 ) :=

(N −1)s1 +1

R2

⎛ Q2 (R1 ) :=

(N −1)s1 +1

− R1 (N − 1)s1 + 1

1/s1 , ⎞p/s2

(1−N )(s2 /p)+1 R 2 ⎝

(1−N )(s2 /p)+1 − R1 ⎠ s N )( p2 ) + 1

.

(15.69)

⏐ v l ⏐ μi ⏐ ⏐ ⏐ ∂R1 f (x) ⏐ri ⏐ ∂R1 f (x) ⏐p ∗ ⏐ ⏐ dx ≤ C ⏐ ⏐ dx , ∂rμi ∂rv A i=1 A

(15.70)

(1 −

Then

where

C ∗ := Q1 (R1 )Q2 (R1 )C1 (R2 − R1 )ρ+(1/s1 ) .

(15.71)

Proof. By Theorem 15.20 for μi > 0 we get that μ

∂Ri1 f (x) ∈ L∞ (A). ∂rμi In general here we get that l ⏐ μi ⏐ ⏐ ∂R1 f (x) ⏐ri ⏐ ⏐ ∈ L1 (A). ∂rμi i=1

Thus, by Proposition 15.3 we have l ⏐ μi l ⏐ R2 ⏐ ∂R1 f (x) ⏐ri μi ri N −1 I1 := |DR1 f (rw)| r dr dw ⏐ ⏐ dx = ∂rμi A i=1 S N −1 R1 i=1

R2

= (S N −1 −K(f ))

R1

l i=1

μ |DRi1 f (rw)|ri

r

N −1

dr dw.

(15.72)

344

15. Fractional Multivariate Opial Inequalities

⏐ ∂ v f (x) ⏐p ⏐ ⏐ 1 Because ⏐ R∂r ⏐ ∈ L1 (A), we also obtain v ⏐ v ⏐ R2 ⏐ ∂R1 f (x) ⏐p r p N −1 I2 := |DR1 f (rw)| r dr dw ⏐ ⏐ dx = ∂rv A S N −1 R1

R2

r |DR f (rw)|p rN −1 dr dw. 1

= (S N −1 −K(f ))

R1

(15.73)

Notice here f (· w) ∈ L1 ([R1 , R2 ]), for all w ∈ S N −1 − K(f ), and λS N −1 (K(f )) = 0. Setting w1 (r) = w2 (r) := rN −1 , r ∈ [R1 , R2 ], we use Theorem 15.34, for every w ∈ S N −1 − K(f ). We get

R2

R1

l

|DRi1 f (rw)|ri rN −1 dr μ

i=1

≤ Q1 (R1 )Q2 (R1 )C1 (R2 − R1 )

R2

ρ+(1/s1 ) R1

v |DR f (rw)|p 1

r

N −1

dr ; (15.74)

that is, we ﬁnd that

R2

R1

C

∗

R2

R1

l

|DRi1 f (rw)|ri rN −1 dr ≤ μ

i=1

v |DR f (rw)|p rN −1 dr, for all w ∈ S N −1 − K(f ). 1

(15.75)

Therefore

(S N −1 −K(f ))

≤C

∗

R1

(S N −1 −K(f ))

That is, thus proving (15.70).

R2

l

μ |DRi1 f (rw)|ri

dr dw

i=1

R2

R1

v |DR f (rw)|p 1

I1 ≤ C ∗ I2 ,

r

N −1

r

N −1

dr dw .

(15.76)

(15.77)

15.5 Main Results

345

We continue with Theorem 15.36. Let f ∈ L1 (a, x), a < x, a, x ∈ R have an L∞ fractional derivative Dav f in [a, x] such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1. Then x w(τ ) a

l

|Daμi f (τ )|ri dτ ≤

i=1

where ρ :=

"l

i=1 (v

w∞ (x − a)ρ Dav f r∞ , &l ρ i=1 Γ(v − μi + 1)ri

(15.78)

− μi )ri + 1.

Proof. This is a transfer of Theorem 2.2 of [65] and its proof. By (15.15) we have τ 1 μi (τ − t)αi |Dav f (τ )|dt, (15.79) |Da f (τ )| ≤ Γ(v − μi ) a implying that

Dav f ∞ (τ − a)v−μi . Γ(v − μi + 1)

(15.80)

Dav f r∞i (τ − a)(v−μi )ri , r (Γ(v − μi + 1)) i

(15.81)

|Daμi f (τ )| ≤ Hence |Daμi f (τ )|ri ≤ and w(τ )

l

"l

|Daμi f (τ )|ri

i=1

w∞ Dav f r∞ (τ − a) i=1 (v−μi )ri ≤ . &l ri i=1 (Γ(v − μi + 1))

(15.82)

Integrating (15.82) over [a, x] we get

x

w(τ ) a

l

w∞ Dav f r∞ ri i=1 (Γ(v − μi + 1))

|Daμi f (τ )|ri dτ ≤ &l

i=1

&l

w∞ Dav f r∞

i=1

proving (15.78).

ri

(Γ(v − μi + 1))

x

We apply Theorem 15.36 to the spherical shell A case.

v f ∂R 1 ∈ L∞ (A). ∂rv

i=1 (v−μi )ri

dτ =

a

(x − a)ρ , ρ

Theorem 15.37. Let f ∈ L1 (A) with

"l

(τ − a)

(15.83)

346

15. Fractional Multivariate Opial Inequalities

v Assume DR f (rw) ∈ R for almost all r ∈ [R1 , R2 ], for each w ∈ S N −1 , 1 v and for these |DR f (rw)| ≤ M1 for some M1 > 0. For each w ∈ S N −1 − 1 v f (· w) K(f ), we assume that f (· w) has an L∞ fractional derivative DR 1 in [R1 , R2 ], and that v−j DR f (R1 w) = 0, f or j = 1, . . . , [v] + 1. 1

We take 0 ≤ μ1 < μ2 ≤ μ3 ≤ . . . ≤ μl < v. If μ1 = 0 we set r1 = 1. Then l ⏐ μi ⏐ 2π N/2 R2N −1 (R2 − R1 )ρ M1r ⏐ ∂R1 f (x) ⏐ri , dx ≤ ⏐ ⏐ & l r ∂rμi ρ i=1 (Γ(v − μi + 1)) i Γ(N/2) A i=1 where ρ :=

"l

i=1 (v

(15.84)

− μi )ri + 1.

Proof. Here

l ⏐ μi ⏐ ⏐ ∂R1 f (x) ⏐ri ⏐ ⏐ ∈ L1 (A). ∂rμi i=1

Hence as before I1 :=

l ⏐ ∂ μi f (x) ⏐ ⏐ R1 ⏐ ri dx = ⏐ ⏐ ∂rμi A i=1 (S N −1 −K(f ))

R2 R1

l

|DRi1 f (rw)|ri r N −1 dr μ

dw.

i=1

(15.85)

Here we set w(r) := rN −1 , r ∈ [R1 , R2 ]. For each w ∈ S N −1 − K(f ) we apply Theorem 15.36. From (15.78) we obtain

R2 R1

l

|DRi1 f (rw)|ri rN −1 dr ≤ μ

i=1

≤

RN −1 (R2 − R1 )ρ v DR f (· w)r∞, [R1 ,R2 ] &l 2 1 ρ i=1 (Γ(v − μi + 1))ri (15.86)

R2N −1 (R2 − R1 )ρ M1r := θ, for all w ∈ S N −1 − K(f ). &l r ρ i=1 (Γ(v − μi + 1)) i

(15.87)

Therefore

I1 =

(S N −1 −K(f ))

R2

R1

l

μ |DRi1 f (rw)|ri

dr dw

i=1

≤θ

r

N −1

dw = θ

(S N −1 −K(f ))

=θ

2π N/2 , Γ(N/2)

dw

(15.88)

S N −1

(15.89)

15.5 Main Results

proving (15.84).

347

We continue with Theorem 15.38. Let f ∈ L1 (a, x), a < x, a, x ∈ R have an L∞ fractional derivative Dav f in [a, x] such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1. Assume also that Dav f has the same sign a.e. in (a, x). For k = 1, 2 , let 0 < sk < 1, let p < 0, and let σ := 1/s2 − 1/p, Q1 (a) and Q2 (a) as in (15.64), C1 as in (15.63). Then

x

w(τ ) a

l

|Daμi f (τ )|ri dτ ≥

i=1

r/p

x

Q1 (a)Q2 (a)C1 (x − a)

ρ+(1/s1 )

w2 (τ )|Dav f (τ )|p dτ

,

(15.90)

a

where ρ :=

l

αi ri + σr.

(15.91)

i=1

Proof. Similar to Theorem 15.34 with transfer of Theorem 2.3 of [65] to anchor point a ∈ R. We give Theorem 15.39. Let s1 , s2 , p ∈ (0, 1), rs1 ≤ 1 and s2 s2 1 1
x

w1 (τ ) a

l

|Daμi f (τ )|ri dτ ≥

i=1

Q1 (a)Q2 (a)C1 (x − a)

ρ+(1/s1 )

r/p

x

w2 (τ )|Dav f (τ )|p dτ a

.

(15.92)

348

15. Fractional Multivariate Opial Inequalities

Proof. Similar transfer of Theorem 2.4 from [65]. We apply Theorem 15.39 on the spherical shell A. Theorem 15.40. All parameters and constants are as in Theorem 15.39. "l We take p = i=1 ri , and 0 ≤ μ1 < μ2 ≤ μ3 ≤ · · · ≤ μl < v. If μ1 = 0 we set r1 = 1. Let f ∈ L1 (A) with v ∂R f 1 ∈ L∞ (A). v ∂r v Assume that DR f (rw) ∈ R for almost all r ∈ [R1 , R2 ], for each w ∈ S N −1 , 1 v and for these |DR1 f (rw)| ≤ M1 for some M1 > 0. For each w ∈ S N −1 − v f (· w) K(F ), we assume that f (· w) has an L∞ fractional derivative DR 1 in [R1 , R2 ], and that v−j DR f (R1 w) = 0, j = 1, . . . , [v] + 1, 1 v also DR (· w) has the same sign a.e. in [R1 , R2 ]. Then 1

⏐ v l ⏐ μi ⏐ ⏐ ⏐ ∂R1 f (x) ⏐ri ⏐ ∂R1 f (x) ⏐p ∗ dx ≥ C dx , ⏐ ⏐ ⏐ ⏐ ∂rμi ∂rv A i=1 A

(15.93)

where C ∗ := Q1 (R1 )Q2 (R1 )C1 (R2 − R1 )ρ+(1/s1 ) .

Proof. Similar to Theorem 15.35 by using (15.92).

(15.94)

We continue with Theorem 15.41. Let f ∈ L1 (a, x), a < x, a, x ∈ R, have an L∞ fractional derivative Dav f in [a, x], such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1. Let v > μ2 ≥ μ1 + 1 ≥ 1. If p, q > 1 : 1/p + 1/q = 1; then a

x

|Daμ1 f (τ )| |Daμ2 f (τ )|dτ ≤ C2 (x − a)2v−μ1 −μ2 −1+(2/q)

2/p

x

|Dav f (τ )|p dτ a

where C2 = C2 (v, μ1 , μ2 , p) is given by

,

(15.95)

15.5 Main Results

349

1 (1/p) C2 :=

2 . Γ(v − μ1 )Γ(v − μ2 + 1) ((v − μ1 )q + 1)1/q ((2v − μ1 − μ2 − 1)q + 2)1/q (15.96)

Proof. Transfer to a ∈ R of Theorem 2.5 [65]. We give Theorem 15.42. Let f ∈ L1 (a, x), a < x, a, x ∈ R, have an L∞ fractional derivative Dav f in [a, x], such that Dav−j f (a) = 0, for j = 1, . . . , [v] + 1; | is decreasing on [a, x]. Let l ≥ 2. If p, q > 1 : 1/p + 1/q = 1 and also |Dav f "l α i=1 i p > −1, then

x

l

1/q

x

|Daμi f (τ )|dτ ≤ C3 (x − a)(γp+lp+1)/p

|Dav f (t)|lq dt

a i=1

,

a

(15.97)

"l

i=1 αi and

where γ :=

C3 = C3 (v, μ, p) :=

p (γp +

1)1/p (γp

+ p + 1)

&l i=1

Γ(v − μi )

.

(15.98)

Proof. Transfer of Theorem 2.6 of [65]. We ﬁnish this subsection with Theorem 15.43. All are as in Theorem 15.42 with p = 1 and q = ∞. Then

x

l

a i=1

where γ :=

"l i=1

|Daμi f (τ )|dτ ≤ C4 (x − a)γ+l+1 Dav f l∞ ,

(15.99)

αi and

C4 = C4 (v, μ) :=

1 (γ + 1)(γ + l + 1)

&l

Proof. Transfer of Theorem 2.7 [65].

i=1

Γ(v − μi )

.

(15.100)

350

15. Fractional Multivariate Opial Inequalities

15.5.2 Riemann–Liouville Fractional Opial-Type Inequalities Involving Two Functions We present Theorem 15.44. Let α1 , α2 ∈ R+ , β > α1 , α2 , β − αi > (1/p), p > 1, i = 1, 2, and let f1 , f2 ∈ L1 (a, x), a, x ∈ R, a < x have, respectively, L∞ fractional derivatives Daβ f1 , Daβ f2 in [a, x], and let Daβ−k fi (a) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Consider also p(t) > 0 and q(t) ≥ 0, with all p(t), 1/p(t), q(t) ∈ L∞ (a, x). Let λβ > 0 and λα1 , λα2 ≥ 0, such that λβ < p. Set

s

Pi (s) :=

(s − t)p(β−αi −1)/(p−1) (p(t + a))−1/(p−1) dt, i = 1, 2; 0 ≤ s ≤ x − a,

0

A(s) :=

(15.101)

q(s + a) (P1 (s))

λα1 ((p−1)/p)

(Γ(β − α1 ))

(P2 (s)) λα1

λα2 ((p−1)/p)

(Γ(β − α2 ))

(p(s + a))

λα2

−λβ /p

,

(15.102)

(p−λβ )/p

x−a

A0 (x − a) :=

(A(s))

p/(p−λβ )

ds

,

(15.103)

0

and

% δ 1 :=

21−((λα1 +λβ )/p) , if λα1 + λβ ≤ p,

(15.104)

1, if λα1 + λβ ≥ p.

If λα2 = 0 we obtain that x

q(s) |Daα1 f1 (s)|λα1 |Daβ f1 (s)|λβ + | Daα1 f2 (s)|λα1 |Daβ f2 (s)|λβ ds ≤ a

+

x

A0 (x − a)

λα2 =0

λβ λ α1 + λ β

λβ /p δ1

,((λα1 +λβ )/p) p(s) |Daβ f1 (s)|p + |Daβ f2 (s)|p ds .

(15.105)

a

Proof. Similar proof to those of Theorems 15.22 and 15.34. Here we transfer Theorem 4 of [48] to an arbitrary anchor point a ∈ R. In fact for

15.5 Main Results

351

a = 0 inequality (15.105) is identical to inequality (8) of Theorem 4 of [48]. We apply it here for the translates f1a := f1 (· + a), f2a := f2 (· + a), p(· + a), q(· + a) and the fractional derivatives Dβ f1a , Dβ f2a , Dαi f1a , Dαi f2a , i = 1, 2, all over [0, x − a]. We use Lemma 15.9, the equivalent deﬁnitions 15.12, 15.13, and (15.13). We also use (15.9) through (15.11). We get the result by Proposition 15.4 applied at the end. We continue with Theorem 15.45. All here are as in Theorem 15.44. Denote λ /λ 2 α2 β − 1, if λα2 ≥ λβ , δ 3 := 1, if λα2 ≤ λβ .

(15.106)

If λα1 = 0, then x

q(s) |Daα2 f2 (s)|λα2 |Daβ f1 (s)|λβ + | Daα2 f1 (s)|λα2 |Daβ f2 (s)|λβ ds ≤ a

A0 (x − a)

λβ /p λβ λ /p δ3 β λα2 + λβ λα1 =0 (λα2 +λβ )/p x

p(s) |Daβ f1 (s)|p + |Daβ f2 (s)|p ds . 2(p−λβ )/p

(15.107)

a

Proof. Transfer of Theorem 5 of [48] to a ∈ R. The proof is similar to that of Theorem 15.44. The complete case λα1 , λα2 = 0 follows. Theorem 15.46. All here are as in Theorem 15.44. Denote ((λ +λ )/λ ) 2 α1 α2 β − 1, if λα1 + λα2 ≥ λβ , γ˜1 := 1, if λα1 + λα2 ≤ λβ ,

(15.108)

and γ˜2 :=

1, if λα1 + λα2 + λβ ≥ p, 21−((λα1 +λα2 +λβ )/p) , if λα1 + λα2 + λβ ≤ p.

(15.109)

352

15. Fractional Multivariate Opial Inequalities

Then, it holds x

+ q(s) |Daα1 f1 (s)|λα1 |Daα2 f2 (s)|λα2 |Daβ f1 (s)|λβ

a

, +| Daα2 f1 (s)|λα2 |Daα1 f2 (s)|λα1 |Daβ f2 (s)|λβ ds ≤ A0 (x − a)

λβ /p λβ · [λλαβ1 /p γ˜2 + 2(p−λβ )/p (γ˜1 λα2 )λβ /p ]· (λα1 + λα2 )(λα1 + λα2 + λβ ) + x ,(λα1 +λα2 +λβ )/p p(s) |Daβ f1 (s)|p + |Daβ f2 (s)|p ds . (15.110) a

Proof. Similar transfer to a ∈ R of Theorem 6 of [48]. We proceed with a special important case. Theorem 15.47. Let β > α1 + 1, α1 ∈ R+ and let f1 , f2 ∈ L1 (a, x), a, x ∈ R, a < x have, respectively, L∞ fractional derivatives Daβ f1 , Daβ f2 in [0, x], and let Daβ−k fi (a) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Consider also p(t) > 0 and q(t) ≥ 0, with p(t), 1/p(t) , q(t) ∈ L∞ (a, x). Let λα ≥ 0, 0 < λα+1 < 1, and p > 1. Denote λ /λ 2 α α+1 − 1, if λα ≥ λα+1 , (15.111) θ3 := 1, if λα ≤ λα+1 , x (1−λα+1 ) λα+1 θ3 λα+1 (1/(1−λα+1 )) L(x − a) := 2 (q(s)) ds , λα + λα+1 a (15.112) and x P1 (x − a) := (x − s)(β−α1 −1)p/(p−1) (p(s))−1/(p−1) ds, (15.113) a

T (x − a) := L(x − a)

P1 (x − a)(p−1/p) Γ(β − α1 )

(λα +λα+1 ) ,

(15.114)

and w1 := 2((p−1)/p)(λα +λα+1 ) ,

(15.115)

15.5 Main Results

353

with Φ(x − a) := T (x − a)w1 .

(15.116)

Then x , + q(s) |Daα1 f1 (s)|λα |Daα1 +1 f2 (s)|λα+1 +|Daα1 f2 (s)|λα |Daα1 +1 f1 (s)|λα+1 ds a

+ ≤ Φ(x−a)

x

,((λα +λα+1 )/p) p(s) |Daβ f1 (s)|p + |Daβ f2 (s)|p ds . (15.117)

a

Proof. Similar transfer to a ∈ R of Theorem 8 of [48]. We give Theorem 15.48. All here are as in Theorem 15.44. Consider the special case of λα2 = λα1 + λβ . Denote T˜(x − a) := A0 (x − a)

Then

x

λβ λα1 + λβ

λβ /p

2(p−2λα1 −3λβ )/p .

(15.118)

+ q(s) |Daα1 f1 (s)|λα1 |Daα2 f2 (s)|λα1 +λβ |Daβ f1 (s)|λβ

a

, +|Daα2 f1 (s)|λα1 +λβ |Daα1 f2 (s)|λα1 |Daβ f2 (s)|λβ ds ≤ ˜ T (x − a)

x

2(λα1 +λβ )/p p(s) |Daβ f1 (s)|p + |Daβ f2 (s)|p ds .

(15.119)

a

Proof. Transfer of Theorem 9 of [48].

Next follow special cases of the last theorems. Corollary 15.49. (to Theorem 15.44) Set λα2 = 0, p(t) = q(t) = 1. Then x+ , |Daα1 f1 (s)|λα1 |Daβ f1 (s)|λβ + |Daα1 f2 (s)|λα1 |Daβ f2 (s)|λβ ds ≤ a

+ C1 (x − a) a

x

β ,(λα1 +λβ )/p |Da f1 (s)|p + |Daβ f2 (s)|p ds ,

(15.120)

354

15. Fractional Multivariate Opial Inequalities

where C1 (x − a) := % δ 1 := We ﬁnd that A0 (x − a)

A0 (x − a)

λα2 =0

λβ λα1 + λβ

λβ /p δ1 ,

21−((λα1 +λβ )/p) , if λα1 + λβ ≤ p, 1, if λα1 + λβ ≥ p.

=

λα2 =0

(15.121)

(15.122)

(p − 1)((λα1 p−λα1 )/p) × (Γ(β − α1 ))λα1 (βp − α1 p − 1)((λα1 p−λα1 )/p)

(p − λβ )((p−λβ )/p) × ((p−λ )/p) β (λα1 βp − λα1 α1 p − λα1 + p − λβ ) (x − a)((λα1 βp−λα1 α1 p−λα1 +p−λβ )/p) .

(15.123)

Proof. Transfer of Corollary 10 of [48]. We continue with Corollary 15.50. (To Theorem 15.44: set λα2 = 0, p(t) = q(t) = 1, λα1 = λβ = 1, p = 2.) In detail; let α1 ∈ R+ , β > α1 , β − α1 > (1/2), and let f1 , f2 ∈ L1 (a, x), a, x ∈ R, a < x have, respectively, L∞ fractional derivatives Daβ f1 , Daβ f2 in [a, x], and let Daβ−k fi (a) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Then

x

+

, |Daα1 f1 (s)| |Daβ f1 (s)| + |Daα1 f2 (s)| |Daβ f2 (s)| ds ≤

a

(x − a)(β−α1 ) √ √ 2Γ(β − α1 ) β − α1 2β − 2α1 − 1

x+

, (Daβ f1 (s))2 + (Daβ f2 (s))2 ds .

a

(15.124) Proof. Transfer of Corollary 11 of [48]. We continue with

15.5 Main Results

355

Corollary 15.51. (To Theorem 15.45; λα1 = 0, p(t) = q(t) = 1.) It holds x+ , |Daα2 f2 (s)|λα2 |Daβ f1 (s)|λβ + |Daα2 f1 (s)|λα2 |Daβ f2 (s)|λβ ds ≤ a

+ C2 (x − a)

x

β ,(λβ +λα2 )/p |Da f1 (s)|p + |Daβ f2 (s)|p ds .

(15.125)

a

Here C2 (x − a) :=

A0 (x − a)

δ 3 := We ﬁnd that A0 (x − a)

λα1 =0

(p−λβ /p)

λα1 =0

2

λβ λα2 + λβ

λβ /p

λ /p

δ3 β , (15.126)

2λα2 /λβ − 1, if λα2 ≥ λβ , 1, if λα2 ≤ λβ .

=

(15.127)

(p − 1)(λα2 p−λα2 )/p × (Γ(β − α2 ))λα2 (βp − α2 p − 1)(λα2 p−λα2 )/p)

(p − λβ )((p−λβ )/p) × (λα2 βp − λα2 α2 p − λα2 + p − λβ )((p−λβ )/p) (x − a)((λα2 βp−λα2 α2 p−λα2 +p−λβ )/p) .

(15.128)

Proof. Transfer of Corollary 12 of [48]. We give Corollary 15.52. (To Theorem 15.45; λα1 = 0, p(t) = q(t) = 1, λα2 = λβ = 1, p = 2.) In detail; let α2 ∈ R+ , β > α2 , β − α2 > (1/2), and let f1 , f2 ∈ L1 (a, x), a, x ∈ R, a < x have, respectively, L∞ fractional derivatives Daβ f1 , Daβ f2 in [a, x], and let Daβ−k fi (a) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Then

x

+

, |Daα2 f2 (s)| |Daβ f1 (s)| + |Daα2 f1 (s)| |Daβ f2 (s)| ds ≤

a

C2∗ (x

− a) a

x

β

2 β 2 (Da f1 (s)) + (Da f2 (s)) ds ,

(15.129)

356

15. Fractional Multivariate Opial Inequalities

where C2∗ (x − a) := √

(x − a)(β−α2 ) . √ √ 2 Γ(β − α2 ) β − α2 2β − 2α2 − 1

(15.130)

Proof. Transfer of Corollary 13 of [48]. We continue with Corollary 15.53. (To Theorem 15.46; λα1 = λα2 = λβ = 1, p = 3, p(t) = q(t) = 1.) It holds

x+

, |Daα1 f1 (s)| |Daα2 f2 (s)| |Daβ f1 (s)| + | Daα2 f1 (s)| |Daα1 f2 (s)| |Daβ f2 (s)| ds ≤

a

√ 1 3 A0 (x − a) 2+ √ 3 6 Here,

x

β 3 β 3 |Da f1 (s)| + |Da f2 (s)| ds .

(15.131)

a

A0 (x − a) = 4(x − a)(2β−α1 −α2 ) ×

1 . Γ(β − α1 ) Γ(β − α2 )[3(3β − 3α1 − 1)(3β − 3α2 − 1)(2β − α1 − α2 )]2/3 (15.132) Proof. Transfer of Corollary 14 of [48]. We give Corollary 15.54. (To Theorem 15.47; here λα = 1, λα+1 = 1/2, p = 3/2, p(t) = q(t) = 1.) In detail: let β > α1 + 1, α1 ∈ R+ , and let f1 , f2 ∈ L1 (a, x), a, x ∈ R, a < x have, respectively, L∞ fractional derivatives Daβ f1 , Daβ f2 in [a, x], and let Daβ−k fi (a) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Then

x

. . + , |Daα1 f1 (s)| |Daα1 +1 f2 (s)| + |Daα1 f2 (s)| |Daα1 +1 f1 (s)| ds ≤

a

Φ∗ (x − a)

)

x

* |Daβ f1 (s)|3/2 + |Daβ f2 (s)|3/2 ds ,

(15.133)

a

where Φ∗ (x − a) :=

2(x − a)(3β−3α1 −1)/2 √ . (Γ(β − α1 ))3/2 3β − 3α1 − 2

(15.134)

15.5 Main Results

357

Proof. Transfer of Corollary 15 of [48]. We continue Corollary 15.55. (To Theorem 15.48; here p = 2(λα1 + λβ ) > 1, p(t) = q(t) = 1.) It holds x+ |Daα1 f1 (s)|λα1 |Daα2 f2 (s)|λα1 +λβ |Daβ f1 (s)|λβ a

, +|Daα2 f1 (s)|λα1 +λβ |Daα1 f2 (s)|λα1 |Daβ f2 (s)|λβ ds ≤ x ˜(x − a) T˜ |Daβ f1 (s)|2(λα1 +λβ ) + |Daβ f2 (s)|2(λα1 +λβ ) ds . (15.135) a

Here,

˜(x − a) := A˜0 (x − a) T˜

(λβ /2(λα1 +λβ )) λβ , 2(λα1 + λβ )

(15.136)

and A˜0 (x − a) := σσ ∗ (x − a)θ ,

(15.137)

where

1 (Γ(β − α1 ))λα1 (Γ(β − α2 ))λα1 +λβ (2λ2α +2λα1 λβ −λα1 )/(2λα1 +2λβ ) 1 2λα1 + 2λβ − 1 × 2λα1 β + 2λβ β − 2λα1 α1 − 2λβ α1 − 1 ((2λα1 +2λβ −1)/2) 2λα1 + 2λβ − 1 × , (15.138) 2λα1 β + 2λβ β − 2λα1 α2 − 2λβ α2 − 1 σ :=

∗

σ :=

2λα1 + λβ S

(2λα1 +λβ )/2(λα1 +λβ ) ,

where S := 4λ2α1 β + 6λα1 λβ β − 2λ2α1 α1 − 2λα1 λβ α1 −2λ2α1 α2 − 4λα1 λβ α2 + 2λβ 2 β − 2λβ 2 α2 and

θ :=

S 2λα1 + 2λβ

.

(15.139)

358

15. Fractional Multivariate Opial Inequalities

Proof. Transfer of Corollary 16 of [48]. We give the following interesting special case. Corollary 15.56. (To Theorem 15.48; here p = 4, λα1 = λβ = 1, p(t) = q(t) = 1.) It holds

x

+

, |Daα1 f1 (s)| (Daα2 f2 (s))2 |Daβ f1 (s)| + (Daα2 f1 (s))2 |Daα1 f2 (s)| |Daβ f2 (s)| ds ≤

a

T ∗ (x − a)

x

β Da f1 (s))4 + (Daβ f2 (s)4 ds .

(15.140)

a

Here, T ∗ (x − a) := and

A∗0 (x − a) √ , 2

(15.141)

1σ 1∗ (x − a)θ , A∗ (x − a) := σ ˜

(15.142)

where σ 1 :=

1 Γ(β − α1 )(Γ(β − α2 ))2 σ 1∗ :=

and

3 4β − 4α1 − 1

3/4

3 12β − 4α1 − 8α2

3 4β − 4α2 − 1

3/2 , (15.143)

3/4 ,

1 θ := 3β − α1 − 2α2 .

(15.144)

(15.145)

Proof. Transfer of Corollary 17 of [48]. We continue with related results regarding the L∞ -norm · ∞ . Theorem 15.57. Let α1 , α2 ∈ R+ , β > α1 , α2 , and let f1 , f2 ∈ L1 (a, x), a, x ∈ R, a < x have, respectively, L∞ fractional derivatives Daβ f1 , Daβ f2 in [a, x], and let Daβ−k fi (a) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Consider also p(s) ≥ 0, p(s) ∈ L∞ (a, x). Let λα1 , λα2 , λβ ≥ 0. Set ρ(x − a) :=

15.5 Main Results

359

p(s)∞ (x − a)(βλα1 −α1 λα1 +βλα2 −α2 λα2 +1) . (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα2 [βλα1 − α1 λα1 + βλα2 − α2 λα2 + 1] (15.146) Then

x

a

+ p(s) |Daα1 f1 (s)|λα1 |Daα2 f2 (s)|λα2 |Daβ f1,(s)|λβ

+| Daα2 f1 (s)|λα2 |Daα1 f2 (s)|λα1 |Daβ f2 (s)|λβ ds ≤ ρ(x − a) + β 2(λ +λ ) 2λ Da f1 ∞ α1 β + Daβ f1 ∞ α2 2 , 2λ 2(λ +λ ) + Daβ f2 ∞ α2 + Daβ f2 ∞ α1 β .

(15.147)

Proof. Transfer of Theorem 18 of [48]. It is similar to the proof of Theorem 15.24, and so on. We give special cases of the last theorem. Theorem 15.58. (All are as in Theorem 15.57; λα2 = 0.) It holds x , + p(s) |Daα1 f1 (s)|λα1 |Daβ f1 (s)|λβ + |Daα1 f2 (s)|λα1 |Daβ f2 (s)|λβ ds ≤ a

, + p(s)∞ (x − a)(βλα1 −α1 λα1 +1) λα1 +λβ λα1 +λβ β β . f + D f · D ∞ ∞ 1 2 a a (Γ(β − α1 + 1))λα1 [βλα1 − α1 λα1 + 1] (15.148) Proof. Transfer of Theorem 19 of [48]. It is similar to the proof of Theorem 15.57. We continue with Theorem 15.59. (All are as in Theorem 15.57; λα2 = λα1 + λβ .) It holds x + p(s) |Daα1 f1 (s)|λα1 |Daα2 f2 (s)|λα1 +λβ |Daβ f1 (s)|λβ a

, +| Daα2 f1 (s)|λα1 +λβ |Daα1 f2 (s)|λα1 |Daβ f2 (s)|λβ ds ≤ ·

p(s)∞ (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα1 +λβ (x − a)(2βλα1 −α1 λα1 +βλβ −α2 λα1 −α2 λβ +1)

(2βλα1 − α1 λα1 + βλβ − α2 λα1 − α2 λβ + 1) + , 2(λ +λ ) 2(λ +λ ) · Daβ f1 ∞ α1 β + Daβ f2 ∞ α1 β .

(15.149)

360

15. Fractional Multivariate Opial Inequalities

Proof. Transfer of Theorem 20 of [48].

We give Theorem 15.60. (All are as in Theorem 15.57; λβ = 0, λα1 = λα2 .) It holds x , + p(s) |Daα1 f1 (s)|λα1 |Daα2 f2 (s)|λα1 + |Daα2 f1 (s)|λα1 |Daα1 f2 (s)|λα1 ds ≤ a

, + 2λ 2λ ρ∗ (x − a) Daβ f1 ∞ α1 + Daβ f2 ∞ α1 ,

(15.150)

where ρ∗ (x−a) =

p(s)∞ (x − a)(2βλα1 −α1 λα1 −α2 λα1 +1) . λα 1 (Γ(β − α1 + 1) Γ(β − α2 + 1)) (2βλα1 − α1 λα1 − α2 λα1 + 1) (15.151)

Proof. Transfer of Theorem 21 of [48].

We give Theorem 15.61. (All are as in Theorem 15.57; λα1 = 0, λα2 = λβ .) It holds x , + p(s) |Daα2 f2 (s)|λα2 |Daβ f1 (s)|λα2 + |Daα2 f1 (s)|λα2 |Daβ f2 (s)|λα2 ds ≤ a

(x − a)(βλα2 −α2 λα2 +1) p(s)∞ (βλα2 − α2 λα2 + 1)(Γ(β − α2 + 1))λα2

+ , 2λ 2λ · Daβ f1 ∞ α2 + Daβ f2 ∞ α2 . (15.152)

Proof. Transfer of Theorem 22 of [48].

We continue with Corollary 15.62. (To Theorem 15.60; all are as in Theorem 15.57; λβ = 0, λα1 = λα2 , α2 = α1 + 1.) It holds x , + p(s) |Daα1 f1 (s)|λα1 |Daα1 +1 f2 (s)|λα1 +|Daα1 +1 f1 (s)|λα1 |Daα1 f2 (s)|λα1 ds ≤ a

(2βλα1

(x − a)(2βλα1 −2α1 λα1 −λα1 +1) p(s)∞ − 2α1 λα1 − λα1 + 1)(β − α1 )λα1 (Γ(β − α1 ))2λα1 + , 2λ 2λ · Daβ f1 ∞ α1 + Daβ f2 ∞ α1 . (15.153)

15.5 Main Results

361

Proof. Transfer of Corollary 23 of [48]. We give Corollary 15.63. (to Corollary 15.62) In detail: let α1 ∈ R+ , β > α1 +1, and let f1 , f2 ∈ L1 (a, x), a, x ∈ R, a < x have, respectively, L∞ fractional derivatives Daβ f1 , Daβ f2 in [a, x], and let Daβ−k fi (a) = 0, for k = 1, . . . , [β] + 1; i = 1, 2. Then

x

+ , |Daα1 f1 (s)| |Daα1 +1 f2 (s)| + |Daα1 +1 f1 (s)| |Daα1 f2 (s)| ds ≤

a

(x − a)2(β−α1 ) 2(β − α1 )2 (Γ(β − α1 ))2

+ , · Daβ f1 2∞ + Daβ f2 2∞ .

(15.154)

Proof. Transfer of Corollary 24 of [48]. We ﬁnally give Proposition 15.64. Inequality (15.154) is sharp; in fact it is attained when f1 = f2 , by f1 (s) = (s − a)β , a ≤ s ≤ x, β > α1 + 1, α1 ≥ 0. Proof. Clearly (15.154), when f1 = f2 , collapses to x (x − a)2(β−α1 ) |Daα1 f1 (s)| |Daα1 +1 f1 (s)|ds ≤ · Daβ f1 2∞ ; 2 2(Γ(β − α + 1)) 1 a (15.155) see Theorem 15.27 and Proposition 15.28. Next we apply the above results on the spherical shell A. We make Assumption 15.65. Let α1 , α2 ∈ R+ , β > α1 , α2 , β − αi > (1/p), p > 1, i = 1, 2, and let f1 , f2 ∈ L1 (A) with β β ∂R f (x) ∂R f (x) 1 1 1 2 , ∈ L∞ (A), x ∈ A; ∂rβ ∂rβ

362

15. Fractional Multivariate Opial Inequalities

A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2, 0 < R1 < R2 . β Further assume that each DR f (rw) ∈ R for almost all r ∈ [R1 , R2 ], for 1 i β N −1 , and for these |DR f (rw)| ≤ Mi for some Mi > 0; i = 1, 2. each w ∈ S 1 i For each w ∈ S N −1 − (K(f1 ) ∪ K(f2 )), we assume that fi (· w) has an L∞ β fractional derivative DR f (· w) in [R1 , R2 ], and that 1 i β−k DR fi (R1 w) = 0, k = 1, . . . , [β] + 1; 1

i = 1, 2. Let λβ > 0 and λα1 , λα2 ≥ 0, such that λβ < p. If α1 = 0 we set λα1 = 1, and if α2 = 0 we set λα2 = 1. We need Notation 15.66. (on Assumption 15.65) Set

s

Pi (s) :=

(s − r)p(β−αi −1)/p−1 (r + R1 )1−N /p−1 dr, i = 1, 2; 0 ≤ s ≤ R2 − R1 ,

0

(15.156)

A(s) :=

(s + R1 )(N −1)(1−(λβ /p)) (P1 (s)) (Γ(β − α1 ))

λα1

λα1 ((p−1)/p)

(Γ(β − α2 ))

(P2 (s))

λα2 ((p−1)/p)

λα2

, (15.157)

and A0 (R2 − R1 ) :=

(p−λβ )/p

R2 −R1

(A(s))

p/(p−λβ )

ds

.

(15.158)

0

We present Theorem 15.67. (All are as in Assumption 15.65 and Notation 15.66.) Here λα1 > 0, λα2 = 0, and p = λα1 + λβ > 1. Then +⏐ α1 ⏐ ⏐ β ⏐ ⏐ ⏐ β ⏐ , ⏐ α1 ⏐ ∂R1 f1 (x) ⏐λα1 ⏐ ∂R1 f1 (x) ⏐λβ ⏐ ∂R1 f2 (x) ⏐λα1 ⏐ ∂R1 f2 (x) ⏐λβ + ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ dx ≤ ⏐ ∂rα1 ∂rβ ∂rα1 ∂rβ A

(λβ /(λα1 +λβ )) λβ λα1 + λβ λα2 =0 ( ' +⏐ ∂ β f1 (x) ⏐p ⏐ ∂ β f2 (x) ⏐p , ⏐ R1 ⏐ ⏐ ⏐ R1 ⏐ +⏐ ⏐ dx . ⏐ ∂rβ ∂rβ A A0 (R2 − R1 )

Proof. By Theorem 15.20 for α1 > 0 we get that α1 ∂R f (x) 1 i ∈ L∞ (A), i = 1, 2. ∂rα1

(15.159)

15.5 Main Results

363

In general here the integrands of both integrals of (15.159) are in L1 (A). Thus, by Proposition 15.3 we have R2 β α1 [ |DR f (rw)|λα1 |DR f (rw)|λβ + I1 := L.H.S(15.159) = 1 1 1 1 S N −1

R1

β α1 |DR f (rw)|λα1 |DR f (rw)|λβ ]rN −1 dr dw = 1 2 1 2

(S N −1 −(K(f

1 )∪K(f2 )))

α1 |DR f (rw)|λα1 1 2

R2

R1

β α1 [ |DR f (rw)|λα1 |DR f (rw)|λβ + 1 1 1 1

β |DR f (rw)|λβ ]rN −1 dr 1 2

dw.

(15.160)

Similarly we have

= S N −1

+⏐ β ⏐ ⏐ , ⏐ β ⏐ ∂R1 f1 (x) ⏐p ⏐ ∂R1 f2 (x) ⏐p I2 := + ⏐ ⏐ dx ⏐ ⏐ ∂rβ ∂rβ A R2 β β p p N −1 [ |DR f (rw)| + |D f (rw)| ]r dr dw = 1 2 R 1 1 R1

(S N −1 −(K(f1 )∪K(f2 )))

R2

R1

β β p p N −1 [ |DR f (rw)| + |D f (rw)| ]r dr dw. 1 2 R 1 1

(15.161) Notice here λS N −1 (K(f1 )∪K(f2 )) = 0. Here for every w ∈ S N −1 −(K(f1 )∪ K(f2 )) and for p(r) = q(r) = rN −1 , r ∈ [R1 , R2 ], N ≥ 2 we apply Theorem 15.44. We obtain

R2 R1

β β α1 α1 [ |DR f1 (rw)|λα1 |DR f1 (rw)|λβ +|DR f2 (rw)|λα1 |DR f2 (rw)|λβ ]rN −1 dr ≤ 1 1 1 1

A0 (R2 − R1 )

R2

R1

λα2 =0

(λβ /λα1 +λβ ) λβ × λα1 + λβ

β β [ |DR f (rw)|p + |DR f (rw)|p ]rN −1 1 1 1 2

dr.

(15.162)

Integrating now (15.162) over S N −1 − (K(f1 ) ∪ K(f2 )) and taking into account (15.160) and (15.161) we derive (15.159). We continue with Theorem 15.68. (All are as in Assumption 15.65 and Notation 15.66.) Here λα1 = 0, λα2 > 0, and p = λβ + λα2 > 1. Denote λ /λ 2 α2 β − 1, if λα2 ≥ λβ , (15.163) δ 3 := 1, if λα2 ≤ λβ .

364

15. Fractional Multivariate Opial Inequalities

Then +⏐ α2 ⏐ ⏐ β ⏐ ⏐ ⏐ β ⏐ , ⏐ α2 ⏐ ∂R1 f2 (x) ⏐λα2 ⏐ ∂R1 f1 (x) ⏐λβ ⏐ ∂R1 f1 (x) ⏐λα2 ⏐ ∂R1 f2 (x) ⏐λβ + ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ dx ≤ ⏐ ⏐ ∂rα2 ∂rβ ∂rα2 ∂rβ A λβ /p λβ λ /p A0 (R2 − R1 ) δ3 β 2λα2 /p p λα1 =0 +⏐ ∂ β f1 (x) ⏐p ⏐ ∂ β f2 (x) ⏐p , ⏐ ⏐ ⏐ R1 ⏐ R1 (15.164) ⏐ +⏐ ⏐ dx . ⏐ ∂rβ ∂rβ A Proof. Based on Theorem 15.45 and similar to the proof of Theorem 15.67. The complete case λα1 , λα2 > 0 follows. Theorem 15.69. (All are as in Assumption 15.65 and Notation 15.66.) Here λα1 , λα2 > 0, p = λα1 + λα2 + λβ > 1. Denote ((λ +λ )/λ ) 2 α1 α2 β − 1, if λα1 + λα2 ≥ λβ , (15.165) γ˜1 := 1, if λα1 + λα2 ≤ λβ . Then

+⏐ α1 ⏐ ⏐ α2 ⏐ ⏐ β ⏐ ⏐ ∂R1 f1 (x) ⏐λα1 ⏐ ∂R1 f2 (x) ⏐λα2 ⏐ ∂R1 f1 (x) ⏐λβ ⏐ ⏐ ⏐ ⏐ ⏐ + ⏐ α α β ∂r 1 ∂r 2 ∂r A ⏐ ∂ α2 f (x) ⏐λα2 ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ R1 1 ⏐ ⏐ R1 2 ⏐ ⏐ R1 2 ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ dx ≤ ∂rα2 ∂rα1 ∂rβ (λβ /p) λβ A0 (R2 − R1 ) [λα1 (λβ /p) + 2(p−λβ )/p (γ˜1 λα2 )(λβ /p) ]· (λα1 + λα2 )p +⏐ ∂ β f1 (x) ⏐p ⏐ ∂ β f2 (x) ⏐p , ⏐ ⏐ ⏐ R1 ⏐ R1 (15.166) ⏐ +⏐ ⏐ dx . ⏐ ∂rβ ∂rβ A

Proof. Based on Theorem 15.46 and similar to the proof of Theorem 15.67. A special important case is next. Theorem 15.70. (All are as in Assumption 15.65.) Here α2 = α1 + 1, λα := λα1 ≥ 0, λα+1 := λα2 ∈ (0, 1), and p = λα + λα+1 > 1. Denote (λ /λ ) 2 α α+1 − 1, if λα ≥ λα+1 , θ3 := (15.167) 1, if λα ≤ λα+1 , L(R2 − R1 ) :=

15.5 Main Results

365

) *(1−λα+1 ) (1 − λα+1 ) N −λα+1 /1−λα+1 R2 − R1 N −λα+1 /1−λα+1 2 (N − λα+1 )

θ3 λα+1 p

λα+1 ,

(15.168)

and P1 (R2 − R1 ) :=

R2

(R2 − r)(β−α1 −1)p/(p−1) r(1−N )/(p−1) dr,

(15.169)

R1

Φ(R2 − R1 ) := L(R2 − R1 )

(P1 (R2 − R1 ))(p−1) (Γ(β − α1 ))p

2(p−1) .

(15.170)

Then +⏐ α1 ⏐ ⏐ α1 +1 ⏐ ⏐ ⏐ α1 +1 ⏐ ⏐ α1 , ⏐ ∂R1 f1 (x) ⏐λα ⏐ ∂R1 f2 (x) ⏐λα+1 ⏐ ∂R1 f2 (x) ⏐λα ⏐ ∂R1 f1 (x) ⏐λα+1 dx + ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ α α +1 α α +1 1 1 1 1 ∂r ∂r ∂r ∂r A

+⏐ ∂ β f1 (x) ⏐p ⏐ ∂ β f2 (x) ⏐p , ⏐ ⏐ ⏐ R1 ⏐ R1 ⏐ +⏐ ⏐ dx . ⏐ ∂rβ ∂rβ A

≤ Φ(R2 − R1 )

(15.171)

Proof. Based on Theorem 15.47 and similar to the proof of Theorem 15.67. We also give Theorem 15.71. (All are as in Assumption 15.65 and Notation 15.66.) Here λα2 = λα1 + λβ and p = 2(λα1 + λβ ) > 1. Denote T˜(R2 − R1 ) := A0 (R2 − R1 )

λβ λα1 + λβ

λβ /p

2−λβ /p .

(15.172)

Then +⏐ α1 ⏐ ⏐ α2 ⏐ ⏐ β ⏐ ⏐ ∂R1 f1 (x) ⏐λα1 ⏐ ∂R1 f2 (x) ⏐λα1 +λβ ⏐ ∂R1 f1 (x) ⏐λβ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ + ∂rα1 ∂rα2 ∂rβ A ⏐ ∂ α2 f (x) ⏐λα1 +λβ ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ ⏐ R1 2 ⏐ ⏐ R1 2 ⏐ ⏐ R1 1 ⏐ ⏐ ⏐ ⏐ ⏐ dx ≤ ⏐ α α β 2 1 ∂r ∂r ∂r +⏐ ∂ β f1 (x) ⏐p ⏐ ∂ β f2 (x) ⏐p , ⏐ ⏐ ⏐ ⏐ R R T˜(R2 − R1 ) (15.173) ⏐ 1 β ⏐ + ⏐ 1 β ⏐ dx . ∂r ∂r A Proof. Based on Theorem 15.48.

366

15. Fractional Multivariate Opial Inequalities

We continue with related L∞ results on the shell A. We make Assumption 15.72. Let α1 , α2 ∈ R+ , β > α1 , α2 and let f1 , f2 ∈ L1 (A) with β β f (x) ∂R f (x) ∂R 1 1 1 2 , ∈ L∞ (A), x ∈ A; β ∂r ∂rβ A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2, 0 < R1 < R2 . β Further assume that each DR f (rw) ∈ R for almost all r ∈ [R1 , R2 ], for 1 i β N −1 , and for these |DR f (rw)| ≤ Mi for some Mi > 0; i = 1, 2. each w ∈ S 1 i N −1 − (K(f1 ) ∪ K(f2 )), we assume that fi (· w) has an L∞ For each w ∈ S β fractional derivative DR f (· w) in [R1 , R2 ], and that 1 i β−k DR fi (R1 w) = 0, k = 1, . . . , [β] + 1; 1

i = 1, 2. Let λα1 , λα2 , λβ ≥ 0. If α1 = 0 we set λα1 = 1, and if α2 = 0 we set λα2 = 1. We present Theorem 15.73. (All here are as in Assumption 15.72.) Set ρ(R2 − R1 ) = R2 N −1 (R2 − R1 )(βλα1 −α1 λα1 +βλα2 −α2 λα2 +1) . (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα2 (βλα1 − α1 λα1 + βλα2 − α2 λα2 + 1) (15.174) Then +⏐ α1 ⏐ ⏐ α2 ⏐ ⏐ β ⏐ ⏐ ∂R1 f1 (x) ⏐λα1 ⏐ ∂R1 f2 (x) ⏐λα2 ⏐ ∂R1 f1 (x) ⏐λβ ⏐ ⏐ ⏐ ⏐ ⏐ + ⏐ ∂rα1 ∂rα2 ∂rβ A ⏐ ∂ α2 f (x) ⏐λα2 ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ R1 1 ⏐ ⏐ R1 2 ⏐ ⏐ R1 2 ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ dx ≤ α α 2 1 ∂r ∂r ∂rβ ρ(R2 − R1 )[M1 2(λα1 +λβ ) + M1 2λα2 + M2 2λα2 + M2 2(λα1 +λβ ) ]

π N/2 . Γ(N/2) (15.175)

Proof. It is based on Theorem 15.57. By Theorem 15.20 for αj > 0 we get that α ∂Rj1 fi ∈ L∞ (A), i = 1, 2; j = 1, 2. ∂rαj In general here the integrand of the integral of (15.175) belongs to L1 (A). Thus by Proposition 15.3 we have

L.H.S(15.175) =

S N −1

R2 R1

α1 α2 β [ |DR f1 (rw)|λα1 |DR f2 (rw)|λα2 |DR f1 (rw)|λβ + 1 1 1

β α2 α1 λ α2 λα1 λβ N −1 |DR f (rw)| |D f (rw)| |D f (rw)| ]r dr dw 1 2 2 R R 1 1 1

(15.176)

15.5 Main Results

=

(S N −1 −(K(f1 )∪K(f2 )))

R2 R1

367

α1 α2 β [ |DR f1 (rw)|λα1 |DR f2 (rw)|λα2 |DR f1 (rw)|λβ + 1 1 1

β α2 α1 λ α2 λα1 λβ N −1 |DR f (rw)| |D f (rw)| |D f (rw)| ]r dr dw 1 2 2 R1 R1 1

(15.177)

(by (15.147) for p(r) = rN −1 , r ∈ [R1 , R2 ]) + ρ(R2 − R1 ) 2(λ 1 +λβ ) β DR ≤ f (· w)∞, α[R + 1 1 1 ,R2 ] 2 N −1 (S −(K(f1 )∪K(f2 ))) 2λα2 [R1 ,R2 ]

β DR f1 (· w)∞, 1

2λα2 [R1 ,R2 ]

β + DR f2 (· w)∞, 1

2(λα1 +λβ ) [R1 ,R2 ]

β + DR f2 (· w)∞, 1

, dw

(15.178)

≤

, ρ(R2 − R1 ) + 2(λα1 +λβ ) M1 + M1 2λα2 + M2 2λα2 + M2 2(λα1 +λβ ) . 2 (15.179) dw = (S N −1 −(K(f1 )∪K(f2 )))

, 2π N/2 ρ(R2 − R1 ) + 2(λα1 +λβ ) M1 , + M1 2λα2 + M2 2λα2 + M2 2(λα1 +λβ ) 2 Γ(N/2) (15.180) by 2π N/2 , (15.181) dw = dw = Γ(N/2) (S N −1 −(K(f1 )∪K(f2 ))) S N −1 because λS N −1 (K(f1 ) ∪ K(f2 )) = 0, thus proving inequality (15.175). We give special cases of the last theorem. Theorem 15.74. (All here are as in Assumption 15.72; here λα2 = 0.) It holds +⏐ α1 ⏐ ⏐ β ⏐ ⏐ ⏐ β ⏐ , ⏐ α1 ⏐ ∂R1 f1 (x) ⏐λα1 ⏐ ∂R1 f1 (x) ⏐λβ ⏐ ∂R1 f2 (x) ⏐λα1 ⏐ ∂R1 f2 (x) ⏐λβ + ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ dx ≤ ⏐ ∂rα1 ∂rβ ∂rα1 ∂rβ A R2 N −1 (R2 − R1 )(βλα1 −α1 λα1 +1) × (Γ(β − α1 + 1))λα1 [βλα1 − α1 λα1 + 1] , 2π N/2 + . M1 (λα1 +λβ ) + M2 (λα1 +λβ ) Γ(N/2)

(15.182)

Proof. Based on Theorem 15.58 and similar to the proof of Theorem 15.73.

368

15. Fractional Multivariate Opial Inequalities

We continue with Theorem 15.75. (All here are as in Assumption 15.72; here λα2 = λα1 + λβ .) It holds +⏐ α1 ⏐ ⏐ α2 ⏐ ⏐ β ⏐ ⏐ ∂R1 f1 (x) ⏐λα1 ⏐ ∂R1 f2 (x) ⏐λα1 +λβ ⏐ ∂R1 f1 (x) ⏐λβ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ + ∂rα1 ∂rα2 ∂rβ A ⏐ ∂ α2 f (x) ⏐λα1 +λβ ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ R1 1 ⏐ ⏐ R1 2 ⏐ ⏐ R1 2 ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ dx ≤ α α 2 1 ∂r ∂r ∂rβ R2 N −1 (Γ(β − α1 + 1)) (Γ(β − α2 + 1))λα1 +λβ λα1

(R2 − R1 )(2βλα1 −α1 λα1 +βλβ −α2 λα1 −α2 λβ +1) [2βλα1 − α1 λα1 + βλβ − α2 λα1 − α2 λβ + 1] , 2π N/2 + . × M1 2(λα1 +λβ ) + M2 2(λα1 +λβ ) Γ(N/2)

(15.183)

Proof. Based on Theorem 15.59 and similar to the proof of Theorem 15.73. We give Theorem 15.76. (All here are as in Assumption 15.72; here λβ = 0, λα1 = λα2 .) It holds +⏐ α1 ⏐ α2 ⏐ ⏐ α2 ⏐ ⏐ ⏐ α1 ⏐ , ⏐ ∂R1 f1 (x) ⏐λα1 ⏐ ∂R1 f2 (x) ⏐λα1 ⏐ ∂R1 f1 (x) ⏐λα1 ⏐ ∂R1 f2 (x) ⏐λα1 dx ≤ + ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ α α α α 1 2 2 1 ∂r ∂r ∂r ∂r A

R2 N −1 (R2 − R1 )(2βλα1 −α1 λα1 −α2 λα1 +1) × (Γ(β − α1 + 1) Γ(β − α2 + 1))λα1 (2βλα1 − α1 λα1 − α2 λα1 + 1) , 2π N/2 + . (15.184) M1 2λα1 + M2 2λα1 Γ(N/2) Proof. Based on Theorem 15.60 and similar to the proof of Theorem 15.73. We give Theorem 15.77. (All here are as in Assumption 15.72; here λα1 = 0, λα2 = λβ .) It holds +⏐ α2 ⏐ α2 ⏐ ⏐ β ⏐ ⏐ ⏐ β ⏐ , ⏐ ∂R1 f2 (x) ⏐λα2 ⏐ ∂R1 f1 (x) ⏐λα2 ⏐ ∂R1 f1 (x) ⏐λα2 ⏐ ∂R1 f2 (x) ⏐λα2 dx ≤ + ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ α β α β 2 2 ∂r ∂r ∂r ∂r A

15.5 Main Results

369

2π N/2 R2 N −1 (R2 − R1 )(βλα2 −α2 λα2 +1) 2λα2 2λα2 . + M M 1 2 Γ(N/2) (βλα2 − α2 λα2 + 1)(Γ(β − α2 + 1))λα2 (15.185) Proof. Based on Theorem 15.61, and so on. We ﬁnish this section with Corollary 15.78. (To Theorem 15.76. All here are as in Assumption 15.72, λβ = 0, λα1 = λα2 , α2 = α1 + 1.) It holds +⏐ ∂ α1 f (x) ⏐ ⏐ α1 +1 ⏐ ⏐ ⏐ α1 ⏐ ⏐ α1 +1 , ⏐ R1 1 ⏐ λα1 ⏐ ∂R1 f2 (x) ⏐ λα1 ⏐ ∂R1 f1 (x) ⏐ λα1 ⏐ ∂R1 f2 (x) ⏐ λα1 + dx ≤ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ α α +1 α +1 α 1 1 1 1 ∂r ∂r ∂r ∂r A

(2βλα1

R2 N −1 (R2 − R1 )(2βλα1 −2α1 λα1 −λα1 +1) × − 2α1 λα1 − λα1 + 1)(β − α1 )λα1 (Γ(β − α1 ))2λα1 +

M1 2λα1 + M2 2λα1

, 2π N/2 . Γ(N/2)

(15.186)

Proof. Based on Corollary 15.62, and so on.

15.5.3 Riemann–Liouville Fractional Opial-Type Inequalities Involving Several Functions We present the following theorem. Theorem 15.79. Let α1 , α2 ∈ R+ , β > α1 , α2 , β − αi > (1/p), p > 1, i = 1, 2, and let fj ∈ L1 (a, x), j = 1, . . . , M ∈ N, a, x ∈ R, a < x, have, respectively, L∞ fractional derivatives Daβ fj in [a, x], and let Daβ−k fj (a) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M. Consider also p(t) > 0 and q(t) ≥ 0, with all p(t), 1/p(t), q(t) ∈ L∞ (a, x). Let λβ > 0 and λα1 , λα2 ≥ 0, such that λβ < p. Set s −1/(p−1) Pi (s) := (s−t)p(β−αi −1)/(p−1) (p(t + a)) dt, i = 1, 2; 0 ≤ s ≤ x−a, 0

(15.187)

370

15. Fractional Multivariate Opial Inequalities

A(s) :=

q(s + a) (P1 (s))

λα1 ((p−1)/p)

(Γ(β − α1 ))

λα1

λα2 ((p−1)/p)

(Γ(β − α2 ))

(p(s + a))

(A(s))

p/(p−λβ )

−λβ /p

,

λα2

(15.188)

(p−λβ )/p

x−a

A0 (x − a) :=

(P2 (s))

ds

,

(15.189)

0

%

and δ ∗1

M 1−((λα1 +λβ )/p) , if λα1 + λβ ≤ p, 2(λα1 +λβ /p)−1 , if λα1 + λβ ≥ p.

:=

(15.190)

Call ϕ1 (x − a) :=

A0 (x − a)

λα2 =0

λβ λα1 + λβ

λβ /p .

(15.191)

If λα2 = 0, we obtain that ⎞ ⎛ x M q(s) ⎝ |Daα1 fj (s)|λα1 |Daβ fj (s)|λβ ⎠ ds ≤ a

j=1

⎡ δ ∗1 ϕ1 (x − a) ⎣

⎛ x

p(s) ⎝

a

M

⎞

⎤((λα1 +λβ )/p)

|Daβ fj (s)|p ⎠ ds⎦

.

(15.192)

j=1

Proof. Similar to Theorem 15.44 and transfer of Theorem 4 of [46]. Next we give Theorem 15.80. (All here are as in Theorem 15.79). Denote λ /λ 2 α2 β − 1, if λα2 ≥ λβ , δ 3 := (15.193) 1, if λα2 ≤ λβ . % 1, if λβ + λα2 ≥ p, (15.194) 2 := M 1−((λβ +λα2 )/p) , if λβ + λα2 ≤ p, and ϕ2 (x − a) :=

A0 (x − a)

((p−λβ )/p)

λα1 =0

2

λβ λα2 + λβ

λβ /p

(λβ /p)

δ3

.

(15.195) If λα1 = 0, then

x a

M

q(s) { [ |Daα2 fj+1 (s)|λα2 |Daβ fj (s)|λβ + |Daα2 fj (s)|λα2 |Daβ fj+1 (s)|λβ ]}+ j=1

15.5 Main Results

371

[ |Daα2 fM (s)|λα2 |Daβ f1 (s)|λβ + |Daα2 f1 (s)|λα2 |Daβ fM (s)|λβ ] ds ≤

(λβ +λα2 )/p) ( 2 ϕ2 (x − a) 2

x

M ((λβ +λα2 )/p) . p(s) |Daβ fj (s)|p ds

a

j=1

(15.196) Proof. Usual transfer of Theorem 5 of [46]. The general case follows. Theorem 15.81. (All here are as in Theorem 15.79). Denote ((λ +λ )/λ ) 2 α1 α2 β − 1, if λα1 + λα2 ≥ λβ , γ˜1 := (15.197) 1, if λα1 + λα2 ≤ λβ , and γ˜2 :=

1, if λα1 + λα2 + λβ ≥ p, 21−((λα1 +λα2 +λβ )/p) , if λα1 + λα2 + λβ ≤ p.

(15.198)

Set ϕ3 (x − a) := A0 (x − a)

(λα1

λβ + λα2 )(λα1 + λα2 + λβ )

λβ /p

· [λλαβ1 /p γ˜2 + 2(p−λβ )/p (γ˜1 λα2 )λβ /p ],

(15.199)

and 3 :=

1, if λα1 + λα2 + λβ ≥ p, M 1−((λα1 +λα2 +λβ )/p) , if λα1 + λα2 + λβ ≤ p.

(15.200)

Then a

x

−1 +M q(s) [ |Daα1 fj (s)|λα1 |Daα2 fj+1 (s)|λα2 |Daβ fj (s)|λβ + j=1

|Daα2 fj (s)|λα2 |Daα1 fj+1 (s)|λα1 |Daβ fj+1 (s)|λβ ]+ [ |Daα1 f1 (s)|λα1 |Daα2 fM (s)|λα2 |Daβ f1 (s)|λβ + , |Daα2 f1 (s)|λα2 |Daα1 fM (s)|λα1 |Daβ fM (s)|λβ ] ds ≤

(λα1 +λα2 +λβ )/p) ( 3 ϕ3 (x−a) 2 a

x

(15.201)

M ((λα1 +λα2 +λβ )/p) p(s) |Daβ fj (s)|p ds . j=1

372

15. Fractional Multivariate Opial Inequalities

Proof. Usual transfer of Theorem 6 of [46]. We continue. Theorem 15.82. Let β > α1 + 1, α1 ∈ R+ and let fj ∈ L1 (a, x), j = 1, . . . , M ∈ N, a, x ∈ R, a < x, have, respectively, L∞ fractional derivatives Daβ fj in [a, x], and let Daβ−k fj (a) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M. Consider also p(t) > 0 and q(t) ≥ 0, with all p(t), 1/p(t), q(t) ∈ L∞ (a, x). Let λα ≥ 0, 0 < λα+1 < 1, and p > 1. Denote λ /(λ ) 2 α α+1 − 1, if λα ≥ λα+1 , θ3 := 1, if λα ≤ λα+1 , x (1−λα+1 ) λα+1 θ3 λα+1 (1/(1−λα+1 )) L(x − a) := 2 (q(s)) ds , λα + λα+1 a (15.202) and x P1 (x − a) := (x − s)(β−α1 −1)p/(p−1) (p(s))−1/(p−1) ds, (15.203) a

T (x − a) := L(x − a)

(P1 (x − a))(p−1/p) Γ(β − α1 )

(λα +λα+1 ) ,

(15.204)

and

Also put

4 :=

Then

a

x

w1 := 2(p−1/p)(λα +λα+1 ) ,

(15.205)

Φ(x − a) := T (x − a)w1 .

(15.206)

1, if λα + λα+1 ≥ p, M 1−(λα +λα+1 /p) , if λα + λα+1 ≤ p.

(15.207)

−1

M q(s) { [ |Daα1 fj (s)|λα |Daα1 +1 fj+1 (s)|λα+1 + j=1

|Daα1 fj+1 (s)|λα |Daα1 +1 fj (s)|λα+1 ]}+

[ |Daα1 f1 (s)|λα |Daα1 +1 fM (s)|λα+1 + |Daα1 fM (s)|λα |Daα1 +1 f1 (s)|λα+1 ] ds ≤ + 2((λα +λα+1 )/p) 4 φ(x − a) a

x

M ,((λα +λα+1 )/p) p(s) |Daβ fj (s)|p ds . j=1

(15.208)

15.5 Main Results

Proof. Transfer of Theorem 7 of [46].

373

Next comes Theorem 15.83. (All are as in Theorem 15.79). Consider the special case of λα2 = λα1 + λβ . Denote λβ /p λβ ˜ T (x − a) := A0 (x − a) 2(p−2λα1 −3λβ )/p . (15.209) λα1 + λβ % 1, if 2(λα1 + λβ ) ≥ p, (15.210) 5 := M 1−(2(λα1 +λβ )/p) , if 2(λα1 + λβ ) ≤ p. Then

x

−1

M q(s) { [ |Daα1 fj (s)|λα1 |Daα2 fj+1 (s)|λα1 +λβ |Daβ fj (s)|λβ +

a

j=1

|Daα2 fj (s)|λα1 +λβ |Daα1 fj+1 (s)|λα1 |Daβ fj+1 (s)|λβ ]}+ [ |Daα1 f1 (s)|λα1 |Daα2 fM (s)|λα1 +λβ |Daβ f1 (s)|λβ + |Daα2 f1 (s)|λα1 +λβ |Daα1 fM (s)|λα1 |Daβ fM (s)|λβ ] ds ≤ + 2(λα1 +λβ )/p) ( ˜ 5 T (x − a) 2 a

x

M ,(2(λα1 +λβ )/p) p(s) |Daβ fj (s)|p ds . j=1

(15.211) Proof. Transfer of Theorem 8 of [46].

Special cases follow. Corollary 15.84. (To Theorem 15.79; λα2 = 0, p(t) = q(t) = 1.) It holds x M |Daα1 fj (s)|λα1 |Daβ fj (s)|λβ ds ≤ a

j=1

+ ∗ δ 1 ϕ1 (x − a) a

In (15.212)

M x

[ |Daβ fj (s)|p ]ds

,((λα1 +λβ )/p)

j=1

A0 (x − a)

λα2 =0

of ϕ1 (x) is given in Corollary 15.49, Equation (15.123). Proof. Transfer of Corollary 9 of [46].

.

(15.212)

374

15. Fractional Multivariate Opial Inequalities

Corollary 15.85. (To Theorem 15.79; λα2 = 0, p(t) = q(t) = 1, λα1 = λβ = 1, p = 2.) In detail: let β > α1 , α1 ∈ R+ , β − α1 > (1/2), and let fj ∈ L1 (a, x), j = 1, . . . , M ∈ N, a, x ∈ R, a < x, have, respectively, L∞ fractional derivatives Daβ fj in [a, x], and let Daβ−k fj (a) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M. Then

a

x

M

|Daα1 fj (s)| |Daβ fj (s)| ds ≤

j=1

(x − a)(β−α1 ) √ √ 2Γ(β − α1 ) β − α1 2β − 2α1 − 1

x

M +

a

, (Daβ fj (s))2 ds . (15.213)

j=1

Proof. Transfer of Corollary 10 of [46]. Corollary 15.86. (To Theorem 15.80; λα1 = 0, p(t) = q(t) = 1.) It holds

x

a

−1

M { [ |Daα2 fj+1 (s)|λα2 |Daβ fj (s)|λβ +|Daα2 fj (s)|λα2 |Daβ fj+1 (s)|λβ ]}+ j=1

[ |Daα2 fM (s)|λα2 |Daβ f1 (s)|λβ + |Daα2 f1 (s)|λα2 |Daβ fM (s)|λβ ] ds ≤

2(λβ +λα2 /p) 2 ϕ2 (x − a)

x

a

In (15.214),

M

((λβ +λα2 )/p) . (15.214) |Daβ fj (s)|p ds

j=1

A0 (x − a)

λα1 =0

of ϕ2 (x − a) is given in Corollary 15.51; see Equation (15.128). Proof. Transfer of Corollary 11 of [46]. Corollary 15.87. (To Theorem 15.80, λα1 = 0, p(t) = q(t) = 1, λα2 = λβ = 1, p = 2.) In detail: let α2 ∈ R+ , β > α2 , β − α2 > (1/2), and let fj ∈ L1 (a, x), j = 1, . . . , M ∈ N, a, x ∈ R, a < x, have, respectively, L∞ fractional derivatives Daβ fj in [a, x], and let Daβ−k fj (a) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M.

15.5 Main Results

375

Then

x

−1

M { [ |Daα2 fj+1 (s)| |Daβ fj (s)| + |Daα2 fj (s)| |Daβ fj+1 (s)|]}+

a

j=1

[ |Daα2 fM (s)| |Daβ f1 (s)| + |Daα2 f1 (s)| |Daβ fM (s)| ] ds √ M

x + , 2(x − a)(β−α2 ) √ √ (Daβ fj (s))2 ds . ≤ Γ(β − α2 ) β − α2 2β − 2α2 − 1 a j=1 (15.215) Proof. Transfer of Corollary 12 of [46]. Corollary 15.88. (To Theorem 15.81; λα1 = λα2 = λβ = 1, p(t) = q(t) = 1, p = 3.) It holds a

x

−1 +M

[ |Daα1 fj (s)| |Daα2 fj+1 (s)| |Daβ fj (s)|+

j=1

|Daα2 fj (s)| |Daα1 fj+1 (s)| |Daβ fj+1 (s)| ]+[ |Daα1 f1 (s)| |Daα2 fM (s)| |Daβ f1 (s)|+ |Daα2 f1 (s)|

|Daα1 fM (s)|

|Daβ fM (s)|

,

] ds ≤

2ϕ∗3 (x

+ − a)

x

M

a

, |Daβ fj (s)|3 ds .

j=1

(15.216)

Here ϕ∗3 (x − a) :=

√ 3

1 2+ √ 3 6

A0 (x − a),

(15.217)

where in this special case A0 (x − a) = 4(x − a)(2β−α1 −α2 ) . Γ(β − α1 ) Γ(β − α2 ) [3(3β − 3α1 − 1)(3β − 3α2 − 1)(2β − α1 − α2 ) ]2/3 (15.218) Proof. Transfer of Corollary 13 of [46]. Corollary 15.89. (To Theorem 15.82; λα = 1, λα+1 = 1/2, p(t) = q(t) = 1, p = 3/2.) In detail: let α1 ∈ R+ , β > α1 + 1, and let fj ∈ L1 (a, x), j = 1, . . . , M ∈ N, a, x ∈ R, a < x, have, respectively, L∞ fractional derivatives Daβ fj in [a, x], and let Daβ−k fj (a) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M.

376

15. Fractional Multivariate Opial Inequalities

Set

∗

Φ (x − a) :=

2 √ 3β − 3α1 − 2

·

(x − a)(3β−3α1 −1/2) . (Γ(β − α1 ))3/2

(15.219)

Then a

x

. . −1

M { [ |Daα1 fj (s)| |Daα1 +1 fj+1 (s)|+|Daα1 fj+1 (s)| |Daα1 +1 fj (s)| ]}+ j=1

[ |Daα1 f1 (s)|

. . |Daα1 +1 fM (s)| + |Daα1 fM (s)| |Daα1 +1 f1 (s)| ] ds ≤ + 2Φ∗ (x − a)

x

a

M

, |Daβ fj (s)|3/2 ds .

(15.220)

j=1

Proof. Transfer of Corollary 14 of [46]. Corollary 15.90. (To Theorem 15.83; p = 2(λα1 + λβ ) > 1, p(t) = q(t) = 1.) It holds a

x

−1

M { [ |Daα1 fj (s)|λα1 |Daα2 fj+1 (s)|λα1 +λβ |Daβ fj (s)|λβ + j=1

|Daα2 fj (s)|λα1 +λβ |Daα1 fj+1 (s)|λα1 |Daβ fj+1 (s)|λβ ]}+ [ |Daα1 f1 (s)|λα1 |Daα2 fM (s)|λα1 +λβ |Daβ f1 (s)|λβ + |Daα2 f1 (s)|λα1 +λβ |Daα1 fM (s)|λα1 |Daβ fM (s)|λβ ] ds ≤ + 2T˜(x − a) a

x

M

, |Daβ fj (s)|2(λα1 +λβ ) ds .

(15.221)

j=1

˜(x − a) in (15.221) is given by (15.209) and in detail by T˜˜(x − a) of Here T˜ Corollary 15.55 and Equations (15.136)–(15.139). Proof. Transfer of Corollary 15 of [46]. Corollary 15.91. (To Theorem 15.83; p = 4, λα1 = λβ = 1, p(t) = q(t) = 1.) It holds a

x

−1

M { [ |Daα1 fj (s)| (Daα2 fj+1 (s))2 |Daβ fj (s)|+ j=1

(Daα2 fj (s))2 |Daα1 fj+1 (s)| |Daβ fj+1 (s)| ]} + [ |Daα1 f1 (s)| (Daα2 fM (s))2 |Daβ f1 (s)|+

15.5 Main Results + (Daα2 f1 (s))2 |Daα1 fM (s)| |Daβ fM (s)| ] ds ≤ 2T˜(x − a)

x a

M

377

, (Daβ fj (s))4 ds .

j=1

(15.222)

Here in (15.222) we have that T˜(x − a) = T ∗ (x − a) of Corollary 15.56; see the Equations (15.141)–(15.145). Proof. Transfer of Corollary 16 of [46]. Next we present the L∞ case. Theorem 15.92. Let α1 , α2 ∈ R+ , β > α1 , α2 , and let fj ∈ L1 (a, x), j = 1, . . . , M ∈ N, a, x ∈ R, a < x, have, respectively, L∞ fractional derivatives Daβ fj in [a, x], and let Daβ−k fj (a) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M. Consider also p(s) ≥ 0, p(s) ∈ L∞ (a, x). Let λα1 , λα2 , λβ ≥ 0. Set ρ(x − a) =

p(s)∞ (x − a)(βλα1 −α1 λα1 +βλα2 −α2 λα2 +1) . (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα2 [βλα1 − α1 λα1 + βλα2 − α2 λα2 + 1] (15.223) Then a

x

−1

M p(s) { [ |Daα1 fj (s)|λα1 |Daα2 fj+1 (s)|λα2 |Daβ fj (s)|λβ + j=1

|Daα2 fj (s)|λα2 |Daα1 fj+1 (s)|λα1 |Daβ fj+1 (s)|λβ ]}+ [ |Daα1 f1 (s)|λα1 |Daα2 fM (s)|λα2 |Daβ f1 (s)|λβ + |Daα2 f1 (s)|λα2 |Daα1 fM (s)|λα1 |Daβ fM (s)|λβ ] ds M

2(λ +λ ) 2λ ≤ ρ(x − a) {Daβ fj ∞ α1 β + Daβ fj ∞ α2 } . j=1

Proof. Transfer of Corollary 17 of [46]. Similarly we give

(15.224)

378

15. Fractional Multivariate Opial Inequalities

Theorem 15.93. (As in Theorem 15.92; λα2 = 0.) It holds

x

M p(s) |Daα1 fj (s)|λα1 |Daβ fj (s)|λβ ds ≤

a

j=1

M p(s)∞ (x − a)(βλα1 −α1 λα1 +1) λα1 +λβ β . (15.225) D f · a j ∞ (Γ(β − α1 + 1))λα1 [βλα1 − α1 λα1 + 1] j=1

Proof. Based on Theorem 18 of [46].

It follows Theorem 15.94. (As in Theorem 15.92; λα2 = λα1 + λβ .) It holds

x

−1

M p(s) { [ |Daα1 fj (s)|λα1 |Daα2 fj+1 (s)|λα1 +λβ |Daβ fj (s)|λβ +

a

j=1

|Daα2 fj (s)|λα1 +λβ |Daα1 fj+1 (s)|λα1 |Daβ fj+1 (s)|λβ ]}+ [ |Daα1 f1 (s)|λα1 |Daα2 fM (s)|λα1 +λβ |Daβ f1 (s)|λβ + |Daα2 f1 (s)|λα1 +λβ |Daα1 fM (s)|λα1 |Daβ fM (s)|λβ ] ds ≤ ≤

p(s)∞ (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))(λα1 +λβ )

2(x − a)(2βλα1 −α1 λα1 +βλβ −α2 λα1 −α2 λβ +1) 2(λ +λ ) · Daβ fj ∞ α1 β . (2βλα1 − α1 λα1 + βλβ − α2 λα1 − α2 λβ + 1) j=1 M

·

(15.226) Proof. By Theorem 19 of [46].

We continue with Theorem 15.95. (As in Theorem 15.92; λβ = 0, λα1 = λα2 .) It holds

x a

−1

M p(s) { [ |Daα1 fj (s)|λα1 |Daα2 fj+1 (s)|λα1 + |Daα2 fj (s)|λα1 |Daα1 fj+1 (s)|λα1 ]}+ j=1

[ |Daα1 f1 (s)|λα1 |Daα2 fM (s)|λα1 + |Daα2 f1 (s)|λα1 |Daα1 fM (s)|λα1 ] ds M , + 2λ ≤ 2ρ∗ (x − a) Daβ fj ∞ α1 . j=1

(15.227)

15.5 Main Results

Here we have

(2βλα1

379

ρ∗ (x − a) :=

(x − a)(2βλα1 −α1 λα1 −α2 λα1 +1) p(s)∞ . − α1 λα1 − α2 λα1 + 1)(Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα1 (15.228)

Proof. Based on Theorem 20 of [46].

Next we give Theorem 15.96. (As in Theorem 15.92; λα1 = 0, λα2 = λβ .) It holds

x

−1

M p(s) { [ |Daα2 fj+1 (s)|λα2 |Daβ fj (s)|λα2 +|Daα2 fj (s)|λα2 |Daβ fj+1 (s)|λα2 ]}+

a

j=1

[ |Daα2 fM (s)|λα2 |Daβ f1 (s)|λα2 + |Daα2 f1 (s)|λα2 |Daβ fM (s)|λα2 ] ds ≤ 2

(x − a)(βλα2 −α2 λα2 +1) p(s)∞ (βλα2 − α2 λα2 + 1)(Γ(β − α2 + 1))λα2

Proof. Based on Theorem 21 of [46].

M 2λ Daβ fj ∞ α2 . (15.229) · j=1

Some special cases follow. Corollary 15.97. (To Theorem 15.95; all are as in Theorem 15.92; λβ = 0, λα1 = λα2 , α2 = α1 + 1.) It holds a

x

−1

M p(s) { [ |Daα1 fj (s)|λα1 |Daα1 +1 fj+1 (s)|λα1 + j=1

|Daα1 +1 fj (s)|λα1 |Daα1 fj+1 (s)|λα1 ]}+

[ |Daα1 f1 (s)|λα1 |Daα1 +1 fM (s)|λα1 +|Daα1 +1 f1 (s)|λα1 |Daα1 fM (s)|λα1 ] ds ≤ 2

(2βλα1

(x − a)(2βλα1 −2α1 λα1 −λα1 +1) p(s)∞ − 2α1 λα1 − λα1 + 1)(β − α1 )λα1 (Γ(β − α1 ))2λα1

+ M , 2λα Daβ fj ∞ 1 . · j=1

(15.230)

Proof. Based on Theorem 22 of [46].

380

15. Fractional Multivariate Opial Inequalities

Corollary 15.98. (to Corollary 15.97) In detail; let α1 ∈ R+ , β > α1 +1, and let fj ∈ L1 (a, x), j = 1, . . . , M ∈ N, a, x ∈ R, a < x, have, respectively, L∞ fractional derivatives Daβ fj in [a, x], and let Daβ−k fj (a) = 0, for k = 1, . . . , [β] + 1; j = 1, . . . , M. Then a

x

−1

M { [ |Daα1 fj (s)| |Daα1 +1 fj+1 (s)| + |Daα1 fj+1 (s)| |Daα1 +1 fj (s)| ]}+ j=1

[ |Daα1 f1 (s)| |Daα1 +1 fM (s)| + |Daα1 fM (s)| |Daα1 +1 f1 (s)| ] ds ≤ M (x − a)2(β−α1 ) β 2 D f j a ∞ . (β − α1 )2 (Γ(β − α1 ))2 j=1

(15.231)

Proof. Based on Corollary 23 of [46]. Corollary 15.99. (to Corollary 15.98) It holds ⎛ ⎞ x M ⎝ |Daα1 fj (s)| |Daα1 +1 fj (s)|⎠ ds ≤ a

j=1

M (x − a)2(β−α1 ) β 2 . D f j a ∞ 2(β − α1 )2 (Γ(β − α1 ))2 j=1

(15.232)

Proof. Based on inequality (15.155) of Proposition 15.64. Next we apply the previous results of this subsection to the spherical shell A . We make Assumption 15.100. Let α1 , α2 ∈ R+ , β > α1 , α2 , β − αi > (1/p), p > 1, i = 1, 2, and for j = 1, . . . , M, M ∈ N, let fj ∈ L1 (A) with β ∂R f (x) 1 j , ∈ L∞ (A), x ∈ A, ∂rβ

A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2, 0 < R1 < R2 .

15.5 Main Results

381

β Further assume that each DR f (rw) ∈ R for almost all r ∈ [R1 , R2 ], for 1 j β f (rw)| ≤ Mj for some Mj > 0; j = each w ∈ S N −1 , and for these |DR 1 j 1, . . . , M . For each w ∈ S N −1 − (∪M j=1 K(fj )) , we assume that fj (· w) has β an L∞ fractional derivative DR1 fj (· w) in [R1 , R2 ], and that β−k DR fj (R1 w) = 0, k = 1, . . . , [β] + 1; 1

j = 1, . . . , M . Let λβ > 0 and λα1 , λα2 ≥ 0, such that λβ < p. If α1 = 0 we set λα1 = 1, and if α2 = 0 we set λα2 = 1. We need Notation 15.101. (on Assumption 15.100) We set s (1−N /p−1) Pi (s) := (s − r)p(β−αi −1)/p−1 (r + R1 ) dr, 0

i = 1, 2; 0 ≤ s ≤ R2 − R1 ,

A(s) :=

(s + R1 )(N −1)(1−(λβ /p)) (P1 (s)) (Γ(β − α1 ))

λα1

(15.233)

λα1 (p−1/p)

(Γ(β − α2 ))

(P2 (s))

λα2 (p−1/p)

λα2

, (15.234)

and A0 (R2 − R1 ) :=

(p−λβ )/p

R2 −R1

(A(s))

p/(p−λβ )

ds

.

(15.235)

0

We present Theorem 15.102. (All are as in Assumption 15.100 and Notation 15.101.) Denote ϕ1 (R2 − R1 ) :=

A0 (R2 − R1 )

λα2 =0

λβ p

(λβ /p) .

(15.236)

Let λα1 > 0, λα2 = 0, and p = λα1 + λβ > 1. Then + M ⏐ α1 ⏐ ⏐ β ⏐ , ⏐ ∂R1 fj (x) ⏐λα1 ⏐ ∂R1 fj (x) ⏐λβ ⏐ ⏐ ⏐ ⏐ dx ≤ ∂rα1 ∂rβ A j=1 M ⏐ β ⏐ , + ⏐ ∂R1 fj (x) ⏐p ϕ1 (R2 − R1 ) ⏐ ⏐ dx . ∂rβ A j=1

(15.237)

382

15. Fractional Multivariate Opial Inequalities

Proof. Based on Theorem 15.79 and similar to the proof of Theorem 15.67; notice here λS N −1 (∪M j=1 K(fj )) = 0. Next we give Theorem 15.103. (All are as in Assumption 15.100 and Notation 15.101.) We denote λ /λ 2 α2 β − 1, if λα2 ≥ λβ , δ 3 := (15.238) 1, if λα2 ≤ λβ , and ϕ2 (R2 − R1 ) :=

A0 (R2 − R1 )

λα1 =0

2(λα2 /p)

λβ p

(λβ /p)

(λβ /p)

δ3

.

(15.239) Here λα1 = 0, λα2 > 0 and p = λβ + λα2 > 1. Then M −1 +⏐ α2 ⏐ ⏐ β ⏐ ⏐ ∂R1 fj+1 (x) ⏐λα2 ⏐ ∂R1 fj (x) ⏐λβ { ⏐ ⏐ ⏐ + ⏐ ∂rα2 ∂rβ A j=1 ⏐ ∂ α2 f (x) ⏐λα2 ⏐ ∂ β f (x) ⏐λβ , ⏐ R1 j ⏐ ⏐ R1 j+1 ⏐ ⏐ ⏐ ⏐ ⏐ }+ ∂rα2 ∂rβ +⏐ ∂ α2 fM (x) ⏐λα2 ⏐ ∂ β f1 (x) ⏐λβ ⏐ ∂ α2 f1 (x) ⏐λα2 ⏐ ∂ β fM (x) ⏐λβ , ⏐ R1 ⏐ ⏐ R1 ⏐ ⏐ ⏐ R1 ⏐ ⏐ R ⏐ ⏐ ⏐ ⏐ +⏐ 1 α2 ⏐ ⏐ ⏐ dx ≤ ∂rα2 ∂rβ ∂r ∂rβ M ⏐ β ⏐ , + ⏐ ∂R1 fj (x) ⏐p 2ϕ2 (R2 − R1 ) (15.240) ⏐ ⏐ dx . ∂rβ A j=1 Proof. Based on Theorem 15.80 and similar to the proof of Theorem 15.67. The general case follows. Theorem 15.104. (All are as in Assumption 15.100 and Notation 15.101.) Here λα1 , λα2 > 0, p = λα1 + λα2 + λβ > 1. Denote ((λ +λ )/λ ) 2 α1 α2 β − 1, if λα1 + λα2 ≥ λβ , (15.241) γ˜1 := 1, if λα1 + λα2 ≤ λβ , and

ϕ3 (R2 − R1 ) := A0 (R2 − R1 )

(λα1

λβ + λα2 )p

(λβ /p)

[λα1 (λβ /p) + 2(λα1 +λα2 )/p (γ˜1 λα2 )(λβ /p) ].

(15.242)

15.5 Main Results

383

Then M −1 +⏐ α1 ⏐ ⏐ α2 ⏐ ⏐ β ⏐ ⏐ ∂R1 fj (x) ⏐λα1 ⏐ ∂R1 fj+1 (x) ⏐λα2 ⏐ ∂R1 fj (x) ⏐λβ ⏐ ⏐ ⏐ ⏐ ⏐ + ⏐ α α β ∂r 1 ∂r 2 ∂r A j=1 ⏐ ∂ α2 f (x) ⏐λα2 ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ R1 j ⏐ ⏐ R1 j+1 ⏐ ⏐ R1 j+1 ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ + ∂rα2 ∂rα1 ∂rβ +⏐ ∂ α1 f1 (x) ⏐λα1 ⏐ ∂ α2 fM (x) ⏐λα2 ⏐ ∂ β f1 (x) ⏐λβ ⏐ ⏐ R1 ⏐ ⏐ R1 ⏐ ⏐ R1 ⏐ ⏐ ⏐ ⏐ ⏐ + ⏐ ∂rα1 ∂rα2 ∂rβ ⏐ ∂ α2 f (x) ⏐λα2 ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ R1 1 ⏐ ⏐ R1 M ⏐ ⏐ R1 M ⏐ dx ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ α α 2 1 ∂r ∂r ∂rβ M ⏐ β ⏐ , + ⏐ ∂R1 fj (x) ⏐p ≤ 2ϕ3 (R2 − R1 ) (15.243) ⏐ ⏐ dx . ∂rβ A j=1 Proof. Based on Theorem 15.81 and similar to the proof of Theorem 15.67. We continue with Theorem 15.105. (All are as in Assumption 15.100.) Here α2 = α1 + 1, λα := λα1 ≥ 0, λα+1 := λα2 ∈ (0, 1) and p = λα + λα1 > 1. Denote (λ /λ ) 2 α α+1 − 1, if λα ≥ λα+1 , (15.244) θ3 := 1, if λα ≤ λα+1 , L(R2 − R1 ) := ) *(1−λα+1 ) (1 − λα+1 ) (N −λα+1 )/(1−λα+1 ) (N −λα+1 )/(1−λα+1 ) R2 − R1 2 (N − λα+1 ) λ θ3 λα+1 α+1 , (15.245) p and

P1 (R2 − R1 ) :=

R2

(R2 − r)(β−α1 −1)p/(p−1) r(1−N )/(p−1) dr,

(15.246)

R1

and

Φ(R2 − R1 ) := L(R2 − R1 )

Then

(P1 (R2 − R1 ))(p−1) (Γ(β − α1 ))p

2(p−1) .

M −1 +⏐ α1 ⏐ ⏐ α1 +1 ⏐ ⏐ ∂R1 fj (x) ⏐λα ⏐ ∂R1 fj+1 (x) ⏐λα+1 { + ⏐ ⏐ ⏐ ⏐ ∂rα1 ∂rα1 +1 A j=1

(15.247)

384

15. Fractional Multivariate Opial Inequalities

⏐ ∂ α1 f (x) ⏐λα ⏐ ∂ α1 +1 f (x) ⏐λα+1 , j ⏐ R1 j+1 ⏐ ⏐ R1 ⏐ } ⏐ ⏐ ⏐ ⏐ ∂rα1 ∂rα1 +1 +⏐ ∂ α1 f1 (x) ⏐ λα ⏐ ∂ α1 +1 fM (x) ⏐ λα+1 ⏐ ∂ α1 fM (x) ⏐ λα ⏐ ∂ α1 +1 f1 (x) ⏐ λα+1 , ⏐ ⏐ R1 ⏐ ⏐ ⏐ R1 ⏐ ⏐ R ⏐ R dx + ⏐ 1 α +⏐ 1 α ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ∂r 1 ∂rα1 +1 ∂r 1 ∂rα1 +1 M ⏐ β ⏐ , + ⏐ ∂R1 fj (x) ⏐p ≤ 2Φ(R2 − R1 ) ⏐ ⏐ dx . ∂rβ A j=1

(15.248)

Proof. Based on Theorem 15.82 and similar to the proof of Theorem 15.67. We also give Theorem 15.106. (All are as in Assumption 15.100 and Notation 15.101.) Here λα2 = λα1 + λβ and p = 2(λα1 + λβ ) > 1. Denote λ /p 2λβ β T˜(R2 − R1 ) := A0 (R2 − R1 ) 2−λβ /p . (15.249) p Then M −1 +⏐ α1 ⏐ ⏐ α2 ⏐ ⏐ β ⏐ ⏐ ∂R1 fj (x) ⏐λα1 ⏐ ∂R1 fj+1 (x) ⏐λα1 +λβ ⏐ ∂R1 fj (x) ⏐λβ { ⏐ ⏐ ⏐ ⏐ ⏐ + ⏐ α α β ∂r 1 ∂r 2 ∂r A j=1 ⏐ ∂ α2 f (x) ⏐λα1 +λβ ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ R1 j ⏐ ⏐ R1 j+1 ⏐ ⏐ R1 j+1 ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ }+ ∂rα2 ∂rα1 ∂rβ +⏐ ∂ α1 f1 (x) ⏐λα1 ⏐ ∂ α2 fM (x) ⏐λα1 +λβ ⏐ ∂ β f1 (x) ⏐λβ ⏐ ⏐ R1 ⏐ ⏐ R1 ⏐ ⏐ R1 ⏐ ⏐ ⏐ ⏐ ⏐ + ⏐ ∂rα1 ∂rα2 ∂rβ ⏐ ∂ α2 f (x) ⏐λα1 +λβ ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ R1 1 ⏐ ⏐ R1 M ⏐ ⏐ R1 M ⏐ dx ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ∂rα2 ∂rα1 ∂rβ ⎤ ⎡ M ⏐ β p ∂R1 fj (x) ⏐ ⏐ ⏐ ≤ 2T˜(R2 − R1 ) ⎣ (15.250) ⏐ ⏐ dx⎦ . ∂rβ A j=1 Proof. Based on Theorem 15.83.

Next we give L∞ results on the shell A involving several functions. We make Assumption 15.107. Let α1 , α2 ∈ R+ , β > α1 , α2 , and for j = 1, . . . , M, M ∈ N, let fj ∈ L1 (A) with β f (x) ∂R 1 j , ∈ L∞ (A), x ∈ A, ∂rβ

15.5 Main Results

385

A := B(0, R2 ) − B(0, R1 ) ⊆ RN , N ≥ 2, 0 < R1 < R2 . β f (rw) ∈ R for almost all r ∈ [R1 , R2 ], for Further assume that each DR 1 j β f (rw)| ≤ Mj for some Mj > 0; j = each w ∈ S N −1 , and for these |DR 1 j N −1 M − (∪j=1 K(fj )) , we assume that fj (· w) has 1, . . . , M . For each w ∈ S β f (· w) in [R1 , R2 ], and that an L∞ fractional derivative DR 1 j β−k DR fj (R1 w) = 0, k = 1, . . . , [β] + 1; 1

j = 1, . . . , M . Let λα1 , λα2 , λβ ≥ 0. If α1 = 0 we set λα1 = 1, and if α2 = 0 we set λα2 = 1. We present Theorem 15.108. (All here are as in Assumption 15.107.) Set ρ(R2 − R1 ) = R2 N −1 (R2 − R1 )(βλα1 −α1 λα1 +βλα2 −α2 λα2 +1) . (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα2 (βλα1 − α1 λα1 + βλα2 − α2 λα2 + 1) (15.251) Then M −1 +⏐ α1 ⏐ ⏐ α2 ⏐ ⏐ β ⏐ ⏐ ∂R1 fj (x) ⏐λα1 ⏐ ∂R1 fj+1 (x) ⏐λα2 ⏐ ∂R1 fj (x) ⏐λβ { ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ + ∂rα1 ∂rα2 ∂rβ A j=1 ⏐ ∂ α2 f (x) ⏐λα2 ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ ⏐ R1 j+1 ⏐ ⏐ R1 j+1 ⏐ ⏐ R1 j ⏐ ⏐ ⏐ ⏐ ⏐ }+ ⏐ ∂rα2 ∂rα1 ∂rβ +⏐ ∂ α1 f1 (x) ⏐λα1 ⏐ ∂ α2 fM (x) ⏐λα2 ⏐ ∂ β f1 (x) ⏐λβ ⏐ R1 ⏐ ⏐ R1 ⏐ ⏐ R1 ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ + ∂rα1 ∂rα2 ∂rβ ⏐ ∂ α2 f (x) ⏐λα2 ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ R1 1 ⏐ ⏐ R1 M ⏐ ⏐ R1 M ⏐ dx ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ∂rα2 ∂rα1 ∂rβ ≤

M

2π N/2 ρ(R2 − R1 ) [Mj 2(λα1 +λβ ) + Mj 2λα2 ] . Γ(N/2) j=1

(15.252)

Proof. Based on Theorem 15.92 and a similar proof to that of Theorem 15.73. Similarly we give

386

15. Fractional Multivariate Opial Inequalities

Theorem 15.109. (All are as in Assumption 15.107; λα2 = 0.) Then + M ⏐ α1 ⏐ ⏐ β ⏐ , ⏐ ∂R1 fj (x) ⏐λα1 ⏐ ∂R1 fj (x) ⏐λβ ⏐ ⏐ ⏐ ⏐ dx ≤ ∂rα1 ∂rβ A j=1 ⎞ ⎛ M N/2 R2 N −1 (R2 − R1 )(βλα1 −α1 λα1 +1) (λα1 +λβ ) ⎠ 2π ⎝ . M j Γ(N/2) (Γ(β − α1 + 1))λα1 [βλα1 − α1 λα1 + 1] j=1 (15.253) Proof. Based on Theorem 15.93, similar to Theorem 15.73.

It follows Theorem 15.110. (All are as in Assumption 15.107; λα2 = λα1 + λβ .) Then M −1 +⏐ α1 ⏐ ⏐ α2 ⏐ ⏐ β ⏐ ⏐ ∂R1 fj (x) ⏐λα1 ⏐ ∂R1 fj+1 (x) ⏐λα1 +λβ ⏐ ∂R1 fj (x) ⏐λβ { ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ + ∂rα1 ∂rα2 ∂rβ A j=1 ⏐ ∂ α2 f (x) ⏐λα1 +λβ ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ R1 j ⏐ ⏐ R1 j+1 ⏐ ⏐ R1 j+1 ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ } α α 2 1 ∂r ∂r ∂rβ +⏐ ∂ α1 f1 (x) ⏐λα1 ⏐ ∂ α2 fM (x) ⏐λα1 +λβ ⏐ ∂ β f1 (x) ⏐λβ ⏐ ⏐ ⏐ R1 ⏐ ⏐ R1 ⏐ + ⏐ R1 α1 ⏐ ⏐ ⏐ ⏐ ⏐ + ∂r ∂rα2 ∂rβ ⏐ ∂ α2 f (x) ⏐λα1 +λβ ⏐ ∂ α1 f (x) ⏐λα1 ⏐ ∂ β f (x) ⏐λβ , ⏐ ⏐ R1 M ⏐ ⏐ R1 M ⏐ ⏐ R1 1 dx ≤ ⏐ ⏐ ⏐ ⏐ ⏐ ⏐ ∂rα2 ∂rα1 ∂rβ R2 N −1 · (Γ(β − α1 + 1))λα1 (Γ(β − α2 + 1))λα1 +λβ ⎡ ⎤ M (2βλα1 −α1 λα1 +βλβ −α2 λα1 −α2 λβ +1) 4π N/2 (R2 − R1 ) ⎣ . Mj 2(λα1 +λβ ) ⎦ [2βλα1 − α1 λα1 + βλβ − α2 λα1 − α2 λβ + 1] j=1 Γ(N/2) (15.254) Proof. Based on Theorem 15.94 and similar to the proof of Theorem 15.73. We continue with Theorem 15.111. (All are as in Assumption 15.107; here λβ = 0, λα1 = λα2 .) Then M −1 +⏐ α1 ⏐ ⏐ α2 ⏐ ⏐ ∂R1 fj (x) ⏐λα1 ⏐ ∂R1 fj+1 (x) ⏐λα1 { ⏐ ⏐ ⏐ ⏐ + ∂rα1 ∂rα2 A j=1

15.5 Main Results

387

⏐ ∂ α2 f (x) ⏐λα1 ⏐ ∂ α1 f (x) ⏐λα1 , ⏐ R1 j ⏐ ⏐ R1 j+1 ⏐ }+ ⏐ ⏐ ⏐ ⏐ α 2 ∂r ∂rα1 +⏐ ∂ α1 f1 (x) ⏐λα1 ⏐ ∂ α2 fM (x) ⏐λα1 ⏐ R1 ⏐ ⏐ R1 ⏐ ⏐ ⏐ ⏐ ⏐ + ∂rα1 ∂rα2 ⏐ ∂ α2 f (x) ⏐λα1 ⏐ ∂ α1 f (x) ⏐λα1 , 4π N/2 ⏐ ⏐ R1 M ⏐ ⏐ R1 1 dx ≤ ⏐ ⏐ ⏐ ⏐ ∂rα2 ∂rα1 Γ(N/2)

R2 N −1 (R2 − R1 )(2βλα1 −α1 λα1 −α2 λα1 +1)

λ

(Γ(β − α1 + 1)Γ(β − α2 + 1)) α1 (2βλα1 − α1 λα1 − α2 λα1 + 1) ⎛ ⎞ M ⎝ (15.255) Mj 2λα1 ⎠ . j=1

Proof. Based on Theorem 15.95.

Next we give Theorem 15.112. (All are as in Assumption 15.107; here λα1 = 0, λα2 = λβ .) Then M −1 +⏐ α2 ⏐ ⏐ β ⏐ ⏐ ∂R1 fj+1 (x) ⏐λα2 ⏐ ∂R1 fj (x) ⏐λα2 { ⏐ ⏐ ⏐ ⏐ + ∂rα2 ∂rβ A j=1 ⏐ ∂ α2 f (x) ⏐λα2 ⏐ ∂ β f (x) ⏐λα2 , ⏐ R1 j ⏐ ⏐ R1 j+1 ⏐ }+ ⏐ ⏐ ⏐ ⏐ ∂rα2 ∂rβ +⏐ ∂ α2 fM (x) ⏐λα2 ⏐ ∂ β f1 (x) ⏐λα2 ⏐ R1 ⏐ ⏐ R1 ⏐ ⏐ ⏐ ⏐ ⏐ + ∂rα2 ∂rβ ⏐ ∂ α2 f (x) ⏐λα2 ⏐ ∂ β f (x) ⏐λα2 , 4π N/2 ⏐ ⏐ R1 M ⏐ ⏐ R1 1 dx ≤ ⏐ ⏐ ⏐ ⏐ ∂rα2 ∂rβ Γ(N/2) ⎞ ⎛ M N −1 (βλα2 −α2 λα2 +1) (R2 − R1 ) R2 ⎝ Mj 2λα2 ⎠ . (βλα2 − α2 λα2 + 1)(Γ(β − α2 + 1))λα2 j=1 Proof. Based on Theorem 15.96. We also mention a special case.

(15.256)

388

15. Fractional Multivariate Opial Inequalities

Corollary 15.113. (to Theorem 15.111) All are as in Assumption 15.107; here λβ = 0, λα1 = λα2 , α2 = α1 + 1. Then M −1 +⏐ α1 ⏐ ⏐ α1 +1 ⏐ ⏐ ∂R1 fj (x) ⏐λα1 ⏐ ∂R1 fj+1 (x) ⏐λα1 { + ⏐ ⏐ ⏐ ⏐ ∂rα1 ∂rα1 +1 A j=1 ⏐ ∂ α1 +1 f (x) ⏐λα1 ⏐ ∂ α1 f (x) ⏐λα1 , j ⏐ ⏐ R1 j+1 ⏐ ⏐ R1 }+ ⏐ ⏐ ⏐ ⏐ ∂rα1 +1 ∂rα1 +⏐ ∂ α1 f1 (x) ⏐λα1 ⏐ ∂ α1 +1 fM (x) ⏐λα1 ⏐ ⏐ R1 ⏐ ⏐ R1 ⏐ ⏐ ⏐ + ⏐ ∂rα1 ∂rα1 +1 ⏐ ∂ α1 +1 f (x) ⏐λα1 ⏐ ∂ α1 f (x) ⏐λα1 , 4π N/2 1 ⏐ ⏐ R1 M ⏐ ⏐ R1 dx ≤ ⏐ ⏐ ⏐ ⏐ α +1 α 1 1 ∂r ∂r Γ(N/2) (2βλα1

R2 N −1 (R2 − R1 )(2βλα1 −2α1 λα1 −λα1 +1) − 2α1 λα1 − λα1 + 1)(β − α1 )λα1 (Γ(β − α1 ))2λα1

M

Mj

2λα1

.

j=1

(15.257)

Proof. Based on Corollary 15.97.

We ﬁnish the chapter with the proof that Dα f of Lemma 15.7 (see (15.8), also other similar fractional derivatives here) are such that Dα f ∈ AC([0, x]) for β − α ≥ 1 and Dα f ∈ C([0, x]), for β − α ∈ (0, 1). The last derives from the next. Proposition 15.114. Let r > 0, F ∈ L∞ (a, b), and s G(s) := (s − t)r−1 F (t)dt,

(15.258)

a

all s ∈ [a, b]. Then G ∈ AC([a, b]) for r ≥ 1 and G ∈ C([a, b]), only for r ∈ (0, 1). Proof. (1) Case r ≥ 1. We use the deﬁnition of absolute continuity. So for every > 0 we need δ > 0: whenever (ai , bi ), i = 1, . . . , n, are disjoint open subintervals of [a, b], then n

(bi − ai ) < δ ⇒

i=1

n

|G(bi ) − G(ai )| < .

i=1

If F ∞ = 0, then G(s) = 0, for all s ∈ [a, b], the trivial case and all fulﬁlled. So we assume F ∞ = 0. Hence we have bi ai (bi − t)r−1 F (t)dt − (ai − t)r−1 F (t)dt = G(bi ) − G(ai ) = a

a

15.5 Main Results

ai

(bi − t)r−1 F (t)dt −

a

ai

(ai − t)r−1 F (t)dt +

a ai

(bi − t)r−1 − (ai − t)r−1 F (t)dt +

a

bi

389

(bi − t)r−1 F (t)dt =

ai bi

(bi − t)r−1 F (t)dt. (15.259)

ai

Call Ii :=

ai

|(bi − t)r−1 − (ai − t)r−1 |dt.

(15.260)

a

Thus ) * (bi − ai )r |G(bi ) − G(ai )| ≤ Ii + F ∞ := Ti . r

(15.261)

If r = 1, then Ii = 0, and |G(bi ) − G(ai )| ≤ F ∞ (bi − ai ),

(15.262)

for all i := 1, . . . , n.

If r > 1, then because (bi − t)r−1 − (ai − t)r−1 ≥ 0, for all t ∈ [a, ai ], we ﬁnd ai (bi − a)r − (ai − a)r − (bi − ai )r (bi − t)r−1 − (ai − t)r−1 dt = Ii = r a r(ξ − a)r−1 (bi − ai ) − (bi − ai )r , r for some ξ ∈ (ai , bi ). Therefore, it holds =

Ii ≤

and

Ii +

r(b − a)r−1 (bi − ai ) − (bi − ai )r , r (bi − ai )r r

(15.263)

(15.264)

≤ (b − a)r−1 (bi − ai ).

(15.265)

That is, Ti ≤ F ∞ (b − a)r−1 (bi − ai ), so that |G(bi ) − G(ai )| ≤ F ∞ (b − a)r−1 (bi − ai ), for all i = 1, . . . , n. (15.266) So in the case of r = 1, and by choosing δ := /F ∞ , we get n n |G(bi ) − G(ai )| ≤(15.262) F ∞ (bi − ai ) ≤ F ∞ δ = , i=1

i=1

(15.267)

390

15. Fractional Multivariate Opial Inequalities

proving for r = 1 that G is absolutely continuous. In the case of r > 1, and by choosing δ := /F ∞ (b − a)r−1 , we get n n (15.266) r−1 |G(bi ) − G(ai )| ≤ F ∞ (b − a) (bi − ai ) (15.268) i=1

i=1

≤ F ∞ (b − a)r−1 δ = , proving for r > 1 that G is absolutely continuous again. (2) Case of 0 < r < 1. Let ai∗ , bi∗ ∈ [a, b] : ai∗ ≤ bi∗ . Then (ai∗ − t)r−1 ≥ (bi∗ − t)r−1 , for all t ∈ [a, ai∗ ]. Then ai∗ (bi − ai∗ )r (ai∗ − t)r−1 − (bi∗ − t)r−1 dt = ∗ Ii∗ = r a (ai∗ − a)r − (bi∗ − a)r (bi − ai∗ )r , (15.269) + ≤ ∗ r r by (ai∗ − a)r − (bi∗ − a)r < 0. Therefore (bi∗ − ai∗ )r r

(15.270)

2(bi∗ − ai∗ )r F ∞ , r

(15.271)

Ii∗ ≤ and Ti∗ ≤ proving that |G(bi∗ ) − G(ai∗ )| ≤

2F ∞ r

(bi∗ − ai∗ )r ,

(15.272)

which proves that G is continuous. Taking the special case of a = 0 and F (t) = 1, for all t ∈ [0, b], we get that G(s) =

sr , all s ∈ [0, b], for 0 < r < 1. r

(15.273)

The last is a Lipschitz function of order r ∈ (0, 1), which is not absolutely continuous. Consequently G for r ∈ (0, 1) in general, cannot be absolutely continuous. That completes the proof.

16 Caputo Fractional Multivariate Opial-Type Inequalities over a Spherical Shell

Here is introduced the concept of the Caputo fractional radial derivative for a function deﬁned on a spherical shell. Using polar coordinates we are able to derive multivariate Opial-type inequalities over a spherical shell of RN , N ≥ 2, by studying the topic in all possibilities. Our results involve one, two, or more functions. We present many univariate Caputo fractional Opial-type inequalities, several of which are used to establish results on the shell. We give an application to prove the uniqueness of solution of a general partial diﬀerential equation on the shell. Also we apply our results for Riemann – Liouville fractional derivatives. This treatment relies on [58].

16.1 Introduction This chapter is inspired by articles of Opial [315], Bessack [80], and Anastassiou, Koliha, and Pecaric [64, 65], and Anastassiou [46, 48]. Opial-type inequalities usually ﬁnd applications in establishing the uniqueness of solution of initial value problems for diﬀerential equations and their systems; see Willett [406]. In this chapter we present a series of various Caputo fractional multivariate Opial type inequalities over spherical shells. To achieve our goal we use polar coordinates, and we introduce and use the Caputo fractional radial derivative. We work on the spherical shell, and not on the ball, because a radial derivative cannot be deﬁned at zero. So, we reduce the problem to a univariate one. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 391 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 16,

392

16. Caputo Fractional Multivariate Opial Inequalities

Therefore we derive and use a large array of univariate Opial-type inequalities involving Caputo fractional derivatives; these are Caputo fractional derivatives deﬁned at arbitrary anchor point a ∈ R. In our results we involve one, two, or several functions. But ﬁrst we need to develop an extensive background in two parts, then follow the main results. At the end we give an application proving the uniqueness of solution for a general PDE initial value problem. Also we re-establish our results by involving Riemann – Liouville fractional derivatives deﬁned at an arbitrary anchor point. In this chapter to build our background regarding the Caputo fractional derivative we use the excellent monograph [134]. The Caputo derivative was introduced in 1967; see [102], and also see [112, 114]. It happens that the Riemann – Liouville fractional derivative has some disadvantages when modeling real-world phenomena with fractional differential equations. One reason is that the initial conditions there involve fractional derivatives that are diﬃcult to connect with actual data, and so on. However, Caputo fractional derivative modeling involves initial conditions that are described by ordinary derivatives, much easier to write based on real-world data. So more and more in recent years the Caputo version is usually preferred when physical models are described, because the physical interpretation of the prescribed data is clear, and therefore it is in general possibly easier to gather these data, for example, by appropriate measurements. Also from the pure mathematics side there are reasons to prefer the Caputo fractional derivative.

16.2 Background—I Here we follow [134]. We start with Deﬁnition 16.1. Let ν ≥ 0; the operator Jaν , deﬁned on L1 [a, b] by x 1 ν−1 Jaν f (x) := (x − t) f (t) dt (16.1) Γ (ν) a for a ≤ x ≤ b, is called the Riemann – Liouville fractional integral operator of order ν. For ν = 0, we set Ja0 := I, the identity operator. Here Γ stands for the gamma function. Theorem 16.2. [134] Let f ∈ L1 [a, b] , ν > 0. Then, the integral Jaν f (x) exists for almost every x ∈ [a, b] .

16.2 Background—I

393

Moreover, Jaν f ∈ L1 ([a, b]). We need Theorem 16.3. [134] Let m, n ≥ 0, Φ ∈ L1 ([a, b]). Then Jam Jan Φ = Jam+n Φ

(16.2)

holds almost everywhere on [a, b] . If additionally Φ ∈ C ([a, b]) or m + n ≥ 1, then the identity holds everywhere on [a, b] . We give Deﬁnition 16.4. [134] Let ν ∈ R+ and m = ν ; · is the ceiling of number. The operator Daν , deﬁned by d , (16.3) dx is called the Riemann – Liouville fractional diﬀerential operator of order ν. For ν = 0, we set Da0 := I, the identity operator. If ν ∈ N then Daν f = f (ν) , the ordinary ν-order derivative. Daν f := Dm Jam−ν f,

D :=

Next we give Deﬁnition 16.5. (p. 37, [134]) Let ν ≥ 0 and n := ν, a ∈ R. Then, we deﬁne the operator ˆ ν f := J n−ν f (n) , D (16.4) a a whenever f (n) ∈ L1 ([a, b]). Also we need Theorem 16.6. (p. 37, [134]) Let ν ≥ 0, n := ν. Moreover assume that f ∈ AC n ([a, b])(the space of functions with absolutely continuous (n − 1)st derivative). Then ˆ aν f = Daν (f − Tn−1 (f ; a)), D

a.e. on [a, b] ,

(16.5)

where Tn−1 (f ; a) (x) :=

n−1 k=0

f (k) (a) k (x − a) , k!

x ∈ [a, b] ,

is the Taylor polynomial of degree n − 1 of f, centered at a. Next we give the deﬁnition of Caputo fractional derivative [134].

(16.6)

394

16. Caputo Fractional Multivariate Opial Inequalities

Deﬁnition 16.7. (p. 38, [134]) Assume that f is such that Daν (f − Tn−1 (f ; a)) (x) exists for some x ∈ [a, b]. Then we deﬁne the Caputo fractional derivative by ν f (x) := Daν (f − Tn−1 (f ; a)) (x). D∗a

(16.7)

So the above deﬁnition applies to all points x ∈ [a, b] : Daν (f − Tn−1 (f ; a)) (x) ∈ R. We have Corollary 16.8. Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]). Then the Caputo fractional derivative x 1 n−ν−1 (n) ν D∗a f (x) = (x − t) f (t) dt (16.8) Γ (n − ν) a exists almost everywhere for x in [a, b]. We have ν Corollary 16.9. Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]). Then, D∗a f ν exists iﬀ Da f exists.

Proof. By linearity of Daν operator and assumption.

We need Lemma 16.10. [134] Let ν ≥ 0, n = ν. Assume that f is such that ν both D∗a f and Daν f exist. Then, ν D∗a f (x) = Daν f (x) −

n−1 k=0

f (k) (a) k−ν (x − a) . Γ (k − ν + 1)

(16.9)

Lemma 16.11. [134] All are as in Lemma 16.10. Additionally assume that f (k) (a) = 0 for k = 0, 1, . . . , n − 1. Then, ν f = Daν f. D∗a

(16.10)

In conclusion ν f exists or Corollary 16.12. Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]), D∗a (k) exists, and f (a) = 0, k = 0, 1, . . . , n − 1. Then

Daν f

ν f. Daν f = D∗a

(16.11)

16.2 Background—I

395

We need the following Taylor – Caputo formula. Theorem 16.13. (p. 40, [134]) Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]). Then n−1 f (k) (a) k ν (x − a) + Jaν D∗a f (x), (16.12) f (x) = k! k=0

∀x ∈ [a, b] . ν f ∈ AC n ([a, b]). Clearly here Jaν D∗a Corollary 16.14. Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]), and f (k) (a) = 0, k = 0, 1, . . . , n − 1. Then x 1 ν−1 ν ν ν f (x) = Ja D∗a f (x) = (x − t) D∗a f (t) dt. (16.13) Γ (ν) a We need Lemma 16.15. Let ν ≥ γ + 1, γ ≥ 0. Call n := ν, m := γ. Then n − m ≥ 1; that is, m ≤ n − 1. Proof. Clearly ν ≥ 1 and ν > γ, ν − γ ≥ 1. By γ + 1 > m we get ν > m, and n > m; that is, ν − m > 0 and n − m > 0. We see that ν ≥ γ + 1 ≥ [γ] + 1, where [·] is the integral part. Thus ν ≥ ([γ] + 1) ∈ N and ν ≥ [ν] ≥ [γ] + 1. Therefore [ν] − [γ] ≥ 1, (16.14) which is used next. We distinguish the following cases. (i) Let ν, γ ∈ / N; then ν = [ν] + 1, γ = [γ] + 1. By (16.14) we get ([ν] + 1) − ([γ] + 1) ≥ 1. Hence n − m ≥ 1. (ii) Let ν, γ ∈ N; then [ν] = ν = ν, [γ] = γ = γ. So by (16.14) n − m ≥ 1. (iii) Let ν ∈ / N, γ ∈ N. Then n = ν = [ν] + 1, m = γ = [γ] = γ. Hence by (16.14) we have (ν − 1) − m ≥ 1, and ν − m ≥ 2 > 1. Hence n − m > 1. (iv) Let ν ∈ N, γ ∈ / N. Then 1+γ < γ+1 = γ + 1, and 1+γ ≤ ν ∈ N by assumption. Therefore γ + 1 ≤ ν, and ν − γ ≥ 1. So that again n − m ≥ 1. The claim is proved in all cases. We present the representation theorem. Theorem 16.16. Let ν ≥ γ + 1, γ ≥ 0. Call n := ν, m := γ. Assume f ∈ AC n ([a, b]), such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and ν f ∈ L∞ (a, b). D∗a

396

16. Caputo Fractional Multivariate Opial Inequalities

Then γ γ D∗a f ∈ C ([a, b]), D∗a f (x) = Jam−γ f (m) (x),

(16.15)

and γ D∗a f (x) =

1 Γ (ν − γ)

x

ν−γ−1

(x − t) a

ν D∗a f (t) dt,

(16.16)

∀x ∈ [a, b] . Proof. If γ = 0 then (16.16) collapses to (16.13) ; also (16.15) is clear. So we assume γ > 0. By Lemma 16.15 we have m ≤ n − 1. By the assumption f ∈ AC n ([a, b]) we get that f ∈ C n−1 ([a, b]) and thus f ∈ C m ([a, b]). γ f = Jam−γ f (m) ∈ C ([a, b]) By Lemma 3.7, p. 41 of [134] we get that D∗a γ / N. Clearly the last statement is true also when and D∗a f (a) = 0, for γ ∈ γ ∈ N, thus proving (16.15) and the ﬁrst claim. Recall that we have ν −m > 0, and ν −1 > 0 by γ > 0. Using Γ (p + 1) = pΓ (p), p > 0, (16.13), and Theorem 7 of [48], we obtain ν f (x), f (m) (x) = Jaν−m D∗a

∀x ∈ [a, b] .

(16.17)

Therefore we get γ D∗a f (x) = Jam−γ f (m) (x)

(16.17)

=

ν Jam−γ Jaν−m D∗a f (x)

(by Theorem 2.2, p. 14 of [134], and ν − γ ≥ 1) ∀x ∈ [a, b] ,

ν f (x), = Jaν−γ D∗a

thus proving (16.16).

We also give the representation theorem. Theorem 16.17. Let ν ≥ γ + 1, γ ≥ 0. Call n := ν, m := γ. Let f ∈ AC n ([a, b]), such that f (k) (a) = 0, k = 0, 1, . . . , n − 1. Assume there exists Daν f (x) ∈ R, ∀x ∈ [a, b] , and Daν f ∈ L∞ (a, b). Then Daγ f ∈ C ([a, b]), Daγ f (x) = Jam−γ f (m) (x),

(16.18)

∀x ∈ [a, b] , Daγ f (x) =

1 Γ (ν − γ)

x

ν−γ−1

(x − t)

Daν f (t) dt,

(16.19)

a

∀x ∈ [a, b] . ν f (x) ∈ R, and Proof. By Corollaries 16.9 and 16.12 we get existing D∗a ν ν ν that D∗a f (x) = Da f (x), ∀x ∈ [a, b] . That is, D∗a f ∈ L∞ (a, b) and by (16.16) we have x 1 ν−γ−1 γ f (x) = (x − t) Daν f (t) dt, ∀x ∈ [a, b] . (16.20) D∗a Γ (ν − γ) a

16.3 Main Results

397

γ γ Because D∗a f ∈ C ([a, b]) we get D∗a f (x) ∈ R, ∀x ∈ [a, b] . And because f ∈ C m ([a, b]) then f ∈ AC m ([a, b]). By Corollary 16.9 Daγ f exists. Also f (k) (a) = 0, for k = 0, 1, . . . , m − 1. Thus by Corollary 16.12 we obγ f (x), ∀x ∈ [a, b] . Now by (16.20) we have established tain Daγ f (x) = D∗a (16.19).

16.3 Main Results 16.3.1 Results Involving One Function We present the following theorem. Theorem 16.18. Let ν ≥ γ + 1, γ ≥ 0. Call n := ν and assume ν f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and D∗a f ∈ L∞ (a, b). Let p, q > 1 such that 1/p + 1/q = 1, a ≤ x ≤ b. Then x a

γ ν |D∗a f (ω)| |(D∗a f ) (ω)| dω ≤ (pν−pγ−p+2)/p

(x − a) √ 1/p q 2 Γ (ν − γ) ((pν − pγ − p + 1) (pν − pγ − p + 2)) x 2/q q ν · |D∗a f (ω)| dω .

(16.21)

a

Proof. Similar to Theorem 25.2, p. 545, [19], and Theorem 2.1 of [64]. A related extreme case comes next. Proposition 16.19. All are as in Theorem 16.18, but with p = 1 and q = ∞, we ﬁnd x γ ν |D∗a f (ω)| |D∗a f (ω)| dω ≤ a

ν−γ+1 2 (x − a) ν D∗a f ∞,(a,x) . Γ (ν − γ + 2)

Proof. Similar to Proposition 25.1, p. 547, [19]. The converse of (16.21) follows.

(16.22)

398

16. Caputo Fractional Multivariate Opial Inequalities

Theorem 16.20. Let ν ≥ γ + 1, γ ≥ 0. Call n := ν and assume ν f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and D∗a f, ν ν f ∈ L∞ (a, b). Suppose that D∗a f is of ﬁxed sign a.e. in [a, b] . Let 1/D∗a p, q be such that 0 < p < 1, q < 0 and 1/p + 1/q = 1, a ≤ x ≤ b. Then x γ ν |D∗a f (ω)| |D∗a f (ω)| dω ≥ a

(pν−pγ−p+2)/p

(x − a) √ 1/p q 2 Γ (ν − γ) ((pν − pγ − p + 1) (pν − pγ − p + 2)) x 2/q q ν · |D∗a f (ω)| dω .

(16.23)

a

Proof. Similar to Theorem 25.3, p. 547, [19], and Theorem 2.3 of [64]. We give Theorem 16.21. Let ν ≥ 2, k ≥ 0, ν ≥ k + 2. Call n := ν and assume f ∈ AC n ([a, b]) such that f (j) (a) = 0, j = 0, 1, . . . , n − 1, and ν f ∈ L∞ (a, b). Let p, q > 1 such that 1/p + 1/q = 1, a ≤ x ≤ b. Then D∗a x k+1 k D∗a f (ω) D∗a f (ω) dω ≤ a

(x − a)

2(pν−pk−p+1)/p

2

2 (Γ (ν − k)) (pν − pk − p + 1) x 2/q q ν · |D∗a f (ω)| dω .

2/p

(16.24)

a

Proof. Similar to Theorem 25.4, p. 549, [19], and Theorem 2.4 of [64]. The extreme case follows. Proposition 16.22. Under the assumptions of Theorem 16.21 when p = 1, q = ∞ we ﬁnd x k+1 k D∗a f (ω) D∗a f (ω) dω ≤ a

2(ν−k)

(x − a)

ν D∗a f ∞,(a,x) 2

2 (Γ (ν − k + 1))

2 .

(16.25)

16.3 Main Results

399

Proof. Similar to Proposition 25.2, p. 551 of [19]. We give the related converse result. Theorem 16.23. Let ν ≥ 2, k ≥ 0, ν ≥ k+2. Call n := ν. Assume f ∈ ν ν AC n ([a, b]) such that f (j) (a) = 0, j = 0, 1, . . . , n − 1, and D∗a f, 1/D∗a f∈ ν L∞ (a, b). Suppose that D∗a f is of ﬁxed sign a.e. in [a, b] . Let p, q be such that 0 < p < 1, q < 0 and 1/p + 1/q = 1, a ≤ x ≤ b. Then x k+1 k D∗a f (ω) D∗a f (ω) dω ≥ a

(x − a)

2(pν−pk−p+1)/p

2

2 (Γ (ν − k)) (pν − pk − p + 1) x 2/q q ν · |D∗a f (ω)| dω .

2/p

(16.26)

a

Proof. Similar to Theorem 25.5, p. 553 of [19]. Next we present Theorem 16.24. Let γ i ≥ 0, ν ≥ 1, ν − γ i ≥ 1; i = 1, . . . , l, n := ν, and f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and ν f ∈ L∞ (a, b). Here a ≤ x ≤ b; q1 (x), q2 (x) continuous functions on D∗a "l [a, b] such that q1 (x) ≥ 0, q2 (x) > 0 on [a, b] , and ri > 0 : i=1 ri = r. Let s1 , s1 > 1 : 1/s1 + 1/s1 = 1 and s2 , s2 > 1 : 1/s2 + 1/s2 = 1, and p > s2 . Denote by x 1/s1 s Q1 (x) := (q1 (ω)) 1 dω (16.27) a

and

Q2 (x) :=

x

(q2 (ω))

−s2 /p

r/s2 dω

,

(16.28)

a

σ :=

p − s2 . ps2

(16.29)

Then

x

q1 (ω) a

l γ D∗ai f (ω)ri dω ≤ Q1 (x) Q2 (x) i=1

l i=1

σ ri σ r r σ (Γ (ν − γ i )) i (ν − γ i − 1 + σ) i

400

16. Caputo Fractional Multivariate Opial Inequalities "l

·

(x − a)(

i=1 (ν−γ i −1)ri +σr+1/s1

)

1/s1 + rs (ν − γ − 1) r s σ + 1 i 1 1 i i=1

"l

· a

r/p

x ν q2 (ω) |D∗a f

p

(ω)| dω

.

(16.30)

Proof. Similar to Theorem 26.1, p. 567 of [19], and Theorem 2.1 of [65]. The counterpart of the last theorem follows. Theorem 16.25. Let γ i ≥ 0, ν ≥ 1, ν − γ i ≥ 1; i = 1, . . . , l, n := ν, and f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and ν ν D∗a f, 1/D∗a f ∈ L∞ (a, b). Here a ≤ x ≤ b; q1 (x), q2 (x) > 0 are continuous "l functions on [a, b] and ri > 0 : i=1 ri = r. Let 0 < s1 , s2 < 1, and s1 , ν f (t) s2 < 0 such that 1/s1 + 1/s1 = 1, 1/s2 + 1/s2 = 1. Assume that D∗a is of ﬁxed sign a.e. in [a, b] . Denote

x

Q1 (x) :=

1/s1

s1

(q1 (ω)) dω

,

(16.31)

a

x

Q2 (x) :=

(q2 (ω))

−s2

r/s2 dω

.

(16.32)

a

Set λ := Then

x

s1 s2 . s1 s2 − 1

(16.33)

l γ D∗ai f (ω)ri q1 (ω)

a

dω ≥

i=1

Q1 (x) Q2 (x)

2 ri 2 s + 1)(ri /s2 s1 ) (Γ (ν − γ )) ((ν − γ − 1) s 1 i i 2 i=1

&l

"

l r (ν−γ i −1)s1 +s−2 2 ))+1}/s1 (x − a){( i=1 i ( · 1/s1 "l −2 (ν − γ r − 1) s + s + 1 i 1 i 2 i=1

· a

x

r/λs2 λs2

ν q2λs2 (ω) |D∗a f (ω)|

dω

.

(16.34)

Proof. Similar to Theorem 26.2, p. 570 of [19], and Theorem 2.3 of [65].

16.3 Main Results

401

A related extreme case comes next for p = 1 and q = ∞. Theorem 16.26. Let ν ≥ γ i + 1, γ i ≥ 0; i = 1, . . . , l. Call n := ν and assume f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, ν f ∈ L∞ (a, b). Here a ≤ x ≤ b, with 0 ≤ q˜ (ω) ∈ L∞ (a, b) and and D∗a "l ri > 0 : i=1 ri = r. Then a

x

l γ D∗ai f (ω)ri dω ≤ q˜ (ω) i=1

r ⎫ ⎧ ν ⎨ ˜ f ∞,(a,x) ⎬ q ∞,(a,x) D∗a . &l ri ⎭ ⎩ i=1 (Γ (ν − γ i + 1)) ⎫ ⎧ " ⎨ (x − a)rν− li=1 ri γ i +1 ⎬ . ⎩ rν − "l r γ + 1 ⎭

(16.35)

i=1 i i

Proof. Similar to Proposition 26.1 of [19], p. 573, and Theorem 2.2 of [65]. We continue with the interesting Theorem 16.27. Let k ≥ 0, γ ≥ 1, ν ≥ 2, n := ν, such that ν − γ ≥ 1, γ − k ≥ 1, and f ∈ AC n ([a, b]) such that f (j) (a) = 0, j = 0, 1, . . . , n − 1, ν f ∈ L∞ (a, b). Here a ≤ x ≤ b, p, q > 1 : 1/p + 1/q = 1. Then and D∗a x k γ |D∗a f (ω)| D∗a f (ω) dω ≤ a

2−1/p (x − a)

(2ν−k−γ−1+2/q)

Γ (ν − k) Γ (ν − γ + 1) ((ν − γ) q + 1) x ν 2/p p |D∗a f (ω)| dω a · . 1/q (2νq − kq − γq − q + 2)

1/q

(16.36)

Proof. Similar to Theorem 26.3, p. 574 of [19], and Theorem 2.5 of [65]. We give Theorem 16.28. Let ν ≥ γ i + 1, γ i ≥ 0, i = 1, . . . , k ∈ N − {1} , n := ν. Assume f ∈ AC n ([a, b]) such that f (j) (a) = 0, j = 0, 1, . . . , n−1, "k ν and D∗a f ∈ L∞ (a, b). Here a ≤ x ≤ b, γ := i=1 γ i . Let p, q > 1 such

402

16. Caputo Fractional Multivariate Opial Inequalities

ν that 1/p + 1/q = 1. Furthermore, suppose that |D∗a f (t)| is decreasing on [a, x] . Then x k γ D∗ai f (ω) dω ≤ a i=1

&k i=1

p (x − a)

(1+kνp−γp/p)

Γ (ν − γ i ) (kνp − γp − kp + 1) 1/q x kq ν |D∗a f (t)| dt a . · (kνp − γp − kp + p + 1)

1/p

(16.37)

Proof. Similar to Theorem 26.6, p. 581 of [19], and Theorem 2.6 of [65]. The extreme case follows. Theorem 16.29. All are as in Theorem 16.28, but p = 1, q = ∞. Then

k x

γi D∗a f (ω) dω ≤

a i=1

· & k i=1

(x − a)

kν−γ+1

ν D∗a f ∞,(a,x)

k

. Γ (ν − γ i ) (kν − γ − k + 1) (kν − γ + 1)

(16.38)

Proof. Similar to Proposition 26.4, p. 582 of [19], and Theorem 2.7 of [65].

16.3.2 Results Involving Two Functions We present the following theorem. Theorem 16.30. Let ν ≥ γ i + 1, γ i ≥ 0, i = 1, 2, n := ν, and f1 , (j) (j) f2 ∈ AC n ([a, b]) such that f1 (a) = f2 (a) = 0, j = 0, 1, . . . , n − 1, a ≤ x ≤ b. Consider also p (t) > 0 and q (t) ≥ 0, with all p (t), 1/p (t), ν fi ∈ L∞ (a, b), i = 1, 2. q (t) ∈ L∞ (a, b). Further assume D∗a Let λν > 0 and λα , λβ ≥ 0 such that λν < p, where p > 1. Set ω (ν−γ k −1)p/(p−1) −1/(p−1) Pk (ω) := (ω − t) (p (t)) dt, (16.39) a

16.3 Main Results

403

k = 1, 2, a ≤ x ≤ b; q (ω) (P1 (ω))

A (ω) :=

λα ((p−1)/p)

λβ ((p−1)/p)

(P2 (ω)) λα

(Γ (ν − γ 1 ))

(Γ (ν − γ 2 ))

−λν /p

(p (ω))

,

λβ

(16.40)

(p−λν )/p

x

p/p−λν

A0 (x) :=

(A (ω))

dω

,

(16.41)

a

and δ 1 :=

21−(λα +λν /p) , if λα + λν ≤ p, 1, if λα + λν ≥ p.

(16.42)

If λβ = 0, we obtain that x + λα ν γ λ q (ω) D∗a1 f1 (ω) |D∗a f1 (ω)| ν + a

, γ λ ν D∗a1 f2 (ω)λα |D∗a f2 (ω)| ν dω ≤ A0 (x) |λβ =0 )

λν λα + λν

λν /p δ1 *((λα +λν )/p)

x ν p (ω) [|D∗a f1

a

(16.43)

p

(ω)| +

ν |D∗a f2

p

(ω)| ] dω

.

Proof. Similar to Theorem 2 of [26] and Theorem 4 of [48]. The counterpart of the last theorem follows. Theorem 16.31. All here are as in Theorem 16.30. Denote λ /λ 2 β ν − 1, if λβ ≥ λν , δ 3 := 1, if λβ ≤ λν . If λα = 0, then a

x

(16.44)

+ λ γ λ ν q (ω) D∗a2 f2 (ω) β |D∗a f1 (ω)| ν +

, γ λ ν D∗a2 f1 (ω)λβ |D∗a f2 (ω)| ν dω ≤ (A0 (x) |λα =0 ) 2

p−λν /p

x

a

all a ≤ x ≤ b.

p

λν λβ + λν

λν /p

p

ν ν p (ω) [|D∗a f1 (ω)| + |D∗a f2 (ω)| ] dω

(16.45) λ /p

δ3 ν

((λν +λβ )/p) ,

404

16. Caputo Fractional Multivariate Opial Inequalities

Proof. Similar to Theorem 3 of [26] and Theorem 5 of [48]. The complete case λα , λβ = 0 follows. Theorem 16.32. All here are as in Theorem 16.30. Denote ((λ +λ )/λ ) 2 α β ν − 1, if λα + λβ ≥ λν , γ˜ 1 := 1, if λα + λβ ≤ λν ,

(16.46)

and γ˜ 2 :=

Then

x

a

1, if λα + λβ + λν ≥ p, 21−(λα +λβ +λν /p) , if λα + λβ + λν ≤ p.

(16.47)

+ λα γ λ γ λ ν q (ω) D∗a1 f1 (ω) D∗a2 f2 (ω) β |D∗a f1 (ω)| ν +

, γ λα ν γ λ 1 D∗a2 f1 (ω)λβ D∗a f2 (ω) |D∗a f2 (ω)| ν dω ≤ A0 (x)

λν (λα + λβ ) (λα + λβ + λν )

· a

x ν p (ω) (|D∗a f1

p

(ω)| +

λν /p +

ν |D∗a f2

(16.48) λν /p

λλαν /p γ˜ 2 + 2p−λν /p (˜ γ 1 λβ )

,

((λα +λβ +λν )/p)

p

(ω)| ) dω

,

all a ≤ x ≤ b. Proof. As for Theorem 4 of [26], and Theorem 6 of [48]. We continue with a special important case. Theorem 16.33. Let ν ≥ γ 1 + 2, γ 1 ≥ 0, n := ν, and f1 , f2 ∈ (j) (j) AC n ([a, b]) such that f1 (a) = f2 (a) = 0, j = 0, 1, . . . , n − 1, a ≤ x ≤ b. Consider also p (t) > 0 and q (t) ≥ 0, with all p (t), 1/p (t), q (t) ∈ L∞ (a, b). ν fi ∈ L∞ (a, b), i = 1, 2. Furthermore assume D∗a Let λα ≥ 0, 0 < λα+1 < 1, and p > 1. Denote λ /λ 2 α α+1 − 1, if λα ≥ λα+1 , θ3 := , (16.49) 1, if λα ≤ λα+1 , L (x) := 2

(1−λα+1 )

x

(1/1−λα+1 )

(q (ω))

dω

a

θ3 λα+1 λα + λα+1

λα+1 , (16.50)

and

x

(ν−γ 1 −1)p/(p−1)

(x − t)

P1 (x) := a

−1/(p−1)

(p (t))

dt,

(16.51)

16.3 Main Results

T (x) := L (x)

((p−1)/p)

P1 (x) Γ (ν − γ 1 )

405

(λα +λα+1 ) ,

(16.52)

ω 1 := 2(p−1/p)(λα +λα+1 ) ,

(16.53)

Φ (x) := T (x) ω 1 .

(16.54)

and

Then

a

x

) λα+1 γ λα γ +1 q (ω) D∗a1 f1 (ω) D∗a1 f2 (ω) +

λα+1 * γ γ +1 1 D∗a1 f2 (ω)λα D∗a f1 (ω) dω ≤ ) Φ (x) a

x

p

ν p (ω) (|D∗a f1 (ω)| + ((λα +λα+1 )/p)

p

ν |D∗a f2 (ω)| ) dω]

,

(16.55)

all a ≤ x ≤ b. Proof. As in Theorem 5 of [26], and Theorem 8 of [48]. We give Corollary 16.34. All here are as in Theorem 16.30, with λβ = 0, p (t) = q (t) = 1. Then x+ γ λ ν D∗a1 f1 (ω)λα |D∗a f1 (ω)| ν + a

, γ λ ν D∗a1 f2 (ω)λα |D∗a f2 (ω)| ν dω ≤

((λα +λν )/p)

x

C1 (x) a

(16.56)

p

p

ν ν (|D∗a f1 (ω)| + |D∗a f2 (ω)| ) dω

,

all a ≤ x ≤ b, where C1 (x) := A0 (x) |λβ =0 δ 1 :=

21−((λα +λν )/p) , 1,

λν λα + λν

λν /p δ1 ,

if λα + λν ≤ p, if λα + λν ≥ p.

(16.57)

(16.58)

We ﬁnd that A0 (x) |λβ =0 =

%

(p − 1) (Γ (ν − γ 1 ))

λα

((λα p−λα )/p)

(νp − γ 1 p − 1)

((λα p−λα )/p)

·

406

16. Caputo Fractional Multivariate Opial Inequalities

(p − λν )

-

((p−λν )/p)

·

((p−λν )/p)

(λα νp − λα γ 1 p − λα + p − λν ) (x − a)

((λα νp−λα γ 1 p−λα +p−λν )/p)

.

(16.59)

Proof. As for Corollary 1 of [26], and Corollary 10 of [48]. Corollary 16.35. (All are as in Theorem 16.30; λβ = 0, p (t) = q (t) = 1, λα = λν = 1, p = 2.) In detail: let ν ≥ γ 1 + 1, γ 1 ≥ 0, n := ν, f1 , (j) (j) f2 ∈ AC n ([a, b]) : f1 (a) = f2 (a) = 0, j = 0, 1, . . . , n − 1, a ≤ x ≤ b, ν fi ∈ L∞ (a, b), i = 1, 2. Then D∗a x γ 1 ν D∗a f1 (ω) |(D∗a f1 ) (ω)| + a

γ

ν D∗a1 f2 (ω) |(D∗a f2 ) (ω)| dω ≤

(16.60)

(ν−γ )

1 (x − a) √ √ 2Γ (ν − γ 1 ) ν − γ 1 2ν − 2γ 1 − 1

+

x

a

2

ν ν ((D∗a f1 ) (ω)) + ((D∗a f2 ) (ω))

2

,

dω ,

all a ≤ x ≤ b. Corollary 16.36. (To Theorem 16.31; λα = 0, p (t) = q (t) = 1.) It holds x+ γ λ ν D∗a2 f2 (ω)λβ |D∗a f1 (ω)| ν + a

, γ λ ν D∗a2 f1 (ω)λβ |D∗a f2 (ω)| ν dω ≤

x

C2 (x) a

p

(16.61)

((λν +λβ )/p)

p

ν ν [|D∗a f1 (ω)| + |D∗a f2 (ω)| ] dω

,

all a ≤ x ≤ b, where C2 (x) := (A0 (x) |λα =0 ) 2

p−λν /p

δ 3 :=

2λβ /λν − 1, 1,

λν λβ + λν

λν /p

if λβ ≥ λν , if λβ ≤ λν

λ /p

(16.62)

.

(16.63)

δ3 ν ,

We ﬁnd that % (A0 (x) |λα =0 ) =

(p − 1) (Γ (ν − γ 2 ))

λβ

((λβ p−λβ )/p)

(νp − γ 2 p − 1)

((λβ p−λβ )/p)

·

16.3 Main Results

(p − λν ) (x − a)

-

((p−λν )/p)

(λβ νp − λβ γ 2 p − λβ + p − λν )

407

·

((p−λν )/p)

((λβ νp−λβ γ 2 p−λβ +p−λν )/p)

.

(16.64)

Corollary 16.37. (To Theorem 16.31; λα = 0, p (t) = q (t) = 1, λβ = λν = 1, p = 2.) In detail: let ν ≥ γ 2 + 1, γ 2 ≥ 0, n := ν, f1 , f2 ∈ (j) (j) AC n ([a, b]) : f1 (a) = f2 (a) = 0, j = 0, 1, . . . , n − 1, a ≤ x ≤ b, ν D∗a fi ∈ L∞ (a, b), i = 1, 2. Then x ν γ 2 D∗a f2 (ω) |D∗a f1 (ω)| + a

ν γ

D∗a2 f1 (ω) |D∗a f2 (ω)| dω ≤ (ν−γ 2 ) (x − a) √ . √ √ 2Γ (ν − γ 2 ) ν − γ 2 2ν − 2γ 2 − 1 x 2 2 ν ν (D∗a f1 (ω)) + (D∗a f2 (ω)) dω ,

(16.65)

a

all a ≤ x ≤ b. We continue with related results regarding ·∞ . Theorem 16.38. Let ν ≥ γ i + 1, γ i ≥ 0, i = 1, 2, n := ν, and f1 , f2 ∈ (j) (j) AC n ([a, b]) such that f1 (a) = f2 (a) = 0, j = 0, 1, . . . , n − 1; a ≤ x ≤ b. ν fi ∈ L∞ (a, b), Consider p (x) ≥ 0 and p (x) ∈ L∞ (a, b), and assume D∗a i = 1, 2. Let λα , λβ , λν ≥ 0. Set (νλ −γ λ +νλ −γ λ +1)

T (x) :=

· Then

a

x

β 2 β (x − a) α 1 α (νλα − γ 1 λα + νλβ − γ 2 λβ + 1)

p (s)∞,(a,x) λα

(Γ (ν − γ 1 + 1))

λβ

(Γ (ν − γ 2 + 1))

.

+ λα γ λ γ λ ν p (ω) D∗a1 f1 (ω) D∗a2 f2 (ω) β |D∗a f1 (ω)| ν +

, γ λα ν γ λ 1 D∗a2 f1 (ω)λβ D∗a f2 (ω) |D∗a f2 (ω)| ν dω ≤ T (x) + ν 2λβ 2(λα +λν ) ν + D∗a f1 ∞,(a,x) + D∗a f1 ∞,(a,x) 2

(16.66)

408

16. Caputo Fractional Multivariate Opial Inequalities

2λ

2(λ +λ )

β α ν ν ν D∗a f2 ∞,(a,x) + D∗a f2 ∞,(a,x)

, ,

(16.67)

all a ≤ x ≤ b. Proof. Similar to Theorem 7 of [26], and Theorem 18 of [48]. We give Corollary 16.39. (to Theorem 16.38) Let ν ≥ γ 1 + 2, γ 1 ≥ 0, n := ν, (j) (j) and f1 , f2 ∈ AC n ([a, b]) such that f1 (a) = f2 (a) = 0, j = 0, 1, . . . , n−1; ν D∗a fi ∈ L∞ (a, b), i = 1, 2. Then x+ γ +1 γ 1 D∗a1 f1 (ω) D∗a f2 (ω) + a

, γ 1 +1 γ D∗a f1 (ω) D∗a1 f2 (ω) dω ≤

+

(x − a)

2(ν−γ 1 ) 2

2 (Γ (ν − γ 1 + 1))

, 2 2 ν ν D∗a f1 ∞,(a,x) + D∗a f2 ∞,(a,x) ,

(16.68)

all a ≤ x ≤ b. Next we give converse results involving two functions. Theorem 16.40. Let γ j ≥ 0, 1 ≤ ν − γ j < 1/p, 0 0 and q (t) ≥ 0, with all ν fi ∈ L∞ (a, b), p (t), 1/p (t), q (t), 1/q (t) ∈ L∞ (a, b). Further assume D∗a i = 1, 2, each of which is of ﬁxed sign a.e. on [a, b] . Let λν > 0 and λα , λβ ≥ 0 such that λν > p. Here Pk (ω), A (ω), A0 (x) are as in (16.39), (16.40), and (16.41), respectively. Set (16.69) δ 1 := 21−((λα +λν )/p) . If λβ = 0, then a

x

+ λα ν γ λ q (ω) D∗a1 f1 (ω) |D∗a f1 (ω)| ν +

, γ λ ν D∗a1 f2 (ω)λα |D∗a f2 (ω)| ν dω ≥ A0 (x) |λβ =0

λν λα + λν

λν /p δ1

16.3 Main Results

)

*(λα +λν /p)

x

a

409

p

p

ν ν p (ω) [|D∗a f1 (ω)| + |D∗a f2 (ω)| ] dω

.

(16.70)

Proof. Similar to Theorem 5 of [51], and Theorem 4 of [47]. We continue with Theorem 16.41. All here are as in Theorem 16.40. Further assume λβ ≥ λν . Denote δ 2 := 21−(λβ /λν ) , δ 3 := (δ 2 − 1) 2−(λβ /λν ) . If λα = 0, then a

x

+ λ γ λ ν q (ω) D∗a2 f2 (ω) β |D∗a f1 (ω)| ν +

, γ λ ν D∗a2 f1 (ω)λβ |D∗a f2 (ω)| ν dω ≥

(A0 (x) |λα =0 ) 2p−λν /p · a

x

(16.71)

p

λν λβ + λν p

λν /p

ν ν p (ω) [|D∗a f1 (ω)| + |D∗a f2 (ω)| ] dω

λ /p

δ3 ν

((λν +λβ )/p) ,

(16.72)

all a ≤ x ≤ b. Proof. Similar to Theorem 6 of [51], and Theorem 5 of [47]. We give Theorem 16.42. Let ν ≥ 2 and γ 1 ≥ 0 such that 2 ≤ ν − γ 1 < 1/p, (l) 0 < p < 1, n := ν. Consider f1 , f2 ∈ AC n ([a, b]) such that f1 (a) = (l) ν fi ∈ L∞ (a, b), f2 (a) = 0, l = 0, 1, . . . , n − 1, a ≤ x ≤ b. Assume that D∗a i = 1, 2, each of which is of ﬁxed sign a.e. on [a, b] . Consider also p (t) > 0 and q (t) ≥ 0, with all p (t), 1/p (t), q (t), 1/q (t) ∈ L∞ (a, b). Let λα ≥ λα+1 > 1. Denote θ3 := 21−(λα /λα+1 ) − 1 2−λα /λα+1 , (16.73) L (x) is as in (16.50), P1 (x) as in (16.51), T (x) as in (16.52), ω 1 as in (16.53), and Φ as in (16.54). Then ) x λα+1 γ λα γ +1 q (ω) D∗a1 f1 (ω) D∗a1 f2 (ω) + a

410

16. Caputo Fractional Multivariate Opial Inequalities

λα+1 * γ +1 γ 1 D∗a1 f2 (ω)λα D∗a f1 (ω) dω ≥ ) Φ (x) a

*((λα +λα+1 )/p)

x ν p (ω) (|D∗a f1

p

(ω)| +

ν |D∗a f2

p

(ω)| ) dω

,

(16.74)

all a ≤ x ≤ b. Proof. Similar to Theorem 7 of [51], and Theorem 7 of [47]. We have Corollary 16.43. (To Theorem 16.40; λβ = 0, p (t) = q (t) = 1.) Then x+ γ λ ν D∗a1 f1 (ω)λα |D∗a f1 (ω)| ν + a

, γ λ ν D∗a1 f2 (ω)λα |D∗a f2 (ω)| ν dω ≥

(16.75)

((λα +λν )/p)

x ν (|D∗a f1

C1 (x) a

p

(ω)| +

ν |D∗a f2

p

(ω)| ) dω

,

all a ≤ x ≤ b, where C1 (x) := A0 (x) |λβ =0

λν λα + λν

λν /p δ1 ,

(16.76)

δ 1 := 21−(λα +λν /p) .

(16.77)

Here A0 (x) |λβ =0 is given by (16.59).

Corollary 16.44. (To Theorem 16.41; λα = 0, p (t) = q (t) = 1, λβ ≥ λν .) Then x+ γ λ ν D∗a2 f2 (ω)λβ |D∗a f1 (ω)| ν + a

, γ λ ν D∗a2 f1 (ω)λβ |D∗a f2 (ω)| ν dω ≥

x

C2 (x) a

p

p

(16.78)

((λν +λβ )/p)

ν ν [|D∗a f1 (ω)| + |D∗a f2 (ω)| ] dω

,

all a ≤ x ≤ b, where C2 (x) := (A0 (x) |λα =0 ) 2

p−λν /p

Here (A0 (x) |λα =0 ) is given by (16.64).

λν λβ + λν

λν /p

λ /p

δ3 ν .

16.3 Main Results

411

16.3.3 Results Involving Several Functions We present Theorem 16.45. Here all notations, terms, and assumptions are as in Theorem 16.30, but for fj ∈ AC n ([a, b]), with j = 1, . . . , M ∈ N. Instead of δ 1 there, we deﬁne here M 1−(λα +λν /p) , if λα + λν ≤ p, (16.79) δ ∗1 := 2(λα +λν /p)−1 , if λα + λν ≥ p. Call

ϕ1 (x) := A0 (x) |λβ =0

λν λα + λν

λν /p .

(16.80)

If λβ = 0, then

x

a

⎞ ⎛ M λα ν γ D∗a1 fj (ω) |D∗a fj (ω)|λν ⎠ dω q (ω) ⎝ j=1

⎡ ≤ δ ∗1 ϕ1 (x) ⎣

⎛ x

p (ω) ⎝

a

M

⎞

⎤((λα +λν )/p)

p ν |D∗a fj (ω)| ⎠ dω ⎦

,

(16.81)

j=1

all a ≤ x ≤ b. Proof. As in Theorem 2 of [27], and Theorem 4 of [46]. We continue with Theorem 16.46. All here are as in Theorem 16.45. Denote λ /λ 2 β ν − 1, if λβ ≥ λν , δ 3 := 1, if λβ ≤ λν , 1, if λν + λβ ≥ p, ε2 := M 1−(λν +λβ /p) if λν + λβ ≤ p,

(16.82) (16.83)

and ϕ2 (x) := (A0 (x) |λα =0 ) 2

(p−λν /p)

If λα = 0, then a

x

λν λβ + λν

λν /p

λ /p

δ3 ν .

⎧⎧ −1 + ⎨⎨M γ λ ν D∗a2 fj+1 (ω)λβ |D∗a q (ω) fj (ω)| ν + ⎩⎩ j=1

(16.84)

412

16. Caputo Fractional Multivariate Opial Inequalities

, γ λ ν D∗a2 fj (ω)λβ |D∗a fj+1 (ω)| ν + + λ γ2 λ ν D∗a fM (ω) β |D∗a f1 (ω)| ν + , γ λ ν D∗a2 f1 (ω)λβ |D∗a fM (ω)| ν dω x ≤ 2(λν +λβ /p) ε2 ϕ2 (x) · p (ω) ⎡ ·⎣

M

⎤ p ν |D∗a fj (ω)| ⎦ dω

j=1

a

⎫((λν +λβ )/p) ⎬ ⎭

,

(16.85)

all a ≤ x ≤ b. Proof. As for Theorem 3 of [27], and Theorem 5 of [46]. We give the general case. Theorem 16.47. All are as in Theorem 16.45. Denote (λ +λ /λ ) 2 α β ν − 1, if λα + λβ ≥ λν , γ˜ 1 := 1, if λα + λβ ≤ λν ,

(16.86)

and γ˜ 2 :=

1, if λα + λβ + λν ≥ p, 21−(λα +λβ +λν /p) , if λα + λβ + λν ≤ p.

Set

λν (λα + λβ ) (λα + λβ + λν ) , + λ /p , · λλαν /p γ˜ 2 + 2(p−λν /p) (˜ γ 1 λβ ) ν

(16.87)

λν /p

ϕ3 (x) := A0 (x) ·

and

ε3 :=

1, M 1−(λα +λβ +λν /p)

if λα + λβ + λν ≥ p, if λα + λβ + λν ≤ p.

(16.88)

(16.89)

Then a

⎡ x

M −1 +

q (ω) ⎣

γ λ γ λ ν 2 D∗a1 fj (ω)λα D∗a fj+1 (ω) β |(D∗a fj ) (ω)| ν

j=1

, λ γ λα γ λ ν + D∗a2 fj (ω) β D∗a1 fj+1 (ω) |(D∗a fj+1 ) (ω)| ν

16.3 Main Results

413

+ λα γ λ γ λ ν + D∗a1 f1 (ω) D∗a2 fM (ω) β |(D∗a f1 ) (ω)| ν ,, λ γ λ α γ λ ν + D∗a2 f1 (ω) β D∗a1 fM (ω) |(D∗a fM ) (ω)| ν dω x (λα +λβ +λν /p) ≤2 ε3 ϕ3 (x) · p (ω) ⎡ ·⎣

M

⎤ p ν |(D∗a fj ) (ω)| ⎦ dω

j=1

a

⎫((λα +λβ +λν )/p) ⎬ ,

⎭

(16.90)

all a ≤ x ≤ b. Proof. As for Theorem 4 of [27], and Theorem 6 of [46]. We give Theorem 16.48. All here are as in Theorem 16.33, but for fj ∈ AC n ([a, b]), j = 1, . . . , M ∈ N. Also put 1, if λα + λα+1 ≥ p, . (16.91) ε4 := M 1−(λα +λα+1 /p) if λα + λα+1 ≤ p. Then a

x

⎧⎧ −1 ) λα+1 ⎨⎨M γ +1 γ 1 D∗a1 fj (ω)λα D∗a q (ω) fj+1 (ω) ⎩⎩ j=1

λα+1 * λα γ +1 γ + D∗a1 fj+1 (ω) D∗a1 fj (ω) ) λα+1 γ λα γ +1 + D∗a1 f1 (ω) D∗a1 fM (ω) λα+1 * λα γ +1 γ + D∗a1 fM (ω) D∗a1 f1 (ω) dω ) ≤2

(λα +λα+1 /p)

⎛ ·⎝

M

ε4 Φ (x) · ⎞

x

p (ω) a

⎤((λα +λα+1 )/p)

p ν |(D∗a fj ) (ω)| ⎠ dω ⎦

,

j=1

all a ≤ x ≤ b. Proof. As for Theorem 5 of [27], and Theorem 7 of [46].

(16.92)

414

16. Caputo Fractional Multivariate Opial Inequalities

We continue with Corollary 16.49. (To Theorem 16.45; λβ = 0, p (t) = q (t) = 1, λα = λν = 1, p = 2 .) In detail: let ν ≥ γ 1 +1, γ 1 ≥ 0, n := ν, fj ∈ AC n ([a, b]), ν fj ∈ L∞ (a, b), j = 1, . . . , M. j = 1, . . . , M ∈ N; a ≤ x ≤ b, and D∗a (l) Here fj (a) = 0, l = 0, 1, . . . , n − 1; j = 1, . . . , M. Then ⎛ ⎞ x M ν γ D∗a1 fj (ω) |D∗a fj (ω)|⎠ dω ≤ ⎝ a

j=1

ν−γ 1 (x − a) √ . √ 2Γ (ν − γ 1 ) ν − γ 1 2ν − 2γ 1 − 1 ⎧ ⎡ ⎤ ⎫ M ⎬ ⎨ x 2 ν ⎣ (D∗a fj (ω)) ⎦ dω , ⎭ ⎩ a

(16.93)

j=1

all a ≤ x ≤ b. Corollary 16.50. (To Theorem 16.46; λα = 0, p (t) = q (t) = 1, λβ = λν = 1, p = 2.) In detail: let ν ≥ γ 2 + 1, γ 2 ≥ 0, n := ν, fj ∈ AC n ([a, b]), ν fj ∈ L∞ (a, b), j = 1, . . . , M ∈ N; a ≤ x ≤ b. D∗a (l) Here fj (a) = 0, l = 0, 1, . . . , n − 1; j = 1, . . . , M. Then ⎧⎧ x ⎨⎨M −1 γ 2 ν D∗a fj+1 (ω) |(D∗a fj ) (ω)| ⎩ ⎩ a j=1

γ

$ ν + D∗a2 fj (ω) |(D∗a fj+1 ) (ω)| γ ν f1 ) (ω)| + D∗a2 fM (ω) |(D∗a γ

$ ν + D∗a2 f1 (ω) |(D∗a fM ) (ω)| dω

√

(ν−γ )

2 2 (x − a) √ ≤ √ Γ (ν − γ 2 ) ν − γ 2 2ν − 2γ 2 − 1 ⎧ ⎡ ⎤ ⎫ M ⎨ x ⎬ 2 ν ⎣ ((D∗a fj ) (ω)) ⎦ dω , ⎩ a ⎭

(16.94)

j=1

all a ≤ x ≤ b. Corollary 16.51. (To Theorem 16.48; λα = 1, λα+1 = 1/2, p = 3/2, p (t) = q (t) = 1.) In detail: let ν ≥ γ 1 + 2, γ 1 ≥ 0, n := ν, and fj ∈

16.3 Main Results

415

(l)

AC n ([a, b]), j = 1, . . . , M ∈ N, such that fj (a) = 0, l = 0, 1, . . . , n − 1, ν fj ∈ L∞ (a, b), j = 1, . . . , M. a ≤ x ≤ b. Assume also D∗a Set (3ν−3γ 1 −1/2) 2 (x − a) ∗ Φ (x) := √ , (16.95) 3/2 3ν − 3γ 1 − 2 (Γ (ν − γ 1 )) all a ≤ x ≤ b. Then ⎧⎧ ' : x ⎨⎨M −1 γ +1 γ 1 D∗a1 fj (ω) D∗a fj+1 (ω) ⎩ ⎩ a j=1

γ + D∗a1 fj+1 (ω) '

γ + D∗a1 f1 (ω) γ + D∗a1 fM (ω) ⎡ ≤ 2Φ∗ (x) · ⎣ a

x

(: γ 1 +1 D∗a fj (ω) : γ 1 +1 D∗a fM (ω)

(: γ 1 +1 dω D∗a f1 (ω)

⎛ ⎞ ⎤ M 3/2 ν ⎝ |(D∗a fj ) (ω)| ⎠ dω ⎦ ,

(16.96)

j=1

all a ≤ x ≤ b. We continue with results regarding ·∞ . Theorem 16.52. All are as in Theorem 16.38 but for fj ∈ AC n ([a, b]), j = 1, . . . , M ∈ N. Then a

x

⎧⎧ −1 + ⎨⎨M γ λ γ λ ν 2 D∗a1 fj (ω)λα D∗a p (ω) fj+1 (ω) β |(D∗a fj ) (ω)| ν ⎩⎩ j=1

, λ γ λ α γ λ ν + D∗a2 fj (ω) β D∗a1 fj+1 (ω) |(D∗a fj+1 ) (ω)| ν + λα γ λ γ λ ν + D∗a1 f1 (ω) D∗a2 fM (ω) β |(D∗a f1 ) (ω)| ν , λ γ λα γ λ ν + D∗a2 f1 (ω) β D∗a1 fM (ω) |(D∗a fM ) (ω)| ν dω ⎧ ⎫ M ⎨ ⎬ 2λβ 2(λα +λν ) ν ν ≤ T (x) D∗a fj ∞,(a,x) + D∗a fj ∞,(a,x) , (16.97) ⎩ ⎭ j=1

all a ≤ x ≤ b.

416

16. Caputo Fractional Multivariate Opial Inequalities

Proof. Based on Theorem 16.38.

We give Corollary 16.53. (to Theorem 16.52) In detail: let ν ≥ γ 1 + 2, γ 1 ≥ 0, (l) n := ν and fj ∈ AC n ([a, b]), j = 1, . . . , M ∈ N, such that fj (a) = 0, ν fj ∈ l = 0, 1, . . . , n − 1; j = 1, . . . , M ; a ≤ x ≤ b. Further suppose that D∗a L∞ (a, b), j = 1, . . . , M. Then ⎧⎧ x ⎨⎨M −1 + γ +1 γ 1 D∗a1 fj (ω) D∗a fj+1 (ω) ⎩ ⎩ a j=1

, γ +1 γ + D∗a1 fj (ω) D∗a1 fj+1 (ω) + γ +1 γ + D∗a1 f1 (ω) D∗a1 fM (ω) , γ +1 γ + D∗a1 f1 (ω) D∗a1 fM (ω) dω ≤

⎞ ⎛ M 2 ν ⎝ D∗a fj ∞ ⎠, 2 (Γ (ν − γ 1 + 1)) j=1 (x − a)

2(ν−γ 1 )

(16.98)

all a ≤ x ≤ b. We continue with the converse results. Theorem 16.54. Let γ j ≥ 0, 1 ≤ ν − γ j < 1/p, 0 0 and q (t) ≥ 0, with all p (t), 1/p (t), q (t), 1/q (t) ∈ L∞ (a, b) . Further assume ν fi ∈ L∞ (a, b), i = 1, . . . , M, each of which is of ﬁxed sign a.e. on [a, b] . D∗a Let λν > 0 and λα , λβ ≥ 0 such that λν > p. Here Pk (ω), k = 1, 2, A (ω), A0 (x) are as in (16.39), (16.40), and (16.41), respectively. Call ϕ1 (x) := A0 (x) |λβ =0

λν λα + λν

λν /p

δ ∗1 := M 1−(λα +λν /p) .

,

(16.99) (16.100)

If λβ = 0, then a

⎞ M λα ν γ D∗a1 fj (ω) |D∗a fj (ω)|λν ⎠ dω q (ω) ⎝ ⎛

x

j=1

(16.101)

16.3 Main Results

⎡ ∗ ⎣ ≥ δ 1 ϕ1 (x)

⎛ x

p (ω) ⎝

a

M

⎞

417

⎤(λα +λν /p)

p ν |D∗a fj (ω)| ⎠ dω ⎦

,

j=1

all a ≤ x ≤ b. Proof. As for Theorem 11 of [47], and Theorem 11 of [51]. We continue with Theorem 16.55. All are as in Theorem 16.54. Assume λβ ≥ λν . Denote ϕ2 (x) := (A0 (x) |λα =0 ) 2

(p−λν /p)

λν λβ + λν

λν /p

λ /p

δ3 ν ,

(16.102)

where δ 3 is as in (16.71). If λα = 0, then ⎧⎧ x −1 + ⎨⎨M γ λ ν D∗a2 fj+1 (ω)λβ |(D∗a q (ω) fj ) (ω)| ν ⎩ ⎩ a j=1

, λ γ λ ν + D∗a2 fj (ω) β |(D∗a fj+1 ) (ω)| ν + λ γ λ ν + D∗a2 fM (ω) β |(D∗a f1 ) (ω)| ν

, λ γ λ ν + D∗a2 f1 (ω) β |(D∗a fM ) (ω)| ν dω ≥ x M 1−(λν +λβ /p) 2(λν +λβ /p) ϕ2 (x) · p (ω) a

⎤ ⎫((λν +λβ )/p) ⎡ M ⎬ p ν ⎣ |(D∗a fj ) (ω)| ⎦ dω , ⎭

(16.103)

j=1

all a ≤ x ≤ b. Proof. As for Theorem 12 of [47], and Theorem 12 of [51]. We give Theorem 16.56. Let ν ≥ 2 and γ 1 ≥ 0 such that 2 ≤ ν − γ 1 < 1/p, 0 < p < 1, n := ν. Consider fi ∈ AC n ([a, b]), i = 1, . . . , M ∈ N, such (l) that fi (a) = 0, l = 0, 1, . . . , n − 1; i = 1, . . . , M ; a ≤ x ≤ b. Assume ν fi ∈ L∞ (a, b), i = 1, . . . , M, each of which is of ﬁxed sign a.e. on that D∗a [a, b] . Consider also p (t) > 0 and q (t) ≥ 0, with all p (t), 1/p (t), q (t), 1/q (t) ∈ L∞ (a, b). Let λα ≥ λα+1 > 1.

418

16. Caputo Fractional Multivariate Opial Inequalities

Here θ3 is as in (16.73), L (x) as in (16.50), P1 (x) as in (16.51), T (x) as in (16.52), ω 1 as in (16.53), and Φ as in (16.54). ⎧⎧ x −1 ) ⎨⎨M λα+1 γ +1 γ 1 D∗a1 fj (ω)λα D∗a q (ω) fj+1 (ω) ⎩⎩ a j=1

λα+1 * λα γ +1 γ 1 1 + D∗a fj+1 (ω) D∗a fj (ω) ) λα+1 λα γ +1 γ + D∗a1 f1 (ω) D∗a1 fM (ω) λα+1 * λα γ +1 γ + D∗a1 fM (ω) D∗a1 f1 (ω) dω ≥ M 1−(λα +λα+1 /p) 2(λα +λα+1 /p) Φ (x) · ⎞ ⎤((λα +λα+1 )/p) ⎛ ⎡ x M p ν ⎣ p (ω) ⎝ |(D∗a fj ) (ω)| ⎠ dω ⎦ , a

(16.104)

j=1

all a ≤ x ≤ b. Proof. As for Theorem 13 of [47] and Theorem 13 of [51]. We have the special cases. Corollary 16.57. (To Theorem 16.54; λβ = 0, p (t) = q (t) = 1.) Then ⎛ ⎞ x M λα ν γ λ D∗a1 fj (ω) |D∗a fj (ω)| ν ⎠ dω ⎝ a

j=1

⎡ ≥ δ ∗1 ϕ1 (x) ⎣

x

a

⎡ ⎤ ⎤((λα +λν )/p) M p ν ⎣ |D∗a fj (ω)| ⎦ dω ⎦ ,

(16.105)

j=1

all a ≤ x ≤ b. In (16.105), A0 (x) |λβ =0 of ϕ1 (x) is given by (16.59). Corollary 16.58. (To Theorem 16.55; λα = 0, p (t) = q (t) = 1.) It holds ⎧⎧ x ⎨⎨M −1 + γ λ ν D∗a2 fj+1 (ω)λβ |(D∗a fj ) (ω)| ν a ⎩⎩ j=1

, λ γ λ ν + D∗a2 fj (ω) β |(D∗a fj+1 ) (ω)| ν

16.4 Background—II

419

+ λ γ λ ν + D∗a2 fM (ω) β |(D∗a f1 ) (ω)| ν , λ γ λ ν + D∗a2 f1 (ω) β |(D∗a fM ) (ω)| ν dω ≥ M 1−(λν +λβ /p) 2(λν +λβ /p) ϕ2 (x) · ⎧ ⎨ ⎩

a

x

⎡ ⎤ ⎫((λν +λβ )/p) M ⎬ p ν ⎣ |(D∗a fj ) (ω)| ⎦ dω , ⎭

(16.106)

j=1

all a ≤ x ≤ b. In (16.106), (A0 (x) |λα =0 ) of ϕ2 (x) is given by (16.64). Next we apply the above results on the spherical shell.

16.4 Background—II Here we initially follow [356, pp. 149–150] and [383, pp. 87–88]. Let us denote by# dx ≡ λRN (dx), $N ∈ N, the Lebesgue measure on RN , and S N −1 := x ∈ RN : |x| = 1 the unit sphere on RN , where |·| stands for the Euclidean norm in RN . Also denote the ball # $ B (0, R) := x ∈ RN : |x| < R ⊆ RN , R > 0, and the spherical shell A := B (0, R2 ) − B (0, R1 ), 0 < R1 < R2 . For x ∈ RN − {0} we can write uniquely x = rω, where r = |x| > 0, and ω = x/r ∈ S N −1 , |ω| = 1. Clearly here RN − {0} = (0, ∞) × S N −1 , and the map Φ : RN − {0} → S N −1 : Φ (x) =

x |x|

is continuous. Also A = [R1 , R2 ] × S N −1 . Let us denote by dω ≡ λS N −1 (ω) the surface measure on S N −1 to be deﬁned as the image under Φ of N · λRN restricted to the Borel class of B (0, 1) − {0} . More precisely the last deﬁnition is as follows. Let A ⊂ S N −1 be a Borel set, and let 1 := {ru : 0 < r < 1, u ∈ A} ⊂ RN . A We deﬁne

1 . λS N −1 (A) = N · λRN A

420

16. Caputo Fractional Multivariate Opial Inequalities

BX here stands for the Borel class on space X. We denote by

ω N ≡ λS N −1 S

N −1

=

dω = S N −1

2π N/2 Γ (N/2)

the surface area of S N −1 and we get the volume |B (0, r)| =

2π N/2 rN ω N rN = , N N Γ (N/2)

so that |B (0, 1)| =

2π N/2 . N Γ (N/2)

Clearly here V ol (A) = |A| = Next, deﬁne

ω N R2N − R1N 2π N/2 R2N − R1N = . N N Γ (N/2)

ψ : (0, ∞) × S N −1 → RN − {0}

by ψ (r, ω) := rω, ψ is a one-to-one and onto function; thus (r, ω) ≡ ψ −1 (x) = (|x| , Φ (x)) are called the polar coordinates of x ∈ RN − {0} . Finally, deﬁne the measure RN on (0, ∞), B(0,∞) by RN (Γ) = rN −1 dr, any Γ ∈ B(0,∞) . Γ

We mention the following very important theorem. Theorem 16.59. (See Exercise 6, pp. 149–150 in [356] and Theorem 5.2.2 pp. 87–88 of [383].) We have that λRN = (RN × λS N −1 ) ◦ ψ −1 on BRN −{0} . In particular, if f is a nonnegative Borel measurable function on RN , BRN , then the Lebesgue integral f (x) dx = rN −1 f (rω) λS N −1 (dω) dr RN

(0,∞)

=

f (rω) r S N −1

(0,∞)

S N −1

N −1

dr λS N −1 (dω) .

(16.107)

16.4 Background—II

421

Clearly (16.107) is true for f a Borel integrable function taking values in R. Based on Theorem 16.59 in [45] we prove the next result which is the main tool of this section. Proposition 16.60. Let f : A → R, be a Lebesgue integrable function, where A := B (0, R2 ) − B (0, R1 ), 0 < R1 < R2 . Then

R2

f (x) dx = A

f (rω) r S N −1

(16.108)

N −1

dr dω.

R1

We make Remark 16.61. Let F : A = [R1 , R2 ] × S N −1 → R and for each ω ∈ deﬁne S gω (r) := F (rω) = F (x), N −1

where x ∈ A, with A := B (0, R2 ) − B (0, R1 ); 0 < R1 ≤ r ≤ R2 , r = |x| , ω = x/r ∈ S N −1 . For each ω ∈ S N −1 we assume that gω ∈ AC n ([R1 , R2 ]), where n := ν, ν ≥ 0. Thus, by Corollary 16.8, for almost all r ∈ [R1 , R2 ] , there exists the Caputo fractional derivative r 1 n−ν−1 (n) ν D∗R g (r) = (r − t) gω (t) dt, (16.109) 1 ω Γ (n − ν) R1 for all ω ∈ S N −1 . Now we are ready to give Deﬁnition 16.62. Let F : A → R, ν ≥ 0, n := ν such that F (·ω) ∈ AC n ([R1 , R2 ]), for all ω ∈ S N −1 . We call the Caputo radial fractional derivative the following function r ν n ∂∗R F (x) 1 n−ν−1 ∂ F (tω) 1 := (r − t) dt, (16.110) ∂rν Γ (n − ν) R1 ∂rn where x ∈ A; that is, x = rω, r ∈ [R1 , R2 ] , ω ∈ S N −1 .

422

16. Caputo Fractional Multivariate Opial Inequalities

Clearly 0 ∂∗R F (x) 1 = F (x), ∂r0 ν ∂∗R F (x) ∂ ν F (x) 1 = , if ν ∈ N. ∂rν ∂rν

The above function (16.110) exists almost everywhere for x ∈ A. We justify this next. Note 16.63. Call ν ∂∗R F (x) 1 does not exist . Λ1 := r ∈ [R1 , R2 ] : ∂rν

(16.111)

We have that Lebesgue measure λR (Λ1 ) = 0. Call ΛN := Λ1 × S N −1 . So there exists a Borel set Λ∗1 ⊂ [R1 , R2 ] , such that Λ1 ⊂ Λ∗1 , λR (Λ∗1 ) = λR (Λ1 ) = 0; thus RN (Λ∗1 ) = 0. Consider now Λ∗N := Λ∗1 × S N −1 ⊂ A, which is a Borel set of RN − {0} . Clearly then by Theorem 16.59, λRN (Λ∗N ) = 0, but ΛN ⊂ Λ∗N , implying λRN (ΛN ) = 0. Consequently (16.110) exists a.e. in x with respect to λRN on A. We give the following fundamental representation result. Theorem 16.64. Let ν ≥ γ + 1, γ ≥ 0, n := ν, F : A → R with F ∈ L1 (A). Assume that F (·ω) ∈ AC n ([R1 , R2 ]) for all ω ∈ S N −1 , and ν F (·ω)/∂rν ∈ L∞ (R1 , R2 ) for all ω ∈ S N −1 . that ∂∗R 1 ν F (x)/∂rν ∈ L∞ (A). More precisely, for these Further assume that ∂∗R 1 ν r ∈ [R1 , R2 ] , and for each ω ∈ S N −1, for which D∗R F (rω) takes real 1 ν ≤ M1 . F (rω) values, there exists M1 > 0 such that D∗R 1 We suppose that ∂ i F (R1 ω)/∂rj = 0, j = 0, 1, . . . , n − 1, for every ω ∈ S N −1 . Then γ F (x) ∂∗R γ 1 = D∗R F (rω) = 1 ∂rγ r 1 ν−γ−1 ν D∗R1 F (tω) dt, (r − t) (16.112) Γ (ν − γ) R1 true ∀x ∈ A; that is, true ∀r ∈ [R1 , R2 ] and ∀ω ∈ S N −1 , γ > 0. Here γ F (·ω) ∈ AC ([R1 , R2 ]), (16.113) D∗R 1 ∀ω ∈ S N −1 , γ > 0. Furthermore

γ F (x) ∂∗R 1 ∈ L∞ (A), γ > 0. ∂rγ

(16.114)

16.4 Background—II

423

In particular, it holds F (x) = F (rω) =

1 Γ (ν)

r

ν−1

(r − t) R1

ν D∗R1 F (tω) dt,

(16.115)

true ∀x ∈ A; that is, true ∀r ∈ [R1 , R2 ] and ∀ω ∈ S N −1 , and F (·ω) ∈ AC ([R1 , R2 ]), ∀ω ∈ S N −1 .

(16.116)

Proof. By our assumptions and Theorem 16.16, Corollary 16.14, we have valid (16.112) and (16.115). Also (16.113) is clear; see [45]. Property (16.116) is easy to prove. Fixing r ∈ [R1 , R2 ] , the function ν−γ−1

δ r (t, ω) := (r − t) is measurable on

ν D∗R F (tω) 1

[R1 , r] × S N −1 , B [R1 ,r] × B S N −1 .

Here B [R1 ,r] × B S N −1 stands for the complete σ-algebra generated by B [R1 ,r] × B S N −1 , where B X stands for the completion of BX . Then we get that r |δ r (t, ω)| dt dω = S N −1

S N −1

r

R1 ν−γ−1

(r − t) R1

ν D∗R F (tω) dt dω ≤ 1

(16.117)

ν r ∂∗R1 F (x) ν−γ−1 (r − t) dt dω ∂rν R1 S N −1 ∞,([R1 ,r]×S N −1 ) (16.118) ν ν−γ ∂∗R1 F (x) 2π N/2 (r − R1 ) = ≤ ∂rν Γ (N/2) (ν − γ) ∞,([R1 ,r]×S N −1 ) ν ν−γ ∂∗R1 F (x) 2π N/2 (R2 − R1 ) < ∞. (16.119) ∂rν Γ (N/2) (ν − γ) ∞,A Hence δ r (t, ω) is integrable on [R1 , r] × S N −1 , B [R1 ,r] × B S N −1 . γ Consequently, by Fubini’s theorem and (16.112), we obtain that D∗R F 1 N −1 (rω), ν ≥ γ + 1, γ > 0 is integrable in ω over S , B S N −1 . So we

424

16. Caputo Fractional Multivariate Opial Inequalities

γ have that D∗R F (rω) is continuous in r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 , and 1 measurable in ω ∈ S N −1 , ∀r ∈ [R1 , R2 ] . So, it is a Carath´eodory function. Here [R1 , R2 ] is a separable metric space and S N −1 is a measurable space, and the function takes values in R∗ := R ∪ {±∞} is a metric γ, which F (rω) is jointly space. Therefore by Theorem 20.15, p. 156 of [10], D∗R 1 B[R1 ,R2 ] × B S N −1 -measurable on [R1 , R2 ] × S N −1 = A, that is, Lebesgue measurable on A. Indeed then we have that γ 1 D F (rω) ≤ ∗R1 Γ (ν − γ) r ν−γ−1 ν (16.120) D∗R1 F (tω) dt (r − t) R1

≤

ν D F (·ω) ∗R1 ∞,[R

1 ,R2 ]

Γ (ν − γ)

(r − t)

ν−γ−1

dt

≤

(16.121)

R1

ν−γ

M1 (r − R1 ) Γ (ν − γ) (ν − γ)

r

≤

M1 ν−γ (R2 − R1 ) := τ < ∞, Γ (ν − γ − 1)

for all ω ∈ S N −1 and for all r ∈ [R1 , R2 ] . We proved that γ D F (rω) ≤ τ < ∞, ∀ω ∈ S N −1 , and ∀r ∈ [R1 , R2 ] , ∗R1

(16.122)

γ hence proving ∂∗R F (x)/∂rγ ∈ L∞ (A), γ > 0. We have completed our 1 proof.

16.5 Main Results on a Spherical Shell 16.5.1 Results Involving One Function We give Theorem 16.65. Let ν ≥ γ i + 1, γ i ≥ 0, i = 1, . . . , l ∈ N, n := ν ; and 0 ≤ γ 1 < γ 2 ≤ γ 3 ≤ . . . ≤ γ l . Here f : A → R is as in Theorem "l 16.64. Let ri > 0 : i=1 ri = p. If γ 1 = 0 we set r1 = 1. Let s1 , s1 > 1 : 1/s1 + 1/s1 = 1, and s2 , s2 > 1 : 1/s2 + 1/s2 = 1, and p > s2 , N ≥ 2. Denote (N −1)s +1 (N −1)s1 +1 1/s1 1 − R1 R2 , (16.123) Q1 (R2 ) := (N − 1) s1 + 1 Q2 (R2 ) :=

(1−N )(s2 /p)+1

R2

(1−N )(s2 /p)+1

− R1 (1 − N ) (s2 /p) + 1

p/s2 ,

(16.124)

16.5 Main Results on a Spherical Shell

and σ :=

p − s2 . ps2

425

(16.125)

Also call C := Q1 (R2 ) Q2 (R2 ) l i=1

"

σ ri σ ri r σ (Γ (ν − γ i )) (ν − γ i − 1 + σ) i

(R2 − R1 )( l i=1

"l

i=1 (ν−γ i −1)ri +p/s2 +1/s1 −1

)

1/s1 . (ν − γ i − 1) ri s1 + s1 sp2 − 1 + 1

Then

(16.126)

ri l γi ∂∗R1 f (x) dx ≤ ∂rγ i A i=1

ν ∂∗R1 f (x) p ∂rν dx. A

C

(16.127)

Proof. Clearly here f (·ω) fulﬁlls all the assumptions of Theorem 16.24, ∀ω ∈ S N −1 . We set there q1 (r) = q2 (r) := rN −1 , r ∈ [R1 , R2 ] . Hence by (16.30) we have

R2

r

N −1

R1

R2

C R1

ri l γi ∂∗R1 f (rω) dr ≤ ∂rγ i i=1

ν ∂ f (rω) p dr, ∀ω ∈ S N −1 . rN −1 ∗R1 ν ∂r

(16.128)

Therefore it holds

N −1

ri l γi ∂∗R1 f (rω) dr dω ≤ ∂rγ i i=1

N −1

ν ∂∗R1 f (rω) p dr dω. ∂rν

R2

r S N −1

R1

R2

C

r S N −1

R1

(16.129)

Using the conclusion of Theorem 16.64 and Proposition 16.60 we derive (16.127) . We continue with the following extreme case.

426

16. Caputo Fractional Multivariate Opial Inequalities

Theorem 16.66. Let ν ≥ γ i + 1, γ i ≥ 0, i = 1, . . . , l ∈ N, n := ν, and 0 ≤ γ 1 < γ 2 ≤ γ 3 ≤ . . . ≤ γ l . Here f : A → R as in Theorem 16.64. Let "l ri > 0 : i=1 ri = r. If γ 1 = 0 we set r1 = 1, N ≥ 2. Call " rν− l

i=1 i i R2N −1 (R2 − R1 ) ; := > 0. M "l &l ri rν − i=1 ri γ i + 1 i=1 (Γ (ν − γ i + 1))

Then

r γ +1

(16.130)

ri l γi ∂∗R1 f (x) dx ≤ ∂rγ i A i=1

N/2 ;M r 2π M . 1 Γ (N/2)

(16.131)

Proof. Clearly here f (·ω) fulﬁlls all the assumptions of Theorem 16.26, ∀ω ∈ S N −1 . We set q1 (r) = rN −1 , r ∈ [R1 , R2 ] . Hence by (16.35) we have

R2

rN −1

R1

l γ D i f (rω)ri dr ≤ ∗R1 i=1

⎛

⎞ " ν rν− li=1 ri γ i +1 N −1 ∂∗R1 f (·ω) (R − R ) R 2 1 2 ⎝ ⎠ "l &l ν ri ∂r ∞,[R1 ,R2 ] rν − i=1 ri γ i + 1 i=1 (Γ (ν − γ i + 1)) ;M r , ∀ω ∈ S N −1 . ≤M 1 Hence

S N −1

R2

R1

(16.132)

l r γ i D i f (rω) dr dω ≤ rN −1 ∗R1 i=1 N/2 ;M1r 2π M . Γ (N/2)

(16.133)

Using the conclusion of Theorem 16.64 and Proposition 16.60 we derive (16.131).

16.5 Main Results on a Spherical Shell

427

16.5.2 Results Involving Two Functions We need to make the following assumption. Assumption 16.67. Let ν ≥ γ i + 1, γ i ≥ 0, i = 1, 2, n := ν , f1 , f2 : A → R with f1 , f2 ∈ L1 (A), where A := B (0, R2 ) − B (0, R1 ), 0 < R1 < R2 . Assume that f1 (·ω), f2 (·ω) ∈ AC n ([R1 , R2 ]) for all ω ∈ S N −1 , ν f (·ω)/∂rν ∈ L∞ (R1 , R2 ), for all ω ∈ S N −1 ; i = 1, 2. Further and that ∂∗R 1 i ν f (x)/∂rν ∈ L∞ (A), i = 1, 2. More precisely, for these assume that ∂∗R 1 i ν r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 , for which D∗R f (rω) takes real values, there 1 i exists Mi > 0 such that ν D∗R fi (rω) ≤ Mi , for i = 1, 2. (16.134) 1 We suppose that ∂ j fi (R1 ω) = 0, j = 0, 1, . . . , n − 1, ∂rj ∀ω ∈ S N −1 ; i = 1, 2. Let λν > 0, and λα , λβ ≥ 0, such that λν < p, where p > 1. If γ 1 = 0 we set λα = 1 and if γ 2 = 0 we set λβ = 1, here N ≥ 2. Assumption 16.67* . (continuation of Assumption 16.67) Set w (ν−γ k −1)p/(p−1) (1−N /p−1) Pk (w) := (w − t) t dt,

(16.135)

R1

k = 1, 2, R1 ≤ w ≤ R2 , A (w) :=

w(N −1)(1−λν /p) (P1 (w)) (Γ (ν − γ 1 ))

R2

A0 (R2 ) :=

λα (p−1/p)

λα

(P2 (w))

λβ ((p−1)/p)

,

λβ

(Γ (ν − γ 2 ))

(16.136)

(p−λν )/p p/(p−λν )

(A (w))

dw

.

(16.137)

R1

We present Theorem 16.68. All here are as in Assumptions 16.67 and 16.67* ; especially assume λα > 0, λβ = 0, and p = λα + λν > 1. Then ⎡ λα λ γ 1 ν ∂∗R1 f1 (x) ∂∗R f (x) ν 1 1 + ⎣ ∂rγ 1 ∂rν A

428

16. Caputo Fractional Multivariate Opial Inequalities

⎤ γ ∂ 1 f (x) λα ∂ ν f (x) λν ∗R1 2 ∗R1 2 ⎦ dx ≤ ∂rγ 1 ∂rν A0 (R2 ) |λβ =0

λν p

(16.138)

(λν /p)

p * ν ) ν ∂∗R1 f1 (x) p ∂∗R f (x) 1 2 + dx. ∂rν ∂rν A Proof. We apply Theorem 16.30 here for every ω ∈ S N −1 ; here p (r) = q (r) = rN −1 , r ∈ [R1 , R2 ] . We use Theorem 16.64 and Proposition 16.60. Thus the proof is similar to the proof of Theorem 16.65. The counterpart of the last theorem follows. Theorem 16.69. All here are as in Assumptions 16.67 and 16.67* ; especially suppose λα = 0, λβ > 0, p = λν + λβ > 1. Denote λ /λ 2 β ν − 1, if λβ ≥ λν , (16.139) δ 3 := 1, if λβ ≤ λν . Then

⎡ ∂ γ 2 f (x) λβ ∂ ν f (x) λν ∗R1 1 + ⎣ ∗R1 2 γ 2 ∂r ∂rν A

⎤ γ ∂ 2 f (x) λβ ∂ ν f (x) λν ∗R1 1 ∗R1 2 ⎦ dx ≤ ∂rγ 2 ∂rν (A0 (R2 ) |λα =0 ) 2λβ /p

λν p

(λν /p)

(16.140)

λ /p

δ3 ν

p * ν ) ν ∂∗R1 f1 (x) p ∂∗R f (x) 1 2 + dx. ∂rν ∂rν A Proof. Based on Theorem 16.31, similar to the proof of Theorem 16.68. Theorem 16.70. All here are as in Assumptions 16.67 and 16.67* ; especially suppose λν , λα , λβ > 0, p = λα + λβ + λν > 1. Denote (λ +λ /λ ) 2 α β ν − 1, if λα + λβ ≥ λν , (16.141) γ 11 := 1, if λα + λβ ≤ λν .

16.5 Main Results on a Spherical Shell

Then

429

⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λβ ∂ ν f (x) λν 1 2 ∗R1 1 ∗R1 + ⎣ ∗R1 γ γ 1 2 ∂r ∂r ∂rν A

⎤ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν ∗R1 1 ∗R1 2 ∗R1 2 ⎦ dx ≤ ∂rγ 2 ∂rγ 1 ∂rν A0 (R2 )

(λν /p) + , λν λ /p λλαν /p + 2(λα +λβ )/p (1 γ 1 λβ ) ν (λα + λβ ) p p ν ν ∂∗R1 f1 (x) p ∂∗R f (x) 1 2 + dx. (16.142) ∂rν ∂rν A

Proof. Based on Theorem 16.32, similar to the proof of Theorem 16.68. We give the next special important case. Theorem 16.71. All are as in Assumption 16.67 without λν there. Here γ 2 = γ 1 + 1, λα ≥ 0, λβ := λα+1 ∈ (0, 1), and p = λα + λα+1 > 1. Denote (λ /λ ) 2 α α+1 − 1 if λα ≥ λα+1 (16.143) θ3 := 1, if λα ≤ λα+1 , ) (1 − λα+1 ) L (R2 ) := 2 (N − λα+1 ) λα+1 ,(1−λα+1 ) θ λ (N −λα+1 )/(1−λα+1 ) (N −λα+1 )/(1−λα+1 ) 3 α+1 R2 − R1 , p (16.144) and R2 (ν−γ 1 −1)p/(p−1) (1−N )/(p−1) P1 (R2 ) := (R2 − t) t dt, (16.145) R1

Φ (R2 ) := L (R2 ) Then

(p−1)

P1 (R2 ) p (Γ (ν − γ 1 ))

2p−1 .

⎡ ∂ γ 1 f (x) λα ∂ γ 1 +1 f (x) λα+1 1 2 ∗R1 ⎣ ∗R1 + γ1 γ 1 +1 ∂r ∂r A

⎤ γ ∂ 1 f (x) λα ∂ γ 1 +1 f (x) λα+1 ∗R1 1 ∗R1 2 ⎦ dx ∂rγ 1 ∂rγ 1 +1

(16.146)

430

16. Caputo Fractional Multivariate Opial Inequalities

p ν ν ∂∗R1 f1 (x) p ∂∗R f (x) 1 2 dx. ≤ Φ (R2 ) + ∂rν ∂rν A

(16.147)

Proof. Based on Theorem 16.33, similar to the proof of Theorem 16.68. We give an L∞ result on the shell. We need to make Assumption 16.72. Let ν ≥ γ i + 1, γ i ≥ 0, i = 1, 2, n := ν , f1 , f2 : A → R with f1 , f2 ∈ L1 (A), where RN ⊇ A := B (0, R2 ) − B (0, R1 ), 0 < R1 < R2 , N ≥ 2. Assume that f1 (·ω), f2 (·ω) ∈ AC n ([R1 , R2 ]) for ν f (·ω)/∂rν ∈ L∞ (R1 , R2 ), for all ω ∈ S N −1 ; all ω ∈ S N −1 , and that ∂∗R 1 i ν f (x)/∂rν ∈ L∞ (A), i = 1, 2. More i = 1, 2. Further assume that ∂∗R 1 i ν precisely, for these r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 , for which D∗R f (rω) takes 1 i real values, there exists Mi > 0 such that ν D∗R fi (rω) ≤ Mi , for i = 1, 2. (16.148) 1 We suppose that ∂ j fi (R1 ω) = 0, j = 0, 1, . . . , n − 1, ∂rj ∀ω ∈ S N −1 ; i = 1, 2. Let λν , λα , λβ ≥ 0. If γ 1 = 0 we set λα = 1 and if γ 2 = 0 we set λβ = 1. We present Theorem 16.73. All are as in Assumption 16.72. Set R2N −1 T (R2 − R1 ) := · (νλα − γ 1 λα + νλβ − γ 2 λβ + 1) (νλα −γ 1 λα +νλβ −γ 2 λβ +1)

(R2 − R1 )

λα

(Γ (ν − γ 1 + 1))

λβ

(Γ (ν − γ 2 + 1))

.

Then ⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λβ ∂ ν f (x) λν 1 2 ∗R1 1 ∗R ∗R 1 1 + ⎣ γ1 γ2 ∂r ∂r ∂rν A

⎤ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν ∗R1 1 ∗R1 2 ∗R1 2 ⎦ dx ∂rγ 2 ∂rγ 1 ∂rν

(16.149)

16.5 Main Results on a Spherical Shell

≤ T (R2 − R1 ) +

2(λα +λν )

M1

2λβ

+ M1

2λβ

+ M2

π N/2 Γ (N/2)

431

,

2(λα +λν )

.

+ M2

(16.150)

Proof. Apply Theorem 16.38 for every ω ∈ S N −1 ; here p (r) = rN −1 , r ∈ [R1 , R2 ] . It follows the proof of Theorem 16.66. Finally use Theorem 16.64 and Proposition 16.60.

16.5.3 Results Involving Several Functions We need to make the following assumption. Assumption 16.74. Let ν ≥ γ i + 1, γ i ≥ 0, i = 1, 2, n := ν, fj : A → R with fj ∈ L1 (A), j = 1, . . . , M, M ∈ N, where RN ⊇ A := B (0, R2 ) − B (0, R1 ), 0 < R1 < R2 , N ≥ 2. Assume that fj (·ω) ∈ AC n ([R1 , R2 ]) for ν f (·ω)/∂rν ∈ L∞ (R1 , R2 ), for all ω ∈ S N −1 ; all ω ∈ S N −1 , and that ∂∗R 1 j ν f (x)/∂rν ∈ L∞ (A), j = 1, . . . , M. j = 1, . . . , M. Further assume that ∂∗R 1 j ν More precisely, for these r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 , for which D∗R f (rω) 1 j takes real values, there exists Mj > 0 such that ν D∗R fj (rω) ≤ Mj , for j = 1, . . . , M. (16.151) 1 We suppose that ∂ j fj (R1 ω) = 0, k = 0, 1, . . . , n − 1, ∂rk ∀ω ∈ S N −1 ; j = 1, . . . , M. Let λν > 0, and λα , λβ ≥ 0, such that λν < p, where p > 1. If γ 1 = 0 we set λα = 1 and if γ 2 = 0 we set λβ = 1. We give Theorem 16.75. Let fj , j = 1, . . . , M, as in Assumption 16.74. Let λν > 0, and λα > 0; λβ ≥ 0, p := λα + λν > 1. Set w (ν−γ k −1)p/(p−1) (1−N /p−1) Pk (w) := (w − t) t dt, (16.152) R1

k = 1, 2, R1 ≤ w ≤ R2 , A (w) :=

w(N −1)(1−(λν /p)) (P1 (w)) (Γ (ν − γ 1 ))

λα ((p−1)/p)

λα

λβ ((p−1)/p)

(P2 (w)) λβ

(Γ (ν − γ 2 ))

, (16.153)

432

16. Caputo Fractional Multivariate Opial Inequalities

λα /p

R2

A0 (R2 ) :=

p/λα

(A (w))

dw

.

(16.154)

R1

Take the case of λβ = 0. Then λα λν M γ1 ν ∂∗R1 fj (x) ∂∗R1 fj (x) dx ∂rγ 1 ∂rν A j=1

≤ A0 (R2 ) |λβ =0

(λν /p)

λν p

⎡ ⎤ M ν ∂∗R1 fj (x) p dx ⎦ . ⎣ ∂rν j=1

(16.155)

A

Proof. As in Theorem 16.68, based on Theorem 16.45.

We continue with Theorem 16.76. All basic assumptions are as in Theorem 16.75. Let λν > 0, λα = 0; λβ > 0, p := λν + λβ > 1, P2 deﬁned by (16.152). Now it is w(N −1)(1−(λν /p)) (P2 (w))

λβ ((p−1)/p)

A (w) :=

(Γ (ν − γ 2 ))

(16.156)

λβ /p

R2

A0 (R2 ) :=

,

λβ

(A (w))

p/λβ

dw

.

(16.157)

R1

Denote δ 3 :=

2λβ /λν − 1, 1,

Call

λβ /p

ϕ2 (R2 ) := A0 (R2 ) 2 Then

if λβ ≥ λν , if λβ ≤ λν . λν p

λν /p

(16.158)

λ /p

δ3 ν .

⎧⎧ ⎡ λβ λ ⎨⎨M −1 γ 2 ν ∂∗R1 fj+1 (x) ∂∗R f (x) ν 1 j + ⎣ ∂rγ 2 ∂rν A ⎩⎩ j=1 ⎤⎫ γ λν ⎬ ∂ 2 f (x) λβ ∂ ν f (x) ∗R1 j ∗R1 j+1 ⎦ + ∂rγ 2 ⎭ ∂rν

(16.159)

16.5 Main Results on a Spherical Shell

433

⎡ ∂ γ 2 f (x) λβ ∂ ν f (x) λν M ∗R1 1 ∗R 1 + ⎣ ∂rγ 2 ∂rν ⎤⎫ γ ∂ 2 f (x) λβ ∂ ν f (x) λν ⎬ ∗R1 1 ∗R1 M ⎦ dx ≤ ∂rγ 2 ⎭ ∂rν ⎡ ⎤ M ν ∂∗R1 fj (x) p dx ⎦ . 2ϕ2 (R2 ) ⎣ ∂rν

(16.160)

A

j=1

Proof. As in Theorem 16.68, based on Theorem 16.46.

We present the general case. Theorem 16.77. All basic assumptions are as in Theorem 16.75. Here λν , λα , λβ > 0, p := λα + λβ + λν > 1, Pk as in (16.152), and A is as in (16.153). Here (λα +λβ )/p R2

A0 (R2 ) :=

p/(λα +λβ )

(A (w))

dw

,

(16.161)

R1

γ 11 :=

2(λα +λβ /λν ) − 1, 1,

Put

if λα + λβ ≥ λν , if λα + λβ ≤ λν .

ϕ3 (R2 ) := A0 (R2 )

λν (λα + λβ ) p

(λν /p)

+ , (λν /p) (λα +λβ /p) ν /p) λ(λ . + 2 (1 γ λ ) β 1 α Then A

⎡

M −1

⎣

j=1

(16.162)

(16.163)

⎡ λβ ∂ γ 1 f (x) λα ∂ γ 2 f ∂ ν f (x) λν j j+1 (x) ∗R ∗R ∗R1 j 1 1 + ⎣ ∂rγ 1 ∂rγ 2 ∂rν

⎤ λα γ λν ∂ 2 f (x) λβ ∂ γ 1 f ν ∗R1 j+1 (x) ∂∗R1 fj+1 (x) ⎦ ∗R1 j + ∂rγ 2 ∂rγ 1 ∂rν ⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λβ ∂ ν f (x) λν ∗R1 M ∗R1 1 + ⎣ ∗R1 1 γ γ 1 2 ∂r ∂r ∂rν ⎤⎤ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν ∗R1 1 ∗R1 M ∗R1 M ⎦⎦ dx ≤ ∂rγ 2 ∂rγ 1 ∂rν

434

16. Caputo Fractional Multivariate Opial Inequalities

⎡ ⎤ M ν ∂∗R1 fj (x) p dx ⎦ . 2ϕ3 (R2 ) ⎣ ∂rν j=1

(16.164)

A

Proof. As in Theorem 16.68, based on Theorem 16.47.

We show the special important case next. Theorem 16.78. Let all be as in Assumption 16.74 without λν there. Here γ 2 = γ 1 + 1, and let λα > 0, λβ := λα+1 , 0 < λα+1 < 1, such that p := λα + λα+1 > 1. Denote (λ /λ ) 2 α α+1 − 1, if λα ≥ λα+1 , θ3 := (16.165) 1, if λα ≤ λα+1 , ) (1 − λα+1 ) L (R2 ) := 2 (N − λα+1 )

(N −λα+1 )/(1−λα+1 )

R2

(N −λα+1 )/(1−λα+1 )

− R1

λα+1 ,(1−λα+1 ) θ λ 3 α+1 , λα + λα+1 (16.166)

and

R2

P (R2 ) :=

(ν−γ 1 −1)(p/p−1) (1−N /p−1)

(R2 − t)

t

dt,

(16.167)

R1

Φ (R2 ) := L (R2 ) Then

(p−1)

P1 (R2 ) p (Γ (ν − γ 1 ))

2(p−1) .

⎧⎧ ⎡ λα γ +1 λα+1 ⎨⎨M −1 γ 1 ∂∗R1 fj (x) ∂∗R1 1 fj+1 (x) ⎣ + ∂rγ 1 ∂rγ 1 +1 A ⎩⎩ j=1 γ λα γ +1 λα+1 ⎤⎫ ∂ 1 f ⎬ 1 ∗R1 j+1 (x) ∂∗R1 fj (x) ⎦ + ∂rγ 1 +1 ⎭ ∂rγ 1 ⎡ ∂ γ 1 f (x) λα ∂ γ 1 +1 f (x) λα+1 1 M ∗R ∗R 1 1 ⎣ + ∂rγ 1 ∂rγ 1 +1 ⎤⎫ γ ∂ 1 f (x) λα ∂ γ 1 +1 f (x) λα+1 ⎬ ∗R1 1 ∗R1 M ⎦ dx ≤ ∂rγ 1 +1 ⎭ ∂rγ 1

(16.168)

16.5 Main Results on a Spherical Shell

⎡ ⎤ M ν ∂∗R1 fj (x) p dx ⎦ . 2Φ (R2 ) ⎣ ∂rν j=1

435

(16.169)

A

Proof. As in Theorem 16.68, based on Theorem 16.48.

We study the L∞ case next. We need to make Assumption 16.79. Let ν ≥ γ i + 1, γ i ≥ 0, i = 1, 2, n := ν, fj : A → R with fj ∈ L1 (A), j = 1, . . . , M, M ∈ N, where RN ⊇ A := B (0, R2 ) − B (0, R1 ), 0 < R1 < R2 , N ≥ 2. Assume that fj (·ω) ∈ AC n ([R1 , R2 ]) for ν f (·ω)/∂rν ∈ L∞ (R1 , R2 ), for all ω ∈ S N −1 ; all ω ∈ S N −1 , and that ∂∗R 1 j ν f (x)/∂rν ∈ L∞ (A), j = 1, . . . , M. j = 1, . . . , M. Further assume that ∂∗R 1 j ν More precisely, for these r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 , for which D∗R f (rω) 1 j takes real values, there exists Mj > 0 such that ν D∗R fj (rω) ≤ Mj , for j = 1, . . . , M. (16.170) 1 We suppose that ∂ k fj (R1 ω) = 0, k = 0, 1, . . . , n − 1, ∂rk ∀ω ∈ S N −1 ; j = 1, . . . , M. Let λν , λα , λβ ≥ 0. If γ 1 = 0 we set λα = 1 and if γ 2 = 0 we set λβ = 1. The last main result follows. Theorem 16.80. All are as in Assumption 16.79. Set R2N −1 T (R2 ) := · (νλα − γ 1 λα + νλβ − γ 2 λβ + 1) (νλα −γ 1 λα +νλβ −γ 2 λβ +1)

(R2 − R1 )

λα

(Γ (ν − γ 1 + 1))

λβ

(Γ (ν − γ 2 + 1))

.

Then

(16.171)

⎧⎧ ⎡ λα γ λβ λ ⎨⎨M −1 γ 1 ν ∂∗R1 fj (x) ∂∗R2 1 fj+1 (x) ∂∗R f (x) ν 1 j + ⎣ ∂rγ 1 ∂rγ 2 ∂rν A ⎩⎩ j=1 ⎤⎫ γ λα λν ⎬ ∂ 2 f (x) λβ ∂ γ 1 f ∂ν f (x) (x) ∗R1 j ∗R1 j+1 ∗R1 j+1 ⎦ + ∂rγ 2 ⎭ ∂rγ 1 ∂rν

436

16. Caputo Fractional Multivariate Opial Inequalities

⎡ ∂ γ 1 f (x) λα ∂ γ 2 f (x) λβ ∂ ν f (x) λν 1 M ∗R1 1 ∗R ∗R 1 1 + ⎣ ∂rγ 1 ∂rγ 2 ∂rν ⎤⎫ γ ∂ 2 f (x) λβ ∂ γ 1 f (x) λα ∂ ν f (x) λν ⎬ ∗R1 1 ∗R1 M ∗R1 M ⎦ dx ≤ ∂rγ 2 ⎭ ∂rγ 1 ∂rν ⎧ ⎫ −1 ⎨M ⎬ 2π N/2 2λ 2(λ +λ ) Mj α ν + Mj β T (R2 ) . ⎩ ⎭ Γ (N/2) j=1

(16.172)

Proof. Based on Theorem 16.52; here p (r) = rN −1 , r ∈ [R1 , R2 ]. Apply (16.97) ∀ω ∈ S N −1 . It follows the proof of Theorem 16.66. Finally use Theorem 16.64 and Proposition 16.60.

16.6 Applications We need the following. Corollary 16.81. (To Theorem 16.68; f1 = f2 .) All are as in Theorem 16.68. It holds λα λν γ 1 ν ∂∗R1 f1 (x) ∂∗R1 f1 (x) dx ≤ ∂rγ 1 ∂rν A

A0 (R2 ) |λβ =0

λν p

(λν /p) ν ∂∗R1 f1 (x) p dx . ∂rν A

(16.173)

So setting λα = λν = 1, p = 2 in (16.173), we obtain in detail Proposition 16.82. Let ν ≥ γ + 1, γ ≥ 0, n := ν, f : A → R with f ∈ L1 (A), where A := B (0, R2 ) − B (0, R1 ) ⊆ RN , N ≥ 2, 0 < R1 < R2 . ν f (·ω)/∂rν Assume that f (·ω) ∈ AC n ([R1 , R2 ]), ∀ω ∈ S N −1 , and that ∂∗R 1 N −1 ν ν ∈ L∞ (R1 , R2 ), ∀ω ∈ S . Further assume that ∂∗R1 f (x)/∂r ∈ L∞ (A). ν More precisely, for these r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 , for which D∗R f (rω) 1 takes real values, ∃ M1 > 0 such that ν D∗R f (rω) ≤ M1 . 1 Suppose that ∂ j f (R1 ω) = 0, j = 0, 1, . . . , n − 1, ∀ω ∈ S N −1 . ∂rj

16.6 Applications

437

Set

r

2(ν−γ−1) (1−N )

(r − t)

P (r) :=

t

dt, R1 ≤ r ≤ R2 ,

(16.174)

R1

r(N −1/2) P (r) , Γ (ν − γ) 1/2

(16.175)

A (r) := 10 (R2 ) := A

R2

2

(A (r)) dr

.

(16.176)

R1

Then

ν γ ∂∗R1 f (x) ∂∗R f (x) 1 ∂rγ ∂rν dx ≤ A 2 ν ∂ f (x) ∗R 1 10 (R2 ) 2−1/2 A dx . ∂rν A

(16.177)

When γ = 0 we get the following in detail.

Proposition 16.83. Let ν ≥ 1, n := ν, f : A → R with f ∈ L1 (A), where A := B (0, R2 ) − B (0, R1 ) ⊆ RN , N ≥ 2, 0 < R1 < R2 . Asν sume that f (·ω) ∈ AC n ([R1 , R2 ]), ∀ω ∈ S N −1 , and that ∂∗R f (·ω)/∂rν 1 N −1 ν ν ∈ L∞ (R1 , R2 ), ∀ω ∈ S . Further assume that ∂∗R1 f (x)/∂r ∈ L∞ (A). ν More precisely, for these r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 , for which D∗R f (rω) 1 takes real values, ∃ M1 > 0 such that ν D∗R f (rω) ≤ M1 . 1 Suppose that ∂ j f (R1 ω) = 0, j = 0, 1, . . . , n − 1, ∀ω ∈ S N −1 . ∂rj Set

r

(r − t)

P0 (r) :=

2(ν−1) (1−N )

t

dt, R1 ≤ r ≤ R2 ,

(16.178)

R1

r(N −1/2) P0 (r) , A∗ (r) := Γ (ν) 1/2

11 A 0 (R2 ) :=

R2

2

(A∗ (r)) dr R1

(16.179)

.

(16.180)

438

16. Caputo Fractional Multivariate Opial Inequalities

ν ∂∗R1 f (x) dx ≤ |f (x)| ∂rν A 2 ν ∂∗R f (x) 11 −1/2 1 A0 (R2 ) 2 dx . ∂rν A

Then

(16.181)

Based on Corollary 16.35 we give Proposition 16.84. Let ν ≥ γ + 1, γ ≥ 0, n := ν, f : A → R with f ∈ L1 (A), where A := B (0, R2 ) − B (0, R1 ) ⊆ RN , N ≥ 2, 0 < R1 < R2 . ν Assume that f (·ω) ∈ AC n ([R1 , R2 ]), ∀ω ∈ S N −1 , and that ∂∗R f (·ω)/∂rν 1 ∈ L∞ (R1 , R2 ), ∀ω ∈ S N −1 . Suppose that ∂ j f (R1 ω) = 0, j = 0, 1, . . . , n − 1, ∀ω ∈ S N −1 . ∂rj

Then

γ D

r

(1)

r

R1

R1

ν (tω) D∗R f (tω) dt ≤ 1

(ν−γ) (r − R1 ) √ √ 2Γ (ν − γ) ν − γ 2ν − 2γ − 1 ν 2 D∗R1 f (tω) dt , all R1 ≤ r ≤ R2 , ∀ω ∈ S N −1 .

∗R1 f

(16.182)

(2) W hen γ = 0 we get ν |f (tω)| D∗R f (tω) dt ≤ 1

r

R1 ν

(r − R1 ) √ √ 2Γ (ν) ν 2ν − 1

r

R1

ν 2 D∗R1 f (tω) dt ,

(16.183)

all R1 ≤ r ≤ R2 , ∀ω ∈ S N −1 . In particular we have R2 ν |f (rω)| D∗R f (rω) dr ≤ (3) 1 R1

ν

(R2 − R1 ) √ √ 2Γ (ν) ν 2ν − 1

R2

R1

ν D∗R f 1

(rω)

2

dr , ∀ω ∈ S N −1 . (16.184)

Next we apply Proposition 16.84; see (16.183) for proving the uniqueness of solution in a PDE initial value problem on A.

16.6 Applications

439

Theorem 16.85. Let ν > 1, ν ∈ / N, n := ν, f : A → R with f ∈ L1 (A), where A := B (0, R2 ) − B (0, R1 ) ⊆ RN , N ≥ 2, 0 < R1 < R2 . ν f (·ω)/∂rν Assume that f (·ω) ∈ AC n ([R1 , R2 ]), ∀ω ∈ S N −1 , and that ∂∗R 1 N −1 ν ν ∈ AC ([R1 , R2 ]), ∀ω ∈ S . Further assume D∗R 1 f (x)/∂r ∈ L∞ (A), ν ≤ M1 , ∀r ∈ [R1 , R2 ] , f (rω) such that there exists M1 > 0 with D∗R 1 N −1 ∀ω ∈ S . Suppose that ∂ j f (R1 ω) = 0, j = 0, 1, . . . , n − 1, ∀ω ∈ S N −1 . ∂rj Consider the PDE ∂ ∂r

ν f (x) ∂∗R 1 ∂rν

= θ (x) f (x),

(16.185)

∀x ∈ A, where 0 = θ : A → R is continuous. If (16.185) has a solution then it is unique. Proof. We rewrite (16.185) as ν ∂∗R1 f (rω) ∂ = θ (rω) f (rω), (16.186) ∂r ∂rν valid ∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 , and 0 = θ : [R1 , R2 ] × S N −1 → R is continuous. Assume f1 and f2 are the solution to (16.185); then ν ∂∗R1 f1 (rω) ∂ (16.187) = θ (rω) f1 (rω), ∂r ∂rν and ∂ ∂r

ν f (rω) ∂∗R 1 2 ∂rν

= θ (rω) f2 (rω),

∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 . Call g := f1 − f2 ; thus by subtraction in (16.187) we get ν ∂∗R1 g (rω) ∂ = θ (rω) g (rω), ∂r ∂rν ∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 . Of course ∂ j g (R1 ω) = 0, j = 0, 1, . . . , n − 1, ∀ω ∈ S N −1 . ∂rj Consequently we have ν ν ∂∗R1 g (rω) ∂ ∂∗R1 g (rω) ∂rν ∂r ∂rν

(16.188)

(16.189)

440

16. Caputo Fractional Multivariate Opial Inequalities

= θ (rω) g (rω)

ν ∂∗R g (rω) 1 , ∂rν

(16.190)

∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 . Hence ν r ν ∂∗R1 g (tω) ∂ ∂∗R1 g (tω) dt ∂rν ∂r ∂rν R1 ν r ∂∗R1 g (tω) θ (tω) g (tω) = dt, ∂rν R1

(16.191)

∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 . Therefore we ﬁnd ∂ ν g(tω) 2 r ∗R1 ν r ∂∗R1 g (tω) ∂r ν = θ (tω) g (tω) dt, 2 ∂rν R1

(16.192)

R1

∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 . ν g (R1 ω)/∂rν = 0, ∀ω ∈ S N −1 ; see (16.110). Notice that ∂∗R 1 Consequently we ﬁnd

ν ∂∗R g (rω) 1 ∂rν

2

= 2

ν ∂∗R g (tω) 1 θ (tω) g (tω) ν ∂r R1 ν r ∂∗R1 g (tω) dt ≤ 2 g∞ |g (tω)| ∂rν R1 ν (16.183) θ∞ (r − R1 ) ≤ √ √ Γ (ν) ν 2ν − 1 r ν 2 D∗R1 g (tω) dt ≤ R1 ν

θ∞ (R2 − R1 ) √ √ Γ (ν) ν 2ν − 1

∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 . Call K :=

r

r

R1

ν 2 D∗R1 g (tω) dt ,

dt (16.193) (16.194)

(16.195)

ν

θ∞ (R2 − R1 ) > 0. √ √ Γ (ν) ν 2ν − 1

(16.196)

So we have proved that

ν ∂∗R g (rω) 1 ∂rν

2

r

≤K R1

ν 2 D∗R1 g (tω) dt ,

(16.197)

ν ∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 . Here D∗R g (·ω) ∈ C ([R1 , R2 ]), ∀ω ∈ S N −1 . 1

16.6 Applications

441

ν 2 Hence by Gr¨ onwall’s inequality we get D∗R g (rω) ≡ 0, so that 1 ν ν N −1 ν g (rω) ≡ 0, ∀r ∈ [R , R ] , ∀ω ∈ S . Thus ∂/∂r ∂ D∗R 1 2 ∗R1 g (rω)/∂r 1 ≡ 0, ∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 . And by (16.188) we have θ (rω) g (rω) ≡ 0, ∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 , implying g (rω) ≡ 0, ∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 . Hence we proved f1 (rω) = f2 (rω), ∀r ∈ [R1 , R2 ] , ∀ω ∈ S N −1 . Thus f1 (x) = f2 (x), ∀x ∈ A, there by proving the claim. We give the very important Remark 16.86. From Corollary 16.12 we saw that: for ν ≥ 0, n := ν, f ∈ AC n ([a, b]), given that Daν f (x) exists in R, ∀x ∈ [a, b] , and f (k) (a) = ν f = Daν f. Also we saw in Theorems 0, k = 0, 1, . . . , n − 1, imply that D∗a 16.16 and 16.17, that by adding to the assumptions of Theorem 16.16 that “there exists Daν f (x) ∈ R, ∀x ∈ [a, b],” we can rewrite the conclusions of Theorem 16.16; that is, we can frame the conclusions of Theorem 16.17 in the language of Riemann – Liouville fractional derivatives. Notice there, γ f (x), ∀x ∈ under the above additional assumption, that Daγ f (x) = D∗a [a, b] also holds. This entire chapter is based on Theorem 16.16. So by adding to the assumptions of all of our results here for all functions involved that “there exists Daν f (x) ∈ R, ∀x ∈ [a, b],” we can rewrite them all in terms of Riemann – Liouville fractional derivatives. Accordingly for the case of the ν f (rω) ∈ R, ∀r ∈ [R1 , R2 ] , spherical shell we need to add “there exists DR 1 N −1 for each ω ∈ S ,” and all can be rewritten in terms of Riemann – Liouville radial fractional derivatives. As examples we next present only a few of all those results that can be rewritten. We present Theorem 16.87. Let ν ≥ γ + 1, γ ≥ 0. Call n := ν and assume f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and ∃Daν f (x) ∈ R, ∀x ∈ [a, b] with Daν f ∈ L∞ (a, b). Let p, q > 1 such that 1/p + 1/q = 1, a ≤ x ≤ b. Then x

|Daγ f (ω)| |(Daν f ) (ω)| dω ≤ a (pν−pγ−p+2)/p

(x − a) √ 1/p q 2 Γ (ν − γ) ((pν − pγ − p + 1) (pν − pγ − p + 2))

442

16. Caputo Fractional Multivariate Opial Inequalities

2/q

x

·

q

|Daν f (ω)| dω

.

(16.198)

a

Proof. Similar to Theorem 16.18.

The converse result follows. Theorem 16.88. Let ν ≥ γ + 1, γ ≥ 0. Call n := ν and assume f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and ∃Daν f (x) ∈ R, ∀x ∈ [a, b] with Daν f, 1/Daν f ∈ L∞ (a, b). Suppose that Daν f is of ﬁxed sign a.e in [a, b] . Let p, q such that 0 < p < 1, q < 0 and 1/p + 1/q = 1, a ≤ x ≤ b. Then x |Daγ f (ω)| |Daν f (ω)| dω ≥ a (pν−pγ−p+2)/p

(x − a) √ 1/p q 2 Γ (ν − γ) ((pν − pγ − p + 1) (pν − pγ − p + 2)) ·

2/q

x

q

|Daν f (ω)| dω

.

(16.199)

a

Proof. As in Theorem 16.20. We present

Theorem 16.89. Let ν ≥ γ i + 1, γ i ≥ 0, i = 1, 2, n := ν, and f1 , (j) (j) f2 ∈ AC n ([a, b]) such that f1 (a) = f2 (a) = 0, j = 0, 1, . . . , n − 1, a ≤ x ≤ b. Consider also p (t) > 0 and q (t) ≥ 0, with all p (t), 1/p (t), q (t) ∈ L∞ (a, b). Further assume ∃Daν fi (x) ∈ R, ∀x ∈ [a, b], and Daν fi ∈ L∞ (a, b), i = 1, 2. Let λν > 0 and λα , λβ ≥ 0 such that λν < p, where p > 1. Here Pk is as in (16.39), A (ω) is as in (16.40), A0 (x) as in (16.41), and δ 1 as in (16.42). If λβ = 0, we obtain that x + λ λ q (ω) |Daγ 1 f1 (ω)| α |Daν f1 (ω)| ν + a

|Daγ 1 f2 (ω)|

λα

λν

|Daν f2 (ω)|

A0 (x) |λβ =0

λν λα + λν

,

dω ≤

λν /p δ1

(16.200)

16.6 Applications

)

443

*((λα +λν )/p)

x

p

p

p (ω) [|Daν f1 (ω)| + |Daν f2 (ω)| ] dω

.

a

Proof. As in Theorem 16.30.

Corollary 16.90. (All are as in Theorem 16.89, λβ = 0, p (t) = q (t) = 1, λα = λν = 1, p = 2.) In detail: let ν ≥ γ 1 + 1, γ 1 ≥ 0, n := ν, f1 , (j) (j) f2 ∈ AC n ([a, b]) : f1 (a) = f2 (a) = 0, j = 0, 1, . . . , n − 1, a ≤ x ≤ b; ν ∃Da fi (x) ∈ R, ∀x ∈ [a, b] with Daν fi ∈ L∞ (a, b), i = 1, 2. Then x [|(Daγ 1 f1 ) (ω)| |(Daν f1 ) (ω)| + a

|(Daγ 1 f2 ) (ω)| |(Daν f2 ) (ω)|] dω ≤ (ν−γ 1 )

(x − a) √ √ 2Γ (ν − γ 1 ) ν − γ 1 2ν − 2γ 1 − 1 x

+

2 ((Daν f1 ) (ω))

+

2 ((Daν f2 ) (ω))

,

(16.201)

dω ,

a

all a ≤ x ≤ b. We need Deﬁnition 16.91. Let F : A → R, ν ≥ 0, n := ν such that F (·ω) ∈ AC n ([R1 , R2 ]), for all ω ∈ S N −1 . We call the Riemann – Liouville radial fractional derivative the following function r ν F (x) ∂R ∂n 1 n−ν−1 1 := (r − t) F (tω) dt, (16.202) ∂rν Γ (n − ν) ∂rn R1 where x ∈ A; that is, x = rω, r ∈ [R1 , R2 ] , ω ∈ S N −1 . Clearly 0 ∂∗R F (x) 1 = F (x), ∂r0 and ν ∂R F (x) ∂ ν F (x) 1 = , if ν ∈ N. ∂rν ∂rν We give Proposition 16.92. Let ν ≥ γ + 1, γ ≥ 0, n := ν, f : A → R with f ∈ L1 (A), where A := B (0, R2 ) − B (0, R1 ) ⊆ RN , N ≥ 2, 0 < R1 < R2 . ν f (·ω)/∂rν Assume that f (·ω) ∈ AC n ([R1 , R2 ]), ∀ω ∈ S N −1 , and that ∂R 1 N −1 ν ∈ L∞ (R1 , R2 ), ∀ω ∈ S . Further assume that ∃DR1 f (rω) ∈ R, ∀r ∈

444

16. Caputo Fractional Multivariate Opial Inequalities

ν [R1 , R2 ] , for each ω ∈ S N −1 , with ∂R f (x)/∂rν ∈ L∞ (A). We suppose 1 N −1 ν that ∃M1 > 0 such that DR f (rω) ≤ M1 . ∀r ∈ [R1 , R2 ] and ∀ω ∈ S 1 Suppose that

∂ j f (R1 ω) = 0, j = 0, 1, . . . n − 1, ∀ω ∈ S N −1 . ∂rj Set

r

2(ν−γ−1) (1−N )

(r − t)

P (r) :=

t

dt, R1 ≤ r ≤ R2 ,

(16.203)

R1

also

r(N −1/2) P (r) , Γ (ν − γ) 1/2

(16.204)

A (r) := 10 (R2 ) := A

R2

2

(A (r)) dr

.

(16.205)

R1

Then

ν γ ∂R1 f (x) ∂R f (x) 1 ∂rγ ∂rν dx ≤ A 2 ν ∂R f (x) −1/2 1 10 (R2 ) 2 A dx . ∂rν A

Proof. Similarly as in Proposition 16.82.

(16.206)

17 Poincar´e-Type Fractional Inequalities

Here we present Poincar´e-type fractional inequalities involving fractional derivatives of Canavati, Riemann–Liouville, and Caputo types. The results are general Lp inequalities forward and reverse, univariate and multivariate, on a spherical shell. We give applications to ODEs and PDEs. We present also mean Poincar´e-type fractional inequalities. This treatment relies on [57].

17.1 Introduction This chapter is motivated by the famous Poincar´e inequality; see [2]. Given a bounded domain Ω ⊂ Rn , n ∈ N, it holds that uLp (Ω) ≤ C ∇uLp (Ω) for functions u with vanishing mean value over Ω, 1 ≤ p ≤ ∞, under very general assumptions on Ω, where ∇uLp (Ω) is deﬁned as the Lp -norm of the Euclidean norm of ∇u. Especially in [2] it is proven for a convex domain Ω ⊂ Rn with diameter d that d uL1 (Ω) ≤ ∇uL1 (Ω) 2 for any u with zero mean value on Ω, with the constant 1/2 also being optimal. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 445 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 17,

446

17. Poincar´e-Type Fractional Inequalities

We are also motivated by [126], where the authors prove the following. Let 1 < p < ∞, −∞ < a < b < ∞. The best constant C (independent of a, b) for which the one-dimensional Poincar´e inequality b f (t) dt 2−1/p ≤ C (b − a) f Lp ([a,b]) f − a b−a 1 L ([a,b])

−1/p

holds for all Lipschitz continuous functions f, is C = 1/2 (1 + p ) , where p > 1 : 1/p + 1/p = 1. Here we present Poincar´e-type inequalities with respect to fractional derivatives of Canavati, Riemann–Liouville, and Caputo types. We give univariate results on a closed interval of R, and multivariate ones on a spherical shell. The surprising fact here is that we require only vanishing initial conditions of the derivatives up to a certain order for the involved function.

17.2 Fractional Poincar´e Inequalities Results Let [a, b] ⊆ R; here see [19, 101]. Let x, x0 ∈ [a, b] such that x ≥ x0 , x0 is ﬁxed, f ∈ C ([a, b]), and deﬁne x 1 ν−1 x0 (x − t) f (t) dt, (Jν f ) (x) := Γ (ν) x0 all x0 ≤ x ≤ b, where ν > 0, n := [ν] (integral part), α := ν − n, the generalized Riemann–Liouville integral, with Γ the gamma function. We consider

x0 f (n) ∈ C 1 ([x0 , b]) . Cxν0 ([a, b]) := f ∈ C n ([a, b]) : J1−α Let f ∈ Cxν0 ([a, b]); we deﬁne the generalized ν-fractional derivative of f over [x0 , b] as x0 Dxν0 f := J1−α f (n) , the derivative with respect to x. Clearly Dxν0 f ∈ C ([x0 , b]). We need Lemma 17.1. ([19, pp. 544–545]) Let ν ≥ γ + 1, γ ≥ 0, n := [ν], and f ∈ Cxν0 ([a, b]), x0 ∈ [a, b]. Assume f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Then x γ 1 ν−γ−1 ν Dx0 f (x) = Dx0 f (t) dt, (x − t) (17.1) Γ (ν − γ) x0

17.2 Fractional Poincar´e Inequalities Results

447

all x ∈ [x0 , b]. We give Poincar´e inequalities with respect to the above-deﬁned Canavatitype fractional derivative. Theorem 17.2. Assumptions are as in Lemma 17.1. Let p, q > 1 : 1/p + 1/q = 1. It holds b γ Dx f (x)q dx ≤ 0

x0

'

(b − x0 )

(

q(ν−γ)

q

(q/p)

(Γ (ν − γ)) (p (ν − γ − 1) + 1)

q (ν − γ)

b

x0

ν Dx f (t)q dt . 0 (17.2)

Proof. We have γ Dx f (x) ≤ 0

1 Γ (ν − γ)

x

ν−γ−1

(x − t) x0

ν Dx f (t) dt 0

(17.3)

(by the H¨ older inequality) 1 ≤ Γ (ν − γ)

x

x0

1/p

x

p(ν−γ−1)

(x − t)

dt

x0

ν Dx f (t)q dt 0

(ν−γ−1)+1/p

(x − x0 ) 1 Γ (ν − γ) (p (ν − γ − 1) + 1)1/p

1/q

x

x0

= ν Dx f (t)q dt 0

(ν−γ−1)+1/p

(x − x0 ) 1 ≤ Γ (ν − γ) (p (ν − γ − 1) + 1)1/p

(17.4)

b

x0

1/q (17.5)

ν Dx f (t)q dt 0

1/q .

That is, we have γ Dx f (x) ≤ 0

(x − x0 )

(ν−γ−1)+1/p

Γ (ν − γ) (p (ν − γ − 1) + 1)

1/p

b

x0

ν Dx f (t)q dt 0

1/q , (17.6)

∀x ∈ [x0 , b]. Consequently we ﬁnd γ Dx f (x)q ≤ 0

q(ν−γ−1)+q/p

(x − x0 ) q

q/p

(Γ (ν − γ)) (p (ν − γ − 1) + 1)

448

17. Poincar´e-Type Fractional Inequalities

b

x0

q ν Dx f (t) dt , 0

(17.7)

∀x ∈ [x0 , b]. Hence we obtain

b

x0

γ Dx f (x)q dx ≤ 0

'

q(ν−γ−1)+(q/p)+1

(b − x0 ) q

(q/p)

(Γ (ν − γ)) (p (ν − γ − 1) + 1)

1 q (ν − γ − 1) +

q p

⎤ b q ν Dx f (t) dt . ⎦ 0 x0 +1

·

(17.8)

Corollary 17.3. (to Theorem 17.2) When γ = 0 we ﬁnd b qν (b − x0 ) q |f (x)| dx ≤ q (q/p) x0 (Γ (ν)) (p (ν − 1) + 1) qν

b

x0

q ν Dx f (t) dt . 0

(17.9)

Let α > 0, f ∈ L1 (a, x), a, x ∈ R, see [45, 64, 134]. We deﬁne the generalized Riemann–Liouville fractional derivative of f of order α by m s d 1 m−α−1 Daα f (s) := (s − t) f (t) dt, Γ (m − α) ds a where m := [α] + 1, s ∈ [a, x], see also later Remark 17.46. In addition, we set Da0 f := f := J0a f, a f := Daα f, if α > 0, J−α

Da−α f := Jαa f, if 0 < α ≤ 1, Dan f = f (n) , for n ∈ N. We need Deﬁnition 17.4. [45] We say that f ∈ L1 (a, x) has an L∞ fractional derivative Daα f (α > 0) in [a, x], a, x ∈ R, iﬀ Daα−k f ∈ C ([a, x]), k = 1, . . . , m := [α] + 1, and Daα−1 f ∈ AC ([a, x]) (absolutely continuous functions) and Daα f ∈ L∞ (a, x).

17.2 Fractional Poincar´e Inequalities Results

449

Lemma 17.5. [45] Let β > α ≥ 0, f ∈ L1 (a, x) , a, x ∈ R, have an L∞ fractional derivative Daβ f in [a, x]; let Daβ−k f (a) = 0 for k = 1, . . . , [β]+1. Then s 1 β−α−1 Daα f (s) = (s − t) Daβ f (t) dt, (17.10) Γ (β − α) a ∀s ∈ [a, x]. Here Daα f ∈ AC ([a, x]) for β − α ≥ 1, and Daα f ∈ C ([a, x]) for β − α ∈ (0, 1). We present Poincar´e inequalities with respect to the above-deﬁned generalized Riemann–Liouville fractional derivative. Theorem 17.6. Let all the assumptions be the same as in Lemma 17.5; p, q > 1 : 1/p + 1/q = 1 with p (β − α − 1) + 1 > 0. Then ' ( x q(β−α) (x − a) q α |Da f (t)| dt ≤ q (q/p) a (Γ (β − α)) (p (β − α − 1) + 1) q (β − α) x β Da f (t)q dt . (17.11) a

Corollary 17.7. All are as in Theorem 17.6. Let α = 0; then it holds that ' ( x qβ (x − a) q |f (t)| dt ≤ q (q/p) a (Γ (β)) (p (β − 1) + 1) qβ x β Da f (t)q dt . (17.12) a

Next we produce Poincar´e inequalities with respect to the Caputo fractional derivative. We mention Deﬁnition 17.8. [134] Let ν ≥ 0, n := ν; · is the ceiling of the number f ∈ AC n ([a, b]). The Caputo fractional derivative is given by x 1 n−ν−1 (n) ν D∗a f (x) := (x − t) f (t) dt, (17.13) Γ (n − ν) a ∀x ∈ [a, b]. ν The above function D∗a f (x) exists almost everywhere for x ∈ [a, b]. We need

450

17. Poincar´e-Type Fractional Inequalities

Proposition 17.9. [134] Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]). Then ν D∗a f exists iﬀ the generalized Riemann–Liouville fractional derivative Daν f exists. Proposition 17.10. [134] Let ν ≥ 0, n = ν. Assume that f is such that ν both D∗a f and Daν f exist. Suppose that f (k) (a) = 0 for k = 0, 1, . . . , n − 1. Then ν f = Daν f. D∗a In conclusion ν Corollary 17.11. [58] Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]), D∗a f exists ν (k) or Da f exists, and f (a) = 0, k = 0, 1, . . . , n − 1. Then ν Daν f = D∗a f.

We need Proposition 17.12. [58] Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]), and f (a) = 0, k = 0, 1, . . . , n − 1. Then x 1 ν−1 ν f (x) = (x − t) D∗a f (t) dt. (17.14) Γ (ν) a (k)

We also need Theorem 17.13. [58] Let ν ≥ γ + 1, γ ≥ 0. Call n := ν. Assume ν f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and D∗a f ∈ γ f ∈ C ([a, b]), and L∞ (a, b). Then D∗a x 1 ν−γ−1 γ ν f (x) = (x − t) D∗a f (t) dt, (17.15) D∗a Γ (ν − γ) a ∀x ∈ [a, b]. Theorem 17.14. [58] Let ν ≥ γ +1, γ ≥ 0, n := ν. Let f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1. Assume ∃Daν f (x) ∈ R, ∀x ∈ [a, b], and Daν f ∈ L∞ (a, b). Then Daγ f ∈ C ([a, b]), and x 1 ν−γ−1 Daγ f (x) = (x − t) Daν f (t) dt, (17.16) Γ (ν − γ) a ∀x ∈ [a, b]. Now we are ready to give

17.2 Fractional Poincar´e Inequalities Results

451

Theorem 17.15. Let ν ≥ γ+1, γ ≥ 0, n := ν. Assume f ∈ AC n ([a, b]) ν such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and D∗a f ∈ L∞ (a, b). Let p, q > 1 : 1/p + 1/q = 1. Then b

q

γ |D∗a f (x)| dx ≤

a

'

(b − a)

(

q(ν−γ)

q

(q/p)

(Γ (ν − γ)) (p (ν − γ − 1) + 1)

q (ν − γ)

b ν |D∗a f

a

q

(x)| dx . (17.17)

Proof. Similar to Theorem 17.2. We also give

Theorem 17.16. Let ν ≥ γ + 1, γ ≥ 0, n := ν. Let f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1. Assume ∃Daν f (x) ∈ R, ∀x ∈ [a, b], and Daν f ∈ L∞ (a, b). Let p, q > 1 : 1/p + 1/q = 1. Then b

q

|Daγ f (x)| dx ≤ a

'

(b − a)

(

q(ν−γ)

q

(q/p)

(Γ (ν − γ)) (p (ν − γ − 1) + 1)

q (ν − γ)

b

|Daν f

q

(x)| dx .

a

(17.18) Proof. Similar to Theorem 17.2.

When γ = 0 we get Proposition 17.17. Same assumptions as in Theorem 17.15; γ = 0. Then b q |f (x)| dx ≤ a

(b − a) q

qν (q/p)

(Γ (ν)) (p (ν − 1) + 1)

qν

a

b

q

ν |D∗a f (x)| dx .

(17.19)

Similarly we have Proposition 17.18. Same assumptions as in Theorem 17.16; γ = 0. Then b q |f (x)| dx ≤ a

452

17. Poincar´e-Type Fractional Inequalities

(b − a)

qν

q

(q/p)

(Γ (ν)) (p (ν − 1) + 1)

b

|Daν f

qν

q

(x)| dx .

(17.20)

a

Special cases of the above results follow. Corollary 17.19. (To Theorem 17.2; p = q = 2.) It holds

b

x0

'

(b − x0 )

2 Dxγ0 f (x) dx ≤ (

2(ν−γ)

2

(Γ (ν − γ)) (2 (ν − γ) − 1) 2 (ν − γ)

b

x0

ν 2 Dx0 f (x) dx .

(17.21)

Corollary 17.20. (To Theorem 17.6; p = q = 2.) Assume β − α > 1/2. Then x 2 (Daα f (t)) dt ≤ a

'

(x − a)

(

2(β−α)

2

(Γ (β − α)) (2 (β − α) − 1) 2 (β − α)

x

β 2 Da f (t) dt .

(17.22)

a

Corollary 17.21. (To Theorem 17.15; p = q = 2.) It holds

b

a

'

(b − a)

2

γ (D∗a f (x)) dx ≤

(

2(ν−γ)

2

(Γ (ν − γ)) (2 (ν − γ) − 1) 2 (ν − γ)

b

2

ν (D∗a f (x)) dx .

a

(17.23)

Corollary 17.22. (To Theorem 17.16; p = q = 2.) It holds

b

2

(Daγ f (x)) dx ≤ '

a

(b − a)

2(ν−γ)

2

(Γ (ν − γ)) (2 (ν − γ) − 1) 2 (ν − γ)

(

b

(Daν f

2

(x)) dx .

(17.24)

a

Next we give converse results. Theorem 17.23. Let ν ≥ γ + 1, γ ≥ 0, n := [ν], f ∈ Cxν0 ([a, b]), x0 ∈ [a, b]. Assume f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Assume that Dxν0 f is of ﬁxed strict sign on [x0 , b]. Let 0 < p < 1, q < 0, such that 1/p + 1/q = 1.

17.2 Fractional Poincar´e Inequalities Results

Then

b

γ Dx f (t)−q dt ≥ 0

453

1 −q

(Γ (ν − γ)) −1 (−q(ν−γ)+2) b (b − x0 ) Dxν f (t)q d . (17.25) 0 (1−q) x0 (p (ν − γ − 1) + 1) (−q (ν − γ) + 2) x0

Proof. Here Dxγ0 f (x0 ) = 0 by (17.1), but by assumption Dxγ0 f (x) = 0, ∀x ∈ (x0 , b]. By Lemma 17.1 and assumptions of the theorem we have x γ 1 ν−γ−1 ν Dx f (x) = (17.26) Dx0 f (t) dt (x − t) 0 Γ (ν − γ) x0 (by reverse H¨older’s inequality) 1 ≥ Γ (ν − γ)

1/p

x

p(ν−γ−1)

(x − t)

x

dt

x0

x0

(ν−γ−1)+1/p

1 (x − x0 ) ≥ Γ (ν − γ) (p (ν − γ − 1) + 1)1/p

b

x0

ν Dx f (t)q dt 0

ν Dx f (t)q dt 0

1/q (17.27)

1/q .

(17.28)

So that γ Dx f (x) ≥ 0

(ν−γ−1)+1/p

1 (x − x0 ) Γ (ν − γ) (p (ν − γ − 1) + 1)1/p

b

x0

ν Dx f (t)q dt 0

1/q , (17.29)

∀x ∈ [x0 , b]. Therefore

γ Dx f (x)−q ≥ 0

(x − x0 )

−q(ν−γ−1)−q/p −q/p

(p (ν − γ − 1) + 1) ∀x ∈ [x0 , b]. Finally we obtain b

−q

(Γ (ν − γ)) b

x0

γ Dx f (x)−q dx ≥ 0

x0

(b − x0 )

−q(ν−γ−1)−q/p+1

−q/p

(p (ν − γ − 1) + 1)

1

ν Dx f (t)q dt 0

−1 ,

1 −q

(Γ (ν − γ))

(−q (ν − γ − 1) − q/p + 1)

(17.30)

b

x0

ν Dx f (t)q dt 0

−1 ,

(17.31) proving the claim.

454

17. Poincar´e-Type Fractional Inequalities

Corollary 17.24. (To Theorem 17.23; γ = 0.) It holds b 1 −q |f (t)| dt ≥ −q (Γ (ν)) x0 −1 (−qν+2) b q (b − x0 ) ν Dx f (t) dt . 0 (1−q) x0 (p (ν − 1) + 1) (−qν + 2)

(17.32)

We continue with Theorem 17.25. Let β > α ≥ 0, f ∈ L1 (a, x), a, x ∈ R, have an L∞ fractional derivative Daβ f in [a, x], and let Daβ−k f (a) = 0 for k = 1, . . . , [β] + 1. Assume that Daβ f has ﬁxed sign a.e. on [a, x], and 1/Daβ f ∈ L∞ (a, x). Let 0 0. Then x 1 −q α |Da f (t)| dt ≥ −q (Γ (β − α)) a −1 x (−q(β−α)+2) β (x − a) D f (t)q dt . (17.33) a (1−q) a (p (β − α − 1) + 1) (−q (β − α) + 2) Proof. Similar to Theorem 17.23; use (17.10).

Corollary 17.26. (To Theorem 17.25; α = 0.) It holds x 1 −q |f (t)| dt ≥ −q (Γ (β)) a −1 x (−qβ+2) β (x − a) Da f (t)q dt . (1−q) a (p (β − 1) + 1) (−qβ + 2)

(17.34)

We present Theorem 17.27. Let ν ≥ γ + 1, γ ≥ 0, n := ν, f ∈ AC n ([a, b]) and ν f is of ﬁxed sign a.e on f (a) = 0, k = 0, 1, . . . , n − 1. Assume that D∗a ν ν [a, b], and D∗a f , 1/D∗a f ∈ L∞ (a, b). Let 0 < p < 1, q < 0, such that 1/p + 1/q = 1. Then b 1 −q γ |D∗a f (t)| dt ≥ −q (Γ (ν − γ)) a −1 (−q(ν−γ)+2) b (b − a) q ν |D∗a f (t)| dt . (17.35) (1−q) a (p (ν − γ − 1) + 1) (−q (ν − γ) + 2) (k)

17.2 Fractional Poincar´e Inequalities Results

455

Proof. Similar to Theorem 17.23; use (17.15).

Corollary 17.28. (To Theorem 17.27; γ = 0.) It holds

b

−q

|f (t)|

dt ≥

(Γ (ν))

a

(b − a)

(−qν+2)

(1−q)

(p (ν − 1) + 1)

(−qν + 2)

1 −q

−1

b ν |D∗a f

a

q

(t)| dt

.

(17.36)

We give Theorem 17.29. Let ν ≥ γ + 1, γ ≥ 0, n := ν, f ∈ AC n ([a, b]), and f (k) (a) = 0, k = 0, 1, . . . , n − 1. Suppose ∃Daν f (x) ∈ R, ∀x ∈ [a, b], and Daν f , 1/Daν f ∈ L∞ (a, b). Here Daν f is of ﬁxed sign a.e on [a, b]. Let 0 < p < 1, q < 0, such that 1/p + 1/q = 1. Then b 1 −q |Daγ f (t)| dt ≥ −q (Γ (ν − γ)) a −1 (−q(ν−γ)+2) b (b − a) q ν |Da f (t)| dt . (17.37) (1−q) a (p (ν − γ − 1) + 1) (−q (ν − γ) + 2)

Proof. As in Theorem 17.23; use (17.16).

Corollary 17.30. (To Theorem 17.29; γ = 0.) It holds

b

−q

|f (t)|

dt ≥

(Γ (ν))

a

(b − a)

(−qν+2)

(1−q)

(p (ν − 1) + 1)

(−qν + 2)

1

−q

−1

b

|Daν f

q

(t)| dt

.

(17.38)

a

Next we treat the easy but important case of p = 1. We present for Canavati-type fractional derivatives Theorem 17.31. Let ν ≥ γ + 1, γ ≥ 0, n := ν, and f ∈ Cxν0 ([a, b]), x0 ∈ [a, b]. Assume f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Then b ν−γ b γ (b − x0 ) Dxν f (t) dt. (17.39) Dx0 f (t) dt ≤ (1) 0 Γ (ν − γ + 1) x0 x0

456

17. Poincar´e-Type Fractional Inequalities

Setting γ = 0 we get

ν

(b − x0 ) |f (t)| dt ≤ Γ (ν + 1) x0

(2)

b

ν Dx f (t) dt. 0

b

x0

(17.40)

Proof. By (17.1) we have γ Dx f (x) ≤ 0

1 Γ (ν − γ)

x

ν−γ−1

(x − t) x0

ν−γ−1

(x − x0 ) ≤ Γ (ν − γ)

b

x0

ν Dx f (t) dt 0

ν Dx f (t) dt, 0

(17.41)

∀x ∈ [x0 , b]. Thus by integrating (17.41) over [x0 , b] we ﬁnd (17.39); setting there γ = 0 we obtain (17.40). Similarly we ﬁnd for Riemann–Liouville fractional derivatives Theorem 17.32. Let β > α + 1, α ≥ 0, f ∈ L1 (a, x), a, x ∈ R, have an L∞ fractional derivative Daβ f in [a, x], and let Daβ−k f (a) = 0 for k = 1, . . . , [β] + 1. Then

β−α

x

|Daα f (t)| dt ≤

(1) a

(x − a) Γ (β − α + 1)

x

β Da f (t) dt.

(17.42)

β Da f (t) dt.

(17.43)

a

When α = 0 we ﬁnd

β

x

|f (t)| dt ≤

(2) a

(x − a) Γ (β + 1)

x

a

Proof. As in Theorem 17.31; based on Lemma 17.5.

Also we have for Caputo derivatives Theorem 17.33. Let ν ≥ γ+1, γ ≥ 0, n := ν. Assume f ∈ AC n ([a, b]) ν f ∈ L∞ (a, b). Then such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and D∗a 1) a

ν−γ

b γ |D∗a f (t)| dt ≤

(b − a) Γ (ν − γ + 1)

b

a

ν |D∗a f (t)| dt.

(17.44)

When γ = 0 we get

ν

b

|f (t)| dt ≤

2) a

(b − a) Γ (ν + 1)

a

b ν |D∗a f (t)| dt.

(17.45)

17.3 Applications of Fractional Poincar´e Inequalities

Proof. As in Theorem 17.31; based on Theorem 17.13.

457

Coming back to Riemann–Liouville derivatives we get Theorem 17.34. Let ν ≥ γ + 1, γ ≥ 0, n := ν. Let f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1. Assume ∃Daν f (x) ∈ R, ∀x ∈ [a, b], and Daν f ∈ L∞ (a, b). Then

ν−γ

b

|Daγ f

(1) a

(b − a) (t)| dt ≤ Γ (ν − γ + 1)

b

|Daν f (t)| dt.

(17.46)

a

When γ = 0 we get b ν b (b − a) |f (t)| dt ≤ |Daν f (t)| dt. (2) Γ (ν + 1) a a Proof. As in Theorem 17.31; based on Theorem 17.14.

(17.47)

17.3 Applications of Fractional Poincar´e Inequalities Next we apply some of the above results to the multivariate case of the spherical shell. # $ Let N ≥ 2, S N −1 := x ∈ RN : |x| = 1 the unit sphere on RN, where |·| stands for the Euclidean norm in RN. Also denote the ball $ # B (0, R) := x ∈ RN : |x| < R ⊆ RN, R > 0, and the spherical shell A := B (0, R2 ) − B (0, R1 ), 0 < R1 < R2 . For the following see [356, pp. 149–150] and [383, pp. 87–88]. For x ∈ RN − {0} we can write uniquely x = rω, where r = |x| > 0, and ω = x/r ∈ S N −1 , |ω| = 1. Clearly here RN − {0} = (0, ∞) × S N −1 , also

A = [R1 , R2 ] × S N −1 .

In the sequel the following theorem is used frequently.

458

17. Poincar´e-Type Fractional Inequalities

Theorem 17.35. Let f : A → R be a Lebesgue integrable function. Then R2

f (x) dx = A

S N −1

f (rω) rN −1 dr dω.

(17.48)

R1

So we are able to write an integral in polar form using the polar coordinates (r, ω). Regarding the Canavati-type fractional derivative we need Deﬁnition 17.36. (see [44]) Let ν > 0, n := [ν], α := ν −n, f ∈ C n A , ν f (x)/∂rν ∈ C A , and A is a spherical shell. Assume that there exists ∂R 1 given by r ν n ∂R f (x) ∂ 1 −α ∂ f (tω) 1 := (r − t) dt , (17.49) ∂rν Γ (1 − α) ∂r ∂rn R1 where x ∈ A; that is, x = rω, r ∈ [R1 , R2 ], ω ∈ S N −1 . ν f /∂rν the radial Canavati-type fractional derivative of f of We call ∂R 1 ν f (x)/∂rν := f (x). order ν. If ν = 0, then set ∂R 1 We also need Lemma 17.37. (see [44]) Let γ ≥ 0, ν ≥1 such that ν − γ ≥ 1. Let f ∈ ν ν C n A and there exists ∂R f (x)/∂r ∈ C A , x ∈ A, and A a spherical 1 shell. Further assume that ∂ j f (R1 ω)/∂rj = 0, j = 0, 1, .. . , n − 1, n := [ν], γ f (x)/∂rγ ∈ C A . and ∀ω ∈ S N −1 . Then there exists ∂R 1 We present the following Canavati-type fractional Poincar´e inequalities on the spherical shell. Theorem 17.38. Let ν ≥ γ + 1, γ ≥ 0, n := [ν], and f ∈ C n A and ν f (x)/∂rν ∈ C A , x ∈ A, and A a spherical shell. Further there exists ∂R 1 assume that ∂ j f (R1 ω)/∂rj = 0, j = 0, 1, . . . , n − 1; ∀ω ∈ S N −1 . Let p, q > 1 : 1/p + 1/q = 1. Then N −1 γ ∂R1 f (x) q dx ≤ R2 1) ∂rγ R1 A ' ( q q(ν−γ) ν ∂R f (x) (R2 − R1 ) 1 ∂rν dx. q q/p A q (ν − γ) (Γ (ν − γ)) (p (ν − γ − 1) + 1) (17.50) When γ = 0 we get N −1 R2 q 2) |f (x)| dx ≤ R1 A

17.3 Applications of Fractional Poincar´e Inequalities

'

( q ν ∂R f (x) 1 ∂rν dx. q q/p A qν (Γ (ν)) (p (ν − 1) + 1)

459

qν

(R2 − R1 )

(17.51)

Proof. Because R1 ≤ r ≤ R2 , N ≥ 2, we have that R1N −1 ≤ rN −1 ≤ and R21−N ≤ r1−N ≤ R11−N . Hence R2 q γ R21−N rN −1 DR f (rω) dr ≤ 1

R2N −1

R1

R2

R1

(17.2)

≤

q γ r1−N rN −1 DR f (rω) dr = 1

R2

R1

γ D f (rω)q dr R1 (R2 − R1 )

q(ν−γ)

q

q/p

(Γ (ν − γ)) (p (ν − γ − 1) + 1) q (ν − γ) R2 q ν r1−N rN −1 DR f (rω) dr ≤ 1 R1

(R2 − R1 )

q(ν−γ)

R11−N

q

q/p

(Γ (ν − γ)) (p (ν − γ − 1) + 1) q (ν − γ) R2 q ν rN −1 DR f (rω) dr . 1

(17.52)

R1

So we have established that N −1 R2 q R2 N −1 γ DR1 f (rω) dr ≤ r R1 R1 q(ν−γ) (R2 − R1 ) q

q/p

(Γ (ν − γ)) (p (ν − γ − 1) + 1) q (ν − γ) R2 q N −1 ν DR1 f (rω) dr , ∀ω ∈ S N −1 . r R1

Then by integration of (17.53) on S N −1 we obtain N −1 R2 q N −1 R2 γ D f (rω) r dr dω ≤ R1 R1 S N −1 R1 '

(R2 − R1 ) q

(

q(ν−γ) q/p

q (ν − γ) (Γ (ν − γ)) (p (ν − γ − 1) + 1)

(17.53)

460

17. Poincar´e-Type Fractional Inequalities

S N −1

R2

R1

q N −1 ν DR f (rω) r dr dω. 1

Using (17.48) in (17.54) we have proven the claim.

(17.54)

Regarding the Riemann–Liouville fractional derivative we need Remark 17.39. Here we follow [383] and denote by λRN (x) ≡ dx the Lebesgue measure on RN , N ≥ 2, and by λS N −1 (ω) ≡ dω the surface class on space X. measure on S N −1 , where BX stands for the Borel Deﬁne the measure RN on (0, ∞) , B(0,∞) by RN (Γ) = rN −1 dr, any Γ ∈ B(0,∞) . Γ

Now let

F ∈ L1 (A) = L1 [R1 , R2 ] × S N −1 .

For a ﬁxed ω ∈ S N −1 , deﬁne gω (r) := F (rω) = F (x), where x ∈ A := B (0, R2 ) − B (0, R1 ), x 0 < R1 ≤ r ≤ R2 , r = |x| , ω = ∈ S N −1 . r By Fubini’s theorem and Theorem 5.2.2, pp. 87 and 88 of [383], we have gω ∈ L1 [R1 , R2 ] , B[R1 ,R2 ] , RN , for λS N −1 − almost every ω ∈ S N −1 . Call # $ K (F ) := ω ∈ S N −1 : gω ∈ / L1 [R1 , R2 ] , B[R1 ,R2 ] , RN # $ = ω ∈ S N −1 : F (·ω) ∈ (17.55) / L1 [R1 , R2 ] , B[R1 ,R2 ] , RN . That is, λS N −1 (K (F )) = 0. Of course, Θ (F ) := [R1 , R2 ] × K (F ) ⊂ A and λRN (Θ (F )) = 0. By the deﬁnition of the generalized Riemann–Liouville fractional derivative we then have (see also Remark 17.46 later) m d 1 β DR1 gω (r) = Γ (m − β) dr

17.3 Applications of Fractional Poincar´e Inequalities

r−R1

m−β−1

(r − R1 − t)

gω (R1 + t) dt,

461

(17.56)

0

where β > 0, m := [β] + 1, r ∈ [R1 , R2 ]. If β = 0, then

β DR g (r) = gω (r). 1 ω

Formula (17.56) is written for all ω ∈ S N −1 − K (F ). We set β DR (r) ≡ 0, ∀ω ∈ K (F ) , ∀r ∈ [R1 , R2 ] , any β > 0. g 1 ω

The above lead to the following deﬁnition (see [45]). Deﬁnition 17.40. Let β > 0, m := [β] + 1, F ∈ L1 (A), and A is the spherical shell. We deﬁne ⎧ β ⎨ F (x) ∂R 1 := ⎩ ∂rβ where

1 Γ(m−β)

d m r−R1 dr 0

m−β−1

(r − R1 − t) F ((R1 + t) ω) dt, for ω ∈ S N −1 − K (F ), 0, for ω ∈ K (F ), (17.57)

x = rω ∈ A, r ∈ [R1 , R2 ] , ω ∈ S N −1 .

If β = 0, deﬁne β ∂R F (x) 1 := F (x). ∂rβ β We call ∂R F (x)/∂rβ the Riemann–Liouville radial fractional derivative 1 of F of order β.

We need Theorem 17.41. (see [45]) Let β > α > 0 and F ∈ L1 (A). Assume that β F (x) ∂R 1 ∈ L∞ (A). ∂rβ β F (rω) takes real values for almost all r ∈ Further suppose that DR 1 N −1 , and for these [R1 , R2 ] , for each ω ∈ S β DR1 F (rω) ≤ M1

for some M1 > 0.

462

17. Poincar´e-Type Fractional Inequalities

For each ω ∈ S N −1 −K (F ), we assume that F (·ω) has an L∞ fractional β derivative DR F (·ω) in [R1 , R2 ], and that 1 β−k F (R1 ω) = 0, k = 1, . . . , [β] + 1. DR 1

Then

α F (x) α ∂R 1 = DR1 F (rω) = ∂rα r 1 β−α−1 β DR (r − t) F (tω) dt, 1 Γ (β − α) R1

(17.58)

true ∀x ∈ A; that is, true ∀r ∈ [R1 , R2 ] and ∀ω ∈ S N −1 . Here α DR1 F (·ω) is in AC ([R1 , R2 ]) for β − α ≥ 1 and

α DR F (·ω) is in C ([R1 , R2 ]) for β − α ∈ (0, 1), 1

∀ω ∈ S N −1 . Furthermore

α F (x) ∂R 1 ∈ L∞ (A) . ∂rα

In particular, it holds F (x) = F (rω) =

1 Γ (β)

r

(r − t) R1

β−1

β DR F (tω) dt, 1

(17.59)

∀r ∈ [R1 , R2 ], ∀ω ∈ S N −1 − K (F ); x = rω, and F (·ω) is in AC ([R1 , R2 ]) for β ≥ 1 and F (·ω) is in C ([R1 , R2 ]) for β ∈ (0, 1), ∀ω ∈ S

N −1

− K (F ).

We present Poincar´e inequalities on the shell involving Riemann – Liouville radial fractional derivatives. Theorem 17.42. Let all terms and assumptions be as in Theorem 17.41. Let p, q > 1 : 1/p + 1/q = 1 with p (β − α − 1) + 1 > 0. Then N −1 α ∂R1 F (x) q dx ≤ R2 (1) ∂rα R1 A ' ( β q(β−α) ∂ F (x) q (R2 − R1 ) R1 dx. β q q/p ∂r A q (β − α) (Γ (β − α)) (p (β − α − 1) + 1) (17.60)

17.3 Applications of Fractional Poincar´e Inequalities

463

When α = 0 we get

N −1 R2 (2) |F (x)| dx ≤ R1 A ' ( β qβ ∂ F (x) q (R2 − R1 ) R1 dx. q q/p ∂rβ A qβ (Γ (β)) (p (β − 1) + 1)

q

Proof. Similar to Theorem 17.38; based on Theorem 17.41.

(17.61)

Regarding the Caputo fractional derivative we mention Deﬁnition 17.43. (see [58]) Let F : A → R, ν ≥ 0, n := ν such that F (·ω) ∈ AC n ([R1 , R2 ]), for all ω ∈ S N −1 . We call the Caputo radial fractional derivative the following function r ν n F (x) ∂∗R 1 n−ν−1 ∂ F (tω) 1 := (r − t) dt, (17.62) ∂rν Γ (n − ν) R1 ∂rn where x ∈ A; that is, x = rω, r ∈ [R1 , R2 ], ω ∈ S N −1 . Clearly 0 ∂∗R F (x) 1 = F (x) , ∂r0 and ν ∂∗R F (x) ∂ ν F (x) 1 = , if ν ∈ N, the usual radial derivative. ν ∂r ∂rν

The above function (17.62) exists almost everywhere for x ∈ A. We mention the following fundamental representation result. Theorem 17.44. (see [58]) Let ν ≥ γ + 1, γ ≥ 0, n := ν, F : A → R with F ∈ L1 (A). Assume that F (·ω) ∈ AC n ([R1 , R2 ]) for all ω ∈ S N −1 , ν F (·ω)/∂rν ∈ L∞ (R1 , R2 ) for all ω ∈ S N −1 . Further assume and that ∂∗R 1 ν that ∂∗R1 F (x)/∂rν ∈ L∞ (A). More precisely, for these r ∈ [R1 , R2 ], for each ω ∈ S N −1 , for which ν F (rω) takes real values, there exists M1 > 0 such that D∗R 1 ν D∗R F (rω) ≤ M1 . 1 We suppose that ∂ j F (R1 ω)/∂rj = 0, j = 0, 1, . . . , n − 1, for every ω ∈ S N −1 . Then γ ∂∗R F (x) γ 1 = D∗R F (rω) = 1 ∂rγ

464

17. Poincar´e-Type Fractional Inequalities

1 Γ (ν − γ)

r

ν−γ−1

(r − t) R1

ν D∗R1 F (tω) dt,

true ∀x ∈ A; that is, true ∀r ∈ [R1 , R2 ] and ∀ω ∈ S N −1 , γ > 0. Here γ F (·ω) ∈ AC ([R1 , R2 ]), D∗R 1

(17.63)

(17.64)

∀ω ∈ S N −1 , γ > 0. Furthermore

γ F (x) ∂∗R 1 ∈ L∞ (A), γ > 0. ∂rγ In particular, it holds r 1 ν−1 ν D∗R1 F (tω) dt, (r − t) F (x) = F (rω) = Γ (ν) R1

(17.65)

(17.66)

true ∀x ∈ A; that is, true ∀r ∈ [R1 , R2 ] and ∀ω ∈ S N −1 , and F (·ω) ∈ AC ([R1 , R2 ]), ∀ω ∈ S N −1 .

(17.67)

We present Poincar´e inequalities on the shell involving Caputo radial fractional derivatives. Theorem 17.45. All terms and assumptions for f are as in Theorem 17.44. Let p, q > 1 : 1/p + 1/q = 1. Then N −1 γ ∂∗R1 f (x) q R2 1) ∂rγ dx ≤ R1 A ' ( q q(ν−γ) ν ∂∗R f (x) (R2 − R1 ) 1 ∂rν dx. q q/p A q (ν − γ) (Γ (ν − γ)) (p (ν − γ − 1) + 1) (17.68) When γ = 0 we get

q

|f (x)| dx ≤

2) A

'

R2 R1

N −1

( q ν ∂∗R f (x) 1 ∂rν dx. q q/p A qν (Γ (ν)) (p (ν − 1) + 1) (R2 − R1 )

qν

(17.69)

We make Remark 17.46. Regarding the two identical forms of the deﬁnition of generalized Riemann–Liouville fractional derivative, here f ∈ L1 (a, x), a < x; a, x ∈ R, iﬀ fa ∈ L1 (0, x − a), where fa (t) := f (a + t).

17.3 Applications of Fractional Poincar´e Inequalities

465

The basic Riemann–Liouville derivative of fa anchored at zero is deﬁned and given by m s d 1 m−ν−1 ν (s − t) f (a + t) dt, (17.70) (D fa ) (s ) = Γ (m − ν) ds 0 s ∈ [0, x − a], m := [ν] + 1, ν > 0. Another way to deﬁne the generalized Riemann–Liouville fractional derivative (see [45, 64]) is (Daν f ) (s) := (Dν fa ) (s − a) , for s ∈ [a, x].

(17.71)

But 0 ≤ s − a ≤ x − a, calling s := s − a, which is a one-to-one and onto map from [a, x] onto [0, x − a]; we have 0 ≤ s ≤ x − a. Consequently we get m d 1 (Dν fa ) (s − a) = Γ (m − ν) d (s − a) s−a m−ν−1 ((s − a) − t) f (a + t) dt = (17.72) 1 Γ (m − n)

0

d ds

m

s−a

m−ν−1

(s − (a + t))

f (a + t) dt

=: (∗).

0

(17.73) In [65, 134], it is proved that s m−ν−1 (s − t) |f (t)| dt < ∞ for a.e. s ∈ [a, x]; a

that is, there exists s

m−ν−1

(s − t)

f (t) dt for a.e. s ∈ [a, x].

a

Therefore from (17.73) we have m s 1 d m−ν−1 (∗) = (s − t) f (t) dt Γ (m − ν) ds 0

(17.74)

(change of variable) true a.e. for s ∈ [a, x]. Because the derivative is deﬁned only for real-valued functions (i.e., here only on existing integrals in (17.74)), we obviously get the identity of the two deﬁnitions. That is, Daν f of (17.71) is such that m s d 1 m−ν−1 (Daν f ) (s) = (s − t) f (t) dt, (17.75) Γ (m − ν) ds a m := [ν] + 1, ν > 0, s ∈ [a, x], as deﬁned earlier in this chapter.

466

17. Poincar´e-Type Fractional Inequalities

So based on the above Remark 17.46 we can redeﬁne the Riemann– Liouville radial fractional derivative as follows. Deﬁnition 17.47. Let β > 0, m := [β] + 1, F ∈ L1 (A), and A is the spherical shell. We deﬁne ⎧ ∂ m r m−β−1 1 β (r − t) F (tω) dt, ⎨ Γ(m−β) ∂r R1 ∂R F (x) 1 N −1 := (17.76) − K (F ), for ω ∈ S ⎩ ∂rβ 0, for ω ∈ K (F ), where x = rω ∈ A, r ∈ [R1 , R2 ] , ω ∈ S N −1 ; K (F ) as in (17.55). If β = 0, deﬁne β ∂R F (x) 1 := F (x). ∂rβ So functions (17.57) and (17.76) are identical.

We need the following important representation result for Riemann– Liouville radial fractional derivatives. Theorem 17.48. Let ν ≥ γ + 1, γ ≥ 0, n := ν, F : A → R with F ∈ L1 (A). Assume that F (·ω) ∈ AC n ([R1 , R2 ]), ∀ω ∈ S N −1 , and ν F (·ω)/∂rν is measurable on [R1 , R2 ], ∀ω ∈ S N −1 . Also assume that ∂R 1 ν ν ∃∂R F (rω)/∂rν ∈ R, ∀r ∈ [R1 , R2 ] and ∀ω ∈ S N −1 , and ∂R F (x)/∂rν is 1 1 measurable on A. Suppose ∃M1 > 0 : ν ∂R1 F (rω) ≤ M1 , ∀ (r, ω) ∈ [R1 , R2 ] × S N −1 . ∂rν We suppose that ∂ j F (R1 ω)/∂rj = 0, j = 0, 1, . . . , n − 1; ∀ω ∈ S N −1 . Then γ F (x) ∂R γ 1 = DR F (rω) = 1 ∂rγ r 1 ν−γ−1 ν DR1 F (tω) dt, (r − t) (17.77) Γ (ν − γ) R1 valid ∀x ∈ A; that is, true ∀r ∈ [R1 , R2 ] and ∀ω ∈ S N −1 ; γ > 0. Here γ F (·ω) ∈ AC ([R1 , R2 ]), DR 1 ∀ω ∈ S N −1 , γ > 0. Furthermore

γ F (x) ∂R 1 ∈ L∞ (A), γ > 0. ∂rγ

(17.78)

(17.79)

17.3 Applications of Fractional Poincar´e Inequalities

467

In particular, it holds F (x) = F (rω) =

1 Γ (ν)

r

ν−1

(r − t) R1

ν DR F (tω) dt, 1

(17.80)

true ∀x ∈ A; that is, true ∀r ∈ [R1 , R2 ] and ∀ω ∈ S N −1 , and F (·ω) ∈ AC ([R1 , R2 ]) , ∀ω ∈ S N −1 .

(17.81)

Proof. By our assumptions and Theorem 17.17, Corollary 14 of [58], we have (17.77) and (17.80). Also (17.78) is clear; see [45]. Property (17.81) is easy to prove. Fixing r ∈ [R1 , R2 ], the function ν−γ−1

δ r (t, ω) := (r − t) is measurable on

ν DR F (tω) 1

[R1 , r] × S N −1 , B [R1 ,r] × B S N −1 .

Here B [R1 ,r] × B S N −1 stands for the complete σ-algebra generated by B [R1 ,r] × B S N −1 , where B X stands for the completion of BX . Then we get that r |δ r (t, ω)| dt dω = S N −1

S N −1

r

R1 ν−γ−1

(r − t) R1

ν DR F (tω) dt dω ≤ 1

(17.82)

ν r ∂R1 F (x) ν−γ−1 (r − t) dt dω ∂rν S N −1 R1 ∞,([R1 ,r]×S N −1 ) ν ν−γ ∂R1 F (x) 2π N/2 (r − R1 ) = ≤ (17.83) ∂rν ∞,([R1 ,r]×S N −1 ) Γ (N/2) (ν − γ) ν ν−γ ∂R1 F (x) 2π N/2 (R2 − R1 ) < ∞. (17.84) ∂rν Γ (N/2) (ν − γ) ∞,A Hence δ r (t, ω) is integrable on [R1 , r] × S N −1 , B [R1 ,r] × B S N −1 . Consequently, by Fubini’s theorem and (17.77), we obtain that N −1 γ DR F (rω), ν ≥ γ + 1, γ > 0, is integrable in ω over S , B S N −1 . 1

468

17. Poincar´e-Type Fractional Inequalities

γ So we have that DR F (rω) is continuous in r ∈ [R1 , R2 ], ∀ω ∈ S N −1 , 1 and measurable in ω ∈ S N −1 , ∀r ∈ [R1 , R2 ] . So, it is a Carath´eodory function. Here [R1 , R2 ] is a separable metric space and S N −1 is a measurable space, and the function takes values in R∗ := R∪ {±∞}, is a metric γ which F (rω) is jointly space. Therefore by Theorem 20.15, p. 156 of [10], DR 1 B[R1 ,R2 ] × B S N −1 -measurable on [R1 , R2 ] × S N −1 = A, that is, Lebesgue measurable on A. Then we have that r γ 1 ν−γ−1 ν D F (rω) ≤ (17.85) DR1 F (tω) dt (r − t) R1 Γ (ν − γ) R1 ν r D F (·ω) R1 ∞,[R1 ,R2 ] ν−γ−1 ≤ (r − t) dt ≤ (17.86) Γ (ν − γ) R1 ν−γ

M1 (r − R1 ) Γ (ν − γ) (ν − γ)

≤

M1 ν−γ (R2 − R1 ) := τ < ∞, Γ (ν − γ + 1)

for all ω ∈ S N −1 and for all r ∈ [R1 , R2 ] . Hence we have shown that γ D F (rω) ≤ τ < ∞, ∀ω ∈ S N −1 and ∀r ∈ [R1 , R2 ] R1

(17.87)

and

γ ∂R F (x) 1 ∈ L∞ (A) , γ > 0. ∂rγ We have completed our proof.

Next we give Poincar´e inequalities on the shell involving Riemann–Liouville radial fractional derivatives. Theorem 17.49. Let all terms and assumptions for f be as in Theorem 17.48, and p, q > 1 : 1/p + 1/q = 1. Then N −1 γ ∂R1 f (x) q dx ≤ R2 (1) ∂rγ R1 A ' ( q q(ν−γ) ν ∂R f (x) (R2 − R1 ) 1 ∂rν dx. q q/p A q (ν − γ) (Γ (ν − γ)) (p (ν − γ − 1) + 1) (17.88) When γ = 0 we get N −1 R2 q (2) |f (x)| dx ≤ R1 A ' ( q qν ν ∂R f (x) (R2 − R1 ) 1 (17.89) ∂rν dx. q q/p A qν (Γ (ν)) (p (ν − 1) + 1)

17.3 Applications of Fractional Poincar´e Inequalities

469

Next we apply (17.69) to a Caputo fractional PDE. Theorem 17.50. Let 1 ≤ ν 1 < ν 2 < · · · < ν k , nk := ν k , k ∈ N, F : A → R with F ∈ L1 (A). Assume that F (·ω) ∈ AC nk ([R1 , R2 ]), ∀ω ∈ νl F (·ω)/∂rν l ∈ L∞ (R1 , R2 ), ∀ω ∈ S N −1 ; l = 1, . . . , k. S N −1 , and that ∂∗R 1 νl Suppose ∂∗R1 F (x)/∂rν l ∈ L∞ (A), l = 1, . . . , k, such that ∃Ml > 0 : ν D l F (rω) ≤ Ml , l = 1, . . . , k, a.e. r ∈ [R1 , R2 ], ∀ω ∈ S N −1 . Also ∗R1 suppose ∂ j F (R1 ω)/∂rj = 0, j = 0, 1, . . . , nk − 1, ∀ω ∈ S N −1 . Let also B, Cl , 1/Cl ∈ L∞ (A), l = 1, . . . , k, with all Cl ≥ 0. The above-described F fulﬁlls k

Cl (x)

l=1

νl ∂∗R F (x) 1 ∂rν l

2 = B (x), ∀x ∈ A.

(17.90)

Deﬁne 1 δ := 2

R2 R1

N −1 max

2ν l

(R2 − R1 ) 2

ν l (Γ (ν l )) (2ν l − 1) 1 . ρ := max 1≤l≤k Cl L (A) ∞ 1≤l≤k

:

Then F L2 (A) ≤

δρ k

B (x) dx

,

1/2 .

(17.91)

(17.92)

(17.93)

A

Proof. Set nl := ν l , l = 1, . . . , k − 1. Clearly, F (·ω) ∈ AC nl ([R1 , R2 ]), ∀ω ∈ S N −1 ; l = 1, . . . , k − 1. Also ∂ j F (R1 ω) = 0, j = 0, 1, . . . , nl − 1, ∀ω ∈ S N −1 , ∂rj as nl ≤ nk , for all l = 1, . . . , k − 1. So all assumptions of Theorem 17.44 are fulﬁlled for F and fractional orders ν l , l = 1, . . . , k. Thus by choosing p = q = 2 we apply (17.69), and we obtain for l = 1, . . . , k that

2

(F (x)) dx ≤ A

'

(R2 − R1 )

2ν l

(

R2 R1

N −1

νl ∂∗R F (x) 1 ∂rν l

2

dx 2 2ν l (Γ (ν l )) (2ν l − 1) A νl 2 ∂∗R1 F (x) −1 ≤δ (Cl (x)) (Cl (x)) dx ∂rν l A

(17.94)

470

17. Poincar´e-Type Fractional Inequalities

νl 2 1 ∂∗R1 F (x) ≤ δ C (x) dx l Cl ∂rν l L∞ (A) A νl 2 ∂∗R1 F (x) ≤ δρ Cl (x) dx. ∂rν l A

(17.95)

(17.96)

That is,

2

(F (x)) dx ≤ δρ

Cl (x)

A

A

νl ∂∗R F (x) 1 ∂rν l

2 dx,

(17.97)

for all l = 1, . . . , k. Consequently, summing (17.97) over all possible l values gives νl 2 k ∂∗R1 F (x) δρ 2 dx (F (x)) dx ≤ Cl (x) k A ∂rν l A l=1

δρ k

= proving the claim.

B (x) dx,

(17.98)

A

In the simpler case of an ordinary diﬀerential equation we apply (17.19) next. Theorem 17.51. Let 1 ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (j) (a) = 0, j = 0, 1, . . . , nk − 1, and νl f ∈ L∞ (a, b), for all l = 1, . . . , k. Let also B, Cl , 1/Cl ∈ L∞ (a, b), D∗a l = 1, . . . , k, with all Cl ≥ 0. The above described f satisﬁes k

2

νl Cl (x) (D∗a f (x)) = B (x) , ∀x ∈ [a, b].

(17.99)

l=1

Call 1 δ ∗ := max 2 1≤l≤k

(b − a)

2ν l

2

ν l (Γ (ν l )) (2ν l − 1) 1 ∗ . ρ := max 1≤l≤k Cl L (a,b) ∞

Then

: f L2 (a,b) ≤

δ ∗ ρ∗ k

,

(17.100) (17.101)

1/2

b

B (x) dx

.

(17.102)

a

Proof. Set nl := ν l , l = 1, . . . , k − 1. Clearly here f ∈ AC nl ([a, b]), l = 1, . . . , k − 1. Also f (j) (a) = 0, j = 0, 1, . . . , nl − 1, l = 1, . . . , k − 1. So

17.3 Applications of Fractional Poincar´e Inequalities

471

all assumptions of Theorem 17.15 are fulﬁlled for f and fractional orders ν l , l = 1, . . . , k − 1. Thus by choosing p = q = 2 we apply (17.19), for l = 1, . . . , k, to ﬁnd b 2ν b (b − a) l 2 2 νl (f (x)) dx ≤ (D∗a f (x)) dx 2 2ν l (Γ (ν l )) (2ν l − 1) a a (17.103) b −1 2 νl ≤ δ∗ (Cl (x)) (Cl (x)) (D∗a f (x)) dx a

≤ δ ∗ ρ∗

b

That is,

b

(f (x)) dx ≤ δ ∗ ρ∗

2

νl Cl (x) (D∗a f (x)) dx.

a

b

2

a

(17.104)

2

νl Cl (x) (D∗a f (x)) dx,

a

(17.105)

for all l = 1, . . . , k. The last imply a

b

δ ∗ ρ∗ (f (x)) dx ≤ k

2

k

a

δ ∗ ρ∗ = k proving the claim.

b

νl Cl (x) (D∗a f

(x))

2

dx

l=1

b

B (x) dx,

(17.106)

a

One can give similar applications to ODEs and PDEs involving fractional derivatives of Canavati-type and Riemann–Liouville-type. We ﬁnish this section with L1 results on the shell. We present for Canavati-type radial fractional derivatives Theorem 17.52. Let ν ≥ γ + 1, γ ≥ 0, n := [ν], and f ∈ C n A and ν f (x)/∂rν ∈ C A , x ∈ A, and A a spherical shell. Further there exists ∂R 1 assume ∂ j f (R1 ω)/∂rj = 0, j = 0, 1, . . . , n − 1; ∀ω ∈ S N −1 . Then N −1 γ ∂R1 f (x) dx ≤ R2 (1) ∂rγ R1 A ν−γ ν ∂ f (x) (R2 − R1 ) R1 dx. (17.107) Γ (ν − γ + 1) A ∂rν When γ = 0 we get N −1 ν ν ∂R1 f (x) R2 (R2 − R1 ) (2) |f (x)| dx ≤ ∂rν dx. R1 Γ (ν + 1) A A

(17.108)

472

17. Poincar´e-Type Fractional Inequalities

Proof. Because R1 ≤ r ≤ R2 , N ≥ 2, we have that R21−N ≤ r1−N ≤ Hence R2 γ rN −1 DR f (rω) dr ≤ R21−N 1

R11−N .

R1

R2

r

1−N N −1

r

R1 (17.39)

≤

γ D f (rω) dr = R1 ν−γ

ν−γ

(R2 − R1 ) Γ (ν − γ + 1)

R11−N (R2 − R1 ) Γ (ν − γ + 1)

R2

R1 R2

R1

R2

R1

γ D f (rω) dr R1

(17.109)

ν r1−N rN −1 DR f (rω) dr ≤ 1 ν rN −1 DR f (rω) dr, 1

(17.110)

∀ω ∈ S N −1 . We have proven so far

R2

R1

γ D f (rω) rN −1 dr ≤ R1

R2

R1

R2 R1

N −1

ν−γ

(R2 − R1 ) Γ (ν − γ + 1)

ν DR f (rω) rN −1 dr, ∀ω ∈ S N −1 . 1

(17.111)

Thus by integration of (17.111) on S N −1 we ﬁnd R2 N −1 γ D f (rω) r dr dω ≤ R1 S N −1

R2 R1

N −1

R1

ν−γ

(R2 − R1 ) Γ (ν − γ + 1)

S N −1

R2

R1

ν DR f (rω) rN −1 dr dω . 1 (17.112)

Using (17.48) on (17.112) we have proven the claim.

We present for Riemann–Liouville radial fractional derivatives Theorem 17.53. Let all terms and assumptions for f be as in Theorem 17.41; here β > α + 1, α ≥ 0. Then N −1 α β−α β ∂R1 f (x) ∂R1 f (x) R (R − R ) 2 2 1 (1) dx. ∂rα dx ≤ R1 Γ (β − α + 1) A ∂rβ A (17.113) When α = 0 we get N −1 β β R2 (R2 − R1 ) ∂R1 f (x) (2) |f (x)| dx ≤ dx. (17.114) R1 Γ (β + 1) A ∂rβ A

17.4 Fractional Mean Poincar´e Inequalities

Proof. Similar to Theorem 17.52, using (17.42).

473

We give for Caputo radial fractional derivatives Theorem 17.54. Let all terms and assumptions for f be as in Theorem 17.44. Then N −1 γ ν−γ ν ∂∗R1 f (x) ∂∗R1 f (x) R2 (R2 − R1 ) dx. (1) ∂rγ dx ≤ R1 Γ (ν − γ + 1) A ∂rν A (17.115) When γ = 0 we obtain

|f (x)| dx ≤

(2) A

R2 R1

N −1

ν

(R2 − R1 ) Γ (ν + 1)

ν ∂∗R1 f (x) ∂rν dx. A

Proof. Similar to Theorem 17.52, using (17.44).

(17.116)

Going back to Riemann–Liouville radial fractional derivatives we have Theorem 17.55. Let all terms and assumptions for f be as in Theorem 17.48. Then N −1 γ ν−γ ν ∂R1 f (x) ∂R1 f (x) R2 (R2 − R1 ) dx. (1) ∂rγ dx ≤ R1 Γ (ν − γ + 1) A ∂rν A (17.117) When γ = 0 we obtain

|f (x)| dx ≤

(2) A

R2 R1

N −1

ν

(R2 − R1 ) Γ (ν + 1)

ν ∂R1 f (x) ∂rν dx. A

Proof. Similar to Theorem 17.52, using (17.46).

(17.118)

17.4 Fractional Mean Poincar´e Inequalities First we give Canavati-type fractional derivatives related results. Theorem 17.56. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := [ν k ],
474

17. Poincar´e-Type Fractional Inequalities

i = 0, 1, . . . , nk − 1; p, q > 1 : 1/p + 1/q = 1. Call ' ( (ν −γ) (b − x0 ) l Λγ := max . 1≤l≤k (Γ (ν − γ)) (p (ν − γ − 1) + 1)1/p (q (ν − γ))1/q l l l (17.119) Then k γ ν Λ γ Dxl f (1) Dx0 f L (x ,b) ≤ . (17.120) 0 Lq (x0 ,b) q 0 k l=1

When γ = 0 we get (2) f Lq (x0 ,b) ≤ where

Λ0 k

k

' Λ0 := max

1≤l≤k

ν Dxl f 0 L

,

q (x0 ,b)

(17.121)

l=1

(b − x0 )

(

νl 1/p

(Γ (ν l )) (p (ν l − 1) + 1)

(qν l )

1/q

.

(17.122)

Proof. Here we apply Theorem 17.2. By (17.2) we obtain ' ( (ν l −γ) γ ) (b − x 0 Dx f ≤ 0 Lq (x0 ,b) 1/p 1/q (Γ (ν l − γ)) (p (ν l − γ − 1) + 1) (q (ν l − γ)) ν Dxl f ≤ Λγ Dxν0l f L (x ,b) . (17.123) 0 L (x ,b) q

0

That is, for l = 1, . . . , k holds γ Dx f 0 L (x q

0 ,b)

q

≤ Λγ Dxν0l f L

q (x0 ,b)

Consequently by addition of (17.124) we have k γ Dxν l f ≤ Λγ k Dx f 0

0

Lq (x0 ,b)

0

Lq (x0 ,b)

.

(17.124)

,

(17.125)

l=1

proving (17.120). We continue with Riemann–Liouville fractional derivatives related results. Theorem 17.57. Let 0 ≤ α < β 1 < β 2 < . . . < β k , k ∈ N; f ∈ β L1 (a, x), a, x ∈ R, have L∞ fractional derivatives Da l f in [a, x], with

17.4 Fractional Mean Poincar´e Inequalities

475

β −k

Da l f (a) = 0 for k = 1, . . . , [β l ] + 1; l = 1, . . . , k. Let p, q > 1 : 1/p + 1/q = 1 with p (β l − α − 1) + 1 > 0, all l = 1, . . . , k. Put ' ( (β −α) (x − a) l Kα := max , 1≤l≤k (Γ (β − α)) (p (β − α − 1) + 1)1/p (q (β − α))1/q l l l (17.126) and ' ( β (x − a) l K0 := max . (17.127) 1≤l≤k (Γ (β )) (p (β − 1) + 1)1/p (qβ )1/q l l l Then (1)

Daα f Lq (a,x)

≤

Kα k

k

β Da l f . Lq (a,x)

(17.128)

l=1

When α = 0 we obtain (2) f Lq (a,x)

K0 ≤ k

k Daβ l f L

q (a,x)

.

(17.129)

l=1

Proof. Similar to Theorem 17.56, now using Lemma 17.5 and Theorem 17.6; see (17.11) there. We continue with Caputo fractional derivative related results. Theorem 17.58. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , nk −1, νl f ∈ L∞ (a, b), l = 1, . . . , k. Let p, q > 1 : 1/p + 1/q = 1. Set and D∗a % (ν −γ) (b − a) l Mγ := max , (1/p) 1/q 1≤l≤k (Γ (ν l − γ)) (p (ν l − γ − 1) + 1) (q (ν l − γ)) (17.130) and % ν (b − a) l M0 := max . (17.131) (1/p) 1/q 1≤l≤k (Γ (ν l )) (p (ν l − 1) + 1) (qν l ) Then (1)

γ D∗a f Lq (a,b)

≤

Mγ k

k l=1

νl D∗a f Lq (a,b)

.

(17.132)

476

17. Poincar´e-Type Fractional Inequalities

When γ = 0 we obtain (2) f Lq (a,b) ≤

M0 k

k

νl D∗a f Lq (a,b)

.

(17.133)

l=1

Proof. Similar to Theorem 17.56; based on (17.17).

We also give a related average result again regarding Riemann–Liouville fractional derivatives. Theorem 17.59. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , nk −1. Assume ∃Daν l f (x) ∈ R, ∀x ∈ [a, b], and Daν l f ∈ L∞ (a, b), for all l = 1, . . . , k. Let p, q > 1 : 1/p + 1/q = 1. Here Mγ , M0 as in (17.130), (17.131), respectively. Then (1) Daγ f Lq (a,b) ≤

Mγ k

k

Daν l f Lq (a,b) .

(17.134)

l=1

When γ = 0 we have (2) f Lq (a,b) ≤

M0 k

k

Daν l f Lq (a,b)

.

(17.135)

l=1

Proof. Similar to Theorem 17.56; based on (17.18).

Next we present converse fractional mean Poincar´e inequalities. First we give Canavati-type fractional derivative related results. Theorem 17.60. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := [ν k ],
1≤l≤k

-

(b − x0 )((ν l −γ)−2/q)

, Γ (ν l − γ) (p (ν l − γ − 1) + 1)(1−1/q) (−q (ν l − γ) + 2)(−1/q) (17.136)

and % θ0 := min

1≤l≤k

-

(ν l −2/q)

(b − x0 )

(1−1/q)

Γ (ν l ) (p (ν l − 1) + 1)

(−qν l + 2)

(−1/q)

.

(17.137)

17.4 Fractional Mean Poincar´e Inequalities

Then (1) Dxγ0 f L

(−q) (x0 ,b)

θγ ≥ k

477

k ν Dxl f 0 L

q (x0 ,b)

.

(17.138)

l=1

When γ = 0 we get (2) f L(−q) (x0 ,b)

θ0 ≥ k

k ν Dxl f 0 L

q (x0 ,b)

.

(17.139)

l=1

Proof. By (17.25) we obtain (b − x0 )((ν l −γ)−2/q) Dx0l f Lq (x0 ,b) ν

Dxγ0 f L

(−q) (x0 ,b)

≥

Γ (ν l − γ) (p (ν l − γ − 1) + 1)(1−1/q) (−q (ν l − γ) + 2)(−1/q)

≥ θγ Dxν0l f L

.

(17.140)

q (x0 ,b)

That is,

γ Dx f 0 L

(−q) (x0 ,b)

≥ θγ Dxν0l f Lq (x0 ,b) ;

for all l = 1, . . . .k. Consequently by addition of (17.141) we have k γ Dxν l f ≥ θγ k Dx f L(−q) (x0 ,b)

0

Lq (x0 ,b)

0

(17.141)

,

(17.142)

l=1

proving the claim.

We continue related results.

with

Riemann–Liouville

fractional

derivatives

Theorem 17.61. Let 0 ≤ α < β 1 < β 2 < . . . < β k , k ∈ N; f ∈ β L1 (a, x), a, x ∈ R, have L∞ fractional derivatives Da l f in [a, x], with β l −k Da f ( a) = 0 for k = 1, . . . , [β l ] + 1; l = 1, . . . , k. Let 0 0, all l = 1, . . . , k. β β Further assume that each Da l f has ﬁxed sign a.e. on [a, x], and 1/Da l f ∈ L∞ (a, x), for all l = 1, . . . , k. Set % θ∗α

:= min

1≤l≤k

and θ∗0

Γ (β l − α) (p (β l − α − 1) + 1)(1−1/q) (−q (β l − α) + 2)(−1/q) (17.143)

% := min

1≤l≤k

-

(x − a)((β l −α)−2/q)

-

(β l −2/q)

(x − a)

(1−1/q)

Γ (β l ) (p (β l − 1) + 1)

(−qβ l + 2)

(−1/q)

.

(17.144)

478

17. Poincar´e-Type Fractional Inequalities

Then (1) Daα f L(−q) (a,x)

θ∗ ≥ α k

When α = 0 we obtain (2) f L(−q) (a,x)

θ∗ ≥ 0 k

k β Da l f . Lq (a,x)

(17.145)

l=1

k β Da l f L

.

q (a,x)

(17.146)

l=1

Proof. Similar to Theorem 17.60; based on (17.33).

We continue with Caputo fractional derivatives related results. Theorem 17.62. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , nk −1, νl νl νl and D∗a f , 1/D∗a f ∈ L∞ (a, b), l = 1, . . . , k. Suppose each D∗a f is of ﬁxed sign a.e. on [a, b] ; all l = 1, . . . , k. Let 0 < p < 1, q < 0, such that 1/p + 1/q = 1. Set % Aγ := min

, Γ (ν l − γ) (p (ν l − γ − 1) + 1)(1−1/q) (−q (ν l − γ) + 2)(−1/q) (17.147)

1≤l≤k

and

%

A0 := min

1≤l≤k

-

(ν l −2/q)

(b − a)

(1−1/q)

Γ (ν l ) (p (ν l − 1) + 1)

Then (1)

-

(b − a)((ν l −γ)−2/q)

γ f L(−q) (a,b) D∗a

Aγ ≥ k

k

(−qν l + 2)

(−1/q)

.

(17.148)

νl D∗a f Lq (a,b)

.

(17.149)

l=1

When γ = 0 we get (2) f L(−q) (a,b)

A0 ≥ k

k

νl D∗a f Lq (a,b)

.

(17.150)

l=1

Proof. Similar to Theorem 17.60; based on (17.35).

We come back to Riemann–Liouville fractional derivatives next. Theorem 17.63. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , nk −1.

17.5 Applications of Fractional Mean Poincar´e Inequalities

479

Suppose ∃Daν l f (x) ∈ R, ∀x ∈ [a, b], and Daν l f , 1/Daν l f ∈ L∞ (a, b) ; l = 1, . . . , k. Assume each Daν l f is of ﬁxed sign a.e. on [a, b], all l = 1, . . . , k. Let 0 < p < 1, q < 0 : 1/p + 1/q = 1. Here Aγ is as in (17.147), and A0 is as in (17.148). Then k A γ (1) Daγ f L(−q) (a,b) ≥ Daν l f Lq (a,b) . (17.151) k l=1

When γ = 0 we get (2) f L(−q) (a,b)

A0 ≥ k

k

Daν l f Lq (a,b)

.

(17.152)

l=1

Proof. Similar to Theorem 17.60; based on (17.37).

17.5 Applications of Fractional Mean Poincar´e Inequalities Next we apply some of the results of Section 17.4 to the multivariate case of the spherical shell. All terminology and symbols are as in Section 17.3. We present the following Canavati-type fractional radial mean Poincar´e inequalities on the spherical shell. Theorem 17.64. ≤ ν 1 < ν 2 < . . .< ν k , nk := [ν k ], Let γ ≥ 0 and 1+γ νl νl k ∈ N. Let f ∈ C n A ; there exist ∂R f (x)/∂r ∈ C A , x ∈ A, A and a 1 spherical shell, for all l = 1, . . . , k. Further assume that ∂ j f (R1 ω)/∂rj = 0, j = 0, 1, . . . , nk − 1; ∀ω ∈ S N −1 . Let p, q > 1 : 1/p + 1/q = 1. Set '

%

1 γ := max Λ

1≤l≤k

-(

(R2 − R1 )(ν l −γ) (q (ν l − γ))1/q (Γ (ν l − γ)) (p (ν l − γ − 1) + 1)1/p

R2 R1

(N −1)/q , (17.153)

and 1 0 := Λ

'

% max

1≤l≤k

-(

νl

(R2 − R1 ) 1/q

(qν l )

1/p

(Γ (ν l )) (p (ν l − 1) + 1)

R2 R1

((N −1)/q) . (17.154)

Then

k ν γ 1γ ∂R1 f ∂Rl1 f Λ 1) γ . ≤ ∂rν l ∂r k Lq (A) Lq (A) l=1

(17.155)

480

17. Poincar´e-Type Fractional Inequalities

When γ = 0 we get 2) f Lq (A)

10 Λ ≤ k

k ν ∂Rl f 1 ∂rν l l=1

.

(17.156)

Lq (A)

Proof. By (17.50) we get γ νl ∂R1 f ∂ f 1 γ R1 ≤ Λ , ∂rγ ∂rν l Lq (A) Lq (A)

(17.157)

for all l = 1, . . . , k. Adding all (17.157) we derive (17.155). Setting γ = 0 in (17.155) we obtain (17.156). Next follow Riemann–Liouville fractional radial mean Poincar´e inequalities on the spherical shell. Theorem 17.65. Let 0 ≤ α < β 1 < β 2 < . . . < β k , k ∈ N; f ∈ L1 (A), β β ∂Rk1 f (x)/∂rβ k ∈ L∞ (A). Further assume that DRl1 f (rω) takes real values β N −1 for almost all r ∈ [R1 , R2 ], for each ω ∈ S , and for these DRl1 f (rω) ≤ Ml , where Ml > 0, for all l = 1, . . . , k. For each ω ∈ S N −1 − K (f ), we β assume that f (·ω) have L∞ fractional derivatives DRl1 f (·ω) in [R1 , R2 ], and that β −ρ DRl1 f (R1 ω) = 0, ρ = 1, . . . , [β l ] + 1, (17.158) for all l = 1, . . . , k. Let p, q > 1 : 1/p + 1/q = 1 with p (β l − α − 1) + 1 > 0, all l = 1, . . . , k. Set (N −1/q) R2 1 Kγ := R1 % % -(β −α) (R2 − R1 ) l max , 1/q 1/p 1≤l≤k (q (β l − α)) (Γ (β l − α)) (p (β l − α − 1) + 1) (17.159) and % - (N −1/q) % βl R − R ) (R 2 2 1 1 0 := K max . 1/q 1/p 1≤l≤k R1 (qβ l ) (Γ (β l )) (p (β l − 1) + 1) (17.160) Then ⎛ ⎞ α k βl 1γ ∂R1 f ∂R1 f K ⎝ ⎠. (1) (17.161) ≤ β ∂rα ∂r l k Lq (A) l=1

Lq (A)

17.5 Applications of Fractional Mean Poincar´e Inequalities

481

When α = 0 we get (2) f Lq (A)

⎛ k βl 10 ∂R1 f K ⎝ ≤ β ∂r l k l=1

⎞ ⎠.

(17.162)

Lq (A)

Proof. As in Theorem 17.64; based on Theorems 17.41 and 17.42. Next are Caputo fractional radial mean Poincar´e inequalities on the spherical shell. Theorem 17.66. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , k ∈ N; f : A → R with f ∈ L1 (A). Assume that f (·ω) ∈ AC nk ([R1 , R2 ]) νl f (·ω)/∂rν l ∈ L∞ (R1 , R2 ) for all ω ∈ for all ω ∈ S N −1 , and that ∂∗R 1 νl F (x)/∂rν l ∈ L∞ (A), for S N −1 , l = 1, . . . , k. Further assume that ∂∗R 1 all l = 1, . . . , k. More precisely, for these r ∈ [R1 , R2 ], for each ω ∈ νl f (rω) takes real values, there exist Ml > 0 such S N −1, for which D∗R 1 νl that D∗R1 f (rω) ≤ Ml , for all l = 1, . . . , k. We suppose that ∂ j f (R1 ω)/ ∂rj = 0, j = 0, 1, . . . , nk − 1, for every ω ∈ S N −1 . Let p, q > 1 such that 1/p + 1/q = 1. 1 0 as in (17.154). Then 1 γ as in (17.153), and Λ Here Λ k ν γ l 1γ ∂∗R1 f ∂∗R f Λ 1 (1) ≤ . ∂rγ ∂rν l k Lq (A) Lq (A)

(17.163)

l=1

When γ = 0 we get (2) f Lq (A)

10 Λ ≤ k

k ν l ∂∗R f 1 ∂rν l l=1

.

(17.164)

Lq (A)

Proof. Similar to Theorem 17.64, using Theorems 17.44 and 17.45.

We ﬁnish this chapter by returning to Riemann–Liouville fractional radial mean Poincar´e inequalities on the spherical shell. Theorem 17.67. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , k ∈ N; f : A → R with f ∈ L1 (A). Assume that f (·ω) ∈ AC nk ([R1 , R2 ]) νl f (·ω)/∂rν l is measurable on [R1 , R2 ], ∀ω ∈ ∀ ω ∈ S N −1 , and that ∂R 1 νl N −1 ; l = 1, . . . , k. Also assume ∃∂R f (rω)/∂rν l ∈ R, ∀r ∈ [R1 , R2 ] S 1 νl N −1 νl and ∀ω ∈ S , and ∂R1 f (x)/∂r is , k. measurable on A; all l = 1, . .N. −1 νl νl ≤ Ml , ∀ (r, ω) ∈ [R1 , R2 ] × S Suppose ∃Ml > 0 : ∂R1 f (rω)/∂r ; all l = 1, . . . , k. We suppose that ∂ j f (R1 ω)/∂rj = 0, j = 0, 1, . . . , nk − 1,

482

17. Poincar´e-Type Fractional Inequalities

1 γ as in (17.153), and Λ 1 0 as ∀ω ∈ S N −1 . Let p, q > 1 : 1/p + 1/q = 1, Λ in (17.154). Then k ν γ 1γ ∂Rl1 f ∂R1 f Λ ≤ . (17.165) (1) γ ∂rν l ∂r k Lq (A) Lq (A) l=1

When γ = 0 we obtain (2) f Lq (A)

10 Λ ≤ k

k ν ∂Rl f 1 ∂rν l l=1

.

(17.166)

Lq (A)

Proof. Similar to Theorem 17.64, using Theorems 17.48 and 17.49.

18 Various Sobolev-Type Fractional Inequalities

Here we present various univariate Sobolev-type fractional inequalities involving fractional derivatives of Canavati, Riemann–Liouville, and Caputo types. The results are general Lp inequalities forward and converse on a closed interval. We give an application to a fractional ODE. We present also the mean Sobolev-type fractional inequalities. This treatment relies on [56].

18.1 Introduction This chapter is motivated by the famous Sobolev-type inequality (see [159, p. 263]), the Gagliardo–Nirenberg–Sobolev inequality: Assume 1 ≤ p < n. Then there exists a constant C, depending only on p and n, such that uLp∗ (Rn ) ≤ C DuLp (Rn ) , (R ) . for all u ∈ Here p∗ := np/n − p, p∗ > p, and Du is the gradient of u. Also we are motivated by the following result (p. 265, [159]): estimates for Sobolev space W 1,p , 1 ≤ p < n : let U be a bounded open subset of Rn , and suppose the boundary ∂U is C 1 . Assume 1 ≤ p < n, and u ∈ W 1,p (U ). ∗ Then u ∈ Lp (U ), with the estimate Cc1

n

uLp∗ (U ) ≤ C uW 1,p (U ) , the constant C depending only on p, n, and U. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 483 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 18,

484

18. Various Sobolev-Type Fractional Inequalities

So here we derive general fractional Sobolev-type inequalities on a closed interval of the real line. As a typical case we mention our results that involve the Caputo fractional derivative. Please see the following results of Theorem 18.21, Corollary 18.22, Theorem 18.23, and Theorem 18.24, the converse inequality in Theorem 18.33, and the application in Theorem 18.39. We also produce corresponding results for the Canavati-type and Riemann–Liouville-type fractional derivatives.

18.2 Various Univariate Sobolev-Type Fractional Inequalities Let [a, b] ⊆ R here; see [19, 101]. Let x, x0 ∈ [a, b] such that x ≥ x0 , x0 is ﬁxed, f ∈ C ([a, b]), and deﬁne x 1 ν−1 (Jνx0 f ) (x) := (x − t) f (t) dt, Γ (ν) x0 all x0 ≤ x ≤ b, where ν > 0, n := [ν] (integral part), α := ν − n, the generalized Riemann–Liouville integral, with Γ the gamma function. We consider

x0 f (n) ∈ C 1 ([x0 , b]) . Cxν0 ([a, b]) := f ∈ C n ([a, b]) : J1−α Let f ∈ Cxν0 ([a, b]); we deﬁne the generalized ν-fractional derivative of f over [x0 , b] as x0 f (n) . Dxν0 f := J1−α Clearly Dxν0 f ∈ C ([x0 , b]) . We need Lemma 18.1. ([19, pp. 544–545]) Let ν ≥ γ + 1, γ ≥ 0, n := [ν], and f ∈ Cxν0 ([a, b]), x0 ∈ [a, b]. Assume f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Then

Dxγ0 f (x) =

1 Γ (ν − γ)

x

ν−γ−1

(x − t) x0

ν Dx0 f (t) dt,

(18.1)

all x ∈ [x0 , b]. We give Sobolev-type inequalities with respect to the above-deﬁned Canavati ([19, 101]) type fractional derivative.

18.2 Various Univariate Sobolev-Type Fractional Inequalities

485

Theorem 18.2. Let ν ≥ γ + 1, γ ≥ 0, n := [ν], and f ∈ Cxν0 ([a, b]), x0 ∈ [a, b] . Assume f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Let r ≥ 1; p, q > 1 : 1/p + 1/q = 1. Then γ Dx f ≤ 0 L (x ,b) r

0

ν Dx f 0 Lq (x0 ,b) + ,1/r . 1/p r ν − γ − 1q + 1 (Γ (ν − γ)) (p (ν − γ − 1) + 1) (b − x0 )

ν−γ+1/r−1/q

Proof. We have γ Dx f (x) ≤ 0

1 Γ (ν − γ)

x

ν Dx f (t) dt 0

ν−γ−1

(x − t) x0

(18.2)

(18.3)

(by the H¨ older inequality) ≤

1 Γ (ν − γ)

p(ν−γ−1)

(x − t)

dt

x0

x

x0

1/p

x

ν Dx f (t)q dt 0

1/q =

(18.4)

(ν−γ−1)+1/p

1 (x − x0 ) Γ (ν − γ) (p (ν − γ − 1) + 1)1/p 1/q x ν Dx f (t)q dt 0 x0

(ν−γ−1)+1/p

≤

(x − x0 ) 1 Γ (ν − γ) (p (ν − γ − 1) + 1)1/p 1/q b q ν Dx f (t) dt . x0

(18.5)

0

That is, γ Dx f (x) ≤ 0

'

(x − x0 )

Γ (ν − γ) (p (ν − γ − 1) + 1)

b

x0

∀x ∈ [x0 , b] .

(

(ν−γ−1)+1/p

ν Dx f (t)q dt 0

1/p

1/q ,

(18.6)

486

18. Various Sobolev-Type Fractional Inequalities

That is, γ Dx f (x) ≤ 0

(x − x0 )

(ν−γ−1/q)

Γ (ν − γ) (p (ν − γ − 1) + 1)

1/p

ν Dx f 0 L

q (x0 ,b)

,

(18.7)

∀x ∈ [x0 , b] . Hence, by r ≥ 1, we obtain γ Dx f (x)r ≤ 0

r(ν−γ−1/q)

(x − x0 ) r

r/p

(Γ (ν − γ)) (p (ν − γ − 1) + 1)

∀x ∈ [x0 , b]. Consequently it holds

b

x0

ν r Dx f 0 L

q (x0 ,b)

, (18.8)

γ Dx f (x)r dx ≤ 0

ν r Dx f 0 L (x ,b) + q 0 ,. r r/p r ν − γ − 1q + 1 (Γ (ν − γ)) (p (ν − γ − 1) + 1) r(ν−γ−1/q)+1

(b − x0 )

(18.9)

So that

γ Dx f ≤ 0 Lr (x0 ,b) ν−γ−1/q+1/r Dxν f (b − x0 ) 0 Lq (x0 ,b) + ,1/r , 1/p 1 r ν−γ− q +1 (Γ (ν − γ)) (p (ν − γ − 1) + 1)

proving the claim.

(18.10)

Corollary 18.3. (to Theorem 18.2) When γ = 0 it holds ν+1/r−1/q Dxν f (b − x0 ) 0 Lq (x0 ,b) f Lr (x0 ,b) ≤ + ,1/r . 1/p (Γ (ν)) (p (ν − 1) + 1) r ν − 1q + 1

(18.11)

Next we give the corresponding L1 result. Theorem 18.4. Let ν ≥ γ + 1, γ ≥ 0, n := [ν], and f ∈ Cxν0 ([a, b]), x0 ∈ [a, b] . Assume f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1; r ≥ 1. Then ν−γ−1+1/r Dxν f (b − x0 ) γ 0 L1 (x0 ,b) (1) Dx0 f L (x ,b) ≤ . (18.12) 1/r r 0 (Γ (ν − γ)) [r (ν − γ − 1) + 1] When γ = 0 we get (2) f Lr (x0 ,b) ≤

(b − x0 )

ν−1+1/r

ν Dx f 0 L

1 (x0 ,b) . 1/r

(Γ (ν)) [r (ν − 1) + 1]

(18.13)

18.2 Various Univariate Sobolev-Type Fractional Inequalities

487

Proof. By (18.1) we have γ Dx f (x) ≤ 0

1 Γ (ν − γ)

x

x0

ν−γ−1

(x − x0 ) ≤ Γ (ν − γ)

ν−γ−1

(x − t)

b

x0

ν Dx f (t) dt 0

ν Dx f (t) dt, 0

(18.14)

∀x ∈ [x0 , b] . That is, ν−γ−1 γ Dx f (x) ≤ (x − x0 ) Dxν f , 0 0 L1 (x0 ,b) Γ (ν − γ)

(18.15)

∀x ∈ [x0 , b] . Consequently r(ν−γ−1) γ Dxν f r Dx f (x)r ≤ (x − x0 ) , r 0 0 L1 (x0 ,b] (Γ (ν − γ))

(18.16)

∀x ∈ [x0 , b] and r(ν−γ−1)+1 Dxν f r (b − x0 ) r γ 0 L1 (x0 ,b) Dx f (x) dx ≤ , r 0 (Γ (ν − γ)) (r (ν − γ − 1) + 1) x0

b

proving the claim.

(18.17)

We continue with the following mean Sobolev-type inequalities result. Theorem 18.5. Let γ ≥ 0 and 1 + γ ≤ ν 1 < ν 2 < . . . < ν k , nk := [ν k ], 1 : 1/p + 1/q = 1. Call Λγ := max 1≤l≤k ⎡ ⎤ ⎢ ⎣

ν l −γ+1/r−1/q

(b − x0 )

⎥ + ,1/r ⎦ . 1/p r ν l − γ − 1q + 1 (Γ (ν l − γ)) [p (ν l − γ − 1) + 1]

Then (1) Dxγ0 f L

r (x0 ,b)

≤

Λγ k

k

(18.18)

ν Dxl f 0 L

q (x0 ,b)

.

(18.19)

l=1

When γ = 0 we ﬁnd (2) f Lr (x0 ,b) ≤

Λ0 k

k

ν Dxl f 0 L

q (x0 ,b)

l=1

,

(18.20)

488

18. Various Sobolev-Type Fractional Inequalities

where Λ0 := max

1≤l≤k

⎡ ⎢ ⎣

⎤

ν l +1/r−1/q

(b − x0 )

1/p

(Γ (ν l )) [p (ν l − 1) + 1]

⎥ + ,1/r ⎦ . 1 r νl − q + 1

(18.21)

Proof. Easy; based on Lemma 18.1, Theorem 18.2, and Corollary 18.3. The corresponding L1 result follows. Theorem 18.6. Let γ ≥ 0 and 1 + γ ≤ ν 1 < ν 2 < . . . < ν k , nk := [ν k ],
and Λ∗0

:= max

1≤l≤k

(

(ν l −1+1/r)

(b − x0 )

1/r

(Γ (ν l )) [r (ν l − 1) + 1]

.

(18.23)

Then (1) Dxγ0 f L

r (x0 ,b)

≤

Λ∗γ k

k

ν Dxl f 0 L

1 (x0 ,b)

.

(18.24)

l=1

When γ = 0 we obtain (2) f Lr (x0 ,b) ≤

Λ∗0 k

k

ν l Dx f . 0 L1 (x0 ,b)

(18.25)

l=1

Proof. Easy; based on Lemma 18.1 and Theorem 18.4.

Let α > 0, f ∈ L1 (a, x), a, x ∈ R; see [45, 64, 134]. We deﬁne the generalized Riemann–Liouville fractional derivative of f of order α by m s d 1 m−α−1 Daα f (s) := (s − t) f (t) dt, Γ (m − α) ds a where m := [α] + 1, s ∈ [a, x] ; see also [57, Remark 46].

18.2 Various Univariate Sobolev-Type Fractional Inequalities

489

In addition, we set Da0 f := f := J0a f, a J−α f := Daα f, if α > 0,

Da−α f := Jαa f, if 0 < α ≤ 1, Dan f = f (n) , for n ∈ N. We need Deﬁnition 18.7. [45] We say that f ∈ L1 (a, x) has an L∞ fractional derivative Daα f (α > 0) in [a, x], a, x ∈ R, iﬀ Daα−k f ∈ C ([a, x]), k = 1, . . . , m := [α] + 1, and Daα−1 f ∈ AC ([a, x]) (absolutely continuous functions) and Daα f ∈ L∞ (a, x). Lemma 18.8. [45] Let β > α ≥ 0, f ∈ L1 (a, x), a, x ∈ R, have an L∞ fractional derivative Daβ f in [a, x], and let Daβ−k f (a) = 0 for k = 1, . . . , [β] + 1. Then s 1 β−α−1 α Da f (s) = (s − t) Daβ f (t) dt, (18.26) Γ (β − α) a ∀s ∈ [a, x] . Here Daα f ∈ AC ([a, x]) for β − α ≥ 1, and Daα f ∈ C ([a, x]) for β − α ∈ (0, 1). We present Sobolev-type inequalities with respect to the above-deﬁned generalized Riemann–Liouville fractional derivative. Theorem 18.9. Let β > α ≥ 0, f ∈ L1 (a, x), a, x ∈ R, have an L∞ fractional derivative Daβ f in [a, x], and let Daβ−k f (a) = 0, for k = 1, . . . , [β]+1. Let r ≥ 1; p, q > 1 : 1/p+1/q = 1 with p (β − α − 1)+1 > 0. Then Daα f Lr (a,x) ≤

(x − a)

β−α+1/r−1/q 1/p

(Γ (β − α)) (p (β − α − 1) + 1) β Da f Lq (a,x) + ,1/r . r β − α − 1q + 1

Proof. Similar to Theorem 18.2; now based on Lemma 18.8. Corollary 18.10. (to Theorem 18.9) When α = 0 it holds β+1/r−1/q Daβ f (x − a) Lq (a,x) f Lr (a,x) ≤ + ,1/r . 1/p 1 (Γ (β)) (p (β − 1) + 1) r β− q +1

(18.27)

(18.28)

490

18. Various Sobolev-Type Fractional Inequalities

Next we give the corresponding L1 result. Theorem 18.11. Let β > α + 1, α ≥ 0, f ∈ L1 (a, x), a, x ∈ R, have an L∞ fractional derivative Daβ f in [a, x] and let Daβ−k f (a) = 0, for k = 1, . . . , [β] + 1; r ≥ 1. Then (β−α−1+1/r) Daβ f (x − a) L1 (a,x) α (1) Da f Lr (a,x) ≤ . (18.29) 1/r (Γ (β − α)) [r (β − α − 1) + 1] When α = 0 we ﬁnd (β−1+1/r)

(2) f Lr (a,x) ≤

(x − a)

β Da f L

1 (a,x) . 1/r

(18.30)

(Γ (β)) [r (β − 1) + 1]

Proof. As in Theorem 18.4; now based on Lemma 18.8.

We continue with mean Sobolev-type inequalities. Theorem 18.12. Let 0 ≤ α < β 1 < β 2 < . . . < β k , k ∈ N; f ∈ L1 (a, x), β β −k a, x ∈ R, have L∞ fractional derivatives Da l f in [a, x], with Da l f (a) = 0 for k = 1, . . . , [β l ] + 1; l = 1, . . . , k. Let p, q > 1 : 1/p + 1/q = 1 with p (β l − α − 1) + 1 > 0, all l = 1, . . . , k; r ≥ 1. Put Kα := max

1≤l≤k

⎡ ⎢ ⎣

⎤ β l −α+1/r−1/q

(x − a)

(Γ (β l − α)) (p (β l − α − 1) + 1)

and

1/p

⎥ + ,1/r ⎦, (18.31) 1 r βl − α − q + 1

⎡

⎢ K0 := max ⎣ 1≤l≤k

⎤ β l +1/r−1/q

(x − a)

⎥ 1/r ⎦ . (18.32) 1/p r β l − 1q + 1 (Γ (β l )) (p (β l − 1) + 1)

Then (1)

Daα f Lr (a,x)

≤

Kα k

k

β Da l f L

.

q (a,x)

(18.33)

l=1

When α = 0 we get (2) f Lr (a,x) ≤

K0 k

k

β Da l f L

q (a,x)

l=1

.

(18.34)

18.2 Various Univariate Sobolev-Type Fractional Inequalities

491

Proof. Based on Theorem 18.9 and Corollary 18.10. The corresponding average L1 result follows.

Theorem 18.13. Let α ≥ 0 and α + 1 < β 1 < β 2 < . . . < β k , k ∈ N; β f ∈ L1 (a, x), a, x ∈ R, have L∞ fractional derivatives Da l f in [a, x], with β −k Da l f (a) = 0 for k = 1, . . . , [β l ] + 1; l = 1, . . . , k, and r ≥ 1. Set ' ( (β l −α−1+1/r) (x − a) , (18.35) Kα∗ := max 1≤l≤k (Γ (β − α)) (r (β − α − 1) + 1)1/r l l ' ( (β l −1+1/r) (x − a) . (18.36) K0∗ := max 1≤l≤k (Γ (β )) (r (β − 1) + 1)1/r l l Then (1)

Daα f Lr (a,x)

≤

Kα∗ k

k

β Da l f L

.

1 (a,x)

(18.37)

l=1

When α = 0 we get (2) f Lr (a,x) ≤

K0∗ k

k

Proof. Based on Theorem 18.11.

β Da l f L

1 (a,x)

.

(18.38)

l=1

Next we produce Sobolev-type inequalities with respect to the Caputo fractional derivative. We mention Deﬁnition 18.14. [134] Let ν ≥ 0, n := ν; · is the ceiling of the number f ∈ AC n ([a, b]). We call the Caputo fractional derivative x 1 n−ν−1 (n) ν D∗a f (x) := (x − t) f (t) dt, (18.39) Γ (n − ν) a ∀x ∈ [a, b] . ν f (x) exists almost everywhere for x ∈ [a, b] . The above function D∗a We need Proposition 18.15. [134] Let ν ≥ 0, n := ν and f ∈ AC n ([a, b]). Then exists iﬀ the generalized Riemann–Liouville fractional derivative Daν f exists.

ν f D∗a

492

18. Various Sobolev-Type Fractional Inequalities

Proposition 18.16. [134] Let ν ≥ 0, n := ν . Assume that f is such ν that both D∗a f and Daν f exist. Suppose that f (k) (a) = 0 for k = 0, 1, . . . , n − 1. Then ν f = Daν f. D∗a In conclusion ν Corollary 18.17. [58] Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]), D∗a f exists or Daν f exists, and f (k) (a) = 0, k = 0, 1, . . . , n − 1. Then ν f. Daν f = D∗a

We need Theorem 18.18. [58] Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]), and f (a) = 0, k = 0, 1, . . . , n − 1. Then x 1 ν−1 ν (x − t) D∗a f (t) dt. (18.40) f (x) = Γ (ν) a (k)

We also need Theorem 18.19. [58] Let ν ≥ γ + 1, γ ≥ 0. Call n := ν . Assume ν f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and D∗a f ∈ γ L∞ (a, b). Then D∗a f ∈ C ([a, b]), and x 1 ν−γ−1 γ ν (x − t) D∗a f (t) dt, (18.41) D∗a f (x) = Γ (ν − γ) a ∀x ∈ [a, b] . Theorem 18.20. [58] Let ν ≥ γ+1, γ ≥ 0, n := ν . Let f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n−1. Assume ∃Daν f (x) ∈ R, ∀x ∈ [a, b], and Daν f ∈ L∞ (a, b). Then Daγ f ∈ C ([a, b]), and x 1 ν−γ−1 (x − t) Daν f (t) dt, (18.42) Daγ f (x) = Γ (ν − γ) a ∀x ∈ [a, b] . Now we are ready to give

18.2 Various Univariate Sobolev-Type Fractional Inequalities

493

Theorem 18.21. Let ν ≥ γ + 1, γ ≥ 0, n := ν . Assume f ∈ ν AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n−1, and D∗a f ∈ L∞ (a, b). Let p, q > 1 : 1/p + 1/q = 1, r ≥ 1. Then γ D∗a f Lr (a,b) ≤

(b − a)

ν−γ+1/r−1/q

(Γ (ν − γ)) (p (ν − γ − 1) + 1)

1/p

ν D∗a f Lq (a,b) + ,1/r . r ν − γ − 1q + 1

(18.43)

Proof. As in Theorem 18.2; based on Theorems 18.18 and 18.19.

Corollary 18.22. (to Theorem 18.21) When γ = 0 it holds ν+1/r−1/q

f Lr (a,b) ≤

ν D∗a f Lq (a,b) 1/r . 1/p 1 r ν− q +1 (Γ (ν)) (p (ν − 1) + 1)

(b − a)

(18.44)

Next we give the corresponding L1 result. Theorem 18.23. Let ν ≥ γ + 1, γ ≥ 0, n := ν . Assume f ∈ ν f ∈ L∞ (a, b), AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n−1, and D∗a r ≥ 1. Then γ f Lr (a,b) ≤ (1) D∗a

(b − a)

(ν−γ−1+1/r)

ν D∗a f L1 (a,b) 1/r

(Γ (ν − γ)) (r (ν − γ − 1) + 1)

.

(18.45)

When γ = 0 we get (2) f Lr (a,b) ≤

(b − a)

(ν−1+1/r)

ν D∗a f L1 (a,b) 1/r

(Γ (ν)) (r (ν − 1) + 1)

.

(18.46)

Proof. Similar to Theorem 18.4; based on Theorems 18.18 and 18.19. We continue with mean Sobolev-type inequalities. Theorem 18.24. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , nk −1, νl f ∈ L∞ (a, b) , l = 1, . . . , k. Let p, q > 1 : 1/p + 1/q = 1, r ≥ 1. and D∗a Call Mγ := max 1≤l≤k

494

18. Various Sobolev-Type Fractional Inequalities

⎡ ⎢ ⎣

⎤ ν l −γ+1/r−1/q

(b − a)

⎥ + ,1/r ⎦, 1/p r ν l − γ − 1q + 1 (Γ (ν l − γ)) [p (ν l − γ − 1) + 1]

and

⎡

⎢ M0 := max ⎣ 1≤l≤k

(18.47)

⎤ (b − a)

ν l +1/r−1/q

⎥ 1/r ⎦ . (18.48) 1/p r ν l − 1q + 1 (Γ (ν l )) [p (ν l − 1) + 1]

Then (1)

γ f Lr (a,b) D∗a

≤

Mγ k

k

νl D∗a f Lq (a,b)

.

(18.49)

l=1

When γ = 0 we obtain (2) f Lr (a,b) ≤

M0 k

k

νl D∗a f Lq (a,b) .

(18.50)

l=1

Proof. Easy; based on Theorems 18.18, 18.19, 18.21 and Corollary 18.22. The corresponding average L1 result follows. Theorem 18.25. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , nk −1, νl f ∈ L∞ (a, b), l = 1, . . . , k, r ≥ 1. Set and D∗a ' ( (ν l −γ−1+1/r) (b − a) , (18.51) Mγ∗ := max 1≤l≤k (Γ (ν − γ)) (r (ν − γ − 1) + 1)1/r l l '

and M0∗

:= max

(b − a)

(

(ν l −1+1/r)

.

(18.52)

νl D∗a f L1 (a,b) .

(18.53)

1/r

(Γ (ν l )) (r (ν l − 1) + 1)

1≤l≤k

Then γ 1) D∗a f Lr (a,b) ≤

Mγ∗ k

k

l=1

When γ = 0 we get 2) f Lr (a,b) ≤

M0∗ k

k l=1

νl D∗a f L1 (a,b)

.

(18.54)

18.2 Various Univariate Sobolev-Type Fractional Inequalities

Proof. Easy; based on Theorem 18.23.

495

Next we return to Riemann–Liouville fractional derivatives related results from another perspective. Theorem 18.26. Let ν ≥ γ + 1, γ ≥ 0, n := ν . Assume f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n−1, and Daν f ∈ L∞ (a, b). Assume ∃Daν f (x) ∈ R, ∀x ∈ [a, b] . Let p, q > 1 : 1/p + 1/q = 1, r ≥ 1. Then Daγ f Lr (a,b) ≤ (b − a)

ν−γ+1/r−1/q

Daν f Lq (a,b) + ,1/r , 1/p r ν − γ − 1q + 1 (Γ (ν − γ)) (p (ν − γ − 1) + 1) Proof. As in Theorem 18.2; based on Theorem 18.20.

(18.55)

Corollary 18.27. (to Theorem 18.26) When γ = 0 it holds f Lr (a,b) ≤

ν+1/r−1/q

Daν f Lq (a,b) 1/r . 1/p r ν − 1q + 1 (Γ (ν)) (p (ν − 1) + 1) (b − a)

(18.56)

Next we give the corresponding L1 result. Theorem 18.28. Let ν ≥ γ + 1, γ ≥ 0, n := ν . Assume f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n−1, and Daν f ∈ L∞ (a, b), r ≥ 1. Assume ∃Daν f (x) ∈ R, ∀x ∈ [a, b] . Then (1) Daγ f Lr (a,b) ≤

(b − a)

(ν−γ−1+1/r)

Daν f L1 (a,b) 1/r

(Γ (ν − γ)) (r (ν − γ − 1) + 1)

.

(18.57)

When γ = 0 we obtain (ν−1+1/r)

(2) f Lr (a,b) ≤

(b − a)

Daν f L1 (a,b) 1/r

(Γ (ν)) (r (ν − 1) + 1)

.

(18.58)

Proof. Similar to Theorem 18.4; based on Theorem 18.20. We continue with mean Sobolev-type inequalities. Theorem 18.29. Let γ ≥ 0 and 1 + γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , and k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . ,

496

18. Various Sobolev-Type Fractional Inequalities

nk − 1, and Daν l f ∈ L∞ (a, b), l = 1, . . . , k. Let p, q > 1 : 1/p + 1/q = 1, r ≥ 1. Further assume ∃Daν f (x) ∈ R, ∀x ∈ [a, b] . Here Mγ is as in (18.47) and M0 as in (18.48). Then k Mγ γ νl (18.59) Da f Lq (a,b) . (1) Da f Lr (a,b) ≤ k l=1

When γ = 0 we obtain (2) f Lr (a,b) ≤

M0 k

k

Daν l f Lq (a,b)

.

(18.60)

l=1

Proof. Easy; based on Theorem 18.26 and Corollary 18.27. The corresponding average L1 result follows. Theorem 18.30. Let γ ≥ 0 and 1 + γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , and k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , nk − 1, and Daν l f ∈ L∞ (a, b), l = 1, . . . , k, r ≥ 1. Further assume ∃Daν f (x) ∈ R, ∀x ∈ [a, b] . Here Mγ∗ is as in (18.51) and M0∗ as in (18.52). Then (1)

Daγ f Lr (a,b)

≤

Mγ∗ k

k

Daν l f L1 (a,b)

.

(18.61)

l=1

When γ = 0 we get (2) f Lr (a,b) ≤

M0∗ k

k

Daν l f L1 (a,b)

.

(18.62)

l=1

Proof. Easy; based on Theorem 18.28.

Next we give converse Sobolev-type results, starting with the Canavatitype fractional derivative. Theorem 18.31. Let ν ≥ γ + 1, γ ≥ 0, n := [ν], f ∈ Cxν0 ([a, b]), x0 ∈ [a, b] . Assume f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Assume that Dxν0 f is of ﬁxed strict sign on [x0 , b] . Let 0 < p < 1, q < 0, such that 1/p + 1/q = 1, r ≥ 1. Then ≥ (1) Dxγ f 0

Lr (x0 ,b)

(ν−γ−1/q+1/r) Dxν f (b − x0 ) 0 Lq (x0 ,b) 1/r . 1/p 1 Γ (ν − γ) (p (ν − γ − 1) + 1) r ν−γ− q +1

(18.63)

18.2 Various Univariate Sobolev-Type Fractional Inequalities

497

When γ = 0 we obtain (2) f Lr (x0 ,b)

ν Dx f 0 Lq (x0 ,b) ≥ 1/r . 1/p r ν − 1q + 1 Γ (ν) (p (ν − 1) + 1) (b − x0 )

(ν−1/q+1/r)

(18.64)

Proof. Here Dxγ0 f (x0 ) = 0 by (18.1), but by assumption Dxγ0 f (x) = 0, ∀x ∈ (x0 , b] . By Lemma 18.1 and assumptions of the theorem we have x γ 1 ν−γ−1 ν Dx f (x) = Dx0 f (t) dt (x − t) 0 Γ (ν − γ) x0 (by the reverse H¨older’s inequality) 1 Γ (ν − γ)

≥

1/p

x

p(ν−γ−1)

(x − t)

dt

x0

ν Dx f (t)q dt 0

x

x0

1/q

(ν−γ−1)+1/p

≥

(x − x0 ) 1 Γ (ν − γ) (p (ν − γ − 1) + 1)1/p

b

x0

So that

ν Dx f (t)q dt 0

γ Dx f (x) ≥ 0 (ν−γ−1)+1/p

(x − x0 )

1/p

(p (ν − γ − 1) + 1) ∀x ∈ [x0 , b] . Thus

b

x0

γ Dx f (x)r ≥ 0

.

(18.65)

1 Γ (ν − γ)

ν Dx f (t)q dt 0

1/q

r(ν−γ−1)+r/p r/p

ν r Dx f 0 L

q (x0 ,b)

∀x ∈ [x0 , b] . Hence ν r Dx f r γ 0 Lq (x0 ,b) Dx f (x) dx ≥ r 0 (Γ (ν − γ)) x0

b

,

(18.66)

1 r (Γ (ν − γ))

(x − x0 )

(p (ν − γ − 1) + 1)

1/q

,

(18.67)

498

18. Various Sobolev-Type Fractional Inequalities

1

b

r(ν−γ−1)+r/p

(x − x0 ) dx r/p x0 (p (ν − γ − 1) + 1) ν r Dx f 1 0 Lq (x0 ,b) = r (Γ (ν − γ)) (p (ν − γ − 1) + 1)r/p

(18.68)

r(ν−γ−1)+r/p+1

(b − x0 ) r (ν − γ − 1) +

r p

. +1

So we have established

ν Dx f γ 1 0 Lq (x0 ,b) Dx f ≥ 0 Lr (x0 ,b) 1/p Γ (ν − γ) (p (ν − γ − 1) + 1) (ν−γ−1)+1/p+1/r

proving the claim.

(b − x0 )

r (ν − γ − 1) +

r p

1/r , +1

(18.69)

Converse results follow related to the Riemann–Liouville fractional derivative. Theorem 18.32. Let β > α ≥ 0, f ∈ L1 (a, x), a, x ∈ R, have L∞ fractional derivative Daβ f in [a, x], and let Daβ−k f (a) = 0 for k = 1, . . . , [β]+1. Assume that Daβ f has ﬁxed sign a.e. on [a, x], and 1/Daβ f ∈ L∞ (a, x). Let 0 0; r ≥ 1. Then (1) Daα f Lr (a,x) ≥ (β−α−1/q+1/r) Daβ f (x − a) Lq (a,x) (18.70) 1/r . 1/p 1 r β−α− q +1 Γ (β − α) (p (β − α − 1) + 1) When α = 0 we get (2) f Lr (a,x)

β Da f Lq (a,x) ≥ 1/r . 1/p r β − 1q + 1 Γ (β) (p (β − 1) + 1) (x − a)

(β−1/q+1/r)

Proof. Similar to Theorem 18.31; based on (18.26).

(18.71)

We continue with converse results related to the Caputo fractional derivative.

18.2 Various Univariate Sobolev-Type Fractional Inequalities

499

Theorem 18.33. Let ν ≥ γ + 1, γ ≥ 0, n := ν, f ∈ AC n ([a, b]), and ν f (k) (a) = 0, k = 0, 1, . . . , n − 1. Assume that D∗a f is of ﬁxed sign a.e. ν ν f, 1/D∗a f ∈ L∞ (a, b). Let 0 < p < 1, q < 0, such that on [a, b], and D∗a 1/p + 1/q = 1; r ≥ 1. Then γ (1) D∗a f Lr (a,b) ≥ (ν−γ−1/q+1/r)

(b − a)

ν D∗a f Lq (a,b) 1/r . 1/p 1 r ν−γ− q +1 Γ (ν − γ) (p (ν − γ − 1) + 1)

(18.72)

When γ = 0 we get (2) f Lr (a,b) ≥

(ν−1/q+1/r)

ν D∗a f Lq (a,b) 1/r . 1/p 1 r ν− q +1 Γ (ν) (p (ν − 1) + 1)

(b − a)

Proof. Similar to Theorem 18.31; based on (18.40), (18.41).

(18.73)

Next we give converse results again for the Riemann–Liouville fractional derivative. Theorem 18.34. Let ν ≥ γ + 1, γ ≥ 0, n := ν, f ∈ AC n ([a, b]), and f (k) (a) = 0, k = 0, 1, . . . , n − 1. Suppose ∃Daν f (x) ∈ R, ∀x ∈ [a, b], and Daν f, 1/Daν f ∈ L∞ (a, b). Here Daν f is of ﬁxed sign a.e. on [a, b] . Let 0 < p < 1, q < 0, such that 1/p + 1/q = 1; r ≥ 1. Then (1) Daγ f Lr (a,b) ≥ (ν−γ−1/q+1/r)

(b − a)

Daν f Lq (a,b) 1/r . 1/p r ν − γ − 1q + 1 Γ (ν − γ) (p (ν − γ − 1) + 1)

(18.74)

When γ = 0 we get (2) f Lr (a,b) ≥

(ν−1/q+1/r)

Daν f Lq (a,b) 1/r . 1/p r ν − 1q + 1 Γ (ν) (p (ν − 1) + 1) (b − a)

(18.75)

Proof. Similar to Theorem 18.31; based on (18.42) . Next we give converse mean Sobolev type inequalities. Theorem 18.35. Let γ ≥ 0 and 1 + γ ≤ ν 1 < ν 2 < . . . < ν k , nk := [ν k ],
500

18. Various Sobolev-Type Fractional Inequalities

i = 0, 1, . . . , nk − 1; 0 < p < 1, q < 0 : 1/p + 1/q = 1; r ≥ 1. Further suppose each Dxν0l f is of ﬁxed strict sign on [x0 , b], all l = 1, . . . , k. Call θγ := min

1≤l≤k

⎧ ⎪ ⎨

⎫ ⎪ ⎬

(ν l −γ−1/q+1/r)

(b − x0 ) 1/r ⎪ , ⎪ ⎩ Γ (ν − γ) (p (ν − γ − 1) + 1)1/p r ν − γ − 1 + 1 ⎭ l l l q and

⎧ ⎪ ⎨ 1≤l≤k ⎪ ⎩

⎫ ⎪ ⎬

(ν l −1/q+1/r)

(b − x0 )

θ0 := min

1/p

Γ (ν l ) (p (ν l − 1) + 1)

Then (1) Dxγ0 f L

r (x0 ,b)

θγ ≥ k

(18.76)

1/r ⎪ . ⎭ r ν l − 1q + 1

(18.77)

k ν Dxl f 0 L

q (x0 ,b)

.

(18.78)

l=1

When γ = 0 we obtain (2) f Lr (x0 ,b)

θ0 ≥ k

k ν Dxl f . 0 Lq (x0 ,b)

(18.79)

l=1

Proof. By (18.63) we have γ Dx f 0 L

r (x0 ,b)

≥

ν Dxl f 0 Lq (x0 ,b) 1/r 1/p r ν l − γ − 1q + 1 Γ (ν l − γ) (p (ν l − γ − 1) + 1) ≥ θγ Dxν0l f L (x ,b) , (18.80) (ν l −γ−1/q+1/r)

(b − x0 )

q

0

for all l = 1, . . . , k. That is, γ Dx f 0 L

r (x0 ,b)

≥ θγ Dxν0l f L

q (x0 ,b)

for all l = 1, . . . , k. Adding the above (18.81) we obtain k γ Dxν l f k Dx f ≥ θγ 0

Lr (x0 ,b)

0

l=1

establishing our claim.

,

Lq (x0 ,b)

(18.81)

,

(18.82)

18.2 Various Univariate Sobolev-Type Fractional Inequalities

501

We continue with Riemann–Liouville fractional derivatives. Theorem 18.36. Let 0 ≤ α < β 1 < β 2 < . . . < β k , k ∈ N; f ∈ β L1 (a, x), a, x ∈ R, have L∞ fractional derivatives Da l f in [a, x], with β l −k Da f (a) = 0 for k = 1, . . . , [β l ] + 1; l = 1, . . . , k. Let 0 0, all l = 1, . . . , k; β r ≥ 1. Further assume that each Da l f has ﬁxed sign a.e. on [a, x], and βl 1/Da f ∈ L∞ (a, x), for all l = 1, . . . , k. Put θ∗α := min 1≤l≤k

⎧ ⎪ ⎨

⎫ ⎪ ⎬

(β l −α−1/q+1/r)

(x − a) 1/r ⎪ , (18.83) ⎪ ⎩ Γ (β − α) (p (β − α − 1) + 1)1/p r β − α − 1 + 1 ⎭ l l l q and θ∗0 := min

⎧ ⎪ ⎨

1≤l≤k ⎪ ⎩

(x − a)

1/p

Γ (β l ) (p (β l − 1) + 1)

Then (1)

Daα f Lr (a,x)

⎫ ⎪ ⎬

(β l −1/q+1/r)

θ∗ ≥ α k

1/r ⎪ . ⎭ r β l − 1q + 1

k Daβ l f L

(18.84)

.

q (a,x)

(18.85)

l=1

When α = 0 we get (2) f Lr (a,x)

θ∗ ≥ 0 k

k Daβ l f L

q (a,x)

.

(18.86)

l=1

Proof. As in Theorem 18.35; based on Theorem 18.32.

We continue with Caputo fractional derivatives. Theorem 18.37. Let γ ≥ 0 and 1+γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , nk −1, νl νl νl f, 1/D∗a f ∈ L∞ (a, b), l = 1, . . . , k. Suppose each D∗a f is of ﬁxed and D∗a sign a.e. on [a, b], all l = 1, . . . , k. Let 0 < p < 1, q < 0, such that 1/p + 1/q = 1; r ≥ 1. Set Aγ := min 1≤l≤k

502

18. Various Sobolev-Type Fractional Inequalities

⎧ ⎪ ⎨

⎫ ⎪ ⎬

(ν l −γ−1/q+1/r)

(b − a) 1/r ⎪ , ⎪ ⎩ Γ (ν − γ) (p (ν − γ − 1) + 1)1/p r ν − γ − 1 + 1 ⎭ l l l q

(18.87)

and ⎧ ⎪ ⎨

(b − a)

A0 := min

1≤l≤k ⎪ ⎩

1/p

Γ (ν l ) (p (ν l − 1) + 1)

Then γ (1) D∗a f Lr (a,b)

Aγ ≥ k

⎫ ⎪ ⎬

(ν l −1/q+1/r)

k

1/r ⎪ . ⎭ r ν l − 1q + 1

(18.88)

νl D∗a f Lq (a,b) .

(18.89)

l=1

When γ = 0 we get (2) f Lr (a,b)

A0 ≥ k

k

νl D∗a f Lq (a,b)

.

(18.90)

l=1

Proof. As in Theorem 18.35; based on Theorem 18.33.

Next we come back to Riemann–Liouville fractional derivatives. Theorem 18.38. Let γ ≥ 0 and 1 + γ ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , and k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , nk − 1; ∃Daν l f (x) ∈ R, ∀x ∈ [a, b], and Daν l f, 1/Daν l f ∈ L∞ (a, b), l = 1, . . . , k. Suppose each Daν l f is of ﬁxed sign a.e. on [a, b], all l = 1, . . . , k. Let 0 < p < 1, q < 0, such that 1/p + 1/q = 1; r ≥ 1. Here Aγ is as in (18.87), and A0 as in (18.88). Then (1)

Daγ f Lr (a,b)

Aγ ≥ k

k

Daν l f Lq (a,b)

.

(18.91)

l=1

When γ = 0 we get (2) f Lr (a,b)

A0 ≥ k

k

Daν l f Lq (a,b) .

(18.92)

l=1

Proof. As in Theorem 18.35; based on Theorem 18.34.

18.3 Applications

503

18.3 Applications Next we give an application to ordinary fractional diﬀerential equations involving Caputo fractional derivatives. Theorem 18.39. Let 1 ≤ ν 1 < ν 2 < . . . < ν k , nk := ν k , and k ∈ N. Assume f ∈ AC nk ([a, b]) such that f (j) (a) = 0, j = 0, 1, . . . , nk − 1, and νl f ∈ L∞ (a, b), all l = 1, . . . , k; r ≥ 1. Let also B, Cl , 1/Cl ∈ L∞ (a, b), D∗a l = 1, . . . , k, with all Cl ≥ 0. The above-described f satisﬁes k

2

νl Cl (x) (D∗a f (x)) = B (x) , ∀x ∈ [a, b] .

(18.93)

l=1

Call

%

(b − a)

∗

-

(2ν l +2/r−1)

2/r 2 (Γ (ν l )) (2ν l − 1) r ν l − 12 + 1 1 ∗ . ρ := max 1≤l≤k Cl L (a,b) ∞

δ := max

1≤l≤k

Then

: f Lr (a,b) ≤

δ ∗ ρ∗ k

,

(18.94)

(18.95)

1/2

b

B (x) dx

.

(18.96)

a

Proof. Set nl := ν l , l = 1, . . . , k − 1. Clearly here f ∈ AC nl ([a, b]), l = 1, . . . , k − 1. Also f (j) (a) = 0, j = 0, 1, . . . , nl − 1, l = 1, . . . , k − 1. So all assumptions of Theorem 18.21 are fulﬁlled for f and fractional orders ν l , l = 1, . . . , k − 1. Thus by choosing p = q = 2 we apply (18.44), for l = 1, . . . , k, to obtain

f Lr (a,b) ≤ Hence it holds

ν l +1/r−1/2

b a

(18.97)

2 νl (D∗a f (x)) dx ≤ 2/r ≤ 2 (Γ (ν l )) (2ν l − 1) r ν l − 12 + 1 (b − a)

2

f Lr (a,b) δ∗

1/2 2 νl (D∗a f (x)) dx 1/r . 1/2 r ν l − 12 + 1 Γ (ν l ) (2ν l − 1)

(b − a)

b

(Cl (x)) a

δ ∗ ρ∗

a

(2ν l +2/r−1)

−1

b

b a

2

νl (Cl (x)) (D∗a f (x)) dx ≤ 2

νl Cl (x) (D∗a f (x)) dx.

(18.98)

504

18. Various Sobolev-Type Fractional Inequalities

That is, f Lr (a,b) ≤ δ ∗ ρ∗

b

2

2

νl Cl (x) (D∗a f (x)) dx,

a

(18.99)

for all l = 1, . . . , k. The last imply by addition that k b δ ∗ ρ∗ 2 2 νl f Lr (a,b) ≤ Cl (x) (D∗a f (x)) dx k a l=1

= proving the claim.

δ ∗ ρ∗ k

b

B (x) dx,

(18.100)

a

One can give similar applications to fractional ODEs involving fractional derivatives of Canavati-type and Riemann–Liouville-type.

19 General Hilbert–Pachpatte-Type Integral Inequalities

In this chapter we present very general weighted Hilbert–Pachpatte-type integral inequalities. These are with regard to ordinary derivatives and fractional derivatives of Riemann–Liouville and Canavati types, and also in regard to general derivatives of Widder-type and linear diﬀerential operators. These results apply to continuous functions and some to integrable functions. This treatment relies on [41].

19.1 Introduction In this chapter we present a series of weighted very general Hilbert– Pachpatte-type integral inequalities regarding ordinary and fractional derivatives, derivatives of Widder-type, and linear diﬀerential operators. First we derive a very general result in Theorem 19.3, for which the rest of the results are applications. The results here are motivated by the original Hilbert double integral inequality. Theorem 19.1. [191, Theorem 316]. If p > 1, q = p/(p − 1) and ∞ ∞ f p (x)dx ≤ F, g q (y)dy ≤ G, 0

then

0

∞

0

0

∞

π f (x)g(y) dxdy < F 1/p G1/q x+y sin(π/p)

(19.1)

G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 505 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 19,

506

19. General Hilbert–Pachpatte-Type Integral Inequalities

where f, g are nonnegative measurable functions, unless f ≡ 0 or g ≡ 0. The constant π cosec(π/p) is the best possible in (19.1). Also the results here are motivated by Theorem 19.2. (Pachpatte [320, Theorem 1]). Let n ≥ 1 and 0 ≤ k ≤ n − 1 be integers. Let u ∈ C n ([0, x]) and v ∈ C n ([0, y]), where x > 0, y > 0, and let u(j) (0) = v (j) (0) = 0 for j ∈ {0, . . . , n − 1}. Then x y |u(k) (s)| |v (k) (t)| dsdt ≤ M (n, k, x, y) 2n−2k−1 + t2n−2k−1 0 0 s 1/2 y 1/2 x (n) 2 (n) 2 (x − s) u (s) ds (y − t) v (t) dt , (19.2) 0

0

√ xy 1 . 2 2 [(n − k − 1)!] (2n − 2k − 1) Also of great motivation to this chapter are the articles [147, 187, 188].

where

M (n, k, x, y) :=

19.2 Main Results We present the following general result. Theorem 19.3. Here for i ∈ {1, . . . , n}, take xi > 0, and assume ui ∈ L1 (0, xi ), gi ∈ L∞ (0, xi )2 , Φi ∈ L∞ (0, xi ), with " gi , Φi ≥ 0. Take ri ≥ 0, pi , qi > 1 : 1/pi + 1/qi = 1, and wi > 0 such n that i=1 wi = Ωn , and ai , bi ∈ [0, 1] such that ai + bi = 1. Call si (ai +bi qi )ri gi (si , τ i ) ϕi (si ) := dτ i . (19.3) 0

Suppose ϕi (si ) > 0,with the exception ϕi (0) = 0. If si ri |ui (si )| ≤ gi (si , τ i ) Φi (τ i )dτ i , si ∈ [0, xi ], i = 1, . . . , n,

(19.4)

0

then I1

:

x1

=

≤

i=1

xn

···

0 n

n &

0

) 1/qi

xi

0

xi

1 Ωn

n " i=1

|ui (si )|

i=1

1/qi wi Ωn ds1 · · · dsn wi ϕi (si )

pi Φi (τ i )

*1/pi ai ri gi (si , τ i ) dsi dτ i (19.5) .

xi

τi

19.2 Main Results

507

Proof. We write ri ai /qi +bi ri ai /pi ri gi (si , τ i ) Φi (τ i ) = gi (si , τ i ) × gi (si , τ i ) Φi (τ i ). Notice here that the two sections of gi belong to L∞ (0, xi ), a.e. Using H¨ older’s inequality we ﬁnd |ui (si )|

≤

si

(ai +bi qi )ri gi (si , τ i ) dτ i

1/qi

0

si

×

ai ri pi gi (si , τ i ) Φi (τ i ) dτ i

1/pi .

0

That is, we have that 1/qi |ui (si )| ≤ ϕi (si )

si

ai ri pi gi (si , τ i ) Φi (τ i ) dτ i

1/pi , i = 1, . . . , n.

0

Set ri∗ = 1/qi wi . By the means inequality ([297, p. 15]) we have

n wi ri∗ ϕi (si )

≤

i=1

Ω n n 1 ri∗ wi ϕi (si ) . Ωn i=1

Therefore we ﬁnd n

|ui (si )|

i=1

Ω n n 1 ri∗ ≤ wi ϕi (si ) Ωn i=1 1/pi n si ai ri pi gi (si , τ i ) Φi (τ i ) dτ i × . i=1

0

That is, n &

1 Ωn

i=1 n " i=1

|ui (si )|

Ωn r∗ wi ϕi i (si )

≤

n i=1

0

si

ai ri pi gi (si , τ i ) Φi (τ i ) dτ i

1/pi ,

508

19. General Hilbert–Pachpatte-Type Integral Inequalities

whenever at least one si > 0, i ∈ {1, . . . , n}, where si ∈ [0, xi ]. Thus we obtain

x1

n &

xn

···

0

0

≤

x1

···

0

=

i=1 n si xn

ai ri pi gi (si , τ i ) Φi (τ i ) dτ i

0

n i=1

Ωn ds1 · · · dsn r∗ wi ϕi i (si )

1 Ωn

|ui (si )|

i=1 n "

0

i=1

xi si

ai ri pi gi (si , τ i ) Φi (τ i ) dτ i

0

1/pi ds1 · · · dsn

1/pi dsi

0

(using again H¨ older’s inequality) ) xi si *1/pi n ai ri pi 1/qi ≤ gi (si , τ i ) Φi (τ i ) dτ i dsi xi 0

i=1

0

(by changing the order of integration) ) xi xi *1/pi n ai ri pi 1/q = gi (si , τ i ) Φi (τ i ) dsi dτ i xi i =

i=1 n

0

) 1/q xi i

xi

pi Φi (τ i )

0

i=1

proving the claim.

τi

*1/pi ai ri gi (si , τ i ) dsi dτ i ,

xi

τi

We give Corollary 19.4. All are as in Theorem 19.3 and 1/p := r > 0. Then ' n % n 1 xi pi 1/qi 1/rp Φi (τ i ) I1 ≤ p xi p 0 i=1 i=1 i -r (1/rp xi ai ri gi (si , τ i ) × dsi dτ i .

"n i=1

1/pi ,

τi

Proof. By the inequality of means, for any Ai ≥ 0 and r > 0, we derive n 1/rp n 1 1/pi Ai ≤ p Ari . p i i=1 i=1 We use here (19.5).

19.2 Main Results

509

Next we apply Theorem 19.3. We give Theorem 19.5. Let ui ∈ C mi ([0, xi ]), xi > 0, mi ∈ N, be such that (j) ui (0) = 0 for j ∈ {0, . . . , mi − 1}, i = 1, . . . , n. Let ki ∈ {0, . . . , mi − 1}; pi , qi" > 1 : 1/pi + 1/qi = 1; ai , bi ∈ [0, 1] : ai + bi = 1. Let also wi > 0 such n that i=1 wi = Ωn . Then I2

:

x1

=

A

xn

···

0

≤

n &

0

n

1/q xi i

xi 0

i=1

where

1 Ωn

'

n

A :=

i=1

(ki )

i=1 n " i=1

s

|ui

(si )| Ωn ds1 · · · dsn

[(ai +bi qi )(mi −ki −1)+1]/qi wi

wi [(ai +b i

(m ) |ui i (τ i )|pi (xi

i qi )(mi −ki −1)+1]

1/qi wi

(1/pi [ai (mi −ki −1)+1]

− τ i)

dτ i

, (19.6)

1 . (mi − ki − 1)![ai (mi − ki − 1) + 1]1/pi

Proof. By [320, Equation (7)] we have si 1 (k ) (m ) ui i (si ) = (si − τ i )mi −ki −1 ui i (τ i )dτ i , (mi − ki − 1)! 0 where si ∈ [0, xi ], i = 1, . . . , n. We use Theorem 19.3. We set (m )

Φi (τ i ) :=

|ui i (τ i )| , all 0 ≤ τ i ≤ xi , i = 1, . . . , n. (mi − ki − 1)!

We also set gi (si , τ i ) := |si − τ i |,

all si , τ i ∈ [0, xi ],

and ri := mi − ki − 1. Clearly then si ϕi (si ) = (si − τ i )(ai +bi qi )(mi −ki −1) dτ i 0 (a +b q )(m −k −1)+1

si i i i i i , si ∈ [0, xi ], i = 1, . . . , n. (ai + bi qi )(mi − ki − 1) + 1

=

We see here that (ki )

|ui

(si )| ≤ 0

si

(si − τ i )mi −ki −1

(m )

|ui i (τ i )| dτ i , (mi − ki − 1)!

510

19. General Hilbert–Pachpatte-Type Integral Inequalities

where si ∈ [0, xi ], i = 1, . . . , n. We also ﬁnd xi (xi − τ i )ai (mi −ki −1)+1 . (si − τ i )ai (mi −ki −1) dsi = ai (mi − ki − 1) + 1 τi

Now the claim is clear. We give

"n

Corollary 19.6. All are as in Theorem 19.5 and 1/p := r > 0. Then ' n n 1 1/q xi i I2 ≤ p1/rp A p i=1 i=1 i

xi

×

(m ) |ui i (τ i )|pi (xi

0

[ai (mi −ki −1)+1]

− τ i)

i=1

1/pi ,

r (1/rp

dτ i

.

Proof. By using inequality of means and (19.6). We present (j)

Theorem 19.7. Let ui ∈ C mi +1 ([0, xi ]) be such that ui (0) = 0 for j ∈ {0, . . . , mi }, and let ρ ∈ C 1 ([0, ∞)), where ρ > 0; xi > 0, mi ∈ N, j ∈ {0, . . . , mi −1}, i = 1, . . . , n. Let ki ∈ {0, . . . , mi −1}; "n pi , qi > 1 : 1/pi + 1/qi = 1; ai , bi ∈ [0, 1] : ai + bi = 1. Also Let wi > 0 : i=1 wi = Ωn . Then I3

:

x1

=

≤

A

xn

···

0 n

n &

0

1/qi

1 Ωn

'

xi

xi 0

i=1

i=1 n " i=1

(ki )

|ui

[(a +b q )(m −k −1)+1]/q w

i i i i s i i i wi [(ai +b q )(m −k −1)+1]1/qi wi i i i i i

τ pi i −1 (ρ(τ i ))pi

τi 0

A :=

Ωn ds1 · · · dsn

(m ) | ρ(σ i )ui i (σ i ) |pi dσ i

(1/pi × (xi − τ i )[ai (mi −ki −1)+1] dτ i

where

(si )|

n i=1

,

1 (mi − ki − 1)![ai (mi − ki − 1) + 1]1/pi

(19.7)

.

Proof. By [320, Equation (14)] we have si 1 (k ) ui i (si ) = (si − τ i )mi −ki −1 (mi − ki − 1)! 0 τi 1 (mi ) ρ(σ i )ui (σ i ) dσ i dτ i . × ρ(τ i ) 0

19.2 Main Results

511

By H¨older’s inequality we obtain

τ i

ρ(σ i )u(mi ) (σ i ) dσ i ≤ τ 1/qi i i

0

τ i

ρ(σ i )u(mi ) (σ i ) pi dσ i i

0

1/pi .

We set here 1/q

τi i 1 Φi (τ i ) := (mi − ki − 1)! ρ(τ i )

τ i

ρ(σ i )u(mi ) (σ i ) pi dσ i i

0

1/pi .

The claim now is clear. We give

Corollary 19.8. All are as in Theorem 19.7 and 1/p := r > 0. Then ≤

I3

1/rp

p

A

n

'

1/q xi i

i=1

n 1 pi

%

i=1

τ pi i −1 (ρ(τ i ))pi -r (1/rp

xi 0

×(xi − τ i )[ai (mi −ki −1)+1] dτ i

"n i=1

1/pi ,

τ i 0

ρ(σ i )u(mi ) (σ i ) pi dσ i i

.

Proof. By using inequality of means and (19.7). We give Theorem 19.9. Let ui ∈ C 2mi ([0, xi ]), ρ ∈ C m ([0, ∞)), ρ > 0, with (j) (j) (m ) m = max mi , be such that ui (0) = 0, and ρ(si )ui i (si ) = 0 at si = 0, for j ∈ {0, . . . , mi − 1}; j ∈ {1, . . . , n}, xi > 0, mi ∈ N. Let ki ∈ {0, . . . , mi −" 1} ; pi , qi > 1 : 1/pi + 1/qi = 1; ai , bi ∈ [0, 1] : ai + bi = 1. n Also let wi > 0 : i=1 wi = Ωn . Then I4

:

x1

=

B

n i=1

xn

···

0

≤

n &

i=1

0

1 Ωn

'

xi

1/qi

xi

0

n " i=1

s

(ki )

|ui

(si )|

[(ai +bi qi )(mi −ki −1)+1]/qi wi

wi [(ai

i +bi qi )(mi −ki −1)+1]

(qi (mi −1)+1)(pi −1)

τ i

×(xi − τ i )[ai (mi −ki −1)+1] dτ i

where B :=

1/qi wi

ρ(σ i )u(mi ) (σ i ) (mi ) pi dσ i i

τi

(ρ(τ i ))pi (1/pi

Ωn ds1 · · · dsn

0

,

n i=1

(19.8)

512

19. General Hilbert–Pachpatte-Type Integral Inequalities

1

1/pi (mi − 1)!(mi − ki − 1)!(qi (mi − 1) + 1)1/qi ai (mi − ki − 1) + 1

.

Proof. By [320, Equation (21)], we get that si 1 (k ) ui i (si ) = (si − τ i )mi −ki −1 (mi − 1)!(mi − ki − 1)! 0 τi (mi ) 1 (mi ) mi −1 ρ(σ i )ui (σ i ) × (τ i − σ i ) dσ i dτ i . ρ(τ i ) 0 Put

(mi ) (m ) . Fi (σ i ) := ρ(σ i )ui i (σ i )

Using H¨ older’s inequality we derive τi (τ i − σ i )mi −1 Fi (σ i )dσ i 0

≤

τi

(τ i − σ i )qi (mi −1) dσ i

0 (q (m −1)+1)/q

i τi i i (qi (mi − 1) + 1)1/qi

=

1/qi

1/pi

τi

Fi (σ i )pi dσ i 0

τi

1/pi

pi

Fi (σ i ) dσ i

.

0

Set now (qi (mi −1)+1)/qi

Φi (τ i ) := Wi

τi

where Wi :=

ρ(τ i )

τ i

ρ(σ i )u(mi ) (σ i ) (mi ) pi dσ i i

0

1/pi ,

1 . (mi − 1)!(mi − ki − 1)!(qi (mi − 1) + 1)1/qi

Now apply Theorem19.3. The claim is now clear.

We give Corollary 19.10. All are as in Theorem 19.9 and 1/p := r > 0. Then I4

≤

p1/rp B

n i=1

' 1/qi

xi

n 1 p i i=1

%

xi 0

"n i=1

1/pi ,

(qi (mi −1)+1)(pi −1)

τi

(ρ(τ i ))pi

-r (1/rp ρ(σ i )u(mi ) (σ i ) (mi ) pi dσ i (xi − τ i )[ai (mi −ki −1)+1] dτ i . i

τ i

× 0

Proof. By using inequality of means and (19.8).

19.2 Main Results

513

Note. Similar theorems to our Theorems 19.5–19.9 appeared earlier in [188]. We need (see [187]) Deﬁnition 19.11. Let α > 0. For any f ∈ L1 (0, x); x > 0 , the Riemann–Liouville fractional integral of f of order α is deﬁned by s 1 (Jα f )(s) := (s − t)α−1 f (t)dt, s ∈ [0, x], (19.9) Γ(α) 0 and the Riemann–Liouville fractional derivative of f of order α by m s d 1 α (s − t)m−α−1 f (t)dt, (19.10) Δ f (s) := Γ(m − α) ds 0 where m := [α] + 1, [·] is the integral part. In addition, we set Δ0 f := f =: J0 f,

J−α f := Δα f if α > 0,

Δ−α f := Jα f if 0 < α ≤ 1.

If α ∈ N, then Δα f = f (α) . Deﬁnition 19.12. [187]. We say that f ∈ L1 (0, x) has an L∞ fractional derivative Δα f in [0, x] iﬀ Δα−k f ∈ C([0, x]), k = 1, . . . , m := [α] + 1; α > 0, and Δα−1 f ∈ AC([0, x]) (absolutely continuous functions) and Δα f ∈ L∞ (0, x). We also need Lemma 19.13. [187]. Let α ≥ 0, β > α, let f ∈ L1 (0, x) have an L∞ fractional derivative Δβ f in [0, x], and let Δβ−k f (0) = 0 for k = 1, . . . , [β] + 1. Then s 1 α Δ f (s) = (s − t)β−α−1 Δβ f (t)dt, s ∈ [0, x]. (19.11) Γ(β − α) 0 Clearly here Δα f ∈ L∞ (0, x), thus Δα f ∈ L1 (0, x). We give Theorem 19.14. Let i ∈ {1, . . . , n}, xi > 0, αi ≥ 0, β i ≥ αi + 1. Let fi ∈ L1 (0, xi ) have L∞ fractional derivatives Δβ i fi in [0, xi ], and let Δβ i −ki fi (0) = 0 for ki = 1, . . . , [β i ] + 1.

514

19. General Hilbert–Pachpatte-Type Integral Inequalities

Let pi , qi > 1 : 1/pi + 1/qi = 1, and wi > 0 such that ai , bi ∈ [0, 1] such that ai + bi = 1. Then I5

:

x1

=

C

xn

···

0

≤

n &

n

1/qi

1 Ωn

)

xi

n " i=1

wi

wi = Ωn , and

i=1

|Δαi fi (si )|

i=1

0

"n

[(ai +bi qi )(β i −αi −1)+1]/qi wi

si

(ai +bi qi )(β i −αi −1)+1

1/qi wi

Ωn ds1 · · · dsn

Δβ i fi (ti )pi (xi − ti )[ai (β i −αi −1)+1] dti

xi

*1/pi ,

(19.12)

0

i=1

where C :=

n

1

1/pi Γ(β i − αi ) ai (β i − αi − 1) + 1

i=1

.

Proof. Here we again apply Theorem 19.3. By Lemma 19.13 we get that Δαi fi ∈ L1 (0, xi ), also Δβ i fi ∈ L∞ (0, xi ), and si 1 |Δαi fi (si )| ≤ (si − ti )β i −αi −1 |Δβ i fi (ti )|dti , ∀si ∈ [0, xi ]. Γ(β i − αi ) 0 We put ui := Δαi fi , gi (si , ti ) := |si − ti |, Φi (ti ) :=

|Δβ i fi (ti )| and ri := β i − αi − 1, i = 1, . . . , n. Γ(β i − αi )

Here

ϕi (si ) :=

si

(si − ti )(ai +bi qi )(β i −αi −1) dti

0 (a +b q )(β −α −1)+1

=

si i i i i i > 0, ∀s ∈ (0, xi ], (ai + bi qi )(β i − αi − 1) + 1

with ϕi (0) = 0. Now the proof is clear. Note. Theorem 19.14 is related and is similar to Theorem 3.2 of [187], but in terms of weights is more general. We give "n Corollary 19.15. All are as in Theorem 19.14 and 1/p := i=1 1/pi , r > 0. Then ' n % n 1 xi 1/qi 1/rp C xi I5 ≤ p p 0 i=1 i=1 i -r (1/rp pi β (a (β −α −1)+1) i i Δ i fi (ti ) (xi − ti ) i dti .

19.2 Main Results

515

Proof. By using inequality of means and (19.12). We need Deﬁnition 19.16. [17, 19, 101, pp. 539–540]. Let ν > 0, n := [ν] and α := ν − n (0 < α < 1). Let f ∈ C([0, x]), s ∈ [0, x]. We consider the Riemann–Liouville fractional integral s 1 (s − t)ν−1 f (t)dt. (Jν f )(s) = Γ(ν) 0 We consider the subspace C ν ([0, x]) of C n ([0, x]): # $ C ν ([0, x]) := f ∈ C n ([0, x]) : J1−α f (n) ∈ C 1 ([0, x]) . Hence, letting f ∈ C ν ([0, x]), we deﬁne the generalized ν -fractional derivative of f over [0, x] as f (ν) := J1−α f (n) . Notice that J1−α f (n) (s) =

1 Γ(1 − α)

s

(s − t)−α f (n) (t)dt

0

exists for f ∈ C ν ([0, x]). We also need Lemma 19.17. [17, 19, pp. 544–545]. Let ν i ≥ 1, γ i ≥ 0, ν i − γ i − 1 ≥ 0, i = 1, . . . , n. Call ni := [ν i ]. Let fi ∈ C ν i ([0, xi ]), xi > 0, such that (j) fi (0) = 0, j = 0, 1, . . . , ni − 1. It follows that fi ∈ C γ i ([0, xi ]) and (γ i )

fi

(si ) =

1 Γ(ν i − γ i )

si 0

(si −ti )ν i −γ i −1 fi

(ν i )

(ti )dti , ∀si ∈ [0, xi ], i = 1, . . . , n. (19.13)

We make Remark 19.18. (on Lemma 19.17). That is, by (19.13) we have (γ ) f i (si ) ≤ i

si

(si − ti )ν i −γ i −1

0

(ν )

|fi i (ti )| dti , ∀si ∈ [0, xi ]; i = 1, . . . , n. Γ(ν i − γ i ) (19.14)

We set here (γ i )

ui := fi

, gi (si , ti ) := |si − ti |, ri := ν i − γ i − 1 ≥ 0,

516

19. General Hilbert–Pachpatte-Type Integral Inequalities (ν )

Φi (ti ) :=

|fi i (ti )| , all si , ti ∈ [0, xi ]; i = 1, . . . , n. Γ(ν i − γ i )

So that (19.4) is fulﬁlled. By applying Theorem 19.3 now we obtain Theorem 19.19. Let i ∈ {1, . . . , n}, xi > 0, ν i ≥ 1, γ i ≥ 0, ν i −γ i −1 ≥ (j) 0, ni := [ν i ]. Let fi ∈ C ν i ([0, xi ]), such that fi (0) = 0," j = 0, 1, . . . , ni −1. n Let pi , qi > 1 : 1/pi + 1/qi = 1, and wi > 0 such that i=1 wi = Ωn , and ai , bi ∈ [0, 1] such that ai + bi = 1. Then I6

:

x1

=

D

xn

···

0

≤

n &

0

n

1/qi

xi

1 Ωn

)

xi 0

i=1

where D :=

i=1

n i=1

n " i=1

(ν i )

|fi

s

(γ i )

|fi

(si )|

((ai +bi qi )(ν i −γ i −1)+1)/qi wi

wi ((ai +b i

i qi )(ν i −γ i −1)+1)

Ωn ds1 · · · dsn

1/qi wi

(ti )|pi (xi − ti )(ai (ν i −γ i −1)+1) dti

1/pi , (19.15)

1 . Γ(ν i − γ i )(ai (ν i − γ i − 1) + 1)1/pi

We give Corollary 19.20. All are as in Theorem 19.19 and 1/p := r > 0. Then ' n % n 1 xi (ν ) 1/qi 1/rp D xi |fi i (ti )|pi I6 ≤ p p i 0 i=1 i=1 -r (1/rp ×(xi − ti )(ai (ν i −γ i −1)+1) dti

"n i=1

1/pi ,

.

Proof. By using inequality of means and (19.15). We make Remark 19.21. The following are taken from [405]. Let f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]), n ≥ 0, and the Wronskians Wi (x) := W [u0 (x), u1 (x), . . . , ui (x)] :=

19.2 Main Results

u0 (x) u0 (x) .. . (i) u0 (x)

u1 (x) u1 (x) (i)

u1 (x)

, i = 0, 1, . . . , n. (i) ui (x)

··· ···

ui (x) ui (x)

···

517

(19.16)

Assume Wi (x) > 0 over [a, b]. For i ≥ 0, the diﬀerential operator of order i (Widder derivative): Li f (x) :=

W [u0 (x), u1 (x), . . . , ui−1 (x), f (x)] , Wi−1 (x)

i = 1, . . . , n + 1; L0 f (x) := f (x), ∀x ∈ [a, b]. Consider also u0 (t) u1 (t) ··· ui (t) u (t) u1 (t) ··· ui (t) 0 1 .. gi (x, t) := . Wi (t) (i−1) (i−1) (i−1) u0 (t) u1 (t) ··· ui (t) u (x) u1 (x) ··· ui (x) 0

(19.17)

, (19.18)

i = 1, 2, . . . , n; g0 (x, t) := u0 (x)/u0 (t), ∀x, t ∈ [a, b]. Example [405] Sets {u0 , u1 , . . . , un } are {1, x, x2 , . . . , xn }, {1, sin x, − cos x, − sin 2x, cos 2x, . . . , (−1)n−1 sin nx, (−1)n cos nx}, and so on. We also mention the generalized Widder–Taylor formula; see [405]. Theorem 19.22. Let the functions f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]), and the Wronskians W0 (x), W1 (x), . . . , Wn (x) > 0 on [a, b], x ∈ [a, b]. Then for t ∈ [a, b] we have f (x) = f (t)

u0 (x) + L1 f (t)g1 (x, t) + · · · + Ln f (t)gn (x, t) + Rn (x), (19.19) u0 (t)

where Rn (x) :=

x

gn (x, s)Ln+1 f (s)ds.

(19.20)

t

For example, one could take u0 (x) = c > 0. If ui (x) = xi , i = 0, 1, . . . , n, deﬁned on [a, b], then Li f (t) = f (i) (t), and gi (x, t) = We need

(x − t)i , t ∈ [a, b]. i!

518

19. General Hilbert–Pachpatte-Type Integral Inequalities

Corollary 19.23. (on Theorem 19.22). By additionally assuming for ﬁxed t ∈ [a, b] that Li f (t) = 0, i = 0, 1, . . . , n , we get that

t

gn (x, s)Ln+1 f (s)ds,

f (x) =

∀x ∈ [a, b].

(19.21)

x

We make Remark 19.24. Here we make use of (19.21). For i = 1, . . . , n let the function sets # $ fi , ui0 , ui1 , . . . , uini ⊆ C ni +1 ([0, xi ]), xi > 0, ni ∈ N. Assume that the Wronskians Wi0 , Wi1 , . . . , Wini > 0 on [0, xi ], i = 1, . . . , n. Also assume that Lij fi (0) = 0; all j = 0, 1, . . . , ni , for all i = 1, . . . , n. By (19.21) we obtain that si fi (si ) = gini (si , ti )Li,ni +1 fi (ti )dti , all i = 1, . . . , n, ∀si ∈ [0, xi ]. 0

(19.22)

Thus |fi (si )| ≤

si

gini (si , ti )Li,ni +1 fi (ti )dti , i = 1, . . . , n, ∀si ∈ [0, xi ].

0

(19.23) From [405] we derive that gini (si , ti ) ≥ 0 when si ≥ ti ≥ 0, i = 1, . . . , n. When si > ti then gini (si , ti ) > 0. Using Theorem 19.3 we derive Theorem 19.25. Here for i ∈ {1, . . . , n}, xi > 0, assume {fi , ui0 , ui1 , . . . , uini } ⊆ C ni +1 ([0, xi ]), ni ∈ N. Suppose that the Wronskians Wi0 , Wi1 , . . . , Wini > 0 on [0, xi ]. Further assume that Lij fi (0) = 0; all j = 0, 1, . . . , ni , for all i = 1, . . . , n. "n Take pi , qi > 1 : 1/pi + 1/qi = 1, and wi > 0 such that i=1 wi = Ωn , and ai , bi ∈ [0, 1] such that ai + bi = 1. Call si (ai +bi qi ) gini (si , ti ) ψ i (si ) := dti . (19.24) 0

19.2 Main Results

519

Then

x1

I7 :=

···

0

≤

n & xn

0

n

1/qi

'

xi

i=1

|fi (si )|

i=1

1 Ωn

n " i=1

xi

Li,n

i +1

wi (ψ i (si

Ωn ds1 · · · dsn ))1/qi wi

pi fi (ti )

0

(1/pi ai gini (si , ti ) dsi dti .

xi

(19.25)

ti

Proof. In Theorem 19.3 put ui := fi , gi := gini , Φi := Li,ni +1 (fi ), ϕi := ψ i , i = 1, . . . , n. Notice here that fi ∈ C([0, xi ]), gin ∈ C([0, xi ]2 ), Li,n +1 (fi ) ∈ C([0, xi ]), ri = 1; all i = 1, . . . , n. i i So (19.4) is fulﬁlled by (19.23). Also for si > 0 we get that ψ i (si ) > 0 with ψ i (0) = 0. Finally we apply (19.5) to get (19.25). We give "n Corollary 19.26. All are as in Theorem 19.25 and 1/p := i=1 1/pi , r > 0. Then ' n % n 1 xi 1/q Li,n +1 fi (ti )pi xi i I7 ≤ p1/rp i p 0 i=1 i=1 i xi -r (1/rp ai gini (si , ti ) dsi dti × . ti

Proof. By using inequality of means and (19.25). We need Setting 19.27. Here we use the notation from [239, pp. 145–154]. Let I be a closed interval of R. Let αj (x), j = 0, 1, . . . , m − 1 (m ∈ N), h(x) be continuous functions on I, and let L = Dm + αm−1 (x)Dm−1 + · · · + α0 (x) be a ﬁxed linear diﬀerential operator on C m (I). Let y1 (x), . . . , ym (x) be a set of linear independent solutions to Ly = 0. Here the associated Green’s function for L is given by

520

19. General Hilbert–Pachpatte-Type Integral Inequalities

y1 (t) y (t) 1 .. H(x, t) := . (m−2) y1 (t) y (x) 1 y1 (t) ··· y (t) ··· 1 = .. . (m−2) (t) ··· y1 y (m−1) (t) ··· 1

(m−2) ··· ym (t) ··· ym (x) ym (t) ym (t) , (m−2) ym (t) (m−1) ym (t) ··· ···

ym (t) ym (t)

which is a continuous function on I 2 . Take a ﬁxed x0 ∈ I; then x H(x, t)h(t)dt, all x ∈ I y(x) =

(19.26)

(19.27)

x0

is the unique solution of the initial value problem Ly = h; y (j) (x0 ) = 0, j = 0, 1, . . . , m − 1.

(19.28)

Based on Setting 19.27 we make Remark 19.28. Let i ∈ {1, . . . , n}, [0, xi ] ⊆ R, xi > 0. Let αi,j (si ), j = 0, 1, . . . , mi − 1 (mi ∈ N), hi (si ) be continuous functions on [0, xi ], and let Li = Dmi + αi,mi −1 (si )Dmi −1 + · · · + αi,0 (si ) be a ﬁxed linear diﬀerential operator on C mi ([0, xi ]). Let yi1 (si ), . . . , yimi (si ) be a set of linear independent solutions to Li y = 0. Here the associated Green’s function for Li is denoted by Hi (si , ti ), which is a continuous function on [0, xi ]2 . Then si Hi (si , ti )hi (ti )dti , all si ∈ [0, xi ] (19.29) yi (si ) = 0

is the unique solution of the initial value problem Li y = hi ; y (j) (0) = 0, j = 0, 1, . . . , mi − 1; all i = 1, . . . , n.

(19.30)

Let εi > 0, then si |Hi (si , ti )| + εi |hi (ti )|dti , i = 1, . . . , n. |yi (si )| ≤

(19.31)

0

We give

19.2 Main Results

521

Theorem 19.29. All elements are as in " Remark 19.28. Take pi , qi > n 1 : 1/pi + 1/qi = 1, and wi > 0 such that i=1 wi = Ωn , and ai , bi ∈ [0, 1] : ai + bi = 1. Call si (ai +bi qi ) ψ i (si ) := |Hi (si , ti )| + εi dti , i = 1, . . . , n. (19.32) 0

Then

x1

···

I8 := 0

≤

n & xn

0

1 Ωn

)

n

1/qi

xi

n " i=1

xi

Ωn ds1 · · · dsn

wi (ψ i (si

))1/qi wi

|hi (ti )|pi

0

i=1

xi

|yi (si )|

i=1

|Hi (si , ti )| + εi

ai

dsi dti

*1/pi .

(19.33)

ti

Proof. In Theorem 19.3 put ui := yi , gi := |Hi (·, ·)| + εi , Φi := |h i |, ϕi := ψ i , i = 1, . . . , n. Notice here that yi ∈ C([0, xi ]), |Hi (·, ·)| + εi ∈ C([0, xi ]2 ), |hi | ∈ C([0, xi ]), ri = 1; all i = 1, . . . , n. So (19.4) is fulﬁlled by (19.31). Also for si > 0 we get ψ i (si ) > 0 with ψ i (0) = 0. Finally we apply (19.5) to get (19.33). We give Corollary 19.30. All are as in Remark 19.28 and Theorem 19.29, and "n 1/p = i=1 1/pi , r > 0. Then ' n % n 1 xi 1/qi 1/rp xi |hi (ti )|pi I8 ≤ p p i 0 i=1 i=1 xi -r (1/rp ai |Hi (si , ti )| + εi dsi dti × . ti

Proof. By using inequality of means and (19.33). Note 19.31. One can give several examples based on all the above results by assuming for instance, that all ai = 1, bi = 0 or ai = 0 and bi = 1, with or without qi = n, wi =

1 n , pi = , ri = r; all i = 1, . . . , n, n n−1

522

19. General Hilbert–Pachpatte-Type Integral Inequalities

also for pi = qi = n = 2. Especially and additionally in Theorems 19.5–19.9 one can assume mi = m, ki = k. Also in Theorem 19.14 we can set αi = α, β i = β; in Theorem 19.19 we can set ν i = ν, γ i = γ, all i = 1, . . . , n, and so on.

20 General Multivariate Hilbert–Pachpatte-Type Integral Inequalities

In this chapter we present general weighted Hilbert–Pachpatte-type multivariate integral inequalities. These are with regard to ordinary partial derivatives and fractional partial derivatives of Canavati and Riemann– Liouville types. These results are applications of a general theorem we present for integrable multivariate functions. This treatment relies on [42].

20.1 Introduction In this chapter we present several weighted general Hilbert–Pachpattetype multidimensional integral inequalities for ordinary partial derivatives and fractional partial derivatives of Canavati and Riemann–Liouville kinds. First we establish a very general multivariate result regarding integrable functions (see Theorem 20.3) to which the rest of the results are applications. Along the way many other interesting results are given. Our results are mainly inspired by [189]. To stimulate the reader’s interest, we give an interesting result from [322]. In this theorem, H(I ×J) denotes the class of functions u ∈ C (n−1,m−1) (I × J) such that D1i u(0, t) = 0, 0 ≤ i ≤ n − 1, t ∈ J, D2j u(s, 0) = 0, 0 ≤ j ≤ m − 1, s ∈ I, and D1n D2m−1 u(s, t) and D1n−1 D2m u(s, t) are absolutely continuous on I × J. Here I, J are intervals of the type Iξ = [0, ξ), ξ ∈ R. Theorem 20.1. [322, Theorem 1] Let u(s, t) ∈ H(Ix × Iy ) and v(k, r) ∈ H(Iz ×Iw ). Then, for 0 ≤ i ≤ n−1, 0 ≤ j ≤ m−1, the following inequality holds. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 523 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 20,

524

20. General Multivariate Hilbert–Pachpatte-Type Inequalities

|D1i D2j u(s, t)D1i D2j v(k, r)|dkdr dsdt 2n−2i−1 t2m−2j−1 + k 2n−2i−1 r 2m−2j−1 0 0 0 0 s x y 1/2 2 1 √ (Ai,j Bi,j )2 xyzw ≤ (x − s)(y − t) D1n D2m u(s, t) dsdt 2 0 0 z w 1/2 n m 2 × (z − k)(w − r) D1 D2 u(k, r) dkdr , (20.1)

x

y

0

z

w

0

where Aij

:

Bij

=

1 , (n − i − 1)!(m − j − 1)! 1 . (2n − 2i − 1)(2m − 2j − 1)

=

(20.2)

This chapter’s results generalize the results of [189, 322], and then mainly treat fractional partial derivatives.

20.2 Symbols and Basics By Z(Z+ ) and R(R+ ) we denote the sets of all (nonnegative) integers and (nonnegative) real numbers. We work with functions of d variables, where d is a ﬁxed positive integer, writing the variable as a vector s = (s1 , . . . , sd ) ∈ Rd . A multi-index m is an element m = (m1 , . . . , md ) of Zd+ . As usual, the factorial of a multi-index m is deﬁned by m! = · · · md !. An integer j may be regarded as the multi-index (j, . . . , j) depending on the context. For vectors in Rd and multi-indices we use the usual operations of vector addition and multiplication of vectors by scalars. We write s ≤ τ (s < τ ) if sj ≤ τ j (sj < τ j ) for 1 ≤ j ≤ d. The same convention applies to multi-indices. In particular, s ≥ 0 (s > 0) means sj ≥ 0 (sj > 0) if 1 ≤ j ≤ d. If s = (s1 , . . . , sd ) ∈ Rd and s > 0, we deﬁne the cell Q(s) = [0, s1 ] × · · · × [0, sj ] × · · · × [0, sd ]; replacing the factor [0, sj ] by {0} in this product, we have the face ∂j Q(s) of Q(s). Let s = (s1 , . . . , sd ), τ = (τ 1 , . . . , τ d ) ∈ Rd , s, τ > 0, let k = (k 1 , . . . , k d ) be a multi-index, and let u : Q(s) → R. Write Dj = ∂/∂sj . We use the following notation: sτ

Dk u(s) s

u(τ )dτ 0

=

1

(s1 )τ · · · (sd )τ d , 1

d

= D1k · · · Ddk u(s), s1 sd = ··· u(τ )dτ 1 · · · dτ d . 0

0

(20.3)

20.2 Symbols and Basics

525

An exponent α ∈ R in the expression sα , where s ∈ Rd , is regarded as a multiexponent; that is, sα = s(α,...,α) . Let N ∈ N be ﬁxed. The following notation and hypotheses are used throughout the chapter. I = {1, . . . , N } mi , i ∈ I

mi = (m1i , . . . , mdi ) ∈ Zd+

xi , i ∈ I

xi = (x1i , . . . , xdi ) ∈ Rd , xi > 0

pi , qi , i ∈ I

pi , qi > 1, "N 1 i=1 p =

p, q ai , bi , i ∈ I wi , i ∈ I

1 pi

+

1 qi

1 1 pi , q

=

=1 "N

1 i=1 qi

a, bi ∈ R+ , ai + bi = 1 "N wi ∈ R, wi > 0, i=1 wi = ΩN .

Throughout this chapter, ui , vi , Φi denote functions from Q(xi ) to R that are measurable with further Lp -type assumptions. If m is a multi-index and x ∈ Rd , x > 0, then C m (Q(x)) denotes the set of all functions u : Q(x) → R that possess continuous derivatives Dk u, where 0 ≤ k ≤ m. We further denote αi := (ai + bi qi )ci ∈ Rd , β i := ai ci ∈ Rd , for ci ∈ Rd , i ∈ I. In particular we observe that 1/qi

xi

= (x1i )1/qi · · · (xdi )1/qi ,

N

(αi + 1)1/qi =

i=1

N d

(αji + 1)1/qi .

i=1 j=1

We give and need the next basic result. Proposition 20.2. In the space C m (Q(x)), m is a multi-index; the order of diﬀerentiation does not matter. That is, Dk u(s) (see (20.3)) remains the 1 d same under any permutation of D1k , . . . , Ddk . Proof. It is enough to prove it for d = 2, which we prove by mathematical ˜ (Q(x)); m, ˜ n ∈ N. We prove in a neighborhood induction. So let f ∈ C (m,n) of (x0 , y0 ) ∈ Q(x) that j ∂ ∂j ∂kf ∂k = P , f, 0 ≤ j ≤ m, ˜ 0 ≤ k ≤ n, ∂xj ∂y k ∂xj ∂y k P stands for any permutation of the involved diﬀerentials. Claim: It holds ∂ ∂k ∂k ∂ f, 0 ≤ k ≤ n. f = ∂x ∂y k ∂y k ∂x

(20.4)

526

20. General Multivariate Hilbert–Pachpatte-Type Inequalities

Proof of Claim: For n = 0 we have (20.4) is true trivially. Using Theorem 6.20, p. 121 of [68] (i.e., if fx , fy , fyx are continuous, then fxy exists and fyx = fxy ), we have that (20.4) is true when n = 1. Next we use mathematical induction. Assume the claim true for n; we ˜ . We prove that prove it for n + 1. So let f ∈ C (m,n+1) ∂ ∂ n+1 ∂ n+1 ∂ f. f= n+1 ∂x ∂y ∂y n+1 ∂x ˜ ˜ Notice that C (m,n+1) (Q(x)) ⊂ C (m,n) (Q(x)), and for 0 ≤ k ≤ n by the induction hypothesis it holds

∂k ∂ ∂ ∂k f. f = ∂x ∂y k ∂y k ∂x ˜ But g := ∂/∂yf ∈ C (m,n) . Hence by the induction hypothesis we have

∂ ∂n ∂n ∂ g; g = ∂x ∂y n ∂y n ∂x That is, ∂ ∂ n+1 f ∂ n+1 ∂f ∂ n ∂ ∂f = , = ∂x ∂y n+1 ∂y n ∂x ∂y ∂y n+1 ∂x proving the claim. Let now 0 ≤ ≤ m, ˜ 0 ≤ k ≤ n. Using the above claim we obtain k ∂ ∂k ∂ ∂ ∂k ∂ ∂ −1 ∂ −1 f f = f = ∂x ∂y k ∂x−1 ∂x ∂y k ∂x−1 ∂y k ∂x ∂ ˜ set h := f ∈ C (m−1,n) ∂x ∂k ∂ ∂k ∂ −2 ∂ −1 h = h = ∂x−1 ∂y k ∂x−2 ∂x ∂y k ∂k ∂ ∂ −2 h = ∂x−2 ∂y k ∂x ∂h ˜ set ρ := ∈ C (m−2,n) ∂x ∂ ∂kρ ∂ −2 ∂ k ρ ∂ −3 = = ∂x−2 ∂xk ∂x−3 ∂x ∂y k ∂k ∂ ∂ −3 ρ ··· = ∂x−3 ∂y k ∂x ∂ ∂k =P , f, true. ∂x ∂y k

20.3 Main Results

527

20.3 Main Results We give our ﬁrst very general result on which the next results rely. Theorem 20.3. Let vi ∈ L1 (Q(xi )), Φi ∈ L∞ (Q(xi )), Φi ≥ 0, ci > −1; ci ∈ Rd , i = 1, . . . , N . Set 1

U :=

N & i=1

.

[(αi + 1)1/qi (β i + 1)1/pi ]

Let Ji := {j ∈ {1, . . . , d} | −1 < cji < 0}. If Ji = ∅ we take (1 + cji ) bi > max . j∈Ji cji (1 − qi ) Suppose

|vi (si )| ≤

si

(si − τ i )ci Φi (τ i )dτ i , ∀si ∈ Q(xi ), i = 1, . . . , N.

(20.5)

0

Then M

:

0

≤ U

xN

···

=

N

N &

x1

0

1/q xi i

i=1

|vi (si )|ds1 · · · dsN

i=1 1 ΩN

N

xi

N " i=1

(α +1)/qi wi wi si i

Ω N 1/pi

(xi − si )

β i +1

pi

Φi (si ) dsi

. (20.6)

0

i=1

Proof. One can write

(si − τ i )ci Φi (τ i ) = (si − τ i ) ai /qi +bi )ci (si − τ i ) ai /pi ci Φi (τ i ).

Next we rewrite (20.5), and apply H¨ older’s inequality and Fubini’s theorem to get |vi (si )|

si

≤

(si − τ i )

ai /qi +bi ci

(si − τ i )

0

≤ =

si

(si − τ i )(ai +bi qi )ci dτ i

0 (αi +1)/qi si (αi + 1)1/qi

1/qi

ai /pi ci

si

× 0

si

(si − τ i )β i Φi (τ i )pi dτ i

Φi (τ i )dτ i (si − τ i )ai ci Φi (τ i )pi dτ i

1/pi

1/pi .

0

That is, we found that (α +1)/q

i s i |vi (si )| ≤ i (αi + 1)1/qi

si

1/pi (si − τ i ) Φi (τ i ) dτ i βi

0

pi

.

(20.7)

528

20. General Multivariate Hilbert–Pachpatte-Type Inequalities

Using the inequality of general means [297, p. 15] we have N

(α +1)/qi si i

≤

i=1

N 1 (α +1)/qi wi wi si i ΩN i=1

ΩN .

(20.8)

We also have by (20.7) that 1/pi N N si (α +1)/qi si i βi pi |vi (si )| ≤ (si − τ i ) Φi (τ i ) dτ i . (20.9) (αi + 1)1/qi i=1 0 i=1 i=1 N

Put 1

W :=

N &

.

(αi + 1)1/qi

i=1

Using (20.9) and (20.8) we ﬁnd

Ω N N 1 (αi +1)/qi wi |vi (si )| ≤ W wi si ΩN i=1 i=1 N

N

si

(si − τ i )β i Φi (τ i )pi dτ i

1/pi .

(20.10)

0

i=1

Thus N &

|vi (si )|

i=1

1 ΩN

N " i=1

Ω N ≤ W

(αi +1)/qi wi

N i=1

wi si

si

1/pi (si − τ i ) Φi (τ i ) dτ i , βi

pi

0

(20.11) whenever at least one si > 0, i ∈ {1, . . . , N }, where si ∈ Q(xi ). Next we integrate (20.11), use H¨older’s inequality and Fubini’s theorem, and at the end we change the order of integration. So we have

x1

xN

···

0

≤ W

N &

0

N i=1

'

|vi (si )|ds1 · · · dsN

i=1

1 ΩN

xi si

N " i=1

(αi +1)/qi wi

0

(

1/pi

(si − τ i ) Φi (τ i ) dτ i βi

0

ΩN

wi si

pi

dsi

20.3 Main Results

≤ W

N

1/q xi i

i=1

=

N & i=1

xi si

0

1/pi

(si − τ i ) Φi (τ i ) dτ i dsi βi

pi

0

W

N

(β i +

1)1/pi i=1

proving (20.6).

529

1/qi

xi

N i=1

xi

1/pi (xi − τ i )β i +1 Φi (τ i )pi dτ i

,

0

Note. Theorem 20.3 and the proof generalize a lot and resemble Theorem 3.1 and its proof in [189]. Corollary 20.4. All are as in Theorem 20.3. Then 'N r (1/rp N 1 xi 1/qi 1/rp β i +1 pi M ≤p U xi (xi − si ) Φi (si ) dsi , p 0 i=1 i=1 i (20.12) where r > 0. Proof. By the inequality of means, for any Ai ≥ 0, r > 0, we obtain N 1/rp N 1 1/pi r Ai ≤ p A . p i i=1 i=1 i Plug into it

Ai :=

xi

(xi − si )β i +1 φi (si )pi dsi .

0

Thus inequality (20.12) is now established. Next we apply Theorem 20.3 to ordinary partial derivatives. Convention 20.5. Here we assume that mi , ki are multi-indices satisfying 0 ≤ ki ≤ mi − 1, and write αi = (ai + bi qi )(mi − ki − 1), β i = ai (mi − ki − 1).

(20.13)

Recall that mi − ki − 1 = (m1i − ki1 − 1, . . . , md1 − kid − 1). We give Theorem 20.6. Under Convention 20.5, let ui ∈ C mi (Q(xi )) be such that Djr ui (si ) = 0 for si ∈ ∂j Q(xi ), 0 ≤ r ≤ mji − 1, 1 ≤ j ≤ d, i ∈ I . Then N & |Dki ui (si )|ds1 · · · dsN x1 xN i=1 M1 := ··· Ω N N " 0 0 (αi +1)/qi wi 1 wi si ΩN i=1

530

20. General Multivariate Hilbert–Pachpatte-Type Inequalities

≤ U1

N

1/q xi i

i=1

N

xi

1/pi (xi − si )

β i +1

|D

mi

pi

ui (si )| dsi

,

(20.14)

0

i=1

where U1 :=

1 N &

(mi − ki − 1)!(αi +

1)1/qi (β

i=1

i

+

1)1/pi

.

(20.15)

Proof. Under the assumptions of the theorem we have from [319] that si 1 ki D ui (si ) = (si − τ i )mi −ki −1 Dmi ui (τ i )dτ i , i ∈ I. (mi − ki − 1)! 0 (20.16) So using (20.6) with vi (si ) = Dki ui (si ), ci = mi − ki − 1, and Φi (si ) = |Dmi ui (si )|/(mi − k1 − 1)! we get (20.14).

We give Corollary 20.7. All are as in Theorem 20.6. Then M1 ≤ p

1/rp

U1

N

' 1/q xi i

i=1

xi r (1/rp N 1 β i +1 mi pi (xi − si ) |D ui (si )| dsi , p 0 i=1 i (20.17)

where r > 0.

Proof. Apply (20.12).

Note. Theorem 20.6 and Corollary 20.7 are similar but generalize Theorem 4.1 and Corollary 4.2 of [189]. We proceed with

Theorem 20.8. Let ui ∈ C 1 (Q(xi )), 1 = (1, . . . , 1) ∈ Nd , be such that ui (si ) = 0 for si ∈ ∂j (Q(si )), 1 ≤ j ≤ d, i ∈ I. Then M2 :

x1

=

···

0

≤

N i=1

N & xN

0

1/qi

xi

|ui (si )|ds1 · · · dsN

i=1

1 ΩN

N xi i=1

0

N " i=1

1/qi wi

Ω N

wi si

pi (xi − si )D1 ui (si ) dsi

1/pi . (20.18)

20.3 Main Results

Proof. By [319] we have that si D1 ui (τ i )dτ i , ui (si ) =

531

i ∈ I.

0

In Theorem 20.3 set vi (si ) = ui (si ), ci = 0, Φi (si ) = D1 ui (si ) (see here αi = β i = 0, U = 1), and the result follows. We have Corollary 20.9. All are as in Theorem 20.8. Then M2 ≤ p1/rp

N

' 1/q xi i

i=1

xi r N 1 pi 1 (xi − si ) D ui (si ) dsi p 0 i=1 i

(1/rp . (20.19)

Proof. Apply (20.12).

Note 20.10. Theorem 20.8 and Corollary 20.9 are similar and generalize Theorem 4.3 and Corollary 4.4 of [189]. Next we apply Theorem 20.3 to fractional partial derivatives. To keep things simple we work at d = 2. We start with those in the Canavati [101] sense. But ﬁrst we need to build some necessary background. We need the following univariate concept. Deﬁnition 20.11. [101, 17, 19, pp. 539–540] Let ν > 0, n ˜ := [ν], [·] the integral part, and α := ν − n ˜ (0 < α < 1). Let f ∈ C([0, x]), x ∈ R+ − {0}, s ∈ [0, x]. We consider the Riemann–Liouville fractional integral s 1 (s − t)ν−1 f (t)dt. (20.20) (Jν f )(s) = Γ(ν) 0 We consider the subspace C ν ([0, x]) of C n˜ ([0, x]): # $ C ν ([0, x]) := f ∈ C n˜ ([0, x]) : J1−α f (˜n) ∈ C 1 ([0, x]) . Hence, let f ∈ C ν ([0, x]); we deﬁne the generalized ν -fractional derivative of f over [0, x] as f (ν) := J1−α f (˜n) . (20.21) Notice that

J1−α f (˜n) =

1 Γ(1 − α)

s

(s − t)−α f (˜n) (t)dt

0

exists for f ∈ C ([0, x]), ∀s ∈ [0, x]; x ∈ R+ − {0}. ν

532

20. General Multivariate Hilbert–Pachpatte-Type Inequalities

We need the following univariate result. Lemma 20.12. [17, 19, pp. 544–545] Let ν ≥ 1, γ ≥ 0, ν − γ − 1 ≥ 0, n ˜ := [ν]. Let f ∈ C ν ([0, x]), x ∈ R+ − {0}, such that f (j) (0) = 0, j = 0, 1, . . . , n ˜ − 1. It follows that f ∈ C γ ([0, x]) and s 1 (γ) f (s) = (s − t)ν−γ−1 f (ν) (t)dt, ∀s ∈ [0, x], x ∈ R+ − {0}. Γ(ν − γ) 0 (20.22) We deﬁne fractional partial derivatives in the Canavati sense. Deﬁnition 20.13. (i) Let ν > 0, n ˜ := [ν], α := ν − n ˜ ; assume here f (·, t) ∈ C ν ([0, x]); x ∈ R+ − {0}, ∀t ∈ [0, y]; y ∈ R+ − {0}. We deﬁne the ν-partial fractional derivative of f with respect to s : ∂ ν f (·, t)/∂sν as ∂ ν f (s, t) ∂ n˜ f (s, t) ∂ := , ∀(s, t) ∈ [0, x] × [0, y]. (20.23) J1−α ∂sν ∂s ∂sn˜ (ii) Let μ > 0, m := [μ], β := μ − m; assume here f (s, ·) ∈ C μ ([0, y]); y ∈ R+ − {0}, ∀s ∈ [0, x]; x ∈ R+ − {0}. We deﬁne the μ-partial fractional derivative of f with respect to t : ∂ μ f (s, ·)/∂tμ as ∂ μ f (s, t) ∂ m f (s, t) ∂ := , ∀(s, t) ∈ [0, x] × [0, y]. (20.24) J 1−β ∂tμ ∂t ∂tm We present Theorem 20.14. Let f : [0, x] × [0, y] → R, x, y ∈ R+ − {0}. Let ν > 0, n := [ν], α := ν − n; ρ > 0, r˜ := [ρ], δ := ρ − r˜. Assume ∂ρf (·, s2 ) ∈ C ν ([0, x]), ∀s2 ∈ [0, y]. ∂sρ2 Suppose also that ∂ ν f /∂sν1 (s1 , ·) ∈ C ρ ([0, y]), ∀s1 ∈ [0, x]. That is, there exist the fractional mixed partial derivatives ∂ ν+ρ f /∂sν1 ∂sρ2 and ∂ ρ+ν f /∂sρ2 ∂sν1 on [0, x] × [0, y], that are meant the obvious iterated way. Suppose further : (1) f ∈ C n,˜r ([0, x] × [0, y]). (2) s2 ∂ r˜f (s2 − s)−δ r˜ (t, s)ds ∈ C n,1 ([0, x] × [0, y]). (20.25) G1 (t, s2 ) := ∂s2 0

20.3 Main Results

533

(3)

s2

M1 (t, s2 ) :=

(s2 − s)−δ

0

∂ n+˜r f (t, s)ds ∈ C 1 ([0, x]) × [0, y]). (20.26) ∂sn1 ∂sr2˜

(4) Let

s1

s2

F (s1 , s2 ) := 0

(s1 − t)−α (s2 − s)−δ

0

∂ n+˜r f (t, s)dsdt. ∂sn1 ∂sr2˜

Assume Fs1 , Fs2 , Fs2 s1 ∈ C([0, x] × [0, y]). (5) s1 ∂n G2 (s1 , s) := (s1 − t)−α n f (t, s)dt ∈ C 1,˜r ([0, x] × [0, y]). ∂s1 0

(20.27)

(20.28)

(6) M2 (s1 , s) :=

s1

(s1 − t)−α

0

Then

∂ n+˜r f (t, s)dt ∈ C 1 ([0, x] × [0, y]). (20.29) ∂sn1 ∂sr2˜

∂ ν+ρ f ∂ ρ+ν f ρ = ν ∂s1 ∂s2 ∂sρ2 ∂sν1

(20.30)

and are continuous on [0, x] × [0, y]. Proof. We observe the following ∂ν ∂sν1

∂ρ f (s1 , s2 ) ∂sρ2

∂ 1 Γ(1 − α)Γ(1 − δ) ∂s1

=

∂ 1 = Γ(1 − α) ∂s1

'

s1

(s1 − t)−α

0

s1

(s1 − t)

−α

0

∂n ∂sn 1

%

∂ ∂s2

s2

∂n ∂sn 1

∂ρ f (t, s2 ) dt ∂sρ2

(s2 − s)−δ

0

∂ r˜ f (t, s)ds ∂sr2˜

- ( dt

(by Proposition 20.2) ∂ 1 = Γ(1 − α)Γ(1 − δ) ∂s1

'

s1

(s1 − t)

−α

0

∂ ∂s2

%

∂n ∂sn 1

s2

(s2 − s)

0

−δ

∂ r˜ f (t, s)ds ∂sr2˜

- ( dt

(by Theorem 24.5, p. 193 of [10]) ∂ 1 = Γ(1 − α)Γ(1 − δ) ∂s1

=

'

∂2 1 Γ(1 − α)Γ(1 − δ) ∂s1 ∂s2

s1

(s1 − t)

0

s1 0

s2 0

−α

∂ ∂s2

s2

(s2 − s)

0

(s1 − t)−α (s2 − s)−δ

−δ

( r ∂ n+˜ f (t, s)ds dt r ˜ ∂sn 1 ∂s2

r ∂ n+˜ f (t, s)dsdt =: A. r ˜ ∂sn 1 ∂s2

534

20. General Multivariate Hilbert–Pachpatte-Type Inequalities

Similarly we have ∂ρ ∂sρ2

∂ν f (s1 , s2 ) ∂sν1

∂ 1 = Γ(1 − α)Γ(1 − δ) ∂s2

= '

∂ 1 Γ(1 − δ) ∂s2

s2

(s2 − s)

−δ

0

s2

(s2 − s)−δ

0

∂ r˜ ∂sr2˜

%

∂ ∂s1

s1

∂ r˜ ∂sr2˜

∂ν f (s , s) ds 1 ∂sν1 −α

- ( ∂n f (t, s)dt ds ∂sn 1

(s1 − t)−α

∂n f (t, s)dt ds ∂sn 1

(s1 − t)

0

(by Proposition 20.2) =

∂ 1 Γ(1 − α)Γ(1 − δ) ∂s2

s2

(s2 − s)−δ

0

∂ ∂s1

∂ r˜ ∂sr2˜

s1 0

(by Theorem 24.5, p. 193 of [10]) =

1 ∂ Γ(1 − α)Γ(1 − δ) ∂s2

s2

(s2 − s)−δ

0

∂ ∂s1

s1

(s1 − t)−α

0

r ∂ n+˜ f (t, s)dt ds r ˜ ∂sn 1 ∂s2

s2 s1 n+˜ r ∂2 1 −δ −α ∂ = (s2 − s) (s1 − t) f (t, s)dtds r ˜ Γ(1 − α)Γ(1 − δ) ∂s2 ∂s1 ∂sn 0 0 1 ∂s2 s1 s2 r ∂ n+˜ ∂2 1 (s1 − t)−α (s2 − s)−δ n r˜ f (t, s)dtds = A, Γ(1 − α)Γ(1 − δ) ∂s1 ∂s2 ∂s1 ∂s2 0 0

proving the claim.

We need Lemma 20.15. Let f ∈ L∞ ([0, A] × [0, B]), where A, B ∈ R+ − {0}. Let

x y

F (x, y) := 0

(x−t)α (y−s)β f (t, s)dtds, ∀x ∈ [0, A], ∀y ∈ [0, B]; α, β ∈ R+ .

0

Then F ∈ C([0, A] × [0, B]). Proof. We notice that

x

y

|x − t|α |y − s|β f (t, s)dtds

F (x, y) = 0

A

0

B

χ[0,x]×[0,y] (t, s)|x − t|α |y − s|β f (t, s)dtds,

= 0

0

∀x ∈ [0, A], ∀y ∈ [0, B]. Let xn ∈ [0, A], yn ∈ [0, B], n ∈ N : (xn , yn ) → (x, y), as n → ∞, x ∈ A, y ∈ B. We have, as n → ∞, that χ[0,xn ]×[0,yn ] (t, s) −→ χ[0,x]×[0,y] (t, s) a.e. in (t, s) ∈ [0, A] × [0, B].

20.3 Main Results

535

Also, as n → ∞, we have that |xn − t|α |yn − s|β f (t, s) → |x − t|α |y − s|β f (t, s), ∀(t, s) ∈ [0, A] × [0, B]. Consequently it holds χ[0,xn ]×[0,yn ] (t, s)|xn − t|α |yn − s|β f (t, s) −→ χ[0,x]×[0,y] (t, s)|x − t|α |y − s|β f (t, s),

a.e. in (t, s) ∈ [0, A] × [0, B]. Furthermore χ[0,xn ]×[0,yn ] (t, s)|xn −t|α |yn −s|β |f (t, s)| ≤ Aα B β |f (t, s)|, ∀(t, s) ∈ [0, A]×[0, B],

and Aα B β |f (t, s)| is integrable. Hence by the dominated convergence theorem we obtain, as n → ∞, that F (xn , yn )

A B

=

0

−→ 0

0

A B 0

χ[0,xn ]×[0,yn ] (t, s)|xn − t|α |yn − s|β f (t, s)dtds χ[0,x]×[0,y] (t, s)|x − t|α |y − s|β f (t, s)dtds = F (x, y),

proving that F is jointly continuous in (x, y).

We present Theorem 20.16. Here i = 1, . . . , N . Let ν i ≥ 1, γ i ≥ 0, ν i − γ i − 1 ≥ 0, ni = [ν i ], and μi , ρi ≥ 1 : μi − ρi − 1 ≥ 0, mi := [μi ], ri := [ρi ]; xi , yi ∈ R+ − {0}. Let fi ∈ C ni ,ri ([0, xi ] × [0, yi ]). Assume that ∂ ρi fi νi ρ (·, si2 ) ∈ C ([0, xi ]), ∀si2 ∈ [0, yi ], ∂si2i and ∂j

∂sji1

∂ ρi fi ρ ∂si2i

(0, si2 ) = 0, all j = 0, 1, . . . , ni − 1; ∀si2 ∈ [0, yi ].

Also suppose that ∂ ν i fi (si1 , ·) ∈ C μi ([0, yi ]), ∀si1 ∈ [0, xi ], ∂sνi1i and ∂j ∂sji2

∂ ν i fi ∂sνi1i

(si1 , 0) = 0, all j = 0, 1, . . . , mi − 1; ∀si1 ∈ [0, xi ]. μ

We suppose∂ μi +ν i fi /∂si2i ∂sνi1i ∈ L∞ ([0, xi ] × [0, yi ]). Further assume that fi fulﬁlls all assumptions of Theorem 20.14 (by indexing all there by i). Then we get the continuous function 1 ∂ γ i +ρi fi (si1 , si2 ) = γ ρ Γ(ν i − γ i )Γ(μi − ρi ) ∂si1i ∂si2i

si1 0

si2 0

(si1 −ti1 )ν i −γ i −1 (si2 −ti2 )μi −ρi −1

536

20. General Multivariate Hilbert–Pachpatte-Type Inequalities

×

∂ μi +ν i fi (ti1 , ti2 )dti2 dti1 , ∀(si1 , si2 ) ∈ [0, xi ] × [0, yi ], μ ∂si2i ∂sνi1i

(20.31)

all i = 1, . . . , N . Proof. By using Lemma 20.12 we obtain that ∂ ρi fi γi ρ (·, si2 ) ∈ C ([0, xi ]), ∀si2 ∈ [0, yi ]. ∂si2i Furthermore we get

∂ ρi fi (si1 , si2 ) ρ ∂si2i si1 ∂ ν i ∂ ρi fi 1 = (si1 − ti1 )ν i −γ i −1 ν i (t , s ) dti1 i1 i2 Γ(ν i − γ i ) 0 ∂si1 ∂sρi2i si1 ∂ ν i +ρi 1 = (si1 − ti1 )ν i −γ i −1 ν i ρi fi (ti1 , si2 )dti1 , Γ(ν i − γ i ) 0 ∂si1 ∂si2 ∂ γ i +ρi fi ∂γi γi ρi (si1 , si2 ) = γ ∂si1 ∂si2 ∂si1i

∀si1 ∈ [0, xi ]; ∀si2 ∈ [0, yi ].

(20.32)

Again applying Lemma 20.12 we get ∂ ν i fi (ti1 , ·) ∈ C ρi ([0, yi ]), ∀ti1 ∈ [0, xi ], ∂sνi1i and

=

1 Γ(μi − ρi )

∂ ρi ρ ∂si2i si2

∂ ν i fi ∂sνi1i

(ti1 , si2 )

(si2 − ti2 )μi −ρi −1

0

∂ μi μ ∂si2i

∂ ν i fi ∂sνi1i

(ti1 , ti2 )dti2 ,

∀si2 ∈ [0, yi ]; ∀ti1 ∈ [0, xi ]. Thus we have 1 ∂ ρi +ν i fi ρi ν i (ti1 , si2 ) = Γ(μ ∂si2 ∂si1 i − ρi ) ×

si2

(si2 − ti2 )μi −ρi −1

0

∂ μi +ν i fi (ti1 , ti2 )dti2 , ∀si2 ∈ [0, yi ]; ∀ti1 ∈ [0, xi ]. μ ∂si2i ∂sνi1i

(20.33)

By Theorem 20.14 we obtain that ∂ ρi +ν i fi ∂ ν i +ρi fi over [0, xi ] × [0, yi ]. ρi ρ νi = ∂si2 ∂si1 ∂sνi1i ∂si2i

(20.34)

20.3 Main Results

537

Using (20.34) and plugging (20.33) into (20.32) we ﬁnally ﬁnd si1 1 = (si1 − ti1 )ν i −γ i −1 Γ(ν i − γ i )Γ(μi − ρi ) 0 si2 × (si2 − ti2 )μi −ρi −1

∂ γ i +ρi fi (si1 , si2 ) γ ρ ∂si1i ∂si2i

0

∂ fi (ti1 , ti2 )dti2 dti1 , μ ∂si2i ∂sνi1i μi +ν i

(20.35)

all i = 1, . . . , N ; there by proving the claim. Continuity of ∂ γ i +ρi fi (si1 , si2 )/ γ ρ ∂si1i ∂si2i is arrived at via Lemma 20.15. We continue with an application of Theorem 20.3. Theorem 20.17. All here are as in Theorem 20.16. Call αi1 := (ai + bi qi )(ν i − γ i − 1), αi2 := (ai + bi qi )(μi − ρi − 1); β i1 := ai (ν i − γ i − 1), β i2 := ai (μi − ρi − 1); all i = 1, . . . , N . Then x1 y1 x2 y2 xN yN ∗ ··· M := 0

0

0

0

0

0

N γ i +ρi & fi (si1 ,si2 ) ∂ ds11 ds12 ds21 ds22 · · · dsN 1 dsN 2 γi ρi i=1

∂si1 ∂si2

1 ΩN

≤U

∗

N

N " i=1

1/qi

(xi yi )

i=1

(α +1)/qi wi (αi2 +1)/qi wi si2

Ω N

wi si1 i1

N i=1

xi

0

yi

(xi − si1 )β i1 +1 (yi − si2 )β i2 +1

0

1/pi pi μ +ν i ∂ i fi × μi ν i (si1 , si2 ) dsi1 dsi2 . ∂si2 ∂si1

(20.36)

Here U ∗ :=

N & i=1

1 . 1/qi 1/pi

(αi1 + 1)(αi2 + 1) (β i1 + 1)(β i2 + 1)

Proof. Here we use (20.31). Call vi (si1 , si2 ) :=

∂ γ i +ρi fi (si1 , si2 ) ∈ C([0, xi ] × [0, yi ]) γ ρ ∂si1i ∂si2i

(20.37)

538

20. General Multivariate Hilbert–Pachpatte-Type Inequalities

and by (20.31), μ +ν ∂ i i fi (si1 , si2 ) 1 Φi (si1 , si2 ) := μ Γ(ν i − γ )Γ(μ − ρ ) ∈ L∞ ([0, xi ]×[0, yi ]), ∂si2i ∂sνi1i i i i and use Theorem 20.3. Then (20.36) is obvious. We give Corollary 20.18. All are as in Theorem 20.17. Then M

∗

'

xi yi N 1 (xi − si1 )β i1 +1 (yi − si2 )β i2 +1 pi 0 0 i=1 i=1 pi r (1/rp μ +ν ∂ i i fi × dsi1 dsi2 , (20.38) ∂sμi ∂sν i (si1 , si2 ) ≤

1/rp

p

U

i2

∗

N

1/qi

(xi yi )

i1

where r ∈ R+ − {0}. Proof. By (20.36) and (20.12).

Finally, we deal with the Riemann–Liouville fractional partial derivatives. Again we work at d = 2. First we need to build the necessary background. Deﬁnition 20.19. (See [187].) Let α ∈ R+ − {0} . For any f ∈ L1 (0, x); x ∈ R+ − {0}, the Riemann–Liouville fractional integral of f of order α is deﬁned by s 1 (s − t)α−1 f (t)dt, ∀s ∈ [0, x], (20.39) (Jα f )(s) := Γ(α) 0 and the Riemann–Liouville fractional derivative of f of order α by m s d 1 (s − t)m−α−1 f (t)dt, (20.40) Δα f (s) := Γ(m − α) ds 0 where m := [α] + 1, [·] is the integral part. In addition, we set Δ0 f := f =: J0 f , J−α f := Δα f if α > 0, and Δ−α f := Jα f , if 0 < α ≤ 1. If α ∈ N, then Δα f = f (α) . Deﬁnition 20.20. [187] We say that f ∈ L1 (0, x) has an L∞ fractional derivative Δα f in [0, x] , x ∈ R+ − {0}, iﬀ Δα−k f ∈ C([0, x]), k = 1, . . . , m := [α] + 1; α ∈ R+ − {0}, and Δα−1 f ∈ AC([0, x]) (absolutely continuous functions) and Δα f ∈ L∞ (0, x). We also need

20.3 Main Results

539

Lemma 20.21. [187] Let α ∈ R+ , β > α, let f ∈ L1 (0, x), x ∈ R+ −{0}, have an L∞ fractional derivative Δβ f in [0, x], and let Δβ−k f (0) = 0 for k = 1, . . . , [β] + 1. Then s 1 α Δ f (s) = (s − t)β−α−1 Δβ f (t)dt, ∀s ∈ [0, x]. (20.41) Γ(β − α) 0 Clearly here Δα f ∈ L∞ (0, x); thus Δα f ∈ L1 (0, x). We need Deﬁnition 20.22. Let f : [0, x] × [0, y] → R, x, y ∈ R+ − {0}, such that f (·, s2 ) ∈ L1 (0, x), ∀s2 ∈ [0, y]; f (s1 , ·) ∈ L1 (0, y), ∀s1 ∈ [0, x]; α, β ∈ R+ − {0}. We deﬁne the Riemann–Liouville fractional partial derivatives with respect to s1 (resp., s2 ) of f of order α (resp., order β) by m s1 ∂ 1 := (s1 − t1 )m−α−1 f (t1 , s2 )dt1 , Γ(m − α) ∂s1 0 (20.42) where m := [α] + 1, ∀s1 ∈ [0, x], all s2 ∈ [0, y]; and Δα s1 f (s1 , s2 )

Δβs2 f (s1 , s2 ) :=

r s2 ∂ 1 (s2 − t2 )r−β−1 f (s1 , t2 )dt2 , (20.43) Γ(r − β) ∂s2 0

where r := [β] + 1, ∀s2 ∈ [0, y], all s1 ∈ [0, x]. We deﬁne the Riemann–Liouville fractional mixed (iterated ) partial derivatives as β β α+β : = Δα Δα s1 Δs2 f (s1 , s2 ) s1 Δs2 f (s1 , s2 ) := Δs1 s2 f (s1 , s2 ), β+α : = Δβs2 Δα Δβs2 Δα s1 f (s1 , s2 ) s1 f (s1 , s2 ) := Δs2 s1 f (s1 , s2 ), ∀(s1 , s2 ) ∈ [0, x] × [0, y], deﬁned the obvious way via (20.40). We need and give a commutativity result for the above partials. Theorem 20.23. Let f ∈ L∞ ((0, x1 ) × (0, x2 )), x1 , x2 ∈ R+ − {0}. Let β 1 , α2 ∈ R+ − {0}, r1 := [β 1 ] + 1, m2 := [α2 ] + 1. Assume that 2 f (s1 , ·) ∈ L1 (0, x2 ) has Δα s2 f (s1 , ·) ∈ L∞ (0, x2 ), ∀s1 ∈ [0, x1 ], and be given β α2 2 Δs2 f (·, s2 ) ∈ L1 (0, x1 ) form Δs11 (Δα s2 f (·, s2 )) over [0, x1 ], ∀s2 ∈ [0, x2 ]. β Also assume that f (·, s2 ) ∈ L1 (0, x1 ) has Δs11 f (·, s2 ) ∈ L∞ (0, x1 ), ∀s2 ∈ β β1 2 [0, x2 ], and be given Δs11 f (s1 , ·) ∈ L1 (0, x2 ) form Δα s2 (Δs1 f (s1 , ·)) over [0, x2 ], ∀s1 ∈ [0, x1 ]. We suppose further :

540

20. General Multivariate Hilbert–Pachpatte-Type Inequalities

(1)

s2

M1 (t1 , s2 ) :=

(s2 − t2 )m2 −α2 −1 f (t1 , t2 )dt2 ∈ C m2 ([0, x1 ] × [0, x2 ]).

0

(20.44)

(2)

s1

M2 (s1 , t2 ) :=

(s1 − t1 )r1 −β 1 −1 f (t1 , t2 )dt1 ∈ C r1 ([0, x1 ] × [0, x2 ]).

0

(20.45)

(3)

s1

λ(s1 , s2 ) := 0

s2

(s1 − t1 )r1 −β 1 −1 (s2 − t2 )m2 −α2 −1 f (t1 , t2 )dt2 dt1

0

∈ C r1 ,m2 ([0, x1 ] × [0, x2 ]). Then there exist

β +α Δs11s2 2 f ,

α +β Δs22s1 1 f ,

(20.46)

are continuous; and

2 2 +β 1 f = Δα Δβs11s+α s2 s1 f 2

(20.47)

on [0, x1 ] × [0, x2 ]. Proof. We observe that 2 2 f (s1 , s2 ) = Δβs11 Δα Δβs11s+α s2 f (s1 , s2 ) 2 =

r1 s1 ∂ 1 2 (s1 − t1 )r1 −β 1 −1 Δα s2 f (t1 , s2 )dt1 Γ(r1 − β 1 ) ∂s1 0 r1 s1 ∂ 1 = (s1 − t1 )r1 −β 1 −1 Γ(r1 − β 1 )Γ(m2 − α2 ) ∂s1 0 m2 s2 ∂ m2 −α2 −1 × (s2 − t2 ) f (t1 , t2 )dt2 dt1 ∂s2 0

(by [10, Theorem 24.5, p. 193]) =

s1

0

s2

r1 m2 ∂ ∂ 1 Γ(r1 − β 1 )Γ(m2 − α2 ) ∂s1 ∂s2 (s1 − t1 )r1 −β 1 −1 (s2 − t2 )m2 −α2 −1 f (t1 , t2 )dt2 dt1

0

= 0

s1

0

s2

∂ r1 +m2 1 2 Γ(r1 − β 1 )Γ(m2 − α2 ) ∂sr11 ∂sm 2

(s1 − t1 )r1 −β 1 −1 (s2 − t2 )m2 −α2 −1 f (t1 , t2 )dt2 dt1 =: B < +∞.

20.3 Main Results

541

Also we have

β α2 2 +β 1 1 Δα s2 s1 f (s1 , s2 ) = Δs2 Δs1 f (s1 , s2 ) m2 s2 ∂ 1 (s2 − t2 )m2 −α2 −1 Δβs11 f (s1 , t2 )dt2 = Γ(m2 − α2 ) ∂s2 0 m2 s2 ∂ 1 = (s2 − t2 )m2 −α2 −1 Γ(r1 − β 1 )Γ(m2 − α2 ) ∂s2 0 r1 s1 ∂ r1 −β 1 −1 (s1 − t1 ) f (t1 , t2 )dt1 dt2 ∂s1 0

(by [10, Theorem 24.5, p. 193]) m2 r1 ∂ ∂ 1 = Γ(r1 − β 1 )Γ(m2 − α2 ) ∂s2 ∂s1

s2 s1

0

m2 −α2 −1

(s2 − t2 )

r1 −β 1 −1

(s1 − t1 )

f (t1 , t2 )dt1 dt2

0

=

s1

0

s2

∂ m2 +r1 1 r1 2 Γ(r1 − β 1 )Γ(m2 − α2 ) ∂sm 2 ∂s1

(s1 − t1 )r1 −β 1 −1 (s2 − t2 )m2 −α2 −1 f (t1 , t2 )dt2 dt1

0

(by Proposition 20.2) = 0

s1

s2

∂ r1 +m2 1 2 Γ(r1 − β 1 )Γ(m2 − α2 ) ∂sr11 ∂sm 2

(s1 − t1 )r1 −β 1 −1 (s2 − t2 )m2 −α2 −1 f (t1 , t2 )dt2 dt1 = B,

0

proving the claim.

We proceed with Theorem 20.24. Here i = 1, . . . , N . Let α i1 , αi2 ∈ R+ ; β i1 ,β i2 ∈ R+ − {0}: β i1 ≥ αi1 + 1, β i2 ≥ αi2 + 1; fi ∈ L∞ (0, xi1 ) × (0, xi2 ) , xi1 , xi2 ∈ i2 R+ −{0}. Assume that fi (si1 , ·) ∈ L1 (0, xi2 ) has Δα si2 fi (si1 , ·) ∈ L∞ (0, xi2 ), αi2 ∀si1 ∈ [0, xi1 ]. Assume that Δsi2 fi (·, si2 ) ∈ L1 (0, xi1 ) has an L∞ partial β i2 fractional derivative Δsi1i1 Δα si2 fi (·, si2 ) in [0, xi1 ], and i2 Δβsi1i1 −ki1 Δα si2 fi (0, si2 ) = 0, for ki1 = 1, . . . , [β i1 ] + 1, ∀si2 ∈ [0, xi2 ]. β

Suppose that fi (·, si2 ) ∈ L1 (0, xi1 ) has Δsi1i1 fi (·, si2 ) ∈ L∞ (0, xi1 ), ∀si2 ∈ β [0, xi2 ]. Suppose that Δsi1i1 fi (ti1 , ·) ∈ L1 (0, xi2 ) has an L∞ fractional deriva β i2 β i1 tive Δsi2 Δsi1 fi (ti1 , ·) in [0, xi2 ], and let Δβsi2i2 −ki2 Δβsi1i1 fi (ti1 , 0) = 0,

542

20. General Multivariate Hilbert–Pachpatte-Type Inequalities β +β

for all ki2 = 1, . . . , [β i2 ]+1, ∀ti1 ∈ [0, xi1 ]. Suppose Δsi2i2si1 i1 fi ∈ L∞ ((0, xi1 ) ×(0, xi2 )). Further assume that fi fulﬁlls all assumptions of Theorem 20.23 (by indexing all there by i). Then we get the continuous function i1 +αi2 Δα si1 si2 fi (si1 , si2 ) =

si1

× 0

0

si2

1 Γ(β i1 − αi1 )Γ(β i2 − αi2 )

i1 f (t , t )dt dt , (si1 −ti1 )β i1 −αi1 −1 (si2 −ti2 )β i2 −αi2 −1 Δβsi2i2s+β i i1 i2 i2 i1 i1

∀(si1 , si2 ) ∈ [0, xi1 ] × [0, xi2 ], all i = 1, . . . , N.

(20.48)

Proof. By using Lemma 20.21 we get αi2 i1 Δα si1 Δsi2 fi (si1 , si2 ) =

1 Γ(β i1 − αi1 )

si1

0

i2 (si1 − ti1 )β i1 −αi1 −1 Δβsi1i1 Δα si2 fi (ti1 , si2 ) dti1 , (20.49)

∀si1 ∈ [0, xi1 ], all si2 ∈ [0, xi2 ]. Also by Lemma 20.21 we have β i2 i1 Δα si2 Δsi1 fi (ti1 , si2 ) 1 = Γ(β i2 − αi2 )

si2

0

(si2 −ti2 )β i2 −αi2 −1 Δβsi2i2 Δβsi1i1 fi (ti1 , ti2 ) dti2 , (20.50)

∀si2 ∈ [0, xi2 ], all ti1 ∈ [0, xi1 ]. That is, i2 +β i1 Δα si2 si1 fi (ti1 , si2 )

=

1 Γ(β i2 − αi2 )

0

si2

i1 f (t , t )dt , (si2 − ti2 )β i2 −αi2 −1 Δβsi2i2s+β i i1 i2 i2 i1

(20.51)

∀si2 ∈ [0, xi2 ], all ti1 ∈ [0, xi1 ]. By Theorem 20.23 we have existence and continuity of partials and i2 i2 +β i1 fi = Δα Δβsi1i1s+α si2 si1 fi , i2

(20.52)

on [0, xi1 ] × [0, xi2 ], all i = 1, . . . , N . Using (20.52) and (20.51) into (20.49) i1 +αi2 we derive (20.48). Continuity of Δα si1 si2 fi (si1 , si2 ) is arrived at via Lemma 20.15. Next we give the ﬁnal application of Theorem 20.3. Theorem 20.25. All here are as in Theorem 20.24. Call α ˜ i1 := (ai + ˜ := ai (β − αi1 − 1), ˜ i2 := (ai + bi qi )(β i2 − αi2 − 1); β bi qi )(β i1 − αi1 − 1), α i1 i1

20.3 Main Results

543

˜ := ai (β − αi2 − 1); all i = 1, . . . , N . Then β i2 i2 x11 x12 x21 x22 xN 1 xN 2 Θ : = ··· 0

N & i=1

0

0

1 ΩN

˜ ≤ U

0

0

0

i1 +αi2 Δα si1 si2 fi (si1 , si2) ds11 ds12 ds21 ds22

N

N " i=1

(˜ α +1)/qi wi (˜ α +1)/qi wi wi si1 i1 si2 i2

1/qi

(xi1 xi2 )

i=1

· · · dsN 1 dsN 2

N i=1

xi1

0

xi2

0

pi i1 f (s , s ) × Δβsi2i2s+β dsi1 dsi2 i i1 i2 i1

Ω N

˜

˜

(xi1 − si1 )β i1 +1 (xi2 − si2 )β i2 +1

1/pi .

(20.53)

Here 1 . N 1/qi 1/pi

& ˜ ˜ (˜ αi1 + 1)(˜ (β i1 + 1)(β i2 + 1) αi2 + 1)

˜ := U

(20.54)

i=1

Proof. Here we use (20.48). Call i1 +αi2 vi (si1 , si2 ) := Δα si1 si2 fi (si1 , si2 ) ∈ C([0, xi1 ] × [0, xi2 ])

by (20.48), and i1 f (s , s ) Φi (si1 , si2 ) := Δβsi2i2s+β i i1 i2 i1 1 ∈ L∞ ([0, xi1 ] × [0, xi2 ]), Γ(β i1 − αi1 )Γ(β i2 − αi2 ) and use Theorem 20.3. Then (20.53) is obvious. We give Corollary 20.26. All are as in Theorem 20.25. Then 'N N 1 xi1 1/rp ˜ 1/qi U θ ≤ p (xi1 xi2 ) p 0 i=1 i=1 i xi2 ˜ ˜ (xi1 − si1 )β i1 +1 (xi2 − si2 )β i2 +1 0

pi ×Δβs i2s+β i1 fi (si1 , si2 ) dsi1 dsi2 i2 i1

r (1/rp ,

(20.55)

544

20. General Multivariate Hilbert–Pachpatte-Type Inequalities

where r ∈ R+ − {0}. Proof. By (20.53) and (20.12).

Note 20.27. One can give several examples based on all the above results by assuming, for instance, that all ai = 1, b1 = 0 or ai = 0 and bi = 1; with or without qi = N , wi = 1/N , pi = N/N − 1, ci = c∗ ; all i = 1, . . . , N . Also for pi = qi = N = 2, all i = 1, . . . , N . Especially and additionally in Theorem 20.6 one can assume mi = m∗ , ki = k ∗ , all i = 1, . . . , N . Also in Theorem 20.17 one can put μi = μ∗ , ρi = ρ∗ , ν i = ν ∗ , γ i = γ ∗ ; all i = 1, . . . , N . ˜ all ˜ , β i2 = β, In Theorem 20.25 one can put αi1 = α∗ , β i1 = β ∗ ; αi2 = α i = 1, . . . , N , and so on. Wishing to keep the length of the chapter short, we do not show these applications here.

21 Other Hilbert–Pachpatte-Type Fractional Integral Inequalities

This is a continuation of Chapters 19 and 20. We present here very general weighted univariate and multivariate Hilbert–Pachpatte-type integral inequalities. These involve Caputo and Riemann–Liouville fractional derivatives and fractional partial derivatives of the mentioned types. This treatment is based on [54]. Of great motivation to write this chapter have been [191, Theorem 316], [320, Theorem 1], [322, Theorem 1], and [147, 187, 188].

21.1 Background Here we follow [134]. We start with Deﬁnition 21.1. Let ν ≥ 0 and the operator Jaν , deﬁned on L1 (a, b) by Jaν f

1 (x) := Γ (ν)

x

ν−1

(x − t)

f (t) dt

(21.1)

a

for a ≤ x ≤ b, be called the Riemann–Liouville fractional integral operator of order ν. For ν = 0, we set Ja0 := I, the identity operator. Here Γ stands for the gamma function. Let α > 0, f ∈ L1 (a, b) , a, b ∈ R; see [134]. Here [·] stands for the integral part of the number. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 545 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 21,

546

Other Hilbert–Pachpatte-Type Fractional Inequalities

We deﬁne the generalized Riemann–Liouville fractional derivative of f of order α by m s d 1 m−α−1 (s − t) f (t) dt, (21.2) Daα f (s) := Γ (m − α) ds a where m := [α] + 1, s ∈ [a, b]; see also [57], Remark 46 there. In addition, we set Da0 f := f, Ja−α f := Daα f, if α > 0, Da−α f := Jaα f, if 0 < α ≤ 1, Dan f = f (n) , for n ∈ N.

(21.3)

We need to mention Deﬁnition 21.2. [134] Let ν ≥ 0, n := ν and · be the ceiling of the number, f ∈ AC n ([a, b]). We call the Caputo fractional derivative x 1 n−ν−1 (n) ν (x − t) f (t) dt, (21.4) D∗a f (x) := Γ (n − ν) a ∀x ∈ [a, b] . ν The above function D∗a f (x) exists almost everywhere for x ∈ [a, b] . Here AC n ([a, b]) is the space of functions with absolutely continuous (n − 1)st derivative, n ∈ N. We need Proposition 21.3. [134] Let ν ≥ 0, n := ν and f ∈ AC n ([a, b]) . Then ν f exists iﬀ the generalized Riemann–Liouville fractional derivative Daν f D∗a exists. Proposition 21.4. [134] Let ν ≥ 0, n := ν . Assume that f is ν such that both D∗a f and Daν f exist. Suppose that f (k) (a) = 0 for k = 0, 1, . . . , n − 1. Then ν f = Daν f. D∗a

(21.5)

In conclusion ν Corollary 21.5. [58] Let ν ≥ 0, n := ν , f ∈ AC n ([a, b]) , D∗a f exists ν (k) or Da f exists, and f (a) = 0, k = 0, 1, . . . , n − 1.

21.1 Background

547

Then ν Daν f = D∗a f.

(21.6)

We need Theorem 21.6. [58] Let ν ≥ 0, n := ν , f ∈ AC n ([a, b]) , and f (a) = 0, k = 0, 1, . . . , n − 1. Then x 1 ν−1 ν (x − t) D∗a f (t) dt. (21.7) f (x) = Γ (ν) a (k)

We also need Theorem 21.7. [58] Let ν ≥ γ + 1, γ ≥ 0. Call n := ν . Assume ν f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and D∗a f ∈ γ L∞ (a, b) . Then D∗a f ∈ AC ([a, b]) , and x 1 ν−γ−1 γ ν D∗a f (x) = (x − t) D∗a f (t) dt, (21.8) Γ (ν − γ) a ∀x ∈ [a, b]. Theorem 21.8. [58] Let ν ≥ γ + 1, γ ≥ 0, and n := ν . Let f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1. Assume ∃Daν f (x) ∈ R, ∀x ∈ [a, b] , and Daν f ∈ L∞ (a, b) . Then Daγ f ∈ AC ([a, b]) , and x 1 ν−γ−1 Daγ f (x) = (x − t) Daν f (t) dt, (21.9) Γ (ν − γ) a ∀x ∈ [a, b] . In [41] we proved the following result which we use here. Theorem 21.9. Here for i ∈ {1, . . . , n} , take xi > 0, and assume 2 ui ∈ L1 (0, xi ) , gi ∈ L∞ (0, xi ) , Φi ∈ L∞ (0, xi ) , with " gi , Φi ≥ 0. Take τ i ≥ 0, pi , qi > 1 : 1/pi + 1/qi = 1, and ω i > 0 such n that i=1 ω i = Ωn , and ai , bi ∈ [0, 1] such that ai + bi = 1. Call si (a +b q )τ ϕi (si ) := (gi (si , τ i )) i i i i dτ i . (21.10) 0

Assume ϕi (si ) > 0, with the exception ϕi (0) = 0. If si τ |ui (si )| ≤ (gi (si , τ i )) i Φi (τ i ) dτ i , si ∈ [0, xi ] , i = 1, . . . , n, (21.11) 0

548

Other Hilbert–Pachpatte-Type Fractional Inequalities

then I1 :=

≤

i=1

0

) 1/q xi i

xn

... 0

n

n &

x1

1 Ωn

n " i=1

xi

pi

Ωn ds1 . . . dsn 1/qi ω i

ω i (ϕi (si ))

xi

(Φi (τ i )) 0

|ui (si )|

i=1

(gi (si , τ i ))

ai τ i

*1/pi

dsi dτ i

. (21.12)

τi

Also from [41] we need Corollary 21.10. All are as in Theorem 21.9 and 1/p := τ > 0. Then ' n n 1 xi 1/q p i 1/τ p xi (Φi (τ i )) i I1 ≤ p p 0 i=1 i=1 i ×

xi

ai τ i

(gi (si , τ i ))

"n i=1

1/pi ,

τ *1/τ p

dsi dτ i

.

(21.13)

τi

For the multivariate results, Section 21.3, we need Symbols 21.11. By Z (Z+ ) and R (R+ ) we denote the sets of all (nonnegative) integers and (nonnegative) real numbers. In the following we work with functions of d variables, 1 whered d is adﬁxed positive integer, writing the variable as a vector s = s , . . . , s ∈ R . A multi-index m is an element m = m1 , . . . , md of Zd+ . An usual, the factorial of a multi-index m is deﬁned by m! = m1 ! . . . md !. An integer j may be regarded as the multi-index (j, . . . , j) depending on the context. For vectors in Rd and multi-indices we use the usual operations of vector addition andmultiplication of vectors by scalars. We write s ≤ τ (s < τ ) if sj ≤ τ j sj < τ j for 1 ≤ j ≤ d. The same convention applies to multi-indices. In particular, s ≥ 0 (s > 0) means sj≥ 0 sj >0 if 1 ≤ j ≤ d. If s = s1 , . . . , sd ∈ Rd and s > 0, we deﬁne the cell

Q (s) = 0, s1 × . . . × 0, sj × . . . × 0, sd ;

replacing the factor 0, sj by {0} in this product, we have the face ∂j Q (s) of Q (s) . Let s = s1 , . . . , sd , τ = τ 1 , . . . , τ d ∈ Rd , s, τ > 0, let k = k 1 , . . . , k d be a multi-index, and let u : Q (s) → R. Write Dj = ∂/∂sj . We use the following notation. τ 1 τ d . . . sd , sτ = s1

21.1 Background 1

d

Dk u (s) = D1k . . . Ddk u (s) , s1 sd s u (τ ) dτ = ... u (τ ) dτ 1 . . . dτ d . 0

0

549

(21.14)

0

An exponent α ∈ R in the expression sα , where s ∈ Rd , is regarded as a multiexponent; that is, sα = s(α,...,α) . Let N ∈ N be ﬁxed. The following notation and hypotheses are used in the following. I = {1, . . . , N } mi , i ∈ I xi , i ∈ I pi , qi , i ∈ I p, q ai , bi , i ∈ I ωi , i ∈ I

mi = m1i , . . . , mdi ∈ Zd+ xi = x1i , . . . , xdi ∈ Rd , xi > 0 pi , qi > 1, p1i + q1i = 1 "N 1 1 "N 1 1 i=1 pi , q = i=1 qi p = a, bi ∈ R+ , ai + bi = 1 "N ω i ∈ R, ω i > 0, i=1 ω i = ΩN .

(21.15)

In the sequel, ui , vi , Φi denote functions from Q (xi ) to R that are measurable with further Lp -type assumptions. If m is a multi-index and x ∈ Rd , x > 0, then C m (Q (x)) denotes the set of all functions u : Q (x) → R that possesses continuous derivatives Dk u, where 0 ≤ k ≤ m. In [42] we proved that in C m (Q (x)) the order of diﬀerentiation does not matter. We further denote αi := (ai + bi qi ) ci ∈ Rd , β i := ai ci ∈ Rd , for ci ∈ Rd , i ∈ I. In particular we observe that 1/qi

xi N

1/qi 1/qi = x1i . . . xdi ,

(αi + 1)

1/qi

i=1

=

d N

1/qi αji + 1 .

(21.16)

i=1 j=1

From [42] we use the following. Theorem 21.12. Let vi ∈ L1 (Q (xi )) , Φi ∈ L∞ (Q (xi )) , Φi ≥ 0, ci > −1; ci ∈ Rd , i = 1, . . . , N. Set 1 (21.17) U := N + ,. & 1/q 1/p (αi + 1) i (β i + 1) i i=1

Let Ji := j ∈ {1, . . . , d} | − 1 < cji < 0 . If Ji = ∅ we take ⎛ bj > max ⎝ j∈Ji

1 + cji

⎞

cji (1 − qi )

⎠.

(21.18)

550

Other Hilbert–Pachpatte-Type Fractional Inequalities

Assume

si

|vi (si )| ≤

c

(si − τ i ) i Φi (τ i ) dτ i , ∀si ∈ Q (xi ) , i = 1, . . . , N. (21.19)

0

Then M1 :=

≤U

1/q xi i

i=1

xN

... 0

N

N &

x1

0

N

xi

|vi (si )| ds1 . . . dsN

i=1 1 ΩN

N " i=1

Ω N (αi +1)/qi ω i

ω i si

1/pi (xi − si )

β i +1

pi

Φi (si ) dsi

.

(21.20)

0

i=1

Finally from [42] we use Corollary 21.13. All are as in Theorem 21.12. Then M1 ≤ p1/τ p U

N

1/qi

xi

i=1

'

xi τ (1/τ p N 1 β i +1 pi (xi − si ) Φi (si ) dsi , p 0 i=1 i

(21.21)

where τ > 0.

21.2 Univariate Results We present the ﬁrst main result of this chapter. Theorem 21.14. Let i ∈ {1, . . . , n} , xi > 0, γ i ≥ 0, ν i ≥ γ i + 1. Let (k ) fi ∈ AC ni ([0, xi ]) , ni := ν i , such that fi i (0) = 0, ki = 0, 1, . . . , ni − 1, νi and D "∗0n fi ∈ L∞ (0, xi ) . Let pi , qi > 1 : 1/pi + 1/qi = 1, and ω i > 0 such that i=1 ω i = Ωn , and ai , bi ∈ [0, 1] such that ai + bi = 1. Then I2 :=

x1

... 0

≤C

n i=1

0

) 1/qi

xi

0

xi

n & Dγ i fi (si ) ds1 . . . dsn

xn

i=1

1 Ωn

n " i=1

p

∗0

[(a +b q )(ν i −γ i −1)+1]/qi ωi s i i i ω i ((ai +b q )(ν −γ −1)+1)1/qi ωi i i i i i

νi |D∗0 fi (ti )| i (xi − ti )

[ai (ν i −γ i −1)+1]

Ω n

*1/pi dti

, (21.22)

21.2 Univariate Results

551

where C :=

n

1 1/pi

i=1

Γ (ν i − γ i ) (ai (ν i − γ i − 1) + 1)

.

(21.23)

Proof. By Theorem 21.7 we get si 1 γ ν −γ −1 νi (si − ti ) i i D∗0 fi (ti ) dti , D∗0i fi (si ) = Γ (ν i − γ i ) 0

(21.24)

∀si ∈ [0, xi ] . γ Here we apply Theorem 21.9. By Theorem 21.7 we have that D∗0i fi ∈ νi AC ([0, xi ]) ; also by assumption D∗0 fi ∈ L∞ (0, xi ) ; and by (21.24) we have si γ 1 ν −γ −1 νi D i fi (si ) ≤ (si − ti ) i i |D∗0 fi (ti )| dti , (21.25) ∗0 Γ (ν i − γ i ) 0 ∀si ∈ [0, xi ] . We set γ

ui := D∗0i fi , gi (si , ti ) := |si − ti | , νi fi (ti )| |D∗0 and Γ (ν i − γ i )

Φi (ti ) :=

τ i := ν i − γ i − 1, i = 1, . . . , n. Here

ϕi (si ) :=

si

(ai +bi qi )(ν i −γ i −1)

(si − ti )

dti =

0 (a +b q )(ν −γ −1)+1

si i i i i i > 0, ∀si ∈ (0, xi ] , (ai + bi qi ) (ν i − γ i − 1) + 1

(21.26)

with ϕi (0) = 0. Now the proof is clear. We give Corollary 21.15. All are as in Theorem 21.14 and 1/p := τ > 0. Then ' n n 1 1/qi 1/τ p C xi I2 ≤ p p i=1 i=1 i 0

xi

νi |D∗0 fi

pi

(ai (ν i −γ i −1)+1)

(ti )| (xi − ti )

"n i=1

1/pi ,

τ *1/τ p dti

.

(21.27)

552

Other Hilbert–Pachpatte-Type Fractional Inequalities

Proof. By using (21.22) and the inequality of means: for any Ai ≥ 0 and τ > 0, we obtain n 1/τ p n 1 1/pi Ai ≤ p Aτi . (21.28) p i i=1 i=1 Here we put

xi

Ai := 0

i = 1, . . . , n.

p

νi |D∗0 fi (ti )| i (xi − ti )

[ai (ν i −γ i −1)+1]

dti ,

(21.29)

We continue with Theorem 21.16. Let i ∈ {1, . . . , n} , xi > 0, γ i ≥ 0, ν i ≥ γ i + 1. Let (k ) fi ∈ AC ni ([0, xi ]) , ni := ν i , such that fi i (0) = 0, ki = 0, 1, . . . , ni − 1, νi D0ν i fi ∈ L∞ (0, xi ) . and there exists D0 fi (si ) ∈ R, all si ∈ [0, xi ] , and " n Let pi , qi > 1 : 1/pi + 1/qi = 1, and ω i > 0 such that i=1 ω i = Ωn , and ai , bi ∈ [0, 1] such that ai + bi = 1. Then n & Dγ i fi (si ) ds1 . . . dsn x1 xn 0 i=1 I3 := ... Ω n ((a +b q )(ν i −γ i −1)+1)/qi ωi n " 0 0 si i i i 1 ω i ((a +b q )(ν −γ −1)+1)1/qi ωi Ωn ≤C

n

) 1/q xi i

i=1

i=1

xi

|D0ν i fi

i

pi

(ti )| (xi − ti )

i i

i

i

[ai (ν i −γ i −1)+1]

*1/pi dti

, (21.30)

0

where C is as in (21.23) . Proof. Very similar to the proof of Theorem 21.14, now using Theorem 21.8 and applying Theorem 21.9. We also give Corollary 21.17. All are as in Theorem 21.16 and 1/p := τ > 0. Then ' n n 1 1/q xi i I3 ≤ p1/τ p C p i=1 i=1 i xi τ *1/τ p p (a (ν −γ −1)+1) |D0ν i fi (ti )| i (xi − ti ) i i i dti .

"n i=1

1/pi ,

(21.31)

0

Proof. By using (21.30) and (21.28) .

One can give a variety of applications of the derived inequalities (21.22) , (21.27) , (21.30) , and (21.31) , for various speciﬁc values of the involved parameters n, pi , qi , ω i , ai , and bi .

21.3 Multivariate Results

553

21.3 Multivariate Results We need the following. Deﬁnition 21.18. (See also [42].) Let f : [0, x] × [0, y] → R, x, y ∈ R+ − {0} , such that f (·, s2 ) ∈ L1 (0, x) , ∀s2 ∈ [0, y] ; f (s1 , ·) ∈ L1 (0, y) , ∀s1 ∈ [0, x] ; α, β ∈ R+ − {0}. We deﬁne the Riemann–Liouville fractional partial derivatives with respect to s1 (resp., s2 ) of f of order α (resp., order β) by m ∂ 1 α Ds1 f (s1 , s2 ) := Γ (m − α) ∂s1 s1 m−α−1 (s1 − t1 ) f (t1 , s2 ) dt1 , (21.32) 0

where m := [α] + 1, ∀s1 ∈ [0, x] , all s2 ∈ [0, y]; and τ ∂ 1 β Ds2 f (s1 , s2 ) := Γ (τ − β) ∂s2 s2 τ −β−1 (s2 − t2 ) f (s1 , t2 ) dt2 ,

(21.33)

0

where τ := [β] + 1, ∀s2 ∈ [0, y] , all s1 ∈ [0, x] . We deﬁne the Riemann–Liouville fractional mixed (iterated) partial derivatives as f (s1 , s2 ) , Dsα1 Dsβ2 f (s1 , s2 ) := Dsα1 Dsβ2 f (s1 , s2 ) := Dsα+β 1 s2 Dsβ2 Dsα1 f (s1 , s2 ) := Dsβ2 Dsα1 f (s1 , s2 ) := Dsβ+α f (s1 , s2 ) , (21.34) 2 s1 ∀ (s1 , s2 ) ∈ [0, x] × [0, y] , deﬁned the obvious iterated way via (21.2) . Clearly in the last (21.34) we assume that Dsβ2 f (s1 , s2 ) ∈ L1 (0, x) , ∀s2 ∈ [0, y], and Dsα1 f (s1 , s2 ) ∈ L1 (0, y) , ∀s1 ∈ [0, x] . We also need Deﬁnition 21.19. Let f : [0, x] × [0, y] → R, x, y ∈ R+ − {0} , such that f (·, s2 ) ∈ AC n ([0, x]) , ∀s2 ∈ [0, y] , where n := ν , ν > 0; f (s1 , ·) ∈ AC m ([0, y]) , ∀s1 ∈ [0, x] , where m := μ , μ > 0. We deﬁne the Caputo fractional partial derivatives with respect to s1 (resp., s2 ) of f of order ν (resp., order μ) by ν D∗s f (s1 , s2 ) := 1

0

s1

(s1 − t1 )

n−ν−1

1 Γ (n − ν)

∂n f (t1 , s2 ) dt1 , ∂sn1

(21.35)

554

Other Hilbert–Pachpatte-Type Fractional Inequalities

∀s1 ∈ [0, x] , all s2 ∈ [0, y] ; and μ D∗s f (s1 , s2 ) := 2

s2

m−μ−1

(s2 − t2 )

0

1 Γ (m − μ)

∂m f (s1 , t2 ) dt2 , ∂sm 2

(21.36)

∀s2 ∈ [0, y] , all s1 ∈ [0, x] . We deﬁne the Caputo fractional mixed (iterated) partial derivatives as μ ν μ ν ν+μ D∗s2 f (s1 , s2 ) := D∗s D∗s f (s1 , s2 ) := D∗s f (s1 , s2 ) , D∗s 1 2 1 1 s2 ν μ ν μ μ+ν D∗s1 f (s1 , s2 ) := D∗s D∗s f (s1 , s2 ) := D∗s f (s1 , s2 ) , D∗s 2 1 2 2 s1

(21.37)

∀ (s1 , s2 ) ∈ [0, x] × [0, y] , deﬁned the obvious iterated way via (21.4) . μ n ([0, x]) , Clearly in the last (21.37) we assume that D∗s 2 f (s1 , s2 ) ∈ AC ν m f (s , s ) ∈ AC ([0, y]) , all s ∈ [0, x] . all s2 ∈ [0, y] , and that D∗s 1 2 1 1 We need Theorem 21.20. Let f be as in Deﬁnition 21.19. Additionally assume that f ∈ C (n,m) ([0, x] × [0, y]) . Then ν+μ μ+ν D∗s f (s1 , s2 ) = D∗s f (s1 , s2 ) , (21.38) 1 s2 2 s1 for all (s1 , s2 ) ∈ [0, x] × [0, y] . Proof. We observe that μ ν+μ ν D∗s2 f (s1 , s2 ) = f (s1 , s2 ) = D∗s D∗s 1 s2 1 1 Γ (n − ν)

s1

n−ν−1

(s1 − t1 )

0

∂n μ D∗s2 f (t1 , s2 ) dt1 = ∂sn1

s1 n 1 n−ν−1 ∂ (s1 − t1 ) Γ (n − ν) Γ (m − μ) 0 ∂sn1 s2 m m−μ−1 ∂ (s2 − t2 ) f (t1 , t2 ) dt2 dt1 ∂sm 0 2

(by [10, Theorem 24.5, p. 193], repeated application) s1 1 = Γ (n − ν) Γ (m − μ) 0 0

s2

n−ν−1

(s1 − t1 )

m−μ−1

(s2 − t2 )

∂ n+m f (t1 , t2 ) dt2 dt1 ∂sn1 ∂sm 2

21.3 Multivariate Results

555

(by Fubini’s theorem and commutativity of ordinary partials) s2 1 = Γ (m − μ) Γ (n − ν) 0 s1 m+n m−μ−1 n−ν−1 ∂ (s2 − t2 ) (s1 − t1 ) f (t , t ) dt dt2 1 2 1 n ∂sm 0 2 ∂s1 (by [10, Theorem 24.5, p. 193], repeated application) s2 m 1 m−μ−1 ∂ = (s2 − t2 ) Γ (m − μ) Γ (n − ν) 0 ∂sm 2 s1 n n−ν−1 ∂ (s1 − t1 ) f (t1 , t2 ) dt1 dt2 = ∂sn1 0 s2 m 1 m−μ−1 ∂ ν D∗s (s2 − t2 ) f (s1 , t2 ) dt2 m 1 Γ (m − μ) 0 ∂s2 μ ν μ+ν D∗s = D∗s f (s1 , s2 ) = D∗s f (s1 , s2 ) , 2 1 2 s1 proving the commutativity of Caputo fractional mixed partials under the assumptions of the theorem. We give the counterpart of the last theorem. Theorem 21.21. All are as in Theorem 21.20. Assume that R Dsμ2 f (s1 , s2 ) exists in s2 ∈ [0, y] , ∀s1 ∈ [0, x] , and ∂ l f /∂sl2 (s1 , 0) = 0, l = 0, 1, . . . , m − 1, ∀s1 ∈ [0,x] . Assume that Dsμ2 f (s1 , s2 ) ∈ AC n ([0, x]) , ∀s2 ∈ [0, y] , and R Dsν1 Dsμ2 f (s1 , s2 ) exists in s1 ∈ [0, x] , ∀s2 ∈ [0, y] , with ∂ k /∂sk1 Dsμ2 f (0, s2 ) = 0, for k = 0, 1, . . . , n − 1, ∀s2 ∈ [0, y] . Also suppose that R Dsν1 f (s1 , s2 ) exists in s1 ∈ [0, x] , ∀s2 ∈ [0, y] , and ∂ k f /∂sk1 (0, s2 ) = 0, k = 0, 1, . . . , n − 1, ∀s2 ∈ [0, y] . And suppose that Dsν1 f (s1 , s2 ) ∈ AC m ([0, y]) , ∀s1 ∈ [0, x] , and R Dsμ2 Dsν1 f (s1 , s2 ) exists in s2 ∈ [0, y] , ∀s1 ∈ [0, x] , and ∂ l /∂sl2 Dsν1 f (s1 , 0) = 0, l = 0, 1, . . . , m − 1, ∀s1 ∈ [0, x] . Then f (s1 , s2 ) = Dsμ+ν f (s1 , s2 ) , (21.39) Dsν+μ 1 s2 2 s1 for all (s1 , s2 ) ∈ [0, x] × [0, y]. Proof. We apply Theorem 21.20 and use Corollary 21.5 repeatedly. Namely we have the following. We notice μ ν+μ ν D∗s2 f (s1 , s2 ) = f (s1 , s2 ) = D∗s D∗s 1 s2 1 μ ν Ds2 f (s1 , s2 ) = Dsν1 Dsμ2 f (s1 , s2 ) = Dsν+μ f (s1 , s2 ) . D∗s 1 1 s2

556

Other Hilbert–Pachpatte-Type Fractional Inequalities

Similarly we have ν μ+ν μ D∗s1 f (s1 , s2 ) = D∗s f (s1 , s2 ) = D∗s 2 s1 2 ν μ Ds1 f (s1 , s2 ) = Dsμ2 Dsν1 f (s1 , s2 ) = Dsμ+ν f (s1 , s2 ) . D∗s 2 2 s1 Now using (21.38) we prove the claim.

We need Lemma 21.22. [42] Let f ∈ L∞ ([0, A] × [0, B]) , where A, B ∈ R+ − {0} . Let x y α β (x − t) (y − s) f (t, s) dtds, (21.40) F (x, y) := 0

0

∀x ∈ [0, A] , ∀y ∈ [0, B] ; α, β ∈ R+ . Then F ∈ C ([0, A] × [0, B]) . We further present Theorem 21.23. Here i = 1, . . . , N. Let αi1 , αi2 ∈ R+ ; β i1 , β i2 ∈ R+ − {0} : β i1 ≥ αi1 + 1, β i2 ≥ αi2 + 1; xi1 , xi2 ∈ R+ − {0} , fi : [0, xi1 ] × [0, xi2 ] −→ R. Call ni1 := β i1 , ni2 := β i2 ; λi1 := αi1 , β +β β +α αi1 +αi2 λi2 := αi2 . The mixed partials D∗s fi , D∗si2i2 si1i1 fi , D∗si1i1 si2i2 fi , and i1 si2 αi2 +β i1 D∗si2 si1 fi are considered according to Deﬁnition 21.19. Assume that fi ∈ C (ni1 ,λi2 ) ([0, xi1 ] × [0, xi2 ]) and are as in Deﬁnition 21.19. αi2 f (·, si2 ) ∈ AC ni1 ([0, xi1 ]) , such that Suppose D∗s i2 i ∂ ki αi2 D∗si2 fi (0, si2 ) = 0, ki = 0, 1, . . . , ni1 − 1, and ki ∂si1 β +α

D∗si1i1 si2i2 fi (·, si2 ) ∈ L∞ (0, xi1 ) , all these ∀si2 ∈ [0, xi2 ] . β

Suppose that D∗si1i1 fi (ti1 , ·) ∈ AC ni2 ([0, xi2 ]) , such that ∂ li β i1 D f (t , 0) = 0, li = 0, 1, . . . , ni2 − 1, and ∗s i i1 i1 ∂slii2 β +β

D∗si2i2 si1i1 fi (ti1 , ·) ∈ L∞ (0, xi2 ) , all these ∀ti1 ∈ [0, xi1 ] . Further suppose β +β

D∗si2i2 si1i1 fi ∈ L∞ ((0, xi1 ) × (0, xi2 )) .

21.3 Multivariate Results

557

Then we obtain the continuous function αi1 +αi2 D∗s fi (si1 , si2 ) = i1 si2

si1

0

si2

β i1 −αi1 −1

(si1 − ti1 )

1 Γ (β i1 − αi1 ) Γ (β i2 − αi2 )

(si2 − ti2 )

β i2 −αi2 −1

β +β

D∗si2i2 si1i1 fi (ti1, ti2 )

0

(21.41)

dti2 dti1 , ∀ (si1 , si2 ) ∈ [0, xi1 ] × [0, xi2 ] , all i = 1, . . . , N. Proof. By using Theorem 21.7 we have αi2 αi1 D∗si2 fi (si1 , si2 ) = D∗s i1

si1

β i1 −αi1 −1

(si1 − ti1 )

0

1 Γ (β i1 − αi1 )

αi2 β D∗si1i1 D∗s f (ti1 , si2 ) dti1 , i2 i

(21.42)

∀si1 ∈ [0, xi1 ] , all si2 ∈ [0, xi2 ] . Again by Theorem 21.7 we obtain β i1 αi2 D∗s D f (t , s ) = ∗s i i1 i2 i1 i2

si2

β i2 −αi2 −1

(si2 − ti2 )

1 Γ (β i2 − αi2 )

β β D∗si2i2 D∗si1i1 fi (ti1 , ti2 ) dti2 ,

(21.43)

0

∀si2 ∈ [0, xi2 ] , all ti1 ∈ [0, xi1 ] . That is, α +β

D∗si2i2 si1i1 fi (ti1 , si2 ) =

si2

(si2 − ti2 )

β i2 −αi2 −1

1 Γ (β i2 − αi2 )

β +β

D∗si2i2 si1i1 fi (ti1 , ti2 ) dti2 ,

(21.44)

0

∀si2 ∈ [0, xi2 ] , all ti1 ∈ [0, xi1 ] . Using Theorem 21.20 we obtain β +α

α +β

D∗si1i1 si2i2 fi = D∗si2i2 si1i1 fi ,

(21.45)

on [0, xi1 ] × [0, xi2 ] , all i = 1, . . . , N. Using (21.45) and (21.44) into (21.42) we derive (21.41) . Continuity of αi1 +αi2 fi (si1 , si2 ) is arrived at via Lemma 21.22. D∗s i1 si2 Next we apply Theorem 21.12 for d = 2 at the fractional level.

558

Other Hilbert–Pachpatte-Type Fractional Inequalities

Theorem 21.24. All here are as in Theorem 21.23. Call α ˜ i1 := (ai + bi qi ) ˜ := ai (β − αi1 − 1) , (β i1 − αi1 − 1) , α ˜ i2 := (ai + bi qi ) (β i2 − αi2 − 1) ; β i1 i1 ˜ := ai (β − αi2 − 1) ; all i = 1, . . . , N. β i2 i2 Then x x x x x x 11

12

21

22

M2 := 0 N & i=1

N2

0

0

0

0

0

α +α D∗si1 s i2 fi (si1 , si2 ) ds11 ds12 ds21 ds22 . . . dsN 1 dsN 2 i1 i2

1 ΩN

˜ ≤U

N1

...

N

N " i=1

Ω N (˜ α +1)/qi ω i (˜ α +1)/qi ω i ω i si1 i1 si2 i2

1/qi

(xi1 xi2 )

i=1

N

˜ +1 β i2

0

i=1

(xi2 − si2 )

xi1

xi2

(xi1 − si1 )

˜ +1 β i1

(21.46)

0

pi 1/pi β i2 +β i1 . D∗si2 si1 fi (si1 , si2 ) dsi1 dsi2

Here ˜ := U

N &

1

)

1/qi

αi2 + 1)) ((˜ αi1 + 1) (˜

i=1

1/pi * . ˜ ˜ β i1 + 1 β i2 + 1

(21.47)

Proof. Here we use (21.41) . Call αi1 +αi2 vi (si1 , si2 ) := D∗s fi (si1 , si2 ) ∈ C ([0, xi1 ] × [0, xi2 ]) i1 si2

by (21.41) , and

Φi (si1 , si2 ) :=

β i2 +β i1 D∗si2 si1 fi (si1 , si2 ) Γ (β i1 − αi1 ) Γ (β i2 − αi2 )

∈ L∞ ([0, xi1 ] × [0, xi2 ])

by assumption; also set ci1 := β i1 − αi1 − 1, ci2 := β i2 − αi2 − 1, and use Theorem 21.12. Then (21.46) is obvious. We give Corollary 21.25. All here are as in Theorem 21.24. Then 'N N 1 xi1 xi2 ˜ +1 1/qi β 1/τ p ˜ M2 ≤ p U (xi1 xi2 ) (xi1 − si1 ) i1 p 0 0 i=1 i=1 i

21.3 Multivariate Results

(xi2 − si2 )

˜ +1 β i2

pi τ *1/τ p β i2 +β i1 , D∗si2 si1 fi (si1 , si2 ) dsi1 dsi2

559

(21.48)

where τ > 0. Proof. By (21.46) and (21.21) . We further give Theorem 21.26. Here i = 1, . . . , N. Let αi1 , αi2 ∈ R+ ; β i1 , β i2 ∈ R+ − {0} : β i1 ≥ αi1 + 1, β i2 ≥ αi2 + 1; xi1 , xi2 ∈ R+ − {0} , fi : [0, xi1 ] × [0, xi2 ] −→ R. Call ni1 := β i1 , ni2 := β i2 ; λi1 := αi1 , β +β β +α i2 λi2 := αi2 . The mixed partials Dsαi1i1s+α fi , Dsi2i2si1 i1 fi , Dsi1i1si2 i2 fi , and i2 α +β Dsi2i2si1 i1 fi are considered according to Deﬁnition 21.18. Assume that i2 i1 f . fi = Dsαi2i2s+β (21.49) Dsβi1i1s+α i i2 i1 (For suﬃcient conditions for (21.49) see Theorem 21.21.) Suppose Dsαi2i2 fi (·, si2 ) ∈ AC ni1 ([0, xi1 ]) , such that ∂ ki αi2 Dsi2 fi (0, si2 ) = 0, ki = 0, 1, . . . , ni1 − 1, ki ∂si1 and there exists in R the next i2 fi (·, si2 ) ∈ L∞ (0, xi1 ) , all these ∀si2 ∈ [0, xi2 ] . Dsβi1i1s+α i2

β

Suppose that Dsi1i1 fi (ti1 , ·) ∈ AC ni2 ([0, xi2 ]) , such that ∂ li β i1 Dsi1 fi (ti1 , 0) = 0, li = 0, 1, . . . , ni2 − 1, li ∂si2 and there exists in R the next i1 f (t , ·) ∈ L Dsβi2i2s+β i i1 ∞ (0, xi2 ) , all these ∀ti1 ∈ [0, xi1 ] . i1

Further assume i1 f ∈ L Dsβi2i2s+β i ∞ ((0, xi1 ) × (0, xi2 )) . i1

Then we obtain the continuous function i2 fi (si1 , si2 ) = Dsαi1i1s+α i2

0

si1

0

si2

β i1 −αi1 −1

(si1 − ti1 )

1 Γ (β i1 − αi1 ) Γ (β i2 − αi2 )

(si2 − ti2 )

β i2 −αi2 −1

i1 f (t Dsβi2i2s+β i i1, ti2 ) i1

560

Other Hilbert–Pachpatte-Type Fractional Inequalities

dti2 dti1 , ∀ (si1 , si2 ) ∈ [0, xi1 ] × [0, xi2 ] , all i = 1, . . . , N.

(21.50)

Proof. By using Theorem 21.8 we have Dsαi1i1 Dsαi2i2 fi (si1 , si2 ) =

si1

β i1 −αi1 −1

(si1 − ti1 )

0

1 Γ (β i1 − αi1 )

Dsβi1i1 Dsαi2i2 fi (ti1 , si2 ) dti1 ,

(21.51)

∀si1 ∈ [0, xi1 ] , all si2 ∈ [0, xi2 ] . Again by Theorem 21.8 we obtain Dsαi2i2 Dsβi1i1 fi (ti1 , si2 ) =

si2

(si2 − ti2 )

β i2 −αi2 −1

0

1 Γ (β i2 − αi2 )

Dsβi2i2 Dsβi1i1 fi (ti1 , ti2 ) dti2 ,

(21.52)

∀si2 ∈ [0, xi2 ] , all ti1 ∈ [0, xi1 ] . That is, i1 f (t , s ) = Dsαi2i2s+β i i1 i2 i1

si2

(si2 − ti2 )

β i2 −αi2 −1

0

1 Γ (β i2 − αi2 )

i1 f (t , t ) dt , Dsβi2i2s+β i i1 i2 i2 i1

(21.53)

∀si2 ∈ [0, xi2 ] , all ti1 ∈ [0, xi1 ] . Using (21.49) and (21.53) into (21.51) we derive (21.50) . Continuity of i2 Dsαi1i1s+α fi (si1 , si2 ) is arrived at via Lemma 21.22. i2 Next we apply Theorem 21.12 for d = 2 at the fractional level again. Theorem 21.27. All here are as in Theorem 21.26. Call α ˜ i1 := (ai + bi qi ) ˜ := ai (β − αi1 − 1) , ˜ i2 := (ai + bi qi ) (β i2 − αi2 − 1) ; β (β i1 − αi1 − 1) , α i1 i1 ˜ := ai (β − αi2 − 1) ; all i = 1, . . . , N. β i2 i2 Then x x x x x x 11

12

21

22

M3 :=

N1

N2

... 0

0

0

0

0

0

N & Dsαi1s+αi2 fi (si1 , si2 ) ds11 ds12 ds21 ds22 . . . dsN 1 dsN 2 i1 i2

i=1

1 ΩN

˜ ≤U

N i=1

N " i=1

Ω N

1/qi

(xi1 xi2 )

(˜ α +1)/qi ω i (˜ α +1)/qi ω i si2 i2

ω i si1 i1

N i=1

0

xi1

0

xi2

(xi1 − si1 )

˜ +1 β i1

(21.54)

21.3 Multivariate Results

(xi2 − si2 )

˜ +1 β i2

561

1/pi β +β Ds i2s i1 fi (si1 , si2 )pi dsi1 dsi2 . i2 i1

Here ˜ := U

N &

1

)

1/qi

αi2 + 1)) ((˜ αi1 + 1) (˜

i=1

1/pi * . ˜ +1 ˜ +1 β β i1 i2

(21.55)

Proof. Here we use (21.50) . Call i2 vi (si1 , si2 ) := Dsαi1i1s+α fi (si1 , si2 ) ∈ C ([0, xi1 ] × [0, xi2 ]) i2

by (21.50) , and

Φi (si1 , si2 ) :=

β i2 +β i1 Dsi2 si1 fi (si1 , si2 ) Γ (β i1 − αi1 ) Γ (β i2 − αi2 )

∈ L∞ ([0, xi1 ] × [0, xi2 ])

by assumption; also set ci1 := β i1 − αi1 − 1, ci2 := β i2 − αi2 − 1, and use Theorem 21.12. Then (21.54) is obvious. We give Corollary 21.28. All here are as in Theorem 21.27. Then 'N N 1 xi1 xi2 ˜ +1 1/qi β 1/τ p ˜ U (xi1 xi2 ) (xi1 − si1 ) i1 M3 ≤ p p 0 0 i=1 i=1 i

(xi2 − si2 )

˜ +1 β i2

τ ,1/τ p β +β Ds i2s i1 fi (si1 , si2 )pi dsi1 dsi2 , i2 i1

(21.56)

where τ > 0. Proof. By (21.54) and (21.21) . One can give a variety of applications of the derived inequalities (21.46) , (21.48) , (21.54), and (21.56) , for various speciﬁc values of the involved parameters, such as N, pi , qi , ai , bi , ω i , αi1 , αi2 , β i1 , and β i2 .

22 Canavati Fractional and Other Approximation of Csiszar’s f -Divergence

Here are presented various sharp and nearly optimal probabilistic inequalities that give best or nearly best estimates for the Csiszar’s f -divergence. These involve Canavati fractional and ordinary derivatives of the directing function f . Also given are lower bounds for the Csiszar’s distance. The Csiszar’s discrimination is the most essential and general measure for the comparison between two probability measures. This treatment is based on [28].

22.1 Preliminaries Throughout this chapter we use the following. (I) Let f be a convex function from (0, +∞) into R that is strictly convex at 1 with f (1) = 0. Let (X, A, λ) be a measure space, where λ is a ﬁnite or a σ-ﬁnite measure on (X, A). And let μ1 , μ2 be two probability measures on (X, A) such that μ1 λ, μ2 λ (absolutely continuous); for example, λ = μ1 + μ2 . Denote by p = dμ1 /dλ, q = dμ2 /dλ the (densities) Radon– Nikodym derivatives of μ1 , μ2 with respect to λ. Here we suppose that p a < a ≤ ≤ b, a.e. on X q and a ≤ 1 ≤ b. The quantity

Γf (μ1 , μ2 ) =

q(x)f X

p(x) q(x)

dλ(x),

(22.1)

G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 563 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 22,

564

22. Canavati Fractional Approximation of Csiszar’s f -Divergence

was introduced by I. Csiszar in 1967 (see [128]), and is called f -divergence of the probability measures μ1 and μ2 . By Lemma 1.1 of [128], the integral (22.1) is well deﬁned and Γf (μ1 , μ2 ) ≥ 0 with equality only when μ1 = μ2 . In [128] the author without proof mentions that Γf (μ1 , μ2 ) does not depend on the choice of λ. We give a proof of that fact in our setting. Lemma 22.1. Call p=

dμ1 , d(μ1 + μ2 )

q=

dμ2 , d(μ1 + μ2 )

p∗ =

dμ1 , dλ

q∗ =

dμ2 . dλ

Then

q(x)f X

p(x) q(x)

d(μ1 + μ2 ) =

q ∗ (x)f

X

p∗ (x) q ∗ (x)

dλ.

(22.2)

Proof. Because f : (0, +∞) → R is a convex function it is continuous there and Borel measurable, making f p(x)/q(x) X-measurable. In general let μ, ν be σ-ﬁnite measures on X such that ν μ and a g-measurable function on X; then it is known that dν g± dν = g± dμ, dμ X X respectively, where g = g+ − g− , g± ≥ 0. Finally we obtain dν g dν = g dμ. dμ X X

(22.3)

Next let E ∈ A be such that λ(E) = 0; then μ1 (E) = μ2 (E) = 0 and (μ1 + μ2 )(E) = 0; that is, μ1 + μ2 λ. Clearly here we have that μ1 μ2

μ1 + μ2 λ, μ1 + μ2 λ.

(22.4) (22.5)

Then it is known that dμ1 dμ1 d(μ1 + μ2 ) = , dλ d(μ1 + μ2 ) dλ

(22.6)

dμ2 d(μ1 + μ2 ) dμ2 = . dλ d(μ1 + μ2 ) dλ

(22.7)

and

That is, p∗ = p

d(μ1 + μ2 ) dλ

(22.8)

22.1 Preliminaries

and q∗ = q

d(μ1 + μ2 ) . dλ

565

(22.9)

Here by assumption 0
p p∗ , ≤ b, q q∗

a.e. on X;

that is, q, q ∗ > 0 a.e. on X. That is, d(μ1 + μ2 )/dλ > 0 a.e. on X. Therefore from the above we get p p∗ a.e. on X. (22.10) = q∗ q Consequently it holds ∗ ∗ p(x) p (x) p (x) ∗ q(x)f f q (x)f d(μ1 +μ2 ) = dμ2 = dλ. q(x) q ∗ (x) q ∗ (x) X X X The concept of f -divergence was introduced ﬁrst in [127] as a generalization of Kullback’s “information for discrimination” or I-divergence ( generalized entropy) [240, 241] and of R´enyi’s “information gain” (I-divergence of order α) [345]. In fact the I-divergence of order 1 equals Γu log2 u (μ1 , μ2 ). The choice f (u) = (u − 1)2 again produces a known measure of diﬀerence of distributions that is called χ2 -divergence; of course the total variation distance |μ1 − μ2 | = X |p(x) − q(x)| dλ(x) equals Γ|u−1| (μ1 , μ2 ). Here by assuming f (1) = 0 we can consider Γf (μ1 , μ2 ) as a measure of the diﬀerence between the probability measures μ1 , μ2 . The f -divergence is in general asymmetric in μ1 and μ2 . But because f is convex and strictly convex at 1 (see Lemma 22.2 next) so is 1 f ∗ (u) = uf (22.11) u and as in [128] we ﬁnd Γf (μ2 , μ1 ) = Γf ∗ (μ1 , μ2 ).

(22.12)

Lemma 22.2. Let f : (0,+∞) → R be a strictly convex function at 1. Then so is f ∗ (u) = uf 1/u , u ∈ (0, +∞). Proof. Let 0 < u1 < 1 < u2 ; then 1/u2 < 1 < 1/u1 . Hence 2 u2 u1 1 1 f + = f u1 + u2 u1 + u2 u1 u1 + u2 u2 u1 1 u2 1 < f + f . u1 + u2 u1 u1 + u2 u2

566

22. Canavati Fractional Approximation of Csiszar’s f -Divergence

Clearly we get

u1 + u2 2

u1 f u11 + u2 f u12 1 . f u1 +u2 < 2 2

Thus we have f∗

u1 + u2 2

<

f ∗ (u1 ) + f ∗ (u2 ) , 2

proving our claim. In information theory and statistics many other concrete divergences are used that are special cases of the above general Csiszar f -divergence, such as Hellinger distance DH , α-divergence Dα , Bhattacharyya distance DB , harmonic distance DHa , Jeﬀrey’s distance DJ , and triangular discrimination DΔ ; for all these see, for example [78, 145]. The problem of ﬁnding and estimating the proper distance (or diﬀerence or discrimination) of two probability distributions is one of the major questions in probability theory. The above f -divergence measures in their various forms have also been applied to anthropology, genetics, ﬁnance, economics, political science, biology, approximation of probability distributions, signal processing, and pattern recognition. A great inspiration for this chapter has been the very important monograph on the topic by S. Dragomir [145]. (II) In the sequel we follow [101]. Let g ∈ C([0, 1]). Let ν be a positive number, n := [ν] and α := ν − n (0 < α < 1). Deﬁne x 1 (x − t)ν−1 g(t) dt, 0 ≤ x ≤ 1, (22.13) (Jν g)(x) := Γ(ν) 0 the Riemann–Liouville integral, where Γ is the gamma function. We deﬁne the subspace C ν ([0, 1]) or C n ([0, 1]) as follows. C ν ([0, 1]) := {g ∈ C n ([0, 1]) : J1−α Dn g ∈ C 1 ([0, 1])}, where D := d/dx. So for g ∈ C ν ([0, 1]), we deﬁne the Canavati ν-fractional derivative of g as (22.14) Dν g := DJ1−α Dn g. When ν ≥ 1 we have the Taylor formula t2 tn−1 + · · · + g (n−1) (0) 2! (n − 1)! + (Jν Dν g)(t), for all t ∈ [0, 1]. (22.15)

g(t) = g(0) + g (0)t + g (0)

When 0 < ν < 1 we ﬁnd g(t) = (Jν Dν g)(t), for all t ∈ [0, 1].

(22.16)

22.1 Preliminaries

567

Next we transfer the above notions over to arbitrary [a, b] ⊆ R (see [17]). Let x, x0 ∈ [a, b] such that x ≥ x0 , where x0 is ﬁxed. Let f ∈ C([a, b]) and deﬁne x 1 x0 (Jν f )(x) := (x − t)ν−1 f (t) dt, x0 ≤ x ≤ b, (22.17) Γ(ν) x0 the generalized Riemann–Liouville integral. We deﬁne the subspace Cxν0 ([a, b]) of C n ([a, b]): x0 Dn f ∈ C 1 ([x0 , b])}. Cxν0 ([a, b]) := {f ∈ C n ([a, b]) : J1−α

For f ∈ Cxν0 ([a, b]), we deﬁne the generalized ν -fractional derivative of f over [x0 , b] as x0 f (n) Dxν0 f := DJ1−α

Observe that x0 (J1−α f (n) )(x) =

1 Γ(1 − α)

(f (n) := Dn f ).

x

(22.18)

(x − t)−α f (n) (t) dt

x0

exists for f ∈ Cxν0 ([a, b]). We recall the following generalization of Taylor’s formula (see [17, 101]). Theorem 22.3. Let f ∈ Cxν0 ([a, b]), x0 ∈ [a, b], ﬁxed. (i) If ν ≥ 1 then f (x) = f (x0 ) + f (x0 )(x − x0 ) + f (x0 ) + · · · + f (n−1) (x0 ) + (Jνx0 Dxν0 f )(x),

(x − x0 )2 2

(x − x0 )n−1 (n − 1)!

for all x ∈ [a, b] : x ≥ x0 .

(22.19)

(ii) If 0 < ν < 1 then f (x) = (Jνx0 Dxν0 f )(x),

for all x ∈ [a, b] : x ≥ x0 .

(22.20)

We make Remark 22.4. (1) (Dxn0 f ) = f (n) , n ∈ N. (2) Let f ∈ Cxν0 ([a, b]), ν ≥ 1 and f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1; n := [ν]. Then by (22.19) f (x) = (Jνx0 Dxν0 f )(x).

568

22. Canavati Fractional Approximation of Csiszar’s f -Divergence

That is, 1 f (x) = Γ(ν)

x

x0

(x − t)ν−1 (Dxν0 f )(t) dt,

(22.21)

for all x ∈ [a, b] with x ≥ x0 . Notice that (22.21) is also true when 0 < ν < 1. We also make Remark 22.5. Let ν ≥ 1, γ ≥ 0, such that ν − γ ≥ 1, so that γ < ν. Call n := [ν], α := ν − n ; m := [γ], ρ := γ − m. Note that ν − m ≥ 1 and n − m ≥ 1. Let f ∈ Cxν0 ([a, b]) be such that f (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. Hence by (22.19) f (x) = (Jνx0 Dxν0 f )(x),

for all x ∈ [a, b] : x ≥ x0 .

Therefore by Leibnitz’s formula and Γ(p + 1) = pΓ(p), p > 0, we get that x0 f (m) (x) = (Jν−m Dxν0 f )(x),

for all x ≥ x0 .

x0 It follows that f ∈ Cxγ0 ([a, b]) and thus (Dxγ0 f )(x) := (DJ1−ρ f (m) )(x) exists for all x ≥ x0 . We easily obtain x 1 x0 f (m) )(x)) = (x − t)(ν−γ)−1 (Dxν0 f )(t) dt, (Dxγ0 f )(x) = D((J1−ρ Γ(ν − γ) x0 (22.22) and thus x0 (Dxν0 f ))(x) (Dxγ0 f )(x) = (Jν−γ

and is continuous in x on [x0 , b].

22.2 Main Results Again f and the whole setting is as in Section 22.1(I). Other notations are as in Section 22.1(II). We present the following. Theorem 22.6. Let a < b, 1 ≤ ν < 2, f ∈ Caν ([a, b]), and a ≤ p(x)/q(x) ≤ b, a.e. on X. Then Daν f ∞,[a,b] Γf (μ1 , μ2 ) ≤ (q(x))1−ν (p(x) − aq(x))ν dλ(x). (22.23) Γ(ν + 1) X Proof. Here n := [ν] = 1 and from (22.21) we have x 1 f (x) = (x − w)ν−1 (Daν f )(w) dw, all a ≤ x ≤ b. Γ(ν) a

(22.24)

22.2 Main Results

Then |f (x)|

≤ ≤ =

x 1 (x − w)ν−1 |(Daν f )(w)| dw Γ(ν) a Daν f ∞,[a,b] x (x − w)ν−1 dw Γ(ν) a Daν f ∞,[a,b] (x − a)ν . Γ(ν + 1)

569

(22.25)

That is, we get |f (x)| ≤

Daν f ∞,[a,b] (x − a)ν , Γ(ν + 1)

for all a ≤ x ≤ b.

(22.26)

Consequently we obtain p(x) p(x) Γf (μ1 , μ2 ) = dλ q(x)f q(x) f dλ ≤ q(x) q(x) X X ν ν Da f ∞,[a,b] p(x) ≤ − a dλ q(x) Γ(ν + 1) q(x) X ν Da f ∞,[a,b] (q(x))1−ν (p(x) − aq(x))ν dλ(x), (22.27) = Γ(ν + 1) X thereby proving (22.23).

The counterpart of the previous result follows. Theorem 22.7. Let a < b, ν ≥ 2, n := [ν], f ∈ Caν ([a, b]), f (i) (a) = 0, i = 0, 1, . . . , n − 1, and a ≤ p(x)/q(x) ≤ b, a.e. on X. Then Daν f ∞,[a,b] (q(x))1−ν (p(x) − aq(x))ν dλ(x). (22.28) Γf (μ1 , μ2 ) ≤ Γ(ν + 1) X Proof. As in the proof of Theorem 22.6.

Next we give an Lα estimate. Theorem 22.8. Let a < b, ν ≥ 1, n := [ν], f ∈ Caν ([a, b]), f (i) (a) = 0, i = 0, 1, . . . , n − 1, and a ≤ p(x)/q(x) ≤ b, a.e. on X. Let α, β > 1: 1/α + 1/β = 1. Then Γf (μ1 , μ2 ) ≤

Daν f α,[a,b] Γ(ν)(β(ν − 1) + 1)1/β

(q(x))2−ν−1/β (p(x)−aq(x))ν−1+1/β dλ(x). X

(22.29)

570

22. Canavati Fractional Approximation of Csiszar’s f -Divergence

Proof. From (22.21) we have x 1 f (x) = (x − w)ν−1 (Daν f )(w) dw, Γ(ν) a

for all a ≤ x ≤ b.

(22.30)

Also (Daν f )(w) ∈ C([a, b]). Hence x 1 (x − w)ν−1 |(Daν f )(w)| dw |f (x)| ≤ Γ(ν) a 1/β x 1/α x 1 ≤ (x − w)β(ν−1) dw |(Daν f )(w)|α dw Γ(ν) a a 1/β x Daν f α,[a,b] ≤ (x − w)β(ν−1) dw Γ(ν) a 1/β Daν f α,[a,b] (x − a)β(ν−1)+1 = . (22.31) Γ(ν) β(ν − 1) + 1 That is, we ﬁnd that |f (x)| ≤

Daν f α,[a,b] ν−1+1/β . (x − a) Γ(ν)(β(ν − 1) + 1)1/β

(22.32)

Therefore Γf (μ1 , μ2 ) ≤ =

ν−1+1/β Daν f α,[a,b] p p dλ ≤ q f q dλ − a q q Γ(ν)(β(ν − 1) + 1)1/β X X

Daν f α,[a,b] Γ(ν)(β(ν − 1) + 1)1/β

proving (22.29).

q 2−ν−1/β (p − aq)ν−1+1/β dλ,

(22.33)

X

An L1 estimate follows. Theorem 22.9. Let a < b, ν ≥ 1, n := [ν], f ∈ Caν ([a, b]), f (i) (a) = 0, i = 0, 1, . . . , n − 1, and a ≤ p(x)/q(x) ≤ b, a.e. on X. Then Daν f 1,[a,b] q(x)2−ν (p(x) − aq(x))ν−1 dλ(x). (22.34) Γf (μ1 , μ2 ) ≤ Γ(ν) X Proof. Again from (22.21) and (22.30) we have x x 1 1 (x−a)ν−1 (x−w)ν−1 |Daν f (w)| dw ≤ |Daν f (w)|dw |f (x)| ≤ Γ(ν) a Γ(ν) a ≤

1 (x−a)ν−1 Γ(ν)

b

|Daν f (w)| dw = a

1 (x−a)ν−1 Daν f 1,[a,b] . (22.35) Γ(ν)

22.2 Main Results

That is, |f (x)| ≤

Daν f 1,[a,b] (x − a)ν−1 , Γ(ν)

571

(22.36)

for all x in [a, b]. Thus ν−1 Daν f 1,[a,b] p p dλ ≤ Γf (μ1 , μ2 ) ≤ − a q f q dλ q q Γ(ν) X X Daν f 1,[a,b] = q 2−ν (p − aq)ν−1 dλ , (22.37) Γ(ν) X

proving (22.34).

Remark 22.10. Let f : [a, b] → R convex, f ∈ C 1 ([a, b]), x0 ∈ [a, b], a = b. Then f (x) ≥ f (x0 ) + f (x0 )(x − x0 ),

for all x ∈ [a, b].

Then one can easily prove that ) * b a+b 1 − x0 ; f (x) dx − f (x0 ) ≥ f (x0 ) Δ(x0 ) := b−a a 2

(22.38)

(22.39)

see also S. Dragomir and C. E. M. Pearce [146, p. 9, Theorem 18]. In this chapter’s setting p(x) ≤ b, a.e. on X, 0

p(x) q(x)

Then

≥ f

p(x) q(x)

)

b

f (x) dx

(22.40)

a

* a + b p(x) − , a.e. on X. 2 q(x)

(22.41)

* (a + b) − p(x) , a.e. on X. 2 (22.42) Here q(x) > 0 a.e. on X. Therefore by integrating (22.42) against λ we obtain p(x) p(x) q(x)(a + b) − p(x) dλ(x). M− q(x)f f dλ(x) ≥ q(x) q(x) 2 X X (22.43) q(x)M − q(x)f

p(x) q(x)

≥ f

p(x) q(x)

)

q(x)

572

22. Canavati Fractional Approximation of Csiszar’s f -Divergence

We have established Theorem 22.11. Let all elements be as in Section 22.1(I), hold, and additionally assume that f ∈ C 1 ([a, b]), a = b. Then Γf (μ1 , μ2 ) ≤

1 b−a

b

f

f (x) dx − a

X

p(x) q(x)

q(x)(a + b) − p(x) dλ. 2 (22.44)

Remark 22.12. Let all elements be as in Section 22.1(I), hold, and f ∈ C 1 ([a, b]). Then by (22.38) we get f (x) ≥ f (1)(x − 1)

and f

p(x) q(x)

≥ f (1)

p(x) −1 , q(x)

(22.45)

a.e. on X.

(22.46)

Thus q(x)f

p(x) q(x)

≥ f (1)(p(x) − q(x)),

a.e. on X.

(22.47)

Integrating (22.47) against λ over X we ﬁnd Γf (μ1 , μ2 ) ≥ 0

(22.48)

which is known and mentioned in [128]. Remark 22.13. Let n ∈ N, f ∈ C n+1 ([a, b]), [a, b] ⊂ R, such that f (n+1) ≥ 0 (≤ 0); then by Taylor’s formula we get that f (x) ≥ (≤)

n i=0

f (i) (x0 )

(x − x0 )i , i!

(22.49)

respectively, for any x, x0 ∈ [a, b] : x ≥ x0 . Inequalities (22.49) are valid also when n is odd and x ≤ x0 . Take 0 < a ≤ 1 ≤ b, a ≤ p/q ≤ b, a.e. on X and f (1) = 0, along with f ∈ C n+1 ([a, b]), n odd, such that f (n+1) ≥ 0 (≤ 0); then n (x − 1)i f (x) ≥ (≤) , (22.50) f (i) (1) i! i=1 for all x in [a, b]. Hence i n q pq − 1 p (i) qf . f (1) ≥ (≤) q i! i=1

(22.51)

22.2 Main Results

573

And ﬁnally we obtain Γf (μ1 , μ2 ) ≥ (≤)

n f (i) (1) i=2

i!

q 1−i (p − q)i dλ,

n odd.

(22.52)

X

Remark 22.14. Let n be odd, f ∈ C n+1 ([a, b]), [a, b] ⊂ R, such that ≥ 0 (≤ 0), x, x0 ∈ [a, b]; then as before we have f (n+1)

f (x) − f (x0 ) ≥ (≤)

n

f (i) (x0 )

i=1

(x − x0 )i . i!

(22.53)

Then

1 b−a

b

f (x) dx − f (x0 )

=

a

1 b−a

b

(f (x) − f (x0 )) dx a

= =

n b

f (i) (x0 ) (x − x0 )i dx i! a i=1 n 1 f (i) (x0 ) b (x − x0 )i dx b − a i=1 i! a

1 ≥ (≤) b−a

n b 1 f (i) (x0 ) (x − x0 )i+1 a b − a i=1 (i + 1)!

1 f (i) (x0 ) (b − x0 )i+1 b − a i=1 (i + 1)!

−(a − x0 )i+1 . n

=

(22.54)

We obtained 1 b−a

b

f (x) dx − f (x0 ) ≥ (≤) a

' i n f (i) (x0 ) i=1

(i + 1)!

( (b − x0 )i−k (a − x0 )k ,

k=0

(22.55) where n is odd. Next let 0 < a ≤ p/q ≤ b, a.e. on X. Then from (22.55) we ﬁnd p f (x)dx − qf q a p ' i i−k k ( n (i) f p p q q b− ≥ (≤) a− (i + 1)! q q i=1 k=0 ( ' n i f (i) pq 1−i i−k k = q (bq − p) (aq − p) . (i + 1)! i=1 q b−a

b

k=0

(22.56)

574

22. Canavati Fractional Approximation of Csiszar’s f -Divergence

So when n is odd we get b 1 f (x) dx − Γf (μ1 , μ2 ) b−a a i n p 1 ≥ (≤) f (i) q 1−i (bq − p)(i−k) (aq − p)k dλ . (i + 1)! q X i=1 k=0 (22.57) We have proved Theorem 22.15. Let n be odd and f ∈ C n+1 ([a, b]), such that f (n+1) ≥ 0 (≤ 0), 0 < a ≤ p/q ≤ b, a.e. on X. Then b 1 Γf (μ1 , μ2 ) ≤ (≥) f (x)dx b−a a i n p 1 (i) 1−i (i−k) k − f (aq − p) dλ . (22.58) q (bq − p) (i + 1)! q X i=1 k=0

Remark 22.16. Let n be odd, f ∈ C n+1 ([a, b]), a, b ∈ R, such that ≥ 0 (≤ 0), x, x0 ∈ [a, b], and μ be any probability measure on [a, b]; f then by (22.53) we get (n+1)

f (x)dμ − f (x0 ) ≥ (≤) [a,b]

n f (i) (x0 ) i=1

i!

(x − x0 )i dμ.

(22.59)

[a,b]

Comment 22.17. Assume all singletons of X belong to A . Let the Dirac measure δ x0 λ, x0 ∈ X; then there exists density fˆ ≥ 0 such that δ x0 (E) = fˆ dλ, for all E ∈ A, E

(here fˆ is the Radon–Nikodym derivative of δ x0 with respect to λ). Clearly 1 = δ x0 ({x0 }) = (22.60) fˆ dλ = fˆ(x0 )λ({x0 }). {x0 }

Because fˆ is real-valued we must have λ({x0 }) = 0 and fˆ(x0 ) =

1 > 0. λ({x0 })

(22.61)

Also 0 = δ x0 (X − {x0 }) =

fˆ dλ. X−{x0 }

22.2 Main Results

Therefore

fˆX−{x

0}

= 0,

a.e. on X.

So let μ2 = δ x0 ; then q = fˆ and p Γf (μ1 , δ x0 ) = fˆ f dλ = f (λ({x0 })p(x0 )), fˆ X

575

(22.62)

(22.63)

is trivial and of no interest for further study. Finally we give Comment 22.18. Let f ∈ Cxν0 ([a, b]), x0 ∈ [a, b] ⊂ R, and ν ≥ 1, the remainder of the fractional Taylor formula (see Theorem 22.3 and (22.19)) is x x0 ν 1 Jν (Dx0 f ) (x) = (x − t)ν−1 (Dxν0 f )(t) dt, (22.64) Γ(ν) x0 for all x0 ≤ x ≤ b. If (Dxν0 f )(t)

≥0 (≤ 0)

over [x0 , b] then x0 ν ≥0 Jν (Dx0 f ) (x) (≤ 0) over [x0 , b]. The last implies (by (22.19)) that f (x) ≥ (≤)

n−1

f (i) (x0 )

i=0

(x − x0 )i , i!

(22.65)

where n := [ν], for all x0 ≤ x ≤ b. According to Section 22.1(I) we take here f to be convex from (0, +∞) into R which is strictly convex at 1 with f (1) = 0. Also we take 0 < a ≤ p/q ≤ b a.e. on X, with a ≤ 1 ≤ b. Let ν ≥ 1 and additionally assume that f ∈ Caν ([a, b]) such that (Daν f )(t) ≥ 0 (≤ 0) over [a, b]; then we ﬁnd f (x) ≥ (≤)

n−1 i=0

f (i) (a)

(x − a)i , i!

(22.66)

for all a ≤ x ≤ b, n := [ν]. Then i n−1 q pq − a p (i) qf , f (a) ≥ (≤) q i! i=0

(22.67)

576

22. Canavati Fractional Approximation of Csiszar’s f -Divergence

a.e. on X. Consequently we obtain Γf (μ1 , μ2 ) ≥ (≤)

n−1 i=0

where n := [ν], ν ≥ 1.

f (i) (a) i!

q 1−i (p − qa)i dλ, X

(22.68)

23 Caputo and Riemann–Liouville Fractional Approximation of Csiszar’s f -Divergence

Here are presented various tight probabilistic inequalities that give nearly best estimates for the Csiszar’s f -divergence. These involve Riemann– Liouville and Caputo fractional derivatives of the directing function f. Also is given a lower bound for the Csiszar’s distance. The Csiszar’s discrimination is the most essential and general measure for the comparison between two probability measures. This is a continuation of Chapter 22 and is based on [55].

23.1 Preliminaries Throughout this chapter we use the following. (I) Let f be a convex function from (0, +∞) into R that is strictly convex at 1 with f (1) = 0. Let (X, A, λ) be a measure space, where λ is a ﬁnite or a σ-ﬁnite measure on (X, A). And let μ1 , μ2 be two probability measures on (X, A) such that μ1 λ, μ2 λ (absolutely continuous); for example, λ = μ1 + μ2 . Denote by p = dμ1 /dλ, q = dμ2 /dλ the (densities) Radon– Nikodym derivatives of μ1 , μ2 with respect to λ. Here we assume that 0
p ≤ b, q

a.e. on X

and a ≤ 1 ≤ b. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 577 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 23,

578

23. Fractional Approximation of Csiszar’s f-Divergence

The quantity

Γf (μ1 , μ2 ) =

q (x) f X

p (x) q (x)

dλ (x),

(23.1)

was introduced by I. Csiszar in 1967 (see [128]), and is called f -divergence of the probability measures μ1 and μ2 . By Lemma 1.1 of [128], the integral (23.1) is well deﬁned and Γf (μ1 , μ2 ) ≥ 0 with equality only when μ1 = μ2 . In [128] the author without proof mentions that Γf (μ1 , μ2 ) does not depend on the choice of λ. For a proof of the last see [28], Lemma 1.1; see here Lemma 22.1. The concept of f -divergence was introduced ﬁrst in [127] as a generalization of Kullback’s “information for discrimination” or I-divergence (generalized entropy) [240, 241] and of R´enyi’s “information gain” (I-divergence of order α) [345]. In fact the I-divergence of order 1 equals Γu log2 u (μ1 , μ2 ). 2

The choice f (u) = (u − 1) again produces a known measure of diﬀerence 2 of distributions called κ - divergence; of course the total variation distance |μ1 − μ2 | = X |p (x) − q (x)| dλ (x) equals Γ|u−1| (μ1 , μ2 ). Here by assuming f (1) = 0 we can consider Γf (μ1 , μ2 ) as a measure of the diﬀerence between the probability measures μ1 , μ2 . The f -divergence is in general asymmetric in μ1 and μ2 . But because f is convex and strictly convex at 1 (see Lemma 2, [28]) so is 1 ∗ f (u) = uf (23.2) u and as in [128] we get Γf (μ2 , μ1 ) = Γf ∗ (μ1 , μ2 ).

(23.3)

In information theory and statistics many other concrete divergences are used that are special cases of the above general Csiszar f -divergence, such as Hellinger distance DH , α-divergence Dα , Bhattacharyya distance DB , harmonic distance DHα , Jeﬀrey’s distance DJ , and triangular discrimination DΔ ; for all these, see, for example [78, 145]. The problem of ﬁnding and estimating the proper distance (or diﬀerence or discrimination) of two probability distributions is one of the major questions in probability theory. The above f -divergence measures in their various forms have also been applied to anthropology, genetics, ﬁnance, economics, political science, biology, approximation of probability distributions, signal processing, and pattern recognition. A great inspiration for this chapter has been the very important monograph on the topic by S. Dragomir [145]. (II) Here we follow [134].

23.1 Preliminaries

579

We start with Deﬁnition 23.1. Let ν ≥ 0; the operator Jaν , deﬁned on L1 (a, b) by x 1 ν−1 Jaν f (x) := (x − t) f (t) dt (23.4) Γ (ν) a for a ≤ x ≤ b, is called the Riemann–Liouville fractional integral operator of order ν. For ν = 0, we set Ja0 := I, the identity operator. Here Γ stands for the gamma function. Let α > 0, f ∈ L1 (a, b), a, b ∈ R; see [134]. Here [·] stands for the integral part of the number. We deﬁne the generalized Riemann–Liouville fractional derivative of f of order α by m s d 1 m−α−1 Daα f (s) := (s − t) f (t) dt, Γ (m − α) ds a where m := [α] + 1, s ∈ [a, b]; see also [57], Remark 46 there. In addition, we set Da0 f := f, Ja−α f := Daα f, Da−α f := Jaα f, Dan f = f (n) ,

if α > 0, if 0 < α ≤ 1, for n ∈ N.

(23.5)

We need Deﬁnition 23.2. [45] We say that f ∈ L1 (a, b) has an L∞ fractional derivative Daα f (α > 0) in [a, b] , a, b ∈ R, iﬀ Daα−k f ∈ C ([a, b]), k = 1, . . . , m := [α] + 1, and Daα−1 f ∈ AC ([a, b]) (absolutely continuous functions) and Daα f ∈ L∞ (a, b). Lemma 23.3. [45] Let β > α ≥ 0, f ∈ L1 (a, b) , a, b ∈ R, have an L∞ fractional derivative Daβ f in [a, b], and let Daβ−k f (a) = 0 for k = 1, . . . , [β] + 1. Then s 1 β−α−1 α (s − t) Daβ f (t) dt, (23.6) Da f (s) = Γ (β − α) a ∀s ∈ [a, b]. Here Daα f ∈ AC ([a, b]) for β − α ≥ 1, and Daα f ∈ C ([a, b]) for β − α ∈ (0, 1). Here AC n ([a, b]) is the space of functions with absolutely continuous (n − 1)st derivative.

580

23. Fractional Approximation of Csiszar’s f-Divergence

We need to mention Deﬁnition 23.4. [134] Let ν ≥ 0, n := ν; · is the ceiling of the number, f ∈ AC n ([a, b]). We call the Caputo fractional derivative x 1 n−ν−1 (n) ν D∗a f (x) := (x − t) f (t) dt, (23.7) Γ (n − ν) a ∀x ∈ [a, b]. ν f (x) exists almost everywhere for x ∈ [a, b]. The above function D∗a We need Proposition 23.5. [134] Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]). Then exists iﬀ the generalized Riemann–Liouville fractional derivative Daν f exists. ν f D∗a

Proposition 23.6. [134] Let ν ≥ 0, n := ν. Assume that f is such that ν f and Daν f exist. Suppose that f (k) (a) = 0 for k = 0, 1, . . . , n−1. both D∗a Then ν f = Daν f. D∗a

(23.8)

In conclusion ν Corollary 23.7. [58] Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]), D∗a f exists ν (k) or Da f exists, and f (a) = 0, k = 0, 1, . . . , n − 1. Then ν f. (23.9) Daν f = D∗a

We need Theorem 23.8. [58] Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]), and f (k) (a) = 0, k = 0, 1, . . . , n − 1. Then x 1 ν−1 ν (x − t) D∗a f (t) dt. (23.10) f (x) = Γ (ν) a We also need Theorem 23.9. [58] Let ν ≥ γ + 1, γ ≥ 0. Call n := ν. Assume ν f ∈ f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n − 1, and D∗a γ L∞ (a, b). Then D∗a f ∈ AC ([a, b]), and x 1 ν−γ−1 γ ν D∗a f (x) = (x − t) D∗a f (t) dt, (23.11) Γ (ν − γ) a ∀x ∈ [a, b].

23.2 Results

581

Theorem 23.10. [58] Let ν ≥ γ+1, γ ≥ 0, n := ν. Let f ∈ AC n ([a, b]) such that f (k) (a) = 0, k = 0, 1, . . . , n−1. Assume ∃Daν f (x) ∈ R, ∀x ∈ [a, b], and Daν f ∈ L∞ (a, b). Then Daγ f ∈ AC ([a, b]), and x 1 ν−γ−1 γ (x − t) Daν f (t) dt, (23.12) Da f (x) = Γ (ν − γ) a ∀x ∈ [a, b].

23.2 Results Here f and the whole setting are as in section 23.1, Preliminaries (I). We present ﬁrst results regarding the Riemann–Liouville fractional derivative. Theorem 23.11. Let β > 0, f ∈ L1 (a, b), have an L∞ fractional derivative Daβ f in [a, b], and let Daβ−k f (a) = 0 for k = 1, . . . , [β] + 1. Also it holds 0 < a ≤ p (x)/q (x) ≤ b, a.e. on X, a < b. Then β Da f ∞,[a,b] Γf (μ1 , μ2 ) ≤ Γ (β + 1) 1−β

q (x)

β

(p (x) − aq (x)) dλ (x).

(23.13)

X

Proof. By (23.6), α = 0, we obtain s 1 β−1 f (s) = (s − t) Daβ f (t) dt, Γ (β) a all a ≤ s ≤ b. Then s 1 β−1 β Da f (t) dt (s − t) Γ (β) a β s Da f ∞,[a,b] β−1 ≤ (s − t) dt Γ (β) a β Da f β ∞,[a,b] (s − a) = Γ (β) β

|f (s)| ≤

(23.14)

582

23. Fractional Approximation of Csiszar’s f-Divergence

β Da f ∞,[a,b]

β

(s − a) , all a ≤ s ≤ b. Γ (β + 1) That is, we have that β Da f ∞,[a,b] β (s − a) , all a ≤ s ≤ b. |f (s)| ≤ Γ (β + 1) Consequently we ﬁnd p (x) Γf (μ1 , μ2 ) = q (x) f dλ (x) q (x) X β β Da f p (x) ∞,[a,b] − a dλ (x) q (x) ≤ Γ (β + 1) q (x) X β Da f ∞,[a,b] 1−β β = q (x) (p (x) − aq (x)) dλ (x), Γ (β + 1) X proving the claim. =

(23.15)

(23.16)

(23.17)

Next we give an Lδ result. Theorem 23.12. Same assumptions as in Theorem 23.11. Let γ, δ > 1 : 1/γ + 1/δ = 1 and γ (β − 1) + 1 > 0. Then β Da f δ,[a,b] Γf (μ1 , μ2 ) ≤ 1/γ Γ (β) (γ (β − 1) + 1) 2−β−1/γ β−1+1/γ q (x) (p (x) − aq (x)) dλ (x). (23.18) X

Proof. By (23.6), α = 0, we get again s 1 β−1 f (s) = (s − t) Daβ f (t) dt, Γ (β) a

(23.19)

all a ≤ s ≤ b. Hence 1 |f (s)| ≤ Γ (β)

s

β−1

(s − t) a

β Da f (t) dt

1/γ s 1/δ s δ β 1 γ(β−1) ≤ Da f (t) dt (s − t) dt Γ (β) a a β Da f β−1+1/γ δ,[a,b] (s − a) ≤ , (23.20) 1/γ Γ (β) (γ (β − 1) + 1)

all a ≤ s ≤ b.

23.2 Results

That is,

β Da f β−1+1/γ δ,[a,b] (s − a) , |f (s)| ≤ 1/γ Γ (β) (γ (β − 1) + 1)

583

(23.21)

all a ≤ s ≤ b. Consequently we ﬁnd p dλ q f q

Γf (μ1 , μ2 ) ≤ X

β Da f δ,[a,b]

≤

q

1/γ

β−1+1/γ p −a dλ q

X Γ (β) (γ (β − 1) + 1) β Da f δ,[a,b] β−1+1/γ = q 2−β−1/γ (p − aq) dλ, 1/γ X Γ (β) (γ (β − 1) + 1)

proving the claim.

(23.22)

It follows an L1 estimate. Theorem 23.13. Same assumptions as in Theorem 23.11. Let β ≥ 1. Then β Da f 1,[a,b] Γf (μ1 , μ2 ) ≤ Γ (β) 2−β β−1 (q (x)) (p (x) − aq (x)) dλ (x) . (23.23) X

Proof. By (23.19) we have 1 Γ (β)

|f (s)| ≤

s

(s − a) Γ (β)

b

β Da f (t) dt

a β−1

=

(s − a) Γ (β)

β Da f . 1,[a,b]

That is, β−1

|f (s)| ≤

β Da f (t) dt

a β−1

≤

β−1

(s − t)

(s − a) Γ (β)

β Da f , 1,[a,b]

for all s in [a, b]. Thus Γf (μ1 , μ2 ) ≤ X

p dλ q f q

(23.24)

(23.25)

584

23. Fractional Approximation of Csiszar’s f-Divergence

≤

=

β Da f 1,[a,b]

Γ (β) β Da f

q

X

1,[a,b]

Γ (β)

β−1 p −a dλ q

q 2−β (p − aq)

β−1

dλ ,

(23.26)

X

proving the claim.

We continue with results regarding the Caputo fractional derivative. Theorem 23.14. Let ν > 0, n := ν, f ∈ AC n ([a, b]), and f (k) (a) = 0, ν f ∈ L∞ (a, b), 0 < a ≤ p (x)/q (x) ≤ b, k = 0, 1, . . . , n − 1. Assume D∗a a.e. on X, a < b. Then ν f ∞,[a,b] D∗a Γf (μ1 , μ2 ) ≤ Γ (ν + 1) 1−ν ν q (x) (p (x) − aq (x)) dλ (x). (23.27) X

Proof. Similar to Theorem 23.11, using Theorem 23.8. Next we give an Lδ result. Theorem 23.15. Assume all are as in Theorem 23.14. Let γ, δ > 1 : 1/γ + 1/δ = 1, and γ (ν − 1) + 1 > 0. Then ν f δ,[a,b] D∗a Γf (μ1 , μ2 ) ≤ 1/γ Γ (ν) (γ (ν − 1) + 1) 2−ν−1/γ ν−1+1/γ q (x) (p (x) − aq (x)) dλ (x). (23.28) X

Proof. Similar to Theorem 23.12, using Theorem 23.8. It follows an L1 estimate. Theorem 23.16. Assume all are as in Theorem 23.14. Let ν ≥ 1. Then Γf (μ1 , μ2 ) ≤

ν f 1,[a,b] D∗a

Γ (ν)

2−ν

(q (x))

(p (x) − aq (x))

ν−1

dλ (x) .

X

Proof. Similar to Theorem 23.13, using Theorem 23.8.

(23.29)

23.2 Results

585

Regarding again the Riemann–Liouville fractional derivative we need: Corollary 23.17. Let ν ≥ 0, n := ν, f ∈ AC n ([a, b]), ∃Daν f (x) ∈ R, ∀x ∈ [a, b], and f (k) (a) = 0, k = 0, 1, . . . , n − 1. Then x 1 ν−1 f (x) = (x − t) Daν f (t) dt. (23.30) Γ (ν) a Proof. By Corollary 23.7 and Theorem 23.8. We continue with results again regarding the Riemann–Liouville fractional derivative. Theorem 23.18. Let ν > 0, n := ν, f ∈ AC n ([a, b]), ∃Daν f (x) ∈ R, ∀x ∈ [a, b], and f (k) (a) = 0, k = 0, 1, . . . , n − 1. Assume Daν f ∈ L∞ (a, b), 0 < a ≤ p (x)/q (x) ≤ b, a.e. on X, a < b. Then Daν f ∞,[a,b] Γf (μ1 , μ2 ) ≤ Γ (ν + 1) 1−ν ν q (x) (p (x) − aq (x)) dλ (x). (23.31) X

Proof. Similar to Theorem 23.11, using Corollary 23.17.

Next we give the corresponding Lδ result. Theorem 23.19. Assume all are as in Theorem 23.18. Let γ, δ > 1 : 1/γ + 1/δ = 1, and γ (ν − 1) + 1 > 0. Then Daν f δ,[a,b] Γf (μ1 , μ2 ) ≤ 1/γ Γ (ν) (γ (ν − 1) + 1) 2−ν−1/γ ν−1+1/γ q (x) (p (x) − aq (x)) dλ (x) . (23.32) X

Proof. Similar to Theorem 23.12, using Corollary 23.17.

The L1 estimate follows. Theorem 23.20. Assume all are as in Theorem 23.18. Let ν ≥ 1. Then Γf (μ1 , μ2 ) ≤

Daν f 1,[a,b] Γ (ν)

586

23. Fractional Approximation of Csiszar’s f-Divergence

2−ν

(q (x))

ν−1

(p (x) − aq (x))

dλ (x) .

(23.33)

X

Proof. Similar to Theorem 23.13, using Corollary 23.17.

We need Theorem 23.21. (Taylor expansion for Caputo derivatives, [134, p. 40]) Assume ν ≥ 0, n = ν, and f ∈ AC n ([a, b]). Then f (x) =

n−1 k=0

f (k) (a) 1 k (x − a) + k! Γ (ν)

x

ν−1

(x − t) a

ν D∗a f (t) dt,

(23.34)

∀x ∈ [a, b]. We make Remark 23.22. Let ν > 0, n = ν, and f ∈ AC n ([a, b]). ν f ≥ 0 over [a, b], then If D∗a (≤0)

x

ν−1

(x − t) a

ν D∗a f (t) dt ≥ 0 on [a, b]. (≤0)

By (23.34) then we obtain f (x) ≥ (≤)

n−1 k=0

f (k) (a) k (x − a) , k!

(23.35)

∀x ∈ [a, b]. Hence qf

k n−1 f (k) (a) p p q −a , ≥ (≤) q k! q

(23.36)

k=0

a.e. on X. Consequently we derive Γf (μ1 , μ2 ) ≥ (≤)

n−1 k=0

f (k) (a) k!

k

q 1−k (p − aq) dλ. X

We have established Theorem 23.23. Let ν > 0, n = ν, and f ∈ AC n ([a, b]).

(23.37)

23.2 Results

587

ν If D∗a f ≥ 0 on [a, b], then (≤0)

Γf (μ1 , μ2 ) ≥ (≤)

n−1 k=0

(q (x))

1−k

f (k) (a) k!

k (p (x) − aq (x)) dλ (x) .

(23.38)

X

We ﬁnish with Remark 23.24. Using Lemma 23.3, Theorem 23.9, and Theorem 23.10 γ and in their settings, for g any of Daα f, D∗a f, Daγ f, that fulﬁll the conditions and assumptions of section 23.1, Preliminaries (I), we can ﬁnd similar estimates as above for Γg (μ1 , μ2 ).

24 Canavati Fractional Ostrowski-Type Inequalities

Optimal upper bounds are given to the deviation of an initial value of a function f ∈ Cxν0 ([a, b]), x0 ∈ [a, b], ν > 0 from the corresponding average of f . These bounds are of type A · Dxν0 f ∞,[x0 ,b] , where A is the smallest universal constant; that is, the produced inequalities are sharp and attained. Here Dxν0 f is the ν-order Canavati-type fractional derivative of f . This chapter was inspired by the work of Ostrowski [318], 1938, and of the author’s [13], 1995. This treatment relies on [24].

24.1 Background In the sequel we follow [101]. Let g ∈ C([0, 1]), n := [ν], ν > 0, and α := ν − n (0 < α < 1). Deﬁne x 1 (x − t)ν−1 g(t)dt, 0 ≤ x ≤ 1, (24.1) (Jν g)(x) := Γ(ν) 0 Riemann–Liouville integral, where Γ is the gamma function Γ(ν) := the ∞ −t ν−1 e t dt. We deﬁne the subspace C ν ([0, 1]) of C n ([0, 1]): 0 C ν ([0, 1]) := {g ∈ C n ([0, 1]) : J1−α g (n) ∈ C 1 ([0, 1])}. Thus letting g ∈ C ν ([0, 1]), we deﬁne the Canavati ν- fractional derivative of g as d . (24.2) Dν g := DJ1−α g (n) , D := dx G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 589 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 24,

590

Canavati Fractional Ostrowski-Type Inequalities

When ν ≥ 1 we have the Taylor formula t2 tn−1 + · · · + g (n−1) (0) 2! (n − 1)! ∀t ∈ [0, 1]. (24.3)

g(t) = g(0) + g (0)t + g (0) + (Jν Dν g)(t), When 0 < ν < 1 we ﬁnd

g(t) = (Jν Dν g)(t),

∀t ∈ [0, 1].

(24.4)

Next we carry the above notions over to an arbitrary [a, b] ⊆ R (see [17]). Let x, x0 ∈ [a, b] such that x ≥ x0 , x0 is ﬁxed. Let f ∈ C([a, b]) and deﬁne x 1 (Jνx0 f )(x) := (x − t)ν−1 f (t)dt, x0 ≤ x ≤ b, (24.5) Γ(ν) x0 the generalized Riemann–Liouville integral. We deﬁne the subspace Cxν0 ([a, b]) of C n ([a, b]): x0 Cxν0 ([a, b]) := {f ∈ C n ([a, b]) : J1−α f (n) ∈ C 1 ([x0 , b])}.

For f ∈ Cxν0 ([a, b]), we deﬁne the generalized ν -fractional derivative of f over [x0 , b] as x0 f (n) . (24.6) Dxν0 f := DJ1−α We observe that Dxn0 f = f (n) , n ∈ N. Notice that x 1 x0 (J1−α f (n) )(x) = (x − t)−α f (n) (t)dt Γ(1 − α) x0 exists for f ∈ Cxν0 ([a, b]). We recall the following fractional generalization of Taylor’s formula (see [17, 101]). Theorem 24.1. Let f ∈ Cxν0 ([a, b]), x0 ∈ [a, b] ﬁxed. (i) If ν ≥ 1 then f (x) = f (x0 ) + f (x0 )(x − x0 ) + f (x0 )

(x − x0 )2 + 2

(x − x0 )n−1 (n − 1)! x0 ν +(Jν Dx0 f )(x), all x ∈ [a, b] : x ≥ x0 . · · · + f (n−1) (x0 )

(24.7)

(ii) If 0 < ν < 1 we get f (x) = (Jνx0 Dxν0 f )(x), Here we use (24.7).

all x ∈ [a, b] : x ≥ x0 .

(24.8)

24.2 Results

591

24.2 Results We present the ﬁrst fractional Ostrowski-type inequality. Theorem 24.2. Let 1 ≤ ν < 2 and f ∈ Cxν0 ([a, b]), a ≤ x0 < b, x0 ﬁxed. Then 1 b Dν f ∞,[x0 ,b] x0 (b − x0 )ν . f (y)dy − f (x0 ) ≤ (24.9) b − x0 x0 Γ(ν + 2) Proof. Here n := [ν] = 1. From (24.7) we have y 1 (y − w)ν−1 (Dxν0 f )(w)dw, f (y) − f (x0 ) = Γ(ν) x0 Thus

∀y ≥ x0 .

1 b 1 b f (y)dy − f (x0 ) = (f (y) − f (x0 ))dy b − x0 x0 b − x0 x0 b 1 |f (y) − f (x0 )|dy b − x0 x0 y b 1 1 ≤ (y − w)ν−1 |Dxν0 f (w)|dw dy b − x0 x0 Γ(ν) x0 y b 1 1 ν ν−1 D f ∞,[x0 ,b] ≤ (y − w) dw dy b − x0 x0 Γ(ν) x0 x0 b Dxν0 f ∞,[x0 ,b] (y − x0 )ν · dy = (b − x0 )Γ(ν) ν x0 ≤

(by Γ(ν + 1) = νΓ(ν), ν > 0) Dxν0 f ∞,[x0 ,b] (b − x0 )ν+1 · (b − x0 )Γ(ν + 1) ν+1

=

= We have proved (24.9).

Dxν0 f ∞,[x0 ,b] · (b − x0 )ν . Γ(ν + 2)

Next we give another more general fractional Ostrowski-type inequality. Theorem 24.3. Let a ≤ x0 < b ﬁxed. Let f ∈ Cxν0 ([a, b]), ν ≥ 2, n := [ν]. Assume f (i) (x0 ) = 0, i = 1, . . . , n − 1. Then 1 b Dν f ∞,[x0 ,b] x0 · (b − x0 )ν . f (y)dy − f (x0 ) ≤ (24.10) b − x0 x0 Γ(ν + 2)

592

Canavati Fractional Ostrowski-Type Inequalities

Proof. Again from (24.7) we have 1 f (y) − f (x0 ) = Γ(ν)

y

x0

(y − w)ν−1 (Dxν0 f )(w)dw,

∀y ≥ x0 .

We observe that |f (y) − f (x0 )|

≤ =

1 Dxν0 f ∞,[x0 ,b] Γ(ν)

y

(y − w)ν−1 dw

x0

Dxν0 f ∞,[x0 ,b] (y − x0 )ν 1 Dxν0 f ∞,[x0 ,b] = · (y − x0 )ν . Γ(ν) ν Γ(ν + 1)

That is, |f (y) − f (x0 )| ≤

Dxν0 f ∞,[x0 ,b] · (y − x0 )ν , Γ(ν + 1)

∀y ≥ x0 .

(24.11)

Therefore we get b 1 f (y)dy − f (x0 ) b − x0 x0

= ≤ (24.11)

≤

= =

That proves (24.10).

b 1 (f (y) − f (x0 ))dy b − x0 x0 b 1 |f (y) − f (x0 )|dy b − x0 x0 b Dxν 0 f ∞,[x0 ,b] 1 · (y − x0 )ν dy b − x0 x0 Γ(ν + 1) Dxν 0 f ∞,[x0 ,b] (b − x0 )ν+1 1 · · b − x0 Γ(ν + 1) ν+1 Dxν 0 f ∞,[x0 ,b] · (b − x0 )ν . Γ(ν + 2)

Remark 24.4. Let μ, ν > 0 such that ν ≤ μ, and n := [ν], α := ν − n. Consider (x − x0 )μ , x0 ≤ x ≤ b. (24.12) φμ (x) := Γ(μ + 1) Then φμ(n) (x) =

μ(μ − 1)(μ − 2) · · · (μ − n + 2)(μ − n + 1)(x − x0 )μ−n , Γ(μ + 1)

and φμ(n) (x) =

(x − x0 )μ−n = φμ−n (x). Γ(μ − n + 1)

24.2 Results

593

Next we ﬁnd that x0 φμ−n )(x) (J1−α

= =

1 Γ(1 − α) 1 Γ(1 − α)

x

(x − t)−α

x0 x

(t − x0 )μ−n dt Γ(μ − n + 1)

(x − t)(1−α)−1 x0

(t − x0 )(μ−n+1)−1 dt Γ(μ − n + 1)

(by [404, p. 256]; notice that 1 − α, μ − n + 1 > 0) 1 · Γ(1 − α)Γ(μ − n + 1) Γ(1 − α)Γ(μ − n + 1) (x − x0 )(1−α+μ−n) Γ(1 − α + μ − n + 1)

=

(x − x0 )(μ−ν+1) . Γ(μ − ν + 2)

= That is,

x0 (J1−α φμ−n )(x) =

(x − x0 )(μ−ν+1) . Γ(μ − ν + 2)

(24.13)

Hence (Dxν0 φμ )(x)

x0 x0 (DJ1−α φ(n) )(x) = (DJ1−α φμ−n )(x) μ−ν+1 (x − x0 ) (x − x0 )μ−ν , = = Γ(μ − ν + 2) Γ(μ − ν + 1)

=

so that (Dxν0 φμ )(x) =

(x − x0 )μ−ν = φμ−ν (x), Γ(μ − ν + 1)

(24.14)

and Γ(μ + 1) (x − x0 )μ−ν , Γ(μ − ν + 1)

Dxν0 (x − x0 )μ =

0 < ν ≤ μ, x0 ≤ x ≤ b. (24.15)

In particular we get Dxν0 (x − x0 )ν = Γ(ν + 1),

(by Γ(1) = 1).

Consequently it holds Dxν0 (x − x0 )ν ∞,[x0 ,b] = Γ(ν + 1).

(24.16)

Proposition 24.5. Inequality (24.9) is sharp; namely it is attained by f (x) := (x − x0 )ν ,

1 ≤ ν < 2, x ∈ [a, b].

594

Canavati Fractional Ostrowski-Type Inequalities

Proof. Observe that R.H.S.(24.9) =

1 b − x0

b

(y − x0 )ν dy = x0

(b − x0 )ν . ν+1

Also we see L.H.S.(24.9)

(24.16)

=

Γ(ν + 1) Γ(ν + 1) (b − x0 )ν (b − x0 )ν = (b − x0 )ν = . Γ(ν + 2) (ν + 1)Γ(ν + 1) ν+1

That is, both sides of (24.9) are equal. Proposition 24.6. Inequality (24.10) is sharp; namely it is attained by f (x) := (x − x0 )ν ,

ν ≥ 2, x ∈ [a, b].

Proof. Observe that f

(i)

(x0 ) = ((x − x0 ) ) ν (i)

= 0,

i = 1, . . . , n − 1.

x=x0

Again we have R.H.S.(24.10) =

(b − x0 )ν , ν+1

L.H.S.(24.10) =

(b − x0 )ν . ν+1

and

That is, both sides of (24.10) are equal. Comment 24.7. In the fractional Ostrowski-type inequalities, under the same initial conditions—assumption, as in the integer ordinary derivative case—one can derive results for higher-order (fractional) derivatives, appearing in the R.H.S.s of the corresponding inequalities. Thus, compare Theorem 24.2, 1 ≤ ν < 2, with an ordinary case of order n = 1 ; see [13]. Also compare Theorem 24.3, ν ≥ 2, where n + 1 ≤ ν < n + 2 , n + 1 = [ν], with an ordinary case of order n + 1; see [13].

25 Multivariate Canavati Fractional Ostrowski-Type Inequalities

Optimal upper bounds are given to the deviation of a value of a multivariate function of a fractional space from its average, over convex and compact subsets of RN , N ≥ 2. In particular we work over rectangles, balls, and spherical shells. These bounds involve the supremum and L∞ norms of related multivariate Canavati fractional derivatives of the involved function. The presented inequalities are sharp; namely they are attained. This chapter has been motivated by the works of Ostrowski [318], 1938, and Anasstasiou [24], 2003, and the chapter is based on [43].

25.1 Background In the sequel we follow Canavati [101]. Let g ∈ C([0, 1]), and α := ν − n (0 < α < 1). Deﬁne 1 (Jν g)(x) := Γ(ν)

n := [ν],

ν > 0,

x

(x − t)ν−1 g(t)dt,

0 ≤ x ≤ 1,

(25.1)

0

the Riemann ∞ – Liouville fractional integral, where Γ is the gamma function Γ(ν) := 0 e−t tν−1 dt. We deﬁne the subspace C ν ([0, 1]) of C n ([0, 1]): C ν ([0, 1]) := {g ∈ C n ([0, 1]) : J1−α g (n) ∈ C 1 ([0, 1])}.

(25.2)

G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 595 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 25,

596

25. Multivariate Canavati Fractional Inequalities

So let g ∈ C ν ([0, 1]); we deﬁne the Canavati ν -fractional derivative of g as (25.3) g (ν) := (J1−α g (n) ) . When ν ≥ 1 we have the fractional Taylor formula ([101]) g(t) = g(0) + g (0)t + g (0)

t2 tn−1 + · · · + g (n−1) (0) + (Jν g (ν) )(t), ∀ t ∈ [0, 1], 2! (n − 1)! (25.4)

and when 0 < ν < 1 we ﬁnd g(t) = (Jν g (ν) )(t), ∀ t ∈ [0, 1].

(25.5)

Next we carry the above notions over to an arbitrary interval [a, b] ⊆ R (see Anastassiou [17]). Let x, x0 ∈ [a, b] such that x ≥ x0 , x0 is ﬁxed. Let f ∈ C([a, b]) and deﬁne x 1 x0 (Jν f )(x) := (x − t)ν−1 f (t)dt, x0 ≤ x ≤ b, (25.6) Γ(ν) x0 the generalized Riemann – Liouville integral. Cxν0 ([a, b]) of C n ([a, b]) :

We deﬁne the subspace

x0 Cxν0 ([a, b]) := {f ∈ C n ([a, b]) : J1−α f (n) ∈ C 1 ([x0 , b])}.

(25.7)

For f ∈ Cxν0 ([a, b]), we deﬁne the generalized ν-fractional derivative of f over [x0 , b], as x0 Dxν0 f := J1−α f (n) . (25.8) We observe that Dxn0 f = f (n) , n ∈ N . Notice that x 1 x0 J1−α f (n) (x) = (x − t)−α f (n) (t)dt Γ(1 − α) x0

(25.9)

exists for f ∈ Cxν0 ([a, b]). We mention the following generalization of the fractional Taylor formula (see Anastassiou [17] and Canavati [101]). Theorem 25.1. Let f ∈ Cxν0 ([a, b]), (i) If ν ≥ 1, then

x0 ∈ [a, b] ﬁxed.

f (x) = f (x0 ) + f (x0 )(x − x0 ) + f (x0 ) f (n−1) (x0 )

(x − x0 )n−1 + (Jνx0 Dxν0 f )(x), (n − 1)!

(x − x0 )2 + ···+ 2!

all

x ∈ [a, b] : x ≥ x0 . (25.10)

25.1 Background

597

(ii) If 0 < ν < 1, we get f (x) = (Jνx0 Dxν0 f )(x),

all

x ∈ [a, b] : x ≥ x0 .

(25.11)

We also mention from Anastassiou [40], the basic multivariate fractional Taylor formula. Theorem 25.2. Let f ∈ C 1 (Q), where Q is convex and compact ⊆ R , N ≥ 2. For ﬁxed x0 , z ∈ Q, assume that as a function of t ∈ [0, 1] : fxi (x0 + t(z − x0 )) ∈ C ν−1 ([0, 1]), all i = 1, . . . , N , where ν ∈ [1, 2). Then N

f (z) = f (x0 ) +

N (zi − x0i ) i=1

Γ(ν)

1

(1 − t)ν−1 (fxi (x0 + t(z − x0 )))(ν−1) dt,

0

(25.12) where z = (z1 , . . . , zN ),

x = (x01 , . . . , x0N ).

The following general multivariate fractional Taylor formula also comes from Anastassiou [40]. Theorem 25.3. Let f ∈ C n (Q), Q compact and convex ⊆ RN , N ≥ 2; here ν ≥ 1 such that n = [ν]. For ﬁxed x0 , z ∈ Q assume that as functions of t ∈ [0, 1] : fα (x0 + t(z − x0 )) ∈ C (ν−n) ([0, 1]), for all "N α := (α1 , . . . , αN ), αi ∈ Z+ , i = 1, . . . , N ; |α| := i=1 αi = n. Then (i) N ∂f f (z) = f (x0 ) + (zi − x0i ) (x0 )+ ∂x i i=1 ) ) 2 * n−1 * "N "N ∂ ∂ f (x0 ) f (x0 ) i=1 (zi − x0i ) ∂xi i=1 (zi − x0i ) ∂xi +· · ·+ + 2! (n − 1)! ⎫ ⎧' n ((ν−n) 1 N ⎬ ⎨ 1 ∂ (1 − t)ν−1 (zi − x0i ) f (x0 + t(z − x0 )) dt. ⎭ ⎩ Γ(ν) 0 ∂xi i=1

(25.13) (ii) If all fα (x0 ) = 0, α := (α1 , . . . , αN ), αi ∈ Z+ , i = 1, . . . , N, |α| := "N l = 1, . . . , n − 1, then i=1 αi = l, f (z) − f (x0 ) = 1 Γ(ν)

1

(1 − t)ν−1 0

⎧' N ⎨ ⎩

∂ (zi − x0i ) ∂xi i=1

n ((ν−n) f

⎫ ⎬ (x0 + t(z − x0 )) dt. ⎭ (25.14)

598

25. Multivariate Canavati Fractional Inequalities

In Anastassiou [24] we proved the following Ostrowski-type results. Theorem 25.4. Let 1 ≤ ν < 2 and f ∈ Cxν0 ([α, b]), α ≤ x0 < b, x0 ﬁxed. Then 1 b Dν f ∞,[x0 ,b] x0 (b − x0 )ν . f (y)dy − f (x0 ) ≤ (25.15) b − x0 x0 Γ(ν + 2) Inequality (25.15) is sharp; namely it is attained by f (x) := (x − x0 )ν , 1 ≤ ν < 2, x ∈ [a, b]. Also in [24] we gave Theorem 25.5. Let α ≤ x0 < b be ﬁxed. Let f ∈ Cxν0 ([a, b]), ν ≥ 2, n := [ν]. Assume f (i) (x0 ) = 0, i = 1, . . . , n − 1. Then 1 b Dν f ∞,[x0 ,b] x0 (b − x0 )ν . f (y)dy − f (x0 ) ≤ (25.16) b − x0 x0 Γ(ν + 2) Inequality (25.16) is sharp; namely it is attained by f (x) := (x − x0 )ν ,

ν ≥ 2,

x ∈ [a, b].

Establishing sharpness in (25.15) and (25.16), we proved ﬁrst that [24] Dxν0 (x − x0 )ν ∞,[x0 ,b] = Γ(ν + 1).

(25.17)

In this chapter, motivated by (25.15) and (25.16), we present various multivariate fractional Ostrowski-type inequalities.

25.2 Results We present the ﬁrst main result of the chapter. Theorem 25.6. Let f ∈ C 1 (Q), where Q is convex and compact ⊆ RN , N ≥ 2. For ﬁxed x0 ∈ Q and any z ∈ Q assume that as a function of t ∈ [0, 1] : fxi (x0 + t(z − x0 )) ∈ C ν−1 ([0, 1]) , all i = 1, . . . , N , where ν ∈ [1, 2). Then f (z)dz Q ≤ f (x0 ) − V ol(Q) max (fxi (x0 + t(z − x0 )))(ν−1) ∞,(t,z)∈[0,1]×Q

1≤i≤N

Γ(ν + 1)V ol(Q)

Q

z − x0 l1 dz. (25.18)

25.2 Results

599

Proof. From (25.12) we obtain N (zi − x0i ) 1 (1 − t)ν−1 (fxi (x0 + t(z − x0 )))(ν−1) dt, f (z) − f (x0 ) = Γ(ν) 0 i=1 (25.19) and N |zi − x0i | 1 (1 − t)ν−1 |(fxi (x0 + t(z − x0 )))(ν−1) |dt ≤ |f (z) − f (x0 )| ≤ Γ(ν) 0 i=1 1 |xi − x0i | (fxi (x0 + t(z − x0 )))(ν−1) ∞,t∈[0,1] . Γ(ν + 1) i=1 N

(25.20)

That is, |f (z) − f (x0 )| ≤

1 z − x0 l1 max (fxi (x0 + t(z − x0 )(ν−1) ∞,(t,z)∈[0,1]×Q , 1≤i≤N Γ(ν + 1) (25.21)

∀ z ∈ Q, x0 ∈ Q ﬁxed. Hence we have f (z)dz (f (z) − f (x0 ))dz 1 Q Q − f (x0 ) = |f (z)−f (x0 )|dz ≤ V ol(Q) Q V ol(Q) V ol(Q) max (25.21) 1≤i≤N ≤

(fxi (x0 + t(z − x0 )))(ν−1) ∞,(t,z)∈[0,1]×Q Γ(ν + 1) V ol(Q)

z − x0 l1 dz,

Q

(25.22) proving the claim.

Next we give Theorem 25.7. Let f ∈ C n (Q), Q compact and convex ⊆ RN , N ≥ 2; here ν ≥ 1 such that n = [ν]. For ﬁxed x0 ∈ Q and any z ∈ Q assume that as functions of t ∈ [0, 1] : fα (x0 + t(z − x0 )) ∈ C ν−n ([0, 1]) , for all "N α : (α1 , ...αN ), αi ∈ Z+ , i = 1, . . . , N ; |α| := i=1 αi = n . Assume fα (x0 ) = 0, all α := (α1 , . . . , αN ), αi ∈ Z+ , i = 1, . . . , N, |α| = l, l = 1, . . . , n − 1. Call Dν−n f (x0 + t(z − x0 ))∞,(t,z)∈[0,1]×Q = max fα(ν−n) (x0 + t(z − x0 ))∞,(t,z)∈[0,1]×Q .

|α|=n

Then

(25.23)

Dν−n f (x0 + t(z − x0 ))∞,(t,z)∈[0,1]×Q Q f (z)dz z−x0 n ≤ f (x0 ) − l1 dz. V ol(Q) Γ(ν + 1) V ol(Q) Q (25.24)

600

25. Multivariate Canavati Fractional Inequalities

Proof. From (25.14) we have |f (z) − f (x0 )| ≤ 1 Γ(ν)

0

1

⎫ ⎧' n ((ν−n) ⎨ N ⎬ ∂ ν−1 (1 − t) (zi − x0i ) f (x0 + t(z − x0 )) dt ⎩ ⎭ ∂x i i=1

⎫ ⎧ ⎡ ⎪ N n ⎤(ν−n) ⎪ ⎨ ⎬ 1 ∂ ⎣ ≤ (zi − x0i ) f⎦ (x0 + t(z − x0 )) ⎪ Γ(ν + 1) ⎪ ∂xi ⎩ ⎭ i=1

≤ ∞,t∈[0,1]

1 (z − x0 l1 )n Dν−n f (x0 + t(z − x0 ))∞,(t,z)∈[0,1]×Q . Γ(ν − 1)

(25.25)

That is, we ﬁnd (z − x0 l1 )n Dν−n f (x0 + t(z − x0 ))∞,(t,z)∈[0,1]×Q , Γ(ν + 1) (25.26) ∀ z ∈ Q, x0 ∈ Q ﬁxed. Therefore as before in (25.22) we have that f (z)dz 1 Q ≤ − f (x ) |f (z) − f (x0 )|dz 0 V ol(Q) V ol(Q) Q |f (z) − f (x0 )| ≤

(25.26)

≤

Dν−n f (x0 + t(z − x0 ))∞,(t,z)∈[0,1]×Q Γ(ν + 1) V ol(Q)

(z − x0 l1 )n dz, (25.27) Q

proving the claim. We continue with Theorem 25.8. Let Q := [x0 , b] × [c, d], x0 ∈ [a, b), and f ∈ C([a, b] × [c, d]). Let 1 ≤ ν < 2 and ∂xν0 f /∂xν ∈ Cxν0 ([a, b]), y0 ∈ [a, b]. Then 1 f (x, y)dx dy − f (x0 , y0 ) ≤ (b − x0 )(d − c) Q ν d 1 (b − x0 )ν ∂x0 f |f (x0 , y) − f (x0 , y0 )|dy + . (25.28) ν d−c c Γ(ν + 2) ∂x ∞,Q Proof. By (25.10) we have 1 f (x, y) − f (x0 , y) = Γ(ν)

x

(x − t)ν−1 x0

∂xν0 f (t, y)dt, ∂xν

(25.29)

25.2 Results

601

x ≥ x0 , all y ∈ [c, d]. That is, |f (x, y) − f (x0 , y)| ≤

ν x 1 ∂x0 f (x − t)ν−1 dt, Γ(ν) ∂xν ∞,Q x0

ν (x − x0 )ν ∂x0 f |f (x, y) − f (x0 , y)| ≤ , Γ(ν + 1) ∂xν ∞,Q

and

(25.30)

(25.31)

all x ≥ x0 , all y ∈ [c, d]. However, it holds |f (x, y) − f (x0 , y0 )| ≤ |f (x, y) − f (x0 , y)| + |f (x0 , y) − f (x0 , y0 )| ≤ ν ∂x0 f 1 |f (x0 , y) − f (x0 , y0 )| + (x − x0 )ν , (25.32) Γ(ν + 1) ∂xν ∞,Q ∀ x ≥ x0 , ∀ y ∈ [c, d]. Consequently we derive 1 f (x, y)dx dy − f (x0 , y0 ) = (b − x0 )(d − c) [x0 ,b]×[c,d] 1 (f (x, y) − f (x0 , y0 ))dx dy ≤ (b − x0 )(d − c) [x0 ,b]×[c,d] 1 |f (x, y) − f (x0 , y0 )|dx dy (25.33) (b − x0 )(d − c) [x0 ,b]×[c,d] ' d (25.32) 1 (b − x0 ) ≤ |f (x0 , y) − f (x0 , y0 )|dy+ (b − x0 )(d − c) c ( ν b (d − c) ν ∂x0 f (x − x0 ) dx Γ(ν + 1) ∂xν ∞,Q x0 ' d 1 = (b − x0 ) |f (x0 , y) − f (x0 , y0 )|dy+ (b − x0 )(d − c) c ( ν ∂ f (b − x0 )ν+1 (d − c) x 0 (25.34) ∂xν Γ(ν + 2) ∞,Q

1 = d−c

c

d

ν (b − x0 )ν ∂x0 f |f (x0 , y) − f (x0 , y0 )|dy + , Γ(ν + 2) ∂xν ∞,Q

proving the claim.

(25.35)

602

25. Multivariate Canavati Fractional Inequalities

We further have Theorem 25.9. Let Q := [x0 , b] × [c, d], x0 ∈ [a, b), and f ∈ C n ([a, b] × [c, d]). Let ν ≥ 2 such that n = [ν] and ∂xν0 f /∂xν ∈ Cxν0 ([a, b]), y0 ∈ [a, b]. We further assume that ∂ j f (x0 , y)/∂xj = 0, j = 1, . . . , n − 1. Then 1 f (x, y)dx dy − f (x , y ) 0 0 ≤ (b − x0 )(d − c) Q ν d 1 (b − x0 )ν ∂x0 f |f (x0 , y) − f (x0 , y0 )|dy + . (25.36) ν d−c c Γ(ν + 2) ∂x ∞,Q Proof. By (25.10) we get again x ∂ν f 1 f (x, y) − f (x0 , y) = (x − t)ν−1 x0ν (t, y)dt, Γ(ν) x0 ∂x

(25.37)

x ≥ x0 , ∀ y ∈ [c, d]. And again ν (x − x0 )ν ∂x0 f |f (x, y) − f (x0 , y)| ≤ , Γ(ν + 1) ∂xν ∞,Q

(25.38)

∀ x ≥ x0 , ∀ y ∈ [c, d]. Also, it holds again

|f (x, y) − f (x0 , y0 )| ≤ |f (x0 , y) − f (x0 , y0 )| +

ν (x − x0 )ν ∂x0 f Γ(ν + 1) ∂xν

,

∞,Q

(25.39) ∀ x ≥ x0 , ∀ y ∈ [c, d]. Integrating (25.39) over Q we prove (25.36). Similar to (25.28) and (25.36) one can prove inequalities in more than two variables. Next we study fractional Ostrowski-type inequalities over balls and spherical shells. For that we make Remark 25.10. We deﬁne the ball B(0, R) := {x ∈ RN : |x| < R} ⊆ R , N ≥ 2, R > 0, and the sphere S N −1 := {x ∈ RN : |x| = 1}, where | · | is the Euclidean norm. Let dω be the element of surface measure on S N −1 and let ω N = S N −1 dω = 2π N/2 /Γ(N/2). For x ∈ RN − {0} we can write uniquely x = rω , where r = |x| > 0 and ω = x/r ∈ S N −1 , |ω| = 1 . Note that B(0,R) dy = N

25.2 Results

603

ω N RN /N is the Lebesgue measure of the ball. For F ∈ C(B(0, R)) we R have B(0,R) F (x)dx = S N −1 0 F (rω)rN −1 dr dω; we use this formula frequently. The function f : B(0, R) → R is radial if there exists a function g such that f (x) = g(r), where r = |x|, r ∈ [0, R], ∀ x ∈ B(0, R). Here we suppose that g ∈ C0ν ([0, R]), 1 ≤ ν < 2. By (25.10) we have s 1 (s − w)ν−1 (D0ν g)(w)dw, (25.40) g(s) − g(0) = Γ(ν) 0 ∀ s ∈ [0, R]. Thus |g(s) − g(0)| ≤

sν Dν g∞,[0,R] , ∀s ∈ [0, R]. Γ(ν + 1) 0

(25.41)

Next we observe that R f (y)dy ( g(s)sN −1 ds)dω B(0,R) S N −1 0 f (0) − = (25.42) = g(0) − R V ol(B(0, R)) sN −1 ds)dω N −1 ( S

0

R R N N g(s)sN −1 ds = N sN −1 (g(0) − g(s))ds ≤ (25.43) g(0) − N R R 0 0 R R (25.41) D ν g N ∞ N 0 N −1 s |g(s) − g(0)|ds ≤ sν+N −1 ds = RN 0 Γ(ν + 1) RN 0 D0ν g∞ N Rν . Γ(ν + 1)(ν + N )

(25.44)

That is, we have proved that R D0ν g∞ N Rν N B(0,R) f (y)dy N −1 g(s)s ds ≤ . = g(0) − N f (0) − V ol(B(0, R)) Γ(ν + 1)(ν + N ) R 0 (25.45)

The last inequality (25.45) is sharp; namely it is attained by g(r) = rν , 1 ≤ ν < 2, r ∈ [0, R]. Indeed by (25.17) we get D0ν xν ∞,[0,R] = Γ(ν + 1). Notice also that R N Rν N = R.H.S.(25.45), (25.46) sν+N −1 ds = L.H.S.(25.45) = N R ν+N 0 proving optimality. We have established

604

25. Multivariate Canavati Fractional Inequalities

Theorem 25.11. Let f : B(0, R) → R that is radial; that is, there exists g such that f (x) = g(r), r = |x|, ∀ x ∈ B(0, R). Assume that g ∈ C0ν ([0, R]), 1 ≤ ν < 2. Then R f (y)dy N D0ν g∞ N Rν B(0,R) N −1 . g(s)s ds ≤ f (0) − = g(0) − N Γ(ν + 1)(ν + N ) V ol(B(0, R)) R 0 (25.47) Inequality (25.47) is sharp; that is, it is attained by g(r) = rν . We continue the previous remark. Remark 25.12. We treat here the general, not necessarily radial, case of f ∈ C(B(0, R)). For any ﬁxed ω ∈ S N −1 the function f (·ω) is radial on [0,R]. We suppose that f (·ω) ∈ C0ν ([0, R]), 1 ≤ ν < 2. That is, ∃∂0ν f (rω)/∂rν and is continuous in r ∈ [0, R], for any ω ∈ S N −1 . Here we have ∂f ∂ ∂0ν f (rω) (·ω) (r) = = J2−ν ∂rν ∂r ∂r r 1 ∂ ∂f (tω)dt . (25.48) (r − t)1−ν Γ(2 − ν) ∂r ∂r 0 For x = 0 (i.e. x = rω, r > 0, ω ∈ S N −1 ), the fractional radial derivative ∂0ν f (x)/∂rν is deﬁned as in (25.48). Clearly ∂0ν f (x) ∂rν x=0 is not deﬁned. We mention Lemma 25.13. All are as in Remark 25.12. The function ∂0ν f (x)/∂rν is measurable over B(0, R) − {0}. Proof. For each n ∈ N deﬁne ) * f r − n1 ω − f (rω) 1 gn (r, ω) := n f r− ω − f (rω) = 1 n n and note that each gn is jointly measurable in (r, ω) because it is jointly continuous in (r, ω) by f ∈ C(B(0, R)); here r ∈ (0, R] and ω ∈ S N −1 . In view of gn (r, ω) → ∂f (rω)/∂r as n → ∞, we get that ∂f (rω)/∂r is jointly measurable in (r, ω) ∈ (0, R] × sN −1 = B(0, R) − {0}. Then ∂f /∂r(r·) is measurable in ω ∈ S N −1 , ∀ r ∈ (0, R]. Thus the r−ε integral Iε (r, ω) = 0 (r − t)1−ν ∂f /∂r(tω)dt, r ∈ (0, R], ω ∈ S N −1 , > 0 small; because it is a limit of Riemann sums, it is measurable in ω ∈

25.2 Results

605

S N −1 . Because (r − t)1−ν ∂f /∂r(tω) is integrable over [0,r], we get that Iε (r, ω) is continuous in r − ε, ∀ ε > 0 small. Thus r ∂f (tω)dt, (r − t)1−ν lim Iε (r, ω) = I(r, ω) := ε→0 ∂r 0 proving I(r, ω) measurable in ω ∈ sN −1 , ∀ r ∈ (0, R]. But, by the assumption f (·ω) ∈ C0ν ([0, R]), we have that I(r, ω) is continuous in r ∈ [0, R], ∀ ω ∈ sN −1 . Therefore by the Carath´eodory theorem (see [10, p. 156]) we get that I(r, ω) is jointly measurable in (r, ω) ∈ (0, R] × S N −1 , as being a Carath´eodory function. Because * ) ∂I(r, ω) 1 = lim n I(r − , ω) − I(r, ω) n→∞ ∂r n and also I(r−1/n, ω) is jointly measurable in (r, ω) ∈ (0, R]×S n−1 , we get that ∂I(r, ω)/∂r is jointly measurable in (r, ω) ∈ B(0, R) − {0}, proving the claim. We need Lemma 25.14. All are as in Remark 25.12. Additionally assume that ∂0ν f (x)/∂rν is continuous on B(0, R) − {0}, and ν ν ∂0 f (x) ∂0 f (x) K := = ess sup < ∞. ∂rν ∂rν L∞ (B(0,R)) B(0,R) Then

ν ∂0 f (rω) ∂rν

∞,(r∈[0,R])

≤ K,

∀ ω ∈ sN −1 .

(25.49)

Proof. In the radial case (25.49) is obvious. Also it is obvious if ν ν ∂0 f (rω) ∂0 f (r0 ω) , = ∂rν ∂rν ∞,[0,R] for some r0 ∈ (0, R] . The only diﬃculty here is if for speciﬁc ω 0 ∈ S N −1 we have that ν ν ∂0 f (0) ∂0 f (rω 0 ) . = ∂rν ∂rν ∞,[0,R]

∗

Then it is evident, for very small r > 0, that by continuity of ∂0ν f (· ω 0 )/∂rν we have ν ∂0 f (r∗ ω 0 ) ∂0ν f (0) ≈ ∂rν . ∂rν If

ν ν ∂0 f (0) > ess sup ∂0 f (x) , ∂rν ∂rν B(0,R)

606

25. Multivariate Canavati Fractional Inequalities

then

ν ν ∂0 f (r∗ ω 0 ) > ess sup ∂0 f (x) ν ν ∂r ∂r

a contradiction.

B(0,R)

ν ∂0 f (x) = , ∂rν ∞,B(0,R)−{0}

Remark 25.15. (continuation) By (25.47) we obtain ν ∂0 f (rω) R N Rν ν ∂r K N Rν ∞,(r∈[0,R]) N −1 f (0) − N ≤ f (sω)s ds ≤ . RN 0 Γ(ν + 1)(ν + N ) Γ(ν + 1)(ν + N )

Consequently we ﬁnd R N K N Rν N −1 . f (sω)s ds dω f (0) − ≤ N Γ(ν + 1)(ν + N ) ωN R 0 S N −1 That proves f (y)dy K N Rν B(0,R) . f (0) − ≤ V ol(B(0, R)) Γ(ν + 1)(ν + N )

(25.50)

We have established Theorem 25.16. Let f ∈ C(B(0, R)) that is not necessarily radial, and assume that f (· ω) ∈ C0ν ([0, R]), 1 ≤ ν < 2, for any ω ∈ S N −1 . Suppose also ∂0ν f (x)/∂rν is continuous on B(0, R) − {0}, and that ν ∂0 f (x) < ∞. ∂rν L∞ (B(0,R)) Then ν f (y)dy ∂0 f (x) N Rν B(0,R) . (25.51) ≤ f (0) − V ol(B(0, R)) Γ(ν + 1)(ν + N ) ∂rν L∞ (B(0,R)) We make Remark 25.17. Let the spherical shell A := B(0, R2 ) − B(0, R1 ), 0 < ¯ Consider f ∈ C 1 (A) ¯ and assume that R1 < R2 , A ⊆ RN , N ≥ 2, x ∈ A. ν ν ¯ there exists ∂R1 f (x)/∂r ∈ C(A), 1 ≤ ν < 2; x = rω, r ∈ [R1 , R2 ], ω ∈ r ν ν 1−ν f (x)/∂r = 1/Γ(2 − ν) ∂/∂r (r − t) ∂f /∂r(tω)dt . S N −1 ; where ∂R R1 1

25.2 Results

607

ν Clearly here f (rω) ∈ C 1 ([R1 , R2 ]) and ∂R f (rω)/∂rν ∈ C([R1 , R2 ]), ∀ ω ∈ 1 R2 ¯ it holds F (x)dx = N −1 F (rω)rN −1 dr dω; S N −1 . For F ∈ C(A) A

S

R1

we often exploit this formula here. Initially we assume that f is radial ; that is, there exists g such that f (x) = g(r). Here V ol(A) = ω N (R2N − R1N )/N . Then we get via the polar method that R2 N N −1 f (R1 ω) − A f (y)dy = g(R1 ) − g(s)s ds V ol(A) R2N − R1N R1 R2 N N −1 = (g(R ) − g(s)) s ds ≤ 1 R2N − R1N R1 R2 N |g(R1 ) − g(s)|sN −1 ds =: (). R2N − R1N R1

Here by (25.10) we get for s ≥ R1 , s 1 ν g(s) − g(R1 ) = (s − w)ν−1 (DR g)(w)dw. 1 Γ(ν) R1 Thus |g(s) − g(R1 )| ≤

ν DR g∞,[R1 ,R2 ] 1 (s − R1 )ν , Γ(ν + 1)

(25.52)

(25.53)

(25.54)

∀ s ≥ R1 . Consequently it holds R2 ν g∞,[R1 ,R2 ] ) N (DR 1 (s − R1 )ν sN −1 ds = () ≤ (R2N − R1N ) Γ(ν + 1) R1 =

ν N DR g∞,[R1 ,R2 ] 1 I =: ( ). N (R2 − R1N ) Γ(ν + 1)

(25.55)

Here

R2

I :=

ν N −1

(s − R1 ) s

R1

N −1

R2

ds = (−1)

(−s)N −1 (s − R1 )ν ds

R1 N −1

R2

= (−1)

(−R2 + R2 − s)N −1 (s − R1 )ν ds =

(25.56)

R1

N −1

R2

(−1)

R1

'N −1 ( N −1 N −1−k k (R2 − s) (s−R1 )ν ds = ((−1)N −1 )2 (−R2 ) k k=0

N −1 R2 N − 1 −k N −k−1 (k+1)−1 (ν+1)−1 (R2 − s) (s − R1 ) ds (−1) R2 k R1 k=0

608

25. Multivariate Canavati Fractional Inequalities

=

N −1 k=0

N −1 Γ(k + 1) Γ(ν + 1) (R2 − R1 )k+ν+1 = (−1)k R2N −k−1 k Γ(k + ν + 2)

Γ(ν + 1)(N − 1)!

N −1 k=0

(−1)k (R2 − R1 )k+ν+1 R2n−k−1 . (n − k − 1)! Γ(k + ν + 2)

That is,

R2

I=

(25.57)

(s − R1 )ν sN −1 ds =

R1

Γ(ν + 1) (N − 1)!

N −1 k=0

(−1)k (R2 − R1 )k+ν+1 R2N −k−1 . (N − k − 1)! Γ(k + ν + 2)

(25.58)

Continuing with (25.55) via (25.58), we have () =

ν DR g∞,[R1 ,R2 ] 1 N R2 − R1N

N!

N −1 k=0

k+ν+1 (−1)k N −k−1 (R2 − R1 ) R . (N − k − 1)! 2 Γ(k + ν + 2) (25.59)

Hence in the radial case we established ν N !DR g∞,[R1 ,R2 ] 1 f (R1 ω) − A f (y)dy ≤ V ol(A) R2N − R1N N −1 k=0

(−1)k (R2 − R1 )k+ν+1 R2N −k−1 (N − k − 1)! Γ(k + ν + 2)

.

(25.60)

Inequality (25.60) is attained by g(s) := (s−R1 )ν , 1 ≤ ν < 2, s ∈ [R1 , R2 ]. Indeed, we observe that L.H.S.(25.60) =

Γ(ν + 1) N ! (R2N − R1N )

N −1 k=0

N N R2 − R1N

R2

(s − R1 )ν sN −1 ds =

R1

(−1)k (R − R1 )k+ν+1 R2N −k−1 2 (N − k − 1)! Γ(k + ν + 2)

= R.H.S.(25.60); (25.61)

by (25.17) that says ν DR (s − R1 )ν ∞,[R1 ,R2 ] = Γ(ν + 1). !

(25.62)

We have established Theorem 25.18. Let A := B(0, R2 ) − B(0, R1 ), 0 < R1 < R2 , A ⊆ RN , N ≥ 2. Consider f : A¯ → R that is radial; that is, there exists g such

25.2 Results

609

¯ Suppose that that f (x) = g(r), x = rω, r ∈ [R1 , R2 ], ω ∈ S N −1 , x ∈ A. ν g ∈ CR ([R , R ]), 1 ≤ ν < 2. 1 2 1 Then R2 f (y)dy N = g(R1 ) − f (R1 ω) − A g(s)sN −1 ds ≤ N N V ol(A) R2 − R1 R1 N −1 ν k+ν+1 N ! DR g∞,[R1 ,R2 ] (−1)k N −k−1 (R2 − R1 ) 1 R . (N − k − 1)! 2 Γ(k + ν + 2) R2N − R1N k=0 (25.63) Inequality (25.63) is sharp; namely it is attained by g(s) = (s − R1 )ν , s ∈ [R1 , R2 ]. We continue the last remark. Remark 25.19. We treat the nonradial case here. For ﬁxed ω ∈ S N −1 the function f (rω) is radial over [R1 , R2 ]. We apply (25.63) for g = f (· ω) to get: R2 N N −1 f (sω)s ds f (R1 ω) − ≤ R2N − R1N R1 ∂ν f ⎛ ⎞ R1 N −1 N ! ∂r ν k+ν+1 ¯⎟ (−1)k ⎜ ∞,A N −k−1 (R2 − R1 ) R . ⎝ ⎠ (N − k − 1)! 2 Γ(k + ν + 2) R2N − R1N k=0

(25.64) Hence it holds R2 N S N −1 f (R1 ω)dω N −1 − N f (sω)s ds dω N ωN (R2 − R1 )ω N S N −1 R1 ⎛ ∂ν f ⎞ R1 N −1 N ! ∂r ν k+ν+1 ¯⎟ (−1)k (R − R ) ⎜ ∞,A 2 1 ≤⎝ RN −k−1 ⎠ (N − k − 1)! 2 Γ(k + ν + 2) R2N − R1N k=0 ν ∂R1 f =: C ∂rν ¯ . ∞,A That is, we proved that ν Γ( N ) ∂R1 f f (y)dy 2 S N −1 f (R1 ω)dω A ≤ C − ∂rν ¯ . N/2 V ol(A) 2π ∞,A

(25.65)

(25.66)

However, we have for x ∈ A¯ : ν N ∂R1 f Γ( f (y)dy ) f (R ω)dω 1 N −1 2 S f (x) − A ≤ f (x) − +C ∂rν ¯ . N/2 V ol(A) 2π ∞,A (25.67)

610

25. Multivariate Canavati Fractional Inequalities

We have established the following result. ¯ such that there exists ∂ ν f (x)/∂rν Theorem 25.20. Consider f ∈ C 1 (A) R1 ¯ ¯ ∈ C(A), 1 ≤ ν < 2, x ∈ A. Then N Γ( f (y)dy ) f (R ω)dω 1 N −1 2 S f (x) − A ≤ f (x) − + V ol(A) 2π N/2

N! R2N − R1N

N −1 k=0

(−1)k (R2 − R1 )k+ν+1 R2N −k−1 (N − k − 1)! Γ(k + ν + 2)

ν ∂R f 1 ∂rν ¯ . ∞,A (25.68)

We make Remark 25.21. This continues Remarks 25.17 and 25.19. Here we establish higher-order multivariate fractional Ostrowski-type inequalities over spherical shells. ¯ which implies Here ν ≥ 2, n := [ν] ≥ 2, α := ν −n. Consider f ∈ C n (A), that f (rω) ∈ C n ([R1 , R2 ]), ∀ ω ∈ S N −1 . Furthermore assume that there ν ¯ x ∈ A; ¯ x = rω, r ∈ [R1 , R2 ], ω ∈ S N −1 , f (x)/∂rν ∈ C(A), exists ∂R 1 r ν f (x)/∂rν = 1/Γ(1 − α)∂/∂r R1 (r − t)−α ∂ n f (t, ω)/∂rn dt . where ∂R 1 ν The last implies ∂R f (rω)/∂rν ∈ C([R1 , R2 ]), ∀ ω ∈ S N −1 . We start 1 ¯ again with f being radial; that is, ∃ g : f (x) = g(r), r ∈ [R1 , R2 ], x ∈ A. We have R2 f (y)dy N N −1 = f (R1 ω) − A (g(s) − g(R ))s ds =: (). 1 V ol(A) R2N − R1N R1 (25.69) By (25.10) we get

g(s) − g(R1 ) =

n−1

g

(k)

k=1

1 (s − R1 )k + (R1 ) k! Γ(ν)

s

R1

ν (s − w)ν−1 (DR g)(w)dw, 1

(25.70) all s ≥ R1 . Consequently it holds () =

1 Γ(ν)

R2

R1

N N R2 − R1N sN −1

'n−1

s

R1

k=1

|g (k) (R1 )| R2 N −1 k s (s − R1 ) ds + R1 k!

* ν ds ≤ ( by (25.58)) (s − w)ν−1 (DR g)(w)dw 1 (25.71)

25.2 Results R2N

N − R1N

' (N − 1)!

n−1

|g

(k)

k=1

N −1 λ+k+1 (−1)λ N −λ−1 (R2 − R1 ) R2 (R1 )| (N − λ − 1)! (λ + k + 1)! λ=0

ν DR g∞,[R1 ,R2 ] 1 + Γ(ν + 1)

N! R2N − R1N

'n−1

|g

k=1

ν (DR g∞,[R1 ,R2 ] ) 1

(k)

611

R2

(

ν N −1

(s − R1 ) s

ds

(25.58)

=

R1

N −1 λ+k+1 (−1)λ N −λ−1 (R2 − R1 ) (R1 )| R + (N − λ − 1)! 2 (λ + k + 1)!

N −1 λ=0

λ=0

(−1)λ (R − R1 )λ+ν+1 R2N −λ−1 2 (N − λ − 1)! Γ(λ + ν + 2)

( . (25.72)

We have established the following result. Theorem 25.22. Here ν ≥ 2, n := [ν]. Suppose that f is radial; that is, ¯ Assume that g ∈ C ν ([R1 , R2 ]). Then f (x) = g(r), r ∈ [R1 , R2 ], ∀ x ∈ A. R1 R2 N N −1 A f (y)dy g(R = E := f (R1 ω) − ) − g(s)s ds ≤ 1 V ol(A) R2N − R1N R1 (25.73) N −1 'n−1 λ+k+1 (k) (−1)λ (R − R ) N! 2 1 N −λ−1 |g (R1 )| R + (N − λ − 1)! 2 (λ + k + 1)! R2N − R1N k=1

ν (DR g∞,[R1 ,R2 ] ) 1

λ=0

N −1 λ=0

(−1)λ (R − R1 )λ+ν+1 R2N −λ−1 2 (N − λ − 1)! Γ(λ + ν + 2)

( .

Inequality (25.73) is sharp; namely it is attained by g ∗ (s) = (s − R1 )ν , s ∈ [R1 , R2 ]. Proof of Sharpness. Again by (25.17) we get ν g ∗ ∞,[R1 ,R2 ] = Γ(ν + 1). DR 1

(25.74)

Also it holds g ∗(k) (R1 ) = 0, k = 1, . . . , n − 1. Thus R2 N (s − R1 )ν sN −1 ds L.H.S.(25.73) = N N R2 − R1 R1 ( 'N −1 λ+ν+1 N ! Γ(ν + 1) (−1)λ N −λ−1 (R2 − R1 ) = R = R.H.S.(25.73). (N − λ − 1)! 2 Γ(λ + ν + 2) R2N − R1N λ=0 (25.75) We give

612

25. Multivariate Canavati Fractional Inequalities

Corollary 25.23. In the terms and assumptions of Theorem 25.22, additionally suppose that g (k) (R1 ) = 0, k = 1, . . . , n − 1. Then N −1 λ+ν+1 (−1)λ N! N −λ−1 (R2 − R1 ) R · E≤ (N − λ − 1)! 2 Γ(λ + ν + 2) R2N − R1N λ=0

ν DR g∞,[R1 ,R2 ] . 1

(25.76)

We continue Remark 25.21 with Remark 25.24. We treat here the general, not necessarily radial, case of f . We apply (25.73) to f (rω), ω ﬁxed, r ∈ [R1 , R2 ]. We then have R2 N! N N −1 f (R ≤ ω) − f (sω)s ds 1 R2N − R1N R2N − R1N R1 N −1 'n−1 λ+k+1 ∂ k f (−1)λ N −λ−1 (R2 − R1 ) R k (R1 ω) ∂r (N − λ − 1)! 2 (λ + k + 1)! k=1

λ=0

N −1 ( ν λ+ν+1 ∂R1 f (−1)λ N −λ−1 (R2 − R1 ) + R . ∂rν ¯ (N − λ − 1)! 2 Γ(λ + ν + 2) ∞,A

(25.77)

λ=0

Therefore R2 N S N −1 f (R1 ω)dω N −1 − f (sω)s ds dω ωN (R2N − R1N )ω N S N −1 R1 ≤

N! R2N − R1N

k ⎞ ⎡n−1 ⎛ ∂ f S N −1 ∂r k (R1 ω) dω ⎝ ⎠ ⎣ ωN k=1

N −1 λ+k+1 (−1)λ N −λ−1 (R2 − R1 ) R (N − λ − 1)! 2 (λ + k + 1)! λ=0

ν ∂R1 f + ∂rν That is,

N −1 ¯ ∞,A

λ=0

(−1)λ (R − R1 )λ+ν+1 R2N −λ−1 2 (N − λ − 1)! (λ + ν + 2)

(

Γ( N ) 2 S N −1 f (R1 ω)dω A f (y)dy − ≤ δ. N/2 V ol(A) 2π

Consequently it holds for x ∈ A¯ that Γ( N f (y)dy 2 ) S N −1 f (R1 ω)dω f (x) − A ≤ f (x) − + δ. V ol(A) 2π N/2 We have established the next result.

=: δ. (25.78)

(25.79)

(25.80)

25.2 Results

613

¯ and assume Theorem 25.25. Here ν ≥ 2, n := [ν]. Consider f ∈ C n (A) ν ν ¯ ¯ Then f (x)/∂r ∈ C( A), x ∈ A. there exists ∂R 1 Γ( N 2 ) S N −1 f (R1 ω)dω A f (y)dy f (x) − ≤ M := f (x) − + V ol(A) 2π N/2

' N! R2N − R1N

n−1 Γ( N 2) 2π N/2

k=1

k ∂ f k (R1 ω) dω N −1 ∂r S

N −1 λ+k+1 (−1)λ N −λ−1 (R2 − R1 ) R + (N − λ − 1)! 2 (λ + k + 1)! λ=0

ν ∂R1 f + ∂rν

N −1

¯ ∞,A

λ=0

(−1)λ (R − R1 )λ+ν+1 R2N −λ−1 2 (N − λ − 1)! (λ + ν + 2)

( .

(25.81)

We ﬁnish with Corollary 25.26. In the terms and assumptions of Theorem 25.25, additionally suppose that ∂ k f /∂rk , k = 1, . . . , n − 1, vanish on ∂B(0, R1 ). Then Γ( N 2) M ≤ f (x) − f (R ω)dω + 1 2π N/2 S N −1 N! N R2 − R1N

N −1 λ=0

(−1)λ (R − R1 )λ+ν+1 R2N −λ−1 2 (N − λ − 1)! Γ(λ + ν + 2)

ν ∂R1 f ∂rν ¯ . ∞,A (25.82)

26 Caputo Fractional Ostrowski-Type Inequalities

Optimal upper bounds are given to the deviation of a value of a univariate or multivariate function of a Caputo fractional derivative related space from its average, over convex and compact subsets of RN , N ≥ 1. In particular we work over closed intervals, rectangles, balls, and spherical shells. These bounds involve the supremum and L∞ norms of related univariate or multivariate Caputo fractional derivatives of the involved functions. The derived inequalities are sharp; namely they are attained by simple functions. This chapter has been motivated by the works of Ostrowski [318], 1938, and of the author’s [24], 2003 and [43], 2007, and the chapter also relies on [52].

26.1 Background We start with Deﬁnition 26.1. [134] Let ν ≥ 0; the operator Jaν , deﬁned on L1 (a, b) by Jaν f (x) :=

1 Γ (ν)

x

(x − t)ν−1 f (t) dt

(26.1)

a

for a ≤ x ≤ b, is called the Riemann–Liouville fractional integral operator of order ν. For ν = 0, we set Ja0 := I, the identity operator. Here Γ stands for the gamma function. By Theorem 2.1 of [134, p. 13], Jaν f (x) , ν > 0, exists for almost all x ∈ [a, b] and Jaν f ∈ L1 (a, b) , where f ∈ L1 (a, b). G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 615 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 26,

616

26. Caputo Fractional Ostrowski-Type Inequalities

Here AC n ([a, b]) is the space of functions with absolutely continuous (n − 1)st derivative. We need to mention Deﬁnition 26.2. [58, 134] Let ν ≥ 0, n := ν ; · is the ceiling of the number, f ∈ AC n ([a, b]) . We call the Caputo fractional derivative x 1 ν f (x) := (x − t)n−ν−1 f (n) (t) dt, (26.2) D∗a Γ (n − ν) a ∀x ∈ [a, b]. ν f (x) exists almost everywhere for x ∈ [a, b]. If ν ∈ N, The above function D∗a ν 0 f = f (ν) the ordinary derivative; it is also D∗a f = f. then D∗a We need Theorem 26.3. (Taylor expansion for Caputo derivatives, [134, p. 40]) Assume ν ≥ 0, n := ν , and f ∈ AC n ([a, b]). Then f (x) =

n−1 k=0

f (k) (a) 1 (x − a)k + k! Γ (ν)

x a

ν (x − t)ν−1 D∗a f (t) dt,

(26.3)

∀x ∈ [a, b]. Additionally assume f (k) (a) = 0, k = 1, . . . , n − 1; then f (x) − f (a) =

1 Γ (ν)

x a

ν (x − t)ν−1 D∗a f (t) dt.

(26.4)

Next we mention the multivariate analogue of (26.3) and (26.4) (see [53]). Remark 26.4. Let Q be a compact and convex subset of Rk , k ≥ 2; z := (z1 , . . . , zk ), x0 := (x01 , . . . , x0k ) ∈ Q. Let f ∈ C n (Q), n ∈ N. Set gz (t) := f (x0 + t (z − x0 )), 0 ≤ t ≤ 1; gz (0) = f (x0 ) , gz (1) = f (z). ⎡

Then

gz (t) = ⎣ (j)

k i=1

∂ (zi − x0i ) ∂xi

(26.5)

j ⎤ f ⎦ (x0 + t (z − x0 )),

(26.6)

n ⎤ f ⎦ (x0 ).

(26.7)

j = 0, 1, 2, . . . , n, and ⎡ (n)

gz

(0) = ⎣

k i=1

∂ (zi − x0i ) ∂xi

26.1 Background If all fα (x0 ) :=

617

∂αf (x0 ) = 0, α := (α1 , . . . , αk ), ∂xα

αi ∈ Z+ , i = 1, . . . , k; |α| :=

k

αi =: l,

i=1

then

(l)

gz (0) = 0, where l ∈ {0, 1, . . . , n}. We quote that gz (t) =

k

(zi − x0i )

i=1

∂f (x0 + t (z − x0 )). ∂xi

(26.8)

When f ∈ C 2 (Q), Q ⊆ R2 , we have gz (t) = (z1 − x01 )2

(z2 − x02 )

∂2f (x0 + t (z − x0 )) + 2 (z1 − x01 ) ∂x21 ∂2f (x0 + t (z − x0 )) ∂x1 ∂x2

+ (z2 − x02 )2

∂2f (x0 + t (z − x0 )), ∂x22

(26.9)

and so on. Clearly here gz ∈ C n ([0, 1]) , hence gz ∈ AC n ([0, 1]). In [53] we proved the following general multivariate fractional Taylor formula. Theorem 26.5. [53] Let ν > 0, n = ν , f ∈ C n (Q), where Q is a compact and convex subset of Rk , k ≥ 2; z := (z1 , . . . , zk ), x0 := (x01 , . . . , x0k ) ∈ Q. Then k ∂f (x0 ) (zi − x0i ) (1) f (z) = f (x0 ) + ∂xi i=1 ) l * "k ∂ (z − x ) f (x0 ) n−1 i 0i i=1 ∂xi + + l! l=2 ⎧ ⎡ ⎡ n 1 k ⎨ ∂ 1 ν−1 ⎣ n−ν ⎣ (1 − t) (zi − x0i ) J0 ⎩ Γ (ν) 0 ∂xi i=1

f ] (x0 + t (z − x0 ))}] dt. Additionally assume that fα (x0 ) = 0, α := (α1 , . . . , αk ) , αi ∈ Z+ , i = 1, . . . , k;

(26.10)

618

26. Caputo Fractional Ostrowski-Type Inequalities |α| :=

k

αi =: r, r = 1, . . . , n − 1; then

i=1

(2) f (z) − f (x0 ) =

1 Γ (ν)

1 0

+ (1 − t)ν−1 J0n−ν

⎧ ⎡ n ⎤ k ⎨ ∂ ⎣ (zi − x0i ) f⎦ ⎩ ∂xi i=1

(x0 + t (z − x0 ))}] dt =: Rν .

(26.11)

Remark 26.6. [53] (on Theorem 26.5) Set ⎧ ⎡ n ⎤ k ⎨ ∂ n−ν ⎣ (zi − x0i ) f⎦ Gν (t) := J0 ⎩ ∂xi i=1

(x0 + t (z − x0 ))}, t ∈ [0, 1],

(26.12)

which shows up in Rν and is continuous; see Proposition 114 of [45] and Rν ∈ R. So we can rewrite 1 1 (1 − t)ν−1 Gν (t) dt, ν > 0. (26.13) Rν = Γ (ν) 0 We mention Theorem 26.7. [53] All are as in Theorem 26.5. Let Rν be the remainder in (26.10) (same as in (26.11)), and Gν (t), t ∈ [0, 1] as in (26.12), ν ≥ 1. Then % Gν Lq ([0,1]) Gν L1 ([0,1]) |Rν | ≤ min , Γ (ν) Γ (ν) (p (ν − 1) + 1)1/p Gν L2 ([0,1]) Gν ∞,[0,1] √ , , (26.14) =: Λν , Γ (ν) 2ν − 1 Γ (ν + 1) where p, q > 1 : 1/p + 1/q = 1.

26.2 Univariate Results We present here our ﬁrst Ostrowski-type result. Theorem 26.8. Let ν ≥ 0, n = ν , and f ∈ AC n ([a, b]). Assume that ν f ∈ L∞ ([a, b]). f (k) (a) = 0, k = 1, . . . , n − 1, and D∗a

26.2 Univariate Results Then

b ν D∗a f ∞,[a,b] 1 f (x) dx − f (a) ≤ (b − a)ν . b − a a Γ (ν + 2)

619

(26.15)

Proof. By (26.4) we get |f (x) − f (a)| ≤ 1 ≤ Γ (ν)

x

1 Γ (ν)

x a

ν−1

(x − t)

a

=

ν (x − t)ν−1 D∗a f (t) dt ν dt D∗a f ∞,[a,b]

(x − a)ν ν D∗a f ∞,[a,b] . Γ (ν + 1)

That is, we have |f (x) − f (a)| ≤

ν D∗a f ∞,[a,b]

Γ (ν + 1)

(x − a)ν ,

(26.16)

∀x ∈ [a, b]. Therefore we get b b 1 1 f (x) dx − f (a) = (f (x) − f (a)) dx b − a a b − a a ≤

b ν (26.16) D∗a f ∞,[a,b] 1 |f (x) − f (a)| dx ≤ b−a a Γ (ν + 1) (b − a) b ν D∗a f ∞,[a,b] (x − a)ν dx = (b − a)ν . Γ (ν + 2) a

This proves (26.15).

We continue with Theorem 26.9. Let ν ≥ 1, n = ν , and f ∈ AC n ([a, b]) . Assume that ν f (k) (a) = 0, k = 1, . . . , n − 1, and D∗a f ∈ L1 ([a, b]).

Then

b ν D∗a f L1 ([a,b]) 1 f (x) dx − f (a) ≤ (b − a)ν−1 . b − a a Γ (ν + 1)

Proof. Again by (26.4) we get |f (x) − f (a)| ≤

1 Γ (ν)

x a

ν (x − t)ν−1 D∗a f (t) dt

(26.17)

620

26. Caputo Fractional Ostrowski-Type Inequalities ≤

1 (x − a)ν−1 Γ (ν) 1 Γ (ν)

ν D∗a f (t) dt ≤

x a

ν f (x − a)ν−1 D∗a

L1 ([a,b])

.

That is, we have |f (x) − f (a)| ≤

ν f L1 ([a,b]) D∗a

Γ (ν)

(x − a)ν−1 ,

(26.18)

∀x ∈ [a, b]. Therefore we get b b (26.18) 1 1 f (x) dx − f (a) ≤ |f (x) − f (a)| dx ≤ b − a a b−a a ν D∗a f L1 ([a,b])

Γ (ν) (b − a)

b

(x − a)ν−1 dx =

a

ν D∗a f L1 ([a,b])

(b − a)ν−1 .

Γ (ν + 1)

This proves (26.17). We also give

Theorem 26.10. Let p, q > 1 : 1/p + 1/q = 1, and ν > 1 − 1/p, n = ν , and f ∈ AC n ([a, b]). Assume that ν f (k) (a) = 0, k = 1, . . . , n − 1, and D∗a f ∈ Lq ([a, b]).

b 1 f (x) dx − f (a) ≤ b − a a

Then

ν D∗a f Lq ([a,b])

(b − a)ν−1+1/p .

Γ (ν) (p (ν − 1) + 1)1/p (ν + 1/p) Proof. Again by (26.4) we obtain |f (x) − f (a)| ≤

≤

1 Γ (ν)

x

1 Γ (ν)

x a

ν (x − t)ν−1 D∗a f (t) dt

1/p (x − t)p(ν−1) dt

a

1/q q ν D∗a f (t) dt ≤

x a

ν (x − a) 1 D∗a f . Lq ([a,b]) Γ (ν) (p (ν − 1) + 1)1/p ν−1+1/p

(26.19)

26.2 Univariate Results

621

That is, we have |f (x) − f (a)| ≤

ν f Lq ([a,b]) D∗a 1/p

Γ (ν) (p (ν − 1) + 1)

(x − a)ν−1+1/p ,

(26.20)

∀x ∈ [a, b] . Consequently we get b b (26.20) 1 1 ≤ f (x) dx − f (a) |f (x) − f (a)| dx ≤ b − a a b−a a ν D∗a f Lq ([a,b])

Γ (ν) (p (ν − 1) + 1)1/p (b − a) =

b

(x − a)ν−1+1/p dx

a

ν f Lq ([a,b]) D∗a

Γ (ν) (p (ν − 1) + 1)1/p (ν + 1/p)

This proves (26.19) .

(b − a)ν−1+1/p .

Corollary 26.11. (To Theorem 26.10; p = q = 2 case.) Let ν > 1/2, n = ν , and f ∈ AC n ([a, b]). Assume that ν f (k) (a) = 0, k = 1, . . . , n − 1, and D∗a f ∈ L2 ([a, b]).

b 1 f (x) dx − f (a) ≤ b − a a

Then

ν D∗a f L2 ([a,b]) (b − a)ν−1/2 . √ Γ (ν) 2ν − 1 ν + 12

Proof. Apply (26.19) .

(26.21)

We ﬁnish this section with Proposition 26.12. Inequality (26.15) is sharp; namely it is attained by / N, x ∈ [a, b] . f (x) = (x − a)ν , ν > 0, ν ∈

Proof. Here the function f (x) = (x − a)ν , ν > 0, ν ∈ / N, x ∈ [a, b], belongs to AC n ([a, b]) , n = ν . We observe that f (x) = ν (x − a)ν−1 , f (x) = ν (ν − 1) (x − a)ν−2 , . . . ,

622

26. Caputo Fractional Ostrowski-Type Inequalities f (n−1) (x) = ν (ν − 1) (ν − 2) . . . (ν − n + 2) (x − a)ν−n+1 f (n) (x) = ν (ν − 1) . . . (ν − n + 1) (x − a)ν−n .

Thus ν f (x) D∗a

(26.2)

=

1 Γ (n − ν)

x

(x − t)(n−ν)−1

a

ν (ν − 1) . . . (ν − n + 1) (t − a)ν−n dt ν (ν − 1) . . . (ν − n + 1) x (x − t)(n−ν)−1 (t − a)(ν−n+1)−1 dt = Γ (n − ν) a (by Whittaker and Watson [404, p. 256]; notice that n − ν, ν − n + 1 > 0) =

ν (ν − 1) . . . (ν − n + 1) Γ (n − ν) Γ (ν − n + 1) Γ (n − ν) Γ (1)

= ν (ν − 1) . . . (ν − n + 1) Γ (ν − n + 1) = Γ (ν + 1) . That is, ν f (x) = Γ (ν + 1) , ∀x ∈ [a, b]. D∗a

(26.22)

Also we see that ν f (k) (a) = 0, k = 0, 1, . . . , n − 1, and D∗a f ∈ L∞ ([a, b]) .

So f fulﬁlls all the assumptions of Theorem 26.8. Next we ﬁnd R.H.S. (26.15) =

(b − a)ν Γ (ν + 1) (b − a)ν = . Γ (ν + 2) (ν + 1)

(26.23)

Furthermore we have b 1 ν (x − a) dx L.H.S. (26.15) = b − a a

=

(b − a)ν+1 (b − a)ν 1 = . (b − a) (ν + 1) (ν + 1)

Clearly R.H.S. (26.15) = L.H.S. (26.15), proving the claim.

(26.24)

26.3 Multivariate Results We present our ﬁrst multivariate Ostrowski-type fractional inequality in the Caputo sense.

26.3 Multivariate Results

623

Theorem 26.13. Let ν > 0, n = ν , f ∈ C n (Q), where Q is a nonempty compact and convex subset of Rk , k ≥ 2; z := (z1 , . . . , zk ), x0 := (x01 , . . . , x0k ) ∈ Q. Further assume that fα (x0 ) = 0, α := (α1 , . . . , αk ) , αi ∈ Z+ , i = 1, . . . , k; |α| :=

k

αi =: r, r = 1, . . . , n − 1,

i=1

where x0 ∈ Q is ﬁxed. Then

Q f (z) dz − f (x0 ) ≤ V ol (Q) ⎛ ⎞ max J0n−ν fα (x0 + t (z − x0 )) ⎜ |α|=n ∞,(t,z)∈[0,1]×Q ⎟ ⎜ ⎟ z − x0 n l1 dz. (26.25) ⎝ ⎠ Γ (ν + 1) V ol (Q) Q

Proof. Notice that (see (26.12)) ⎡ n ⎤ k ∂ n−ν (zi − x0i ) J0 f⎦ |Gν (t)| = ⎣ ∂xi i=1 n (x0 + t (z − x0 ))| ≤ z − x0 l1 max J0n−ν fα (x0 + t (z − x0 ))

(26.26)

∞,(t,z)∈[0,1]×Q

|α|=n

,

∀z ∈ Q, x0 ∈ Q being ﬁxed, 0 ≤ t ≤ 1. Hence

n Gν ∞,t∈[0,1] ≤ z − x0 l1 max J0n−ν fα (x0 + t (z − x0 ))

∞,(t,z)∈[0,1]×Q

|α|=n

,

(26.27)

and by (26.11) and (26.14) we get |f (z) − f (x0 )| ≤ Λν ≤

Gν ∞,t∈[0,1] Γ (ν + 1)

max J0n−ν fα (x0 + t (z − x0 ))

|α|=n

(26.27)

≤

z − x0 l1

n

Γ (ν + 1)

∞,(t,z)∈[0,1]×Q

,

∀z ∈ Q. Consequently we obtain Q (f (z) − f (x0 )) dz Q f (z) dz = ) − f (x ≤ 0 V ol (Q) V ol (Q)

(26.28)

624

26. Caputo Fractional Ostrowski-Type Inequalities 1 V ol (Q)

⎛

|f (z) − f (x0 )| dz ≤ Q

max J0n−ν fα (x0 + t (z − x0 ))

⎜ |α|=n ⎜ ⎝

⎞ ∞,(t,z)∈[0,1]×Q ⎟

⎟ ⎠

Γ (ν + 1) V ol (Q)

proving the claim.

Q

z − x0 n l1 dz, (26.29)

We further have Theorem 26.14. Let f : [a, b] × [c, d] → R be Lebesgue integrable, also f (a, ·) integrable on [c, d] and f (·, y) ∈ AC n ([a, b]), ∀y ∈ [c, d], where n = ν , ν ≥ 0, and ∂ k /∂xk f (a, y) = 0, k = 1, . . . , n − 1, ∀y ∈ [c, d]. Here ·∞,[a,b]×[c,d] is the supremum norm. ν f (x, y)/∂xν ∈ B ([a, b] × [c, d]) bounded functions, Further suppose that ∂∗a (x, y) ∈ [a, b] × [c, d]. Then 1 f (x, y) dxdy − f (a, c) ≤ (b − a) (d − c) [a,b]×[c,d] ν (b − a)ν 1 ∂∗a f |f (a, y) − f (a, c)| dy + . (26.30) ν (d − c) [c,d] Γ (ν + 2) ∂x ∞,[a,b]×[c,d] Proof. Because f (·, y) ∈ AC n ([a, b]), ∀y ∈ [c, d], and ∂ k /∂xk f (a, y) = 0, k = 1, . . . , n − 1, ∀y ∈ [c, d], by (26.4) we have f (x, y) − f (a, y) = ∀x ∈ [a, b] , ∀y ∈ [c, d]. Because

1 Γ (ν)

x a

(x − t)ν−1

ν ∂∗a f (t, y) dt, ∂xν

(26.31)

ν f ∂∗a (x, y) ∈ B ([a, b] × [c, d]), ∂xν

we obtain |f (x, y) − f (a, y)| ≤

ν (x − a)ν ∂∗a f , ν Γ (ν + 1) ∂x ∞,[a,b]×[c,d]

(26.32)

∀x ∈ [a, b], ∀y ∈ [c, d]. But it holds |f (x, y) − f (a, c)| ≤ |f (x, y) − f (a, y)| + |f (a, y) − f (a, c)| ≤ |f (a, y) − f (a, c)| + ν (x − a)ν ∂∗a f , Γ (ν + 1) ∂xν ∞,[a,b]×[c,d] ∀x ∈ [a, b], ∀y ∈ [c, d].

(26.33)

26.3 Multivariate Results

625

Consequently we have 1 f (x, y) dxdy − f (a, c) = (b − a) (d − c) [a,b]×[c,d] 1 (f (x, y) − f (a, c)) dxdy ≤ (b − a) (d − c) [a,b]×[c,d] 1 (b − a) (d − c)

|f (x, y) − f (a, c)| dxdy

(26.34)

(26.33)

≤

[a,b]×[c,d]

1 (b − a) (d − c)

|f (a, y) − f (a, c)| dxdy +

ν ∂∗a f 1 1 (b − a) (d − c) Γ (ν + 1) ∂xν ∞,[a,b]×[c,d] 1 (x − a)ν dxdy = |f (a, y) − f (a, c)| dy (d − c) [c,d] [a,b]×[c,d] ν (b − a)ν ∂∗a f , + ν Γ (ν + 2) ∂x ∞,[a,b]×[c,d] proving the claim.

(26.35)

[a,b]×[c,d]

(26.36)

Similar to (26.30) one can prove inequalities in more than two variables. Next we study fractional Ostrowski-type inequalities over balls and spherical shells. For that we need to make Remark 26.15. We deﬁne the ball

B (0, R) := x ∈ RN : |x| < R ⊆ RN , N ≥ 2, R > 0, and the sphere

S N −1 := x ∈ RN : |x| = 1 ,

where |·| is the Euclidean norm. Let dω be the element of surface measure on S N −1 and let ωN =

S N −1

dω =

2π N/2 . Γ (N/2)

For x ∈ RN − {0} we can write uniquely x = rω, where r = |x| > 0 and ω = x/r ∈ S N −1 , |ω| = 1. Note that dy = B(0,R)

ω N RN N

626

26. Caputo Fractional Ostrowski-Type Inequalities

is the Lebesgue measure of the ball. Following [356, pp. 149–150, Exercise 6] and [383, pp. 87–88, Theorem 5.2.2] we can write for F : B (0, R) → R a Lebesgue integrable function that

F (x) dx = B(0,R)

R

F (rω) r

S N −1

N −1

dr

dω;

(26.37)

0

we use this formula often. Initially the function f : B (0, R) → R is radial; that is, there exists a function g such that f (x) = g (r), where r = |x|, r ∈ [0, R], ∀x ∈ B (0, R). Here we assume that g ∈ AC n ([0, R]), n = ν , ν ≥ 0, and g (k) (0) = 0, k = 1, . . . , n − 1. By (26.4) we get g (s) − g (0) =

1 Γ (ν)

s 0

ν (s − t)ν−1 D∗0 g (t) dt,

(26.38)

∀s ∈ [0, R]. ν g ∈ L∞ ([0, R]). Further assume that D∗0 By (26.38) we obtain |g (s) − g (0)| ≤

ν sν D∗0 g , ∞,[0,R] Γ (ν + 1)

(26.39)

∀s ∈ [0, R]. Next we observe that B(0,R) f (y) dy = f (0) − V ol (B (0, R)) R N −1 ds dω S N −1 0 g (s) s g (0) − = R N −1 s ds dω N −1 S 0

(26.40)

R N N N −1 g (s) s ds = N g (0) − N R R 0 R N −1 s (g (0) − g (s)) ds ≤ 0 N RN

R

sN −1 |g (0) − g (s)| ds

(26.41)

(26.39)

≤

0

ν N D∗0 g∞,[0,R] R N −1 ν s s ds = Γ (ν + 1) RN 0 ν ν ν ν+N g D D g ∗0 ∞,[0,R] N R ∗0 ∞,[0,R] R N = . Γ (ν + 1) (ν + N ) Γ (ν + 1) (ν + N ) RN

(26.42)

(26.43)

26.3 Multivariate Results

627

That is, we have proved that B(0,R) f (y) dy = f (0) − V ol (B (0, R)) ν R D N Rν g ∗0 ∞,[0,R] N g (s) sN −1 ds ≤ . g (0) − N Γ (ν + 1) (ν + N ) R 0

(26.44)

The last inequality (26.44) is sharp; namely it is attained by g (r) = rν , ν > 0, r ∈ [0, R]. Indeed by (26.2) we get ν g (r) = Γ (ν + 1) , D∗0

and

ν D∗0 g

∞,[0,R]

= Γ (ν + 1) .

So that in that case R.H.S (26.44) = But L.H.S (26.44) = =

N RN

N Rν . ν+N

R

(26.45)

sν+N −1 ds

0

N Rν+N N Rν = , N ν+N R (ν + N )

(26.46)

proving equality in (26.44). We have established the following. Theorem 26.16. Let f : B (0, R) → R that is radial; that is, there exists g such that f (x) = g (r), r = |x|, ∀x ∈ B (0, R). Assume that g ∈ AC n ([0, R]), n = ν , ν ≥ 0, and ν g ∈ L∞ ([0, R]). g (k) (0) = 0, k = 1, . . . , n − 1, and D∗0

Then

B(0,R) f (y) dy f (0) − = V ol (B (0, R)) ν R D∗0 g∞,[0,R] N Rν N N −1 g (s) s ds ≤ . g (0) − N Γ (ν + 1) (ν + N ) R 0

(26.47)

Inequality (26.47) is sharp; it is attained by g (r) = rν , ν > 0. We make Remark 26.17. Let the spherical shell A := B (0, R2 ) − B (0, R1 ), 0 < R1 < R2 , A ⊆ RN , N ≥ 2, x ∈ A. Consider again that f : A → R is radial; that is,

628

26. Caputo Fractional Ostrowski-Type Inequalities

there exists g such that f (x) = g (r), r = |x|, r ∈ [R1 , R2 ], ∀x ∈ A. Here again x can be written uniquely as x = rω, where r = |x| > 0, and ω = x/r ∈ S N −1 , |ω| = 1. Following [356, pp. 149–150, Exercise 6] and [383, pp. 87–88, Theorem 5.2.2] we can write for F : A → R a Lebesgue integrable function that R2

F (x) dx = A

S N −1

Here V ol (A) =

F (rω) rN −1 dr

dω.

(26.48)

R1

ω N R2N − R1N N

,

and we assume that g ∈ AC n ([R1 , R2 ]), n = ν , ν ≥ 0, and g (k) (R1 ) = 0, k = 1, . . . , n − 1. By (26.4) we get g (s) − g (R1 ) =

1 Γ (ν)

s R1

ν (s − t)ν−1 D∗R g (t) dt, 1

(26.49)

∀s ∈ [R1 , R2 ]. ν g ∈ L∞ ([R1 , R2 ]) . By (26.49) we obtain Further assume that D∗R 1 |g (s) − g (R1 )| ≤

(s − R1 )ν ν D∗R g ∞,[R ,R ] , 1 1 2 Γ (ν + 1)

(26.50)

∀s ∈ [R1 , R2 ]. Next we observe that f (y) dy = f (R1 ω) − A V ol (A)

R2N

R2 N N −1 g (R ) − g (s) s ds = 1 N N R2 − R1 R1 R2 N N −1 (g (R ) − g (s)) s ds ≤ 1 R2N − R1N R1 R2 (26.50) N |g (R1 ) − g (s)| sN −1 ds ≤ R2N − R1N R1 ν D R2 ∗R1 g ∞,[R1 ,R2 ] N (s − R1 )ν sN −1 ds Γ (ν + 1) − R1N R1

(by [43], there (2.41)) =

R2N

N − R1N

ν D∗R g ∞,[R 1

1 ,R2 ]

Γ (ν + 1)

(Γ (ν + 1) (N − 1)!

(26.51)

(26.52)

(26.53)

26.3 Multivariate Results N −1 k=0

(−1)k (R − R1 )k+ν+1 R2N −k−1 2 (N − k − 1)! Γ (k + ν + 2) ⎛ =⎝

N −1 k=0

ν g ∞,[R N ! D∗R 1

629

(26.54)

⎞

1 ,R2 ]

R2N − R1N

⎠

(−1)k (R − R1 )k+ν+1 R2N −k−1 2 (N − k − 1)! Γ (k + ν + 2)

.

(26.55)

.

(26.56)

Hence in the radial case we proved f (y) dy = f (R1 ω) − A V ol (A) R2 N N −1 g (s) s ds ≤ g (R1 ) − N N R2 − R1 R1 ⎛ ⎞ ν N ! D∗R g ∞,[R ,R ] 1 1 2 ⎝ ⎠ R2N − R1N N −1 k=0

(−1)k (R − R1 )k+ν+1 R2N −k−1 2 (N − k − 1)! Γ (k + ν + 2)

Inequality (26.56) is attained by g (s) := (s − R1 )ν , ν > 0, s ∈ [R1 , R2 ]. Indeed we observe that L.H.S. (26.56) =

Γ (ν + 1) N ! N R2 − R1N

N −1 k=0

R2N

N − R1N

R2

(s − R1 )ν sN −1 ds =

R1

(−1)k (R − R1 )k+ν+1 R2N −k−1 2 (N − k − 1)! Γ (k + ν + 2)

.

(26.57)

But by (26.22) we get ν g (s) = Γ (ν + 1) , ∀s ∈ [R1 , R2 ]. D∗R 1

That is,

ν D∗R g 1 ∞,[R

1 ,R2 ]

= Γ (ν + 1).

The last implies that

R.H.S. (26.56) =

Γ (ν + 1) N ! R2N − R1N

(26.58)

(26.59)

630

26. Caputo Fractional Ostrowski-Type Inequalities N −1 k=0

(−1)k (R − R1 )k+ν+1 R2N −k−1 2 (N − k − 1)! Γ (k + ν + 2)

.

(26.60)

So by (26.57) and (26.60) we get L.H.S. (26.56) = R.H.S. (26.56),

(26.61)

proving sharpness of (26.56). So putting the above together we derive Theorem 26.18. Let f : A → R be radial; that is, there exists g such that f (x) = g (r), r = |x|, ∀x ∈ A; ω ∈ S N −1 . Assume that g ∈ AC n ([R1 , R2 ]), n = ν , ν ≥ 0, and ν g ∈ L∞ ([R1 , R2 ]). g (k) (R1 ) = 0, k = 1, . . . , n − 1, and D∗R 1

f (y) dy f (R1 ω) − A = V ol (A) R2 N N −1 g (s) s ds ≤ g (R1 ) − N N R2 − R1 R1 ν N ! D∗R1 g ∞,[R ,R ]

Then

1

2

R2N − R1N N −1 k=0

(−1)k (R − R1 )k+ν+1 R2N −k−1 2 (N − k − 1)! Γ (k + ν + 2)

.

(26.62)

Inequality (26.62) is sharp; namely it is attained by g (s) := (s − R1 )ν , ν > 0, s ∈ [R1 , R2 ] .

(26.63)

We need Deﬁnition 26.19. (See [58].) Let F : A → R, ν ≥ 0, n := ν such that F (·ω) ∈ AC n ([R1 , R2 ]), for all ω ∈ S N −1 . We call the Caputo radial fractional derivative the following function ν F (x) ∂∗R 1 1 := ∂rν Γ (n − ν)

r R1

(r − t)n−ν−1

∂ n F (tω) dt, ∂rn

where x ∈ A; that is, x = rω, r ∈ [R1 , R2 ] , ω ∈ S N −1 . Clearly 0 ∂∗R F (x) 1 = F (x) , ∂r0

(26.64)

26.3 Multivariate Results ν F (x) ∂∗R ∂ ν F (x) 1 = , if ν ∈ N. ν ∂r ∂rν

631 (26.65)

The above function (26.64) exists almost everywhere for x ∈ A (see [58]). We continue Remark 26.17; we make Remark 26.20. We treat here the general, not necessarily radial, case of f. We apply (26.62) to f (rω) , ω ﬁxed, r ∈ [R1 , R2 ], under the following assumptions: f (·ω) ∈ AC n ([R1 , R2 ]) , for all ω ∈ S N −1 , ν ≥ 0, n := ν , where f : A → R is Lebesgue integrable; ∂kf , k = 1, . . . , n − 1 vanish on ∂B (0, R1 ) ; ∂rk and

ν f ∂∗R 1 ∈ B A , along with ∂rν ν D∗R f (·ω) ∈ L∞ ([R1 , R2 ]), ∀ω ∈ S N −1 . 1

R2 N N −1 f (sω) s ds ≤ f (R1 ω) − N N R2 − R1 R1 ν ⎛ ⎞ ∂∗R f 1 −1 ⎜ N ! ∂rν ⎟ N R2N −k−1 (R2 − R1 )k+ν+1 (−1)k ⎜ ∞,A ⎟ ⎜ ⎟ (N − k − 1)! Γ (k + ν + 2) ⎝ R2N − R1N ⎠

So we have

k=0

=: λ1 .

(26.66)

Consequently it holds N S N −1 f (R1 ω) dω − N ωN R2 − R1N ω N S N −1

R2

f (sω) s R1

That is,

ds dω ≤ λ1 .

N −1

Γ (N/2) A f (x) dx ≤ λ1 . f (R ω) dω − 1 2π N/2 V ol (A) S N −1

Therefore it holds for x ∈ A that f (x) dx ≤ f (x) − Γ (N/2) f (x) − A f (R1 ω) dω + λ1 . V ol (A) 2π N/2 S N −1 We have established the next result.

(26.67)

(26.68)

(26.69)

632

26. Caputo Fractional Ostrowski-Type Inequalities

Theorem 26.21. Let f : A → R be Lebesgue integrable with f (·ω) ∈ AC n ([R1 , R2 ]), ν ≥ 0, n := ν , ∀ω ∈ S N −1 ; ∂ k f /∂rk , k = 1, . . . , n − 1 vanish on ∂B (0, R1 ); ν N −1 ν ν f (·ω) ∈ L ([R , R ]), ∀ω ∈ S ; and ∂ f /∂r ∈ B A (bounded D∗R ∞ 1 2 ∗R1 1 functions on A). Then for x ∈ A we have f (x) dx ≤ f (x) − Γ (N/2) f (x) − A f (R ω) dω 1 N/2 V ol (A) 2π S N −1 ν ⎛ ⎞ ∂∗R f 1 −1 N ! ∂rν ⎜ ⎟ N R2N −k−1 (R2 − R1 )k+ν+1 (−1)k ⎜ ∞,A ⎟ +⎜ . ⎟ (N − k − 1)! Γ (k + ν + 2) ⎝ R2N − R1N ⎠ k=0 (26.70) We make Remark 26.22. Let f : B (0, R) → R be a Lebesgue integrable function, that is not necessarily a radial function. Assume f (·ω) ∈ AC 1 ([0, R]), ∀ω ∈ S N −1 ; ν f (·ω) ∈ L∞ ([0, R]), ∀ω ∈ S N −1 . 0 ≤ ν < 1, and D∗0 Clearly here by (26.3) we obtain 1 Γ (ν)

f (sω) − f (0) =

s 0

ν (s − t)ν−1 D∗0 f (tω) dt,

(26.71)

∀ω ∈ S N −1 , ∀s ∈ [0, R] . We further assume that ν D∗0 f (tω)

∞,(t∈[0,R])

≤ K, ∀ω ∈ S N −1 ,

(26.72)

where K > 0. So we apply (26.47) for f (·ω), ∀ω ∈ S N −1 . We obtain R N N −1 f (sω) s ds ≤ f (0) − N R 0 ν D∗0 f (tω)∞,(t∈[0,R]) N Rν (26.72) ≤ Γ (ν + 1) (ν + N ) KN Rν . Γ (ν + 1) (ν + N )

(26.73)

Consequently we ﬁnd R N N −1 f (sω) s ds dω f (0) − N ωN R 0 S N −1 ≤

KN Rν . Γ (ν + 1) (ν + N )

(26.74)

26.3 Multivariate Results That proves

KN Rν B(0,R) f (x) dx . ≤ f (0) − V ol (B (0, R)) Γ (ν + 1) (ν + N )

633

(26.75)

We have established our last result. Theorem 26.23. Let f : B (0, R) → R be a Lebesgue integrable function, not necessarily radial. Assume f (·ω) ∈ AC 1 ([0, R]), R > 0, ∀ω ∈ S N −1 ; 0 ≤ ν < 1, ν f (·ω) ∈ L∞ ([0, R]), ∀ω ∈ S N −1 . and D∗0 Suppose also that ν D∗0 f (tω) ≤ K, (26.76) ∞,(t∈[0,R]) ∀ω ∈ S N −1 , where K > 0. Then KN Rν B(0,R) f (x) dx . ≤ f (0) − V ol (B (0, R)) Γ (ν + 1) (ν + N )

(26.77)

27 Appendix

27.1 Conversion Formulae for Diﬀerent Kinds of Fractional Derivatives Without loss of generality we work on [0, 1] and the anchor point is zero. ν the Riemann–Liouville Let ν > 0, n ∈ / N, n = [ν] . We denote by DR−L ν fractional derivative of order ν > 0, by Dc the Canavati fractional derivative of order ν > 0, and by c ∂ ν /∂xν the Caputo fractional derivative of order ν > 0. Consider

C0n+1 ([0, 1]) = f ∈ C n+1 ([0, 1]) : f (0) = f (0) = · · · f (n) (0) = 0 . By Lemma 6, p. 64 of [101], we obtain that if f ∈ C0n+1 ([0, 1]), then Dcν f (x) exists and is continuous for any x ∈ [0, 1]. We also have that C n+1 ([0, 1]) ⊂ AC n+1 ([0, 1]) . By Lemma 3.7, p. 41 of [134], for f ∈ C n+1 ([0, 1]) we have that c ν ∂ f (x)/∂xν exists and is continuous for any x ∈ [0, 1], and c ∂ ν f (0)/∂xν = 0. ν f (x) By our Corollary 16.9 for the last case, we get equivalently that DR−L n+1 ([0, 1]) all three derivatives exist. exists for any x ∈ [0, 1]. So for f ∈ C0 When f ∈ C0n+1 ([0, 1]), by Lemma 3.3, p. 39 of [134] we get that ν DR−L f (x) =

c ν

∂ f (x) , ∂xν

for any x ∈ [0, 1]. So we conclude that studying conversion formulae among the above three types of fractional derivatives makes perfect sense. G.A. Anastassiou, Fractional Diﬀerentiation Inequalities, 635 c Springer Science+Business Media, LLC 2009 DOI 10.1007/978-0-387-98128-4 27,

636

27. Appendix

We mention again that ν f DR−L

1 (x) = Γ (n + 1 − ν)

Dcν f (x) = and

d dx

n+1

∂ f (x) 1 = ∂xν Γ (n + 1 − ν)

(x − t)n−ν f (t) dt,

(27.1)

0

d 1 Γ (n + 1 − ν) dx

c ν

x

x

(x − t)n−ν f (n) (t) dt,

(27.2)

(x − t)n−ν f (n+1) (t) dt.

(27.3)

0

x 0

Let us assume ﬁrst that f ∈ C0n+1 ([0, 1]). We set

x

y (x) = 0

(x − t)n−1 f (t) dt = (n − 1)!

x x1

xn−1

... 0

0

f (xn ) dxn . . . dx1 ,

(27.4)

0

for any x ∈ [0, 1]. Thus y (n) (x) = f (x), for all x ∈ [0, 1], and y (i) (0) = 0, i = 0, 1, . . . , n. Clearly y ∈ C0n+1 ([0, 1]), so that Dcν y exists and is continuous. We observe that n x d d 1 ν f (x) = (x − t)n−ν f (t) dt DR−L Γ (n + 1 − ν) dx dx 0 =

1 Γ (n + 1 − ν)

n x d d (x − t)n−ν y (n) (t) dt dx dx 0 n d = Dcν y (x). dx

(27.5)

So we have proved Proposition 27.1. For f ∈ C0n+1 ([0, 1]) it holds that ν DR−L f

(x) =

d dx

n

Dcν y (x),

(27.6)

for any x ∈ [0, 1], where y is as in (27.4). Next assume that f ∈ C0n+2 ([0, 1]). We observe that the following function exists and is continuous, Dcν f (x) =

d 1 Γ (n + 1 − ν) dx =

So we have

d dx

c

x

(x − t)n−ν f (n+1) (t) dt

0

∂ ν f (x) . ∂xν

(27.7)

27.1 Conversion Formulae for Diﬀerent Kinds of Fractional Derivatives Proposition 27.2. For f ∈ C0n+2 ([0, 1]) it holds c ν ∂ f (x) Dcν f (x) = , ∂xν for any x ∈ [0, 1], and c ∂ ν f (0)/∂xν = 0. From the last we get x c ν ∂ f (x) = Dcν f (t) dt. ∂xν 0 x Call F = f ; that is, f (x) = 0 F (t) dt. Thus by (27.9) we obtain x c ν x ν ∂ 0 F (t) dt Dc F (t) dt, (x) = ν ∂x 0 for any x ∈ [0, 1] . Again by (27.8) we derive Dcν F

(x) =

c ν

∂

F (t) dt (x) , ∂xν

637

(27.8)

(27.9)

(27.10)

x 0

(27.11)

for any x ∈ [0, 1]; here F ∈ C0n+1 ([0, 1]) . Next we assume that f ∈ C n+1 ([0, 1]) . We put x (x − t)n f (t) dt; g (x) = n! 0

(27.12)

that is, g (n+1) (x) = f (x), for any x ∈ [0, 1], and g ∈ C n+1 ([0, 1]). Therefore there exists c ∂ ν g/∂xν and it is continuous on [0, 1] . We observe that n+1 x d 1 ν f (x) = (x − t)n−ν g (n+1) (t) dt DR−L Γ (n + 1 − ν) dx 0 n+1 c ν ∂ g (x) d = . (27.13) dx ∂xν We have established Proposition 27.3. Let f ∈ C n+1 ([0, 1]) . Then n+1 c ν ∂ g (x) d ν DR−L f (x) = , dx ∂xν

(27.14)

for any x ∈ [0, 1], where g is as in (27.12). Conclusion 27.4. Any two kinds of the above main fractional derivatives of the literature are connected indirectly via the above established formulae (27.6), (27.8), and (27.14); see also our Lemma 16.10, which is Lemma 3.2, p. 39 of [134]. Consequently treating all, but each one separately, regarding establishing fractional diﬀerentiation inequalities in this monograph, it makes perfect sense and is well justiﬁed by their many applications, for example, in diﬀerential equations, and their importance in mathematics.

638

27. Appendix

27.2 Some Basic Fractional Derivatives Here we mention the formulae of some basic functions’ fractional derivatives. The anchor points are again zero. Regarding the Canavati fractional derivative from [101] we get Γ (μ + 1) xμ−ν , 0 < ν ≤ μ. Γ (μ − ν + 1)

Dcν xμ =

(27.15)

Let μ = m an integer and ν = 1/2; we have 1/2 m

x

Dc =

m−

=

1 2

m! xm−(1/2) Γ (m + (1/2))

m! √ xm−(1/2) . m − 32 . . . 12 π

(27.16)

Regarding the Riemann–Liouville fractional derivative, from [333, pp. 309– 310], with function variable t > 0, order α ∈ R, we get α (H (t)) = DR−L

t−α , Γ (1 − α)

(27.17)

where H is the Heavyside function, % α (H DR−L

(t − a)) =

(t−a)−α , Γ(1−α)

and α (δ (t)) = DR−L

0,

t>a 0 ≤ t ≤ a,

t−α−1 , Γ (−α)

(27.18)

(27.19)

where δ is the delta function. We continue with α tν = DR−L

Γ (ν + 1) ν+α , ν > −1, t Γ (ν + 1 − α)

α DR−L eλt = t−α E1,1−α (λt),

(27.20) (27.21)

where Ea,b stands for the two-parameter Mittag–Leﬄer function; see p. 17 of [333], √ α (27.22) DR−L cosh λt = t−α E2,1−α λt2 , α DR−L ln t =

t−α (ln t + Ψ (1) − Ψ (1 − α)), Γ (1 − α)

(27.23)

where Ψ is the digamma function, and α tβ−1 ln t = DR−L

Γ (β) tβ−α−1 (ln t + Ψ (β) − Ψ (β − α)), Re β > 0. Γ (β − α)

(27.24)

27.2 Some Basic Fractional Derivatives

639

Regarding the Caputo fractional derivative from [134, pp. 155–156], with order √ n > 0, n ∈ / N, m := n , i = −1, N0 = N∪ {0}, we get ⎧ 0, if j ∈ N0 and j < m, ⎪ ⎨ Γ(j+1) n j j−n x = x , if j ∈ N0 and j ≥ m, (27.25) D∗0 Γ(j+1−n) ⎪ ⎩ or j ∈ / N and j > m − 1. Next let c > 0, j ∈ R; then n D∗0 (x + c)j = 2 F1

Γ (j + 1) cj−m−1 xm−n Γ (j + 1 − m) Γ (m − n + 1)

1, m − j; m − n + 1; −

x , c

(27.26)

where 2 F1 is the hypergeometric series. For j ∈ R we have n ejx = j m xm−n Em−n+1 (jx), D∗0

(27.27)

where Ea stands for the Mittag–Leﬄer function; see [333, p. 16], and

n D∗0 (sin jx) =

⎧ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎩

j m i(−1)m/2 xm−n 2Γ(m−n+1)

[−1 F1 (1; m − n + 1; ijx) +1 F1 (1; m − n + 1; −ijx)] ;

m even,

j m (−1)(m−1)/2 xm−n [1 F1 (1; m − n + 1; ijx) 2Γ(m−n+1)

+1 F1 (1; m − n + 1; −ijx)] ;

m odd, (27.28)

where 1 F1 is the conﬂuent hypergeometric function, and so on. As we see above, computing the fractional derivatives exactly leads to highly complicated formulae, hard to handle. Therefore in applications they numerically approximate the fractional derivatives, and numerically solve the fractional diﬀerential equations, and so on. Excellent sources for the above are [134, 140] and [333].

References

[1] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions, Dover, New York, 1964. [2] G. Acosta and R. G. Dur´ an, An Optimal Poincar ´e inequality in L for convex domains, Proc. AMS 132 (2003), No. 1, 195–202. [3] R. P. Agarwal, Sharp Opial-type inequalities involving r−derivatives and their applications, Tˆ ohoku Math. J. 47 (1995), 567–593. [4] R. P. Agarwal and P. Y. H. Pang, Opial Inequalities with Applications in Diﬀerential and Diﬀerence Equations, Kluwer Academic Publishers, Dordrecht, Boston, London, 1995. [5] O. P. Agrawal, Response of a diﬀusion-wave system subjected to deterministic and stochastic ﬁelds, ZAMM, Z. Angew, Math. Mech. 83 (2003) No. 4, 265–274. [6] O. P. Agrawal, Fractional variational calculus in terms of Riesz fractional derivatives, J. Phys. A: Math. Theor. 40 (2007) 6287–6303. [7] H. Ahn, Y. Q. Chen and I. Podlubny, Robust stability test of a class of linear time-invariant interval fractional-order system using Lyapunov inequality, Appl. Math. Comput. 187 (2007), No. 1, 27–34. [8] M. A. Al-Bassam, On fractional analysis and its applications. In: H. L. Manocha (ed.), Modern Analysis and its Applications, Prentice Hall of India Ltd., New Delhi 1986, 269–307. 641

642

References

[9] I. Ali, V. Kiryakova and S. L. Kalla, Solutions of fractional multi-order integral and diﬀerential equations using a Poisson-type transform, J. Math. Anal. Appl. 269 (2002), No. 1, 172–199. [10] C. D. Aliprantis and B. Owen, Principles of Real Analysis, Third edition, Academic Press, Inc., San Diego, CA, 1998. [11] B. N. Al-Saqabi and V. K. Tuan, Solution of a fractional diﬀerintegral equation, Integral Transform. Spec. Funct. 4 (1996), No. 4, 321–326. [12] T. J. Anastasio, The fractional-order dynamics of brainstern vestibulooculomotor neurons, Biological Cybernetics, 72 (1994), 69–79. [13] G. A. Anastassiou, Ostrowski type inequalities, Proc. Amer. Math. Soc. 123 (1995), No. 12, 3775–3781. [14] G. A. Anastassiou, Multivariate Ostrowski type inequalities, Acta Math. Hungar. 76 (1997), No. 4, 267–278. [15] G. A. Anastassiou, General fractional Opial type inequalities, Acta Appl. Math. 54 (1998), 303–317. [16] G. A. Anastassiou, Opial type inequalities for linear diﬀerential operators, Math. Inequal. Appl. 1 (1998), No. 2, 193–200. [17] G. A. Anastassiou, Opial type inequalities involving fractional derivatives of functions, Nonlinear Stud. 6 (1999), No. 2, 207–230. [18] G. A. Anastassiou, Opial type inequalities involving functions and their ordinary and fractional derivatives, Comm. Appl. Anal. 4 (2000), No. 4, 547–560. [19] G. A. Anastassiou, Quantitative Approximations, Chapman & Hall/CRC, Boca Raton, New York, 2001. [20] G. A. Anastassiou, Probabilistic Ostrowski type inequalities, Stochastic Anal. Appl. 20 (2002), No. 6, 1177–1189. [21] G. A. Anastassiou, Multivariate Montgomery identities and Ostrowski inequalities, Numer. Funct. Anal. Optim. 23 (2002), No. 3–4, 247–263. [22] G. A. Anastassiou, Univariate Ostrowski inequalities, revisited, Monatsh. Math. 135 (2002), No. 3, 175–189. [23] G. A. Anastassiou, Multidimensional Ostrowski inequalities, revisited, Acta Math. Hungar. 97 (2002), No. 4, 339–353. [24] G. A. Anastassiou, Fractional Ostrowski type inequalities, Commun. Appl. Anal. 7 (2003), No. 2, 203–208. [25] G. A. Anastassiou, Fuzzy Ostrowski type inequalities, Comput. Appl. Math. 22 (2003), No. 2, 279–292.

References

643

[26] G. A. Anastassiou, Opial type inequalities involving fractional derivatives of two functions and applications, Comput. Math. Appl. 48 (2004), 1701–1731. [27] G. A. Anastassiou, Fractional Opial inequalities for several functions with applications, J. Comput. Anal. Appl. 7 (2005), No. 3, 233–259. [28] G. A. Anastassiou, Fractional and other approximation of Csiszar’s f −divergence, (Proc. FAAT 04-Ed. F. Altomare), in Rendiconti Del Circolo Matemitico Di Palermo, Serie II, Suppl. 76 (2005), 197–212. [29] G. A. Anastassiou, Diﬀerence of general integral means, JIPAM. J. Inequal. Pure Appl. Math. 7 (2006), No. 5, Article 185, 13pp. (electronic). [30] G. A. Anastassiou, Multivariate Euler type identity and Ostrowski type inequalities, Numerical analysis and approximation theory, Casa cartii de stiinta, Cluj-Napoca, 2006, 27–54. [31] G. A. Anastassiou, Csiszar and Ostrowski type inequalities via Euler-type and Fink identities, Panamer. Math. J. 16 (2006), No. 2, 77–91. [32] G. A. Anastassiou, Ostrowski type inequalities over balls and shells via a Taylor-Widder formula, JIPAM. J. Inequal. Pure Appl. Math. 8 (2007), No. 4, Article 106, 13pp. [33] G. A. Anastassiou, Opial type inequalities for Widder derivatives, Panamer. Math. J. 17 (2007), No. 4, 59–69. [34] G. A. Anastassiou, Optimal multivariate Ostrowski Euler type inequalities, Stud. Univ. Babes-Bolyai Math. 52 (2007), No. 1, 25–61. [35] G. A. Anastassiou, Multivariate Fink type identitty and multivariate Ostrowski, comparison of means and Gr¨ uss type inequalities, Math. Comput. Modelling 46 (2007), No. 3–4, 351–374. [36] G. A. Anastassiou, High order Ostrowski type inequalities, Appl. Math. Lett. 20 (2007), No. 6, 616–621. [37] G. A. Anastassiou, Opial type inequalities for semigroups, Semigroup Forum 75 (2007), No. 3, 625–634. [38] G. A. Anastassiou, Taylor-Widder representation formulae and Ostrowski, Gr¨ uss, integral means and Csiszar type inequalities, Comput. Math. Appl. 54 (2007), No. 1, 9–23. [39] G. A. Anastassiou, Multivariate Euler type identitty and optimal multivariate Ostrowski type inequalities, Adv. Nonlinear Var. Inequal. 10 (2007), No. 2, 51–104. [40] G. A. Anastassiou, Multivariate Fractional Taylor’s formula, Commun. Appl. Anal. 11 (2007), No. 2, 189–199. [41] G. A. Anastassiou, Hilbert-Pachpatte type general Integral inequalities, Appl. Anal. 86, (2007), No. 8, 945–961.

644

References

[42] G. A. Anastassiou, Hilbert-Pachpatte type general multivariate integral inequalities, Intern. J. Appl. Math. 20, (2007), No. 4, 549–573. [43] G. A. Anastassiou, Multivariate fractional Ostrowski type inequalities, Comput. Math. Appl. 54 (2007), 434–447. [44] G. A. Anastassiou, Fractional multivariate Opial type inequalities over spherical shells, Commun. Appl. Anal. 11 (2007), No. 2, 201–233. [45] G. A. Anastassiou, Riemann-Liouville Fractional Multivariate Opial type inequalities over Spherical Shells, Bulletin of Allahabad Math. Soc., India 23 (2008), 65–140. [46] G. A. Anastassiou, Riemann-Liouville fractional Opial inequalities for several functions with Applications, Commun. Appl. Anal. 12 (2008), No. 4, 377–398. [47] G. A. Anastassiou, Reverse Riemann-Liouville fractional Opial inequalities for several functions, Complex Variables and Elliptic Equations, 53, (2008), No. 6, 523–544. [48] G. A. Anastassiou, Opial type inequalities involving Riemann-Liouville fractional derivatives of two functions with applications, Math. Comput. Model. 48 (2008), 344–374. [49] G. A. Anastassiou, Opial type inequalities for cosine and sine operator functions, Semigroup Forum 76 (2008), No. 1, 149–158. [50] G. A. Anastassiou, Riemann-Liouville Fractional Multivariate Opial Inequalities on Spherical Shell, RGMIA Monographs, 2008, http://rgmia.vu.edu.au/ [51] G. A. Anastassiou, Converse fractional opial inequalities for several functions, CUBO 10, (2008), No. 1, 117–142. [52] G. A. Anastassiou, Caputo fractional Ostrowski type inequalities, J. Appl. Func. Anal. 4 (2009), No. 2, 218–236. [53] G. A. Anastassiou, Multivariate fractional Taylor’s formula revisited, Sarajevo J. of Math. to appear 2009. [54] G. A. Anastassiou, Hilbert-Pachpatte type fractional integral inequalities, Math. Comput. Model. 49 (2009) 1539–1550. [55] G. A. Anastassiou, Riemann-Liouville and caputo fractional approximation of Csiszar’s f −divergence, Sarajevo J. of Math. to appear 2009. [56] G. A. Anastassiou, Fractional Sobolev type inequalities, Appl. Anal. 87 (2008), No. 5, 607–624. [57] G. A. Anastassiou, Fractional Poincar´e type inequalities, Indian J. Math. 50, (2008), No. 3, 533–571.

References

645

[58] G. A. Anastassiou, Caputo fractional multivariate Opial type inequalities on spherical shells, Accepted 2008, Proceedings of AMAT 2008, In “J. Concrete and Appl. Math.”, “Internat. Conf. on Approx. Th. and Appl. Math.”, G. Anastassiou (ed.), held at U. Memphis, Memphis, TN, USA, October 11-13, 2008. [59] G. A. Anastassiou, G. R. Goldstein and J. A. Goldstein, Multidimensional Opial inequalities for functions vanishing at an interior point, Atti Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl. 15 (2004), No. 1, 5–15. [60] G. A. Anastassiou, G. R. Goldstein and J. A. Goldstein, Multidimensional weighted Opial inequalities, Appl. Anal. 85 (2006), No. 5, 579–591. [61] G. A. Anastassiou and J. A. Goldstein, Fractional Opial type inequalities and fractional diﬀerential equations, Result. Math. 41 (2002), 197–212. [62] G. A. Anastassiou and J. A. Goldstein, Ostrowski type inequalities over Euclidean domains, Atti Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl. 18 (2007), No. 3, 305–310. [63] G. A. Anastassiou and J. A. Goldstein, Higher order Ostrowski type inequalities over Euclidean domains, J. Math. Anal. Appl. 337 (2008), 962–968. [64] G. A. Anastassiou, J. Koliha and J. Pecaric, Opial inequalities for fractional derivatives, Dynam Systems Appl. 10, (2001) 395–406. [65] G. A. Anastassiou, J. Koliha and J. Pecaric, Opial type Lp −inequalities for fractional derivatives, Intern. J. Math. & Math. Sci. 31 (2002), No. 2, 85–95. [66] G. A. Anastassiou and J. Pecaric, General weighted Opial inequalities for linear diﬀerential operators, J. Math. Anal. Appl. 239 (1999), No. 2, 402–418. [67] M. Annaby and P. Butzer, Sampling in Paley-Wiener spaces associated with fractional integro-diﬀerential operators, J. Comput. Appl. Math. 171 (2004), No. 1–2, 39–57. [68] T. Apostol, Mathematical Analysis, Addison-Wesley Publ. Co., London, 1969. [69] R. Askey, Inequalities via fractional integration, Lect. Notes Math. 457 (1975), 106–115. [70] M. Axtell and M. E. Bise, Fractional calculus applications in control systems, Proc. of the IEEE 1990 Nat. Aerospace and Electronics Conf., New York, 1990, 563–566. [71] R. L. Bagley and R. A. Calico, Fractional order state equations for the control of viscoelastically damped structures, J. Guid. Contr. Dynam. 14 (1991), 304–311.

646

References

[72] R. L. Bagley and P. J. Torvik, Fractional calculus-a diﬀerent approach to the analysis of viscoelastically damped structures, AIAA J. 21, (1983), No. 5, 741–748. [73] R. L. Bagley and P. J. Torvik, A theoretical basis for the application of fractional calculus to viscoelasticity, J. Rheol 27 (1983), No. 3, 201–210. [74] R. L. Bagley and P. J. Torvik, On the appearence of the fractional derivative in the behavior of real materials, J. Appl. Mech. 51 (1984), 294–298. [75] R. L. Bagley and P. J. Torvik, Fractional calculus in the transient analysis of viscoelastically damped structures, AIAA J. 23, (1985), No. 6, 918–925. [76] D. Bainov and P. Simeonov, Integral Inequalities and Applications, Kluwer Academic Publishers, Dordrecht, Boston, London, 1992. [77] E. Barkai, R. Metzler and J. Klafter, From continuous time random walks to the fractional Fokker-Planck equation, Phys. Rev. E 61 (2000), 132–138. [78] N. S. Barnett, P. Cerone, S. S. Dragomir and A. Sofo, Approximating Csiszar’s f −divergence by the use of Taylor’s formula with integral remainder (paper # 10, pp. 16). In: S. S. Dragomir (ed.), Inequalities for Csiszar f −Divergence in Information Theory, Victoria University, Melbourne, Australia, 2000. On line: http://rgmia.vu.edu.au [79] H. Bauer, Maβ−und Integrations-Theorie, de Gruyter, Berlin, 1990. [80] P. R. Beesack, On an integral inequality of Z. Opial, Trans. Amer. Math. Soc. 104 (1962), 470–475. [81] V. A. Belavin, R. Sh. Nigmatullin, A. I. Miroshnikov and N. K. Luiskaya, Fractional diﬀerentiation of oscillographic polarograms by means of an electrochemical two-terminal network (in Russian), Trudy Kazan Aviatsion Institutel 5 (1964), 144–152. [82] F. Bella, P. F. Biagi, M. Caputo, G. Della Monica, A. Ermini, P. Manjgaladze, V. Sgrigna and D. Zilpimiani, Very slow-moving crustal strain disturbances, Tectonophysics 179 (1990), 131–139. [83] D. A. Benson, S. W. Wheatcraft and M. M. Meerachaert, Application of a fractional advection dispersion equation, Water Resour. Res. 36 (2000), 1403–1412. [84] D. A. Benson, S. W. Wheatcraft and M. M. Meerachaert, The fractionalorder governing equation of L´evy motion, Water Resour. Res. 36 (2000), 1413–1424. [85] H. Beyer and S. Kempﬂe, Deﬁnition of physically consistent damping laws with fractional derivatives, ZAMM. 75 (1995), 623–635. [86] B. Bittner, Irregular sampling based on the fractional Fourier transform, Proceedings of the 1997 workshop on Sampling Theory and Applications, Samp TA-97, Aveiro, Portugal, 1997, 431–436.

References

647

[87] B. Bonilla, J. J. Trujillo and M. Rivero, Fractional order continuity and some properties about integrability and diﬀerentiability of real functions, J. Math. Anal. Appl. 231 (1999), No. 1, 205–212. [88] B. Bonilla, M. Rivero and J. J. Trujillo, On systems of linear fractional diﬀerential equations with constant coeﬃcients, Appl. Math. Comput. 187 (2007), No. 1, 68–78. [89] B. Bonilla, M. Rivero, L. Rodriguez-Germ´ a and J. J. Trujillo, Fractional diﬀerential equations as alternative models to nonlinear diﬀerential equations, Appl. Math. Comput. 187 (2007), No. 1, 79–88. [90] P. L. Butzer, H. Dyckhoﬀ, E. G¨ orlich and R. L. Stens, Best trigonometric approximation, fractional order derivatives and Lipschitz classes, Canad. J. Math. 29 (1977), 781–793. [91] P. L. Butzer, M. Hauss and M. Schmidt, Factorial functions and Stirling numbers of fractional order, Resultate Math. 16 (1989), 16–48. [92] P. L. Butzer, A. A. Kilbas and J. J. Trujillo, Fractional calculus in the Mellin setting and Hadamard-type fractional integrals, J. Math. Anal. Appl. 269 (2002), 1–27. [93] P. L. Butzer, A. A. Kilbas and J. J. Trujillo, Composition of Hadamardtype fractional integration operators and the semigroup property, J. Math. Anal. Appl. 269 (2002), No. 2, 387–400. [94] P. L. Butzer, A. A. Kilbas and J. J. Trujillo, Mellin transform analysis and integration by parts for Hadamard-type fractional integrals, J. Math. Anal. Appl. 270 (2002), No. 1, 1–15. [95] P. L. Butzer, A. A. Kilbas and J. J. Trujillo, Stirling functions of the second kind in the setting of diﬀerence and fractional calculus, Numer. Funct. Anal. Optim. 24 (2003), No. 7–8, 673–711. [96] P. L. Butzer, A. A. Kilbas, L. Rodriguez-Germ´ a and J. J. Trujillo, Stirling functions of ﬁrst kind in the setting of fractional calculus and generalized diﬀerences, J. Diﬀerence Equ. Appl. 13 (2007), No. 8–9, 683–721. [97] P. L. Butzer and R. L. Stens, Fractional Chebyshev operational calculus and best algebraic approximation, Approximation theory, II (Proc. Internat. Sympos., Univ. Texas, Austin, Tex., 1976), Academic Press, New York, 1976, 315–319. [98] P. L. Butzer and R. L. Stens, Chebyshev transform methods in the solution of the fundamental theorem of best algebraic approximation in the fractional case. In: Fourier Analysis and Approximation Theory (Colloquia Math. Soc. J´ anos Bolyai, 19, Proc. Conf. Budapest, August 1976, Eds. G. Alexits-P. Tur´ an) Amsterdam 1978, 191–212. [99] P. L. Butzer and U. Westphal, An access to fractional diﬀerentiation via fractional diﬀerence quotients. In: Fractional Calculus and its Applications. Lecture Notes in Mathematics, Springer, Berlin 1975, 116–145.

648

References

[100] P. L. Butzer and U. Westphal, An introduction to fractional calculus, Ch. 1. In: R. Hilfer (ed.), Applications of Fractional Calculus in Physics, Singapore, World Scientiﬁc, 2000, 1–85. [101] J. A. Canavati, The Riemann-Liouville integral, Nieuw Archief Voor Wiskunde 5 (1987), No. 1, 53–75. [102] M. Caputo, Linear models of dissipation whose Q is almost frequency Independent-II. Geophys, J. R. Astr. Soc. 13 (1967), 529–539. [103] M. Caputo, Elasticit` a e Dissipazione, Zanichelli, Bologna, 1969. [104] M. Caputo, Wave-number-independent rheology in a sphere, Atti Accad. Naz. Lincei Rend. Cl. Sci. Fis. Math. Natur. (8) 81 (1987), No. 2, 175–207. [105] M. Caputo, The rheology of an anelastic medium studied by means of the observation of the splitting of its eigenfrequencies, J. Acoust. Soc. Am. 86 (1989), No. 5, 1984–1987. [106] M. Caputo, The splitting of the free oscillations of the Earth caused by the rheology, Rend. Fis. Acc. Lincei, Ser. 9, 1 (1990), 119–125. [107] M. Caputo, Lectures on Seismology and Rheological Tectonics, Univ. degli studi di Roma “La Sapienza”, 1992–1993. [108] M. Caputo, Free modes splitting and alterations of electrochemically polarizable media, Rend. Fis. Acc. Lincei, Ser. 9, 4 (1993), 89–98. [109] M. Caputo, Mean fractional-order-derivatives diﬀerential equations and ﬁlters, Ann. Univ. Ferrara Sez. VII (N.S.) 41 (1995), 73–84. [110] M. Caputo, 3-dimensional physically consistent diﬀusion in anisotropic media with memory, Atti. Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl. 9 (1998), No. 2, 131–143. [111] M. Caputo, Distributed order diﬀerential equations modelling dielectric induction and diﬀusion, Fract. Calc. Appl. Anal. 4 (2001), No. 4, 421–442. [112] M. Caputo and F. Mainardi, A new dissipation model based on memory mechanism, Pure Appl. Geophys. 91 (1971), No. 8, 134–147. [113] M. Caputo and F. Mainardi, A new dissipation model based on memory mechanism, Pure Appl. Geophys. 91 (1971a), 134–147. [114] M. Caputo and F. Mainardi, Linear models of dissipation in anelastic solid, Rivista del Nuovo Cimento 1 (1971b), 161–198. [115] G. E. Carlson, Investigation of fractional capacitor approximations by means of regular Newton processes, Kansas State University Bulletin, Vol. 48, No. 1, Special report No. 42, 1964. [116] G. E. Carlson and C. A. Halijak, Simulation of the fractional derivative √ √ operator s and the fractional integral operator 1/ s. Kansas State University Bulletin, Vol. 45, No. 7 (1961), 1–22.

References

649

[117] G. E. Carlson and C. A. Halijak, Approximation of fractional capacitors (1/s)1/n by a regular Newton process, IEEE Trans. Circuit Theory, Vol. CT-10, No. 2 (1964), 210–213. [118] A. Carpinteri and F. Mainardi (eds.), Fractals and Fractional Calculus in Continuum Mechanics, Springer Verlag, Vienna-New York, 1997. [119] A. Chaves, A fractional diﬀusion equation to describe L´evy ﬂights, Phys. Lett. A 239 (1998), 13–16. [120] A. V. Chechkin, R. Gorenﬂo, I. M. Sokolov and V. Yu. Gonchar, Distributed order time fractional diﬀusion equation, Fract. Calc. Appl. Anal. 6 (2003), No. 3, 259–279. [121] A. V. Chechkin, R. Gorenﬂo and I. M. Sokolov, Fractional diﬀusion in inhomogeneous media, J. Phys. A 38 (2005), No. 42, L679–L684. [122] M.-P. Chen, H. Irmak and H. M. Srivastava, A certain subclass of analytic functions involving operators of fractional calculus, Comput. Math. Appl. 35 (1998), No. 5, 83–91. [123] M.-P. Chen, H. M. Srivastava and C.-S. Yu, Some operators of fractional calculus and their applications involving a novel class of analytic functions, Appl. Math. Comput. 91 (1998), No. 2–3, 285–296. [124] Y. Q. Chen and K. L. Moore, Discretization schemes for fractional-order diﬀerentiators and integrators, IEEE Trans. On Circuits and Systems-I. Fundamental Theory and Applications 49 (2002), No. 3, 363–367. [125] Y. Chen, B. M. Vinagre and I. Podlubny, Continued fraction expansion approaches to discretizing fractional order derivatives-an expository review, Nonlinear Dynam. 38 (2004), No. 1–4, 155–170. [126] S.-K. Chua and R. L. Wheeden, A note on sharp 1−dimensional Poincar´e inequalities, Proc. AMS 134 (2006), No. 8, 2309–2316. [127] I. Csiszar, Eine Informationstheoretische Ungleichung und ihre Anwendung auf den Beweis der Ergodizit¨ at von Markoﬀschen Ketten, Magyar Trud. Akad. Mat. Kutato Int. K¨ ozl. 8 (1963), 85–108. [128] I. Csiszar, Information-type measures of diﬀerence of probability distributions and indirect observations, Studia Math. Hungarica 2 (1967), 299–318. [129] J. B. Diaz and T. J. Osler, Diﬀerences of fractional order, Math. Comput. 28 (1974), 185–202. [130] K. Diethelm, An algorithm for the numerical solution of diﬀerential equations of fractional order, Elec. Trans. Numer. Anal. 5 (1997), 1–6, ISSN 1068–9613. [131] K. Diethelm, Fractional error constants for quadrature formulas, Approximation theory IX, Vol. I. (Nashville, TN, 1998), 113–118, Innov. Appl. Math., Vanderbilt Univ. Press, Nashville, TN, 1998.

650

References

[132] K. Diethelm, Estimation of Quadrature errors in terms of Caputo-type fractional derivatives, Fract. Calc. Appl. Anal. 2 (1999), No. 3, 313–327. [133] K. Diethelm, Predictor-Corrector Strategies for Single- and Multi-Term Fractional Diﬀerential Equations. In: E. A. Lipitakis (ed.), Proceedings of the 5th Hellenic-European Conference on Computer Mathematics and its Applications, LEA Press, Athens, 2002, 117–122. [134] K. Diethelm, Fractional Diﬀerential Equations, On line: http://www.tu-bs.de/∼diethelm/lehre/f-dgl02/fde-skript.ps.gz,2003. [135] K. Diethelm, Smoothness properties of solutions of Caputo-type fractional diﬀerential equations, Fract. Calc. Appl. Anal. 10 (2007), No. 2, 151–160. [136] K. Diethelm, J. M. Ford, N. J. Ford and M. Weilbeer, Pitfalls in fast numerical solvers for fractional diﬀerential equations, J. Comput. Appl. Math. 186 (2006), No. 2, 482–503. [137] K. Diethelm and N. J. Ford, Analysis of fractional diﬀerential equations, J. Math. Anal. Appl. 265 (2002), No. 2, 229–248. [138] K. Diethelm and N. J. Ford, Multi-order fractional diﬀerential equations and their numerical solution, App. Math. Comput. 154 (2004), No. 3, 621–640. [139] K. Diethelm, N. J. Ford and A. D. Freed, A predictor-corrector approach for the numerical solution of fractional diﬀerential equations, Fractional order calculus and its applications. Nonlinear Dynam. 29 (2002), No. 1–4, 3–22. [140] K. Diethelm, N. J. Ford, A. D. Freed and Y. Luchko, Algorithms for the fractional calculus: A selection of numerical methods, Comput. Meth. Appl. Mech. Engrg. 194 (2005), No. 6–8, 743–773. [141] K. Diethelm and A. D. Freed, On the solution of nonlinear fractional differential equations used in the modeling of viscoplasticity. In: F. Keil, W. Mackens, H. Voβ and J. Werther (eds.), Scientiﬁc Computing in Chemical Engineering II: Computational Fluid Dynamics, Reaction Engineering, and Molecular Properties, Springer, Heidelberg (1999a), 217–224. [142] K. Diethelm and Y. Luchko, Numerical solution of linear multi-term initial value problems of fractional order, J. Comput. Anal. Appl. 6 (2004), No. 3, 243–263. [143] K. Diethelm and G. Walz, Numerical solution of fractional order diﬀerential equations by extrapolation, Numer. Algorithms 16 (1997), 231–253. [144] S. S. Dragomir, On some Gronwall type lemmas, Studia Univ. Babes-Bolyai Math. 33 (1988), 29–36. [145] S. S. Dragomir (ed.), Inequalities for Csiszar f -Divergence in Information Theory, Victoria University, Melbourne, Australia, 2000. On line: http://rgmia.vu.edu.au

References

651

[146] S. S. Dragomir and C. E. M. Pearce, Selected Topics on Hermite-Hadamard Inequalities and Applications, Monograph, Victoria University, Melbourne, Adelaide, Australia, 2000. On line: http://rgmia.vu.edu.au [147] S. S. Dragomir and Y.-H. Kim, Hilbert-Pachpatte type integral inequalities and their improvement, J. Inequal. Pure Appl. Math. 4 (2003), No. 1, Article 16, http://jipam.vu.edu.au/ [148] D. P. Drianov, Equivalence between fractional average modulus of smoothness and fractional K-functional, C. R. Acad. Bulg. Sci. 38 (1985), 1609–1612. [149] M. M. Dzhrbashyan and A. B. Nersesyan, Fractional derivatives and the Cauchy problem for diﬀerential equations of fractional order, Izv. Akademii Nauk Arm. SSR 3 (1968), No. 1, 3–29. [150] E. C. Eckstein, J. A. Goldstein and M. Leggas, The mathematics of suspensions: Kac walks and asymptotic analyticity, Fourth Mississippi State Conference on Diﬀerential Equations and Computational Simulations, Electronic J. Diﬀerential Equations, Conf. 03, 1999, 39–50, 2000. [151] E. C. Eckstein, B. Ma, M. Leggas and J. A. Goldstein, Linking theory and measurements of tracer particle position in suspension ﬂows, 1–8, Proceedings of ASME FEDSM’00 ASME 2000 Fluids Engineering Division Summer Meeting, June 11–15, 2000, Boston, MA. [152] J. T. Edwards, N. J. Ford and A. C. Simpson, The numerical solution of linear multi-term fractional diﬀerential equations: Systems of equations, J. Comput. Appl. Math. 148 (2002), 401–418. [153] A. M. A. El-Sayed, Fractional derivative and fractional diﬀerential equations, Bull Fac. Sci., Alexandria Univ. 28 (1998), 18–22. [154] A. M. A. El-Sayed, Fractional order evolution equations, J. Fract. Calc. 7 (1995), 89–100. [155] A. M. A. El-Sayed, Fractional order diﬀusion-wave equation, Int. J. Theor. Phys. 35 (1996), 311–322. [156] N. Engheta, On fractional calculus and fractional multipoles in electromagnetism, IEEE Trans. on Antennas and Propagations, 44, (1996), No. 4, 554–566. [157] A. Erd´elyi, On fractional integration and its applications to the theory of Hankel transforms, Quart. J. Math. (Oxford), 11 (1940), 293–303. [158] L. Euler, Memoire dans le tome V des Comment, Saint Petersberg Ann´ees, 55, 1730. [159] L. C. Evans, Partial Diﬀerential Equations, Graduate studies in Mathematics Vol. 19, American Math. Soc., Providence, R. I., 1998.

652

References

[160] A. M. Fink, On Opial’s inequality for f (n) , Proc. Amer. Math. Soc. 115 (1992), 177–181. [161] N. J. Ford and A. C. Simpson, The numerical solution of fractional differential equations: Speed versus accuracy, Numer. Algorithms 26 (2001), 333–346. [162] J. B. J. Fourier, Th´eorie analytique de la chaleur, Ocuvres de Fourier, 1, Didot, Paria, 1822, 508. [163] Ch. Friedrich, Rheological material functions for associating comb-shaped or H-shaped polymers: A fractional calculus approach, Psilosophical Magazine Lett. 66, (1992), No. 6, 287–292. [164] L. Gaul, P. Klein and S. Kempﬂe, Damping description involving fractional operators, Mech. Syst. Signal Process. 5 (1991), 81–88. [165] M. Giona, S. Gerbelli and H. E. Roman, Fractional diﬀusion equation and relaxation in complex viscoelastic materials, Physica A 191 (1992), 449–453. [166] W. G. Gl¨ ockle and T. F. Nonnenmacher, Fractional integral operators and Fox functions in the theory of viscoelasticity, Macro-molecules 24 (1991), 6426–6436. [167] W. G. Gl¨ ockle and T. F. Nonnenmacher, A fractional calculus approach to self-similar protein dynamics, Biophysical J. 68 (1995), 46–53. [168] R. Gorenﬂo, Fractional calculus: Some numerical methods, CISM Lecture Notes, Udine, Italy, 1996. [169] R. Gorenﬂo, Fractional calculus: Some numerical methods. In: A. Carpinteri, F. Mainardi (eds.), Fractals and Fractional Calculus in Continuum Mechanics, Springer, Wien, 1997, 277–290. [170] R. Gorenﬂo and E. A. Abdel-Rehim, Discrete models of time-fractional diﬀusion in a potential well, Fract. Calc. Appl. Anal. 8 (2005), No. 2, 173–200. [171] R. Gorenﬂo and E. A. Abdel-Rehim, Convergence of the Gr¨ unwaldLetnikov scheme for time-fractional diﬀusion, J. Comput. Appl. Math. 205 (2007), No. 2, 871–881. [172] R. Gorenﬂo, Y. Luchko and F. Mainardi, Wright functions as scale-invariant solutions of the diﬀusion-wave equation. Higher transcendental functions and their applications, J. Comput. Appl. Math. 118 (2000), No. 1–2, 175–191. [173] R. Gorenﬂo, Y. F. Luchko and S. R. Umarov, On some boundary value problems for pseudo-diﬀerential equations with boundary operators of fractional order, Fract. Calc. Appl. Anal. 3 (2000), No. 4, 453–468.

References

653

[174] R. Gorenﬂo and F. Mainardi, Fractional calculus: Integral and diﬀerential equations of fractional order, Fractals and Fractional Calculus in Continuum Mechanics (Udine, 1996), 223–276, CISM Courses and Lectures, 378, Springer, Vienna, 1997. [175] R. Gorenﬂo and F. Mainardi, Fractional calculus and stable probability distributions, Fourth Meeting on Current Ideas in Mechanics and Related Fields (Krak´ ow, 1997). Arch. Mech. (Arch. Mech. Stos.) 50 (1998), No. 3, 377–388. [176] R. Gorenﬂo and F. Mainardi, Fractional relaxation of distributed order, Complexus mundi, 33–42, World Sci. Publ., Hackensack, NJ, 2006. [177] R. Gorenﬂo, F. Mainardi, D. Moretti, G. Pagnini and P. Paradisi, Fractional diﬀusion: Probability distributions and random walk models, Non extensive thermodynamics and physical applications (Villasimius, 2001), Phys. A 305 (2002), No. 1–2, 106–112. [178] R. Gorenﬂo, F. Mainardi, D. Moretti and P. Paradisi, Time fractional diffusion: A discrete random walk approach, Fractional order calculus and its applications, Nonlinear Dynam. 29 (2002), No. 1–4, 129–143. [179] R. Gorenﬂo, F. Mainardi, E. Scalas and M. Raberto, Fractional calculus and continuous-time ﬁnance III. The diﬀusion limit, Mathematical ﬁnance (Konstanz, 2000), 171–180, Trends Math., Birkh¨ auser, Basel, 2001. [180] R. Gorenﬂo, F. Mainardi and A. Vivoli, Continuous-time random walk and parametric subordination in fractional diﬀusion, Chaos Solitons Fractals 34 (2007), No. 1, 87–103. [181] R. Gorenﬂo, A. Vivoli and F. Mainardi, Discrete and continuous random walk models for space-time fractional diﬀusion, Nonlinear Dynam. 38 (2004), No. 1–4, 101–116. [182] R. Gorenﬂo, A. Vivoli and F. Mainardi, Discrete and continuous random walk models for space-time fractional diﬀusion, J. Math. Sci. (NY) 132 (2006), No. 5, 614–628. [183] I. S. Gradstein and I. M. Ryzhik, Tables of Integrals, Series and Products, Academic Press., New York, 1980. [184] A. Graham, G. W. Scott Blair and R. F. J. Withers, A methodological problem in rheology, Br. J. Philos. Sci. 11 (1961), 265–278. [185] A. P. Grinprimenko and A. A. Kilbas, On compositions of generalized fractional integrals and evaluation of deﬁnite integrals with Gauss hypergeometric functions, J. Math. Res. Exposition 11 (1991), No. 3, 443–446. [186] S. B. Hadid and Yu. F. Luchko, An operational method for solving fractional diﬀerential equations of an arbitrary real order, Panamer. Math. J. 6 (1996), No. 1, 57–73.

654

References

[187] G. D. Handley, J. J. Koliha and J. Pecaric, Hilbert-Pachpatte type integral inequalities for fractional derivatives, Fract. Calc. Appl. Anal. 4 (2001), No. 1, 37–46. [188] G. D. Handley, J. J. Koliha and J. Pecaric, New Hilbert-Pachpatte type integral inequalities, J. Math. Anal. Appl. 257 (2001), 238–250. [189] G. D. Handley, J. J. Koliha and J. Pecaric, Hilbert-Pachpatte type multidimensional integral inequalities, J. Inequal. Pure Appl. Math. 5 (2004), No. 2, Article 34, http://jipam.vu.edu.au/ [190] G. H. Hardy and J. E. Littlewood, Some properties of fractional integral, I. Math. Z., 27 (1928), 565. [191] G. H. Hardy, J. E. Littlewood and G. Polya, Inequalities, Cambridge University Press, Cambridge, UK, 1934. [192] H. Hayakawa, Fractional dynamics in phase ordering processes, Fractals 1 (1993), 947–953. [193] N. Hayek, J. Trujillo, M. Rivero, B. Bonilla and J. C. Moreno, An extension of Picard-Lindel¨ oﬀ theorem to fractional diﬀerential equations, Appl. Anal. 70 (1999), No. 3–4, 347–361. [194] R. Hilfer, Foundations of fractional dynamics, Fractals 3 (1995), 549–556. [195] R. Hilfer, On fractional diﬀusion and its relation with continuous time random walks, In: R. Kutner, A. Pekalski and K. Sznajd-Weron (eds.), Anomalous Diﬀusion: From Basis to Applications, Springer Verlag, Berlin 1999, 77–82. [196] R. Hilfer, Fractional calculus and regular variation in thermodynamics, Applications of Fractional Calculus in Physics, World Sci. Publ., River Edge, NJ, 2000, 429–463. [197] R. Hilfer (ed.), Applications of Fractional Calculus in Physics, World Scientiﬁc, Singapore, 2000. [198] R. Hilfer, Fractional time evolution, Applications of Fractional Calculus in Physics, World Sci. Publ., River Edge, NJ, 2000, 87–130. [199] R. Hilfer, On fractional diﬀusion and continuous time random walks, Phys. A 329 (2003), No. 1–2, 35–40. [200] R. Hilfer, Remarks on fractional time, Time, Quantum and Information, Springer, Berlin, 2004, 235–241. [201] R. Hilfer and L. Anton, Fractional master equations and fractal time random walks, Phys. Rev. E 51 (1995), R848–R851. [202] H. J. Holmgren, Om diﬀerentialkalkylen med indices of hvad nature sam helst, Kgl. Sv. Vetenskapsakademia Handl, 11, 1864, 1.

References

655

[203] T. Y. Hu and K. S. Lau, Fractal dimensions and singularities of the Weierstrass type functions, Trans. Amer. Math. Soc. 335 (1993), No. 2, 649–665. [204] R. N. Kalia, Fractional calculus and bilateral expansions of incomplete beta and elliptic functions, Intern. J. Math. Ed. Sci. Tech. 22 (1991), No. 2, 203–206. [205] R. N. Kalia, Fractional calculus: Its brief history and recent advances, Recent Advances in Fractional Calculus, Global Res. Notes Ser. Math., Global, Sauk Rapids, MN, 1993, 1–30. [206] R. N. Kalia (ed.), Recent Advances in Fractional Calculus, Global Publ. Co., Sauk Rapids, MN, 1993. [207] R. N. Kalia and S. Kalia, Integrals derivable from Feynman integrals, fractional calculus applications to the H-function, and certain Gaussian model in statistical mechanics, Math Sci. Res. Hot-Line 2 (1998), No. 7, 19–24. [208] R. N. Kalia and S. Keith, Fractional calculus and expansions of incomplete gamma functions, Appl. Math. Lett. 3 (1990), No. 1, 19–21. [209] A. A. Kilbas, Asymptotic expansions of fractional integrals and of the solutions of the Euler-Poisson-Darboux equations (Russian), Diﬀer. Uravn. 24 (1988), No. 10, 1764–1778, 1837–1838; Translation in Diﬀer. Equ. 24 (1988), No. 10, 1174–1185. [210] A. A. Kilbas, Asymptotic properties of fractional integrals and their applications, Boundary Value Problems, Special Functions and Fractional Calculus (Russian), Belorus, Gos. Univ., Minsk, 1996, 141–158. [211] A. A. Kilbas, B. Bonilla, Kh. Rodriges, Kh. Trukhillo and M. Rivero, Compositions of fractional integrals and derivatives with a Bessel-type function and the solution of diﬀerential and integral equations (Russian), Dokl. Nats. Akad. Nauk Belarusi 42 (1998), No. 2, 25–29, 123. [212] A. A. Kilbas, V. Bonilla and Kh. Trukhillo, Nonlinear diﬀerential equations of fractional order in the space of integrable functions (Russian), Dokl. Akad. Nauk 374 (2000), No. 4, 445–449. [213] A. A. Kilbas, B. Bonilla and J. J. Trujillo, Existence and uniqueness theorems for nonlinear fractional diﬀerential equations, Demonstratio Math. 33 (2000), No. 3, 583–602. [214] A. A. Kilbas and A. A. Koroleva, Integral transform with the extended generalized Mittag-Leﬄer function, Math. Model. Anal. 11 (2006), No. 2, 173–186. [215] A. A. Kilbas and S. A. Marzan, The Cauchy Problem for Diﬀerential Equations with Fractional Caputo Derivative, Doklady Mathematics 70 (2004), No. 3, 841–845, Translated from Doklady Akademii Nauk 399 (2004), No. 1, 7–11.

656

References

[216] A. A. Kilbas and S. A. Marzan, Nonlinear Diﬀerential Equations with the Caputo Fractional Derivative in the space of Continuously Diﬀerentiable Functions, Diﬀerential Equations 41 (2005), No. 1, 84–89, Translated from Diﬀerentsialnye Uravneniya 41 (2005), No. 1, 82–86. [217] A. A. Kilbas, T. Pierantozzi, J. J. Trujillo and L. Vazquez, On the solution of fractional evolution equations, J. Phys. A: Math. Gen. 37 (2004), 3271–3283. [218] A. A. Kilbas, M. Rivero and J. J. Trujillo, Existence and uniqueness theorems for diﬀerential equations of fractional order in weighted spaces of continuous functions, Fract. Calc. Appl. Anal. 6 (2003), No. 4, 363–399. [219] A. A. Kilbas, L. Rodriguez and J. J. Trujillo, Asymptotic representations for hypergeometric-Bessel type function and fractional integrals, J. Comput. Appl. Math. 149 (2002), No. 2, 469–487. [220] A. A. Kilbas and M. Saigo, On Mittag-Leﬄer type function, fractional calculus operators and solutions of integral equations, Integral Transform. Spec. Funct. 4 (1996), No. 4, 355–370. [221] A. A. Kilbas, H. M. Srivastava and J. J. Trujillo, Fractional diﬀerential equations: An emergent ﬁeld in applied and mathematical sciences, Factorization, Singular Operators and Related Problems (Funchal, 2002), 151–173, Kluwer Acad. Publ., Dordrecht, 2003. [222] A. A. Kilbas, H. M. Srivastava and J. J. Trujillo, Theory and Applications of Fractional Diﬀerential Equations, Elsevier, Amsterdam, Netherlands, 2006. [223] A. A. Kilbas and A. A. Tityura, A Marchaud-Hadamard-type fractional derivative and the inversion of the Hadamard-type fractional integrals (Russian), Dokl. Nats. Akad. Nauk Belarusi 50 (2006), No. 4, 5–10, 125. [224] A. A. Kilbas and J. J. Trujillo, Computation of fractional integrals via functions of hypergeometric and Bessel type, Higher transcendental functions and their applications, J. Comput. Appl. Math. 118 (2000), No. 1–2, 223–239. [225] A. A. Kilbas and J. J. Trujillo, Diﬀerential equations of fractional order: methods, results and problems, I. Appl. Anal. 78 (2001), No. 1–2, 153–192. [226] A. A. Kilbas and J. J. Trujillo, Diﬀerential equations of fractional order: methods, results and problems, II. Appl. Anal. 81 (2002), No. 2, 435–493. [227] A. A. Kilbas, J. J. Trujillo and A. A. Voroshilov, Cauchy-type problem for diﬀusion-wave equation with the Riemann-Liouville partial derivative, Fract. Calc. Appl. Anal. 8 (2005), No. 4, 403–430. [228] Y. C. Kim and H. M. Srivastava, Fractional integral and other linear operators associated with the Gaussian hypergeometric function, Complex Variables Theory Appl. 34 (1997), No. 3, 293–312.

References

657

[229] V. S. Kiryakova, Generalized fractional integral and fractional derivative representations of hypergeometric functions p Fq for p = q or p = q + 1, Constructive Theory of Functions (Varna, 1987), 260–269, Publ. House Bulgar. Acad. Sci., Soﬁa, 1988. [230] V. Kiryakova, Generalized fractional calculus and applications, Pitman Research Notes in Math. Series, 301. Longman Scientiﬁc & Technical, Harlow, copublished in U.S.A with John Wiley & Sons, Inc., New York, 1994. [231] V. S. Kiryakova, Multiple (multiindex) Mittag-Leﬄer functions and relations to generalized fractional calculus, Higher transcendental functions and their applications, J. Comput. Appl. Math. 118 (2000), No. 1–2, 241–259. [232] V. S. Kiryakova and B. Al-Saqabi, Solutions of Erdelyi-Kober fractional integral, diﬀerential and diﬀerintegral equations of second kind, C. R. Acad. Bulgare Sci. 50 (1997), No. 1, 27–30. [233] V. S. Kiryakova and G. P. Ivanov, Fractional calculus and Mittag-Leﬄer functions: Some applications in control theory, Applications of Mathematics in Engineering (Sozopol, 1998), 117–123, Heron Press, Soﬁa, 1999. [234] H. Kober, On fractional integrals and derivatives, Quart. J. Math. (Oxford), 11 (1940), 193–211. [235] R. C. Koeller, Applications of fractional calculus to the theory of viscoelasticity, J. Appl. Mech. 51 (1984), 299–307. [236] R. C. Koeller, Polynomial operators, Stieltjes convolution, and fractional calculus in hereditary mechanics, Acta. Mech. 58 (1986), 251–264. [237] V. Kokilashvili and S. Samko, On Sobolev theorem for Riesz-type potentials in Lebesgue spaces with variable exponent, Z. Anal. Anwendungen 22 (2003), No. 4, 899–910. [238] K. M. Kolwankar and A. D. Gangal, Fractional diﬀerentiability of nowhere diﬀerentiable functions and dimensions, Chaos 6 (1996), No. 4, 505–513. [239] D. Kreider, R. Kuller, D. Ostberg and F. Perkins, An Introduction to Linear Analysis, Addison-Wesley, Reading, MA, 1966. [240] S. Kullback and R. Leibler, On information and suﬃciency, Ann. Math. Statist. 22 (1951), 79–86. [241] S. Kullback, Information Theory and Statistics, Wiley, New York, 1959. [242] B. Kuttner, Some theorems on fractional derivatives, Proc. London Math. Soc., 3, (1953), 480. [243] S. F. Lacroix, Trait´e du Calcul Diﬀerential et du Calcul Integral, Courcier, Paris, 3 (1819), 409–410. [244] J. L. Lagrange, Sur une nouvelle esp´ee de calcul relatif a la diﬀerentiation et a l’ integration des quantit´es variables, Ouvres de Lagrange, GauthierVillars, Paris, 3 (1772), 441.

658

References

[245] J. L. Lavoie, T. J. Osler and R. Tremblay, Fractional derivatives and special functions, SIAM Rev. 18 (1976), 240–268. [246] G. W. Leibnitz, Letter from Hanover, Germany, September 30, 1695, to G. A. L’ Hospital, Leibnizen Mathematische Schriften 2, pp. 301–302. Olms Verlag, Hildesheim, Germany, 1962, First published in 1849. [247] A. V. Letnikov, On the historical development of the theory of diﬀerentiation of an arbitrary order, Mat. Sb. 3 (1868), 85–112 (in Russian). [248] Y. S. Liang and W. Y. Su, Connection between the order of fractional calculus and fractional dimensions of a type of fractal functions, Anal. Theory Appl. 23 (2007), No. 4, 354–362. [249] Y. S. Liang and W. Y. Su, The relations between the fractal dimensions of a type of fractal functions and the order of their fractional calculus, Chaos, Solitons Fractals 34 (2007), 682–692. [250] S.-D. Lin, J.-C. Shyu, K. Nishimoto and H. M. Srivastava, Explicit solutions of some general families of ordinary and partial diﬀerential equations associated with the Bessel equation by means of fractional calculus, J. Fract. Calc. 25 (2004), 33–45. [251] S.-D. Lin and H. M. Srivastava, Some families of the Hurwitz-Lerch zeta functions and associated derivative and other integral representations, Appl. Math. Comput. 154 (2004), No. 3, 725–733. [252] S.-D. Lin, H. M. Srivastava, S.-T. Tu and P.-Y. Yang, Some families of linear ordinary and partial diﬀerential equations solvable by means of fractional calculus, Int. J. Diﬀer. Equ. Appl. 4 (2002), No. 4, 405–421. [253] S.-D. Lin, S.-T. Tu and H. M. Srivastava, Explicit solutions of certain ordinary diﬀerential equations by means of fractional calculus, J. Fract. Calc. 20 (2001), 35–43. [254] S.-D. Lin, S.-T. Tu, H. M. Srivastava and P.-Y. Wang, Some families of multiple inﬁnite sums and associated fractional diﬀerintegral formulas for power and composite functions, J. Fract. Calc. 30 (2006), 45–58. [255] J. Liouville, Memoire sur quelques qu´estions de g´eometrie et de m´ecanique, et sur un noveau genre pour r´espondre ces qu´estions, Jour. Ecole Polytech. 13 (1832), 1–69. [256] P. I. Lizorkin, Bounds for trigonometrical integrals and the Bernstein inequality for fractional derivatives (Russian), Izv. Akad. Nauk SSSR, Ser. Mat. 29 (1965), No. 1, 109–126. [257] E. R. Love, Fractional derivatives of imaginary order, J. London Math. Soc. (2) 3 (1971), 241–259. [258] E. R. Love, Two index laws for fractional integrals and derivatives, J. Australian Math. Soc. 14 (1972), 385–410.

References

659

[259] E. R. Love, Some inequalities for fractional integrals, Linear spaces and approximation (Proc. Conf., Math. Res. Inst., Oberwolfach, 1977), 177–184. Intern. Ser. Numer. Math., Vol. 40, Birkh¨ auser, Basel, 1978. [260] E. R. Love, A third index law for fractional integrals and derivatives, Fractional calculus (Glasgow, 1984), 63–74, Res. Notes in Math., 138, Pitman, Boston, MA, 1985. [261] E. R. Love, Inequalities like Opial’s inequality, Rocznik naukowo-dydaktyczny WSP w krakowie, Pr. Mat. 97 (1985), 109–118. [262] E. R. Love, Lebesgue points of fractional integrals, Real Anal. Exchange 12 (1986/87), No. 1, 327–336. [263] E. R. Love and Y. C. Young, On fractional integration by parts, Proc. Lond. Math. Soc. 44 (1938), 1–35. [264] Ch. Lubich, Discretized fractional calculus, SIAM J. Math. Anal. 17, (1986), No. 3, 704–719. [265] Y. Luchko and R. Gorenﬂo, The initial value problem for some fractional diﬀerential equations with the Caputo derivatives, ftp://ftp.math.fu-berlin.de/pub/math/publ/pre/1998/Pr-A-98-08.ps [266] Y. Luchko and R. Gorenﬂo, An operational method for solving fractional diﬀerential equations with the Caputo derivatives, Acta. Math. Vietnam. 24 (1999), No. 2, 207–233. [267] Yu. F. Luchko and H. M. Srivastava, The exact solution of certain diﬀerential equations of fractional order by using operational calculus, Comput. Math. Appl. 29 (1995), No. 8, 73–85. [268] Y. Luchko and J. J. Trujillo, Caputo-type modiﬁcation of the Erd´elyi-Kober fractional derivative, Fract. Calc. Appl. Anal. 10 (2007), No. 3, 249–267. [269] Yu. F. Luchko and S. B. Yakubovich, Convolutions of the generalized fractional integration operator, Complex Analysis and Generalized Functions (Varna, 1991), 199–211, Publ. House Bulgar. Acad. Sci., Soﬁa, 1993. [270] J. L¨ utzen, Joseph Liouville 1809–1882, Master of Pure and Applied Mathematics, Springer, New York, Berlin, Heidelberg, 1990. [271] J. A. T. Machado, Analysis and design of fractional-order digital control systems, J. Syst. Anal. Model. Simul. 27 (1997), 107–122. [272] F. Mainardi, On the initial value problem for the fractional diﬀusion-wave equation. In: S. Rionero and T. Ruggeri (eds.), Waves and Stability in Continuous Media, World Scientiﬁc, Singapore, 1994, 246–251. [273] F. Mainardi, Fractional relaxation in anelastic solids, J. Alloys Compd., Vol. 211/212 (1994), 534–538.

660

References

[274] F. Mainardi, Fractional diﬀusive waves in viscoelastic solids. In: J. L. Wegner and F. R. Norwood (eds.), Nonlinear Waves in Solids, ASME/AMR, Fairﬁeld NJ, 1995, 93–97. [275] F. Mainardi, Fractional relaxation-oscillation and fractional diﬀusion-wave phenomena, Chaos, Solitons Fractals 7 (1996), 1461–1477. [276] F. Mainardi, The fundamental solutions for the fractional diﬀusion-wave equation, Appl. Math. Lett. 9 (1996), No. 6, 23–28. [277] F. Mainardi, Fractional calculus: Some basic problems in continuum and statistical mechanics, Fractals and Fractional Calculus in Continuum Mechanics (Udine, 1996), 291–348, CISM Courses and Lectures, 378, Springer, Vienna, 1997. [278] F. Mainardi, Applications of fractional calculus in mechanics. In: P. Rusev, I. Dimovski and V. Kiryakova (eds.), Transform Methods and Special Functions, Varna ’96, SCT Publishers, Singapore, 1997. [279] F. Mainardi, Linear viscoelasticity, Acoustic interactions with submerged elastic structures, Part IV, 97–126, Ser. Stab. Vib. Control Syst. Ser. B, 5. World Sci. Publ., River Edge, NJ, 2002. [280] F. Mainardi and R. Gorenﬂo, The Mittag-Leﬄer function in the RiemannLiouville fractional calculus, Boundary value problems, special functions and fractional calculus (Russian) (Minsk, 1996), 215–225, Belorus. Gos. Univ., Minsk. [281] F. Mainardi and R. Gorenﬂo, Fractional calculus: Special functions and applications, Advanced special functions and applications (Melﬁ, 1999), 165–188. Proc. Melﬁ Sch. Adv. Top. Math. Phys., 1, Aracne, Rome, 2000. [282] F. Mainardi, M. Roberto, R. Gorenﬂo and E. Scalas, Fractional calculus and continuous-time ﬁnance II: The waiting-time distribution, Physica A 287 (2000), 468–481. [283] F. Mainardi, R. Gorenﬂo and A. Vivoli, Renewal processes of Mittag-Leﬄer and Wright type, Fract. Calc. Appl. Anal. 8 (2005), No. 1, 7–38. [284] F. Mainardi and G. Pagnini, The Wright functions as solutions of the timefractional diﬀusion equation, Advanced special functions and related topics in diﬀerential equations (Melﬁ, 2001), Appl. Math. Comput. 141 (2003), No. 1, 51–62. [285] F. Mainardi, G. Pagnini and R. Gorenﬂo, Some aspects of fractional diﬀusion equations of single and distributed order, Appl. Math. Comput. 187 (2007), No. 1, 295–305. [286] R. J. H. Marks and M. W. Hall, Diﬀerintegral Interpolation from a Bandlimited Signal’s Samples, IEEE Transact Acoust Speech Signal Process 29 (1981), 872–877.

References

661

[287] C. Martinez, M. Sanz and M. D. Martinez, About fractional integrals in the space of locally integrable functions, J. Math. Anal. Appl. 167 (1992), 111–122. [288] D. Matignon and G. Montseny (eds.), Fractional Diﬀerential Systems: Models, Methods, and Applications, ESAIM Proceedings, Vol. 5, SMAI, Paris (http://www.emath.fr/Maths/Proc/Vol.5/index.htm), 1998. [289] A. C. McBride, Fractional Calculus and Integral Fransforms of Generalalized Functions, Pitman Research Notes in Mathematics, # 31, Pitman, London, 1979. [290] A. C. McBride, Fractional Calculus, New York, Halsted Press, 1986. [291] A. C. McBride and G. F. Roach, Fractional Calculus, Pitman, Research Notes in Mathematics, 138, 1985. [292] M. M. Meerschaert, D. A. Benson and B. Baeumer, Multidimensional advection and fractional dispersion, Phys. Rev. E 59 (1999), 5026–5028. [293] M. M. Meerschaert, D. A. Benson, H. P. Scheﬄer and B. Baeumer, Stochastic solution of space-time fractional diﬀusion equations, Phys. Rev. E 65 (2002), 1103–1106. [294] J. Mikusinski, Operational Calculus, Pergamon Press, New York, 1959. [295] K. Miller and B. Ross, An Introduction to the Fractional Calculus and Fractional Diﬀerential Equations, John Wiley and Sons, Inc., New York, 1993. [296] K. S. Miller, Derivatives of noninteger order, Math. Mag. 68 (1995), 183–192. [297] D. S. Mitrinovi´c, J. E. Peˇcari´c and A. M. Fink, Classical and New Inequalities in Analysis, Kluwer Acad. Publ., Dordrecht, 1993. [298] G. M. Mittag-Leﬄer, Sur la nouvelle fonction Eα (x) , C. R. Acad. Sci. Paris 137 (1903), 554–558. [299] D. Mo, Y. Y. Lin, J. H. Tan, Z. X. Yu, G. Z. Zhou, K. C. Gong, G. P. Zhang and X.-F. He, Ellipsometric spectra and fractional derivative spectrum analysis of polyaniline ﬁlms, Thin Solid Films 234 (1993), 468–470. [300] G. Montseny, J. Audounet and D. Matignon, Fractional integro-diﬀerential boundary control of the Euler-Bernoulli beam, 36th IEEE Conference on Decision and Control, San Diego, California, December 1997, IEEE-CSS, SIAM, 4973–4978. [301] R. R. Nigmatullin, Fractional integral and its physical interpretation, Soviet J. Theor. and Math. Phys. 90, (1992), No. 3, 354–367. [302] K. Nishimoto, Fractional Calculus, New Haven, CT University of New Haven Press, 1989.

662

References

[303] K. Nishimoto (ed.), Fractional Calculus and its Applications, Nihon University, Koriyama, 1990. [304] T. F. Nonnenmacher, Fractional integral and diﬀerential equations for a class of L´evy-type probability densities, J. Phys. A: Math. Gen. 23 (1990), L697–L700. [305] T. F. Nonnenmacher and W. G. Gl¨ ockle, A fractional model for mechanical stress relaxation, Philosophical Magazine Lett. 64, (1991), No. 2, 89–93. [306] T. F. Nonnenmacher and R. Metzler, On the Riemann-Liouville Fractional Calculus and some Recent Applications, Fractals 3 (1995), 557–566. [307] Z. M. Odibat, Rectangular decomposition method for fractional diﬀusionwave equations, Appl. Math. Comput. 179 (2006), 92–97. [308] Z. M. Odibat, Approximations of fractional integrals and Caputo fractional derivatives, Appl. Math. Comput. 178 (2006), 527–533. [309] Z. M. Odibat and N. T. Shawagfeh, Generalized Taylor’s formula, Appl. Math. Comput. 186 (2007), 286–293. [310] K. B. Oldham, The reformulation of an inﬁnite sum via semi-integration, SIAM J. Math. Anal. 14 (1983), No. 5, 974–981. [311] K. B. Oldham and J. Spanier, The Fractional Calculus, Academic Press, New York, 1959. [312] K. B. Oldham and J. Spanier, The replacement of Fick’s law by a formulation involving semidiﬀerentiation, J. Electroanal. Chem. Interfacial Electrochem. 26 (1970), 331–341. [313] K. B. Oldham and J. Spanier, The Fractional Calculus: Integrations and Diﬀerentiations of Arbitrary Order, New York, Academic Press, 1974. [314] K. B. Oldham and J. Spanier, The Fractional Calculus: Theory and Applications of Diﬀerentiation and Integration to Arbitrary Order, Dover Publications, New York, 2006. [315] Z. Opial, Sur une in´egalit´e, Ann. Polon. Math. 8 (1960), 29–32. [316] T. J. Osler, Leibniz rule for fractional derivatives, generalized and an application to inﬁnite series, SIAM J. Appl. Math. 18 (1970), 658–674. [317] T. J. Osler, A further extension of the Leibniz rule to fractional derivatives and its relation to Parseval’s formula, SIAM J. Math. Anal. 3 (1972), 1–16. ¨ [318] A. Ostrowski, Uber die Absolutabweichung einer diﬀerentiebaren Funktion von ihrem Integralmittelwert, Comment. Math. Helv. 10 (1938), 226–227. [319] B. G. Pachpatte, Existence and uniqueness of solutions of higher order hyperbolic partial diﬀerential equations, Chinese J. Math. 17 (1989), 181–189.

References

663

[320] B. G. Pachpatte, Inequalities similar to the integral analogue of Hilbert’s inequalities, Tamkang J. Math. 30 (1999), No. 1, 139–146. [321] B. G. Pachpatte, On Hilbert type inequality in several variables, An. Stiint. Univ. Al. I. Cuza Iasi. Mat. (N.S.), 46 (2000), 245–250. [322] B. G. Pachpatte, On two new multidimensional integral inequalities of the Hilbert type, Tamkang J. Math. 31 (2000), 123–129. [323] P. Y. H. Pang and R. P. Agarwal, On an Opial type inequality due to Fink, J. Math. Anal. Appl. 196 (1995), 748–753. [324] I. Petras and L. Dorcak, The frequency methods for stability investigation of fractional control systems, In SACTA Journal 2 (1999), No. 1–2, 75–85, Durban, South Africa. [325] I. Podlubny, Fractional derivatives: A new stage in process modelling and control, 4th International DAAAM Symposium, Brno, Czech Republic, September 16–18, 1993, 263–264. [326] I. Podlubny, Fractional-Order Systems and Fractional-Order Controllers, Technical Report UEF-03-94, Institute for Experimental Physics, Slovak Acad. Sci, 1994. [327] I. Podlubny, Numerical methods of the fractional calculus, Transactions of the Technical University of Kosice, 4 (1994), No. 3–4, 200–208. [328] I. Podlubny, Numerical solution of initial value problems for ordinary fractional-order diﬀerential equations, Proceedings of the 14th World Congress on Computation and Applied Mathematics, W. F. Ames (ed.), July 11–15, 1994, Atlanta, Georgia, USA, Late Papers volume, pp. 107–111. [329] I. Podlubny, Solution of linear fractional diﬀerential equations with constant coeﬃcients. In: P. Rusev, I. Dimovski and V. Kiryakova (eds.), Transform Methods and Special Functions, SCT Publishers, Singapore, 1995, pp. 217–228. [330] I. Podlubny, Numerical solution of ordinary fractional diﬀerential equations by the fractional diﬀerence method, Advances in Diﬀerence Equations (Veszpr´em, 1995), 507–515, Gordon and Breach, Amsterdam, 1997. [331] I. Podlubny, Riesz potential and Riemann-Liouville fractional integrals and derivatives of Jacobi polynomials, Appl. Math. Lett. 10, (1997), No. 1, 103–108. [332] I. Podlubny, Application of orthogonal polynomials to solution of fractional integral equations, Transform Methods & Special Functions, Varna ’96, 350–359, Bulgarian Acad. Sci., Soﬁa, 1998. [333] I. Podlubny, Fractional diﬀerential equations, An introduction to fractional derivatives, fractional diﬀerential equations, to methods of their solution and some of their applications, Mathematics in Science and Engineering, 198. Academic Press, Inc., San Diego, CA, 1999.

664

References

[334] I. Podlubny, Matrix approach to discrete fractional calculus, Fract. Calc. Appl. Anal. 3 (2000), No. 4, 359–386. [335] I. Podlubny, Geometric and physical interpretation of fractional integration and fractional diﬀerentiation, Dedicated to the 60th anniversary of Prof. Francesco Mainardi, Fract. Calc. Appl. Anal. 5 (2002), No. 4, 367–386. [336] I. Podlubny and I. Kostial, Fractional derivative based process models and their applications, 4th International DAAAM Symposium, Brno, Czech Republic, September 16–18, 1993, 265–266. [337] I. Podlubny and J. Misanek, The use of fractional derivatives for solution of heat conduction problems, Proceedings of the 9th Conference on Process Control, Tatranske Matliare, May 1993, STU Bratislava, 270–273. [338] I. Podlubny, L. Dorcak and J. Misanek, Application of Fractional-Order Derivatives to Calculation of Heat Load Intensity Change in Blast Furnace Walls, Transact. Tech. Univ. Kodice 5 (1995), 137–144. [339] V. G. Ponomarenko, Modulus of smoothness of fractional order and the best approximation in Lp , 1 < p < ∞ (Russian); In: Constructive Function Theory ’81, (Proc. Conf. Varna, June 1981), Publ. House Bulg. Acad. Sci., Soﬁa 1983, 129–133. [340] J. K. Prajapat, R. K. Raina and H. M. Srivastava, Some inclusion properties for certain subclasses of strongly starlike and strongly convex functions involving a family of fractional integral operators, Integral Transform. Spec. Funct. 18 (2007), No. 9–10, 639–651. [341] A. I. Priesto, S. Salinas de Romero and H. M. Srivastava, Some fractionalcalculus results involving the generalized Lommel-Wright and related functions, Appl. Math. Lett. 20 (2007), No. 1, 17–22. [342] A. V. Pskhu, Solution of a Boundary Value Problem for a fractional Partial diﬀerential equation, Diﬀerential Equations 39 2003, No. 8, 1150–1158, Translated from Diﬀerentsial’nye Uravneniya 39 2003, No. 8, 1092–1099. [343] H. Rafeiro and S. Samko, On a class of fractional type integral equations in variable exponent spaces, Fract. Calc. Appl. Anal. 10 (2007), No. 4, 399–421. [344] R. K. Raina and H. M. Srivastava, A certain subclass of analytic functions associated with operators of fractional calculus, Comput. Math. Appl. 32 (1996), No. 7, 13–19. [345] A. R´enyi, On measures of entropy and information. In Proceedings of the 4th Berkeley Symposium on Mathematical Statistic and Probability, I, Berkekey, CA, 1960, 547–561. [346] B. Riemann, Versuch einer allgemeinen Auﬀassung der Integration und Diﬀerentiation, Gesammelte Mathematische Werke und Wissenschaftlicher Nachlass. Teubner, Leipzig 1876 (Dover, New York, 1953), 331–344.

References

665

[347] M. Riesz, L’int´egrale de Riemann-Liouville et le probl`eme de Cauchy, Acta. Math. 81 (1949), 1–223. [348] F. Riewe, Mechanics with fractional derivatives, Phys. Rev. E 55 (1997), No. 3, 3581–3592. [349] B. Ross, Fractional Calculus and its Applications, Springer-Verlag, Berlin, New York, 1975. [350] B. Ross, Fractional calculus: An historical apologia for the development of a calculus using diﬀerentiation and antidiﬀerentiation of non integral orders, Mathematics Magazine 50 (1977), No. 3, 115–122. [351] B. Ross, The development of fractional calculus 1695–1900, Historia Math. 4 (1977), 75–89. [352] B. Ross, S. G. Samko, E. R. Love, Functions that have no ﬁrst order derivative might have fractional derivatives of all orders less than one, Real Anal. Exchange 20 (1994/95), No. 1, 140–157. [353] Yu. A. Rossikhin and M. V. Shitikova, Applications of fractional calculus to dynamic problems of linear and nonlinear hereditary mechanics of solids, Appl. Mech. Rev. 50 (1997), No. 1, 15–67. [354] H. L. Royden, Real Analysis, Second edition, Macmillan, New York, 1968. [355] B. Rubin, Fractional Integrals and Potentials, Pitman Monographs and Surveys in Pure Appl. Math. 82, Longman, Harlow, 1996. [356] W. Rudin, Real and Complex Analysis, International Student Edition, Mc Graw Hill, London, New York, 1970. [357] Th. Runst and W. Sickel, Sobolev spaces of fractional orders, Nemytskij Operators, and Nonlinear Partial Diﬀerential Equations, De Gruyter, Berlin, 1996. [358] R. S. Rutman and R. Gorenﬂo, Simulation and inversion of fractional integration: A systems theory approach, Inverse Problems and Applications to Geophysics, Industry, Medicine and Technology (Ho Chi Minh City, 1995), 141–148, Publ. HoChiMinh City Math. Soc., 2, HoChiMinh City Math. Soc., Ho Chi Minh City, 1995. [359] S. Samadi, M. O. Ahmad and M. N. S. Swamy, Exact fractional-order diﬀerentiators for polynomial signals, IEEE Signal Process. Lett. 11 (2004), No. 6, 529–532. [360] S. G. Samko, The coincidence of Gr¨ u nwald-Letnikov diﬀerentiation with other forms of fractional diﬀerentiation, The periodic and non-periodic cases (Russian). In: Reports of the Extended Session of the Seminar of the I. N. Vekua Inst. Appl. Math., Tbilisi 1985. Tbiliss. Gos. Univ. 1. [361] S. G. Samko, Fractional integration and diﬀerentiation of variable order, Anal. Math. 21 (1995), No. 3, 213–236.

666

References

[362] S. Samko, Hardy-Littlewood-Stein-Weiss inequality in the Lebesgue spaces with variable exponent, Fract. Calc. Appl. Anal. 6 (2003), No. 4, 421–440. [363] S. G. Samko, Fractional Weyl-Riesz integrodiﬀerentiation of periodic functions of two variables via the periodization of the Riesz kernel, Appl. Anal. 82 (2003), No. 3, 269–299. [364] S. Samko, On inversion of fractional spherical potentials by spherical hypersingular operators, Singular integral operators, factorization and applications, 357–368. Oper. Theory Adv. Appl. 142, Birkh¨ auser, Basel, 2003. [365] S. G. Samko, A. A. Kilbas and O. I. Marichev, Fractional Integrals and Derivatives and Some of their Applications, Nauka i technika, Minsk, 1987. [366] S. G. Samko, A. A. Kilbas and O. I. Marichev, Fractional integrals and derivatives, Theory and applications, Edited and with a foreword by S. M. Nikolskii, Translated from the 1987 Russian origina,. Revised by the authors, Gordon and Breach Science Publishers, Yverdon, 1993. [367] S. G. Samko and A. Ya. Yakubov, Zygmund estimate for moduli of continuity of fractional order of a conjugate function (Russian), Izv. Vyssh. Uchebn. Zaved. Mat. 77 (1985), No. 12, 49–53. [368] S. G. Samko and A. Ya. Yakubov, A Zygmund estimate for hypersingular integrals in the case of moduli of continuity of fractional order (Russian), Izv. Severo-Kavkaz. Nauchn. Tsentra Vyssh. Shkoly Estestv. Nauk. 140 (1986), No. 3, 42–47. [369] R. K. Saxena, On fractional integration operators, Math. Z. 96 (1967), 288–291. [370] R. K. Saxena, S. L. Kalla and V. S. Kiryakova, Relations connecting multiindex Mittag-Leﬄer functions and Riemann-Liouville fractional calculus, Algebras Groups Geom. 20 (2003), No. 4, 363–385. [371] T. Sekine, K. Tsurumi, S. Owa and H. M. Srivastava, Integral means inequalities for fractional derivatives of some general subclasses of analytic functions, Inequalities in univalent function theory and their applications (Japanese) (Kyoto, 2002) Surikaisekikenkyusho Kokyuroku No. 1276 (2002), 79–88. [372] E. Scalas, R. Gorenﬂo and F. Mainardi, Fractional calculus and continuoustime ﬁnance, Phys. A 284 (2000), No. 1–4, 376–384. [373] E. Scalas, R. Gorenﬂo, F. Mainardi and M. Raberto, Revisiting the derivation of the fractional diﬀusion equation, Scaling and disordered systems (Paris, 2000), Fractals 11 (2003), Suppl. 281–289. [374] W. R. Schneider and W. Wyss, Fractional diﬀusion and wave equations, J. Math. Phys. 30 (1989), 134–144. [375] V. B. Shakhmurov and A. I. Zayed, Fractional Wigner distribution and ambiguity functions, Fract. Calc. Appl. Anal. 6 (2003), No. 4, 473–490.

References

667

[376] W. Smit and H. de Vries, Rheological models containing fractional derivatives, Rheologica Acta 9 (1970), 525–534. [377] W. Specht, Zur Theorie der elementaren Mittel, Math. Z. 74 (1960), 91–98. [378] H. M. Srivastava, Fractional calculus and its applications, Cubo Mat. Educ. 5 (2003), No. 1, 33–48. [379] H. M. Srivastava and M. K. Aouf, A certain fractional derivative operator and its applications to a new class of analytic and multivalent functions with negative coeﬃcients II, J. Math. Anal. Appl. 192 (1995), No. 3, 673–688. [380] H. M. Srivastava, S.-D. Lin, Y.-T. Chao, P.-Y. Wang, Explicit solutions of a certain class of diﬀerential equations by means of fractional calculus, Russ J. Math. Phys. 14 (2007), No. 3, 357–365. [381] H. M. Srivastava, A. K. Mishra and M. K. Das, A class of parabolic starlike functions deﬁned by means of a certain fractional derivative operator, Fract. Calc. Appl. Anal. 6 (2003), No. 3, 281–298. [382] H. M. Srivastava and S. Owa (eds.), Univalent Functions, Fractional Calculus, and their Applications, Halsted (Horwood), Chichester, Wiley, New York, 1989. [383] D. Stroock, A Concise Introduction to the Theory of Integration, Third Edition, Birk¨ auser, Boston, Basel, Berlin, 1999. [384] R. Taberski, Contribution to fractional calculus and exponential approximation, Funct. Approx. Comment. Math. 15 (1986), 81–106. [385] F. B. Tatom, The relationship between fractional calculus and fractals, Factals 3 (1995), No. 1, 217–229. [386] J. J. Trujillo, M. Rivero and B. Bonilla, On a new generalized Taylor’s formula, Transform methods & spectral functions, Varna ’96, 508–516. Bulgarian Acad. Sci., Soﬁa, 1998. [387] J. J. Trujillo, M. Rivero and B. Bonilla, On a Riemann-Liouville generalized Taylor’s formula, J. Math. Anal. Appl. 231 (1999), 255–265. [388] C. C. Tseng, Design of fractional order digital FIR diﬀerentiators, IEEE Signal Process. Lett. 8 (2001), 77–79. [389] C. C. Tseng, S. C. Pei and S. C. Hsia, Computation of fractional derivatives using Fourier transform and digital FIR diﬀerentiator, Signal Process. 80 (2000), 151–159. [390] S.-T. Tu, T.-C. Wu and H. M. Srivastava, Comommutativity of the Leibniz rules in fractional calculus, Comput. Math. Appl. 40 (2000), No. 2–3, 303–312.

668

References

[391] V. K. Tuan and R. Gorenﬂo, The Grunwald-Letnikov diﬀerence operator and regularization of the Weyl fractional diﬀerentiation, Z. Anal. Anwendungen 13 (1994), 537–545. [392] V. K. Tuan and R. Gorenﬂo, On the regularization of fractional diﬀerentiation of arbitrary positive order, Numer. Funct. Anal. Optim. 15 (1994), No. 5–6, 695–711. [393] V. K. Tuan and R. Gorenﬂo, Extrapolation to the limit for the numerical fractional diﬀerentiation, Z. angew. Math. Mech. 75 (1995), 646–648. [394] V. K. Tuan, R. K. Raina and M. Saigo, Multidimensional fractional calculus operators involving the Gauss hypergeometric function, Int. J. Math. Sci. 5 (1996), No. 2, 141–160. [395] V. K. Tuan and M. Saigo, Multidimensional modiﬁed fractional calculus operators, Math. Nachr. 161 (1993), 253–270. [396] M. Unser and T. Blu, Fractional splines and wavelets, SIAM Review 42 (2000), No. 1, 43–67. [397] B. M. Vinagre and Y. Q. Chen, Lecture notes on fractional calculus applications in automatic control and robotics, The 41st IEEE CDC2002 Tutorial Workshop No. 2, B. M. Vinagre and Y. Q. Chen (eds.), retrieved from http://mechatronics.ece.usu.edu/foc/cdc02 tw2 ln.pdf, Las Vegas, Nevada, 2002, 1–310. [398] B. M. Vinagre, L. Podlubny, A. Hernandez and V. Feliu, Some approximations of fractional order operators used in control theory and applications, Fract. Calc. Appl. Anal. 3 (2000), No. 3, 231–248. [399] A. A. Voroshilov and A. A. Kilbas, The Cauchy problem for diﬀusion-wave equation with the Caputo partial derivative (Russian), Diﬀer. Uravn. 42 (2006), No. 5, 599–609, 717. Translation in Diﬀer. Equ. 42 (2006) No. 5, 638–649. [400] P.-Y. Wang, S.-D. Lin and H. M. Srivastava, Remarks on a simple fractional-calculus approach to the solutions of the Bessel diﬀerential equation of the general order and some of its applications, Comput. Math. Appl. 51 (2006), No. 1, 105–114. [401] B. J. West, P. Grigolini, R. Metzler and T. F. Nonnenmacher, Fractional diﬀusion and L´evy stable processes, Physical Review E, 55 (1997), 99–106. [402] U. Westphal, An approach to fractional powers of operators via fractional diﬀerences, Proc. London Math. Soc. (3) 29 (1974), 557–576. [403] H. Weyl, Bemerkungen zum Begriﬀ des Diﬀerentialquotienten gebrochener Ordnung. Vierteljschr. Naturforsch. Gesellsch., Z¨ urich, 62, 1917, 296. [404] E. T. Whittaker and G. N. Watson, A Course in Modern Analysis, Cambridge University Press, Cambridge, UK, 1927.

References

669

[405] D. V. Widder, A generalization of Taylor’s series, Trans. Am. Math. Soc. 30 (1928), No. 1, 126–154. [406] D. Willett, The existence-uniqueness theorem for an nth order linear ordinary diﬀerential equation, Amer. Math. Monthly 75 (1968), 174–178. [407] W. Wyss, The fractional diﬀusion equation, J. Math. Phys. 27, (1986), No. 11, 2782–2785. [408] K. Yao, W. Y. Su and S. P. Zhou, On the Fractional Calculus of a Type of Weierstrass Function, Chinese Annals of Mathematics 25:A (2004), 711–716. [409] K. Yao, W. Y. Su and S. P. Zhou, On the connection between the order of fractional calculus and the dimensions of a fractal function, Chaos, Solitons Fractals. 23 (2005), 621–629. [410] M. Z¨ ahle and H. Ziezold, Fractional derivatives of Weierstrass-type functions, J. Computat. Appl. Math., 76 (1996), 265–275. [411] A. I. Zayed, Fractional fourier transform of generalized functions, Integral Transform. Spec. Funct. 7 (1998), No. 3–4, 299–312. [412] A. I. Zayed, A class of fractional integral transforms: a generalizaton of the fractional Fourier transform, IEEE Trans. Signal Process. 50 (2002), No. 3, 619–627.

List of Symbols

[·] , integral part of a number, 7 Jν , 7 Cν , 7 Jνx0 , 8 Cxν0 , 8 Dν , 8 Dxν 0 , 8 I ν , 42 Daν , 42 AC m , 42 AC, 42 C m , 42 L∞ (0, x), 42 I −ν , 42 D−ν , 42 I ν (L1 (0, x)), 42 L1 (0, x), 54 Jα , 108 Dα , 108 L1 (0, x), 108 R+ , nonnegative real numbers, 181 Jv , 206 C v , 206 Dv , 206 671

672

List of Symbols

Cxv0 , 207 Dxv 0 , 207 Dav , 251 L∞ (a, w), 252 D0ν , 258 I, 269 Γ, gamma function, 281 AC n , 269 ν , 270 D∗a n C (Q), 270 γ /∂rγ , 287 ∂R 1 B (0, R), 280 S N −1 , 280 J0 , 513 J−α , 513 B[R1 ,R2 ] , Borel σ-algebra on [R1 , R2 ], 329 ν >α D , 393

Tn−1 , Taylor polynomial of degree n − 1, 393 · , ceiling of the number, 393 RN , N –dimensional Euclidean space, 320 BRN , Borel σ-algebra on RN , 321 ∇, gradient, 445 Lp , 445 A, 280 W 1,p , 483 Δα , 513 W, Wronskian, 517 Wi , 517 C (n−1,m−1) (I × J), 523 Z+ , nonnegative integers, 524 β Δα s1 Δs2 , 539 α+β Δs1 s2 , 539 Γf (μ1 , μ2 ), 563 C0n+1 , 635 ν , 636 DR−L Dcν , 636 c ν ∂ f /∂xν , 636 Ea,b , 638 Ψ, 638 2 F1 , 638 Ea , 639 1 F1 , 639

Index

absolutely continuous function, 42 absolutely continuous measures, 563 anchor point, 42 attained inequality, 593 ball, 320 Borel measurable function, 321 Borel set, 320 Canavati ν-fractional derivative, 8 Caputo fractional derivative, 270 Caputo fractional mixed (iterated) partial derivative, 553 Caputo fractional PDE, 469 Caputo radial fractional derivative, 421 Carath´eodory function, 424 ceiling of the number, 270 cell, 524 chain rule, 268 Chebychev’s inequality, 64 compact and convex subset, 259 conﬂuent hypergeometric function, 639 continuous function, 7 converse, 205 Csiszar’s f -divergence, 563 delta function, 638 digamma function, 638 673

674

Index

dominated convergence theorem, 288 essential supremum, 45 Euclidean norm, 320 fractional bivariate Taylor formula, 274 fractional diﬀerential equation, 23 fractional mean Poincar´e inequality, 473 fractional mixed partial derivative, 281, 283 fractional partial derivative, 281 fractional Taylor formula, 8 gamma function, 7 generalized Riemann–Liouville integral, 8 generalized ν-fractional derivative, 24 gradient, 2 Heavyside function, 638 Hilbert inequality, 3 Hilbert–Pachpatte inequality, 506 Hilbert–Pachpatte multidimensional inequality, 523 H¨ older inequality, 12 hypergeometric series, 639 information for discrimination, 563 information gain, 563 initial value problem, 20 integrable fractional derivative, 323 integral part, 7 L∞ fractional derivative, 108 Lebesgue integrable function, 322 Lebesgue measurable function, 322 Lebesgue measure, 320 Lipschitz condition, 18 mean Sobolev-type inequality, 493 Mittag–Leﬄer two-parameters function, 638 Mittag–Leﬄer function, 639 mixed Canavati fractional partial derivative, 281 multivariate Caputo fractional Taylor formula, 276 multivariate fractional Taylor formula, 257, 264 Opial inequality, 2 ordinary fractional diﬀerential equation, 503 Ostrowski inequality, 3 PDE initial value problem, 469 Poincar´e inequality, 2 polar coordinates, 280 probability measure, 563 radial fractional derivative, 466 radial, 604 representation theorem, 396 Riemann–Liouville fractional derivative, 42

Index Riemann–Liouville fractional integral, 42 Riemann–Liouville fractional mixed (iterated) partial derivative, 539 Riemann–Liouville integral, 7 Riemann–Liouville radial fractional derivative, 330 sharp inequality, 593 Sobolev inequality, 483 sphere, 320 spherical shell, 320 strictly convex, 563 surface measure, 419 Taylor formula, 8 Taylor–Caputo formula, 270 total variation, 565 uniqueness of solution, 97 upper bound, 102 volume, 280 weak fractional derivative, 317 weight function, 109 Widder–Taylor formula, 517 Wronskian, 516 χ2 -divergence, 565

675