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!
uniformly with respect to ~ , Eh Il when I3 = exp(ihS),one dyd< 2 0, V w E L (BnxIRn), where, as usual, < , > stands for the inner product in Cp. We relate w(y,<) and u by setting ,
-
finds, using (2.8.49), that
IuliS)
,E,h 5
where the constant
C
CI
jlul I
( S )
.E,h'
V u E
1 H ( S ) ,E,h(%,+)
does not depend on ( ~ , h ) E Il
5 c inf
IUITs),E,h
1
and the extension 1. EO rho
The last inequality yields:
1 llul I ( s ) ,E,h = c I'I
'ls)
,E,h'
V (E,h) E Jl
1 V u E H (s),~,h(%,+)
EO,ho'
.
One uses the open mapping theorem and its corollary (see, for instance, [Rud, p.49, Corollary 2.12 (d)1 for showing that 1 I V u E H ( s ) , ~ , h ( ~ h , +' ) (Erh) 'Eo,h0
V h
1 jul I (s).E,h +
We emphasize, one more time, that 'lI is an orthogonal projector, 2 1 h 2 1
> 0, in L (T ) onto the functions in L (T ) which can be extended
1 analytically on the exterior of T1 in C , including the infinity: of course, II- is an orthogonal projector in L2 (T1 ) , V h > 0, onto the functions in h 1 L2(T1), which can be extended analytically on the interior of T 1 Let U = (x+,x-) C R be a finite interval and let Uh = U n % . We
.
assume that x+ E Uh.
-
Consider first the case s2 L 0, covering of
s3 L
by two open intervals U
0 integer. Let
and
U-
the corresponding partition of unity, i.e. X? E X,(X)+X_(X)
: 1,
v
x E
5,
U+ U U-
be a
and denote by {X+(x) ,X- (x)1 C
m
o (U* )
and
X+(X+) = 0. - - = 1, X+(X-) - +
2 . Sobolev Spaces of Vedorial Order
210
For a meshfunction u(E,h,-)
:
Uh
+
, denote
C, ( ~ , h )E JIE
0rh0
U+(E,h,X)
=
(X+u)(E,h,x++x), u-(E,~,x)= (X-
and define the norm of u of order s E R
[
where
1 - 1 1' ( s ) ,E,h are
3
,
U)
(E,h,x--x)
as follows
the norms (2.8.49).
Definition 2.8.24.
The space
H (s).E,h(Uh ) is d e f i n e d t o c o n s i s t o f a l l meshfunctions
n o m s (2.8.50)
.
finite f o r each ( ~ , h )E Il
The space
H
u with
€Ofh0 (u) c o n s i s t s o f a l l meshfunctions u E
(S)
H
(u
(s),E,h h
)
such
that
One defines the quotient-space H where
Ho( S ) (U)
' ''
(S)
is again the subspace of
functions u whose norms
For functions Q(e,h,-)
:
( S ),5
,h,Uh
aUh
C
+
(u)
=
ff
H ( S ) (U),
( S )
(U)/Hys) (U) as previously,
consisting of all mesh-
vanish When (E,h) -t ( 0 , O ) .
the norms of order 1 E IR are
introduced as previously
H1 , ~ ,(aU h h ) the family of spaces of functions Q with norms finite, V (&,h) E JIE
We denote by (2.8.52)
0gho
Further, introducing the norm
H1(aU) the space of Q with norm previously H (au) stands for the quotient 1 denote by
(2.8.53)
space
the subspace of functions Q whose norms (2.8.52)
finite, while, as
0 H1(au)/Hl(au)
with
0 H,(au)
vanish, when (E,h) + ( 0 , O ) .
Remark 2.8.25. The norms, defined by (2.8.50)
depend on the partition of unity {X+,X-}.
However, it can be shown that the norms, corresponding to different choice of partition of unity are all equivalent uniformly with respect to
, so that they define the same topology in the corresponding
(E.h) E Il spaces.
'0tho
Notes
21 1
Notes Weighted spaces with a small parameter of vectorial order (0,k ,k ) with 1 2 integer k l 2 0, k 2 0, were already implicitly present in the work by 2 Vishik and Lyusternik 1 1 1 , where they appear through quadratic forms related to singularly perturbed strongly elliptic operators with homogeneous Dirichlet boundary conditions (see also Huet [l]) . This kind of spaces is also useful Lions [l]).
in singularly perturbed control problems, (see
Spaces of order (0,sl,s2)with s1 E IR,
s2
E R , have been
introduced in Demidov [l] and used for the investigation of the first boundary value problem for some classes of elliptic pseudodifferential singular perturbations
.
HBlder type spaces with a small parameter have been used in Fife [11 for establishing one-sided a priori estimates for elliptic singular perturbations with constant coefficients. 3 Sobolev type weighted spaces of vectorial order s = ( s 1 ’ s 2 ’ s 3 ) E R have been introduced in Frank [221 in order to establish two-sided a priori estimates (uniform with respect to the small parameter) for the solutions of coercive (elliptic up to the boundary) singular perturbations. Results concerning equivalence of norms in spaces of vectorial order (Lemma 2.4.2) have been obtained in Frank and Wendt [lo], as well as theones concerning the extension by zero (Theorem 2.6.6.)
(see also Frank and Heijstek [l]).
Sobolev type meshfunction spaces of any vectorial order have been introduced in Frank 171 in order to establish two-sided a priori estimates (uniform with respect to the mesh-size) for elliptic and coercive finite difference operators (Frank [8,9,11,13], see a l s o Thomee and Westergren [l], where meshfunction spaces have been used in order to get interior regularity results for a subclass of elliptic finite difference operators). Sobolev type spaces of meshfunctions, which depend on two small parameters, have been introduced in Frank [16,21,23] and ussd in order to establish two-sided a priori estimates for solutions of elliptic finite difference equations, which approximate elliptic and coercive singular perturbations (Frank [15,19,22]).
This Page Intentionally Left Blank
CHAPTER 3 SINGULAR PERTURBATIONS ON SMOOTH MANIFOLDS WITHOUT BOUNDARY
Several classes of pseudodifferential and difference singular perturbations are introduced here and the elliptic theory of these parameter dependent operators developed. A special attention is given to parameter dependent hyperbolic pseudodifferential and difference singular perturbations and corresponding classes of Fourier integral operators. 3.1.
Singular perturbations with constant symbols
With any differential operator Q ( D ) , D
=
(D1,
...,D
),
Dk = -ia/axk, with
constant coefficients one can associate the polynomial Q(<), called also symbol, so that
where F
and F - l
X-tS
are the direct and inverse Fourier transforms defined
S’X
i n Chapter 1.
The splitting of Q ( 5 ) intci a sum of homogeneous polynomials Q ( 5 ) of k degree k,
with m
=
deg Q ( E ; ) the degree of Q , is a natural graduation of Q ( 5 ) where
especially the homogeneous polynomial Q ( 5 ) of highest order,called also m principal symbol, is well defined and plays an important role in the classical elliptic theory of partial differential operators. The main object of the investigation here are differential operators Q(E,D)
whose symbols may depend upon a small parameter
that their degree deg Q depends on deg Q ( E , S ) = m, V
E
E
E
E [O,E
0
1 and such
in the following fashion:
E ( 0 . ~ ~ 1deg , Q ( 0 , S ) = m 1 < m.
A typical example of such an operator is the one appearing in the
linear Elasticity theory:
3 . Singular Perturbations on Smooth Manifolds without Boundary
214
2 2 (3.1.3)
QO(E,D)
2
= E A -A,
f o r which m = 4, m
=
A
=
-
Dk
C
lZk
2.
1 The c o r r e s p o n d i n g symbol Q
5)
(E,
O
f u n c t i o n of d e g r e e 2 i n v a r i a b l e s
(E
=
-1
E
2
15 1 4+1 5 I
,F) E
(3.1.3) can be r e p r e s e n t e d as a product Q
Rn.
R + X
(E,C) 0
i s a homogeneous
Besides, Q ( E , ~ ) i n
2 2 2 151 ( 1 + ~151 ) so t h a t
=
it is n a t u r a l t o a s s o c i a t e w i t h Q i t s v e c t o r i a l o r d e r v = ( 0 , 2 , 2 ) , which i s
t h e l e a s t v e c t o r i a l o r d e r such t h a t Q ( E , D ) 0 uniformly with r e s p e c t t o E .
:
H
(s), E ( R n )
H(S-V) , E
(W")
I f one c o n s i d e r s a s l i g h t l y d i f f e r e n t d i f f e r e n t i a l o p e r a t o r
(3.1.4)
Q(E,D) = QO(€,D)+Q1(D),
then t h e r e p r e s e n t ati o n sum of homogeneous i n
v(l)
=
(E
(3.1.4)
-1
Ql(D)
=
i s a n a t u r a l s p l i t t i n g of Q ( E , D )
,5) t e r m s
of o r d e r v(O)
> v"),
r e s p e c t i v e l y , b e s i d e s v")
(O,l,O),
1 b D l
(0,2,2) and
=
so t h a t a g a i n one c a n
view Q O ( ~ , S ) a s t h e p r i n c i p a l symbol of Q ( E , ~ ) and s h o u l d e x p e c t Q p l a y thedominant
0 to r o l e i n t h e e l l i p t i c t h e o r y of d i f f e r e n t i a l o p e r a t o r s
a f f e c t e d by t h e p r e s e n c e o f a small p a r a m e t e r .
> i s o n l y p a r t i a l , it i s n o t s u r p r i s i n g t h a t a , 5 ) principalsymboldoesnotalwaysexistforagivensymbol
Since t h e o r d e r i n g geneousin
(E
-1
4
I n f a c t , t h e polynomial Q ( E , ~ ) =
E
s p l i t i n a sum of homogeneous i n
(E
Q,
= E
4
151 -1
5
+E
2
homoQ(E,<)
4
151 + / 5 I 2 + < b . 5 > can b e s t i l l
,5) t e r m s
Q
= E
2
1514+1512
and
1 5 j 5 + < b . < > o f d e g r e e 2 and 1 r e s p e c t i v e l y , b u t t h e i r v e c t o r i a l
o r d e r s v")
= ( 0 , 2 , 2 ) and
n e i t h e r v(O)
> v"),
v(l) = (0,1,4) c a n n o t b e compared, s i n c e
n o r v(l)
> v").
A l l t h i s j u s t i f i e s t h e following
D e f i n i t i o n 3.1.1.
A function Q
:
( 0 . ~ ~x 1x n + c i s s a i d t o belong t o t h e class Pv w i t h
v = (v1,v2,v3) E R ~ i ,f ( i )Q ( E , ~ )i
s a polynomial i n 5 E
R~
w i t h c o e f f i c i e n t s belonging t o
m
c
((O,EO1).
( i i )There
e x i s t s a decovposition Q
=
such t h a t
Q+,R
Q,
can be extended a s
a homogeneous f u n c t i o n of ( E - I , ~ )E R + x mn of degree v1+v2, (3.1.5)
-1 Q o ( tE , t C )
= t
satisfying the inequality
v 1+v 2
Q,,(E,S),
t/ t
E R+, t/
(E,c)
E
R + x Rn,
3 . 1 . Singular Perturbations with Constant Symbols
and t h e remainder
R
215
s a t i s f i e s the inequality -V
(3.1.7)
151 v 2 < E c > v 3 ~ 1 5 ~ - 1 + E ) v,
/ R ( E , S L 5 CE
w i t h some p o s i t i v e c o n s t a n t
E
E
v 5
(O,Eo1,
E En,
15)
2 1,
C.
The f u n c t i o n Q ~ ( E , ~ ): IR
x
IR~.-+
c i s c a l l e d t h e p r i n c i p a l symbol o f
Q ( E , ~ ) .
- (v +u2)
Putting t =
E
in
go(€,<)=
one f i n d s :
[3.1.5),
Qo(l,~g).
E
For d e r i v i n g a p r i o r i estimates f o r e l l i p t i c s i n g u l a r p e r t u r b a t i o n s w e
i
s h a l l u s e symbols of t h e form Q ( E , ~ ; ) w i t h L(E,:',~,)
=
with
<5'>1c'/-l5',
=
<5>1-'5
and even of t h e form
which a r e w e l l d e f i n e d f o r
6 E
En-,
i s a s u b s e t of measure z e r o .
where S c IRn
D e f i n i t i o n 3.1.2.
A function L
:
( 0 . ~ ~x 1m n \ s +
t o be a symbol i n t h e c l a s s Lv
c w i t h s a c o n i c s u b s e t of measure zero, i s s a i d
=
LV(xn) with v
= (vl,v2,v3),
i f
( i ) t h e r e e x i s t s c o n s t a n t s C,m such t h a t
v
-v (3.1.8)
'is>
/L(E,S)I 5 CE
'<EF,>'
( i i ) t h e r e e x i s t s a decomposition L a homogeneous f u n c t i o n of (E-',S) E
(3.1.10)
Lo(t
-1
3
,V
E ( 0 . ~ ~X 1l R n \ S ,
= L~+R, IR+X
such t h a t L 0 can be extended a s of degree v1+v2,
(R~,{o})
v +v ~ , t 5 )= t
2Lo(~,S), V t
E IR+,
V
(E,E)
€ IR+
x
(IR~\{O})
s a t i s f y i n g t h e ineyuazity -v (3.1.11)
iLO(E,S)l
and t h e remainder
R
5 CE
v
IR(E,S)I
x
(lRn\{O})
s a t i s f i e s t h e estimate: -v
(3.1.12)
3
'151 2 < ~ 5 > v , V (E,S) E R+
5 CE
v
'151
2 < ~ 5 > v 3 ( 1 5 1 - 1 + t ) ,V E E ( 0 . ~ ~ 1V, 5 E Rn\S,
151 21 w i t h some p o s i t i v e c o n s t a n t c . The f u n c t i o n L O ( ~ , S ) : E+ x I R ~+ c i s c a l l e d t h e principaZ symbol of L.
3 . Singular Perturbations on Smooth Manifolds without Boundary
216 Lemma 3 . 1 . 3 . (1)
If L . E l u ( j )j,
( i i )If
1 L~ E
1
=
1,2, then L1L2 E L" w i t h v
IL1(E,S)I 6 C I E
w i t h a p o s i t i v e c o n s t a n t cl, (iii)L
(1)
( i v ) 1" ( 1 )
uL
n
" (2) 5 L\,
1"
V
P r o o f . With -
u(1)+v(2).
s a t i s f i e s the inequality: -V
(3.1.13)
=
(1) 1
u(l)
<5>
<ES>
" (1) 3
,v
E (O,Eo1
(E,5)
t h e n L ~ ( E , ~ ) - 'E 1 w i t h u
with the least u
>
.(I),
=
x R
n y
-u ( 1 ) .
j = 1,2.
c 1 w i t h t h e l e a s t v =< u ( j ) , j = 1,2. (2) - I, b e t h e s u b s e t where L . i s n o t d e f i n e d , m . 1 1 and l e t L . = L + R . be t h e decomposition accor1 jrO I ( i i )of D e f i n i t i o n 3 . 1 . 2 . I t i s immediate t h a t t h e symbol
j = 1 , 2 l e t S . c Rn
3
the constants in
(3.1.9)
ding t o t h e p a r t
L = L1L2 s a t i s f i e s ( 3 . 1 . 1 0 ) w i t h v = v ( 1 ) t v ( 2 ) on Rn\S
For showing
(3.1.9)
with S
= S
1
w e e s t i m a t e L ( E , S ) - L ( E , ~ )a s f o l l o w s :
"
s2-
Further, the inequality
214<5-n>l4,
5
<S>a
so t h a t ( 3 . 1 . 5 ) h o l d s f o r L
m
=
= L L
1 2
v (5,n)
with u
v a E
E n n x lRn,
R
= u ( ~ ) + v ( ~ and )
rnaxtm2 'm1 + ~ v2( ~ ) l + l v ~ ~ ) l } . O b v i o u s l y , t h e f u n c t i o n Lo
=
LloLz0 s a t i s f i e s t h e c o n d i t i o n s ( 3 . 1 . 1 0 ) ,
(3.1.11). The r e m a i n d e r R (3.1.14)
IH(E,E)
One h a s
I
6
=
L-Lo
can b e e s t i m a t e d a s f o l l o w s :
lR1(E,5)R2(E,5) /+IR1(E,5)L2,0(E,c)
I+IR2(E,c)L
3.1. Singular Perturbations with Constant Symbols -v
/R1(E,5)R2(E,t)~5
cE
217
v llcj 2<Et>v3(~5~-i+i)8
v
(E,c)
E
(o,Eol
nn\S,
151 2
1
w i t h v , = v ! 1 ) + v ! 2 ) s i n c e ( / ~ l - l + E ) -5~ C ~ S I - ~ + fEo r 151 2 1 , E 5 E 0' 1 3 3 F u r t h e r , it i s q u i t e o b v i o u s t h a t t h e t w o l a s t t e r m s i n t h e r i g h t hand -V
s i d e of
15 1 u 2 < ~ c > u( 31 5
( 3 . 1 . 1 4 ) a r e bounded by C E
a s Well.
T h i s p r o v e s t h e f i r s t p a r t of Lemma 3 . 1 . 3 .
If L 1 s a t i s f i e s (3 1 . 1 3 ) t h e n o b v i o u s l y ( 3 . 1 . 8 ) h o l d s f o r -1 L ( E , ~ )= L 1 ( c , < ) with v =
-u(').
The d i f f e r e n c e L ( E , S ) - L ( E , I ? ) c a n b e
estimated a s follows:
where C 1 and C a r e t h e same p o s i t i v e c o n s t a n t s a s i n ( 3 . 1 . 1 3 ) respectively.
v w i t h some p o s i t i v e c o n s t a n t c
2
.
E
E
R+,
v
5 E nn\iO})
and (3.1.12),
3. Singular Perturbations on Smooth Manifolds without Boundary
218
Thus, t h e f u n c t i o n L (E,S) = L l 0 ( t , t ) - ' 0 (1)
with v =
s a t i s f i e s (3.1.10).
(3.1.11)
.
-w
The remainder R = L-L
0
can b e e s t i m a t e d a s f o l l o w s :
I t i s l e f t t o t h e r e a d e r t o check
( i i i ) and
I
(iv).
D e f i n i t i o n 3.1.4.
If
L 6
then t h e family of operators
L
(3.1.15)
L(E,D)u(x)
E
-f
-1
V u
= FEex L ( E , ~ )Fx+[u,
d e f i n e d by t h e formula
L(E,D)
E
S(Rn),
i s s a i d t o belong t o O P L ~ and i s c a l l e d a s i n g u l a r p e r t u r b a t i o n . Obviously,
L(c,D)
:
S(iRn)
+
S' (Rn).
D e f i n i t i o n 3.1.5.
A singular perturbation L ( E , D )
:
i f for
v
v e c t o r i a l order v E IR' L(€,D)
: H
( s ) ,E ( n n )
S ( I R ~ -f ) S'
(nn) is s a i d t o have t h e
s € 1~~ i t can be exLended as
( m n ) uniformly w i t h r e s p e c t t o + H(s-v) ,E
E
E ( 0 , ~ ~ l .
Proposition 3.1.6.
If
L ( E , ~ )
E Lv,
then
: s(#)
L(E,D)
-f
s 8 ( m n ) has t h e v e c t o r i a L order w.
P r o_ of. _
L
If L ( E , ~ )E
(3.1.8)
then
yields L(E,D)
E
V s E Rn u n i f o r m l y w i t h r e s p e c t t o E Along w i t h L ( E , D )
E OPLw
: S(R)
+
S'(R)
D e f i n i t i o n 3.1.7. 0
(3.1.17) L
0
(E
v1
<5>
-v
0
(Rn)
1.
-f
,
0
( € , e l )
-f
L(E,E',D,),
i s c a l l e d a reduced symbol of
-V
Lv
e = (l,-l,l),
( 5 ) being a reduced symbol of L ( E , ~ ) t, h e operator OpLO OPL = L (E,D).
L(E,E) E
-V
2 < ~ < > 3 L ( ~ , 5 ) - < 5 > 2 L o ( 5 ) ) E L-,,
t h e reduced operator o f
H(s-v),E ( I R n )
d e f i n e d by t h e f o r m u l a :
The f u n c t i o n L ( 5 ) E L ( o , v 2 , 0 )
if
(O,E
( s ) ,E
w e s h a l l a l s o c o n s i d e r l a t e r t h e f a m i l y of
one d i m e n s i o n a l s i n g u l a r p e r t u r b a t i o n s L(E,<',D,)
: H
= L'(D)
i s called
3.1. Singular Perturbations with Constant Symbols
219
Lemma 3 . 1 . 8 .
( i )I f f o r L E lv a reduced symbol L
0
E 1
unique. ( i i )I f L E
Lv w i t h v 3 -2 0 has t h e reduced symbol
Proof. __ ( i ) The i n c l u s i o n ( 3 . 1 . 1 7 ) y i e l d s L
def ined
0
(6) =
e x i s t s then i t i s
(0,v2,O) L
0
then
E L(o,v2,0)
l i m E"L(€,~).
Thus Lo i s w e l l
€4-0
.
( i i )The symbol cV1L-Lo c a n b e r e w r i t t e n a s f o l l o w s :
E
v1
L(E,S)-L
0
(5)
=
-V -v < p v 2 < E p 3 ( E"1
"
3 E Since < p V 2 < ~ < > V
1 (o,v2,v3),
( 3 . 1 . 1 7 ) and Lemma 3 . 1 . 3
imply t h a t t h e
f i r s t t e r m i n t h e r i g h t hand side of t h e l a s t i d e n t i t y belongs t o For v3 2 0 one h a s : ( < E S > ' ~ - ~ E ) L(-l,l,v - 1 ) , so t h a t +l,v -1). 3 3~ a t h e symbol L ( t ) ( < ~ S > ~ ~ - lb)e l o n g s t o l(-l,v + l , v - 1 ) , a s w e l l . 2 3 D e f i n i t i o n 3.1.9. 5-1,"
The symbol L~
L E
Lv i s c a l l e d e l l i p t i c of order v i f i t s p r i n c i p a l symbol
satisfies the inequality:
(3.1.18)
-vl
IL ( E , S ) ~ 2 C E
0
w i t h some p o s i t i v e c o n s t a n t
"
J '
151 2 < ~ E > ',
V (E,S)
E
1R+
x
(Rn\{O)).
C.
The c l a s s of a l l e l l i p t i c symbols o f order v i s denoted by OPE\,
E ~ w , hile
stands for t h e corresponding c l a s s o f e l l i p t i c s i n g u l a r p e r t u r b a t i o n s
o f order v. def {w
1
w E R n , / w I = 1 ) . Using t h e homogeneity of L 0 Of d e g r e e v +u2, one c a n r e f o r m u l a t e t h e c o n d i t i o n ( 3 . 1 . 1 8 ) e q u i v a l e n t l y a s 1 follows :
Let Q
-v (3.1.19)
ILo(p,w)I
2 Cp
v '
',
V (p,w)
E 0
R+x
I n t r o d u c i n g t h e r e d u c e d p r i n c i p a l symbol L o ( C ) ,
and t h e homogeneous p r i n c i p a l symbol L o o ( c ) ,
nn.
3 . Singular Perturbations on Smooth Manifolds without Bounda y
220
- ( v +v )
(3.1.21)
L
00
(5)
2
= l i m t
3
L0(l,tS), 5 E R n ~ i O l
tJ-m
one c a n r e f o r m u l a t e e q u i v a l e n t l y t h e e l l i p t i c i t y c o n d i t i o n a s f o l l o w s : P r o p o s i t i o n 3.1.10.
0
L
If L E
admits t h e reduced p r i n c i p a l symbol L ~ ( C ) and t h e homogeneous
p r i n c i p a l symbol ~
~
~ d e(f i n5e d ) by ( 3 . 1 . 2 0 1 , ( 3 . 1 . 2 1 ) , then
L
is e l l i p t i c
( L E E )iff t h e f o l l o w i n g c o n d i t i o n s a r e s a t i s f i e d :
LOO(,)
(i)
( i i )~
~
~
v
# 0, #( 0 w,
) V
# 0, V
( i i i )L O ( p , w )
w E w
E R E R+x an.
(p,w)
Proof. -
L
If L E
i s e l l i p t i c of o r d e r v t h e n (3.1.19) h o l d s f o r V (p,w)
Multip1y':ng
letting
151
+
E
.
n
0 one g e t s ( i ) -(v2+v3) = 1 , m u l t i p l y i n g it by 151 and
( 3 . 1 . 1 9 ) by p V 1 a n d l e t t i n g p
Further, taking i n (3.1.18)
E R + x S2
+
one g e t s ( i i ).
m,
( E , C ) of o r d e r v1+v2 and ( 3 . 1 . 1 2 ) imply (iii). 0 -V ( i ) - ( i i ih )o l d s , t h e n ( i ) y i e l d s / L ( p , o ) /2 C p
The homogeneity of L Conversely, i f
0
E (O,po]
f o r V (p,w)
PO
Rn w i t h s o m e p o s i t i v e c o n s t a n t C which depends 0 0 PO '3-'1 o n l y on p o and y o = min IL (IJI) while (ii) i m p l i e s I L o ( p , w ) 2 C p wERn 0 p1 which depends f o r V (p,wl E [p,,-) X Qn w i t h some p o s i t i v e c o n s t a n t C x
I
1;
o n l y on p 1 a n d y o 0
=
min
IL
00
wERn
((0)
~. The
41
Qn b e i n g compact
s u b s e t [pO,pl]x
R ,
one can f i n d , u s i n g ( i i i ) a c o n s t a n t C > 0 s u c h t h a t n 2 C P - ~ ' < P > " ~which , i s p r e c i s e l y ( 3 . 1 . 1 9 ) and by homogeneity of 0 -1 L ( e , < ) of o r d e r v +v2 i n ( E ,C) is equivalent t o (3.1.18). I 0 1 in R IL
+
X
(p,w)
P r o p o s i t i o n 3.1.11.
Let
L ( E , D ) E O P L , ~ . Then L is e l l i p t i c o f order following e s t i m a t e holds f o r E~ small enough:
(3.1.22)
/ I ~ l l ( S ) , E ~ C ( / I L U / I ( S+ -" l l)u,lE[ ( s , ) , E ) ,v
v where the c o n s t a n t cdoes n o t depend on
alrs
=
is'
v ( L ( E , D ) E OPE")
1
SI
3
E R ,
S I
<
S ,
S;
E E
n3,
v
so E
E
E (O,E01,
a3,
$ 1
iff t h e
v
uEs(Rn),
< s, s' E airs,
and u and where
= s1}.
Proof. Let L(t,D)
be e l l i p t i c of o r d e r v and l e t L 0 ( € , C )
of L ( E , ~ ) .Witn
i
=
< 5 > \ 5 \ -15 t h e e l l i p t i c i t y o f
b e t h e p r i n c i p a l symbol L implies:
3.1. Singular Perturbations with Constant Symbols
22 1
1
C1. Denote L O ( ~ , D )= OpL ( E , E ) . Then, u s i n g t h e P a r c e v a l ' s i d e n t i t y and 0 t h e i n e q u a l i t y C 5 - P 5 1,one g e t s
The l a s t i n e q u a l i t y l e a d s t o (3.1.22) w i t h s'
=
s-e2 i f
E~
< C/(2C ) .
1
Now,
u s i n g t h e i n t e r p o l a t i o n i n e q u a l i t y (2.1.11) f o r Sobolev norms o f v e c t o r i a l o r d e r , one g e t s (3.1.22) w i t h V s'< s , s'
airs.
E
C o n v e r s e l y , assume t h a t (3.1.22) h o l d s w i t h s o m e s and s' = s-e2. Then t h e Parceval's i d e n t i t y implies:
for
v
(E,s)
E
(o,E01
x
(R~,s).
The l a s t i n e q u a l i t y i m p l i e s :
for (E,<)
E ( 0 . ~ ~x 1 ( R n \ S ) ,
positive constant C
16) 2
r w i t h r l a r g e enough, where t h e
depends o n l y on C , v and r .
The homogeneity o f L O ( ~ , S )i n
(E
-1
o f o r d e r v +v2 i m p l i e s t h a t t h e
,<)
1
l a s t i n e q u a l i t y , perhaps with a d i f f e r e n t p o s i t i v e c o n s t a n t , h o l d s f o r
v
(E,S)
E R+
x
I
(W"\IO}).
Example 3.1.12. The s i n g u l a r p e r t u r b a t i o n s
(3.1.3), (3.1.4) a r e e l l i p t i c o f o r d e r (0,2,2),
while t h e s i n g u l a r p e r t u r b a t i o n Q ( E , D ) =
E
2
4
I D 1 -1DI2 i s n o t e l l i p t i c .
Remark 3.1.13. I t i s l e f t t o t h e r e a d e r t o check t h a t any s i n g u l a r p e r t u r b a t i o n Q ( E , D )
w i t h polynomial symbol Q ( E , ~ )E p a d m i t s t h e r e d u c e d p r i n c i p a l symbol 0 and t h e homogeneous p r i n c i p a l symbol Q o o ( 5 )
Qo(S)
p~o,v2,01
The same i s t r u e f o r symbols which a r e r a t i o n a l f u n c t i o n s of 5 .
p(o,v2+v3,0)
'
3 . Singular Perturbations on Smooth Manifolds without Boundary
222
Remark 3 . 1 . 1 4 . some s ' < s s u c h t h a t s' < s l , t h e n f o r 1 s u f f i c i e n t l y s m a l l t h e s i n g u l a r p e r t u r b a t i o n L ( & , D ) , E € (@,€,,I, If
(3.1.22) h o l d s for
E
-
equiisomorphism, i . e . L -1 -1 L := Op L(E,S) : H
:= L ( s , O )
(s-")
: H
0 is m
(IRn ) and
( s ), E
(IRn),
E
E
(u,E@~,
,E
are b o t h
e q u i c o n t i n u o u s (see D e f i n i t i o n 1 . 2 . 3 3 ) .
D e f i n i t i o n 3.1.15.
An e l l i p t i c s i n g u l a r p e r t u r b a t i o n
L(E,D)
E
OPEv
is caZZed s t r o n g l y e L l i p t i c
of order u if w i t h some c o n s t a n t s 6 E [ 0 , 2 n ) , y E R+ hoZds: -v (3.1.23)
where
Re(eiCLo(s,F)) L
L ~ ( E , ~i )s
YE
u
' / 5 ( 2 < ~ 5 >3v, V
t h e p r i n c i p a l symboZ of
(E,<)
E
(IRn\{O)),
R+x
L.
Example 3 . 1 . 1 6 . The s i n g u l a r p e r t u r b a t i o n s order u
=
(3.1.131,
(3.1.14) a r e s t r o n g l y e l l i p t i c of
( 0 , 2 , 2 ) , t h e c o n s t a n t s i n ( 3 . 1 . 2 3 ) b e i n g 8 = 0 , y = 1; t h e
i D 2 + D 2 ) 4 ( 1 + ~ 2I: D;)" is e l l i p t i c I: k n llksn-1 l
= (
P r o p o s i t i o n 3.1.17.
Let
L(E,D)
E
OPE
be s t r o n g l y e l l i p t i c of order v and Zet 6, y be t h e
corresponding c o n s t a n t s in ( 3 . 1 . 2 3 ) . Then t h e r e e x i s t c o n s t a n t s
Co
and
Cl
such t h a t (3.1.24)
i6 Re<e
L(E.D)U,U> 2
(Y-C 0 E )
1 IuI 1
2 ( v / 2 ) ,E-Cll
V u E H
( v / 2 ) ,E
where < , > denotes t h e d u a l i t y betweenIl(u,2) , E
, 2 lU/
(Rn)
,
l((v-e2),2) v
(Bn)and H
E
,€,
E (O,EO1'
(-u/2) ,E
(En).
Proof. Without r e s t r i c t i o n of g e n e r a l i t y one c a n assume t h a t 8 L (6,F)
0
=
0, y
b e i n g t h e same a s i n t h e p r o o f of P r o p o s i t i o n 3 . 1 . 1 1 ,
u s i n g t h e same argument a s i n t h e p r o o f of t h i s p r o p o s i t i o n :
=
1.
one g e t s
3.1. Singular Perturbations with Constant Symbols
223
P r o p o s i t i o n 3.1.18.
Let
L ( E , D ) E OPE be s t r o n g l y e l l i p t i c o f order v w i t h p r i n c i p a l symbol L ~ ( E , ~ )Assume . t h a t L admits t h e reduced symbol L0 ( 5 ) E L and l e t
(0 'V2 , O )
0 L~
( 5 ) be t h e corresponding reduced p r i n c i p a l symbol. L e t
(3.1.25)
0 -1 0 P ( F , ~ =) L ( E , S ) - L ~ ( E , ~ ) L ~ ( LS ) ( 5 ) .
Then t h e r e e x i s t s a c o n s t a n t
c such t h a t
Proof. ___ O b v i o u s l y , t h e symbol P E identically: P
0 (3.1.12), P ( E . 5 )
consequence of
(E,S)
Z
Lu and i t s p r i n c i p a l and r e d u c e d symbols v a n i s h
0 , P o ( S ) E 0 . T h e r e f o r e , a s a consequence of
E L(u-el)
u L(u-e2),
(3.1.17), P ( E , S )
s i n c e P0 ( E , S )
E L(u-e),
. . T h i s l e a d s to t h e c o n c l u s i o n t h a t P ( E , ~ E) ).
e
=
L
Z
0. F u r t h e r , a s a
(1,-1,1),
(we)
"
0
s i n c e P ( 5 ) :0 .
('("-el)
"
L(u-e2)) =
I t i s immediate t h a t
w i t h some c o n s t a n t C > 0 and t h e l a t t e r h a s ( 3 . 1 . 2 6 ) a s i t s immediate
I
consequence. C o r o l l a r y 3.1.19.
Let
L(E,D)
E
OPEv be s t r o n g l y e l l i p t i c o f order u and l e t L O ( ~ , c )be i t s 0
0
principaZ symbo2, L (D) i t s reduced symbol and
L ~ ( D )i
symbol. Assume t h a t t h e r e i s a c o n s t a n t yo
such t h a t
Then t h e r e e x i s t s such a c o n s t a n t
c that
> 0
t s principal reduced
3. Singular Perturbations on Smooth Manifolds without Boundary
224
(3.1.28)
2 1 I U I ](v,2),E~
0 Re<eieL(E,D)u,u> 2 (YoY -CE)
V u E H
(v/2)
,E
(2) ,v
E
E
(O,Eo1.
I n f a c t , (3.1.27) is an immediate consequence o f (3.1.25)-(3.1.27). Remark 3. 1.20.
0
I f L ( E , D ) 6 OPE
a d m i t s t h e r e d u c e d symbol L ( D ) E OPE
(0,v2I O )
'
then
i n t r o d u c i n g t h e symbols
(3.1.29)
-1
0
R(E,<) = LO(~,S)LO(S)
R ( E , ~ )=
anddefining
,
S ( e , S ) = R(E,S)-~
R ( E , ~ ) ,; ( € , < I
= S(E,<),
i
with
=
< < > I t-1/ 5 ,
one
c a n w r i t e down: L(E,D)
(3.1.30) L
where
0
(D)
=
R ( E , D ) L 0 ( D ) + E Q(~E , D )
= ;(E,D)L(E,D)+EQ~(E,D),
Q 1 ( ~ , S )E L v l Q ~ ( E , S )E L ( O ,"*,o ) .
I n f a c t , t h e argument u s e d i n t h e p r o o f of P r o p o s i t i o n 3.1.18 i s 0 a p p l i c a b l e t o t h e symbols p l ( ~ , < ) L ( E , ~ ) - ~ ( E , < ) L( 5 ) and
p2(€,6) = L0 ( < ) - S-( E , < ) L ( E , S ) , s i n c e t h e i r p r i n c i p a l and r e d u c e d symbols 0 v a n i s h i d e n t i c a l l y : P . ( E , S ) 5 0 , ? . ( 5 ) :0 . JO 3 T h i s remark w i l l be u s e f u l l a t e r f o r t h e r e d u c t i o n o f s i n g u l a r
e l l i p t i c p e r t u r b a t i o n s w i t h v a r i a b l e symbols t o t h e r e g u l a r o n e s .
3.2. S i n g u l a r p e r t u r b a t i o n s w i t h homogeneous symbols Let
E 1". Then c l e a r l y t h e f a m i l y ( 0 . ~ ~3 1E
a(E,S)
-f
a ( € , . ) E S'(R;) i s
a c o n t i n u o u s l i n e a r map and one c a n r e w r i t e a ( ~ , D ) ua s t h e f o l l c w i n g
f a m i l y of c o n v o l u t i o n o p e r a t o r s :
(3.2.1)
a ( E , D ) u = (AE
where A (x) For V
=
E
F
-1
5-
*
v
U),
a ( E , c ) , ( 0 . ~ ~3 1E
-f
u E S(IRn)
,
Ac E S'(lR:).
E ( O , E ~ ] and V u E S ( R n ) t h e f u n c t i o n
r a p i d l y decreasing as
6
-f
-, so t h a t a ( ~ , D ) E u
<
m
C (IRn)
-f
,
a(E,S);(<)
V
E
E
is
(0,s01.
S i n g u l a r p e r t u r b a t i o n s whose symbols a r e homogeneous f u n c t i o n s o f (E
-1
,6) a r e
of s p e c i a l i n t e r e s t f o r t h e a p p l i c a t i o n s .
D e f i n i t i o n 3.2.1.
A function:
R+X
(R~\[o})3
(E,s)
+
a0(E,<)
E c is said to belong to t h e
3.2. Singular Perturbations with Homogeneous Symbols
symbol c l a s s Hm
V'
(i) . a
E c
i f t h e fo%%owingc o n d i t i o n s a r e s a t i s f i e d :
o i ~ x+ (IR~\IO}))
(ii) t h e r e e x i s t s a c o n s t a n t (3.2.2)
lao(E,~)I5 CE
-1
c
If a
o
-1
ao(t
E,t<)
=
> 0
such t h a t
151 2<E5>v3, v
(iii) ao(E,t) i s homogeneous i n (3.2.3)
225
(E,c)
E
R+X
(IRn\{O}),
( E - ~ , C )of degree u l + u 2 :
u +u 1 2 t ao(E,c), v t E R + , V
(€,
E IR+X(R*\IO}).
E H-u is a polynomial in 5 ,
then a (6,D) is the differential singular perturbation, 0
E
If a .
f , v2
> -n,
then R + 3
E
-f
a0 (E,.) E S'(R;)
i s a family of
regular distributions and the corresponding singular perturbation can be written as the Fourier integral (3.2.5)
aO(E,D)u(x) = (2n)-" 1 ei<x*s'
. a
(E,
5 ) ( 5 )d5,
Rn
v
E
E R + ,v u E S ( R n ) .
v 2 < 0, -(n+l) < v +v < 0. Let a E H: with u E IR3 such that -n 0 2 3 We shall compute the family of distributions IR+ 3 E + A. (E,. 1 E S ' (Rz),
(3.2.6)
-1
A (€,XI = F5+x ao(E,S),
0
which are the kernels of the convolutions a (E,D)u(x)= (A ( E , . ) * u ) (x), 0 0
u E
S(Rn).
The homogeneity of degree v + u 2 of a ( E , S ) implies: 1 0 - ( v +u +n) (3.2.7) AO(€,x) = 6 AO(X/E),
*
3 . Singular Perturbations on Smooth Manifolds without Boundary
226
Let
]5,/
=
E
-1
b0(5)
where S
=
and l e t
R
=
{y 1
-1
a n d , moreover,
v +v 2 3
(3.2.2)
=
N
,E),
E, E
( x O , x ) . Denote
(nfl+',s),
, 5, =_0 o r 5 = 01. E H yields: 0 v
E Wn+'
bO(ty) = t
(C0,5), x
=
a0(1501
/EO1
The homogeneity of a (3.2.9)
5
v -v
N
(3.2.8)
N
N
, 50 E
-
bo(5),
V t E R+, V
y
E (IRn+'\S),
implies:
w i t h some c o n s t a n t C > 0 . one c a n w r i t e
Obviously,
v -v
15,/
(3.2.11)
-1 1A0(1501
-
-1 -1 ,x) = Fg+xbO(So,C) = FX -tS F - - b o ( 5 ) . 0 X0' 5
A s a consequence of t h e assumption n+u2 > 0 , n + l + v 2 + v 3 > 0 , t h e d i s t r i b u -
tion b
0
E
S'
(R:+')
i s a r e g u l a r one a n d , t h u s ,
5
t h e i n t e g r a l i n t h e r i g h t hand s i d e o f
(3.2.13) being convergent f o r
N
x E (E?n+l\{O}) and V 6 > 0 , s i n c e ( 3 . 2 . 1 0 ) h o l d s and n+v2 > 0 , n+l+u +V > 0. 2 3 One f i n d s , u s i n g t h e homogeneity of b ( 5 ) of d e g r e e 0 N
-I-
where z + T(z) i s t h e r - f u n c t i o n , n+l = {w w E IR , = 11.
=
lTl-'T,?
=
V
2
+v 3
lxl-lz and
Therefore,
u -v (3.2.14)
A
0
(E,X)
= E
3
i ( n / 2 ) (u2+u3+n+l) r ( v 2 +v3 + n + l ) 1(2n)-(n+l)e
*
3.2. Singular Perturbations with Homogeneous Symbols
227
-
- ( v +v + n + l ) (<;',;>+is)
A f t e r t h e change of v a r i a b l e : xo
=
Ixlt,
dwdx
0'
(3.2.14) y i e l d s t h e following
formula: (3.2.15) -(n+l),
.
r ( v +u + n + l ) [ x / 2
-1 J
2
3
3
E
- ( v +v + n t l )
-1
.-iE
J
- ( v +v + n ) v3-vl
i ( n / 2 ) (v2+v3+n+l)
= (2n)
A (E,x) 0
dwdt,
'n+1 where y
=
/ x / - l x and t h e d i s t r i b u t i o n ( t + i O ) h
the distributions ( t + i G ) '
6
when
i s t h e l i m i t i n S ' ( R ) of
(see [Sch
J. 0
1
1,
[G-Sh 1
1).
Example 3 . 2 . 2 . Denote by A z ( c , x ) t h e f a m i l y of d i s t r i b u t i o n s E
0
A Z ( ~ , x =) F-l
0
X;(n)
X;(E~),
=
where z
E
-f
C,
A'(€,.) 0
E
S'(R:)
,
0 > Re z > - ( n + l ) .
X'S
Then t h e f o r m u l a ( 3 . 2 . 1 5 ) y i e l d s : (3.2.16)
- I
.
- ( n + l ) i ( n / 2 ) ( z + n + l ) r ( z + n + l~) ~ j - ( z + n ) ~ z
A'(E,x) 0
= ( 2 ~ )
-iE-l
J e
1x1 t ( t w + < Y , W > + l O )
0
- ( z + n + ld)r d t .
'n+1
-
The i n t e g r a l 1
z,n
=
( t < t > - l w 0 +
J
-(z+n+l) dw
'n+l -1 and y = 1x1 x E -1 Indeed, i n t r o d u c i n g y = < t > ( t , y ) ,
d o e s n o t depend on t
E
R
-
N
s u c h t h a t y becomes eo
=
(l,O,
...,0) E
an. = 1 and making a r o t a t i o n
one f i n d s :
W e s h a l l n o t compute h e r e t h e c o n s t a n t I and s h a l l g i v e o n l y t h e z,n f i n a l formula f o r A Z ( c , x ) w i t h 0 > R e z > - ( n + l ) :
0
A s a consequence of i t s d e f i n i t i o n ,
the function
3. Singular Perturbations on Smooth Manifolds without Bounda y
228
c a n b e e x t e n d e d a s a n e n t i r e f u n c t i o n of z E C v a l u e d i n S ' ( R n )
v
E
,
for
E R+. Of c o u r s e , one a l w a y s h a s : l i m A'(E,x) 0 €40
= 6(x)where 6(x) is the Dirac's
6-function. For z
=
1-n one c a n compute t h e i n t e g r a l o n t h e r i g h t hand s i d e o f
( 3 . 2 . 1 7 ) so t h a t
For z = 2m-n-lwhere hand s i d e of
m i s a n i n t e g e r , 1s m < ( n + 1 ) / 2 , t h e i n t e g r a l o n t h e r i g h t
( 3 . 2 . 1 7 ) c a n be a g a i n computed, u s i n g t h e r e s i d u u m
CalCUlUS,
so t h a t one f i n d s i n t h i s case:
(3.2.20)
Azm-n-l(~,x) = 21-n 2m-n-1
-
-1 2-E
1x1
.(n-1)/2r((n-1)/2-m) For Rez +n < 0 one c a n e x p r e s s A ' ( E , x )
-
f u n c t i o n s ( s e e [Grad.
If z
=
A;(E~)
0
Ryz.,
1
c 2' (2m- j - 2 ) OSjsm-1 c ' j ! ( m - l - j ) !
i n t e r m s of t h e m o d i f i e d H a n k e l ' s
1)
2m+a w i t h i n t e g e r rn > 0 a n d a E ( - 2 , 0 ) , t h e n , o b v i o u s l y , 2m = Xo ( E ~ ) X ; ( E ~ ) , so t h a t i n t h i s c a s e one g e t s t h e f o l l o w i n g
r e p r e s e n t a t i o n f o r t h e s i n g u l a r p e r t u r b a t i o n X'(ED):
0
(3.2.22)
XE (E,D)u (x) =
where A
= ( ~ - E ~ A In ~ )A; ( E , x - ~ ) u ( y )dy =
R 2 fnA;(~,x-Y) (I-€ A ) m u(y)dy, Y R
v
C a 2 / a x i i s t h e Laplace o p e r a t o r . 16kSn I n p a r t i c u l a r , u s i n g ( 3 . 2 . 1 7 ) one g e t s t h e f o r m u l a e =
A 0- ' ( E , x )
=
(3.2.23)
and f o r n = 3 t h e f o r m u l a ( 3 . 2 . 1 9 ) y i e l d s : (3.2.24)
.A -2
(E,x)
=
2 -1 - E - l ( 4 n ~1x1) e 1x1,
x
E
IR
3
.
u E S(IR?,
3.2. Singular Perturbations with Homogeneous Symbols Notice t h a t t h e d i s t r i b u t i o n s
E
.A
j.
-2
(E,x)
229
are fundamental s o l u t i o n s
2
f o r t h e d i f f e r e n t i a l s i n g u l a r p e r t u r b a t i o n 1-E A : (3.2.25)
2 (1-E A
-2
)Ao ( € , X I
V
6(x),
=
E R + , x E Bn.
E
We s h a l l e x i b i t a f o r m u l a f o r t h e s p h e r i c a l l y symmetric fundamental L
L
s o l u t i o n s f o r t h e d i f f e r e n t i a l s i n g u l a r p e r t u r b a t i o n E A -Ax,
appearing
i n t h e l i n e a r e l a s t i c i t y theory. Example 3 . 2 . 3 . Denote by
A”’-2
(E,x)
t h e f a m i l y of d i s t r i b u t i o n s F 2
xo- 2 ,
-2 (E,Sl
=
-1 x - 2 , - 2 0
S X ’
(E,S)
where
S i n c e one c a n a l s o w r i t e
I 1-
2
-E
2
<E5>
-2
,
one f i n d s , u s i n g ( 3 . 2 . 1 7 ) and t h e f o r m u l a f o r t h e s p h e r i c a l l y symmetric fundamental s o l u t i o n f o r t h e L a p l a c e o p e r a t o r : (3.2.26)
.A
-2,-2
(€,XI
=
T ( ( n - 2 ) / 2 ) l x 1 2 - n 2 -2 -E A. n/2
(E,X),
E
E R+,
X
E Rn.
4Tl I n p a r t i c u l a r , one g e t s (3.2.27)
-1 A 0 2 f - 2 ( ~ , x )= ( 4 ~ 1 x 1 ) ( l - e - €
-1 E
E R+, x E R
3
.
Example 3 . 2 . 4 . E -f Y ( E D ) , Y (E Dl = O P ( E 5 < E C > - ~ )w i l l be k k k c a l l e d s i n g u l a r l y p e r t u r b e d Riesz t y p e o p e r a t o r . W e s h a l l denote -1 (E E k < ~ F , > - ’ ) ,so t h a t one c a n w r i t e E j. Y (E,D) a s a f a m i l y Y (E,x) = F k S-tX k of s i n g u l a r i n t e g r a l o p e r a t o r s :
The s i n g u l a r p e r t u r b a t i o n
(3.2.28)
Y
k
(E,D)u(x)
I
= l i m
Yk(€,x-y)u(y)dy, V u E S(IRn).
640 I x - y [ > 6
W e s h a l l compute t h e k e r n e l d i s t r i b u t i o n Y
(E,x).
k C o n s i d e r f i r s t t h e c l a s s i c a l R i e s z o p e r a t o r Yk
=
Op(CklCl
-1
I.
One h a s : (Y u )
k
(x)
=
Op(lC
-1
)Dku(x).
1
The k e r n e l 12-l (x) o f t h e i n t e g r a l c o n v o l u t i o n o p e r a t o r Op( 15 - l ) c a n b e
0
e a s i l y computed:
230
3 . Singular Perturbations on Smooth Manifolds without Boundary
where the constant
Tk.e
C
can be found using the formula:
last formula yields:
Therefore, the integration by part
implies:
+
ic lim i 6-("+l) (xk-Yk)u( y "6J-0 lx-yl=6 - (n+l = -ic (n-1)v.p. J (x -y lx-yl Rn k k
+ ic lim i u(6w+x)w dw = k n6J-0 Rn /x-y~-(~+l = -icfi(n-l)v.p.i (x -y =n k k
Further, denoting, as previously,
E
=
/cO1-1 , -5 =
N
(C0,5), x
=
(xO,x), one
gets after the change of variable: xo = lxlt the following formula for the Riesz operator with a small parameter
:
-ix 5 0 On-(n+2)/2T((n+2)/2) v.p. xkiXl-(n+2)dx0 = =-i Je R
The operators Y (ED) can be also expressed in terms ofthe modified Hankel's k functions (see [Grad. Ryz., 1 1 ) :
-
3.2. Singular Perturbations with Homogeneous Symbols
The l a s t f o r m u l a c a n b e a l s o d e r i v e d from ( 3 . 2 . 1 7 ) , and u s i n g t h e r e l a t i o n : K ( r ) = K
231
(3.2.18), s e t t i n g z = 1
( r ), V v > 0 .
-V
Example 3.2.5. The s i n g u l a r p e r t u r b a t i o n A ( E D ) 0
=
O p < ~ pi s c a l l e d t h e
Riesz o p e r a t o r
w i t h s m a l l p a r a m e t e r . One f i n d s :
=
A -1 ( E D ) u ( x ) 0
-
C
i E
Yk(€D)
l5kSn
au ( x ) = axk
w y ayk The f o l l o w i n g s i n g u l a r p e r t u r b a t i o n of t h e one-dimensional
d
t
.
Riesz o p e r a t o r
(which i s t h e H i l b e r t t r a n s f o r m ) a p p e a r s i n t h e d i s l o c a t i o n t h e o r y ( s e e , f o r instance [ Hirth
-
Lothe,
1
1).
Example 3.2.6. Let
oE
= o p ( s g n c ( l + e x p ( - E / t / ) ) ,E,
E
R
. One
finds easily that
E +
OE i s
t h e f a m i l y of c o n v o l u t i o n o p e r a t o r s : OEU(X)
=
i
4v.p.
The m u l t i - d i m e n s i o n a l
(E,x)
u (y) ---dy x-Y
+
a n a l o g u e s of OE a r e t h e o p e r a t o r s :
One f i n d s t h e k e r n e l O k,n
0 k,n
1
= D 0
(E,x)
k n
o f t h e c o n v o l u t i o n o p e r a t o r s OE: k
(E,x),
where
For n
=
2 , e v a l u a t i n g t h e l a s t i n t e g r a l by means of t h e residuum c a l c u l u s ,
3 . Singular Perturbations on Smooth Manifolds without Boundary
232
so t h a t
(3.2.32)
0
k,2
(€,XI
= 2iT
-1
2 -3/2
2
x k ( c +41xI )
,
k = 1.2.
For n 2 2 one h a s : (3.2.33)
a
(E,x)
=
(n-1) ! 7
(2n)
71
J s i nn-2e ( ~ - i / ~ j ec) 1-n ~ sd e , 0
.
i s t h e area o f t h e u n i t s p h e r e i n R n - l where R n- 1 Hence, an e a s y c o m p u t a t i o n y i e l d s :
Of c o u r s e , i n any case, one h a s : (3.2.35)
lim
(E,x-y)u(y)dy
0
€40 Rn
=
2Yku(x),
V u E S(IR~).
krn
W e are g o i n g t o i n d i c a t e one m o r e c l a s s of s i n g u l a r p e r t u r b a t i o n s
with
homogeneous symbols, which i s i n t e r e s t i n g f o r a p p l i c a t i o n s . Example 3.2.7. Denote by AZ(ED',&D ) t h e s i n g u l a r p e r t u r b a t i o n w i t h t h e symbol
where y > 0 and z E C a r e g i v e n . F i r s t assume t h a t R e
2
< 0 . Using t h e formula
ni2/2 (3.2.37)
(-a+iB)'
=
r (-2)
t-z-l
-8t-iat
dt,
Re z < 0 ,
=+
we find that
where H(x ) i s t h e H e a v i s i d e ' s f u n c t i o n One f i n d s e a s i l y t h a t
I n f a c t , u s i n g t h e r e s i d i u m c a l c u l u s ( o r (3.2.37)
w i t h z = -11,
one f i n d s
3.2. Singular Perturbations with Homogeneous Symbols
233
Therefore,
which l e a d s t o ( 3 . 2.39) when x The f o r m u l a e ( 3 . 2 . 3 8 ) .
> 0.
( 3 . 2 . 3 9 ) imply f o r R e z < 0
so t h a t
I n t h e same manner one f i n d s f o r R e z < 0
Let R e z > 0 and l e t z
=
r + <w i t h i n t e g e r r > 0 a n d R e 5 < 0 .
W e c a n w r i t e down:
Hence.
and i n o r d e r t o f i n d t h e k e r n e l of t h e o p e r a t o r s compute t h e i n v e r s e F o u r i e r t r a n s f o r m of f i n d s:
ED' , E D -
)
one h a s t o
15' / X c ( c S ' , ~ C n ) , a s w e l l . One
*
3 . Singular Perturbations on Smooth Manifolds without Boundary
234
Using t h e same argument a s p r e v i o u s l y , one f i n d s :
so t h a t
> 0 , then t h e l a s t formula y i e l d s :
If x
we can again rewrite t h e s i n g u l a r p e r t u r b a t i o n s
Combining ( 3 . 2 . 4 0 ) - ( 3 . 2 . 4 2 ) ,
a s a f a m i l y of i n t e g r o - d i f f e r e n t i a l
,ED,)
A:(ED'
o p e r a t o r s . The s a m e argument
l e a d s t o a s i m i l a r r e p r e s e n t a t i o n f o r X-Z ( s D ' , s D n ) . F i n a l l y , i f z = r > 0 i s i n t e g e r , t h e n one u s e s o n l y ( 3 . 2 . 4 1 ) , f o r g e t t i n g a r e p r e s e n t a t i o n of A T ( E D ' , E D
)
-
(3.2.42)
a s a f a m i l y of i n t e g r o - d i f f e r e n -
t i a l operators. Example 3 . 2 . 8 . Denote by S ( E , D ) t h e s i n g u l a r p e r t u r b a t i o n , whose symbol i s t h e f o l l o w i n g
e
matrix
where 0 E ( - n / 2 , n / 2 )
is given.
The s i n g u l a r p e r t u r b a t i o n S ( E , D ) w i t h t h e symbol ( 3 . 2 . 4 3 ) f o l l o w i n g S t o k e ' s problem i n R n (E
(3.2.44)
-2 2 i 8 e -A
where u : R n
,u> -f
)u
+ v p
= f,
= 0,
Cn, p : W n +
Ax i s t h e L a p l a c i a n , B
solves the
:
x
E xtn
x
E
IRn
C a r e t o be d e t e r m i n e d ,
: En+ C is
f
i s t h e g r a d i e n t and
=
given,
au/ax.. l<j
'
One c h e c k s e a s i l y u s i n g t h e F o u r i e r t r a n s f o r m t h a t u = S (E,D I f . e x 2 -2i8 2 Since I+E e 151
(3.2.45)
z
O,V
(E,s)
ER+
x
nn,
if
e
E (-n/2,T/2),
it i s
3.3. Singular Perturbations with Variable Symbols q u i t e o b v i o u s t h a t t h e f o r m u l a (3.2.17) h o l d s f o r
235
r e p l a c e d by
E
Ee
-iB
.
I n d e e d , b o t h l e f t and r i g h t hand s i d e i n (3.2.17) a d m i t a n a n a l y t i c extension with r e s p ect t o E
E R
onto the half-plane
larg
€1
< ~/2.
Using t h e second formula (3.2.23) and t h e c l a s s i c a l R i e s z o p e r a t o r s Yk
(Example 3.2.4), one c a n r e p r e s e n t S e ( € , D ) as t h e c o m p o s i t i o n o f two
operators:
(3.2.46) where, a s For n = 3
(3.2.47)
N o t i c e t h a t f o r each 8 E (-n/2,n/2)one h a s :
(3.2.48) l i m
E - ~ S( E , D ) f ( x ) = e - 2 i e ( f ( x ) - ( Y , e
E++O
x
for
v
f
D
Y.)f)(x)), 7
E ( s ' ( R ~ ) ) "= s ~ ( R X ~ . . . X)S ' ( R " )
F u r t h e r m o r e , S ( E , D ) c a n be e x t e n d e d a s
e
(3.2.49) S B ( ~ , D x ):
x (H(s), E ( R n ) I n
+
(H(s-v),e (Rn)) n
w i t h v = (-2,0,-2),univormly w i t h r e s p e c t t o E E ( 0 . ~ ~ 1 ,v E~ < *, i . e . t h e norms of t h e o p e r a t o r s (3.2.49) a r e bounded by a c o n s t a n t which depends o n l y on
and n . I n d e e d , t h i s i s a n immediate consequence o f
E~
(3.2.43) and
Parceval's identity. Using a r e s u l t from [ M i k h . l , Z I , o n e c a n show t h a t (3.2.49) s t i l l h o l d s i n spaces spaces
H
( R n ) w i t h L -norms 1 < p < a, 5 s ),2,E( R n ) b e i n g t h e P ( R n ) ( s e e 3.12 f o r m o r e d e t a i l s ) .
(s)rP,E
H (s),E
3.3. S i n g u l a r p e r t u r b a t i o n s w i t h v a r i a b l e s y m b G W e c o n s i d e r h e r e c l a s s e s of symbols which are € u n c t i o n s of
( x , E , ~ ) and
i n t r o d u c e t h e c o r r e s p o n d i n g c l a s s e s of s i n g u l a r p e r t u r b a t i o n s which i n c l u d e s i n g u l a r l y perturbed d i f f e r e n t i a l o p e r a t o r s with v a r i a b l e c o e f f i c i e n t s ,
as w e l l . D e f i n i t i o n 3.3.1.
Let u be an open s u b s e t of cZass
S'
1,o a(x,E,c) E
R~
and Zet v
(u) i s d e f i n e d t o c o n s i s t
=
(vl,v2,v3)E R
3
.
The syrnboZ
o f a22 f u n c t i o n s
s a t i s f y i n g t h e c o n d i t i o n : f o r any compact K c u such t h a t and any couple of m Z t i - i n d i c e s a , t~h e r e e x i s t s a c o n s t a n t c
c
(UX ( o , E ~ ]x
Ka I3
.
3. Singular Perturbations on Smooth Manifolds without Bounda y
236
D e f i n i t i o n 3.3.2.
The symboZ a E
S'
1,o
(u) belongs t o t h e symbol c l a s s s'(u)
if t h e r e e x i s t s a
representation (3.3.2)
a ( x , ~ , t ; )= a ( x , ~ , < ) + r ( x , ~ , S ) 0
where a O ( x , ~ , E is ) homogeneous i n v +v2
1
(E-',s)
for
E
E ( O , c o l , 151 2 1 , of degree
i.e. - ( v +v )
a (x,t-lE,tS), 0
a0(x,E,S) = t
(3.3.3)
v
t 2 I,
v
E
E (o,EOl,
v 5 E mn, and where t h e remainder r ( x . E . 6 ) on each compact
K c
151
'
u s a t i s f i e s the
inequalities
for a l l x E K ,
E ( O , E ~ ] , E, E B~ and all m u l t i - i n d i c e s
E
a,B
P r o p o s i t i o n 3.3.3.
Let a E
Sv
1,o
(U),
b
E 'S
1,o
.
(U) Then
Proof. The s t a t e m e n t s of t h i s p r o p o s i t i o n are an immediate consequence of t h e chain rule. D e f i n i t i o n 3.3.4.
Let a E sV
op a
=
(u),( r e s p e c t i v e l y , a E sv (u)) . The s i n g u l a r p e r t u r b a t i o n
1,o a ( x , ~ , ~ is)
d e f i n e d as f o l l o v s
1,
231
3.3. Singular Perturbations with Variable Symbols
The s i n g u l a r p e r t u r b a t i o n a ( x , ~ , i~s) s a i d t o belong t o t h e c l a s s ( r e s p e c t i v e l y , t o t h e c l a s s op
op S;,,(U)
Sv(U)).
Theorem 3.3.5.
If
a
E
m
s ~ , ~ ( U ) ,t h e n a ( x , c , D ) : C o ( U ) + C"(Ui
a ( x , ~ , ~: )E ' ( u ) E"
-t
D'(u),
m 1 a(x,E,D) : Co(U)
+
V
E
( 0 , ~I . 0
E
and E"'a(x,E,D)
Cm(U)
and i t can be eztended a s
Moreover, : E'(U)
-t
D'(U)
E
(E
(O,En])
are e q u i c o n t i n u o u s . Proof. If a
rn
E sv
1,o
(U), u E C o ( U ) ,
then t h e function
Evla(x,E,D)u(x) = (2nl-n
e
i < x , S > '1
mn
E
a (x,E ,5)
( 5 )d5
c a n b e d i f f e r e n t i a t e d under t h e i n t e g r a l s i g n and t h e c o r r e s p o n d i n g i n t e g r a l s are a b s o l u t e l y and u n i f o r m l y c o n v e r g e n t w i t h respect t o x on any compact K c U , Further, l e t v
E
E
E
(O,E~]. m
The i n t e g r a t i o n by p a r t y i e l d s :
Co(U).
Using t h e l a s t i n e q u a l i t i e s , one shows t h a t t h e f u n c t i o n a l s
are w e l l d e f i n e d , i f u
E
V
E'(U),
E
E
( O , E ~ ] and, moreover, t h i s f a m i l y of
f u n c t i o n a l s i s equibounded i n t h e s e n s e o f D e f i n i t i o n 1.2.33.
In f a c t ,
w r i t i n g formally (3.3.12)
<E v
la ( x , E , D ) u , v > = ( 2 ~ ) I- ~ I
v(x)E"1a(x,E,S);(S)ei<x'5>
dsdx =
IRn lRn =
I ~ ( c (I ) mn mn
(2.rrl-n
and u s i n g ( 3 . 3 . 1 0 ) , defined f o r V u
E
Ev
1 i<x,S> a(x,E,<)v(x)e dx)dS
one g e t s t h e c o n c l u s i o n t h a t t h e l a s t i n t e g r a l i s w e l l
E'(U),
s i n c e t h e family o f f u n c t i o n s
i s r a p i d l y d e c r e a s i n g when
5
+
m
uniformly with r e s p e c t t o
E
E (O,E~],
3. Singular Perturbations on Smooth Manifolds without Bounda y
238
f o r any u E E ' ( U )
so t h a t t h e f a m i l y of f u n c t i o n a l s (3.3.11)
bounded w i t h respect t o
E
E (O,E
0
1.
i s uniformly
I
Remark 3.3.6.
(0) (0)
1
L e t a E S v ( " ) ( U ) where i n t e g e r v . t 0, j = 2,3, and l e t E a ( x , E , c ) be 3 polynomial i n ( ~ ~ 5 L)e .t a , ( x , E , S ) b e t h e p r i n c i p a l symbol o f a ,
a-a,
,
(0) ( 0 ) > v(l)+v(l) 1 2 v 1 +v2 I f a t e a c h s t e p o f t h i s p r o c e s s t h e remainder r k = (a-aoa ) E k- 1 (k) ( U ) , t h e n t h e symbol a ( x , E , S ) c a n b e r e p r e s e n t e d a s a f i n i t e sum:
where, of c o u r s e , u(O) > v(l) and moreover,
...-
Sv
a ( x , ~ , C )=
(3.3.13)
C ak(x,E,<) O
where ak have t h e v e c t o r i a l o r d e r v ( ~ ) a, r e homogeneous i n d e g r e e V : ~ ) + V ; ~ ) and, moreover, t h e o r d e r s v ( ~ ) ,0 5 k
=< N,
(E
-1
, 5 ) of
form a
s t r i c t l y d e c r e a s i n g sequence i n t h e f o l l o w i n g s e n s e :
The s p l i t t i n g (3.3.13)
i n t o a sum of homogeneous t e r m s ak, whose o r d e r s
v ( k ) s a t i s f y (3.3.14),
c a n be viewed a s a n a t u r a l g r a d u a t i o n of a ,
a n a l o g o u s t o (3.1.2). Of
c o u r s e , any symbol a
which i s w e l l d e f i n e d f o r
E
h a s a z e r o t e r m a.
Sv(U)
E (O,E~],
E
e x t e n d e d a s a homogeneous f u n c t i o n o f
of graduation,
151 2 1, x E U and can b e u n i q u e l y -1 , 5 ) E IR+x (lRn\{Oj) f o r V x E
(E
U
Example 3.3.7. The symbol a ( x , E , C ) c(x) : U
-f
C,
= E
2
j5j4+1512++c(x), w i t h b ( x ) : U
I t admits a n a t u r a l g r a d u a t i o n :
a (x,E,~) 0
+
Cn,
smooth f u n c t i o n s , b e l o n g s t o S V ( O ) ( U ) w i t h v ( O ) = (0,2,2).
2 = E
a
=
4
a o + a l + a 2 , where
IS/ +l5I2, al(x,E,5)
v ( O ) = (0,2,2), v ( l ) = (O,l,O), v")
= cb(x),C>, a2(x,€,S) = c ( x ) ,
= (O,O,O),
t h e c o n d i t i o n (3.3.14)
being
3.4. Continuity of Singular Perturbations
239
obviously s a t i s f i e d .
~ ~ / < / ~ + 1 < 1b ~e l o+n g~s ~ / c 1 ~
On t h e o t h e r hand, t h e symbol a ( x , ~ , E ; )= ~n ( R n ) , with v = ( 0 , 2 , 3 ) b u t does n o t belong t o S ( R ) t o Sv 1# O admit a n a t r u a l g r a d u a t i o n .
and d o e s n o t
3 . 4 . C o n t i n u i t y of s i n g u l a r p e r t u r b a t i o n s i n Sobolev s p a c e s of v e c t o r i a l o r d_ er _ The a i m of t h i s s e c t i o n i s t o p r o v e t h e f o l l o w i n g Theorem 3.4.1.
L e t a ( x , E , c ) E sV
1,O
the
( m n ) and Let
a ( x , s , c ) E s(IR:). Then f o r
v
s E R
3
foZlowingfamiLy of Linear mappings i s equicontinuous ( s e e D e f i n i t i o n 1 . 2 . 3 3 )
(3.4.1)
a ( x , E , D ) : H ( s ),E (7Rn)
-f
H(S-V)
,E
(Rn)
,
(o,Eol.
E
Proof. F i r s t assume t h a t s = v = 0 , s i n c e t h e argument i n t h i s c a s e i s p a r t i c u l a r l y simple.
The
assumption
a ( x , E , S ) E So 1no
(mn) n
S ( m n ) yields the
inequalities
/ a ( r l , ~ , S I) 5 CN
(3.4.2)
v
N
since
naa(n,~,E)
D"a(x,E,S). x-trl x
= F
Therefore,
II
(3.4.3)
I
~ ( ~ , E , D ) (~0 I) , E
Since a(x.E.5)
= F
-1 a(rl.E.5) n+x
5
c ~ < ~ >I U-I ~ I ( oI)
and j e x p ( i x . n )
/
;E
G
+
5 1 one g e t s u s i n g ( 3 . 4 . 3 ) :
i s equivalent t o the inequality
Now c o n s i d e r any s and v . Then ( 3 . 4 . 1 )
where
V N.
L2(IRn),
5
KG i s a n i n t e g r a l o p e r a t o r w i t h t h e k e r n e l
and t h e c o n s t a n t C d o e s n o t depend on
E
and u.
Again t h e a s s u m p t i o n s on a ( x , E , S ) imply t h a t
:
3 . Singular Perturbations on Smooth Manifolds without Bounda y
240
The f o l l o w i n g i n e q u a l i t y h o l d s :
<E>T
(3.4.7)
4 21TI<E-T)>lTl
(3.4.7) h o l d s with (3.4.7)
with
Thus, (3.4.8)
N
T
and t h a t i m p l i e s t h a t
< 0 , one s w i t c h e s
5 and n and a p p l i e s
> 0.
-T
(3.4.7)
(3.4.6), /K(~,E,TI)
where C
Z 0. I f
T
2 4 2
<5>'
I n d e e d , e v i d e n t l y one h a s :
E W.
t/ T
I
lead t o the inequality:
5 CN<S-n>-N,
depends o n l y on N ; h e r e N = N
-1
1
s
2
-v
2
I-1
s
3-v 3
I.
T h e r e f o r e , f o r V N one h a s :
where f ( 5 ) = C
N
<5>-N and f *
N
g s t a n d s f o r t h e c o n v o l u t i o n of f and g .
Applying t h e i n e q u a l i t y :
and t a k i n g N > n , one g e t s t h e c o n c l u s i o n t h a t t h e f a m i l y o f i n t e g r a l 2 n o p e r a t o r s w i t h k e r n e l s K ( C , E , ~ ) are c o n t i n u o u s i n L ( W ) and t h e i r norms
a r e u n i f o r m l y bounded w i t h r e s p e c t t o
E
E (O,E
0
1.
I
C o r o l l a r y 3.4.2.
Then t h e foZZowing f a m i l y o f l i n e a r mappings i s equicontinuous for each 3 s = (Sl,S2,S3) E R : a(X,E,D)
In f a c t , a ' ( x , E , D )
: II(s)
( S )P E
u n i f o r m l y bounded i n am(E,D)
,E(IRn)
: H
+
(Wn)
13
(s-u) , €
H(S-V) ,E
(2) , E (IR
E
(O,Eo1.
) and have t h e i r norms
E ( 0 ,3 ~a c c o r d i n g t o Theorem 3.3.1,
E
O
: H
n
(IR )
( s ) r E ( R n ) + H(S-V) ,E a s f o l l o w s from P r o p o s i t i o n 3.1.6.
while
uniformly with r e s p e c t t o
D e f i n i t i o n 3.4.3.
A singtilar pel-tilrbation
a(X,E,D)
v = ( v ~ P ~ ~ i ,f "f o~r )each
a(x,E,D)
: H
( S ) ,E
given
i s s a i d t o have v e c t o r i a l order E~
m
the family
( I R R n ) + H(S-V)
is equicontinuous and, moreover, f o r each
,E
(Wn),
(E
E
(o,Eol)
< v the mppings
E
E (O,E~I,
3.5. Pseudolocality of Singular Perturbations (Rn)
a(x,E,D) : H ( s )
H
+
(s-v)
o r have unbounded norms when
E
(E
24 1
E (O,E?I) e i t h e r are n o t continuous
,E
0.
+
C o r o l l a r y 3.4.4.
A f a m i l y o f p e r t u r b a t i o n s a ( x , E , D ) has order v E R
and a ( x , ~ , c )$?
S'
1 ro
( n n ), V
3
.
zf a ( x , E , c ) E s"
1,o
p < v.
(Rn)
Example 3.4.5. with A t h e Laplace o p e r a t o r , belongs
( i )A f a m i l y of p e r t u r b a t i o n s E'A'-A
( R n ) with V
t o S'
>
LI
and i t s o r d e r i s
(0,2,2)
1r o ( i i )A f a m i l y of p e r t u r b a t i o n s
a
1
E
= (0,2,2).
(x)D D with a ( x ) E C z ( R n ) kj k 1 kj
lsk,j
belongs t o S
Remark 3.4.6. Using t h e s a m e argument a s i n t h e p r o o f of Theorem 3.4.1,
one shows t h a t
m
(U) and g E C o ( U ) , t h e n H (u) 3 u + g O p a u E i f a ( x , E , E ) E S" 1,o ( s ) ,E H f U ) i s a f a m i l y of c o n t i n u o u s l i n e a r mappings, whose norms a r e ( s - v ) ,E u n i f o r m l y bounded w i t h r e s p e c t t o E E ( O , E ~ ] . I n o t h e r words, f o r each E~
> 0, s
E
R 3 t h e f o l l o w i n g f a m i l y o f l i n e a r mappings i s e q u i c o n t i n u o u s :
a(x,E,D) where
E +
gu E H
3.5.
H
1oc
H
:
loc
(u)
( s ) ,E
+
H ( s - " ), E
(U),
( U ) i s t h e f a m i l y of s p a c e s c f a l l f u n c t i o n s u s u c h t h a t
( s ) ,E
(U) f o r ( 5 ),E
v
g
E
m
Co(U).
The p s e u d o l o c a l i t y of s i n g u l a r p e r t u r b a t i o n s
W e s t a r t with t h e following
D e f i n i t i o n 3.5.1.
A f a m i l y of d i s t r i b u t i o n s
on an open s e t
v 5u
(O,E~I3
E + U(E,X)
i f f o r any g i v e n
E
E D' (u) i s s a i d t o be smooth
E ( 0 ,0 I~ it c o i n c i d e s idith some
) f o r any compact s e t f u n c t i o n u ( x ) E c ~ ( v and,moreover, (3.5.1)
max xEK
1
I D ~ ~ ( E , X )5
where t h e c o n s t a n t s
C
K,a
c ~ , ~ v ,a, v
KCC
do n o t depend on
E
KCC
v
one has:
v
-
E ( 0 , ~ 01
Definition 3.5.2.
The complement of t h e maximal open s e t on u h i c h a f a m i l y o f d i s t r i b u t i o n s E + U(E,X)
i s smooth, i s c a l l e d t h e s i n g u l a r support of u and denoted by
s i n g supp u ( E , x )
.
3. Singular Perturbations on Smooth Manifolds without Bounda y
2 42
Example 3.5.3. IR+ 3
The f a m i l y o f d i s t r i b u t i o n s :
singular support t h e o r i g i n x
=
E
+
Eaexp(-/xl
2
/E
2
)
E S ' (Xn) h a s as i t s
0 , V a E IR.
Our p u r p o s e i s t o prove t h e f o l l o w i n g Theorem 3.5.4.
Let ( o , E 3 E~ + I u(x,E) (3.5.2)
E
(u) and l e t
€ 1
(u). Then a ( x , E , ~ )E S" 1,o
s i n g s u p p ( ~ " a ( x , ~ , D ) u ( ~ , x5 ) ) s i n g supp u ( E , x ) .
Proof. W e associate w i t h E"la(x,E,D)
which i s a f a m i l y of
its kernel K(E,x,Y),
distributions
d e f i n e d by t h e r e l a t i o n
According t o Theorem 3.3.5,
the kernel-distribution m
f a m i l y of c o n t i n u o u s l i n e a r maps K : C o ( U ) W e are g o i n g t o show t h a t t h e f a m i l y (O,E
0
+
1
i n (3.5.3)
Cm(U) and
3
E +
K
:
defines a
E' (U)
+
D ' (u).
K ( E , x , ~ ) i s smooth o f f t h e
d i a g o n a l x = y i n UxU. Indeed, u s i n g (3.5.3)
and t h e d e f i n i t i o n of s i n g u l a r p e r t u r b a t i o n s ,
one f i n d s e a s i l y t h a t
Consequently,
The l a s t f o r m u l a f o r l a / > u +u +n c a n b e r e w r i t t e n as a c o n v e r g e n t i n t e g r a l 2 3
a n d , a s a consequence o f t h e u n i f o r m convergence of t h e i n t e g r a l on t h e r i g h t hand s i d e , one g e t s t h e c o n c l u s i o n t h a t ( x - ~ ) ~ K ( E , x , ~ E) C J ( U W ) i f
/ a / > v 2 +v3 +n+j. Moreover, a l l t h e d e r i v a t i v e s up t o t h e o r d e r j of (~-y)~K(~,xa ,y r e) u n i f o r m l y bounded w i t h r e s p e c t t o E E ( 0 . ~ ~ 1T. h a t p r o v e s t h e claim t h a t
E
+ K(E,x,~)
h a s as i t s s i n g u l a r support a c l o s e d set
243
3.6. Asymptotic Expansions of Symbols belonging t o t h e diagonal x = y i n UXU. I n o r d e r t o end t h e p r o o f of t h e theorem, w e need o n l y t o a p p l y t h e following Lemma 3.5.5.
Let
( O , E ~ ] 3 E + K ( ~ , x , y )E
D ' ( u x u ) . Suppose t h a t t h e famiZy of Zinear
maps m
m
(3.5.7)
K : Co(U)
+
C (U), K : E ' ( U )
is u n i f o r m l y bounded i n t h e diagonal in
E
+
D'(U)
E (O,c0], and, moreover, t h a t K is smooth off
UXU.
Then s i n g supp Ku c - s i n g supp u, V u E E ' ( U )
.
Proof of Lemma 3 . 5 . 5 . Let
v
and W b e open s u b s e t s of U s u c h t h a t W C C V. m
Let $ E
co(v),
Let
u E E ' ( U ) b e such t h a t u i s smooth i n V . Then Ku = K ( $ u ) + K ( ( l - $ ) u )
E -t
$
1 i n a neighborhood of t h e c l o s u r e of W. m
Since E
E +
$u i s smooth on U , and E
E (O,E~],
K(@)
$u E C (U) u n i f o r m l y w i t h r e s p e c t t o
+
0
i s smooth on U . On t h e o t h e r h a n d , w e c a n w r i t e
K ( ( l - $ ) u ) as an i n t e g r a l K((l-$)u) =
1
K(E,x,Y)
(I-$(y))u(y)dy.
U I f x E W , y E s u p p ( 1 - $ ( y ) ) , t h e n ( x , y ) b e l o n g s t o t h e complement o f some s u i t a b l e neighborhood of t h e d i a g o n a l i n U X U , t h a t i s ( x , y ) b e l o n g s t o a m
s u b s e t where K ( E , x , ~ ) i s smooth ( C
in (x,y) with a l l the derivatives
E ( 0 ,0 ~1 ) . Hence, K ( ( 1 - $ ) u ) i s smooth on V a s w e l l . S i n c e V w a s any open s u b s e t of U, one g e t s t h e c o n c l u s i o n
u n i f o r m l y bounded w i t h r e s p e c t t o
t h a t s i n g supp Ku C_ s i n g supp
E
U.
1
3 . 6 . Asymptotic Expansions of symbols A s it h a s been mentioned i n Remark 3 . 3 . 6 ,
admit a n a t u r a l g r a d u a t i o n
(3.3.13),
some p o l y n o m i a l i n (E,E)
symbols
(3.3.14).
One c a n c o n s i d e r f o r m a l i n f i n i t e sums o f symbols of t h e form ( 3 . 3 . 1 3 ) whose o r d e r s s a t i s f y ( 3 . 3 . 1 4 ) . One o f t h e b a s i c m o t i v a t i o n s t o c o n s i d e r such f o r m a l sums i s t h e f a c t t h a t t h e y a p p e a r i f one expands t h e i n v e r s e symbols t o e l l i p t i c polynomial symbols of t h e form ( 3 . 3 . 1 3 1 ,
(3.3.14)
into
a n a s y m p t o t i c series. On t h e o t h e r hand, it i s c o n v e n i e n t t o have a symbol f u n c t i o n f o r which a g i v e n f o r m a l sum of symbols c a n b e c o n s i d e r e d a s an
3. Singular Perturbations on Smooth Manifolds without Boundary
244
a p p r o p r i a t e l y i n t e r p r e t e d asymptotic expansion. Theorem 3 . 6 . 1 .
i)
L e t a . E syl0 ( u ) , v (1")
< v(j), j
7
v (11 ) J.
(3.6.1)
-a
0,1, and l e t
=
for j + ( 01
Then t h e r e e x i s t s a E s V
1.0
a(x,E,c) -
(3.6.2)
such t h a t (N) (U), V N t 0
C a . ( x , ~ , E ;E ) Sv 1r o O=<j
and moreover, f o r each p a i r of m u l t i - i n d i c e s a , @ and every compact s e t K c U
t h e f o l l o w i n g asymptotic r e l a t i o n s hold:
unifomnZy w i t h r e s p e c t t o (x,c;) E
K x
mn.
Proof. L e t + ( t )E
crn(K+) ,
Let K .
C K
K
c
o
E
...,
for t E
[o,tI, 4
z 1 for t t I.
b e a sequence of compact s e t s e x h a u s t i n g U, i . e .
3' 1 2 t h e union o f a l l K . c o i n c i d e s w i t h U. 3 We choose 6, so small t h a t
f o r V j t 1, V
E
E ( 0 . ~ ~ 1V, 5 E mn,
V x E Krr
v
a,B, r w i t h
l a i + l f I + r5 j .
T h i s i s p o s s i b l e s i n c e v!J)
+
-m.
I n f a c t , one h a s
Let
W e choose 6,
> 0 t o b e so s m a l l t h a t
I n t r o d u c i n g t h e numbers m. =
max t +lt
1
(1-1) 1
+ ( t I)
one c a n s a t i s f y t h e i n e q u a l i t y ( 3 . 6 . 6 ) by c h o o s i n g 6 . > 0 s u c h t h a t 7
3.6. Asymptotic Expansions of Symbols
245
Y. w i t h y , = ( v ~ J ) - v ~ J - ' ) ) - <' 0 .
6 . 4 ( 2 J C .m . ) 3
3
3
3
Moreover, w e r e q u i r e :
6.
1
+
j +
for
0
m.
Now we d e f i n e t h e symbol a ( x , ~ , E )t o be (3.6.7)
j
a(x,e,E)
-1
@ ( 6 . ~ ) a .( x , E , S ) ,
C
=
j>o
3
V (x,E,S)
E u
x
( 0 . ~ ~x 1R ~ .
One c h e c k s e a s i l y t h a t f o r a ( x , E , C ) d e f i n e d by ( 3 . 6 . 7 ) w i t h 6 . 3 s a t i s f y i n g (3.6.4), the r e l a t i o n s (3.6.2), (3.6.3) hold.
+
0,
+ m,
I n f a c t , i t i s enough t o check ( 3 . 6 . 3 ) s i n c e ( 3 . 6 . 2 ) i s a consequence of
( 3 . 6 . 3 ) . One g e t s u s i n g ( 3 . 6 . 5 ) w i t h j
for V
E
E (0,6N-1],
E
V x
Kr,
V
=
N and ( 3 . 6 . 4 ) :
5 E Rn , where t h e c o n s t a n t s C'
a,B,N,j
depend on t h e i r s u b s c r i p t s , and t h i s i s p r e c i s e l y ( 3 . 6 . 3 ) .
(3.6.8)
+
~ ( 1 ) 2
-m
for
Then t h e r e e x i s t s a E
Sv
j
--t
might
I
m.
(0) (U) such t h a t
1t o
(N)
(3.6.9)
a ( x , ~ , S -)
a.(x,E,S) 06j
E Sv
1t o
(U), V N t 0 ,
and moreover, for each p a i r o f m u l t i - i n d i c e s
uniformly w i t h r e s p e c t t o
E
a,B
and each compact K c u
E ( 0 . ~ ~ 1x , E K
Proof. The f u n c t i o n $
for
I a1 + / B I +r
E Cm(F+) b e i n g t h e same a s i n t h e p r o o f o f Theorem 3 . 6 . 1 ,
S j, x
E
Kr,
Kr,
r
=
sequence of compacts a s p r e v i o u s l y .
0,1,.
. ., b e i n g
t h e same e x h a u s t i n g U
3. Singular Perturbations on Smooth Manifolds without Boundary
246
S i n c e v ~ J )J-
for j +
-m
m,
one c h e c k s e a s i l y , u s i n g t h e L e i b n i t z
f o r m u l a and t h e same argument a s i n t h e p r o o f o f Theorem 3 . 6 . 1 ,
t h a t such
a c h o i c e of 6 . i s p o s s i b l e . A f t e r w a r d s , one d e f i n e s a ( x , ~ , S )t o b e : 1
I
and ( 3 . 6 . 9 ) , ( 3 . 6 . 1 0 ) c a n b e checked a s p r e v i o u s l y . [ v ] = v2+v3, \ v [
W e r e c a l l the notation:
=
vl+v2, V v
=
(vl,V2,v3) E R
3
.
Theorem 3.6.3.
Let a . E 3
(3.6.13)
"yl0
j), v
[v")]
iv
(I+') C
--,
'j),
for
j
j +
=
o , ~ ... , and l e t
m.
(0)
Then t h e r e e x i s t s a E a ( x , ~ , E -)
(3.6.14)
Sv
( U ) such t h a t
I:
a . ( x , ~ , F )E Svl,o
1 to
(N)
, V N
2 0
OSj
and, moreover, for each p a i r of m u l t i - i n d i c e s
a,B
and each compact
K c
u
t h e f o l l o w i n g asymptotic r e l a t i o n s h o l d : (j) (3.6.15)
E"
'
I: a , ) O<j
I-lalD6Da(a-
uniformly w i t h r e s p e c t t o
E
E
= O(lEC1
["(N) ] ) f o r / € E l
-f
m
( O , E ~ ] ,x E K .
Proof. __ With t h e same c h o i c e of t h e f u n c t i o n $ for j
-f
m
E
Cm(%
)
a s p r e v i o u s l y and 6 . C 0 , 1
such t h a t
f o r / a / + l B l + rS j , x E K r ,
K
1
c K
2
c
. _ .e x h a u s t i n g
U, one d e f i n e s a t o b e :
-
Xa. and c a l l Za. t h e f o r m a l symbol, i f a . have o r d e r s 3 I 3 s a t i s f y e i t h e r ( 3 . 6 . 1 ) o r ( 3 . 6 . 8 ) o r (3.6.13) and Ea, i s a s y m p t o t i c a l l y 3 c o n v e r g e n t t o a i n t h e s e n s e of (3.6.3) o r (3.6.10) o r ( 3 . 6 . 1 5 ) , r e s p e c t i v e l y . We shall w r i t e a
D e f i n i t i o n 3.6.4. (0)
The symbol a E sV
(0) (u) i (j)
s s a i d t o belong t o t h e c l a s s K~
(u) i f t h e r e
.
(UT) w i t h v (O) > v ( l ) t...> v ( J ) t . . , 1 u (1)1 c --, e r i s t symbols a . E s" 7 1.0 1 [ v ( J ) ] c -m f o r j + m, which a r e homogeneous of degree / v ( J ) I i n (E- , S ) ' g o
3.6. Asymptotic Expansions of SymOols for
E
x n , 151 t
E IR+, 5 E
I , i.e. ( j1
-1
a.(x,t 3
(3.6.18)
~ , t 5 )= t
241
+”( j )
v1
2
a . (x,E,t), 3
v
t t 1,
v
(E,<)
151
E
R+X
IR~,
2 1,
and such t h a t (N) (3.6.19)
(a
-
a,) E
C
S1)
(U),
V N 2 0.
O<j
With a .
u
-1 (E
x
,5)
t h e e x t e n s i o n of a . ( x , E , < ) from t h e s e t
( x , E , ~ )
70
R+
x
x
(mn\{
15 I
3
a s a homogeneous f u n c t i o n in v a r i a b l e
< 1))
R+ x ( I R ~ \ { o I o) f degree ( v ( J ) l , a ,
30
E
C m ( U x R+ x ( I R ~ \ { o I ) ) ,
t h e formal sum (3.6.20)
1 a. j t o 3’
(x,E,<)
i s c a l l e d t h e graded symbol and i s denoted b?j
ar(x,€,<).
D e f i n i t i o n 3.6.5.
A function a 0 ( x , € , t )
:
symbol c l a s s HV(u), v
u
=
x
R+
x
(Rn\{Ol)
(vl,v2,v3) E
i s homogeneous i n ( E - ~ , < )E R+
a0
f o r each p a i r o f m u l t i - i n d i c e s
where t h e c o n s t a n t s c
clBK
x
m3
+
c i s s a i d t o belong t o t h e
i f a O E c m ( ux R +
x
( n n \ 1 0 ) ) of degree l v l
(nnl{o})), =
vl+v2 and
a,B and each compact s u b s e t K c u h o l d :
depend only on t h e i r s u b s c r i p t s .
Remark 3.6.6. I t i s immediate t h a t t h e symbol a . i n D e f i n i t i o n 3.6.4 b e l o n g s t o 10 Hv(I) (U). I n f a c t , one h a s o n l y t o check t h e v a l i d i t y of (3.6.21) w i t h
v
=
v (I),
( E -1 , < )
which i s a consequence of (3.3.1) and t h e homogeneity of a , i n JO E R + x ( n n \ { 0 } ) of d e g r e e Iv(1)/.
Theorem 3.6.7.
L e t a E KV(’) (u) and l e t ar b e i t s graduate symbol. Then f o r V f E c:(u), t/
( x , ~E )
u
x
( R ~ \ { O } ) t h e following asymptotic f o m l a holds: -1
(3.6.22)
e-iE
iE-l<x,n>f(x)) < X ’ n > a ( x , c , D )( e
where a ( x , ~ , D )= O p a .
3. Singular Perturbations on Smooth Manifolds without Boundary
248
Moreover if K1 c u ,
K2
c (lRn\iO}) a r e any compact s u b s e t s and B is
any bounded s e t in c;(u), then ( 3 . 6 . 2 2 ) hoZds uniformly w i t h r e s p e c t t o E K~
(X,S)
x K~
and f E
B.
Proof. __
One h a s a c c o r d i n g t o t h e d e f i n i t i o n of a s i n g u l a r p e r t u r b a t i o n a ( x , E , D ) : -iE
-1
<x,r~>,(~
,E,D) (e
i E
-1
<x
'n>f(x)) =
-1 =
(2n)-n
.i<x,S-~
'>a(x,E,S)f(S-E
-1
q)dg =
mn =
(2n)-n J
e
i < x , E>
a (x,E
,S+E
-1
rl)
2 ( 5 )d5.
lRn
We a r e g o i n g t o u s e T a y l o r ' s f o r m u l a (3.6.23)
-1 a ( x , ~ , c + n) ~
where a ( a )
=
aaa
5
=
'a1 a
i'
1
c
=
a(")
-1 (X,E,E
n)cn
+ qN(x,E,5,q)
(0)
l 4 < N
D a and qN i s t h e r e m a i n d e r . S i n c e a E
5
Kv
(U) one
has def (3.6.24)
rN
=
(N)
(a-
z a,) OSj
E
S'
(u),
1 ,O
Therefore,
F u r t h e r , one h a s : (3.6.26)
if
E
a!") ( x , E , E - ' ~ ) 3
=
a!")
(X,E,E
-1 n)
=
E l " l - l v ( J ) ' a j(O a )( x , l , n ) ,
10
E (0,InjI. Using ( 3 . 6 . 2 5 ) ,
( 3 . 5 . 2 6 ) one g e t s t h e f o l l o w i n g a s y m p t o t i c r e l a t i o n
which means t h a t f o r V N t h e remainder
N
(X,E,E
-1
n)
d e f i n e d by ( 3 . 6 . 2 4 )
s a t i s f i e s (3.6.25). I n o r d e r t o f i n i s h t h e p r o o f w e have o n l y t o e s t i m a t e t h e t e r m
3.6. Asymptotic Expunsions of Symbols corresponding to 'the remainder q
249
N'
where 6 E ( 0 , l ) Denote
and let
depends only on its subscripts. where the constant C arK1tK2 Further for 2 ~ 1 5 12lrl and la1 (0) S" (U): 1to
=
N 2
via)
one gets, using one more
time the fact that a E la
(01)
(X,E,E
-1
-V
rl+65)j 5 c
(0)
E
Therefore, one finds for Q (f):
so that the last integral in the right hand side of (3.6.29) is
exponentially small when
E
-f
0:
(0)
3 . Singular Perturbations on Smooth Manifolds without Bounda y
250
S i n c e y depends o n l y on t h e d i a m e t e r of supp f , t h e i n t e g r a l (3.6.30) is e x p o n e n t i a l l y s m a l l u n i f o r m l y w i t h r e s p e c t t o f i n a g i v e n bounded s e t B
c C m ( U ) and w i t h r e s p e c t t o 0
T'
1
E K2 c U , and t h a t ends t h e p r o o f .
Remark 3.6.8.
E S v ( U ) t h e n t h e same argument as i n t h e p r o o f of Theorem 3.6.7.
If a
shows t h a t
-1
(3.6.31)
e-iE-'<xfn>
a ( x , E , D ) ( ei
= E-"'aO(X,l,T1)
f o r ri f R n , In1 2 1, x E K
C
<x*ri>f ( x ) )
E
+
o(E1-'v'),
U, f E B C C ; ( U ) ,
=
E +
0,
where a 0 ( x , E , C ) i s d e f i n e d
by (3.3.2), (3.3.3). One h a s t h e same a s y m p t o t i c formula (3.6.31) a l s o f o r ( u n i f o r m l y w i t h r e s p e c t t o q i n any g i v e n compact K2
C
V r- E IRn\{O]
IRn\{O}), b u t , o f
c o u r s e , a o ( x , E , C ) s h o u l d b e r e p l a c e d i n t h i s c a s e by i t s e x t e n s i o n t o
U
x
R+ X ( R n \ { O j ) as a homogeneous f u n c t i o n of (c-l,C) of d e g r e e
1\11
v +v
2'
1
3.7. Amplitudes, A d j o i n t s and P r o d u c t s of S i n g u l a r P e r t u r b a t i o n s W e s t a r t w i t h i n t r o d u c i n g c l a s s e s of s i n g u l a r p e r t u r b a t i o n s , which
w i l l l o o k a s more g e n e r a l , b u t i n f a c t , c o i n c i d e w i t h t h e c l a s s e s d e f i n e d p r e v i o u s l y i f some n a t u r a l c o n d i t i o n s a r e s a t i s f i e d . D e f i n i t i o n 3.7.1.
The f u n c t i o n a ( x , y , E , < ) E cm(u x u order v
=
u
e x i s t s a constant c
v
( O , E ~ ]x
nn)i s c a l l e d amplitude of
(vl,v2,v3) E m3 and i s said t o belong t o t h e c l a s s s v
i f f o r each compact K c
for
x
( x , y ) E K,
v
u and
x
> 0
(E,c)
The s i n g u l a r p e r t u r b a t i o n
(U
x
U)
there
such t h a t
E (o,cO1 A~
1 ,O
each t r i p l e of i n d i c e s a,B,y,
x
nn.
w i t h an amplitude a E S y , o ( U
x
U) i s d e f i n e d
a s t h e double i n t e g r a l
f o r V u E C:(U), in
<.
where t h e i n t e g r a t i o n i s done f i r s t i n y and afterwards
25 1
3.7. Amplitudes, Adjoints and Products m
m
The p r o o f o f Theorem 3 . 3 . 5 . shows t h a t A : Co(U) + C ( u ) , v E E ( o n E O 1 , "1 m m AE : C o ( U ) + C (U) a r e u n i f o r m l y b o u n d ed with r e s p e c t t o E E (O,soI. and t h a t E
its distributional
One can a s s o c i a t e w i t h t h e s i n g u l a r p e r t u r b a t i o n A kernel A(x,Y,E)
which c a n be f o r m a l l y d e f i n e d a s t h e f o l l o w i n g F o u r i e r
inteqral ~ ( x , y , E )= ( 2 7 1 1 - ~ i
(3.7.3)
mn
e
i<x-y
,5>
a (x,Y, E t S)dS.
A p r e c i s e d e f i n i t i o n o f t h e d i s t r i b u t i o n A ( x , y , ~ )c a n b e g i v e n , i f w e ,
i n s t a n c e , i n t e r p r e t e t h e r i g h t hand s i d e o f
for
(3.7.3) a s an o s c i l l a t o r y
i n t e g r a l i n the following sense. We may w r i t e ( s t i l l f o r m a l l y ) V N E Z+:
Now we u s e t h e L e i b n i t z f o r m u l a (see C h a p t e r 1 ) :
-
w i t h a s u i t a b l e cho c e o f t h e c o e f f i c i e n t s q
(3.7.6)
n,B'
( 3 . 7 . 5 ) a l l o w u s t o r e w r i t e A ( x , y , ~ )a s f o l l o w s :
Hence ( 3 . 7 . 4 ) , A(x,y,S
where
a r e w e l l defined continuous functions of
(x,y) E U
X
U, V
E
E ( 0 . ~ ~i f1 N
i s c h o s e n l a r g e enough.
Now,
(3.7.6),
( 3 . 7 . 7 ) can be t a k e n a s a d e f i n i t i o n of t h e d i s t r i b u t i o n
t h e d e r i v a t i v e s Dh i n ( 3 . 7 . 6 ) b e i n g i n t e r p r e t e d i n d i s t r i b u t i o n a l
A(x,Y,E),
sense. Example 3 . 7 . 2 . 2 2 2 -1 L e t a ( x , Y , E . t ) = (1+E q ( x , y ) c ) where q ( x , y )
E
m
C
(U
a E S"(U
X
U) w i t h v
=
(0,0,-2).
One f i n d s u s i n g
x
-'.
some c o n s t a n t qo > 0 t h e i n e q u a l i t i e s qo < q ( x , y ) < qo
U) s a t i s f i e s w i t h Obviously,
Example 3 . 2 . 2 .
that the
c o r r e s p o n d i n g d i s t r i b u t i o n a l k e r n e l A ( x , y , ~ )i s g i v e n by t h e f o r m u l a
3 . Singular Perturbations on Smooth Manifolds without Boundary
252
-2 where ho ( E , x ) i s g i v e n by ( 3 . 2 . 2 3 ) . m
The s i n g u l a r p e r t u r b a t i o n A
m
: Co(U)
+ C
(U) w i t h k e r n e l d i s t r i b u t i o n
(3.7.8), A u(x) =
I
A(x,y,~)u(y)dy,
U can not. b e r e d u c e d t o a s i n g u l a r p e r t u r b a t i o n of t h e form a ( x , ~ , D )w i t h symbol a ( x , e , S ) .
However, it c a n b e r e p r e s e n t e d a s a sum (3.7.9)
=
where a ( x , E , D )
a
N
(x,E,D)
+ RN,
,
V N > O
i s t h e s i n g u l a r p e r t u r b a t i o n w i t h t h e symbol
has i t s amplitude i n t h e c l a s s and t h e s i n g u l a r p e r t u r b a t i o n R (N1 N, E S" ( U X U) w i t h "(N) = (-N,O,-2-N). 1,o Moreover, t h e r e e x i s t s a s i n g u l a r p e r t u r b a t i o n a(x,E,D) w i t h t h e a r a d e d svmbol
and a s i n g u l a r p e r t u r b a t i o n R
with amplitude r ( x , y , E , S ) E S
f o r V u E E'(U) one h a s REuECm(U)and A
=a(x,E,D)+R
. Infact,one
R u =O(sm)
u s e s (3.6.7)
when E + O )
(--,
0 ,- m )
1, o
(i.e
such t h a t
i n o r d e r t o c o n s t r u c t a symbol
a ( x , E , C ) h a v i n g ( 3 . 7 . 1 0 ) a s i t s g r a d u a t e symbol. W e s h a l l show a l l t h i s l a t e r i n a more g e n e r a l s i t u a t i o n . T h e r e i s one i m p o r t a n t c l a s s o f s i n g u l a r p e r t u r b a t i o n s i n O p S " ( U which c a n be r e d u c e d t o t h e form a ( x , E , D ) E O p S " ( U 1 .
X
U)
These a r e p r o p e r l y
supported s i n g u l a r p e r t u r b a t i o n s . D e f i n i t i o n 3.7.3.
A singular perturbation
supported if
m
E"'A~
:
A
F o p sv (uxu), ( v = ( v 1 , ~ ,2u 3 ) ) , v +
c o ( u ) E ' ( u ) and its transpose -f
E
"AE
is c a l l e d properly m
: Co(Ul
-+
E ' (U),
( E E ( o , c 0 ] ) arebothequicontinuous. V1
One c a n show t h a t A E i s p r o p e r l y s u p p o r t e d i f f i t s k e r n e l A ( x , y , E ) h a s t h e A(. , ,E ) c u x u h a s compact p i - n p e r t y : t h e c l o s u r e of t h e u n i o n (iO<E$EO s ~ p p i n t e r s e c t i o n w i t h KxU and UXK f o r any compact K c U .
.
3.7. Amplitudes, Adjoints and Products intersection with
KXU
and UXK, f o r any compact s e t K c U .
I f AE i s p r o p e r l y s u p p o r t e d , t h e n A I n f a c t , AE
m : CO (U)
-t
C m ( U ) and A
9E
K t U be a compact. W e choose
i s contained i n K
253
x K1 w i t h K
1
: Cm(U)
: Cm(U) E
+
-t
D' (U).
Cm(U), V Let u
E
E
m
Co(U)
E ( 0 . ~ ~ 1 .
C m ( U ) and l e t
such t h a t supp A ( x , Y , E )
n
(U x K )
c U a n o t h e r compact and @ E 1 i n a
. T h e r e f o r e , one h a s : A u ( x ) = A E ( @ u()x ) , v x E K , 1 m so t h a t A u ( x ) E C ( i n t K ) , V E E ( O , E 1. S i n c e K i s any
neighborhood of K
V
E
E
(O,E~],
compact, it p r o v e s t h a t A
: Cm(U)
+
Cm(U), V
E
0 E (0.~~1.
Theorem 3 . 7 . 4 .
E op s"
Let
1 .o
(u
x
u ) be properly supported and l e t
a ( x , y , E , E ) be i t s
amp E t u d e . Then (3.7.11)
A
=
a(x,c,~),a
E op
S"
1 ,o (U)
'
where
and t h e forma2 symbo2 of a ( x , E , ~ )is given by t h e f o m 2 a
Proof. ___ Since A
: Cm(U) + Cm(U),
t h e symbol a ( x , E , g ) = e - i < x , S > A E(,i<x,S> ) i s a
w e l l d e f i n e d smooth f u n c t i o n of
(x,s) E U
x
Rn
depending smoothly on
E E ( 0 . ~ ~ 1 Fu . rther,
u(x)
=
IT-^
J
ei<x'5';(C)dC
Bn
and A
: C m ( U ) + C m ( U ) , one c a n a p p l y A
under t h e i n t e g r a l s i g n , so t h a t
E S" (U) and t h a t ( 3 . 7 . 1 3 ) h o l d s . 1 ,O (U x U) For d o i n g t h i s w e n o t i c e f i r s t t h a t f o r any a m p l i t u d e b E S' 1# O and any m u l t i - i n d i c e a t h e a m p l i t u d e s ( y - x ) " b ( x , y , ~ , c ) and D " b ( x , y , E , S ) 5 pa ( U x u ) w i t h urn = p-(O,lal,O). d e f i n e t h e same s i n g u l a r p e r t u r b a t i o n i n Op S and w e have o n l y t o show t h a t a
I n f a c t , one f i n d s f o r B
E,a
=
Op(y-x)"b(x,y,~,g):
3 . Singular Perturbations on Smooth Manifolds without Bounda y
254
Now w e w r i t e t h e f i n i t e T a y l o r e x p a n s i o n of a ( x , y , ~ , S )w i t h r e s p e c t t o y about t h e p o i n t x:
where
T h e r e f o r e one g e t s t h e c o n c l u s i o n t h a t a l o n g w i t h t h e a m p l i t u d e a ( x , y , € , t ) a l s o t h e amplitude
w i t h R g i v e n by ( 3 . 7 . 1 5 ) . d e f i n e s t h e same s i n g u l a r p e r t u r b a t i o n A N more o v e r ,
and,
for V N > 0. But t h a t means p r e c i s e l y t h a t t h e f o r m a l symbol of A ( 3 . 7 . 1 3 ) , so t h a t if a ( x , ~ , S )i s any symbol in S"
1 to
(U)
is g i v e n by
with t h e asymptotic
expansi o n
( s u c h a symbol e x i s t s a s a consequence o f Theorem cv1(A - a ( x , E , D ) )
f op s
(O'-mr-m)
1 r0
3.6.2),
then
uniformly with r e s p e c t t o E E ( 0 . ~ ~ 1 .
Theorem 3 . 7 . 5 .
If a ( x , c , D ) E op sv
1 ,O
(u
x
U)
is p o p e r l y supported, then its transpose
3.7. Amplitudes, Adjoints and Prodiicts
and i t s a d j o i n t a ( x , E , D ) * are again s i n g u l a r p e r t u r b a t i o n s i n
a ( x ,E , D )
s;,o(u
x
U). t
Moreover t h e i r formal symbols a , a
*
a r e g i v e n by t h e f o l l o w i n g
formulae:
(3.7.16) a*-X-
114 a!
D’DZa 5
( x ,E , )
*,
where a ( x , t ) * i s t h e (complezi conjugate o f a ( x , C ) . Proof. __ One can w r i t e :
=
(2.rr)-”
/,u(x) IR X
J
i
e
i<x-y
,c>
nn IRn 5 Y
a ( y , E , - 5 ) v ( y )d y d t d x
t so t h a t a ( x , ~ , D ) h a s a ( x , ~ , - E )a s one of i t s a m p l i t u d e s .
In t h e s a m e way one f i n d s
where a g a i n t h e u p p e r
*
means t h e (complex) c o n j u g a t e .
T h e r e f o r e , a ( x , E , D ) * h a s a s one of i t s a m p l i t u d e s t h e f u n c t i o n
a ( y , E , t )*. Applying Theorem 3 . 7 . 4 t o t h e a m p l i t u d e s a ( y , E , - C ) , a ( y , ~ , E , ) * one , gets
I
formulae ( 3 . 7 . 1 6 ) . Theorem 3 . 7 . 6 .
Let a(x,E,D) E op
’S
1,o
(u) and l e t
b(x,E,D)
E
Op S’
supported. Then (3.7.17)
a(x,E,D)
b(x,E,D)
E
Op S”+’(U 1r 0
x
U)
1n
o
(U) be p r o p e r l y
255
3. Singular Perturbations on Smooth Manifolds without Boundary
256
and, moreover, t h e forma2 symbol c (3.7.18)
c
besides c ,
-
1 l " l
Zc. 3'
=
l a l = j I:-- a!
-
Zc.
I
of a
b
Daa ( x , E ,C) D>
F
is given by the f o m l a ( x ,E , E , )
,
sv+~-(O,l,O) (U). 1.0
3
P r o_ of. _ We s h a l l f i r s t assume t h a t b ( x , E , F ) h a s compact s u p p o r t i n x b ( x , ~ , S )? 0 , V x E W K ,
E
E ( O , E ~ ] , 5 E l R n , where K
C
E
U, 1.e.
U i s some compact
set.
I n t h a t c a s e F u b i n i ' s theorem y i e l d s (3.7.19)
(a(x,E,D)
0
b ( x , E , D ) u ) ( x ) = (271
where (3.7.20)
c ( x , E , ~ )= (2n)-nJnei<X'TI'
a ( x , ~<+r?)b , (TI,€
R and, a s usual ; ( F ) Since x
-f
=
F
X+5
u(x), i)(n,~,c)
= F
X+r?
b(x,E,S
b ( x , E , S ) h a s compact s u p p o r t , c ( x , ~ , F )i n ( 3 . 7 . 2 0 ) i s w e l l
( X , E , F ) E S"+' (U) 1,o Now we a r e g o i n g t o p r o v e ( 3 . 7 . 1 8 ) . Using t h e f i n i t e T a y l o r e x p a n s i o n
d e f i n e d and one c h e c k s e a s i l y t h a t c
for a(x,E,F+q),
with
one c a n rewrite
(3.7.19) a s follwos:
where t h e f o l l o w i n g f o r m u l a h a s been u s e d :
3 . 7 . Amplitudes, Adjoints and Products
257
One c h e c k s e a s i l y , u s i n g t h e f a c t t h a t b ( x , E , S ) h a s compact s u p p o r t i n x , that cN(x,E,c)
=
( 2 n P J
ei<xr”
RN (x,E , 5 , n ) b (ri,E , 5 ) drl
mn b e l o n g s t o t h e symbol c l a s s Sv+U-(O,N+1,0) (U), V N 2 0 , and t h a t e n d s t h e 1r o p r o o f of ( 3 . 7 . 1 7 ) , ( 3 . 7 . 1 8 ) i n t h e c a s e , when x + b ( x , E , S ) h a s compact support. Now we d r o p t h e a s s u m p t i o n t h a t x
+
b ( x , E , S ) h a s compact s u p p o r t
a n d assume onLy t h a t b ( x , t , D ) i s p r o p e r l y s u p p o r t e d . A s a consequence of Theorem 3 . 7 . 5 t h e a d j o i n t b * ( x , E , D ) b e l o n g s t o Op S’
1 ,O
(U
X
U)
and i t s
a m p l i t u d e b* ( x , E , ~ h ) as t h e asymptotic expansion
Using t h e a m p l i t u d e b * ( x , c , c ) one c a n rewrite b ( x , ~ , D ) ua s f o l l o w s :
**
b ( x , ~ , D ) u ( x )= ( b ( x , ~ , D ) u(x) = =
( 2 n P kn;.e
i<x-y,S>
*
b ( y ,E,
5 ) *u ( y )dyd5.
Hence,
so t h a t c ( x , Y , E , < ) c(x,E,D)
=
a(x,c,D)
= a ( x , ~ , 5 ) b * ( y , ~ , E i) s* a n a m p l i t u d e o f 0
b(x,~,D).
Applying ( 3 . 7 . 1 3 ) and u s i n g ( 3 . 7 . 2 3 ) one f i n d s t h e f o l l o w i n g a s y m p t o t i c e x p a n s i o n f o r t h e symbol of c ( x , E , D ) :
3. Singular Perturbations on Smooth Manifolds without Bounda y
258
S i n c e e a c h t e r m i n t h e l a s t sum i s a n t i s y m m e t r i c w i t h r e s p e c t t o B,y when B+y > 0 , t h i s sum v a n i s h e s a n d w e a r e l e f t w i t h ( 3 . 7 . 1 8 ) i n t h i s c a s e , a s
I
well. Remark 3 . 7 . 1 .
A l l t h e r e s u l t s e s t a b l i s h e d h e r e f o r t h e S i n g u l a r P e r t u r b a t i o n s i n OP
a r e a l s o v a l i d f o r t h e classes Op s"(U) and Op K"(U)
sy,o(U)
(see D e f i n i t i o n 3 . 3 . 2
and 3 . 6 . 4 ) . 3 . 8 . The S t a t i o n a r y P h a s e , L a p l a c e and S a d d l e P o i n t Methods I n t h i s s e c t i o n we s h a l l i n v e s t i g a t e t h e a s y m p t o t i c b e h a v i o r o f i n t e g r a l s o f t h e form: I(p) =
(3.8.1)
J
f ( x ) ei p g ( x ) d x ,
p
E B + ,p
+
+a,
U where U C Bn i s bounded, f E Ci(U), g E C m ( U ) , I m g ( x ) :0 , We s h a l l s t a r t w i t h t h e c a s e n
=
v
-
x E U.
1, U = (x-,x+) c R.
Proposition 3.8.1.
Let g ' ( x ) # 0 , V x E
c n . T h e n f o r any non-negative i n t e g e r N and K
holds (3.8.2)
p
N
K
(d/dp) I ( p )
-f
0
when
P r o o f . An i n t e r g r a t i o n by p a r t
p
+
-.
yields:
~
A f t e r N p a r t i a l i n t e g r a t i o n s one g e t s : K
N
p
(d/dp) I ( p )
=
J
u
m
I
fN(x)eipg(X)dx, fN E Co(U).
D e f i n i t i o n 3.8.2.
A p o i n t x0 E g'(xo)
=
u i s s a i d t o be s t a t i o n a r y f o r t h e integraZ
(3.8.1)
i f
0.
The s t a t i o n a r y p o i n t x0 i s s a i d t o be r e g u l a r , i f g " ( x o ) # 0 . Theorem 3 . 8 . 3 .
Assume t h a t t h e r e i s only one s t a t i o n a r y p o i n t x0 f o r t h e i n t e g r a l ( 3 . 8 . 1 ) and t h a t it i s r e g u l a r . Then t h e f o l l o w i n g asymptotic formula h o l d s :
3.8. The S t a t m a r y Phase
259
and the coefficients a . ( f , g ) , j t 1, depend only on f ( x O ), g ( x o ) and their I derivatives at xo up to the order 2 j at most. The asymptotic formuZa ( 3 . 8 . 3 ) can be differentiated with respect to p
~
infinitely many times.
P r o o f . W e s h a l l assume t h a t g " ( x > 0. If g"(x < 0 , then f o r t h e 0 0 complex c o n j u g a t e I * ( p ) one h a s : g " ( x ) > 0. L e t $ E C" ( U ) b e such t h a t
0 0 < a / 2 , j , Z 0 f o r Ix-xOl 2 a , where t h e c o n s t a n t a i s 0 s u f f i c i e n t l y s m a l l and w i l l be chosen l a t e r on. R e w r i t i n g I ( p ) a s f o l l o w s :
j,
Z
1 f o r Ix-x
(3.8.5)
I(@)
I
1
=
ipg(x)dx f(x)$(x)e
+ i f ( x ) ( l - $ ( x ) ) ei p g ( x )dx
U
U
and n o t i c i n g t h a t t h e second i n t e g r a l o n r a p i d l y d e c r e a s i n g when p
+ m,
t h e r i g h t hand s i d e of
(3.8.5) is
a s a consequence of P r o p o s i t i o n 3.8.1,
g e t s t h e c o n c l u s i o n t h a t it s u f f i c e s t o show ( 3 . 8 . 3 ) ,
(3.8.4) f o r t h e
integral
I1(p)
=
J f (x)@(x)eipg(x)dx. U
The l a t t e r can b e r e w r i t t e n a s f o l l o w s : (3.8.6)
i
I(p) = eipg(xo)
i p h (x) f(x)$(x)e dx ,
R where (3.8.7)
h(x)
=
g(x)-g(x
0
A s a consequence of
(3.8.8)
y(x)
2
for x
+
x
0'
( 3 . 8 . 7 ) t h e change of v a r i a b l e s
( x - x O ) ( h ( x ) (x-x 1
=
3
= ~ g " ( x o ) ( x - x o )+ O ( / x - x o l ) ,
0
-2
1 ) 2 ,
x
=
x ( y ) , lx-xOl 5 a
i s a diffeomorphism of t h e c o r r e s p o n d i n g i n t e r v a l s , p r o v i d e d t h a t a i s
s u f f i c i e n t l y s m a l l . With such a c h o i c e o f a and u s i n g ( 3 . 8 . 8 1 , one c a n r e w r i t e I ( p ) , as follows: 1 (3.8.9)
I1(p)
=
e ipg
(xo) I R
ipg (xo) = e where (3.8.10)
q(y)
= f (X
2 f (x ( y ) ) $ (x ( y )) x ' ( y )eipy dy = 2
70 q ( y ) e i p Y d y ,
( y ) ) $ ( X ( y ) )x' ( y )+f (X ( - y ) ) V I ( x ( - y ) ) x ' ( - y )
.
one
3. Singular Perturbations on Smooth Manifolds without Bounda y
260 If
-
f(x(y))x'(y)
C b.y' jdo
is the Taylor expansion of f(x(y))x'(y) at y = 0 , then for q(y) given by ( 3 . 8 . 1 0 ) holds: (3.8.11)
q(y)
-
y2J, y + 0,
2 C b jto 2j
since q(y) is even and $(x(y)) : 1 in some small neighbourhood of zero. Notice that as a consequence of ( 3 . 8 . 7 ) ,
( 3 . 8 . 8 ) , the following formulae
hold: (3.8.12)
x'(y) ly=o
=
f. . r % " ( ~ O ) ) fbo , = f(xo)(~g"(xo))
!?-
Furthermore, denote
where
r
Y
is the following contour in
r
Y
I
= { z
1 z
E c
, Z
C':
:i = y+oe , u 2 0 1 ,
and the integration in ( 3 . 8 . 1 4 ) goes from
m
to the pointy
Rewriting P .( p ,y), 3
one can estimate it in the following fashion:
Using the functions ( 3 . 8 . 1 4 ) , integrating 2N+2 times by part in ( 3 . 8 . 1 3 ) and using the estimates ( 3 . 8 . 1 6 ) , one gets the following asymp-
totic formula for Q ( p ) : (3.8.17)
Q(p)
Z
=
15j 52N+2
(-1) j-1
(j-1)
9
1
m
(Y)P.(P,Y) 3
+ O(P
-N- 1
0
The Taylor expansion ( 3 . 8 . 1 1 ) for q(y) and ( 3 . 8 . 1 7 ) yield: (3.8.18)
Q(p) =
-
1 2 ( 2 j ) ! b 2jp2j + l ( P . 0 ) + O ( P - N - l ) OSjSN
I
P
-f
a,
),
P
-f
m.
3.8. The Stationary Phase
26 1
s i n c e q ( y ) :0 f o r y s u f f i c i e n t l y l a r g e . One e v a l u a t e s e a s i l y P 2 j + l ( p , 0 ) u s i n g ( 3 . 8 . 1 5 ) : - l e i ( ~ r / 4 )( 2 j + l )p - j - +
P2j+l(p,0) = - ( f ) T ( j + t ) ( ( 2 j ) ! )
(3.8.19)
so t h a t ( 3 . 8 . 1 8 ) ,
(3.8.19),
(3.8.9) y i e l d (3.8.3).
Using ( 3 . 8 . 1 2 ) and ( 3 . 8 . 1 9 ) w i t h j
=
0 , one g e t s t h e f o r m u l a ( 3 . 8 . 4 ) .
One c h e c k s e a s i l y t h e s t a t e m e n t c o n c e r n i n g t h e c o e f f i c i e n t s a . ( f , g ) , 3 j t 1 , and t h e f a c t t h a t ( 3 . 8 . 3 ) c a n b e d i f f e r e n t i a t e d i n f i n i t e l y many times with respect t o p .
I n d e e d , it s u f f i c e s t o show t h a t t h e a s y m p t o t i c
relation (3.8.20)
-
Z
~ ( p )
i ( ~ / 4 ()2 j + l ) p - j - + , r(j+?)bZje
~
j20
can b e t e r m by t e r m d i f f e r e n t i a t e d . One p r o c e e d s a s f o l l o w s . D i f f e r e n t i a t i n g k t i m e s ( 3 . 8 . 1 3 ) , one f i n d s : m
k (d/dp) Q ( P ) =
1
2 2 k ( i y 1 q(y)eiPY dy.
0 k A f t e r w a r d s , u s i n g t h e p r e v i o u s a r g u m e n t , one g e t s f o r ( d / d p ) Q ( p ) a n
a s y m p t o t i c e x p a n s i o n , which c o i n c i d e s w i t h t h e f o r m a l t e r m by t e r m d e r i v a t i v e o f o r d e r k w i t h r e s p e c t t o p o f t h e r i g h t hand s i d e i n ( 3 . 8 . 2 0 ) . The l a s t c l a i m f o l l o w s immediately i f one n o t i c e s t h a t 2 k (iy 1 q(y)
-
2i
k
b 2 j y2 ( J + k ) , f o r y + 0 ,
C
j20
k a n d , t h u s , f o r ( d / d p ) Q ( p ) , h o l d s t h e same k i n d o f a s y m p t o t i c e x p a n s i o n as (3.8.20) with c o e f f i c i e n t s c instead of b such t h a t c 21 21 21 k I = i b 2 ( j - k ) r 1 2 k. 21
=
0, j < k
and c
Now we d r o p t h e a s s u m p t i o n t h a t n = 1 and s h a l l c o n s i d e r t h e m u l t i d i m e n s i o n a l i n t e g r a l s ( 3 . 8 . 1 ) f o r any d i m e n s i o n . We s t a r t w i t h a s t a t e m e n t a n a l o g o u s t o P r o p o s i t i o n 3 . 8 . 1
in the
multidimensional case. Proposition 3.8.4.
L e t f E C:(U), V x E
E c r n ( U ) ,I m g ( x ) : 0 , V x E
U. Then for t h e i n t e g r a l
(3.8.21) Proof. ___
g
p N ( d / d p ) K I ( p )+ 0 Denote by
L
~ ( p given ) by
when
p
-f
a,
and assume t h a t Vg(x) # 0 , ( 3 . 8 . 1 ) holds:
V N 2 0, V K Z 0.
(x,a) t h e d i f f e r e n t i a l operator 9
3. Singular Perturbations on Smooth Manifolds without Boundary
262
and by L * ( x , a ) i t s f o r m a l a d j o i n t : 9
Hence, f o r any i n t e g e r K 2 0 , N 2 0 one f i n d s : K
( d / d p ) I(P) = ( i p ) - '
(3.8.24)
I U
K
f(x1(ig(x))
L ( x , a ) ei ? g ( x )dx g
K
( i p ) -1
K
eipg(x) L*(x,a) ( f ( x ) i g ( x ) dx = U g N+ 1 K K - (ip)-(N+l)i .ipg(x) ( L z ( x , a ) ) ( f ( x ) i g ( x ) )dx. =
U A s a consequence of
(3.8.24)
one g e t s ( 3 . 8 . 2 1 ) .
Definition 3.8.5.
A p o i n t xo E
u i . s~a i d t o be s t a t i o n a r y f o r t h e i n t e g r a l
Vg(x ) = 0. The s t a t i o n a q p o i n t xo
0
of t h e second d e r i v a t i v e s 2
2
D g(xo) =
d e t D g ( x ) # 0.
I
0
-
(3.8.1) i f
i s s a i d t o be r e g u l a r i f t h e m a t r i x is non-singular: K
I1
The i n v e s t i g a t i o n o f t h e a s y m p t o t i c b e h a v i o u r of t h e i n t e g r a l ( 3 . 8 . 1 ) for p +
c a n b e reduced t o t h e one d i m e n s i o n a l c a s e u s i n g t h e f o l l o w i n g
M o r s e ' s Lemma.
L e t g ( x ) , x E u b e r e a l valued and assume t h a t g ( x ) i s cm in some neighbourhood o f i t s r e g u l a r s t a t i o n a r y p o i n t xo E U. Then t h e r e e x i s t neighbourhoods Ox , Oo of t h e p o i n t s x
= xo
and y
=
0, r e s p e c t i v e Z 2 , and
0
a cm-diffeomorphism
h :
Ox + O o such t h a t 0
where u j ,
1 5 j 5 n,
are t h e eigenvalues of t h e m a t r i x
2
D g(x ) .
0
Proof o f Morse's Lemma Without r e s t r i c t i o n o f g e n e r a l i t y one c a n assume t h a t xo = 0, 2 D g ( 0 ) = d i a g ( v l , . . . , p n ) , s i n c e an o r t h o g o n a l t r a n s f o r m a t i o n r e d u c e s t h e X
g e n e r a l c a s e t o t h i s one. L e t 0
0
b e convex and s u f f i c i e n t l y s m a l l ,
3.8. The Stationay Phase
00
3 0. Since Vg(0)
=
0 , one h a s :
=
(
1/2 ) <s ( x )X I x> ,
where t h e m a t r i x S ( x )
=
I / S k j ( x ) I15k,jsn
263
i s d e f i n e d by t h e f o r m u l a e :
S ( x ) b e i n g , o f c o u r s e , symmetric, S ( O 1 = d i a g ( p l , ..., u n )
00’
function in
I d e n t i f y i n g t h e symmetric nxn m a t r i c e s w i t h R c o n s i d e r S ( x ) a s a Cm-map from
0,
m
matrix-
and C
n ( n + 1 ) / 2 one can
i n t o R n ( n + 1 ) / 2 . W e s e e k a smooth ( C m )
upper t r i a n g u l a r m a t r i x - f u n c t i o n Q ( X )
:
0O + f l R n ( n + 1 ) / 2 s u c h t h a t
where Q ( x ) * i s t h e c o n j u g a t e of Q ( x - ) . Denote F(Q,X)
=
Q*S(O)Q-S(X)
where Q i s any u p p e r t r i a n g u l a r m a t r i x and x E m
00 ’
One c a n c o n s i d e r F ( Q , x ) a s a C -map from R
n(n+1)/2
X
0,
into the
s p a c e o f a l l symmetric m a t r i c e s . One h a s F(Id,O)
=
and, given t h a t a l l
0
u,, I
1 5 j 6 n a r e d i f f e r e n t from z e r o , one f i n d s
easily that
QF ( Q r x ) / Q = I d , x = O = where J i s t h e m a t r i x which h a s t h e e n t r i e s : kj a k j = 1 , alm = 0 , V ( l , m )
# (k,j).
O b v i o u s l y , t h e m a t r i c e s J k j + J j kform a b a s i s i n t h e s p a c e o f a l l symmetric nxn m a t r i c e s , so t h a t
3 . Singular Perturbations on Smooth Manifolds without Boundary
264
t h e I m p l i c i t F u n c t i o n Theorem c a n b e a p p l i e d , which g u a r a n t e e s t h e e x i s tence of a (well-defined) C
m
upper t r i a n g u l a r m a t r i x Q ( x ) i n a
oo
s ~ i f i c i e n t l ys m a l l neighbourhood
of t h e o r i g i n , t h a t s a t i s f i e s t h e
conditions :
Obviously t h e map y
h(x)
=
def -
Q(x)x
i s a Cm diffeomorphism of in
0,
i n Rn o n t o some neighbourhood of t h e o r i g i n
which r e d u c e s g ( x ) t o t h e form
pln
Y’
(g
h)(y) = g ( O ) + $
0
2 C U.Y. I<j
’
Theholomorphversioiiofthislemmaistrue,too.Theproofis
Theorem 3.8
l e f t t o t h e reader.
&.
L e t u c R~ be bounded and l e t f t c ~ ( u ) ,g E cm(G), t h e f u n c t i o n g(x), x E
u, being
r e a l valued. Assume t h a t g ( x ) has p r e c i s e l y one
s t a t i o n a r y p o i n t xo in
5 and
t h a t xo is r e g u t a r .
Then for each i n t e g e r K b 1 one has: (3.8.26)
f ( x ) ei P 9 ( x ) d x
I(p) =
~
U - j-n/2
= ei p g ( x ~ ) c a.(f,g)p 0; j ‘K
’
where a . ( f , g ) 3
operators w i t h
m > K
=
+QK(p),
~ . ( x , a ) f ( x ) ~~ ~, (=x ~, a~being ), some l i n e a r d i f f e r e n t i a l 3 2j a t most, and whsre
m’
c - c o e f f i c i e n t s of order
o being some i n t e g e r s and
c K ( g ) > 0 being some constants,which may depend
on K and 9. The asymptotic f o m m * ~ a
can be d i f f e r e n 6 i a t e d wi+h m s p e c t t o
p
i n f i n i t e l y many t i m e s .
3.8. The Stationary Phase Proof. -
265
I t s u f f i c e s t o c o n s i d e r t h e c a s e when supp f b e l o n g s t o some s m a l l
0
neighbourhood
o f t h e p o i n t x o , s i n c e t h e p a r t i t i o n o f u n i t y and
XQ
P r o p o s i t i o n 3.8.4
can be used i n o r d e r t o reduce t h e g e n e r a l s i t u a t i o n
t o t h i s case. Assuming t h a t
0
i s so s m a l l t h a t Morse's Lemma c a n b e
XO
a p p l i e d , w e can rewrite I ( p ) a s follows:
Using r e p e a t e d l y t h e one d i m e n s i o n a l s t a t i o n a r y phase method n t i m e s , one g e t s (3.8.27) f o r I ( p ) and t h e c o r r e s p o n d i n g f o r m u l a e f o r i t s derivatives
.
I
W e are g o i n g t o g e t e x p l i c i t
formulae f o r t h e c o e f f i c i e n t s a . ( f , g ) i n 1
(3.8.27). Denote by Q
(3) the differential operator 4 (xo)
2 -1 . 2 i s t h e i n v e r s e m a t r i x f o r t h e m a t r i x D g ( x ) of t h e where D g ( x o ) 0 second d e r i v a t i v e s of g ( x ) a t x o , and i n t r o d u c e
2
u 9 ( x0 1
where
i s t h e s i g n a t u r e of D g ( x
).
0
Theorem 3.8.7.
Under t h e assumptions o f Theorem 3.8.6. t h e foZZowing asymptotic formula holds for
where
Q
g i v e n by (3.8.1):
I(p)
9 (xo)
(a),
h (x,xo), y 9
a r e given r e s p e c t i v e z y by (3.8.28), (3.8.29),
(3.8.30), where (3.8.32)
\
and where f o r
=
p
n/2 + k
-
[2k/3]
>= 1 holds:
w i t h some i n t e g e r
M k ' O
3. Singular Perturbations on Smooth Manifolds without Boundary
266 Proof. -
Denote iph (x,xo) v(p,x0,x) = f(x)e g
and
For a symmetric regular matrix A one has the following formula
where
- (2T)-n/21det Al-f e(in/4) sign A 'n,A
-
sign A being the signature of A. Using Parceval's identity and the last formula, one can rewrite I ( p ) as follows: 1
Expanding the exponential under the integral sign into Taylor's series, one gets:
and that is precisely (3.8.31). Since h ( x , x ~ )has at x = x0 a zero of order 3 at least, the degree 4
of the polynomial
is [2j/?] at most. Writing I ( p ) in the form
using the fact that the coefficients of an asymptotic expansion are uniquely d e f i n e d and that the degree of q . ( p ) is at most [2j/3 , one gets the 7 p-' with p 5 n/Z+k-[Z(k-1)/31
conclusion that all the terms containing have to be included in the sum
c
-1
L-
Sj'P),
O<j
(3.8.33).
I
3.8. The Stationary Phase
267
Remark 3 . 8 .-8.. Taking o n l y t h e f i r s t t e r m i n t h e e x p a n s i o n ( 3 . 8 . 3 1 ) , one g e t s t h e f o l l o w i n g a s y m p t o t i c formula f o r I ( p )
when p
+ m.
Remark 3 . 8 . 9 . I f g ( x ) h a s a f i n i t e number o f c r i t i c a l p o i n t s x l , ...,x regular, then t h e localization principle
cl
which all a r e
( p a r t i t i o n o f u n i t y and
Proposition 3.8.4)
l e a d s t o t h e following asymptotic formula f o r I ( p )
(3.8.35)
X
I(p) =
I . (p) + O ( P - ~ ) , for p
:
+ m,
l<j
0'
I
I n a p p l i c a t i o n s f and g may a l s o depend ( c o n t i n u o u s l y ) o f some m
p a r a m e t e r w b e l o n g i n g t o a compact C -manifold w i t h o u t boundary ( f o r
R
instance t h e u n i t sphere
i n Wn ) ; b e s i d e s f may a l s o h a v e some weak
(polynomial l i k e ) dependence upon p E IR+. More p r e c i s e l y , w e consi.der now i n t e g r a l s o f t h e form i p g ( x ,w ) dx f(x,w,p)e U where, a s p r e v i o u s l y , U i s bounded, U C IRn,
(3.8.36)
I(p,w)
1
=
o E M , M b e i n g a compact
m
riemannian C -manifold of f i n i t e dimension 9, and p E IR+, a s p r e v i o u s l y . The f u n c t i o n f , c a l l e d a l s o t h e a m p l i t u d e , and t h e f u n c t i o n g , c a l . l e d a l s o t h e p h a s e , are supposed t o s a t i s f y t h e f o l l o w i n g c o n d i t i o n s . 1'
f E Cm(U
Z 0 For e a c h
X
M X R+), g
E M
(wo,po)
x
E Cm(U
x M),
a n d , moreover, g i s r e a l v a l u e d .
R+ , h o l d s : supp f ( x , w 0 , p 0 )
i s some compact which d o e s n o t depend on
C K C U,
where K
(wo,po).
3" For e a c h t r i p l e of m u l t i - i n d i c e s a , B , y h o l d s :
where m i s some g i v e n number and t h e c o n s t a n t s C
Cr,B,Y
depend o n l y on t h e i r
subscripts. 4 " For each g i v e n
c r i t i c a l point x
iz
=
E M t h e f u n c t i o n g(x,w)
x (w) E U, 0
:
U
i . e . t h e equation
-t
R h a s p r e c i s e l y one
3 . Singular Perturbations on Smooth Manifolds without Boundary
268 (3.8.38)
Vxg(x,o)
=
0
h a s p r e c i s e l y one z e r o x = x o ( w ) ,
v
E
w
M.
2
5" The e i g e n v a l u e s p I, (a), 1 6 j 6 n , of t h e m a t r i x Dx g ( xo ( w ) ,w) the inequalities:
(3.8.39)
inf
0, V j = 1 , 2 ,..., n
( p . ( w ) ( 2 po
wEM
satisfy
'
Introduce the d i f f e r e n t i a l operator
2 -1 2 t h e i n v e r s e m a t r i x of D g ( x o ( w ), w ) , w i t h D g ( x o ( w ),w)
and d e n o t e , a s
p r e v i o u s l y by y = y ( w ) t h e c o n s t a n t ( 3 . 8 . 3 0 ) w i t h xo = xo(w) t h e o n l y n n c i r t i c a l p o i n t of g i n U. Denote by h (x,w) t h e f u n c t i o n ( 3 . 8 . 2 9 ) where 9 xo = x ( 0 ) i s t h e c r i t i c a l p o i n t of g i n U. 0
Theroem 3.8.10.
Assume t h e c o n d i t i o n s l o k 2 1
holds:
where
%
= n/2tk-[Zk/3]-m
I%(P,~)
I
t o be f u l f i l l e d . Then for each i n t e g e r
5 O
and
5 Ck, V
( 0 , ~E)
M
x
IR,
w i t h some c o n s t a n t Ck which depends only on i t s s u b s c r i p t . The a s y m p t o t i c expunsion
D + -
can be d i f f e r e n t i a t e d i n f i n i t e l y many t i m e s w i t h r e s p e c t t o and holds uniformly w i t h r e s p e c t t o
E
w
(w,p)
E M
x
M.
P r o_ o f . A s a consequence of t h e c o n d i t i o n s 4 " , 5 O , t h e c r i t i c a l p o i n t _ x ( w ) of g i s r e g u l a r , V w E M a n d , moreover, t h e s i g n a t u r e a ( x ( w ) ,w) O 2 ¶ O of D g ( x ( w ) , w ) x o
d o e s n o t depend on w
E
M.
The i m p l i c i t f u n c t i o n s theorem
R+
3.8. The Stationary Phase m
i m p l i e s t h a t xo(w) E C ( M ) .
r(
,
)
=
{w
I
w
w ) t h e length of t h e P P' and w and 6 > 0 t o be chosen l a t e r o n . L e t
P c M b e one o f s u c h domains and l e t w l ,
Ro. I f x ( w ) f
K, V w E Q
0
2O,
distance
E M , r ( w , w ) < 6) w i t h r ( w
geodesic j o i n i n g w
0
m
M b e i n g equipped w i t h a C -riemannian
and compact, one c a n c o v e r M by a f i n i t e number o f domains
R(wp,6)
R
269
0
...,wcl
b e l o c a l c o o r d i n a t e s on
w i t h K t h e compact d e f i n e d i n t h e c o n d i t i o n
t h e n f o r any i n t e g e r k t 0 and any p a i r o f m u l t i - i n d i c e s one h a s :
I D U~ D ~PI ( ~ , L I ) I
(3.8.43)
5
ck,a,B
P
i n f a c t , using again the operator
-k
,v
L (w,x,a ¶
t h e f a c t t h a t V g(x,w) # 0 , V (x,w) E
0
)
x
R+.
d e f i n e d by ( 3 . 8 . 2 2 ) and
c x co one
same p r o c e d u r e as i n P r o p o s i t i o n 3 . 8 . 4 . Now, l e t R
E no
(W,P)
g e t s (3.8.43) u s i n g t h e
and ( 3 . 8 . 3 7 ) .
b e such a domain t h a t t h e s e t Uo
=
x (R ) h a s a non-
0
0
empty i n t e r s e c t i o n w i t h K i n t h e c o n d i t i o n 2 " . Without r e s t r i c t i o n o f g e n e r a l i t y one c a n assume t h a t wo = 0 , x 0 ( w 0 ) = 0 , g ( x0 ( w 0 ) , w o ) = 0 . I f 6 i n t h e d e f i n i t i o n o f Ro i s s u f f i c i e n t l y small, t h e n one c a n a g a i n u s e Morse's Lemma i n o r d e r t o r e d u c e by an a p p r o p r i a t e d i f f e o m o r p h i s m x = h(y,w) the function g(x,w) t o t h e q u a d r a t i c function:
with
u1
Uo and T , = * I . 3 we u s e t h e p a r t i t i o n of u n i t y ~i ( X I + $ ( x )
2
-1
-
The i n t e g r a l I ( p , w ) 1
where f l ( y , w , p )
=
2
1 in
fi,
0 for 1 E 1 f o r x E U with U2 c U o , U 2 c U 1 . D e n o t e b y I . ( p , w ) , j = 1 , 2 x !$ 2 1 t h e c o r r e s p o n d i n g c o n t r i b u t i o n s i n t o I (p , w ) , so t h a t i (p,W) = I 1 ( p ,W) + I 2( P , w ) NOW
VJ
can be r e w r i t t e n as follows:
f ( h ( y , w ), w , p ) ($1 o h ) ( y , w ) . d e t h ' y ( y , w ) .
The f u n c t i o n f l h a s compact s u p p o r t i n y E R n , and s a t i s f i e s ( 3 . 8 . 3 7 ) i n h((U1
x
Go))
x
Ro
x
R + . Applying t h e s t a t i o n a r y p h a s e method
r e p e a t e d l y i n v a r i a b l e s y1,y2, . . . , y n ,
one g e t s f o r I 1 ( p , w )
asymptotic
e x p a n s i o n s o f t h e form ( 3 . 8 . 3 1 ) and t h e ( u n i f o r m i n w , p ) e s t i m a t e f o r
the remainder of the form (3.8.33) w i t h
\
= n/2tk-[2k/3]-m,
since t h e
amplitude f s a t i s f i e s (3.8.37). The same argument a s p r e v i o u s l y i n t h e p r o o f of Theorem 3.8.3 y i e l d s t h e c o r r e s p o n d i n g a s y m p t o t i c formulae f o r t h e d e r i v a t i v e s o f I ( p , w ) .
3 . Singular Perturbations on Smooth Manifolds without Boundaty
270
C o r o l l a r y 3.8.11.
Assume t h a t f,g s a t i s f y t h e c o n d i t i o n s o f Theorem 3.8.10 and assume, furthermore, t h a t f belongs t o some bounded s e t B i n ci(U
x
M
x
R + ) . Then
is uniform w i t h r e s p e c t t o f E
t h e e s t i m a t e of t h e remainder in (3.8.42)
B.
To c o n c l u d e t h i s p a r a g r a p h w e s h a l l c o n s i d e r t h e a s y m p t o t i c b e h a v i o u r
of I ( p ) when p i s a complex p a r a m e t e r , p
+
-.
Theorem 3.8.12.
The assumptions o f Theorem 3.8.6 (3.8.44)
being f u l f i l l e d , assume a d d i t i o n a l l y t h a t
g ( x o ) > g ( x ) , V x E v\{xo}.
Then t h e asymptotic expansions ( 3 . 8 . 2 6 1 ,
and t h e corresponding
(3.8.31)
e s t i m a t e s f o r t h e remainders are s t i l l v a l i d when such t h a t I m p
5 0,
/p
2 1,
p
is a complez parameter,
uniformly w i t h r e s p e c t t o a r g
p.
If (3.8.45)
g(xo) < g(x)
v
x
E v\{xol,
then t h e s t a t e m e n t s ab ve a r e t r u e f o r I m p L
o,
Ip
1
2 1.
Proof. W e s h a l l assume t h a t ( 3 . 8 . 4 4 ) h o l d s , t h e case when one h a s ( 3 . 8 . 4 5 ) __
c a n b e t r e a t e d i n t h e c o m p l e t e l y analoguous way. i t s u f f i c e s t o p r o v e t h i s theorem f o r n
Evidently,
=
1 , s i n c e Morse’s
Lemma r e d u c e s t h e m u l t i d i m e n s i o n a l case t o n = 1. I n one d i m e n s i o n a l
s i t u a t i o n i t i s enough t o show t h a t ( 3 . 8 . 1 6 ) h o l d s f o r complex p s u c h t h a t I m p S 0 . For d o i n g t h a t one i n t e g r a t e s i n ( 3 . 8 . 1 4 ) a l o n g t h e r a y
r
Y
= {z
I
z E
c,
i0 z=y+oe , a 2 0, 8
where y = a r g p . Then f o r z E
r
Y
one f i n d s
and t h a t y i e l d s ( 3 . 8 . 1 6 ) .
I
Remark 3.8.13. If
( 3 . 8 . 4 4 ) h o l d s , t h e n one h a s f o r p +
m
=
n/4-y/2},
3.8. The Stationary Phase
27 1
uniformly with r e s p e c t t o a r g p . The p r o c e d u r e above f o r computing a s y m p t o t i c s o f t h e form ( 3 . 8 . 4 6 ) i s c a l l e d t h e L a p l a c e method. I f f , g a r e a n a l y t i c and U i s a m a n i f o l d i n Cn of r e a l dimension n ,
t h e n t h e p r o c e d u r e above h a s t h e name: t h e mountain p a s s method ( o r t h e s a d d l e p o i n t method). W e s h a l l s t a t e t h e main r e s u l t i n t h i s case which w i l l b e u s e d l a t e r on for d e s c r i b i n g a s y m p t o t i c b e h a v i o u r of s i n g u l a r l y p e r t u r b e d Poisson type o p e r a t o r s .
E C n , b e holomorphic a t t h e p o i n t z o , which i s supposed t o
Let g ( z ) , z
b e a r e g u l a r c r i t i c a l p o i n t f o r g ( z ) . L e t U be a s u f f i c i e n t l y s m a l l neighbourhood o f z o , U
=
{z
1
Iz-zo/
<
6) and l e t
G~ =
u n
{Re(ig(z)-ig(zo))>
-ul,
LG
u n
{Re(ig(z)-ig(z ) )
-GI,
=
0
=
where u > 0 i s s u f f i c i e n t l y s m a l l . The r e l a t i v e homology g r o u p H ( Gu , L u )
i s isomorphic t o Z . Denote
by y t h e g e n e r a t i n g c y c l e o f t h i s t r o u p . I f g ( z ) =
C 22 t h e n one l Z j 6 n I'
can d e f i n e Y a s f o l l o w s
(3.8.47)
y
=
=...=
{yl
yn
=
01 n u
Theorem 3.8.14.
L e t f ( z ) , g ( z ) be holomorphic a t z o E y, z o being a reguZar s t a t i o n a r y p o i n t f o r g ( z ) and l e t J f ( z ) eiPg(')dz, Y y being t h e generating c y c l e d e f i n e d above. ~ ( p )=
Then one has f o r (3.8.48)
I(p)
p
-
-f
-:
p -n/2
eipg(zo)
I: a . p - 1 . jto
'
The main term i n t h e asymptotic expansion ( 3 . 8 . 4 8 ) has t h e form: (3.8.49)
I(p)
-
f ( z o ) ( 2 n / p ) " l 2 ( d e t DZq(zO))-' 2
rJhere ( d e t D2g(z z
o
))
eipg(zo),
' I 2i s d e f i n e d by t h e o r i e n t a t i o n o f
y.
3 . Singular Perturbations on Smooth Manifolds without Bounda y
212 Proof. -
I n a s u f f i c i e n t l y s m a l l neighbourhood of zo one c a n u s e a holomor-
p h i c d i f f e o m o r p h i s m z = h ( w ) (holomorph v e r s i o n o f Morse's Lemma) which r e d u c e s g ( z ) t o a sum of squares: (g
0
h ) (w) = g ( z o ) +
b e s i d e s , one h a s 2
h'(0)
I: 1sj
=
2 w. J'
-1
(det DZg(zO))'.
N
L e t y b e t h e image of y u n d e r t h e holomorphic diffeomorphism h . The Cauchy theorem y i e l d s :
/ (f /f ( 2 ) e i p g ( z ) d z = N
Y
o
i p (g
h) (w)h'(w)e
0
h ) (w)dw
Y
/ (f~h)(w)h'(w)eip(goh)(W)d+ w / ( f ~ h ) ( w ) h ' ( W ) . i p ( g o h ) ( W ) dw, YO Y1 where yo i s t h e c y c l e d e f i n e d by ( 3 . 8 . 4 7 ) and R e i ( g , h ) (w) = -u f o r w E yl, =
so t h a t t h e i n t e g r a l o v e r y1 h a s t h e o r d e r
when p
-t m
and i s , t h u s , e x p o n e n t i a l l y s m a l l compared t o t h e i n t e g r a l o v e r
y o . Now t o t h e i n t e g r a l o v e r yo t h e L a p l a c e method c a n b e a p p l i e d , which l e a d s t o t h e asymptotic formula ( 3 . 8 . 4 9 ) .
I
Example 3.8.15. Let h
E (O,hO] be
(3.8.50)
a s m a l l parameter. Consider t h e following i n t e g r a l
Ph ( x ,t ) = (Znh)-'J
where ( x , t ) E R
2
e
3 ih-' ( ( x - t ) <+ ( t / 6 ) 5 )
d<
IR
.
The d i s t r i b u t i o n ( 3 . 8 . 5 0 )
i s t h e s o l u t i o n of t h e Cauchy problem:
l i m P (x,t) = 6(x),
t+o
where 6 ( x ) i s D i r a c ' s m a s s a t t h e o r i g i n . We a r e g o i n g t o f i n d t h e a s y m p t o t i c b e h a v i o u r o f P h ( x , t ) when h Introduce
-f
0.
3.8. The Stationary Phase
Assume f i r s t t h a t a = x / t > 1. Then 5
+
213
g ( x , t ; C ) d o e s n o t have s t a t i o n a r y
p o i n t s on R and t h e p a r t i a l i n t e g r a t i o n i n t3.8.50)
yields:
u n i f o r m l y w i t h r e s p e c t t o ( x , t ) i n any compact s e t K b e l o n g i n g t o t h e halfplane x / t
> 1.
let a
Now,
=
x / t < 1. Then
5
+
g ( x , t ; c ) h a s two s t a t i o n a r y p o i n t s :
which are r e g u l a r , s i n c e a c: 1.
,f
L e t f ( 5 ) E C:(lR)
:
R
+
R b e such t h a t f ( 5 ) : 1 f o r
5 E [<-(a)-1,5+(a)+l]. Then a g a i n t h e p a r t i a l i n t e g r a t i o n y i e l d s : 0 0 3 i h - l ( ( x - t ) S - ( t / 6 ) < )d5 = O(hN ) , ( 2 a h ) - l / (1-f ( 5 ) ) e R V N 2 0,
v
t # 0,
so t h a t one h a s i n t h i s c a s e :
(3.8.51)
P h ( x , t ) = (2nh)
-1
3 ih-' ( ( x - t ) 5+ ( t / 6 ) 5 ) d S
J f(5)e R
Applying ( 3 . 8 . 3 1 ) t o t h e i n t e g r a l o n
+
O(hm).
t h e r i g h t hand s i d e o f
one g e t s t h e f u l l a s y m p t o t i c e x p a n s i o n f o r P
h
( x , t ) when h
+
(3.8.51)
0.
W e w r i t e here only t h e f i r s t t e r m of t h i s asymptotics:
Ph(x,t) = 2 ) (ah)-'(tl-'(l-a)
(3.8.52)
-'cos
(n/4-Z3j2 ( 3 h ) - l t ( l - a ) 3 / 2 ) ( 1 + 0 ( h )) h + O
where a = x / t < 1; thus one h a s : s i n g s u p p Ph = { x S t j . One c a n a l s o e x p r e s s P ( x , t ) i n t e r m s of A i r y ' s f u n c t i o n ( s e e , f o r h 1 ) . Namely, one f i n d s e a s i l y :
i n s t a n c e [Was, 1
P ( x , t ) = c h-5/6t-1/3
h
0
( t - x ) ' l 6 A i ( c h-2/3t-1/3 ( t - x ) ) 1
,
2-5/3rr - 1/ 2 , c = - 61/3 = 0 1 Using t h e s a d d l e p o i n t method ( o r t h e a s y m p t o t i c e x p a n s i o n f o r A i r y ' s
where c
f u n c t i o n ) one c a n show, t h a t f o r a = x / t > 1 t h e f u n c t i o n Ph d e c r e a s e s a s e-h-ly(xft) NOW,
w i t h some y ( x , t ) > 0 , when h + 0 .
assume t h a t s t i l l a
=
x / t < 1 and t h a t
,
3 . Singular Perturbations on Smooth Manifolds without Bounda y
214
(1-a)
-
,
h'
f o r h + 0,
where u < 2 / 3 .
5
A f t e r t h e change of v a r i a b l e
L
= ( l - a ) 2 ~t ,h e i n t e g r a l
(3.8.50) takes
t h e form: P ( x , t ) = (Zah)-'(l-a)
h
f 1 ei t h - ' ( I - a ) 3 / 2 R
(-q+
( 1 / 6 ) n 3 ) dll
where p =
h-1(l-a)3/2
30/2-1 O(h
=
for h
m
-f
+
0.
Hence, one c a n a p p l y a g a i n t h e s t a t i o n a r y p h a s e method f o r computing t h e asymptotics f o r P ( x , t ) with p as a l a r g e parameter. h T h e r e f o r e , t h e a s y m p t o t i c s above are v a l i d f o r P ( x , t ) i n t h e r e g i o n h x / t < 1-Ch' f o r any c o n s t a n t s C > 0 , o E ( 0 , 2 / 3 ) . N o t i c e t h a t i n t h e r e g i o n 1-a
=
1 +u
O(h
)
w i t h any g i v e n
0
> 0 t h e main
t e r m i n t h e a s y m p t o t i c e x p a n s i o n f o r P ( x , t ) h a s t h e form: h Ph(X,t)
-c
t
-1/3h-2/3
where
m
@ - 2 / 3 cos
= 3-2/321/3n-1
e de
0 Example 3.8.16. Again l e t h
For t
E
IR,
E
(0,h
x E
0
1
\=
b e a s m a l l parameter. Consider t h e i n t e g r a l :
{x
1
x
E IR,
x/h
E
Z}
t h e function E ( x , t ) is the h
s o l u t i o n o f t h e Cauchy problem:
LE
at
h
( x , t ) + (2h)
l i m E (x,t)
=
-1
(Eh(x+h,t)-E(x-h,t) = 0 ,
6h(x),
t-to where 6 (0) = h-', 6h(x) = 0 f o r x h The p h a s e f u n c t i o n
h a s two s t a t i o n a r y p o i n t s
E %..{0}
3.8. The Stationary Phase
+
(3.8.54) if
Eg(a) =
a
arccos a ,
_+
275
= x/t
la1 C I , moreover, f o r la1 < 1 t h e s t a t i o n a r y p o i n t s ( 3 . 8 . 5 4 ) are r e g u l a r . m
Hence, f o r la1 > 1 , i . e .
1x1 > I t / , on: has: E ( x , t ) = O ( h ) , h h u n i f o r m l y w i t h r e s p e c t t o ( x , t ) i n any compact b e l o n g i n g t o t h e s e t IIxj
'
-f
0
It\}.
Assume now t h a t 1x1 < Itl. L e t f ( < ) E C m ( R ) be s u c h t h a t 0
supp f c ( - 2 , 2 ) ,
E
f E 1 for 5
[-3/2,3/2].
Then one h a s :
One c a n c o n s i d e r t h e i n t e g r a l
2 m a s a n e x t e n s i o n o f E ( x , t ) t o R up t o a n e r r o r O(h ) , f o r h + 0 . h Applying t h e s t a t i o n a r y p h a s e method ( f o r m u l a ( 3 . 8 . 3 1 ) ) t o t h e i n t e g r a l ( 3 . 8 . 5 6 ) one g e t s t h e f u l l a s u m p t o t i c e x p a n s i o n f o r Eh ( x , t ) when h + 0 . Again w e w r i t e h e r e o n l y t h e f i r s t t e r m i n t h i s e x p a n s i o n : (3.8.57)
Eh ( x , t )
where la1 < 1 , a
=
-1 2 -d 2*(nh) ' / t / - f ( l - a ) cos(h-'t(aarccosa-Jl-a
x/t;
t h u s , s i n g s u p p Eh =
Now assume t h a t (1-a ) -ha
with
0
{\XI
2
)+n/4),
5 Itl}.
E (2/5,2/3).
Then ( 3 . 8 . 5 3 ) c a n b e r e w r i t t e n a s f o l l o w s :
where $ ( S )
= 5-(1/6)5
3
-sin
5
5
= O(5 )
A f t e r t h e change o f v a r i a b l e
Since p
=
( 1 - a ) 3/2h-1
each i n t e g r a l
+ m
(h
+
5
=
when
5
(1-a)
t n,
+ 0.
( 3 . 8 . 5 8 ) becomes
0 ) i s a l a r g e p a r a m e t e r , one c a n a p p l y t o
3. Singular Perturbations on Smooth Manifolds without Boundary
216
t h e S t a t i o n a r y Phase Method. Hence, w i t h 1-a = Ch',
2/5 < u < 2/3, one f i n d s :
and moreover,
T h e r e f o r e , t h e main t e r m i n t h e a s y m p t o t i c b e h a v i o u r of E ( x , t ) i n t h e h c a s e c o n s i d e r e d i s g i v e n by t h e f o r m u l a :
-
The a p p l i c a t i o n o f t h e S t a t i o n a r y Phase Method t o t h e i n t e g r a l o n 3/2 h-l + the large ( 3 . 8 . 5 9 ) w i t h P = (1-a)
t h e r i g h t hand s i d e o f
parameter l e a d s t o t h e following asymptotic formula for E ( x , t ) : h
where c . > 0 , j = 1 , 2 are any g i v e n c o n s t a n t s 7 y = m i n I 1 - 3 ~ / 2 , 5u/2-11 > 0.
< c 2 , 2/5 < a < 2/3 and
C
Example 3.8 . 1 7 . Consider t h e i n t e g r a l
0 i s an i n t e g e r . where h E ( 0 , h ] i s a s m a l l p a r a m e t e r and p 0 The f u n c t i o n ( 3 . 8 . 6 1 ) i s t h e s o l u t i o n of t h e Cauchy problem:
lim
t++o
G
(x,t) = 6(x).
Using t h e change of v a r i a b l e 5
where
1 -
prh +
(x/t-1)
2p- 1
,€ one f i n d s :
3.8. The Stationary Phase
277
2p (3.8.63)
a
= t(x/t-l)
2p- 1
For p = 1 u s i n g Cauchy‘s theorem one f i n d s e a s i l y : -
( 2 n t h ) - f e - ( x - t ) 2/ ( 2 t h )
-
Gl,h(~,t)
W e a r e g o i n g t o compute t h e f i r s t t e r m i n t h e a s y m p t o t i c e x p a n s i o n of G
Pth
( x , t ) when h
0 , u s i n g t h e S a d d l e P o i n t Method.
-f
The polynomial
g (5) P
=
i ~ - ( 2 p ) - ~ 5 ’ P , gp(5) :
c
1
+
c
1
has t w o regular stationary points n i_ ni -__ 2 (2p-1) Z(Zp-1) (3.8.64) = e , C 1 = -e
c0
w i t h t h e same imaginary p a r t , I m 5 .
I
g
P
(Lo)
= g
P
(ill
=
=
sin(;
~i 2(2p-1)
)
> 0 , where
i ( 1 - i ) e 2 ( 2 p - 1 ) , R e g (5.)< 0 , 2P P I
and f o r any o t h e r s t a t i o n a r y p o i n t 5
one h a s :
I t i s e a s i l y seen t h a t
max R e g ( 5 ) = R e g ( < , ) , P I
j = 0,l.
<EY
I n d e e d , t h e s t a t i o n a r y p o i n t s of t h e f u n c t i o n R e g ( 5 ) : y + R a r e t h e 2p- 1 P p o i n t s , where R e 5 = 0, i . e . the points C j , j = 0 , l i n (3.8.64) o n l y , so t h a t one h a s
The Cauchy theorem y i e l d s :
Applying t h e S a d d l e P o i n t Method t o t h e i n t e g r a l on t h e r i g h t hand
3 . Singular Perturbations on Smooth Manifolds without Boundary
278
s i d e o f ( 3 . 8 . 6 5 ) a t each s t a t i o n a r y p o i n t < , , j = 0 , 1 , one g e t s t h e f u l l 3 asymptotic expansion f o r G ( x , t ) when h + O . Wewritedown e x p l i c i t l y the P,h f i r s t t e r m of t h i s e x p a n s i o n : I
-1 2p-1 -f (3.8.66)G ( x , t ) = 2(2nh)-+(xt-l-l) a R e { J ; ; - - e a g P ( S O ) ] ( l + O ( h a - l ) )= 9 ( C 0 ) P,h P _ _P_ 1 2p- 1 -f -c h - l a = (2nth) 2 ( x / t - l ) (2p-1) w p ( c o s ( b h-'a+a ) ( l + O ( h a - l ) ) ~
P
P
Note t h a t one h a s : s i n g supp G
P*h
= ix = t}.
Example 3 . 8 . 1 8 . C o n s i d e r t h e Cauchy problem
d K (x,t) +h-l(%(x,t)-%(x-h,t)) dt
= 0, x
E \,
t
E
W+
h
(3.8.67) 1i m
t++o where
%(x,t)
= Ah(X)
"i, and 6h ( x ) a r e d e f i n e d i n Example 3.8.16. Using t h e F o u r i e r t r a n s f o r m i n x E "i, , one f i n d s e a s i l y :
(3.8.68)
% ( x , t ) = (2nh)-'
J
e
h
-1
(ix<-t(l-e
-i<
))dC.
/
R e w r i t i n g t h e l a s t i n t e g r a l i n t h e form:
one g e t s :
W e s h a l l f i n d t h e a s y m p t o t i c b e h a v i o u r of K ( x , t ) f o r x > 0 , x E h 0.
when h
-f
The f u n c t i o n (3.8.71 h a s one s t a t i o n a r y p o i n t
c0
= i In
a.
%,
t > 0,
3.8. The Stationary Phase
279
Denote (3.8-72)
Y = i~
1
5 E c
1
,
Im g(a,i)
One f i n d s e a s i l y t h a t t h e c u r v e y i n
=
I m g(a,c 1,
c1
<
(Re
0
71.
i s g i v e n by t h e e q u a t i o n
Furthermore, denoting
Y,-
=
Ii
I
c
E
1
c ,c
= fn+in,
rl
E IR+},
the integrals (2nh)-l J
e
t h - I ( i a c - ( l-e-ii
Yf e x i s t and c o i n c i d e . Hence, t h e Cauchy theorem y i e l d s : (3.8.73)
%(x,t)
= (2nh)
-1
J e
th-l(iac-(l-e-i'
))dc
Y A s t r a i g h t f o r w a r d computation
shows t h a t
Applying t o t h e l a s t i n t e g r a l t h e s a d d l e p o i n t method one g e t s t h e f u l l a s y m p t o t i c e x p a n s i o n f o r % ( x , t ) , x > 0 , when h
-f
0. W e w r i t e only the
f i r s t term i n t h i s e x p a n s i o n : (3.8.74)
Kh(x,t) =
(2nh)
-
(27ihx/t)
-1/2
(x/t)
-1/2
e
-(x/h+1/2)
(3.8.75)
Y,(x,t)
-h
ln(x/t))
-1
w i t h some a
Now assume t h a t ( t - x ) - h a i n t e g r a l (3.8.68)
-h-'(t-x+x
(t-x)
E
,
-
x > 0, t > 0, h
(1/3,1/2).
-f
0.
Rewriting t h e
i n t h e form =
(21rh)-'(a-l)
<
e
t h - I (a-1 )
(is- ( 1/2) t; * ) +th-lJ, ( (a-1 5 ) as
"a
one g e t s ( a s p r e c i o u s l y i n Example 3 . 8 . 1 6 )
t h e conclusion,
that in the
case c o n s i d e r e d t h e main c o n t r i b u t i o n i n t h e a s y m p t o t i c b e h a v i o u r o f %(x,t) for x - t
-
h'
,
u E
(1/3,1/2),
i s g i v e n by t h e i n t e g r a l
3 . Singular Perturbations on Smooth Manifolds without Boundary
280
i n t h e p r e v i o u s example The f u n c t i o n
where
$(5)
=
(4/3)sin 5
-
(1/6) s i n 25 - ( 1 - c o s
5)2
i s t h e s o l u t i o n of t h e Cauchy problem
i s t h e s h i f t o p e r a t o r 8 u ( x ) = u ( x + h ). h One f i n d s i n t h e same way a s p r e v i o u s l y t h a t f o r
(x-t)-ha,
t h e main t e r m i n t h e a s y m p t o t i c b e h a v i o u r o f K
( x , t ) when h
where Bh
c o i n c i d e s w i t h t h e r i g h t hand s i d e o f $(c)
=
U E (2/5,2/3), +
2rh (3.8.66) with p = 2 , since
~ - ( 1 / 4 ) < ~ + 0 (
+
0
0.
Example 3.8.19. Consider t h e f u n c t i o n
where
E
E ( 0 . ~ ~i s1 a small p a r a m e t e r .
The f u n c t i o n P ( x ) i s t h e s o l u t i o n o f t h e boundary v a l u e problem: E
2 (1-E A ) P ( x )
=
0,
2
x E W + = {x
1
x E W 2 ,x 2 > 01
(3.8.77)
P ( x ) = 6 ( x 1 ) , l i m P ( x ) = 0. x ++a ++o 2 2 We s h a l l a p p l y t h e s a d d l e p o i n t method f o r computing t h e a s y m p t o t i c lim
x
e x p a n s i o n o f P ( x ) f o r E + 0 , and a g i v e n x E R:. The f u n c t i o n
3.8. The Stationary Phase
28 1
c a n be e x t e n d e d a n a l y t i c a l l y i n t o t h e complex p l a n e 5 c u t s along t h e r a y s 1
1-
{5
=
[
5
=
in, n
domain where
=
1
{C
< -1).
5 = in,
E
C1 w i t h two
=
1+ U 1-, t h e
ri > 1) arid
W e s h a l l d e n o t e by C 1 \ l ,
1
is analytic.
The f u n c t i o n
g(a,<)
(3.8.78)
1 h a s i n C \{I}
2 1/2 ia<-(1+< )
.
one s t a t i o n a r y p o i n t
5 (a) 0
(3.8.79)
=
=
ia(l+a2 ) -4
2 1/2 and g ( a , s ( a ) ) = - ( l + a
0
.
The s e t
i s c o n n e c t e d , c o n t a i n s t h e r e a l a x e s and i t s boundary 20'
C+an 5
=
+ m ,
0 a s i t s asymptote when
<
+-, 5
>
has t h e l i n e
0 and t h e l i n e <-an
=
0 when
5 < 0.
A c t u a l l y , a f t e r an e l e m e n t a r y Computation, one f i n d s t h a t f o r a # 0
w h i l e f o r a = 0 one h a s : Uc = C(')\{l].
0
Denote
One f i n d s e a s i l y t h a t f o r a # 0 t h e c o n t o u r
r
i s a p i e c e o f hyper-
bola defined as follows:
while f o r a = 0 t h e contour The c u r v e I' when 5
+
-, n
ro
c o i n c i d e s w i t h t h e r e a l a x e s R.
l i e s i n D+ and h a s t h e l i n e at-q
> 0 and t h e l i n e a<+q = 0 when 5
d i s t a n c e from a p o i n t 5
E r
+
=
-,
0 as i t s asymptote rl < 0 .
t o aD+ goes t o i n f i n i t y when 5
Hence, t h e +
-.
Thus,
3 . Singular Perturbations on Smooth Manifolds without Bounda y
282
t h e Cauchy t h e o r e m y i e l d s : 2 1/z
-1 (3.8.82)
PE(x) = ( Z n ~ ) - l J eE
(ix1'-x2(1+5
)
'dc,
x
E R 2+ .
ra Since g ( a , < ) = R e g(c(,L), V Re g ( O r , < ) on
r
< c ra,
e a c h s t a t i o n a r y p o i n t of
i s also a s t a t i o n a r y p o i n t of g ( a , < ) . Hence
by ( 3 . 8 . 7 9 ) i s t h e o n l y s t a t i o n a r y p o i n t of Re g ( a , < ) on ly Re g(a,<) i n (3.8.79)
+
--
when i
+
-, < E Ta
<
r .
( a ) defined 0 Since obvious-
one g e t s t h e c o n c l u s i o n , t h a t i 0 ( a )
i s t h e maximal p o i n t f o r Re g ( a , < ) on
r
a'
so t h a t
3.8. The Stationary Phase
283
x E which f o r n
=
q =i x
>
0},
2 c o i n c i d e s w i t h P ( x ) i n t h e p r e v i o u s example.
The f u n c t i o n P n,E
( x ) i s t h e s o l u t i o n of t h e boundary v a l u e problem:
2
x E Ry
(1-E A)P (xi = 0 , nrE
W e a r e g o i n g t o a p p l y t h e S a d d l e P o i n t Method f o r computina
t h e asymptotic expansion of P
( x ) f o r ~ 1 x I - l+ 0 .
n.E
The f u n c t i o n (3.8.89)
g (x,5 )
is t h e a n a l y t i c a l extension of g ( x , S ' ) ,
which c o n t a i n s Rn-'
domain i n Cn-'
1/ 2
i < x ' , 5 '>-xn( 1+<<' , 5 '>I
=
5 ' E Rn-' i n t o some c o n n e c t e d
and t h e o n l y s t a t i o n a r y p o i n t 5 '
s'
5 ; = ixvlxl-' of g ( x , 5 ' ) . One f i n d s : g ( x . 5 ' )
=
-1~1.
Again i n t r o d u c e
D+
=
rx
=
{C'
(3.8.90)
I
5' E
I
5 ' E D,:
C
n- 1
,
Re g ( x , < ' ) < 01,
Im g(x,<')
=
I m g(x,L*) =
i n t h e p r e v i o u s Example 3.8.19 one can show t h a t Rn- 1 so t h a t t h e Cauchy theorem y i e l d s :
AS
-1 (3.8.91)
=
n,E
(2.rrE)l-n
eE
rX
01. i s homological t o
.
( i < x ' , c ' > - x n ( 1 + < ~ ' , < * > )dg ~ /' *
rX
Again, t h e s a m e argument a s i n Example 3 . 8 . 1 9 l e a d s t o t h e c o n c l u s i o n -1 . i s t h e o n l y maximal p o i n t of t h e f u n c t i o n Re q ( x , L ' ) on t h a t 5; = i x ' / x j
rx,
t h a t i s Re g(x,<') < R e g ( x , < ' ) , V 5 E rx\{c;}.
I n o r d e r t o a p p l y t o t h e i n t e g r a l ( 3 . 8 . 9 1 ) t h e S a d d l e P o i n t Method, one h a s t o compute (3.8.92)
A 9 (x)
=
2 -det D 5 , g , g ( x , 5 t ) \
<'=
3. Singular Perturbations on Smooth Manifolds without Bounda y
284 One finds: (3.8.93)
A
9
where y' = x
t (x) = Ixln-' detl)Id+y' y'll
-1
x'.
Denoting t
D(E,Y') = det( ( I d + ~ y y'I '
1
2 2 one finds: D(o,~')z 1, (a/aE)D(E,Y')lE=O = l y~l , ( a 2 / a E )D(E,~'): 0, so that D(1,y') = 1 + ly'l
2
and (3.8.94)
A
9
(x)
=
X:~]X]~.
Now we can write the first term in the asymptotic expansion of
Again, as in the previous example one finds that for
/XI/€
-+
0 the
(x) is nothing else but the first term in the asymptotic expansion for P n,E Poisson kernel for the Laplace operator in En with the Dirichlet boundary condition at x
where
=
0, i.e.
is the area of the unit sphere in IRn.
The two asymptotics (3.8.95),
(3.8.96) can be written as one
asymptotic formula:
for ( ( x / / E + E / ~ x l )+ - ~m .
Let A be symmetric positive definite nxn matrix. Denote by P (X) n,A,E 2 E
the Poisson kernel for the operator 1 -
a
7
=
a/ax., with the Dirichlet boundary condition at x 7
...,
= 0.
A similar argument leads to the following asymptotic formula:
3.8. The Stationary Phase
for
-1
(IxI/E+E/IxI)
-f
m,
where
01
= A
-1
,
285
Ix1f =
i s t h e s o l u t i o n o f t h e Cauchy problem:
l i m E (x,t) = 0, l i m
t+o
a/at
E
(x,t) = 6(x)
t-to
One h a s : (3.8.99)
E ( x , t ) E 0 f o r 1x1 > t .
I n d e e d , assume, f o r i n s t a n c e , t h a t x > t . Using t h e a n a l y t i c e x t e n s i o n of
<E> i n t o t h e domain C1\{l} from Example 3 . 8 . 1 9 a n d t h e Cauchy
theorem, one c a n s h i f t t h e i n t e g r a t i o n on t h e r i g h t hand s i d e o f ( 3 . 8 . 9 7 ) 1 and i s homological t o t h e o v e r t o any c o n t o u r r , which l i e s i n C \(l} r e a l a x i s IR.
It is e a s i l y seen t h a t f o r x > t t h e contour
r
c o n s i s t i n g of
I
two s i d e s o f t h e c u t 1 = ( 5 5 E C1, 5 = i n , ri > 1 1 run i n t h e o p p o s i t e lj2 d i r e c t i o n s , where ( l + g L ) h a s t h e imaginary p a r t of o p p o s i t e s i g n s , i s
*
homological t o IR. One c h e c k s e a s i l y , t h a t t h e c o r r e s p o n d i n g i n t e g r a l o v e r i s i d e n t i c a l l y z e r o . I f x < - t , t h e n t h e c o r r e s p o n d i n g homological c o n t o u r
c o n s i s t s o f t h e Cut 1
= {<
1
5 E
C1,
5
=
in,
ri < -1)
r u n t w i c e from t w o
d i f f e r e n t s i d e s i n two o p p o s i t e d i r e c t i o n s . Now assume t h a t 1x1 < t . W e a p p l y t h e s t a t i o n a r y p h a s e method f o r computing t h e f i r s t t e r m i n t h e a s y m p t o t i c e x p a n s i o n of E ( x , t ) when
E
-f
One f i n d s i n t h a t case f o r E ( x , t ) :
N o t i c e t h a t t h e s i n g u l a r s u p p o r t of t h e f a m i l y of d i s t r i b u t i o n s ( 3 . 8 . 9 7 ) c o i n c i d e s w i t h t h e c l o s u r e o f t h e cone V+
=
((x,t)
I
(x,t) E R x
IR+
,
1x1 < t } , as a consequence o f t h e a s y m p t o t i c f o r m u l a ( 3 . 8 . 1 0 0 ) . A s i m i l a r a s y m p t o t i c formula h o l d s a l s o i n t h e c a s e when x
E lRn , n > 1.
The p r e v i o u s argument y i e l d s ( 3 . 8 . 9 9 ) i n t h e m u l t i d i m e n s i o n a l case x E
mn,
as w e l l .
0.
r
286
3. Singular Perturbations on Smooth Manifolds without Boundary
Example 3.8.22 W e s h a l l c o n s i d e r h e r e a one p a r a m e t e r f a m i l y of f i n i t e d i f f e r e n c e
a p p r o x i m a t i o n s ( i n s p a c e v a r i a b l e ) of a h y p e r b o l i c f i r s t o r d e r d i f f e r e n t i a l o p e r a t o r , t h e mesh-size h b e i n g t h e s m a l l p a r a m e t e r . The s i n g u l a r s u p p o r t and t h e a s y m p t o t i c b e h a v i o u r
(for h
-t
0 ) of t h e
Fundamental S o l u t i o n o f t h e Cauchy problem f o r t h e s e a p p r o x i m a t i o n s , w i l l b e i n v e s t i g a t e d under t h e a s s u m p t i o n t h a t t h e y s a t i s f y t h e von Neumann's s t a b i l i t y condition. L e t a ( s ) : C1 + C1 b e a n a n a l y t i c f u n c t i o n of
s =
n+iT,
which s a t i s f i e s
the conditions: in q E R ,
( i )a ( s ) i s Z v - p e r i o d i c
( i i )a ( 0 ) = 0 , a ' ( 0 ) = i w w i t h s o m e w 6 R ,
E R.
( i i i )Re a ( q ) 5 0 , V r' Let
(3.8.101) a ( s )
iws+ C
=
k2p
a s k
k
b e t h e T a y l o r e x p a n s i o n f o r a ( s ) i n a neighbourhood of z e r o . D e f i n i t i o n 1.
1 " . The f u n c t i o n a ( s ) i s s a i d t o b e p - p a r a b o l i c if
i n t h e sense of Petrovsky
(i)- ( i i i )a r e f u l f i l l e d and, moreover, i f
(3.8.102)
R e a ( q ) < 0 , V I)
p = 2 r i s even and R e a
P
E
37,
r' # 0 (mod 2 n ) ,
< 0.
2". The f u n c t i o n a ( s ) i s s a i d t o b e ( p , q ) - p a r a b o l i c i n t h e s e n s e of S h i l o v if ( i ) - ( i i i )(,3 . 8 . 1 0 2 ) are f u l l f i l l e d and t h e c o e f f i c i e n t s ak i n
(3.8.101)
s a t i s f y the conditions: ( 3 . 8 . 1 0 3 ) a # 0 , R e ak = 0 , p i k < q , R e a < 0 , q = 2 r i s even. P q
3 " . The f u n c t i o n a ( s ) i s s a i d t o b e p - h y p e r b o l i c i f ( i ) - ( i i ia) re f u l f i l l e d a n t ( 3 . 8 . 1 0 4 ) Re a ( r l )
Let
5'
R
Z
0 , V II
E IR,
a
P
# 0.
b e a one p a r a m e t e r f a m i l y o f meshes i n I R , c o n t a i n i n g t h e
o r i g i n , h b e i n g t h e mesh s i z e . Denote by Ah t h e f i n i t e d i f f e r e n c e o p e r a t o r a s s o c i a t e d w i t h t h e f u n c t i o n a ( s ) above a c c o r d i n g t o t h e f o r m u l a (3.8.105)
= h-la(hDx),
D~ = - i a / a x
3.8. The Stationary Phase
287
t h e f u n c t i o n a ( n ) b e i n g a l s o c a l l e d t h e symbol o f A Since a ( n ) i s 2n-periodic, a(n)
z c
=
h
.
it c a n b e expanded i n t o a F o u r i e r series,
eikq,
kEZ so t h a t t h e o p e r a t o r Ah i s a l s o w e l l - d e f i n e d on t h e m e s h - f u n c t i o n s U(X)
:
R
h
C1 w i t h compact s u p p o r t by t h e f o r m u l a :
-f
(Ah u ) (x) L e t E ( x , t ) : IRh
h
=
1 h-lCku(x+kh), kEZ
R+ b e t h e s o l u t i o n o f t h e Cauchy problem:
X
h
(3.8.106)
x E R h
(D + i A ) E h ( x , t ) = 0 ,
( x , t ) E Rh x
lim
x E \,
t
E
t++o
h
(x,O) = 6 h ( x ) ,
where, a s p r e v i o u s l y , 6 ( x ) = 0 , V x # 0 , d h ( 0 ) h
The o p e r a t o r D
+inh t
=
R+
h
-1
,
D
=-ia/at.
t i s a f i n i t e d i f f e r e n c e approximation ( i n space
v a r i a b l e x ) of t h e h y p e r b o l i c o p e r a t o r Dt-uDX,
t h e approximation e r r o r
being 0(hP-'). Definition 2. The o p e r a t o r D t + i A h
i s s a i d t o be p-parabolic
or ( p , q ) - p a r a b o l i c o r p-
h y p e r b o l i c if t h e symbol a ( n ) of Ah i s p - p a r a b o l i c ,
or ( p , q ) - p a r a b o l i c o r
p-hyperbolic. Definition 3.
E R x %+ i s s a i d t o b e r e g u l a r f o r E ( x , t ) i f t h e r e (x h O'tO) e x i s t c o n s t a n t s Ckl such t h a t
A point
k l
IDtDx,hEh(~,t)I
5 Ckl,
f o r e a c h map ( x , t ) : (O,ho] + R h
V k,l,
V h E (O,hol,
X
R+ w i t h ( x , t )
+
(x0,t0) for h
-+
0;
here D = - i ( B - l ) / h , where 0 i s t h e s h i f t o p e r a t o r 8' u ( x ) = u ( x + h ) . x,h h h h The c l o s u r e o f t h e complement i n R X of t h e s e t of a l l r e g u l a r p o i n t s of E ( x , t ) i s c a l l e d t h e s i n g u l a r s u p p o r t o f Eh and i s d e n o t e d h s i n g supp E h' C l a i m 1.
I f t h e approximation D t + i A h
i s ( p , q )- p a r a b o l i c , t h e n
( 3 . 8 . 1 0 7 ) s i n g supp Eh = { ( x , t ) E R
X
z+1
x + u t = 0)
3 . Singular Perturbations on Smooth Manifolds without Boundary
288
Proof of Claim 1. Using the discrete Fourier transform in space variable x E l€i,,one gets for E (x,t) the formula h
Analyticity of a(<) and 27-periodicity yield
as a consequence of Cauchy's theorem.
We shall choose Then for
<
=
sufficiently small and such that sgn T =sgn(x+wt)
T
q+iT with l q l 5 6, using the formula
one gets the conclusion that for 1171 5 6,
] T I 5 TO,
sgn
T
=
sgn(x+ut) with
sufficiently small, the following inequality holds: (3.8.110) Re(ix<+ta(<))
v
for
17
E [-6,61,
<-\TI
Ix+wt//2
V T E [ - T ~ , T ~ ] , sgn T = sgn(x+wt).
since Re a(n)
s -c6, v
q E [-rr,nI,
J r i J2
s
with some positive
constant C6 (this is a consequence of (p,q)-parabolicity), one can choose so small that Re a(<) S -(1/2)Cs,
T~
1.r
I
v
ri with
6 2
6 T ~ sgn , T = sgn(x+wt) , so that combining
lql
$ 7,
t/ T
with
t with (3.8.110) one will
have : (3.8.111) Re(ix<+ta(<)) 5 for
v
11 E
[-x,al,
V T
E
-IT\
/x+wtl/2
t-TO,~Ol,
sgn
T =
sgn(x At) with
T~
sufficiently
small. Taking T =
T~
sgn(x+wt) in (3.8.109) and using (3.8.1111, one gets
the following estimate for Eh (x,t): -To I X+Wt I / (2h) /Eh(x,t)) 5 h-'e k 1 For the derivatives Dx,hDtEh(~,t)one gets, using the same argument, the estimates:
k
1
h-(k+l+l)e-~0/~+~t)/)2h)
3.8. The Stationary Phase Now l e t x t w t
0, t > 0. As a consequence o f
=
289 (p,q)-parabolicity,
the
main c o n t r i b u t i o n i n t o t h e a s y m p t o t i c b e h a v i o u r o f t h e i n t e g r a l (3.8.108)
i s g i v e n by a neighbourhood o f z e r o R e a ( q ) 5 -C6,
in1
5 6 s i n c e f o r 1q1 2 6 one h a s :
w i t h some p o s i t i v e c o n s t a n t C 6 , depending on 6 . Hence,
E ( x , t ) = (2rrh)
-1
h
J
h - l ( i x q + t a ( q ) ) d r lt O ( e - t h - l C
6).
lrlIS6
One h a s i n t h e case c o n s i d e r e d : ixq+ta(q)
t a rl Pb ( q ) , P
=
E C"([-6,61), b ( 0 )
where b ( n )
R e a
P
P
# O
= 1.
I n t r o d u c i n g a new v a r i a b l e
5 w i t h zl/'
=
lip,
q(b(rl))
rl =
t h e main b r a n c h ( z 1 /P
Q ( 5 ) , rl'(0)
l Z Z1
=
1,
= 1) of t h i s multivalued f u n c t i o n ,
one c a n r e w r i t e t h e l a s t a s y m p t o t i c f o r m u l a f o r Eh ( x , t ) w i t h x+wt = 0 and 6 so s m a l l t h a t 5 f-t rl t o b e a b i h o l o m o r p h i c d i f f e o m o r p h i s m , i n t h e following fashion:
'I
Y6 where t h e c u r v e y6 i n a s m a l l complex neighbourhood o f t h e o r i g i n i s t h e image of [-S,S] under t h e b i h o l o m o r p h i c d i f f e o m o r p h i s m rl
5.
++
N o t i c e t h a t R e ( a g P ) < 0 , b' g E y6\iO}. I f 6 i s s u f f i c i e n t l y s m a l l P t h e n one w i l l have R e ( a iP)< -C6 < 0 on t h e segments P 1* = {Z, E C 5 = Re < ( ? 6 ) + i . r , C I I m <(?6)\}. Using t h e Cauchy theorem,
\TI
1
one c a n d i f f o r m t h e c u r v e y 6 i n t o t h e c o n t o u r which i s t h e u n i o n of t h e i n t e r v a l [Re 5 ( - 6 ) ,
Re
<(&)I
C
R
and t h e c o r r e s p o n d i n g p i e c e s o f t h e
segments 1
2-
Obviously, t h e i n t e g r a l s o v e r t h e s e p i e c e s of 1+ g i v e e x p o n e n t i a l l y -
s m a l l terms when h
+
0 , so t h a t t h e main c o n t r i b u t i o n i n t o a s y m p t o t i c
b e h a v i o u r of E ( x , t ) i n t h e case c o n s i d e r e d ( x + o t = 0 , t > 0 ) i s g i v e n by h t h e i n t e g r a l o v e r t h e i c t e r v a l I = [Re 5 ( - 6 ) , R e 5 ( 6 ) 1 C W , -1 t a h-lSP - t h C6 E ( x , t ) = (2nh)-' I q ' ( 5 ) e dS+O(e ) = h I
t a h-lEP =
(2nh)-' J e I
d5+0((h/t)(2/p)-1)
=
3. Singular Perturbations on Smooth Manifolds without Boundary
290
R
where
a c
P
=
cp
dg # 0,
J e
(an)-'
w
t h e l a t t e r i n t e g r a l b e i n g c o n v e r g e n t s i n c e Re a
P
5 0, a
P
f C, p = 2 r
2
2.
T h a t e n d s t h e p r o o f of Claim 1. C l a i m 2.
If t h e a p p r o x i m a t i o n D t + i A h
for (x,t) E \ x
i s p-hyperbolic,
then E ( x , t ) is w e l l defined h
R and
( 3 . 8 . 1 1 2 ) s i n g s u p p Eh
=
{(x,t) E R
2
1
3 rl E R s . ~ .a ' ( n )
= -ix/t}.
N o t i c e t h a t t h e n u l l - c h a r a c t e r i s t i c x+wt = 0 i s c o n t a i n e d i n s i n g supp E h' since a ' ( 0 ) = i w . Proof of C l a i m 2 . since a ( n )
=
iw(n), w
:
[-n,n]
-Z
IR b e i n g 2 n - p e r i o d i c r e a l v a l u e d a n a l y t i c
f u n c t i o n , one can r e w r i t e ( 3 . 8 . 1 0 8 ) a s f o l l o w s
and a p p l y t o t h e i n t e g r a l on t h e r i g h t - h a n d s i d e of if w'(n)
3.8.1,
(3.8.113) P r o p o s i t i o n
# - x / t , V n € [-ii,n]
S i n c e w ( n ) i s a n a n a l y t i c f u n c t i o n , t h e r e may b e o n l y a f i n i t e number of 1 5 j 5 r , where w " ( g . ) = 0 . F o r t h e same r e a s o n f o r p o i n t s E . E [-n,.rr], 3 3 e a c h g i v e n v a l u e of x / t t h e r e may b e o n l y f i n i t e l y many p o i n t s n . ( x / t )
where w ' ( n )
= -x/t.
s t a t i o n a r y p o i n t ri with 5 .
3'
7 L e t x / t b e such t h a t t h e r e e x i s t s a t l e a s t one
f o r x r l + t w ( n ) which i s r e g u l a r , i . e . d o e s n o t c o i n c i d e
1 5 j 5 r d e f i n e d above a s z e r o s of w " ( n ) .
Then t h e c o r r e s p o n d i n g
c o n t r i b u t i o n i n t o t h e a s y m p t o t i c b e h a v i o u r of E ( x , t ) f o r h j. 0 from t h i s h s t a t i o n a r y p o i n t c a n b e computed by u s i n g t h e S t a t i o n a r y Phase method, so t h a t t h e l i n e s x t t w ' ( n ) = O b e l o n g t o s i n g s u p p Eh if n s ( x / t )
ft 1 5 , ,..., 5,).
I f min a ' ( n ) < - x O / t O < max a ' ( n ) t h e n f o r any ( x , t ) s u c h t h a t Ix / t - x / t l 0
0
< 6 with 6 s u f f i c i e n t l y s m a l l t h e r e a r e s t i l l s t a t i o n a r y p o i n t s
n s ( x / t ) , s o l u t i o n s o f t h e e q u a t i o n a'(n)
=
-x/t,
so t h a t i n f a c t t h e r e a r e
continuum l i n e s x t t w ' ( q s ( x / t ) ) = 0 i n R2 b e l o n g i n g t o s i n g supp Eh,
ns(x/t)
3.8. The S t a i m a y Phase
291
b e i n g a r e g u l a r s t a t i o n a r y p o i n t , and t h e r e a r e o n l y f i n i t e l y many l i n e s i n t h e s e t on t h e r i g h t hand s i d e o f
( 3 . 8 . 1 1 2 ) f o r which t h e r e may b e no
r e g u l a r s t a t i o n a r y p o i n t s f o r t h e i n t e g r a l d e f i n i n g Eh, c h a r a c t e r i s t i c x+wt = 0 b e i n g one o f them, u n l e s s p
=
the null-
2 . Hence ( 3 . 8 . 1 1 2 )
i s proved. W e state here
(without a f u l l proof) t h e following r e s u l t concerning
t h e a s y m p t o t i c b e h a v i o u r of E ( x , t ) ( f o r h + 0) i n s o m e s t r i p h a ( h ) 5 x+wt [ < a 2 ( h ) w i t h a . ( h ) -+ 0 f o r h -f 0 , s a t i s f y i n g a p p r o p r i a t e c o n d i t i o n s 1 3
I
Claim 3 . 1 " . The a p p r o x i m a t i o n D + i A h b e i n g p - p a r a b o l i c , t formula h o l d s : (3.8.114) E ( x , t ) = h h
-1/2 -l/G (p-1))
t
t h e following asymptotic
I x+wt 1 - (p-2) / (p-1)
V ( x , t ) such t h a t h~x+otl-P(P-l)+h-l/x+wtl (p+l)'(p-l) Aj,
Bjr
-+
0 when h
+
0; here
j = 1 , 2 a r e some c o n s t a n t s , R e A . > 0 , j = 1 , 2 , which depend o n l y 1 i n (3.8.101). P
on w and t h e c o e f f i c i e n t a 2'.
+inh b e i n g p - h y p e r b o l i c , assume a d d i t i o n a l l y t h a t t # 0 (mod Z I T ) . Then t h e f o l l o w i n g a s y m p t o t i c f o r m u l a e
The a p p r o x i m a t i o n D
# 0, V n
a'(n)-iw hold:
a ) f o r p even
V ( x , t ) s u c h t h a t h l x + w t /- o / ( P - l ) + h - l A,
I'
B., 1
j
=
Ix+wt/
1 , 2 , are some non-vanishing
which depend o n l y on w and a
P
(p+l)'(p-l)
+ 0 when h
constants, I m A . 1 i n (3.8.101).
=
-f
0; h e r e
0 , j = 1,2,
b ) f o r p odd, one h a s t h e same f o r m u l a ( 3 . 8 . 1 1 4 ) a s i n t h e p - p a r a b o l i c case f o r sgn(x+wt) t h e p-parabolic
=
-sgn I m ap, a n d , t h e same f o r m u l a ( 3 . 8 . 1 1 5 ) a s i n
c a s e w i t h even p , when s n g ( x + w t ) = sgn I m a
P
.
The f o l l o w i n g argument g i v e s an i d e a how t o p r o v e t h e f o r m u l a e ( 3 . 8 . 1 1 4 ) , (3.8.115).
3 . Singular Perturbations on Smooth Manifolds without Bounda y
292
Introducing the new variable n = /(x/t)+o(l’(p-l)E,
one rewrites the
integral (3.8.108) in the form where the phase function in the exponent is as follows:
where $ ( n ) = a(q)-iwn-a np
P
=
for rl
O(np+l),
-f
0.
One can consider O(x,t,E) as a small perturbation of the phase function
when a =
I (x/t)+wIP-l
is small, the remainder being of order
+1 p+l $ ( a S ) = O(ap 5 ) , when a
One has the natural conditions on large parameter when h
+
+
0.
1 (X/t)+ W /
:
h-ll (X/t)+W lp’(p-l) has to be a
0 in order to apply saddle point or stationay
phase method to the corresponding integral with the phase function iS+a SP, while h-ll(x/t)+wl(p+l)/(p-l) has to be a small parameter when
P
h
+
0, for being able to use the Taylor expansion for the function
exp(ih-’t$(aE)) as in Example 3.8.16. Thus, the computation of the main term in the asymptotic expansion of E (x,t) for (x,t) such that -1
h 1 (x/t)+wl p/(p-l) + m, h-ll (x/t)+wl t p + l ) / ( p - l ) + O the case of the polynomial phase function iS+a 5’.
P
(h+O) is reduced to The use of the saddle
point (as in Example 3.8.17) or stationary phase method (in the corresponding p-hyperbolic case), leads to the formulae (3.8.114), (3.8.115) above. Example 3.8.23. 1 Let 5 E C , 151 < 1 and let:
where 5
*
is the complex conjugate of 5.
Introduce s;(c;e)Je=l = r,
o
<
r 5 I.
With $(x,t,<,w) = xw+(t/r)arg S(r;,eiw) consider the following integral
3.8. The Stationary Phase
293
It i s e a s i l y seen t h a t E ( 5 ; x . t ) is t h e s o l u t ion of t h e following f i n i t e h d i f f e r e n c e i n i t i a l v a l u e problem:
E (5;x,t) h where R
1
8
U ( t )
E
= {kh, k
h o p e r a t o r on
1
6h(x),
=
Z},
lR1
T ,+
= {mT,
m E
8 u ( x ) = u ( x + h ), and 6
%,
i s the s h i f t h 1 i s t h e s h i f t o p e r a t o r on IR T,+
Z+>,
h
T = rh,
8
‘
= U(t+T).
The problem ( 3 . 8 . 1 1 7 ) i s a n i m p l i c i t d i f f e r e n c e a p p r o x i m a t i o n o f t h e f o l l o w i n g d i f f e r e n t i a l Cauchy problem: (D -D ) E ( x , t ) t x
=
l i m E(x,t) t++o
6(x)
=
0,
x E IR, t E Rt
where 6 ( x ) i s t h e Dirac 6 - f u n c t i o n . Obviously, E ( x , t )
=
6 ( x + t ) , so t h a t s i n g supp
E
=
{x+t=O}. W e a r e
g o i n g t o i n v e s t i g a t e t h e s i n g u l a r s u p p o r t o f t h e mesh-function Obviously, f o r 5
=
0
(1- =
l ) , one h a s :
E (O;x,t)
=
Eh ( 5 ; x . t ) .
6h ( x t t ) , so t h a t
h s i n g supp E ( O ; x , t ) = s i n g supp E ( x , t ) . F u r t h e r , s i n c e h iw
IS(<;e
)I
=
1,
v
w E [-n,nI,
so t h a t it i s e a s i l y s e e n t h a t t h e s t a t i o n a r y p o i n t s of t h e p h a s e f u n c t i o n $ a r e s o l u t i o n s of t h e e q u a t i o n :
/eiw-5/’
= a -1 j 1 - 5 l 2 ,
a
=
-x/t.
The l a s t e q u a t i o n h a s p r e c i s e l y two z e r o s w . ( < ; a ) E [ - n , n ] , 3 -1 c1 E [ X - ( c ) ,X+(C) ] , w i t h
provided t h a t
A s t r a i g h t f o r w a r d computation shows t h a t b o t h s t a t i o n a r y p o i n t s
w . ( ~ ; x , t ) ,j = 1 , 2 , are r e g u l a r , p r o v i d e d t h a t 5 # 0 . Applying t h e 3 s t a t i o n a r y p h a s e method t o t h e i n t e g r a l ( 3 . 8 . 1 1 6 ) one g e t s a f u l l a s y m p t o t i c e x p a n s i o n f o r Eh ( < ; x , t ) ,
3. Singular Perturbations on Smooth Manifolds without Bounda y
294
f o r each ( x , t ) E
B
X
R + such t h a t ( - t / x )
E
[x-(<),x+(<)].
T h e r e f o r e , one h a s
T h i s phenomemon o f s p r e a d i n g o u t of t h e s i n g u l a r s u p p o r t o c c u r s f o r a l l h y p e r b o l i c f i n i t e d i f f e r e n c e a p p r o x i m a t i o n s ( i . e . such a p p r o x i m a t i o n s iw t h a t jS(<,e ) E 1 ) of h y p e r b o l i c d i f f e r e n t i a l o p e r a t o r s .
I
One h a s
a n d , moreover, one c h e c k s e a s i l y t h a t
where
F u r t h e r , w i t h wo
=
-2 a r g
<
one f i n d s :
Using t h e s a m e argument a s i n Example 3.8.18 one f i n d s t h e f i r s t t e r m i n t h e a s y m p t o t i c e x p a n s i o n f o r E ( x , t ) i n a neighbourhood of t h e z e r o h 2 -1 -f a, (x+t)3h-1 + 0 for h c h a r a c t e r i s t i c x + t = 0 such t h a t ( x + t ) h
+
0.
W e w r i t e it h e r e down:
N o t i c e t h a t t h e f i r s t e x p o n e n t i a l term on t h e r i g h t hand s i d e of t h e l a s t formula (which i s t h e c o n t r i b u t i o n from w ( < ; a ) ) c o n v e r g e s t o 6 ( x + t ) a s 1 h + 0 and i t s b e h a v i o u r i s l i k e t h e one of t h e fundamental s o l u t i o n f o r S c h r o d i n g e r ' s e q u a t i o n . The second e x p o n e n t i a l t e r m (which i s t h e
3.9. The Fourier Integral Singular Perturbations c o n t r i b u t i o n from w when h
+
2
295
(<;a)) c o n v e r g e s t o z e r o i n t h e d i s t r i b u t i o n s e n s e ,
0.
3.9.
The F o u r i e r I n t e g r a l S i n g u l a r P e r t u r b a t i o n s n. L e t U . E IR , j = 1,2. 3
'
Definition 3.9.1.
The f u n c t i o n $ ( x , y , h , S ) , ( x , y , h , S ) E u1 x U2 x IR x I R ~i s s a i d t o be a phase f u n c t i o n i f t h e f o l l o w i n g c o n d i t i o n s a r e s a t i s f i e d : l o $ i s reaZ valued, I $ ( X , Y , E - ' , < ) E S : ~ 6 ° " ' 2"
f o r each compact s e t K c u1
151 Let
(U1 x U2)
there e x i s t s a constant c
K
l o , 2" t h e f u n c t i o n ei'
A s a consequence o f
when
x U2
such t h a t
is rapidlyoscillating
-t m .
x
E
m
C
0
(IRn)
x
and
E 1 for
151
5 R w i t h some R <
m.
D e f i n i t i o n 3.9.2.
L e t $ ( x , y , X , S ) be a phase f u n c t i o n and l e t a ( x , y , e , < ) E S;,,(U, an amplitude (see D e f i n i t i o n 3 . 7 . 1 ) . The operator
x
u 2 ) be
i s c a l l e d t h e Fourier i n t e g r a l s i n g u l a r p e r t u r b a t i o n (F.I.S.P.) provided t h a t t h e r i g h t hand s i d e i n ( 3 . 9 . 2 ) i s p o i n t w i s e convergent. I f a ( x , y , ~ , S )E
su(U1
X
( s e e D e f i n i t i o n 3.3.2) t h e n , as u s u a l ,
U2)
a O ( x , y , ~ , S )s t a n d s f o r t h e c o r r e s p o n d i n g p r i n c i p a l symbol. Theorem 3 . 9 . 3 .
Let AE be a FISP. Then @
@
: cm(U )
0
2
+
Cm(ul)
f o r each given
E
> 0.
P r o_ o f . W e s h a l l p r o v e somewhat more. Namely, t h a t f o r e a c h E > 0 and e a c h _ p a i r o f i n t e g e r s K > 0 , 1 > 0 s u c h t h a t u +u + n + l < K o n e h a s :
A :
:
c1 0 ( u 2 )-t cK ( u l ) .
integral
2 3 L e t f i r s t K = 0 . Then u +u + n i l < 0 , so t h a t t h e 2 3
3. Singular Perturbations on Smooth Manifolds without Boundary
296
i s a b s o l u t e l y (and u n i f o r m l y w i t h r e s p e c t to x E U 1 ) c o n v e r g e n t , a n d , moreover, one c a n d i f f e r e n t i a t e it 1 t i m e s u n d e r t h e i n t e g r a l s i g n , t h e
c o r r e s p o n d i n g i n t e g r a l s b e i n g a b s o l u t e l y and u n i f o r m l y i n x E U 1 converg e n t , t o o . That proves t h e c l a i m f o r K = 0. W e u s e t h e i n d u c t i o n f o r p r o v i n g i t f o r any i n t e g e r K
0. L e t K = 1 .
Denote
so t h a t a f t e r one p a r t i a l i n t e g r a t i o n i n t h e r i g h t hand s i d e of
(3.9.2),
one g e t s : dy dS =
where L
* is b
t h e ( f o r r a l l y ) a d j o i n t o p e r a t o r of
consequence o f belong t o
sV - ( O , 1t o
Lb,
and where, a s a
( 3 . 9 . 1 ) , t h e a m p l i t u d e s of t h e o p e r a t o r s AE 0 5 j 5 n, $,I' 1, O ) (U1 x U 2 ) . 3
Example 3 . 9 . 4 . Consider t h e following F o u r i e r I n t e g r a l S i n g u l a r P e r t u r b a t i o n : (3.9.61
(E:(t)u) ( x )
=
(2i)-'((E~,+[t)u(x)-(EE (t)u)( X I ) , n,-
where
One c h e c k s e a s i l y t h a t v ( E , t , x )
=
( E z ( t ) u ) ( x ) i s t h e s o l u t i o n of t h e
f o l l o w i n g s i n g u l a r l y p e r t u r b e d Cauchy problem: (E
2
(3.9.8) lim
t++o
a 2 / a t L.- €
2
A + i ) v ( E , t , x=) 0, t > 0 , x n
a -v t++o at
v ( E , ~ , x ) = 0, l i m
E R ~ ,
( ~ , t , x )= 6 ( x ) .
One h a s t h e f o l l o w i n g formula f o r t h e d i s t r i b u t i o n a l k e r n e l s E n ( € , t ; x - y ) of t h e f a m i l y o f o p e r a t o r s
E + EE(t):
3.9. The Fourier Integral Singular Perturbations
(3.9.9)
E (E,t;x) = F -1
E<Ec>-l sin(t
X'S
Introducing
to
E R , 15,/
= E
-1
-
, 5
=
297
E -1 < E s > ) .
(Co,5) E R
n+l
-
, x
=
(xo,x) E Rn+',
one can rewrite (3.9.9) as follows:
Let n = 2m+2, where rn 2 0 is integer. One has in that case (see [ G - Sh, I ] ) : (3.9. where C12m+2
6 /:(I
is the area of the unit sphere in R
t) is the &-function on the sphere
(3.9.12)
(6(l;l-t),$)
=
t
t
=
{y 1
and the distribution
x E iR2m+3, 1y1
= t}, i.e.
.
J $(:)do_, S
S
2m+3
V Ji E C m ( ~ R 2 m + 3 ) X
Using (3.9.10), (3.9.11), one finds:
Furthermore, one has
where, as usual, s
=
max {s,O}.
Hence, (3.9.131, (3.9.14) yields: -1
(3.9.15) E 2 m + 2 ( ~ , t ;=~) and for m = 0 one finds:
Therefore the FISP (3.9.6), (3.9.7) with n equivalent representation:
=
2m+2 admits the following
3 . Singular Perturbations on Smooth Manifolds without Boundary
298 and f o r n
2 m + l one c a n u s e t h e c l a s s i c a l d e s c e n t method f o r g e t t i n g a
=
s i m i l a r r e p r e s e n t a t i o n f o r ( E : (~ t )~u )~( x ) ( s e e , [Cour.,
1
I.
One f i n d s t h e f o l l o w i n g f o r m u l a f o r t h e c o r r e s p o n d i n g k e r n e l d i s t r i b u t i o n :
where J ( p ) i s t h e B e s s e l f u n c t i o n of o r d e r z e r o and H ( T ) , T C R , i s 0 H e a v i s i d e ' s f u n c t i o n (see, f o r i n s t a n c e , [Grad.-Ryz., 11 f o r t h e B e s s e l f u n c t i o n s . N
Indeed, l e t a g a i n 5 =
(5,,5) E R
2m+2
-
, x
= (xO,x) C R
Using t h e d e s c e n t method, one f i n d s f o r
t h e following formula
Again a p p l y i n g ( 3 . 9 . 1 0 ) , one g e t s :
since
1
1
e-ipe(1-82)-4d8
=
71
Jo(p),
V p 6 IR
-1 (see [ G r a d . - R y z . ,
1
1).
I n p a r t i c u l a r , one h a s :
L e t B b e symmetric p o s i t i v e d e f i n i t e nxn m a t r i x . The s o l u t i o n of t h e s i n g u l a r l y p e r t u r b e d Cauchy problem: (E
2 2
at-E
lim
two
2
v = 0,
a
x' x
0,
>+l)V
lim
t++O
a tv
=
t > 0,
u(x)
X
6
mn
2m+2
.
3.9. The Fourier Integral Singular Perturbations
299
c a n b e w r i t t e n a g a i n a s a sum o f S i n g u l a r F o u r i e r I n t e g r a l O p e r a t o r s :
and one c h e c k s e a s i l y u s i n g t h e s u b s t i t u t i o n y = B-*x,
that the
d i s t r i b u t i o n a l k e r n e l E n , B ( ~ , t ; x - y ) of EE ( t ) f o r n = 2m+2 i s g i v e n by n,B t h e formula
The cone V = { ( p , 5 ) E R
n+l
,
*
=
u2}
i s n a t u r a l l y connected with t h e
d i f f e r e n t i a l o p e r a t o r above. L e t V b e t h e d u a l cone of V , n +1 2 V* = { ( t , x ) E R .
2m+2,B whose boundary i s
c o i n c i d e s with t h e set
*
v
Dv*
=
( t , x ) E R n + ' ,
.
We a r e g o i n g t o r e s t r i c t s l i g h t l y t h e c l a s s of p h a s e f u n c t i o n s and s h a l l c o n s i d e r , from now on t h e p h a s e f u n c t i o n s $ ( x , y , A . S ) which s a t i s f y t h e following Condition 3.9.5.
The phase f u n c t i o n s @ ( x , y , X , E ) are p o s i t i v e l y homogeneous f u n c t i o n s of degree 1 i n v a r i a b l e s ( h . 5 ) f o r h
2
2
2 f.
For such p h a s e f u n c t i o n s ( 3 . 9 . 1 ) can b e s t a t e d i n t h e form
O f c o u r s e , i n t h i s case @ ( X , Y , E
-1
,<)
€ .S(lfo'l)(Ul
x
U2).
On t h e o t h e r hand, w e s h a l l e x t e n d t h e cl.ass of t h e o p e r a t o r s c o n s i d e r e d i n t h e f o l l o w i n g way. We s h a l l d e n o t e by K t h e c l a s s of i n t e g r a l o p e r a t o r s (3.9.18)
(Ku) ( E , x )
=
1
K(E,x,y)u(y)dy
Rn such t h a t ~ ( E , x , y )E C ~ ( U x u2) u n i f o r m l y w i t h r e s p e c t t o E E ( o , E o ] 1 W e s h a l l use t h e n o t ati o n ( 4 ) f o r t h e f a m i l i e s of c o n t i n u o u s _1,0 l i n e a r mappings A C m ( U 1 + C ( U 1 ) , V E E (O,E 1, which c a n be I$: 0 2 0 r e p r e s e n t e d i n t h e form ( 3 . 9 . 2 ) w i t h t h e phase f u n c t i o n $ s a t i s f y i n g
s"
2
1,
3 . Singular Perturbations on Smooth Manifolds without Boundary
300
Condition 3.9.5 and with an amplitude a E Sv ( u x u), a being defined up to 1,o an element i n K . We shall need later on the following Corollary 3.9.6. Let A
A
$,I
a2
'j
@
E S y , o ( $ ) . Then A
$
can be represented in the form (3.9.5) with
C SV-e2($) where e 2 = ( 0 , 1 , 0 ) . Furthermore, the amplitudes ia a and 1,O 61 4, with a = a/aC., define the same singular Fourier Integral Operator 3
'j
according to the formula (3.9.2). As a consequence of Schwartz's Kernel Theorem (see [Sch, l]), one
can associate with A -V E
'A(E,x,Y) E D ' ( U ,
(3.9.19)
: Cm(U )
0 X
+ Cm(U
0 2 U 1 for each 2
E
1
a family of distributions E ( 0 . ~ ~such 1 that
e " l ( ~ u ) (E,x) = , @
v
u E c;(u2).
We are going to localize the singularities of the kernel distribution A(E,x,Y) associated with E"'A
4-
With @(x,y,h,E) positively homogeneous of degree 1 in ( h , c ) for h
2
+ ] < I2
t
phase function, satisfying (3.9.17), introduce the set
and denote
Theorem 3.9.7.
The singular support of the family E + A(E,X,YI of distributional kernels defined by (3.9.19) is contained in the set Q defined by (3.9.201, 4 (3.9.21): (3.9.22)
5 Q+. sing supp (A,~,x,y)
Proof. As
a consequence of Corollary 3.9.6, it suffices to consider the case,
when v + n < 0 , v +u +n < 0. Then E + A(~,x,y)is a family of continuous 2 2 3 in (x,y) E u l x u2 functions uniformly with respect to E E ( O , E I . More0 over, one has for A(~,x,y)the formula: (3.9.23)
A(~,x,y)=
I mn
E'
-1 a(x,y,~,<)e~'(~'~'~ ")d<.
1
3.9. The Fourier Integral Singular Perturbations W e show t h a t u n i f o r m l y w i t h r e s p e c t t o
E
A(€,.,.)
u C C (U1
Indeed, l e t
V (x,y) E
a.
E
30 1
one h a s :
C ( U 1x U 2 ) \ Q $ ) .
X
X 2 + / E l 2 2 1 one h a s V r$ # 0 , 5
Then f o r
U2)\Q4.
Using t h e o p e r a t o r ( 3 . 8 . 2 2 ) w i t h $ i n s t e a d o f g , one g e t s
t h e following representation f o r A ( ~ , x , y ) : A(E,x,~) = Jn
E"
1a l ( x , y , t , S ) e i r $ ( X ' Y ' E
-1
")dS,
V ( x , y ) Ea
R with
E
so t h a t , i n f a c t , A ( E , . , . )
C
1
(u)u n i f o r m l y
with respect t o
E
E ( 0 , ~1. 0
Using r e p e a t e d l y t h e o p e r a t o r ( 3 . 8 . 2 2 ) a n d t h e p a r t i a l i n t e g r a t i o n , one g e t s t h e c o n c l u s i o n t h a t A ( € , . , . ) to
E C
(u)u n i f o r m l y
with r e s p e c t
8
E (O,Eo1.
E
m
Example 3 . 9 . 8 . Let n
=
n2 = n , U1 = U2 and $ ( x , y , A , S ) = < x - y , F > , so t h a t t h e c o r r e s -
ponding S i n g u l a r F o u r i e r I n t e g r a l O p e r a t o r i s j u s t a s i n g u l a r p e r t u r -
sv ( U ) i n t r o d u c e d i n D e f i n i t i o n 3 . 3 . 1 . 1,o One f i n d s e a s i l y i n t h i s case
b a t i o n of t h e c l a s s
A s a consequence, one g e t s a g a i n t h e p s e u d o l o c a l i t y p r o p e r t y o f
singular perturbations: s i n g supp(E"la(x,E,D)u)
5
s i n g supp u .
Example 3 . 9 . 9 . Let n
=
(3.9.24)
n2
=
n + l , U1 = U2
$(x,y,t,X,E)
=
=
{(x,t) E R
n+ 1
},
and l e t
< x - y , C > + ( t - T ) ( A 2+
where B i s symmetric p o s i t i v e d e f i n i t e nxn m a t r i x . One h a s :
Q (X,C) r$
=
{(X,t,y,T) E U
X
2 U, X-y+(t-T) ( A +
Hence i n t r o d u c i n g rl = B t 5 , one f i n d s e a s i l y :
3. Singular Perturbations on Smooth Manifolds without Boundary
302
The phase f u n c t i o n ( 3 . 9 . 2 4 ) a p p e a r s w h i l e s o l v i n g t h e Cauchy problem f o r t h e s i n g u l a r l y p e r t u r b e d wave o p e r a t o r : 2 2 2 a t - € CBa X '
ax>+]
(see Example 3 . 9 . 4 ) , so t h a t t h e s i n g u l a r i t i e s o f t h e c o r r e s p o n d i n g
fundamental s o l u t i o n
a r e c o n t a i n e d i n Q $ . A c t u a l l y , f o r n = 2mc3, t h e
s i n g u l a r s u p p o r t of t h e fundamental s o l u t i o n i s e x a c t l y t h e whole Q$, a s Example 3.9.4
shows.
Now w e s h a l l i n t r o d u c e t h e c l a s s of
hyperbolic singular perturbations
and i n v e s t i g a t e t h e s i n g u l a r i t i e s of t h e k e r n e l d i s t r i b u t i o n s a s s o c i a t e d w i t h t h e s i n g u l a r F o u r i e r i n t e g r a l o p e r a t o r s which s o l v e t h e Cauchy problem f o r h y p e r b o l i c s i n g u l a r p e r t u r b a t i o n s . D e f i n i t i o n 3.9.10.
With a ( x , t , E , c , < ) E S v ( ~ y + l,) R:+'=
R
~
x
XR
t,+'
O p a i s s a i d t o be a
s t r i c t l y hyperboZic s i n g u l a r p e r t u r b a t i o n i f v 2 2 0 , v3 2 0 a r e i n t e g e r , a ( x , t , c , S , < ) i s polynomial i n 5
of degree v 2 + v 3 and t h e principa2 symbol
a ( x , t , ~ , S , < )s a t i s f i e s t h e c o n d i t i o n : 0 The zeros < . ( x , t , ~ , S ) ,1 5 j 6 v2+v3,0f t h e equation 1
(3.9.25)
a0(x,t,E,S, 5 . )
I
=
0
n+ 1
are r e a l and d i s t i n c t , v ( x , t ) E R +
,v
(E,c)
E R + x ( B ~ l\ o } ) .
H y p e r b o l i c s i n g u l a r p e r t u r b a t i o n s whose symbols a r e polynomial i n ( E , t ) , a r e o f s p e c i a l i n t e r e s t . W e s h a l l c o n s i d e r t h e Cauchy problem f o r t h e c o r r e s p o n d i n g h y p e r b o l i c d i f f e r e n t i a l s i n g u l a r p e r t u r b a t i o n s whose symbols have c o n s t a n t c o e f f i c i e n t s and c o i n c i d e w i t h t h e i r p r i n c i p a l symbols. L e t ao(t,S,<)
E P
b e s t r i c t l y h y p e r b o l i c . Without r e s t r i c t i o n of
g e n e r a l i t y , one c a n assume t h a t v 1 = 0 . I f v 3 = 0 , t h e n a. on
E
d o e s n o t depend
and becomes j u s t t h e c l a s s i c a l s t r i c t l y h y p e r b o l i c polynomial w i t h
3.9. The Fourier lntegral Singular Perturbations
303
c o n s t a n t c o e f f i c i e n t s ( s e e [ P e t r . 1 1 ) . T h e r e f o r e , w e s h a l l assume t h a t
v3 > 0 , v2 2 0. C o n s i d e r t h e Cauchy problem:
n+l
aO(E,Dx,Dt)u = 0 ,
( x , t ) E R+
(3.9.26)
where 6 . = 0 , 1 5 j < m, bm = 1 i s t h e Kronecker symbol and v E C m ( R n ) 1m 0 Let <.(E,<), 1 5 j 5 m = v 2 + v 3 be t h e z e r o s of t h e e q u a t i o n
.
1
(3.9.27)
aO(E,S,<.(E,S)) = 0. 3
homogeneous f u n c t i o n s o f ( E -1 , S ) o f d e g r e e 1 . 0 F u r t h e r m o r e , d e n o t i n g by < . ( c ) , 1 6 j 6 v2 t h e z e r o s of t h e r e d u c e d
Obviously,
<.(E,S) a r e 1
3
equation: (3.9.28)
0 <,(c)
0
0
ao(S,<.(S)) 3
=
0,
b e i n g homogeneous i n
6 o f d e g r e e 1 , one h a s t h e a s y m p t o t i c f o r m u l a e :
J
(3.9.29)
< . ( E , S ) = <0. ( E ) + O ( E 2~ )S, ~ f o r 3
3
and f o l l o w i n g i n e q u a l i t i e s f o r C , ( E , ~ ) 3
(3.9.30)
l?j(~,c)I
1 5 j 5 v2
5 C/SI,
where C > 0 i s some c o n s t a n t .
<,(E,E)
For t h e o t h e r v3 z e r o s
3
of
(3.9.27) t h e following i n e q u a l i t i e s
hold (3.9.31)
C-lE-I
5
IC.(E,S)
3
I
5 C(E
-2+1~/2)t,
v2
<
j 5 v 3'
where a g a i n C i s some c o n s t a n t . The s o l u t i o n of (3.9.32) where
u ( E , x , ~ )=
( 3 . 9 . 2 6 ) c a n b e w r i t t e n a s f o l l o w s ( s e e [Gel-Sh,
I
I
R
~
C
15jsv R ~ +v 2 3
a.
11):
3. Singular Perturbations on Smooth Manifolds without Boundary
304
One h a s f o r e a c h g i v e n t > 0
I: (A v ) ( E , x , ~ ) , l s j s v +v 'j 2 3 i s t h e S i n g u l a r F o u r i e r I n t e g r a l O p e r a t o r w i t h t h e p h a s e $ . and where A I' 1 t h e amplitude a , , i n t r o d u c e d above. 1 One f i n d s : =
U(E,X,t)
Q
(h,E;t)
E Rn
= { (x,y)
x Rn
,
(x-y)t
-1
=
}.
-vn<j(l,'l)
$3
rl=A-l 5 : Rn\{O}
L e t K , be t h e c l o s u r e o f t h e r a n g e of V 5 . ( 1 , n ) I '13 K = u K Denote j' (x,y,t)
K* =
Rn
x
Rn
x
R+
,
(x-y)/t
E
-f
nn
and l e t
K}.
so t h a t K*
U
=
(t).
Q
t>O,j
'j
A s a consequence of Theorem 3 . 9 . 7 t h e s i n g u l a r s u p p o r t of t h e k e r n e l
d i s t r i b u t i o n a s s o c i a t e d w i t h t h e s o l u t i o n u ( E , x , ~ )o f
(3.9.26),
is
For c l a s s i c a l h y p e r b o l i c o p e r a t o r s ( v 3 = O ) , t h e s e t K*
c o n t a i n e d i n K*.
i s n o t h i n g e l s e b u t t h e d u a l c o n e of r a y s f o r t h e c h a r a c t e r i s t i c cone of
(see f o r i n s t a n c e [Cour I]).
t h e o p e r a t o r a,(E)
I n t h i s c a s e dim K* = 2n,
i
w h i l e i n t h e g e n e r a l case of h y p e r b o l i c s i n g u l a r p e r t u r b a t i o n s onemay h a v e : dim K
*
=
2 n + l , a s it was t h e case i n Example 3 . 9 . 4 .
Using ( 3 . 9 . 3 2 ) one g e t s a f t e r t h e change o f v a r i a b l e s E C
+
5 the
f o l l o w i n g formula f o r t h e d i s t r i b u t i o n a l k e r n e l o f t h e o p e r a t o r , t h a t s o l v e s t h e Cauchy problem ( 3 . 9 . 2 6 ) (3.9.341
:
-1 im-1(2n)-nE1-n/n c a . (l,c)eiE ' j ' t ~ x ~ y ~ l ~ ' ) d ~ , R 1 c j s v +v 2 3 a r e g i v e n by ( 3 . 9 . 3 3 ) .
E(E,x-y,t)
=
where a . and 4 . 7 7 Let E . ( ( x - y ) / t ) be a s t a t i o n a r y p o i n t f o r 1 (3.9.35)
V5cj(l,S.) 1
=
(x-y)/t,
0.
1' l - e .
1 5 j 5 m.
Definition 3.9.11.
The p o i n t ( x , y , t ) E
K*
i s s a i d t o be non-focaZ i f t h e corresponding
s t a t i o n a r y p o i n t s 5 . ( ( x - y ) / t ) d e f i n e d by ( 3 . 9 . 3 5 ) are r e g u l a r , i . e . t h e 5
3.9. The Fourier Integral Singular Perturbations
matrices
D
2
305
5 ( 1 , c . ) are non-singular.
5 ,
I
Applying t h e s t a t i o n a r y p h a s e method t o t h e i n t e g r a l ( 3 . 9 . 3 4 ) a t e a c h n o n - f o c a l p o i n t ( x , y , t ) a n d assuming a 6 l i t i o n a l l y t h a t t h e c o r r e s ponding s t a t i o n a r y p o i n t s S . ( ( x - y ) / t ) a r e w e l l - d e f i n e d by ( 3 . 9 . 3 5 ) , one
I
g e t s t h e following assymptotic formula f o r E ( E , x - Y , ~ ) a t each such p o i n t (x,y,t): (3.9.36)
E(~,x-y,t)
N
v
+I,
-1 (2Tr)
i
-n/2
t
-1/2
E
1-n/2
~
q . ( ( x - y ) / t )e 1 6 j 6 v +v 7 2 3
itE-lAj
((x-y)/t)
A . ( 2 ), z E R n , i s t h e Legendre t r a n s f o r m of t h e f u n c t i o n 5 , ( 1 , c ) , 5 E 3 1 i . e . A.(z) and < . ( l , E ) a r e r e l a t e d by t h e Lagrange t r a n s f o r m a t i o n 1 3
where
(3.9.37)
A.(z) = < Z , S > + < . ( l , S ) , 3
1
v65 3 (1,5)
mn,
= -2,
and where
2
X . ( 5 . ) b e i n g t h e s i g n a t u r e o f t h e m a t r i x D E E < , ( 1, 5 , ) . I
1 ( 3 . 9 . 3 6 ) , E ( ~ , x - y , t )a t e a c h n o n - f o c a l p o i n t i s
3
A s a consequence o f
r a p i d l y o s c i l l a t i n g when For e a c h ( x , y , t ) multi-indice
c1
K
*
E +
0.
one f i n d s , a p p l y i n g P r o p o s i t i o n 3 . 8 . 4 ,
t h a t f o r each
t h e following asymptotic formula h o l d s :
2 2 For t h e wave o p e r a t o r a ( E . S . ~ ) = 5 -151 one h a s c1 = 151, c 2 = -1c1, 0 $ l ( t , x , y , l , S ) = < x - y , C > + t / < I , $ 2 ( t , x , y , 1 , 5 ) = < x - y , C > - t / E and t h e
s t a t i o n a r y p o i n t s a r e n o t w e l l - d e f i n e d by ( 3 . 9 . 3 5 ) so t h a t ( 3 . 9 . 3 6 ) c a n not hold. Notice t h a t i n t h i s case, i n f a c t , $
E
s(orlto)(UxU).
For t h e s i n g u l a r l y p e r t u r b e d wave o p e r a t o r w i t h t h e symbol ao(E,S,S)
c2
= E 2 5 2 -E 2 [ 5 1 2 + 1 ( s e e Example 3 . 9 . 4 )
= - E - ~ < E ~ >and
<
=
€-kc>,
t h e c o r r e s p o n d i n g s t a t i o n a r y p o i n t s a r e w e l l - d e f i n e d by
( 3 . 9 . 3 5 ) p r o v i d e d t h a t Ix-yl
and
one h a s :
< t . One f i n d s e a s i l y i n t h i s c a s e
3 . Singular Perturbations on Smooth Manifolds without Boundary
306
2 2 -2 I D s S 5 j ( l , c3. ) I = ( l - l x - y ] t
Hence, one has K* (x,y,t)
E
K
*
=
{ ( x , y , t ) , Ix-yI
l+n/2
< t} i n t h i s c a s e , e a c h p o i n t
i s n o n - f o c a l and t h e f o r m u l a e ( 3 . 9 . 3 6 ) - ( 3 . 9 . 3 8 ) a r e v a l i d
f o r t h e d i s t r i b u t i o n a l k e r n e l E ( ~ , x - y , t )of t h e corresponding o p e r a t o r s o l v i n g t h e Cauchy problem ( 3 . 9 . 2 6 ) f o r t h e S i n g u l a r l y P e r t u r b e d Wave 2 2 2 O p e r a t o r E at-€ A + l ( s e e ( 3 . 8 . 1 0 0 ) where t h e f i r s t t e r m i n t h e a s y m p t o t i c e x p a n s i o n f o r E ( E , x , t ) i s e x p l i c i t e l y w r i t t e n down). N o t i c e t h a t i n t h e c a s e of t h e s i n g u l a r l y p e r t u r b e d wave o p e r a t o r
a t- & 2 A x +1 -
2 2
t h e c o r r e s p o n d i n g p h a s e f u n c t i o n s a r e $ ( t , x , y , ~ , < )= 1 < x - y , S > + t ~' < E S > , G 2 ( t , x , y , & , S ) = < x - y , S > - - t ~ - ~ < & and c > $ j € S ( l ' o r l ) ( U x U) E
but
$jtr
S (o.l,o)
(UX
U).
Hence, it i s n a t u r a l t o i n t r o d u c e t h e f o l l o w i n g D e f i n i t i o n 3.9.12.
A Fourier I n t e g r a l S i n g u l a r Perturbation A'
$
i s s a i d t o be a Proper Fourier
I n t e g r a l S i n g u l a r Perturbation (PFISP) i f i n ( 3 . 9 . 2 ) t h e corresponding phase -1
function $ ( x , Y , E , t ) s a t i s f i e s t h e conditions: 1'
$
b u t $(X,Y,E 2"
i s real-valued, $ ( x , Y , E -1 , c ) E S ( l ' o ' l ) ( u l -1
, S ) ?! S(o,l,o)
(x,y,h,t!
variabzes ( A , 5 ) f o r
(U1
x
x
u2),
U2).
i s p o s i t i v e l y homogeneous o f degree 1 w i t h r e s p e c t t o
1 A I 2+ 15 I
2
+ and, moreover, $ s a t i s f i e s
vx,x,S$ f 0 , vy,h,<$ # 0 ,
v
(x,y) €
u1
x
u2, v
the condition
( h , t ) E IRn+l,
2 h + ( < I 2 = 1.
I f $ s a t i s f i e s l o , 2" t h e n i t i s c a l l e d a proper phase f u n c t i o n s
family. Remark 3.9.13. I f $ ( x , y , A , t ) i s a p r o p e r phase f u n c t i o n t h e n t h e f o l l o w i n g i n e q u a l i t i e s
3.10. Diffeornorphisms and Singular Perturbations
v
( x , y , C ) E U1
U2
X
x
R n and when 5 +
The F I S P ' s i n Example 3.9.4 D e f i n i t i o n 3.9.14.
(Vl.0,V3)
m,
V (x,y,X) E U 1 x U
307
2
x IRR\{O}.
are p r o p e r . n+l
with a ( x , t , E , c , c ) E S (R+ 1 , R y t l = nn x R ~ , + , op a is s a i d t o be a Proper S t r i c t l y Hyperbolic S i n g u l a r P e r t u r b a t i o n i f v3 > 0 is i n t e g e r a ( x , t , E , S , c ) i s polynomial i n 5 of degree v3 and t h e p r i n c i p a l symbol a o ( x , t , ~ , S , c )s a t i s f i e s t h e c o n d i t i o n :
The z e r o s 5 . ( x , t , c ) , 1 5 j 5 v3, of t h e equation: 7
ao(x,t,l,S,c)
=
0
a r e r e a l and d i s t i n c t , V ( x , t , < ) E
I < ,3 ( x . t . 6 ) I
b 6 > 0,
I R ~x IR
x
R;, and, moreover,
t.+ 1 5 j 5 v3, V ( x , t , c ) , where 6 i s some c o n s t a n t .
Remark 3.9.15. L e t a ( E , D ,Dt) b e a Proper S t r i c t l y Hyperbolic S i n g u l a r P e r t u r b a t i o n o x whose symbol a O ( E , E , L ) E S ( 0 ' 0 ' v 3 ) (R:+l ) i s polynomial i n (5,C.) and -1 homogeneous i n ( E ,c,c) of d e g r e e 0 . L e t E ( ~ , x - y , t ) b e t h e k e r n e l d i s t r i b u t i o n of t h e o p e r a t o r E ( t ) t h a t s o l v e s t h e c o r r e s p o n d i n g Cauchy Problem ( 3 . 9 . 2 6 ) ( w i t h v2 = 0 ) . * n c R X R n x I7 b e d e f i n e d , a s p r e v i o u s l y i n t h e c a s e v 2 b 0 , so x Y t,+ * t h a t s i n g supp E ( ~ , x - y , t ) E K Then f o r P r o p e r H y p e r b o l i c S i n g u l a r
Let K
.
P e r t u r b a t i o n s w i t h polynomial symbols a ( E , E , < ) always h a s dim K formula ( 3 . 9 . 3 4 )
0
*
= 2n+l
( w i t h v2
E S ( 0 f 0 ' v 3 ) ( m y + ' ) one
(see a l s o Example 3 . 9 . 4 ) . =
In t h i s case i n t h e
0 ) f o r E ( ~ , x - y , t ) a l l $ j , 1 5 j Iv 3 , a r e
p r o p e r phase f u n c t i o n s a n d , i t c a n b e shown t h a t t h e measure of t h e s e t of a l l f o c a l p o i n t s i n K
i s z e r o , so t h a t t h e s t a t i o n a r y p h a s e method c a n
be a p p l i e d , t h e a s y m p t o t i c f o r m u l a ( 3 . 9 . 3 6 ) b e i n g v a l i d a . e . i n K
*
.
3 . 1 0 . Diffeomorphisms of S i n g u l a r P e r t L r b a t i o n s on m a n i f o l d s W e s t a r t w i t h some a s y m p t o t i c f o r m u l a f o r S i n g u l a r P e r t u r b a t i o n s
which p l a y s t h e r o l e comparable w i t h L e i b n i t z ' f o r m u l a f o r D i f f e r e n t i a l Operators. Theorem 3 . 1 0 . 1 .
Let
f
E c;(u), g E C m ( U ) , g being real-vaZued and dg ( x ) # 0 , V x E supp f .
L e t a ( x , E , ~ )E o p s;,,(u), N 2 o one has:
a ( x , E , ~ ):
m
c o ( u ) + c m ( u ) . Then f o r each i n t e g e r
3. Singular Perturbations on Smooth Manifolds without Bounda y
308
where h (x,Y) g
= g ( x ) - g ( y l -
and where t h e remainder s a t i s f i e s on any compact s e t K c
c
depending onZy on
N,K
(3.10.2)
N
u (with a constant
and KJ the foZZowing e s t i m a t e
/ R N ( x , ~ , p )5 / C
N.K
E
-U1
p
-N+[N/2]+V2
<EP>
v3
.
Proof. Without r e s t r i c t i o n of g e n e r a l i t y , one can assume t h a t u z + v 3 < -n,
as t h e
c a s e v2+u3 2 -n can b e reduced t o t h i s one u s i n g t h e i n t e g r a t i o n by p a r t . Denote
With v2+u3 < -n,
Q ( x , ~ , p )c a n b e w r i t t e n a s a d o u b l e i n t e g r a l and a f t e r
t h e change o f v a r i a b l e s p 5 (3.10.4)
+
5 one g e t s f o r
1 1
= (2.ir)-"pn
Q(x,E,P)
R
Q(x,E,P)
t h e f o l l o w i n g formula:
q ( x , y , ~ , p , 6 ) e ~ WdS, ~ ~ ( ~ ' ~ ~ ~ )
U
where
The p o i n t (3.10.6)
M(x) = ( x , V g ( x ) ) E lRn
Y
X
lRn
5
i s t h e o n l y s t a t i o n a r y p o i n t o f t h e phase f u n c t i o n il,
: R n x IRn y 6 + R . W e s h a l l u s e a p a r t i t i o n of u n i t y i n 6 - v a r i a b l e and a f t e r w a r d s a p p l y
Theorem 3 . 8 . 1 0 i n o r d e r t o f i n d a n a s y m p t o t i c e x p a n s i o n f o r Q ( x , E , P ) by ( 3 . 1 0 . 4 ) .
( 3 . 1 0 . 5 ) when p
+
F i r s t w e show t h a t t h e c o n t r i b u t i o n i n t o Q ( x , E , P ) o v e r t h e areas for p
-f
-,
{[c/
< 6, y
given
-.
E U} and
> R, y
from t h e i n t e g r a t i o n
E Ul i s of o r d e r
-m
O(p
)
p r o v i d e d t h a t 6 i s s u f f i c i e n t l y s m a l l and R i s s u f f i c i e n t l y
l a r g e . S i n c e dg # 0 on supp f , one c a n choose 6 ' > 0 , R ' > 0 , such t h a t 6 ' 5 IVg(y)l 5 R ' ,
V y C supp f . L e t 6 < 6 ' .
x 1 ( 5 ) : 0 f o r 151 > 6 , x , ( S )
!
1 for
and l e t x , ( S )
151 < 6 / 2 . Denote
E C;(lRn)
,
3.10. Diffeomorphisms and Singular Perturbations
Q 1 ( x , ~ , p )=
J 1 mn u
309
qx,eip'dydS.
L e t K C U b e any g i v e n compact. W e s h a l l show t h a t f o r any m u l t i i n d i c e s
a,B and any i n t e g e r N 2 0 h o l d s :
Denote
Given t h e c h o i c e o f t h e c u t - o f f
x l ( S ) , one o b v i o u s l y h a s :
function
IVg(y)-SI 2 c o > 0 , V ( x , y , S ) E K
X
X
supp
xl.
Therefore, the
c o e f f i c i e n t s o f t h e o p e r a t o r L ( y , C , a ) a r e smooth and bounded on Y U X supp x . F u r t h e r m o r e , s i n c e a ( x , E , S ) E sv (U) one o b v i o u s l y h a s f o r 1 1,o any a : -V
(3.10.9)
lD;q(x,y,E,P,S)\
5 C
a,K
E
l
V2
<EPS>
v3
,
V ( x , y , ~ , p , S )E K x
X
(O,E
0
1
x
l R + x lRn,
depends o n l y on a and K .
where t h e c o n s t a n t C U,K
Using t h e f o r m u l a
L ( Y , s , a y ) ei P J l
= ipeiP',
and d e n o t i n g by Lt(y,S,a t
L (y,c,a
Y
)
=
Y
t h e f o r m a l a d j o i n t o p e r a t o r of L ,
-
~(Y,s)>,
one g e t s , a f t e r one i n t e g r a t i o n by p a r t f o r Q ( x , E , P )
1
AS a consequence of
estimate :
t h e following formula:
( 3 . 1 0 . 9 ) , one g e t s f o r Q l ( x , & , p ) t h e f o l l o w i n g
3 . Singular Perturbations on Smooth Manifolds without Boundary
310
A f t e r N i n t e g r a t i o n s b y p a r t u s i n g t h e o p e r a t o r L ( y , S , a ) , one f i n d s Y -W
I
IQ~(X,E,P)
=
3, IpV2-N<Ep>v
'N,K€
'd (x,E,P) E K
X
( 0 . ~ ~X 1R + , p 2 1.
D i f f e r e n t i a t i n g Q ( x , E , ~ )w i t h r e s p e c t t o x and p, one g e t s t h e same k i n d
1
of i n t e g r a l s , so t h a t ( 3 . 1 0 . 7 )
for p
-f
-,
u n i f o r m l y w i t h r e s p e c t t o x E K C C U and
Now, l e t R > R ' 5 IVyg(y)
x2
0 for
i s proved. N o t i c e , t h a t
151
5 R,
x2
Z
1,
V y 6 supp
E
f and l e t
vl
X,E,P) Q1( E [ O , E 1. E
=
O(p
-m
)
0
x,(S)
E Cm(lRn)
,
1 for 151 2 2 R , b e t h e c o r r e s p o n d i n g c u t - o f f
f u n c t i o n . Denote by Q ~ ( X , E , P )t h e f u n c t i o n
Q2(x,~,p) =
IqX2eiP'dydc, Rn U
t h e d o u b l e i n t e g r a l i n t h e r i g h t hand s i d e b e i n g c o n v e r g e n t s i n c e v2+v3
-n.
Using a g a i n t h e o p e r a t o r L ( y , S , a ) i n t r o d u c e d by ( 3 . 1 0 . 8 ) , Y Q2(x,€,p) i n t h e form:
S i n c e Vg(y)-C
# 0,
V (y,C) E
V
supp
x2
(y,S)
1 .( y , < )
1
E
X
c s(o,-l,o) l,o (U)
x
supp
x2,
one h a s i n f a c t ,
one c a n r e w r i t e
lv g(v)-cj
w i t h s o m e p o s i t i v e c o n s t a n t C , so t h a t
Y
5
c
a n d , t h e r e f o r e , Q2(x,€,p) c a n b e e s t i m a t e d , a s
follows:
-vl v2-1 I Q ~ ( x , E , P )5~ CE
P
R e i t e r a t i n g t h e p a r t i a l i n t e g r a t i o n , one g e t f o r Q 2 ( x , ~ , p ) t h e s a m e
estimate a s f o r Q ( x , E , P ) . D e r i v a t i v e s of Q2 w i t h respect t o x and p b e i n g 1 e x p r e s s e d by i n t e g r a l s o f t h e same k i n d , a s Q, i t s e l f , one g e t s f i n a l l y ,
f o r V (x,E,P) E K
x
( 0 , 1~ X R + , p t 1, and any i n t e g e r N t 0.
N o t i c e t h a t E V ' D a D B i ( x , E , P ) = O(p X P 2 r e s p e c t t o x E K C C U, and E E [ O , E 1 . 0
Let
x3(S)
=
1 - x 1 ( S ) - x 2 ( S ) and l e t
-a
)
for p
+
m,
uniformly with
3.10. Diffeomorphisms and Singular Perturbations
31 1
I n o r d e r t o b e a b l e t o a p p l y t h e s t a t i o n a r y p h a s e method t o t h e i n t e g r a l ( 3 . 1 0 . 1 1 ) one h a s t o check t h a t t h e o n l y s t a t i o n a r y p o i n t M(x) = ( x , V g ( x ) ) , M(x) E supp qx3, of t h e p h a s e f u n c t i o n $,
i s regular.
Indeed one f i n d s :
so t h a t a t t h e s t a t i o n a r y p o i n t M(x) = ( x , V g ( x ) ) , one f i n d s :
I n a neighbourhood of t h e s t a t i o n a r y p o i n t M(x) one h a s :
I n t r o d u c i n g t h e v a r i a b l e s z = y-x, 2 =
1
0
n
=
S - V g ( x ) - f g x x ( x ) ( y - X I , one f i n d s t h a t
, -Id
-Id,
0
1
so t h a t
Applying t o ( 3 . 1 0 . 1 1 ) t h e s t a t i o n a r y phase method and t a k i n g i n t o a c c o u n t t h e estimates f o r Q . ( x , E , P ) , j 1 e x p a n s i o n f o r Q ( x , E , P ) when p + m.
=
1 , 2 , one g e t s a n a s y m p t o t i c
I t r e m a i n s t o compute t h e c o e f f i c i e n t s of t h i s e x p a n s i o n . For d o i n g
t h a t , w e n o t i c e f i r s t t h a t f o r g ( x ) = < x , n > w i t h I? E lRn and f o r any i n t e g e r N > 0 one h a s :
where t h e remainder i s g i v e n by t h e formula
a given vector
3. Singular Perturbations on Smooth Manifolds without Boundary
312
m
w i t h some 8 E ( 0 , l ) and E l E C O ( U ) . ( U ), one g e t s f o r R N f t h e f o l l o w i n g and a E Sv 1 ,o
E C;(U)
S i n c e Daf e st i m a t e :
-V
(3.10.13)
IRN(x,E,p,rl,Dx)f(x)I 5 C N ( f l ) E
w i t h some c o n s t a n t C ( f l ) > 0 N S e t t i n g n = Vg(x), f l (x)
=
V
-N
V <EP>
P'
3
f ( x ) ei p h 9 ( x ' y ) and a p p l y i n g t h e f o r m u l a
(3.10.12) t o t h e i d e n t i t y e-ipg ( x ) a( x , E , D X ) ( ei W f )
~
ip
I
= e
one g e t s t h e f o r m u l a ( 3 . 1 0 . 1 ) . The e s t i m a t e ( 3 . 1 0 . 2 ) f o r t h e r e m a i n d e r R N ( x , E , p ) f o l l o w s from
I
grows a t most l i k e ( 3 . 1 0 . 1 3 ) and t h e f a c t t h a t t h e c o n s t a n t C ( f e i p h l N Y=x P "/21 f o r p + m, g i v e n t h a t t h e f u n c t i o n h ( x , y ) h a s a z e r o o f o r d e r 2 when y-x
+
9
I
0.
C o r o l l a r y 3.10.2. Taking N = 1 i n ( 3 . 1 0 . 1 ) , ( 3 . 1 0 . 2 ) , one g e t s t h e f o l l o w i n g aSymptOtiC formula : ( 3 . 1 0 . 1 4 ) e - i p g ( x )a ( x , E , D
) (f
( x ) ei p g ( x ) )
=
f ( x ) a(x,E,pvxg(x))
-
2 2 - i {
5
-v
+
O(E
1p"2-2<Ep>v
C o r o l l a r y 3.10.3. -1 . Taking p = E i n (3.10.1),
3
).
(3.10.21,
one f i n d s
iE-lg(X) ( 3 . 1 0 . 1 5 ) e-iE - l g ( x ) a ( x , E , D x ) ( f ( x ) e ) =
N-[N/2]-V
+
O(E
-V2
1
)
u n i f o r m l y w i t h r e s p e c t t o x i n any compact s e t K C C U and w i t h r e s p e c t t o
3.10. Diffeornorphisnzs and Singular Perturbations
313
m
f in any bounded set in
Co(U).
= v, In particular, if a E Kv(U) and ar = C a . ord a . = v(1), v(') j z o I' 1 is its graduate symbol, then the following asymptotic formula holds:
where, as usual a(a)= aaa, the asymptotic relation (3.10.16) being uniform
5
with respect to x I n any compact set K c c
U,
with respect to f in any
m
given bounded set in C ( U ) . 0 We are going to apply Theorem 3.10.1 in order to show how the symbols of singular perturbations are transformed under the coordinates diffeomorphisms. This will enable us to introduce singular perturbations on smooth manifolds without boundary. Theorem 3.10.1 will also be applied for giving a different proof of the formulae for the symbols of product of properly supported singular perturbations.
We shall need the following useful auxiliary result. Theorem 3.10.4. (j)
(u), j b 0 w i t h v ( 1 ) 1,o L e t ( O , E ~ I 3 E + a(x,~,c)E cm(ux
Let a E
1
E Sv
=
(j) (vl,v2 ,v3), ViJ) c
::
R~ ) be continuous w i t h r e s p e c t t o
E ( 0 . ~ ~and 1 assume t h a t t h e r e e x i s t c o n s t a n t s
c
a. 5
-a
andp
for
=
j
+
m.
u ( a , B ) such
that
f o r any x E U,
E
E (o,c01,
5 E R ~ .
Further, assume t h a t t h e r e e x i s t u ( N )
J-
-m
for
N +
and such t h a t on every
compact s u b s e t K cc u one has:
uniformly w i t h r e s p e c t t o x on any compact s u b s e t K
a U and
6
E ( 0 . ~ ~ 1
3. Singular Perturbations on Smooth Manifolds without Boundary
314 Proof. -
(01
According t o Theorem 3 . 6 . 2 ,
t h e r e e x i s t s a symbol b ( x , ~ , S )C Sv
1# O
(U) such
that
(N) (b(x,E,S) -
Z O<=j
and such t h a t ( 3 . 1 0 . 1 9 )
a . ( x , ~ , S ) C)
'
S"
1 to
(U)
h o l d s w i t h b i n s t e a d of a . m
I t r e m a i n s t o show t h a t r = ( a - b ) E sy,o(U), where
v
m
=
(v1,-m,v3). The i n e q u a l i t i e s ( 3 . 1 0 . 1 8 ) imply:
(3.10.20)
Ir(x,E,S)
I
-V =
< 'N,KE
V
l < E ; > - N < ~ S >', V x E
K C C U, V E
E (O,E~],
V N Z O . W e have t o check t h a t (3.10.20) h o l d s f o r t h e d e r i v a t i v e s B w D D r ( x , E , S ) . F o r d o i n g t h a t , o n e may u s e t h e Kolmoqorov i n e q u a l i t y
x s
z
SUP
/ a l = l xEK 1
2 IDa43(x)( 5 C s u p I43(x)( Z 1Da@(x)I, xEK2 1452
where K , a r e compact s e t s , K CC i n t K 2' 7 1 O b v i o u s l y , i n o r d e r t o show t h e v a l i d i t y of t h i s l a s t i n e q u a l i t y i t s u f f i c e s t o c o n s i d e r t h e one d i m e n s i o n a l case. Using t h e T a y l o r formula i n t h e form:
43 ' ( x )
= 'clp
26
6
( x + 6 )-43 (x-6) 1 + -(a" ( y + 6 )- $ ' I 4
and c h o o s i n g 6 = ( 2 s u p I @ ( x )I/sup
I
(y-6) )
,
I
I$"(x) ) ' ,
one g e t s t h e Kolmogorov
i n e q u a l i t y above. Indeed, a p p l y i n g t h i s l a s t i n e q u a l i t y t o t h e f u n c t i o n s @c(X,E,T))
with K
r(X,&,c+rl)
1 = K X { O } , K 2 = K 6 X Irl g e t s , using (3.10.17), (3.10.20)
/ 1111
5 61, K6 = { x
I
d i s t ( x , K ) 6 6}, one
3.10. Diffeornorphisrns and Singular Perturbations Now w e g i v e a d i f f e r e n t p r o o f of Theorem 3.7.6,
315
which i s b a s e d on t h e
s t a t i o n a r y phase method. Theorem 3.10.5.
L e t a ( x , ~ , ~: )c;(u)
+
c;(u),
a(x,~,<E )
m
b(x,E,D) :
c;(u)
+
C o ( u ) , b ( x , E , c ) E S'
1 .o
Sv
(u), and
1 PO ( U ) be p r o p e r l y supported
s i n g u l a r p e r t u r b a t i o n s . Then t h e i r product c ( x , E , D )
=
a ( x , ~ , D 0) b ( x , ~ , D ) ,
i s a properZy supported singuZar p e r t u r b a t i o n , whose symbol E Sv+' ( u ) has t h e f o l l o w i n g asymptotic expansion: 1,o
c(x,E,S)
t h e asymptotic r e l a t i o n (3.10.21) being understood in t h e f o l l o w i n g sense
u n i f o m l y w i t h r e s p e c t t o x i n any compact s u b s e t K C C to
E
E
U and w i t h r e s p e c t
(O,Eo1.
Proof. Let u
m
E C o ( U ) . S i n c e a ( x , ~ , D ) :C;(U)
Applying (3.10.1) w i t h p
=
151,
+
C E ( U ) , one h a s
g ( x ) = <x,w> w
=
i n order t o
I
i l ~<x,w>
compute t h e a s y m p t o t i c e x p a n s i o n o f a ( x , 6 , D ) ( b ( x , E , < ) e -f
m,
one g e t s t h e formulae
c ( x , E , ~ )of
)
when
(3.10.21), (3.10.20) f o r t h e symbol
the singular perturbation c = a
o
b . Now Theorem 3.10.4 i m p l i e s
E S"+'(U). I 1P O For some classes of s i n g u l a r p e r t u r b a t i o n s t h e c o m p o s i t i o n f o r m u l a C(X,E,S)
N
(3.10.21) i s v a l i d w i t h a remainder which i s of o r d e r O ( E < E S > - ~ )when 6 ' 0 ,
5'm.
D e f i n i t i o n 3.10.6.
A singular perturbation A o p J~
1 ,o
(u) i f
v
E op
= (vl,0,v3) and
( u ) i s s a i d t o b e in t h e c l a s s 1,o i t s symbol i s a cm-function of t h e form
Sv
a ( x , s , E ( ) , which s a t i s f i e s t h e i n e q u a l i t i e s
3 . Singular Perturbations on Smooth Manifolds without Boundary
316
and f o r V ( x , E , ~ ) E K
x
( 0 . ~ ~x 1iRn
where K is any compact s e t in
U.
Theorem 3.10.7.
L e t AE
=
Op a E Jv ( u ) , BE = Op b E Sy,o(u) be properly supported. Then 1r 0 C = AE o BE is a p r o p e r l y supported singuZar p e r t u r b a t i o n
t h e i r composition
i n OP Sv+' ( U l whose symbol c(x,E,~)has t h e f o l l o w i n g asymptotic expansion: 1,O
where t h e asymptotic r e l a t i o n (3.10.24) i s i n t e r p r e t e d in t h e f o l l o w i n g sense: f o r each N
= O,l,
... t h e
remainder
moreover,
Proof. ~
m
m
Cn
Let u E C (K). Since A
: CO(U)
0
-f
C
(U)
, one has
so that (3.10.1), (3.10.2) yield
+ (271)-n
where r
N
i rN(x,E,5)ei<X"> mn
u ( 5 )dS ,
satisfies the inequality IrN(X,E,S)I 5
N-(V +Ul) 1
cN , K ~
<5'
V +U
V
+!.I
v
-N
, V x E K ,
2<E5>
E E
(0,E0It v 5 E R n ,
I
as a consequence of (3.10.2), (3.10.23). Corollary 3.10.8. If a(x,~,D)E Op J"
1 to
a(x,E,D)
o
(U)
with v = (v1,0,v2),b ( x , E , D )
E
Op
b(x,E,D) = Op(a(x,E,~)b(x,E,~)) + ERE
(u) sV 1r o
then
3.20. Diffeornovhisrns and Singular Perturbations where R
E Op sl,o(U) (Y) with y = v+~1-(0,0,1), y
317
< v+u
The following result shows how the symbols of singular perturbations are transformed after a diffeomorphic change of variables.
,
lRn be bounded domains and let x : U Y diffeomorphism. Let AE = Op a, a ( x , ~ , 5 )E Sv ( u ) and let Let U
C
IR:
V
C
-
m
+
V be a C
1#O
Theorem 3.10.9.
The operator AE
m
:
Co(V)
-f
Cm(V) defined by (3.10.27) is a singular
X
perturbation whose symbol a (x,E,S) E Sv (v) has the following asymptotic X 1.0 expansion,
where (3.10.29) qa t X
and where
(x,c) = Da(,i<X(Y)-X(x)-X'(x) (y-x), 5 > Y
t x' is the transpose of
the Jacobian matrix
x i(x).
Proof. It suffices to consider properly supported singular perturbations. Let m
u E Co(V) and denote y
= x(x), v(x) = ( u
X) (x). One finds
where we have denoted (3.10.30) b (x,E,<)= e-i<X(x)rS>a(x,E,Dx)(ei<x(x)rS>).
X
Notice that Vx<x(x),S> = tx'(x)5 # 0, V (x,S) E U Hence, formula (3.10.1) with p = 151, g(x)
=
x
(lRn\{O)).
<x(x),w>, w
=
5 / 1 5 ] yields
(3.10.28), (3.10.29), while Theorem 3.10.4 implies that m
a (x,E,5) E sy,o(V). Therefore, any C -diffeomorphism
X
bijection
x,
:
x
:
U
+
V induces a
Sy,o(U) ++ sy,o(V) according to the formula (3.10.27), the
corresponding symbols being transformed according to the formulae (3.10.281, (3.10.29).
I
3. Singular Perturbations on Smooth Manifolds without Bounda y
318
C o r o l l a r y 3.10.10.
E S"(U)
L e t AE = O p a , a ( x , E , < )
of a ( x , € , D x ) . L e t
and l e t a o ( x , E , E ) b e t h e p r i n c i p a l symbol m
x
: U
V b e a C -diffeomorphism and l e t a X , O ( x , ~ , Eb)e
-f
t h e p r i n c i p a l symbol o f t h e t r a n s f o r m e d s i n g u l a r p e r t u r b a t i o n d e f i n e d by ( 3 . 1 0 . 2 7 ) . Then ( 3 . 1 0 . 2 8 ) ,
( 3 . 1 0 . 2 9 ) imply:
C o r o l l a r y 3.10.11. Let
m
x
: Rn
Rn b e a C -diffeomorphism which r e d u c e s t o a l i n e a r map
-t
o u t s i d e of some b a l l i n Rn Then f o r V s
E
R3
.
t h e diffeomorphism
x
i n d u c e s a b i j e c t i o n of each s p a c e m
H (S)
( R n ) o n t o i t s e l f . I n d e e d , u n d e r t h e C -diffeomorphism t h e i d e n t i t y
o p e r a t o r i s t r a n s f o r m e d i n t o a n o p e r a t o r J = op j , whose symbol j ( x , < ) E So ( E n ) and i s c o n s t a n t f o r 1x1 s u f f i c i e n t l y l a r g e . T h e r e f o r e , 1,o w i t h v ( x ) = (u o x ) ( x ) Theorem 3 . 4 . 1 i m p l i e s t h a t v E H ( I R n ) whenever
.
u E Hs(Rn)
Hence, t h i s p r o v e s Lemma 2 . 6 . 1 ,
w h i l e Lemma 2.6.2
is a direct
consequence o f Theorem 3 . 4 . 1 . Remark 3.10.12. m
x
A C -diffeomorphism
i s o m o r p h i c mappings: (3.10.32)
x
*
: U
x*$
V w i t h U and V open sets i n R n i n d u c e s * -1 m = $ 0 x , x*$ = ( x ) $, V $ E C (U), V IJJ E C m ( u ) , -f
m
m
: Co(V)
+
Co(U),
x*
: C m ( U ) + Co0(V)
m
a n d , by d u a l i t y between C ( V ) and D'(V) t h e isomorphism 0 (3.10.33)
x,
:
D'(U)
*
D'(V).
F u r t h e r m o r e , it g i v e s r i s e t o a map t o t h e formula ( 3 . 1 0 . 2 7 ) ,
i.e.
( 3 . 1 0 . 3 4 ) X,a
x -1 ,
= (a
o
X)
D
x,
:
op
V a(x,E,Dx)
S y , o ( u ) * Op S"
1,o
E
(V) according
Op . q y , , ( U ) ,
a s w e l l a s a symbol homomorphism (which i s a n isomorphism u p t o t h e smoothing symbols o f o r d e r (3.10.35)
-m):
(X,a) ( y , ~ , n )= a ( y , ~ , n ) , X
where a ( y , ~ , q )i s w e l l d e f i n e d a s y m p t o t i c a l l y by ( 3 . 1 0 . 2 8 )
X
smoothing symbol of o r d e r
-m.
up t o a
3.10. Diffeomorphisms and Singular Perturbations
319
F u r t h e r m o r e , t h e p r i n c i p a l symbols a r e t r a n s f o r m e d a c c o r d i n g t o t h e formula ( 3 . 1 0 . 3 1 ) , which c a n b e r e w r i t t e n a s f o l l o w s :
(x,ao) ( y , ~ , r ~ = )a o ( x , E , S ) l x = x - ' ( y ) . 5 = tX ' ( X ) l l .
(3.10.36)
m
W e c a n now d e f i n e s i n g u l a r p e r t u r b a t i o n s on a C
paracompact m a n i f o l d M.
D e f i n i t i o n 3.10.13. m
A f a m i l y of operators A €
i s s a i d t o be a s i n g u l a r p e r t u r -
: c ~ ( M ) -f c-(M)
b a t i o n i n t h e c l a s s o p Sv
1 to
and a cm diffeomorphism x
i f f o r any coordinate neighbourhood x c
(MI,
:
x
+
u c n n , t h e f a m i l y o f operators
M
A€
U
d e f i n e d by t h e commutative diagram
i s a singular p e r t u r b a t i o n i n op Sv
1to
I n o t h e r words A' -1 .
A E = (A'
o
x)
x
0
: C;(M)
+ Cm(M)
: X + U,
i s i n t h e c l a s s Op
1 s a s i n g u l a r p e r t u r b a t i o n i n Op
U consequence of Theorem 3 . 1 0 . 9 , :A
x
(u).
i s i n Op
sv
sv 1,o
sv (M), 1,o
if
(U). A s a
( U ) f o r any C -diffeomorphism
1 PO so t h a t D e f i n i t i o n 3.10.13 i s c o r r e c t , i . e . t h e c o n c e p t of
s i n g u l a r p e r t u r b a t i o n on m a n i f o l d d o e s n o t depend of s p e c i f i c c h o i c e of neighbourhoods and d i f f e o m o r p h i s m s . Furthermore, given t h e t r a n s f o r m a t i o n r u l e (3.10.36) f o r t h e p r i n c i p a l
*
symbols, t h e y a r e w e l l d e f i n e d f u n c t i o n s on t h e c o t a n g e n t b u n d l e T ( M ) . I n t h e same way one d e f i n e s s i n g u l a r p e r t u r b a t i o n s classes O p S " ( M ) and Op K v ( M ) ,
u s i n g t h e d i a g r a m above and t h e i n v a r i a n c e of the classes
S"(U), K V ( U ) (see D e f i n i t i o n s 3.3.2 and 3 . 6 . 4 ) w i t h r e s p e c t t o t h e
Cm-
diffeomorphisms. However,
one c a n g i v e an i n d e p e n d e n t i n t r i n s i c d e f i n i t i o n o f s i n g u l a r
p e r t u r b a t i o n s on a smooth m a n i f o l d . D e f i n i t i o n 3.10.14.
Let
M
be a
Cm
paracompact manifold. A f a m i l y of operators m
E
-f
: C;(M)
-f
c
(M)
i s s a i d t o be a s i n g u l a r p e r t u r b a t i o n on
i f t h e f o l l o u i n g c o n d i t i o n s are s a t i s f i e d : 1
'.
There e x i s t s a sequence k
,
7
J.
--, f o r
j
+
m,
such t h a t
M
3. Singular Perturbations on Smooth Manifolds without Bounda y
320
2O. E
There e x i s t s a sequence s . 3
+
-m
for j
+ m,
such t h a t f o r each given
E ( 0 . ~ ~one 1 has:
( 3 . 1 0 . 3 9 ) e- ip g (x)AE ( f ( x )e i p g ( x ) )
f o r v f E c;(M), v g E
-
S .
’
C b E ( f , g ) p I, p + m , j>o c ~ ( M ) , dg f 0 on supp f , g being r e a l
3O. There e x i s t s a sequence m . C 3
for
-m
j +
a,
valued.
such t h a t
E c ~ ( M ) , V g E c - ( M ) , dg # 0 on supp f , g being r e a l vaZued. Furthermore, t h e order v = ( v l , v 2 , v 3 ) of t h e singuZar p e r t u r b a t i o n is t h e l e a s t v E m3 such t h a t
for V
f
> v (’I dEf
(3.10.41) v
(k.,m.-k.,s.-m.+k.), V j 1 3 3 3 3 1
=
0,1,
...
Furthermore, t h e f u n c t i o n ( 3 . 1 0 . 4 2 ) uA ( f , g ) = e
-ig(x)A (feig(x)
E
) #
E
which is asymptoticalZy represented by (3.10.39) when I V g ( x ) [ each g i v e n
+
m
for
i s s a i d t o b e t h e symbol o f t h e operator A ~ . E R n and i n t r o d u c i n g t h e f u n c t i o n
E,
Taking g ( x ) = < x , < > , 5
one c a n r e w r i t e t h e o p e r a t o r A (3.10.44)
m
: CO(M) + C
m
(M)
i n t h e following fashion:
4
( ~ ~ u ) ( =x )( 2 7 ~ ) -1~ e IRn
where i t i s u n d e r s t o o d t h a t supp u b e l o n g s t o some p a t h V
x
: V
-f
C
U i s a diffeomorphism o f V c M o n t o some open s e t U
Furthermore,
homogeneous f u n c t i o n s o f
f u n c t i o n s on
M
X
Taking g ( x ) a
<
of d e g r e e s .
a j o b e i n g smooth
{Rn\{O}} which depend c o n t i n u o u s l y of E E ( 0 . ~ ~ 1 . = E-~<x,<>,
(f,E,S) AE
Rn.
( 3 . 1 0 . 3 9 ) becomes
with aE(f,:) 3
M and C
w i t h 5 E Rn\{O} g i v e n , and i n t r o d u c i n g
= 0A ( f , E - l S ) ,
3.10. Diffeomorphisms aird Singular Perturbations
32 1
( 3 . 1 0 . 4 0 ) becomes: (3.10.46)
uA (f,E,C)
-
Z a.(f,E,S),
0
E +
j>o
E
-1 , < ) f u n c t i o n s o f d e g r e e m . which where a . ( x , E , S ) are homogeneous i n ( E 3 3' a r e smooth f o r x E M , E Rn\{O}, f E R + .
I f v ( O ) = ( k , m -k ,s -m +k ) s a t i s f i e s t h e c o n d i t i o n 0 0 0 0 0 0 (3.10.47)
v(O) t "(1)
=
(k.,m.-k. s.-m.+k.), 3 3 3 ' 3 3 3
j = 1,2
,...,
t h e n t h e zero c o e f f i c i e n t on t h e r i g h t hand s i d e o f ( 3 . 1 0 . 4 6 ) i s s a i d t o be t h e p r i n c i p a l symbol o f A (3.10.48)
If
and i s denoted by o
AE.O'
U , ~ , ~ ( ~ , E , C )= a O ( f , E , 5 ) .
(3.10.47)
i s well defined,
i s s a t i s f i e d i . e . t h e p r i n c i p a l symbol o f
t h e n A~ E Op Sv
(0) (M)
.
I f , moreover, one has (3.10.50)
v ( O ) t v ( l ) t...> v ( 1 ) t
.. . ,
One h a s t o u s e Theorem 3 . 1 0 . 1 i n o r d e r t o show t h a t e a c h s i n g u l a r i n t h e sense of D e f i n i t i o n 3.10.13 i s a s i n g u l a r
perturbation A
p e r t u r b a t i o n i n t h e sense of D e f i n i t i o n 3.10.14 a l s o . W e s h a l l n o t e l a b o r a t e on t h i s p o i n t and l e a v e t h e d e t a i l s t o t h e r e a d e r . Example 3 . 1 0 . 1 5 . L e t R 1 denote t h e u n i t c i r c l e ,
nl
=
{eie E C
1
1
101 5 n } .
Define t h e
operator
n+
:
c;(nl)
by t h e formula (3.10.51)
( I I + ~ )( 8 )
1 lim 211
= -
6++0
and l e t
n-
= Id
71
Jr1-e
i (e-y+iS)
-1
1
u(y)dy
-Tr
-.'II
Consider t h e family o f o p e r a t o r s E
-f
A
:
Cm(nl)
+
Cm(Ql),
3. Singular Perturbations on Smooth Manifolds without Bounda y
322 (3.10.52)
A
=
Id
+
+
-
E D ~ (- ~Il ) ,
where I d i s t h e i d e n t i t y o p e r a t o r and D
e
=
-ia/ae.
The F o u r i e r series e x p a n s i o n
e s t a b l i s h e s a n isomorphism of C m ( R ) o n t o t h e s p a c e s of r a p i d l y d e c a y i n g 1 s e q u e n c e s , {u 1 -m
AEu = F - l ( l + E / n l ) F u .
We s h a l l see t h a t AE i s a s i n g u l a r p e r t u r b a t i o n i n t h e c l a s s Op S ( o r o ' l ) ( R 1 ) and even i n O p K
( O r O r l ) (R1).
L e t U C R1 b e any p a t h
U
22
( i t s u f f i c e s t o t a k e f o r i n s t a n c e (-n/2,n/2)
E
m
m
x E
x
(R 1 , : 1 on some p a t h 0 1 U , one c a n r e w r i t e t h e o p e r a t o r AE a s f o l l o w s ( w e d e n o t e 8 and y by
and i t s r o t a t e s ) and l e t u
Co
(U). With
C
1 x and y , r e s p e c t i v e l y ) :
Introducing b(x,y)
=
-1 i(x-yl (i(x-y)) [l-e IX(Y)
I
one h a s : (3.10.55)
A
u
=
u(x)
+
+ ( ~ I T ) - ~ E Dl i m
I
i b ( x , y ) [ ( x - y + i S ) - l - ( x - y - i G ) -1 ] u ( y ) d y .
6+0 R Since
m
(x-y+i6)-1
= -i/ ei(x-y+iG)c d c ,
0 one c a n r e w r i t e ( 3 . 1 0 . 5 5 ) a s f o l l o w s :
(x-y-id)
-1
=
O
i J e -m
i(x-y-iS)c dE
3.10. Difleomorphisms arid Singular Perturbations
323
where
N o t i c e t h a t t h e p r i n c i p a l symbol o f t h e s i n g u l a r p e r t u r b a t i o n A
is
a o ( x , E , C ) = 1+E151. Using ( 3 . 1 0 . 5 4 ) , one c a n d e f i n e t h e i n v e r s e o p e r a t o r A - l :
O f course, A
R1
: A-lu(f?) =
-1
i s a f a m i l y of c o n v o l u t i o n o p e r a t o r s on
(rE * u ) ( f ? ) ,
where t h e c o n v o l u t i o n k e r n e l r (8) i s g i v e n
by t h e f o r m u l a :
rE(e)
=
z
(~n)-l
(l+Elnl)-leinf?.
nE Z Introducing t h e d i s t r i b u t i o n
m
one checks t h a t r
=
(e)-E-'r(f?/E)
EqE(f?), where q ( 0 ) E C
(R 1) u n i f o r m l y
w i t h r e s p e c t t o E. One c h e c k s i n t h e s a m e way as p r e v i o u s l y f o r A p e r t u r b a t i o n i n Op
s ( o r o ' - l )(R1)
Notice t h a t A i l
E'
-1 . t h a t AE i s a singular
whose p r i n c i p a l symbol i s ( l + ~ I f l ) - ' .
allows t o solve t h e following s i n g u l a r l y perturbed
boundary v a l u e problem Av(x) = 0 , (3.10.58)
x E B1 = { X E R
a ( 1 - E--)V(X) aNe
lQl
2
, 1x1
< l},
= u(f?)8
where Nf? i s t h e inward normal a t e
if?
E
Ql,
and E > 0 .
I n d e e d , u s i n g t h e F o u r i e r series e x p a n s i o n , one f i n d s t h a t
(3.10.59)
v(x)
=
?
(211)-'
--71
where x E B ' ,
x
=
1-1x1 2
I I
1-2 x cos ( e-y)
+
I
2 ( . % i 1 u )( y ) d y ,
if? (xie
.
N o t i c e t h a t t h e problem ( 3 . 1 0 . 5 8 ) where t h e inward normal i s r e p l a c e d by t h e outward o n e , i s n o t w e l l posed f o r a l l E > 0 , s i n c e i n t h a t case 1 t h e o p e r a t o r A = F- ( l - E l n l ) F i s n o t always i n v e r t i b l e when E + 0 . I t i s -E
3. Singular Perturbations on Smooth Manifolds without Boundary
324
s t i l l a s i n g u l a r p e r t u r b a t i o n i n Op l-ElC1
i s no l o n g e r i n v e r t i b l e ,
s ( O ' O t l ) (n,)
N o w we check t h a t t h e s i n g u l a r p e r t u r b a t i o n
i n t h e c l a s s Op K ( o n o f 1 ) ( S 2
E
A
(f)
=
E S"(n,)
i . e . t h e r e i s no symbol r
such
1 w i t h ro t h e p r i n c i p a l symbol o f r .
t h a t ( l - ~ l c l )0r( x , E , ~ )
one h a s f o r e a c h f
whose p r i n c i p a l symbol
1
)
is, i n f a c t ,
(3.10.52)
i n t h e s e n s e o f D e f i n i t i o n 3.10.14.
Indeed,
Cm(al)
f(8)+Efl(e),f,(e)
so t h a t ( 3 . 1 0 . 3 8 ) h o l d s w i t h ko Now, w e check ( 3 . 1 0 . 3 9 ) .
=
= D
e (n+-n-)f(e) E cm(nl),
0 , k l = -1, k . I
First, we notice t h a t
=
--, j
> 1.
(3.10.52) f o r u
E
m
C
(R1)
can be r e w r i t t e n as f o l l o w s
f o r each g Moreover,
E
,
Cm(lR)
g' ( 8 )
# 0 on supp f
( 8 ) , g being real valued.
(3.10.61) can b e d i f f e r e n t i a t e d with r e s p e c t t o
I n d e e d , assuming t h a t g ' ( 8 ) > 0 on supp f
e
and p .
( t h e case g ' ( 8 ) < 0 c a n b e
t r e a t e d i n a c o m p l e t e l y analoguous w a y ) , i n t r o d u c e t h e new v a r i a b l e t = g ( y ) - g ( 8 ) and l e t y
=
x e ( t ) .Then
where
al(e,t) al(B,-)
E
C:(R)
=
,V 0
F u r t h e r , with $ ( t )
E
t(e-y)-'a(e,y)
E
Q1,
m
CO!R)
(g*(y))-lf(y), y =
and a l ( B , O )
,
=
xe(t),
f(8).
$(t) F 1 f o r It1 S a , $ ( t )
0 f o r It1 2 2 a ,
a b e i n g s u f f i c i e n t l y s m a l l , one c a n r e w r i t e I ( 8 , p ) as f o l l o w s
3.10. Diffeomorphisms and Singular Perturbations
325
The function
-03
Therefore, one has I1(e,p) = O ( p part (or Proposition 3.8.1).
)
for
p +
m,
as shows the integration by
Furthermore, the function
is analytic for complex t # 0 , It1 < a. Using the Cauchy theorem, one can rewrite I ( 8 , p )
2
in the following
fashion Il(e,p) = e ipg(8)f(8) lim [
6++0
J t-'$(t)eiPtdt + lt/>6
dt + nil,
+ 6+6 t-leiPt where
C i =
{t E
C
1
I
It1 = 6, Im t > 0).
Therefore, with
ri
the contour consisting of intervals [-2a,-6],
[6,2a] and the half circle C i , one has (3.10.63) I 1 ( B , p )
=
eipg(e)f(8) [ J + t-'$(t)eiptdt +nil.
r6 The integration by part yields: (3.10.64) /+ t-l (t)eit dt
-m
= O(p
),
forp
+
m,
r6 so that one gets combining (3.10.62)-(3.10.64) the asymptotic formula
(3.10.61). Differentiation with respect to p or 8 leads to the same kind of integrals. Combining (3.10.611, (3.10.62), one find -m
(3.10.65) ewipgA (feipg) = (l+Epjgl(e)j)f(e)+EDef(e)+O(p
),
so that (3.10.39) holds, as well, with s = 1, s1 = 0, s . 0 3 -1 in . Setting p = E (3.10.611, one finds
=
p + m
-m,
j > 1.
3. Singular Perturbations on Smooth Manifolds without Boundary
326 when
E +
0 , so t h a t (3.10.40)
h o l d s w i t h mo
=
0, ml = -1, m .
3
= -a,
3 > 1.
One c o u l d have u s e d t h e f a c t t h a t t h e c o n v o l u t i o n o p e r a t o r ( n i )
-1
v . p . x-'*
can be considered a s t h e s i n g u l a r p e r t u r b a t i o n ( p s e u d o d i f f e r e n t i a l o p e r s t o r of o r d e r z e r o ) w i t h symbol sgn
c,
and have a p p l i e d Theorem 3.10.1
However,wehavepreferedtogivehere
i n o r d e r t o g e t t h e expansion (3.10.65).
a n i n d e p e n d e n t argument u s i n g s o m e d i f f e r e n t The o r d e r of A "(1) =
(-m,--,-m),
technique.
i s " ( O ) = ( 0 , 0 , 1 ) , f u r t h e r m o r e , one h a s " ( l )
7"
>
1. Hence, A E
E op
P ( O ) (fill,
=
(-l,O,O),
i t s p r i n c i p a l symbol
being the function
and i t s symbol b e i n g
u
(f,c)
E/clf(B)+(f(e)+EDef(e)).
=
AE
The r e a s o n f o r which t h e e x p a n s i o n s ( 3 . 1 0 . 3 8 ) - ( 3 . 1 0 . 4 0 ) perturbation A
d e f i n e d by (3.10.52)
f o r the singular
( o r e q u i v a l e n t l y , by ( 3 . 1 0 . 5 4 ) )
c o n t a i n o n l y two t e r m s i s t h a t it d i f f e r s o n l y by a smoothing o p e r a t o r
i s a l i n e a r f u n c t i o n of
1)
( 1 + I~D
from t h e s i n g u l a r p e r t u r b a t i o n
E Op
s (o'ofl)(pi) ,
whose symbol
and a p i e c e - l i n e a r f u n c t i o n of 5 , so t h a t
6
0 f o r l a [ > 1 and 5 # 0. 1 m 1 = {z(s) E C , s E [ 0 , 1 ] } b e any c l o s e d Jordan C c u r v e i n C . One
Da(l+EISl)
5
Let
r
can c o n s i d e r t h e corresponding o p e r a t o r A
: Cm(r) + Cm(r)
g i v e n by t h e
formula (3.10.66)
where D
(A u ) ( z )
=
-1 -1 (ri) I(z-5) u ( < ) d < , z E
u (z)+ E a(z)D v.p.
r
is the tangential derivative a t z E
r.
The same c o n c l u s i o n s are v a l i d f o r t h e o p e r a t o r (3.10.661,
K(ofo'l)
a s i n g u l a r p e r t u r b a t i o n i n Op
(r)
r
A
being
w i t h t h e p r i n c i p a l symbEl
/el.
a o ( z , ~ c )= 1 + E a ( z )
I-' 5 Cis>-1 , V ( z , n ) E r x R , t h e n a Op s ( o f o ' -(lr)) w i t h t h e p r i n c i p a l symbol
F u r t h e r m o r e , i f jao(z,c) singular perturbation R
E
r ( Z , E ~ )= ( l + ~ a ( zlt1)41 ) and s u c h t h a t i t s symbol r ( z , ~ < s) a t i s f i e s t h e 0 c o n d i t i o n s o f Theorem 3.10.7 h a s t h e f o l l o w i n g p r o p e r t y : (3.10.67)
A
0
E
R
E
=
I d + EQA'),
R
0
€
A
6
=
Id
+ EQ ( 2 ) E
'
3.10. Diffeornovphisrns and Singular Perturbations
E
where Q ( 1 )
Indeed,
r
if
s ( o ' o r o () r )
and I d i s t h e i d e n t i t y . 1,o ( 3 . 1 0 . 6 7 ) i s a consequence of C o r o l l a r y 3.10.8.
Op
= R , a(z) f a
(3.10.67)
i f RE
321
=
# 0 , l+alnl # 0 , V
Op ( 1 + E l C l ) - ' .
convolution o p e r a t ors R
E
I n t h e l a t t e r c a s e R:
= E-lr(x/E)
*
Notice t h a t
R , t h e n Q ( 1 ) : 0 , j = 1.2 i n
u , V u E C;(R)
i s a f a m i l y of
,
where t h e
d i s t r i b u t i o n a l k e r n e l r ( x ) h a s t h e form (3.10.68)
r(x)
=
(2n)
-1
J (l+lg])
-1 i x g e dg.
R One c a n u s e ( 3 . 1 0 . 6 8 ) f o r c o n s t r a c t i n g an o p e r a t o r R (3.10.67)
on a smooth c l o s e d c u r v e
r
of a p i e c e of z E
r
r
with p r o p e r t y
above. I n d e e d , t a k i n g t h e l e n g t h t
between a g i v e n f i x e d p o i n t zo
a s a g l o b a l c o o r d i n a t e , so t h a t z
E r
and a g e n e r i c p o i n t
z ( t ) , one r e w r i t e s A
=
i n the
following fashion ( A v ) ( t ) = v ( t ) + E a ( z ( t ) ) D tv . p .
J (t-s)-lK(t,s)v(s)ds,
(Ti)-'
R
where v ( t )
(u
=
m
o
K(t,s)
z) ( t ) E C O ( R ) , and =
(t-s)( z ( t ) - Z ( S ) ) - l Z ' ( S ) .
I n t r o d u c inq
one d e f i n e s t h e o p e r a t o r R
(REu
o
2)
a s follows:
(t)= J R
The p r i n c i p a l symbol of A p r i n c i p a l symbol of R
K E ( t , s ) (U
o
Z)
(s)ds.
b e i n g t h e f u n c t i o n a.
b e i n g ro
=
(1+Ea( 2 )
=
( l + ~ a ( z151 )
and t h e
15 I ) - l , C o r o l l a r y 3.10.8 i m p l i e s
the property (3.10.67). I t i s l e f t t o t h e r e a d e r t o check t h a t t h e s i n g u l a r p e r t u r b a t i o n
i s t i g h t l y connected with t h e o p e r a t o r B
(3.10.66)
U
C
R2
t h e i n t e r i o r of
problem: Aw z
E
r,
=
where N
mappings.
)
0 i n U, w
d e f i n e d a s f o l l o w s . With
r
and w i t h w t h e s o l u t i o n Ef t h e D i r i c h l e t
=
u on
r,
l e t (B u)
i s t h e inward normal a t z E
(2)
r.
=
(l-Ea(z)a/aNZ)u(z), ( H i n t : u s e conformal
3. Singular Perturbations on Smooth Manifolds without Boundary
328
3 . 1 1 . An a l g e b r a o f D i f f e r e n c e O p e r a t o r s I n t h i s s e c t i o n s e v e r a l c l a s s e s o f one p a r a m e t e r f a m i l i e s o f d i f f e r e n c e o p e r a t o r s are c o n s i d e r e d .
I n some r e s p e c t t h e y are s i m i l a r t o t h e c o r r e s -
ponding c l a s s e s o f s i n g u l a r p e r t u r b a t i o n s introduced i n Sections 3.1
-
( p s e u d o d i f f e r e n t i a l operators1
3 . 3 , 3 . 9 and a r e i n t e n d e d f o r n u m e r i c a l
t r e a t m e n t o f boundary and i n i t i a l v a l u e problems f o r t h e s e o p e r a t o r s . Throughout h i s s e c t i o n h
E (O,ho]
w i l l b e a s m a l l p a r a m e t e r and U
h a one p a r a m e t e r f a m i l y of u n i f o r m meshes ( g r i d s ) w i t h meshsize h i n an open s e t
u c - mn.
W e s t a r t w i t h t h e c a s e when U = R n
and Uh
=
<<
{ek}lsk=
,
= hZn,
.ih".
1 5 k S n be the s h i f t operator i n the positive x -direction k
a l o n g t h e c o o r d i n a t e v e c t o r e k , so t h a t f o r any mesh-function u : I€?n+ C one h a s Q u = u ( x + h e 1 . h k k Using t h e d i s c r e t e F o u r i e r t r a n s f o r m F
x+S , h
and i t s i n v e r s e F
-1 E-fx,h '
one c a n w r i t e
(0 u ) (x) k
where 8
k
(n)
= F
<+XI
v
FJk(h<) F
-1
h
X+E
k
so t h a t
I f a ( x , h , Q ) i s a polynomial i n O k ,
(3.11.1)
,
= exp(i<ek,rp) = exp(irik).
Denote by 0-l t h e i n v e r s e o f Elk,
which a r e Cm
u E C;(IRn)
,bur
-1
0 1 2 k 5 n , with c o e f f i c i e n t s , k ' f u n c t i o n s of x E R n and depend c o n t i n u o u s l y on (x,h)E R n X ( O , h o ] ,
a(x.h,@) =
Z
aa(x,h)Oa,
ClEZ"
t h e sum on t h e r i g h t hand s i d e o f
( 3 . 1 1 . 1 ) b e i n g f i n i t e , t h e n one can
a s s o c i a t e w i t h t h i s d i f f e r e n c e o p e r a t o r t h e 2n/h p e r i o d i c i n e a c h 1 6 j 5 n, function a(x,h,FJ(hS I ) ,
e(n)
< I'
= ( ' J l ( n ),..., e n ( r l ) ) , so t h a t one
has
-1
,
is i r r e l e v a n t , and one can u s e Bk ( 3 . 1 1 . 2 ) f o r d e f i n i n g a c o r r e s p o n d i n g d i f f e r e n c e o p e r a t o r f o r any " d e c e n t "
The f a c t t h a t a i s polynomial i n 9k ,
21r/h p e r i o d i c i n e a c h E j ,
1 C j 2 n , f u n c t i o n a ( x , h , h < ) . Now w e s h a l l g i v e
a p r e c i s e meaning to t h e word " d e c e n t " by i n t r o d u c i n g t h e c o r r e s p o n d i n g
i . t
3.11. A n Algebra of Difference Operators
329
classes of functions a(x,h,hC) which will be also called symbols Definition 3 . 1 1 . 1 . A f u n c t i o n a(x,h,n)
FY,o(u),
:
U
v = ( v l , v 2 ) E IR
zf f o r each compact
c
u,a there e x i s t s a constant
idhere 5
a,B,K
-vl
I
,...,cn), sk
= (il
<5>
+I
,
u and each m u l t i - i n d i c e s
V (x,h,5) E K
X
(O,hol
Bn,
x
(exp(ihCk)-l).
,...,w
1, w
w(c)
h
= (wl
<
introduced ;reviously
We shall denote by The partial ordering <,
-1
(ih)
=
K c
such t h a t one has:
IDBDaa(x,h,h<) 5 C h x s a,B,K
(3.11.3)
s a i d t o b e a symbol i n t h e c l a s s
is
(O,hol x : T 2 .
x
=
the vector 5 with h = 1 .
for vectors in W3
induces,
in a natural way, an order relation in R2 , if one identifies the vectors v
=
E
( v l , v 2 ) E B2 and ( v l , v 2 , 0 )
If a E
FY,,(U)
but a f F'
order of the symbol a: ord a
=
1 to v.
W
3
.
(U), V 1-1 < v, then v is said to be the
Let I$ (t) E C " ( K + ) be a cut-off function such that I$ (t) :0 for
6
6
4 (t) : 1 for t
t E [0,61,
6
2 26,
4 (t) t 0, v t E K + . 6
Definition 3 . 1 1 . 2 . A symbol a C Fv
1,o
(u) i s s a i d t o belong t o t h e c l a s s
a f u n c t i o n ao(x,n) E set
K c
u
x
r'7
one has:
( T ~\{o}) 1,rl
and moreover, i f f o r some 6
>
0 one has
< v. The j u n c t i o n ao(x,q) : I R ~x (T~\IO:)
f o r some Symbol
Fv (u), i f t h e r e e x i s t s
such t h a t uniformly on each compact
(TY \ { O j ) )
Cm(U x
1-1
Of
+
c is c a l l e d t h e p r i n c i p a l
F V ( 2 ) , h r l - ( v 1 + v 2 )a0 (x,hE) being
t h e symbol a E
homogeneous i n
(h-l,F) of degree v l + v 2 . Proposition 3 . 1 1 . 3 .
If a(x,h,hS) C F V ( u ) ther, t h e p r i n c i p a l symbol a (x,n) 0
:
U
x
(Tn \{01) 1,n
+
c is a cm-function which on every compact
and for each m u l t i - i n d i c e s a,B s a t i s f i e s t h e i n e q u a l i t i e s : (3.11.6)
ID:D:ao(x,n)
1
5 Ca , O , K I w ( q )
I
v2-lal
, v (X,ri) E
K
(T>{Oj),
K c
u
3. Singular Perturbations on Smooth Manifolds without Boundary
3 30
where t h e c o n s t a n t s c
depend onZy on t h e i r s u b s c r i p t s . v +v -In1
a,B,K
Proof. Multiplying ( 3 . 1 1 . 3 ) by h -
and letting h
-f
0, ( 3 . 1 1 . 4 )
I
yields ( 3 . 1 1 . 6 ) .
m
The class of functions a (x,n) E C ( U
0
.
(TY,n\{O})) satisfying (3.11.6)
X
is denoted by HV2 (U)
Definition 3.11.4.
A symbol a E F V ( u ) i s s a i d t o beZong t o t h e cZass G' ( u ) , i f t h e r e exist a (j)= (j) (j)) =2 , v = v (0)> v ( l ) > . . _ ,v ( j ) + v ( J )4. -m, sequence v ( v l ,v2 1 2 .(I)
for j
and f u n c t i o n s a .(x,n) E H
+ m,
(u) such t h a t f o r any given
7
0 holds:
integer N
>
(3.11.7)
(a(x,h,h5)-$6(1) C
-
(v; j
h
) +V ( j) )
2
(N) aj(x,hS)) E
Fv
1,o
O<j
(U).
The forma2 sum
i s ca2Zed t h e graded def a .(x,h,hS) = 1
order
=
9'1)
(x,h,F,)E
u
x
symbo2 a s s o c i a t e d w i t h a E G ' ( u ) , whiZe -(,,;J)+,,(I) 2 h a.(x,hC) i s s a i d t o be homogeneous symboZ of 3
, a ,(x,h,h5) b e i n g a we22 d e f i n e d cm function of 1 , ht # 0 (mod 2 n ) , homogewous of degree v 1( 1 ) + v2( J ) Rn , hc # 0 (mod 2 7 ~ ) . The class of a l l homogeneous
( V ~ J ), v : y
R+
i n ( I I - ~ , ~E) IR+
x x
)
R
symboi's a(x,h,hS) of order v
=
(v1,v2) and such t h a t a(x,l,n) E H
v2
(U),
i s denoted by H'(U).
Theorem 3 . 1 1 . 5
L e t ar be agraded symbol, -(V1
(3.11.9) a (x,h,q) =
h
C
( j)
+ v (1)
(j)
a .(x,rl), 7
j 20
w i t h v = v") > v ( ' ) t ..., and w i t h v ~ J ) + v ~ J4.) Then t h e r e e x i s t s a symbol a(x,h,h<) E G ' ( u ) ,
a. E 7
-- f o r
j
H~ -f
(u)
-.
for which ( 3 . 1 1 . 8 ) i s
t h e graded symbol. ~
Proof. With 4 (t) the cut-off function introduced above and {tj}j20 6 t . C 0 for j + let b.(x,h,hc)=$6(tj1<1)a.(x,h,hS)and let 3 3 1 de f ( 3 . 1 1 . 1 0 ) a(x,h,hE) = C b,(x,h,hC). (a,
jro
J
,
33 I
3.21. An Algebra of Difference Operators For e a c h g i v e n h E (O,ho!
and f o r each g i v e n ( x , S ) E lRn
Let
%,
k = 0,1,2,...,
lR;
X
,
t h e r e are
(3.11.10).
o n l y f i n i t e 1 . y many t e r m s on t h e r i g h t hand s i d e o f
b e a Sequence of compact s u b s e t s , e x h a u s t i n o U .
V (x,h,S)
E
%
(O,hol
X
X
mn,
f o r lal+lBi+k S j . L e t t o = 1. The L e i b n i t z formula y i e l d s :
O b v i o u s l y , one h a s f o r x E LY.:
S i n c e $ , ( t , l <El )0 f o r t . 1 5 1 5 6 , one f i n d s f o r x E I 7
If la-yl
=
5
1 > 0 i n t h e sum on t h e r i g h t hand s i d e o f
using the l a s t
(3.11.12) then a f t e r
q time d i f f e r e n t i a t i n g @ S ( t . I < q l 5) ,1, one g e t s tqhl-q a s a f a c t o r . Taking I 1 i n t o a c c o u n t t h a t t h e s u p p o r t of any d e r i v a t i v e of b e l o n g s to t h e s e t -1 , one g e t s f o r each t e r m i n t h e sum ( 3 . 1 1 . 1 2 ) {6 6 tkl
k
t h e same e s t i m a t e
Therefore,one has f o r x (j-l)-"(j) -"(I) "(j-i)-lUl 1 2 v2 t. 2 h <5>
(3.11.13) since h 5 C l i 1 - l .
/ D R D C l b .( x , h , h S )
x 5 1
5 C
a,B,k
E
3
when Denoting by m , t h e maximum of a l l C 3 a,B.k t . t o be
I a1 +I R 1 +k
5 j , and c h o o s i n g
I
one g e t s t h e i n e q u a l i t i e s
%:
(3.11.11).
One c h e c k s e a s i l y t h a t t h e f u n c t i o n a ( x , h , h C ) d e f i n e d by
(3.11.10) with
332
3 . Singular Perturbations on Smooth Manifolds without Boundary (j)
b ( x , h , h 1 E F" (U) s a t i s f y i n g ( 3 . 1 1 . 1 1 1 , 3 1# O ( 3 . 1 1 . 3 1 , so t h a t a ( x , h , h c ) E F" (U). 1 to
s a t i s f i e s the inequalities
C o r o l l a r y 3.11.6.
k
E
L e t a (x,h,hC)
k
Gv(U),
1 , 2 , have t h e same g r a d u a t e symbol a
=
r'
(j)
ar(x,h,n)
Z
=
a.(x,h,n),
v
=
v(O)
>
"(l)
...,
>
Then k ( x , h , h < )
=
E ffv
a.
I
j20
I
(U),
" ( J ) + " ( j ) + -m f o r j + m. l1 2 a ( x , h , h c ) - a ( x , h , h c ) i s a symbol i n
*
s 1(-a,-) ,o
(U), i . e .
f o r e a c h i n t e g e r N t 0 , e a c h compact K c U and e a c h p a i r of m u l t i - i n d i c e s
a,R t h e r e e x i s t s a constant C (3.11.14)
a , B ,N,K
such t h a t -N
,
5 C n,0,N,K<5>
/D:D;k(x,h,hc)]
I n d e e d , t h i s i s a n immediate consequence o f
symbol k
r
V (x,h,C) E K
x
(O,hol
x
R;.
( 3 . 1 1 . 7 ) , s i n c e t h e graded
of k v a n i s h e s i d e n t i c a l l y .
P r o p o s i t i o n 3.11.7.
L e t a ( x , h , h ( ) E F"
1t o
(U), b ( x , h , h E ) E F'
1 ,O
(U). Then
and (3.11.16) a b
E
F"+'(U). 1,o
Furthermore, i f w i t h some c o n s t a n t c (3.11.17) la(x,h,hC)-'l
5 C
h
0 one has:
-v 2 , V ( x , h , C ) E U
vl
x
(O,h,,]
X
Rn,
then ( 3 . 1 1 . 1 8 ) a ( x , h , h C ) - l E F;Yo(U). P r o o f . All t h e s t a t e m e n t s above a r e a n immediate consequence o f t h e c h a i n ___ rule. D e f i n i t i o n 3.11.8.
L e t a ( x , h , h S ) E Fv a(x,h,hD)
m
:
Co(u)
1 ,o +
(U). The corresponding f a m i l y o f d i f f e r e n c e operators
D ' ( u ) i s d e f i n e d by t h e formula
(3.11.19) a ( x , h , h D ) u = F
where
5
-1 C+X
a(x,h,h<)F
X-S
u+\u,
i s a one parameter f a m i l y o f operators t h a t can b e extended a s a
(continuous in h E [O,hO])f a m i l y of continuous l i n e a r mappings from E ' ( u )
333
3.1 1. An Algebra of Difference Operators
into
Crn(U1.
The n o t a t i o n Op a w i l l a l s o b e u s e d f o r d e n o t i n g t h e d i f f e r e n c e o p e r a t o r a s s o c i a t e d w i t h t h e symbol a E
Fv
1 to
(U).
Remark 3 . 1 1 . 9 . Using t h e P o i s s o n formula r e l a t i n g i n t e g r a l and d i s c r e t e F o u r i e r t r a n s f o r m s and t h e 2 1 r - p e r i o d i c i t y of q (3.11.20) a ( x , h , h D ) u
=
F
-f
a ( x , h , q ) , one c a n r e w r i t e ( 3 . 1 1 . 1 9 ) a s f o l l o w s :
-1 a(x,h,hS)F u+K u E+x,h x-fS.h h
-1 t h e d i s c r e t e F o u r i e r t r a n s f o r m and FS-fx,h i t s i n v e r s e . with F x+S , h makes s e n s e f o r any meshThe d e f i n i t i o n ( 3 . 1 1 . 2 0 ) (modulo t h e t e r m \ ) function u ( h ; x )
:
RE+ C , whose F o u r i e r t r a n s f o r m i s a 2n/h p e r i o d i c
d i s t r i b u t i o n on t h e t o r u s Tn S ,h. Examples 3.11.10. 1. L e t a = ( al , . . . , a n )
<
E
Z n and l e t e"(h5)
=
e x p ( i < h a , S > ) , Ba(hS) E FY,oAIRn)
The c o r r e s p o n d i n g d i f f e r e n c e o p e r a t o r i s t h e f a m i l y o f s h i f t o p e r a t o r s 0
h hZn, Oau(x) = u ( x + h a ) . Of c o u r s e , t o t h e complex c o n j u g a t e symbol h Ba(hS) * = e x p ( - i < h a , S > ) c o r r e s p o n d s t h e i n v e r s e s h i f t o p e r a t o r h ' O-au(x) = u ( x - h a ) . h : E Op Fo (U), V a E Z n . Obviously, 0
on
=
a. 2 . L e t < . ( h , h S ) = ( i h ) - l ( e J ( h < ) - l ) and l e t S * ( h , h S ) b e i t s complex 7 c o n j u g a t e . Obviously < , , c * E F ( O ' ' ) ( R n ) and t h e c o r r e s p o n d i n g f a m i l i e s 3 7 1;o of d i f f e r e n c e o p e r a t o r s D ,., h , D j , h a:-e ( m u l t i p l i e d by - i ) f o r w a r d and back-
ward f i n i t e d i f f e r e n c e d e r i v a t l v e s , D
3th
u = ( i h ) - l ( u ( x + h e . ) - u ( x ) ) ,D* u = ( i h ) J J rh
The symbols 5
a,h
=
-1 (ih) (e"(hS)-l),
-1
(u(x)-u(x-he.)). 7
define
t h e ( m u l t i p l i e d by -i) f o r w a r d and backward f i n i t e d i f f e r e n c e d e r i v a t i v e s i n t h e d i r e c t i o n a E Zn. 3.
To
]
t h e symbol
One h a s : =
Obviously, (-Ah) 4 . With
E
=
h
-2
a E Zn.
F(Or2)
of - A ,
A b e i n g t h e Laplace o p e r a t o r ,
(2u( x )-u ( x + h e , -u ( x - h e , 1 ) 1
16j5n
E op
E F ( O r l ) (&,V
I<.<*E F'Or2' (R*')i s a s s o c i a t e d t h e u s u a l
1 7 f i n i t e d i f f e r e n c e a p p r o x i m a t i o n -Ah -A u ( X I h
(mn).
1
3. Singular Perturbations on Smooth Manifolds without Bounda y
334
<
d e f i n e s a n o t h e r f i n i t e d i f f e r e n c e a p p r o x i m a t i o n of t h e L a p l a c e e q u a t i o n 4 o n t h e mesh f o r which t h e a p p r o x i m a t i o n e r r o r i s O ( h ) when h + 0 . With
t h e symbol (h,hC)
a:')
=
(-151
2
+a(h,hS)-b(h,hS))
E
F(o'2)(iK*1j
1, o
d e f i n e s a f i n i t e d i f f e r e n c e a p p r o x i m a t i o n of t h e Laplace e q u a t i o n on 6 w i t h t h e a p p r o x i m a t i o n e r r o r O(h ) when h -f 0 . One c h e c k s e a s i l y t h a t a ( ] ) (h,hC) E F ( O r 2 '
a
5. L e t a ( h , h E )
=
(mn).
(l+I
'-*) E IR. O i v i o u s l y , a E F (1O,o
.
(BR)
The c o r r e s p o n d i n g d i f f e r e n c e o p e r a t o r a ( h , h D ) s o l v e s t h e e q u a t i o n on (3.11.21)
(1+D D * ) u ( x ) h h
=
5:
f(x),
I?)
i.e. u
= a ( h , h D ) f = F-l - l Fx+5,h f . (1+16 S-fx, h E v i d e n t l y , one c a n r e w r i t e a ( h , h D ) a s a f a m i l y of c o n v o l u t i o n
operators,
We a r e g o i n g t o f i n d an e x p l i c i t formula f o r t h e k e r n e l G ( x ) which i s h nothing e l s e b u t t h e Green's function f o r t h e d i f f e r e n c e equation (3.11.21) One h a s
'h,C The l a s t i n t e g r a l c a n a l s o b e r e w r i t t e n a s f o l l o w s :
With e + ( h )
=
2
l + ( h / 2 ) - h ( 1 + h 2 / 4 ) ' t h e z e r o of t h e e q u a t i o n
e2 -
2 (2+h ) e + 1 = O
i n s i d e t h e u n i t c i r c l e 181 = 1 , one f i n d s u s i n g t h e residuum c a l c u l u s :
3.12. An Algebra of Difference Operators Notice, t h a t t h e r e i s a point-wise s o l u t i o n G(x)
=
335
convergence of G ( x ) t o t h e fundamental
9
( 1 / 2 ) e x p ( - [ x [ ) f o r t h e o p e r a t o r 1-d / d x 2 , when h
I t i s e a s i l y seen t h a t ( 1 + / < l 2 ) - ' E
-f
0.
F ( 0 r - 2 ) (U). 4
6. L e t a ( h , h E ) = a l ( h , h S ) + a 2 ( h , h 5 ) w i t h a l ( h , h S ) = hi and a 2 ( h , h S ) F(-1,4)
Obviously, a
( u ) , a2 E
~ ( O t ~ ) ( u S) i.n c e n e i t h e r ( - 1 , 4 )
>
=
151
(0,2)
2
.
1 ( - 1 , 4 ) , t h e symbol a = a + a 2 does n o t have a w e l l d e f i n e d 1 p r i n c i p a l symbol, s o t h a t a E Fv (U) w i t h v E iR2 t h e l e a s t v e c t o r such 1 ,0 t h a t v > ( - 1 , 4 ) , v > ( 0 , 2 ) , b u t a B F v ( U ) . Obviously, i n t h e c a s e c o n s i d e r e d , nor (0,Z)
v
=
>
(0,3).
Theorem 3.11.11.
L e t a ( x , h , h S ) E Fv
1.0
(3.11.23) h
v1
Then uniformly w i t h r e s p e c t t o h E [O,h 0] one has:
(U).
a(x,h,hD)
:
C;(U)
-f
Cm(U).
m
P r o o f . One h a s f o r e a c h u E C (U):
0
~
Since
G ( S ) i s rapidly decreasing for I S ]
-f
= and a ( x , n , h t ; ) s a t i s f i e s
( 3 . 1 1 . 3 ) one f i n d s immediately t h a t h V 1 a ( x , h , h D ) u b e l o n g s t o a bounded s e t
I
i n C"(U), V h E [O,h,,I.
Theorem 3 . 1 1 . 1 2 . de f L e t Ah = a ( x , h , h D ) E Op Gv and
be i t s graduate symbol, a , ( x , ~ E) v = v(O) > v ( l ) t
1
_ _.. Then
-(v;j)+v(J)) k t ay(x,h,n) (j) v2
H
f o r each
i h - <x
1 h jt0
(u), ( v ~ J ) + v ~ J )4)
rl
a.(x,n) 7 -m
E Tn\{O) and each f
tQ>f ( x ) )
P r o o f . The d e f i n i t i o n ( 3 . 1 1 . 2 0 ) y i e l d s : __
The T a v l o r f o r m u l a v i e l d s :
=
-
-, E c i (U) hoZds
for j
+
3 . Singular Perturbations on Smooth Manifolds without Bounda y
336
where R
M
i s t h e remainder,
(3.11.25) R (x,h,n,hS) = M
C
hM ( a ) a (x,h,rl+yhS)Sa, y E [ 0 , 1 1 .
IaI=M '!
Furthermore, s i n c e a ( x , h , h S ) E G v ,
(3.11.7) y i e l d s :
v a
V x E K C C U , V h E (O,ho],
E
Zy,
v
N 2 0
Now w e e s t i m a t e t h e r e m a i n d e r . On t h e s e t
one h a s f o r
1011
= M >
v
x E
2'
K C C U, h
f (O,hol:
Further, ( 3 . 1 1 . 2 8 ) / h M a ( a )( x , h , n + y h S )
-V
/
5 CM,K h
V
'<<>
-M
,
V x E K, V h, V
5 E
En
( 3 . 1 1 . 2 8 ) , one c a n e s t i m a t e t h e r e m a i n d e r t e r m
Combining ( 3 . 1 1 . 2 7 ) ,
i n t h e following fashion: (3.11.29)
IRM(f)I 5
m
Since f E C o ( U ) , O ( h m ) when h
+
t h e l a s t i n t e g r a l on t h e r i g h t hand s i d e o f
0.
Hence, M-(V
(3.11.30)
RM(f) = O(h
+V
1
2
) ),
h + O
(3.11.29) i s
I
3.11. A n Algebra of Difference Operators
u
u n i f o r m l y on any compact s e t ( x , q ) E
337
101.
x Tn\
11
I n o r d e r t o f i n i s h t h e p r o o f , one h a s o n l y t o show t h a t (3.11.31) e-ih
-1
<x,n>
ih-l<xtl^l>f( x ) )
'i,( e
u n i f o r m l y on any compact s e t i n
u
x
T~
O,,,a,,
=
~
\{oI.
1,ri One p r o c e e d s as f o l l o w s . By S c h w a r t z ' s k e r n e l t h e o r e m , one c a n w r i t e
%
:
E ~ ( u -+) crn(u)a s f o i i o w s
(3.11.32) K u ( x ) h
=
V u E E'(U), m
where t h e k e r n e l f u n c t i o n k
h
( x , y ) E C (U
U) uniformly with r e s p e c t t o
X
h E [O,hol. Let
x
m
E Co(U),
x
: 1 on some open subset U
kh(XrE)
1
s u p p f . Then i n t r o d u c i n g
2
Fy+s k h ( x r Y ) X ( Y )
=
one can w r i t e , a s p r e v i o u s l y f o r a ( x , h , h < ) : -ih-l<x,n> K
ih-l<x,ri> f ( x ) )
=
h(e
S i n c e k h ( x , s ) i s r a p i d l y d e c r e a s i n g (uniformlywithrespecttohandx)w h e n 5 - t m , o n e g e t s conclusion ( 3 . 1 1 . 3 1 ) , u s i n g t h e sameargumentas i n t h e e s t i m a t e o f t h e remainder R M
.
Now a n a n a l o g u e o f t h e a s y m p t o t i c f o r m u l a ( 3 . 1 0 . 1 ) w i l l b e s t a t e d and proved f o r d i f f e r e n c e o p e r a t o r s . Theorem 3.11.13.
Let f E
E
Cm(U), g
0
Cm(E),
LeR a ( x , h , h D ) E Op F"
1.0
integer
N 2 0
: C:(U)
holds: -1
( 3 . 1 1 . 3 3 ) e-ih
and d g ( x ) # 0 , V x E s u p p f . + C m ( U ) . Then for each
g being real-valued
(U), a(x,h,hD)
9(X)a(x,h,hD
(f(x) e
ih-lg (x)
)
=
where (3.11.34) $ (x,Y) 9
= g(X)-g(y)-
and where t h e remainder s a t i s f i e s on any compact K c u ( w i t h a c o n s t a n t
3. Singular Perturbations on Smooth Manifolds without Boundary
338
c
depending only on i t s s u b s c r i p t s ) t h e f o l l o w i n g e s t i m a t e
NrK
(3.11.35)
( R (x,h) N
1
5 C
N,K
,
hN-"/21-vl-"2
P r o o f . A f t e r t h e change o f v a r i a b l e s ~
5
-f
V (x,h) E K
x
(O,hol.
h c , one g e t s t h e f o r m u l a
-1 ( 3 . 1 1 . 3 6 ) e-ih
g ( X ) a ( x , h , h D x )( f ( x ) e i h - l g ( x ) ) =
where
and t h e i n t e g r a t i o n on t h e r i g h t hand s i d e o f
5.
r e s p e c t t o y and a f t e r w a r d w i t h r e s p e c t t o Then f o r
151 2 R ,
(3.11.35) i s f i r s t with Let
IV g(y)[ 5
C,
Y
V y E
u.
R sufficiently large, the operator
i s w e l l d e f i n e d a n d , moreover, one h a s :
S p l i t t i n g t h e i n t e g r a l on t h e r i g h t hand s i d e of where t h e i n t e g r a t i o n w i t h r e s p e c t t o {
151
< R} and
> R},
(3.11.36)
i n t o two p a r t s
5 i s r e s p e c t i v e l y over t h e sets
u s i n g t h e o p e r a t o r ( 3 . 1 1 . 3 8 ) and t h e f o r m u l a
( 3 . 1 1 . 3 9 ) , one g e t s , a f t e r [ n / 2 ] + l p a r t i a l i n t e g r a t i o n s i n t h e second p a r t , a r e p r e s e n t a t i o n o f t h e l e f t hand s i d e o f
( 3 . 8 . 3 6 ) by a b s o l u t e l y and
uniformly convergent i n t e g r a l s . Using t h e same argument a s i n t h e p r o o f o f Theorem 3 . 1 0 . 1 ,
one g e t s t h e
c o n c l u s i o n t h a t t h e s t a t i o n a r y p h a s e method i s a p p l i c a b l e t o t h e i n t e g r a l ( 3 . 1 1 . 3 6 ) , t h e o n l y s t a t i o n a r y p o i n t of t h e p h a s e
on t h e r i g h t hand s i d e of function
+
( d e f i n e d by ( 3 . 1 1 . 3 7 ) ) b e i n g t h e p o i n t M(x) = ( x , V x g ( x ) ) ; b e s i d e s ,
t h e same argument a s p r e v i o u s l y i n t h e p r o o f of Theorem 3 . 1 0 . 1 M(x) i s r e g u l a r . Now u s i n g ( 3 . 1 0 . 2 3 ) w i t h q proof of
(3.10.11,
C o r o l l a r y 3.11.14. graded
t h e formula (3.11.33). If
4,
=
=
V x g ( x ) , one g e t s , a s i n t h e
I
a ( x , h , h D ) , with a(x,h,hS)
symbol a r i s w e l l - d e f i n e d .
shows t h a t
E G"(U),
then i t s
Indeed, t h e c o e f f i c i e n t s of t h e
asymptotic expansion (3.11.24) a r e uniquely determined.
3.11. An Algebra of Difference Operators
339
The n e x t r e s u l t i s concerned w i t h t h e c o n t i n u i t y p r o p e r t i e s of d i f f e r e n c e o p e r a t o r s as mappings between s p a c e s ff Theorem 3.11.15.
v
-v
L e t a E Fv
1t o
(Rn),
v
(s),h("h) * = ( v , , v 2 ) E R2 , and
h l < < > 2 a ( - , h , h S ) E S ( R z ) uniformly w i t h r e s p e c t t o ( h . 5 ) E (O,hol x Rn
Then t h e f a m i l y (3.11.40)
is
5 '
%,
,,(.lf:)
a(x,h,hD) : H(s+V) 2 equicontinuous, V s E R :=
-f
H
.
( s ), h
h E (O,hol,
(<),
P r o o f . W e u s e more o r l e s s t h e same argument a s i n t h e p r o o f of Theorem __
F
After t h e d i s c r e t e Fourier transform
3.4.1.
x+S , h
applied t o z
= A
h
u,
one f i n d s :
(2n)-nJ
-n
where
-h a
h ( 5 ) = Fx,5,hz(x),
ah (5-rl , h ,hn
)
,h Fx+5,ha(x,-,.),
(n)d o ,
u-h
'rl
(5,.,-)
=
Gh(t)
= Fx+5,hu
a r e t h e corresponding d i s c r e t e Fourier transforms. Denote C e = < ( h , h < ) . I n t r o d u c i n g Gh(rl)
one can r e w r i t e
-(sl+vl) = h
2
+v 2 -h
-h u (ri), w
--s
(5)
'
= h
5
>
s2 -h z
(5),
(3.11.41) i n t h e following fashion def
(3.11.42) Gh(h)
s
=
(ylGh)
(5)
=
J
(2n)-"
K(h,c,n)Gh(n)dn
Tn
n th
where
Kh
so t h a t ( 3 . 1 1 . 4 0 ) i s e q u i v a l e n t w i t h
: L2(Ti,h)
+
5 ,h ) uniformly
L2(Tn
with r e s p e c t t o h E [O,hol. S i n c e x + a ( x , . , . ) E S(R:)
,
a E
Fv1 ,o
( E n ), t h e f o l l o w i n g i n e q u a l i t i e s
h o l d w i t h any i n t e g e r N 1 L 0 :
Indeed,
( 3 . 1 1 . 4 4 ) i s a consequence of
(1.3.10) w i t h k
Using ( 3 . 1 1 . 4 4 ) and P e e t r e ' s i n e q u a l i t y << >'<<
S
r
l
>-'
=
0
5 2"'<<
j = N1'
E-n
>, V y E R ,
one f i n d s f o r any i n t e g e r N 2 0 : (3.11.45) where C
lK(h,c,q)l 5 C
N rho
Ntho
<<5-n>
-N
,
V ( h , t , n ) E [O,hol
may depend o n l y on N and h o .
T h e r e f o r e , w i t h any N 2 0 h o l d s :
l R5n x
xn n'
3 . Singular Perturbations on Smooth Manifolds without Boundary
340
Applying t h e i n e q u a l i t y
and t a k i n g N = n + l , one g e t s t h e c o n c l u s i o n t h a t t h e norms o f t h e o p e r a t o r s
K
: L
~ ( T ) ~+ L 2 ( T n
5 rh h by t h e c o n s t a n t
Remark 3.11.16.
5 .h
a r e u n i f o r m l y bounded w i t h r e s p e c t t o h E [O,ho]
)
E F" (Bn) , a(x,h,hE) = am(h,hC)+ n1 ,o E S(IRx) u n i f o r m l y w i t h r e s p e c t t o h
I t i s obvious t h a t i f a
a ' ( x , h , h c ) where h V 1 < p - Y 2 a ' ( * , h , h S ) and 5 , t h e n 9 , = Op a , h
E
(0,h
0
1 i s stillequicontinuousbetweenthesarnespaces:
A s i m i l a r argument shows t h a t i f a
E F"
1.0
(U) t h e n u n i f o r m l y i n h
E (O,ho]
holds : (3.11.46)
a(x,h,hD)
:
H
~
(s+u),h('h)
H(l0C)
( s ) ,h
(Uh),
v
s E R
2
.
Now we c o n s i d e r t h e c o m p o s i t i o n of d i f f e r e n c e o p e r a t o r s . Theorem 3.11.17.
Let
9, =
Gp a ( x , h , h C ) , a E
Fu
1 ,o
(U) and Let Bh
Then f o r each $I E cm(u) t h e composition ch 0
=
= A,
F' (U). 1t o i s a difference
Gp b ( x , h , h C ) , b E 0
I)
0
B~
operator i n Gp F"+'(u)
and for each x such t h a t $I s 1 i n some neighbourhood 1 ,o symbol c ( x , h , h S ) has t h e asymptotic expansion:
of x i t s (3.11.47)
c(x,h,hC)
-
C
l
a
z ( a C a ( x , h , h 5 )( D > ( x , h , h < ) ) ,
/a(?O
where t h e asymptotic r e 2 a t i o n (3.11.47) i s i n t e r p r e t e d in t h e fol2owing sense: f o r each i n t e g e r N > 0 one has: (3.11.48)
with hN
=
rN(x,h,h5)
def = (c(x,h,hS)-
Z / 4 < N
1
=aa
5
a(x,h,hS)D" b ( x , h , h E ) ) E X
V+U-(~,N).
We g i v e h e r e a n h e u r i s t i c argument, which c a n b e e a s i l y made r i g o u r o u s by a p p l y i n g t h e s t a t i o n a r y p h a s e method w i t h p = h - l I 5 I
as a l a r g e parameter
I n d e e d , one h a s f o r e a c h u E C m ( U ) and h s u f f i c i e n t l y s m a l l
0
3.1 1. A n Algebra of Difference Operators
34 1
The complete proof of (3.11.47), that is the justification of i3.11.48), can be done in the same way as the proof of Theorem 3.10.5. It is left to the reader (see also [Fr, 9 , l o ] ) , where the proof is given without use of the stationary phase method). A different commut'ition formula for the symbol c(x,h,h<) of the operator Ah
D
$J
0
Bh may be useful, especially, when the symbol a(x,h,ht)
of Ah is a polynomial in
. Namely,
(<,<*)
holds for each x E U such that
)I
5
the following asymptotic formula
1 in some neighbourhood of x:
*
where iD and iD are difference forward and backward derivatives, x,h x,h respectively. Of course the sum on the right hand side of (3.11.49) is finite, when a is polynomial in ( < , < * I . We shall not give here a complete proof of (3.11.49), leaving it to the reader, and shall only present a formal argument which leads to (3.11.49). -h One has for the difference Fourier transform v ( 5 ) of the function v
= C
u the formula
h
where 5
<(h,hq) and the upper -h stands for the difference Fourier
=
transform of corresponding functions. Obviously, one has: 5,
with
eT
=
=
( e T ,...,eT
Notice, that
1
<;
c,+GT
n
eT
=
<:+e,
-1
*
cq-T,
= exp(ihT ) ,
k
k
so that the Taylor formula yields:
3. Singular Perturbations on Smooth Manifolds without Bounda y
342
where F i s t h e d i f f e r e n c e Fourier transform. x+5 , h Therefore(after t h e inverse Fourier transform F
-1 1, ( 3 . 1 1 . 5 0 ) E+x,h
can
be formally r e w r i t t e n i n t h e following f a s h i o n
t h e sum on t h e r i g h t hand s i d e of t h e l a s t a s y m p t o t i c f o r m u l a b e i n g a s y m p t o t i c a l l y c o n v e r g e n t i n t h e same s e n s e a s i n ( 3 . 1 1 . 4 8 ) . W e d o n ' t p r o v e t h i s l a s t s t a t e m e n t , l e a v i n g i t s proof t o t h e r e a d e r . Example 3.11.18. With a ( x , h , h < )
=
< ,
A
h
= D
x,h'
b ( x , h , h S ) :b ( x ) , t h e f o r m u l a ( 3 . 1 1 . 4 7 )
becomes
g i v e s i n t h i s case
w h i l e (3.11.49)
c(x,h,h5) = b(x)i(h,hS)+D b(x)O(hS). x.h Now t h e c o r r e s p o n d i n g r e s u l t w i l l b e s t a t e d f o r t h e a d j o i n t o f a d i f f e r e n c e o p e r a t o r , t h e p r o o f of t h i s r e s u l t b e i n g i n any r e s p e c t , v e r y s i m i l a r t o t h e one i n t h e c a s e of s i n g u l a r p e r t u r b a t i o n s . With
9, =
Op a ( x , h , h E ) , a E
F"1 ,o (U),
l e t tA
h
be i t s a d j o i n t defined
by t h e r e l a t i o n : (3.11.51)
(
4,u , v I h
= ( u , ~ A ~ v ) ~V, u , v E C i ( U )
( w i t h h s u f f i c i e n t l y s m a l l , where ( family of l a t t i c e s
( u , v ) ~=
,
jh
i s t h e i n n e r p r o d u c t on t h e
g h r
C u(x)v(x)*hn. xERn x,h t The f o l l o w i n g a s y m p t o t i c formula h o l d s f o r t h e symbol a ( x , h , h E ) o f
3.11. An Algebra of Difference Operators
where as usual, a
*
is the complex conjugate of a.
a(x,h,S, S * ) t asymptotic formula for a Again, if a
343
5
then one can also write a different
the latter being more convenient in applications, since, usually, difference approximations of differential operators have symbols which are polynomial in
(<,<*).
Example 3.11.19. With a(x,h,hS)
=
a(x)i(h,hS), 9,
t =
~ ( X ) D ~ one , ~ has ,
9,
*
*
= Dx,h a (x).
Formula (3.11.52) qives in this case
while using the formula (3.11.521, one finds:
Remark 3.11.20. As in the case of singular perturbations, one can use amplitudes ( U X U ) for introducing difference operators. Under the a(x,y,h,hSi F F" 1 ,o assumption that the corresponding operators are properly supported (see
Definition 3.7.3, which can be adequately modified in this case), the (UxU) coincide with the classes Op F" ( U ) , the relation classes Op F" 1 .o 1,o between an amplitude a(x,y,h,hS) E F" (UXU) and the corresponding symbol 180
a(x,h,hE) E F"
(U) being the one given by formula (3.7.12) and the formal 1 ,o symbol of a(x,h,hD) being given by the asymptotic expansion (3.7.13).
Example 3.11.21. Let q
=
q(x,y) E C m ( R
x
a(x.y,h,hE)
=
0
Obviously, a E F 1 , O ( ~x One finds easily
9,
IR) and let
-2 . 2 hc -1 (1+4q(x,y) sin -) 2
.
IR). = Op a(x,y,h,hS). Indeed,
3 . Singular Perturbations on Smooth Manifolds without Boundary
344 where
%(x,y)
=
-1 Fc+(x-y) ,h a(xrYchrh5).
Introducing f3 = exp(ihc), one finds using the residuum calculus:
Applying (3.7.13), one finds an asymptotic expansion of the symbol a(x,h,hE;)of Ah: a(x,h,hS)
1
-
C
N
+o
a aDaa(x,y,h,hS)
I
a! 5 y
Y=x
,
so that
where r(x,y,h,hS) E F0
1 to
(R x
a).
Introducing the difference operator 2
p(x,h,hD) = l+q(x) (2-0-0-') with 0 the shift operator, Ou(x1 = u(x+h), putting q(x,y) = q(y) one finds, applying the commutation formula (3.11.47) that =
Id
+ hR(l)
=
Id
+
h
'
hR(2) h '
where Id is the identity operator and R(1) are difference operators with h their amplitudes in Fo ( R x IR) (thus, RLJ) have their symbols in 1 .o
FY,O(IR)
and under the assumption that q(x)
=
q+q'(x) with q' E S ( R ) , Rh( 3 )
are uniformly with respect to h E (0,h ] bounded mappings from H (IR' ) 0 (s),h x,h into itself, as a consequence of Theorem 3.11.15). The formula (3.11.53) means that p(x,h,hD) and a(x,y,h,hD) with q(x,y) = q(y) are quasi-inverse of each other (inverse up to a small operator for h
+
O), so that using the Neuman's series one can construct -1 -1
the inverse operators p(x,h,hD) h
and a(x,y,h,hD)
when h E (O,hol with
sufficiently small. Of course, the fact that p(x,h,hE;) + Const (1 in
0 this case) when h
+
0, h5
+
0 is the reason that the quasi-inverse of
p(x,h,hD) can be constructed using the symbol
calculus. The same is true
3.11. An Algebra of Difference Operators
345
for the multidimensional difference operator
with q(x) symmetric positive definite matrix, V x E R n , q(x) E Cm(wn) ,
*
(x), q' (x) E S(Rn) , and iDx,h, -iD the difference forward x,h andbackward gradients, respectively.
q(x) = q,+q'
In this case one can take as an amplitude of the quasi-inverse difference operator the same symbol-function a, a(y,h,hS) where 5
=
2
(l+h )-1,
=
* -1 ( 5 l,...,
* k of 5 ) are symbols of D and Dx,h, respectively. x,h Now we are going to consider the group of the linear automorphisms of
<
and the corresponding transformations of symbols of difference operators. Denote by ISO
(q< I ;
the group of linear transformations mapping
onto itself, and by SL(n;Z) the group of nxn matrices, whose entries are integral and the determinant 21. Proposition 3.11.22. SL(n;Z) and I S O ( R ~ ;<) are isomorph.
Proof.
If a E SL(n;Z) then ax
as well, the mapping x
-f
<
E
,
V x E
<.
Given that a-1 E SL(n;Z),
ax is onto Rn. h and let {he,}lcj5n be the standard
Reciprocally, let a E ISO(Ri; <)
<
1
basis on RE, e . being the j-th coordinate vector. 3
One has a(he.) E 3
, Vj,
1 6 j S n and, thus, the matrix ( a ) of a with
respect to the basis {he.} < , < nhas integer entries. So it is also for a-l, J 1=]= -1 E ISO(R~; RE) , too. Therefore, det a and det a-' are both since a
I
integral and necessarily equal to +1. Let
4,
=
F"1 r o (Rn) . is periodic and smooth in n , it can be
Op a(x,h,hC), a(x,h,hE) €
Since a(x,h,n)
:
U
x
(O,ho] x Tn
n
+ C
expanded into a Fourier series
so that using the expansion (3.11.541, one can write
(3.11.55) (%u) (x) =
C
vEZn
a (x,h)u(x+hy), V u E C;(Rn).
3. Singular Perturbations on Smooth Manifolds without Boundary
346
P r o p o s i t i o n 3.11.23.
L e t A,, x
a
=
=
Fv ( R ~.)Then a f t e r t h e change of v a r i a b l e s 1 ,o t -1 E S L ( n ; z ) t h e symbol of A,, becomes a ( a y , h , h ( a ) 5 ) , where
Op a ( x , h , h o , a E
ay w i t h
c1
t .
z-s t h e transpose o f a .
P r o o f . Denote v ( y ) -
(3.11.56)
(\u)
=
(x) =
u ( a y ) and u s e (3.11.54),
Z
(3.11.55). I t g i v e s :
-1 a ( a y , h ) u ( a ( y + h a y))
-1 a ( a y , h ) v ( y + h a y).
C
=
yEz"
yEZn
The symbol of t h e d i f f e r e n c e o p e r a t o r on t h e r i g h t hand s i d e o f
(3.11.56)
is
We a r e g o i n g t o i n t r o d u c e some d i f f e r e n c e a n a l o g u e s of t h e F o u r i e r I n t e g r a l Singular Perturbations. Let
u.
t
3 -
,
RnJ
j = 1,2.
D e f i n i t i o n 3.11.24.
The f u n c t i o n g ( x , y , h , n ) , ( x , y , h , q ) E U 1
X
U2 x R +
x
Tn
n
i s s a i d t o be a
phase f u n c t i o n if t h e f o l l o w i n g c o n d i t i o n s are s a t i s f i e d 1". $J
is r e a l valued, $ ( x , y , h , h E ) E F::;') ( U 1 x U2) u1 x u2 t h e r e e x i s t s a c o n s t a n t cK such t h a t
2O. f o r each c o q u e t s e t K c
A s a consequence o f
h
+
0,
lhEl 2
no
(3.11.57) t h e f u n c t i o n ei'
> 0 f o r some g i v e n
i s r a p i d l y o s c i l l a t i n g when
no.
D e f i n i t i o n 3.11.25.
L e t $ ( x , y , h , h c ) be a phase f u n c t i o n and l e t a ( x , y , h , h E ) be an amplitude in FY,o ( U 1 x u2). The operator
is c a l l e d t h e d i f f e r e n c e Fourier operator. I f a ( x , h , h c ) E F"(U1 ao(x,y,h,hS!
x
U2) ( D e f i n i t i o n 3.11.2) t h e n , a s p r e v i o u s l y ,
s t a n d s f o r t h e c o r r e s p o n d i n g p r i n c i p a l symbol.
Using t h e same argument as i n t h e p r o o f o f Theorem 3.9.3, t h e following
one p r o v e s
3.11. An Algebra of Difference Operators
341
Theorem 3.11.26. h
Let A
@
be a difference Fourier operator. Then uniformly with respect t o
h E [O,hol hoZds
Indeed for establishing (3.11.59) , one has to use, instead of the operator the one defined by the formula:
(3.9.3),
L @ (x,y,h,hS,Dy,h,ag)= =
-i( ID
$ 1 2+1
Y ,h
where Dy,h
5 (h,h5)I
2
/ v 5 $ / ')-'
(
5 $,Vg>),
(DYl,h,...,Dyn,h), iDyj,h being the forward difference
=
derivative with respect to y . .
I
Again, one has in this case
and an integration by part with so that a summation by part in y E Rn Y th respect to 5 lead to a representation of Ah in the form similar to (3.9.5) @ h (A (u))(x) $
=
(Ah
+to
(u))(x) +
Z Ah u) (x), 15j5n $.j'"xjrh
where the amplitudes of the difference Fourier operators A @ , j , 0 5 j S n, FV-(O,l) (U1 x U2). 1 ,o
belong to
Example 3.11.27. Consider the following difference Fourier operators
The functions
solve the Cauchy problem:
3. Singular Perturbations on Smooth Manifolds without Boundary
348 where
Ah
-
=
* -2 * C D D = h 1 (0).-2+0.) 1s jsn xj f h xj ph lsjsn J
'
is the difference Laplace operator. h The distributional kernels of E+(t) are -
(3.11.60) E+(h;x,y,t) -
=
( 2 1 ~ h ) - ~eih-l(<x-y'n>'t~u(n)l)dn, /
t 2 0,
Tn 17
where, as previously, w ( n )
=
L(1,n).
One can apply the stationary phase method for computing the asymptotic expansion of E+ when h
0.
-f
-
An
easy computation leads to the following equation for the stationary points
on T ~ :
n
lw(q)/-lRe w ( n )
=
T t-'(x-y)
so that for Ix-yl > t there are no stationary points and the singular 2n+1 support of E+ is contained in the cone = { (x,y,t) E R + , Ix-yl 5 t}. -
One can show that in fact,sing supp E+ -
=
V, while for the corresponding
distributional kernels of the wave operation (h aV = {(x,y,t) E R 2n+l j jx-yj = tI.
=
0 ) the singular support is
precisely
Let A be a symmetric positive definite nxn matrix and let * l
A (3.11.61) E+(h;x,y,t)
-
=
(21~h)-"/ e Tn
ih-I (<x-y,n>+t
t 2 0.
One checks easily that (3.11.62) s i n g sup
Ed
=
i(x,y,t) E IRINC1
I
A Of course, E+ are solutions of the Cauchy problems: -
2
2
<,
*
+
>)Eh(h;x,y,t) = 0, x E t > 0 x ,hIDx,h at aE; L Ei(h;x,y,O) = 6 (x-y), -(h;x,y,O) = f OP < A < , < > 2 6h(X-Y). h at (-
2
Example 3.11.28. Let w ( n ) ,
n E
T" be as previously defined and let
n
2 2 (3.11.63) B ( p ) = 1-(p /2)+ip(l-p /4)',
2 p 2 = p2(n) = r <w(n),w*(n)>,
where the parameter r satisfies the (hyperbolicity) condition:
-t (3.11.64) 0 5 r 5 n
.
3.11. A n Algebra of Difference Operators A s a consequence o f
v
ri
349
( 3 . 1 1 . 6 4 ) , one h a s p 2 5 4 , so t h a t I f 3 ( p ) f
1
=
1,
E T ~ v, r E [ o , n t I . rl
Consider t h e f o l l o w i n g d i f f e r e n c e F o u r i e r o p e r a t o r s :
where (3.11.66)
++- ( x , y , t ; h , h S )
=
-1
<x-y,C>
I n f3 ( p (hg)1 .
f t (irh)
h
h
-
-
I t i s e a s i l y s e e n t h a t t h e f u n c t i o n s v + ( x , t ) = ( E + ( t ) u )( x ) s o l v e t h e
f o l l o w i n g Cauchy problems
where, a s p r e v i o u s l y , IR+
=
+
TZ ,
Z
+
=
{k E Z
t , T
I
k > 01.
One c h e c k s e a s i l y u s i n g t h e s t a t i o n a r y p h a s e method, t h a t a g a i n i n t h i s
case t h e s i n g u l a r s u p p o r t of t h e c o r r e s p o n d i n g k e r n e l s of t h e o p e r a t o r s
E:(t), -
i s contained i n
(3.11.68)
7
=
( a c t u a l l y , c o i n c i d e s with) t h e cone:
2n+l { ( x , y , t ) E iR+
1
Ix-yl
5 t}.
N o t i c e t h a t when r + 0 one g e t s a g a i n t h e d i f f e r e n c e F o u r i e r o p e r a t o r s h E + ( t ) from Example 3.11.27.
Eh ( t ) t h e d i f f e r e n c e F o u r i e r o p e r a t o r s d e f i n e d by ( 3 . 1 1 . 6 3 ) , f ,A (3.11.66) with
Denote by (3.11.65), (3.11.69)
p2 = p
2 A
(ri)
=
2
r
where A i s a symmetric p o s i t i v e d e f i n i t e nxn m a t r i x and r s a t i s f i e s t h e (hyperbolicity) condition: (3.11.70)
o
5 r 5
(nj (A1
I)-+.
The c o r r e s p o n d i n g f u n c t i o n s vh
f ,A
of t h e d i f f e r e n c e equation r e p l a c e d by
*
(x,t)
=
(Eh f
,+A
( t ) u )( x ) a r e s o l u t i o n s
where
x ,h ‘Dx,h>. Our c o n j e c t u r e i s t h a t t h e s i n g u l a r s u p p o r t of t h e k e r n e l s
E+ A ( h ; x , y , t ) of - I
the operators
Eh+,A ( t ) c o i n c i d e s w i t h t h e cone ( 3 . 1 1 . 6 2 ) .
is
3 . Singular Perturbations on Smooth Manifolds without Boundary
350
If n = 1 , r = 1 , one f i n d s an e x p l i c i t formula f o r t h e k e r n e l s
E+(h;x, y, t) of the d i fferen ce Fourier opeators
-
(3.11.65), (3.11.66).
h
E -+ ( t ) d e f i n e d by ( 3 . 1 1 . 6 3 1 ,
Indeed, i n t h i s c a s e O + ( p ( q ) ) = c o s q f i l s i n q l , -
SO
t h a t a n e a s y c o m p u t a t i o n shows t h a t (3.11.71) E + ( h , x , y , t )
=
(ZTih)-'
;(e ,.
-ih
-1
(x-y-t)n+eih
-1
(x-y+t)q
)
dn
and t h e same formula ( u p t o t h e s i g n ) h o l d s f o r E - ( h , x , y , t ) . Of c o u r s e , i n t h i s case t h e c o r r e s p o n d i n g d i f f e r e n c e F o u r i e r o p e r a t o r c a n b e r e w r i t t e n a s a d i s c r e t e c o n v o l u t i o n o p e r a t o r i n t h e form:
w i t h E + ( h , x , y , t ) g i v e n by ( 3 . 1 1 . 7 1 ) . AS a consequence o f
h
-t
( 3 . 1 1 . 7 1 ) , E + ( h , x , y , t ) converges i n D ' ( I R )
when
0 t o the distribution
t h e l a t t e r b e i n g t h e s o l u t i o n of t h e Cauchy problem
I
= Op(lSI) i s t h e s i n g u l a r p e r t u r b a t i o n ( i n f a c t , t h e where o f c o u r s e ID p s e u d o - d i f f e r e n t i a l operator) w i t h t h e symbol 151 E S ( 0 , l r O ) ( R ) .
W e a r e g o i n g t o r e s t r i c t t h e c l a s s of p h a s e f u n c t i o n s and s h a l l
c o n s i d e r , from now on t h e p h a s e f u n c t i o n s # ( x , y , h , h c ) which s a t i s f y t h e following C o n d i t i o n 3.11.29.
There e x i s t s a c o n s t a n t 6 ( h , n ) E IR+
x Tn
> 0
w i t h I < ( h , q )I
such t h a t for any ( x , y ) E u1 2 6
x
u2 and any
holds:
-1 (3.11.72) $ ( x , y , h , q ) = h $ ( x , y , l , n ) .
For such phase f u n c t i o n s t h e c o n d i t i o n (3.11.57) w i l l be s t a t e d a s foZlows:
3.11. An Algebra of Difference Operators
35 1
On t h e o t h e r hand, we s h a l l e x t e n d t h e c l a s s o f t h e o p e r a t o r s c o n s i d e r e d i n t h e f o l l o w i n g way. We s h a l l d e n o t e by
Kh t h e c l a s s o f o p e r a t o r s o f t h e form
where t h e f a m i l y o f k e r n e l f u n c t i o n s h s e t in Cm(U1
x
U2)
E
for h
: Cm(U
4
0
2
K ( h , x , y ) b e l o n g s t o a bounded
[O,hol w i t h a g i v e n h
We s h a l l u s e t h e n o t a t i o n l i n e a r mappings Ah
+
F"
m l,o
+ C
0' f o r t h e f a m i l i e s of continuous
(4)
(U1),
v
h C ( 0 , h 1, which c a n b e 0 4 satisfying
r e p r e s e n t e d i n t h e form ( 3 . 1 1 . 5 8 ) w i t h t h e p h a s e f u n c t i o n C o n d i t i o n 3.11.29,
i.e.
a ( x , y , h , h S ) E Fy,,CU,
x
(3.11.72),
(3.11.731, and with an amplitude
U2) (mod K h ) .
h One can a s s o c i a t e w i t h a d i f f e r e n c e F o u r i e r o p e r a t o r A + E -V
its distribution kernel h
FY,O(d)
'A4 ( h , x , y ) , where
V ( x , y , h ) E U1
x
U2
(O,hol.
X
W e are g o i n g t o l o c a l i z e t h e s i n g u l a r i t i e s o f A ( h , x , y ) .
d
F i r s t , introduce t h e set
where t h e u p p e r b a r , a s u s u a l d e n o t e s t h e c l o s u r e o f t h e c o r r e s p o n d i n g s e t . I n t h e same way, a s p r e v i o u s l y f o r Theorem 3 . 9 . 7 ,
one p r o v e s t h e
following Theorem 3.11.30.
The s i n g u l a r support o f t h e f a m i l y h
+
A (h,x,y)
0
of t h e d i s t r i b u t i o n a l
k e r n e l s d e f i n e d by ( 3 . 1 1 . 7 5 ) i s contained i n t h e s e t (3.11.76), (3.11.78)
(3.11.77): s i n g supp A ( h , x , y )
d
5
Q4.
Q
4
d e f i n e d by
3. Singular Perturbations on Smooth Manifolds without Boundary
352
Now t h e c l a s s of H y p e r b o l i c D i f f e r e n c e O p e r a t o r s w i l l b e i n t r o d u c e d and t h e c o r r e s p o n d i n g F o u r i e r D i f f e r e n c e O p e r a t o r s w i l l b e c o n s i d e r e d f o r Hyperbolic D i f f e r e n c e Operators with c o n s t a n t c o e f f i c i e n t s . One d i s t i n g u i s h e d ( t i m e - l i k e ) v a r i a b l e i s g o i n g t o p l a y a s p e c i a l n+l - R n X R and c o n s i d e r g r i d s x,t x + = (hZn) x ( T Z ) w i t h two mesh-sizes h and T = r h , where r > 0 i s a
r o l e . Hence, w e s h a l l d e n o t e W IRE::
g i v e n c o n s t a n t . The d u a l v a r i a b l e s w i l l b e d e n o t e d by ( S , E ; ) 0
A s usual, 0
E
Rn x R
5
50
and 0-1 a r e , r e s p e c t i v e l y , t h e f o r w a r d and backward s h i f t
o p e r a t o r s on t h e g r i d R
while 8
= TZ,
t , T
symbols. F u r t h e r , D
=
(iT)-'
t t T
-1
( T S ~ ) ,B0 ( T S0 )
*o ( e T - l ) , Dt,-,
-1
= (iT)
stand f o r t h e i r are
(I-@-')
( m u l t i p l i e d by -i) forward and backward d i f f e r e n c e d e r i v a t i v e s and
A(T,TS
*
-
(T,?< ) d e n o t e t h e i r r e s p e c t i v e symbols. A s u s u a l , 0 ORn x n+l = with R = { t E R t > 0 ) and x t,+ t ,+ n+ 1 n+l w i t h R+ t h e c l o s u r e o f R ),
*;+I=
I
%,-,,+
~
.
D e f i n i t i o n 3.11.31. n+ 1
A symbol a ( x , t , h , h E , ~ < ~E ) Fv(P(+
)
,
is s a i d t o be s t r i c t l y
v = (vl,v2)
hyperbolic i f t h e following conditions are s a t i s f i e d :
i s an i n t e g e r ;
1". v2 > 0 2'.
a can be r e p r e s e n t e d i n t h e form:
(3.11.79) a ( x , t , h , h S , T S O ) = h
'1
*
-k
( ~ E ; ~ ) p ( x , t , h , X , C) ,, <
O0
(0,v2)
where k 2 0 i s some i n t e g e r and t h e symbol p E F
(R:+l)
i s polynomial
i n A o f degree v 2 , 3 " . w i t h p o ( x , t , h , X , C , < * ) t h e p r i n c i p a l symbol o f p , t h e z e r o s 1 5 j 5 v2, (3.11.80)
of t h e equation
*
po(x,t,l,~,w,w)
a r e aZl d i s t i n c t f o r (3.11.81)
u 1. ( x , t , ~ ) ,
=
0, w
IE T>{O},
11+irU.(x,t,q)l = I , 7
=
(ul,
(x,t) E
v
...,w n ) ,
= -i(e
w
iqk
-l),
7 -
and, moreover,
( x , t , q ) E x;+l
x T ~ .
II
D i f f e r e n c e o p e r a t o r s w i t h s t r i c t l y h y p e r b o l i c symbols a r e s a i d t o be s t r i c t l y hyperbolic. It i s e a s i l y seen t h a t t h e o p e r a t o r
s t r i c t l y hyperbolic with v
=
(O,l), n
=
(3.8.117)
i n Example 3 . 8 . 2 3 i s
1.
F u r t h e r , t h e d i f f e r e n c e a p p r o x i m a t i o n ( 3 . 1 1 . 6 7 ) of t h e wave o p e r a t o r A -2:
i s s t r i c t l y h y p e r b o l i c w i t h v = ( 0 , 2 ) , p r o v i d e d t h a t (3.11.64)
is
3.11. An Algebra of 3ifference Operators
*
s a t i s f i e d a s w e l l a s Dt , T
353
*
D t,T -
d e f i n i t e matrix i s a s t r i c t l y h y p e r b o l i c d i f f e r e n c e approximation of t h e wave o p e r a t o r D 2
t
=
x
x
We s h a l l c o n s i d e r t h e Cauchy problem f o r s t r i c t l y h y p e r b o l i c d i f f e r e n c e o p e r a t o r s w i t h symbols which do n o t depend on x , t and s u c h t h a t i n t h e r e p r e s e n t a t i o n ( 3 . 1 1 . 7 9 ) one h a s : v 1 p(x,t,h,X,<,<*)
po(h,h,c,<*) = Am
supposed t o b e p o l y n o m i a l i n
*
=
0, k
=
0 , v2
m > 0,
=
+ terms o f lower o r d e r i n
<,< , a s
A,
po b e i n g
well.
Hence, c o n s i d e r t h e Cauchy problem:
m
where v E C (R) i s g i v e n .
0
Denote a g a i n by
u.(n)
(3.11.83) po(l,u,w,w*)
3
=
the zeros of the equation
0 , w = (wl
,...,w n ) ,
I t i s e a s i l y s e e n t h a t t h e s o l u t i o n of
in mk = - i ( e
k-l).
( 3 . 1 1 . 8 2 ) i s g i v e n by t h e f o r m u l a
C L: a ,(h,hg)ei~l'x~yft'h'hS)v(y)hndg, (3.11.84) u ( x , t ) = J h T~ y ~ m f : l<j<m 1 ht5 where
(3.11.85) $ . ( x , y , t , h , h S )
l
=
-1 <x-y,S>+t(Ti) l n ( l + i u .( h S ) ) , 3
and where
I n d e e d , w i t h $ . d e f i n e d by ( 3 . 1 1 . 8 3 ) , one f i n d s u s i n g t h e f a c t t h a t -1 po i s homogeneous i n ( h ,g) o f o r d e r m:
since
u , (n) s a t i s f y ( 3 . 1 1 . 8 3 ) . 1 Hence, t h e f u n c t i o n ( 3 . 1 1 . 8 4 ) i s a s o l u t i o n o f t h e d i f f e r e n c e e q u a t i o n
3. Singular Perturbations on Smooth Manifolds without Bounda y
354 (3.11.82).
F u r t h e r , one h a s f o r any polynomial P ( A ) o f d e g r e e m w i t h s i m p l e z e r o s A L and w i t h l e a d i n g c o e f f i c i e n t 1 t h e f o l l o w i n g f o r m u l a e :
where y i s a c o n t o u r i n C p'(A)
=
which e n c l o s e s t h e z e r o s h l ,
...,A rn
of p(X) and
dp(A)/dX.
Using a g a i n t h e f a c t t h a t w i t h $ , d e f i n e d by ( 3 . 1 1 . 8 3 ) one h a s :
I
and a p p l y i n g (3.11.871, (3.11.84)
one g e t s t h e c o n c l u s i o n t h a t u (x,t) d e f i n e d by h s a t i s f i e s t h e i n i t i a l c o n d i t i o n s i n ( 3 . 1 1 . 8 2 ) , a s w e l l , so t h a t
it i s t h e s o l u t i o n o f t h e Cauchy problem ( 3 . 1 1 . 8 2 ) .
Using t h e homogeneity of a . ( o f o r d e r m-1) and 4 , ( o f o r d e r 1 ) i n 3 I -1 ( h , < ) , one c a n r e w r i t e ( 3 . 8 . 8 4 ) i n t h e f o l l o w i n g f a s h i o n : (3.11.88) u ( x , t ) = h
C lsjsm
(A
h
v)(x),
'j
where t h e d i f f e r e n c e F o u r i e r o p e r a t o r Ah function j '
has as i t s k e r n e l t h e following
One h a s f o r e a c h t
where
AS a consequence of t h e h y p e r b o l i c i t y c o n d i t i o n t h e v e c t o r
is real-valued. and l e t K =
L e t K . b e t h e c l o s u r e of t h e range of F . 1 I
U K.. 16j<m J
(n)
: Tn\{O}
n
+
Rn
3.1 1. An Algebra of Difference Operators
355
Introduce K* = {(x,y,t)
E IRn
En
x
X
1
R+
t-'(x-y)
E K},
so that K*
Q
U
=
(t).
tzo,j j' As a consequence of Theorem 3.11.30, the singular support of the kernel of
the operator, which solves the Cauchy problem (3.11.821, belongs to K*. Obviously, K* is a conic set in Rn
x
Rn
x
R+
.
Assume that p (h,D 0 t,T ,Dx,h,Di,h) is a difference approximation of a hyperbolic differential operator pO(Dt,Dx) which is necessarily homogeneous in (D D ) of order m. Moreover, assume that the set po(So,<) = 0, 5 E l R n , t' x 151 = 1, consists of a finite number of ovals containing each other. In that case the smallest oval V is always convex (see [John, 11 and also for the sufficient conditions for such a situation, which are that m and n to be even and po(O,E) # 0 , V 5 E R n , 161 = 1 ) . Let K be the cone in
Rn+' having the smallest oval V as its base. Let V* be the dual cone of V (the cone of rays). We conjecture, that the singular support of the kernel of the operator which solves the Cauchy problem (3.11.81) for po (h,Dt,T,Dx,h,D:,h)
in this case is always contained in the closure of
the interior of the dual cone V*. This is precisely the case for the hyperbolic approximation (3.11.67) of the wave operator, as discussed in Example 3.11.28. Further,we consider one class of hyperbolic difference operators with variable symbols and investigate the asymptotic behaviour of their oscillatory solutions when the mesh-size goes to zero. Let lR;+l
= {x =
(xo,x')
I
xo E R+ , x' E Rn-'1 and let
R,,+ n+l
be the
with the mesh-size hk = rkh, h E W+ with rk > 0 , 0 5 k 5 n,
greed in
given positive constants. Further, let Vn = { ( h , C ' ) E Let X(x,h,<') 1".
X E
E + X
n+1
: R+
x
Vn
Cn +
1
Il+ih 5 1 = 1, 1 2 k 2 n}. k k
C have the following properties:
m
C
in variables (x,h,C') for
<' #
0;
Z 0 . X(x,h,<') s X(-,h,<') for 1x1 sufficiently large;
3'.
X(x,t-'h,K')
5
--n+l
th(x,h,<'), V t > 0, V (x,h,<') E R +
4 " . h(x,l,o') with w '
= (wl
satisfies the conditions:
,...,w n ) ,
l+irkwk = exp(ir 5 1 , k k
x
Vn;
1 5 k 5 n,
3 . Singulav Perturbations on Smooth Manifolds without Boundary
356
and
These conditions mean that def A(x,h,<) = G o + $6(15'1)h(x.h,5') with C 0 = ( i h o )
-1 (exp(ihOcO)-l)and $,(t)
above ( $ 6 ( tZ) 0 for t
E
[0,6],
the cut-off function introduced
$&(t) E 1 for t t 26) is a hyperbolic
symbol in F(OS1) (my+1),whose principal symbol is A 0(x,h,<)=
0 2 k 5 n, and using 3 O ,
in the form:
-1 AO(x,h,<) E h o(x,S), 5 = ( c o , 5 ' ) , rkSk = nk, Oskin, rl E T;+l. Consider the Cauchy problem on the greed
("(xA
I
Dxo,h,DX , , ) uh ( X) = 0 ,
-1
uh (0,~') = f(x') exp(ih
where f E
m
c0
g:: x E Iqt
$o(x')),
(Rn)and $o E Cm(Rn), $q : Rn
+
R , are given.
We seek uh as an asymptotic series:
-1 k uh(xo,x') = exp(ih $(xo,x')) C h pk(xo,x') k>O with Q(xq,x') and pk(xo,x) t o be determined as smooth functions of (xo,x') n+ 1 in R . Substitution of uh(xo,x') given as a formal asymptotic series above and the use of Theorem 3.11.13 yields the relations:
4
and pk being subject to the initial conditions: $(O,X') = $,(x),
po(O,x') = f(x'), pk(O,x') = 0, k = 1,2,...i
here o(x,<) = hA 0 as defined above, the coefficients Ck (x) are expressed
3.11. A n Algebra of Difference Operators recursively in terms of u and $, Lk variables x', L-l
=
357
are differential operators in
0.
A s a consequence of the hyperbolicity condition, the Cauchy problem for
with $o (x') real-valued, has a well-defined smooth solution for
$ (xo, X I )
xo E ( 0 , 6 ) , x' E Rn with 6 > 0 sufficiently small. The same hyperbolicity condition guarantees that -1 (x,Vx$)) , 1 2 k S n, are real-valued, V x E 10,s)
x Rn , as 50 ' k it will be shown below, so that the Cauchy problem for pk, k 2 0, is well0
(x,Vx$)( U
posed. It can be shown (but will be not done here) that the formal asymptotic m
expansion thus obtained,is asymptotically convergent in C (K) with -n+ 1 K C W+ any compact. Hence, one can determine successively pk(xo,x') for (xo,x')F [0,6)X Rn. We shall discuss a little more in detail the non-linear Cauchy problem for $(xo,x') stated as follows: U(X,V
$1 = 0, $(O,X') = $o(x').
One finds differentiating the equation for $ with respect to xk, 0 5 k 6 n: 0 =
v
u(x,V$) = [ $
v
xx ll
I
u(x,ll) + v x u ( x , l l ~ l
ll=Vy$
is the matrix of second derivatives of 9. x I'O
where $
xx
=
I l ' x
k j
0 0
A s a consequence of the hyperbolicity, the equation u(x,So,S')=O has the
real zero <,(x,S')
(well-defined up to an additive term 2rLri1, L E Z), so
that differentiating this equation with respect to that
I?5
-'
k ~ c o=-
-aco/ac,
5,.
1 4 k 5 n, one finds
are real-valued. Hence, the system of ordinary
differential equations
with the initial condition x(0) = ( O , s ) , s = (sl,. .. ,sn) E Rn defines the n+ 1 , passing through the point unique curve without singularities in R (0,s)
With S(t) = Vx$(x(t)), x(t) being the curve defined above, one finds
3. Singular Perturbations on Smooth Manifolds ioithout Boundary
358
'Th-refore,x(t),E,(t)
is the solution of the following Hamiltonian type
system:
and along the trajectory x(t),<(t) one has:
Taking the initial condition for x(t) and $(t) in the form x(0) = ( 0 , s ) E Rn+l, $'
0
= $'o(s),
i t is readily seen that the initial conditions for C ( t ) are
0 0
where 5 ( s ) is the zero of the equation
The Hamiltonian type system for x(t),S(t) along with the equation for $ and the initial conditions above, define the curve x = x(t,s),
in iR
2n+2
5
=
S(t,s)
and the function $(t,s) :
The equations x = (xo,x') = x(t,s), $I = @(t,s) give a parametric representation for $. One can eliminate (t,s) by solving the equation x(t,s) = x and find 0 as a function of x, provided that the Jacobian J = D(x0,x')/D(t,s) is non-singular.
It is readily seen that = 1, det J (t=O = det D(x')/D(s)~~=~
so that for t sufficiently small det J # 0 and one can determine $ as a
3.11. An Algebra of Difference Operators function of x by eliminating (t,s). Therefore, for t E [ 0 , 6 ) with 6 sufficiently small the asymptotic series for the solution uh of the Cauchy problem above is well-defined. For the first coefficient po(xo,x) in the asymptotic expansion for
uh one has the following equation (cf Theorem 3.11.13):
Introducing the vector field v tangential to the solutions of the system of ordinary differential equations (the bicharacteristics of the equation for p ) 0
:
dx /ds = 1, dx'/ds = ( u 0
50
(x,V$))-~V ,u(x,Vd),
5
one can rewrite the equation for po in the form:
Using the Liouville's theorem for first order autonomous systems of n+ 1 ordinary differential equations in R
which says that the determinant of the Jacobian
satisfies the differential equation: (d/ds) In det J = div h(x), one finds for Fo:
Further, given that a-1 50
=
exp(-iroEO), one gets
359
3. Singular Perturbations on Smooth Manifolds without Boundary
360
=
given
-ir
exp(-ir @ ) u 5 (x,V$) = ir u exp(-ir$ ) , OXO k O xo xO
C @ 06k6n xO%
that the differentiation of the equation u(x,V@) = 0 yields +
0
xO
z u (x,vQ)@x = 0 OSkSn k ' k O
Therefore, one finds :
Hence, one gets for Po with some constant C:
If SP(UxS) + ox
= 0,
0
then po = C(det J)-'
with some constant C z 0 (see [Mas,
1
1,
where this
kind of computation is done for pseudodifferential equations). This argument can be easily extended to higher order strict hyperbolic difference operators which are polynomial in D
xo
or to first Ih0
Id + A(x,h,D' with A(x,h,S') a pxp order systems of the form D xo rho x',h) matrix-symbol in F(O'l) (IRy+l), whose principal symbol AO(x,l,w') has real eigenvalues A , (x,1 ,w 7
')
such that
[w'~-l/A.(x,l,w')-~ 9. (x,l,w'P 2 6
n+ 1 with some 6 > 0 , V x E IR+
3
,...,w
V w ' = (wl
),
l+ir w =exp(ir 5 k k k k
,
# 1.
Consider several specific examples. With u
=
exp(i<0)-exp(i51), which corresponds to the approximation of the
-u = 0 by the finite difference equation uh (xO+h,x1 x1 0, one finds for the solution uh of this equation with the -1 initial condition u (0,x ) = exp(ih Q (x ) ) , the following equation for h 1 0 1 Q(xo,x): equation u
xo uh(xO,xl+h) =
exp(i$
)-exp(i$x xO
)
= 0,
2
x E R+
1
with the initial condition $(O,x1 ) = @ 0 (x11 . For the first coefficient Po(xo,xl) in the asymptotic expansion of u h'
3.1 1 . A n Algebra of Difference Operators
36 1
one has the Cauchy problem:
One finds easily: $(x) = $ (x +x ) + 2nkxO, 0 (x) = f(xo+xl), 0 0 1 0 2
so that on the greed IF$,
one has: r+
uh(x) = exp(ih
-1
$ (x +x ) ) (f(x +x )+O(h)),
0 0 1
0 1
h
+
0.
In fact, in the case considered O(h) : 0 . Now consider the following Cauchy problem on
( ( f ) (l+r)+(l-r)Oxl,h)Dx ,h -D h)\=O, 0 0 xl' where ho = rh, hl
=
2
%,+ :
uh (0,x1 ) =exp(ih-l$O(xl))f(xl),
h.
One has: u
=
-1 (2r) ((l+r)+(l-r)exp(iS1))(exp(irSO)-l)-(exp(i~l)-l),
and the following equation for $:
tJX
=
0
-1 r arg((p+exp(itJx ) ) (l+p exp(i4x ))-l), 1 1
p
=
-1 (1-r)(l+r)
where arg z is the argument of z E C-{01. Let $(O,x ) = wxl with some w E IR 1
. Then
+(x) =y(w)xO+uxl, y ( w ) =r-l arg((p+exp(io)) (l+p exp(iw))-'). Further, one finds easily that
(u
-1
u
)
(V$)
=
I-'.
-41 (l+r)exp(i$x )+(l-r)
50 51
1
Hence, with $ = y(w)xo+wxl as above, one gets for po the equation (polx - a(@) (polx = 0 , 0 1
a(w)
=
41 (l+r)exp(iw)+(l+r)
so that Po(x) = f(a(w)xo+xl) and u h (x) = exp(ih-'(y(w)x0+ox1)) (f(a(w)xo+xl)+O(h)),
h -+ 0 .
The argument above can be extended to higher order hyperbolic operators.
,
362
3. Singular Perturbations on Smooth Manifolds without Boundary
Consider as an illustration, the wave equation u
-u = 0 and its x x 1 1 finite difference approximation (D D* -D = 0, on the xO,ho xo,h0 x1 ,hDxl,h)% x x
oo*
greed
2
I$,+ in Rf
with the meshsize ho = rh, hl = h, the latter being
hyperbolic provided that r E (0,l) One finds easily:
u
=
2 2 4(r-2sin (rc0/2)-sin (c1/2)) = o+o-,
where
u'(~)
= 2(r-lsin(r~~/2)3sin(51/2)).
Consider for instance, the asymptotic behaviour of the wave associated with
. ' 0
One finds for the phase function @ the following equation: -1 r sin(r9 /2)-~in($~/2) = 0. xO 1
Again assume that $(O,x
1
) =
axl, I w / < a / 2 . Then
~(x) = y(w)xo+wxl, y ( w ) = 2r-larc sin(r sin(w/2)). For po one gets in the same way as hereabove: po(x) = f(y'(a)xO+xl),
y'(w)
= dy/dw,
so that for the wave moving to the right one has:
uh (x)
=
-1
exp(ih
(y(w)xo+wxl)f(y'(w)x0+x1) .
As another illustrative example, consider a hyperbolic difference approximation of the system atu+pglaxp = 0 ,
atp+p o o2 a xu
= 0,
(x,t) E
E x
R+ =
2
n+ ,
which describes the propagation of plane acoustic waves of small amplitudes in a medium
at rest, u(x,t) being the speed of the propagation, p(x,t) the
presure, the constants
p o > 0,
co > 0 being the density and the
compressibility characteristic of the medium, respectively; here, as previously a = a/at, ax = a/ax. 2 t Let %,+ be the greed in =
z:
(h,T),
T =
(x,t) E R
X
z+1
with the meshsize
rh, r > 0 being a given constant.
Consider the following finite difference approximation of the system 2 above on the greed B,,+ :
3.11. An Algebra of Difference Operators
363
where, as previously, 0 is the shift operator in the greed h I$, = {x E IR I x =kh, k E Z } , (0 v) (x) = v(x+h), and iDt,,, iDx,h are h forward finite difference derivatives in t and x, respectively, on the 2
greed ph,+ * Multiplying the second equation by ( p c
-1
0 0
,
adding it to and after-
wards subtracting it from the first one, one gets the following finite 2 + -1 difference equations on If$,+for the Riemann'sinvariants u- = u+(poco) p of the differential system above:
where we have denoted: L2(rc0,Oh,Dt,, ,Dx,h) = if ( 4 ) (l?rco)+(lfrco)@h)Dt,TfcoDx,h}, the latter being hyperbolic finite difference approximations of the operators at+coax. If the initial conditions for u and p are highly oscillatory functions 0)
of the form f (x)exp(ih-l$o(x)) with $o E C ( I R ) , fo(x) = p 0(x) for
o
m
f (x) = u (x) in CO(R 1 , then the initial conditions for the Riemann's 0 O + (x)). invariants u- will be of the same form u:(x)exp(ih-'$ 0 Seeking the asymptotic expansion of u'(xlt)
one gets for $'(x,t)
as h
+
0 in the form
the following Hamilton-Jacobi equations:
+
with the initial conditions $-(x,O) = $ (x). 0
The transport equations for u'(x,t) k as above.
are derived in the same way
Finally, for the multidimensional wave equation U
x0x0
-
16k,j(n
akjUx x k j
with positive definite matrix A approximation
=
= o
I [ akj 1 I
and for its finite difference
3. Singular Perturbations on Smooth Manifolds without Boundary
364
1 akjDxk,hDx,,h)uh=O,h0 =rh,h1=...=h =h, h D* x ,h (DxO' 0 0 0 l
Again consider the wave associated with o+ = 2(r-' sin(r~~/2)-p(~')).
If f (x') = (planewave), then p (x) = f (
uh = exp(ih-l(<w,x'>+y(w)x
)
0
(f(<X,x'+xoVwy(w))+O(h)).
We finish this section by introducing some classes of CiEference singular perturbations affected by the pr-eseme OF two small uarameters
E
and h.
Definition 3.11.32 A f u n c t i o n a(x,E,h,n)
u
x
( 0 . ~ ~x 1 (0,h 1 0
i s said t o be a symbol in (vl,v2,v3) E n3 , i f for each compact x
Tn
rl
( u ) , with vo E m , v = 1,O u and each multi-indices ci,B there e x i s t s a constant c
the class F K c
:
V0,V
a,B ,K
such t h a t
holds:
V (x,E,h,S) E K
If v0 = 0 we shall write
Fv1 r 0 (U) instead of
X
( 0 . ~ ~ X 1 (O,hol X Rn
Forv(U).
1r o Analogously to Definitions3.11.2,3.11.4oneintroduces
symbolclasses
FvO'"(U) and Gvorv(U),as well as the graded symbols in the latter class.
.
3.1 1. A n Algebra of Difference Operators
365
Examples 3.11.33.
1. Let
U C
W
and let q E C m ( u ) , q(x) 2 qo > 0. With -2
-p
a1(x,€,h,n) = (1-e and
)
-2
a2(x,E,h,n) = 1+4p
11-e
p =
h/E the functions
in-(p/q(x)) 2
I
2 q (x)sin2 n/2
vo,v
VO,"
are symbols in F
(U)
(U)) with v,
(and even in G
=
0, v = (0,0,2);
besides, they coincide with their graduate symbols,which contain in this case just one term. Both symbols al and a2 are approximations (when h + 0) of the symbol 2 2 1 + ~q ( x ) C 2 in the classes s ( o r o ' 2 ) (U) and K ( o p o r 2 ) (U) defined in Section 3.6. 2
Vo,V
I < I 4+1 < /
2 . The symbol E
belongs to the class G
v = (0.0.2). It approximates (when h
+
(U)
,
U
C
Rn , with v
0
=0,
0 ) the symbol E ~ ~ S ! ~ + in ! S the ! ~
class K(or2r2)(U) in Section 3.6. The symbols Q ( x , E , ~ , ~ ) 0 ,= ( 0 l,.. . ,en), Bk = exp(ihSk) which are -1 polynomial in 8 k, Bk , k = 1,2 ,...,n, Q E F v o r v ( U ) with v 2 , v 3 nonnegative integer, will be of special interest in Chapter 4 where boundary value problems for elliptic difference singular perturbations will be studied. The difference singular perturbations with symbols in corresponding F v o r v ( U ) and GVo"(U)
classes F V o " ( U ) ,
are defined as in Definition 3.11.8.
The proof of the following statement is just a copy of the one in Theorem 3.11.15.
-
v ,v
vorv
( W n ) the subclass of symbols in F
Denote by F 0
f o r each symbol a(x,~,h,hS)E Fvo
1v
(Wn) such that
(Rn) there exists a symbol a_(e,h,hE)
such that a-a-, as a function of x E lRn , belongs to S(lRn) Proposition 3.11.34.
Let
Ah,,
=
vo,v
op a(x,h,~,hE)w h e r e a E F
1 to
.
-
(w"). T h e n u n i f o r m l y w i t h r e s p e c t
to h E (O,hol, E E (O,ho] h o l d s : (3.11.91)
V O% , E :
Here the spaces H
(Rt)
H(s+v) , ~ , h
+ H(s) ,E,h( < I .
are the ones introduced in Definition 2.8.1.
(s),E.h (<)
The following statement can be proved using the same argument as in the proof of Theorem 3.11.5. Theorem 3.11.35.
.(j)
L e t a.(x,E,h,hS) E F 3
where v i 1 ) + v i J )
J.
180
-- f o r
(u), j
+
j = 0,1, m.
..., w h e r e
v = v")
t v(')
>
... and
3. Singular Perturbations on Smooth Manifolds without Boundary
366
Then t h e r e e x i s t s a symbol a ( x , E , h , h < ) E Fv
1 #O
(u) such t h a t for each
i n t e g e r N > 0 holds:
The proof of t h i s statement i s l e f t t o t h e r e a d e r . One can a l s o formulate (and prove i n e x a c t l y t h e same way) a statement analogous t o Theorem 3.11.12
f o r difference singular perturbations i n
Op G v ( U ) , v = (vl,v2,v3). Theorem 3.11.36.
LEA
a ( x , ~ , h , h D )E Op G v ( U ) , v = (vl,v2,v3) and l e t
:=
- (v:j)+v a r =
1
"(1)
(1)
2
h
)
a , ( x , p , n ) , p = h/E, b e i t s graduate 3
(l+p-')
j>O
symbol, where a . s a t i s f i e s on each compact K c c u t h e i n e q u a l i t y 3 (j) .(I) la4(x,p,n) J
and where v
= v")
I
cK l w ( r i )
1'"
3
,
>v
g i v e n p E R+ , P = h/E ho Zds :
3.12. E l l i p t i c Singular P e r t u r b a t i o n s In t h i s s e c t i o n a p r i o r i e s t i m a t e s a r e e s t a b l i s h e d and parametrix c o n s t r u c t i o n s a r e c a r r i e d o u t f o r E l l i p t i c p s e u d o d i f f e r e n t i a l and d i f f e r e n c e s i n g u l a r perturbations. We s t a r t with a g l o b a l v e r s i o n of t h e d e f i n i t i o n of symbols S v ( U ) introduced i n Section 3 . 3 . D e f i n i t i o n 3.12.1. a ( x , E , < ) i s s a i d t o be i n t h e symbol class L'(R") following conditions are s a t i s f i e d
A function
i f the
( i )a E s'(R") ( i i )t h e r e
x
-f
e x i s t s a symbo2 a,(E,c) E s'(R") such t h a t t h e @ n e t i o n : = a ( x , ~ , c ) - a - ( ~ , cbeZongs ) t o s(R:), i . e . t h e folZowing
a'(x,c,c)
i n e q u a l i t i e s hold:
3.12. Elliptic Singular Perturbations
367
where C > 0 are some constantswhichmay depend on t h e i r s u b s c r i p t s . With a (x,E,S) the corresponding symbol in the representation (3.3.2) (with 0 a . and r satisfying (3.3.3), (3.3.4)),we shall denote (with some abuse of notation) also by a (x.E.~)the homogeneous extension of a . (as a function 0 of ( E -, C~) ) to (x,E,S) E Rn X IR+ x Rn , which is also called the principal symbol of a. Definition 3.12.2.
A symbol a E L'(IR~)
i s s a i d t o be e l l i p t i c of order v E
m3
i f i t s principal
symbol a0(x,E,<) s a t i s f i e s t h e c o n d i t i o n :
where c
0
i s some c o n s t a n t .
The corresponding s i n g u l a r p e r t u r b a t i o n a(x,E,D)
= op a
E op LV(I??") i s
s a i d t o be e l l i p t i c of order v. The following definition is a natural extension of Definition 3.1.7. to v n symbol classes L (IR ) . Definition 3.12.3. 0
The symbol a (x,E) E 1 a(x,E,c) E L'(IR") (3.12.3)
-v2
<<>
(O,V2,O) ( 7 ~i ~s )s a i d t o be a reduced symbol o f
i f v1
(E
<E<>
2 0 a(x,~,S)-a(x,S))E L-e,
e
=
(l,-l,l),
0 a (x,C) being a reduced symbol o f a(x,~,C),t h e operator Op ao
=
0 a (x,D)
i s s a i d t o be a reduced operator of t h e s i n g u l a r p e r t u r b a t i o n op a v
= ~(X,E,D)
n
Lemma 3.1.8. extends automatically to the symbol classes L (I?? ) . (0,V2.O) If a E LV(Rn) has the reduced symbol ao E L (Rn) , then we shall denote by aOO(x,S) the principal homogeneous in 5 (of order v-+v L 3 0 part of its principal symbol aO(x,E,<) and by a (x,<) the principal homo0 0 geneous in 5 (of order v2) part of a (x,S). Remark 3.12.4. v n As a consequence of Proposition 3.3.3 a symbol a E L (I?? ) . order v iff its principal symbol a
is elliptic of
satisfies the condition
3. Singular Perturbations on Smooth Manifolds without Bounda y
368
x ( S ) aO(x,E,S) that
x
5
-1
E L-"(Rn) , where
0 for 151 5 6,
If a E Lv(IRn)
x
m n E C (IR )
x /Si
E 1 for
is a cut-off function such 0 2 2 6 with some 6 > 0 .
L(orv2po) (Rn) , then
h a s the reduced symbol ao E
the
ellipticity condition for a can be stated in the following equivalent fashion: Proposition 3.12.5.
L e t a E Lv(Rn) and l e t aO,aO0 and ao be its p r i n c i p a 2 , p r i n c i p a l homo0
geneous and p r i n c i p a l reduced symbols, r e s p e c t i v e l y . Then a i s e l l i p t i c of order v i f f t h e f o l l o w i n g c o n d i t i o n s are s a t i s f i e d :
z
v (x,w) E mn x an (ii) agO(x,u) P 0 , v (x,w) E nn x an E xn x (R~\{oI); (iii) ao(x,l,S) z 0 , v (x,~;) here an = { C E mn 1 151 = 11. (i) aOO(x,w)
0,
Proof. ___ It is quite obvious that ( 3 . 1 2 . 2 ) (3.12.2)
by
and letting
E~~
E +
implies (i)-(iii). Indeed multiplying
0, one gets
(i)i afterwards, multi~lying
v -v (3.12.2)
by
and letting
E
E +
immediate consequence of ( 3 . 1 2 . 2 ) of order vl+v2.
one gets (ii); finally (iii) is an
m,
and homogeneity of a (x,E,C) in 0
Reciprocally, using again the homogeneity of . a aO(x,E,S)
z
v (x,E,c)E mn
0,
x
R+
x
( E - 1, 5 )
and (iii), one finds:
(IR~\{oI).
Further, using (ii) one gets for E~ sufficiently small: -v 1 jao(X,E,W)j 2 CE , d' (x,6,w) E Rn X ( O , E ~ ] X an, and,as a consequence of (i), one finds: ja0(x,E,w)1 t C provided that
~
, ~v
(-X , E~, W )~
E IRn x [El,m)
x
nn,
is sufficiently large. Thus, one gets that
E
1-V
I
E
la (X,E,W) b c e 0
1<E>v3,
v
(X,E,<) E
mn
x
w+
x
an, where
C > 0 is some -1
(x,E,S) in 0
constant. Using one more time the homogeneity of a
(E
,S)
of order v +v2, one gets finally ( 3 . 1 2 . 2 ) . 1
Proposition 3.12.6.
Let
x
for
1XI
:
mn X
+
Rn be a diffeomorphism such t h a t X(X) Y
s AX w i t h A
E ISO(R~,IR~)
s u f f i c i e n t l y l a r g e . Then t h e e l l i p t i c i t y of s i n g u l a r p e r t u r b a t i o n s
i n op LV(mn) i s an i n v a r i a n t p r o p e r t y w i t h r e s p e c t t o such diffeomorphisms. Proof. This statement is an immediate consequence of ( 3 . 1 0 . 3 1 ) . Our purpose now is to establish a two-sided
a
priori estimate for
3.12. Elliptic Singular Perturbations
369
elliptic singular perturbations in Op L V (Rn) which is their characteristic property. It has been already proved (see Corollary 3.4.2) that if a E then uniformly with respect to op a
: H
(123
( s ),E
-t
E
E
(O,E
1
holds:
E
R 3 , i.e. one has with some
0
H ( ~ - ~ )(XIn) , ~ ,V s
constant C, which does not depend on
Lv (Rn ) ,
and u,the following inequality:
E
It turns out, that a kind of reciprocal inequality holds for elliptic singular perturbations of order v and that the ellipticity condition is also necessary for such inequality to hold. To establish this property some auxiliary results are needed. Lemma 3.12.7.
For each
(s2,s3)
E
R and each
R X
E
E (0,11 t h e following i n e q u a l i t y
0
holds: 1-s (3.12.5)
where c and N
0
depend only on s2 and
s
N +1
-S
5
2<Eq>
/<5>s2<E5>s3-
C
, v
E
E
(o,Eo]
3‘
Proof. It is immediate that I V
5
<S>’I
6 lp1<5>p-1 ,so that the Lagrange formula for
the difference on the left hand side of (3.12.5) and this last inequality yield:
1
1 <5>s2<E5>s3-s2<Er)>s3 5
s 2 -1
<<-n>(ls21
1-s
-S
s3 <Ece>
5
2< E I ) >
+€IS
3
I<<
s -1 3
>’ e
<ECe>
1-s
)
-S 2<Eri>
3
,
= <-e(n-5) with some 0 E ( 0 , l ) . and Now using Peetre’s inequality to estimate e >P1
where C e
gets (3.12.5) with
c = 2maxI 1 s 2 12
I
I S 2 - i / + j s3 I
N~ = max{ s2-l
Is2/+ls3-1
I + I s3 I 1 s 2 I +Is3-l I }
noticing also that ~ < n > 5 <EW, V
E
E
(0.11.
I },
,/s3/2 I
I
In several applications the following auxiliary statement (known as Schur‘s
3. Singular Perturbations on Smooth Manifolds without Bounda y
370
lemma) turns out to be very useful Lemma 3.12.8. 0
If s(x,y) E C ( R ~x wn) and supn I Is(x,y)Idx 4 C, yElR lRn
then t h e i n t e g r a l operator s norm
:
sup x d
L'(R~)
+
I
Is(x,y)Idy 5 C,
wn
L'(IR~)
w i t h k e r n e l s(x,y) has i t s
SC:
I Is1
2
Proof. Cauchy-Schwarz inequality yields:
I (SU Estimating the last integral by C and integrating with respect to x, one gets
I Isu I Lemma 3.12.9.
-s
L e t a E l"(Rn) and l e t pS(~,S)= with respect t o (3.12.6)
E
E
(O,E
0
s
I<<> 2<~<>s3,s E m 3 . Then uniformly
E
I one has:
[ps(~,D),a(x,~,D)l: H
(s+v-e2)
,E
(wn)
+
H
(0),E
(Rn) ,
where [p (E,D),~(X,E,D)]stands f o r t h e c o m t a n t of t h e singuLar p e r t u r b a t i o n s p (E,D) = op p Proof. _ _ Since [p (€,D),am (E,D)I that x
+
and a(x,e,~)= op a, and e2
=
(0,1,0).
0, one can assume without loss of generality
a(x,E,S) belongs to S(R:)
,
as a function of x, V (E,E;)
Denote K(x,€,D) = [p (E,D),a(x,~,D)1. It is easily seen that
is an integral operator with the kernel
E ( O , E ~X] an
3.12. Elliptic Singular Perturbations where a(C,E,n)
=
37 1
Fx+C a(x,S,rl).
With V(c) = Ps+u-e ( E R E ) 2
U(<),
(3.12.6) is equivalent to the following inequality:
and where the constant C does not depend on Given that a E L'(Rn)
E
E
P-u(E,rl)/a(<,E,rl)I
,
and a(.,E,S) E S(iR:)
one has the following inequalities for ; ( C , E , ~ ) (3.12.9)
(O,E
5 CN<<>-N,
v
0
V (E,€,)
v
5
E
(O,E~]
x
Rn,
a(.,~,rl):
= Fx+<
N 2 0,
2 E L (Rn).
1 and
(E,n),
C may depend o n l y on N. N Now, (3.12.8), Lemma 3.12.7 and (3.12.9) with N
where the constant
(3.12.10)
J j;(S,~,n) / d S 5
J Ig(S,~,rl)/dn5
C,
N +n+2 yield:
=
0
C,
Rn
Rn
where the constant C does not depend on E , S , ~ I . Hence, as a consequence of Lemma 3.12.8, one has (3.12.7), so that
I
(3.12.6) holds, as well.
The following statement is the local version of a priori estimates for the solutions of elliptic singular perturbations with sufficiently small support. We recall that for each s E Rn
,
V
and a . V are defined by (2.1.13), 1 s
(2.1.15). Lemma 3.12.10.
L e t a E 1' (mn)be e l l i p t i c of order u and l e t u, ( 0 <
E
5 E ~ belong ) t o a bounded
s e t i n c t ( u 6 ) , where t h e open s e t u 6 does not depend on
E
and has diameter
6. 3
Then f o r each s E R and each such t h a t f o r 6 and holds:
E~
s'
E
a,rs, t h e r e e z i s t s a c o n s t a n t c s , s '
s u f f i c i e n t l y s m a l l , t h e fo'olowing a p r i o r i e s t i m a t e
3. Singular Perturbations on Smooth Manifoolds without Bounda y
372
Proof. L e t xo be any g i v e n p o i n t i n U6. I n t r o d u c i n g t h e n o t a t i o n
a (x) = a (x,E,D), 0
0
R
0
a ( x ) = a(x,E,D),
(x) = a (x )-a ( x ) , R(x) = a ( x ) - a ( x ) ,
0
0
0
0
one c a n w r i t e (3.12.12) a ( x ) = a o ( x o ) + R o ( x ) + R ( x ) . By ( 3 . 3 . 4 ) w i t h ( a = @ = 0 ) and as a consequence o f C o r o l l a r y 3 . 4 . 2 , has
(X-x0)
-1 6
Ro(X)
: H
uniformly w i t h r e s p e c t t o
E
(v)
,E
E (0, 1~. 0
(Rn)
,
one
3.12. Elliptic Singular Perturbations
373
By Proposition 3.1.11 one has:
I
1 lul I ( s ) , E
(3.12.17)
+I
5 C ( / la0(Xo)UI (s-v)
for each s ' E a,Vs, where
C
IuI I ( s ' )
1
may depend only on s and s ' but not on
E
and u.
Finally, combining (3.12.13). (3.12.15)-(3.12.17) and using interpolation inequality 2.1.12 (with 61 instead of 6) in order to estimate the
I
I
term l u l (s-e2) , E by A l l IUJ I ( s ) ,E and CI l u l on 6,, s and s ' , one gets the estimate :
I
with
( s t )
C
which may depend
I
(3.12.18)
I (s-v)
1 Iui 1 ( s ) , E
where the constant Now choosing
5 c ( j ja(x,E.~)ul
does not depend on
C
E
~
6 , and 6
E
,E
and u.
sufficiently small, (3.12.18) gives (3
1
Remark 3.12.11. If a E Lv (Rn) is elliptic of order v and u Ur
=
m
E L
m
( (o,Eo] ; Co(U,)
{x E Rn , 1x1 > r}, then for r large enough and
E~
)
,where
sufficiently small,
(3.12.11) still holds. Indeed, the fact that the diameter of supp u is < 6 was used in the proof of Lemma 3.12.10 only in order to estimate the term with R (x) by 0
Noticing that R (x) 0
=
(a (x)-a 0
0
(-))
C6.
E S(Rn) as a function of x, the same
argument as previously, yields the inequality (3.12.16). Now we are in a position to prove the global version of the a priori estimate (3.12.11) for elliptic singular perturbations of order v . Theorem 3.12.12.
L e t a E L'(IR") 3 each s E IR ,
be e l l i p t i c of order v. Then f o r s'
E~
s u f f i c i e n t l y small and
E alVs one has:
where t h e c o n s t a n t c does n o t depend on
E
and u.
Proof. Let B6
=
{x
E
Rn
I
1x1
C
6-l) and let {UjllSjsJ be a finite covering of B 6 n -
by the open sets U . whose diameters are < 6, so that, with U j + l
=nRxyB6,
the collection of open sets IUj}15jsJ+l is a finite covering of
JK
1
.
Let {$j}lsjsJ+l be the partition of unity subordinate to the covering
3 . Singular Perturbations on Smooth Manifolds without Boundary
314 "j
m n 15j<J+1' so that for each u E L ~ ( ( O , E ~ ]Co(Rx)) ; ,one has:
(3.12.20) u
=
I: Jlju. 15j<J+l
If 6 is chosen to be sufficiently small, then, as a consequence of
Lemma 3.12.10 and Remark 3.12.11, one finds, using (3.12.20):
Applying again Lemma 3.12.9 and interpolation inequality (2.1 gets the estimate:
with some constant s,s',v,6 and
C
which does not depend on
E
and u , but may depend on
E
0' Now, (3.12.21), (3.12.22) yield (3.12.19).
I
We show the necessity of the ellipticity condition for the validity of the a priori estimate (3.12.19). Theorem 3.12.13.
If f o r a E L'(R") and
E alvs,
s'
the a priori estimate (3.12.19) holds with some s'
<
s,
s
E R
then a is elliptic of order v .
Proof. Let
rl
E R3 and let $ E Ci(Rn)
.
We shall substitute into (3.12.19) the
functions u = exp(ic-'<x,Q>)$(x). using (3.6.31), one finds for each 3 q s E R :
and for each s E R',
s'
E alVs,
depends only on @ and
where C m,rl
s
rl.
<
s':
3
3.12. Elliptic Singular Perturbations
375
One can show also directly that (3.12.23), (3.12.24) hold. Indeed,
the brackets on the right hand side of
Denoting the difference within
(3.12.25), by g(E,n), using the triangle inequality for the norms and Lemma 3.12.7, one finds: s -1
(3.12.26) (g(EITl)I5 C E ( I E
S
-1
<En>
s N +1
Noticing that
one gets, using (3.12.25)-(3.12.27),the formula (3.12.23) with
A similar argument can be used for proving directly (3.12.24).
Now, using (3.12.23), (3.12.24). the a priori estimate (3.12.19) with u
rl
as chosen above, yields,after letting
where C does not depend on
n E
Wn\{O>
E
+
0:
2 and $ E L (Rn).
The inequality (3.12.28) is equivalent with (3.12.2), since a E -1 and a . is homogeneous in ( E , 5 ) E lR+ X Rn of order v +v I 1 2'
L v (Rn )
Our next step is to present the elliptic theory and, particularly, a parametrix construction for elliptic singular perturbations with their (U), S " ( u ) or K " ( U ) , which will be subsequently carried out to symbols in S" 1 PO m elliptic singular perturbations on C -manifolds without boundary.
We start with the concept of local ellipticity.
3 . Singular Perturbations on Smooth Manifolds without Bounda y
316
Definition 3.12.14.
A symbol a E s ~ , ~ ( ui s) said t o be e Z l i p t i c of order s t r i c t l y p o s i t i v e continuous on
u functions
v i f
there e x i s t
6(x), r(x) and q(x) such t h a t
for each x E u one has: -v (3.12.29)
la(x,E,S)\ 2 q(x)E
v
2<~5>v3, V
'151
E
E (0,6(x)1, V 5 E n n , 151 >r(x)
The corresponding singular perturbation op a = ~(x,E,D)i s called e l l i p t i c o f order v . For symbols a E S v ( U ) the ellipticity condition (3.12.29)
can be
stated in the following equivalent way by using the homogeneous extension of a principal symbol a0 (x,E,<) of a to R~
X
R+
X
(IRg\{Ol)
(by an abuse
of notation this homogeneous extension, which is well defined, will also be denoted by . a (x,E ,C)
.
Proposition 3.12.15.
A symbol a E
sV(u)
aO(x,E,<) onto
i s e l l i p t i c of order v i f f the homogeneous extension
xn
x R+ x (R~\{o}) of i t s principal symboz s a t i s f i e s the 5 condition: there e x i s t s a s t r i c t l y p o s i t i v e continuous on u f u n c t i o n q(x)
such t h a t f o r each x E
u holds: -v
(3.12.30)
/aO(x,E,S)\t q(x)E
v
'151 2 < ~ S > v 3 , V
(E,<)
E R+
x
(Rn\{Ol)
Proof. It is quite obvious that (3.12.29)
implies the same inequality for
/
2 r(x). Now the homogeneity of a (x,E,<) a0(X,E,~) when 0 < E 6 6(x), 0 in ( E - ~ , < ) of order vl+v2 implies ( 3 . 1 2 . 3 0 ) .
conversely, since la-a
0
I
5 Cla
0
I
when 0 <
E 6
6(x),
151
b r(x) with
a constant C which can be chosen as small as one wishes when 6 ( x ) is sufficiently small and r(x) sufficiently large, it is quite obvious that (3.12.30)
implies ( 3 . 1 2 . 2 9 ) .
Before the next statement will be formulated, we introduce a useful notation. For symbols a and b in Sv
180
(U) denote by a-b a symbol in Sv 1,o(")
whose asymptotic expansion is
In fact, as a consequence of Theorem 3.6.2,
formula (3.12.31)
the (non-commutative) multiplicative group structure on
defines
11 Sv (U), the VER3 l r 0
3.12. Elliptic Singular Perturbations product of two symbols or order
u
377
and v being well defined modulo a symbol
of order ( v +p1,--,v3+u3). Thus, with the natural additive group structure 1
and topology on
u
(U) defined by the semi-norms (3.3.1)' U Sv (U) becomes 1.0 v 1,o an algebra with addition and multiplication as continuous algebraic operations Sv
L,
with respect to its topology. Proposition 3.12.16.
Let a E
The f o l l o w i n g c o n d i t i o n s a r e e q u i v a l e n t :
S' l,o(u).
(i) a i s e l l i p t i c of order v. (ii) There e x i s t s a r i g h t i n v e r s e a-l E S-'(U)
(iii) There e x i s t s a l e f t i n v e r s e ' ; a
E
(3.12.33) (a;'*a-l)
S 1( ,o o f - m r o()u )
E s-'
1to
( u ) of a, i.e.
.
-1
Moreover, a l can be chosen i n such a way,that -1 a E
of a, i . e .
-1 -1 -1 al =ar = a , any i n v e r s e symbol
s;~,(u) being w e l l d e f i n e d i n t h e f a c t o r aZgebra
(-V1,-",-V3) (U) . s-v 1 .o (u)/sl,o Proof. -
Let a E
Sv
1.0 as follows:
(U) be elliptic of order v . Define the functions a-l recursively, k
and for k = 1,2,..
Since a E when 0 <
(U) is elliptic of order v, the symbols a;'(x,E,<) are welldefined 1 to -1 6 6 ( x ) , 151 Z r(x). We shall denote by ak ( x , E , S ) any smooth
Sv E
extension of these symbols onto
E ( 0 . ~ ~X 1lRn.
An elementary -1 v(k) computation shows (using induction in k) that a E S1,o ( u ) with - (-V1,-v2-k,-v ) . Thus, Theorem 3.6.2 guarantees the existence of a V(k) 3 -1 symbol a ( X , E , ~ )E S-' (U) , having the asymptotic expansion: 1to (3.12.36) a-'
-
X k>O
(E,<)
-1 a k'
Now, (3.12.34), (3.12.35) are derived to satisfy (3.12.32) for a-l defined by (3.12.36), and that ends the proof of (ii).
3. Singular Perturbations on Smooth Manifolds without Boundary
378
Now, suppose that (ii) holds. This implies that -1 (U), so that for each compact K (a(x,E.E)a (x,E,E)-I) E s 1 ,o there exists a constant CK such that la(x,E,S)a-'(x,E,S)-l~ 5
cK< ~ > - l ,v
(x,E,<) E K
x
C
U,
(o,Eo] x
mn
2 rK, one gets for Choosing rK > 0 ,ko large to have C i0-l < f , V K such 5 and V E E (O,E~],V x E K with (perhaps a different) constant Ck
the inequality:
+
-1
5 la(x,E,<)a
( ~ , E , Ej )5 c;ja(x,E,E)
u1 / E
<s>
-v
-V
2<E~>
3
and that implies the ellipticity of a of order v. We have to show the equivalency of (i) and (ii). Using the same argument as previously and seeking again the asymptotic expansion of -1 E S1),o(U) in the form a 1 a-l
N
c k>O
a-l k
-1 and the one gets, using (3.12.31), the same formula (3.12.34) for . a -1
following recurrence relations for ak , k > 0:
to satisfy (3.12.33). Now, the previous argument applies. To end the proof, we have to show that one can choose the same right and
left inverse symbol in S-v (U) for an elliptic symbol in Sv
1 ,o
180
(U). This is
an immediate consequence of the associativity of product (3.12.31). Indeed,
Corollary 3.12.17.
Let a ( x , ~ , ~be) in op sv
( u ) . Then the following conditions are equivalent:
1 ,o (i) a(x,E,D) is e l l i p t i c of order v.
(ii) There e x i s t s a-'(x,c,~) i n op s-'
1 to
(u) such t h a t
a(x,c,D) 'a -1 (x,€.D)-Id : 0 (mod Op S (0,-",O) (U)) 1 ,O
3.12. Elliptic Singular Perturbations
where Id is the identity operator and
0
379
stands for the composition
(multiplication) of singular perturbations, as defined in Section 3.7.
I
Definition 3.12.18.
A s i n g u l a r p e r t u r b a t i o n a-'(x,~,D) s a t i s f y i n g (3.12.37) i s s a i d t o be a
p a r m e t r i x f o r a(x,e,D) E op
S~,~(UI.
Corollary 3.12.19.
( u ) . Then a(x,~,D)has a pararnetrix a-l -1 (x,E,D)E op s-' 1 .o I,O(U) i f f a(x,E,c) i s e l l i p t i c of order v. Furthermore, a (x,E,D)i s e l l i p t i c of L e t a E sv
order -v and i s w e l l d e f i n e d i n S;~o(~)/~:~~l'-m'-v 3) (U).
I
Corollary 3.12.20.
Let a E E
Sv
1 to
(U).
Then a(x,E,D) i s e l l i p t i c i f f uniformly w i t h r e s p e c t t o
E ( 0 . ~ ~one 1 has:
(3.12.38)
-1 where a (x,E,D)i s any pararnetrix of a(x,~,D),HComP ( u ) i s t h e subspace ( s ) ,E (U) o f a l l U(X,E) w i t h t h e i r support i n some compact K c U and i f l H H
V
loc(s) , E (U) i s t h e space of a l l U(X,E) such t h a t x(x)u E H (S),E
x E
u do n o t depend on
Co(U); here K and
( s ) ,E
(U)
,
E.
Indeed, the top line in (3.12.38) i s a consequence of Theorem 3.4.1, since one can extend x
X
1 on an open set
a(x,~,S)to x E Rn by introducing x(x)a(x,~,S),
+
U',
U 3 3
U'
3
K,
continuous (uniformly with respect to
for any given compact K
C
x E)
m
E C0( U ) ;
further, there is a natural
embedding:
U.
The bottom line in (3.12.38) follows by the same argument, given that the -1 existence of a parametrix a (~,E,D) E op s;y0(u) of a(x,E,~)E op s Y,O(U) is equivalent with the ellipticity of a E S y , o ( U )
of order v.
Corollary 3.12.21.
If a E s ~ , ~ ( iu s) e l l i p t i c of order
v, t h e n any s o l u t i o n uE E m
E ' ( u ) of t h e
c (u) t o o , and moreover, i f EvlfE,O < E 2 E ~ ,belongs t o a bounded s e t i n cm(u) and uE, 0 < E 2 E 0 belongs t o a bounded s e t i n E ' ( u ) , t h e n f o r any g i v e n compact K c u one has: equation a(x,E,D)uE
=
fE w i t h f E E cm(u) i s , i n f a c t , i n
3. Singular Perturbations on Smooth Manifolds without Boundary
380 uE, 0 <
E
5
E
beZongs to a bounded s e t in Cm(K).
0'
--Y
Indeed, writing u b(x,E,D) f Op
S
1 -1
a
(x,E,D) r"la(x,E,D)u+b(x,&,D)u with
( o f - m f o )(U) (as 1#O
a consequence of (3.12.33)), one gets
= 6
a3
immediately the conclusion that {u}o<EcE belongs to a bounded set in C ( K ) , 0
since {b(x,E,D)u } O < E L E and -V
Cm(U) and
1
6
{ E V 1 a(x,s,D)u)O<E,E
0
-1 a (x,E,D)E Op
are in a bounded set in 0
(0,-V2,-V3) S
(U) has the pseudolocality
1#O
I
property (see Theorem 3.5.4). Remark 3.12.22.
The elliptic singular perturbation theory extends automatically to systems of singularly perturbed pseudodifferential operators, whose
symbols are pxp matrices depending on (x,E,<)E
u
(O,E
x
0
1
x
En.
In all
the definitions and results above the absolute value of symbols should be interpreted as the matrix-norm in Hom(CP;CP). Remark 3.12.23. Let M be a smooth paracompact manifold without boundary and let a(x,E,D) E Op
Sv
1
(M) be a singular perturbation of order v on M (see
Definition 3.10.13). If a(x,E,D) f Op S"(M), then its principal symbol a ( x , E , ~ is ) well defined 0
and, as a consequence of (3.10.31). aO(x,E,5) is a function on R (E
E 1R+
,
*
X
T (M)
(x,S) E T*(M), T*(M) being the cotangent bundle on M) with values
in Hom(CP;CP). Thus, the ellipticity concept can be extended to singular perturbations on M, the ellipticity condition for a E S v ( M ) being the
v requirement: a (x,E,<) E ISO(C~;C~), (ao(x,E,<)I 2
CE
E
R+
-" 16 I "2<E5>"3, v x E M, where
,v
(x,~) E T*(M), 5
C > 0
+ 0,
is some constant. The
same argument as previously in the case M = R n , shows that the ellipticity condition is equivalent with the a priori estimate (3.12.19) where the norms are in corresponding spaces H
( s ), E
(M).
One can also show (but it will not be done here), that if a(x,&,D) E Op S"(M)
is elliptic and M is compact, then the kernel of m
a(x,E,D)
:
H
V 6 E (O,Eol;
( s ) , )E'(
+ H ( s - v ) ,E
(M) belongs to a bounded set in C (M),
furthermore, the range of a(x,~,D) is the orthogonalcomplement
of thekernel ofthe adjoint a(x,~,D)*. Therefore, theindex dim coker a of an elliptic singular perturbation a(x,&,D) H
(s-v)
fact, 0
( M ) does not depend on s
X(E) :
H
=dim ker a ( s ),E
(M)
*
E 1R3. It will be shown later, that, in
,E
X(E)
5
X ( 0 ) if a(x,s,D) has the reduced pseudodifferential operator 0
) the index of a (x,D). a (x,D), where ~ ( 0 is
3.12. Elliptic Singular Perturbations
38 1
Example 3.12.24. The singular perturbation (3.10.52) on the unit circle
a1
is elliptic of
order v = (0,0,1), since its principal symbol a (x,E,S) = l+EIE;I 0
-
<ES>.
As a consequence of (3.10.54) kernel and cokernel of this singular
perturbation are both trivial and its index
X(E)
=
0 , t/
E
2 0.
The next example is concerned with Stokes' singular perturbation (3.2.43). Example 3.12.25. Consider the singular perturbation S (E,D)whose symbol is the following
e
nxn matrix s
E
(E,S):
S ~ ( E , S )= where w = 5 / 1 5 / , w
E
2ie+E212)-1(Id (e
w X w T) ,
T .
is the transpose of w and wxwT is the matrix with the
entries wjok, 1 5 j , k 5 n. Any vector u E Cn such that <w,u> = 0 is an eigenvector of s (E,S) e 2 2iE 2 = E (e + 15 ) , while w is an eigenvector with zero
I
with the eigenvalue u eigenvalue. Thus,
is not elliptic. However, considered on the
se(E,S)
orthogonal complement of
w
in Cn it becomes elliptic of order (-2,0,-2).
We are going to introduce a more general ellipticity concept for systems of singular perturbations in order to include in the elliptic theory Stokes' singular perturbation (3.2.44). Introduce again the notation
and introduce
Definition 3.12.26.
A matrix symbol a(x,E,S) = 1 lakj(x,E,S) 1 1 with a E LvkJ(mn) is said to kj be elliptic in the sense of Douglis-Nirenberg, if there exist vectors vk E IR 3 , LI; E n 3 , 1 5 k 5 p, such that the symbol ;(X,E,~), N
(3.12.40) a(x,E,S) = q,,(E,~)a(x,E,S)q,(E,~)
is elliptic of some order v in the sense of (3.12.2) where the norm in Hom(CP;CP) .
I. 1
stands f o r
3. Singular Perturbations on Smooth Manifolds without Boundary
382
A matrix singular perturbation a(x,e,~)is said to be elliptic in the sense of Douglis-Nirenberg, if its symbol a(x,~,S)is elliptic i n the same sense. The following result is proved by repeating word for word the proof of Theorem 3.12.12.
Theorem 3.12.27.
Let a(x,s,D) be e2Ziptic in the sense of Douglis-Nirenberg and let 1-1k' 1 5 k 5 p and v be the corresponding vectors in Definition 3.12.26. Then uniformly with respect to (3.12.41) 119
-P
where s' E R
3
(E,D)u/
1 (s)
E
,E
E (O,E I 0
-j
one has:
I
Iqu,(E,D)a(x,E,D)ul ( s - v )
is any order s.t. s' < s-uk, 1
,E
+
I lul I
(s'), €
2 k 5 p.
Example 3.12.28. Consider Stokes' singular perturbation the symbol a(A,S):
where, as previously,
ST
is the row-vector transpose of 5 f lRn
.
With any u = ( u t,O)T such that <S,u'> = 0, one finds that 2 T a(A,S)u = ( A + l < l ) u , so that on the orthogonal complement of Span{(S,O) ,en+l] 2 n+ 1 a(A,S) reduces to ( A + / < ] )I in C (n-l) . Furthermore, an easy (n-l) computation shows that the other two eigenvalues of a(A,S) are, respectively,
Hence, if u - = (S,p-) is an eigenvector associated with the eigenvalue restricted to the orthogonal complement of u- in CnTi has 2 (X+lSl 2), while X/El2 when its norm of order (X+lS ) , since x+
X-,
then a ( A , S )
I
(A+l<12)
-t a.
-
-
This suggests that a(X,<) might be elliptic in the sense of
Definition 3.12.26. It is easily seen that for larg X I < 77 ( 1 8 1 < IT/^), 5 E Rn\{O} -1 inverse matrix a(X,S) exists and is given by the formula
the
3.12. Elliptic Singular Perturbations where, a s previously, w
=
5 / 1 5 ] and
w
T
383
1s the transpose of o.
1 5 k 5 n+l in Definition
The latter suggests a choice of pk,L$, 3.12.26 as follows: (3.12.44)
uk
= p i = 0, 1 5 k 5
n, !Jn+l=
=
(Z,-1,2),
a reasonable candidate for v in the definition being v = (2,O.Z). Indeed, with p . chosen as in (3.12.44) and with 7
C2 =
A-',
one finds,
using (3.12.39),(3.12.40):
N
The matrix a(A,c) on the orthogonal zomplement of Span{ ( 5 , O )
,en+l
is s t i l 1 the same multiple of the (n-l)x(n-1) identity i.e. (X+/S/2)I(n-l)x(n-1), N
and an easy computation shows the two other eigenvalues of a(X,c) are, respectively,
( € , E l above by introducing the
Modifying slightly the definition of q
N
factor i on the right hand side of (3.12.39::
one gets a(h.6) which is the
following symmetric matrix
whose eigenvalues are ( A + 15 1 Y
a(A,C)
2
)
and ( 3 ) (1+45)(X+ 15
can be estimated by ( f ) (fikl)( A + ( S /
2
)
1 2)
so that the norm of
frombelowandabove, respectively.
N
Thus, a ( h , S ) is elliptic of order (2,0,2) in the sense of Definition 3.12.2. The estimate (3.12.41) becomes in this case: (3012.45)
c
lSjSn
+
I bjI I ( s ) , 6
+
I
I (s-pn+l)
lUn+ll
N
,€
I IUI I (s'),E
with any s' < s,
s'
< s-pnCl, where e . is the j-th coordinate vector in Cn 3
3 . Singular Perturbations on Smooth Manifolds without Bounda y
384
With u
=
(u',u~+~) a(A,D)u ~, = (f',fn+l)T, one can rewrite
(3.12.45)
in the following fashion:
where, as previously, w = (2,O.Z) and e2 The reduced symbol for :(A
while the inverse :(A,<)-'
=
(O,l,O).
, 5 ) is :
has the form:
Introducing
one can write (3.12.50)
a(X,S)
=
(A+l5l
2
)
0
S(h,S)a ( 5 ) S ( h , S ) ,
ao(C) being the reduced symbol for a(A,S). An easy computation shows that
Hence, (3.12.52)
a(A,S)-'=
( A + 1 5 / 2 ) -1 diag(Inxn,(A+15/ , 2))a 0 (5)-'diag(In,,,(X+~S~*)),
v
E
C,
/arg A / <
T,
v c
E Rn\tO},
and the multiplication from the left by (A+1S12)-1S(h,S)-1 right by
S(A,S)-l
and from the
reduces a(A.5) to a0 ( 5 ) .
In x-representation the left reducing operator is the convolution with diag(A -1 A. -2 (h-fIxl)Inxn,l)
3.12. Elliptic Sinplar Perturbations
385
2 and the right one is the convolution with diag(1 nxn' (X+ ID1 ) 6 (x)) , where A;(E~X~J is the family of distributions, defined by (3.2.21) with E E C, /arg € 1 < ~ / 2 ,E =
-1
x '.
Example 3.12.29. Consider the Oseen linearization of the system of partial differential equations in the Nonstationary Magnetohydrodynamics under the noncompessibility assumption upon the moving fluid (see, for instance [La-Li, 21). It has the following form:
at
- aAv +
+ VP
=
0
T (3.12.53) VTv = 0 , V H = 0
aH at
BAH +
=
0
where v(x,t) is the velocity field, P(x,t) is the pressure, H(x,t) is the magnetic field and where the linearization is taken at the stationary magnetic field H
0 and the stationary velocity field vm : Const,the
density of the fluid being p
?
1; here, as usual V is the gradient operator
and the upper index T stands for the transpose, so that VT is the divergence operator; further a > 0, 6 > 0 are given. If the initial value H (x) for H(x,t) satisfies the condition 0 T V H (x) = 0, then, as a consequence of the last equation in (3.12.53) one 0 T gets that V H(x,t) E 0, V t 10, so that under this assunotion upon the initial value for H in the overdetermined system (3.12.53) one can drop the equation V'H
=
0, thus getting a determined system of equations.
After the Laplace transform with respect to t one gets the Stokes-Oseen problem in Magnetohydrodynamics, whose symbol a(X,c) has the form:
(3.12.54) a(X,E) =
Further, introducing new unknown functions
one reduces (2.12.53) to the Stokes-Oseen problem with the principal symbol aO(X,5),
3 . Singular Perturbations on Smooth Manifolds without Boundary
386
With
1x1
=
E - ~ ,/arg XI <
T,
a (h,D) is an elliptic singular 0
perturbation in the sense of Definition 3.12.26, as it follows from Example 3.12.28. Notice that for each given A E constant
Co >
for B = C a with some 0 in (3.12.55) is also elliptic in
0%-and
0, a << 1, the symbol a .
the sense of Definition 3.12.26, the order v appearing in this definition being v = (0,0,2), while for a and 6 fixed and A
-t
m
this order is (2,0,2).
The next example is concerned with von Karman's equations for the buckled state of a thin elastic plate. Example 3.12.30. The following system of equations
describes the buckled steady state of a thin elastic plate; here h is the thickness of the p1ate.h << 1, <(x) is the deflection and x(x) is the stress -1 = Po(x)h , P (x) being a given exterior forces'density.
function, P(x,h)
0
The linearization of (3.12.56) at 5 following singular perturbation (h = (3.12.57) (E21DI4+)c XO
=
E)
=
0,
x
=
x0 (x) leads
to the
for the deflection:
P
where
If
( x ) 2 y(x)Id with some continuous y(x)
A
0 , then the singular
XO
perturbation (3.12.57) is elliptic in the sense of Definition 3.12.14 of order v = (0,2,2). The ellipticity of (3.12.57) is equivalent to the strict convexity of the given deflection
x 0 (x)
at each point x C
U.
3.12. Elliptic Singiilar Perturbations
387
Next the elliptic theory of difference singular perturbations will be considered. We start with one parameter families of difference operators introduced in Section 3.11. Definition 3.12.31. A symbol a E F v ( u ) i s s a i d t o be e l i p t i c of order v
=
(vl,v2) i f i t s
p r i n c i p a l symbol ao(x,rl) s a t i s f i e s t h e c o n d i t i o n (3.12.58) ]ao(x,n)I 2 m(x) /w(n)I
v2
where m(x) i s some continuous s t r i c t l y p o s i t i v e f u n c t i o n on U. A f a m i l y of d i f f e r e n c e operator,. a(x,h,D) E op Fv(u) i s s a i d t o be
e l l i p t i c of order v, i f i t s symbol a(x,h,hS) i s e l l i p t i c of order v. We shall specifically treat the case U = lRn, using the following global version of symbol classes
Fv ( U ), Gv ( U ).
Definition 3.12.32.
-
A symbol a E FV(mn) ( r e s p . a E G'(IR") ~
FV(wn) ( r e s p .
G'(IR~)
I is
s a i d t o b e in t h e c l a s s
I if t h e r e e x i s t s a symbol a (h,hC) E F'(E~)
am(h,hE) E GV(Rn)I such t h a t t h e symbol
(resp.
hu1
h~'<~>-"2(a(,,h,h5)-a_(h,h~)), a s a f u n c t i o n of xERnbelongs t o S(Rz) UnifOYW?ly %n (h,S).Asymbol aEFV(mn) i s s a i d t o b e ( g l o b a l l y ) e l l z p t i c o f order v = ( v , , v ~ ) ,
=
if t h c r e e x i s t s a p o s i t i v e c o n s t a n t m such t h a t f o r t h e p r i n c i p a l symbol a (x,n) of a(x,h,h<) one has: 0
(3.12.59) la0 (x,rl)I 2 m/w(rl)I
v2
, V (x,rl) E IRn
X
(T>{O}).
The l a r g e s t m such t h a t (3.12.59) i s v a l i d , i s c a l l e d t h e e l l i p t i c i t y c o n s t a n t f o r a(x,h,hS). Remark 3.12.33. As a consequence of Proposition 3.11.23, ellipticity is an invariant
property of difference operators with respect to the group of transformations SL(n;Z). The following statement can be proved in exactly the same way as the analoguous statement in Theorem 3.12.12 (so that its proof is left to the reader) : Theorem 3.12.34.
L e t a E Fv(En) be g l o b a l l y e l l i p t i c of order v. Then f o r each couple of orders
(s,s')
E m2x IR'
w i t h s' < s, s;
= sl,
there e x i s t s a constant c
s,s'
3 . Singular Perturbations on Smooth Manifolds without Boundary
388 such t h a t
Now a parametrix construction (analogous to the one for elliptic singular perturbations) will be briefly sketched for families of elliptic difference operators. First, using (3.12.31), one provides U Fv (U) = Fl,o(U)with a multiv 1,o plicative group structure , the product a1*a2 of two symbols ( 1 ) +\) (2) k (U) with the (u), k = 1,2, being any symbol c E Fv a E FYj:) 1 #O asymptotic expansion (3.12.31) in the following sense : for each integer E J > 0 the difference
where
pN =
~(')+v(~)-Ne~, e2 = (0.1)
Furthermore, Proposition 3.12.16 carries over to the elliptic s y h o l s in classes Fv (U) with the only difference that instead of Theorem 3.6.2 1 PO one will have to use the corresponding statement, analoguous to the one in
Fv'(k) (U) with 1.0 Further, for a -1 and a-l in given elliptic symbol a E Fv (U) the corresponding symbols a 1 1,o F;Yo(U) are defined by the same recurrence procedure (3.12.34), (3.12.35). (-v1 , m ) -1 -1 (U) and Corollary modulo a symbol in F Again, one has a = al 1n o 3.12.17 carries over to the elliptic difference operators with their Theorem 3.11.5, which extends to any sequence of symbols in
> v(l) > v ( ~ )
... > v ( ~ )> ...,
V:~)+V;~) +
-m
for k
+ m.
symbols in the classes Fv (U), as well as all the other corollaries and 1 .o remarks which follow Proposition 3.12.16, with the only difference that the spaces H
(s)
(U) should be replaced ,P' H(s) ,E
by the corresponding
spaces of families of meshfunctions H (s),h(If) . Next, several examples of elliptic difference operators will be considered. Example 3.12.35. h Consider the family of difference operators aZ whose symbo-1 h o(a-) = (cl ,h+ic2,h) E F("') (Rn) approximates the Cauchy-Riemann operator z h a = wl+iw2. This approximation of a_, the principal symbol of o(a-1 being . z
2
a-
is not elliptic, since the circles Il+iXI = I and Il-XI in the complex
plane C
always have an intersection at a point X o # 0, so that there is
always a point
T.
o
E T2\{O} r
l
such that aO(rlO) = 0 .
3.12. Elliptic Singular Perturbations
389
Rewriting the Cauchy-Riemann operator 2- as a system for the real and imaginary parts of the unknown complex-valued function, one can indicate h a whole family At,depending upon the parameter t E R , o f approximations of
a-2' h At
=
it
all of them being elliptic, but the one corresponding to t
=
*
t ; here, as
previously, D are the (formal) adjoint operators for D x . ,h x .,h' 1 1 h Indeed, one finds immediately that /det(-iAt)I = 2 2 2 *2 21 z2 = I(t t(1-t) )1#+t(l-t)(5 +5 2 (2t-1) 1 5 : .
)I
Considering a more general system for pseudoanalytical functions (3.12.62) $x -T(X)$x -0(X)ix = 0 , $x +0(X)Qx -T(X)$x 1 1 2 2 1 2
=
0
> 0 (see [Bers, 1 I ) , One checks easily that 0 the approximation by finite differences of (3.12.62) with the symbol
which is elliptic if u(x) 2 u
is elliptic and one can indicate in this case, as well, a one parameter family of elliptic approximations of (3.12.62) similar to the case of the Cauchy-Riemann system considered above. Example 3.12.36. Consider the following approximations by finite differences of - A . Obviously, the approximation with the symbol u1 = / 5 1 2 is elliptic and it corresponds to the classical approximation of -A. Next, consider the approximation with the symbol u2 = [ 5 ' principal symbol lw'12+w;
I
2+
vanishes when
it is not elliptic, since the
ri
approximation of -A with the symbol u 3 =
= IT
(wn
=
2i) and I w ' /
nl 5 I 2- (h2/6)
=
2. The
2
I Ckl I Sj I
Z
is
l(k<j$n elliptic if n 5 3; for n > 3 it is not elliptic, since its principal
Iw(q)
[ sufficiently small and strictly negative when
lw(rl)
I
=
It is easily seen that the approximation of -A with the symbol
2 , n > 3.
3 . Singular Perturbations on Smooth Manifolds without Boundary
390
is elliptic in any dimension, the accuracy of this approximation being 6
O ( h ) when h
-f
0.
Example 3.12.37. The difference operator p(x,h,hD)
=
l+q(~)~(2-0-0-~) with 0 = exp(ihD) the
1
shift operator on F$, , is elliptic,of order ( O , O ) , provided that 2 2 /Re q(x) i qo > 0 , V x E IR . It approximates the identity operator with the 2 accuracy O(h when h + 0. Besides, as discussed in Example 3.11.21, the 2 2 -1 difference operator a(x,h,hD) with the symbol a(x,h,h<) = (l+q(x) /w(h5)) )
I
is a quasi inverse of p(x,h,hD), i.e. p where
41)
0
are some difference operators with the symbols o(R(1)) in Fl,O(IR). h More precisely, using Theorem 3.11.17, one finds the asymptotic expansions for the symbols of a(RL1))
0
-
c
ia
a>O
(2)) (Rh
ia a>O
where, as previously, for each symbol b(x,h,hE) we have denoted: b(a) (x,h,q) = Dab(x,h,rl), b (x,h,n) = D>(x,h,n), D = -id/dx. (a)
Using the recurrence procedure as in (3.12.341, (3.12.351, and the Corresponding analogue of Theorem 3.6.2 for symbols in Fv (U), one can 1 to find a symbol p-'(x,h,ht) such that for the corresponding difference -1 N (1) -1 operator p-l(x,h,hD) holds p o p -Id = h Rh,Nr P D p - Id = hN$yA, with any integer N > 0 and with some R(k) E Op FG ( W ) , k = 1,2. Of course, h,N 1 ,o -1 (x,h,hD) is also elliptic of order ( 0 , O ) .
p
Example 3.12.38. The following system appears in the linear elasticity theory: (3.12.63) P.(D
)
=
u~~ nx n+ ( p + ~ ) v x v ~ ,vxvT
I / a x , a x I I16k,j6n,
=
I
k
where I is the identity in Hom(Cn,Cn), V is the gradient and VT is the nxn divergence, !J > 0, X 2 0 being given coefficients. The system (3.12.63) is elliptic in the classical Petrovsky sense, since n 7n det A ( S ) t !J 151- , v 5 E IRn. T
Denote by Vh and Vh the forward difference approximation on the grid
.If: of the gradient V
and the divergence V
T
,
respectively, and let
3.12. Elliptic Singular Perturbations
*
391
T *
be the corresponding backward approximations of V and V Vh, (Vh following two approximations of (3.12.63) are elliptic
T
. The
where Ah is the classical approximation of the Laplacian with the symbol
12.
li(h,hS)
It is also easily seen that the whole family of approximations tAh(Dx,h)+(l-t)<(D x,h) , t E [O,ll, is elliptic, the accuracy of the approximation for t = f being higher than that for other values of t.
Ah(Dx,h) t
=
Example 3.12.39. The system (3.12.63) is a specific example of strongly elliptic operators in the sense of M.I. Vishik, the general form of these systems being (3.12.65) A(Dx)
z
=
la1
with A
a,B
=I 51-
Dz(Aa,6(x)D6) + terms of lower order,
(x) symmetric matrix satisfying the condition:
with some constant y o > 0. It is quite obvious that the family of difference operators
is an elliptic approximation of the principal part of A(D
)
in (3.12.65).
To end this paragraph, elliptic difference singular perturbations will be considered. W e start with the corresponding symbol classes with two small parameters
E
E (0, 1~ and h E (O,hol. 0
Definition 3.12.40.
A f u n c t i o n a(x,c,h,q) : u x ( 0 . ~ ~x1 (O,ho] x T i C is s a i d t o be a symbol i n t h e cZass F Y , o ( u ) , v = (vl,v2,v3)E ~ 3 i ,f for each compact K c u and each p a i r of m u l t i - i n d i c e s a,@ t h e r e e x i s t s a c o n s t a n t c +
a,B,K
t h a t holds:
such
3. Singular Perturbations on Smooth Manifolds without Bounda y
392
If a E
but a B F;,,(U),
FY,,(U)
V
u < v, then v i s said t o be the order
of a.
i s said to belong t o the c l a s s FV(u) i f there
A symbol a E F;,,(U)
e x i s t s a f i n e t i o n ao(x,p,n) a .
E
(Tl,n\{O})
Cm(u x
each compact s e t
w i t h some u $6(t) E $6(t)
=
K c
u
:
0
n C (u x
R+
u
x
x
R+
R+
;l,il + c r (T~,~\{O}) such t h a t uniformly on
x
one has:
(T:,~\{o})
x
(pl,p2,u3)s: v such t h a t
IA + p
1
v 1+v 2; here, as previously,
<
2
cm(E+)i s a cut-off f u n c t i o n such t h a t $6(t) : 0 f o r t E [ o , S l , 1 f o r t t 26, $ 6 ( t2) 0 , v t E E,, 6 > o being a given s u f f i c i e n t l y
small number. The f u n c t i o n aO(x,p,n) s a t i s f y i n g (3.12.67) i s called the principal symbol of a E FV(u). Proposition 3.12.41.
L e t a E F v ( U ) . Then i t s principal symbol ao(x,p,n) i s well defined and f o r
each compact cowstalzt c
u and each pair of m l t i - i n d i c e s a,8 there exists a such t h a t hold:
KCC
V (x,p,n)
E K
X
R+
X
(TY
rn
\{o}).
Proof. _ _ Rewriting (3.12.67) in the form -V
(3.12.69) ((s+h)v1
-V
2a(x,E,h,h5)-h
-V -V 2 $ 6 ( 1 5 / ) < 5 > 2ao(x,h/E,hS))<EC>
E
FV-V
1 ,O(U)' setting
I- =
hc,
p =
h/E and letting h
+
0,
E
+
0 in such a say that
=
h/E
be constant, one gets the conclusion that the limit on the left hand side of (3.12.69) is zero, since
u1
5 vl,
ul+u2
<
vl+v2, u < v, so that ao(x,p,n)
is well defined by the formula:
v +v (3.12.70) ao(x,p,q) = lim h h+O
2(1+p-1)v
at each given point (x,p,n) E
U
X
R+
X
1
a(x,hp
-1
,h,n),
(TY,n\{O}).
The same argument leads to the following formula
v
+V
(3.12.71) D BDaa ( x , p , n ) = h 1 x n o
-la/
2
(l+p-l)' l D ; D : a V (x,p,n)
(x,hp-l, h , 0 )
E
U x
R+
X
,
(TY,n\{O}).
3.12. Elliptic Singular Perturbations
393
Now,using (3.12.661, (3.12.71), one g e t s (3.12.68).
I
D e f i n i t i o n 3.12.42.
A symboZ a E F V ( u ) i s s a i d t o be e l l i p t i c of order v
p r i n c i p a l symbol a
0
(x,P,I)
=
(vl,v2,v3), if i t s
s a t i s f i e s the condition: there e x i s t s a
s t r i c t l y p o s i t i u e continuous on u f u n c t i o n q ( x ) such t h a t for each ho Zds : (3.12.72) / a o ( x , p , n )
I
2 q(x)]w(ri)
x
E u
1 v 2<~-1w(n)>v3
The g l o b a l v e r s i o n of D e f i n i t i o n s 3.12.40, 3.12.42 and o f P r o p o s i t i o n
3.12.41 f o r U
=
b a t i o n s i n Op
L v ( Rn )
Rn
i s t h e same, as p r e v i o u s l y for t h e s i n g u l a r p e r t u r -
-
and d i f f e r e n c e o p e r a t o r s i n Op Fv(Hln)
.
In particular,
t h e g l o b a l e l l i p t i c i t y means t h a t i n (3.12.72) one c a n t a k e a p o s i t i v e c o n s t a n t q i n s t e a d of a p o s i t i v e c o n t i n u o u s f u n c t i o n q ( x ) .
W e a r e g o i n g t o g i v e a n o t h e r e q u i v a l e n t d e f i n i t i o n of t h e e l l i p t i c difference singular perturbations
which have a reduced symbol.
D e f i n i t i o n 3.12.43.
A symboL a E Fv
1 ,o
0
a (x,h,hE,) E
FV’
1,o
-V
(3.12.73) <<>
where e
=
2
(u),
v = (v1,v2,v3), i
,
(U)
v ‘ = (v -V
((E+h)”<E<>
v ),
2’ 3 3
s s a i d t o have a reduced symbol
i f holds:
a(x,E,h,hS)-a
0
(x,h,hS))
E Fi:O(U),
(I,-I,~).
I t i s q u i t e o b v i o u s t h a t t h e reduced symbol i s u n i q u e l y d e f i n e d ,
i f it
e x i s t s , a n d , moreover, one h a s
0
(3.12.74) a ( x , h , h S )
= lirn
(E+h)’l
a(x,E,h,hS)
E+O
for e a c h g i v e n x E U , h E ( 0 , h
1,
0
ri = i 5 E Ty,n.
L e t a E Fv(U) be e l l i p t i c o f o r d e r u and assume t h a t a h a s t h e reduced symbol a’. a.
L e t a o ( x , p , n ) b e t h e p r i n c i p a l symbol of a . O b v i o u s l y , 0 h a s t h e reduced p r i n c i p a l symbol a o, a s w e l l . B e s i d e s , i t i s e a s i l y
seen t h a t
0 (3.12.75) a,,(x,n) U
=
1
l i m ( h V 2 l i m (ECh)” h+O E O ’
a ( x , ~ , h , n ) )=
3. Singular Perturbations on Smooth Manifolds without Boundary
394
Furthermore, introduce the principal quasi-homogeneous symbol aO0(X,n) by the formula: (3.12.76)
aOO(x,n) = lim h
v +v 2 3
ao(x,h,n).
h+O
It is quite obvious that if an elliptic symbol a E F " ( U ) a
of order
V
has
reduced symbol, then its reduced principal symbol and principal quasi-
homegeneous symbol are both elliptic of order v 2 and V2+v3, respectively, in the sense of Definition 3 . 1 2 . 3 1 . The global version of the previous definitions and Proposition 3 . 1 2 . 4 1 in the case of U = Bn is the same, as in the case of the symbol classes Lv(lRn)
,
-
FV(IRn)
,
yet considered in this section.
Remark 3 . 1 2 . 4 4 . It is easily seen that if a E F v ( U ) is elliptic of order v , then its principal symbol a ( x , o , n )
0
satisfies the inequalities:
'd (x,E,h,S) E U
(O,EO1
X
X
(O,hol
Bn,
where q(x) is a strictly positive continuous on U function. Our aim now is to establish two-sided a priori estimates for elliptic difference singular perturbations with their symbols in F " ( U )
v-
or F (Rn)
(global version) and to outline the corresponding parametrix construction. A
difference singular perturbation op a E op
F"(U)
is defined by its
symbol a E F u ( U ) as in ( 3 . 1 2 . 2 ) . using the discrete Fourier transform. It is said to be ezziptic of order v , if its symbol is elliptic of order v . Example 3 . 1 2 . 4 5 . The symbols a.(x,E,h,n), x E 1
3.11.33
u c
n a n d the symbol
E
2
4 151
+1<12
from Examples
are elliptic of order ( 0 , 0 , 2 ) and ( 0 , 2 , 2 ) , respectively. The
corresponding elliptic difference singular perturbations are finite difference approximations of the differential singular perturbations 2 l + s q(x)2D:
and E ~ A ~ - A ,respectively, where, as usual, Dx = -id/dx and A
is the Laplace operator. Proposition 3 . 1 2 . 4 6 .
The eZZipticity of a difference singular perturbation in op F V ( u ) is an
invariant property w i t h respect
to
the autornorphisms of
<.
3.12. Elliptic Singular Perturbations Proof. -
395
Indeed, this statement is an immediate consequence of Proposition
I
3.11.23.
Proposition 3 . 1 2 . 4 7 .
L e t a E F V ( u ) be e l l i p t i c of order v. Then f o r E ~ ,ho s u f f i c i e n t l y small and R s u f f i c i e n t l y large h o l d s : v
-v
ja(x,e,h,hS)I Zq(x) (E+h)
(3.12.78)
2<~<>v3, V x V h
E (O,ho],
V
E U,
E (O,E~]
V E
5 E mn with 151
b R,
where q(x) > 0 i s continuous f u n c t i o n of x E U. Proof. -
Let a (x,p,n) be the principal symbol of a, so that ( 3 . 1 2 . 7 7 ) holds, 0
on the one hand, and, on the other hand, one has:
with some li < v such that ~ ~ + < l v i1+v2 ~ and with some continuous q 1 ( x ) > 0, V x E U, as it follows from ( 3 . 1 2 . 6 7 ) .
Now, combining ( 3 . 1 2 . 7 7 ) that ho,
E~
and R
-1
with ( 3 . 1 2 . 7 9 ) , one gets ( 3 . 1 2 . 7 8 1 ,
provided
are sufficiently small.
Corollary 3 . 1 2 . 4 8 .
L e t a E F V ( u ) be e l l i p t i c o f order v. Then f o r (s;,s;)
w i t h any
= (vl,s;,s;)
s'
E m2 one has: -V
(3.12.80)
V x
E U,
-v
S'
(/a(x,E,h,hSj+(E+h) '
3,
2 q(x) (E+h)
v 3 '<<> 2<E<>v ,
E (O,E~], V h E (O,hol, V 5 E Rn ,
provzdea t n a t cO arid ho are s u f f i c z e n t l y smal2; h e r e , a s p r e v i o u s l y q(x) > 0
is some continuous f u n c t i o n on u. 2 R , then ( 3 . 1 2 . 8 0 )
Indeed, if
]<]
follows from ( 3 . 1 2 . 7 8 ) , >
6 R one can always find a constant CR , s ' , v
s'
s'
< p 2<E<>
V
L
cR , s ' ,v<'>
V
2 < ~ < >3 , V
<
and for
0 such that
with 151 5 R .
Proposition 3 . 1 2 . 4 9 .
L e t a(E,h,hC) E F V ( m n ) be e l l i p t i c of order s and any s' E
v
=
(vl,v2,v3).
~ , that a 1v s t h e r e e x i s t s a c o n s t a n t c ~ , such
Then f o r any
3. Singular Perturbations on Smooth Manifolds without Boundary
396
prcvided t h a t Proof. -
As
E~
and h0 a r e s u f f i c i e n t l y small.
a consequence of Parceval's identity, (3.12.81) is equivalent with
the inequality
with some other constant
C'
s,s"
the latter being equivalent with (3.12.801,
since <(c+h)<> can be estimated from below and above by
Remark 3.12.50.
If (3.12.81) holds for an elliptic difference singular perturbation Op(a) Of order v with some s ' < s such that s ; < sl, then, in fact, the term
I I.'
( s ' ) ,E,h
on the right hand side of (3.12.81) can be deleted.
Lemma 3.12.51.
For any (s2,s3) E R and (E,h) E (0,tl
x
(0,41
the following inequality
holds:
6
c<<
5-11
= <(h,h<) and where c and N~ depend only on s 2 and s3. E Proof. One checks easily that -
where 5
where 7
5
is the gradient with respect to 5 .
Further, denoting the left-hand side in (3.12.82) by g(E,h,E,n), the inequality (3.12.83) and the Lagrange formula for the difference on the left-hand side of (3.12.82) yield:
3.12. Elliptic Singular Perturbations p1 P to estimate << >
X
X
P1 = s -1 and with 2
p2 =
s3 or
p2 =
397
with
p
= s
1 2 OK s -1, and using also the obvious 3 11
inequality
N~ = maxI Is2-l/+Is31,ls21+Is3-1l},
noticing also that (E+h)
n
Lemma 3.12.52.
-
X
[O,+]
.
L e t R(x,~,h,hC)E F’ (mn) Then f o r any s = (sl,s2,s3)E n3 and 1 .o u n i f o m Z y w i t h r e s p e c t t o E E lo,+], h E (o,?] h o l d s : -S
(3.12.85) [ E
s2 s3 <(E+h)Dx,h> ,Op(R)]
1
:
H (s+iJ-e2), ~ ,(<) h
-f
where [A,B] is t h e c o r n t a t o r of o p e r a t o r s A and B and where e2
H ( 0 ),E,h(<)’ = (O,l,O). -
Proof. We recall that a symbol R(x,~,h,hS)is in the class F’ (Rn) if it 1r o is in the class F’ (Rn)and there exists a symbol Rm(~,h,hF)E F’ (Rn) such 1,o 1 no that ( E + ~ ) ’ ~ < C > - ~ ~ < E ~ > - ’ ~ ( R -as R ~a)function , of x , belongs to S(IR:) uniformly in ~ , andc. h F u r t h e r m o r e , w i t h o u t r e s t r i c t i o n o f generality, one canassume that h R_(~,h,hc)E 0. Denote by KE the commutator on the left-hand side of
kt
its difference Fourier transform, i.e. the operator, (3.12.85) and by -h u = F ( Khu ) . It is easily which acts according to the formula: K,.? x+C,h x+S,h E . a farr.ily of integral operators with the kernels seen, that K-h 1s
i:
Tn 5rh
Tn
X
Ilth
+ C
given by the fornula:
(3.12.86) k(E,h,E;, n ) = -S
=
(E+h) 1[<
3
-s313h(5-n.~,h,hn),
-h where R (S,E,h,hn) = Fx+5,hR(x,~,h,hn). -h with the kernels Introduce the family of integral operators S
s
:
Tn Srh
X
Tn
Qrh
-L
C, given by the formula: 1-s
(3.12.87) s(E,h,S,rl) = ih,E(~,h,5,n)<< n> where k is defined by (3.12.86).
-’2
2
<(E+h)Cn>
-s
-’3
3
3 . Singular Perturbations on Smooth Manifolds without Boundary
398
Obviously, (3.12.85) will follow if one shows that uniformly with respect to (E,h) E (3.12.88)
-h
Sc
: L
2
(0,tI
n (T5,h)
X
-f
(0,tl
holds:
2 n L (T5,h).
1-1
Since (Eth) 1<5>-v2<~5>-’3R, as a function of x, belongs to S (R:) -h uniformly in c,h and 5 , one has for R = Fx-t(<-n),hR with any integer N > 0 the following inequality (see (1.3.10)):
where the constant CN may depend only on N. Now, (3.12.87), Lemma 3.12.51 and (3.12.89) with N = N0+n+2 (No being the same as in Lemma 3.12.51), yield:
where the constant C does not depend on E,h,S and rl. Hence, Lemma 3.12.8 and (3.12.90) yield (3.12.88), and that proves (3.12.85). Lemma 3.12.53.
-
(mn)be such (s),E,~ h = {x E mn 1 /x-xol < 61, V (e,h) E ( 0 . ~ ~X 1 (O,hol. t h a t supp u c B 6 txo Then f o r any s ‘ E a v and for 6,e0,h0 s u f f i c i e n t l y s m a l l , t h e r e e x i s t s a 1 s c o n s t a n t c which m y depend o n l y on s , s ’ , 6 , ~ ~ , such h~, that L e t a E Fv(mn) be e l l i p t i c o f order v and l e t u E H
(3.12.91)
1 i u l 1 ( s ) ,E,h 5 V
E
E
C(l lOp(a)ul 1 (s-v),E,h t
(O,E~], V h
=
1
(s‘)
,~,h)~
E (O,hol, V u E H(S),E,h (E$,
__ Proof. Introduce the notation: ao(x) r0 (x) = a0 (x)-a0 (x0) , r(x)
lul
=
SUPP U = B
ao(x,E,h.h6), a(x) = a(x.E.h.hE),
a(x)-a(xo), so that one has:
(3.12.92) a(x) = a (x )+rO(x)+r(x).
0 0 Since r(x) E F’(mn) , Proposition 3.1 .34 yields:
=< (3.12.93) )/Op(r)u// (s-V),E,h
CJ/uJ
(s+v-v), ~ , h .
m
Now, let + & E CO(B26,xO) be such that $6(x) 6 Denote r0 (x)
=
6,xo
5
1, V x E B
+6(x)ro(x). Lemma 3.12. 52 yields for each
(36)/2 ,xo
3.12. Elliptic Singular Perturbations
=
Op(r6)(E+h)s - v 2<(c+h)Dx.h >
s
399
-v 3u+K:u,
with some family of operators Kh which satisfies the inequality
where the constant
C6
may depend only on 6 and, as usual, e2 = (O,l,O).
6
Since r (x) is a smooth function of x with its support in B
6
26,x and 0
r (xo) E 0, one gets immediately, that 6
(3.12.96)
s
>
v - s '
I lOp(r )(E+h)
-v
x.h
2<(E+h)D > x,h
s -v
ulI
(O),E,h
'
where C does not depend on 6,E.h and u. By Proposition 3.12.49, one has
where C may depend only on s , s ' , ~ ~ , V h ~s' , E
a
V
1 s' Hence, (3.12.92)-(3.12.97) yield for 6 sufficiently small:
Since li <
ul+u2
V,
< v1+v2, one can use the interpolation inequality
(2.8.18) (with 6 1 sufficiently small instead of 6) in order to estimate
I I 'I Order
1 1 . 1 1 (s),E.h with a small coefficient (of 1 1 . 1 I ( ~ 1 ) , ~ ,with h a large one. Applying the same one, finally, gets (3.12.91). I estimate I 1 . 1 I (s-e,) ,E,h'
(s+~-v) ,E,h '1) and Of
inequality to
by the sum of
Remark 3.12.54. If a(x,s,h,hS) E
supp u
C
Rn\B
R,O'
v n F ( m ) is elliptic of order v and if u E H
(R;) , ( s ) ,E,h V E,h, with R sufficiently large, then the same argument
3 . Singular Perturbations on Smooth Manifolds without Boundary
400
as in the proof of Lemma 3.12.53, shows that with some constant C (which may depend on R ) holds the same inequality (3.12.91),V (E,h)E ( O , E ~X] (O,ho], V U€H(~),~,~(<) , supp u c Rn\BR,O.
Indeed, with 6 = R-l and
ro(x) = ao(x)-ao("), (3.12.96) is still valid, since a0 (x)-ao(m), as a
.
function of x E Rn , belongs to S(IRn)
Now, we are in a position to prove the following statement: Theorem 3.12.55. L e t a(x,s,h,hE) E FV(mn) be gZobaZZy e l l i p t i c of order v (i.e. (3.12.72)
holds w i t h some q > 0 , which does not depend on x E mn). Then f o r each S,S'
with
(3.12.99)
< s,
s'
a constant
C,
I
IUI
s' E
alvs, and
for ~
~ s u f,f i c ihe n t l~y s m a l l , t h e r e e x i s t s
which may depend on s , s ' , E ~ , ~such ~ , that
I ( s ),E,h 5
C(l
I
/Op(a)ul ( s - v ) ,€,h +
V E E (0,c01,
n Proof. Let Uo = R \Bl,s,o,
I lul I ( s ' ) &,h),
V h E (O,hol, V u E H(S),E,h
.
U j = Bs,x,, 1 Ij 2 J, { UJ, }0-1< , < J being a finite 3
covering of R n , where the parameter 6 will be chosen later on. Let
{ $J, IO=J= < . < J be the partition of unity subbordinate to the covering
{U,} 3 O<j<J'
so that one has:
u
x
=
$ju,
v
u.
OSjSJ
If 6 is sufficiently small, then by Lemma 3.12.53 and Remark 3.12.54, one has:
Applying Lemma 3.12.52 and the interpolation inequality (2.8.18), one can write down:
Furthermore, using Proposition 3.11.34 (or again Lemma 3.12.52), one gets the estimate:
3.12. Elliptic Singular Perturbations Now, (3.12.100)- (3.12.102) yield (3.12.99) .
40 1
I
Now, we prove the necessity of the ellipticity condition in order to have the estimate (3.12.99) for a E FU(zn)
.
Theorem 3.12.56.
-
If f o r a E FU(nn) t h e a p r i o r i e s t i m a t e (3.12.99) h o l d s w i t h some s'
s ' is ,
E a1vs, t h e n a(x,E,h,hS) is g l o b a l l y e l l i p t i c o f order v.
Proof. Let
ri
E
'{O})
(T:
.
and let $ E C;(Bn)
We are goinq to substitute into (3.12.99) the functions (3.12.103) u = $(x) exp(ih-l<x,rp and to let
0, h
E -t
0 under the condition that the ratio p = h/E is a
-f
given constant, p E R+. Applying (3.11.24) to Op(a)u with u given by (3.12.103) and with p =
h/E a given positive constant, one gets the following asymptotic
formula: (3.12.104) a(x,e,h,hD)u = -(U
= h where a .
+v
1
)
2
-V
'a (x,p,ri)$+hy$(h,x)1 , h 0
exp(ih-l<x,n>)[ (l+p-')
is the principal symbol of a, y = v 1+U 2- ( ! J 1+ p 2
being the same as in (3.12.67), and where
[ l$(h, - )
I lo
>
+
0,
0, p = (p1,p2,u3)
is uniformly bounded
with respect to h E (O,hol. Furthermore, a straightforward computation (or one more application of (3.11.24)) shows that with any u E R 3 the following asymptotic formula holds f o r u given by ( 3 . 1 2 . 1 0 3 ) :
Therefore, (3.12.99), (3.12.104), (3.12.105) yield: (3.12.106)
Iw(ri)
I " 2
3 1181 ! o 5
CI
lao(-,P,ri)qIIo+C+h6 ,
where C is the same constant as in (3.12.99), may depend on $ and 6 = min{y,sl+s2-s;-s;,l} s'
E alVs, i.e. s1+s 2
> s'+s'
1 2 '
>
C
$
is some constant, which
0, given that s ' < s ,
3 . Singular Perturbations on Smooth Manifolds without Bouliday
402
Now, letting
0, h
E -+
-f
0, p = h/E being a given positive constant,
(3.12.106) yields:
the constant
C
here being the same, as on the right-hand side of (3.12.99).
Thus, (3.12.107) yields (3.12.72) with q(x)
S
C-l > 0, i.e. a is
globally elliptic. Examples 3.12.57. 1”. Let q(x)
E
m
C
(U), U _C R1 and let q(x) > 0, V x E U.
The symbol b(x,h/E,hE),
is elliptic of order
I ,
(0,0,1) and it is globally elliptic of the same
order if q(x) 2 qo
0, V x E
=
(in the case U = R1 , in addition, it is 1
assumed that q(x) = q_+q’(x) with q‘(x) E S(R
))
.
The corresponding difference singular perturbation b(x,h/E,hD) is an approximation by finite differences of the elliptic singular perturbation (l+iEq(x)Dx) E op s”(u), w = (o,o,I). Denote by b*(x,h/E,hD) the formal adjoint of b(x,h/e,hD).
The difference
singular perturbation (3.12.109) a(x,E,h,hD) = b*(x,h/E,hD)
0
b(x,h/E,hD)
is elliptic of order u = ( 0 , 0 , 2 ) , a E Op F W ( u ) , its principal symbol being ao(x,P,n)
=
l2
(b(x,p,q)
with b(x,p,n) given by (3.12.108); furthermore, the
difference singular perturbation (3.12.109) is a three point approximation by finite differences of the formally self-adjoint differential singular perturbation a(x,E,D) = (1-iED q(x)) (l+ieD q(x)) E Op S ( o ’ 0 ’ 2 ) ( U ) .
Besides,
if U is a finite interval, say U = (O,l), then the equations a(x,s,D)u = 0, 1 x E U and a(x,E,h,hD)u = 0, x E Uh = &$,fl U, have the same asymptotic x ’ E a U , in the following sense: solutions uo = exp(-jx-x’//(q(x’)~)), supla(x,~,D)u x o XEU with some constant C
I
S CE,
sup[a(x,E,h,D)u,-l6 C(E+h) XEU
0 which depends only on q(x).
Notice that a(x,y,~,h,hS)= b(y,h/E,hS)* b(x,h/E,hE) with b(y,p,n)* the complex conjugate of b ( y , p , n )
is the amplitude of the difference
sinqular perturbation (3.12.109) (see Remark 3.11.20).
3.12. Elliptic Singular Perturbations
403
The difference singular perturbation b+(h/E,hD) with the symbol -1 = l+p (exp(in)-l), b+ E F(of081)(~) , is a non-elliptic
b+(p,n)
approximation by finite differences of the singular perturbation l+i-ED. Indeed, b+(p,n) being also the principal symbol of b+(h/E,hD), one has: b+(2,1~)= 0 . Notice that b-(h/E,hD) = l+p-'(l-CI-') operator to the left) whose symbol is b- ( p , r l )
=
(with 0-1 the shift l+p-' (l-exp(-iq)) is an
elliptic approximation of l+iED. Consider a family of difference singular perturbations: t [0,1] 3 t + a (~,h,hD)defined by their symbols
2'.
(3.12.110) at (~,h,h5)= -~l
*
0
It is quite obvious that a (&,h,hS) = ir; (l+isr;): 1 E i<*(l+(E/h)(exp(ihE)-l)) is not elliptic, while a ( ~ , h , h < ) = i<(l+iEr;*)f 5
ir;(l+(-E/h) (l-exp(-ihE))) is elliptic of order (O,l,l),
It is easily seen, that for each t E (f,1] the difference singular perturbation at (~,h,hD)with the symbol (3.12.110) is an elliptic approximation of -ED2+iD, while for t E [O,f1 it is not elliptic. However, the approximation a t (E,h,hD) = --ED D* +(i/2) (Dx,h+D:,h) deserves x,h x,h some attention,since this is the only one of the family at which has the 2
accuracy O(h
),
as h
+
0. Besides, the ellipticity condition (3.12.72)
(with q(x) E q > 0) is only violated at p find a constant q ( p
0
t
t
q(po)
I w ( n ) I
where ao(p,rl) = a (l,p,n) = -p of a f
m,
i.e. for each
po >
0 one can
such that
)
(3.12.111) lao(p,n)I 2
t
=
-1
t/ rl
E
1
T
ri'
V p E (O,pol,
lw(n)I*+iRe w ( n ) is the principal symbol
. Indeed, since ao(p,n) t # 0, v
p
E (O,pol
'd
n
E (T:{O})
and -1
1
lao(p,n) f I 2 Cn for n 0 , so that infimum of
1
since a-(=,n) = iRe w ( n ) = 0 = sin n and the corresponding difference operator (i/2)(D +D* ) is a x,h x,h non-elliptic approximation of iDx, the ellipticity condition being violated On the other hand, it can't hold for
at n
po =
m,
= T.
The last example motivates the following
3. Singular Perturbations on Smooth Manifolds without Boundary
404
Definition 3.12.58.
A symbol a E F v ( u ) and t h e corresponding d i f f e r e n c e s i n g u l a r p e r t u r b a t i o n m there o p ( a ) a r e s a i d t o be weakZy e l l i p t i c of order v i f for each p o e x i s t s a continuous s t r i c t l y p o s i t i v e f u n c t i o n q(x), which may depend on po,
such t h a t t h e p r i n c i p a l symbol ao(x,p,rl) of a s a t i s f i e s t h e c o n d i t i o n
(3.12.72) w i t h
pE (O,pol.
A symbol a E Fv (mn)
(and t h e corresponding d i f f e r e n c e s i n g u l a r p e r t u r b a t i o n )
i s s a i d t o be g l o b a l l y weakly e l l i p t i c , i f i n (3.12.72) one can choose as q(x) a p o s i t i v e c o n s t a n t , which may depend on p o .
Using the previous argument, one proves in exactly the same way the following statement. Theorem 3.12.59.
v n The a p r i o r i e s t i m a t e (3.12.99) holds f o r a E op F ( R ) and for p E ( O , p o l , V p 0 < m , w i t h a c o n s t a n t C, which may depend on po, i f f a i s g l o b a l l y weakly e l l i p t i c . For elliptic difference singular perturbations the statements, similar to Proposition 3.12.16 and Corollary 3.12.17 are valid, the recurrence procedure for constructing the symbol of the left and right parametrix being exactly the same as in (3.12.341, (3.12.35). Furthermore, a statement similar to Theorem 3.6.2 (or Theorem 3.11.5) can be used in order to produce a symbol, having the asymptotic expansion defined by the recurrence procedure as in (3.12.34), (3.12.35). One can also use (3.11.49) as a definition (asymptotically equivalent to (3.11.47)) of the product of two symbols in F
(U) = U Fv (U), which 1#O v 1,o leads to a slightly different recurrence procedure for constructing the
symbol of the left and right parametrix for the elliptic difference singular
the right parametrix the terms
-
-1 Z: ak (x,~,h,hg)where for k>O of this asymptotic expansion are defined
perturbations. Namely one seeks a(x,E,h,hg)-l
recursively as follows (for 15 1 sufficiently large) : -1 . a (x,E,h,hS) = a(x,E,h,hE)-',
(3.12.112)
and for the left parametrix one has to switch a and a-l k-lal
on the right-hand
3.12. Elliptic Singular Perturbations
405
side of the second line in (3.12.112). but one can also use the same recurrence procedure for both parametrices, since all the symbols are, anyway, defined modulo a symbol of order (v1,-m,v3)with some vl,v3 E R . An example will be considered in order to illustrate the paranetrix construction for elliptic difference singular perturbations just sketched above. Example 3.12.60. 2
Let a(x,E,h,hS) = 1 + ~, where A(x) =
1 lakj (x)1 1 l ~ kj,~
is symmetric n
.
m
positive definite matrix with C -entries a (x), x E U _C Rn Since kj (u)~ isf a ~polynomial ' in ( c , c * ) it is more convenient to use a E ~ ( ~ f (3.12.112) for constructing a parametrix for a(x,~,h,hD)E Op
F(0,0,2)
(u).
However, still making use of (3.11.49), we shall proceed differently in order to produce a parametrix for the difference singular perturbation 2 -1 a(x,~,h,hD)= 1 + ~ . We seek a parametrix a (x,E,p,hD) for x ,h- 1x,h E F(o'or-2)(u) can be represented a(x,~,h,hD)whose symbol a (x,~,p,hS) asymptotically as a sum (3.12.113) a-l(x,E,p,n)
-
C
E
k -1 ak (x,p,n),
E ' O
k2O
uniformly with respect to x on each compact K C C U, p 2 p 0 > 0 with any given p o and n E TnWrite the ker:el
of the integral o2erator a-'(x,E,p,hD)
in the for-:
Using (3.11.49) (or a straightforward computation), one finds:
Now the coefficients in the asymptotic expansion (3.12.113) will be determined by requiring r(x,E,p,pq) defined by (3.12.114) to satisfy the asymptotic relation: r(x,E,p,pn)-l
N
= O(E )
for any integer N > 0.
The latter requirement leads to the following recurrence formulae for the coefficients a;'(x,o,q)
in (3.12.113):
3 . Singular Perturbations on Smooth Manifolds without Boundary
406
-1 where a-l = 0.
Now, choosing a cut-off function $(t) and a sequence 6 . 7
J.
0 as in the
proof of Theorem 3.6.1, we can write the following formula for the symbol of a parametrix for a(x,~,h,hD): (3.12.116) a-'(x,c,h,hS)
C
=
-1 k -1 + ( 6 k ~ )E ak (x,h/~,hS),
k>O with a ; '
defined by recurrence formulae (3.12.115).
It is left to the reader to check that for any integer N > 0 one has:
-1 (3.12.117) a(x,~,h,hD)o a (x,E,h,hD) - Id = E ~ R ~ ' ~ N '
(i.e., if A(x) = Am+A'(x) with A'(x) E S(lRn)) and is globally elliptic (of -1
order v = (0,0,2)),then a
(x,~,h,hD) constructed above is, in fact, a m
quasi-inverse operator for a with accuracy O ( E
),
that is (3.12.17) holds
for any integer N > 0 with R E P h which is uniformly bounded with respect to N E ( 0 .~ ~ 1 h ,E (O,hol fromH(s),E,h(<) into H (S-(O,O,-N)) ,E,h VsEz33.
'<'
E
I
The parametrix construction discussed in Example 3.12.60 can be extended to the class of elliptic singular perturbations of order ( O , O , v ) , whose symbol a E F(oronv) (U) has the property r := (a-ao) E F(-l'o'v-l) (U) ,
a (x,h/~,hS)being the principal symbol of a. 0
Indeed, one can use (3.11.47) (or, equivalently, (3.11.49)) for defining the (non-commutative) product a'b of two symbols a and b in F1,o(U) = U v
Fv1,o (U),
as a symbol a * b with the asymptotic expansion (3.11.47)
(or, equivalently, (3.11.49)). Now, using again recurrence procedure (3.12.115) with a (x,p,rl) instead of l+ , 0
-
-1
one can find a
symbol ai'(x,E,p,n) such that 1-a (x,h/~,hE) . a (x,~,h/~,hS) = 0 N = O ( E
3.12. Elliptic Singular Perturbations
a(x,E,h,hS)
-1 -1 a . (x,~,h/~,hS) = l+r(x,~,h,hS) - a o (x,~,p,hS)
*
so that modulo an operator in Op F(-"'o'-'")
( U ) one has:
1 r0
(3.12.118) Op a
o
407
Op a ; '
=
Id + Op r
-1 Op . a
o
.
Since ~~l~ := op r op a;1 E op F(-lr0r-l) ( U ) , the operator (1+KEfh)-l 1 ,o exists and can be represented as a convergent Neumann series, so that E Op
(I+K'lh)-'
Fo
1 to
Let l+k-'(x,e,h,hS)
(U).
be the symbol of (I+KEfh)-l.
Since (3.12.119) Op a
-1
o
Op a .
1 Op(l+k- ) = Id
o
and there is in this case a one to one correspondence between the symbols and the operators in (3.12.119) (modulo a symbol of order ( - - , O , - - )
),
one
gets the conclusion that the singular difference perturbation with the -1 (1+k (x,e,h,hS))is a parametrix and at the same symbol a-'(x,E,h/E,hS)
-
0
time a quasi-inverse with accuracy
N
O(E ),
V N
> 0, for the elliptic
singular perturbation a(x,E,h,hD) above. Finally, we outline the extension of the elliptic theory Of singular perturbations in spaces ff
( s ) .E
to elliptic differential singularperturbations
in analoguous spaces with L -structure, 1 < p <
P
-,
which will be defined
below. We start with a result on Fourier multipliers in L (El") due to Hsrmander.
P
Definition 3.12.61. A tempered d i s t r i b u t i o n f E L
P
S'
( R ~ )is s a i d t o be a Fourier m u l t i p l i e r i n
(mn) if F ~ : ~ ~ F E~'L~ uand t h e r e e x i s t s a c o n s t a n t M such t h a t
The l e a s t c o n s t a n t x such t h a t (3.12.120) holds f o r a g i v e n f E s ' (Rn) and p E (1,~) is denoted by M (f) and t h e s e t of a l l Fourier m u l t i p l i e r s i n
P
L (mn)
P
is denoted by M
P
.
Our aim is to present the proof of the following result of HcSrmander: Theorem 3.12.62.
L e t f E L _ ( R ~ ) and l e t B > 0 be a c o n s t a n t such t h a t (3.12.121)
1
fRS15 I <2R
iRla'Daff~)fdS5 B2Rn,VR > 0 , V a with la1 5 [n/21+1,
5
3. Singular Perturbations on Smooth Manifolds without Bounda y
408
where [n/2] i s t h e i n t e g e r p a r t of n/2. Then f E M , V p E ( 0 , ~ )and, moreover, P (3.12.122) M ( f ) 5 C B, P P,n
where t h e c o n s t a n t c
depends only on i t s s u b s c r i p t s .
P,n The following consequence of Theorem 3.12.62 will be used
consistently later on. Corollary 3.12.63.
L e t f E L_(B~) and l e t
Bo > 0
be a c o n s t a n t such t h a t f o r a l l m u l t i -
i n d i c e s a w i t h l a / 6 [n/21+1 h o l d s : (3.12.123) IDaf(S) I 5 B o / t I -
v 5 E Rn\IO}.
5
Then f E M
P'
V p E (1,m) and, moreover,
(3.12.124) M (f) 5 C' B P p,n 0 '
where t h e c o n s t a n t C'
depends o n l y on i t s s u b s c r i p t s .
P#n Proof of Corollary 3.12.63.
It is easily seen that if f E Lw(Rn) satisfies (3.12.1231, then it satisfies (3.12.121), as well. Indeed,
the right-hand side of the last inequality being defined for n the limit for (n-2lal)+ 0.
=
21al as
I
With f E L_(Rn) satisfying (3.12.123) for la1 6 n the result stated in Corollary 3.12.63 was first established in [Mich., 1
, 21.
The proof of Theorem 3.12.62 requires a considerable effort and will be split into several steps. We start with the following statement: Theorem 3.12.64.
L e t p E ( 1 , ~ )and l e t p-'+p'-' (3.12.125) M ,(f) = M (f). P P
=
and 1. If f E M then f E M P P'
3.12. Elliptic Singular Perturbations
409
Proof of Theorem 3.12.64.
,
With u E S(lRn)
v E S ( R n ) , Parseval's identity and Holder's inequality
yield:
1
6 M (f) Iul
P
1 L (Rn)
P where, as usual, the upper
*
1 IVI I
(Rn) ' P' stands for the complex conjugate. L
Since S(lRn) is dense in L (Rn) , one can choose a sequence
P
{uklkZl, uk E S(Rn) which converges in L (Rn) to
P
1
/Fg,xf
p'-2 -1 (FS+xf Fx+Sv) * E L (Rn) as k Fx+SVI F Furthermore, one finds
-f
-.
Hence using (3.12.126) (with uk chosen as above) and the last two formulas, one gets
so that f
E M
and M P'
P'
(f) <= M ( f ) .
P
I
Now, by exchanging p and p ' , one gets (3.12.125).
Next we shall prove an interpolation result due to Riesz and Thorin (see [Zygm.,
1
1.
Theorem 3.12.65.
Let 1 Let T
<
p < r < q <
:
L (W") P
-f
m
and Zet
e E
L ( I R ~ ), T : L
suck c o n s t a n t s t h a t
P
9
-1
(0,1) be such t h a t r-l = ep-1+(1-8)q (xn)
+
L
4
(w") and l e t c
P
(T),
c
9
(T)
be
.
3 . Singular Perturbations on Smooth Manifolds without Boundary
410
Then
the following version of Phragmen-LindelBf's
For proving Theorem 3 . 1 2 . 6 5 maximum principle is needed: Theorem 3 . 1 2 . 6 6 .
E c 1 a- 5 Re z S a } and l e t f : Z + c1 be continuous and bounded on 2 and a n a l y t i c on t h e i n t e r i o r of 5. If
Let Z
= { z
t h e n one has: (3.12.1301
/f(z)( 5 M ,
V z
E 2.
Proof of Theorem 3 . 1 2 . 6 6 . If lim y+im
max a 5xSa
-
I
/f(x+iy) = 0 ,
+
then one gets ( 3 . 1 2 . 1 3 0 )
by applying the classical maximum principle to
f(z) in a rectangle { a - 5 Re z 6 a + , 11m z I 5 A} with A sufficiently large. Otherwise, one applies the classical maximum principle to the function f(z) exp(6z2), 6 > 0, and let 6
+
0.
I
Corollary 3 . 1 2 . 6 7 .
L e t Z and f be a s i n Theorem 3.12.66 and assume t h a t
Then for z, = ea- + ( 1 - 6 ) a + + iy,
e E
( 0 , 1 ) , holds:
Indeed, one has only to apply Theorem 3 . 1 2 . 6 6
to the function
3.12. Elliptic Singular Perturbations
41 1
i n order t o g e t (3.12.131). Proof o f Theorem 3.12.65. Denote by S t ( i R n ) t h e s e t of f u n c t i o n s on R n which t a k e o n l y a f i n i t e number of complex v a l u e s and v a n i s h o u t s i d e a b a l l w i t h s u f f i c i e n t l y l a r g e r a d i u s . A s a consequence o f t h e d e f i n i t i o n of L e b e s g u e ' s i n t e g r a t i o n , with 1 5 r <
S t ( i R n ) i s d e n s e i n any L ( W n )
m.
F u r t h e r , s i n c e t h e d u a l s p a c e ( L ( E n ) )' Lr, (Rn)
,
where r-'+r'-'
(3.12.133)
IIfl
IL
= 1 , one h a s r f o r e a c h f ErL
sup gESt(R
=
r
of L ( R n ) c o i n c i d e s w i t h
,I
IL
:
11 f ( x ) g ( x ) * d x l . mRn
=I
[gl
(Rn)
r' For z E C (3.12.134)
, 0 ~
5 Re z 5 1 , d e f i n e t h e f u n c t i o n r ( z ) by t h e e q u a l i t y :
+ -1-2
2 1 _- r(z) p
q
and t h e f u n c t i o n r ' ( z ) as f o l l o w s : (3.12.135)
1 +-= r'(z) r(z)
For f E S t ( R n )
,
g
1.
E
S t ( R n ) with j l f l l L
the functions: fz =
= 1,
r l f l r / r ( z ) ei a r g f I
g2
=
lgl
]
= 1 introduce /glIL r'
r'/r'(z)
i arg g
(3.12.136) .O(z) =
1
( T f Z )(x). g Z ( x ) d x .
Rn Notice t h a t (3.12.137)
O(8) =
1
(Tf) (x).g(x)dx. RRn
C Rn Denoting by 1 ( x ) t h e c h a r a c t e r i s t i c f u n c t i o n o f a s e t A A
w r i t e f o r f and g i n S t ( l R n ) :
Hence,
,
one c a n
3. Singular Perturbations on Smooth Manifolds without Bounda y
412
where Xk E C
1
and a > 0 are some numbers, which are well-defined by f and k
9.
As a consequence of the last representation for @ z ) , it is bounded and continuous on the strip ? = { z E C1
1
0 5 Re z 5 1
Further for Re z = 0 one finds that f (x)Irdx) ''l
=
1,
and, in the same way, that 119iy/lL = 1 9'
Now, applying HBlder's inequality and using the second line in (3.12.127), one finds:
Using the same argument, one gets the inequality: (3.12.139) l@(l+iy)
I
5
c
P
v y E
(T),
W.
Applying Corollary 3.12.67 and using (3.12.1381, (3.12.139), one gets:
1
(3.12.140) l @ ( z ) 6 C (T)Re Z(Cq(T))l-Re
P
,
1 V z E C , O S R e z S l .
The definition of @ ( z ) by (3.12.136), (3.12.137) and (3.12.140) yield: (3.12.141) 11
En
(Tf)(x)g(x)dxl 5
C
P
e
(T) (Cq(T))l-e, V f,g E St(Rn)
.
Now, with V f E St(Rn) and using (3.12.133), (3.12.141), one gets (after taking the supremum on the left-hand side of (3.12.141) over all
and that completes the proof, since St(Rn) is dense in L (Wn) .
I
Corollary 3.12.68.
L e t p E (1,m) and l e t p-'+p'-'
=
1. If f E M
P'
then f E
M
q'
V q
such t h a t
min(p,p') 2 q I max(p,p'). Indeed, this is an immediate consequence of Theorems 3.12.64 and 3.12.65.
I
3.12. Elliptic Singular Perturbations
413
Next, i-'ormander's q e n e r a l resu1.t c o r x e r n i n a _ F o u r i e r m u l t i p l i e r s ( i n
L (lRn)) P proved.
,
which a r e l o c a l l y i n t e g r a b l e f u n c t i o n s , w i l l b e s t a t e d and
k f o r t h e s e t o f a l l FE:,f, 'd f P (F-l f ) = M ( f ) . I t i s q u i t e o b v i o u s , t h a t coincides p e x P P s e t of d i r e c t F o u r i e r t r a n s f o r m s o f f E M as w e l l . P' 1 n I n t r o d u c e t h e f o l l o w i n g c l a s s K o f f u n c t i o n s i n L loc(R) 1 k E Lloc(Rn) i s s a i d t o b e l o n g t o t h e c l a s s K i f t h e r e e x i s t a We s h a l l u s e t h e n o t a t i o n
write
E M and P with t h e
.A
s e t A C IRn
,
a neighbourhood
0 of zero i n
function
bounded
and a c o n s t a n t C > 0 s u c h
lRn
t h a t f o r a l l s > 0 holds:
i
(3.12.142)
/ k ( x - y ) - k ( x ) Idx 5 C ,
v
y E so,
s(Rn\A) where by sB i s d e n o t e d t h e s e t : SB = { y E En Let
1
I
y = sx, x
b e c o l l e c t i o n o f a l l open c u b e s i n lRn
p a r a l l e l t o t h e c o o r d i n a t e a x e s . Choosing I. a t t h e o r i g i n and s u c h t h a t I.
C
0,
1; 3 A ,
C
E
B}.
with t h e i r edges,
1 , I:
C
I
with t h e i r c e n t r e s
w e d e f i n e f o r each I
C
I
a n o t h e r cube I * c 1 w i t h t h e same c e n t e r as I s u c h t h a t u(I*)
=
) L I ( I )where ,
(p(I;)/u(I0)
u
i s t h e Lebesgue measure on IRE.
Lemma 3.12.69.
1
n
L e t k E K and l e t I c I and I* a s d e f i n e d above. Then f o r V u E L ( l ~) , supp u c I , i u ( x ) d x
=
0, t h e f o l l o w i n g i n e q u a l i t y h o l d s :
I
where t h e c o n s t a n t
c i s t h e same a s i n
(3.12.142),
and where :c*u stands
f o r t h e convolution of k and u. Proof o f Lemma 3.12.69. L e t z b e t h e c e n t r e o f I and d e n o t e s = u ( I ) / p ( I 0 ) .
J u(x)dx
Since supp u
= 0 , one f i n d s u s i n g t h e d e f i n i t i o n o f t h e c l a s s
I (k*u)( x ) / d x =
(3.12.144)
IR \ I *
K:
C
I and
3. Singular Perturbations on Smooth Manifolds without Boundary
414
f
2 C
1-2
ju(y+z))dy = C J ju(x)]dx. I for the shifts of the corresponding sets by
Here (Rn\I*)-z and 1-2 stand
I
vector z .
Next, we prove the following Theorem 3.12.70.
L e t k E K fl
f o r some p E (1,-I. Then t h e r e e x i s t s a c o n s t a n t P t h a t f o r aZZ u E L ~ ( R ~w)i t h compact support holds:
c1 such
One needs for proving the last theorem, the following Lemma due to Calderon and Zygmund (see, for instance, [Her. 1 1 ) : Lemma 3.12.71.
L e t u E L ~ ( R ,~ )1 IuI I L1 ( Rn) a s a sum: (3.12.146) u
=
> 0,
and Zet s > 0 . Then u can be represented
v+ Z w k' kt 1
where v and wk s a t i s f y t h e foZZowing c o n d i t i o n s : (i) v E L (Rn) , wk E L1(Rn) 1
(ii)
,
k = 1.2 ,...,
1
/v( + IIWkll 5 31lull L1(Rn) k2l L1 (Rn) L1(R
I
)
(iii) Iv(x)] 5 2"s azmost everywhere, (iv) f o r c e r t a i n d i s j o i n t cubes
c I holds: supp w
Moreover, i f u has compact support, then v and w
k
c
and
k 2 1, can be chosen
t o have t h e i r supports contained i n some f i x e d compact s e t . Proof of Lemma 3.12.71. Divide Rn into a mesh of cubes from >
.-lll~ll
, so
and denote by I s.
with volume of each cube
1
over every cube of
L1(R 1
this mesh becomes
t
I
that the mean value of ju(x)
c
S.
Further, divide each cube ofthe mesh into 2
. . those of them, over which
,Il2,.
Obviously, one has:
n
equal Cuke
1.1
the mean value of
is
3.12. Elliptic Singular Perturbations
415
1 lu(x) /dx < 2nsui(Ilk).
(3.12.147) sv(Ilk) 5
Ilk Indeed, if Ilk is obtained as a result of the subdivision of a cube I' from the mesh above, then, according to the construction, one has: SU(Ilk)
5
J Ilk
lU(x) Idx 2 1 lu(x) [ d x <su(1') I'
=
Znsu(Ilk).
Define
wherexI V x
(XI is the characteristic function of Ilk, i.e. xlk(x) E 1, lk
E Ilk and
xI
(x) 0 elsewhere. lk Next, making again the same subdivision of the cubes from the mesh
above, which are not among I ll,I12,..., one singles out those new cubes over which the mean value of lu(x) I is L s and extends the
..
I Z 1 ,I22,.
definition by (3.12.148) of w ~ ~ ( xto) these cubes. Continuation of this process leads to theconstruction of disjoint cubes I . and corresponding 3k
functions w . ~ ( x ) ;one rearranges them, for convenience, as a sequepce, 3
denoted again by Ik, wk(x), and defines v(x) by setting v(x) V x f! Q
=
=
u(x),
U I . It is clear, that (3.12.146) holds and (i) is valid, as k
k
well. For showing (ii), first notice that
1 (Iv(x)I+lwk(x) 1 )dx
5 31 /u(x)ldx.
I
k
k '
Indeed, the last inequality is an immediate consequence of (3.12.148) and the definition of v(x). Since the cubes Ik, k = 1,2,..., are disjoint, supp wk
J Rn\Q
Iv(x) [dx =
In R ' 9
one gets immediately (ii):
lu(x) ldx,
5 Ik
and
3 . Singular Perturbations on Smooth Manifolds without Bounda y
416
Further, (iii) follows from (3.12.147) if x E Q. On the other hand, if x $! Q, then there are arbitrary small cubes containing x, over which the mean value of lu(x) 1 is <
so that lu(x) I 5 s almost everywhere on mny?,
s,
and that proves (iii). Further, as a consequence of (3.12.148), one has:
1 wk(x)dx
= 0,
...,
k = 1,2,
I k
and summing up in (3.12.147) with respect to k, one gets that
since I ~ k,
=
1,2,..., are disjoint.
1
Proof of Theorem 3.12.70. N
Denoting
U(X) =
(k*u)(x) and using the decomposition (3.12.146) of u(x), one
finds that almost everywhere
and, thus, almost everywhere holds:
Applying Lemma 3.12.69, one gets
*
*
Further, denoting Q* = U I with Ik defined by Ik as above and using (iv) k from Lemma 3.12.71, one gets
N
Introducing w(x)
Z Iwk(x) 1 , one finds, using again Lemma 3.12.69 N
=
k> 1 and (ii) from Lemma 3.12.71:
3.12. Elliptic Singular Perturbations
417
This last inequality yields:
1
(3.12.150) u{x E Rn\Q*
I
1 w(x) t (4)s) 5 6Cs- /uII
Hence, it follows from (3.12.149), (3.12.150) that w(x) < ( f ) sexcept on a set of measure at most
1
The assumption k E
P
w th some p E (1,m) and (ii), (iii) from Lemma
3.12.71 yield:
Therefore, one gets,using the last inequality:
N
Since the measure of the set, where w(x) 2 s/2 is bounded by constant (3.12.151). one findsrusing the last inequality, that UIX E Rn
I
jc;(x)
I
> s
and that is precisely (3.12.145) with a = s . with C3 = (2C2)p+6C+u(I~)/~(10),
I Next, using Theorem 3.12.70
an argument of Marcinkiewicz (see [ Z , l ] )
and Corollary 3.12.68, the following statement will be proved: Theorem 3.12.72.
f o r a l l p E ( l , m ) , o r e l s e f o r no such p. P for some p E (1,m). It will be shown that k E fir,
L e t k E K. Then e i t h e r k E Proof. Let k E __
v
k
P
r E (1,pl. For u E L (Rn) define v
E Lr(Rn) , ws E L (Rn) such that u
I
u(x) = ws(x) if /u(x) > s and u(x) = vs(x) if lu(x)
I
u +v s s' 6 s, where s > 0 is a =
given fixed number. Obviously, w
has compact support, ws E L1(Rn) ,
I lwsl
5
I1uI ILr(Rn)
3. Singular Perturbations on Smooth Manifolds without Bounda y
418
and, moreover, one has:
E Lp(Rn), j lvs 1 ]
By a similar argument one finds:v
5
ll~ll
and moreover,
For a given Lebesque-measurable function f ( x ) denote by m(t) the measure of the set where I f ( x )
I
>
t. Then for each f E L (Rn) with q E 1 one has: 9
m
m
(3.12.152)
1 If ( x ) Iqdx
=
Rn
-1 tqdm(t) 0
=
q
tq-'m(t)dt. 0
Further, one has for each f E L (Rn) : 9
so that
(3.12.153) m(t) = u{x E Rn
I
I
/f(x) > t} S I ' t
If1
1'
L q The last inequality will be used for estimating Introduce
!Jaw = (3.12.154) uls(t)
VIX =
E Rn 1
V{x E
JZ(X) 1 > Rn 1 ] G s ( X ) 1
N
N
where, as previously, u = k*u, w
t), >
t],
N
=
k*ws, v
=
k*vc. I
Applying Theorem 3.12.70 to k E
Since by assumption k E M
bSlI
P'
p, ws E L 1 ( R n ) , one finds:
one has with some constant C
0:
, so that using (3.12.153) one finds for L (mn) P ! ~ ~ ~ defined (t) by (3.12.154), the following inequality: 5 CI
L (R")
P
3.12. Elliptic Singiilar Perturbations
419
N
Now applying (3.12.152) with f = u , one finds: m
Further, since Iw (x)[ 5 t, Iys(x) 1 5 t implies lu(x) 1 5 2t. one has: N
N
p0(2t) 9 ~ ~ ~ ( t ) + p ~ so ~ (that t ) ,the following inequality holds: m
(3.12.156) 1
Iy(x)lrdx
Eln
=
2rr 1 tr-lp0(2t)dt 9 0 m
5 r2r(7 tr-'uls(t)dt+ 1 tr-1u2s(t)dt). 0 0
These two integrals on the right-hand side of the last inequality will be estimated by setting s = t in the definition of w
and v
Using again (3.12.152) and (3.12.155). one finds:
By definition of w (x t m
1 0 m
p t x E IRn
.
3. Singular Perturbations on Smooth Manifolds without Boundary
420
Combining (3.12.156)-(3.12.158), one gets finally, that
with some constant C > 0 , and that proves k E Now, Corollary 3.12.68 yields: k E
fir, V
Mr,
V r E (l,p].
r E (1,m).
I
Now, one needs only two more lemmas in order to be able to prove Theorem 3.12.62. Lemma 3.12.73.
There e x i s t s a f u n c t i o n
+ E Cm(Rn) , supp $ c 15 E 0
Rn 1 0) 6 IS1 < 21,
such t h a t
c
(3.12.159)
+(2-k0 E 1 ,
v
5 E Rn\{Ol.
kEZ
Proof of Lemma 3.12.73. Let 0 2 0 be a function in Cm(R+) , supp 0 0 O(r) # 0, V r E [1//2,/2]. Defining
+ ( S ) = @ ( / 5 /( )
z
r
C
Q(2-klcl))-1,
v
I
f < r < 2), and let
5 E Rn\t01,
+(O)
=
0,
kEZ
it is easily seen, that + ( 5 ) satisfies (3.12.159).
a
Lemma 3.12.74.
L e t T be c d i s t r i b u t i o n i n S'(xn) and l e t p E (1,m). Then t h e following t w o c o n d i t i o n s a r e equivaZent:
(3.12.160)
and (3.12.161)
3.12. Elliptic Singular Perturbations where
c is a constant and, a s u s u a l ,
p-'+p'-'
=
1.
proof of Lemma 3.12.74. HGlder's inequality yields:
Proof of Theorem 3.12.62, The proof consists of three steps. First it will be shown that an approximation of f(6) with compact support is in M2. Next, it will be proved that this approximation belongs to the class K defined above by (3.12.142). Finally, one shows that it is allowed to take the limit of these approximations. Step 1. Let $ ( 5 ) be the function constructed in Lemma 3.12.73 and let fk(F)
=
f(5)$(2-kg). Using Leibnltz forxula, one can estkate the
derivatives of f ( 5 ) by those of f(5) : k
Hence, (3.12.1211, (3.12.162) yield:
0 5 j 5 2, are some constants which depend only on n. where C n,j'
Introducing gk(x) = (Fgixfk)(x), Parceval's identity along with (3.12.163) yields:
42 1
3. Singular Perturbations on Smooth Manifolds without Boundary
422
(3.12.164)
1 (1+22k 1x1 2) X Igk(x) I 2dx
5
Rn
wiiere we have denoted X = [n/2]+1 and where C n,j' which depend only on n .
j
=
3,4, are some constants,
Using Cauchy-Schwarz' inequality and (3.12.164), one finds: (3.12.165)
lgk(x) Idx 5
mn
where, as previously, Since fk(5)
=
X = [n/21+1 and C depends only on n. n,5 (F g ( 5 1 , one can estimate f ( 5 ) in the following x+S k k
fashion:
almost everywhere on IRn. Introduce
Since at each point 5 E Rn at most two functions f ( 5 ) do not vanish, k one gets, using 3.12.1661,the following estimate almost everywhere on Rn : IFN(<
I
5 2Cn,5B = Cn,6B.
-1 Hence, for GN(x) = (FS+xFN) (x), one finds
that is to say that G
E M2, FN E M2, V N
=
O,l,
:
L2(Rn)
... and, moreover, the
L (Rn) is bounded 2 B, where C depends only on n. uniformly with respect to N by C n,6 n,6 Step 2. We show that G E K , V N, by estimating the integral norm of the convolution operator GN
+
N
1 IGN(x-y)-GN(x)Idx, (x(>2t
for Iyi 5 t.
First, we esti-nate such ar. integral for each gk(x) in the definition
3.12. Elliptic Singular Perturbations
423
of GN(x). Dropping the additive term 1 between the paranthesis on the left-hand side of (3.12.164), and using again Cauchy-Schwarz' inequality, one finds:
with the same X = [n/2]+1 and some constant
which depends only on n. n.7 Further, using (3.12.167), one gets for IyI < t the estimate: C
k Since n/2-X = n/2-[n/2]-1 5 -f, (3.12.168) can be used when 2 t t 1. k If 2 t < 1 , then one uses Cauchy-Schwarz' inequality, Parceval's identity and (3.12.162), respectively, in order to estimate the integrals on the left-hand side of (3.12.168). One has, using Cauchy-Schwarz' inequality with X = [n/2]+1:
Further, Parceval's identity and (3.12.162) yield:
Hence, using (3.12.169) and the last inequality, one gets:
3. Singular Perturbations on Smooth Manifolds without Bounda y
424
N
where t h e c o n s t a n t C
depends o n l y on n.
Combining ( 3 . 1 2 . 1 6 8 ) ,
( 3 . 1 2 . 1 7 0 ) , one f i n d s f o r G
I:
=
Ikl6N
I
/ G (x-y)-GN(x) IdxSCnB
)x/>2t
= {y
E Rn
1
min{t2k,(t2k)n'2-X1
C
= C(l)B.
-m
The l a s t i n e q u a l i t y shows t h a t GN E
0
gk :
K with
A = { x E Rn
~
1x1 6 2 1 and
/ y / < 11 i n t h e d e f i n i t i o n of K . y i e l d s GN E
N o w , Theorem 3.12.72
fip,
V p E (1,m) a n d , moreover,
where t h e c o n s t a n t C may depend o n l y on p and n. Ptn T h e r e f o r e , FN E M V p E (1,m) and M (FN) = fi ( G ) . P' P P N S t e p 3. W e have t o show t h a t one may l e t N + S i n c e FN(C) + f ( < ) , V 5
m.
E Rn\{O}, p o i n t w i s e and
IFN(:)
I
i s bounded, one
.
h a s : F + f i n S ' ( R n ) F u r t h e r , s i n c e t h e F o u r i e r t r a n s f o r m i s an isomorN phism o f S ' ( R n ) o n t o i t s e l f , one h a s : GN + G = F-' f i n S ' ( R n ) , t o o . W e S+X
a r e g o i n g t o a p p l y Lemma 3.12.74 i n o r d e r t o show t h a t ( 3 . 1 2 . 1 7 1 ) h o l d s for N
+ m,
i . e . Lor G .
Consider t h e s e t S '
B
C
S'(Rn)
d e f i n e d as f o l l o w s :
v Obviously, S ; ' I '
-f
is a closed set i n S ' ( R n )
,
. Hence,
then
I (GN*u*v)( 0 ) I
+
I (G*u*v) ( 0 ) 1 ,
v
u , v E S(7Rn)
T h e r e f o r e , one h a s f o r a l l u , v E S ( R n )
(3.12.172)
I.
, s i n c e f o r each given u , v E S ( R n ) ,
(T*U*v) ( 0 ) i s a c o n t i n u o u s l i n e a r form on S ' ( W n )
S'(Rn)
Now,
u,v E S(Rn)
:
and Lemma 3.12.74 y i e l d :
and t h a t ends t h e p r o o f o f Theorem 3.12.64.
I
.
if GN
+
G in
3.12. Elliptic Singular Perturbations
425
The proof presented here is due to Hormander (see [Her, 11). Now, we introduce analogues H (IR~), p E (1, - I , of spaces ( s ) ,P.E H ( s ) ,€(Rn) with L -structure and point out their properties which are
P necessary for the development of the corresponding L -theory of elliptic P singular perturbations. LogicalLy this material belongs to Chapter 2, but,
since the derivation of the basic properties of the spaces
H
( s ) ,PIE
is
based upon the use of the Fourier-multipliers result stated in Corollary 3.12.63, it is included in this section. Definition 3.12.75.
For each
s =
(s1,s2,s3)E n3 and each p E (I,-) d e f i n e
and denote (3.12.174) H
3 d ( (s),P,E
=
1
Iu
u
A f a m i l y o f d i s t r i b u t i o n s U(E,. )
space
:
(O,Eo1
-f
: (O,E~] +
f l l u l I ( s ) ,P,E <
S'(RRn)
-1.
s'(nn) i s s a i d t o belong t o t h e
H ( s ) ,P( R ~ ), if
Remark 3.12.76. With s
=
(O,s,O)
the spaces H
( s ) rP
(Rn) are nothing else but the classical
Bessel-potential spaces, denoted usually HS (IRn) . They have the following
P
duality property: (Hs(Rn))
=
Hz,(mn) where p-lcp * - l =1 (cf., for instance,
P
[Tr, 11). Remark 3.12.77. It is easily seen that for each
s
E R n , p E ( 1 , ~ )the family
( 0 . ~ ~3 1E + H (an)is a family of Banach spaces, (s), P . E of course, a Banach space, as well.
f f
( s ),P
(Rn) being,
Now, an analogue of Proposition 2.1.8 will be stated and proved using the Fourier-multiplier result in Corollary 3.12.63. Proposition 3.12.78.
L e t s' 6 s , p E (I,-). Then t h e r e e x i s t s a c o n s t a n t c only on i t s s u b s c r i p t s , such t h a t
s ' ,s ,P'
which depends
3. Singular Perturbations on Smooth Manifolds without Boundary
426
Proof. First, we check that <ES>-' in L (mn) ,
P
v
<E<>'<S>-'
and
are Fourier-multipliers
a 2 0.
One shows easily,using the induction argument, that for each multiindex a holds:
where Q ( 5 ) is some polynomial whose degree is at most
[ a1,
so that one
has
Now, using (3.12.177), one finds
I
(3.12.178) D"<E~>-'
s
5
I
= E
I a I Da
S
s c 1 I 5 l - ~ a ~ ,v 5
C€lal<,S> -u-/ul
E Rn-.IO},
provided that u 2 0. Further, the Leibnitz formula yields: (3.12.179) /Da(<€5><5>-')'l
5
=
1
I: $53
(i) (DB<E<>') 5
(Da-'<S>-'
5
)I s
s c
v 5 #
0,
and the same computation shows that (3.12.180) (Da(E<6><E5>-1)u/2 C2)5)-Ia', v 5 # 0, v a .
5
Hence, as a consequence of Corollary 3.12.63, both <EE'-' are Fourier-multipliers in L (Rn) ,
<EC>'<~>-'
P
Therefore, one has with u = s - s ' u2+rJ3 2 0:
>
v
U
and
2 0.
0 , i.e. U 1 2 0 , a +u 2 0 , 1 2 -
v a,
3.12. Elliptic Singular Perturbations
421
Corollary 3.12.79.
If
s
> s'
then H
H (s).P(Bn) 5
(IRn)5 H ( s ' ) ,P,E(mn), V E
(s),P,E H ( s ' ,p ) (R").
E ( 0 , ~ and ~l
Lemma 3.12.80.
Let s
c
=
3
(sl,s2,s3)E IR , s2 > 0 , p E ( 1 , m ) .
Then t h e r e e x i s t s a c o n s t a n t
> 0 such t h a t holds:
Proof. Without restriction of generality, one can assume that -
s1 = 0.
Further, as a consequence of (3.12.177) and Corollary 3.12.63,
fl(c)
-S
=
<5>
is a Fourier-multiplier in L (IR")
using again the Leibnitz f2(5) = <5>s2(1+151s2)-1
P
, t/
s2 2 0 . Furthermore,
formula, one gets immediately that and f3(<) =
151s2
that f ( 5 ) and f ( 5 ) are Fourier-multipliers, as well, by virtue of the 2 3 same Corollary 3.12.63. Hence, one has with s = (0,s2,s3), s2 > 0:
On the other hand, one finds:
3 . Singular Perturbations on Smooth Manifolds without Boundary
428
and
Lemma 3.12.81.
Let
c
s = (sl,s2,s3)E R
> 0
3
,
s 3 > 0 , p E (I,-).
-S
Proof. (E
to
One checks easily that <ES>
15 I <EC>-') E
Then t h e r e e x i s t s a c o n s t a n t
such t h a t h o l d s :
E (O,E
3
, <E<>'~ (1+( E 15 I ) s 3 )
-'
and
s3 are Fourier-multipliers in L ( R n ) uniformly with respect 0
1
Lemma 3.12.80.
P
and uses the same argument as previously in the proof of
I
For s = (sl,s2,s3)E R
X
R+
X
R + , p E ( 1 , ~ )and u E S ( R n ) introduce
the family of norms
Corollary. L e t s E R x R+ x R + , p E (I,-). Then t h e f a m i l i e s of norms
-
and I I I I I I ( s ) there e x i s t s a constant c >
a r e e q u i v a l e n t uniformly w i t h r e s p e c t t o
o
E
I I .I 1 ( S ) , P , E E (O,E
0
I i.e.
such t h a t
Indeed, using the definition of
1 1 . I 1 ( s ) .P,E
and the previous two
3.72. Elliptic Singular Perturbations
429
lemmas, one finds:
Therefore, in order to prove (3.12.184). one has only to show, that with some constant C > 0 one has:
is defined by (3.12.183). For that purpose, it suffices to show that f ( 5 ) = 15
I s3 (I+ 15 I
s +s
2
3 -1 )
with V s2 2 0, s 3 2 0 , is a Fourier-multiplier in L (Rn) . One checks easily, using the Leibniz
formula, that the condition (3.12.123) for
fs(c) is satisfied, so that Corollary 3.12.63 applies. This gives:
1 Corollary 3.12.83.
L e t s E R+
z+, p E (I,-). Then t h e r e e x i s t s a c o n s t a n t C
x Z+ x
> 0
such
t h a t one has: -S
(3.12.186) C-lI lul s I I E
+
I
5
E
( s ) ,P,E s +s
3D 2 3 x , UI
I
1
(1
/u/
I
s2
I
+ 1 (/IDx,UI L (IR*') l < = j s n 2 L
P
P
I
) )
I
5 CI lul (.-)
(w")
+
,P,EI
L (Rn)
P
v
E
E
(o,Eo],
v
U
E H
( s ) rP,E
(Rn) .
Indeed, one uses Corollary 3.12.82 and Corollary 3.12.63 for showing (3.12.186). For doing that, one has to show that for each function f
.(F)
or1
=
<"SI-" 1
0
>
0 the
satisfies (3.12.123). This can be easily seen by
3 . Singular Perturbations on Smooth Manifolds without Boundary
430
induction argument. Therefore, one has for any positive integer a:
On the other hand, with any positive even integer a holds:
and for a positive odd integer one has:
using (3.12.187)-(3.12.189), one gets immediately (3.12.186).
I
Remark 3.12.84. For s = (sl,s2,s3)E R families of norms 1 1 1 . 1 are equivalent with to
E
x
lR+
x
R+
,
s
,
1 I I ( s ) ,E,P without
1 1 . I 1 ( s ) ,P.E
E Z+, j = 2,3, one can define
using the Fourier-transform, which
introduced above, uniformly with respect
I
E (O,Eol-
-1 Let u E (0,l) and let 2n(n+a) < p <
m.
equivalence (see, for instance, [Ste, 1,21):
where
Replacing
U ;l)u
by
Then one has the following
3.12. Elliptic Singular Perturbations
43 1
the equivalency (3.12.190) holds for u E ( 0 , Z ) . NOW, if s
=
m+u with some integer m > 0 and some u E (O,l), then one
has the following equivalency (see [Ste, 1,2])
Now, let u
=
(0,u2,u3) with u , E 3
(O,l), j
=
2,3. Then holds:
-
\-.
I
and it is quite obvious, how the last equivalency extends to the case when s = (sl,s2,s3),s1 E I R , s . = m.+u. j = 2,3 with integer m . 2 0 , m 2+m3 > 0
1
and
U.
E (O,l), j
=
3
2,3.
I' I
3
3
Now we are in a position to outline the proof of the two-sided a priori estimates in spaces H
(s),P,E
(Rn) for elliptic singular perturbations
We need the following technical Lemma 3.12.85.
L e t a E L'(R")
c and
be e l l i p t i c of order v . Then t h e r e e x i s t p o s i t i v e c o n s t a n t s
R such t h a t
(3.12.192) la(x,E,E)I 2 CE
-ul
151 "2<65>'3,
v
x E
Rn, V
provided t h a t E~ i s s u f f i c i e n t l y s m a l l . Moreover w i t h xR(S) E Cm(Rn) , xR t 0, f o r 1 5 ) 2 ZR, t h e symbo2 xR(S)a(x,E,5)-'
couple of m u l t i - i n d i c e s
a,B
holds:
xR
E
o
v
E
E
(o,EO1.
5 E R n , 151
f o r 151 5 R,
beZongs t o L-~(R")
2 R,
xR
, i.e.
:1
for each
3. Singular Perturbations on Smooth Manifolds without Boundary
432 Proof. Since -
v n (see D e f i n i t i o n 3 . 1 2 . 1 ) , one h a s a E S ( R )
a E L'(R")
.
F u r t h e r m o r e , a ( x , E , S ) b e i n g e l l i p t i c , i t s p r i n c i p a l symbol a 0 ( x , E , S ) s a t i s f i e s ( 3 . 1 2 . 2 ) , so t h a t one g e t s u s i n g ( 3 . 1 2 . 2 ) and ( 3 . 3 . 4 ) w i t h a = B = O :
la(x,E,c)
1
-v 2 (C/~)E
2 lao(x,c,S)
1
1-1
(a-ao) ( x , E , < )
1
2
-vl
151
CE
2<E<>
v3
( 1 - c l ( E + < ~ > - l ) )2
v
151 2<EF,>v3, V x E
V
IRn,
E
E
(o,Eo],
v 5 E
En,
151
b R,
( E +RP1) S f. 1 0 N o w , w e check (3.12.193) u s i n g i n d u c t i o n w i t h r e s p e c t t o a and 5. With-
provided t h a t C
o u t r e s t r i c t i o n o f g e n e r a l i t y , one c a n assume t h a t v1 = 0 . F o r a = B = 0 (3.12.193)
i s an immediate consequence o f
(3.12.192). W e use i n d u c t i o n i n
o r d e r t o show, t h a t f o r 151 > 2R h o l d s :
I n d e e d , assuming t h a t ( 3 . 1 2 . 1 9 4 ) ,
(3.12.195)
hold for l a l + ( B (
=
m, one
finds that
where b a + e k , B
=
aD
5b a , g - ( l + l a ~ + ~ B l ) b a , B D S ksaa t i s f i e s
i n s t e a d o f a , as it c a n e a s i l y b e s e e n . Bcek a -1 The Same argument a p p l i e s t o D D a
x
s
,
(3.12.195)
w i t h a+ek
t h e d e t a i l s being l e f t t o t h e
reader. Next, w e p r o v e t h e main r e s u l t , s t a t e d i n t h e f o l l o w i n g Theorem 3.12.86.
Let a E
L'(R~)
and p E
(l,m),
E
E
( 0 , ~ ]
0
(3.12.196)
be e l l i p t i c o f order v. Then f o r each s E I R ~ ,s ' E alvs t h e foZZowing equivalency holds uniformly w i t h r e s p e c t t o
w i t h c o n s t a n t s , which may depend on s , s ' , p and
1 / u / 1 (s),P,E
provided t h a t
E~
- j lop
I
( a ) u l ( s - v ) ,P,E
i s s u f f i c i e n t l y small.
+
E
~
:
I l u l I ( s ' ) ,P,E'
3.12. Elliptic Singular Perturbations
433
Proof. First, we prove (3.12.196) for a = a(E,S) which does not depend on ___ x. Since s' E a l V s , i.e. s; = sl, one can assume without restriction of generality that s ' = s1 = u1 = 0. With a E a(E,F,),s = (0,s2,s3), 1 v = (0,u2,u3) and x ( 5 ) as in Lemma 3.12.85, one can write R
Now, using Leibnitz' formula and (3.12.193), it is easily seen that f
(E,
5)
=
xR ( 5 ) a( E, 5 ) -' I < 1 v2<~S>u3satisfies the
condition (3.12.123)
uniformly with respect to E E ( O , E 01 , i.e. with a constant Bo on the righthand side of (3.12.123), which does not depend on E E ( O , E 01 . Hence, Corollary 3.12.63 appliy to fl(E,S). The same, obviously, holds for s -s s3-sj m , since f2(E,S) E C 0 (Rn) . Therefore, f2(E,S) = (l-xR)<S> '<ES> Corollary 3.12.63 yields:
where the constant C depends only on s,s',p and
E 0' On the other hand, Proposition 3.12.78 yields:
I lul I
( s t )
,P,E 5 CI
lul I ( s ) ,P,E-
Furthermore, one uses again the splitting
xR as in Lemma 3.12.85, and the previous argument,in order to show that as a consequence of Corollary 3.12.63, the following inequality holds:
with
I lo?(a) with a constant
C,
UI
I (s-") ,P,E 5
I
cl lul (s),P,E
which does not depend on
E
and u.
Now, if a=a(x,E,S), then one has to use a partition of unity argument, that is to prove analogues of Lemma 3.12.9
and Lemma 3.12.10 in
3 . Singular Perturbations on Smooth Manifolds without Boundary
434 spaces H
( s ) ,PrE
(Rn) , which can be shown again by means of Corollary
I
3.12.63, but will not be done here. Example 3.12.87.
Consider the singular perturbation which appears in the dislocation theory and whose symbol a(E,S) is
5, (3.12.197) a(E.5) = (l+exp(-~/Sl))sgn
E
> 0, 5 E R
(see also Example 3.2.6). This singular perturbation is elliptic of order zero but a f L0 (IR1 ) 1) , since it is not smooth at 5 = 0. However, with x6 E Cm (R x6 E 0 for
151
5 6,
x6
E 1 for I S / 2 26 the singular perturbation O p ( x (S)a(E,<))with
6
a ( € , < ) given by (3.12.1971, is an elliptic singular perturbation of order
0 1 zero in Op L (R ) , so that (3.12.196) holds with v = 0 . A
straightforward computation shows that (3.12.123) is satisfied for
a(E,S) given by (3.12.197) with Bo = 2 . Hence, uniformly with respect to E
holds:
(3.12.198) Op(a(E,<)) : L (R1
1
,
L (R )
+
P
P
V p E (1,~).
Furthermore, a straightforward computation shows that (3.12.123) holds for a(E,S)-'
=
(l+exp(-E:Sj))-lsgn5 , as well, with Bo = 1, so that
uniformly with respect to (3.12.199) Op(a(E,C)-l)
E
: L
P
E (O,E
1
(R )
1,
0
V
E~
1
+
L (IR )
P
<
m,
holds:
.
Hence, combining (3.12.198), (3.12.1991, one gets that the singular perturbation with symbol (3.12.197) is uniformly with respect to E E ( O , E 1 , 0 an isomorphic mapping of L (I71 ) onto itself, V p E (1,m). Notice that the
P
reduced operator
(E =
0) for Op a(E,S) given by (3.12.197) is the multiple
of the Hilbert transform (@(a
0
( S ) ) u ) (x)
=
2v.p.
i
7
J1 (x-y)-1 u(y)dy.
I
R
Example 3.12.88. Let k(x) be locally integrable function, which satisfies the following conditions: (i) k(x) E 0 for
1x1 < 1;
(ii) k E K , i.e. k satisfies (3.12.142);
3.12. Elliptic Singular Perturbations
(iii) there exists a function k (w), w E Rn, k 0
0
435 1
E L (R
)
such that with
some constants C > 0 , y > 0 holds a.e.:
(iv) the function k
0
J ko(w)dw
=
( 0 )has
zero mean value over R
:
0.
On We shall denote by k (x) the function, which is identically zero for 1x1 < 1 0 and coincides with k (w) for 1x1 2 1. 0
IxI-~
We are going to show that the family of convolution operators
E
-f
K
with kernels
is uniformly bounded in L
P
One can view K
(mn),
1 < p <
-,
as a singular perturbation of the convolution operator K 0
with the kernel ko(w)/ x / -for ~ 1x1 2 1 and identically zero for 1x1 < 1, this convolution operator having been considered previously in [Cal-Zig, 1 1 and [Hbr, 11. We have to show that GE(€,) = F
X-fE5
belongs to K
(3.12.142) with a constant, which does not depend on
n k2 and E
satisfies
E (0,lI. Then
(3.12.200) will follow from Theorem 3.12.72. First, we show that 1; ( 5 ) E Lm(Rn) uniformly with respect to so that uniformly with respect to E E (0,1] will hold:
E
E (0,11,
E E fi2. Notice that
if y E 0, then kE(x-y)-kE(x) E L1(Rn) , as a function of x E IRn, since k
E K , for each
E
E (0,1]. Therefore its Fourier transform
) (exp(-i
-f
m
(uniformly with respect to
E (0,ll). Furthermore, obviously one has (as a consequence of (iii), (iv)) 1
k(tEE)-E(ES) =
1 1 ko(w)exp(-i
+
[wj=l t
Denoting the first and second integrals on the right hand side of the last
3 . Singular Perturbations on Smooth Manifolds without Boundary
436
formula by Il(5,t) and 12(€,5,t), respectively, one finds that
On the other hand, if (iv) is satisfied, then
When 5 E Qn, t E (0,1], the right hand side of the last formula is bounded by the constant C
21~1 /k
=
0
1 I Limn)'
Hence, since I1(E,t) = io(t<)-co(6)
1
with i o= F k (x), one gets the conclusion that / g o ( < ) is bounded for xt5 0 5 # 0. Therefore, 1; ( 5 ) is the sum of a bounded on Rn function and, 0
possibly, a linear conbination of 6-function and its derivatives supported at the origin. It is easily seen that the latter singular part in the sum representing k0(5), must be zero. Indeed, it suffices to show that for each $ 6 ( x ) = $(x/6), with $ E C;(Rn)
1;0 ( 46 ) = k0(4&)
=
, holds:
0 for 6
+
+
0, where
J k (x)J(6x)6"dx.
wn
O
I
E S(Rn) , so that ]i(y) 5 C
Using the fact that
1 Iko(rw)Idw
5 Cr-n,
n' one finds, introducing polar coordinates, that
lCo(b6)l
5 ~ 6 "ln(1+6-l).
Hence, indeed, kO(E) coincides, as a distribution, with some function in L,(Rn)
. Now, using (iii), we can estimate the second integral I2 ( ~ , < , t ) in
the following fashion m
1 J
l12(~,5,t)I2 C
-1 r (r/tE)-Ydrdw 5 C
Iw/=lt
v (E,E,t) E In particular, for the function k(x)
=
-1
1'
[0,1]
-2
F k (5)<5> s+x 0
x
, for
wn
x
w+.
1x1 2 1 , k(x)
5
0
for 1x1 < 1, where 1; ( 5 ) is homogeneous of order zero and satisfies (iii), 0
(iv), the conditions (i)-(iv) are fulfilled, and the corresponding family of convolution operators K
with kernels kE(x) = E-'k(x/E)
bounded in L (Rn) , 1 < p <
m,
P
V
E
E (0.11, i.e. iE(5)
=
are uniformly
];(ES) are
3.13. Girding's Inequality (uniformly with respect to
E
437
E (0,1]) Fourier-multipliers in L (En)
P
Notice that the convolution operator -1 io(:)<:>-2 k (x) = E-"k(x/E), k(x) = F X'C -1 1; ( 5 ) oEerators with kernels k (x) = F 0 X'[ 0 latter being the exponentially decreasing
.
with the distributional kernel is product of convolution and GE(x)
= F-l < E C > - ~ , the C-tx at infinity fundamental solution
for the elliptic singular perturbation l-cLA with A the Laplace operator.
I 3.13. Garding's inequality
This inequality, first proved in [Vish, 1
1
for strongly elliptic
differential operators in domains with small diameter and properly extended in [G&,
1
1
to strongly elliptic differential operators in
arbitrary domains and in [Cal-Zig, 1
1
to singular integral operators,
plays an important role in different fields of the theory of pseudodifferential operators. Its sharp form, first proved in [Lax-Nir, 1
]
for pseudodifferential operators and for one parameter families of difference operators appearing in the stable approximations of well posed evolution problems, is one of the central results in the linear theory of pseudodifferential operators. A simplified proof of the sharp form of G&ding's [Vai, 1
1
inequality was given in [Friedr, 1
1
and later on in
where the method introduced by Friedrichs (the use of a smooth
mollifier) was slightly simplified. For classes F"
of difference (s)
operators (see Definition 3.12.32) the sharp form of Giirding's inequality was first established in [Fr,8,91 in corresponding spaces H
(Rt) (s),h
,
using the method introduced by Friedrichs. We shall present here the proof of Gzrding's inequality for the classes of singular and difference singular perturbations introduced above and shall give several applications of this inequality in the elliptic theory and to some classes of singularly perturbed problems of evolutionary type. We start with the simplest form of Garding's inequality Concerning strongly elliptic singular perturbations. We consider here singular perturbations, whose symbols are pxp matrices, the corresponding symbol classes S" ( U ), S v ( U ) , K " ( U ) , H " ( u ) , 1 to FY,o(U), F v ( U ) being defined a s previously in sections 3.3, 3.6, 3.11, 3.12 with the only difference that in all these definitions the absolute values in the corresponding inequalities should be replaced by the norms of corresponding matrices in Hom(CP;CP).
3. Singular Perturbations on Smooth Manifolds without Boundary
438
Definition 3.13.1. A symbol a(x,E,S) E s v ( U ) , a
:
U
(O,E~]
X
lRn+ Hom(CP;CP), i s s a i d t o
X
be s t r o n g l y e l l i p t i c , i f i t s p r i n c i p a l symbol ao(x,c,S) s a t i s f i e s t h e condition -v
Re a (x,E,<) 2 y
(3.13.1)
0
E
v
'151 2
where Y i s some p o s i t i v e c o n s t a n t , Re . a
=
( O , E ~ ~ (Wn\tO}), X
( 4 ) (ao+a;)) i s t h e ( s e l f - a d j o i n t )
and Id i s t h e i d e n t i t y i n Hom(CP;CP), r e a l p a r t of t h e m a t r i x a ( x , E , ~ ) 0
t h e i n e q u a l i t y B b 0 for any s e l f - a d j o i n t m a t r i x B i n Hom(CP;CP) being
E cp. op a E op sv(u) i s s a i d t o be s t r o n g l y
i n t e r p r e t e d i n t h e usual way:
=
20, V z
e l l i p t i c o f order v, i f i t s symbol a(x,E,S) E sv(u) i s s t r o n g l y e l l i p t i c . Lemma 3.13.2.
L e t a E s V ( u ) b e s t r o n g l y e l l i p t i c . Then t h e r e e x i s t s a symbol b E Sv'2(u) such t h a t (3.13.2)
(Re
where b*(x,~,5)i s t h e conjugate symbol of t h e symbol b(x,s,<) whose asymptotic expansion i s g i v e n by (3.7.16),
i.e.
w i t h b(x,E,S)* t h e conjugate m a t r i x of b(x,E,S), and where
-
stands f o r
t h e (non-commutative) product of symbols, d e f i n e d by (3.12.311, i . e . b*(x,E,S).b(x,E,S) i s a symbol w i t h t h e asymptotic expansion
Without restriction of generality one can assume that v = 0, since one can v1
consider
E
-v
<S>
-V
*<ES>
a principal symbol .
3
a(x,E,5) instead of a(x,E,C). Furthermore, if the 0
of a E S (U) satisfies (3.13.11, then for I S 1
sufficiently large the same inequality holds for a(x,~,C)itself with y/2 instead of y. Hence, one can find a symbol r(x,s,5) E that
S(or-m'O)
1.0
(U) such
3.13. Girding's Inequality
439 2
Indeed, (3.13.5) is valid with r(x,E,S) constant
C >
=
Ce-lC1 , provided that the
0 is sufficiently large.
Therefore, one can assume from the very beginning that a E S0 (U) satisfies the inequality: (3.13.6)
Re a(x,E,S) 2 y Id, V 1
( x , E , ~ )
E ux (O,E~] x lRn,
with some constant y1 > 0 ; here Re a = ( 1 / 2 ) (a+a*). We shall define the symbol b(x,E,C) E So ( U ) recursively to satisfy 1#O (3.13.2). Let (3.13.7) where
r
1 i Xt(XI-Re a(x,~,S))-ldX, bo(x,E,S) = 2~ri ~
is any contour surrounding the spectrum of the matrix Re a(x,~,S)
and excluding the origin. Since a E (x,E,~).
S
0
(U) satisfies (3.12.6), one can choose
r
independent of
Therefore, bo(x,E,C) is again a smooth function of (x,S) with
values in Hom(CP;CP), and moreover, one has: (3.13.8)
bO(x,E,S)* = bO(x,E,S), bO(x,~,S)2 = Re a(x,~,<).
Further, a straightforward computation shows that b (x,s,S) E 0 formulae for the adjoint symbol bi(x,E,E) and the product
So
1 to
(U), the
bi(x,E,C)'bO(x,~,S)lead to the conclusion that
and that the principal symbol of r ( X , E , ~ ) is a self-adjoint matrix. 1 Using the induction argument, assume that we have already found bO(x,E,S), bl(x,E,S),...,bk(x,E,S) such that
As a consequence of (3.13.10), the principal symbol of r ( X , E , ~ ) is a selfk
adjoint matrix. We have to determine a matrix bk+f(x,~,e) in such a way that def (3.13.11) rk+l(x,E,S)
=
(Re a(x,E,S)-( Z b*(x,E,S)).( Z bj(x,E,S))) 05 j5k+l OSj
3 . Singular Perturbations on Smooth Manifolds without Bounda y
440
Inclusions (3.13.10), (3.13.11) yield the following equation for bk+l(x,~,S):
'
C b*(x,E,S)-bk+l(x,E,<). O<jZk+l
If bk+l(x,~,S)is chosen to satisfy the equat.ion
Moreover, since bO(x,E,<) and the principal symbol of r (x,E,S) are k both self-adjoint, so it is also for the principal symbol of bk+l(x,~,<). Furthermore, bO(x,E,5) t y Id, with some constant y o > 0. 0 The matrix b ( x , E , < ) , given by k+l (x,E, 5 ) --tbO(x,E,S) (3.13.14) bkil(x,€,S) = 1 e dt rk (x,E , 5 ) e 0
-tho
is well defined (since bo is positive definite uniformly with respect to x,E,S) and satisfies (3.13.13). Indeed, one has:
Furthermore, it is obvious that the principal symbol of bk+l(x,~,E)is self-adjoint. It is easily seen that bk+l(x,~,S)(given by (3.13.14)) is the only solution of the matrix-equation (3.13.13). Indeed, one has to show that the only matrix X E Hom(CP;CP), that satisfies the equation
0, provided that b (x,E,S) is positive definite. 0 Multiplying (3.13.15) by exp(-tbo(x,E,S)) from the left and from the right, one gets the equation, which can be rewritten in the following form: is X
- -(e dt
-tb -tb Oxe O)
=
0,
v
t 2 0.
3.13. Girding’s Inequality -tbo Hence, e-tboXe
X,
5
v
t 2 0 . Now letting t
-f
44 1
+-,
one gets the
conclusion that X E 0. Hence, bk+l(x,~,E) defined by (3.13.14) is the only solution of (3.13.13), so that the tefms b (x,E,S) in an asymptotic expansion of b(x,E,S) are
k
uniquely defined by the recursive procedure above. Therefore, the symbol b(x,E,S) which satisfies (3.13.2) is well defined modulo a symbol in (u, ,--,v3) I (U). S 1 ,o Remark 3.13.3. Using precisely the same argument, one can define recursively by the same formulae (3.13.7), (3.13.9), (3.13.lo), (3.13.131, (3.13.14) with a(x,E,S) itself instead of Re a(x,~,S),a symbol c(x,E,~)such that t (3.13.16) (a(x,E,S)-c(x,E,S) -c(x,E,S))E
(v1,-m,u3) (U),
s1 t o
t where C ( X , E , ~ ) is the transpose symbol of
C ( X , E , ~ ) ,
defined (asymptotic-
ally) by the first of formulae (3.7.16). Furthermore, c(x,E,~)is well defined modulo a symbol in
(V1?,V3) S
:u).
1 ,O
Corollary 3.13.4.
L e t a(x,E,S) E Sv(U) be s t r o n g l y e l l i p t i c i n t h e sense of Definition 3.13.1 and l e t A~ = op a be t h e corresponding s t r o n g l y e Z l i p t i c s i n g u l a r p e r t u r b a t i o n . Then t h e r e e x i s t s a s t r o n g l y e l l i p t i c s i n g u l a r p e r t u r b a t i o n B
E op S~’~(U)such t h a t h o l d s :
(vl,-,,v3)
(3.13.17) (Re AE - B*B E
where
BZ
E Op S
) E
i s t h e a d j o i n t of m
B~
1 to
with respect t o the duality relation <
1
between CE(u) and (co(u)) and Re A E Indeed, with b(x.E.6) E Sv’2(U) 3.13.2 one gets for B
=
=
,
>
(1/2)(AE+A:).
as defined in the proof of Lemma
Op b immediately (3.13.17),as a consequence of
this lemma. Further, with
=
C
Op c(x,~,S)with C ( X , E , ~ ) constructed as discussed
in Remark 3.13.3, one has the inclusion (u1,-m,u3) (3.13.18) (AE ) -,)C
E
OP S
1 ,o
(U),
L
where C L is the transpose of the family of singular perturbations CE E opEsv/2(u), CE
:
CpJ)
+
Cm(U).
3 . Singular Perturbations on Smooth Manifolds without Bounda y
442
Theorem 3.13.5. 3
be strongly e l l i p t i c . Then f o r any s E R , f o r any c : ~ c # cK tc u there e x i s t constants c0 > 0 and c1 > 0 such t h a t Let a ( x , ~ , ~E) op s'(u)
(3.13.19) Re(a(x.E,D)u,u) 2 coI ( U I I 2(v/2) ,E-c11
where
(
,
)
I u I 1 ( s ) , E , . v ~ c:(K),vEE E
(o,E01,
2
stands for the inner product in L (K).
Proof. __ Corollary 3.13.4 guarantees the existence of a singular perturbation B
=
Op b(x,E,S) E Op S"/2(U) such that (3.13.7)
is valid with
AE = Op(a(x,E,~)-(y/2)E-"1<S>u2<ES>"3 Id), where the constant y is the one, appearing in (3.13.1). Denoting Rr = Re Ac-(y/2)€ -"1
E'
one can write for
v
u E
m
c0 (K):
I IUI (Y/2) I l u !
Re(AEu,u) = (Y/2) 2
(V/2),E
+(B u,B u)+(REu,u) k E
E
3
V s E l R ,
I
where c1 may depend on s and K. Remark 3.13.6.
Definition 3.13.1 carries over in an obvious way to singular perturbations ona smooth manifold M without boundary, their principal symbols being well defined for V
E
E R+ as a mapping from the cotangent bundle T*(M)
into Hom(CP;CP) in the case of matrix-valued symbols. Hence for a strongly 3 elliptic singular perturbation a(x,E,D) : Cm(M) + Cm(M) of order u E R 3 one has for each s E R :
where the constant co depends only on y in (3.13.1) and the constant C
1
does not depend on u E Cm(M). Example 3.13.7. Consider again the singular perturbation A
from Example 3.10.15 defined
by (3.10.54) or, equivalently by (3.10.58) on the unit circle R
1
C
C.
Its principal symbol a0 ( x , E , S ) = 1 + E / < 1 satisfies (3.13.1) with u = (0,0,1) is strongly elliptic of order (O,O,l). Defining
and yo = 1, so that A
the singular perturbation B
=
F-l(l+Elnl)fF with F the Fourier series E
one has in this case (A u,u)
=
(B u,B u) 2 c E
E
0
1
2
and AE = BE. Besides, 2 with some /U/ 1
expansion operator, it is easily seen that B* = B
E
(0,Ort)
,E
3.13. Gdrding's Inequality
443
constant co > 0 which may depend upon the partition of unity in the definition of the norms 1 if3 E c I I e l 5 TI. fil = Ie
1. I [
on the circle
(O,O,!)
,E
Singular perturbation (3.10.66) is strongly elliptic of the same order v = (0,0,1) BE
iff Re a(z) t . a
having ( I + € Re a(z)
/
>
r,
0, V z E
the corresponding operator
as its principal symbol.
It will be shown later that for the singular perturbation A
defined
by (3.10.66) the following inequality holds:
where the constant C does not depend on u and
E.
Now several forms of sharp Gzrding's type inequalities will be considered and we start with the one for one parameter families of difference operators introduced in 3.11. We follow essentially [Friedr, 1 1
1
and [Vai, 1
for proving this form of sharp Gzrdinq's inequality.
The symbols considered are valued in Hom(CP;CP), i.e. are pxp matrixsymbols.
As usual, we denote by a,(x,n)
the principal symbol of a symbol
a E F"(U), ao(x,n) = lim h"l:(x,h,n). h+O Theorem 3.13.8. O-ii
Let p E F ( n
)
Assume t h a t t h e primcipal symboZ pO(x,ri) E n e g a t i v e hermitian m a t r i x :
1,I?
)
and is a non-
(3.13.22) Po(X,I1) 2 0 , v (x,r?)E lRnxTy,n.
Moreover, assume t h a t t h e r e e x i s t s a c o n s t a n t
B
c such t h a t 1
I
(3.13.23)<~>~*~lD~p(x,h,ri)-D~p,(x,n) B 6 Clh, V (x,h,n) E Rn x(O,ho]xTy,ri,
IB/
v 8, Then f o r ph
=
op(p) one has:
2
(3.13.24) Re(Phu,u)t -C2h/l u l
lo,
V u
m
n
E CO(R
),
V h
E (O,hOl
where c 2 i s some c o n s t a n t , which may depend on ho, and where
I 1. I 1
5 n+l
2
are t h e i n n e r prod-uct and t h e norm in L ( n n)
.
(
,
)
and
3 . Singular Perturbations on Smooth Manifolds without Boundavy
444
Proof. As
a consequence of assumption (3.13.23),it suffices to prove (3.13.24)
for the difference operator Ph,D
-
Op po, whose acting on a function
u E Cm(IRn) is given by the formula: 0
Indeed, if p
pm(h,hE) then, as a consequence of (3.13.23) and Parceval's
identity, one has:
I I Ph-Ph,ol lL2,L2
5 C h. Thus, without restriction of 1
generality, one can assume, that pm(h,h5) x E IRn , belongs to S(IR:)
5
0, so that p, as a function of
and so it is for po, as well. We use again the
same argument, as in the proof ot Theorem 3.11.15. Denote r(x,h,hE) One has for
=
V(5)
F
x+S
=
=
%".
u the following representation:
J
=
p(x,h,hS)-po(x,hC), Rh = Op r, v
G(E-n,h,hn)G(Il)dIl,
IRn where ;(C,h,hrl)
=
Fx+Er.
For any function 41 E S(Rn
)
and for each multi-index a one has the
following inequality, as a consequence of Parceval's identity:
where R
is the area of the unit spheere in IRn. n Applying the last inequality with 161 5 n+l to r(x,h,n) (as a function
of x E lRn
)
and using (3.13.23), one gets the conclusion that there exists
a constant C such that
Therefore, one has
I"(<) 1
6 Ch
f Rn
Applying the inequality:
3.13. Girding's Inequality one finds that
I / P -P
I
445
5 C h, where the constant C
h,o L2(Rn) +L2(iRn) depends only on n and C 1 in (3.13.23). Thus, it remains to prove (3.13.24) for P defined by (3.12.25). h,O If po did not depend on x, then, obviously, one would have: Re(Ph,Ou,u) 2 0 . Freezing the symbol p (x,n) at a point x = y and writing 0
the corresponding difference operator with the constant symbol p (y,hS) as 0
pO(y,hDx), one still has, of course, that (3.13.26) Re(po(y,hDx)u,u) = Re J n u(x)*po(y,hD )u(x)dx 2 0 , IR b u t , the latter quadratic form has very little to do with the original
one, which, using the notation Ph,o
=
po(x,hDx) can be written as follows:
(3.13.27) Re(po(x,hDx)u,u) = Re J u(x)*po(x,hDx)u(x)dx. iRn As previously, here u(x)
*
is the complex conjugate of u(x).
However, formally multiplying (3.13.26) by 6(x-y) and integrating it with respect to y E IRn, one gets (3.13.27). The idea of Friedrich's mollifier is to replace 6(x-y) by a specially chosen family of test functions $6 such that @
-t
6fx-y), as 6
-+
0.
Hence, the quadratic form Re
1
J
46(x-y)u(x)*po(y,hDx)u(x)dx dy
Rn IRn could be eventually a suitable candidate for approximating (3.13.27). Unfortunately, it is not defined by a symmetric kernel, so that trying to achieve the symmetry as one of the goals inapproximating (3.13.27), onecomes 2 with a, a family of non-negative test functions 6 and with the following quadratic form Qh as an approximation of (3.13.27). to the choice of 46 as
a6,
whose symmetry and non-negativity are obvious. Indeed, using the Parceval's identity and the notation v(c,y) = Fx+c($6 u), one can rewrite (3.13.28) in the following equivalent form:
6 = (21r)-" 1 (3.13.29) Qh
I v(5,y)*PO(y,h5)v(5,y)dSdy 2
nniRn
0.
3. Singular Perturbations on Smooth Manifolds without Boundary
446
6
Quadratic form.(3.13.28) can be rewritten in the form (Phu,u), where
6
(3.13.30) (Phu)(XI
Ji (x-Y)pO(Y,hDx)(Ji6(x-Y)u(x))dy.
=
Rn -n/2 a n We choose Ji (x) = 6 $(x/6), where $ E C ( R ) , Ji 2 0 , 6 0 2 supp $ C B 1 = {x E Rn 1x1 < 1 1 , J $ (x)dx = 1, V 6 > 0 .
1
Further requirements on Ji and 6 will later naturally come out. With this choice of Ji6 and using the substitution x-y = 62, one can rewrite (3.13.30) in the following fashion: 6 (3.13.31) (Phu)(X)
=
-1
/nJi(Z)Po(~-6~,hDx+h6Dz) (Ji(z)u(x))dz. R
Indeed, the Leibnitz' formula yields: pO(y,Dx)(Ji6(x-y)u(x))=
C
hI4 -(pAa)
(y,hDx)u(x))(D:Ji6(x-y))
=
a
=
6-n/2po(y,hDx+h6-1Dz) ($(z)u(x)).
6 Naturally, P can be viewed as a family of difference operators defined by h the following mollified symbol
6
(3.13.32)po (x,rl) = J
Ji ( z ) p o(x-6z,n+h6-1DZ)Ji(z)dz.
R
Representing p (x,n) as a Fourier series 0
Po(X.T1) =
POk(x) exp(i
one can rewrite (3.13.32) as follows: 1 J Ji(y)Pok(x-h6-'y)Ji(y+h6-lk)dy exp(i
P;(X,n)
=
0
Using (3.13.32) and Taylor's expansion up to the order two with remainder in integral form, one finds:
3.13. Girding’s Inequality
447
where
Here, as usual,
(
)x,
(
) 5 are gradients with respect to x and 5 ,
( )xx’ stand for the corresponding matrix of second derivatives and
( ) 5 5 ’ ( )xs < , > denotes the inner product in Rn.
Obviously, one has:
Further, choosing J, such that
/
2
$ ( z ) z dz = 0,
k
R
1 5 k 5 n
(for instance, choosing $ ( z ) to be even), one finds, that
6
R (u) and, as a consequence of (3.13.34), the optimal h choice of 6 is: 6 = h f . f 2 : L2(lRn) -t L ( R n ) in the same Now, estimating the norm of ((F’:-Ph,o)u)
(x)
=
<
) : L2(Rn) L2(Rn fashion as it has been done for the norm of (P -P h h,O one gets the conclusion, that there exists a constant C such that -f
I
hf 2 (Rh u,u) I 5 Chi IuI l o ,
V u E C:(Rn
),
V h
)
,
E (O,hol.
h’ Therefore, given that (Ph u,u) b 0, one finds: hf hf (P u,u) +Re((Ph,O-Ph ) u , u ) +Re((Ph-Ph,O)u,u) 2 h 2 I t -Chi lu/1 0’
Re(P u,u) h
=
Remark 3.13.9. As
a consequence of the assumption p
0
E C2 ( R n X TY
) r n
and (3.13.13) , the
difference operators in Theorem 3.13.8 are, in fact, finite difference 0 0
approximations of the multiplication operator by the matrix p (x) = po(x,O),
3. Singular Perturbations on Smooth Manifolds without Boundary
448
which is the reduced operator for the family Ph. That explains the presence of a small parameter h on the right hand side of (3.13.241, the latter leading for h
0
-f
0 to the obvious inequality: (p u , u ) 2 0, V u E CO(Rn) 0
.
This kind of operators appears in stable discretizations of problems of 0 being in the latter case the identity. 0
evolutionary type, p Remark 3.13.10.
P hrP belonging to some closed parameter set E (i.e.
Obviously, (3.13.24) holds for any family of difference operators p uniformly with respect to
p
-f
with C 2 which does not depend on p either) if the conditions of Theorem 3.13.9 are satisfied uniformly with respect to p E E. Corollary 3.13.11. 0-
op q, q E F 2 qo(x,q) E c ( x n x T~
L e t Q~
=
(fl) . Assume that i t s principaZ syrnbo2 ),
s a t i s f i e s (3.13.13) and has i t s norm i n
It11
Hom(CP;CP) 5 1: (3.13.35) lqo(x,rl)I
cp+cp
5 1,
v
(x,rl) E lRn
X
Tn
1,o-
Then t h e r e e x i s t s a c o n s t a n t c such t h a t
Indeed, (3.13.36) is equivalent with the inequality: 2 m n (Phu , u ) 2 -ChjlujjO, V u E C O ( R ) , where Ph = Id-Q;Qh
(with
Qi the adjoint in L
2
(W") of Qh) has as its
principal symbol p0(x,n) the following hermitian non-negative matrix: po(x,rl) = l-qo(x,q)*q (x,rl), the latter being a consequence of (3.11.471, 0
and the non-negativity of po following immediately from (3.13.35).
1
Remark 3.13.12. The conclusion of Theorem 3.13.8 is still true for the difference operators
(g)
is a family of meshf f ( o ) ,,(W:) where ff ( 0 ) ,h Ph : f f ( o ) ,h(lRi) function spaces with the inner products and norms defined by (2.7.1), -+
(2.7.2),
respectively. In other words, under the assumptions of Theorem
3.13.8 the following inequality holds with some constant C which does not depend on h and u:
3.13. Ga"rding"s Inequality
449
Indeed, one proves (3.13.37) using the same argument as in the proof of Theorem 3.13.8 and combining it with the one in the proof of Theorem 3.11.15 (with s = v = 0). Example 3.13.13. Let a(x)
:
IR
+
R be such that (a(x)-a(m)) E S ( R
Consider the difference operator Q h (3.13.38) qr(x,n)
=
Op q
)
, and let 0
5 a(x) 2 a
0'
with the symbol
l-ra(x)+ra(x)exp(irl),
=
where r E (0,r 1 is a given parameter. 0 The difference operator Qh with symbol (3.13.38) solves the following Cauchy problem
where, as previously introduced, D and D are forward difference t,T x.h approximations of Dt and Dx, respectively and r = T/h. Indeed, for each t = N? with integer N > 0 the solution of (3.13.39) is given by the formula
If ro satisfies the stability condition: rOao S 1, for each rl r = T/h E (0,r ] onehas Iqr(x,rl) I S 1, t/ (x,n) E R x T1. 0 Hence, as a consequence of Corollary 3.13.11, inequality (3.13.36)
holds for Qh, so that one has:
-_
and the L -norm of the solutions u of (3.13.39) are uniformly bounded with 2
respect to h E (0,h 1 on each finite time interval 0 4 t 4 T < 0
Further, consider the semi-discretized Cauchy problem -1
atu(x,t)+h (3.13.40)
(P u ) (x,t) = 0 , h
U(X,O) = uo(x) where Ph
=
Op p has as its symbol p(x,hE) = a(x) (l-exp(-ihC)), so that (Phv)(x)
=
a(x) (v(x)-v(x-h)).
3. Singular Perturbations on Smooth Manifolds without Boundary
450
One finds easily in this case
so that
v,v)+(~,v,v), Re(P v,v) = Re(P h h,O where we have denoted:
Obviously, one has:
where C
=
sup
axa(x) 1 .
xElR
a
E Cm(Wx T
Furthermore, p,,
) , po(x,n) t 0, V ( x , n ) E R x T l so 1 .rl ,n' one gets:
that applying Theorem 3.13.8, (3.13.42)
Re(P V,V) 2 -Chi h
I v / lo,2
with some constant C, which does not depend on h and v. On the other hand, an easy straightforward computation yields:
with
5, is
defined by the second line in ( 3 . 1 3 . 4 1 ) .
Using ( 3 . 1 3 . 4 3 )
and the fact that ( 1 - 0 ) *
2Re(P v,v) h
=
=
(1-0
-1
((P +P*)v,v) = (a(x)(1-0 -1 )v, h h
),
one finds:
(l-O-l)v)+(\O-lv,v).
Hence, yiven that a(x) 10, the last identity yields: (3.13.44)
(+)
(mah21ID:,h~I
2 < ( f ) (Mah /D:,hVI
I
where a'(x)
=
a
) ;1
Io+ma,h/ 2 /vI
l o2
6 Re(Phv,v) 4 2
+
M a t h /IVI l o )
a(x) and the notation mf and Mf stands for infinimum and
supremum, respectively, of a given function f(x) over W . Thus, since m
L
0, one finds that the best possible choice of the
constant C on the right hand side of ( 3 . 1 3 . 4 2 )
is:
3.13. Girding's Inequality C =
45 1
inf a a(x). XEIR
-(+)
2
Furthermore, taking the L (R)-inner product in (3.13.40) and using the first of inequalities (3.13.44), one finds:
so that Gronwall's lemma yields:
Example 3.13.14. Let A E Horn(CP;CP) have real eigenvalues A . 1 5 j 5 p , and be similar to 7' the diagonal matrix diag(A l,...,A ) .
P
Consider the following strictly hyperbolic Cauchy problem: u
(3.13.45)
{u(x,O) -AU
(x,t) E
=
0,
=
uo(x)
nXB+
.
We are qoint to exhibit a class of high order two-layer stable difference schemes suitable for the numerical treatment of (3.13.45). These schemes can be written as follows rh
T =
V(t+T,X) = (Qh,"V)(ttx),
where r > 0 is a given constant and where the difference operator Qh,v - O P v' has the symbol: (3.13.46) qv
=
C
Ikl
a
k
exp(ikhS)
with matrix-coefficients to be chosen in order to provide a high order approximation of (3.13.45). Since the solution of (3.13.45) has the form:
one has to approximate the exponent exp(irh5A) by the symbols sf fcrm (3.13.46) with given integer v > 0, i.e. the coefficients ak on the right han? side of (3.13.46) are to be chosen to satisfy the condition: (3.13.47) exp(irqA)-
1 ak exp(ikrl) (k(5v
=
o[n
2 -v
)
for
rl
+
0.
3 . Singular Perturbations on Smooth Manifolds without Boundary
452
Taylor‘s expansion for exponents in (3.13.47) yields the following system of equations for the matrix coefficients ak:
It is readily checked that ak are given by the formula: (-l)v-k (3.13.49) a = k (V-k) ! (V+k)!
n (rA-pI), IPlZV PZk where I is the identity matrix in Hom(CP;CP) Indeed, since ak are polynomials in A, it suffices to show the identities
It is sufficient to show (3.13.40) for any 2V+l distinct values of h ,
since the left hand side of (3.13.50) is a polynomial of degree at most 2v. Taking h
=
m/r, m = @,fl, ...,+v
(3.13.51) qv = q (rA;rl) =
‘
Ik(
,
one finds:
(-l)V-k (v-k)! (v+k)!
Il
(rA-pI)exp(ikq).
IPl=
Pfk We are going to show that for each r such that (3.13.52) IrX.1 5 1, 3
1 5 j Z p,
the difference approximation of (3.13.45) given by (3.13.51) is stable in 2
L (R) in von Neumann’s sense, i.e. for t = NT E [-T,T] where N E Z and
0 < T < -,one has:
where the constant C depends only on T,r and the matrix A. Since q
is a polynomial in A, it suffices to show that
(3.13.53) lqv(C;?) 1 S 1, V 5 E [-l,lI, V r~ E Denote
1
3.13. Girding's Inequality (3.13.54)
II
( 5 ) = 2115 II
453
(1-52/p 2) .
lSp5w One has:
As a consequence of (3.13.55), q v ( 5 ; r i ) is the solution of the following
initial value problem:
(3.13.56)
{dq /dn
=
igq -i(21~)-'f~(g;q)
qv(5;0) = 1 where
Hence,
where we have denoted:
In the last formula one can take 5 to be complex, as well. Taking Im 5 < 0 and letting lim
il
+
+-,
one has, as a consequence of (3.13.55):
exp(-ign)qw(g;il) = 0,
q++m
so that (3.13.58) and the last formula yields: m
icw(g)/(sin(8/2))2wexp(-igB)dB = 1.
0 Therefore, one finds for Im 5 < 0: m
q v ( 5 ; n ) = icv(g)/(sin(8/2))2vexp(iC(a-8))dO = a
= Cv(5)/(2
m
sin IIg)/(eing-e-iI15)e-i5e(sin -)e+a 2vdo = n
2
3 . Singular Perturbations on Smootn Manifolds without Boundary
454
The uniqueness of the analytic extension leads to the conclusion:
Given the representation of sin 1 ~ gas an infinite product
z
sin 1 ~ g= 711
" (l-<'/pL),
pL 1 one finds that n v ( c ) / ( 2 sin ~ g 5) I ,
t/
5 E [-1,11.
2v 2 Furthermore, one has 2 ( v ! ) / ( 2 v ) !
< 1 , so that the last two
inequalities and (3.13.59) yield (3.13.53).
If some of the eigenvalues of A are complex numbers, then the Cauchy problem is ill-posed. Nevertheless, the difference schemes with symbols qv(rA;hC) are stable in the sense, introduced in [Fr.-Ch,i], [Fr., 1 1 (see also [Fr., 2
1
where their stability is proved).
With v = 1, one gets the following difference scheme for solving numerically (3.13.45) : (3.13.60) v(x,t+T) = [ (+)rA(rA-I)@-'+(I-r
2 2
A )+(t)rA(rA+I)Olv(x,t)
2
which is stable in L (1R) for IrX.I 5 1, 1 5 j 5 p with A . E R the eigen3
1
values of A and has the viscosity term of the form:
T Dx,hD*x,hv(x,t),
i.e.
(3.13.50) can be rewritten as follows:
-(L)
[iD (D +D* )+?Dx,hD:,h]~(~,t) = 0 . t , ~2 x,h x,h Now, let A
=
A(x) E Hom(Cp;CP) be symmetric and assume that there
exists Am E Hom(CP;CP) such that (A(x)-A_) E S ( R )
. Consider
again
(3.13.60) where A depends on x and let Qh be the corresponding difference operator on the right hand side of (3.13.60), whose symbol q is given by the formula: +(I-rA(x)) (I+rA(x))+
3.13. Girding's Inequality
455
Since q(x,hS) = q(x,hS)*, the estimate [ql 5 1 holds, if for each c+c eigenvalue h(x,n) of q(x,n) one has: Ih(x,n) S 1; the latter is valid.
I
provided that there exists a constant r > 0 such that /rh, (x)1 5 1, 3 1 5 j 5 p, V x E R , where A . (x) E R , 1 5 j 5 p are the eigenvalues of 3 A(x). Since q(x,hE) coincides with its principal symbol and satisfies all the conditions of Corollary 3.13.11, one gets the conclusion that (3 13.36) holds for Qh defined by (3.13.60). Assume now that in A in (3.13.45) is hermitian: A = A* and let P+ and P
be projection operators in Cp on the subspaces corresponding to non-
negative and strictly negative eigenvalues of A respectively. Hence A has the following block-structure: A = A 8 (-Am) with A+ =P+A, A-=-P-A in a suitable basis in Cp. Denote by B+ the Cayley transformation of rA+, V r > 0, i.e. B+_= (rA+-I)(rA++I)=l, the latter being well-defined, since -
-
rA2 are non-negative hermitian operators in Cp. Consider the following finite difference approximation of the hyperbolic problem (3.13.45) with Hermitian A: [P+-B+Oh)O(O P -B )8 -(8 P -B ) @ ( P - - O B )]V(x,t) h - - T h + + h -
=
0,
where, as previously, 0 and 0 are the shift operators in x and t, h respectively, (8 v) (x,t) = v(x+h,t), ( 8 v) (x,t) = v(x,t+T), and r = T/h. h It is readily seen that the solution of this finite difference problem is given by the formula:
where Qh is the difference operator with the symbol q, qo (r;0 )
=
(exp(in)P+-B+) (P+-exp(in)B+)-'o (P--exp(iq)B-)(exp(in)P- -B- ) - ' ,
1 q0 (r;ri) being a unitary operator in Cp, V (r,n) E n + x T n (see also
Example 3.8.83) . Hence,
3. Singular Perturbations on Smooth Manifolds without Bounday
456 If A
=
A(x) depends on x
E
W
and i s s t i l l h e r m i t i a n , t h e n f o r t h e S a m e
a p p r o x i m a t i o n above w i t h A = A ( x ) , t h e p r i n c i p a l symbol q 0 ( r , x , n ) o f t h e c o r r e s p o n d i n g o p e r a t o r Q i s s t i l l a u n i t a r y matrix i n C p , s o t h a t , u s i n g h a g a i n C o r o l l a r y 3.13.11, one comes t o t h e c o n c l u s i o n t h a t (3.13.36) h o l d s
fo; t h i s Qh, \J r > 0 . Example 3.13.15. L e t a ( x , ~ D) = Op a ( x , E S ) , a
E L(o’ofu)
Consider t h e e q u a t i o n a(x,ED)u assumed t h a t f
E C;(Rn
)
= f ,
x
( R n ) b e e l l i p t i c of o r d e r (O,O,u).
E
Wn
, where, f o r s i m p l i c i t y
it i s
.
L e t a ( x , p , h D ) b e an e l l i p t i c f i n i t e d i f f e r e n c e a p p r o x i m a t i o n o f o r d e r (O,O,U)
of a ? x , t D x ) , whose symbol i s a ( x , p , h < ) , R n
p = h/E,
(3.13.61)
x R+ X Tn + Hom(CP;CP), hrS and c o n s i d e r t h e c o r r e s p o n d i n g d i f f e r e n c e e q u a t i o n on Rn x,h .
.
(a(x,p,hD ) v ) ( x ) = f ( x ) ,
x E
$.
One o f t h e problems a r i s i n g i n n u m e r i c a l t r e a t m e n t o f l i n e a r e q u a t i o n s (3.13.61)
i.e. thedependenceofthe relative
i s t h e one o f w e l l - c o n d i t i o n i n g ,
e r r o r i n t h e s o l u t i o n v upon t h e r e l a t i v e e r r o r i n t h e datum f . The l a t t e r c z n bc e x p r e s s e d i n terms o f t h e upper and l o w e r bounds o f t h e s p e c t r u m o f the operator b = a*
0
a . Indeed, i f y ( a ) 2 i s t h e r a t i o
o f t h e upper and
l o w e r bounds f o r t h e s p e c t r u m o f b , t h e n t h e r e l a t i v e e r r o r i n t h e s o l u t i o n 2 measured i n L (an ) i s bounded by y ( b ) t i m e s t h e r e l a t i v e e r r o r i n f a xrh measured i n LL (IRn ) (see f o r i n s t a n c e [ B e r . - Z h i , 1 1 ) . I n o t h e r words, i n X rh o r d e r t o estimate t h e c o n d i t i o n number y(a) one h a s t o estimate t h e u p p e r and lower bounds f o r t h e q u a d r a t i c form ( b v , v )h , where ( , ) h i s t h e i n n e r 2 product i n L (Rn x,h) Applying Theorem 3.13.8 and u s i n g Remark 3.13.12, one f i n d s t h e
.
f o l l o w i n g estimates f o r t h e u p p e r and lower bounds r e s p e c t i v e l y , o f t h e q u a d r a t i c form ( b v , v )
h
A
max
( p , h ) and Xmin(p,h),
:
where c and C a r e some c o n s t a n t s a n d , f o r a b b r e v i a t i o n , t h e n o t a t i o n
i s u s e d f o r t h e norm i n H o m ( C P ; C P ) . T h e r e f o r e , t h e r e e x i s t s a c o n s t a n t C1 s u c h t h a t
1 1
3.13. Girding's Inequality (3.13.62)
457
y ( a I 2 5 C h+ 1
[ a(x,P , n) *a (x,p ,n)l /inf I a(x.p .n)*a(x,p,n) 1. (x,n)EBnxTn ,n 1r n As a consequence of the ellipticity condition, for each p o > 0 there
+sup (x,n)ERnX T Y
exists a constant C
such that PO
< c
,
v p t p
PO
Indeed, ( 3 . 1 3 . 6 3 ) take v 2
=
0, v 3
=
0'
follows immediately from ( 3 . 1 2 . 7 2 ) where one should
v.
2 2
2
Consider, for instance, the singular perturbation 1 + q~ (x)Dx, x E B , 2
2
which is elliptic of order ( 0 , 0 , 2 ) if q (x) L qo > 0, V x E B . Consider 2 2
2
first the following finite difference approximation a of 1 + ~q (x)D
:
2 2
a(x,p,~~,~,~:, = ~I+€ ) q (X)Dx,hD* x,h' which is elliptic of the same order. Applying ( 3 . 1 3 . 6 2 ) , one finds the following estimate for the conditioning number for the corresponding equation ( 3 . 1 3 . 6 1 ) y(a) 5 Ch+sup (x,n)EIR x T where
saxXER sup =
(I+% 1r
-2 2
2
q (x)sin ( r l / 2 ) )
=
in this case:
Ch+ ( 4 p
-2
2
sax+' )
n
q(x).
NOW consider the singular perturbation a(x,E,D
)
=
(l+iEDxq(x))~(l-Eq(x)Dx), which is formally self-adjoint and
, > 0, elliptic of order ( 0 , 0 , 2 ) if q(x) 2 qmln
difference singular perturbation M (3.13.64)
Mq(x)
=
v
x E R . Introduce the
q(x) '
-1 (l-exp(-p/q(x)) (l-exp(-p/q(x))8), P = h/E,
whith 0 the shift operator (Ov)(x) = v(x+h). Difference operator ( 3 . 1 3 . 6 4 ) being a finite difference approximation of the singular perturbation l-iEq(x)D , its formal adjoint M* q'
M* 9
=
(1-0
-1
exp(-p/q(x)) (l-exp(-pq(x))) - '
is a finite difference approximation of the formal adjoint of (l-iEq(x)D which is (l+iED q(x)).
)
3 . Singular Perturbations on Smooth Manifolds without Boundary
458
Hence, the operator a(x,E,h,o), (3.13.65) a(x,s,h,O) = M*
4
M
4
is a finite difference approximation of a(x,c,D
)
(l+iED q(x))o(l-isq(x)Dx).
=
The principal symbol of a(x,s,h,0) being -2 2 a0(x,p,n) = (l-exp(-p/q(x))) Il-exp(in-p/q(x)) I , it is elliptic of order (0,0,2). Furthermore, (3.13.62) yields the following estimate for the conditioning number y(a) in this case:
Hence, for each p 0 > 0 the corresponding problem (3.13.61) is wellconditioned for the difference operator (3.13.65) uniformly with respect toptp
0'
Example 3.13.16. Let a(x,sD
)
=
Op a, a E !-(ororv)(Rn) be again as in Example 3.13.15 and
let again a(x,p,hD a(x,eD
)
be an elliptic finite difference approximation of
)
on the family of grids R:,h.
We consider again the equation
(3.13.61). which can be solved numerically by using the following iterative procedure: (3.13.66) uv+l = u
-T<ED
-2v > a(x,p,hDx)*(a(x,p,hDx)uv-f), x,h
>-2v is the difference singular perturbation with the symbol where <ED -2" xrh = ( 1 + ~ ~ ] < 1 ~and ) - ~T is a parameter to be chosen to guarantee the <E<> convergence of the iterations (3.13.66). As a consequence of ellipticity condition (3.12.721, one has:
def (3.13.67) 0 < m
=
inf (x,p,rl)xRn X
s sup
R
(x,p,n)ERnx R + XTn 1,n Here, as previously, w ( n ) = C(h,E) with n = ht, h
= 1.
-1 Hence choosing a positive number r E ((M-m)(M+m) ,I) and T
E [(l-r)M,(l+r)M], one gets the conclusion that the principal symbol
q0(x,n) I
3.13. Girding's Inequality
459
of the difference operator Q h,E' Qh,€
=
Id-7<ED > x,h
-2v
a(x,p,hDx)* o a(x,p,hDx)
satisfies the inequality: lqo(x,q) I 5 r,
v
(x,q) E
xn
x T:,,.
Therefore, as a consequence of the corresponding version of Corollary
,
3.13.11 in L2(lRE)
one gets the conclusion that
where C is some constant, which may depend on p . It can be shown, but will not be done here, that (3-13-68)I IQh,Eil
5 r+c(h+E), V (h,E) E (O,ho] X (O,Eol,
L ( W h )L'
(Bh)
where the constant c does not depend on h and E. In fact, one establishes (3.13.68) by using the argument in the proof of Theorem 3.13.8 and estimating carefully the second derivatives of the symbol q of Q
by constants, which in the case considered can be chosen h,E independent on p. This analysis will be done later in connection with the
corresponding version of Theorem 3.13.8 for singular perturbations of order
(O,O,V)
-
For h and
sufficiently small and with the choice of y and
E
T
indicated above, (3.13.68) yields the converqence of the iterative procedure (3.13.66), the asymptotically (for E
-f
0, h
-+
0 ) best choice of
r being r
=
(M-m)(M+m)
-1
with m and M defined by (3.13.67) and the rate of convergence being rv as
v
+
w.
For a(x,ED
)
=
(l+iED q(x))
o
(l-iEq(x)D
)
from Example 3.13.15, one
can use the following iterative procedure: (1+E2Dx,hD:,h)UV+l with the choice of
T
=
[(I+€ 2Dx,hD~,h)-Ta(x,E,h,O)lu+Tf V
as indicated above.
3. Singular Perturbations on Smooth Manifolds without Bounda y
460
Remark 3.13.17. O n L e t p E S (37 )
, p b e i n g v a l u e d i n Hom(CP,CP), i . e .
P(X,E,~)
i s a pxp
m a t r i x . L e t t h e p r i n c i p a l symbol p ( x , & < ) o f p be a n o n - n e g a t i v e h e r m i t i a n 0
m a t r i x : po(x,I1) 2
po E C ( B n x R n )
=
p o ( x , n ) * 2 0, V (x,Tl) E R n x R n and assume t h a t
. Furthermore,
assume t h a t w i t h s o m e c o n s t a n t C > 0 t h e
following inequality holds: ( 3 . 1 3 . 6 9 ) <x>n + l < E ~ > ] D B ( p ( x , E , S ) - p O ( x r ~ , S ) )SI C E ,
v
(X,E,E)
E
mn
5 n+l,
x (O,Eol
x IRn
.
Then e x a c t l y t h e s a m e argument a s i n t h e p r o o f o f Theorem 3 . 1 3 . 8 y i e l d s t h e following inequality ( 3 . 1 3 . 7 0 ) R e ( P u , u ) 2 -cE/ ( u I
1 2(o,o,o)
V
E
E
(O,Eo],
V u E H
(O,O,O) , E
(R?,
= O p ( p ) a n d c i s s o m e c o n s t a n t , which d o e s n o t depend on u and E .
where P
I n e q u a l i t y (3.13.70)
c a n b e s h a r p e n e d : namely t h e norm
o n i t s r i g h t hand side c a n b e r e p l a c e d by a weaker norm
I I . 1 1 (o,o,o)
,€
11. I
(O.O,-f) , E ' However t o show t h i s r e q u i r e s a d i f f e r e n t k i n d of m o l l i f y i n g t h a n t h e one u s e d i n t h e p r o o f o f Theorem 3.13.8. C o r o l l a r y 3.13.18.
Let a E
S ( o t o r v )( R n )
constant, i . e .
be s t r o n g l y e l l i p t i c w i t h y a s t h e strong e l l i p t i c i t y
R e a (x,n)
0
2 y , V ( x , q ) E I R ~ ~ I where R ~ , a O ( x , E < ) is t h e
p r i n c i p a l symb02 of a. Furthermore, assume t h a t a-a
0
s a t i s f i e s (3.13.69).
Consider t h e i n i t i a l value problem
'
u +A u = f ( x , t ) , (3.13.71)
where ut
= au/at
and
(x,t)
E
IRnxn+,
A~ = o p a .
Then t h e r e e x i s t s a c o n s t a n t c , which does n e t depend on u and such t h a t for t h e s o l u t i o n u o f ( 3 . 1 3 . 7 1 ) holds
E,
:
I n d e e d , f i r s t , c o n s i d e r ( 3 . 1 3 . 7 1 ) w i t h f Z 0 . Then a f t e r t a k i n g t h e i n n e r p r o d u c t w i t h u i n t h e d i f f e r e n t i a l e q u a t i o n ( 3 . 1 3 . 7 1 ) , one f i n d s , u s i n g (3.13.70) :
3.13. Ghding's Inequality
46 1
Hence, G r o n w a l l ' s lemma y i e l d s :
I lu(.tt) I I(0)
2
I l U ( . J O ) I I ( 0 ) ,E
2
,€
e x p ( (-y+cE) t )
.
I n o t h e r words, t h e semigroup o f o p e r a t o r s :
*
IR+ 3 t
e x p ( - t A E ) E Hom(H ( 0 ) , E ( R n ) i H
i s a contructionuniformly with respect t o
E
E
(0,E
0
1,
provided t h a t
E~
is
s u f f i c i e n t l y s m a l l , a n d , moreover, one h a s : (3.13.73)
1 lexp(-tAE) I I H ( 0 ), E
S i n c e t h e s o l u t i o n of
(IRn)
6 e x p ( (-Y+cE) t ) , V t E R + ,
v
&
E (O,Eol.
( 3 . 1 3 . 7 1 ) i s g i v e n by t h e f o r m u l a :
t U(X,t)
=
( e x p ( - t A )'$)
(X)
eXp(-(t-T)AE)f(X,T)dT,
+
0 one g e t s i m m e d i a t e l y ( 3 . 1 3 . 7 2 ) , u s i n g ( 3 . 1 3 . 7 3 ) . I n p a r t i c u l a r , c o n s i d e r (3.13.71) f o r A_ = Op(a) w i t h f ( x , t )
Z
f(x)
and a ( x , E , c ) = <E
1
V x E Wn
I
w i t h some c o n s t a n t
support i n Rn.
y > 0 , b ( m ) e x i s t s and b ( x ) - b ( m ) h a s compact
Of c o u r s e , w i t h t h i s s p e c i f i c c h o i c e o f A E, one can
r e f o r m u l a t e (3.13.71)
i n t h i s case a s a d i f f e r e n t i a l i n i t i a l v a l u e problem
of Sobolev t y p e i n t h e f o l l o w i n g f a s h i o n : (1-E
2
Ax) ( u t - f ) + ( l + E 2
C b k j ( x ) Dk D j ) u = 0 , 1 6 j ,k5n
(3.13.74)
A s a consequence of
(3.13.721,
( x , t ) E IRn
X
W,
t h e s o l u t i o n u m ( x ) of t h e c o r r e s p o n d i n g
s t a t i o n a r y problem (3.13.75) ( l + E with g
=
2
c bkj(x)D D )um(x) = g(x) k j lsk, j
( l - ~ ~ A ) if s, t h e o n l y a t t r a c t o r f o r ( 3 . 1 3 . 7 4 ) . which i s , moreover,
asymptotically s t a b l e f o r (3.13.74), a s t
+
9.
3 . Singular Perturbations on Smooth Manifolds without Boundary
462
Furthermore, (3.13.72) yields the following estimate for the difference v = u-u,:
min pk(x) > 0, pk(x) > 0 being the min{l,p} with p = inf xEIRn l
=
One can use (3.13.74) for solving numerically (by using the iterations) the stationary problem (3.13.75), as it is indicated in Examples 3.13.15, 3.13.16 in the case n = 1. Now a corresponding version of the sharp Ggrding's inequality,the LaxNirenberg theorem, will be proved for classical pseudodifferential operators in Op S'(pin)
with
V
E
IR
. Again
the sLymbols considered are valued in
Horn(CP;CP), i.e. they are p x p matrices. We follow here essentially the mollif ier method introduced in [ Friedr., 2 over in [Vai
r
1
1 with the simplifications brought
I.
We start with the case of pseudodifferential operators of order zero. Theorem 3.13.19.
. Then
L e t p(x,E;) be hermitian non-negative d e f i n i t e m a t r i x , p E
f o r t h e corresponding p s e u d o d i f f e r e n t i a l operator P (3.13.76) Re(Pu,u) 2
where
I / . I l -t
-C\ luj
I",
V u
E C:(IRn)
= op(p)
one has:
,
is t h e cZassicaZ Soboieu-Slobodetski
norm of order - t and
where c is a c o n s t a n t , which does n o t depend on u. Proof. ___ The idea is to produce a symmetric non-negative quadratic form Q ( u , u ) such that with some constant C > 0 holds:
First, we notice that
2 (3.13.78) I/
3.13. Girding's Inequality where t h e f a m i l y of symbols: lRn 3 5
-f
463
@ ( 5 , < ) E lR w i l l be chosen l a t e r on
a s a consequence o f n a t u r a l r e q u i r e m e n t s l e a d i n g t o ( 3 . 1 3 . 7 7 ) . One h a s :
p*u
=
(2n)-nj/ei<x-Y1S' P ( Y t 5 )
( Y ) dYd5 r
so t h a t an e a s y c o m p u t a t i o n y i e l d s :
F i r s t , we e s t i m a t e t h e d i f f e r e n c e Q ( u , u ) - ( P u , u ) 0
Rewrite Q ( u , u ) as follows: 0
w e c o m e t o t h e c o n c l u s i o n t h a t 0 ( 5 , < ) s h o u l d b e chosen t o g u a r a n t e e a good a s y m p t o t i c a p p r o x i m a t i o n o f p ( y , < ) by t h e i n t e g r a l
1
de f (3.13.81)
PQ( Y
J0 ( 5 , , < )
=
2
p (y,5 )d < .
For d o i n g t h a t w e i n t r o d u c e 0 ( 5 , 5 ) a s f o l l o w s : (3.13.82) 0 ( E , < ) = 6 n ' 2 $ ( 6 ( < - 5 ) ) , where 6 = 6 ( 5 ) + 0 , a s 5 m n c B~ @ E c o ( ) ~, supp (3.13.83)
/ @ ( e 2) d8
=
-f
a,
i s t o b e d e f i n e d l a t e r on and where
=
~e
E nin
I lei
< 11,
0 2 o and
1,
t h e l a t t e r c o n d i t i o n b e i n g n e c e s s a r y i n o r d e r t o have an a s y m p t o t i c a p p r o x i m a t i o n of p ( y , S ) by p , ( y , < )
as
5
-f
m.
3. Singular Perturbations on Smooth Manifolds without Boundary
464
I n d e e d , w i t h t h e c h o i c e o f O(E,C,) t h e change of v a r i a b l e s 6 ( < - 5 ) (3.13.84) p,(y,S)
=
/$(el
2
a n d , as a consequence o f
a s i n ( 3 . 1 3 . 8 2 ) . one f i n d s u s i n g
= 8:
p(y,5+6-'a)dB
(3.13.83),
( 3 . 1 3 . 8 4 ) and h a v i n g i n mind a n
a s y m p t o t i c a p p r o x i m a t i o n o f p ( y , s ) by p ( y , S ) a s €2 O
m,
-f
one r e a c h e s t h e
c o n c l u s i o n , t h a t 6(5) s h o u l d b e c h o s e n t o s a t i s f y t h e c o n d i t i o n : ( 3 . 1 3 . 8 5 ) 6(<)
+ m,
as
<
-f
m.
F u r t h e r m o r e , u s i n g T a y l o r ' s f o r m u l a f o r p(y,c+6-1'8) w i t h t h e r e m a i n d e r of s e c o n d o r d e r i n t h e i n t e g r a l form, one g e t s f o r p ( y , < ) t h e f o l l o w i n g O
r e p r e s e n t a ti o n :
2 ( 3 . 1 3 . 8 6 ) p O ( y , s ) = P ( Y , S ) + ~ ( C ) - ~$/( e ) < e , v 5 p ( y , t ) > d e +
nn
l
2
+ 6 ( ~ ) - ~ . r ~4 (8) ( i - t ) < p E 5 ( y , ~ + t ~ - l ae ,)e > d t d e . Rn
0
Hence, one g e t s a b e t t e r a s y m p t o t i c a p p r o x i m a t i o n o f p ( y , 5 ) by r e q u i r i n g that
2
( 3 . 1 3 . 8 7 ) / n 8 $ ( 8 ) dB
=
0.
Q7
F o r i n s t a n c e , c h o o s i n g $ ( a ) t o be e v e n : $ ( - € I )
=
( 3 . 1 3 . 8 7 ) i s a u t o m a t i c a l l y s a t i s f i e d , so t h a t i f 6 ( 5 ) w i t h some
CY
$
('a), V 8 E
-
IRn
-a <<> as 5
-f
, a,
E ( 0 , 1 ) , then
( 3 . 1 3 . 8 8 ) ( p -p) E s 2 ( a - 1 ) (IRn)
,
a s a n immediate consequence o f
(3.13.86),
(3.13.87).
N o w w e p r o c e e d t o e s t i m a t e t h e q u a d r a t i c form
(3.13.89) R ( u , u ) =
( 2711 -nJ/J/O ( 5
d ef =
, 5 ) ( Q ( 5 2, C, ) -0 ( 5
,c ) )
Again, w e u s e T a y l o r ' s f o r m u l a :
Furthermore,
(3.13.82) y i e l d s :
ei<'
1-52>
(5 )
, (5
)> d5 d5 2dydL
3.13. Gdrding's Inequality
where we have denoted by
R 1 (u,u) the
465
quadratic form, which corresponds to
the remainder in the integral form on the right-hand side of (3.13.90) Integrating by parts in the first integral on the right hand side of (3.13.93), after having rewritten $$
5
as ( f )
2
($ )<,
one gets:
(3.13.94) R(u,u) =
+ R(u,u) + R1(u,u),
where we have denoted by R(u,u) the quadratic form defined by the second integral on the right-hand side of (3.13.93). As
previously, one gets the conclusion that
(3.13.95) 1 0 (5,5)2 1 P ~ , ~ , ( Y , < E) ~S-l(Rn) < l<jSn J
.
Furthermore, one finds (after the change of variables 6(5-E,)
=
8):
so that (3.13.88), (3.13.97) yield the optimal choice of a by equating 2(a-1)
=
-1, i.e. a
=
f.
Notice that for each pseudodifferential operator a
E Op S-'(Rn)
holds:
3. Singular Perturbations on Smooth Manifolds without Bounda y
466
I Indeed
one gets immediately the last inequality by introducing
-1
v =
E Op
S ( O f O ' O ) (mn)
with s = (O,O,O). Thus, the last inequality, (3.13.88), (3.13.97) with a =
+
yield:
where we have denoted by Po the pseudodifferential operator with symbol pa given by (3.13.84).
Now we proceed to estimate quadratic form R (u,u), 1 (3.13.99) ? ? , ( u , u ) =
3.13. Girding's Inequality
467
one can rewrite (3.13.99) as follows (after the integration by parts with respect to y) :
(3.13.102)
/I
(1-A
Y
)
k r1(y,clrL2) Idy 5 C, V (5,,5,)
E RnxRn.
Hence, (3.13.101), (3.13.102) yield: IR1(utu)I 5
CI
/uI
2 I-+,
so that one finally gets: (3.13.103) IQo(u,u)-P(u,u)1 6
CI
lul I:+,
V u
E C;(Rn).
Now, since (P*u,u) = (Pu,u)* and QQ(u,u) is real-valued, one gets, as a consequence of (3.13.103), the same inequality for the adjoint P*: (3.13.104) lQ (u.U)-(P*U,U) Q
I
5 CI
l2
v
-f'
u E C;(Rn),
so that (3.13.103). (3.13.104) yield: 2
(3.13.105) IQQ(u,u)-Re(Pu,u)I 5 CI
1 -4'
using the non-negativity of 2 ( u , u ) and applying (3.13.105), one Q finds: -Re(Pu,u) 5 Q (u,U)-Re(PU,U) 5 CI 0
so that
Re(Pu,u) t -C(/uI
/-*f '
v
lU1
u E C:(Rn).
2 I-+,
I
Remark 3.13.20. One is tempted to take 6(5)
G
1 (a
= O),
by (3.13.92), vanishes: R ( < , S ) E 0.
since in that case R(S,<), defined
3. Singular Perturbations on Smooth Manifolds without Boundary
468
However, while choosing a = 0, one looses control over the quadratic form -a R (u,u) defined by (3.13.99), which, on the other hand, with 6(5) = <E,> , 1
2
a > 0 , can be estimated by CI /uI1 -
. Thus, actually, the optimal choice of
a is assigned by (3.13.88) and by the estimate of R1(u,u), i.e. by equating 2(a-1) = -2a, which yields a = $ .
#
The following result is a combined version of Theorem 3.13.8 and Theorem 3.13.19 for singular perturbations of order ( O , O , O ) .
It can be
established in the same manner as Theorem 3.13.19 with the choice of
6
= E
f <ES> -l’, so that its proof is left to the reader.
Theorem 3.13.21.
Let P
=
P(x,E,S). p E s ( o ‘ o r o ) ( x n )
. Assume
t h a t t h e p r i n c i p a l symbol
c2(xnxxn)and i s a non-negative hermitian m a t r i x . Furthermore, assume t h a t t h e r e e x i s t s a c o n s t a n t c > 0 such t h a t pO(x,~S)of p(x,c,<) belongs to
(3.13.106) <x> n+l< E S > ~ DB ~ ( P ( X , E , S ) - ~ ~ ( X ,IE S )C)E ,
Then f o r P
=
! B i 5 ncl,
op(p) holds:
where t h e c o n s t a n t k does n o t depend on
E
and u .
The following result is a finite difference version of Theorem 3.13.19 for difference operators of order (O,O), whose full proof for difference operators of any order (with the corresponding necessary modifications in the formulation of Gsrding’s inequality in this case) can be found in [Fr.,9,101. Theorem 3.13.22.
L e t p(x,h,h<) E F ( o ’ o f o ’ 1,o
(an) b e Hermitian non-negative m a t r i x . Then f o r
t h e corresponding d i f f e r e n c e operator Ph = p(x,h,hD) h o l d s :
where t h e c o n s t a n t X does n o t depend on h and u
.
Proof. ~
One has just to repeat the proof of Theorem 3.13.19, a suitable choice of the finite difference substitute for the family of pseudodifferential operators 5 + O(D , 5 ) in (3.13.79) being the family of difference operators
3.13. Girding's Inequality
where
$(e)
i s t h e s a m e f u n c t i o n a s i n t h e p r o o f o f Theorem 3.13.17 and
where, a s p r e v i o u s l y , w ( r l ) Re w(nk )
=
469
(u(rll),.,.,u(n
=
) ) with
u(n
k
)
=
i(l-exp(iqk)),
I
s i n qk.
One s h o u l d stress t h e d i f f e r e n c e between Theorems 3.13.8
and 3.13.22,
0 which c o n s i s t s i n t h e f a c t t h a t t h e c l a s s o f symbols p E F ( I R ~) , s a t i s f y i n g i s more r e s t r i c t i v e t h a n t h e one i n
t h e a s s u m p t i o n s of Theorem 3 . 1 3 . 8 ,
Theorem 3 . 1 3 . 2 2 , t h e l a t t e r g i v i n g r i s e t o a s t r o n g e r r e s u l t ( 3 . 1 3 . 1 4 ) . 2 2 2 ( N o t i c e t h a t one h a s a l w a y s h / O , h 5 (hO+4n) V h E (O,hol . I lul
1111 1
I
See a l s o Remark 3 . 1 3 . 9 . C o r o l l a r y 3.13.23.
Let b ( x , S ) E S 2 m ( l R n ) , m E R , b e such t h a t i t s p r i n c i p a l symbol b (x,S! i s a non-negative hermitian m a t r i x : b o ( x , S )
2 0 , V (x,C)
0 E IRnx(Rn,{O}).
Further, assume t h a t t h e order of b ( x , E ) - b ( x , < ) is 5 2m-1. Then t h e r e 0
e x i s t s a c o n s t a n t k such t h a t f o r t h e p s e u d o d i f f e r e n t i a 2 operator
B = Op(b)
ho Zds (3.13.110) R e ( B u , u )
where
, ) s,
(
(u,vlS
=
s
E R
2 -kl lul
,
2 1 s+m-f ' v
u E C;(Rn
),
stands f o r t h e i n n e r product d e f i n e d by
(
with
,
(
)
=
0
.
i.e.
L ~ ( I )R ~
-s-m
s+m
u , P =
I n d e e d , i n t r o d u c i n g v =
t h e i n n e r product in
0
h e r m i t i a n m a t r i x , one g e t s (3.13.110) by a p p l y i n g Theorem 3 . 1 3 . 1 9 t o t h e 0 p s e u d o d i f f e r e n t i a l o p e r a t o r Po w i t h t h e symbol P o ( x , 5 ) x ( 5 ) E S ( R n) , where x E Cm(IRn ) i s a c u t - o f f f u n c t i o n , which v a n i s h e s i d e n t i c a l l y f o r 1 5 ) 5 f , is identically 1 for
(c(2
1 and i s n o n - n e g a t i v e o t h e r w i s e :
The f o l l o w i n g s l i g h t g e n e r a l i z a t i o n o f Theorem 3.13.19
x
2 0, V
5 E Rn
c a n be proved
by t h e s a m e argument.
Let p
=
p ( x , e , ~ ) ,p E
p r i n c i p a l symbol p o ( x , n ) , n
s ( o ~ o( ~ m nu) )w i t h some u E R . Assume t h a t t h e 2 = E E , o ~ P ( x , E , ~ )belongs t o c ( I R ~ X I R ~ ) and is x r l
a non-negative hermitian m a t r i x . Furthemore, assume t h a t
1
~ x > n + l < ~ ~ ~ lB- u ~ ~ X ( P ( X , ~ , ~ ) - p5 Oc E( ,x , 1E8~1 )5) n + l ,
v
(X,E,S)
E ~ n x ( O , E o ]E n .
.
470
3. Singular Perturbations on Smooth Manifolds without Bounda y
Then f o r pE
= op(p)
Re(P u,u) E
and f o r each
(S),E
s
3
E IR holds: 2 (s1,s2,s3+(v-1)/2)
2 -csl/u//
where t h e c o n s t a n t c does n o t depend on
E
.El
v
E
E (O,EO1'
and u.
Corollary 3.13.24. (0.2m) m E R be such t h a t i t s p r i n c i p a l symbol Let b(x,h.W) E F l , o ho(x,n) i s a non-negative h e m i t i a n m a t r i x : h (x,n) 2 0, 0
{O}). Further, assume t h a t t h e order o f b(x,h,hc)i s 5 2m-1. Then t h e r e e x i s t s a c o n s t a n t k such t h a t f o r t h e
V ( x , n ) E IRn x(T:
h-2%o(x,h<)
\
corresponding d i f f e r e n c e operator B~ (3.13.111)
where
(
=
op (h) holds:
Re(B u,u) h s,h 2 -kl IuI 1 2 s+m-! ,h' V h E (O,hol, v u E Hs+m,h(J()
, )s,h d e f i n e d by
(2.7.6)
with
z S ~ , < ~ x , h > S ~ )whi,t h ( = (
s1 = 0 , s2 = s ,
,
,
i.e.
)h t h e i n n e r product (2.7.1)
Indeed, one uses the same argument as in Corollary 3.13.23
and Theorem
to show ( 3 . 1 3 . 1 1 1 ) .
3.13.22,
Corollary 3.13.25.
L e t a(x,<) E sm(Rn), m E i ~ be , e l l i p t i c o f order m , i . e . f o r t h e p r i n c i p a l symboZ a of a h o l d s : lao(x,<)1 0
w i t h some c o n s t c n t y
>
2
m y l < / , V (x,<)EIR~~(R"\IO})
0, t h e b e s t p o s s i b l e value yo o f y being t h e
e l l i p t i c i t y c o n s t a n t . Further, ussume ( h a t t h e order o f a(x,c)-a (x,<) i s 0
5 m-1.
Then t h e r e e x i s t s a c o n s t a n t c such t h a t f o r t h e corresponding
p s e u d o d i f f e r e n t i a l operator A
=
op (a)h o l d s :
Indeed, applying Corollary 3 . 1 3 . 2 1 to B = A*
0
A-y
immediately ( 3 . 1 3 . 1 1 2 ) . With
4,
=
Op (a)an elliptic difference operator of order m (a E
Fm1,o
whose ellipticity constant is y o , the difference analogue of ( 3 . 1 3 . 1 1 2 ) true, too, i.e.
(IRn)), is
3.13. Girding's Inequality
47 1
Example 3-L3-.2. The following matrix differential operator B(D
)
appears in the linear
elasticity theory (see, for instance, [Land.-Lif. , 1 1 ) : (3.13.114) B(D
)
=
2 UD I d + ( V + A )
I IDx .Dx 1 1 lij,kcn' 3
k
'-
C
DX -
l
2 D x ' k
where Id E Hom(Cn;Cn) is the identity and p > 0, X 1 0 are given constants. It is easily szen that the matrix R([) non-negative hermitian. Indeed, for each z
=
1
= =
I ( [ 7.[ k 1 1 l<j,k
5 E I R ~ ,is
E Cn holds:
2
Sjzj( , V 5 E R n ,
Z l'j
where, as usual, <
,>
Hence, B([) -p 15 I 'Id
is the inner product in C". 2 0,
5 E Rn and using the Fourier transform, one
finds easily that ( B ( D x ) ~ , ~ )2s l i / ( 2~ ( l ~v+ s~ E , R , V u E Ci(IRn). If p
=
p(x) 2
u0
>
0, V x E m n , X = X(x) 2 0, V x E m n , X E S ( R n ) and
= u_+u'(x), with p ' E S(Rn), then applying Corollary 3.13.21 to ~(x) 2 B(x,D )-!~(x)It) Id, one finds:
2 2 (3.13.115) R ~ ( B ( X , D ~ ) U , U2 )u~o \ ( u !/s+l-C1/ U (Is++,
'd
*, v
E
S
U
E Ci(Rn),
where the constant C does not depend on u. The following finite difference approximation B(D
x,h
,D:,h)
of B(Dx)
given by (3.13.114) is elliptic of order 2.
u(x)
=
u_+u'(x), p' E S(IRn):
then the difference analogue of (3.13.115)
holds uniformly with respect to h E (0.h 1 with the same uo and a constant 0
C, which does not depend on h E (O,hol and u E Hs+l,h(<
),
b u t may depend
I
on s E 37. Example 3.13. 27. Let a(x,t) a(x,t)
=
=
\lakj(x,t)l/,a(x,t)* = a(x,t) t a0Id, . a
a_+a'(x,t), a(x,t) E C;(mRn
xE+)
> 0, V (x,t)ERnx%+,
and let V(x,t) 2 vo
0,
3 . Singular Perturbations on Smooth Manifolds without Boundary
412
v
(x,t) E lRnxiii+, V(X,t) = V,+V'
(x,t), V' (x,t) E
c p nG+).
Denote At =
-
IR+3 t
+
At E Op
S
2
n
(IR
being a family of second order elliptic
)
differential operators with the ellipticity constant a
0'
As a consequence of Corollary 3.13.23, for each s E IR there exists a constant C > 0 such that
On the other hand, the integration by parts yields 2
(3.13.117) (Atu,u) 2 yoI lul 11, V u E C;(lRn), where yo = rnin(ao,Vo) > 0. Consider the initial value problem
( at (-A
x
+I)+A )u(x,t) = 0 , t
(x,t) E
mnx=
(3.13.118) U(X,O)
=
u
0
x E IRn
(XI I
a n where uo E C (IR ) and A is the Laplace operator in x E lRn. 0 We are going to show that for each integer m > 0 there exists a constant
C
m
such that
After taking the inner product, (3.13.117), (3.13.118) yield (f)d/dt//u(.,t)([,+y,llu(.,t)ll, 2 2 5 0,
V t 2 0.
so that
(3.13.120)
I lu(.,t) I l 1
5
.-Yoti
luol 11, V t 2 0.
( a t (1-Ax )+At )u = 0 and taking the inner product with one finds using (3.13.116):
Applying
2 0 = (+)d/dt(/u(.,t)[ Is+l+Re(Atu,~)s 2 Z
(?)d/dt/lu(.,t) I
Is+l2
+aoj lu(.,t) I
I s +2l
2 I Is++,
-CI lu(.,t)
3.23. Girding's Inequality
473
so that Gronwall's lemma yields:
Taking s
=
f in (3.13.121) and using (3.13.120) to estimate the
integral on the right hand side of (3.13.121), one finds:
We have used the obvious inequality:
1
e
-2a
t
O( -T)-2yo'dT
v
5te-2yot,
t 2 0, . a
2 yo.
0 Now taking s
=
3/2 in (3.13.121) and using (3.13.121), (3.13.122), one
gets the inequality
By induction argument, one gets for each integer m > 0 the following estimate
+ (ct1~~-~/((2m-2) !)e2Yot//uOl.:1 and this last inequality yields (3.13.119) with C m
max Ck/k! 0< k < 2m- 1 Using the interpolation, one can show(but it will not be done here), =
that, in fact, (3.13.119) holds for V m L f . With x E R , y E R and A = D2+D2 equations similar to (3.13.118) t X Y have been previously considered in [Sob. 3 1. I
The corresponding evolution problems being not of Cauchy-Kowslevski's
3 . Singular Perturbations on Smooth Manifolds without Boundary
414
type, they are known as Sobolev type initial (and initial boundary) value problems (see also [Galp., 1
1,
1).
[Fr., 1
I
Example 3.13.28. The following singular perturbation appears in the theory of gaussian perturbations of dynamical systems (see, for instance, [Freid-Wentz. , 1 (3.13.123) a(x,E,D where b
=
-EAX-,
x E IRn, m
m
: lRn
-f
I):
Rn is a C -vector field, b E C (Rn) , and where, as usual,
Ax is the Laplace operator and V
is the gradient in x E Rn
.
Assume that b(x) admits the representation: (3.13.124) b(x) = -2Vx$(x)+g(x),
with $ E Cm(IRn), g € Cm(lRn), 4 : Rn
+
IR, g
:
Rn
Assume, furthermore, that there exists a vector
+
$m
En.
E Rn such that
(b(x)-g(x)+$m) E S(IRn), i.e. $(x) admits the representation (x), $ ' (x) E s(lRn).
$(x) = <$,,X>+$' Let us assume that qo = inf
xE mn
points, and denote yo
=
IVx$(x) 1
2
> 0, i.e. $(x) does not have critical
min{l,qo}, yo > 0.
Then the following inequality holds: (3.13.125)
I
exp(-E-'$ (x)) u (x)*a (x,E ,Dx)(exp(E-l$ (x)) u (x)) dx 2
R
where the constant C does not depend on
E
and u .
Indeed, using the equality b(x) = -2V $(x)+g(x), one easily finds: (3.13.126)b (x,E ,Dx
def =
6
= -E
exp ( -E
-1
2 AX-E
with g(x) defined by (3.13.124). Notice, that
$ (x)) a (x,E ,Dx)exp ( E-'$ (x)) =
3.13. Girding‘s Inequality with C = ( f ) sup
475
I
,Emn Hence, applying the second part of Corollary 3.13.21 to P ( x , E , ~ )= E
2
2 1512+/$,(X) 1 2-EA,$(X)-Y~(~+E 151 2) , p E S(0f0‘2)(X7n) with p (x,rl) =
0 2 2 2 111 +Iox(X) -Yo
1
(3.13.125). Consider the corresponding parabolic initial value problem, which describes theMarkov process with the infinitesimal generator a(x,E,Dx) given by (3.13.123) :
Using (3.13.125) one gets the following estimate for the solution u(x,t) of (3.13.128) : (3.13.129) Jnexp(-2~-l$(x)) (u(x,t)1 2dx 5 IR
-1 5 exp(-2~ (yo-CE)t) J
n
2 exp(-2~-’$(x)) luo(x) I dx,
R where yo,C and $(x) are as defined here above and where $(x) is still supposed to have no critical points: $” # 0, V x E Rn. -1 Indeed, putting u = v e x p ( $ ~ ) , v = v(x,t) is the solution of the ‘b
Cauchy problem: -1 (at+€ ~ ( X , E , D))v(x,t)
=
0,
(x,t) E I R ~ X R +
(3.13.130) V(X,O) where v
=
=
uo exp(-s
v (x), 0
-1
0) and b ( x , ~ , D) is defined by (3.13.126)
Now, (3.13.125) yields: (3.13.131) Re(b(x,~,D)v,v) 2 (y -CE)1 /v1I 2 0
Thus, taking the inner product in the equation (3.13.130), using (3.13.131) and Gronwall’s lemma, one gets (3.13.129). The condition $ (x) # 0, V x E lRn , $(x) being linear function for 1x1 sufficiently large, means that, up to the orthogonal to $
component
g(x), the vector field b(x) is diffeomorphic to a constant vector field .0, Of course, the most interesting situations in the theory of random
3 . Singular Perturbations on Smooth Manifolds without Boundary
476
p e r t u r b a t i o n s o f dynamical s y s t e m s (see [Freid-Wentz
,
1)
1
and o t h e r
f i e l d s of a p p l i c a t i o n s of s i n g u l a r p e r t u r b a t i o n s ( f o r i n s t a n c e , r e a c t i o n d i f f u s i o n p r o c e s s e s w i t h l a r g e d r i f t v e c t o r f i e l d s ) , are p r e c i s e l y t h e o n e s when b ( x ) ( a n d , t h u s , a l s o $ ( x ) ) h a s c r i t i c a l p o i n t s . I n t h a t case t h e b e h a v i o u r o f t h e t r a j e c t o r i e s o f t h e s y s t e m of o r d i n a r y d i f f e r e n t i a l equations A ( t )
=
b ( x ( t ) ) i n t h e neighbourhood of t h e c r i t i c a l p o i n t s of
b ( x ) ( s t a t i o n a r y s o l u t i o n s ) p l a y s t h e c r u c i a l r o l e i n t h e i n v e s t i g a t i o n of t h e a s y m p t o t i c b e h a v i o u r o f t h e s o l u t i o n s t o (3.13.128)
and t h e c o r r e s p o n d i n g
s t a t i o n a r y e l l i p t i c problem w i t h a s m a l l p a r a m e t e r i n domains i n Wn 2 I n p a r t i c u l a r , f o r n = 1 , uo E L ( W ) , supp uo 5 IR+ , b ( x ) : 1 ,
.
i n e q u a l i t y (3.13.129) y i e l d s f o r t h e s o l u t i o n u of (3.13.128):
1
exp(-2~-'X)/ u ( x , t )
=-
I 2 dx
-1
5 exp(-2~ (l-Ce)t)I
exp(-2e-'x)
2 / u ( x ) ; dx 0
IR+
i . e . u ( x , t ) f o r t > 0 , x E IR-
w i t h r e s p e c t to ( x , t )
E IR-
xW+
i s exponentially s m a l l a s
E
+
0 , uniformly
w i t h x + t 2 y > 0.
0
N o t i c e t h a t i n t h e l a t t e r case t h e ( g e n e r a l i z e d ) s o l u t i o n u ( x , t ) o f 0 t h e reduced problem ( E = 0 ) a s s o c i a t e d w i t h ( 3 . 1 3 . 1 2 8 ) , i s u ( x , t ) = uo ( x + t ) , 0 so t h a t u ( x , t ) Z 0 f o r x + t < 0 . I Example 3 . 1 3 . 2 9 . L e t m > 0 b e i n t e g e r and l e t a E S ( o ' o r 2 m )( R n) ,
be a s t r o n g l y e l l i p t i c d i f f e r e n t i a l s i n g u l a r p e r t u r b a t i o n of o r d e r (0,0,2m), i . e . t h e p r i n c i p a l symbol a ( x , q ) = a ( x , O , q ) , q = € 5 , o f a ,
0
s a t i s f i e s t h e c o n d i t i o n : R e a ( x , n ) 2 y
0
2m
,
V (x,n) E WnxWn
w i t h some
c o n s t a n t y > 0 , t h e b e s t p o s s i b l e value yo of y being c a l l e d t h e s t r o n g e l l i p t i c i t y constant f o r a. F u r t h e r m o r e , assume t h a t t h e c o e f f i c i e n t s a ( x , E ) s i d e of
(3.13.132)
a r e L i p s c h i t z continuous i n
E
E [O,E
on t h e r i g h t hand
0
1
uniformly with
r e s p e c t t o x E R n , so t h a t , i n p a r t i c u l a r , t h e r e e x i s t s a c o n s t a n t C s u c h that (3.13.133)
]a(x,E,n)-a(x,O,q)
1
5 Ce
2m
,
V ( x , q ) E mnxlRn
C o n s i d e r t h e p a r a b o l i c s i n g u l a r l y p e r t u r b e d i n i t i a l v a l u e problem:
3.13. Girding's Inequality
(a/at+a(x,~ED ))u(x,t)
=
f(x,t),
477
(x,t) E IRnxlR+
(3.13.134) U(X,O)
=
u
0
x)
.
Then, there exists a constant k E IR such that the following estimate holds for the solution u(x,t) of (3.13.134):
where y o is the strong ellipticity constant for a(x,E,ED 1 . Indeed, first consider (3.13.134) with f E 0. Let p
=
Re a, q
Im a. One finds for q(x,E,EDX)
=
(3.13.136) (Re(iq(x,E,~D )u,u) 1 =
+I
=
Op(q)using (3.7.16)
(q(x,E,tD )-q(x,E,ED )*)u,u) I 5
since, as a consequence of (3.17.16), one has:
Furthermore, using again the second part of Corollary 3.13.23 and applying it to a (x,ED )-y ( ~ - E )m ~ Awith Ax the Laplace operator, one 0 x o finds:
where k l is some constant which does not depend on
E
and u,
Combining (3.13.133), (3.13.136), (3.13.137), one gets immediately the inequality:
Hence, (3.13.138), (3.13.134) (after taking the inner product in differential equation (3.13.134) with f F 0) and Gronwall's lemma yield (3.13.135) with f F 0. Now, the same argument as in Corollary 3.13.18, leads to (3.13.135) with V f(.,t) E H
(0),E
(lRn), v t 2 0.
3 . Singular Perturbations on Smooth Manifolds without Boundary
478
One can show in the same way that, in fact, for each s E R3 , the solution u of (3.13.134) one has:
Example 3.13.30. Let A(x)
:
Rn
-f
(3.13.139) a(x,D
Hom(R )
n
,R
n
b(x)
),
:
Rn
-f
R n , c(x) : Rn
= ++c(x),
-f
IR and let
x E R n , (with < , > the
real inner product) be an elliptic second order differential operator with m
C
-coefficients and the ellipticity constant y o > 0. As usual, we assume that there exists Am, bm and cm such that
(A(x)-A_) E S ( R n ) , (b(x)-bm) E S(Rn), (c(x)-cm) E S(IRn) which means that each component of the corresponding matrix- or vector-function belongs to Schwartz's space S(Rn). Consider the initial value problem: (2/2t+a(x,Dx))u(x,t) = 0,
(x,t) E RnxWt
u(x.0) = uo(x),
x E Rn
(3.13.140)
and its discretization in time variable t: (3.13.141) {(Id+rTa(x,DX ))OT -(Id-(I-r)Ta(x,D X ))lvT (x,t) = 0 , vT(x,o) = uo(x),
-
(x,t) E lRnxml T,+
where r E (0,1] is a parameter to be determined later on, and where, as -1
= It = kT, k E Z, k 2 0 ) and OT is the shift operator on T,+ -1 the mesh g:,+, (OTv)(t) = v(t+.r), V t E R T , + , Id being the identity
previously, IR
operator. Let For
v
E
E
T = E*,
O,E
0
3
b(x,E,D
with
E~
)
=
2 Id+rE a(x,D
),
b E Op S(o'or2) ( R n )
.
sufficiently small, there exists a singular
perturbation b-l (x,€,DX) E Op S(080r-2) ( R n ) , such that (b 0 b-l-Id) E op s(-l,O,l) (Rn ) . Indeed, one can define b-l(x,c,Dx) in the following fashion: b-'(x,E,Dx)
=
-1 Op b0(x,€<) with b ( x , q ) = l+r
(3.7.17), (3.7.18) will yield:
0
3.13. Gdrding’s Inequality -1 (b(x,E,Dx)
b
(b-l(x,E,Dx) Therefore for
v
419
(x,€,Dx)-Id)E Op S(-lro’l) ( R n1
I
b(x,€,Dx)-Id) E Op S ( - l r o r l ) ( R n ) .
E ( O , F ~ ] with c0 sufficiently small, there exists the
E
-’ which is a pseudodifferential singular
inverse operator b ( x ,,Dx) ~
perturbation in S ( o ’ o r - 2 1 (IRn ) . 2 Introducing QE = (Idirc a(x,Dx))-l
2 (Id-(l-r)s a(x,D
E Op S ( O r O ’ O ) ( ~ n :
))
one can write the solution of (3.13.141) in the form:
One finds easily that the principal symbol q ( X , E ~ E ) S(o’ofo)(IRn) 0
of Q, is given by the formula: qo (x,n) =
(
l+r
-’ I(
a n ,rp)
( 1 -r)
1
so that /qo(x,n) 6 1, V (x,n) F IRnxIRn, provided that r E [t,l].
Hence, as a consequence of (3.13.26) (with E~ instead of h), one finds: (3.13.142)
1 I (QE)t”l 1
5
L ~ ( I R+~L)~ ( I R ~ )
l I Q E 1 1 t/T
5 ( l+CT ) t/T 5
L2 (nn)+L2 ( R n )
5 exp(Ct),
v
t E IR
1
.
T ,+
It can be easily shown, but will not be done here, that a s a consequence of the stability result (3.13.142), the family of solutions + vT(x,t) of (3.13.141) converges in L2 (IRn ) to the solution u(x,t) of
T
(3.13.140) and, moreover, one has: I (u(.,tl-vT(.,tl 1 1 5 CTI ( U 1 1 0 2’ 1 L2 ( Rn) V t E ]RT , +, where I 1 . 1 is the classical Sobolev norm of order 2 and
I
C > 0 is a constant, which does not depend on u, vT, uo and T . Instead of (3.13.141), one can consider a full discretization of (3.13.140) on IR:,hx$,+,
using any elliptic finite difference
approximation of a(x,D 1 , for instance, the following one: a (x,h,Dx,h,,:D
h) =
.
h>+ ( + )
h>+c (x)
Then, an argument similar to the one used above, leads to the corresponding to (3.13.142) stability result for the family of difference singular perturbations Q T
QT,h
=
,h ( I d + r ~ a ( x , h , D ~ , ~ , D : , ~ )o- ~(id-(l-r)Ta(x,h,D
x,h ,D* x,h)
3 . Singular Perturbations on Smooth Manifolds without Boundary
480 in n
( 0 ) ,h
that r
(q)uniformly with respect to ( ~ , h E)
E [$,I] and ~
( O , ’ T ~ I X ( O ,provided ~~I
~ are, sufficiently h ~ small.
I
Example 3.13 .Xl . Let A(x,t)
:
IR x[-T,T]
+
E Cm(IRx x[-T,T]) and let
Hom(Rp ;lRp), A
A(x,t) = Am(t)+A’(X,t), A’(.&)
E S ( R x ) , v t E [-T,T].
The operator (3.13.143)L(x,Dt,Dx)
def =
i (DtId+A(x,t)D
)
is said to be strictly hyperbolic for (x,t) E Rx[-T,T],
if the eigenvalues
A.(x,t), 1 5 j 5 p, of A(x,t) are real, A . E I R , V (x,t) E IRX[-T,TI and, 7 1 moreover, there exists a non-singular matrix C(x,t) : R x[-T,T] + ISO(IRp;IRp), C E Cm(lRxx[-T,T])
such that
(3.13.144) A(x,t) = C(x,t)-ldiag(A.(x,t)) < , < C(x,t), V (x,t) E IRxX[-T,T]. 3 l=J,p For instance, if there exists a constant 6 > 0 such that /A.(x,t)-Ak(x,t)(2 6 > 0, V (x,t) E IRx[-T,T], V j # k , then (3.13.143) is
I
strictly hyperbolic. Also, if A(x,t) is symmetric, V (x,t) E lRxX[-T,T], then (3.13.143) is hyperbolic. Consider the initial value problem
x
u(x,O) = uo(x),
E
9.
Let Pk(X’t), P (x,t) k
=
-1 (2ni)
J
/X-Ak(x,t)j = p
( AId-A(x,t)) -‘dh
with p > 0 sufficiently small, be the projector on the one-dimensional invariant subspace of A(x,t) in IRp which corresponds to the eigenvalue
X (x,t) of A(x,t). k
It is
readily seen that (3.13.144)
implies that
P (x,t) depends smoothly on (x,t) E Rx[-T,T] and, moreover, k P (x,t) = P (t)+P;(x,t), V (x,t) E RxX[-T,T], with P;(.,t) E S ( I R ) , k krm V t E [-T,T].
Thus, with Pk(x,t)* the adjoint of Pk(x,t) , the matrix M(x,t) M(x,t)
def
Pk(x,t)*Pk(x,t)
=
1(k
3.13. Girding's Inequality
48 1
has the same regularity properties and, moreover, M(x,t) V (x,t) E BxX[-T,T] with some constant m
0
=
M(x,t)* 2 moId,
> 0.
Furthermore, one finds: iM(x,t)A(x,t)D + (iM(x,t)A(x,t)Dx)* =
=
i(M(x,t)A(x,t)-A(x,t)*M(x,t)*)D -(A(x,t)*M(x,t)* ) X'
where ( A * M * ) = ~ a/ax(A*M*) and, as usual, the upper
* stands for the
adjoint of the corresponding operator or matrix. Since Ak(x,t), 1 S k 6 p, are real-valued, one has: M(x,t)A(x,t) -A(x,t)*M(x,t)* = =
C
(Pk (x,t)*Pk (x,t)A(x,t)-A(x,t)*Pk(x,t)*Pk (x,t)) =
l6k5p =
c A k (x,t)(Pk (x,t)*Pk (x,t)-Pk(x,t)*Pk (x,t))
=
0.
l5k5p Hence,
(3.13.146) iM(x,t)A(x,t)Dx+ (iM(x,t)A(x,t)Dx)* = B(x,t), where B(x,t)
dEf -(A(~,t)*M(x,t)*)~: iRx[-T,T]
+
HOm(IRP ;iRP)*
Using (3.13.146), one finds for the solution u(x,t) of (3.13.145):
Therefore, Gronwall's lemma and the strict positivity of M(x,t) yield: 5 exp(Clt/)I / u o / 1 2 + (3.13.147) 1 lu(.,t) 1 1 2 L2 (IR) L2 (IR)
+ C'
It1
0
2 lIf(.,T)l\
eXp(C((t!-T))dT
L (R)
3. Singular Perturbations on Smooth Manifolds without Bounda y
482
where C , C ' and C" are some positive constants. Using the same argument one can easily show that, in fact, for each s
E
IR3
the following a priori estimate holds for the solution u of
(3.13.145) :
V t E [-T,T],
with some constant Ct which may depend on t. Introducing the operators a(x,t,~,D) , a(x,t,E,DX) = A(x,t)Dx<EDx>-l E Op S(orlr-l)
( R X)
r
and considering the initial value problem:
(3.13.149)
i(D +a(x,t,€,Dx))uE(x,t) = f(x,t) t u (X,O) = u (x),
0
one shows in the same way, as previously, that the same a priori estimate, as ( 3 . 1 3 . 1 4 8 ) , with C
holds for u (x,t) uniformly with respect to
which does not depend on u,uo,f and
t Since for each given
E
E
E (O,E
0
1,
i.e.
6.
0
> 0 one has: a(x,t,E,Dx) E Op S ( R x ) , one
shows using the Picard method and the contruction mappinq arqument that the solution u (x,t) to ( 3 . 1 3 . 1 4 9 ) unique. Now, using ( 3 . 1 3 . 1 4 8 )
exists for t E [-T,T] and is
with s = (O,r,O) with appropriate r E R , and
E 0 that for each 2 uo E H ( R x ) , each f E L ([-T,T];H (R ) ) there exists a unique solution
the compactness argument, one shows by letting
u E
o
O'
-f
x
([-T,T];H (IR ))of initial value problem ( 3 . 1 3 . 1 4 5 ) . a x We are going to consider the following finite difference approximation
C
of ( 3 . 1 3 . 1 4 5 )
which can be successfully used for its numerical treatment.
Let, as previously, let I
= imT1
sT,T
5
= hZ C IR be a grid with meshsize h >
0 and
be the grid on [-T,T] with meshsize T .
Consider the two parameter family of solutions u (x,t), (x,t)E Rh XIT h,T of the following discrete initial value problem
and where the difference operators B hrT following symbols:
qnrT have,respectively, the
3.13. Girding's Inequality
483
b(x,t,h,~,h<= ) Id+ir(2h)-lA(x,t) s i n h<, (3.13.151) a ( x , t , h , . r , h < ) = b ( x , t , h , ,~h c ) -iTh
-1
2 A ( x , t ) s i n h < (1-4y s i n ( h < / 2 ))
y E IR b e i n g a p a r a m e t e r , whose a d m i s s i b l e v a l u e s a r e t o b e d e t e r m i n e d t o g u a r a n t e e t h e s t a b i l i t y of
.
(3.13.150)
i s s a i d t o be a s t a b l e
The i n i t i a l v a l u e problem (3.13.150) approximation of (3.13.145) i n
H(o,,h(%)
where t h e c o n s t a n t C d o e s n o t depend on
E
for t
I
,t
> 0,
i f one h a s :
h , u ~ , uo ~ ,and f .
T,
N o t i c e t h a t f o r e a c h T = r h w i t h r > 0 a g i v e n c o n s t a n t , t h e f a m i l y of def d i f f e r e n c e o p e r a t o r s Bh = B H ( 0 ), h ( % ) has an h , r h ' Bh ( 0 ) ,h(iRh) inverse B-l(iR ) whose norm i s u n i f o r m l y bounded, h . H ( 0 ), h ( % ) -t H ( 0 ) , h h w i t h r e s p e c t t o h E ( 0 , h 1, p r o v i d e d t h a t ho i s s u f f i c i e n t l y s m a l l .
'
'
0
I n d e e d , f o r e a c h pxp m a t r i x $ ( x ) f1 f o r t h e commutator [$,Oh I d ] = $ ( x ) O f l
E
C1
-Oil$
(z) , $
.
=
I / @ j k ( x1) 1 ,
one h a s
(x) t h e following estimate:
where $ ' ( x ) i s t h e m a t r i x , whose e n t r i e s are ( d / d x ) $ . ( x ) a n d , a s u s u a l , lk s t a n d s f o r t h e norm i n Hom(CP;CP).
1 1
Hence, one f i n d s , u s i n g ( 3 . 1 3 . 1 5 2 ) :
(1) h '
= Id+TR
and t h e s a m e argument y i e l d s : i Im B
h
=
( f ) ( B -B*) h h
= =
r ( A ( x , t ) 8 +O A ( x , t ) - A ( x , t ) o - ' 4 h h h (2) TR h '
where t h e d i f f e r e n c e o p e r a t o r s R i J )
:
H(o) ,h(%
norm u n i f o r m l y bounded w i t h r e s p e c t t o h Thus, f o r h o ,
To
E
(O,h
0
H(?,
)
-f
1
by C / A ' ( x , t )
S u f f i c i e n t l y s m a l l Bh = B h , r h
]) i n v e r s e Br e s p e c t t o ( h , T ) E ( O , h o l ~ ( O , ~ obounded h'
,h(lF$,
OhL 1 A ( x , t ) ) =
have t h e i r
has a (uniformly with
'
H ( 0 ),h("h)
H ( 0 ), h ( % -1 . N o t i c e , t h a t t h e p r i n c i p a l symbol b i l ( x , t , r , h < ) of Bh i s g i v e n by t h e formula:
,
3 . Singular Perturbations on Smooth Manifolds without Boundary
484
-1 (3.13.154) bo (x,t,r,h<) = (Id+i(r/2)sin h< A(x,t))-l the right hand side on (3.13.154) being well defined, V (x,t,h<),since the eigenvalues of A(x,t) are real. Let (3.13.155) Qh,T
-1 Bh,T%,T.
=
being the matrix The principal symbol q (x,h,~,h<) 0
(3.13.156) q (x,t,h,~,h<) = 0
2
Id-ir sin hS(1-4~sin (hS/n))A(x,t)(Id+i(r/2) sin h5 A(x,t))-',
=
one finds that the eigenvalues u,(x,t,r,y,n) of q (x,t,~,h,q)are 0
3
(3.13.157) u,(x,t,r,y,n) = 3
=
1-ir sin n(1-4y sin2 (n/2))X.(x,t)(I+i(r/?)(sinq)Ai(x,t))-' 3
with A.(x,t) the corresponding eigenvalue of A(x,t), so that an easy 3
computation yields: (3.13.158) 1 p . I
2
7
=
=
2 2 1-4yr2sin2q sin (q/2) (1-4y sin (q/2))X Thus,
1 3~ I . 5
j
2 2 -1 (~,t)~(l+(r'/4)(sin n)A.(x,t)
1
1, V (x,t,r,q), if
(3.13.159) 0 < y 6 1/4. Using (3.13.144). (3.13.150), (3.13.155) and introducing (x,t), one gets for v (x,t) the difference (x,t) = C(x,t)-lu hrT h,T h,T equation: v
-1
(3.13.160) V
(X,t+T) = C(X,t) h.7
c(X,t)V
'h,T
Obviously, (3.13.152) will hold for
h,T
(xrt)+T(<:Tf)
(X,t).
if it is valid for v h.T Qh,TC(x,t) is diag(p.(x,t,r,y,q)
-FcT
Since the principal symbol of C(x,t) and 1 p . l 2 1 if y 3
E
(O,i),
its discrete version with conclusion that
3
using Corollary 3.13.11 (or, rather
H ( o ) ,h(\
)
instead of L2(iR)) one gets the
.
3.13. Gdrding's Inequality
485
satisfies the inequality
T
=
rh, V t E [-T,Tl,
where the constant C does not depend on h. Therefore, using (3.13.160), one finds:
The last inequality now yields:
where the constant C2 may depend on T and r = T/h, but does not depend on h, v ~ , uo ~ ,and f. One shows in exactly the same way that (3.13.161) holds with t replaced by -t.
The difference scheme as in (3.13.150), (3.13.151) turns out to be useful for the numerical solution of the initial value problem for the systems of conservation laws of the form (3.13.163) with A
=
a au + -f(u) ax at
-
= 0,
u(x,O) = u,(x)
f'(u) the Jacobian matrix for the vector-function f(U)
However, in the regions where the (generalized) solution u(x,t) to (3.13.163) has discontinuities, scheme (3.13.150), (3.13.151) being not of monotone type, its use give numerical solutions, which might oscillate near the discontinuity lines (shocks and others) of u(x,t), instead of giving just a numerical jump, which should reflect the singularity of u(x,t).
I
3 . Singular Perturbations on Smooth Manifolds without Boundary
486
3.14. Reduction of Elliptic Singular Perturbations to Regular Perturbations Given an elliptic singular perturbation a(x,E,D) in Rn or on a compact smooth manifold M, an explicit construction is provided here for a sit:qularly perturbed pseudodifferential operator s(x,E,D) which reduces a(x,E,D) to a regular perturbation, i.e. the product (soa)(x,E,D) is a family of elliptic pseudodifferential operators of the form:(soa) (x,E,D)= 0
0
a (x,D)+ EYQ(x,€,D), where a (x,D) is the reduced operator for a(x,E,D), y is
some positive constant and the order of Q(x,E,D) is at most the same 0
as the one of a (x,D). The same kind of a left reducing operator can be constructed for any elliptic difference singular perturbation on the greed Rn x,h’ A left reducing operator turns out to be a right one for both elliptic pseudodifferential and difference singular perturbations. Some applications to singularly perturbed boundary value problems are also discussed. First, the reduction procedure will be explained using a specific singular perturbation comingfrom thelinear theory of thin elastic plates. Consider the singular perturbation (3.14.1)
=
a(s,a/ax) :=
E
7.2 A -
A€,
__
c a a a +1, l
I
where the matrix 1 lakj I I&, jSn is supposed to be symmetric and positive definite, A being the Laplace operator. Obviously, one has uniformly with respect to (3.14.2)
AE : H
(0,2,0)
E
E
(0,ll
:
(d-5.
,E
With ao(E,S) the principal symbol of AE,
define the symbol ~ ( € 6 ) : (3.14.4)
~ ( € 6 =) ao(O,S) (a0(€,C))-’,
( E , S ) E K + x (R”\IO})
= Op s , the latter being a -1 family of convolution operators with the kernels S(E,X) = F
and the corresponding singular perturbation SE
S-tXS1
(3.14.5)
( s E u ) ( x ) = JnS(~,x-y)u(y)dy, V u E CE(Rn). IR
487
3.14. Reduction of Elliptic Singular Perturbations
-2 where ho (E,x) is defined by (3.2.23). Since AE given by (3.14.1) plays an important role in the linear elasticity theory, we shall give here a formula for the kernel S(s,x) of the reducing operator S E = Op s ( ~ E . 1for any symmetric positive definite
I
in the case of the dimension n=3.
-1
where S(x) = FS+x s ( S ) . Using the spherical coordinates 5 = pw, with p = 151, w E fin, Rn being the unit sphere in I R F , one finds easily when n=3: S(x) =
I
I(w;x)dw,
fi3 where we have denoted:
2 with 0 (w) =
1 akjwkwj, the limit on the right hand side of 15k,j53 3 the last formula being interpreted in the distributional sense in S ' ( R ) .
Obviously, one has: m
-3 2 2 2 -1 I(w;x) = (2n) {iQ ( w ) (<x,w>+i0)-'-O4(w)l(p +Q ( w ) ) exp(ip<x,w>)dp}. 0
Furthermore, since I(w;-x) = I(w;x), one can rewrite the last formula in the form: I(w;x) =
(+)
(2n)-3~2(w)~i( (<x,w>+io)-l-(<x,w>-io)-l
-
m
-Q2(w) /(p2+02(w)
exp(ip<x,w>)dp},
-m
for the distributions -1 (xfiO) and the residuum calculus for computing the integral on the right
so that the Plemelj-Sokhotski formulae
hand side of the last formula, one finds: 2 3 2 -1 ) I20 (w)A(<x,w>)-Q (w)exp(-Q(w)I<x,w>\)}, I(w;x) = (161~
3. Singular Perturbations on Smooth Manifolds without Boundary
488
where 6(y), y E R is the Dirac's 6-function. Further, noticing that 2 -(d/dy)2 exp(-@(o)/yl) = 4 (w) exp(-@(w)/yl) + 2@(w)6(y), and using the fact, that
=
Zwk2
1 , one gets finally the following
=
formula for I (w;x): I(w;x) where A
=
=
-(l6n2)-'A
O(w)
exp(-@(w) I<x.w>/),
Z(a/ax.)L is the Laplace operator.
I
Thus, one finds for S(x) when n=3: S(x) = -(16n2 -1 A For a = Id, one has: @(a)
exp(-
n3 1 and the last formula yields:
5
the latter coinciding with A02(1,x) defined by (3.2.23) when n=3. The same kind of computation carries over for any odd dimension n, since it allows to evaluate I(w;x) by replacing the integral over R
= {p>O) in its expression by the one over R = {-m
and by using the
residuum calculus.
. some constant co > 0 (which Since c - ~ < E ~ >5 -~~( € 5 4) c <ES> -2 with 0
0
depends only on the smallest and greatest eigenvalues of the matrix
I 1 akj 1 1 ) , one gets (3.14.6)
SE :
H
the conclusion that (Rn)
(0,0,-2), E
uniformly with respect to
E
E
( 0 . ~ ~ 1V,
e0 <
m.
Moreover, actually, the family of linear mappings (3.14.6) is an isomorphism uniformly with respect to Further, the product SE (3.14.7) where Ao
SE
=
0
A'
= A
0
+ E
2 E (I
0
AE
E
E (0.~~1.
can be represented in the form:
,
a(O,a/ax) is the reduced operator for
op s(or2r-2)( R n ) ,
AE
and
3.14. Reduction of Elliptic Singular Perturbations (3.14.8)
QE
:
H
489
( Rn)
(0,2,0),E
is a family of continuous (uniformly with respect to
E
E [O,E~])
convolution operators with the (distributional) kernels q(E,x),
1
q(E,x) = F-1
E*X
~
1
~
(aO(E,E))-l.
(Rn)with H2(Rn) and H(o,O,O),e(Rn)with (0,2,0), E one comes to the conclusion that the multiplication of A from
Identifying H L2(Rn
),
the left by
SE
reduces elliptic singular perturbation (3.14.1), (3.14.2)
to a regular perturbation of A (3.14.9)
Ao
:
H2(Rn)
*
0
,
L2(Rn).
It will be shown in this section that for any elliptic singular perturbation A E in Rn or on a smooth manifold M without border, an explicit algebraic construction can be given for a left reducing operator SE, which reduces A'
to a regular perturbation of the operator A
0
.
Now the construction of a reducing operator for any elliptic singular perturbation will be presented systematically. From now on all the symbols and the corresponding singular perturbations are supposed to possess the reduced symbols and the corresponding reduced operators. Let a E Op
L v ( Rn ) (see Definition 3.12.1) be elliptic of order
v = (vl,V2,V3)and assume that it has a reduced operator
ao E op L ( 0 1 v 2 ' o ) (Rn)(see Definition 3.12.3). Our goal now is to produce a singular perturbation r E O p (to be called factorizing) and a singular perturbation s E O p (to be called reducing) such that (3.14.10) a(x,E,D) = r(x,E,D) 0
(3.14.11) a (x,D) = s(x,E,D)
o
0 a (x,D)+EQ~(x,E,D),
o
a(x,E,D) +EQ~(x,E,D),
where for each s = (sl,s2,s3) E R3 the linear mappings:
are bounded uniformly with respect to
E
E (0.~~1.
L
(vl,0,v31 ( Rn)
L (-v1,O,-v3) ( Rn)
3 . Singular Perturbations on Smooth Manifolds without Boundary
490
Lv ( Rn )
It is readily seen that for each elliptic symbol a E
of order
v there exist constants C > 0, R > 0 such that
V (x,E)
E
Bnx
( O , E ~ ] ,V
5 E
Rn,
Given an elliptic symbol a E LV(Rn)satisfying (3.14.13),
x
E Cm(Rn) be a cut-off function such that x(S) 0
x(S)
5
5
(51
2 R.
let
0 for 151 2 2R,
I for 151 5 R, and denote @ ( S ) = 1-x(5). As a consequence of (3.14.13), the following functions are well5
defined :
{s(x,E,~)
0
r(x,c,S) = Q(E)a(x,c,c) (a (x,<))-'+
(3.14.14)
=
-V E
1 x(~),
0 Vl Q(c)a ( x , ~(a(x,E,C))-l+ ) E x(~),
where a0 (x,S) is the reduced symbol of a(x,e,S). Indeed, since (3.14.13)
is valid uniformly with respect to
E,
it
0
holds for the reduced symbol a (x,S), as well.
The symbol a E LV(mn)being e l l i p t i c of order v
Lemma 3.14.1.
=
(v1,v2,v3)
( 0 , V2,O)
and ao E L ( R ~ ) being i t s reduced symbol, t h e f u n c t i o n s r(x,E,S) and s(x,e,S) defined by (3.14.14)are e l l i p t i c symbols of order (v1,0,v3) and (-vl,0,-v3) belonging t o t h e c l a s s e s ~ ( ~ 1 ' ~ ' (Rn) ~ 3 ) and
L
(-V1 ,O,-V3)
so
(w" ), r e s p e c t i v e l y ; furthermore, t h e i r reduced symbols ro and
are i d e n t i c a l l y
1.
_ Proof. _ The latter statement is immediate, the ellipticity of r(x,E,S) and s(x,e,S) (of order (vl,0,v3) and (-v 1 ,O,-v3), respectively) being a direct consequence of the fact that their principal symbols are 0 -1 0 rO(x,EpS) = aO(x,E.5) (ao(x,5)) and sO(x,~,S)= ao(x,5) (a0 ( X , E , S ) ) - ' , respectively. Furthermore a straightforward computation (using the chain rule and the ellipticity of a of order v) shows that r(x,E,E) and s(x,E,<) satisfy
(3.3.1) with (vl,O,v ) and (-vl,0,-v3), respectively. 3
I
Lemma 3.14.2.
The symbol a E L'(R") ao E
being e l l i p t i c of order v = (vl,v2,v3) and
L(o'v2ro) (xn) being its reduced symbol, one has w i t h any
N > 0:
3.14. Reduction of Elliptic Singular Perturbations
49 1
Proof. It is readily seen that with r(x,E,E) and s(x,E,S) defined by (3.14.14), one has:
Now (3.14.15) is an immediate consequence of Definition 3.12.3 and Proposition 3.3.3, since
x
E ~ ( ~ l - ~ (cE-n )~,)v N
> 0.
Remark 3.14.3. Using the following definition of r and s : -V
0
1 X(S),
r(x,E,S) = a(E)aO(x,E,c)(a0 ( x , ~ ) ) - ' + c 0
s(x,s,~)= a(E)ao(x,t) (a0 (~,E,s))-'+ s where . a
0
and . a
v1
x(<),
are the principal symbols of a and 'a
respectively, it is
readily seen that (3.14.16) (a-ra ) E L
"
( v 1-1,v 2 ' 3 )
(-l,V2,O)
( R ~ ) ,(sa-a ) E L
(Rn).
Sometimes, it is more convenient to use principal symbols for constructing r and s . In that case one can choose a fixed R (for instance, R = 1). while choosing the cut-off function
x(S).
I
The following lemma plays an important role in the reduction procedure and is proved by using a more or less standard argument based on Schur's lemma. 3.14.4. Lemma _ _ _ ~ L e t j(x,~,S)E L'(R")
with v
symbol j0 ( x , ~ ): I . v
(x,<)
= (vl,0,v2) and a s s m e t h a t t h e reduced E mRnx R". L e t a(x,c,<) E L"(R~).
Then one has: def (3.14.17)
QE
=
(Op(j)
o
Op(a) - Op(ja)) E Op L V+u-(l,O,O) (Rn),
3
i . e . f o r each s E R t h e folZowing i n e q u a l i t y holds: (3.14.18)
1 1 QEuI I ( s-v,-F1) ,
wi t h a constant c
> 0,
I (s)
v
E
E
(o,Eol,
which does n o t depend on
E
and u.
5 C E / IuI
,E'
v
u E H
( s ) ,E
(Rn)
3. Singular Perturbations on Smooth Manifolds without Boundary
492 Proof. -
As a consequence of Theorem 3.7.5 (with U = lRn, so that the
condition on Op(j), %(a)
of being properly supported is automatically v+u-(O,l,O) n ( R ) . The specific property
satisfied), one has: QE E Op L
(3.14,17)of Qe is a consequence of the fact that Op(j) = ~(x,E,D) has identity as its reduced operator. As
a consequence of Definition 3.12.1 of the symbol class
L v ( Rn ) ,
one has the following decomposition: j(x,E,S) = j,(E,<)+j'(x,E,S), where the functions x
+
j'(x,e,S), x
+
a(x,e,<) = a,(e,<)+a'(x,c,c), a'(x,E,S) are in S(mn), X
v
(E,S)
E
(O,Eo1
x R;.
Given that Op(j)
o
Op(a,)
- Op(jam)
= 0,
one gets the conclusion that
for proving (3.14.17) it suffices to show that (3.14.19) Op(j')
o
Op(a') - Op(j'a') = E Q ~ ,
(3.14.20) Op(jm)
o
Op(a')
where QE E LV+'(lRn),
-
Op(j,a')
=
EQ~,
i.e. (uniformly with respect to
E
E (O,Eo]) for each
s E R 3 'holds:
(3.14.21) Q? J
:
H ( S ),E
First, (3.14.19) (with QE satisfying (3.14.21)) will be proved. 1 One has the following Fourier representation for QEu:
~ ' ( s - T , E , =~ IF~+~-,~'(X,E,~), ;1(7-n,~,n)
=
~~+~-~a'(x,~,n).
By Schur's lemma on boundedness of integral operators in L2 (see Lemma 3.12.8) for proving (3.14.19), (3.13.21), it suffices to show that s
(3.14.23) and
E
-v
-s
-Ll
3<,1>
-S
2<E11> 31K(S,E,n)/dS5 CE
3.14. Reduction of Elliptic Singular Perturbations v +lJ
(3.14.24)
E
s
with some constant As
-v
I <S> nn C >
s -v -lJ
-s
3
2<ES>
-S
2<~n1>
0, which does not depend of
3
493
I K ( S , E , ~ )/dn 5 CE,
(S,E,~). 0
a consequence of Definition 3.12.3 and the assumption: j (x,S)
1,
one has:
Furthermore, one finds : -V
jDx(I'(x,E,~)-j'(x,~,rl))/ a . 5 IDa(b'(x,€,~)-b'(x.€,~))€ '<ET>'
3 It
Using the representation I
bl(X,E,~)--b'(~,E,n)=
I
(a/ae)bi(X,~,q+e(T-n))de,
0
the inequalities (3.12.1) with v
=
(-l,l,-1) and inequality (3.4.7), one
gets:
V ( x , E , T , ~ ) ,V a E N n ,V k L 0,
with m = 3 + / v 1 , and with some constants Ca,k > 0, which may depend only 3 on their subscripts. Hence, one has for j inequalities: E,O) (3.14.25) I~'(y,s,~)-~'(y
I
1-v
2 C < T - ~ > ~ E'<ET>'~<~>-~, t/ y,E , T
k
,n ,k.
Similarly, one has:
Using (3.14.25) and the estimate
' :1
(y,E,n)1 5 CkE
-lJ
lJ lJ '
V y , ~ , ,k, n
(which follows from the fact that a'(.,~,n) E S(lR:)
and from (3.12.1)),
one can estimate the integral on the left hand side of (3.14.23) in the following fashion:
3 . Singular Perturbations on Smooth Manifolds without Boundary
494
where
One chooses k > 0 to be sufficiently large in order to guarantee convergence of the integral on the right hand side of (3.14.27), and that ends the proof of (3.14.23). The same argument can be used to show (3.14.24). Hence, we have proved (3.14.191, (3.14.211. Using (3.14.26) instead of (3.14.25), one gets (3.14.23), (3.14.24) with K replaced by the kernel: K,(~,E,I?) = (j,(E,5)-j,(€,~))at(5-n,E,q).
This yields (3.14.20) with Q, satisfying (3.14.21).
1
Remark 3.14.5. A slight modification of the proof shows that Lemma 3.14.4 is still true if
u2 is not necessarily zero
(j(x,E,<)
=
1+Ea(x)1512 is an example of a 0
symbol, whose order is (O,l,l) and whose reduced symbol j (x,S)
5
1).
How-
ever, in applications the most incountered situation is the one, where the order of j is (vl,0,v3). Assume that j E L ( o r o ’ ” ) ( E n ) and j = j (x,E<). Then the following argument gives a simple explanation to the assertion in Lemma 3.14.4. One has in that case (see Theorem 3.7.6):
where
QE
is the singular perturbation whose symbol has the asymptotic
expansion
3.14. Reduction of Elliptic Singular Perturbations
the symbol of QE being in Lv+' (En )
;
495
here the following notation has been
used:
Notice, that nowhere in the proof of Lemma 3.14.4
the fact that the
symbols j(x,E,S) and a(x.E,S) are smooth in 5 has been used. In fact, the assertion of Lemma 3.14.4
is still true for symbols a which satisfy a kind and still admit the decomposition
of Lipschitz condition of the form (3.1.9) (ii) in Definition 3.12.1
with a' satisfying (3.12.1)
for V@ and cx = 0.
In fact, the same informal argument (based on the use of Theorem 3.7.6)
LV(En ) ,
shows that for each j E @(a)
one has:
OOp(j) =a(x,E,D) 0 j(x,E,D) =
-
1
C pp(a(U)(x,Er5) ) ' ( j
(x,E,~))=
Op(aj) + EQ',
where Q"
E - ~
N
0
Since j ( x , S )
/'I
Z
1, one has, according to Definition 3.12.3
for each
> 0: IE-lj(al
(~,E,s) =
v
-vl
>c
E
<5>
*
1'
-1 a
D~(I(X,E,S)-I) I 5 ,
v3+1
-1 <ES>
i.e. Q" is a continuous linear mapping (uniformly with respect to
n ( S ),E (mn ) into H (s-v-')
A
,E
E)
from
( W n ) for each s E E3 :
rigourous proof of the formula Op(a)
o
Op(j) -Op(aj)
= cQE
with QE
satisfying the above continuity condition, consists of repeating alnost word after word the proof of Lemma 3.14.4. Hence r(x,E,D) and s(x,E,D) defined by their symbols (3.14.14)
are
not only left but also right factorizing and reducing operators, respectively, for a given elliptic singular perturbation a(x,E,D) having 0
the reduced operator a (x,D) (see [Fr-W, 1,3] statement).
for a detailed proof of this
3. Singular Perturbations on Smooth Manifolds without Bounda y
496
Theorem 3.14.6.
L e t a € LV(nn) b e e l l i p t i c of order v = (vl,v2,v3) a admits t h e reduced symbol ao E 1
'
(-v1 ,o,-v.
)
and s E 1
Then f o r each
(0.V2
to)
3 (R )
s
be d e f i n e d by E x3 t h e diagram
€
n3 and assume t h a t
(1~"). L e t r E 1
(Vl'O,V3)
3 (E )
(3.14.14).
(3.14.28)
i s commutative modulo o p e r a t o r s , whose norm i s bounded by
CE
w i t h some
p o s i t i v e constant C.
In o t h e r words, t h e f o l l o w i n g e s t i m a t e s hold (3.14.29)
I /Op(a)-Op(r)
0
0
Op(a )
11
1 IOp(r)
0
Op(s) - Id1 I H(s-(0,v2,0))
(3.14.31)
I /Op(s)
o
lH
Op(r) - Id/
(S-V) , E (3.14.32)
I IOp(s)
0
0 Op(a) - Op(a )
+H
1 IH (s) ,E
(vl ,0,v3)
Proof. Since r E
L
~
C >
0 does not depend on
Finally (3.14.32)
+H
5 CE,
(s-(O,V2,O))
,E
5 CE, (S-V)
+H
,E
5 CE.
(s-(O,V2,O))
(-v1,0,-v (3.14.31)
Furthermore, the Lemmas 3.14.2,
where
,E
,E
3.14.4
,E
)
(lRn), s E 1
one gets immediately (3.14.30),
E + 0:
5 CE,
H ( s ),E+H(s-v) (3.14.30)
as
( X n ) and ro E 1, s o E 1, by applying Lemma 3.14.4. imply:
E.
is an immediate consequence of ( 3 . 1 4 . 2 9 ) - ( 3 . 1 4 . 3 1 ) .
I Remark 3.14.7. Obviously, Lemma 3.14.4
and Theorem 3.14.6 are still valid for matrixv n valued symbols in Hom(Cm;Cm), a E L (R ) being elliptic of order V, i.e.
3.14. Reduction of Elliptic Singular Perturbations satisfying (3.14.13) with la(x,E,S) being defined as in (3.14.14) with
I
497
the norm in Cm and r(x,E,S), s(x,E,S) -V
8
'~(5) replaced by E-'~x(S)I~, where
Id E Hom(Cm;Cm) is the identity. A s a first application of the reduction procedure, we give here a
proof of the statement in Remark 3.13.17, based on the use of Lemma 3.14.4. Theorem 3.14.8.
L e t p E Lo(Rn ) , p being valued in Hom(Cm;Cm). L e t t h e principaZ symboZ pO(x,~S)of P(X.E,~) be
a hermitian non-negative m a t r k : p
0
2
v (x,n) E d'x
(X
'
n)
= p (X, n ) * 2 0 ,
and assume t h a t po E c ( m n x m n ) . Furthermore, assume t h a t w i t h some constant c > 0 t h e fo-dloming i n e q u a l i t i e s hold: (3.14.33) <x>~+'<ES>ID 6 (p(x,~,S)-p~(x,~S)) I 5 CE,
v Then one has for
PE =
161 5 n+l,
(X,E,S) E mnx
(O,Eo1
x
2.
)
stands
P(X,E,D~)= op(p):
where t h e c o n s t a n t c does not depend on
E
and u and vhere
(
,
for t h e i n n e r product i n L2 (nn). Proof. We give here the main steps in the proof of (3.14.34) without going ~
into detail. First, we notice, that on the ground of the same argument, as in the proof of Theorem 3.13.6, one can assume, without restriction of generality, that p(x,~.S) 5 p,(x,~S) and, thus, one has to show (3.14.34) only for P:
=
P~(X,ED)= op(po~Furthermore, since po(x,n) E C2(Rn x R n
)
, the
reduced symbol
0
0
P (x,S) Z Po(x,O), i.e. the reduced symbol P (x) is j u s t the multiplication 0 by the hermitian non-negative matrix P (x,O)OE 1o ( Rn ) . 0
Further, introduce the symbols:
Obviously, one has: r(x,E,S) * = r(x,E,S), s(x,E,S)* and
=
s(x,E,<), rs
=
sr = Id, rtId, s>Id,
3. Singular Perturbations on Smooth Manifolds without Bounday
498 0
0
r (x,S) = s (x,S) = Id, where, as usual, ro and so are the reduced symbols of r and s , respectively. Applying Lemma 3.14.4
(with j
r or j = s ) , or using an argument very
=
similar to the one in its proof, one finds with RE = r(x,E,D) = Op(r), SE = s(x,E,D) = Op(s):
REoSE = Id+EQ:, € = P~+EQ:, 0 (3.14.36) sE"pO
( sE ) * -sE =
SEoRE = Id+EQg,
: P
En:,
= R'~P~+EQ~, O E
(R€)*-R€ = E Q ~ , 0
where Po is the multiplication operator by Po(x), and Id, QS, 16j56, are 0
identity and (uniformly with respect to mappings in the corresponding spaces H
E
E
1) continuous linear
(O,E
0
( W n)
(s) ,E
.
Now, using (3.14.361, one finds:
2
.
Furthermore, let r(x,E,S) = (b(x,E,S)) wlth b = b* 2 Id the positive definite square root from r = r* t Id. Denote BE = b(x,E,D) = Op(b); one has also for BE (as a consequence of Lemma 3.14.4):
where again QE
J'
j = 7,8, are (uniformly with respect to
linear mappings in corresponding spaces H
(s)
,E
(Wn )
E)
continuous
.
Thus, using (3.14.37), (3.14.381, one finds: Re(PEu,u) = Re(BEoBE(P:U)
,U)
1
I
+ O ( E ) j u / 2( o ) , E =
0
0
=
Re(BE(PEu),(BE)*U) +
=
Re(Po(x) (BEu),BEu) +
0
o
since P (x) = (P0(x)) 0
*
O ( E ) \ / U I /2 ( 0 ) , E= O(E)
[
jul
[ 2( o ) , €
b -cEI / u l
I 2( 0 ) ,El
t 0.
In this argument, besides Lemma 3.14.4, also the following statement has been systematically used:
3.14. Reduction of Elliptic Singular Perturbations Let J~ = op(j), j E 0
(JE)*-JE = EQ',
j (x,S) E Id. Then QE
: H
(XIn)
( s ) ,E
respect to
E
E
+
L (O,O,v)(=n), (j(x,E,S))* = j(x,E,S)
H(s
(O,E
0
))
,s
2,S3-V),E
499
2 Id,
where J E is the L -adjoint of JE, and 2 (mn)is a family of (uniformly with
continuous linear mappings.
The proof of this statement is very similar to the one of Lemma 3.14.4
n
and will not be given here. Remark 3.14.9.
Obviously the statements in Lemma 3.14.4 and Theorem 3.14.6 carry over to m
the case, when Rn is replaced by a compact C
manifold M, dim M = n.
Indeed, let {(Ui,ai), lSi5NI be an atlas on M. i.e. {Ui'lZiQJ aimi)
C
and
mn . l5iSNI be a partition of unity subordinate to the covering
Let
{UiIlsisN. One can choose Ui, lSi
with def Qi,k,l
=
-
Op(Oij) oOp($ka)O O P ( Q ~ ) - ~ I ? ( $ ~ $ ~ $ ~ ~ ~ )
If
maps { u E H ( s ) ,E (M), supp u C U1 into (M), supp f C Uil. In the local coordinate system H(sl,S2-"2,S3),E
aikl
:
Each
QE
lrktl
vikl
+
m n , one can view Q?
lrkr1
as a singular perturbation in m n .
Similarly to (3.14.22), one has the following Fourier representation for Q? i,k,l'
where
It is readily seen that (3.14.23), (3.14.24) hold for Ki,k, since the argument in the proof of Lemma 3.14.4 for the kernel K given by (3.14.22) is still applicable when a' is replaced by $ a and j' is replaced by $.j. k
3. Singular Perturbations on Smooth Manifolds without Bounda y
500
Therefore, we conclude that (3.14.18) holds for each Q?
l,k,l
fore, also for
QE =
Op(j).Op(a)-Op(ja)
and, there-
considered as a singular perturbation
on M . Furthermore, defining locally r(x,~,E),s(x,E,E) by the same formulae (3.14.14) and using the local coordinates and a subordinate partition of
unity, one defines globally the corresponding factorizing r(x,s,D) and reducing s(x,E,D) singular perturbations such that (3.14.29)-(3.14.32) in corresponding spaces H uses only Lemma 3.14.4,
( s ) ,E
(M),
hold
given that the proof of Theorem 3.14.6
the latter being true with
mn
replaced by M .
Corollary 3.14.10.
If A€ E op
i s e l l i p t i c of order w
s'(M)
= (v1,v2,v3) and
t h e reduced operator
i s an (algebraic and topological) isomorphism uniformly w i t h respect t o E
E ( 0 . ~ ~ 1then , f o r zO s u f f i c i e n t l y small t h e family of l i n e a r mappings
AE
:
H
( s ),E
*
(')
H(s-V) ,E ( M ) ,
E
E (o,Eol,
i s an (algebraic and topological) isomorphism u n i f o m l y w i t h respect t o E
E ( 0 . ~ ~ 1as , well.
Indeed, the multiplication from the left by a reducing operator reduces A E to a regular perturbation of Ao
:
H
( s ),E(M)
SE
+ H(s1,s2-V2,S3) ,E
(M)
Actually, it suffices to assume that 'A
:
H
(MI
-f
L2(~)
v2
is an isomorphism in order to reach the same conclusion for 3 V s E R (see [Fr.-W, 1,3 1 ) . AE : H (M) + H(s-"), € ( M ) , ( S ) ,E
Example 3.14.11. Let B1 = {x E R
2
, 1x1
< 11, Rl = aB1 = {x
E
R 2 , 1x1 = 11. Consider the
following singularly perturbed boundary value problem:
where N is the inward unit normal at x' E Rl, and a(x'), g(x') are given X' smooth functions on R 1' Using the polar coordinates x
=
(
1x1 ,XI) = (r,B), one can represent
.
3.14. Reduction of Elliptic Singular Perturbations
50 1
any harmonic function u(x) in the disk B1 as the following Fourier series: (3.14.40)
uE(x) = (2Tr)-l
GE(k)r1k1exp(ik8),
Z
kEZ where u (k) are the Fourier coefficients (discrete Fourier transform) of u
(XI.
Substituting ( 3 . 1 4 . 4 0 )
on R l , one
into the boundary condition for u
finds for u (0) the following singularly perturbed equation on f i l :
since in the case considered -a/aN
X‘
a/ar.
=
Using the projection operator Il+ defined by ( 3 . 1 0 . 5 1 ) , one can rewrite ( 3 . 1 4 . 4 1 ) in the following form: (3.14.42)
where II-
(E~,(n+-n-)+a(e))u€(e) = g ( e ) , Id - II+ and, as usual, D
=
e
=
eE
fil,
-ia/ae.
As it has been shown in Example 3 . 1 0 . 1 5 ,
the operator EDe(n+-n-)+Id
is a singular perturbation in Op S ( o r o r l ) (R 1 ) whose principal symbol is 1 +15~1 . Hence, the singular perturbation
is a singular perturbation in Op S ( o f o ’ l ) (fi 1) , as well, whose principal symbol is: a0 (e,Ec) = a ( e ) + E / c I . Hence AE E Op Ia(e)+El5l
v e E
I
S(ofonl)
(ill) is elliptic of order (0,0,1)
iff
2 C < E with ~ some constant C > 0 , i.e. iff Re a ( 0 ) > 0 ,
fi1.
one can use ( 3 . 1 4 . 4 )
Assuming AE to be elliptic of order (0,0,1),
in
order to reduce ( 3 . 1 4 . 4 2 ) to a regular perturbation equation on R 1 , the a reduced symbol here being, evidently, :
=
a (8).
However, it is more convenient in this case to use the construction of a reducing operator indicated in Remark 3.14.3.
Actually, one can take
as a reducing operator here any singular perturbation whose principal symbol is: so(e,Ec)
= a ( e )( a ( e ) + E / c I )
-1
,
SE
E Op
S(oro‘-l)
(a1)
3 . Singular Perturbations on Smooth Manifolds without Boundary
502
and whose reduced symbol
s o F 1.
One of the possible definitions is the following one:
c
(sEv)( 8 ) = ( 2 n I - l
a ( e ) (a(e)+EikI)-' G(kjexp(ik8)
=
kEZ = v(B)-E(2n)
-1
X Ik/(cx(e)+E/kl
G(k)exp(ike)
kEZ where G(k) are the Fourier coefficients of v. Anyhow, since the reduced operator Ao (which is the multiplication by
a ( @ )with Re a ( e ) > 0) is invertible, one gets the conclusion (see Corollary 3 . 1 4 . 1 2 ) v
=
(0,0,1),
that A' given by ( 3 . 1 4 . 4 3 1 ,
AE
:H
(S),E
is an isomorphism uniformly with respect to
(0 ) + H E
1 (S-V),E E (O,EO1,
(a1) I
provided that E~ is sufficiently small. m
If g E Cm(R1), then the solution u ( B ) of ( 3 . 1 4 . 4 2 ) with respect ot
E)
is in C (fl,)
(uniformly
and, moreover, it can be represented by the following
asymptotically convergent series:
so that, for instance,
One can consider for a harmonic function u in the unit disk B
1
another
boundary condition on R1 of the form: (3.14.44)
where II,
B(8)
(-a/ak,+a(e)-EB(e)a =
2
/ae 2 ) u E ( e ) =
g(e),
E
E ( 0 . ~ ~ 1
( E 1 ( 8 ) , I 1 2 ( 8 ) ) is a given smooth vector field on R1,
, g(B)
and a ( B ) ,
are given smooth functions.
Using the same argument, one can easily show that under the conditions:
a ( B ) > 0 , B ( B ) > 0, II ( 8 ) case+ !L2(8) sine c 0 , V 8 E R 1 , there exists a 1 uniquely defined harmonic function u satisfying boundary condition ( 3 . 1 4 . 4 4 ) , V
E
E (O,E
0
1, provided that
E~
is sufficiently small. In the latter case
the corresponding singular perturbation on the boundary turns out to be elliptic of order ( O , l , l ) EB(e)c
2
-(.tl(e)
cos e
+
and to have the following principal symbol:
i2(e)
sine) 151.
Furthermore, the reduced boundary condition
3.14. Reduction of Elliptic Singular Perturbations
503
for a harmonic function u in B1, defines an isomorphism from H (B1) onto s2 (R if a ( 9 ) > 0 (see [Fr.-W.,1,31), which means that reduced equation Hs2-3/2 1 (3.14.45) is uniquely solvable in Hs2-+(R1) for each g E Hs2-3/2 (R1). Hence, under the assumptions above on R e ,
(R
H(sl,s2-3/2,s3) 1 small.
V
),
B ( e ) , a ( B ) , the perturbed equation
(R1) (s1,s2-t,s3+1)
(3.14.44) is solvable in H
for each
E ( 0 . ~ ~ 1provided , that
E
E~
is sufficiently
It is obvious that a convenient choice of the spaces H
( s ) ,E
( B 1 ) for
the harmonic functions u , satisfying the boundary condition in (3.14.39) (respectively,boundary condition (3.14.44)) is the one, where s = (s1,s2,s3) satisfies the conditions s2 > f, s2+s3 > 3/2 (respectively,s2 > 3/2, s2+s3 > 3/2) (see also Theorem 2.2.21, since this choice of s guarantees
existence of traces of u and its first transversal derivatives on R 1 = aB1 uniformly with respect to E). Consider for a harmonic function uE in B1 the following boundary condition on a. = aB 1'. (3.14.46) (-Ea/aNe+i)(exp(iKe)n++n-)uE(9)= g(e), where TI+ and II- = Id-II'
are the same, as above, and
Since the trace operator (-Ea/aN +l)u(r,B)
e
singular perturbation AE E Op (AEv)( 9 )
=
(2rrI-l
S(o'otl)
K
E Z.
I r=l defines an elliptic
(R,),
(I+Elkj);(k)exp(ikB)
C
kEZ (see Example 3.10.15),
equation (3.14.46) can be rewritten in the equivalent
form:
where gE(e) = AElg(9)
def =
C
1
(l+Ejkj)- g(k)exp(ik9).
kEZ The operator on the left hand side of (3.14.47) (well known in the literature as Noether's example of an operator with index), has a nontrivial kernel of dimension dimension
K
iff
K
> 0.
-K
iff
K
< 0 and a non-trivial cokernel of
In both cases its index (difference between the
dimension of its kernel and the one of its cokernel) is
-K.
The reduction of
3. Singular Perturbations on Smooth Manifolds without Boundary
504
(3.14.46) to (3.14.47) shows that the singularly perturbed operator on
the left-hand side of (3.14.46) has the same index, VE. The same is true, when one considers the boundary condition for the harmonic function uE in B1 of the form:
(3.14.48) (-Ea(e)a/aNe+B(e))(exp(iKe)n++K-)u ( 8 ) = 9(e),
provided that a ( e ) ,
B(e)
are smooth functions satisfying the ellipticity
condition:
where C is some positive constant. Indeed, if (3.14.49) is satisfied, then the singular perturbation in Op S ( o r o r l(Q,) ) associated with the trace operator defined by (3.14.48) on harmonic functions uE, is elliptic of order (O,O,l), its principal symbol being
with H ( C ) the Heaviside function. Since the singular perturbation A
(-Ea(e)a/aNe+B(e))uE(r,e) of order (0,0,1)
jr=l
associated with the trace operator
on harmonic functions u in B ~ is , elliptic
and has an invertible reduced operator (which is just the
multiplication by B ( 8 ) # 0, V 0 E R1), the inverse A-’ exists and is an elliptic singular perturbation in Op S ( o c o r - l(al), ) V that
E~
E
E
( 0 . ~ ~ 1provided ,
is sufficiently small (see Corollary 3.14.12).
Thus, the index of the singular perturbation defined by (3.14.48) is the same as the one for the reduced operator, the latter being again that of Noether on the left hand side of (3.14.47). It is readily seen that (3.14.17) can be reformulated in the following equivalent way: find a function $ + ( z ) analytic in B1 = { z E C, I z I < 1 1 and a function $ - ( z ) analytic in CB1 = { z E C, / z / > 1 ) such that exp(i~e)~+(exp(iB))+$-(exp(ie))= g ( € J ) ,
which is the Riemann-Hilbert problem for the circle Q l [:4uskh.,
1
C
C (seefor instance
I,.
Example 3.14.12. We come back to the singular perturbation from Example 3.13.26:
3.14. Reduction of Elliptic Singular Perturbations (3.14.50) a(x,s,D
)
505
-~A~-, x E R n .
=
With +(x) defined by (3.13.124) and b(x,E,DX) defined by (3.13.126), one reduces the equation (3.14.51) a(x,E,D )uE(x) = fE(x) to the following one: (3.14.52) b(x,E,D )vE(x) = gE(X),
X
E Rn,
where
-1 -1 (3.14.53) v (x) = u (x)exp(E +(x)), gE(x) = Ef€(x)exp(~ Q(x)). Assuming that g v
E H E
(s1,s2,s3+2) ,E
E
E H
( s ) ,E
( R n)
,
one gets the conclusion that
(R"), since b(x,E,S) E !-(o'of2)(Rn) is elliptic of
order ( 0 . 0 . 2 ) . 0
Furthermore, the reduced operator b (x,Dx) is just the multiplication =f by q ( x ) de jVX+(x) 2 qo > 0 , V x E R n , q(x) = l$m\2+qo(x) with
l2
q' E S(Rn). Thus, using (3.14.14), one defines a reducing operator
E s(x,E,D~)
L (o'of-2)(Rn) and gets the conclusion that (3.14.52) has a well--
( 1 ~ ~for ) any g E H (En). (s1,s2,s3+2), E E ( S ) ,E One can also use as a reduclng operator the singular perturbation
defined solution v E H E
s (x,ED~) with the symbol: 0
. (3.14.54) so(x,~S)= q(x) ( 1 ~ 521+l
E
+ so
(x,€DX) is a family of integral
3 . Singular Perturbations on Smooth Manifolds without Bounda y
506 where
P(X) and A.
-2
=
2' (IVXtJ(X) I2+(i)/g(x)I 1 2 ,
(E,x) is given by (3.2.23) (it can be also expressed in terms of
Hankel's functions). For n
=
3, one finds using (3.2.24):
2 2 (3.14.56) S0(x,y) = (4dyl)-l(p(x)) /VxO(x)1 exp(-(t)(P(x)lyl+
Since the reduced operator b (x,D
1 Vx$ (x)1 2, the
)
reduction of the singul:r
is just the multiplication by perturbation given by (3.14.50)
to a regular one, allows one to solve (3.14.51) asymptotically, i.e. to represent u (x) as an asymptotically convergent series:
with some constant C > 0. The following matrix singular perturbation a(x,E,D
),
plays a special role in fluid dynamics with hydromagnetic effects; here H(x) : R3
+ It3
is the magnetic field,
E
is the magnetic viscosity which
is supposed to be a small parameter and v(x)
:
R3
+
R3 is the velocity
field of the fluid. Besides, for both v(x) and u(x) holds: div v(x) = X 3 = divxu(x) = 0. The cross x stands for the vectorial product in R
, VxX
being the usual curl operator on functions w [Lan.-Lif. , 2
R3
-+
R3 (see for instance,
I).
Since div H(x) X
:
=
div v(x)
=
0, one finds easily:
X(v(x)XH(x)) =
vX
=
Id the identity in Hom(C3;C3).
In some situations (for instance, in some problems of cosmic fluid dynamics with hydromagnetical effects) the influence of the magnetic field on the hydrodynamical characteristics of the stream are very weak,
3.14. Reduction of Elliptic Singular Perturbations
507
so that one can determine the velocity field v(x) independently of ff(x). Besides, v(x) turns out to be practically constant when 1x1 is sufficiently large. Assuming that v(x) satisfies the conditions imposed on b(x) in Example 3.13.26, one can use the singular perturbation s 0 (x,E,D ) defined by (3.14.54). (3.14.56), in order to reduce the singular perturbation given by (3.14.57) to a regular one, i.e. to solve the corresponding equation a(x,E,D )H(x) = f(x) asymptotically when X
E +
0 (see [Br.-Fr., 1
11,
where an asymptotic analysis of this hydromagnetical problem has been done and used for the interpretation of the solar wind phenomenon around the earth). Example 3.14.13. The following singular perturbation appears as a suitable mathematical description of a semiconductor device in the situation, when it can be considered as a one-dimensional stationary pn-junction (see
[Mo, 1
1):
(3.14.58)
-E
2 $"+n-p
f(x), -n"+(n$')' = g(n,p),-p"-(p$')' = g(n,p),
=
where $,n,p are unknown functions (normalized potential and electron and hole densities), upper prime stands for the derivative with respect to x E U 5 R and f(x) , g(n,p) are given functions, g(n,p) =
-1
2
(4)(co-np)(n+p+2co) ,
with co > 0 a given constant, E being a small dimensionless parameter. Introducing new unknown functions u
c0n exp(-ji), v = cop exp($), one can rewrite (3.14.58) in the following form: -E
(3.14.59)
2
$"+c (u exp($) - v exp(-$)) 0
=
=
f(x)
-(e*u')' = G($,u,v) -(e-*vI)*
=
Notice that u E v
~($,u,v)
5
1 is a solution of (3.14.60) if $ 1s a solution
of the following singularly perturbed equation with a given second member: -E
2
*" +
2c
sh $ = f(x),
3. Singular Perturbations on Smooth Manifolds without Boundary
508
whose reduced equation has the following solution ji0 (x) = sh-l(f (x)/2c0). Consider the linearization of (3.14.59) at the point (IJJo(x),1,l), whose principal symbol is the following matrix:
(3.14.60) aO(x,E,6)=
0
where q(x) = 2c0 ch(+O(x)), the reduced principal symbol a0 (x,S) being just a0(x,0,5). It is readily seen that (3.14.60) is elliptic and the singular perturbation with the symbol s (x,E,~), 0
0
0 can be used for reducing the linearization of (3.14.59) at the point ji
=
I$~(X),u = 1, v
=
1, to a regular perturbation when x E R and w
f E C i ( R ) or when x E TI, f E C ( T
1
)
(periodicity conditions).
We shall come back later on to (3.14.58) while we will be considering boundary value problems for elliptic singular perturbations. Let M be a smooth manifold withoutboundary and let a(x,~,D)EOp L V ( M ) be elliptic of order u. Assume that it has a
( M ) for some u 2 E R . Then
(M) onto Ha -v
Hu
2
2
0
reduced operator a (x,D) which maps isomorphically
2
0
a (x,D) : H
(M)
s2
-f
H
s
-v
2
(M)
2
is an isomorphism for V s2 E R . Indeed, one gets this conclusion by using the classical elliptic regularity results for s 2 >
f12,
the duality argument for s2 =< - 0 +v
and
.
interpolation between -u2+v2 and a2 , if necessary (see [Fr -W. ,1,31 for more detail). Let s(x,E,D) be a reducing singular perturbation for a(x,~,D) (as defined above). The singular perturbation q(x,E,D), def 0 q(x,~,D) = 1-(a (x,D))-'s(x,~,D)a(x,E,D),
3.14. Reduction of Elliptic Singular Perturbations
is well defined, provided that
E~
is sufficiently small, and, moreover, as
a consequence of Theorem 3.14.6 and Remark 3.14.9, one has for the norm of the singular perturbation (3.14.61):
where the constant C does not depend on
E (O,E~].
E
Therefore, one can represent the inverse singular perturbation -1
(a(x,E,D))
by the following Neumann’s series: k
O
-1
Z (q(x,E,D)) (a (x.D))
(3.14.62) (a(x,E,D))-l =
s(x,E,D),
k>O C ( q ( x , ~ , D ) being )~ convergent in the Banach space k>O (M)) of all continuous linear mappings in H (M) L(H(s) ,E(M);H (s), E ( s ) .E uniformly with respect to E E (O,E 1.
the series
0
Hence, for each f E H
(s-V)
formula for the solution u E H (3.14.63) u =
,E
(M), one has the following asymptotic
( s ) ,E
(M) of the equation a(x,s,D)u = f:
k O N (q(x,C,D)) (a (x,D))-lS(X,E,D)f+ E g,
Z
O5k
E
E (O,E
If f E Cm(M) uniformly with respect to
0
1
E
with some constant C > 0.
E ( O , E ~ ] and symbols
a(x,~,c)admits an asymptotic expansion: a(x,E,5)
-
z
E
k k a (x,S)
k>O
then (3.14.63) can be simplified. We give here an example to illustrate this situation. Example 3.14.14. On the unit circle R 1 with 8 E [ - i ~ , n ) as a coordinate consider the following singular perturbation a(e,E,D,),
m
where q E C (R1) and satisfies the ellipticity Re(q(0))L > 0, tl 0 E R1. We choose the branch q(0)
=
509
((q(0))2)1’2 such that Re q(0) > O , V 8 E R l .
3 . Singular Perturbations on Smooth Manifolds without Bounda y
510
The reduced operator ao being identity, the singular perturbation defined by (3.14.64) is a family of isomorphic linear mappings uniformly with respect to that E~ given
6
E
( O , E ~ ] of H
(R
(S),E
1
onto itself, V s E R 3 , provided
)
is sufficiently small (if q(8) > 0, V 8 E SZ
E~
> 0 uniformly with respect to
E
1
then it holds for any
E (O,E~]).
One can use a partition of unity and the definitlon of singular perturbations on a smooth manifold without boundary for constructing a reducing operator for a(e,~,D) given by (3.14.64) (see Remark 3.14.9).
e
However, given that 8 E [ - r , n I isa coordinateonR1, one can give a different definition of a left reducing operator for a(B,E,DA) without using a partition of unity. Indeed, define the kernel s(@,E,e-e')
=
as follows:
S(e,E,e-e')
(2n)-1 c (1+E2(q(e))2k2)-1exp(ik('t3-8')). kEZ
Using the Poisson formula (1.2.180), one can represent S ( O , E , f 3 - 8 ' )
in
the form: (3.14.65)
S(e,E,e-e')
=
(2€q(e))-l I: exp(-1e-e'-2nk(/(Eq(e))), kEZ
given that with Re q ( 8 ) > 0 one has:
Define the singular perturbation
SE =
s ( ~ , ED
e)
as an integral
operator with the kernel S: def (3.14.66) (S(8,E,~e)g) (8) = J S(0,E,e-ei)g(e')de, v g E
Crn(Ql).
12 1
Since the principal symbol s
0
0
E s ( o , o , - 2 ) (R 1 ) , one has:
( 8 , 6 , < ) of
s(@,c,Da
is: s,=
"
2 2 2 -1 ( l + ~ (q(8)) ) ,
s(S,E,D ) E Op S(ogo*-2) (nl). Furthermore,
e
SE = s(B,E,De)has identity as its reduced operator. Therefore, as a
consequence of Theorem 3.14.6 and Remark 3.14.9, SE = s(B,E,De ) defined by (3.14.651, (3.14.66), is a reducing operator for a(e,6,De) given by (3.14.64) (in fact, it is a quasiinverse operator for a(B,E,Dg)),i.e. Id+ q (~B ) ,E,D S ( ~ , E , D ~ ) ~ ~ ( ~=~ , E,D where
e) ,
3.14. Reduction of Elliptic Singular Perturbations
51 1
is a family of continuous linear mappings uniformly with respect to E
E
(O,E~], V s
E
R 3 , provided that
E~
is sufficiently small.
Besides, (a(e,E,D,)) -1 =
X (Id-s(e,E,De)oa(B,E,De))k .,s(8,E,De), ktO
the series on the right-hand side of the last formula being convergent ( Q ) ) of all continuous linear 1 mappings in H ( Q ) uniformly with respect to E E (O,E 1 , V s E R3 (s),E 1 0 Furthermore, if fE E C m ( Q ) (uniformly with respect to E E (O,E 1 ) 1 0
in the Banach algebra L ( H
( s ) , E ( Q l ) ' H ( s ) ,E
.
then the solution u (8) of the equation ~(@,E, D (8) = fE(e), e~)U
E
E
(O,E
01, m
is well defined and belongs to a bounded set in C ( i l l ) , V provided that
E~
E
E
(0.~~1,
is sufficiently small (see Remark 3 . 1 2 . 2 3 ) .
Moreover, if f (8) admits an asymptotic expansion,
then so it is also for uE(8), (3.14.67)
(8)
-
c
E
k
uk(e), uk E
m
c (a1),
kZO
where the coefficients uk can be defined by recursion. For instance with f (8)
5
f (8), one finds easily: 0
Besides, the asymptotic convergence in ( 3 . 1 4 . 6 7 ) , space H
( 3 ) ,E
(R
1
(3.14.68)
in each
is an immediate consequence of the a priori estimate
with a constant C which might depend only on s and q, the latter resulting from Theorem 3.14.6 and Remark 3.14.9. A
special attention deserves the case when f(8) = fs(8)+g(8) where
3 . Singular Perturbations on Smooth Manifolds without Boundary
512 m
g E C (Q1) and f (8) is the singular part of f, having a finite number of points in R1 as its singular support. For instance, consider the case, when f (8) = sgn 8, 8 E Ql. Since sgn 6 E H
(0),E = sgn 8:
au
(Q ) ,
1
one has for the solution u (6) of the equation
Using the reducing operator s ( € l , ~ , D ) defined by (3.14.65), (3.14.66), 6 one can represent u (6) in the form:
u where on
E
(el
= s ( e , E , ~ )f (e)+EVE(e), 6 s
I
/v C E 1 1 (0,0,2) ,E E (O,eo1.
C
with some constant
C >
0, which does not depend
Using (3.14.65), (3.14.66), one finds:
, def (3.14.69) ~ ( ~ , e= l
s ( e , E , ~ )f
6
= (2Eq(e))-l c
s
sgn
(6) =
el
exp(-/e-e'-2nk//(Eq(e)))de',
e E nl.
kEZ Ql
Computing integrals on the right-hand side of (3.14.69), one finds explicitly the first term u ( E , ~ ) in the asymptotic expansion u (6) the following formula:
the point
e
= IT
being identified with 8 =
-IT.
Hence, also for u(E,e) the singular support consists of the same two points as a consequence of the pseudolocality property of singular perturbations (see Theorem 3.5.4). Now the construction of a right reducing operator for a given elliptic singular perturbation will be given and some applications will be considered.
3.14. Reduction of Elliptic Singular Perturbations
513
We are going to show that for a given elliptic singular perturbation a(x,E,D 1 E Op Sv(M) the left factorizing and reducing operators r(x,E,D and s(x,E,D
)
defined by their symbols ( 3 . 1 4 . 1 4 ) ,
)
are also right factorizing
and reducing operators, respectively. The essential part here is the following analogue of Lemma 3 . 1 4 . 4 : Lemma 3 . 1 4 . 1 5 .
Let j(x,~,S)E Lv(Rn) w i t h w 0
j (x,~) E I,
v (x,~) E
L e t a(x,c,S) E L ! ' ( R ~ ) .
R~
x
=
( 0 , 0 , v 3 ) and assume t h a t t h e reduced symbol
mn. Then one has:
i . e . f o r each s E m3 t h e folZowing i n e q u a l i t y h o l d s :
w i t h a c o n s t a n t c > 0 , which does n o t depend on
E
and u.
Proof. First a transparent heuristic argument will be presented, which explains why ( 3 . 1 4 . 7 0 1 ,
(3.14.71)
should be true.
Using Theorem 3 . 7 . 6 , one finds:
5
as a consequence of Definition 3 . 1 2 . 3 . Thus,
1, one has:
3. Singular Perturbations on Smooth Manifolds without Bounda y
514
given t h a t t h e o r d e r of t h e symbol on t h e l e f t hand s i d e of t h e l a s t
< v+p, v
inclusion i s v+p-(o,/crj , o ) + ( o , I , I )
IcrI > 0.
QF,
Besides, t h e amplitude of t h e o p e r a t o r which i s t h e k e r n e l of t h e o p e r a t o r
QE,
(see Definition 3.7.1),
coincides with t h e k e r n e l of t h e
operator
t h e o r d e r of t h i s amplitude being a t most v+p. Now a rigorous proof of t h e lemma w i l l be given, which i s very s i m i l a r t o t h e one of Lemma 3.14.4.
Without r e s t r i c t i o n of g e n e r a l i t y , one can
assume t h a t v1 = p l = 0. L e t a'(X,E,S)
def = a(x,E,<)-a,(~,S), j'(x,E,S) = j ( x , c , t ) - j m ( ~ , S ) ,
where a ' , a m and j ' , j m a r e as required i n D e f i n i t i o n 3.12.1 of c l a s s e s Lv(iRn).
Since j m does not depend on x , one has:
Therefore, it s u f f i c e s t o show t h a t (3.14.72) O p ( a ' ) . O p ( j ' ) - O p ( a ' j ' )
= EQ€. 2
( 3.14.73 ) Op (arn) .Op ( j ' ) -0p (amj
= €QE 3'
where
QE
I
'1
E Op LV+'(Xn), j = 2,3.
First,
(3.14.72) w i l l be proved. One has:
where
and t h e h a t
s t a n d s f o r t h e Fourier transform with r e s p e c t t o x
E Rn.
By Schur's lemma on boundedness of i n t e g r a l o p e r a t o r s i n L 2 ( s e e Lemma 3 . 1 2 . 8 ) , i n o r d e r t o p r o v e t h e lemma, it s u f f i c e s t o show t h a t f o r each s = (0,s2,s3)
E
n3 s -p
(3.14.75) f n < q > iR
onehas: -v
s -p
2<En2
-v
-s 3
-s 2<ET>
v
' IE
3/K(ll,E,T)
iRn,
v
E
Idr? 6 CE,
E
(o,Eolp
3.14. Reduction of Elliptic Singular Perturbations
-v
s -p (3.14.76) In<'l> R
s -p
-v
-s
3
2<Erl>
515
-S
31K(Q,E,T) /dT 5 CE.
2<ET>
v
rl
E mn, v
E
E
(O,Eo1,
where the constant C > 0 might depend only on s2,s3,n,v2,v3 and the symbols a and j, but not on As
0 j (x,S)
E,T
and rl.
a consequence of Definition 3.12.3
and the assumption:
1, one finds:
Let b(x,E,S) = b,(E,<)+b'(x,E,S)
with bm and b' as in Definition 3.12.1,
where Ck > 0 does not depend on <,E,x. The same argument shows that D a j ' (x,E,S) satisfy the inequalities (3.14.77) with different constants which may depend on the multi-index a.
Hence, one gets:
with perhaps different constants Ck > 0. Now the mean value theorem and the inequalities in Definition 3.12.1
of classes Lp(Rn ) yield:
where we have used Peetre's inequality: <S>p
5
2'pl<<-T>lp'
,
V < , T
E lRn,
v p E R.
The same inequalities being true for Daa' (x,€,S) with some constants 'k,a'
one gets the conclusion
516
3 . Singular Perturbations on Smooth Manifolds without Bounda y Obviously, the inequality
is satisfied as well. Now we have all the elements necessary to prove (3.14.75). Using (3.14.74), (3.14.78), (3.14.79) and the inequality
< n - O 2 2
where ml = 1+1u2-11+/u31+js21+1s31+1u 2 +v2 1 and m2 =
l u 2+v 2 I + / s 2 ] + / s 3 / .
We choose k to be sufficiently large in order to guarantee the convergence of the integral on the right hand side of (3.14.81), and that
.
yields (3.14.75)
The same argument can be used in order to prove (3.14.76). Hence, we have proved (3.14.72). In order to show (3.14.73), one will have to estimate the kernel
-
def K _ ( ~ , E , T ) = j'(T-n,&,n) (arn(~,T)-arn(&,n)). Using (3.14.80) instead of (3.14.79), one gets (3.14.75), (3.14.76) with
K replaced by Km. This yields (3.14.73). We have shown that the singular perturbations Q ? , j = 2,3, in (3.14.72),(3.14.73) have atmostorder I that QE = Q:+Q~ E Op LU+' (iRn1, as well.
v+p,
so
Now, using exactly the same argument, as in the proof of Theorem 3.14.6, one shows its analogue with r(x,&,D), s(x,E,D) as right factorizing and reducing operators:
the same kind of inequalities being also true for elliptic singular perturbations in S'(M)
with M a smooth manifold without boundary.
3.14. Reduction of Elliptic Singular Perturbations
517
Corollary 3.14.16.
L e t M be a smooth manifold w i t h o u t boundary and l e t a E o p s'(M) of order v
=
f a c t , a0
H
be e l l i p t i c
(v1,v2,v3)and a s s m e t h a t t h e reduced operator ( 0 '"2 , O ) (M! maps isomorphicaZZy M (M) o n t o L~(M), so t h a t , i n ao E op s V-
:
-v
( M ) + HS
s2
2
(M)
is an isomo$icm
f o r any giuen
s2 E
w.
2
Consider t h e equation a(x,~,D)u(x) = f (x) w i t h f E H
(s-v), E
(M)
giuen.
Then U(X) can be represented i n t h e form: 0
(3.14.84) u(x) = S(x,E,D) (a (x,D))-'f(x)
1 IvI 1 ( s ) , E
where
depend on
E;X
+ EV(X),
w i t h some c o n s t a n t c > 0 , which does n o t
and f.
Indeed, using (3.14.82) and the corresponding version of (3.14.30),
(3.14.31) on a smooth manifold (see Remark 3.14.9), one gets immediately (3.14.84). Example 3.14.17. Let Tn be the n-dimensional torus, Tn = Rn/Zn, with the global coordinates m n x . E [0,1], 1 5 j 6 n. Let q(x) > 0, q E C ( T ) be given, and consider the 7 equation on T ~ :
(3.14.85)
(E
2 2 A -A+q(x))u'(x)
=
f(x),
where f E L2(Tn) is given and A is the Laplace operator on Tn. Using (3.14.84), one can represent the solution u(x) of (3.14.85) in the form:
where s(x,E,D) is a right reducing operator for the singular perturbation on the left hand side of (3.14.15) and
v
=
(0,2,2)
I / v s I I (v),E
I
, with
5 CI/fl
and a constant C > 0 which does not depend on
L (T
E,U
and f.
Moreover, as a right reducing operator, one can choose the one, which is globally defined as follows:
where S(E,x) =
Z kEZn
2 2
(1+4rr E [kI2)-l exp(2rri
3. Singular Perturbations on Smooth Manifolds without Bounda y
518
Using again the Poisson formula ( 1 . 2 . 1 W ) , S(E,X) =
A.
Z
-2
one finds:
(E,x-k),
kEZn where A.
-2
-1
(€,XI = FS,x(1+c252)-1
is given by ( 3 . 2 . 2 3 )
F o r n = 3 , one finds using ( 3 . 2 . 2 4 ) :
and for the solution u of ( 3 . 1 4 . 8 5 )
one gets the asymptotic formula:
0
where (lu ) (x) is the periodic extension of uo with a constant 5ClIfll 2 L (T )
6,U.f.
c
:
Tn
-f
C onto Rn and
> 0, which does not depend on
Example 3 . 1 4 . 1 8 . Let U aU
C
R2 be a bounded convex domain, whose boundary aU is a Cm-curve,
{x=y(s), 0 5
=
s 5
L}, the parameter s being the arc length on aU
between a fixedpint xo = y(0) and the current point y(s). Let u E ( x ) be a harmonic function in U, which satisfies the following boundary condition on
au
(3.14.87)-Eau~(x)/aNs+a(s)UE(X) = g'(s),
where N
x=y(s), 0 5 s 5 L,
is the unit inward normal to aU at x = x(s), a ( s ) > 0 ,
a ( 0 ) = a(L) and the periodic extension of a
:
v
S
E [o,L],
[O,L] -+ R+ Onto R iS
m
supposed to be in
C (R).
We shall also assume that gE E H ( o ,s2-f ,s3-l) s2+s3> 3/Z
sought in H
caul with
s2 > f ,
(3), E
will be
(U), so that the trace on the left hand side of ( 3 . 1 4 . 8 7 )
is well-defined, V
E
>
0 (see Theorem 2 . 2 . 3 ) .
Denote by P the following integral operator: (Pa)(x) where (3.14.88)
,€
and the harmonic functions uE(x) satisfying ( 3 . 1 4 . 8 7 )
P(s,x-Y
-2
3.14. Reduction of Elliptic Singular Perturbations
519
Introducing
with
T
=
dy/ds the tangential unit vector to aU at y(s), one can rewrite
P(s,x-y(s)) in the form: P(s,x-y(s)) = F-lI~x
(3.14.89)
I I)
exp ( -P 5
1
the right hand side of (3.14.89)
#
being well-defined, since p > 0 , V x E
i,
x#y(s), as a consequence of the convexity of U. Thus, one finds for x
p(s',s)
> 0,
for
s'
+
y(s'),
# s.
Notice that 2
x'(s',s) =s'-s+(s'-s) a(s',s), p ( s ' , s ) =tk(s',s)( s ' - s ) ~k(s',s) , > 0, s'
where k(s,s) = k ( s ) b 0 is the curvature of m
k(s',s) being C
au
#
s,
at y(s), a(s',s) and
in ( s ' , s ) .
Further, since y(0) = y(L), one can consider the operator Q,
with the distributional kernel Q(s,s'-s) defined by (3.14.901, operator acting on functions $ defined on the circle R R , R
=
as an L/2r
parametrized by s E [O,LI. Let {6k}15ksm be a covering of RR by intervals Ak of length 6 sufficiently small and let {
x
~ be} a subordinate ~ ~ ~ partition ~ ~ of unity.
Then one has:
where (Q$
k
) (s')
is the operator of the form:
3 . Singular Perturbations on Smooth Manifolds without Boundary
520
f f qk(s ' , s , S ')exp(i ( s ' - s ) 5 ' ) $k ( s ')dE'ds
(Qa,) ( s ' ) = (2r)-'
R3R
with the amplitude q given by the formula: k (-.14.91)qk(s',s,€,') =
as, ,NS> 1 5
1 exp (i( s ' - s 1 2 a ( s ' ,s ) 5
I-
(f )k( s
I ,
s ) ( s '- s )
It is readily seen that with any cut-off function $ ( S ' ) ,
jc'l
2
15
$(<')
]). 5
0 for
1 for 15'1 2 26, $ E C m ( R 5 , 1 , the amplitude JI(c')qk(s',s,C')
2 6, $ ( 6 ' )
satisfies the conditions of Definition 3.7.1 with v = (O,l,O), i.e. the operator Q with the distributional kernel Q(s,s'-s) defined by (3.14.90) is a singular perturbation in Op qO(t',S')
=
(Or'
S
1r
(62,)
whose principal symbol
o
1511.
Therefore, the operator (3.14.92) (a(S',E,Ds,)@) (S')
def =
is a singular perturbation in Op
(-Ea/aNs,+a(S'))( P o ) (S') S(orofl)
1 ro
(62
R
whose principal symbol
ao(s',E,5') is: (3.14.93) a0(s',€,5') = ~ ( S ' / + a ( s ' ) . Since a(s') > 0 , V s ' E [ O , L J ,
singular perturbation (3.14.92) is
elliptic of order (O,O,l), its reduced operator being just the multiplication by a ( s ' ) . Define a reducing operator as a singular perturbation b in (O,O,-l)
OP
s1,o
(62,)
given by the formula
(3.14.94) (b(s',s,Ds,)$)( s ' ) =
(2w)-'
Z
1
where
Ok
=
I
def =
( l+E
(a(s'))-l<<'>)-lexp(i ( s ' - s ) 5 ' ) $,(s)dg'ds
R R
xk$.
The singular perturbation (3.14.94) is at the same time a left and a right reducing operator for a(s',&.D
,)
given by (3.14.92).
One can also use as a reducing (left and right) operator in the case considered.the integral operator with the kernel (3.14.95)
S (E ,S
' , S ) de=f a ( S ' ) (TE)-1
m
(1+<
) -'Cos
(a(s
) E - l ( S' - 5 )
0
which is the inverse Fourier transform of
( 1+
(E/a( s ' ) ) I ,€ '
1)
:
5 I ) d5
3.14. Reduction of EUiptic Singular Perturbations
Seeking a harmonic function u'(x) the form uE
in U satisfying (3.14.87) on aU, in
with P defined as above, one gets for
= PQE
52 1
QE
the following
singularly perturbed equation on aU: (3.14.96) (a(s',E,Ds,)QE)( s ' ) = gE(s') with gE E H
(0,s2-f,s3-l),E
(au) given.
A s a consequence of Theorem 3.14.6 and Remark 3.14.9, equation
(3.14.95) has a well-defined solution
QE
consequence of Theorem 2.2.3, u E - PQE
H(0,s2+f,s3),E H(O,S2,S3),E
(au).
AS a
(u) for s 2 > f ,
s2+s3 > f . If g E Cm(aU) then to E
(O,E~], V
E~
<
+'
-,
E cm(au) and uE E
uE being
Cm(fi)
uniformly with respect
well-defined, as a consequence of
Theorem 3.12.12 and Remark 3.12.23. 0
Let u (x) be the harmonic function in U which takes the value g ( y ( s )) y ( s ) E au, on
au.
,
Rewriting (3.14.87) in the form:
(-Ea/aNs+a(s))uE(x)-u0 (x)
=
0, for x = Y(s)
E au
and writing the formal asymptotic series
one gets the following formal asymptotic expansion for uE(x),x uE(x)
-
(a(S))-l
c
=
y(S) E aU:
Ek((a(S))-la/aNS) k O (x), x=y(s).
k>O Denote by PD the operator which associates with a given function f on aU the harmonic function in U which takes the value f on aU, so that
The asymptotic series on the right-hand side of the last formula is asymptotically convergent ot uE (x). Indeed, introducing
3. Singular Perturbations on Smooth Manifolds without Bounda y
522
one f i n d s e a s i l y t h a t t h e d i f f e r e n c e 6
v P
def 6 = u
-
z
k k
E U ( X )
O
i s t h e harmonic f u n c t i o n i n U which s a t i s f i e s on aU t h e following boundary condition:
Thus, a s it has been y e t shown, one has f o r vE t h e following e s t i m a t e :
Since up E
m -
c (u),
p = 0.1,
...,
for g
E cm(au)
(classical result for
e l l i p t i c boundary value problems without small p a r a m e t e r ) , one g e t s t h e conclusion t h a t vE
E
E ( 0 , s 0 1 and, 5 CsP+l, V p = 1,2 where
C m ( f i ) uniformly with r e s p e c t t o E
moreover, f o r each m t 0 holds:
1 I vE1 I
t h e constant C may depend on m and p.
,...,
Cm(3
3ne can consider t h e problem of d e f i n i n g a harmonic f u n c t i o n uE i n U, which s a t i s f i e s t h e boundary condition: (3.14.93
- a u E / a i s + a ( s ) u 6 ( x ) - E B o2au/as 2
=
E
( s ) , x = y ( ~ E)
au
where Is i s a smooth u n i t v e c t o r f i e l d on a U , > 0 , V s E
au,
and
S
cx(s),B(s)
v s E
a r e given smooth f u n c t i o n s with r e a l v a l u e s , a ( s ) > 0 , 5 ( s ) > 0 ,
10,Ll. Using t h e same o p e r a t o r P a s above and t h e same argument, one g e t s on
t h e boundary
au
perturbation a
t h e equation of t h e form (3.14.96) with a s i n g u l a r
E Op
S("'")
1 PO and whose p r i n c i p a l symbol a. ao(S',E;')
(0,)
which i s e l l i p t i c of o r d e r v
=
(0,1,1)
has t h e form:
=
Besides, t h e reduced o p e r a t o r a ( s ' , D
,)
. being an e l l i p t i c pseudo-
d i f f e r e n t i a l o p e r a t o r i n S ( 0 ~ 1 ~(R,0) ) of ozder ( 0 , 1 , 0 ) , 1,o 0 (a ( s ' , D s , ) b ) ( s ' ) = (-
x=y(S')
t h e r e e x i s t s a well-
defined harmonic i n U f u n c t i o n , which s a t i s f i e s (3.14.97) on a U .
3.14. Reduction of Elliptic Singular Perturbations Moreover, for each 9 ' E H (o,s2-3,2,s3-1),E(U) $'
of the corresponding equation (3.14.96) :
uE = P$'
$€
523
one has for the solution
E H
(O,s,-f,s,) L J
(U), so that .E
E H(0,s2,S3),E(U) provided that s2 > f, s2+s3 > f .
As a reducing operator one can take the one with the symbol (1+c(8(s')/
operator with the kernel (3.14.95) with
with a(s,s'), k(s,s') defined as above. Notice, that each term $ ( < ' ) (s'-s)2P(ia(s',s)S'-(f)k(s',s) lS'I)p with
$ ( E l ) the cut-off function, as defined above, is an amplitude which defines
(nR 1 , so that the operator a singular perturbation in S(or-pro) 1 ,o $ + Q$ = -a/aN (P$)(x), x=y(s'), is locally defined by an amplitude q S' k' which is an asymptotic sum of amplitudes, whose orders -p decrease to -m as p
+
m.
Obviously, the argument above extends to any dimension n > 2, the kernel of the operator P being in that case defined in an obvious way as -1 -n P(X,Y') = 2n <x-Y',N ,>Ix-Y*/ , x E Y
being the unit inward normal at y' E aU and N Y' unit sphere in Rn
u,
nn
au,
y' E
being the area of the
.
In the multidimensional case the operator -2 2/as2 in (3.14.97) is replaced by any elliptic second order differential operator in tangential variables y' E aU, whose principal symbol has the form: with a (n-l)X(n-l) positive definite matrix A(y'), V y' E aU
,
the ellipticity
condition being fulfilled, provided that
1 the given smooth vector-field on aU. Y' We give here two examples in order to illustrate the reduction procedure and its applications to elliptic difference operators. Example 3.14.19. m
Let q1 (x) 5 0, q 1 E CO(R) and let q(x) = l+q (x). Consider the family of 1
524
3. Singular Perturbations on Smooth Manifolds without Boundary
difference operators a(x,h,hD) on
%1
with the symbols a(x,h,hS),
With iDh and iDE the forward and backward finite difference derivatives on 1 2 the greed % , one has: a(x,h,hD) = h (D D*l2+q(x)DhD; (see also Examples h h 3.11.10). Obviously, a(x,h,hS) is a finite difference approximation of the 2 2 differential operator p(x)D2 = -p(x) d / d x on W , so that (3.14.98) can be considered as a singular perturbation (with the meshsize h as the small 0
parameter) of the difference operator a (x,h,hD) = p(x)D D*, the latter h h being the usual elliptic three point finite difference approximation of the differential operator -p(x) dL/dxL on W
.
One has: 0
a(x,h,hD) = r(x,h,hD) a (x,h,hD) with r(x,h,h3) the following factorizing operator: 2 (3.14.99) r(x,h,hD) = q(x) + h DhD:
=
(2+q(x))-(8 h
h
h being the shift operator on E( : (Bhu)(x) = u(x+h) . Introduce the following operator s(x,h,hD):
f3
s (x,h,hD) =
Op(r (x,h,hS))
Obviously, s(x,h,hD) is a discrete convolution operator on
4 with the
kernel S (x,x-y) given by the formula: h
It is readily seen that s(x,h,hD), acting on the mesh functions u(x) with a compact support in
4
according to the formula:
3.14. Reduction of Elliptic Singular Perturbations with
525
(x,x-y) given by (3.14.100), is a left and a right reducing operator h for a(x,h,hD) S
.
Indeed, one gets this conclusion by applying Theorem 3.11.17 to s(x,h,hD)
a(x,h,hD) and a(x,h,hD)
o
,.
s(x,h,hD), respectively. In both
cases a(x,h,hD) after the multiplication from the left or from the right 0
by s(x,h,hD) is reduced to a regular perturbation of a (x,h,hD) by terms of the form hb(x,h,hD) with b(x,h,hS) E F ( O r 2 ) (IR) of order (0,2) at most. 1P O Example 3.14.20. m
Let q E C (IR) be as in the previous example and let -Ah be the usual five point elliptic approximation by finite differences of -A on the greed in B2 (see Examples 3.11.10). Consider the following discrete boundary value problem in 2 E \ , x2 2 01:
2 %,+= tx
(3.14.101)
{
(l-Ah)u(x) = f(x),
'(,DX
hD:
1'
x E
%2 ? + x2 ,
,h-iq(xl)Dx ,,)u(xl,x2)
2
1
> 0,
Ix2=o
=
$(x,),
x1 E
4,
(3.14.101) being a finite difference approximation of the following differential boundary value problem:
We are going to show that problem (3.14.101) with 2 has a unique ' ( s 1, s 2-2,s3),~,h(%,+) and '(S~,S~-~/~,S~-~),E,~
'
solution u E
H ( s ) ,E,h(%,+
)
, provided
that s2 2 2, s3 b 0
and s
3
being integer. AS a consequence of Lemma 2.8.18 (applied in the variable x2
there exists an extension if E
ff ( s
f in (3.14.101).
,S
1
-2,s3),s,h(<) 2
Introducing v = op(1+/<j2)-lif,v E H ( s ) ,c,h(%
1
E (X$,+
),
of the second member
)'
it is readily seen
that
Thus, introducing w = u-v, problem (3.14.101) is reduced to the case, when f E 0, so that from the very beginning one can consider this situation, without restriction of generality.
3. Singular Perturbations on Smooth Manifolds without Boundary
526
Each solution of the equation (l-Ah)u(x) = 0 in for x2
+
+-, has the
with some $(x
1
)
g,+, which decreases
form:
and 6+ given by the formula:
where, as usual,
2
=
-1 1+1S1I2, C1 = (ih) (exp(ihS1)-l).
Thus, $(x ) is a solution of the following singularly perturbed 1 finite difference equation on
4:
1 (3.14.102) (a(xl,E,h,hol)Q)(x,) = $(xl), x1 E Rh ,
where a(xl,E,h,hD1)has the following symbol: 2 (3.14.103) a(xl,E,h.hC1) = ~ 1 5 ~ 1 Obviously, a E
F;Yb1")
(R
and is elliptic of order ( O , l , l ) ,
since its
principal symbol a
0'
def -1 (3.14.104) a0(x1,p,nl) = p Iwl12+q(xl)/wll( ( I + iwl = exp satisfies the inequality: -1 (3.14.105) ao(xl,p,n1 ) 2 coIwl/
1
V (xl,p,nl)E R 1 x R + x Tl,nl
(see Definition 3.12.42). 0
Besides the reduced operator a (xl,h,hD1), (3.14.106) a0 (xl,h,hD1)=q(xl)
>/2)2)f -haxl,h>/2)
1' lfh 0 -1 having ao(xl,h,nl) = h q(xl)lwll ((1+( lw11/2)2)t-(w11/2) asitsprincipal symbol and being elliptic of order ( 0 , l ) in the class Op F ( o ' l ) ( B ) ,
is
also invertible, its inverse being defined by the formula:
Therefore, the perturbed operator has an inverse as well, provided that E
E ( O , E ~ ] with
E~
sufficiently small.
3.14. Reduction of Elliptic Singular Perturbations
527
Furthermore, a(xl,h,hD1) being elliptic of order (O,l,l), one gets the conclusion that the solution I) of (3.14.102) satisfies the condition:
Hence, an easy computation (similar to the proof of the second part of Theorem 2.7.28 concerning the lifting operator Lh) shows that uniformly with respect to (E,h)E
(0,ll
x
(0,1] holds:
As a left and a right reducing operator for a(xl,h,hD1 ) one can take the one defined by the symbol:
An easy computation shows that s(xl,E,h,hD1 ) is a discrete convolution operator with the kernel S(p,xl,xl-yl)given by the formula:
Since k = (x -y )h-' E Z for (xl,yl)E
1 %X
%1 ,
the integrand on the right
hand side of (3.14.11), contains Tschebyshev's polynomials cos(2k arcsint) of degree 2k.
,l, Of course, one has for each $(x 1) with a compact support in m
and one finds easily that S(-,xl,xl-y1 ) 0 otherwise.
I
=
:
6 , i.e. 1 for x1 = y1 and x1 'Y1
3 . Singular Perturbations on Smooth Manifolds without Boundary
528 Notes ___
The concept of vectorial order v E R3 and classes Op P
of singular
perturbations with smooth coefficients were introduced in [Frank, 15,19,22 1 but were implicitly present also in [Vishik - Luysternik, 1
1.
The classes
op Lv (also in the case of variable symbols) were considered in [Frank, 221, [Frank - Wendt, 1,5,101. (U), Op S v ( U ) are natural extensions of the corres1 ,o ponding classes of pseudodifferential operators to singular perturbations,
Classes O p
Sv
as well as the concept of uniform pseudolocality and the results concerning
the C*-algebra structure (adjoints and products of singular perturbations are still in the union with respect to v E iR3 of the classes mentioned above 1 . While presenting the stationary phase, Laplace and saddle point methods, we follow essentially [Fedorjuk, 1 in [Dieudonns, 1 1, [Hbrmander, 4 Examples 2.8.15
-
1. Other presentations are to be found 1 , [Melin - Sjdstrand, 1,2 1 and others.
3.8.18 (also in a slightly more general situation) were
considered in [Frank, 14
1
(see also [Guillermin - Sternberg, 1
1
as far
as Example 3.8.15 is concerned). Examples 3.8.19, 3.8.20 are useful for the theory of coercive singular perturbations (see [Frank, 22 1, [Frank - Wendt, 1,3,10 1) which are the subject of vol. I1 of this book. Example 3.8.22 is taken from [Frank, 14
1.
Example 3.8.23 and general classes of hyper-
bolic difference operators were considered in [Frank,5,7] (see also [Strang,
1
1
for first order systems of hyperbolic difference operators). Fourier integral singular perturbations, as presented in 2.9, are a
natural extension of the local theory of such operators without a small
, 1 1 , [Ludwig , 1 1, [Hbrmander, 3 1 , [Duistermaat , 1 1, [Duistermaat , 1 1 , [Trgves, 1 1 , [Taylor , 1 1 ) . On
parameter (see [Lax Hormander
the other hand they are tightly connected with the local theory of Maslov's canonical operators (see [Maslov
,
1 I).
The idea to use the stationary phase method in order to establish the transformation formulae for the symbols of the classical pseudodifferential operators after smooth diffeomorphisms belongs to M. Fedorjuk (see [Fedorjuk,l,3]). However, earlier this kind of formulae was established in [Hbrmander, 2 ] without using the stationary phase method. In 3.10 the procedure indicated in [Fedorjuk, 1 1 is extended to the classes of singular (U) and Op S v ( U ) , as defined in 3.3. The idea to use Sv 1to an equivalent global definition of singular perturbations on a smooth manifold
perturbations Op
529
Notes without boundary (Definition 3.10.14) comes from [Hdrmander, 2
1
where such
a definition was introduced for classical pseudodifferential operators. Essentially, the difference between both cases consists in the presence of two large
independent parameters 151 and
E
-1
when one considers singular
perturbations, while for the classical operators only 151 ( 5 E Rn
being
the cotangent variables) is relevant. Example 3.10.15 is of interest for the diffraction theory, where the small parameter characterizes the loss of energy near the boundary (see vol. I1 of this book where the eigenvalue problems for such and more general coercive singular perturbations are considered). The C*-algebra of difference operators considered in 3.11 was introduced in [Frank, 4,9,10
1.
A
different kind of one parameter families of difference
operators, adapted to the approximation of first order hyperbolic systems, where considered in [Yamaguti - Nogi
, 1 1.
operators introduced in [Lax - Nirenberg, 1
The algebra of difference
1
was the main tool for proving
the Von N e m n n conjecture concerning the stability of difference evolution equations (Von Neumann's conjecture had been an open problem for many year before the appearance of [Lax-Nirenberg, 1
I).
The difference version of the
local theory of Fourier integral operators, as considered in 3.11, is
1,
useful for hyperbolic finite difference approximations (see [Strang, 1
1)
[Frank, 5
of hyperbolic differential and pseudodifferential operators.
In terms of this kind of Fourier integral operators, one can reformulate the Courant-Friedrichs-Levy stability condition also in the case of general hyperbolic operators. Definition 3.11.31 was introduced in [Frank, 5
1.
The
class of hyperbolic difference operators and asymptotic formulae for their solutions, as the mesh-size vanishes, are to be found in [Frank, 7
1.
The
difference version of Fourier integral operators with a small parameter 0 (thus, they are affected by the presence of two small parameters,
E
>
E
> 0 and the mesh-size h > 0 ) is useful for the numerical treatment of
singularly pertrubed hyperbolic operators. There is a hope to be able to elaborate on this topic in forthcoming volumes of this book. The ellipticity concept for singular perturbations was introduced in [Fife, 1
1
and, independently in the form as it is formulated in 3.12
(Definition 3.12.2) in [Frank,15,19,221. C h a r a c t e r i s t i c t w o - s i d e d a priori estimates for elliptic and coercive singular perturbations (uniform with respect to the small parameter) were stated in [Frank, 15,19,22] and proved in [Frank, 22
1.
The strong ellipticity concept for finite difference
3. Singular Perturbations on Smooth Manifolds without Boundary
530
1.
operators is already present in [Lax - Nirenberg, 1
The concept of
ellipticity for one parameter families of difference operators was introduced in [Frank, 4
1
and independently (for a subclass of finite difference schemes,
which approximate elliptic differential operators) in [Thomee - Westergren, 1
1.
Two-sided a priori estimates for elliptic finite difference operators
were stated in [Frank, 41 and proved in [Frank, 13
1.
For difference
singular perturbations analoguous concepts were introduced in [Frank, 16, 17,21
1
and corresponding two-sided estimates established in [Frank, 23
1.
A priori estimates (uniform with respect to the small parameter) in L P norms for elliptic singular perturbations were established in [Sweers, 1.21. Theorem 3.12.62 is due to Hdrmander and we follow esstentially the scheme
1
in [Hdrmander, 1
for proving it. In order to establish uniform a priori
estimates in L -norms for elliptic singular perturbations, the classical
P
result in [Michlin,l,2 1 (Corollary 3.12.63) suffices. The guide lines for proving results similar to Theorem 3.12.62 are to be traced back, for instance, in [Zygmund, 1
1.
As far as L -estimates for classical pseudo-
P
differential operators and more general classes of operators are concerned, the reader is refered to [Beals, 2
1
1 ,
[Triebel, 1
1
1,
[Fefferman - Stein, 1
1,
[Strichartz,
and others.
For the strongly elliptic differential systems introduced in [Vishik,
11 M.I. Vishik proved that the symmetric quadratic forms associated with the real part of such a system of order 2m (in a bounded domain R
C
lRn
with sufficiently small diameter) defines an equivalent norm in the Sobolev space Hm(fi). A proper extension of this result to arbitrary bounded domains was given in [Gsrding, 1.2
1
and came into use in the mathematical
literature under the name of Ggrding's inequality. This kind of inequality
1.
for singular integral operators was established in [Calderon - Zygmund, 1 Several important results toward the sharp form of G&ding's are to be found in [Seely, 1
1.
inequality
The sharp form of Girding's inequality for
the classical pseudodifferential operators, as introduced in [Untergerger Bokobza, 1 1, [Kohn-Nirenberg, 1
1
-
and for one parameter families of
difference operators (appearing in the stable approximations of well posed evolution problems), was first established in [Lax - Nirenberg, 1
1.
A
simplified proof of the sharp form of Ggrding's inequality was given in [ Friedrichs, 2
1
and [Vaillancourt, 1
3.
For classes F"(Zn
)
(see Definition
3.12.32) of one parameter families of difference operators the sharp form
of Gsrding's inequality was proved in [Frank, 9,lO
1.
For singular
Notes
53 1
perturbations whose reduced symbol is identity it was stated in [Frank, 241 (see also [Helffer
-
SjGstrand, 1
1
for the case of semi-classical pseudo-
differential operators). As far as Gsrding's inequality is concerned in the case of pseudodifferential (and more general) operators without small or large parameters, the reader is also refered to [Beals - Fefferman, 1 [Fefferman - Phong, 1
1
1,
and others. Example 3.13.14 is taken from [Frank, 3
and the last part of Example 3.13.15 from [Frank, 24
1
1. For proving the
Lax-Nirenberg theorem (Theorem 3.13.19), we follow essentially [Friedrichs, 2
1,
[Vaillancourt, 1
relevant
1
(with some minor modifications).Example 3.13.28 is
for the probability theory (see [Friedlin - Wentzel, 1
1
and for
quantum mechanics (see [Helffer - Robert, 1 I). A s far as the difference methods for the conservation law systems are concerned, the reader is refered to [Lax, 2,3 1, [Godunov, 1 1, [Harten - Hyman - Lax, 1 1, [Osher,
1
1,
[Van Leer, 1
1
and others.
The idea of a constructive reduction of coercive singular perturbations to regular perturbations was put forward in [Frank, 19
out in [Frank - Wendt, 1-5,6,10
1
and fully worked
1 and [Wendt, 1,2 1. The Wiener-Hopf
factorization was used in [Eskin, 1
1
in order to reduce an elliptic pseudo-
differential singular perturbation with homogeneous Dirichlet boundary conditions to a regular perturbation. Example 3.14.11 is taken from [Frank - Wendt, 1
1.
devices (see [Mock, 1
Example 3.14.13 comes from the theory of semiconductor
1,
[Markowich, 1 1, [Smith, 1
1,
[Sze, 1 1). Example
3.14.17 is of interest for the linear elasticity theory and Example 3.14.18 is relevant for the diffraction theory. The idea of reducing elliptic finite difference singular perturbations to regular perturbations, put forward in Examples 3.14.19 and 3.14.20, can be consistently worked out for the general elliptic finite difference operators in the same way and spirit as it has been done for the elliptic singular perturbations in Op L v ( M ) .
This Page Intentionally Left Blank
533
Bibliography
Adams, R.A.: [l] Sobolev spaces. Academic Press, 1975. Agmon, S.: [l] Lectures on elliptic boundary value problems. Van Nostrand, Princeton-London-Toronto 1965. Agmon, S . , A. Douglis and L. Nirenberg: [l]Estimates near the boundary for solutions of elliptic partial differential equations satisfying general boundary conditions. I. Comm. Pure Appl. Math. 12,623-727 (1959); 11. Comm. Pure Appl. Math. 17, 35-92 (1964). Agranovich, M.S. and M.I. Vishik: [l]Elliptic problems with a parameter and parabolic problems of general type. Usp. Mat. Nauk, 19 (3), 53-161 (1964) (Russian). Airy, G.B.: [l]On the intensity of light in a neighborhood of a caustic. Trans. Cambr. Phil. SOC.6, 379-402 (1838). Andronov, A.A., A.A. Witt and S.E. Hailin: [l]Theory of oscillations.“Nauka”, Moscow 1981 (Russian). Arnold, V.I.: [l] Mathematical methods in classical mechanics.
“Nauka”,
Moscow, 1974 (Russian). English transl.: Springer-Verlag, Berlin-Heidelberg-New York 1978. Aronszajn, N. and K.T. Smits: [l]Theory of Bessel potentials I. Ann. Inst. Fourier (Grenoble) 11, 385-475 (196 1). Aubin, J.P.: [l] Approximation of elliptic boundary value problems. Wiley, New York, 1972. Bachvalov, N.S.: [l]On the optimization of methods for the solution of ordinary diffeential equations with highly oscillating solutions. J . of Comp. Math. and Math. Phys., 11:5, 1318-1322 (1971) (Russian). Beak, R.: [l]A general calculus of pseudo-differential operators. Duke Math. J. 42, 1-42 (1975).
534
Bibliography
- [2] LP boundedness of Fourier integral operators. Memoirs Amer. Math.
SOC.,Providence, R.I. 1982. Beals, R. and C. Fefferman: [l]On local solvability of linear partial differential equations. Ann. of Math. 97, 482-498 (1973). Beals, R., C. Fefferman and R. Grossman: [l] Strictly pseudoconvex domains in 43”. Bull. Amer. Math. SOC.8, (2), 125-322 (1983). Bellman, R.: [l]Perturbation techniques in mathematics, physics and engineering. Holt, New York 1964. Bensoussan, A., J.L. Lions and G. Papanicolaou: [l] Asymptotic methods in periodic structures. North-Holland, Amsterdam-New York-Oxford 1978. Bensoussan, A., J.L. Lions and R. Temam: [l]Sur les methodes de decomposition, de d6centralisation et de coordination.
Cahiers IRIA, 11, 5-190
(1972). Berezin, I.S. and N.P. Zhidkov: [l]Computing methods. Pergamon Press, 1965. Berger, M.S. and L.E. F’raenkel: [l]On the asymptotic solution of a nonlinear Dirichlet problem. J . of Math. and Mech., 19, 553-585 (1970). Bers, L.: [l] Mathematical aspects of subsonic and transonic gas dynamics. Surveys in applied mathematics. 111. J. Wiley, New York, 1958. Besjes, J .G.: [l] Singular perturbation problems for linear elliptic differential operators of arbitrary order, I , Degeneration to elliptic operators. J . Math. Anal. Appl. 49, 34-46 (1975); 11, Degeneration to first order operators. J . Math. Anal. Appl., 49, 324-346 (1975). Birkhoff, D.: [l]Quantum mechanics and asymptotic series. Bull. Amer. Math. SOC.39, 681-700 (1933). Birkhoff, G.D.: [l]On the asymptotic character of the solutions of certain linear differential equations containing a parameter. Trans. Amer. Math. SOC. 9, 219-231 (1908). Bogolubov, N.N. and Yu.A. Mitropolskii.1.: [l] Asymptotic methods in the theory of nonlinear oscillations. “Nauka” , Moscow 1974, (Russian), English transl.: Gordon and Breach, New York 1961. Borel, E.: [l] Sur quelques points de la thkorie des fonctions. Ann. Sci. Ecole Norm. Sup. 12 (3), 9-55 (1895).
535
Bibliography
Boutet de Monvel, L.: [l]Boundary problems for pseudo-differential operators. Acta Math. 126, 11-51 (1971). Brailovsky, M. and L.S. Frank: [l] Stationavy flow by solar wind around a multipole. Report No. 77, 1-35, Institute of Space Research of USSR Academy of Sciences, 1971. Brauner, C. and B. Nicolaenko: [l] Singular perturbations and free boundary problems. Computing methods in Applied Sciences and Engineering, North-Holland, Amsterdam (1980). BrCzis, H.: [l]Analyse fonctionelle. Masson. Paris, 1983. Burgers, T.M.: [l] The nonlinear diffusion equation: asymptotic solution and statistical problem. North-Dordrecht-Holland, Amsterdam] 1974. Calderbn, A.P. and A. Zygmund: [l]On the existence of certain singular integrals. Acta Math. 88, 85-139 (1952). - [2] Singular integral operators and differential equations. Amer.
J . Math.
79, 901-921 (1957). Carleman, T.: [l] PropriCtCs asymptotiques des fonctions fondamentales des membranes vibrantes. C.R. Cong. des Math. Scand. Stockholm 1934, 34-44, Lund (1935). Chazarain, J. and A. Piriou: [l]Introduction
la thCorie des Cquations aux
dkrivCes partielles linkaires. Gauthier-Villars, 1981. Ciarlet, P.G. and H. le Dret: [l]Justification of the boundary conditions of a clamped plate by an asymptotic analysis. Asympt. Anal., 2: 2, 257-278, (1989). Ciarlet, P.G. and P. Rabier: [l] Les equations de von KBrmBn. Springer-Verlag, Berlin-Heidelberg-New York, 1980. Cole, J.D.: [l] Perturbation methods in applied mathematics. Blaisdell Publ., New York 1968. Comstock, C.: [l] Singular perturbations of elliptic equations. SIAM J . Appl. Math., 20, 491-502 (1971). Courant, R.: [l]Partial differential equations. Interscience, New York, 1962.
-
[2] Variational methods for the solution of problems of equilibrium and vibrations. Bull. Amer. Math. SOC.49, 1-23 (1943).
536
Bibliography
Courant, R., K.O. Friedrichs and H. Lewy: [I] Uber die partiellen Differentialgleichungen der Mathematischen Physik. Math. Ann., 100, 32-74 (1928). Daletski, Yu.L.: [l] An asymptotic method for the solution of some differential equations with oscillatory coefficients. Dokl. Acad. Nauk USSR, 143, no. 5, 1026-1029 (1962) (Russian). Daletski, Yu.L. and M.G. Krein: [l]Stability of solutions of differential equations in Banach spaces. “Nauka”, Moscow 1970 (Russian). Davis, R.B.: [l]Asymptotic solutions of the first boundary value problem for a fourth-order elliptic partial differential equation. J. Rat. Mech. and Anal., 5, 605-620 (1956). De Bruijn, N.G.: [l]Asymptotic methods in analysis. North-Holland, Amsterdam, 1958. De Giorgi, E. and T. Franzoni: [l] Su un tip0 di convergenza variazionale. Rend. Sem. Mat. Brescia, 3, 63-101 (1979). Demidov, A.: [l] Elliptic pseudo-differential boundary value problems with a small parameter in the coefficients of the leading operator. Math. USSR
Sb., 20, 439-463 (1973). Dieudonnk, J.: [l]Elkments d’Analyse, Gauthier-Villars, 1978. - [2] Foundation of modern analysis. Academic Press, New York, 1964.
Dixmier, J.: [l]Les C*-algBbres et leurs reprksentations. Gauthier-Villars, Paris 1964. Dobrokhotov, S.Yu. and V.P. Maslov: [l] Finite gap almost periodic solutions in asymptotic expansions. Anal. and Numer. approaches to Asympt. Problems (0. Axelsson, L.S. Frank, A. van der Sluis, eds.), 1-15, NorthHolland, Amsterdam-New York-Oxford (1981). Duistermaat, J.J.: [l] Fourier integral operators, Lecture Notes, Courant Inst. of Math. Sci., New York, 1973. Duistermaat, J.J. and L. Hormander: [l]Fourier integral operators. 11. Acata Math. 128, 183-269 (1972). Dunford, N. and J.T. Schwarz; [l]Linear operators 1-11. Interscience Publishers, New York-Lon don, 1958-1963. Eckhaus, W.: [l]Boundary layers in linear elliptic singular perturbation prob-
Bibliography
537
lems. SIAM Rev. 14, 225-270 (1972). Eckhaus, W. and E.M. De Jager: [l]Asymptotic solutions of singular perturbation problems for linear differential equations of elliptic type. Arch. Rat. Mech. and Anal., 23, 26-86 (1966). Egorov, Ju.V.: [l]The canonical transformations of pseudedifferential operators. Uspekhi Mat. Nauk 24 : 5, 235-236 (1969). Erdklyi, A.: [l]Asymptotic representation of Fourier integrals and the method of stationary phase. SIAM J . 3 : 1, 17-27 (1955). - [2] General asymptotic expansions of Laplace integrals. Arch. Rat. Mech.
and Anal., 7 : 1, 1-20 (1961). Eskin, G.I.: [l] Boundary value problems for elliptic pseudo-differential equations. Amer. Marth. SOC.Transl. of Math. Monographs 52, Providence,
R.I., 1981. Evgrafov, M.A. and M.V. Fedorjuk: [l] Asymptotics for the solutions of the equation
w N ( z ) - p ( z ,X)w(z) = 0 for X Nauk, 21 : 1, 3-50 (1966).
+ 00
in the complex domain. Uspekhi Mat.
Fedorjuk, M.V.: [l]The saddle point method, “Nauka”, Moscow, 1977. - [a] The stationary phase method for multidimensional integrals. Journ.
Vichisl. Mat. and Mat. Fisica 4 : 4, 671-681 (1964) (Russian).
- [3] The stationary phase method and pseudo-differential operators. Uspekhi Mat. Nauk 26 : 1, 67-122 (1971) (Russian). Fefferman, C. and D.H. Phong: [l]The uncertainty principle and sharp Gdrding inequalities. Comm. Pure Appl. Math. 34, 285-331 (1981). Fefferman, C. and F.M. Stein: [l]HP spaces of several variables. Acta Math., 129, 137-193 (1972). Fife, P.: [l]Singularly perturbed elliptic boundary value problems I: Poisson kernels and potential theory. Annali di Mat. Pura Appl. Ser. 4, 90, 99-148 (1971). - [2] Dynamics of internal layers and diffusive interfaces. SIAM, Philadelphia
1988. Flatto, L. and N. Levinson: [l]Periodic solutions of singularly perturbed sys-
Bibliography
538
tems. J . Rat. Mech. and Analysis, 4, 943-950 (1955). Fleming, W.H.: [l] Stochastic control for small noise intensities. SIAM J . Control, 9, 472-517 (1971). Fok, V.A.: [l] Problems of diffraction and propagation of electremagnetic waves. “Sovietsk. Radio”, 1970. Frank, L.S.: [l]On Cauchy’s problem for some equations with variable coefficients which are not of Kovalevski’s type. Vestnik Moscov. Univ., 5, 13-23 (1963) (Russian, English summary).
- [2] Difference methods for the solution of the ill-posed Cauchy problem systems of the first order. Vestnik Moscov. Univ., 21,3-11 (1966) (Russian, English summary).
- [3] The difference schemes of high order of accuracy for systems of first order. Vestnik Moscov. Univ., 21, 26-33 (1966) (Russian, English summary). - [4] Difference operators in convolutions. Dokl. Akad. Nauk SSSR, 181, no.
2, 286-289 (1968); English transl. Soviet Math. Dokl., 9, 831-840 (1968).
- [5] Hyperbolic difference operators. Dokl. Akad. Nauk SSSR, 189, no. 3, 486-488 (1968); English transl.: Soviet Math. Dokl., 10, no. 6, 1449-1451 (1969). - [6] Coercive boundary value problems for difference operators. DokLAkad.
Nauk SSSR, 192, 42-46 (1969); English transl.: Soviet Math. Dokl., 11, 591-594 (1970). - [7] The asymptotic behavior of rapidly oscillating solutions of hyperbolic
difference equations. Uspekhi Mat. Nauk, 25, 252-254 (1970) (Russian).
[8] Spaces of network functions. Mat. Sb., 86(128) : 2(10), 187-233 (1971); English transl.: Math. USSR Sbornik, 15, 183-226 (1971). - [9] An algebra of difference operators. Institute of Space Research of USSR, Acadmy of Sciences, Moscow, 1971. -
- [lo] Algkbre des opCrateurs aux diffCrence finies. Israel J. Math., 13, 24-55
(1972). - [ll]Fonctions de mailles et thdorie elliptique des opkrateurs aux diffkrences
finies. Seminaire Goulaouic-Schwartz (1972-1973): equations aux dCrivkes partielles et analyse fonctionnelle: exposks Nos. 5-6, Paris, Centre de
5 39
Bibliography
mathhmatique, Ecole Polytechnique, 1-15 (1973). - [la] Sur la fonction spectrale d’un ophrateur aux diffhrences finies elliptique.
C.R. Acad. Sci. Paris, 276, 1521-1523 (1976). - [13] OpCrateurs elliptiques aux diffkrences finies.
J . Math. Anal. Appl.,
45, 260-273 (1974). - [14] Comportement asymptotique et support singulier de la fonction de
Green des opkrateurs diffhrence - diffkrentiels stables. C.R. Acad. Sci. Paris, 278, 1051-1054 (1974). - [15] Problbmes aux limites coercifs avec un petit parambtre. C.R. Acad.
Sci. Paris, 282, 1109-1111 (1976). - [16] Perturbation singulibres et difftkences finies. C.R. Acad. Sci. Paris,
283, 859-862 (1976). - [17] Perturbazioni singolari alle differenze finite.
Ren. Sem. Univ. di
Torino, 35, 129-142 (1976/77). - [l8] General boundary value problems for ordinary differential equations
with a small parameter. Annali Mat. Pura Appl., 114, 26-67 (1977). - [19] Perturbzioni Singolari Ellittiche. Ren. Sem. Mat. Milanepavia, 47,
135-163 (1977).
- [20] Factorisation for difference operators. J . Math. Anal. Appl., 62, 170-185 (1978). -
[all Coercive singular perturbations:
stability and convergence. Numeri-
cal analysis of singular perturbation problems, 181-198, Academic Press, London (1979). - [22] Coercive singular perturbations. I, A priori estimates. Annali Mat.
Pura Appl., 119, 41-113 (1979). - [23] Elliptic difference singular perturbations. Annali Math. Pura Appl.,
122, 315-363 (1979). - [24] InCgalitC de Gkding et perturbations singuli6res elliptiques aux diffh-
rences finies. C.R. Acad. Sci. Paris, 296, 93-96 (1983).
- [25] Perturbation singulibres coercives IV: probkme de valeurs propres. C.R. Adac. Sci. Paris, 301, 69-72 (1985).
540
Bibliography
- [26] Remarque sur le phCnombne de bifurcation pour les perturbations sin-
gulibres coercives quasi-lin6aires. C.R. Acad. Sci. Paris, 302, 17-20 (1986). - [27] On a singular perturbation in the kinetic theory of enzymes and the
corresponding nonlinear time optimal control. Analysis and optimization of systems (A. Bensoussan and J.-L. Lions, eds.), 653-667, Springer-Verlag,
Berlin-Heidelberg-New York-Tokyo (1986). - [28] Perturbations singulikres coercives: m6thode de factorisation et appli-
cations. Equations aux d6rivCes partielles, expos6 110.18, Ecole Polytechnique, Palaiseau, 1986-87. - [29] Coercive singular perturbations: eigenvalue problems and bifurcation
phenomena. Annali Mat. Pura Appl., (IV), CXLVIII, 367-395 (1987). - [30] Sharp error estimates in the reduction procedure for the coercive sin-
gular perturbations. Asympt. Analysis, 3 : 1, 37-42 (1990). - [31] Toeplitz operators and the theory of one parameter families of differ-
ence operators. Rep.9008, Univ. Nijmegen, February 1990. Frank, L.S. and L.A. Chudov: [l] Difference methods for solving the incorrect Cauchy problem. Numerical methods in Gas Dynamics, Moscow, nr. 4, 3-27 (1965) (Russian). Frank, L.S. and E. van Groesen: [l] Singular perturbations of an elliptic operator with discontinuous nonlinearity. Analytical and numerical approaches to asymptotic problems in analysis, North-Holland, Amsterdam, 289-303 (1981). Frank, L.S. and J.J. Heijstek; [l]On the continuation by zero in Sobolev spaces with a small parameter. Asympt. Analysis, 1 : 1, 51-60 (1988). - [2] On the reduction of coercive singular perturbations to regular per-
turbations. Operator Theory: Advances and Applications, 41, 279-297, Birkhauser-Verlag, Base1 (1989). Frank, L.S. and H. Norde: [l]On a singular perturbation in the linear soliton theory. Rep. Univ. Nijmegen, 1990 (to appear in Asympt. Anal.) Frank, L.S. and W.D. Wendt: [l] Isomorphic elliptic singular perturbations. Rep. no. 8003, Univ. Nijmegen, 1980. - [2] Coercive singular perturbations: asymptotics and reduction to regularly
Bibliography
541
perturbed boundary value problems. Analytical and numerical approaches to asymptotic problems in analysis, 305-318, North-Holland, Amsterdam
(1981). -
[3] Coercive singular perturbations: reduction to regular perturbations and applications. Comm. Partial Diff. Equations, 7, 469-535 (1982).
-
[4] Coercive singular perturbations: reduction and convergence, Theory and applications of singular perturbations. Lecture notes in mathemat-
ics,942, 54-64, Berlin-Heidelberg-New York (1982). -
[5] Coercive singular perturbations 11: reduction and convergence. J. Math. Anal. Appl., 88, 464504 (1982).
-
[6] Sur une perturbation singulikre en thCorie des enzymes. C.R. Acad. Sci. Paris, 294, 741-744 (1982).
- [7] Solutions asymptotiques pour une classe de perturbations singulikres elliptiques semilinkaires. C.R. Acad. Sci. Paris, 295, 451-454 (1982). -
[8] Sur une perturbartion singulikre parabolique en thCorie cinCtique des enzymes. C.R. Acad. Sci. Paris, 295, 731-734 (1982).
[9] On an elliptic operator with discontinuous nonlinearity. J. Diff. Equations, 54, 1-18 (1984). - [lo] Coercive singular perturbations 111: Wiener-Hopf operators. J. d’Analyse Math., 43, 88-135 (1983/84). - [ll] On a singular perturbation in the kinetic theory of enzymes. Delft -
Technische Hogeschool, Delft Progress, 53-66 (1985).
[12] Elliptic and parabolic singular perturbations in the kinetic theory of enzymes. Nonlinear partial differential equations and their applications: Collkge de France seminar, Vol. VII, 84141, (H. BrCzis and J.L. Lions (eds.)), Pitman, Boston 1985. - [13] Elliptic and parabolic singular perturbations in the kinetic theory of enzymes. Annali Mat. Pura Appl., 144, 261-301 (1986).
-
Freidlin, M.I. and A.T. Wentzel; [l] Random perturbations of dynamical systems. Springer-Verlag, Berlin-Heidelberg-New York, 1984. Friedman, A.: [l]Singular perturbations for the Cauchy problem and for boundary value problems. J. Diff. Equations, 5, 226261 (1969).
542
Bibliography
Friedrichs,K.O.: [l] Asymptotic phenomena in mathematical physics. Bull. Amer. Math. SOC.,61 : 6, 79-94 (1955).
- [2] Pseudo-differential operators: an introduction. Courant Inst. of Math. Sci., Lecture Notes, New York 1970. Friedrichs, K.O. and J . Stoker: [l] The nonlineair boundary value problem of the buckled plate. Amer. J. Math. 63, 839-888 (1941). Galpern, S.A.: [l] Cauchy problem for general systems of linear partial differential equations. Dokl. Akad. Nauk SSSR (N.S.), 119, 640-643 (1958). Gkding, L. [l] Linear hyperbolic partial differential equations with constant coefficients. Acta Math. 85, 1-62 (1951).
- [2] Dirichlet’s problem for linear elliptic partial differential equations.Math. Scand. 1, 55-72 (1953). - [3] On the asymptotic distribution of the eigenvalues and eigenfunctions of
elliptic differential operators. Math. Scand. 1, 237-255 (1953).
- [4] Transformation de Fourier des distributions homogknes. Bull. SOC. Math. France, 89, 381-428 (1961). Gelfand, I.M. and G.E. Shilov: [l]Generalized functions, vol. 1-111, “Fizmatgis”, 1958. English transl.: Academic Press, New York, 1964. Godunov, S.K.: [l]A difference method for numerical calculation of discontinuous solutions of the equation of hydrodynamics. Mat. Sb. (N.S.) 47 (89), 271-306 (1959) (Russian). Gohberg, I. and I.A. Feldman: [l] Convolution equations and projection methods for their solution. “Nauka”, Moscow 1971 (Russian). Gohberg, I. and M.G. Krein: [l]Systems of integral equations on a half-line with kernels depending on the difference of arguments. Uspekhi Mat. Nauk, 13 : 2, 3-72 (1958). - [2] Defect numbers, root numbers and indecies of linear operators. Amer.
Math. SOC.transl. Ser. 2, 13, 185-264 (1960). Goldenveizer, A.A.: [l] Thin elastic shells. Gos. Izdat. Tehn.-Ter.
Lit.,
MOSCOW, 1953 (Russian). Grabmuller, H.: [l]Singular perturbation techiniques applied to integro-differential equations. Pitman, San Francisco, 1978.
Bib Ziography
543
Gradshtein, I.S.: [l] Linear equations with variable coefficients and small parameters in the highest derivatives. Mat. Sb. (N.S.) 27 (69), 47-68 (1950) (Russian). Gradshtein, I S . and I.M. Ryzhik; [l] Table of integrals, series and products. Academic Press, 1980. Greenlee, W.M.: [l]Singular perturbations of eigenvalues. Arch. Rat. Mech. and Anal., 34, 143-164 (1969). Groen, P.P.N.: [l]Singular perturbations of spectra. Asymptotic Analysis (F. Verhuls, ed.), Springer-Verlag, Berlin-Heidelberg-New York (1979). Grubb, G.: [l] Functional calculus of pseudo-differential boundary problems. Progress in Mathematics, 65, Birkhauser, Boston 1986. Hadamard, J.: [l]Le problkme de Cauchy et les Cquations aux dCrivCes partielles 1inCaires hyperboliques. Paris 1932. Handelman, G.H., J.B. Keller and R.E. O’Malley Jr.: [l]Loss of boundary conditions in the asymptotic solution of linear ordinary differential equations. I. Eigenvalue problems. Comm. Pure Appl. Math. XXI, 243-261 (1968). Harris Jr., W.A.: [l]Singular perturbations of eigenvalue problems. Arch. Rat. Mech. and Anal., 7, 224-241 (1961). Harten, A. van: [l] Nonlinear singular perturbation problems: proof of correctness of a formal approximation based on a contruction principle in a Banach space. J . Math. Anal. Appl. 65, no. 1, 126-168 (1978). Harten, A., J.M. Hyman and P.D. Lax: 111 On finite difference approximations and entropy conditions for shocks. Comm. Pure Appl. Math. 29, 297-322, (1976). Heijstek, J.J.: [l]On a singular perturbation in the theory of elastic rods. Ph.D. Thesis, Univ. of Nijmegen, 1989. Helffer, B. and D. Robert: [l]Calcul fonctionel par la transformCe de Mellin et applications. J. Funct. Anal. 53, (1983). - [2] Riesz means of bound states and semiclassical limit connected with a
Lieb-Thirring’s conjecture. Asympt. Anal. 3 : 1, (1990). Helffer, B. and J. Sjostrand: [l]Puits multiples en limite semi-classique. Ann. Inst. H. PoincarC (section physique), 42, no. 2, 127-212 (1985).
5 44
Bibliography
Herglotz, G.: [l]Uber die Integration h e a r e r partieller Differentialgleichungen mit konstanten Koeffizienten 1-111. Berichte Sachs. Akad. d. Wiss. 78, 93-126, 287-318 (1926); 80, 69-114 (1928). Hirth, J.-P. and J . Lothe: [l]Theory of dislocations. Mc Graw-Hill, New York 1967. Hoppensteadt, F.: [l] On quasi-linear parabolic equations with a small parameter. Comm. Pure Appl. Math., XXIV, 17-38 (1971). IJormander, L.: [I] Estimates for translation invariant operators in LP spaces. Acta Math. 104, 93-140 (1960).
- [2] Pseudedifferential operators. Comm. Pure Appl. Math. 18, 501-517 (1965).
- [3] The spectral function of an elliptic operator. Acta Math. 121, 193-218 (1968).
- [4] The analysis of linear partial differential operators. Vol. I-IV, SpringerVerlag, Berlin-Beidelberg-New York, 1985. Huet, D.: [l] Singular perturbations of elliptic problems. Annali di Mat. Pura Appl. XCV, 77-114 (1973). - [2] Dkcomposition spectrale et opkrateurs lint5aires. Press Universitaire de
France, 1976. - [3] StabilitC et convergence dans Ies problkmes de perturbation singulikre.
Asympt. Anal. 2 : 1, 5-20 (1989). - [4] Remarque sur un thCor5me d’hgmon et applications b quelques problk-
mes de perturbation singulikre. Boll. Unione Mat. Ital., 3, XXI, 219-227 (1966). Il’yn, A.M.: [l]Difference scheme for the differential equation whose highest derivative is affected by the presence of a small parameter. Math. Notices, 6 : 2, 237-248 (196s) (Rcssian).
- [2] Matching of asymptotic expansions for solutions of boundary value problems. ”Nauka”, Moscow, 1989 (Russian). John, F.: [l] Plane waves and spherical means applied t o partial differential equations. New York 1955. - [2] A priori estimates, geometric effects and asymptotic behavior. Bull.
Bibliography
545
Amer. Math. SOC.,81 : 6, 1013-1023 (1975). Kamin (Kamenomostskaya), S.L.: [l]The first boundary problem for equations of elliptic type with a small parameter in the highest derivatives. Izv. Akad. Nauk SSSR, Ser. Mat., 19, 345-360 (1955) (Russian).
- [2] Perturbation elliptique d’un opCrateur du premier ordre avec un point singulier. C.R. Acad. Sci. Paris, 285, 677-680 (1977). Kaplun, S.: [l]Fluid mechanics and singular perturbations. Academic Press, New York 1967. KBrmAn, von T.: [l] Festigkeitsprobleme im Maschinenbau. “Encyklopadie der Mathematischen Wissenschaften”, vol. IV-4, C, 14-15, Leipzig (1907). Kato, T.: [l]Perturbation theory for linear operators. Springer-Verlag, BerlinHeidelberg-New York 1976. Keller, J.: [l]Corrected Bohr-Sommerfeld quantum conditions for nonseparable systems. Ann. Physics, 4, 100-188 (1958). Keller, J.B. and R.E. O’Malley Jr.: [l] Loss of boundary conditions in the asymptotic solution of linear ordinary differential equations. 11. Boundary value problems. Comm. Pure Appl. Math., XXI, 263-270 (1968). ICifer, J u . I.: [l] On the spectral stability of invariant tori under small random perturbations of dynamical systems. Dokl. Akad. Nauk SSSR, 235, 512515 (1977); Sov. Math. Dokl. 18, 952-956 (1977). Kohn, J.J. and L. Nirenberg: [l]On the algebra of pseudo-differential operators. Comm. Pure Appl. Math. 18, 269-305 (1965). Kolmogorov, A.N.: [l] Zufdlige Bewegungen. Ann. of Math., 35, 116-117 (1934). Korteweg, D.J. and G. de Vries: [l]On the change of form of long waves acting in a rectangular channel and on a new type of long stationary waves. Philos. Mag., 39, 422-443 (1895). Krein, M.G.: [l]Integral equations on a half-line with kernels, depending on the difference of arguments. Uspekhi Mat. Nauk, 13 : 5, 3-120 (1958). Krein, S.G.: [l] Linear differential equations in Banach spaces.
“Nauka”,
Moscow 1967 (Russian). Kucherenko, V.V.: [l] Asymptotic solutions of equations with complex char-
5 46
Bibliography
acteristics. Mat. Sb. 95, 164-213 (1974); Math. USSR Sbornik 24 : 2,
159-207 (1974). Kuman-go, H.: [l]Algebras of pseudodifferential operators. J. Fac. Sci. Tokyo
17, 31-50 (1970). Ladyzhenskaya, O.A.: [l] On equations with a small parameter in the highest order derivatives in linear partial differential equations. Vestnik Leningrad Univ. Ser. Mat. Mech. Astr., 12 : 7, 104-120 (1957) (Russian, English summary). Landau, L.D. and E.M. Lifschitz: [l] Elasticity theory. “Nauka”, Moscow 1965 (Russian); ThCorie de 1’ClasticitC. “Mir” , Moscou 1967.
- [2] Electrodynamics in continuous media. Gosizdat, Moscow, 1957 (Russian). Lang, S.: [I] Differential manifolds. Addison Wesley Publ., 1972. Lax, P.D.: [l] Asymptotic solutions of oscillatory initial value problems. Duke Math. J. 24, 627-646 (1957).
[2] Hyperbolic systems of conservation laws. 11. Comm. Pure Appl. Math., X,537-566 (1957). - [3] Shock waves and entropy. Contributions t o nonlineair functional analysis, Academic press, New York, (1971). -
Lax, P.D. and C.D. Lawrence; [l]The small dispersion limit of the Kortweg-de Vries equation. I. Comm. Pure Appl. Math., XXXVI, 253-290 (1983). Lax, P.D. and L. Nirenberg: [l]On stability of difference schemes: a sharp form of Giirding’s inequality. Comm. Pure Appl. Math., 19, 473-492 (1966). Lax, P.D. and R.D. Richtmyer: [l]Survey of the stability of linear finite difference equations. Comm. Pure Appl. Math. 9, 267-293 (1956). Lax, P.D. and B. Wendroff: [l] Systems of conservation laws. Comm. Pure Appl. Math. 13, 363-375 (1960). Leray, J.: [l] Analyse lagrangienne et mhcanique quantique. Univ. L. Pasteur, Strasbourg 1978; English transl.: MIT Press, Cambridge, Mass., 1982. Levin, J.J. and N. Levinson: [l]Singular perturbations of nonlineair systems of differential equations and an associated boundary layer equation. J. Rat. Mech. and Analysis, 3, 247-270 (1954).
Bibliography
5 47
Levinson, N.: [l]The first boundary value problem for EAU+ A(z,y)uz + B ( z ,y)u, C(z, y)u = D ( z ,y) for small E . Ann. of Math. (2) 51, 428-445
+
(1950). Lions, J.-L.: [l]Perturbations singulikres dans les probkmes aux limites et en control optimal. Springer-Verlag, Berlin-Heidelberg-New York, 1973.
- [2] ContrblabilitC exacte et homogCnCisation. Asympt. Anal., 1 : 1, 3-11 (1988). Lions, J.-L. and E. Magenes: [l]Problkmes aux limites non homogcnes et applications. 1-111. Dunod, Paris 1968-1970; English transl.: Springer, BerlinHeidelberg-New York, 1972-1973. Loomis, L.H. and S. Sternberg: [l]Advanced calculus. Addison-Wesley, %ading-London-Amsterdam-Don Mills, 1980. Lopatinski, Ya.B.: [l] On a method of reducintg boundary problems for a system of differential equations of elliptic type to regular integral equations. Ukrain. Math. Z. 5, 123-151 (1953). Amer. Math. SOC.Transl. (2) 89, 149-183 (1970). Ludwig, D.: [l]Exact and asymptotic solutions of the Cauchy problem. Comm. Pure Appl. Math. 13, 473-508 (1960). Lyusternik, L.A.: [l] On difference approximations of the Laplace operator. Uspekhi Mat. Nauk 9, No. 2 (60), 3-66 (1954) (Russian). Lyusternik, L.A. and V.I. Sobolev: [l]Elements of functional analysis. “Nauka”, MOSCOW, 1965 (Russian). Markovich, P.A.: [l]The stationary semiconductor devices equations. SpringerVerlag, Wien-New York, 1985. Maslov, V.P.: [l]Theory of perturbations and asymptotic methods. MOSCOV.,
Gos. Univ., Moscow 1965 (Russian); Thkorie des perturbations et m & h e des asymptotiques. Dunod, Paris 1968; English transl.: Springer-Verlag, 181-182, 1972. - [2] Operational methods.
“Nauka”,Moscow 1973 (Russian); MCthodes
opkrationnelles. “Mir” , Moscou 1987.
- [3] Complex WKB method for nonlinear equations “Nauka” , Moscow 1977 (Russian).
548
Bibliography
Maslov, V.P. and M.V. Fedorjuk: [l] Semi-classical approximations for equations of quantum mechanics. “Nauka”, Moscow 1976 (Russian; English transl.: D. Reidel Publ., vol. 7, Dordrecht 1981). Matkowsky, B.J. and E.L. Reiss; [l] On the asymptotic theory of dissipative wave motion. Arch. Rat. Mech. and Anal., 43(3), 194212 (1971). Mauss, J.: [l] Approximation asymptotique uniforme de la solution d’un probkme de perturbation singulihe de type elliptique. J. de MCcanique, 8,
373-391 (1969). Melin, A. and J. Sjostrand: [l] Fourier integral operators with complex phase functions, Springer Lecture Notes no. 459, 120-223 (1974). - [2] Fourier integral operators with complex phase and application to an
interior boundary problem. Comm. Partial Diff. Eqns., 1, no. 4 , 313-400
(1976). Michelson, D.: [l] Stability theory of difference approximations for multidimensional initial boundary value problems. Math. Comp. 40, 1-45
(1983). Mikhlin, S.G.: [l] On the multipliers of Fourier integrals. Dokl. Akad. Nauk
SSSR 109, 701-703 (1956) (Russian). -
[2] Multidimensional singular integrals and integral equations. Pergamon Press, 1965.
Mimura, M., M. Tabata and Y. Hosono: [l] Multiple solutions of twepoint boundary value problems of Neumann type with a small parameter. SIAM J. Math. Anal., 11, 613-631 (1980). Miranker, W.L.: [l] The computational theory of stiff differential equations. Rome 1975. Mises, von R.: [l] Grenzschichte in der Theorie der gewonlichen Differentialgleichungen. Acta Univ. Szeged., Sect. Sci. Math., 12, 29-34 (1950). Mityagin, B.S.: [l] Differential equations with a small parameter in the Banach space. Izv. Azerbaijan Acad. of Sciences, 1, 23-28 (1961) (Russian). Mock, M.S.: [l] Analysis of mathematical models of semiconductor devices.
Boole Press, Dublin 1983. Morawetz, C. and D. Ludwig: [l]An inequality for reduced wave operator and
Bib liogaphy
549
the justification of geometrical optics. Comm. Pure Appl. Math. 21 : 2,
187-203 (1968). Morgenstern, D.: [l] Singulare storungstheorie partieller Differentialgleichungen. J. Rat. Mech. and Anal., 5, 203-216 (1956). Moser, J .: [l]Singular perturbation of eigenvalue problems for linear differential equations of even order. Comm. Pure Appl. Math., vol. VIII, 251-278
(1955). Muskhelishvili, N.I.: [l] Singular integral equations. Third ed. “Nauka” 1968 (Russian); English transl. of the first ed.: P. Noordhoff N.V., GroningenHolland, 1953.
+
f(z,y,y’,A) = 0 0. Proc. Phys. Math. SOC.Japan, 21, 529-534 (1939). Naimark, M.A.: [l] Linear differential operators. “Nauka”, Moscow 1969 (Russian).
Nagumo, M.: [l] Uber das Verhalten des Integrals von Ay” fur A
Nayfeh, A.H.: [l] Introduction t o perturbation techniques. Wiley Interscience Publ., New York-Toronto 1981. Nirenberg, L.: [l] Remarks on strongly elliptic partial differential equations. Comm. Pure Appl. Math. 8, 648-675 (1955). Nishiura, Y. and H. Fujii: [l] Stability of singularly perturbed solutions to systems of reaction-diffusion equations. SIAM J . Math. Anl. 18, 1726
1770 (1987). Nother, F.: [l] Eine Klasse singularer Integralgleichungen. Math. Ann., 82,
42-63 (1920). Oleinik, O.A.: [l] On equations of elliptic type with small parameter. Mat. Sb. (N.S.), 31 (73), 104117 (1952) (Russian).
- [2] Mathematical problems of boundary layer theory. Uspekhi Mat. Nauk, 23, 3-65 (1968); English transl.: Univ. of Minnesota Lecture Notes, 1969. Oleinik, O.A. and A.I. Zhizhina: [l] On the boundary value problem for the equation EY” = f(z, y, y’) for small
E.
Math. Sb., 31 (73), no. 3, 707-717
(1952) (Russian). Olver, F.W.J.: [l] Asymptotics and special functions. Academic Press, New York-London 1974.
Bibliography
550
Osher, S.: [l] Numerical solution of singular perturbation problems and hyperbolic systems of convervation laws. Anal. and Numer. Approaches to Asymptotic Problems, North-Holland, Amsterdam-New York, 179-204 (1981). Ott, H.: [l]Die Sattelpunktsmethode in der Umgebung eines Pols. Ann. Phys. 43, 393-394 (1943). Palais, R.S.: [l] Natural operations on differential forms. Trans. Amer. Math. SOC.,92, 125-141 (1959). Pazy, A.: [l] Asymptotic expansions of solutions of ordinary differential equations in Hilbert space. Arch. Rat. Mech. and Anal. 24, 193-218 (1967). Peetre, J.: [l] Another approach to elliptic boundary problems. C o r n . Pure Appl. Math. 14, (1961).
- [2] Mixed problems for higher order elliptic equations in two variables. I. Ann. Scuola Norm. Sup. Pisa, 15, 337-353 (1961). Petrovsky, I.G.: [l] Uber das Cauchysche Problem fur ein system h e a r e r partieller Differentialgleichungen im Gebiete der nichtanalytischen Funktie nen. Vestnik Univ. Moscow SCr. Int. 1, No. 7, 1-74 (1938).
- [2] Sur I’analyticitC des solutions des syst5mes d’kquations diffCrentielles. Mat. Sb. 5 (47), 3-70 (1939). Poincark, H.: [l] Les mCthodes nouvelles de la mCcanique celeste. GauthierVillars , 1892- 1893. - [2] LeGon de mCcanique cCleste. Gauthier-Villars, Paris, 1905-1910.
Raviart, P.A.: [l]Sur la rksolution numirique de 1’Qquationu~+uu,--E((u,(u,),
= 0. J. Diff. Equations, 8 (l),56-94 (1970). Rayleigh (Lord Rayleigh): [l]On waves propagated along the plane surface of an elastic solid. Proc. London Math. SOC.17, 4-11 (1885). - [2] Theory of sound.
I. London, 1937.
Rellich, F.: [l]Perturbation theory of eigenvalue problems. Gordon and Breach, New York 1969. Rempel, S. and B.-W. Schulze: [l]Index theory of elliptic boundary problems, Akademie-Verlag, Berlin 1982. Richardson, L.: [l]Weather prediction by numerical process. Cambridge 1922.
551
BiblioTaphy
Richtmyer, R.D. and K.W. Morton: [l] Difference methods for initial value problems. Interscience, New York 1967. Riesz, F. and B.Sz.-Nagy: [l]LeCons d’analyse fonctionelle. Acad. des sciences de Hongrie, Budapest, 1953. Roseau, M.: [l] Asymptotic wave theory. North-Holland, Amsterdam-New York-Oxford, 1976. Rothe, E.: [l]Uber asymptotische Entwiklungen bei Randwertaufgaben elliptischer partieller differentialgleichungen. Math. Ann., 108, 578-594 (1933). Rudin, W.: [I] Functional analysis. Tata Mc Graw-Hill, New Delhi, 1973. Saito, Y.: [l]On the asymptotic behavior of the solutions of the Schrodinger equation (-A+Q(y)-k2)V
= F . Osaka J . Math., 14, 11-35 (1977).
Sanchez-Palencia, E.: [l]Non-homogeneous media and vibration theory. Springer-Verlag, Berlin-Heidelberg-New York, 1980. Schlichting, H.: [l]Boundary layer theory. Mc Graw Hill, 1968. Schuss, Z.: [l] Asymptotic expansions for parabolic systems. J . Math. Anal. and Appl., 44, 136-159 (1973). Schur, J .: [l] Bemerkungen zur Theorie der beschrakten Bilinearformen mit unendlich vielen Verinderlichen. J . fur reine und angewandte Mathematik, Bd. 140, 1-28 (1911). Schwartz, L.: [l]Thdorie des distributions, 1-2. Hermann, Paris 1950-1951). - [2] Thdorie des noyaux. Proc. Int. Congr. Math.,
I, 220-230, Cambrdige
(1950). - [3] Transformation de Laplace des distributtions. Comm. SCm. Math. Univ. Lund, Tome suppl., 196-206 (1952). Seeley, R.T.: [l] Extension of Cw-functions defined on a half-space. Proc. Amer. Math. SOC.15, 625-626 (1964).
- [2] Refinement of the functional calculus of Calderon and Zygmund. Koninkl. Nederl. Akad. v. Wet. Proceedings, Serf. A, 68, 521-532 (1965).
- [3] Complex powers of an elliptic operator. Proc. Symp. Pure Math., 10, 288-307 (1968). Serrin, J.: [l]Recent developments in the mathematical aspects of boundary layer theory. Int. J. Engng. Sci. 9, 233-240 (1971).
Bibliography
552
Shamir, E.: [l] Mixed boundary value problems for elliptic equations in the plane, the &-theory. Ann. Scuola Norm. Super. Pisa, 17, 117-139 (1963).
- [2] Elliptic systems of singular integral operators I. Trans. Amer. Math. SOC.,127, 107-124 (1967). Shapiro, Z.Ya.: [l]On general boundary problems for equations of elliptic type. Izv. Akad. Nauk SSSR, 17, 539-562 (1953) (Russian). Shilov, G.E.: [l]Mathematical analysis. I. Fizmatgiz, Moscow, 1961 (Russian). -
[2] Mathematical analysis. 11. “Nauka”, MOSCOW,1965 (Russian).
- [3] Generalized functions and partial differential equations. Gordon and Breach, Science Publ. New York, 1968 (Russian original, 1965). Sirovich, L.: [l] Techniques of asymptotic analysis. Springer-Verlag, BerlinHeidelberg-New York, 1971. Smith, R.A.: [l] Semiconductors, Cambridge University Press 1978. Smoller, J.A.: [l]Recent developments in the mathematical aspects of boundary layer theory. Int. J . Engng. Sci. 9, 233-240 (1971). Sobolev, S.L.: [l] MQthodenouvelle A rQsoudrele probkme de Cauchy pour les dquations linkaires hyperboliques normales. Mat. Sb. 1 (43), 39-72 (1936). -
[2] Some applications of functional analysis in mathematical physics.Leningrad, 1950 (Russian); English transl.: Amer. Math. SOC.Transl., Providence, R.I. 1963.
- [3] Cauchy’s problem for a special case of systems not belonging to the Kowalevsky type. Dokl. Akad. Nauk SSSR, 82, 205-208 (1952) (Russian). Sobolevskii, P.E.: [l] On second order equations with a small parameter multiplying the highest order derivative. Uspekhi Mat. Nauk, 19 : 6, 217-219
(1964) (Russian). Sommerfeld, A.: [l] Optics. Lectures on theoretical physics. Academic Press, New York, 1969. Srubschchik, L.S. and V.1 Yudovich: [l] The asymptotic integration of the system of equations for the large deflection of symmetrically loaded shells of revolution. J. of Appl. Math. and Mech. (P.M.M.) 26, 913-922 (1962)
(Russian). Stein, E.M.: [l] Note on singular integrals. Proceedings of Amer. Math. SOC.
Bibliography
553
8, 250-254 (1957). - [2] Singular integrals and differentiability properties of functions. Prince-
ton Univ. Press 1970. Strang. G.: [l]On strong hyperbolicity. J. Math. Kyoto Univ., 6, 397-417 (1967). Strichartz, R.S.: [l] Multipliers on fractional Sobolev spaces J . Math. Mech. 16, 1031-1060 (1967). Sweers, G . : [l] Estimates for elliptic singular perturbations in Lp-spaces. Doctoraalscriptie, Univ. of Nijmegen, 1984.
-
[2] Estimates for elliptic singular perturbations in Lp-spaces. Asympt. Analysis, 2 : 2, 101-138 (1989).
Sze, S.M.: [l]Physics of semiconductor devices. Wiley, New York, 1981. Taylor, M.E.: [l] Pseudodifferential operators. Princeton University Press, 1981. Temam, R.: [l] Sur la stabiliti et la convergence de la mithode des pas fractionnaires. Ann. Mat. Pura Appl., 4, LXXIV, 191-180 (1968). Thomie, V. and B. Westergren: [l] Elliptic difference equations and interior regularity. Numer. Math., 11 : 3, 196-210 (1968). Tikhonov, A.N.: [l] On the dependence of the solutions of differential equations upon a small parameter. Mat. Sb., 22 (64), no. 2, 193-204 (1948) (Russian). Trkves, F.: [l] Introduction to pseudodifferential and Fourier integral operators. I, 11. Plenum Press, New York-London, 1980. Triebel, H.: [l] Interpolation theory, function spaces, differential operators. North-Holland, Amsterdam, 1978. Unterberger, A. and J. Bokobza: [l] Opkrateurs de Caldkron-Zygmund prkcisks. C.R. Acad. Sci. Paris, 259, 1612-1614 (1964). Vaillancourt, R.: [l] A simple proof of Lax-Nirenberg theorems. Comm. Pure Appl. Math., 23, 151-163 (1970). Vainberg, B.R.: [l]Asymptotic methods for equations of mathematical physics. Moscow State Univ., 1982 (Russian). Van Dijke, M.: [l] Perturbation method in fluid mechapics. Academic Press,
554
Bibliography
London-New York, 1964. Van Leer, B.: [l] Upwind differencing for hyperbolic systems of conservation laws. Numerical methods for engineering, I, 137-149, Dunod, Paris 1980. Viahik, M.I.: [l]On strongly elliptic systems of differential equations. Mat. Sb. 29, 615-676 (1951) (Russian). Vishik, M.I. and G.I. Eskin: [l] Normally solvable problems for elliptic systems of equations in convolutions. Mat. Sb. 78, 326-356 (1967) (Russian).
Vishik, M.I. and L.A. Lyusternik: [l] Regular degeneration and boundary layers for linear differential equations with a small parmeter. Uspekhi Mat. Nauk 12, no. 5 (77), 3-122 (1957); English transl.: Amer. Math. SOC. Transl., (2), 20, 239-364 (1962). - [2] Asymptotic behavior of solutions to differential equations, whose co-
efficients or these ones of the boundary conditions are large or vary fast. Uspekhi Mat. Nauk, 15, 3-80 (1960). Volk, I.M.: [l] On periodic solutions of autonomous systems. Prikl. Mat. Mekh., 12, no. 1, 29-38 (1948) (Russian). Volosov, V.M.: [l]The averaging for systems of ordinary differential equations. Uspekhi Mat. Nauk, 17, no. 6, 3-126 (1962) (Russian). Wasow, W.: [l] Asymptotic expansions for ordinary differential equations. Wiley, Interscience Publ., New York-London-Sidney, 1965. Weinstein, A.: [l]The order and symbol of a distribution. Trans. Amer. Math. SOC.,241, 1-54 (1978). - [2] On Maslov’s quantization condition.
Fourier integral operators and
partial differential equations. Springer Lecture Notes in Math. 459, 341372 (1974). Wendt, W.D.: [l] Coercive singularly perturbed Wiener-Hopf operators and applications. Ph.D. Thesis, Univ. of Nijmegen, 1983. - [2] The reduced operators of singularly perturbed Wiener-Hopf matrices.
Asympt. Anal., 1 : 1, 61-93 (1988). - [3] Stability and convergence results for a class of one-dimensional homog-
enization problems. Preprint no. 701, Univ. Bonn, 1984. Wilcox, C.H.: [l] Perturbation theory and its applications in quantum mechan-
Bibliography
555
ics. Wiley, New York, 1966. Yamaguti, M. and T. Nogi: [l]An algebra of pseudodifference schemes and its application. Publ. Res. Inst. Math. Sci. Ser A.3, 151-166 (1967). Zygmund, A.: [l]Trigonometric series. I, 11. The University Press, Cambridge, 1959.
This Page Intentionally Left Blank