Algebras of Pseudodifferential Operators

Mathematics and Its Applications (Soviet Series)

Managing Editor:

M. HAZEWINKEL
Centre for Mathematics and Computer Science, Amsterdam, The Netherlands

Editorial Board:

A. A. KIRILLOV,MGU,Moscow, U.S.s.R.
Yu. I. MANIN, Steklav Institute of Mathematics, Moscow, U.S.S.R.
N. N. MOISEEV, Computing Centre, Academy of Sciences, Moscow, U.S.S.R.
S. P. NOVIKOV, Landau Institute of Theoretical Physics, Moscow, U.S.S.R.
M. C. POLYV ANOV, Steklov Institute of Mathematics, Moscow, U.S.S.R.
Yu. A. ROZANOV, Steklov Institute of Mathematics, Moscow, U.S.S.R.

Volume 43
B. A. Plamenevskii
Leningrad Electrical Engineering Institute,
Leningrad, U.S,S.R.

Algebras of
Pseudodifferential
Operators

KLUWER ACADEMIC PUBLISHERS

DORDRECHT / BOSTON / LONDON
Library of Congress Cataloging in Publication Data
Plamenevskil. B. A.
[Algebry psevdodifferentSial 'nykh operatorov. Engl ish]
Algebras of pseudodlfferential operators I by B.A. Plamenevskii
translated from the RussIan by R.A.M. Hoksbergen.
p. cm. -- (MathematICs and ItS applIcations. Soviet series
43)
Translation of: Algebry psevdodlfferentSlal 'nykh operatorov.
Includes bIbliographIcal references.

1. Pseudodifferentlal operators. 2. Co-algebras. I. Title.

II. Series: Mathematics and its applicatio~s (Kluwer Acade.lc
Publlshers). Sovlet serles; 43.
OA329.7.P5313 1989
515' .7242--dc20 89-19849

ISBN-13: 978-94-010-7564-0 e-ISBN-13: 978-94-009-2364-5

DOl: 10.1007/978-94-009-2364-5

Published by Kluwer Academic Publishers,

P.O. Box 17,3300 AA Dordrecht, The Netherlands.

Kluwer Academic Publishers incorporates

the publishing programmes of
D. Reidel, Martinus Nijhoff, Dr W. Junk and MTP Press.

Sold and distributed in the U.S.A. and Canada

by Kluwer Academic Publishers,
101 Philip Drive, Norwell, MA 02061, U.S.A.

In all other countries, sold and distributed

by Kluwer Academic Publishers Group,
P.O. Box 322, 3300 AH Dordrecht, The Netherlands.

Printed on acid-free paper

This is the translation of the original work

AnI'EBPbI nCEBA0,n;HCllCllEPEHIJ;HAJlbHbIX OnEPATOPOB
Published by Nauka, Moscow, © 1986

Translated from the Russian by R. A. M. Hoksbergen

All Rights Reserved

This English Edition © 1989 by Kluwer Academic Publishers
Softcover reprint ofthe hardcover 1st edition 1989
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical
including photocopying, recording, or by any information storage and
retrieval system, without written permission from the copyright owner.
SERIES EDITOR'S PREFACE

'Et moi, ... , si j'avait su comment en revenir, One service mathematics has rendered the
je n'y serais point alle.' human race. It has put common sense back
Jules Verne where it belongs, on the topmost shelf next
to the dusty canister labelled 'discarded non-
The series is divergent; therefore we may be sense'.
able to do something with it. Eric 1'. Bell
O. Heaviside

Mathematics is a tool for thought. A highly necessary tool in a world where both feedback and non-
linearities abound. Similarly, all kinds of parts of mathematics serve as tools for other parts and for
other sciences.
Applying a simple rewriting rule to the quote on the right above one finds such statements as:
'One service topology has rendered mathematical physics .. .'; 'One service logic has rendered com-
puter science .. .'; 'One service category theory has rendered mathematics .. .'. All arguably true. And
all statements obtainable this way form part of the raison d'etre of this series.
This series, Mathematics and Its Applications, started in 1977. Now that over one hundred
volumes have appeared it seems opportune to reexamine its scope. At the time I wrote
"Growing specialization and diversification have brought a host of monographs and
textbooks on increasingly specialized topics. However, the 'tree' of knowledge of
mathematics and related fields does not grow only by putting forth new branches. It
also happens, quite often in fact, that branches which were thought to be completely
disparate are suddenly seen to be related. Further, the kind and level of sophistication
of mathematics applied in various sciences has changed drastically in recent years:
measure theory is used (non-trivially) in regional and theoretical economics; algebraic
geometry interacts with physics; the Minkowsky lemma, coding theory and the structure
of water meet one another in packing and covering theory; quantum fields, crystal
defects and mathematical programming profit from homotopy theory; Lie algebras are
relevant to filtering; and prediction and electrical engineering can use Stein spaces. And
in addition to this there are such new emerging subdisciplines as 'experimental
mathematics', 'CFD', 'completely integrable systems', 'chaos, synergetics and large-scale
order', which are almost impossible to fit into the existing classification schemes. They
draw upon widely different sections of mathematics."
By and large, all this still applies today. It is still true that at first sight mathematics seems rather
fragmented and that to find, see, and exploit the deeper underlying interrelations more effort is
needed and so are books that can help mathematicians and scientists do so. Accordingly MIA will
continue to try to make such books available.
If anything, the description I gave in 1977 is now an understatement. To the examples of
interaction areas one should add string theory where Riemann surfaces, algebraic geometry, modu-
lar functions, knots, quantum field theory, Kac-Moody algebras, monstrous moonshine (and more)
all come together. And to the examples of things which can be usefully applied let me add the topic
'finite geometry'; a combination of words which sounds like it might not even exist, let alone be
applicable. And yet it is being applied: to statistics via designs, to radar/sonar detection arrays (via
finite projective planes), and to bus connections of VLSI chips (via difference sets). There seems to
be no part of (so-called pure) mathematics that is not in immediate danger of being applied. And,
accordingly, the applied mathematician needs to be aware of much more. Besides analysis and
numerics, the traditional workhorses, he may need all kinds of combinatorics, algebra, probability,
and so on.
In addition, the applied scientist needs to cope increasingly with the nonlinear world and the

v
vi Series Editor's Preface

extra mathematical sophistication that this requires. For that is where the rewards are. Linear
models are honest and a bit sad and depressing: proportional efforts and results. It is in the non-
linear world that infinitesimal inputs may result in macroscopic outputs (or vice versa). To appreci-
ate what I am hinting at: if electronics were linear we would have no fun with transistors and com-
puters; we would have no TV; in fact you would not be reading these lines.
There is also no safety in ignoring such outlandish things as nonstandard analysis, superspace
and anticommuting integration, p-adic and ultrametric space. All three have applications in both
electrical engineering and physics. Once, complex numbers were equally outlandish, but they fre-
quently proved the shortest path between 'real' results. Similarly, the first two topics named have
already provided a number of 'wormhole' paths. There is no telling where all this is leading -
fortunately.
Thus the original scope of the series, which for various (sound) reasons now comprises five sub-
series: white (Japan), yellow (China), red (USSR), blue (Eastern Europe), and green (everything
else), still applies. It has been enlarged a bit to include books treating of the tools from one subdis-
cipline which are used in others. Thus the series still aims at books dealing with:
- a central concept which plays an important role in several different mathematical and/or
scientific specialization areas;
- new applications of the results and ideas from one area of scientific endeavour into another;
- influences which the results, problems and concepts of one field of enquiry have, and have had,
on the development of another.

Does an analyst need to know about such things as algebras, Noetherianess, homology, and
representation theory? The answer would appear to be an emphatic yes. (He also needs to know
about topology (and not only in the context of topological vector spaces, but also differential topol-
ogy).) And the more complicated the setting of the problems in the analysis concerned, the more
this is needed to sort out the phenomena.
It has been said before (by myself, for example), but bears to be repeated: "Mathematics is a
tool for thought" (at least as long as our brains are rather spectacularly finite in their abilities to
instantly oversee all the consequences and implications of, say, a certain set of assumptions). This
implies that any distinct part of mathematics provides tools for thought for any other. Mostly,
mathematicians are not slow to point out the importance and applicability of their science to other
fields; I wonder whether we follow our own advice with equal vigour intra-mathematically.
One fairly recently created tool, of considerable interest in itself and of vast applicability, is that
of algebras of pseudodifferential operators. The present book is devoted to that topic; more
specifically it is devoted to the case of the theory of pseudodifferential operators on manifolds of
higher dimension (;;;02) with isolated singularities. As such, it is unique with virtually no overlap
with other existing monographs; a more than worthy addition to this series.
Pseudo-differential operators are a topic from analysis, of course; in the book they are studied
from the point of view of C' -algebras, and the attentive and interested reader will become aware of
the importance and power of algebraic, topological and, perhaps especially, representation theoretic
considerations in this field.

The shortest path between two truths in the Never lend books, for no one ever returns
real domain passes through the complex them; the only books I have in my library
domain. are books that other folk have lent me.
J. Hadamard Anatole France

La physique ne nous donne pas seulement The function of an expert is not to be more
I'occasion de resoudre des problemes ... elle right than other people, but to be wrong for
nous fait pressentir la solution. more sophisticated reasons.
H. Poincare David Butler

Bussum, September 1989 Michiel Hazewinkel

 Table of Contents

Series Editor's Preface, ................................................................................................................................. v

Introduction ...... .... ............................. .................. ....... ..... ...................... ............... ........................ ......... ...... 1

Chapter 1.
Integral transforms on a sphere ................................................................................................................... 5

1. The generalized kernels (.\)' )~, (±xy + i 0)1' ............................................................................... 5

2. The operator E(A), its relation with the Fourier and Mellin transform ...................................... 8
3. The action of E(A) on spherical functions .................................................................................... 11
4. Operators related with the transform E(A) .................................................................................. 16
5. The spaces HS(A,Sn-'). The operator E(A) on the spaces Hs(A,Sn-') .............................. 22
6. An analog of the Paley· Wiener theorem for the operator E(A) .................................................. 29

Chapter 2.
The Fourier transform and convolution operators on spaces with weighted norms ................................ 37

1. The spaces Hp(R·) ....................................................................................................................... 38

2. The Fourier transform on the spaces Hp(Rn) ............................................................................ 39
3. Convolution operator on the spaces Hp(Rn) ............................................................................... 43
4. The spaces Hp(Rm ,Hm -.) .......................................................................................................... 47
5. Transversal operators and special representations ...................................................................... 51
6. Estimates for the convolution operator on the spaces Hp(Rm ,H m -n) ...................................... 60

Chapter 3.
Meromorphic pseudodifferential operators ................................................................................................ 73

1. Canonical meromorphic pseudodifferential operators ............................................................. 74

2. Operations on canonical meromorphic pseudodifferential operators .......................................... 83
3. General meromorphic pseudodifferential operators ..................................................................... 88
4. Traces of meromorphic pseudodifferential operators ................................................................... 97
5. Meromorphic pseudodifferential operators on strongly oscillating functions ........................... 106
6. Estimates for meromorphic pseudodifferential operators .......................................................... 116
7. Periodic meromorphic pseudodifferential operators .................................................................. 122
8. Change of variables in meromorphic pseudodifferential operators ........................................... 130
Vlll Table of COli fell fs

Chapter 4.
Pseudodifferelltial operators with discontinuous symbols
on manifolds with conical singularities .................................................................................................... 133

1. Pseudodifferelltial operators on R n .••.••...••.••••.•.••.....•..•.••..•..••.••.•....•.••••••.•••.••••••.••.•••••••..•.....•.•.• 134

2. Pseudodifferential operators on a conic manifold ..................................................................... 146
3. Pseudodifferential operators on manifolds with conical points ................................................. 163
4. Algehras generated b}' pseudodifferential operators of order zero ............................................ 168

Chapter 5.
The spectrum of a C' -algebra of pseudodifferential operators
with discontinuous symbols on a closed manifold ............................................................. _._._ ...... _.......... 181

1. Results from the theory of C' -algebras ..................................................................................... 182

2. The spectrum of a C' -algebra of pseudodifferential operators with discontinuities of the
first kind in the symbols on a smooth closed manifold (statement of the main theorem} ....... 187
3. Representations of the algebra @CA) generated by the operators ECA)" 1 <l>C<j>,W)ECA) ............. 191
4. Representations of an algebra (\')(Ix) .......................................................................................... 196
5. Proof of theorem 2.1 ................................................................................................................... 202
6. Ideals in the algebra of pseudodifferential operators with discontinuous symbols .................. 209
7. Spectra of C· -algehras of pseudodifferential operators on a manifold with conical points .... 213
8. The spectrum of a C· -algehra of pseudodijJerential operators with oscillating symbols ........ 228

Chapter 6.
The spectrum of a C· -algehra of pseudodifferential operators on a manifold with boundary ............. 235

1. The algebras (I),~ CA) ......•........•..............•...••.••.... _....................................................................... 235

2. The algebras l~ - CA) ..................................................................•................................................. 245
3. The algebras (I),' (I fl) .................................................................................................................. 257
4. The algebras (I) ~ (/ fl) .................................................................................................................. 265
5. The spectrum of an algebra of pseudodifferential operators on a manifold with boundary .... 273

Bibliographical sketch ............................................................................................................................. 279

References ................................................................................................................................................ 281

Index ......................................................................................................................................................... 287

 Introduction

The theory of pseudodifferential operators ('l'DOs), created in the last two-three

decades, has become both an object of application and a generally used instru-
ment in various branches of mathematics, such as mathematical physics, func-
tional analysis, and topology. Several monographs have been devoted to this
theory ([40], [67], [68], [69], [73], [74], etc.); in them pseudodifferential operators
arise in the 'smooth' case - with smooth symbols (coefficients), on smooth mani-
folds. In all branches mentioned above, quite a lot of problems appear related to
the study of pseudodifferential operators with singularities. In [9], [12], [17], [41],
[71] one-dimensional singular integral operators with discontinuous coefficients on
composite contours are treated. The corresponding problems for pseudodifferential
operators on manifolds of dimensions n ;;. 2 have not been reflected upon in
monographs. The present book is devoted to this very theme. The results
expounded in it form, it seems, the core of the theory of pseudodifferential opera-
tors with discontinuous symbols, which is being created. In essence there is no
overlap with other books.
We consider algebras of pseudodifferential operators on manifolds with a finite
number of isolated singularities. The symbols of the operators are allowed to have
discontinuities at a finite number of points. The set of points of discontinuity
depends on the operator, and by closing the algebras, operators whose symbols
have an everywhere dense set of singularities arise. Different versions of symbolic
calculus are considered, and necessary and sufficient conditions for a
pseudodifferential operator to be Fredholm are clarified.
One of the basic aims of the book is the study of pseudodifferential operators
from the point of view of the theory of C* -algebras. The scalar pseudodifferential
operators of order zero and with smooth symbols on a smooth manifold ~)TL
without boundary generate an algebra a, which after factorization by the ideal ~J;:
of compact operators becomes a commutative algebra. Hence all irreducible
representations of the quotient algebra Ct'/~\: are one-dimensional, and its spectrum
(the set of equivalence classes of irreducible representations) can be identified with
2 Introduction

the space of maximal ideals, which coincides with the bundle S' (~) of unit
cotangent vectors. For an A E ce, [A] E cU:J{ denotes its residue class, and if 'IT
runs through the spectrum of if/~ then the map 'IT~'1T[A] can be realized as the
function S' (01L) :3 'IT ~ <1>('IT), where <I> is the symbol of A.
The situation becomes more complicated if the manifold has a boundary or
singularities and (or) if the symbols of the operators become discontinuous. The
corresponding quotient algebra if/:J{ is, in general, not commutative. Among the
irreducible representations of if;:J{ there appear infinite-dimensional ones. In this
book we describe the spectra of algebras generated by pseudodifferential opera-
tors: all equivalence classes of irreducible representations are listed, a topology on
the spectrum is elucidated (the so-called Jacobson topology), and a realization of
the irreducible representations is given, i.e. a map te 3 'IT ~ '/T(A), where A E te and
if is the spectrum of if. The dependence of the spectrum on the kind of discon-
tinuity of the symbols and on the choice of function space is studied.
In the first Chapter we introduce and study the integral transform E(;\) on the
(n -I)-dimensional sphere Sri -1. The Fourier transform can be decomposed in a
product of three operators: the Mellin transform, E(;\), and the inverse Mellin
transform. We compute the values of E(;\) on spherical functions, and indicate a
space on which E(;\) acts as a continuous operator. The results of this Chapter are
constantly used in the sequel.
Pseudodifferential operators with discontinuous symbols are naturally regarded
in spaces with weighted norms. In the second Chapter we define the 'weighted'
classes Hp(lR m ), Hp (IR m , IR m -rl). On these spaces we study the Fourier transform
and operators of convolution with a homogeneous function. We prove bounded-
ness theorems for these operators, and derive special representations for convolu-
tions in terms of E(;\) and the Mellin transform. These representations play an
important part in the study of algebras of pseudodifferential operators. The third
Chapter is devoted to the special class of meromorphic operator-functions which
are called meromorphic pseudodifferential operators. It will later turn out that
meromorphic pseudodifferential operators are values of 'operator symbols'. The
algebras of these symbols is isomorphic to the quotient algebra if;:J{. Usin!! mero-
morphic pseudodifferential operators the irreducible representations of te are real-
ized (in Chapters 5, 6). The first three Chapters form the technical basis for the
subsequent exposition.
Introduction 3

In the fourth Chapter we study pseudodifferential operators with discontinuous

symbols on Euclidean space IR n and on manifolds. In particular, we introduce the
algebra ® of operator symbols and establish the isomorphism ® R:i (Xl ~{ On the
manifolds we allow for isolated singularities - 'conical' points. In defining
pseudodifferential operators on a smooth manifold one commonly uses the fact
that locally a pseudodifferential operator on a manifold coincides with a
pseudodifferential operator on a Euclidean space. In a neighborhood of a singular
point similar considerations are not immediately applicable. Thus we have to
define pseudodifferential operators on a manifold with conical points in a special
manner. This is done in several stages. First we introduce meromorphic
pseudodifferential operators on IR n , and then on a cone. The result of the Mellin
transform applied to a meromorphic pseudodifferential operator will be
pseudodifferential operator on a cone. After this, a pseudodifferential operator on
a manifold with conical singularities is glued together using partition of unity.
The spectrum of a C* -algebra of pseudodifferential operators is described in
the last two Chapters, the fifth and sixth. In Chapter 5 we consider algebras gen-
erated by pseudodifferential operators on closed manifolds, and in Chapter 6 - on
manifolds with boundary. The necessary information from the theory of C·-
algebras is given in §l of Chapter 5.
The contents of the book does not exhaust all known material concerning
pseudodifferential operators with discontinuous symbols. Aside remained the
theory of boundary value problems (which is sufficiently well-developed for elliptic
pseudodifferential operators with isolated singularities), algebras with multidimen-
sional discontinuities in the symbols (see [61]), and some scattered results. Some
articles concerned with these problems are included in the list of references.
 Chapter 1

Integral transforms on a sphere

In this Chapter we introduce and study the operator E(;\), which acts on the space
of functions on the sphere SrI -] . It is used in all subsequent Chapters.

§1. The generalized kernels (xy)~, (-+- xy + i 0)"

Let x = (X1, ···,x,,), Y = (yj, ... ,y,,) be unit vectors in ~", let
xy = x 1Y 1 + ... + x"y", and let SrI -1 be the (n - I)-dimensional unit sphere
with center at the coordinate origin. We introduce for Re f.L > - I the following
transforms on functions u of class Coo (S" - 1):

(I:u)(x) = f (xy)~u(y)dy.
s" -I
(1.1)

Here dy is the volume element on SrI -], and, as usual, t~ ° for t,,:;; 0,
t~ = ell]n I if t > 0, and t~ = (- t)~ .

Pro po sit ion 1.1. The maps IJf: Coo(S"-I)~coo(S"-1) are continuous.
The operatorjunction 1-''''''IJf
is ana(ytic in the halfplane Re f.L > - 1. Each of these
functions can be analytical(}' extended onto the whole I-'-plane, except for the points
I-' = - 1, - 2, ... , which are poles of the first order.

Proof. Let {1)j} be a partition of unity on SrI -I, i.e. a finite collection of non-
negative functions from coo(S,,-I) such that 2.1)) = 1. It suffices to verify the
statement for the operators 1)) I~. We may assume that on the support SUpP1)) of
1)j the angular coordinates form a regular coordinate net. Denote by gx some

5
6 Chapter 1.

rotation of Sn - I transforming the point 9L = (0, ... ,0,1) to the point x. Deter-
mine a family {gx}, x E SUPP11j, so that the matrix entries, giving rotations, are
smooth functions of the angular coordinates of x. We have

Denote by OJ, ... ,On - I the angular coordinates of a point Z E Sn -1, related to
Cartesian coordinates (z I, . . . ,Zn) by

(1.2)

where °
~ 0 1 < 27T, °
~ Ok ~ 7T, k =i= 1. Introduce the function vxCOn -I) =
11/X)X!U(gx(z)(sinon_2)n-3 ... sin02 dOl'" dOn - 2. Then
.,,/2
(11j I;u)(x) = j (cosOn_IY(sin°n_l)n-2vxC°n_ddOn_l.
o
Putting t = cos On -I, W X , + (t) = vxCarccos t), we extend the function JL t-> 11j I; u
to the halfplane Re JL > - k - 1 using the formulas
I
(11j I;u)(x) = jtp.{l-t 2){n-3)!2wx, + (t)dt = (1.3)
o
I
j tp.(l- t2){n - 3)/2 {
w k
+ (I)-
- I
"" q
_t_w{q)+ (0) } dt +
x, ~ q' x,
o q=O .

+k~l_l
~
(q) (O)..lB[JL+Q+l n-l]
, w x, + 2 2' 2 .
q=o q.

Recall that the beta-function B can be expressed in terms of the gamma-function

r, B (A, v) = r(A)r( v) I rCA + v), and that r{}L) is a meromorphic function on the
§1. The generalized kernels (xy)~,(-+-xy + iO)Jl 7

whole ,u-plane. The poles of the T-function are simple, are located at the points
,u = 0, -1, ... , and

(1.4)

Put now t = -COS(Jn-l, wx , -(t) = vA7T-arccost), and define the extension

of ,u t-> TI} I;: u to the halfplane Re,u > - k - 1 by
I
(TIjI;:u)(x) = jt Jl (1-t 2 )(n-3)/2 wx ,_(t)dt = (1.5)
o

I
jtJl(1- {k-I q
t 2 )Cn -3)/2 wx, -(t)- ~ ~w~~)_ (0)} dt +
o q =0 q .

+ k~l_l
~ ,w(q)x, - (O)J.-B[,u+q+l n-l]
2 2' 2 .
q=o q.

In view of formulas (1.3), (1.5) and the properties of the T-function indicated, the
proposition can be immediately verified . •
For Re,u > -1 and for u E coo(sn -I) we introduce the operators

(J:u)(x) = j (-+-xy +iOYu(y)dy, (1.6)

sn-l

where (xy +iO)Jl = (xy)Jt +eiJl'IT(xy)~, (-xy+iO)Jl = eiJl'IT(xy)Jt +(xy)~.

Pro po sit ion 1.2. The maps J:: coo(sn-I)~coo(sn-I) are continuous.
The operatorfunction ,u t-> J: can be analytically extended to the whole ,u-plane.

Proof. By proposition 1.1 it suffices to convince ourselves of holomorphy of

the extensions of ,u t-> J:,
which are obtained using the extensions of ,u t-> I:. For
this we need to take into account formulas (1.3) and (1.5), and note that the
singularities mutually cancel each other. •
8 Chapter 1.

§2. The operator E(A), its relation with the Fourier and Mellin transfonn
We define for arbitrary complex "A except "A = i(k +nI2), k = 0, I, the fol-
lowing operators on functions u from Coo(Sn -I):

1 i~(iA+nI2)
(E("A)u)(x) = 12 e 2 r(i"A+nI2) X (2.1)
(2'ITt

X j (-xy+iO)-iA-nl2 u (Y)t:{y.
s" -I

Proposition 1.2 implies that the operator-function "A I-> E("A):

Coo (Sn -]) ~ Coo (Sn -]) is analytic everywhere except at the points indicated, at
which it has poles of the first order. The residue at the pole "A = i (k + n I 2) is a
finite-dimensional operator; using (1.4) we obtain

(-i/ +] 1
resE("A)u IA=i(k+nl2) = -(2'IT)n/2 ~ -:l xY j yYu(y)~y, (2.2)
IY I ~ k I· s" - I

where x,y E IR n , Ix I = Ly I = 1, and, as usual, xY = xii ... x~" for

x = (x]," ·,xn),Y = (Y], ···,Yn)·
Let now u be a function from Coo(sn -]) satisfying

f y Y U (y)dy = 0 (2.3)
s" -I

for all multi-indices y for which Iy I = k. Formula (2.2) implies that the function
"A I-> E("A)u remains regular at the point "A = i (k +n I 2). Using the Taylor series
expansion

(- xy+iO)-iA-nl2 = {-xy+iO/+
-(-~y +iO)kln(-~y +iO)(k +i"A+nI2)+

and (1.4) we obtain

lim (E("A)u)(x) = (2.4)

A-->i (k+" 12)

- 1'2 bt.-j(Xy)kln(-xv+iO)U(v)dv
(2'IT)" I k!· ~ .

I /2
(2'IT)"
(-il j (xy) k[ In -1--1
k!·
I
~v
.
'-I'lTO(xy)
.
1u(y)dy,.
§2. The operator E(A) 9

where O(z) = I in case z > 0 and O(z) = 0 for z < O. Since

I k _ I
-k'.(·xy) - L -,xYyY,
Iyl ~k y.

by (2.3) the relation (2.4) can be replaced by

lim (E(A)U)(X) = (2.5)

A~i(k +nI2)

= I
(27Tt I2
.i=iL
k!
f (Xy)k{ln-l--~sgn(x}')}u(y)dy.
sn-l Ixy I 2 . -

We will denote this operator by E (i (k + n / 2).

In the remainder of this section, x is an arbitrary vector from IR n, r = Ix I,
q, = x / Ix I, ~ is a vector from the dual space, p = I~ I, -.f; = ~ / I~ I.
We write the Fourier transform as

(Fu)(~) = 1 fe-i~Xu(x)dx
(27T)nl2 '

and the Mellin transform (for functions from Co (IRn \ 0» as

(Mu)(A,q,) _ ii(A,q,) = _b-fr-iA-lu(r,q,)dr, A E C.

V27T 0

For the Mellin transform the inversion formula

(2.6)

holds, as well as Parseval's equality

00
1
f I u(A,q,) 12 dA = f r2f31 u(r,q,) I 2 dr, T = {3+- (2.7)
2·
ImA='T 0

(It is obtained by the change of variable r = e l from the corresponding property

of the one-dimensional Fourier transform.)

Pro p 0 sit ion 2.1. For functions u of class Co (IRn \ 0) the equali~}'

(2.8)
10 Chapter 1.

Proof. Using the inversion formula (2.6) we write the Fourier transform Fu as

(Fu)(p,lj.) = (2'7T)(n + 1)12 X

+00
X f fe-irp#rn-Idrdcp f r°l.- n12 u(}.+inl2,cp)dA.
-00

In order to substantiate the interchange of the integrals in this formula we

introduce a parameter T > 0:
+00
f f e-irpcj:Af-r"rn-1drdcp X (2.9)
o S"-1

+00
X f u(J...+inI2,cp)r iJ...-n/2dJ... =
-00

1 +00
lim (+1)12 f f u("A+ in I 2,cp)dNlcp X
T~+O(2'7T)n -ooS"-1

+00
X f e- ir !#-r"r il\+n/2- l dr.
o
Further,
00
f e -irpcj:Af-r"r O,-I+n/2dr = (2.10)
o

(The calculation of this integral was given in, e.g., [5].) Using the last equality
and (2.9) we obtain (2.8) . •
§3. The action oj E(A) on spherical Junctions 11

§3. The action of E('\) on spherical functions

Denote by % a sequence of integers (k l ,k 2,'" ,-+-kn - 2) for which
m = ko ~ kl ~ . . . ~ k n -2 ~ 0, and by Ymx the following spherical function
of order m:

YmX -- A mxe ::,:,

·k ,n-3
.-2'i'1 II('sm'l'n
.r,
- j - 1
)
k
J+l X (3.1)
j =0

XCkJ+1+(n-2-j)/2( .r, )
kJ - kJ + 1 cos 'l'n - j - 1 ,

where I/; E Sn -I, 1/;1, . . . ,I/;n -I are the angular coordinates of I/; (cf. (1.2», C~
are the Gegenbauer polynomials, and

2 _ 1
Amx - r(nI2) X (3.2)

n -3 2
XII
2kJ + 1 +n -j -4(k -k· + l)!(n -J' +2k -2)r 2
J J J
«n -J' - 2)/2+ k+ I)
!
j =0 y:;;r(kj + k j + 1 + n - j - 2)

(cf., e.g., [lO]). The functions YmX form a complete orthogonal system in the space
L 2 (sn-l).

Pro p 0 sit ion 3.1. The Jormula

(3.3)

1 i!!...(iA+nI2)
--=-n-I-=-2 e
(2w)
2 r{iA+n/2) f (_qi}+iO)-iA-nI2YmX(O)dO =
S.-I

r m +iA+nI2]
= (_ i)m2iA 2 Ym:K:(CP)
r [ m -i~+nl2]

holds.

Proof. Consider the integral

(3.4)

Denote by g 4> some rotation of Sn - I mapping the point 9L = (0, ... ,0,1) to the
12 Chapter 1.

point cpo Rewrite the integral (3.4) as

1(-g1>(0L)-8-fiO),"Ym'X(8)d8 =

= IYm'X(g1>l/I)(-9Uf;+iOYdl/l = IYm'X(g1>l/I)(-Cosl/ln-l +iOYdl/l.

Under rotations of the sphere, the spherical functions transform by the rule (cf.
[10)).

Y m'X(g1>l/I) = ~t~h(g;I)Ym''1dl/l),
0lL

where t0h are the matrix entries of the irreducible representations of the rotation
group SO(n), g E SO(n). In the sequel we only need the following properties of
the functions t~h:

m (g-I)
to'X 1> = (3.5)

= (m!r(n -1»1!2(r(m +n -2)(2m +11 -2»-1!2Ym :J;(.p);

here 0 = (0, ... ,0). Using this rule we obtain

(3.6)

We compute the last integral. Formula (3.1) implies that

I
= A m0n. e ±1I11n
. , n-3
-'YI II (sm't'n
.. 1. -). - J
)ml II X
j ~o

X C ml-m"1 +'0)1'
m j,I+(n-2- J )I2(.1. X
cos't'n-;-J )( -COSYn-l
I
I

. n - 2.1.
X sIn . n -- 3.1.
't'n-I'Sln 't'n-2'" Slll't'2 't'l ... d.l.
. .1. d· l.
't'n-J·

This integral is equal to the product of the following integrals:

I e-
27T
-+. ,
1mn 2Yldl/l1, (3.7)
o
' -} - J )m
1(sml/ln
7T
jl l +n- j --2 mj ,I+(n-2- j )/2(.L
C mJ -mJ I I COS't'n -,--1
)d· l.
't'n -;-
.
J,
(38)
.
o
§3. The action of E(A) on spherical functions 13

1 ~j ~n -3,

"'. )ml+n~2Cml+Cn~2)!2(
71

f( sm'rn~1 m~ml
,I,
COS'rn~1
)X (3.9)
o
x ( - cos !/In ~] + i oy d!/ln ~ ].

We show that all integrals (3.7), (3.8) are distinct from zero only if
m1 = ... = mn ~2 = O. Clearly, the integrals (3.7) are distinct from zero only
if mn ~ 2 = O. Suppose now that mj + I = O. Then the integral

' 'rn
71

f( sm ,I,
~j ~ ]
)mJ+I+n~j~2CmJ+I+(n~2~j)!2(
m ~m + J J
,I,
cos 'rn ~ j
I ~I
)d'"
'rn ~ j ~ I
o

' 'rn ~ j ~ ] )n ~ j ~2C(nm ~2~ j)!2 (,I,

71

f( sm ,f,
J cos 'rn ~ j ~ ] )d'"'rn ~ j ~ 1
o
is distinct from zero if and only if mj = 0, which follows from the orthogonality
of the Gegenbauer polynomials:
71

fcr(cosO)Cfn(cosO)sin2pO dO =0 for m =1= I

o
([10]), and the equation q)(cosO) _ 1.
Thus, m 1 = ... = mn ~2 = O. In this case the product of the integrals (3.7),
(3.8) is equal to

Now we consider the integral

f (sin !/I)n ~ 2C~ ~ 2)/2(cos!/l)( - cos !/I + i O)"d!/l =

71

o
I
f (1- t 2 )Cn ~ 3)!2C~ ~2)!2(t)( - t + iOYdt,
~I

which coincides with the integral (3.9) for m] = O. Taking into account that ([6],
1l.l, (25»
14 Chapter 1.

C~ -2)/2(t) = f«n - 1) 12)f(m + n - 2)(1- t 2p-n)/2 X

(- 2)mf(m +(n -1)/2)f(n -2)

x d m (1_t 2)m+(n-3)/2
dtm '

we integrate by parts m times, and obtain

1
i
j (1- t 2 n -3)I2C~ -2)/2(t)( - t +i oy dt =
-I

= v(v -1) ... (v -m + l)f«n -1)/2)f(m +n -2) X

(-2)mm!f(n -2)f(m +(n -1)/2)
1
X j (1- (2)m +(n -3)/2(t + iOY -mdt.
-I

Since (t +ioy-m = t+- m +ei1T(v-m)t~-m, we have

1
j(1-t 2)m+(n-3)/2(t+iOy- mdt = (3.10)
-I

1
= (1+ei1T(v-m)jtv-m(1_t2)m+(n-3)/2dt =
o
= (1 +e i1T(v-m) f(m +(n -1)/2)f«v -m + 1)12)
2f«m +n +v)12)

Combining the equation (cf. (3.2»

AmO = (m!f(n-2)(2m+n-2»1/2(f(n+m-2)(n-2»-1I2

and formulas (3.5), (3.6), (3.10) leads to the relation

j( -.p8+iO)vYm'X(8)d8 =

_ 'IT(n-I)/2 v ... (v -m + l)f«v -m + 1)12)(1 +e i1T(v-m)

- (-2)mf«m +n +v)/2) YmX(cf»·

Putting v = -}l.-n12 and calling to mind formula (2.1) we find

= f(-v)v··· (v -m+ l)f«v -m + 1)12)e-iv1T12(l +e i1T(v-m) X

(-1)m2m +n12 y:;f«m + n +v)/2)
§3. The action of E(;\) on spherical functions 15

This and the equation f(z + 1) = z f(z) imply that

(3.11)

where

ILm(;\) = (3.12)

(- i)mf{i;\+m +n/2)f«l- i;\ -m -n/2)/2) X

2m -1+nI2 V;f«m -j;\+n/2)/2)

X (i;\+nI2+m)'IT
cos 2 .

Applying the formula f(l-z)f(z) = 'lTlsin'ITz for z = i;\+m +n/2+ 1, we find

f«(1- i;\-m -n/2)/2) =
'IT
f{{i;\+m +nl2+ 1)/2)sin('IT(i;\ +m +nI2+ 1)/2)·

Hence (3.12) may be rewritten as

imV;f(i;\+m +nI2)
2m -1+nI2f«m +n/2-i;\)/2)f«m +n/2+ 1+i;\) 12) .

Now we use the duplication formula

f(2z) = 22z -1'IT- 1/2 f(z)f(z + 112)
for z = (i;\+m +nI2)/2. We have
f{i;\+m +n/2)
f((i;\+m +n/2+1)/2)
= 2iHm +nl2- I 'IT- 1I2 f«(i;\ +nI2+m)/2).
So that

which was required . •

16 Chapter 1.

We now consider the operator E(i (k + n 12», defined by (2.5) for functions
satisfying the condi tions (2.3). We first note that the conditions (2.3) are
equivalent to

ju(x)Ym'X(x)dx = 0, O.s:;m ~k, m k(mod2). (3.13)

Indeed, any homogeneous polynomial Pk of degree k in the variables XI, .•. ,Xn

has on Sn - I the canonical expansion

[k!2]
Pdx) = ~ hk -2s(X),
s -0

where the h; are homogeneous harmonic polynomials of degree j (see [10]). There-
fore (3.13) implies (2.3). Conversely, by (2.3) f u(x)xYh k -2sC'C)dx = °for every
multi-index y for which 1 y 1 = 2.1'. This clearly implies (3.13).

Pro p 0 sit ion 3.2. The subspace of all functions from Coo(Sn -I) satisfving
(3.13) (or, equivalent?v, (2.3») is invariant under the operator E(i (k + n / 2)) (see
(2.5)). The formula

( E(i(k +n/2))Y ,')("')

mX 'f'
r«m+n-k)/2)
= (_i)m2- n12 - k r«m Y,,("')
+k)/2) mX 'f'

holds.

Proof. Obtained by applying proposition 3.1 . •

§4. Operators related with the transform E(t\)

1. The operator E(t\)-I. We define for arbitrary complex ;... except
;... = - i(k + n / 2), k = 0, I, ... , the following operator on functions v from
coo(sn-I):

(4.1)

I i..'!..(nl2-i!l.) . _
--'---e 2 r(nI2-iA)j(XV +iOY!l. nl2v(J,)dv
(2'7T)n!2 . :T'

where x,y E IR n, 1x 1 = Iy 1 = 1, and dy is the volume element on Sn -I. Pro-

position 1.2 implies that the function ;... ..... E(;...)-I: coo(sn-I)~coo(sn-I) is
§4. Operators related with the transform E(A) 17

analytic throughout the A-plane, with the exception of the points indicated. at
which it has poles of the first order. The residue at A = - i (k + n I 2) is a finite-
dimensional operator,

res(E(A) - Iv) IA~ -i(k +nl2) (4.2)

= 1 ik+1 ~ _1 xYjyYv(y)dy.
(2'7Tyf2 lyl =k Y! - -

If a function v satisfies (2.3), then

-1 ·k
lim
A->-i(k +n(2)
E(A)-I v = (2'7Ty/2 kl '. j(xy)k In(~y +iO)v(y)dy.

We denote this operator by :B( - i (k + n I 2» - I. We have

E(-i(k+nI2»-l v = (4.3)

i k j (xy) k { In--1 - +1-

1nl2 k' . '7T }
=
(2'7T). I.~y I 2 sgn(~y) v (y)dy.

Pro po sit ion 4.1. For A =1= -+-i(k +nI2) the operators E(A) and E(A)-I are
inverse to each other.

Proof. By formula (2.8),

We apply the inverse Fourier transform. Then

co
X j p-iA-n/2E<j>->y(A)u(A+in12,<j»dA
-co

Changing the order of integration we obtain, because of (2.10),

u(r,lJ) = 1 +jCO j r iA - ni2 / ; (ni2-iA) X

( 2'7T)(n+I)f2 -x;s n l ·
18 Chapter 1.

x r(n 12 - iA)(fAp+ ioiA-nI2E1>->~(A)uCA + inl2,cp)dAdI/; =

= M-I(A+inf2)~' 4-->0 (A) -I Eq.-->y(A) U(A + in 12,cp).

After applying the Mellin transform M r -->A+inI2 we obtain the equation

U(A + in 12,') = E(A) - I E(A)u(A + in 12, .),

which was required.

So, E(A) -I E(A) = 1. By interchanging the roles of the transforms F and F- 1
we can verify the equality E(A) E(A) -I = I. •

Pro p 0 sit ion 4.2. The formula

(E(A)-I YmX)(CP) = i m 2-i>'f«m +nI2-iA)I2) Y "(cp) (4.4)

r«m +nI2+ iA)/2) mX

holds.
For the proof it suffices to compare propositions 3.1 and 4.1 . •
We denote by Sjk the space spanned on the spherical harmonics {YmX} for
which m';;;; k, m k mod 2, and by Sj t
the subspace of functions from
coo(sn-I) satisfying the conditions (3.13) (or, equivalently, the conditions (2.3».
The subspace Sj f is invariant under the operators E(A)± 1 for all A, and also
under the operators E(ik+inl2) and E(-ik-inI2)-1 given by (2.5) and (4.3).
We yet introduce the restrictions EUk + in /2) - 1 and E( - ik - in 12) onto the sub-
space Sjf.

Pro p 0 sit ion 4.3.

1) The operators EUk + in 12) - 1 and E( - ik - in 12) are inverses of EUk + in /2) and
E( - ik - in /2) - I, respectively, and are defined by the following formulas:

E(-+i(k +nI2»+l v = (4.5)

= I in+kr(n +k)j(-+cpO+io)-n-kv(O)dO.
(27T)nI2

2) The operators E( - ik - in 12) and E(ik + in /2) - 1 annihilate the subspace Sjk'
Thus,
§4. Operators related with the transform E(;\) 19

where prku is the orthogonal projection to ffJk (in the sense of L 2 (sn -1») of the
element u.

Proof. The first assertion of the theorem follows from proposition 4.1 and the
definitions of the operators (cf. (2.5) and (4.3»). It remains to convince ourselves
that E( - ik - in l2)ffJk = 0 and E(ik + in 12) -1 ffJk = O. In order to verify, e.g.,
the first of these equations one must use formula (3.3) and take into account that
at ;\ = -ik-inI2 the function f«m-i;\+nI2)/2) has a pole if m~k,

m _ k mod2. The second equation is verified using formula (4.4) . •

2. The operator E(A)* adjoint to E(A). We denote by E(;\)* the operator adjoint to
E(;\) with respect to the scalar product in L 2 (sn -1).

Pro p 0 sit ion 4.4. The equation

(4.6)

holds.

Proof. The definition of the generalized function ( - </>w + i O)JI implies that for
arbitrary complex /-t,

jv(</»d</>j(-</>w+iOYu(w)dw =

= jv(</»d</>j[ei7TJI(<j>w)~ +(<j>w)~]u(w)dw =

j u (w)dw j[ e - i7TI1 (<j>w)'t + (<j>w)~]v (</»d</> =

j u(w)dw j e -i7TI1 (</>W+ ioiv (</»d</>.

Since feu) = f(U), formulas (2.1) and (4.1) imply (4.6) . •

3. The Fourier transfonn of a homogeneous function. Let the homogeneous func-

tion have the form G(x) = raj (</», where r = Ix I, <p = xlix I, and j is a
smooth function on the sphere Sn -1. It suffices to calculate the Fourier transform
of this homogeneous function for Re a > - n; for remaining a the formula is
20 Chapter 1.

obtained as the result of analytic extension. We have

= 1 lim /J(cp)dCP/e~irPW~Trr"+n~ldr,
(27Tt I2 T-'+O

where p = I~ I, l/; = ~I I~I· This and (2.10) implies

(FG)(~) = 1 p~" ~nei(" + n)'1T12 r(a + n )/( -CPl/;+ iO)~" ~nJ (cp)dcp.
(27T)1l12

Put <P(l/;) = p" +n(FG)(~). Recalling formula (2.1) we are led to

<P = E(-i(a +nI2»j. (4.7)

If a = - n - k, where k is a nonnegative integer, then in order for G to be a

homogeneous function it is necessary and sufficient for J to satisfy (2.3). In this
case (4.7) can be changed to

<P(l/;) = E¢---->y(ik +inI2)J(cp) = (4.8)

-
-
1
(27T)nI2
(-it
k!
/ (cpl/;) k{ In I#T
1
-12 sgn (#) f(CP)dcp.
.7T }

In drawing this paragraph to an end we give some remarks.

Rem ark 4.5. In the theory of singular integral operators the function J is
called a characteristic, and cI> a symbol (cf. [40]). Formulas (4.7), (4.8) (as weII as
(2.1), (4.1), and (4.3» express one of these functions in terms of the other, for
arbitrary a. In case k = 0 equation (4.8) is called the Calderon-Zygmund Jormula.

Rem ark 4.6. Extend u +, u ~ E ccc(sn ~ 1) to 1R1l \ 0 as (positively) homo-

geneous functions of degrees iA - n 12 and - iA - n 12, respectively, i.e. assume
that u + (~y) = t iA - n12 u + 0-'), u ~ (ty) = t ~iA~nI2tL (y) for all y E 1R1l \ 0 and
t > O. Introduce the exterior differential form
w(y) = Yl~Y2/\'" /\~Yn+ ... +(_1)n-l YndYl/\ ... /\dYn~I' Denote by
S an arbitrary (n --- I)-dimensional smooth surface in IR n encircling the coordinate
origin. Then the operators E(A)± 1 allow the representations
§4. Operators related with the transform E(A) 21

(E(A)±lu±)(x) =
1
12 e
i; (nl2±iA) f(nI2-+-iA) X (4.9)
(27T)n

X !(-+-xy+io)-nl2+iA u ±(y)w(y).
s

Since the integrand is homogeneous of degree 0 (in .F), the 'integral' is indepen-
dent of the choice of S. It is clear that the operator E(A) (resp. E(A) - I) thus
defined maps homogeneous functions of degree iA - n 12 (resp. - iA - n /2) to
homogeneous functions of degree - iA - n 12 (resp. iA - n 12).

Rem ark 4.7. Consider the operators E(A)±I for n = 1. In the one-
dimensional case x and y take the two values -+-1, and the 'integral' over the
zero-dimensional sphere is the sum u (1) + u ( - 1). Hence for n = 1 we have the
formulas

[ (E(A)U)(l) 1 f(l/2+iA)
X (4.10)
(E(A)U)(-1) -
&
e -2
iw [2+
1 IA ] e i; [-i-+iA]
X
iw [1 ] iw [2+
1 iA]
[ u(l) 1
u( -1) ,
e 2 2+ iA e -2

1=
-I
[ (E(A) -I u)(I) r{l/2-iA2
X (4.11)
(E(A) u)( - 1) V2;;

e i; [+-iA] e- i; [+-iA]
1
X
iw [2-
e -2 1 IAJ e 2iw [I2- IA
"J
[ u(l)
u(-l)'

Rem ark 4.8. Let

J:'«x) = ! (-xy+iO),",u(y)dy.
sn - I n {y : I xy I >(}

If for small positive £ this integral can be represented as

22 Chapter 1.

where ReAk > 0, and if a finite limit ao(x) = limHo a«x) exists, then we say
that the principal value (of the) integral

j ( - xy + iO)Jl. u (y )dy

exists. By definition

p.v. j(-xy +iO)Jl.u(y)try = ao(x).

It turns out that the principal value exists for =t= -1, - 2, ... , and coincides
J: defined in §l (formula (1.6». For details see [62].
jJ-
with the integral

§5. The spaces HS(J>.,sn-I). The operator E(J>.) on the spaces HS(J>.,sn-l)
In this paragraph we introduce for an arbitrary real number s the spaces
HS(A,sn-l) of (generalized) functions on the sphere, with norms depending on
the parameter A E C. For each definite value of this parameter the norm in
HS(A,sn -I) is equivalent to the norm in the Sobolev-Slobodetskii space
w~(sn-I) Hs(sn-I). It turns out that the maps E(A)±I:
HS(A,sn -1) --,) Hs±ImA(A,sn -1) are continuous.

1. The Sobolev-Slobodetskii spaces HS(~) on a manifold ''JlL For an arbitrary real

s the space HS(Rn) is defined as the completion of the set Cif (Rn) with respect to
the norm

(5.1)

For nonnegative integer s this norm is equivalent to the norm

and for arbitrary s > 0 it is equivalent to the norm

~ I(D"'u)(x)-(Dau)(y) I 2 dx dylll2
[ IIU ;H[sJ(Rn)112 + j j
lal =[s] IR"IR" I.X -}' I n + 2s - 2[s]
§5. The spaces HS(}..,sn -I). The operator E(A) 23

Let ~ be an n-dimensional compact manifold without boundary, of class Coo,

and let {U, X} be some atlas on it, i.e. {U} is a finite open cover of ~ and
X: U --i> IR n are coordinate maps. Let {f} be a partition of unity subordinate to
this cover. The space HS(~) is the completion of the set Coo(~) with respect to
the norm

(5.2)

°
where ~xu = ~UOX-I on X(U), and ~xu = outside X(U). Another partition of
unity and another, equivalent, atlas lead to an equivalent norm in HS(''Jrl). For
s 1 < s 2 the space H S2 (~) compactly belongs to H S1 (~). The spaces H S(~) and
H-S(~) are dual with respect to the scalar product in HO(~).

2. Expansion in spherical hannonics. In the sequel we conveniently denote the

spherical harmonics of order m (m =0,1 ... ) by Ymk . The index k enumerates
the harmonics of the same order, k = 1, ... ,km' where k m = (2m + n - 2) X
X(m +n -3)![(n -2)!m!)-I, i.e. k m = O(m n - 2 ) as m --i>00. The functions Y mk
form a complete orthonormal system in HO(sn-l) = L 2 (sn-I). Hence for
U E HO(sn -I) the series U = 2.k,m Umk Y mk converges in HO(sn -I), and the Par-

seval equality Ilu ;HO(sn - I )11 2 = 1

2. m ,k Umk 12 holds. Here Umk = (u, Y mk) (scalar
product in HO(sn - I )).
Let 8 be the Laplace-Beltrami operator on Sn -I:

8 = -5:= .' n - ; - 10
j 1 qjsm n-j
.' aona-j. [sinn - j - IOn - j aona_j.] ,

where 0 1,'" ,On-I are the angular coordinates (see (1.2)), ql = 1,

qj = (sin On -I ... sin On _ j + 1)2. The Laplace operator t:. on IR n, written in spher-
ical coordinates, has the form

(5.3)

On HO(sn -I), the operator 8 is nonnegative and has a selfadjoint Friedrichs

extension, which we will also denote by 8. The spectrum of 8 consists of the eigen-
values Am = m (m + n - 2), m = 0, 1, .... To each Am correspond the k m ortho-
normal eigenfunctions Ymk. The domain of definition of a power (I + 8Y, s > 0,
24 Chapter 1.

consists of only those functions v for which the series

11(1 + 8Yv; HO(sn - 1)11 2 = ~ (l + m (m +n - 2))2s 1Vmk 12 (5.4)

m,k

converges. The (generalized) functions v on the sphere for which (5.4) converges
belong to the domain of definition of (I + oy also for s < O.
The aim of the present paragraph is to prove the following assertion.

Pro p 0 sit ion 5.1. For arbitrary s E IR the inequalities

hold.

The verification of these inequalities is subdivided into several lemmas.

L e m m a 5.2. For u E C(j (IRn) the formula

(I +oYFu = F(I +oYu

holds.

Proof. It suffices to check oFu = Fou. Using (5.3) we obtain

o'ifx->~u(x) = Fx->~{[(iI]J#)2+i(n -l)I]J#-6.x r 2 ]u(x)}, (5.5)

where r Ix I, cp = x 1 Ix I, I/; = ~ 1 I~ I . Since

a2 n -1
6.(r 2u) = - 2 (r 2u)+---a (r 2u)-ou,
a
ar r r

it remains to convince ourselves that

For this we must integrate by parts the two last terms . •

Let 1) E C""(IR), 1) = 1 on the (closed) interval [3/4, 514] and 1) = 0 outside
the interval [112, 3/2]. For v E c",,(sn-l), put C'\{rcp) = 1)(r)v(cp). It is easy to
check that
§5. The spaces H S CA, Sn -1). The operator E(A) 25

(5.6)

with constants c1,c2 not depending on v.

Le ill ill a 5.3. For 0 < s < 1 the estimates

hold.

Proof. By Parseval's equality and Lemma 5.2,

00

11(1 +8r'"'V; HO(lR n)11 2 = J pn -1dp J I(I +8rF'Y12d~. (5.8)

° S"-I

Taking into account the assertions concerning the spectrum of the operator 8
we have

r
J I(1-8rF'Y12d~ =

~ ~ [(1 +m(m +n -2))'~ 1(F"\)md [~I (P"Vlmk I' 2 r-'·

Now the Holder inequality with exponents s -1 and (1- s) -1 gives

J I (I +8?F'Y12d~:S;; [J I(I +8)F'Y12d~r [J IF'Y12d~r-s. (5.9)

Formula (5.5) implies the estimate

1(1 + 8)F'Y1 2:s;; (1 + p2)2 ~ IFx/YI2, (5.10)

j

where Xj E coo(lRn) and the sum is finite. By (5.8) - (5.10),

II(I +8)2'Y; HO(lR n )11 2 :s;;

:s;; 1(1+p2)2spn-1dP [7JIFXj'Y12d~r [JIF'Y12d~r-s:s;;

'" c 1(1 + p')"p" -Idp f [~:I Fx/VI' + F~l'l dol

1
26 Chapter 1.

(we have used Young's inequality aSb 1-s ~ sa +(1-s)b). Hence

11(1 HY'v: HO(R")II' "" C [lie" H"(R")II' + 117X/V: H"(R')II'l.

Since Xj E Cco(lRn) and "\~r,cp) = 0 for r > 312, we have

11(1 +8Y'Y; HO(lRn)11 ~ cll'\; H 2s (lR n)ll. (5.11)

One of the inequalities in (5.7) has been proved. Let us verify the second. We
have
co
J = II'Y; H 2s (lR n)11 2 ~ 1(1 + p2)2s pn -I dp 1 I(I + 8)-SF(1 + 8)S(YI2d~.
° S·-l

The integral over the sphere is at most

[I I(I + 8)-1 F(I +8Y'Y12d~] S [J I F(I +8Y'Y12d~] I-s

(this estimate can be derived similarly to (5.9)). By Holder's inequality,

~ [l(l+p2)2pn-IdPII(I+8)-IF(I+8Y'Y12d~r X

r' '"
J (5.12)

X [1 p.-ldp [1 IF(l HY'YI'd';1

.;;; [J(l+ 1~12)2IF(I+8y-I("(12d~r [J1(I+8Y''Vfd~r-s.

It is clear that

Using (5.3) we find that the righthand side does not exceed

c [J I(I +8Y -Iv 12d~+ 1I(I +8Y -18v 12d~] .;;; ell (I +8Y'YI2dx.(5.14)

Combining (5.12) - (5.14) we are led to the estimate J .;;; 11(1 +8Y,,\; HO(lR n)11 2,
which together with (5.11) gives (5.7) . •

L e m m a 5.4. For (1 > 0 the inequalities

§5. The spaces HS(J...,sn-l). TheoperatorE(}..) 27

hold.

Proof. Formula 5.3 and the equation '"'v = T/V imply that (I + ~)'\~r, 1/» =
= x(r)v(I/»+(I-r2.6.)'\~r,I/», where X E C~(IR'+). Hence

111(1 + 8)'"'V; H 2o (lR n)II-II(1 - r2 .6.)'11; H 2°(lR n)111 ~ c Ilxv; H 2o(lR n)ll.

Since 'Y vanishes outside the annulus 1/2 < r < 3/2,

c 111(1 - .6.)'Y; H 2o(lR n)11 ~

,,;;;;; II(I - r2 .6.)'11; H 2o(lR n)11 ,,;;;;; c211(1 - .6.)'1[; H 2o(lR n)ll.

Hence

11(1 +~)'II; H2o(lRn)11 ,,;;;;; c(II'II; H 2o +2(lR n)11 + Ilxv; H 2o(lR n)II), (5.16)

II'"'V; H 2o +2(lR n)11 ~ c(ll(I + 8)'11; H2o(lRn)11 + Ilxv; H 2o(lR n)II). (5.17)

Using inequalities of the type of (5.6) we obtain

Ilxv; H 2o(lR n)11 ~ cllv; H 2o(sn -1)11 ~ cll'l[; H 2°(lRn)ll.

Together with (5.16) this leads to the estimate 1I(I+8)'II;H 2o(lR n)ll,,;;;;;
,,;;;;; cll'll; H 2o + 2(lR n)ll, i.e. to the inequality on the right in (5.15).
Let us now prove the left inequality. We have

(5.18)

The inequality II'll; HO(lRn)11 ~ cll(I +8)'1[; HO(lRn)11 and estimate (5.18) imply

Ilxv; H 2°(lR n)11 ,,;;;;; c II'll; H2o+2(lRn)llo I (0+ I) 11(1 + 8)'Y; H 2o(lR n)11 1I (o+ I).

By using (5.17) and Young's inequality aSb l - s ~s£llsa+(l-s)bc-1 for

s = (1 I «(1 + 1) and a sufficiently small £ > 0 the proof is finished . •

Proof 0 f pro p 0 sit ion 5.1. Let initially s ~ O. By (5.6) it suffices to

convince ourselves that
28 Chapter 1.

Put s = [s]+t. By lemma 5.3,

c l ll(I+o)[sl'Y;H 21 (lR n )11 ,;;;;

,,;;:; 11(1+~Y,\;HO(Rn)11 ';;;;c211(I+~)[slcV;H2t(Rn)ll.

Lemma 5.4 implies

This and the previous inequalities give (5.19). The case s < 0 is obtained by tran-
sition to the dual spaces with respect to duality in HO(sn -I). •

3. The operator E(A) on the spaces HS(A,sn-I). We denote by HS(A,sn-l) the

space of functions on the sphere obtained by completing the set Coo (S n - I) with
respect to the norm

(5.20)

Equation (5.4) and proposition 5.1 imply that for each fixed A E C the norm
(5.20) is equivalent to the usual norm in the Sobolev-Slobodetskii space
Hs(sn-I).

Pro p 0 s i t ion 5.5. If A =1= i(k +n!2) (or A =1= -i(k +nI2)), where
k=O,l,···, then the map E(A):Hs(A,sn-I)~Hs+ImA(A,sn-l) (resp.
E(A)-I:Hs(A,sn-I)~Hs-ImA (A,sn-I)) is continuous. On every closed set go
lying in a strip I1m A I < h and not containing the points A = i (k + n I 2) (resp.
A = -i(k +n!2)), k = 0, I, ... , the estimate

IIE(A); HS(A, Sn -I) ~ H S+ ImA(A,Sn -I )11 ,;;;; c(go) (5.21)

(resp.

(5.22)

holds.

Proof. By proposition 3.1,

§5. The spaces HS(J. .,Sn -I). The operator Ef}..) 29

E(A)V = ~ ILm(A)Vmk Ymk, (5.23)

m,k

where

(A) = (_i)mziA r«m +nI2+iA)12). (5.24)

ILm r(m +n12-iA)/2)

Put A = o+i7". Since r(z) = r(Z), we have Ir«m +n12-iA)/2)1 =

I r«m + n I 2 + io + 7") I 2) I. Using a well-known asymptotic formula for a quotient
of r-functions (cf. [5]) we obtain on 'R as 1m +;0 I ~ 00,
I (A) I = 2-"1 r«m +nI2+;0-7")/2) 1= (5.25)
ILm r«m +n12+io+7")12)

= Im + i 0 + n 121 -"(1 + 0 ( Im +i0 I - I».

Because for A ¥= i (k + n 12) none of the functions ILm' m = 0,1, ... , can have
a pole, (5.23) - (5.25) imply that the map E(A):Hs(A,sn-I)~Hs+Im\A,Sn-l) is
continuous, and that (5.21) holds. The corresponding properties for E(A)-I are
verified in a similar manner. •
Combining propositions 4.4 and 5.5 gives

Cor 0 11 a r y 5.6. For ImA = 0 the operators E(A):HO(sn-I)~HO(sn-l)

are unitary.

§6. An analog of the Paley-Wiener theorem for the operator E(Ar)

Put S"t-- I = {x=(x',xn) E IR n : Ix I = 1, Xn >O}, S~-I = {x=(x',xn) E

II" n: Ix I = 1, Xn < O}. The main result in this paragraph consists, roughly speak-
ing, of the fact that the operator Ef}..) establishes an isomorphism between the set
of functions with support in the halfsphere S"t--I (or in S~ - I) and the set of
homogeneous functions of degree - iA - n 12 having an analytic extension in the
last coordinate to the lower (resp. upper) complex halfplane.
First we will prove several helpful propositions. Let w = (w',w n) E IR n,
Iwl = 1, w'¥=O. Put w = w'/lw'l, s = wn/lw'l. The map wt->(w,s)
.) This paragraph is used in Chapter 6 only.
30 Chapter 1.

transforms the sphere Sn -I with poles (0, -+-1) deleted onto the cylinder
Sn -2 X IR. We denote the inverse map by K. Clearly, w = (w',w n) = K(W,S) =
(W(I +s2)-1/2, s(1 +s2)-II2).

Lemma 6.1. Foru E COO(Sn-l)andlo. E IR the equation

(6.1)

f
+00
X e sgn(SI)(>"-P.-i/2)-rrI2p _>..(t)-llt li(p.->")-1/2(uoK)(~,t)dt
-00

holds, where p>..(s) = (1 +s2)(iA+nI2)/2 and E' 0) is 'the operator E(lo.) on the sphere
Sn -2 = {(w',w n) E sn -I: wn =O}', i.e. E' (lo.) is defined by formula (2.1), in which
n must be replaced by n - 1.

Proof. We denote by F' and IJ the Fourier transform with respect to the coor-
dinates x' = (XI, ... ,Xn-I) and Xn, respectively.
Let f E CO' (IR + ), J(lo. + in 12) = 1. Applying proposition 2.1 we find

4->w(lo.)u(-.{I) = MI~I->inI2->..Fx->~M;.?inI2->lx IJ(p.+inl2)u(-.{I), (6.2)

, -1 I

F x'->E' = Mi(n -1)/2-v-> If I E,p~' Ilfl (v)M lx' I->V+i(n -1)/2'

Putting (f,~n) = 1~1(1+s2)-1/2(w,s) we rewrite the equation F = IJF' in the

form

Substitute this in (6.2). Then

-I '
X Mi(n -1)/2-v-> lEI (1 +s2)-112 E,p->w(v)M I x' I->v +i(n -1)/2 X

X M;.? in I 2-> Ix If(p.+inI2)u(K(w,x n 1 Ix'I)).

§6. A Paley- Wiener theorem for E(A) 31

By the change of variables p = Ig1(1 +s2)-1I2 we are led to the relation

X E~-->w(v)Mlx'l-->v+i(n-l)/2 X

X M;J in I 2--> Ix If{JL + in 12)u (IC({1,xn I Ix' I)).

As the result of the new change x = p - ly this equation takes the form

(4-->w(A)U)OIC(W,S) = (6.3)

= P}..(S)M p-->i(inI2-A)p-lqJYn-->sM i(n1-l)12-v-->p X

Recall the inversion formula (1.2.6). We see that composition of the operators
M=;-->ppiV is integration with respect to v and multiplication by (277)-112.
Further,

M;jinI2-->IY I p-ip. = M=~-->p Iy Ii(p.+inI2).

The transforms M p-->_}.. and M=~-->p are inverses of each other. Represent
M IY 'I-->v+i(n-l)/2 as
the limit, as £~ +0, of the operator
MIY'I-->vly'l(n-l)12e-fIY'I. For a fixed £>0 the operators MIy'I-->ve-fIY'1 and
M =~-->p can be interchanged. Hence (6.3) implies

+00
X lim f 1/ I -iv +(n -3) /2 e -fly' I Iy IiA-nl2 X
f-->+O 0
32 Chapter 1.

x U(K(tt-,Yn I Iy' I»d Iy' I,

which takes, after the change of variable Yn = t Iy' I, the form
(E(A)U)OK(W,S) = (6.4)
+00 +00
= <~~0(2'1T~3I2P'\(S) _~ E'(v)dv _~p=~(t)u(K(tt-,t»dt X
+00
X f e-illy'ls-<ly'IIy'li(.\-v)-ll2dly'l·
o

We have (compare with (2.10»

+00
lim f eiW-<rriu-1!2dr = (6.5)
<..... +0 0

= f(ia + 1/2) 111 I- ia -I f2 e (-u +if2)7TSgn1jf2.

Applying this formula for a = A-v, 11 = - ts, (6.1) derives from (6.4) . •
Denote by X± the characteristic function of the halfaxis IR ±. For functions
given on IR (on sn-I), put (P±v)(t) = X±(t)v(t) (resp. (P±u) = X±(wn)u(w),
where W = (w',w n), Iwi = 1). For functions U E Coo(Sn -I) we define
(T(A)U)(W,o) = 'Js-lop.\(s)(Ey ..... w(A)u)(K(w,s», (w,o) E Sn -2 X IR.

L e m m a 6.2. For A E IR, U E coo(sn -I) the equations

(P±T(A)U)(W,O) = T(A)(P±U)(w,o) = (6.6)

hold.

Proof. Using (6.1) we write the operator T(A) in the form

(T(A)U)(W,O) = (6.7)
§6. A Paley- Wiener theorem for E(t..) 33

1
f
+00,
= --2 E (J.t)f(i(A-,u)+ 1I2)d,u X
(2'17") -00

+00
X f p =~ (t) t I I i(j.t-A)-1/2(uo/()(~,t)dt X
-00

X lim f
+00
ei(Js -( IS I +sgn(SI)(A-I'-i/2)7T12 s I Ii(j.t-A)-1 12 ds.
( ..... +0
-00

By (6.5),

lim f
+00
ei(Js -( Is I +sgn(sl)(A-I'-i/2)7T/2 1 S Ii(j.t-A)- I12 ds (6.8)
( ..... +0 -00

X (e(A-I')7T(sgn(J+sgnl)l2+i7T(sgn(J-sgnl)/4 +

+ e - (A -1')7T(sgn(J+ sgn I) 12 - i7T(sgn(J- sgnl) 14).

For sgno+sgnt = 0 the expression between brackets (on the right) vanishes,
while for sgno = sgnt it equals (e 7T(A-/l)+e- 7T(A-I'))sgno. Since
f(i(A-,u)+ 11 2)f(i(J.t-A)+ 1/2) = 2'17"(e7T(A-I'J +e-I'(A-I'J)-I, formulas (6.7) and
(6.8) imply (6.6) . •
We introduce the space H~(A,sn-l) as the closure of the set CO(S~-I) with
respect to the norm of HS(A,Sn -1). Denote by H~ (lR) the completion of the set
Co (IR:±:) with respect to the norm of HS(IR). Let, further, L 2(sn-2,B) be the
space of functions given on Sn-2 = {w=(W',W n)ES n - 1 :Wn =0} with values in
the normed space B, and let, moreover,

Pro p 0 sit ion 6.3. For A =¥= i (k + n 12), k = 0,1, ... , the operator

T(A): H"±ImA(A,sn -1) ~L2(sn -2,H"±ImA(IR)) (6.9)

realizes an isomorphism (algebraical~y and topologicalry).

Proof. Recall that by proposition 5.5 the map

34 Chapter 1.

E()..): H;ImA (A,sn -I) --7 HO(sn -I) is continuous. The volume element dw on
sn-I can be written in the form dw = (l+t 2)- n12 dtdw, where, as before,
w = (w',w n) = (1 + t 2)-1!2(w,t). This implies the inequality

IIT(A)U ;L2(sn -2 ,H - Im?(IR»11 ,,;:;; c Ilu ;H"± 1m?' (A,Sn -I )11. (6.10)

By lemma 6.2, for U E CO' (S~-I) and real i\.,

P::!::T(A)U = T(A)P::!::U = T(A)U. (6.11)

Using analytic extension, relation (6.11) can be extended to all complex

A = i (k + n 12), k = 0,1, .... Together with (6.10) this means that T(A) realizes
a continuous map (6.9), which clearly is a monomorphism.
We will show that the operator (6.9) is epimorphic. Applying proposition 5.5
we obtain that the operator T(A) - 1 = E(A) - IpA 0'":
L2(sn-2,H-ImA(IR»--7H-ImA(i\.,sn-2) is continuous. Put U = T(A)-Ij for a
given j E L 2(sn-2,H"±lmA(IR». If ImA";:;;O, then U E HO(sn-I), and we can
apply the projection p::!:: to u. From lemma 6.2 we obtain that
p::!::u = P::!::T(A)-Ij = T(A)-Ip::!::j = T(A)-Ij = u, i.e. U E H"±ImA(A,Sn-I).
Thus it has been proved that T(A) is epimorphic if ImA";:;; 0. If, however ImA > 0,
we choose a sequence {fk} converging to j in L 2(sn -2 ,H"± ImA (IR» such that
T(A)-IA E HO(sn-I). Since P::!::T(A)-Ijk = T(A)-Ip::!::jk = T(A)-Ijk. we have
U = T(A)-Ij = limT(A)-ljk E H"±lmA(A,sn-I) . •

In essence, the next assertion is well-known (the Paley-Wiener theorem; cf.,

e.g., [74]).

Pro p 0 s t o n 6.4. Let j E L2(sn -2,H,,+ (IR». Then

Tt-+0'"a~sCj(w,(J)eaT) = (0'"j)(w,s+iT) is,jorT";:;;O, a continuous junction with values
in L 2(sn-2,0'"H"+(IR». For almost all WE sn-2 the function s+iTt-+(qJf)
(w,s + iT) is ana?ytic in the halfplane T < 0, and
+00

f 1(0'"f)(w,s + iT) 12(1 +s2 +~)'ds ~ c(W) < 00,

-00

where c (w) is independent of T ~ 0. The inequality

f dw f
+00

1(0'"j)(w,s + iT) 12(1 +s2T2)' ds ~ C (6.12)

sn-2 -00
§6. A Paley- Wiener theorem for E(A) 35

holds, in which the constant c < 00 is also independent of 7' ~ o. Converse~y, if for
almost all W E S n- 2 the function ('fI/)( w,·) is anarytic in the halfplane indicated and
satisfies (6.12), then f E L 2 (sn -2,H+ (IR)). The proposition remains valid if H + is
replaced by H _ and 7' < 0 by 7' > 0, everywhere.
Let X+(,u) (resp. X-(,u)) be the subspace of HO(sn-l) formed by the elements
v satisfying the conditions: a) for almost all f / 1fiE sn -2 the homogeneous
function g = (g',gn) t-+ 1 g Il'v (V 1 g I), g' =I=- 0, can be analytically extended to the
halfplane Imgn > 0 (resp. Imgn < 0) (this is equivalent to the possibility of
analytically extending the function s t-+ w (w, s) = (1 + S 2 )1' 12 v 0 IC( W, s) to the
corresponding halfplane); b) the inequality
+00
f dw f 1w(w,s + i7') 12(1 +s2 +7'2)- Rel'-n12ds ~ C < 00
S·-2 -00

is fulfilled with a constant c which is independent of 7' ~ 0 (resp. 7' ~ 0).

The 0 rem 6.5. For A =I=- i (k + n / 2), k = 0,1, ... , the operator E(A) maps
H;ImA(A,Sn-l) onto X:::i=C-iA-nl2). This map is an isomorphism (algebraical~y
and topologicalry).

Proof. By proposition 5.5 the operator E(A):H;ImA(A,sn-l)~HO(Sn-l) is

continuous. By proposition 6.3 the membership v E H;lmAOI.,sn-l) implies that
T(A)V E L 2 (sn -2,H;lmA(IR)). By applying proposition 6.4 we find that the
function s t-+ ('!IT(A)V )(w,s) can be analytically extended to the halfplane Ims :§; 0
for almost all w. This means that the function
Wt-+(PA'!IT(A)V)OIC-1(W) = (E(A)V)(W), W = g/ Igl, can be analytically extended as
a homogeneous function of degree - iA - n / 2 to the halfplane 1m gn :§; O. Thus we
have a monomorphism E(A):H;lmA(A,Sn-l)~X+(-iA-nl2). The fact that
the map is epimorphic follows from propositions 6.3 and 6.4. •
 Chapter 2

The Fourier transfonn and convolution operators on spaces

with weighted nOnDS

In this Chapter we consider operators of the form A = Fi1x«P(~)Fy~~, where F is

the Fourier transform on IRm, and «P is a positively homogeneous function of com-
plex degree a, i.e. «P(t~ = ta«p(~ for ~ E IR m \ 0, t > 0. We study the problem of
continuity of the maps A:Hs(lRm) ~Hp -Rea (IRm) and A:Hp(lRm,lRm -n) ~
Hp - Rea (IRm ,lR m -n).

The space Hp (IR m), for nonnegative integers s, /3 E IR, is endowed with the
norm

while in the space Hp(lRm,lRm-n) we have the norm

rj~mlal± Ix(1)12(,8-s+lal)ID~U(X)12dXll!2,
=0

where n ';;;;m, x = (x(l),x(2)),

x(l) = (xj, ... ,xn), x(2) = (Xn+l, ... ,xm). For
m
n = m the space Hp(lRm,lR -n) coincides with Hp(lRn).
In §l we introduce, for arbitrary real numbers sand /3, the space Hp(lRn). In
§2 it will turn out that the Fourier transform F, at first defined on a suitable

*
dense subset of Hp(lRn), can be extended to the whole space Hp(lRn), and realizes
a continuous map F:H:B(lRn)~H:B(Hn) (if /3-s k+nI2, k = 0,1, ... ). For
k + n 12 < /3 - s < k + 1 + n 12 the operator F is at first defined on the set

{v E C't (IRn \ O)j.v(x)xYdx = 0, Iy I = 0,1, ... 'k}.

37
38 Chapter 2.

while in case {J - s < n 12 we may assume that F is given on C[) (Ikln \ 0). §3 is
devoted to the map A:Hp(lkln)~Hp-Rea(lkln). If {J-s=/=k+nI2,
{J-s =/= -k - Rea -n12 (k =0,1, ... ), then the operator A (at first given on
some dense subset of Hp(lkln» is continuous, hence its closure A is an operator
defined on all of Hp (Ikln). The representation for A obtained in §3 is the starting
point for the study of the algebras of pseudodifferential operators with discontinu-
ous symbols in the next Chapters. In §4 - §6 analogous problems for the map
A:H:e (Ikl m, Ikl m -n) ~ H:e - Rea (Iklm,lklm -n) are considered.

§1. The spaces H~(lkln)

Let, initially, s be a nonnegative integer and {J E Ikl. We denote by Hp(lkln) the

completion of the set C[) (Ikln \ 0) with respect to the norm

(Ll)

This norm is equivalent to the norm

where, as before, r = 1x I. Using the Parseval equality (1.2.7) for the Mellin
transform, we obtain yet another equivalent norm:

Proposition 1.5. I and the obvious truths

s
IAI 2j(l+m 2y-j ~(l+ IA12+m 2y ~c ~ IAI 2j(l+m 2y-j
j =0

imply that the norm (Ll) is equivalent to the norm

(1.2)

The righthand side of (1.2) makes sense for arbitrary real s, cf. (1.5.20). For all
 39

s,{3 E IR the space Hfi (IRn) is defined as the completion of withCo (IRn \ 0)
respect to the norm (1.2). The following assertions are readily derived from the
definitions of the norms.

Pro p 0 sit ion 1.1. The operator of multiplication by a homogeneous function

of degree a, IP(~) = pUf(I/I), realizes a continuous map Hfi(lR n ) ~H~-Reu(lRn) if
and only iff is a multiplier in Hs(Sn -I) (i.e. if the operator of multiplication ~Y f is
continuous on Hs(Sn -I)).
It is well-known that for 2s > n - I the space Hs(Sn -I) is a ring under ordi-
nary multiplication. Hence for 2s >n -1 the operator IP:Hfi(lRn)~Hfi-Reu(lRn)
is continuous only if f E Hs(Sn -I). (A description of the multipliers in the
Sobolev-Slobodetskii spaces can be found in, e.g., [37].)

P r o p 0 s i t i o n 1.2. For arbitrary pOSitive l) the inclusion

H~t~(lRn) C H~(lRn) holds; the inequality Ilu;H~(lRn)11 ~ Ilu;H~t~(lRn)11 holds
also.

Pro p 0 sit ion 1.3. The spaces Hfi(lR n) and H=fi(lR n ) are the duals of each
other under the extension of the scalar product in L 2(lR n) = Hg (IRn).

§2. The Fourier transform on the spaces Hp(lRn)

For u E Co (IRn \ 0) we put

(F;=lu)(r,cp) = ,b-
2'1T
V
f
ImA=(J
ri(in/2+A)~~<p(A)±lu(in/2+A,I/I)dA; (2.1)

the operator F (J (resp. F;; I) is defined under the condition that (J =1= k + n / 2
(resp. (J =1= - k - n / 2), where k = 0, I, . . .. It is obvious that for
(J =1= +(k + n / 2) the operators F (J and F;; 1 are the inverses of each other, and Fo

coincides with the Fourier transform (proposition 1.2. I).

Pro p 0 sit ion 2.1. Let (J and (J' be real numbers subject to the inequalities
(J< n / 2, P + n / 2 < (J' < P + I + n / 2, where p is a certain nonnegative integer. Then
the formulas
40 Chapter 2.

(F:=f!u)(x) = (2.2)
1 (+i)I'Y1 +2
(2'IT)
Cn-l)12
.
p
~
I'YI =0 -y. S"-l
f -+-1
,X'Y y'Y u (Y)4J;+(F±o'u)(X)

hold, in which the upper (lower) sign must be chosen everywhere. Moreover, for arbi-
trary h < 0,

(F:=f!u)(x) = 0 (I x Ih) as x -HO, (2.3)

(F:=f!,u)(x) = O(lx IP+I) as x~o. (2.4)

Thus, the formulas (2.2) give for x ~ 0 the asymptotic behavior of F:=f! u, and for
x ~ 00 those ofF~!,u.

Proof. We verify the assertion for, e.g., Fou. Since u E CO' (~n \ 0), the func-
tion A...... U(A + in 12,') is analytic in the whole A-plane and decreases faster than
any power of IAI in every strip lImA I < N. This and inequality (1.5.21) imply
(2.3) and the estimate

(2.5)

Further,

res Ey->cp(A)U(A + in 12,1/;) IX=iCk +nI2) (2.6)

= (~it+1 £ 'Yu d
(2'ITin + 1)12 ~ "" fy
I'YI =k I'R"
(y)~.
Therefore, replacing in (2.1) the integration curve ImA = a by ImA = 0' we are
led to (2.2). If a" E (p + 1 + n / 2,p + 2 + n / 2), then the difference F o'U ~ F o"U is
equal to 2'ITi Ix IP + 1 times the residue (2.6) for k = P + I. Together with (2.5) (in
which a' must be replaced by 0") this leads to (2.4) . •

Pro p 0 sit ion 2.2.

1) Let f3 ~ s =1= k + n /2, k = 0, 1, . . . . Then the maps

Ff3-s:Hft(~n) ~ Hf(~n), Fs--.!f3:Hft(~n) ~Hf(~n) (2.7)

are continuous. Under the condition f3 ~ s =1= ~k ~n/ 2, k = 0, 1, ... , each of

the maps (2.7) is an isomorphism.
§2. Fourier transform on the spaces Hfi(lJin) 41

2)Let /3-s = k +n12 and let Hfi,k(R n) (resp. Hf k(Rn)) be the subspaces of
Hfi(R n) (resp. Hf(R n)) obtained by closure of the set of functions from
Co (Rn \ 0) satisfying the conditions

f v(r,q,)Ym,q(q,)dq, -
sn-l
0, m ~k, m _ k(mod2). (2.8)

Then every map

F ,B-s :Hfi,k(R n) ~ H~k(Rn), F s-~,B:Hfi.k(Rn) ~ H~k(Rn)

is an isomorphism.

Proof. 1) It suffices to convince ourselves of the continuity of the operators

(2.7). By (2.1), ~(-A+inI2),·) = E(A) u(A+ini2,·). Using proposition
1.5.5 and the definition of the norm (1.2), we are led to the inequality
IIF,B-su ;Hf(Rn)11 ~ cllu ;Hfi(Rn)ll. Continuity of the first map (2.7) has been
proved; continuity of the second is verified similarly. 2) The restriction of E(A)
(resp. E(A)-I) to the lineal et
does not have poles on the line ImA = k +n12
(resp. ImA = -k -nI2); moreover, et
is invariant under these operators (see
proposition 1.4.3). Therefore we can reason similarly to the proof of assertion
1). •

Let Mp be the set of functions v from Co (Rn \ 0) subject to the condition

v(i(n +q),q,) 0, q = 0, ... ,p.

Pro p 0 sit ion 2.3. The set Mp is dense in Hfi (Rn) for any nonnegative
integer p, and for /3-s *- k +ni2, k = 0,1, ....

P r o o f . We first verify that M 0 is dense. For an arbitrary function

v E Co (Rn \ 0) we introduce the sequence {vd k= 1 for which
vk(r,q,) = ev(kr,q,) if /3-s >n12 and vk(r,q,) = k-nv(k-1r,q,) if /3-s <nI2.
Clearly v -Vk E Mo. Further, vk(A+inI2,') = k:::':::(iHnI2)v(A+inI2,') if
/3 - s § n 12. This and (1.2) imply that Vk ~ 0 in Hfi (Rn). We have proved that
M is dense. We now assume that Mp -I is dense also. It suffices to convince
0
ourselves that every function in Mp -I is the limit of a sequence of functions from
Mp. Let w E M p _ l , wk(r,q,) = k:::':::(n+p)w(k:::':::lr,q,) for /3-s §p +nI2. It
42 Chapter 2.

remains to note that W-Wk E Mp and that Ilwk;Hp(Rn)II-,>O . •

Note that proposition 2.3 remains valid for /3-s = k +n12 too, we only need
modify the definitions of the sequences {Vd,{Wk} (d. [44]).
We denote by ~ the set of functions from Co (Rn \ 0) satisfying the condi-
tions

jv(x)x'Ydx = 0, /y/ = 0,1, ... ,po (2.9)

The 0 rem 2.4.

1)Let /3-s =1= k +nl2, k = 0,1, ... , and let F be the Fourier transform defined
on ~for /3-s E (p +nI2,p + 1 +nI2) and on CO'(R n \ 0) for /3-s <nl2.
Then the domain of definition, 6j)(F), of F is dense in Hp (Rn). For all u E 6j)(F)
the estimate

(2.10)

holds. The operator F (the closure of F), defined on all of Hp(Rn), realizes a con-
tinuous map F:Hp(Rn) -,> Hf(R n). All assertions made remain valid if /3 and s
are interchanged and if F- 1 is written instead of F. The formulas F= F fi-s'
F- 1 = F.B~s (cf (2.1)) hold for the operators F and F- 1, which are defined on
Hp(Rn) and Hf(R n), respectively. Under the additional restriction
/3 - s =1= - (k + n 12) the operators F and F- 1 are inverse to each other.
2) Let /3-s = k+nl2 and let F be the Fourier transform defined on the set offunc-
tions from CO' (Rn \ 0) satisfying (2.9) for I y I = 0, ... ,k -1 and (2.8) (if
k = 0, condition (2.9) can be omitted). The closure of this set of functions coin-
cides with the subspace Hp,k(R n), and the map F:Hp,k(R n) -,> Hf,k(R n) is an iso-
morphism. The operators F and F fi-s coincide on Hfi,k(R n). All assertions
remain valid if /3 and s are interchanged and F- 1 is written instead ofF.

Proof. 1) That 6j)(F':::I) is dense follows from proposition (2.3). On 6j)(F) (resp.
6j)(F- 1» the operator F (resp. F- 1) coincides with F fi-s (resp. F .B~s) (proposition
2.1). The estimate (2.10) follows from proposition 2.2.
2) Note that the denseness of the set of functions on which the Fourier
transform F is initially defined follows from the proof of proposition 2.3. The
remaining assertions are verified as is done in the first part of the theorem . •
§2. Fourier transform on the spaces H;HRn) 43

Rem ark 2.5. If the numbers fl-s, fl'-s' belong to distinct intervals
(-oo,nI2), (q+nI2,q+l+nI2), where q = 0,1, " ' , then the operators F and
F' defined as the closure of the Fourier transform on Hfi(R n) and H~'(Rn) are,
generally speaking, distinct on functions u from Co (Rn \ 0). The equality
Fu = F'u holds if u E <JRp where p = max{[fl'- s'- n 12], [fl- s - n I 2]}.

§3. The convolution operator on the spaces Hp(Rn)

Let (> be a homogeneous function of (complex) degree a on R n \ 0 which is
infinitely differentiable on the sphere Sn -I. We consider the operator

A = F-I(>(~F. (3.1)

The aim of the present paragraph lies in clarifying the problem of the bounded-
ness of the map A:Hfi(Rn)~Hfi-Reu(Rn).

Pro po sit ion 3.l. Let Rea> -nI2. Then the representation

(Au)(r,cp) = _ b-
2
VL.'TT
+00
f r i (A+iU+inI2)E
-00
w.....<j>(X+ia)-1 X (3.2)

holds for the operator (3.1), defined on Co (JR n \ 0).

Proof. By (l.28),

pU(>(w)(Fu)(p,w) = _1_ f pi(inI2-A)(>(w) X (3.3)

...,!2;lmA=Reu

X E(X-ia)u(A-ia + in I 2,)dJ\.
Now note that in the strip between the lines ImX = 0, ImX = Rea the integrand
is holomorphic, since Re a > - n I 2, while the poles of the function IL 1-+ E(p.) are
located at IL = i(k+nI2), k = 0,1, .... The membership u E Co(Rn\O)
implies that the function X1-+ u(A + in I 2,) is rapidly decreasing as X ~ 00 in any
strip IImX I < h. Together with (1.5.21) this allows us to replace in (3.3) the
integration line ImX = Rea by the line ImX = O. By now applying the inverse
Fourier transform F- I = FOI (cf. (2.1» to (3.3) we obtain
44 Chapter 2.

Put J.t = A - ia and then replace the line 1m J.t = - Re a by the line 1m J.t = 0. As
the result we obtain (3.2). •

Pro p 0 s i t ion 3.2. Let 7"=I=k+nl2, 7"=1= -k-Rea-nl2, where

k = 0,1, ... , and let A be the operator defined on COO (jRn \ 0) by (3.1) for
Re a > - n I 2. The following representation holds for this operator (or for its analytic
extension in a) when a =1= -I - n, where I = 0, 1, ... ."

(Au)(r,cp) = _1_ f r i(A+ia+inl2) X (3.4)

y'2;ImA=T

X f (xw+io)-n-a-kcp(w) ~ w~ dw fyYu(y)dy +
S"-l lyl =k y. IRI"

[-T-nI2- Rea 1
- ~ (-l)kr(n+a+k)X
k =0

X ~ x~ f
fU(y)dy (-YW+io)-n-a-kcp(W)WYdW},
lyl =k y· lRI " S"-l

where there is no summation if the corresponding upper limit is negative. If a is a

nonnegative integer and cP is a homogeneous polynomial of degree a, then A is a
homogeneous differential operator of order a. For such an operator a representation
(3.4) in which only the first righthand term occurs holds.

P r o o f . The integrand in (3.2) has singularities at A = i (k + n 12),

A = - i (k + a + n I 2), k = 0,1, ... , only, and these singular points are simple
poles. Changing in (3.2) the integration line and writing out explicitly the residues
leads to (3.4). If a is a nonnegative integer and cP is a homogeneous polynomial of
degree a, the residues are annihilated (see proposition 1.4.3) . •
§3. Convolution operator on the spaces Hp (Rn) 45

Now we consider the case a = -I - n. In general, for such an a formula (3.4)

cannot be written down since certain terms between the brackets have a pole at a.
If we additionally require (l to satisfy

f (l(w)wYdw =
S·-I
° (3.5)

for all multi-indices y such that Iy I = 0, ... ,I, then the residues of the terms in
(3.4) are annihilated for a = -I - n. As the result we obtain a representation for
A also in case a = -I - n. It coincides with (3.4), in which we must substitute
a = -1- n and transform the terms for 0,,;;;;; k ,,;;;;; I taking into account (1.2.5) and
(1.4.3).
Assume that /3-s =1= k +nl2, /3-s =1= -k - Rea -nI2, k = 0, I, ... , and
denote by ~(f3,s,a) the set of functions u from Co
(Rn \ 0) such that
u(i(q+n),q,)_O for q = 0, ···,[/3-s-nI2] and u(-i(k+a),q,)_O for
q = 0, ···,[s-/3-Rea-nI2] (if one of [/3-s-nI2], [s-/3-Rea-nI2] is
negative, the corresponding condition is omitted). In the following theorem, A
denotes the operator (3.1) or the operator obtained from (3.1) by analytic exten-
sion in a.

The 0 rem 3.3. Let /3 - s =1= k + n 12, /3 - s =1= - k - Re a - n 12, where

k = 0,1, .... Then on the set ~(f3,s,a) the following representation is valid for A:

(Au)(x) = _1_ fr;(A+;a+;n/2)Ew~.p(A+ia)-1 X (3.6)

Y;);;ImA.=P-s
X (l(W)E8~w(A)u(A+inI2,8)dA.

This representation can be written as

(3.7)

here Fp-s (resp. F.8~s + Rea) is the closure of the Fourier transform (resp. the
inverse Fourier transform) on Hp(Jl n) (resp. H~_Rea(Rn») described in theorem 2.4.
On ~(f3,s,a) the estimate

(3.8)

holds. The set ~(f3,s,a) is dense in Hp(R n), hence A (the closure of A) is defined on
all of Hp(Rn) and realizes a continuous map A:Hp(Rn)~Hp-Rea(Rn). The
46 Chapter 2.

representations (3.6) and (3.7) hold for A.

Proof. Formula (3.6) follows immediately from (3.4) and the definition of
1Jl(ft,s,a). We show that (3.6) can be rewritten as (3.7). By (3.6), (2.1) and
F- 1 = F.B~s+Rea, which holds on H~-Re(J(lRn) (theorem 2.4), we have
Au = F.B~s+ReaV, where v(-l\.-ia+inI2,·) = «IIE(l\.)u(l\.+inI2,), Iml\. = /3-s.
Inverting the Mellin transform (cf. (1.2.6)) and taking into account (2.1) we obtain
v(P,w) = pa«ll(w)(Fp_su)(P,w). This implies (3.7). In order to obtain (3.8) it
suffices to use (3.7) and to recall that each map Fp_s:Hp(JRn)-?Hff(JR n),
«II:H~(JRn) -?H~ -Rea(JR n), F.B~s + Rea: Hff _Rea(JR n) -? Hp -Rea(JR n) is continuous
(theorem 2.4, proposition 1.1). The denseness of 1Jl(ft,s,a) is verified similarly as in
the proof of proposition 2.3. The representations (3.6), (3.7) for A follow from the
corresponding formulas for A and the continuity of the three maps just listed. •

Rem ark 3.4. The relation between the operators (3.1) and A for a = -/ - n
can be established under the additional conditions (3.5) only. However, A is itself
defined without these conditions. Unless stated otherwise, A is considered for
arbitrary functions «II.

R e ill ark 3.5. If /3-s =1= k +nI2, /3-s =1= -k -Rea-nI2, the operator
(3.6) realizes a continuous map Hp (JR n) -? Hp - Rea (JR n) if and only if the function
Sn-I 3 ()t-+«II(fJ) is a multiplier in the Sobolev-Slobodetskii space HP(sn-l) (cf.
proposition 1.1). In particular, for 1/31 > (n - 1) I 2 the condition
«II E HIP I (Sn -I) is necessary and sufficient for the operator (3.6) to be continu-
ous.

Rem ark 3.6. The operators (3.6) and F- 1 «II(~)F coincide also on a set larger
than 1Jl(ft,s,a); it is only necessary that the terms between the brackets in (3.4)
cancel each other.
 47

§4. The spaces Hp(Rm ,R m -n)*)

Let 1";;;;'n ";;;;'m, x(l) = (Xl, .•. ,xn ) ERn, X(2) = (Xn+l, •.. ,Xm ) E Rm - n,
X = (x(1),x(2» E Rm. Put CQ(Rm,R m - n) = CQ (R m \{x=(x(1),x(2»:
X(l)=O}). We first assume that s is a nonnegative integer and that f3 E R, and we
introduce the space Hp(Rm,R m - n) of functions on R m as the completion of the
set CQ(Rm,R m - n) with respect to the norm

(4.1)

For n = m the space Hp(Rm,R m -n) coincides with Hp(Rm).

We denote by it = §"u the Fourier transform of a function u with respect to the
X (2) = (Xn+l, ···,xm ) variable, and we put Z = Ix(l) 111,

V(Z,l1) = 111ls-p-nl2it(zlll1l,l1). The norm (4.1) is equivalent to the norm

Unless otherwise said, all spaces occurring in the sequel are defined as comple-
tions of COO (Rn \ 0) or COO (Rm,R m -n) in corresponding norms.
We introduce the space tti:8(Rn) of functions on Rn, by taking

Taking into consideration the definition of the norm in H~ (Rn) (see (1.1» we
find that an equivalent norm in ttip(Rn) is given by

We now define the space Hp(Rm,Rm -n) for an arbitrary nonnegative index s.
For functions u with support in the set {x=(x(l),x(2» E R m ;2- l < Ix(I)1 <2}
the norm in H:8(R m,R m -n) will be taken to be equivalent to the norm in the
*)§4 - §6 are not used in the sequel.
48 Chapter 2.

Sobolev-Slobodetskii space HS(lRm). As is well-known [28], for such functions u

the norm in HS(lRm) has the form

V.laf=oID~U(X)12dX+ la~[sJ i ,dx(2) //ID;u(y,x(2)

R + (4.2)

- D~u(z,x(2) 12 1y - z I-n + 2([sl-s)l{vdz +

We assume that also for arbitrary s the norm in H~(lRm,lRm-n) is homogene-

ous of degree f3-s+mI2. This means that the norm is multiplied by tf3-s+mI2
as the result of a similitude transformation x .... tx in IRm. Using (4.2) and homo-
geneity of the norm, we define the norm in H:e
(IR m, IR m -II) for functions 1I from
err m
(IRm,lR -II) with arbitrary support by

(4.3)

+ f dx(2) f f ~ Ily I f3 D;u(y,x(2)-lz If3D~u(z,x(2)12 X

R"'-' R' R' lal =[s]

Let t E "), f :;II; O. We will assume that the norm of a function

CO' (\R m -
u(x(l),x(2) = f(x(2)w(x(l) in H~(lRm,lRm-lI) is equivalent to the norm of w in
0:HlR n ). Substituting the expression for u in (4.3) we are led to the following
expression for the norm in 0-p(1R"):

(4.4)
 49

We rewrite the expression for 11·;6ip(lR n )11 in terms of the Mellin transform.
Replace (4.4) by the equivalent norm

[IIW ;Hj,(R")II' +} ~olIW ;Hj, (R")II'] u,

Recalling (1.2) we obtain

Ilw;6i p(lR n)11 2 = (4.5)

f Ilw(A + inl2, ·);Hs(A,sn -1)11 2 dA +

Imh=,8-s
[s 1
+ ~ fIlw(A+inl2,);HJ(A,Sn-l)1I 2 dll..
J =0 Imh=,8- J

Note that for /3 E IR, s ~ 0,

(4.6)

Pro p 0 sit ion 4.1. Let v E CO' (IRn \ 0), s, /3 E R s ~ 0. For arbitrary
I E [O,s 1the inequality

(4.7)

holds, with a constant c independent of v.

°
Proof. Let p be an arbitrary number subject to the condition
p > max {/32, (f3 - s i }. Then the inequality Re (A2+ p) > holds for
/3-s ,,;;;; ImA";;;; /3. Put U(A) = (1-<5)-U>'+,8)/2(1 +A2 +p -<5)(i>'+,8)/2, where
/3-s ,,;;;; ImA";;;; /3 and <5 is the Laplace-Beltrami operator. For ImA = /3 the opera-
tor U(A) is unitary on L 2 (Sn -I), while for ImA = /3-s the map
U(A): HS(A,sn -I) ~ Hs(sn -I) is an isomorphism; moreover,
IIU(A)II+IIU(A)-III,,;;;;const. This implies that the righthand side of (4.7) is
equivalent to
50 Chapter 2.

f IIU(A)V(A + in 12,);Hs(sn -I )11 2dA + (4.8)

ImA= f3-s

+ f IIU(A)v(A+inI2,-);L 2(Sn-I)11 2dA.

ImA=f3

Let A~/(A) E coo(Sn-l) be a function which IS analytic in the strip

f3 - S < 1m A < f3 and continuous in the closure of this strip. Assume that the ine-
quality

f 11/(A);Hs (sn-I)11 2dA+ f 11/(A);L 2 (sn-I)11 2 dA< 00

ImA=f3-s ImA=f3

holds. Putting z = x+i(f3-(l-8)s), 0<0< 1, we can write/(z) as a Poisson

integral:

(z) = sin7TO f[ [(o+i(j3-s» + [(o+i{3) jdO =

I 2 R cosh7T(x -S)-COS7TO cosh7T(x -O)+COS7TO -

where 10 and II are the traces of I on the lower and the upper edge of the strip,
respectively.
As is well-known, the space H'(sn -I) is interpolating with respect to the pair
(H s (sn-I),L 2 (sn-I», Hl(sn-I) = (H s (sn-I),L 2(sn-I»O,2 for 1= (l-O)s,
and the inequality

III (z);H'(sn -1)11 2 ~ (4.9)

~ c (I I}J{) (0, -)*10 ;Hs (sn -I )11 2 + II!LI (0, -)*/I;L 2(Sn -I )112)

holds.
Substitute in (4.9) f (z) = U(z)v(z + in 12, -) and subsequently integrate over
the line Imz = 13-/ = f3-(1-0)s. Since the kernels !L/B,·), j = 1,2, are
smooth and rapidly decreasing, the quantity

f IIU(z)v(z + in 12, ');H'(sn -1 )11 2 dz

ImA=f3-'

can be bounded from above by (4.8). Taking into account the properties of the
operator-function U thus leads to (4.7) . •
This proposition allows us to define an equivalent norm in Sfi(/Rn) by
 51

§5. Transversal operators and their special representations

1. A transversal operator. Let «P be positively homogeneous function of degree a
on IR m \ 0 that is infinitely differentiable on Sm -I. We denote the operator of
multiplication by «P by the same letter «P. Consider the operator

Au = F-I«pFu, (5.1)

where F is the Fourier transform on IR m and u E Co (Rm,lR m - n). We denote, as

before, by u=
§it the partial Fourier transform with respect to the variables
Xn+l, ···,Xm . We put 0 = r 2)/lr2)1, X = x(l)lr 2)1, Y = y(l)lr 2)1. By
using the fact that «P is homogeneous we obtain that the quantity
Ir2) l- a(A0(X Ir2) I-I, r2) is equal to
-I-n jeiXT/«P(T/,O)dT/je-iYT/u(Y Ir2) 1-I,r2)dy.
(217") IR" R"

In the sequel we will denote X and Y by x and y (E IR n), and u(Y I r 2) I - 1, r 2)

and Ir2) l- a (Ar;)(Xlr 2) 1-I,r2» by v(y) and (A(8)v)(x). Thus,

(5.2)

For every 8 E sm -n -1 we call the operator A(8) a transversal operator

(corresponding to the operator (5.1».

2. Special representations of a transversal operator. In the investigation in §3 of

the boundedness of a convolution operator in the scale Hp
(IRn) we needed a spe-
cial representation of such an operator (cf. (3.4». Our aim in the present section
is to derive corresponding representations for A(8). We will need them in order to
give estimates in the scale 0p(lR n ).
Let v E Co (IRn \ 0) and T =1= k +n!2, k = 0,1, .... Put
52 Chapter 2.

J ,,(t, W) = j e -iP.I~-4w(p,)V(p,+ in 12,tf;)dp.. (5.3)

Imp.="

L e m m a 5.1. Let 1/- E Coo(R) with, moreover, 1/-(t) = °for t> 112 and
1/ - (t) = 1 for t < - 112. The integral
+00
j e iN 1/_(t)iP(e 1w,O)J,,(t,w)dt, Iwl = 101 = 1, (5.4)
-00

defined for ImA < n12, has an analytic extension to the whole A-plane as a mero-
morphic function; only the points A = i (p + n 12), where p = p ", p" + 1, ... , and
p" = max {[T-nI2]+ I,O}, can be Singular. At these points the integral (5.4) can
have poles of order one only. The residue at A = i (p + n 12) is equal to
(_i)ly+ll
(n-l)12 ~ wY+K " 8K iP(O,0) jyYv(y)dy, (5.5)
(217') IYI+IKI=p y./(. IR"

where (8 K iP)(x,z) (8~iP)(x,z). As A tends to 00 while remaining in some strip

{A: 11m'\' I < h}, then the integral (5.4) (its analytic extension) decreases faster than
any power of IA I·

Proof. Note that J,,(t,w) = y'2; pnI2(F"v)(p,w), where p = e l and F" is

defined by (2.1). By (2.2) we have

1 M Y
J (t w) = ~ (-i)IYI e(nl2+ IYI)I~ X (5.6)
" , (2 17' )(n - 1) I 2
lyl"""=p, y.,

X jyYv(y)dy +J".(t, w),

IR"

with T E (M +nI2,M + 1+nI2). Expand iP by Taylor's formula and rewrite

(5.4) as
1 M+N (-i)IYlwY+K X
-1)12 ~ ~ (5.7)
(2 )(n
17' p=p,IKI+lyl=p y!/(!

+00
X j y Yv (y)dy-3 K iP(O,0) j e(iHnl2+ IKI + IYI)I1/_(t)dt+
IR" - 00
+00 +00
+ j eiAI1/_(t)iPN+l(t,W,O)J,,(t,w)dt+ j e iAI 1/_(t)iP(e 1w,0)J".(t,w)dt.
-00 -00
§5. Transversal operators and special representations 53

In view of the estimates <PN + I (t, w,1J) = 0 (e(N + 1)1),

JT'(t,w) = O(e(N+1+nI2)1), JT(t,w) = O(e(P,+nl2)l) as t~ -00 (cf. (2.4)), the
last two terms in (5.7) are functions analytic in the halfplane
ImA<ho = min{N+p",M}+I+nI2 and rapidly decreasing as IAI ~OO,
I 1m A I ~ h < h o. In order to convince ourselves of the required behavior of the
remaining terms of (5.7), it suffices to apply the formula for integration by parts
to the integrals with respect to t . •

Pro po sit ion 5.2. Let -n12 < T+ Rea, q+nl2 < T < q + I +nI2, where q
is a nonnegative integer. Then we have for the operator A(IJ), defined by (5.2), and
v E Co (IRn \ 0) the representation

(A(lJ)v)(x) = (5.8)

= _1_
,f2;; f r i (inI2+,\+iu)E f\
w-..tj> \"
+ ia)-I<p11--->'\ (w , 0) X
'7TIm'\=T

Here r = I x I, cp = x I Ix I,

<P/L-..,\(w,O)w(}L,w) = (5.9)

1 +00 .
= 2'7T f
e'N<P(w,e-IlJ)dt f e-i/Llw(}L,w)dp"
- 00 Im/L=T

Z (x 0) =
Y'
(-i)lyl
(2'7T)(n + 1)/2",1
/.
{f
my=h
r i (inI2H)E
w--->tj>
(A)-lwYdA X

+00 +00
X f '1'/+ (t)e(iHnl2+ lyl)I<P(elw,lJ)dt+ f r i (inl2+'\)E(A)-l wYdA X
-00 -00

the '1'/ ~ are nonnegative functions from Coo (IR), '1'/- (t) = 0 as t > 112, '1'/- (t) =
as t < - 112, '1'/ + (t) = 1 - '1'/- (t); and h is an arbitrary number satisfying
h >max{-nI2, 1'(1 +Rea+nI2}. Under the conditions -nI2<T+Rea,
T<n12 the formula obtained from (5.8) by discarding the terms containing Zy
54 Chapter 2.

holds.

Proof. Applying proposition 1.2.1 we rewrite (5.2) in the form

(5.10)

+00

X f eiAt.p(e1w,O)J oCt, w)dt.

-00

Our aim is to replace J 0 by J T (cf. (5.3» and the line Imi\ = 0 by Imi\ = T.
Represent (5.10) as the sum I + (x) + I _ (x), where the I ± are equal to the right-
hand side of (5.10) with .p replaced by "1±(t).p. By the equation
JT(t,w) = yi2;pn12 (FTv)(p,w) and estimate (2.3), the inner integral in 1+, con-
taining "1 +, is an analytic function on the whole i\-plane and decreases faster than
any power of Ii\ I in every strip I1m i\ I < N. Hence for T + Re a > - n 12,

I +(x) = \12 f r i (inl2+A+ia)E(i\+ia)-ldi\ X (5.11)

(2'1T) ImA==T

+00

X f e iAt "1+ (t).p(w,e-l())Jo(t, w)dt.

-00

Note that (2.4) implies that the integral (5.4) IS analytic in the halfplane
Imi\<q+l+nI2 (under the condition T>q+nI2). For
-nI2<T+Rea <q+l+nI2, by using (5.6) (in which we must put T = 0,
T' = T) we obtain

(5.12)

f
+00
X eiA-ia)I"1_(t).p(w,e-I())JT(t,w)dt+
-00

+ I f
( 2'TT)(n + 1) 12 ImA==O
r i (inI2+A)E(i\)-Jdi\ X

Substituting in (5.11) the expression for J 0(1, w) from (5.6) (where, again, we put
§5. Transversal operators and special representations 55

T = 0, T' = T) and using (5.12) leads to (5.8).

For T + Re a > q + 1 + n 12 the poles of the integral (5.4) fall within the strip
0< ImA < T+ Rea. However, proposition 1.4.3 and formula (5.5) imply that the
poles disappear after an application of E(A) -1 to this integral. Hence (5.12)
remains valid also under the single restriction - n 12 < T + Re a. The proof is
finished by combining (5.11) and (5.12); in advance it is necessary to replace
J oCt, w) by (5.6), in which T = 0, T' = T . •

Pro po sit ion 5.3. Let T<nl2, -p -I-nI2<T+Rea < -p -nl2, p a

nonnegative integer. Then we have for the operator (5.2) and v E Co (~n \ 0) the
representation (5.8) in which the sum of the terms containing Z-y must be replaced ~Y
p
~ x 'I (v, t-y(-,O).
I'll =0
Here (j,g) is the scalar product in L2(~n) and t-y(x, 8) coincides with the righthand
side of (5.9) in which <I> must be replaced by <I>.

Proof. Lemma 5.1 implies that the integral

f
+00

eiN<I>(etw,O)Jo(t,w)dt
-00

is an analytic function in the halfplane ImA<nI2. Recall that Jo(t,w) = Jr(t,w)

for T<nI2. We will first assume that a is such that -nI2<T+Rea <8-nI2,
with 8 a sufficiently small positive number. Then the equation obtained from (5.8)
by discarding the terms containing Z-y, I"Y I = 0, ... ,q, holds. This and (1.4.2)
imply that

f
+00

X e iN <I>(w,e- t 8)Jr(t,w)dt+
-00

f f
+00

-----;-=---::-::- i dw etnI2<I>(et w,8)J r(t, w)dt.

(2'7T)(n + 1) 12 s" ~ 1 _ 00

Clearly, equality also holds under the condition 8 - I - n 12 < T + Re a < - n 12. If
56 Chapter 2.

this condition is satisfied, we may again replace the line 1m;\. = T-~ by
1m;\. = T. Repeating in a suitable manner these considerations we are led to the
formula (for T< nl2, -p -l-nl2 < T+ Rea < -p -nI2):

(A(O)v)(x) = 1
(27T)312 Im'\=T
f
r i (inl2+'\+ia)E(;\'+ia)-ld;\. X

+00
X _~ ei.\t<P(w,e-10).!T(t,w)dt- (27Ti n\I)/2 X

f wYdw f
P 'lyl+1 +00
X ~ I , xY e(nl2+IYI)I<P(e 1w,O)Jo(t,w)dt.
1 y =0
1 "y. s· -1 - 00

Consider the integral

f w dw f e(nl2+ Iy 1)/'/)+ (t)<P(e w,O).! o(t, w)dt

Y l

f w dw f e(nl2+ I y I)/,/)+ (t)<P(e1w,O)dt X

Y

X f e- iIlI E{JL)v{JL+inI2,')dp"
Imll= -h

where h is chosen to satisfy the inequality h > max { - n 12, I"Y I + Re a + n 12}.
Changing the order of integration, we rewrite this expression in the form

f v (x)dx f r i (inI2- Il )Ew--->y{JL) X

R' Imll=-h

The expression with '/)_ instead of '/)+ can be written similarly; we must only
put h = 0 in this case. Recalling (1.2.1) and (1.4.1) we obtain
_ -1--
Ew--->y{JL)1 (w) - Ew--->yGU) I (w).

It remains to combine the equations given. •

Using a reasoning close to the above proves

P r o p 0 s i t i o n 5.4. Let -p-1-nI2<T+Rea<-p-nI2,

q + n 12 < T< q + 1 + n 12, p and q nonnegative integers. Then we have lor the
§5. Transversal operators and special representations 57

operator (5.2) and v E Co (II~n \ 0) the representation (5.8) in which at the righthand
side we must add the sum

where
(-i)lyl
J ri(inf2+P.)E(p.)-lwYdt-t X (5.13)
Imp.= -T

J e i p.t+(nf2+ lyl)tq,(e t w,8)dt.

+00
X
-00

3. Some properties of the special representations for A(9).

Pro p 0 sit ion 5.5. The functional

CD(Rn\O) 3 v~ Jv(y)yYdy (5.14)

R"

is continuous on ~p(Rn) if and only if I'Y I +n12 < {3 < s + Iy I +nI2.

Proof. The definition implies that 0p(Rn) is the intersection of the spaces
H~(Rn), j = 0, ... ,[sj, and Hp(Rn). Since the inclusion H~-:;'~(Rn) C H~(Rn)
holds for all I, {3 and arbitrary 8> 0, the functional is continuous on ~p(Rn) if it
is continuous on the intersection of the spaces H~_ /Rn), j = 0, ... ,[s j, and
H~_s(Rn). The latter
property is guaranteed by the relations
xy Y E (~_s(Rn)f = and
Hl_p(Rn) (l-x)yY E (H~(Rn»* = H~p(Rn),
where X E CD (Rn) with X = 1 in a neighborhood of the origin. It is obvious that
these relations follow from the inequalities to which {3, y and s are subjected. It
has been established that the conditions of the proposition are sufficient: neces-
sity, in particular, is contained in the following proposition.

Pro p 0 sit ion 5.6. Let PI,P2 be integers, ° ~PI ~P2' and let
GJR.{p1,P2) = {vECD(R n \O):v(i(q+n),Ifl)_O,q =PI,···,P2}. In ~p(Rn)
each of the following sets is dense: '!JR.(O,p 2) if {3 > s +P 2, GJR.{p I ,P 2) if
{3 <PI +nI2.
58 Chapter 2.

Proof. The proof is obtained from the proof of proposition 2.3 by making
some obvious changes in it. •
We now turn to the functionals (v,f y ) and (v,7]y,q) from propositions 5.3 and
5.4.

Pro p 0 sit ion 5.7. The functions x ~ fy(x, 0), defined in proposition 5.3, are

°
infinitely differentiable outside the origin. As x ~ 00 they decrease faster than any
power of / x /' uniformly in 0 E Sm - n - I. As x ~ the estimates

O(1) for Rea ~ -n -y, a +n + /y / =1= 0,

f.,.{x,O) = { °(In /x /) for a + n + /y / = 0, (5.15)
O(/x/-n-Rea-Iyl) for Rea>-n-/y/,

hold. Everything said remains valid for the functions Z y(, 0) also.

Proof. The function fy(-,O) coincides with the righthand side of (5.9) in which
cP must be replaced by CPo Since in the first term of (5.9) the number h can be
taken arbitrary large, using proposition 1.5.5 we find that this term decreases fas-
ter than any power of / x / as x ~ 00.
We now turn to the second term in (5.9). The inner integral has a meromorphic
extension onto the whole A-plane. The points A = i ( / y / + /K / +n 12),
/ K / = 0,1, ... , turn out to be (first order) poles, and the corresponding residues

are equal to i (K!) -I aKcp(O, O)w K • As A~ 00 such that / 1m A/ < N for some N, then
this extension decreases faster than any power of / A /. After multiplication of the
inner integral by wYand an application of the operator E(A)-I the poles disap-
pear (by proposition 1.4.3). This allows us to replace in the second term of (5.9)
°
the line of integration ImA = by the line ImA = h. From this we obtain that
the second term also decreases rapidly as x ~ 00.
We now verify formula (5.15). Using the fact that the poles of E(A)-I are
located at the points A = -i(k +nI2), k = 0,1, ... , and by moving down-
wards the line of integration of the second term in (5.9) we find that this term is
0(1). Consider the first term. The inner integral extends as a meromorphic func-
tion onto the whole A-plane, and only the points A = i (/ y / + a - / a/ + n 12) (are
first order) poles (in order to see this we must expand CP(w, e -/0) by Taylor's for-
mula at (w,O». By moving the line of integration downwards, we find that the first
§5. Transversal operators and special representations 59

°
term in (5.9) is O( Ix I -a - 1y I-n) as x ~ (if a=/=-- Iy I - n) or O(ln Ix I) (if
a = - Iy I -n; in this case the integrand has a second order pole at
A = -in!2) . •
A similar reasoning proves

Pro p 0 sit ion Let q+nI2<T<q+l+nI2 and

5.S.
- p - 1 - n I 2 < T + Re a < - p - n I 2, where p, q are nonnegative integers. The
functions 1)y,q('O), defined by (5.13), are infinitely differentiable outside the origin.
As x ~ 00 we have, for any N,
(_)IYI-1
1)y, q(x ) + (2)nI _ 112,
q
~
~
{;vV'
EL-
,
f wY+"d W
X
7T y. 1"1 =0 K. sn-I

f
+00
X e(n+I"I+lyl)t<ll(etw,O)dt = O(lx IN).
-00

As x ~ ° the estimate

1)y,q(x,O) =

o ( Ix Iq + 1) for q + 1 .;;;; - Re a - n !2 - Iy I,
q+I=/=- -a-n!2-lyl,
o ( Ix Iq + lIn Ix I)
for q + 1 = - a - n 12- Iy I,
O(lxl-Rea-nI2-IYI) for q+I>-Rea-nI2-lyl

holds.
In order to verify the following two assertions it is necessary to use proposi-
tions 5.7 and 5.S (compare with the proof of proposition 5.5).

Pro po sit ion 5.9. If /3-s < min{n!2, - Rea -I y I -nI2}, then the func-
tionals v ~ (v, Ky) are continuous on 6fi(R n).

P r o p 0 s i t i o n 5.10. Let q+nI2</3-s<q+I+nl2,

-p -I-nI2 < /3-s + Rea < -p -nI2, where p,q are nonnegative integers.
Then the funetionals v ~ (v, 1)y,q), where Iy I = 0, ... ,p, are continuous on 6fi(R n).
The closure 9C in 6fi(R n) of the set 9C = {v E c~ (Rn \ O):(v, 1)y,q) = 0,
Iyl = 0,'" ,p, (V,yK) = 0, IKI = 0, ... ,q} coincides with the subspace of
60 Chapter 2.

common zeros of the functionals (v,1]y,q)'

§6. Estimates for the convolution operator on the spaces Hh(lRm,~m -n)
We first give a boundedness theorem for the transversal operators A(e) (cf. (5.2»
in the scale 0:B(~n). As a corollary we subsequently obtain a statement concerning
continuity of the original operator (5.1) in the scale H:B(~m,~m -n).

The 0 rem 6.1. Let a,b and s be such that f3-s+Rea > -nI2, Rea ~s,
s ;;;. 0, and f3-s =1= k +nl2, f3 =1= k +nI2, k = 0,1, .... For f3-s > nl2 we put

S3 = {vECO(~n\o)j.v(y)yycry = 0, iyi = 0, "',[f3-s-nI21},

and for f3 - s < n 12 we will assume S3 = Co (~n \ 0).

Then the operator A(e), given on S3 by (5.3), allows the estimate

(6.1)

with a constant c which is independent of e E sm -n -I. The set S3 is dense in

0:B(~n), and for A(e) (the closure of A(O)), which is defined on all of 0:B(~n), the
representation

(A(e)v)(x) = _1_ f ri(in/2+Hia)E",~</>(A+ia)-1 X (6.2)

,fi;;ImA= {J-s

holds.

Proof. By proposition 5.2, for v E S3 the expression A(e)v coincides with the
righthand side of (6.2). Proposition 5.6 implies that S3 is dense. Hence it suffices to
verify the estimate (6.1).
Put w = A(e)v. After application of the Mellin transform this equation takes
the form (for 1m7\. = ImfL = f3-s)

w(A + ia + in 12,cp) = (6.3)

= E",~<j> (7\. + ia) - 1 cI> P.~A (w, O)Ey~", {JL)ii (f.L + in 12, l/;),
§6. Estimates for the convolution operator 61

where the expression () J.I~A (w, O)E(p.)ii(p. + in 12,) is to be understood as in proposi-

tion 5.2. We show that Al-+w(A+ia +nl2,), defined for ImA = /3-s by (6.3), has
an analytic extension to the halfplane ImA ~ /3-s.
We have

(6.4)

f
+00

= 1/- (t)eiAt()(w,e -to)'! p-s(t, w)dt +

-00

+co
+ f 1/+(t)e iN ()(w,e- tO).!p_s(t,w)dt
-co

(the notations are those of §5.2). Formula (5.6) and proposition 2.1 imply that
Jp-it,w) = O(e ht ) as t~+oo and h = [/3-s-n/2]+n/2. Recall that the
operator function A1-+ E(A + ia) - 1 depends analytically on A in the halfplane
1m A+ Re a > - n 12. Hence the result of applying E(A + ia) - I to the second term
on the right in (6.4) is also an analytic function on the halfplane ImA ~ /3-s.
The first term has a meromorphic extension to the whole A-plane (cf. lemma 5.1).
It was shown in the proof of proposition 5.2 that the value of E(A+ia)-1 on the
first term is also an analytic function in the halfplane 1m A+ Re a > - n 12.
Let's turn to estimating the norm

We denote the first and the second term on the right in (6.4) by g_(A,w,O) and
g+ (A,w,O), respectively. We have
w(A+ia +inI2,q,) = (6.5)

=E W ->4>(A + ia)-I g- (A,w, 0) + E W ->4>(A + ia)-I g+ (A,w,O).

We first consider the expression E(A + ia) - 1g+ (A, W, 0). By assumption
/3-s+Rea> -n/2, hence for T = /3-s or T = /3-Rea the line ImA = T does
62 Chapter 2.

not contain poles of the operator-function At->E(A+ia)-1. This and proposition

1.5.5 imply that

f IIE(A + ia) -1 i + (;\, ',0); Hl("A.,sn -I )11 2dA';;;; (6.6)

Im;\=-r

.;;;; c f Iii + (A, ',0); H.B("A.,sn -1)11 2dA,

Im;\=T

where I = /3--r-Rea. By putting!{jL,w) = ~-yw{jL)v{jL+inl2,l/;) we obtain

i+(g+i-r,w,O) = (6.7)

f f
+00 +00
= ei~ITJ+(t)e(ft-s-T)II'P(w,e-IO)dt e- i(JI!(a+i(f3-s),w)da.
-00 -00

Obviously, /3-s --r';;;; O. Assume for the moment that /3 ~ O. Then the norm

is equivalent to the norm

(6.8)

1!2'
+ Iii + (g+ i-r, ',0); HP(sn - I )11 2 ]dg}

where, as usual, HP(sn -I) is a Sobolev-Slobodetskii space. Further, the operator

of multiplication by the function t t-> TJ + (t)e(ft-s -T)I l'P(w,e -10) is continuous on
HfJ(lR) and, moreover, its norm is uniformly bounded in wand e,
IeI = Iw I = l. Hence the integral of the first term in (6.8) is at most
+00
C f (l+a 2 )PII!(a+i(f3-s),·);L 2 (sn-I)11 2 da.
-00

The same operator of multiplication is continuous also on the space

L 2 (1R ;HP(sn -I)) of square-integrable functions with values in HP(sn -I). Thus,
the integral of the second term in (6.8) is at most
+00
C f Ilf(a+i(f3-s),·);HP(sn-I)11 2 da.
-00
§6. Estimates for the convolution operator 63

Thus,

f Ilg + (A" O);H,B(A, Sn - I )11 2 dX';;;;; (6.9)

Imh='T

.;;;;; c f Ilj(P,);H,B(p,sn -1)11 2dlL-

Imh=,B-s

The operator adjoint to the operator j 1-+ g+ with respect to the duality

<u,v> = f f u(~,w)v(~,w)d~dw
R S·-I

is obtained from (6.7) by replacing (l by 41. The estimates proved for 13 ~ 0 for
the adjoint operator imply estimates for the operator (6.7) for 13 < 0 also.
By the requirements in the theorem, there are no poles of E(P) on the line
Imp. = /3-s. By applying proposition 1.5.5 we derive from (6.9) that

f IIg + (A, ·,O);H,B(A,sn -I )11 2dX.;;;;; (6.lO)

Imh=T

.; ; ; f Ilii(p+inI2,);H s(p,Sn -1)11 2dp..

Imp.=,B-s

We will now consider the expression E",-t>(X+ia)-lg_(A,w,O). By (5.6), on the

line ImX = 7" (7" = 13- Rea or T = f3-s) we have

(6.11)

f
+00
= E",~tj>(A+ia)-1 ei(h+ia)ITJ_(t)(l(e1w,0).! ,B(t,w)dt+
-00

f
+00
X TJ_(t)e(i(h+ia)+nl2+ l'YI)lq,(e1w,O)dt,
-00

where ko = max{[T-nI2] + 1,O}, kl = [/3-nI2]. By proposItIon 2.1, as

t~-oo we haveJ,B(t,w) = O(e([,Bl+I)I) (if /3>nI2) orJ,B(t,w) = O(e 1nI2 ) (if
13 < nI2). Hence the integral in the first term on the right in (6.11) converges. The
inner integral in the remaining terms can be regarded as the analytic extension of
the convergent integral defined by
64 Chapter 2.

f
+00
1)- (t)e(iA-a +nl2+ Ir I)1 «II(e 1 ,w,O)dt =
-00

f
+00
1)_(t)e(iA-a+nl2+ Irl)I«IIN+I(t,w,O)dt +
-00

f
+00
X e(iA-a+nl2+ Irl + IKI)I1)~(t)dt.
-00

Here «II N + 1 is the remainder in the Taylor expansion of «II. If 1 1m AI < const and
A ~ 00, then this extension decreases faster than any power of 1A I. An applica-
tion of E(A + ia)-I preserves the character of decrease (proposition 1.5.5), while
the poles disappear (proposition 1.4.3).
The functionals (5.14) occur in (6.11) only if Irl +n12 </3. Thus, these func-
tionals are continuous on 0p(lR n ). Taking into account that the terms in (6.11)
containing them are rapidly decreasing as A~ 00, we find that the norm in
0:8 -Rea(lRn) of each such term is at most cllv;0:8(lR n)ll.
Denote the first term on the right in (6.11) by EW ->4>(A+ia)-lh_(A,w,8). The
inequality

f IIE(A + ia) - I h_ (A, ',8);H'(A,Sn - I )11 2 dA ~

IlIlA=T

~c f Ilh_(A,·,0);H.B(A,sn-I)11 2 dA
ImA=T

holds. Putting, as before, /(JL,w) = E.p->w{JL)v{JL + in 12,1[J) we are led to the formula

f f
+00 +00
= ei~I+(fl-T-Rea)I1)_(t)«II(elw,O)dt e i (Jlj(o+i/3,w)do.
-00 -00

In order to estimate h _ we repeat the considerations by which we derived

(6.10). We also learn that E{JL) does not have poles on the line Im/-t = /3. As the
result we obtain
§6. Estimates for the convolution operator 65

j IIE('\ +ia) -} h_(,\, ., 0); HI(,\, Sn -} )11 2 dA ~ (6.12)

Im;\=T

~c j Ilj(,u,·);H.B(,u,sn-I)112d~~
Imp.={3

~c f Ilv(,u+inI2,·);H°(,u,sn-I)112d~.
Imp.={3

By combining (6.5), (6.6), (6.10) - (6.12) we thus have

Ilw;Sfi - Reu(lRn)11 ~

~ C{IIV ;Sp(lR n)11 2+ f Ilv(,u+ in 12, ');Hs(,u,sn -I )112d~+

Imp.={3-s

Since both integrals on the right occur in the expression for the norm Ilv,Sp(lRn)11
(cf. (4.5», the last inequality immediately leads to (6.1) . •
If fj-s + Rea <
-nl2, the operator A(O), generally speaking, 'gets out of
hand'. In order to make the map A(O):Sp(lRn) ~Sp -Reu(lRll) bounded, we must
restrict it to a subspace with finite-dimensional deficiency and (or) add to it a
finite-dimensional term. We give two theorems of this kind.
Let p,q be nonnegative integers, p ~ q, let s be an arbitrary nonnegative
number, and let a,fj be such that

Rea ~s, -nI2-q -1 <fj-s + Rea < -nl2-q. (6.13)

Let, also, r.l,lJ) be the function defined in proposition 5.3. If P < q we put

~ = {VECO(lR n \O):jv(Y)r.I'(Y'O)4Y = 0, I'YI =p+1, ... ,q}, (6.14)

while if P = q we will assume ~ = Co (1R1l \ 0).

We introduce on the operator (1(0), defined for
fj E ( - P - 1 - n 12, - P - n I 2) by

p
«(1(O)v)(x) = (A(O)v)(x)- ~ x'Y(v,rycO),
I'YI =0
66 Chapter 2.

while for 'T) > - n 12 we put (i(U) = A(U), where A(U) is the operator (5.2).

The 0 rem 6.2. Assume 13 =I=- k +nI2, f3-[s - Rea] =I=- -k-nI2, where
k = 0,1,"', and f3-Rea <nI2. Let, moreover, conditions (6.13) be fulfilled.
Then thefunctionals (v,fy) occurring in the definition ofCI are continuous on 0p(lR n).
For v EO CI the estimate

(6.15)

holds.
For the operator (f(B) (the closure of (i(U)), given on the subspace CI C 0p(lRn)
(which coincides for 13 EO ( - q - 1- n 12, - q - n 12) with the whole space 0p(lRn»),
the representation

«(i(U)v)(x) = _1_ f
r i (in12+Jo..+ia) X (6.16)
V);;rffiJo..=p-S

X Ew~¢(A + ia) -1 q,1'~Jo..(w,U)~~w{j.LYii{j.L + in 12,I[;)dA

is valid.

Proof. The assumption 13 - Re a < n 12 and conditions (6.13) imply the ine-
quality f3-s < min{nl2, -q - Rea -nI2}. Continuity of the functionals (v,fy)
from (6.14) is guaranteed by proposition 5.9. For a function v EO (IRn \ 0) the Co
representation (6.16) (with the replacement of ImA = f3-s by ImA = 13- Rea)
follows from proposition 5.3. Put w = (i(U)v. After Mellin transformation we
obtain on the line limA = 13- Rea equation (6.3), in which Im,u = 13- Rea. We
show that for v EO CI the function A1-+ W(A + fa + in 12,') is analytic in the strip
f3-s ~ ImA ~ 13- Rea. By lemma 5.1 the integral
+00
f eiJo..tq,(w,e-tB)Jp_Rea(t,w)dt (6.17)
-00

is analytic in the halfplane 1m A < - Re a + n 12. Hence only the poles

A = - i (a + k + n 12), k = 0,1, ... , of the operator-function E(A + ia) -1 can be
singular points for W(A + ia + in 12, -) if 13 < n 12; moreover, their order is one. The
residue of w(A+ia +inl2,.) at A = -i(a +k +nI2) is zero if (v,fy) = 0 for all
multi-indices y such that I y I = k. Hence v EO CI implies that W is analytic in the
§6. Estimates for the convolution operator 67

strip f3-s .;;;; ImA';;;; 13- Rea if 13.;;;; n12. For 13 > nl2 there do not occur new
singularities of E(A+ia)-1 in the given strip. By lemma 5.1, the integral (6.17) has
a meromorphic extension onto the A-plane. The poles of this extension disappear
after an application of E(A + ia) -1. So, if v E CI, then A ~ W(A + ia + in 12,·) is an
analytic function in the strip mentioned above. This leads to the representation
(6.16).
We now turn to the estimation of the norm Ilw;Sfi-Rea(~n)ll. Put
+00

g±(A,W,O) = j TJ± (t)e iN ip(w, e -1(/).1 f3-Rea(t, w)dt;

-00

then (6.5) holds for W(A + ia + in 12, .). In view of the conditions of the theorem,
there are no poles of E(A + ia) -1 on the line 1m A = 1" for 1" = 13 - Re a or
1" = f3-s. This implies inequality (6.6). By puttingf(,u,) = E(,u)ii(,u+inI2,) and
using the fact that J f3-Rea(t,W) = J f3-s(t,w) (since 13- Rea < n12) we obtain
(6.7), in which, according to the above said, f3-s -1" E;;; 0. Thus, (6.10) holds. The
remaining considerations coincide literally with the corresponding part of the
proof of theorem 6.1. •

Rem ark 6.3. We view the operator (£(0) from a different perspective. Let
s = a = 0, p + n 12 < 13 <p + 1 + n 12, and let A(O):s~(~n) ~ s~(~n) be the con-
tinuous map described in theorem 6.1. Clearly, s~(~n) = H~(~n) and
(s~(~n))* = S~f3(~n). The operator (A(8))*:S~f3(~n)~S~f3(~n), adjoint to A(O)
with respect to the scalar product in L2(~n), coincides with the righthand side of
(6.16) in which 13 is replaced by -13, ip by ip and in which we have put
a = s = 0. By theorem 6.2, (A(8)), is the closure of (£(8) (in which also ip IS

replaced by ip). The latter operator is 'formally adjoint' to the operator

p
v I-+A(O)v - ~ Zy(·,O) jyYv(y)dy,
lyl =0 Rn

which coincides with A(O) on a dense set in s~(~n) (see proposition 5.2 and
theorem 6.1).

The 0 rem 6.4. Let q+nI2<f3-s<q+1+nI2 and

-p -1- nl2 < f3-s + Rea < -p -n12, where p,q are nonnegative integers.
Assume that .B-[s - Rea] -k -n12, .B *- *-
k +n12 (k=O,l, ... ). Define on
68 Chapter 2.

the set

W = {VECO(lRn\O):jV(YJrJk,q(Y,e)cry = 0,

IK I = 0, ... ,p; j v (y)y 'Y cry = 0, Iy I = 0, ... ,q}.

where the functions T)",q are given by (5.25) for l' = /3~s, the operator A(O) by for-
mula (5.3).
Then for v E <r the estimate (6.2) holds. Representation (6.2) is valid for the
operator A(O) (the closure of A(O»), given on the subspace

9'i = {VE0 P(lR n):jV(Y)T)",q(y,O)dY = 0, IKI = 0, ... ,p} (6.18)

Proof. Boundedness of the functionals from (6.18), and equation (6.18) itself,
follow from proposition 5.10. By applying proposition 5.4 we obtain representa-
tion (6.2) for the operator A(O) on the set W. The remaining considerations are
completely analogous to those given in the proof of theorems 6.1 and 6.2 . •
Let's turn to the initial operator (5.1). Boundedness results for this operator in
the scale of spaces Hp(lRm,lR m -n) will be based on boundedness theorems for the
operator A(O) in the scale 0p(lR n).

The 0 rem 6.5. Let the numbers a E C, °

/3 E IR, and s ;;;. be such that
Re a ~ s, /3 ~ s + Re a > ~ n 12, and /3 ~ s =1= k + n 12, /3 =1= k + n 12,
k = 0,1, .... For /3~s > nl2 put

53 = {uECO(lRm'lRm-n):j(§'u)(y,~y'Ycry = 0, Iyl = 0, ... ,[/3~s-nI2l}

and for /3-s < nl2 assume that 53 = Co (IRm,lR m -n).
Then the estimate

(6.19)

holdsfor the operator A given on 53 by (5.1). The set 53 is dense in Hp(lRm,lR m - n),
and for the operator A (the closure of A), defined on the whole space
§6. Estimates for the convolution operator 69

Hp(Rm,R m -n), the representation

(Au)(x(l),x(2) = (6.20)

= iJft~X(2) f r i (inl2+,\+ ia)Ew~cp(A + ia) -1 X

Im'\=,B-s

X <P1'~,\(w,O) I~2) Ii('\-I')~~w(,u)~(,u+inI2,~;~2)dA

holds. Here, as before, iJ is the Fourier transform of x(2) ~u(x(1),x(2),
0fu5(,u+inI2,~;~2) is the Mellin transform of r~(iJu) (r,~;~2), r = Ix(l) I,
cp = x(l) II x(l) I, 0 = ~2) I I ~2) I, ~ = y(l) I Iy(l) I, and the operator <P1'->,\(w,O) is
the same as in proposition 5.2.

Proof. Put w = A u. Instead of ~2) we simply write ~. By (4.6) it suffices to

verify the inequality

111~I-a(~~~w)('/I~I,~);0p-Rea(Rn)11 ~ (6.21)

~ cll(iJy~~u)U I ~I ,~;0p(Rn)ll.

The estimate (6.19) is obtained by multiplication of (6.21) by 1~ls-,B-nI2 and

subsequent integration with respect to ~. Recalling the definition of A(O) we derive
(6.21) immediately from (6.1). It must here also be taken into account that
v (y) _ (iJu) (y I I ~ I ,~), (A(O), v)(x) _ I ~ l-a('!f(Au»(x I I ~ I ,~). The representation
(6.20) clearly follows from (6.2).
We turn to the proof of the fact that S3 is dense. Passing to the variables
r = Iy I, ~ = Y I Iy I, we write (iJu )(y,~) in the form (iJu )(r, ~;~). We first verify
that the set of functions u from COO (R m , R m -n) subject to the condition

f (iJu )(r, ~;~)rn - 1dr

00

= 0 (6.22)
o
is dense in Hp(Rm,R m -n).
For an arbitrary function u E COO (R m ,lijm -n) we introduce the sequence {ud
for which (iJuk)(y'~) = k n('5u)(Iry, ~). It is obvious that the difference u - Uk
satisfies (6.22). Furthermore, ~(A+in!2,~;~ = ki'\+nl20fu5(A+inI2,~;~), and
by (4.5),
70 Chapter 2.

where the dot indicates the y variables. This and (4.6) imply

IIUk;Hp(Rm,R m-n)11 ::;;:;; k n -2(,8-.I')llu ;Hp(Rm,R m -n)ll.

Hence Uk ~O in Hp(Rm,R m- n) as k ~ 00, and it has been proved that the set of
functions satisfying (6.22) is dense. The remaining reasoning is obvious (cf. the
proof of proposition 2.3). •
We now give a statement concerning boundedness of the operator A; this state-
ment follows from theorem 6.2 concerning the operator A(fJ). Let p,q be nonnega-
tive integers, p::;;:;; q. If P < q we denote by )ii the set of functions from
COO (Rm ,R m-n) satisfying for all ~ E R m-n the conditions

here y =y(l), ~ =
(2). If p = q we assume )ii = COO(Rm,R m- n). It is easy to
prove that

r'Y(YI~I,V I~I) =
= I~I-a- I'YI (-i)I'Y1 f
(2'1T) (n + I) I 2.",
r i (in/2+p.hu")-I,,,'Ydll. X
LJ\I" r
r ·Imp.=h

f pip.+n/2+ I'YI-Itp(p",,~p,
00
X
o
where r = Iy I, tp = Y 1 Iy I, h is an arbitrary number satisfying the inequality
h > max { - n 12, I"Y I + Re a + n 12}, and the inner integral is understood as the
analytic extension (with respect to p,) of the corresponding convergent integral.
Put

(the Fourier transform is understood in the sense of the theory of generalized

functions). We introduce on )ii the operator cr, defined for
{J E ( -P - I - n / 2, - p - n 12) by

(CiU)(x(l),x(2» = (Au)(x(l),x(2»+
§6. Estimates for the convolution operator 71

while for 13 > -n12 we put if, = A, with A the operator (5.1).

The 0 rem 6.6. Assume that 13 =1= k +nl2, f3-[s - Rea] =1= -k -nI2, where
k = 0, 1, .. " and 13 - Re a < n I 2. Assume also that conditions (6.13) are
satisfied Then for u E § the estimate

holds.
A representation of the form (6.20) holds for the operator if, (the closure of if,),
given on the subspace §.
We will not halt in order to formulate a boundedness theorem for A following
-- -
from theorem 6.4 on A(U). We finally note that the representations for A(O) and A
obtained in this paragraph can be written using the 'generalized' Fourier
transform from theorem 2.4 (compare with theorem 3.5).
 Chapter 3

Meromorphic pseudodifferential operators

Meromorphic pseudodifferential operators arise, e.g., after applying the Mellin

transform on the right and on the left to the operator of convolution with a
homogeneous function. A 'canonical' meromorphic pseudodifferential operator of
order a has the form E8~(A+ia)-1cI>(cp,0)E.;t.....8(A). Thus, the operator E(A) plays
the same role for meromorphic pseudodifferential operators as does the Fourier
transform for ordinary pseudodifferential operators.
In §1 we define canonical meromorphic pseudodifferential operators, and show
some simplest properties of them. In particular, a meromorphic pseudodifferential
operator is represented as an integral operator, whose kernel is described. In §2
we study actions on canonical operators: composition, transition to the adjoint
operator, shift, and differentiation with respect to a parameter. These operations
lead, in general, outside the set of canonical meromorphic pseudodifferential
operators. A larger class of meromorphic pseudodifferential operators that is
invariant under these actions is discussed in §3. In §4 we define for a mero-
morphic pseudodifferential operator of arbitrary order the trace and prove formu-
las expressing the trace of an operator in terms of its symbol. The asymptotic
behavior of the values of a meromorphic pseudodiffere~tial operator on strongly
oscillating functions is clarified in §5. §6 is devoted to the generalization to mero-
morphic pseudodifferential operators of certain inequalities which are well-known
for classical pseudodifferential operators (Garding's inequality, etc.). In §7 we
study 'periodic' meromorphic pseudodifferential operators, and the final §8 con-
tains rules for change of variables in meromorphic pseudodifferential operators.
73
74 Chapter 3.

§I. Canonical meromorphic pseudodifferential operators

I. Definitions of operators. Continuity. Let a be an arbitrary complex number.
For A=I=i(k+nI2), A =1= -i(k+a+nI2), k = 0,1, ... , we introduce the
operator

(1.1)

By proposition 1.5.5 the maps E(A.):Hs(A.,sn-I)~Hs+Im\A.,sn-l) and

E(A+ia)-I:Hs+ImA(A.,Sn-I)~Hs-Rea(A.,Sn-l) are continuous. Hence the esti-
mate

IIU(A.) II ~ cllell;H s + 1m A(A., Sn -I) ~Hs +Im\A.,Sn -1)11

holds for the norm of the operator U(A):Hs(A.,sn-I)~Hs-Rea(A.,sn-I), where

on the right we have the norm of the operator of multiplication by the function ell
on Hs+ImA(A.,Sn-I). Using the inequalities (1.5.21) and (1.5.22) we obtain the
following

Pro p 0 sit ion 1.1. Let B be an arbitrary closed set located in a strip
I ImAI < h and not containing poles of the merom orphic operatorfunction A~ U(A)
(which can be located only at the points A = i (k + n 12), A = - i (k + a + n 12),
k = 0,1, ... ). Then the estimate

holds, where A E Band q is a number depending on h only.

In the sequel it is assumed that ell E coo(sn -I), and hence the choice of a
concrete value of q does not play a role.
The operator (1.1) made its appearance already in the previous Chapter (cf.
formula (2.3.6)); it is also the simplest meromorphic pseudodifferential operator of
order a. An operator of the form

(1.3)

will be called a canonical merom orphic pseudodifferential operator of order a.

Proposition 1.5.1 and formula (1.5.4) imply the following assertion

Proposition 1.2. Let ell E coo(sn-IXsn-l) and let

§l. Canonical merom orphic itDOs 75

00 km
cI>(q"O) =}: }: amk(q,)Ymk(O). (1.4)
m =0 k =1

Then for arbitrary "y,8 = 0, I, ... , the estimate

holds.

Pro po sit ion 1.3. Let c:P E coo(sn -1 Xsn -1) and let U(A) be the operator
defined by (1.3) for A =1= i(k +nI2), A =1= -i(k +a +nI2), k = 0, I, .. '. Then
the map U(A): HS(A,sn -1) ~ H S- Rea(A,sn -1) is continuous. On every closed set ~
located in a strip I1m AI < h and not containing poles of the operator junction
;\ 1-+ U(;\) the estimate

(1.5)

holds.

Proof. Expand c:P in a series (1.4), and denote by Umk(A) the operator
E(J .....<P(A+ia)-IYmk(O)~ ..... (J(O). By proposition l.l, inequality (1.2) with c:P
replaced by Ymk holds for the operator U mk . In view of the estimate
Ilymk ;cq(sn-l)ll,;;;; O(m-l+ q + nI2 ) (cf. [29] or [40]), this and proposition 1.2
imply that the series

converges, and thus imply the relation (1.5). •

The operator-function ;\ ~ U(;\), given by (1.3), is meromorphic in the whole
A-plane. If a =1= -/ - n (l = 0,1, ... ), then this function can have poles at
;\ = i(k +nI2), ;\ = -i(k +a +nI2) of the first order only, and there are no
other poles. The residue of the function A1-+ U(A)u at ;\ = i (k + n I 2) is

_1_e i '1T(n + a) 12 r(n +k +a) X (1.6)

C2'7T)n

X jCq,8+iO)-k-n-ac:pcq"O) ~ J,fJYdOj1/;Y u (1/;)d1/;,

lyl =k"Y'
76 Chapter 3.

and the residue at A = -i(k +a +nI2) is

_l_e i1TCn +a)/2f(n + k +a) ~ (-ll cpY X (1.7)

(2?Tt Iyl =k Y!

If a is a nonnegative integer and if the function 0 ~ «f.J(cp, 0) coincides with the

values on Sn-l of a homogeneous polynomial x~P(cp,x) of degree
Ie (O=x1Ix I, the coefficients of the polynomial depend on cp), with moreover
Ie';::; a, Ie _ a(mod2), then these residues are equal to zero; the operator-function

U(A) is then holomorphic in the whole plane (cf. proposition 1.4.3).

In case a = -I-n, the points A = i(k +nI2) for k = 0, ... ,I are poles of
order two, and the points A = i (k + n 12), A = i (1- k +n /2) for
k = 1+ 1,1 + 2, . .. are poles of order one. The residues at poles of order one
can be computed by formulas (1.6), (1.7). The coefficient in front of
(iA + k + n 12) - 2 in the Laurent expansion of the function A~ U(A)U is equal to
( l) k + 1·/
- I j(-UI\I-k«f.J( O)dOj(tI.'l (o")d!J;
k!(/-k)!(2?Tt '1"') cp, V'f U 't' ,

and the coefficient in front of (iA + k + n 12) - 1 coincides with

( l)k+J.I [
- I {II;(l-k + l)-!J;(k + I)} X
k !(l- k) !(2?Tt

X j (cp8)I- k «f.J(cp,8)dO j{tAf)k u (!J;)d!J; +

+ j (cpO)' - k«f.J(cp,8)dO j (tAf)kln( - tAf +i O)u (!J;)d!J; +
- j (cp8)I- k «f.J(cp,8)ln(cpO + i O)dO j (tAfl u (!J;)d!J;) ,

where I/I{z) = p{z)/r{z). These formulas can be obtained by a direct computa-

tion; we only have to use the representation ([5], 1.17 (11»
r(z) = (-ll{(z + k)-l +I/I(k + 1)+ O(z +k)} 1k! and the expansion

(-+-cp8+iO)-Z = (-+-cp8+iO)k_(z +k)(-+-cp8+iOlln(-+-cp8+iO) + .. '.

Thus, at the points A = - i (k + a + n 12) the function U is, in general, not finite-
meromorphic, while at the other poles it is finite-meromorphic. (A meromorphic
operator-function is called finite-meromorphic at a pole A = Ao if the coefficients
§1. Canonical meromorphic i'DOs 77

in the principal part of its Laurent expansion are finite-dimensional operators.) At

the poles A = - i (k + a + n / 2) the coefficients in the principal part of the expan-
sion of the function (1.3) are integral operators with Coo (Sn -I X Sn -I) kernels.

2. One-dimensional meromorphic pseudodifferential operators. In the one-

dimensional case t/J, 0, l/; take the two values +1, and the integral of a function u
over the zero-dimensional sphere is the sum u (1) + u ( - 1). In this case the opera-
tors E(A)±I are defined by (1.4.10) and (1.4.11), and the vector
v(t/J) = Eo-->.p(A+ia)-l4l(t/J,O)E.p-->o(A)u(l/;) is given by

v(1)
[ v(-l)
1= r(l!2+iA)r(1l2-iA+a) \B(A) [ u(1)
211' u(-l)'
1
where the entries bjk of the matrix \B(A) are as follows:
b ll = ei"IT(a-2iA)/24l(l,1)+e-i"IT(a-2iA)/24l(l,_1),

b 12 = e i"IT(l+a) /2 4l(1 , 1)+e- i"IT(l+a) /2 4l(1 , -1) ,

b21 = e-i"IT(a+I)/24l(-1,1)+ei"IT(a+1)/24l(-1,-1),

b 22 = e- i"IT(a-2iA)4l(-1,1)+e i"IT(a-2iA)4l(_1,_1).

Clearly, 4l(t/J,0) = c(t/J)+d(t/J)sgnO, where c(t/J) = (4l(cp, 1)+4l(t/J, -1»/2,

d(CP) = (4l(t/J, l)-4l(cp, -1»12. Hence a canonical meromorphic pseudodifferentia1
operator of order a has the form

r(l!2+iA)r(ll2-iA+a) X
11'

X { c(l)cos; (a -2iA) -c(l)sin;a

+
-c(-l)sin..!.a c( - 1)oos ; (a - 2iA)
2

d(l)Sin; (a -2iA) d(l)oos"2a

+i [ 11' }
11'
-d( -l)oos-a -d( -l)sin; (a - 2iA) .
2

Since r(iA + 1 / 2)r( - iA + 1/2) = 11' / COS(11'iA), for a = 0 we are led to the follow-
ing expression for U(A):
78 Chapter 3.

[
c (1) + d (1 )tanh?TA id (1) I cosh?TA 1
-id(-l)/cosh?TA c(-l)-d(-l)tanh?TA·

3. The operator U(.\) as an integral operator, and its kernel. Consider the operator
(1.1). We first assume that the number a is subject to the inequalities
- nl2 < Rea < O. We put 1 = E(ia +inI2)-I4J, and introduce the function G,
which is homogeneous of degree - a - n, by

G(x) = r-n-IJ/(tP), r = Ix I, tP = xlix I·

For a function u E Co (JR n \ 0) we define

(G*u)(x) = jG(x -y)u(y)dy.

Since (FG)(g) = I~ IIJEq, .....o(ia + in 12)1 (tP) = I~ 11J4J(8) (cf. (1.4.7», we have
G*u = F- 1 I~ IlJcp(8)Fu. By proposition 2.3.1 we have for the operator G*u the
representation
+00
(G*u)(x) = _1_ j r i(HilJ+inl2)U(A)u(A+inl2,·)dA, (1.8)
y'2; -00
where U(A) is the operator (1.1).
We apply to the convolution G*u the Mellin transform Mr .....,\+ilJ +inl2 for
ImA = O. Putting x = rtP, y = fJ\/; we have
+00
_1_ j r -i(HilJ +inl2)-I(G*u)(x)dr = (1.9)
y'2; 0

I +00 . . . +00
= - - j r- 1 (HIIJ+ml2)-ldr j dl/; j G(rtP-fJ\/;)u(p,l/;)pn-1dp =
....(2; 0 S"~l 0

+00
= j dl/; j r- i (HilJ+inl2)-IG(tcp_l/;)u(A+inl2,l/;)dt.
s" ~l 0

Now (1.8) and (1.9) imply

U.p .....q,(A)u(A+inI2,l/;) = jG(tP,l/;;A)u(A+inl2,l/;)dl/;,

where
+00
G(tP,l/;;A) = j t- i (HilJ+in/2)-IG(tq,-l/;)dt. (1.10)
o
§1. Canonical merom orphic vDOs 79

Thus, for an arbitrary function v E coo(sn -1),

(1.11)

Equation (1.11) has been proved for real A and under the restriction
- n I 2 < Re a < O. By analytic extension it can be proved for all complex a and A,
except for the poles.
We clarify a manner for analytically extending the integral (1.10). Let XbX2 be
nonnegative functions of class COO[O,oo), Xl(t)+X2(t) =
1, where moreover Xl
has compact support and equals one near t = O. The integral

= f t- i(Hia+inI2)-lG(tq,-I/;)Xl(t)dt
00
Gl(q"I/;;A)
o
is, for q, =1= 1/;, an analytic function in the halfplane Im.\ > - Re a - n I 2. Analytic
extension of Gl(q"I/;;A) to the domain ImA> -Rea-p-nl2 (p a natural
number) is realized using the formula

G (q,I/;'A)
1 , ,
= (-It
(a + n 12- iA) ... (a + n I 2 + P - 1 - iA)
X (112)
.

X f t -i(A+ia +inI2)+p - I _
00 dP - 1
_ (x (t)G(tq,-I/;»dt
dtP -I 1
o
(the points A = -i(a +nI2), ... , -i(a +p -1 +n/2) tum out to be poles).
The function

= f t- i (Hia+inI2)-1G(tq,-I/;)X2(t)dt
00
G2 (q"I/;;A)
o
is analytic in the halfplane 1m A < n I 2. After the change of variable t = T -1 it
takes the form

f T-iHnI2-1G(<p-r.f;)X2(lIT)dT.
00
G2(<P,~;A) =
o
The analytic extension of G 2 (q"I/;;A) to the domain ImA <p +nl2 is con-
structed according to the formula

G ("".I,·A) - (-It X (1.13)

2 't','I', - {iA+nl2)'" (iA+nI2+p -1)
80 Chapter 3.

By G(cp,l/;;A) we will understand the function defined by

Note that for a :::: 0,1, .. " the operator E(ia +in I 2) -I is not an isomor-
phism (it annihilates the subspace 53 a ; cf. proposillon 1.4.3), while for
a :::: -I - n, I :::: 0,1, .. " the operator E(ia + in I 2) - 1 is simply not defined.
Summarizing all this leads to the following assertion.

Pro p sit ion 1.4. Let a =1= 0, I, . .. and a =1= -1- n, where I :::: 0, I, ....
0

Let also A =1= i (k + n I 2), A =1= - i (k +a + nl2), k :::: 0, I, .... Then formula (1.9)
holds for the operator (l.l). The kernel G(cp,l/;;A) is defined by (1.8); moreover,
G(x):::: r- n - a E w.....<j>(ia+inI2)-I4>(w), r :::: Ix I, cp:::: xl Ix I, and the integral
(LlO) has to be understood in the sense of the analytic extension described above (ef
(I.l2) and (Ll3)).
We will now consider the operator (1.1) for those values of a that are excluded
in proposition 1.4. Let a be a nonnegative integer. Represent 4> as a sum 4>0 + 4> 1>
where 4>0 is subject to the conditions

f 4>o(O')fJY dO :::: °
s" -I

for all multi-indices y such that Iy I :::: a, and 4>1 has the form
[a/2]
4>1 (0):::: ~ ha -2j«());
j=o

as before, hj is a harmonic polynomial of degree j. For A =1= i(k +nI2),

A =1= -i(k +a +nI2) the operator (l.l) splits into two terms:

(l.l4)

Pro p 0 sit ion 1.5. For the first term on the right in (Ll4) a formula of the
form (Lll) holds with G(x) = r - n - a Ew .....<j>(ia + in 12) - 14>(w), and the second term
is a differential operator of order a on the sphere Sn -1 in which the parameter A
enters as a polynomial of degree a.
§1. Canonical meromorphic itDOs 81

Proof. By proposition 1.4.3, E(ia +in/2)-Iq,1 = 0. Hence E(ia +in/2)-1

CP = E(ia +in/2)-lcpo, and representation (1.11) for E(X+ia)-1 cpoE(X) follows
from proposition 104.
Extend q,1 to R n by means of the equation
[a/2]
q,1(g) = 2
~ Igl Jh a - 2 /g).
j =0

Thus, g I-> q,1 (g) is a homogeneous polynomial of degree a. Consider the differential
operator peel/ax) = F-Iq,I(g)F. By proposition 2.3.1,
pea 1 ax)u =
f
+00
= _1_ ri(A+ia +in/2)E(A +ia)-ICPI (O)E(X)«(X + in 12, ·)dA.
...n;;; -00

Note that pea I ax) = r -a01(ra 1 ar, I a acp), where aI acp is a differentiation operator
in the local coordinates of a point cp = x I Ix I on the sphere and 0l is a polyno-
mial. Substituting this expression for P in (1.15) and applying the Mellin
transform we find

0l(i (A + in I 2), aI acp)u(A + in 12,CP) =

= EO-->tj>(X + ia) -1q,1 (O)~-->ou(X + in/2,~),

which was required. •

Finally, let a = -I - n, where 1 = 0,1, .. '. We require q, to satisfy the con-
ditions

f q,(0)(JY dO = °
for all multi-indices y such that Iyl = I. Put G(x) =
= r-n-aEo-->tj>(ia +in/2)-lcp(0), with E(ia +inl2) the operator introduced by
equation (1.4.3). Then representation (1.11) holds for the operator U(X).
Everything that has been said in this section concerning the operator (1.1) can,
with obvious changes, be transferred to the operator (l.3) also. In particular, for
a =F -/ - n the function G IS defined by
G(CP,x) = r -n -aEo-->tj>(ia + in 12) -I cp(CP,O), and the kernel G (CP,I/J;X) is defined by

f t- i(A+ialin/2)-IG(cp,tCP-I/J)dt.
00
G(CP,~;X) = (1.16)
o
82 Chapter 3.

Pro p 0 sit ion 1.6. Let TI,t E Coo(Sn -1) with sUPPTl n suppt = 0. Then
for the operator (1.3) we have the estimate

(1.17)

where 'J is an arbitrary closed set in a strip I1m AI < h not containing poles of U(A),
p is an arbitrary real number, and Cqy is a constant depending on ~s and p.

Proof. We will first assume that a =1= -I - n, a =1= I, where I = 0, 1, ....

Then U(A) allows a representation (1.11), in which the kernel G (cp,I/!;A) is defined
by (1.16). By integration by parts in formulas (1.12), (1.13) we can show that on 'J
the kernel t(cp)G(cp,I/!;A)T}(I/!) (as well as its derivatives with respect to the coordi-
nates CP,I/!) decreases faster than any power of IA I, uniformly with respect to CP,I/!.
This implies (1.17). Let now a 0 be one of the numbers -I - n or I. If ( is
sufficiently small, I(tu(A)T}u, v) I ~cqyllu;Hs(A,sn-l)II·llv;H-S-p(A,sn-l)ll, with
one and the same constant Cqy for U(A) = E(A+ia)-ICP(cp,w)E(A) for all a subject
to Ia - a 0 I = t:. The maximum principle gives the latter inequality for all a in
the disk Ia -aD I ~ (if U(A) remains analytic in a) . •

4. An expression for the operator (1.3) in terms of the exterior form w. Let, as in
§1.4,

and let (x,z),....CP(x,z) be a positively homogeneous function of degree a in x as

well as in z; here a is a complex number and x,z E IR n \ 0. We will assume that
the restriction of cP is of class Coo (sn -] X Sn -]). Let also y ,.... u (y, A) be a posi-
tively homogeneous function of degree ill. - n 12. Put

Uy~AA)u(y),) = Ez~x(A+ia)-I«I>(x,z)Ey->AA)u(y,A), (1.18)

where E±] are understood to mean the operators defined by (1.4.9). It is clear
that the operator (1.18) does not depend on the choice of the surface S entering
(1.4.9), and maps homogeneous functions of degree iA-nl2 to homogeneous
functions of the same degree. This way of defining the operator U(A) (on homo-
geneous functions) turns out to be convenient under transition to a manifold.
§2. Operations on canonical meromorphic 'It DOs 83

§2. Operations on canonical meromorphic pseudodifferential operators

In this paragraph we study composition and the operations of taking the adjoint,
differentiation, and shift by a parameter, for canonical meromorphic
pseudodifferentialoperators.
We also consider meromorphic operator-functions that have in every strip
IIm;\ I < h < 00 at most finitely many poles. Any closed set located in a strip
lImA I < h and not containing poles of the operators considered will be called
admissible.
Let 0 E coo(sn-I) and «P E coo(sn-IXsn-I). Extend 0 to Rn\O as a
homogeneous function of degree v, and extend the function 01-+ «P(cp, (}) as a homo-
geneous function of degree a; the numbers a and v are taken arbitrary complex.
Introduce the notations

where (CPI, ... ,CPn), (OJ, ... ,On) are the Cartesian coordinates of the vectors cp,O.

Pro p 0 sit ion 2.1. For arbitrary nonnegative N the formula

U(;\)o = (2.1)

holds for the operator (1.3), where, moreover, the operatorfunction 9T is subject to the
inequality

(2.2)

on every admissible set '5.

(Here and in the sequel the letter 9T refers to distinct operators - remainders in
'asymptotic' formulas.)

Proof. Let {fj} be a partition of unity on Sn - I subordinate to a sufficiently

fine cover of the sphere. Then

U(;\)o = ~JjU(;\)~kO. (2.3)

j,k
84 Chapter 3.

If supp rj n supp rk = 0, by proposition 1.6 the estimate

IIrjU(A)rk;HS(A,sn -I) ~ H S+p(A,sn -1)11 .s:;; cg{p,s) (2.4)

holds on every admissible set 1f, where p is an arbitrary number.

Consider those terms in (2.3) for which the supports of rj and rk intersect. For
arbitrary t > 0,

N (-1)11'1
oCt/;) = tVo(t/;!-I) = tV [ ~ t a1'o(cp)(cp-t/;rl)"Y + (2.5)
11'1 =0 y.

+ ~ ICicp,t/;t -l)(cp_t/;t- I )1'] =

11'1 =N+I

N ( 1)11'1
= ~- t v - 1' a1'o(cp)(tcp-t/;)1'+
I l

11'1 =0 y!

+t v - N - 1 ~ IC y(CP,t/;t- 1)(tcp-t/;)1',
11'1 =N+I
where

We multiply both sides of (2.5) by r/cp)rk(t/;)G(CP,tcp-t/;) and thus obtain, by

applying the Mellin transform with respect to t,

ft-
00

r/cp)rk(t/;)O(t/;) i(H ia + inI2)-1 G (CP,tCP-t/;)dt = (2.7)

o

X G(cp,tcp-t/;)(tcp-t/;)1'dt + ~ r/cp)rk(t/;) X
11'1 =N+I

f t -i(Hiv +i(a - N -
00
X 1+nI2))-1 ICicp,t/;t - I)G (cp,tcp-t/;)(/cp- t/;)1' dl.
o

Here, each integral is understood to be its regularization using analytic exten-

sion, described above proposition 1.4.
§2. Operations on canonical merom orphic 'l'DOs 85

Formulas (1.10) and (1.11) imply that the expression

Jt -i(A+iv +i(a -I rl +n/2))-IG(tcp-1J;)(tCP-1J;)'Y dt

00
(-1)1 rl
o
is the kernel of the operator

(2.8)

Therefore (2.7) implies the relation

~jU(A)~k(J =
= :i:
Irl =0
~j~ar(J(cp)E(A+iv+i(a-lyl))-IDrc{)(CP,w)E(A+iV)~k+
y.

where 9l denotes the operator with kernel

R(cp,1/;;A) = ~ f t- i(A+iv+i(a-N-1+n/2))-1 X
00

(2.9)
iri=N+lo
X ICy(CP,lJ;! - I )G(cp,tcp-1J;)(tcp-1J;)'Y dt.

Let l1q, q = 1,2,3, be nonnegative functions of class Coo [0,(0),

111 +112 +113 = 1, with SUPP111 C [0,112), SUPP112 C (1/4,4) and
sUPP113 C [2, (0). We denote by ~lCP)Rq(CP,1J;;A)~k(1J;) the expression obtained from
the righthand side of (2.9) by multiplying the integrand by ~lCP)l1q(t)~k(1J;). Using
the method for regularization of the integrals occurring in the expression ~jRq~k
for q = 1,2,3 (cf. formulas (1.12) and (1.13)), we are led to the estimates

(2.10)

where iff is an arbitrary admissible set, A E §'; and Cj 9l q(A)Ck denotes the operator
with kernel Cj(CP)Rq(CP,1J;;A)Ck(1J;).
In order to estimate Cj 9l 2Ck we note that by using Taylor's formula of order
M (M > N) instead of (2.5) we can write the operator 9l as a finite sum of opera-
tors of the form (2.8) (for N < Iy I < M) and a term with kernel (2.9), in which N
is replaced by M. It now suffices to apply proposition 1.3 and use inequality (2.10)
and the smoothness of the function
86 Chapter 3.

~ "y(cp,tPt -I)G (tCP-tP)(tCP-tP)YT/2(t)ticp)tk(tP). •

lyl=M+l
Pro p 0 sit ion 2.2. Let IL be an arbitrary complex number, and cp = x / Ix I.
Then for the operator U defined by (1.3) the formula

U(X + IL) - U(X) = (2.11)

= ~ J,(ailx liJ4 )ll x l=lEo--><p(A+i(a-lyl»-1 X

lyl=1 y.

holds, where the operator 9l is subject to the inequality

on every admissible set 'If. If IlL I < 8, with 8 an arbitrary small number, then a set §'
which is admissible for the operators E(X+i(a -lyJ))- l D9«P(CP,O)E(X),
0.;;:; Iy I .;;:; N, is admissible also for all operators X1-+ 9l{X,IL), and the constant c'iJ in
(2.12) may be assumed to be independent of J.L.

Proof. It suffices to put v = ilL, = (J I x 1 iJ4 in proposition 2.1 and then to

replace X-IL by X. •
It follows from (2.11) that, in particular, if U is an operator of order a, then the
difference U(X + IL) - U(X) is an operator of order a - 1.

Pro p 0 sit ion 2.3. The derivative aAU _ au / ax of the operator (1.3) has, for
arbitrary natural N, the representation
N 1
aAU(X) = ~ - , pix) 1 1x 1= 1 X (2.13)
lyl=1 y.

where pix) = limJ4 -+oIL - 1 all x 1iJ4, while the estimate (2.2) is valid for the operator
9l(X) (which depends on N).
The proof follows from formulas (2.11) and (2.12) . •
We now consider composition of canonical meromorphic pseudodifferential
operators. Let
§2. Operations on canonical meromorphic 'l'DOs 87

U(X) = E(X + ia) -1 «P(cp, O)E(A),

~(X) = E(X+ib)-1'l'(cp,O)E(X).

We will assume that the function i' is extended to IIl n \ 0 in each argument cp,O as
a homogeneous function of degree b, and «P - as a homogeneous function of
degree a.

Pro p 0 sit ion 2.4. For arbitrary nonnegative integers N the formula

U(X)~(X) = i J,E(A + i
Irl =0 'Y.
(a + b - Iy I»- 1 X (2.14)

X DO «P(cp, O)a~i'(cp,O)E(X) + 9t(A)

holds; on every admissible set 'J we have the estimate

Proof. Expand 'l' is a series in spherical harmonics, i'(cp,O) = ~amk(CP)Ymk(CP).

The coefficients amk are taken to be homogeneous functions of degree b. It is clear
that

(2.15)

moreover, this series converges in the norm of operators from H S (A, S n -1) to
H S - Reb(A,sn -1). By formula (2.1),

U(X)amk = i J,a~amk(CP)E(X+i(a+b-I'YI»-1
Irl =0 'Y.
X (2.16)

X DO «P(cp,O)E(X + ib)+ 9t(X).

It remains to combine (2.15) and (2.16) . •
Finalising this paragraph, we consider operators adjoint to canonical mero-
morphic pseudodifferential operators with respect to the scalar product in
L2(Sn -1).

Pro po sit ion 2.5. Let U(X) = E(X+ia)-1«P(cp,O)E(X) and let U(X)* be the
operator a4Joint to U(X) with respect to duality in L 2 (Sn -1). Then
88 Chapter 3.

where N is an arbitrary nonnegative integer and the operator )R is subject to inequal-

ity (2.2). The function cI> is taken to be extended to R n \ 0 in both arguments cp,O as
a homogeneous function of degree a.

Proof. Expand cI> in a series

where the amk can be extended to R n \ 0 as homogeneous functions of degree a.

Then

Clearly,

By proposition 1.4.4,

This and proposition 2.1 imply that

- 1 - - 1- - --
E(AfYmk(E(A+ia)- )*amk = E(A)- YmkE(A-ia)amk =
Nl- - - - -
~ -, E(A+ i(ci -I y 1»-la~amk(cp)DBYmk(fJ)E(A)+)R(A).
I'YI =oy·
Summation over m and k leads to (2.17). •

§3. General meromorphic pseudodifferential operators

The results of the previous paragraph imply that a composite of canonical mero-
morphic pseudodifferential operators is not, in general, a canonical meromorphic
pseudodifferential operator. The operations of taking the adjoint and
differentiation with respect to a parameter also lead outside the class of canonical
operators. In this paragraph we define a larger class of meromorphic
pseudodifferential operators which is invariant already under composition,
§3. General meromorphic itDOs 89

differentiation, and the operation U<A) r-. U0)' .

1. Definition of general meromorphic pseudodifferential operators. Let a 0, a I, .. .

be a sequence of complex numbers, let aj = Reaj' and let ao > al ;;;;. a2;;;;' ... ;
moreover, let aj~ -00. We denote by {cflj}j=o a sequence of functions of class
coo(sn-IXsn-I). We will assume that each function (tP,IJ)r-.cfl/tP,IJ) is extend-
able in each argument tP and 0 to R n \ 0 as a homogeneous function of degree aj.

D e fin i t ion 3.1. An operator-function Ar-. U<A) that is meromorphic in the

complex plane is a merom orphic pseudodifferential operator of order a 0 if every
strip of the form lImA I < h contains only finitely many poles of U and if for all
nonnegative integers N the inequality

holds an every set ~ that is admissible for the operators U and Uj , j = 1, ... ,N.
Here Uj<A) = Eo-->cj> (A +iaj) -1 (llcp, O)Et-->O(A) are canonical meromorphic
pseudodifferential operators and C<J is a constant depending on iff,N and s.
The formal series
00

o(tP, 8) = ~ cfl/tP, 8) (3.2)

j =0

will be called the (complete) symbol of U, and the function cfl o will be called the
principal symbol.
A series };j'=oUj consisting of canonical meromorphic pseudodifferential opera-
tors is called an asymptotic series for the meromorphic pseudodifferential operator
U if inequalities of the type (3.1) hold for all Nand §: In the sequel the notation
U ~ };Uj means that };Uj is an asymptotic series for U.
Note that by proposition 2.2 the function Ar-.U(A+,u), with U defined by (1.3),
is a meromorphic pseudodifferential operator of order a with symbol

(the function cfl is homogeneous of degree a both in x and in 8).

90 Chapter 3.

By proposition 2.3 the derivative 0" U is also a meromorphic pseudodifferential

operator, and its symbol is
00 1
~ - , py(x)D"b «P(x, 0).
Iyl =0 y.

Pro p 0 sit ion 3.2. In definition 3.1 the expansion

00
U(;\) ~ ~ E(A + iaj) -I «Piq"O)E(;\) (3.3)
j =0

may be replaced by an expansion

00
U(;\)-- ~ E(;\)-li'iq"w)E(;\-ibj ), (3.4)
j =0

where b o = ao, f30 > f31 ;;;. f32;;;' ... , f3j = Rebj ~ - 00 and {i'j} is a sequence
ojcoo(sn-I XSn-l)junctions, '1'0 = «Po.

Proof. Suppose, e.g., expansion (3.3) holds. By proposition 2.2, for

py(q"p.) = Ix IIYI-l'oIlx II' we have

E(;\ + iaj)-I«Pj(q"w)E(A) -- E(;\)-I«Piq"w)E(A - ia;)+

+ ~ py(cp, ~aj) E(;\ - i Iy I) -I DL «Pj(cp,w)E(;\ - iaj)'

lyl=1 y.

By successively applying this proposition to operators of the form E(A - ik)-I

X(cp,w)E(A - iq) we obtain the expansion
00
E(;\+ia)-I«Pj(q"w)E(;\)-- ~ E(A)-IXq(q"w)E(;\-i(aj-q)). (3.5)
q =0

Hence, in order to pass from (3.3) to (3.4) it suffices to replace the terms in the
series (3.3) by their asymptotic expansions of the form (3.5). Transition from (3.4)
to (3.5) is performed similarly . •
§3. General meromorphic itDOs 91

2. Existence of meromorphic pseudodifferential operators.

The 0 rem 3.3. For every given series (3.2) there is a meromorphic
pseudodifferentiai operator having it as symbol.

Proof. By proposition 3.2 it suffices to prove that there is a meromorphic

operator-function A~ U(A) having only finitely many poles in any strip
I1m AI < h and allowing the asymptotic expansion
00

U(A) -- ~ 'B/A),
j=o

where 'B/A) = E(A) -1(>/q"w)E(A - ib), {itj } IS a given sequence from

coo(sn-I XS n - l ) and {bj} is a sequence of complex numbers such that
/30 >/31 ~/32 ~ "', /3j = Rebj~ -00.

The proof is divided into several steps

A) We introduce auxiliary operators. Denote by IT/ (l a real number) the pro-
jection operator that annihilates the spherical harmonics of degree at most I and
that is the identity on the remaining harmonics.
Recall that the function A~ E(A)U is singularity free at the point
A = i(k +n!2) if

jU(O)Ym'KJO)dO = 0, O';;;;m ,;;;;k, m k(mod2)

(cf. (1.2.2); the fact that conditions (1.2.3) and (1.3.13) are equivalent must be
taken into account here). Hence the function A~ E(A - ibj)IT/u is singularity free
in the halfplane ImA < h if I ~ h - Rebj -nI2. Put Ij = j - Rebj -nI2. Clearly,
Ij ,;;;; ~ + 1 and Ij ~ + 00. The operator-function A~ 'B/A)IT,) is analytic every-
where in the halfplane 1m A < j, possibly except at the points A = - i (k + n 12),
k = 0,1, ....
So, in the halfplane ImA < q the operators 'BiA)IT/), j = 0, ... ,q, can have
poles at the points A = - i (k + n 12) only. Hence, every strip of the form
lImA I < h < 00 contains at most finitely many such points, which coincide with
the poles of at least one of the operators 'BjIT/i , j = 0,1, ....
Let qr be an admissible set for the sequence {'BjIT/Jj=o. Denote by A~) the
poles of the operator E(A - ibj)(I- ITI) and put
92 Chapter 3.

(Note that the function A1-+ exp( -I5A.2) decreases exponentially as A~ 00 within
angles 1 arg AI < '7T I 4, 1 arg A- '7T 1 < '7T I 4, and that it differs by an arbitrarily
small amount from the identity in any disk 1A 1 < R, for 15 = I5(R) sufficiently
small.)
On 'J only the points A = -i(k +nI2), k = 0, I, .. " can be poles of
~lA)qlA;6J,I5).

B) We now prove that we can choose an increasing sequence {Ij} such that
Ij ~ Ij and

where N is an arbitrary natural number.

Let 'Jj be the set obtained from 'J by deleting from the latter small neighbor-
hoods of the poles of ~j that are not poles of ~jIIIJ' Let also c('Jj) be a constant
such that

For Ij ~ Ij and A E 'Jj we have

II~lA)III;V;Hs-fJJ-I(A,sn-I)112 ~ (3.7)

..;; c('Jj)211II/;V ;H s -I (A,sn - I )11 2 =

00 k",
= c('Jji ~, ~ (l+m 2 + IAl 2y- 1 lv m k 12 ~
m =1) k = 1

'J is a pole of ~j(A), then using the maximum principle, applied to the
If J.t E
function (~j(A)III;V, w) (duality in L2(Sn -I»,
and estimate (3.7), the inequality

II~lA)II/;v ;H s -fJ)-\A,sn -1)11 2 ..;; (3.8)

~ c('Jji(1 + 1?)-lllv ;Hs(A,sn -1)11 2

§3. General meromorphic i'DOs 93

can be proved for all i\ in a neighborhood of JL.

Choose 1; such that the condition

(1 +1?)-l c(lJj)2 < 2- 2j , j = 0,1, ... ,

holds.
Note that for all j = N + 1, N + 2, ... , with the possible exception of some
initial terms, the inequalities Pj + 1 < PN hold. Hence by (3.8),

11'B/i\)II1;V ;H s -PH(i\,sn -1)11 ~

11'B/i\)II1;V ;H s -p,-I(i\,sn -1)11 ~ 2- jllv ;Hs(i\,sn -1)11.

This implies inequality (3.6).

C) In this step we determine a sequence of positive numbers 8j = 8ilJ) such
that the estimate

~ c(GJ,N)

holds for all natural numbers N.

For i\ E IJ we have

II'Bii\)(I - IIt)qii\;GJ,8j )v ;Hs - Pr 1(i\,sn -1 )11 2 ~ (3.10)

~ 1qj(i\, GJ,8j ) 12(1 + 1i\ 12)-1 c(lJj)21Iv ;Hs(i\,Sn -1 )112.

Choose a number 8j = 8j (IJ) small such that for i\ E IJ the inequality

1qj(i\;GJ,8j ) 12(1 + 1i\ 12)-lc(lJj)2 < 2- 2j

holds.
Under the condition Pj + 1 < PN we obtain
II'Bii\)(I - II1)qjv;H s - P." (i\,sn -1)11..;;;

~ II'Bj(i\)(I - IIt)qjv;HS -Pj-I (i\;Sn -1)11 ..;;;

~ 2-jllv ;Hs(i\,sn -1)11.

94 Chapter 3.

This implies the estimate (3.9).

D) We define the operator U. Let {':fd k=O be an increasing sequence of sets
admissible for the operators is/A)II/j , and having the following properties:
1. The union Uk Cffk coincides with the complex plane from which the poles of
the operators is;CA)II/j are deleted;

2. Whatever a set Cff admissible for the sequence {is/A)II/j }, there is a number
N = N(Cff) such that all j ;;;. N the inclusion Cff C ~ holds. (E.g., we may take

Cffk = {A: IImAI ~k}\ IAqUl,,;;k{A: IA-Aql <k- I },

where {A q } is a set of points each of which is pole of at least one operator

is/A)II/j , j = 0,1, .... Recall that every strip lImA I ~ k contains only finitely
many points Aq , by the choice of Ij .)

Denote by {l;(Cffk )}, {ll/Cffk )} sequences as constructed in B) and C) for the set
Cffp • Put L j = I;(~), !1j = ll/~). The operators is/A)(I - II/)q/A;~,!1j) and
is/A)II Lj , j ;;;. N, do not have poles in Cff and satisfy the inequalities

Il ~Mis/A)II LJ ;HS(A,S" -I) ---+ H S-PM (A,S" - I )11 ~ c ('!f,M), (3.11)

Il ~Mis/A)(I - II L)q/A;'!f,!1);H S(,\,S" -I) ---+ H S-P". (,\,S" -I )11 ~

~ c('!f,M)

for all natural numbers M.

We introduce the operators
00 00

U(A) = ~ is/A)II Lj + ~ is/A)(I - IIL)q/A;~,Ll). (3.13)

j =0 j =0

E) We will show that the symbol of (3.13) is ~i'j. Let M be an arbitrary

natural number and let Cff be a set admissible for the operators U and isj ,
j = 0,··· ,M-l. We have
M-I 00

U(A) - ~ is/A) = ~ iSj(A)II Lj + (3.14)

j =0 j=M
§3. General merom orphic 'YDOs 95

For A E '?f we have

M-1
,..;; ~ cJ,§"II(I-Ih)(qlA;§j'~j)-1)u;Hs+,B,-,BM(A.,sn-1)112,..;;
j=O
M-1 L, k ..
,..;; ~ c],§" ~ ~ (1 +m 2 + 1AI2)S +,B,-,BM 1qlA;§j'~j)-1121 Umk 12 ,..;;
j=O m=Ok=1

where
M-1
c~M)2 = .~ c],§"~~~(1+m2+IAI2l,-,BMlqiA;§j'~j)-112
J -0 O<m<L,

Combining (3.11), (3.12), (3.15), and formula (3.14) leads to the inequality

IIU(A)-7~01mj(A.);HS(A.,sn -1) ~ H S-,BM(A.,sn -1 )11,..;;

,..;; c(GJ,M) = const. •

3. The algebra of meromorphic pseudodifferential operators. The following two

theorems follow immediately from propositions 2.4, 2.5, and definition 3.1.

The 0 rem 3.4. Let U and m be meromorphic pseudodi./ferential operators of

orders a 0 and b 0 with symbols
00 00

ou(tp,w) = ~ 'Pj(tp,w), o~(tp,w) = ~ 'Yk(tp,W),

j=O k=O
where 'P j and 'Yk are functions homogeneous in tp of degree aj and in w of degree aj
(resp. bk). Then u(A.)m(A.) is a meromorphic pseudodi./ferential operator of order
a 0 + b 0.. whose symbol is defined by
96 Chapter 3.

The 0 rem 3.5. Let U be a meromorphic pseudodifferential operator of order ao

with symbol
00

ou(,p,w) = ~ <P/,p,w),
j=o

where the <Pj are homogeneous functions of degree aj in ,p and in w. Let U(A)* be the
operator adjoint to U with respect to the scalar product in L 2 (sn -]). Then the
operator U*(A) defined by U*0) = U(A)* is a meromorphic pseudodifferential opera-
tor of order £lo, and the asymptotic expansion

U"(A)-- ~ ~ ~E(A+i(aj-lyl»-]a~DL<P/,p,w)E(A)
j=o Irl=oY.

holds.

D e fin i t ion 3.6. A meromorphic pseudodifferential operator U is called

elliptic if its principal symbol <Po does not vanish anywhere on Sn -] X SIl -] .

D e fin i t ion 3.7. We say that the order of a meromorphic pseudodifferential

operator ~ is equal to - 00 if the estimate

holds on every admissible set £!f, for any positive p.

Note that any strip lImA I :s;;; h contains only finitely many poles of ~ and that
the complete symbol of ~ vanishes identically.

The 0 rem 3.8. Let U be an elliptic meromorphic pseudodifferential operator of

order a. Then there is a meromorphic pseudodifferential operator jJl of order - a such
that both differences jJlU - I and UjJl- L where I is the identity operator, are mero-
morphic pseudodifferential operators of order - 00.
If jJl' is an arbitrary meromorphic pseudodifferential operator 'inverting' U in this
sense from one side, then it inverts U from the other side also, and
jJl'(A) = jJl(A) + ~(A), where ~ is a meromorphic pseudodifferential operator of order
-00.

Proof. Put \.B(A) = E(A - ia) - ] <Po(,p, w) -] E(A). Then \.B(A)U(A) = 1- 0(A),
§3. General merom orphic i'DOs 97

and ~ is a meromorphic pseudodifferential operator whose order has negative real

part. Denote by o~ the complete symbol of ~ and introduce the formal series
~f=o(o~)f, where the operation of taking a power is understood in the sense of
multiplication of symbols under composition of operators. It is clear that this
series can be regarded as the complete symbol of a meromorphic
pseudodifferential operator ~(A). Now put mieft = ~(A)~(A). We have

mIeft(A)U(A) = ~(A)~(A)U(A) = ~(A)(I - ~(A)) = 1+ 2 left (A),

where 2 Ieft is a meromorphic pseudodifferential operator of order - 00.

We can similarly construct an operator mright> 'inverting' U from the right.

Let m{eft and m~ght be arbitrary meromorphic pseudodifferential operators
inverting (in the same sense) U from the left and from the right, respectively. Then

m{eft(A)U(A)m~ght(A) = m{eft(A)(I + 2~ght(A)) = (I + 2Ieft(A))m~ght(A),

where 2{eft and 2~ght are meromorphic pseudodifferential operators of order - 00.
Hence the order of the difference m{eft -m~ght is also - 00. This implies both
assertions in the theorem (we may put, e.g., meA) = mIeft(A)) . •

§4. Traces of meromorphic pseudodifferential operators

1. The trace of a canonical meromorphic pseudodifferential operator. We first
recall some well-known notions and facts (see, e.g., [14], [16]). A linear operator A
on a Hilbert space is called nuclear if for arbitrary orthonormal systems {fn}, {gn}
the series ~(Afn>gn) converges. For a nuclear operator the sum of the series
~(Afn,fn) does not depend on the choice of an orthonormal basis {fn}; it is called

the trace of A, and is denoted by SpA.*) If A and B are bounded operators such
that AB and BA are nuclear, then Sp BA = Sp AB. The equality Sp A = ~An'
where {An} are the eigenvalues of A counted with multiplicities, holds. Finally, if
an integral operator A is nuclear,

(Au)(x) = jG (x,y)u (y)dy,

Q

and G is a continuous function on ~ X ~, then

'lIn English literature often Tr A is used instead of Sp A.
98 Chapter 3.

SpA = fG(x,x)dx. (4.1)

~

(In the sequel this result is applied with Q = Sn -I.) An imbedding operator
HI(sn-l) C H1-N(sn-I), where H S is a Sobolev-Slobodetskii space, is nuclear
for N >n-1.

Pro p 0 sit ion 4.1. Let G(cp,x) = r-a-nf(cp,l/,-}, where

f E coo(sn-1Xs n - 1), 1/1 = xlix I, r = Ix I, cp E Sn-I. Let also

00
G(cp,I/I;A) = f t -i(Hia +in!2)-l G(cp,tcp-I/I)dt (4.2)
o
(the integral is understood as the analytic extension with respect to A; cf (1.12),
(1.13»). If Rea < I-n, then the function (cp,I/I)~G(CP,I/I;A) is continuous on
sn-IXsn-l.

Proof. Let X be a nonnegative smooth function, equal to 1 in a small neigh-

borhood of t = 1 and equal to 0 outside an interval [1-(,1 +(]. We have
00
G(cp,I/I;A) = f (I-X(t»t -i(Hia +inI2)-1 G(cp,tcp-I/I)dt +
o
1 +<
+ f X(t)t -i(Hia +in!2)-1 G (cp,tcp-I/I)dt.
1-<

The first integral is a smooth function on Sn -I X Sn -I. Hence it suffices to

prove that for small ( the second term is small.
Rewrite the second integral as
1
f t -i(H ia +inl2)-I(l_ t) -n -aX(t)G(cp, (</>-1/1)(1- t) -1 - cp)dt +
1-<

1 +.
+ f t -i(H ia + inI2)-I(t -1) - n -aX(t)G(cp,cp+ (cp-I/I)(t -1) -1 )dt.
1

If 21 t - 1 I < Icp -1/11 or if Icp -1/11 < It - 1 I 12, then

I(cp-I/I)(l_t)-l_cpl ;;;;'112 and IG(cp,+(cp-I/I)(1_t)-I+cp)1 ~const. If now
It - 1 I 12 < Icp -1/11 < 21 t - 1 I, It - 1 I < (, the vectors </> and cp -1/1 are almost
§4. Traces of meromorphic itDOs 99

orthogonal, hence again I(cp -1/1)(1 - t) - 1- cp I ~ 112 and I G I .;;;; const. It

remains to note that the integrals
1
j t -i(X+ia +in(2)-I(1_ t)-n -adt,
1-<

1+<
j t -i(X+ia +inI2)-I(t _l)-n -adt
1

converge. •
Let U(A) = E/J .....</> (A + ia)-I~(cp,0)4..... /J(A) be a canonical meromorphic
pseudodifferential operator with symbol ~, Rea < n -1. The map U(A):
L 2(Sn -I) ~H-Rea(}\,sn -I) is continuous, hence U(A) viewed as an operator on
L 2(sn-l) is nuclear. By §1.3,

(U(A)u)(tp) = j G(tp,l/;;A)u(l/;)dl/;,

where G(tp,l/;;A) is defined by (4.2), G(tp,x) = r- a -nf(tp,l/;),

f(tp,l/;) = Eo .....t(ia +in!2)-ICP(tp,8). By proposItIon 4.1 the function
(tp,I/I)~G(tp,l/;;A) is continuous on Sn-I X sn-I, hence formula (4.l) applies.
Thus,

SpU(A) = jG(tp,tp;A)dtp.

It is clear that
00
G(tp,tp;A) = j t -i(A+ia +inI2)-IG(tp,(t -l)tp)dt =
o
1
= j t -i(A+ia +inI2)-1 (1- t)-n -adtG(tp, - tp) +
o
00
+ j t -i(A+ia +inI2)-I(t -l)-n -adtG(tp,tp).
1

Therefore,

SpU(A) = rea -iA+n!2)r(l-n -a) jG(tp -tp)dtp+ (4.3)

r(l-iA-n!2) ,

+ r(iA+nI2)r(l-n -a) jG(tp tp)dtp

r(iA+1-a-nI2) ,.
100 Chapter 3.

Using analytic extension this formula can be generalized to all complex a and
A, except for the poles.
We will now express the trace of an operator U{A) in terms of its symbol.

Pro p sit ion 4.2. Let U{A) = E/J-></> {A + ia) -l~{tP,O)E.r->/J{A) be a canonical
0

merom orphic pseudodifJerential operator. For the trace Sp U{A), defined using ana-
lytic extension, the formula

S U{A) 1 e i7T(a+n)/2 7T [r(a-iA+nI2)x (4.4)

p = (27T)nl2 sin7T{a +n) f(l-iA-nI2)

x f dtP f (-cp8+ io)-a -n~{tP,O)dO+

+ . f{iA+nI2) fdtPf{cp8+io)-a-n~{tP,o)dOl
r(IA+1-a-nI2)

holds for a =1= -+k -n, k = 0,1, ... ; while for a = -k -n, k = 0,1, ... , the
formula

(4.5)
SpU{A) = (27Tt /2 (iA+nI2) ... (iA+nI2+k) X

X [i fdtPf{tPO)ksgn{tPO)~{tP,O)dO-cotan7T{iA+nI2) X

X f dtP f (tPO)k~{tP,O) dO]

holds.

Proof. If a =1= -+k - n, then, substituting in (4.3) the expression

= (27T;n/2 e i7T(a + n)/2 f{a + n) f (!J;O+ iO) -a -n~(cp,O)dO,

we obtain (4.4). (We must use the fact f(z)f(l-z) = 7T/sin7Tz.) It is relatively
easy to see that in (4.4) the factor in front of 11 sin 7T(a + n) vanishes at the points
a = -k -n, k = 0,1, .... Hence by limit transition in (4.4) as a ~ -k -n we
obtain

S U(A) = [f(iA+nI2)f'(iA+nI2+k + 1) + (4.6)

P f2(iA+nl2+k + 1)
§4. Traces of meromorphic i'DOs 101

+( - l l P( - 1
k -:- iA - n 12) j dcp j (<PfJ)k<p(CP,8)dO+
f(I-IA - n12)

_ [ f(iA+nI2) +(_I)k+l f(-k-iA-nl2)] X

f(iA+nI2+k + 1) f(1-iA-nI2)

Further (d., e.g., [5]),

f(iA+nI2) = (_I)k f(-k -iA-nI2)

f(iA+nI2+k + 1) f(1-iA-nI2)

(iA+nI2) ... (iA+nI2+k)'

P{iA+nI2+k + 1) _ P(-k -iA-nI2)
- '7Tcotan '7T(iA + n 12).
f(iA+nI2+k + 1) f( -k -iA-nI2)

Using these equations (4.5) can be derived from (4.6) . •

We note that also the residue in (4.4) at the points a = I-n, 2-n, ... van-
ishes, under certain restrictions regarding the symbol (of the type
j dcp j8(k)(<PfJ)I(CP,8)dO = 0). If such restrictions are satisfied, an expression for
Sp U(A) at a = 1 - n, 2 - n, . .. can be obtained from (4.4) also.

2. Traces of general meromorphic pseudodifferential operators. Let now U be a

general meromorphic pseudodifferential operator of order ao in the sense of
definition 3.1. Then for arbitrary N the formula
N~l

U(A) = ~ U/A)+ffiN(A)
) =0

holds, and for ffiN the estimate

(4.7)

holds an every admissible set iJ. Here cx) = Rea) ~ - 00, aj being the order of the
canonical meromorphic pseudodifferential operator Uj . It is clear that for
sufficiently large N the operator ffiN(A) is nuclear on L 2 (sn ~ 1).
We define the trace of U by
102 Chapter 3.

N-]
SpU(A) = ~ SpU/A)+SpffiN(A).
j =0

§4.1 implies that Sp U is a meromorphic function on the A-plane.

Let U and ~ be meromorphic pseudodifferential operators of orders ao and boo
Then the operators U~ and ~U are also meromorphic pseudodifferential opera-
tors (theorem 3.4).

The 0 rem 4.3. The equality

Sp (U(A)~(A)) = Sp(~(A)U(A)) (4.8)

holds.

Proof. Write U and ~ as

N-]
U(A) = ~ U/A)+ffiN(A),
j =0

N-]
~(A) = ~ ~k(A)+G£N(A),
k =0

where Uj ' ~k are canonical meromorphic pseudodifferential operators of orders

aj' bk ; for ffiN the estimate (4.7) holds, while for fIN the analogous estimate holds
with aN replaced by fiN = RebN. Choose N large such that fiN -ao > n -1,
aN-fio > n -1. Then the composites of ~j and ffi N, Uj and G£N, and ffiN and G£N
are nuclear, so their traces do not depend on the orders of the factors. Hence it
suffices to prove (4.8) under the assumption that U and ~ are canonical operators.
So, let U and ~ be canonical meromorphic pseudodifferential operators of
orders a and b with symbols «P and '1', which may be assumed to be homogeneous
of orders a and b, resp., in each argument <p and O.
By the rule for composition (proposition 3.2.4)

U(A)~(A) = (4.9)

~ E(A+i(a +b -IYI))-]-.!,-DO«P(<p,8)a~'1'(<p,O)E(A)+9R(A),
o.:;;lrl.:;;N y.

~(A)U(A) =
§4. Traces of meromorphic i'DOs lO3

~ E(A+i(a +b -IYI))-1J,Dhi'(<f>,8)a~~(<f>,O)E(A)+ilC(A),
O,,;:;I-rI,,;:;N y.

while estimates of the type (4.7) hold for [Q and ilC.

We are to convince ourselves that for k = 0,1, ... ,

Sp [ ~ E(A+i(a +b -k))-I~aO~(<f>,O)a~i'(<f>,8)E(A)l (4.lO)

l-rl =k y.

= Sp [ ~ E(A+i(a +b -k))-I~aOi'(<f>,8)a~~(<f>'o)E(A)1.
l-rl =k y.

Since both sides of this equation are meromorphic functions of, say, the param-
eter a we may assume that k - a - b - n + 1 is not a natural number and apply
formula (4.4). Hence the situation reduces to checking the relation

~ f dcp f (-+--cp8+io)k-a-b-n~a8q,(cp,8)a~i'(cp,8)d8 = (4.11)

l-rl =k sn-I sn-I y.

= ~ f d<f> f (-+--cp8+ iol- a -b -n~aOi'(cp,f})a~q,(cp,8)d8.

l-rl =k sn-I sn-I y.

Note that the integrands in (4.11) are homogeneous of degree -n in <f> and in
8, and that (4.11) is equivalent to

f
I<l x l,lyl<2
(-+--~+iol-a-b-n ±
i.···.ik=IaXi l
ak~~~,y)
aXik
X (4.12)

X ak~~~,y) dxdy = f (-+--xy +iol- a - b - n X

aY11 aYik 1< Ix 1.ly 1<2

X ± aki'(x,y).
i],'" .ik=I aXil' .. aXik aYi l '" aYik
ak~(x,y) dxdy.

We show, e.g., the 'plus' relation (4.12). Put

J m•k - m = f (xy+io)k-a-b-n X
1< 1x I. Iy 1<2
akq, aki'
X ------------------- --------------------dxdy
aXi I . • • aXimaYi I • • • aY)k -m aYi I • . • aYim ax) I • . • aX)k_m
104 Chapter 3.

(the summation indices run over 1, ... ,n). The result of partial integration gives
for 1 :E;; m :E;; k,

Jm,k-m =Jm-l,k-m+I+(k-a-b-n)X (4.13)

ak -I «P
XI<lxl~YI<2 iI, ':~'im-' (lXi, ... aXim_,aYi, ... (lYik-m X
iI, ... ,ik-m

The function (Ay) ...... i'(x,y) is homogeneous of degree b both in X and in y.

Hence (4.13) can be written as

Jm,k-m = J m - l ,k-m+I+(k-a-b-n)(2m-k-1)hm, (4.14)

where

hm f (xy+io)k-a-b-n-I X
1< 1x I, Iy 1<2
(lk - 1«p ak - 1i'
X ~ ~.
i,," ',i",_, (lXi, ... (lXi m_, (lYi, ... (lYik-m 0Yi, ... OYi",_,(lxi, ... OXik_m

From this we derive that

k
h,O-JO,k = (k-a-b-n) ~ (2m-k-1)h m (4.15)
m=1

[(k + 1)/2]
= (k-a-b-n) ~ (2m-k-1)(h m -h k - m + I ).
m=1

The equality

Jp,O = Jo,p (4.16)

for p = 1 follows from (4.14). Assume that the relations (4.16) hold for
p = 1, ... ,k -1. We show that (4.16) also holds for p = k. The induction
hypothesis implies that hm = hk -m + 1 for m = 1, ... ,[(k + 1)/2]. Indeed, put
a2m - 2«p
«Pm = ,
aXil ... aXim_, aYi, ... OYim_'
§4. Traces of meromorphic itDOs 105

where iI, ... ,im -1 ,j 1, . . . ,jm -1 is an arbitrary set of indices. The functions «Pm
and itm are homogeneous of degrees am = a -m + 1 and bm = b -m + 1, resp.
(in each of the arguments x and y). The formula hm = hk - m + 1 follows from the
equations

f (xy + iO)k -2m + l-a m -bm -n X

1< Ix I, Iy I <2

-
ak - 2m + 1 «P a k - 2m + 1 it 1dxdy = 0,
a"./v / 1 .,. a.."v /k-2m+1 aX/ 1 ... ax/k-2m+l

which are satisfied by assumption. By now applying (4.15) we are led to (4.12),
which was required. So, (4.12) holds, and with it (4.10).
It remains to convince ourselves that the traces of the operators IDe and 91 from
(4.9) coincide. The number N in these formulas IS subject to
N > Re(a + b) + n - 2, ensuring that IDe and 91 are nuclear.
Let {Jj} be some orthogonal basis in L 2 (sn -1). Since the operator

91(A) = U(A)~(A) +
~ E(A+i(a +b _lyJ))-I~D8<<P(4),8)a~it(4>,8)E(A)
00;;; Iyl o;;;N y.

is nuclear, the series -:i/91(A)!J,!J) converges. Every term in it is a meromorphic

function of a in the halfplane N > Re(a + b) + n - 2, hence the sum of the series,
i.e. Sp 91(A), has the same property. Everything said remains true for the operator
IDe(A) also. The equality

Sp 91(A) = Sp IDe(A) (4.17)

holds for Re(a + b) < 1 - n. In fact, this equality ensures that the composite of
U(A) and ~(A) is nuclear, hence that Sp (U(A)~(A» and Sp(~(A)U(A» coincide.
This and (4.9), (4.10) imply (4.17). Uniqueness of analytic extension implies that
(4.17) holds for Re(a +b) < N -n +2 also. •
106 Chapter 3.

Rem ark 4.4. Theorem 4.3 holds also for matrix operators U and lB. Indeed,
if U is a canonical matrix operator, then (4.4) or (4.5) holds for its trace (replacing
«P(cp,O) by Sp «p(cp,lI)). Equality (4.12) is also preserved, and the sign of Sp must be
specified in it.

§5. Meromorphic pseudodifferential operators on strongly oscillating functions

Here we clarify the asymptotic behavior of an expression
e -i/Lg(q,)U>f~<I>(A)ei/Lg(>f)u(I}), where U(A) is a meromorphic pseudodifferential opera-
tor, as IA12 + 1-'2 ~ 00, I1m AI ~ h, I-' E IR. In particular, theorem 5.1 justifies the
terminology 'pseudodifferential' for operators U(A) (cf. the definition of
pseudodifferentia1 operator in the article [72]). This theorem will be repeatedly
used in the sequel.
Let g,u E c oo (sn-1), with g a real-valued function. Extend u and g to IR n \0
as homogeneous functions of O. degree Let also
U(A) = Ew~q,(A+ia)-l«p(cp,w)Ey->w(A) be a canonical meromorphic
pseudodifferential operator of order a, let 0 = ReA and I-' E IR.

The 0 rem 5.1. For any P ;:;:;. 0 there are nonnegative integers Nand Q such that
the inequality

Ie -i/Lg(</»U>f~</>(A)ei/LgC>f)u - lal.:;;;N
~ ~«pca)(cp'I-'Vg(cp) + ocp) X
0:.
(5.1)

X D; {u (l})piCA+inI2) ei/L(gC>/J)- g(</»)-i(y -</>,/L"Vg(</»+o</»} I;~~ I ~

~ c(p.2 +(2)- P12 llu ;cQ(sn -1)11

holds, where p = Iy I, I} = ylly I, «p(a)(cp,w) = D~«p(cp,w).

The proof is divided into several steps. In order to simplify notations we
write «P(w) instead of «p(cp, w).
A) In this step we derive the basic asymptotic formula, but not yet estimate the
remainder. Let ~ E Co (IR) with nt) = 1 for It I < 1/4 and nt) = 0 for
It I> 112. Put
v(r,p,cp,I},I-') =
§5. M eromorphic i'DOs on strongly oscillating functions 107

= u(t/J) f pi(Hinl2)~(vdist(cp,t/J»~(v (1- r / p»w(A + in/ 2)dA,

y'2;1m ~'=-T
where v = (j.t2+o2)l!4, 0 = ReA, p. E IR, dist(cp,l/-') is the distance on sn-I
between two points cp and 1/-', and W E C(f (IR + ). Permuting the operator
Ay-+x = Fi--lxCP(~)Fy-+~ and the integral we obtain

A (e ipg (t)v (r, p,CP,t/J,p.» = _ b-

v2'1T
X (5.2)

X f A[e ipg (t)pi(HinI2)u(t/J)w(A+inI2) X

ImA=T

X nvdist(cp,t/J»~(v(1-r/ p»]dA.

Rewrite the integrand as

j j ei(x -y)xCP(x)ei(xev)-x(x)-ev -x,~x» ei(x(x)+ev - x, ~x)) q (A, p., x,y )dyd~, (5.3)

where X(Y) = p.g(t/J)+olnp, q(A,p.,x,y) = u(t/J)w(A+inI2) X

X t(vdist(cp,t/J»t(v(l- r I p»p -T-nI2, ~x = \lX(x) = p.\l g(x) + ox I 1x 12 =
= r- 1{j.t\lg(cp)+ocp). Put h(x,y) = X(Y)-X(x)-(y -x,~x) and note that (5.3) is
equal to

We will assume that Rea is a sufficiently large positive number. By expanding

+ '11) using Taylor's formula we obtain
CP(~x

A [e ipg (>¥)pi(HinI2)u(l/-')w(A+inI2)f(vdist(CP,I/-')f(v(l-rl p»] = (5.4)

= jjei(X-Y'Jrleih(x,y)+iX(X) }: J,cp(a)(~x)'I1aq(A,p.,x,y)dyd'l1+
lal";;NY'
1
+ }: j(1-O)NR a(x,A,p.,O)d(),
lal =N+l 0

where

Consider the terms with 1 a 1 .;;;:; N. Their sum can be written as

108 Chapter 3.

~ ~«I>(a)Cxx)eiX(X) D;(eih(X,y) q(A,ji-,x,y» iy =x'

lal";;'N u ,

Recall that for sufficiently close f/J and \fI (r and p) the quantity
r(vdist(f/J,\fI»r(v(l-rlp» equals 1. Moreover,

Hence

D;(e ih (x,Y)q (A,ji-,x,y» iy =x =

= D;(eih(X'Y)u(\fI)(rl py+nI2) iy =xW(A + inl2)

(we assume that the function \fit-> u(\fI) has been extended to IR n \ 0 as a homo-
geneous function of degree zero). Thus, the righthand side of (5.5) takes the form

ei/Lg«j»r- a ~ ~ri(Hinl2)«I>(a){p.'Vg(f/J)+of/J) X (5.6)

lal";;'N u ,

Taking (5.4) into consideration, we can rewrite (5.2) as

rae -i/Lg(<j» Ay ..... XCei/Lg(>i')v (r, P,f/J, \fI,ji-» = (5.7)

= _1_ f ,i(;>"'+inI2) ~ _1 q,(a){p.'Vg(q,)+oq,),lal X

y'2;Im;>"'=T lal";;'N u !

Note that (5.6) becomes, by multiplication with rae-ipg(<j», a homogeneous

function of degree zero (in the variable x). Hence, after applying the Mellin
transform M r .....Hinl2 to (5.7), the terms at the righthand side with I u I ,,;;,; N coin-
cide with

~ ~«I>(a){p.'Vg(f/J)+of/J)rlal X (5.8)
lal";;'N u .
§5. Meromorphic i'DOs on strongly oscillating functions 109

In view of the fact that this expression is independent of r (it is a homogeneous

function of degree zero) we may assume that r = 1, hence Ix I = Iy I = 1,
q, = I/; (after differentiation). Finally we take into account that
eih(x,y) = eil'g(t)-il'g(<j»-r -I(y -x. l''Vg (<j»+ a<j»(r I p)-ia.

As the result the sum (5.8) takes the form

}: ~«p(a)(,u\7g(q,)+oq,) X (5.9)
lal";;N u ,

X Da[u(l/;)pi(HinI2)eil'(g(t)-g(<j»)e-i(y-<j>,I''Vg(<j»+a<j»llly I =1 X
y t=<j>

X w(A+inI2).

Now we have to clarify what happens with the remaining quantities in (5.7)
under the action of the Mellin transform. We first consider the 1efthand side.
Write A as convolution with the function G = p-I «P,

(Af)(x) = fG(x -y)f(y)dy.

Then

A(eil'g(t)v (r, p,q"I/;,p,» = _1_ f u (I/;)eil'g(t)dl/; X

V2;;S·-I

f f pi(A+inI2)+n
00
X t(vdist(q"I/;»)dl\. - I t(v(l- rl (p»G(rq,- pt/J)dp.
ImA=T 0

(In this and the previous equality we may assume that Re a is a negative number
whose absolute value is sufficiently large, so that all transformations are justified.
The final formulas are generalized to the values of a required by analytic exten-
sion.) Putting t = rip we obtain

f piHnl2-1 t(v(l- rl p»G(rq,- pl/;)dp =

00

= ri(A+ia +in/2) f G (tq,-I/;)t -i(Hinl2+ia)-1 f(v(1- t»dt.

00

o
This implies that the lefthand side of (5.7) is equal to
110 Chapter 3.

e- i/1g(<!»_l_ j r i (>..+inI2)11{A+inI2)d"A X
-J2;ImA=".

X j u(l})e i/1g(>/;)f(pdist(CP,I}))dl} X
s. -1
00
X j G (tCP-I})t - i (A+ inl2+iu)-1 t(V (1- t))dt.

o
After Mellin transformation this expression takes the form

e- i/1g(<!»w(A+inI2) X (5.10)

X j [7G(tcp-I})t(P(l-t))t-i(A+inl2+iU)-ldt] X
s· -1 0

We tum to the remainder in (5.7). By putting

1
ir a(ex,1)) = (N + l)j(l-8)Nrp(a)(ex+fh))dfJ
o
and by taking into account that eix(x) = ei/1g(<!»ri(A+inI2)rdnI2, we can rewrite it
as

_l_r u j r i (A+inI2)w(A+in 12)d"A X (5.11)

-J2; ImA=".

X j j ei(x -Y)T/+ih(x'Y)u(l})t(pdist(CP,I}))t(p(l- rl p))(rl pr+ n'2 X

X ~ ~1)air a(ex ,1))dyd1).

lal=N+l a .

Since ex = r-1{jL'Vg(CP)+ocp), we have

1
ir a(ex,1)) = (N + l)r lal-u j(1_8)Nrp(a){j.t'Vg(cp)+ocp+fh)r)dfJ =
o
= r I a I -uir a{j.t'V g(cp) + ocp, 1)Y).
By denoting 1)r by 1) again and rl p by t, we see that (5.11) equals

_l_e- i/1g(<!» j r i (>..+inI2)w(A+nI2)dA X

yI2; ImA='"
§5. Meromorphic 'I'DOs on strongly oscillating functions III

00

X f t -i(J..-n/2)-1 r(v(l- t»dt X

l
o

X si/I'g(t) r(vdist(q,,~»ej(q,-t -I t )(1/+ l'Y'g(q,)+aq,)d~

The expression between brackets is independent of r. Hence we obtain after Mel-

lin transformation,

W(A+inI2)e- i l'g(q,) f ~ ~'T/a'I'a(p,V'g(q,)+oq,,'T/)d'T/ X (5.12)

R"lal=N+l a .
1
X ft-i(J..-in/2)-Ir(v(l-t»dt X
o
X f e il'g(t) u(~)t(vdist(q,,~»ei (q,- t -I t )(1/+ l'Y'g (q,) + aq,) d~.
S"-I

So, as the result of Mellin transformation, the Iefthand side of (5.7) coincides
with (5.10), and the righthand side transforms to the sum of (5.9) and (5.12). Con-
sequently, after division by w(A + in 12) we are led to

e -il'g(q,)Ut-+q,(A)eil'g(t)u(~) = ~ ~q,(a)(p,V'g(q,)+oq,) X (5.13)

lal,,;;;N a .

X D a {u (~)pi(H in 12) eil'g(t)- g(q,»-i(y -q,.I'Y'g(q,)+aq,)} I p= 1 +

Y q,=t

+ J(l_U)N
o
~
lal =N+l a.
J,!!Ra(U)dlHe-il'g(q,) f
S"-I
[7 G(tq,-~)
0
X

X t -i(Hinl2+ia)-I[l_ nvdist(q,,~)H(p(I- t»]dt) eil'g(t)u(~)d~,

where

!!Ra(8) = (5.14)

= e -il'g(q,)(N + 1) ~ f ~'T/aq,(a)(p,V'g(q,)+ oq,+8'Tj)d'T/ X

lal =N+l a.

f t -i(J..-inI2)-1 t(p(l- t»dt f e i(q,-t- t)(1/+I'V'g(q,)+aq,) X

00
X 1

o
112 Chapter 3.

B) We will prove that the last term on the righthand side in (5.13) allows the
estimate 0 «(p,2 + 0 2) -q) as p.2 + 0 2 ~ 00, for any q. Recall that the inner integral
(with respect to the variable t) is understood as the analytic extension of a conver-
gent integral, i.e. it equals the sum of the expressions (1.l2) and (1.l3) in which
Xj(t) must be replaced by X/t)[l-r(pdist(q"I/;))r(p(1-t))Y, j = 1,2. We consider
one of these expressions, e.g.

(-It X (5.15)
(a + inl2- iA) ... (a +nI2+p -1- iA)
X f f t -i(A+in/2+ia)+p -1--_-1
[
00 dP-l
X
sn-l 0 dtP

X {XI (t)[ 1- r(pdist(q"I/;))t(p(l- t))]G (tq,-I/;) }dt] eil'(g(>/,)-g(</»)u (I/;)dl/;,

where p can be assumed sufficiently large. We introduce the new variables

s = -In t, Z = g(l[;) (without loss of generality we assume that the z variables are
local coordinates on the support of the function u). Using the equality

e-iCJlnl+ipg(l/;) = (_1)q(02+p.2)-q~fzeiCJs+iJ.lz,

we rewrite (5.15) as

f f (q"z,s, A'P.)(02 + p.2) -q M,ze iCJS + il'z dsdz;

sn-l

here f (q"z,s, A,p.) = 0 for s < - N and sufficiently large N (in view of the pres-
ence of the factor Xl (e -S)), while f is rapidly decreasing as s ~ + 00 (since P is a
large number). Transferring the operator t:..q to f and recalling that
p = (p,2+~)1/4 we obtain the estimate required for (5.15). The second expression

(containing X2) is treated similarly.

C) We now tum to the integral
I
j(l_O)N ~ J.,6Jva (O)d(}, (5.16)
o lal=N+l u ,

where 6Jva is defined by (5.14). Represent in (5.14) the function q,(a) as a sum
Xlq,(a)+X2q,(a), where Xj E Coo(R n), Xj;:;;"O (j=1,2), Xl+X2 = 1, and
XI(X) = 1 for Ix 1< 112 and Xl(X) = 0 for Ix I> 3/4. Then 6Jva = 6Jv~1)+0t~2),
§5. Meromorphic i!DOs on strongly oscillatingfWlctions 113

where '!Jl~) is obtained from '!Jla by replacing cI>(a) by XjcI>(a). In this step we esti-
mate the integral (5.16) with '!Jl~) instead of '!Jla .
Putz = (t-ll/;J, ···,t- 1I/;n) E IR n, 11/;1 = 1,t E IR+. Wehave

(N+l)-I'!Jl~2)(0) = (5.17)

= J ~ ~1ja(x2cI>(a»(}JS7g(CP)+ocp+ih)d1j X
lal=N+ICl.·

X J ei(t/>-zJr/+ih(t/>,z) I z I -T- n12 t(v(1- I z I -I »f(vdist(cp,I/;»u (z)dz,

where h(cp,z) = ,.,.(g(If;)-g(cp»-(z -cp)C/L'Vg(cp)+ocp) + o1n I z I, while u is, as

before, a homogeneous function of degree zero. Introduce new coordinates
g= (cp-z)v, Ie = 1jlv and rewrite (5.17) as

(N + 1)-1'!Jl~)(0) = vN+ 1Je iEK Cx2cI>(a»C/L'Vg(cp) + ocp+ (JVIe) X

X DHeih(t/>,Z)t(v(l-lz 1-1»t(vdist(cp,I/;»u(z)lz I- T- nf2 }dgdle,

with z = cp - gv - 1. Represent this expression as a sum

(N + 1) -1'!Jl~~)1 + (N + 1) -1'!Jl~~, by regarding lie I < (2v) -I I,.,.'Vg(cp) + ocp I in the
first and llel >(2V)-II,.,.'Vg(cp)+ocpl in the second term. We first consider the
first term. Obviously, 1,.,.'V(cp)+ocp+(Jvk I >c'? for a certain constant c >0 and
all 0 E [0,1]. Hence

I Cx2cI>(a»C/L'Vg(cp) + ocp+(Jvk) I ~ c I,.,.'Vg(cp) +ocp+(Jvk I Rea -I a 1 ~

This implies the inequality

1'!Jl~»)(0)1 ';;;c,?Rea-N+n X

X JIDg{eih(</>,z)t(v(l-lz 1-1»f(vdist(cp,I/;»lz I- T- n12 u(z)} Idf

The estimate h (cp, z) = ,? 0 ( I cp - Z 12 ), the definition of f, and the relation

D ~ = - V -I D z imply that the derivative

(5.19)

has support not containing the point z = °

and located in a ball I~ I ,;;; K, and
remains bounded as v~ 00. Together with (5.18) this leads to the inequality
114 Chapter 3.

(5.20)

Further,

(N + 1) -10t~:~(lJ) =

= 0' + I j ei~1C I KI -2M(x21f.>(a)(p,Vg(CP) + Ocp+ lJVK) X

X dr DH eih(CP,z)K(v(l- Iz 1-1»t(vdist(CP,I/J» IZ I- T - nI2 U(z) }d~dK,

where M is an arbitrary natural number and integration is over the domain

{(K,~): ~ E IR n, IKI > IIlVg(CP)+oCPI/(2v)}. Taking into account that
Rea - N -] ,,;;;; 0, we obtain

10t~:~(lJ) I ~ c0' + I X

X j IKI-2MdKjldrDHeih(<I>.z)K(v(l-lz I-I» X
IICI >cv

Using the property of the derivatives (5.19) already mentioned we can thus derive
the inequality

10t~:~(lJ) I , ; ; cv- 2M + n +N + Illu ;C 2M + N + I(sn -1)11 (5.21)

for all lJ E [0, 1].

So, choosing first N in (5.20) and then Min (5.21) we see that

)(l_lJ)N ~ J,~~2)(lJ)dlJl,,;;;; cv- 2P llu ;cQ(sn -1)11 (5.22)

o
1
lal=N+l a .

for any given P and certain Q.

D) We estimate the other 'part' of the integral (5.16), obtained by replacing 0ta
with 0l.~1). Recall that at the begin of the proof the number Re a was assumed to
be large. The estimates in the previous step do not depend on this assumption. In
order to finish the proof it suffices to perform an analytic extension with respect
to a of (5.16) (with 0ta replaced by 0t~) and to obtain an estimate analogous to
(5.22) for this extension.
If Rea - N -] > -n then, as in C), we have
1
j(l-lJ)N ~ J,0t~I)(lJ)d(J = (5.23)
o lal =N+I a.
§5. M eromorphic it DOs on strongly oscillating functions 115

I
= (N + 1)v N + 1 j(1-0)NdO j ~ ~ I K 1- 2M (xIIl>Ca) X
o lal =N+I a.

where
V(X,V,K) = Jeil(~6.rDHeihCcJ>,z) Iz l- r - n12 t(v(1-lz 1-1»nvdist(<p,I/J»u(z)}d~. Put
e (v) = J.t v g (<p) + o<p and introduce a new variable s = e (v) + Ov K, after which the
righthand side of (5.23) takes the form
1
(N + 1)( Ie(v) I / v)-2M~ + I-n j(1_O)N 02M -ndO X (5.24)
o
1
X j ~ - , (xIIl>Ca)(s)w(x, v,O,s)ds,
lal =N+I a.

where v (x, v,O,s) = Il-s / Ie (v) 11- 2M v(x, v,(s - e(v»e- I v-I). The inner
integral is equal to

(r + 1) j ~ ~(xlll>(a)(s) ~ S Yds X (5.25)

lal =N+I a. Iyl =f+1

1 1
X j(1-t)Y_, wCY)(x,v,O,ts)dt+
o y.

1
+ ~ ~ -'-I w(Y)(x,v,O,O)j(xIIl>Ca)(s)sYds,
lal=N+I Iyl=oa.y.

where w(Y) is the derivative of w(x, v,O,s) with respect to s.

The first integral converges for Re a - N + 1 + r > - n, while the integrals of
the product (xIIl>Ca)(s)s y can be extended in the usual way as meromorphic func-
tions to the whole a-plane (see, e.g., [13]). Thus, substituting (5.25) in (5.24) we
obtain the looked-for analytic extension of the integral (5.16) (with 0t~I) instead of
01,a )·

The properties of the derivatives (5.19) (indicated after (5.19» ensure the esti-
mate

IvSY) (x, V,K) I ",;; cllu ;C 2M +N + I(sn - I )11.

Hence
116 Chapter 3.

(5.26)

for Is I ",;;; 1. Note that c 1 v"';;; Ie(v) I I v",;;; C2V for v ~ 1 (with positive constants CI
and C2). Choose r such that the inequality Rea - N -1- r> -n holds for a
given a. Combining (5.23) - (5.26) we obtain for sufficiently large M relation (5.22)
with 'iit~2) replaced by 'iit~I). This implies that (5.22) remains valid if 'iita is written
instead of 'iit~2). It remains to use the estimate proved in part B) and formula
(5.13) . •

§6. Estimates for meromorphic pseudodifferential operators

In this paragraph we will prove analogs for meromorphic pseudodifferential
operators of inequalities which are well-known for classic pseudodifferential opera-
tors. An analogue of Giirding's inequality will be established, the essential norm
of an operator will be computed, etc.

The 0 rem 6.1 (generalized Giirding's inequality). Let

U(A) = EO .....cj> (A + ia)-Icf>(cp,O)~ .....o(A) be a canonical meromorphic
pseudodifJerential operator, let a be a real number, let Recf>(cp,O) ~ c = const for
cp,O E S n -I. Then for every positive number f and every set ~ that is admissible for
U there is a constant c' = C'(f,~) such that the inequality

Re(U(A)u,U) + c'llu ;H(a -1)/2(A,sn -I )11 2 ~ (6.1)

~ (c -f)IIE«1-iaI2)u ;HO(sn -1)11 2

holds for all A E ~ and u E Coo (S n - I); here (1 = Re A if I Re A I ~ 1 and (1 = 1 if

IReAI < 1.
(Taking IIE«1- ial 2)u ;HO(sn -1)11 as (equivalent) norm in H af2 (A,sn -I), rela-
tion (6.1) can be given a form better resembling the usual Giirding inequality.)

Proof. Denote by U/(A) the canonical meromorphic pseudodifferential opera-

tor with symbol iIm cf>(CP,O). Then the adjoint operator U/(A)* equals, by proposi-
tion 2.5, -iE(X+ia)-IImcf>(cp,O)E0) (up to terms of order a -1). Applying pro-
position 2.2 and again neglecting terms whose orders do not exceed a - 1, we may
change Xto A in this formula. So, up to such terms, U/(Ar = - iE(A + ia)-I X
X Imcf>(cp,O)E(A). This and the equality 2Re(U[(A)u,u) = (U(A)u,u)+(U[(A)*u,u)
§6. Estimates for meromorphic it DOs 117

implies the estimate

on every admissible set 'if. Hence the symbol 4> may be regarded as being real in
the proof of (6.1). Moreover, proposition 2.2 implies that it suffices to check (6.1)
for real A satisfying I A I ;;. I.
We denote by '1' the positive function defined by '1'2($,0) = 4>($,O)-c +£. Let
meA) be the canonical meromorphic pseudodifferential operator of order a 12 with
symbol '1'. The operator m(A.r coincides (up to terms of order aI2-1) with the
operator E(A.+iaI2)-lit(cp,O)E(A.). By theorem 3.4 the asymptotic series for
m(A)'m(A) has as initial term E(A.+ia)-lit 2(cp,O)E(A), while the second term is an
operator of order a - 1. Hence the order of

is also a - I. Since the symbol of U(A) is real, proposition 2.5 applied to U(A)
leads to the inequality

Re(U(A.)u, u) -(c -£)(E(A + ial2) -I E(A - iaI2)u, u)- (m(A)*m(A.)u,u) ;;.

;;. c'llu ;H(a -1)/2(A.,sn -I )11 2

(in order to replace E(A. + ia) -I E(A) by E(A + ia 12) - 1E(A - ia 12) we use proposi-
tion 2.2). It remains to note that (E(A+ial2)-lr = E(A.-iaI2) (proposition
1.4.4) and that (m(A.)*m(A.)u,u) = Ilm(A)u ;HO(sn -I )11 2 ;;. O. •

The 0 rem 6.2. Let 4> E coo(sn-I XS n - I ), K = 114>;c(sn-1 XSn-I)II, and

U(A) = E(A + ia) - 14>(cp, O)E(A), where a is a real number. Then for every positive
number { and every set 'if that is admissible for U there is a constant c such that

IIU(A.)u;HO(sn-I)II.s:;; (6.2)

.s:;; (K +£)IIE(o-ia)u ;HO(sn -1)11 +c Ilu ;H a -1!2(A,sn -1)11;

here, as before, 0 = ReA. if I ReA. I ;;. 1, and 0 = 1 if I ReA. I < 1.

Proof. Proposition 2.2 implies that it suffices to check (6.2) for real A. satisfying
IA.I ;;. 1 only. Put it(CP,lJ) = 4>(CP,O)4>(CP,lJ) and m(A.) = E(A.+ia)-lit(cp,O)E(A.-ia).
By theorem 3.4 and propositions 2.2 and 2.5, the order of m(A) - U(A)*U(A.) is
118 Chapter 3.

2a -1. Hence

Since K2 - i.'(cp,tl) ~ 0 for cp,O E Sn -I, by applying theorem 3.1 to

E(A + ia) -1(K2 - i.'(CP,tl))E(A - ia) we obtain the inequality

K2Re(E(A + ia)-I E(A -ia)u,u)- Re(SE(A)u,U)+ (6.4)

+c'llu;H u - I12 (A,sn-I)11 ~ -€IIE(a-ia)u;H o(Sn-I)1I 2 .

Since

we finish the proof by combining formulas (6.3) and (6.4) . •

Rem ark 6.3. If a =1= +(k +nI2), where k = 0,1, ... , then we may put
a = ReA in (6.1) and (6.2) also if IReAI < 1.

Pro p 0 sit ion 6.4. Let U(A) be a canonical merom orphic pseudodifferential
operator of order a, with symbol Ill, and let Co = IIll(c/>o, 0o) I, with (CPo,Oo) an arbi-
trary point of sn - I X sn -I. Then for any sequence {€j} of positive arbitrarily small
numbers there are a sequence {Aj} of complex numbers and a sequence {Uj} of
COO(Sn -I)functions such that ImAj = 1" (1" a given number), IAj I ~ 00, and

IIIU(Aj)u ;HO(sn -I )11- Co IIE(aj - ia)uj;Ho(Sn - I )111 ~ (6.5)

~ €jlluj;Hu(Aj,sn -1)11,

where aj = ReA.i.
Proof. First of all we note that it suffices to prove (6.5) for 7' = O. Indeed, for
A = a+;1" we have U(A) = U(a) + T(A) , where
T(A):Hs(A,sn-I)~Hs-u+I(A,sn-l) is a continuous map with bounded norm
on the line ImA = T (proposition 2.2). Further,
IIT(A)U;HO(sn -1)11 ~ c lIu ;H u -I (Aj,sn -I )11 ~ c IAj I -lllu ;Hu(Aj,sn -I )11. This
and the truth of the proposition for T = 0 imply that the proposition holds for all
T.

Let first U be a canonical meromorphic pseudodifferential operator of order

§6. Estimates for meromorphic i'DOs 119

zero. Choose a real-valued function g homogeneous of degree zero such that the
vector V'g(<</>o) lies in the (<</>o,Oo)-plane and is nonzero. Obviously, the vectors </>0
and V'g(<f>o) are orthogonal. Expand 00 with respect to the vectors <f>o and
V'g(<f>o),Oo = p,oV'g(</>o)+ao</>o, and put /.t = t/.to, a = tao, A = a+iT. By apply-
ing theorem 5.1 we find that for a function U E coo(sn-I) with support in a
small neighborhood of <f>o the inequality

holds for sufficiently large positive t and given f > O. TIris means that for any
sequence {fj } of arbitrarily small numbers there are a sequence
{Uj} C Coo (sn -I) and a sequence {tj} of arbitrarily large positive numbers such
that Iluj;HO(sn -1)11 = 1 and

Ie -i/ljg(</»Uo/---><p(Aj)//ljg(o/)u/1/I)-(I(<f>o,Oo)Uj I < fj'

where r//."j = tj"//.o j = tj

r ' a" ' A') =
"(10 (1
j + iT . TIris implies that for the sequence
{Vj = ei/l}g Uj} the relation

(6.6)

holds, which immediately leads to (6.5) with a = O.

Let now a be an arbitrary number. Using proposition 2.2 the operator
U{A) = E{A+ia)(I(<f>,O)E(A) is written as

U(A) = E(A)-I(I(<f>,O)E(A-ia)+T{A),

where ImA = 0, while the operator T allows the estimate

1IT(A);Hs(A, sn -I) ~ H S -a + 1(A,sn -I )11 ~ const. TIris allows us to check (6.5) for
E(A)-I(I(<f>,O)E{A-ia) (cf. the begin of the proof). Clearly, the latter operator
equals UO(A)E(A)-IE{A-ia), where UO(A) = E(A)-I(I(<f>,O)E(A) is a canonical
meromorphic pseudodifferential operator of order zero. Put
Wj = E(Aj-ia)-IE{Aj)vj, where Aj and Vj are the same as in (6.6). Since E{Aj) is
unitary for ImAj = 0 we have

I IIU{Aj)Wj;HO{sn-"I)II-coIIE{Aj-ia)wj;HO(sn-I)11 I ~
~ II(Uo(A.i) -(I(<f>o,{}O»E(Aj)-1 E{Aj - ia)wj;HO(sn -I )11 =

= 11(Uo(Aj) -(I(<f>o,Oo»vj;H O(sn - I )11.

120 Chapter 3.

Using formula (6.6) for Uo (it holds for this operator) we are led to the estimate

I IIU(A.J)wj;Ho(Sn-I)II-coIIE(i\j-ia)wj;Ho(Sn-I)11 I .;;;
:;;;; f)lvj;HO(sn -1)11.

Note that

IIVj;HO(sn-I)11 = IIE(i\j)-IE(i\j-ia)wj;Ho(Sn-I)11 =

= IIE(i\j - ia)wj;Ho(Sn -1)11 .;;; cllwj;Hu(i\j,sn -I )11

(proposition 1.5.5). Hence {Wj} can play the role of the sequence whose existence
is asserted in the theorem. •

The 0 rem 6.5. Let U(i\) = E(i\+ia)-I«fI(cp,8)E(i\), let a be a real number, and
let K = 1I«fI;C(sn-l,sn-I)II. Assume also that the line {i\:Imi\='T} does not con-
tain poles of U. Then for every positive f there is a uniformly bounded operator
'0(i\):Hs(i\,sn -I) ~ H S -u + I(i\,sn -I) such that on this line

where 0 is defined as in theorems 6.1, 6.2 (cf remark 6.3); the constant K is sharp,
i.e. it cannot be replaced by a smaller number while preserving inequality (6.7).

Proof. Proposition 2.2 implies that it suffices to prove (6.7) for Imi\ = O.
Denote by X a function from Co (IR) such that X(t) = 1 for It I < Nand
X(t) = 0 for It I ~ 2N. Introduce the operator X(i\) by
00 km
X(i\)u = ~ ~ x(m 2 +i\2)Um k Y mk
m =0 k = 1

and put '0(i\) = U(i\)x(i\). By theorem 2.2 we have

11(U(i\) - '0(i\»u;HO(sn -I )11 = 11U(i\)(l- x(i\»u;HO(Sn -I )11 ,,;;; (6.8)

:;;;; (K +f! 2)IIE(o - ia)(l- x(i\»u ;HO(sn - I )11 +

+ c 11(1- X(i\»u ;H u -1!2(i\,sn -I )11.

Taking into account that E(i\) has the representation

§6. Estimates for meromorphic itDOs 121

E(A) = ~ ILm(A)Umk Y mb
m,k

where the ILm are complex-valued functions, we obtain

IIE(o- ia)(I-X(A»u;HO(sn -I )11 =

= 1I(I-x(A»E(o-ia)u;HO(sn-I)11 ~ IIE(o-ia)u;HO(sn-I»II.
The definitions of the norm and the function X imply the estimate

By proposition 1.5.5,

Ilu ;Ha(A,Sn -1)11 ~ cIIE(o- ia)u ;HO(sn -1)11.

This and (6.8), (6.9) imply for sufficiently large N inequality (6.7).
It remains to convince ourselves that K is sharp. Suppose this is not so, and
that for some operator T inequality (6.7) holds with K replaced by Q < K. Then
we have for the sequence {Uj} from proposition 6.4,

(co -£j)IIE(Aj - ia)uj;HO(sn -1)11 ~

~ IIU(A)u ;HO(sn -I )11 ~

~ 1I(U(Aj)0(Aj»uj;HO(sn -1)11 + 110(Aj)U ;HO(sn -1)11 ~

~ (Q +£)IIE(Aj -ia)uj;HO(sn -1)11 +clluj;H a -1{A,sn -1)11.

From the definition of the norms in the spaces HS(A,Sn -I) and proposition
1.5.5 we obtain that for sufficiently large j,

clluj;H a -1{Aj,sn -1)11 ~ C I Aj I-Iliu ;Ha(Aj,sn -1)11 ~

~ £IIE(Aj -ia)uj;HO(sn -1)11.

Thus,

Taking Co > Q, the number j large, and £ small, we obtain a contradiction. •

122 Chapter 3.

§7. Periodic meromorphic pseudodifferential operators

A canonical meromorphic pseudodifferential operator arises as the result of apply-
ing the Mellin transform to an operator Fi=1xIP(<p,g)Fy_+ where <p = xlix I (cf.,
e.g., (2.3.6». Dependence of the symbol on <p means that it has a discontinuity 'of
the first kind' at the coordinate origin (the limit as x ~ 0 depends on the direction
of approach). Now we consider an operator Fi=1xIP(r,<P,g)Fy .....~, r = Ix I. We will
assume that for any value of r > 0 the equality fP(re,<p,g) = fP(r,<p,~) holds, where
e is the base of the natural logarithm. Thus, fP can have a discontinuity 'Qf the
second kind' at x = O. Applying a certain integral transform to
Fi=1xIP(r,<p,g)Fy-->( will lead to a periodic meromorphic pseudodifferential opera-
tor. This transform is obtained by a change of variable from the transform used
by I.M. Gel'fand in the study of differential operators with periodic coefficients
(I5] (cf. also [42]).
We will first describe this integral transform. Using it we are led to a periodic
meromorphic pseudodifferential operator, starting from an operator
F-1fP(r, <p,g)F. We then establish for periodic meromorphic pseudodifferential
operators analogs of the results given in §3 for meromorphic pseudodifferential
operators (concerning composition, transition to the adjoint operator, etc.).

1. The transform G. For a v E Co (R +) we put

1 +00
(Gv)(.u,r) = 1/2 ~ (rei)-illv(re i ), (7.1)
(2?T) i=-oo

where I-' is an arbitrary complex number, and 0 < r < 00. It is obvious that
V(.u,r) = (Gv)(.u,r) is a 'periodic' function, i.e. V(.u,re) = V(.u,r). The inversion
formula for (7.1) can be verified immediately:

vCr) = (G- 1 V)(r) = (2?T;1/2/ riIlV (.u,r)dl-'; (7.2)

where by IT we denote the set {I-' E C: Iml-'='T, O,,;;;;Rel-',,;;;;2?T}.

A Plancherel formula holds for G:

(7.3)

Let's prove this formula. Without loss of generality we may assume that 'T = O.
§7. Periodic meromorphic i'DOs 123

Putting I/I(t) = v (e l ) we can rewrite (7.3) as

2'IT 1 + 00
j dp. j Ii'(p.,t) 12dt = j Ii'(t) 12dt,
o 0 -00

where
1 +00
i'f",t) = "-"' e -ij.t{1 + j)\{I(t +J").
\J" (2?T)1/2 j=~OO

Clearly, i'(p.,t + 1) = i'(p.,t). Denote by i'n(p.) the Fourier coefficients of

i'(p., .):
1
i'n(p.) = ji'(p.,t)e- 2'ITinl dt =
o

\/2 j j "}:OO l/I(t+j)e- iP.(I+j)-2'ITinl dt =

(2?T)
1

0 = - 00

+00
1
j l/I(t)e-i(p.+2'1Tn)ldt.
(2?T)1I2 _ 00

We have
2'IT 1 + 00 2'IT
j dp.jli'(p.,t)1 2dt = ~ j Ii'n(p.) I2dp. =
o 0 n=-ooo

which was required.

2. Canonical operators. In chapter 2 we have proved that the equation

Fi=lx41(~)Fy--+~v(y) = _1_ j r i (Hia+in/2)U(A)v(A+inI2,)dA,

...f);;IffiA=P-s

holds on an everywhere dense set (in Hp(lRn». Here

U(A) = E/J4>(A+ia)-l41(lJ)~--+/J(A), r = Ix I, 1/1 = yl Iy I, q, = xl Ix I· Taking
into consideration operators of more general nature we introduce the operator

(Av)(x) = _1_ j ri(Hia +in/2)~--+4>(r,A)v(A+inI2,I/I)dA; (7.4)

...f);;IffiA=T
124 Chapter 3.

here U(r,A) = E(A +ia)-I«P(r,q,,8)E(A), «P(re,·) = «P(r, ').

Proposition 7.1. For v E C(f(Rn\O) the equation

Av= (Gr--+(r,!,+ia +inl2» -I A(r, p.)Gr--+(r,!,+inI2) v (7.5)

holds, where 1m p. = '1',

A.J,--+<j>(r, p.)V(p.+ in 12,r, 1/1) = (7.6)

+00
~ r 2'ITif U",--+<j>(r, p.+ 2'1Tj)V;(p.+ in/2,1/I),
j=-oo

e
Vj(v,l/I) = jr- 2'ITij V(v,r,l/I)r- 1dr, V(v,r,l/I) = Gr--+(r,v)v(r,I/I).
1

Proof. Using the 'periodicity' of «P in r and formulas (7.1), (7.4) we find

Gr--+(r,p.+ia +inI2)Av = (7.7)

1 ~
= -2 (rej)-ip. f (re j )i.\U(r,A)v(A+inI2,·)dA =
'1T j = - 00 Im.\=T

= -2
1 j r;('\-p.) ~ e i (.\-p.)jU(r,A)v(A+in12,)dA.
'1T1m '\=T j=-OO

It is well-known (cf. e.g., [13]) that

+00 +00
~ ei(.\-p.)j = 2'1T ~ 8(A-p.-2'1Tj).
j=-OO j= -00

Substituting this in (7.7) we obtain

(7.8)
+00
~ r 2'ITijU(r, p.+2'1Tj)v(p.+2'1Tj + in 12,).
j=-oo

The definition of Mellin transform and formula (7.2) imply that

(7.9)
§7. Periodic meromorphic 'I!DOs 125

Expanding V in a series,
+00
V(A.,r,·) = ~ V k (A,)r 2'ITik,
k=-oo

we obtain

Jr-i(P-X)-1 V(A.,r,)dr
00 +00
= 27T ~ Vk(A.,·)~(A-/L+2k7T).
o k=-oo

Hence (7.9) implies v<IL,) = V k <IL-2k7T,), where k is such that

Re/L-2k7T E [0,27T]. Hence v<IL+27Tj+inI2,·) = Y.i<IL+in12,·). Now formula
(7.8) takes the form
+00
~ r 2wiJ U(r, /L+ 27Tj) Y.i<IL+ in I 2, ').
j =-00

From this we can derive (7.5) immediately. •

Denote by H S (IT), where II = [1, e] X S n - 1, the Sobolev-Slobodetskii space
obtained by completion of the set of smooth functions V satisfying the condition
V(1,O) = V(e,O), 8 E Sn -I.

Pro p 0 sit ion 7.2. Let '!f be an arbitrary closed set lying within the rectangle
{/L: 0.;;;; Re W:;;;27T, I Im/L I .;;;; h}, h < 00, and not containing the points
27Tj-i(a+k+nI2), 27Tj+i(k+nI2), where k = 0,1,"', j = 0,+1,···.
Then the estimate

IIAC/L)V;H S -Rea(II)II';;;; c('!f)IIV;HS(II)11 (7.10)

holds for the operator A defined by (7.6) on the elements V E HS(II).

Proof. If U is independent of r (i.e. cfI is independent of r), then (7.10) follows

from proposition 1.3. lbis estimate can be generalized to the case of an arbitrary
function cfI E Coo([l,elXSn-IXSn-l) satisfying cfI(r,') = cfI(re,') using the
series expansion cfI(r,q"O) = ~jr2'ITiJcfllq,,8) (cf. the proof of proposition 1.3). •
The operator A defined by (7.6) on smooth functions II :3 (r, I{;) 1-+ V(r,l{;)
126 Chapter 3.

satisfying V(1,~) = V(e,~) will be called a canonical periodic meromorphic

pseudodifferential operator of order a; the function 4> is called the symbol of this
operator.
Formula (7.6) implies the relation

(7.11)

3. General periodic meromorphic pseudodifferential operators. Next we will con-

sider meromorphic operator-functions which either have finitely many poles in
any strip I1m A I < h < 00, or satisfy (7.11) and have finitely many poles in any
rectangle {J.! EO C: 0.;;;;;;ReW,,;;2'1T, I ImJ.! I < h}. Every closed set lying in a strip
(rectangle) of this kind and not containing poles of an operator considered will be
called admissible.
A meromorphic operator-function J.! ..... A(-,J.!) satisfying (7.11), with r EO [I,e], is
called a periodic meromorphic pseudodifferential operator of order a 0 if for any
N = 0,1, ... , the inequality

holds on every set 'J that is admissible for the operators A(r, J.!), Air, J.!),
j = 0, ... ,N. Here Aj is the canonical periodic meromorphic pseudodifferential
operators of order aj with symbol 4>(j)(r, cp,O), Uj = Reaj' ao ;;;. U1 ;;;. u2;;;' ... ,
Uj --:> - 00 as j --:> 00. We will assume that the symbol 4>0) of a canonical operator
can be extended to R n \ 0 in each argument cp,O as a homogeneous function of
degree aj. The formal series 2.j4>(j)(r, cp,O) is called the symbol of A(r, J.!).

The 0 rem 7.3. For a given series (J there is a periodic meromorphic

pseudodifferential operator A with (J as symbol.

Proof. Using theorem 3.3 we determine a meromorphic pseudodifferential

operator U(r,;\) with symbol (J(r, cp,O) (the variable r is regarded as parameter).
The periodic meromorphic pseudodifferential operator looked for is defined, using
these operators U, by (7.6) . •
§7. Periodic meromorphic it DOs 127

4. The algebra of periodic meromorphic pseudodifferential operators.

The 0 rem 7.4. Let A,B be periodic meromorphic pseudodifJerential operators of
orders ao,b o, with symbols
00

o A (r, cp,O) -- ~ <p (J) (r, cp,O),

) =0
00
OB(r,cp,O) = ~ it(k)(r,cp,O),
k =0

where <p(J) and it(k) are functions homogeneous in cp and 0 of degree a) and bb
respectively. Then AB is a periodic meromorphic pseudodifJerential operator, of order
ao +b o and with symbol defined by

00 1 lib
OAB(r,cp,O) = ~ ~ - , Db<p(J\r,cp,O)r Y - kaiit(k)(r,x,O). (7.12)
j,k =01 Y 1>0 y.

Here ay = a lyl laxI I ••• aY'xm Db = (-i)IYI ab, aiit(k)(r,x,8) =

= L.)aHr 27Tij 'JrY)(x,8)), r = Ix I, cp = xl Ix I; moreover, 'Jr(k) is homogeneous of
degree b k both in x and in O. Thus, the terms of the series (7.12) are periodic func-
tions in r, i.e. oAB(r,cp,O) = oAB(re,cp,O).

Proof. It suffices to assume A and B to be canonical operators of orders a and

b with symbols <P(r, cp,8) and 'Jr(r, cp,8) respectively. We have
+00
A(r,p,)B(r,p,)V = ~ r 27Tl}U(r,p,+2j?T) X (7.13)
j= -00

+00
X ~ '.8) -k(p.+2k?T)Vk>
k=-oo

where '.8 q(p.) = E(p.+ib)itq(cp,O)E(p.). By proposition 2.2 the asymptotic series

expansion

U(r,p,+2j?T) ~ U(r,p,+2k?T)+ (7.14)

+ ~ ~py(cp,2i(j -k)?T)Eo-><j>(p.+2k?T+ i(a - Iy I))-1 X

lyl >1 y.

X Db <P(r, cp, 0) E,p->o (p. + 2k?T)

is valid, where Py(cp,p,) = Ix IIYI-I-'ai Ix II-'. Applying proposition 2.4 we obtain

128 Chapter 3.

E8 ......p{j.L+2k'IT+i(a -lyi))- I DecI>(r,cp,0)£v, ..... 8{j.L+2k'IT) X (7.15)

X Ew ......y{j.L+2k'IT+ib)-li'j_k(~,"')Ex-->w{j.L+2k'IT) ~

~ ~ E{j.L+2k'IT+i(a+b-IY+KI))-1 X
1"1 ;;;.0

Taking into account cancellation because of the homogeneity of the extension

i'j -k, we have

(7.16)

= "'"
~_I_rIYI+IKI-b-2iv-k)'lTaYr2iv-k)'lTa"i'.
, , x x ]-k
(x 0) =
,
y.K.

1 r IY+KI-b-2iv-k)'ITaY+"(r2iv-k)'lTi'._ (x 0))
(Y+K)! x ] k , .

Relations (7.13) - (7.16) imply

+00
A(r,J.L)B(r,J.L) ~ ~ ~ r 2k'ITi E8-->.p{j.L+2k'IT + i(a +b -lyl))-1 X
k= -00 lyl;;;'O

Denote by A(r, J.LY the operator adjoint to A(r, J.L) with respect to the scalar pro-
duct

---drdp
<U,V> = fe s·f
1 -I
U(r,cp)V(r,cp) .
r
(7.17)

Put A*(r,jJ) = A(r,J.Lr.

The 0 rem 7.5. Let A(r, J.L) be the periodic meromorphic pseudodifferential opera-
tor of order a 0 with symbol (JA (r, cp,O) = 2.cI>v) (r, cp,O), where the cI>v) are functions
homogeneous of degree aj both in cp and in O. Then A*(r,J.L) is a periodic mero-
morphic pseudodifferential operator of order 00, and its symbol is

i: ~
j=olyl;;;'o y.
J,rIYI-aJarDecI>V)(r,x,O)
§7. Periodic meromorphic vDOs 129

(the operations ar and D"'6 are understood as is done in theorem 7.4).

Proof. Assume, without loss of generality, that A is the canonical operator of
order a with symbol (l(r, q,,0). Then
00

<A(-,IL)U,V> = ~ (Uk-j<IL+ 2jw)Uj ,Vk ), (7.18)

j,k=O

where (j,g) is the scalar product in L2(Sn -1),

Uk<IL) = E8-->cp<IL+ia)-1(lk(q"U}~-+iI<IL), and (lk are the Fourier coefficients of
(l(.,q,,0). By proposition 2.5 the asymptotic formula

Uk-j<IL+2jW) -- ~ J,E<IL+2jW+i(a-!y!»-1 X (7.19)

1'Y1;;;'0 y.

X a~D"'6(lk-iq"o)E<IL+2j7T)

holds for Uk-j<IL+27Tj)* = Uk-j(jL+2wj). Applying proposition 2.2 we obtain

E<IL+2j7T+i(a-!yl))-la~D"'6(lk-iq"o)E<IL+2j7T) --

-- ~ J,pl(q,,2i (j - k)w) X
I I ;;;.0 /C.
I(

Substitute this formula in (7.19) and use a relation of the form (7.16). Then

Uk-j<IL+2jw) -- ~ J,r l'Yl-a-2i(;-k)71 X (7.20)

1'YI;;;'oY.

X E<IL+ 2k7T+ i (a - !Y 1)-lal(r 2i (; -k}71 D"'6(lk - j(x, 0)E<IL + 2kw).

It follows from (7.18) that

A*(r,IL)V = ~r2ij71Uk_j<IL+2j7T)Vk =
j,k
= ~r2ih~r2i(;-k)71Uk_j<IL+2jw)Vk.
k j

It remains to combine this equation with (7.20) . •

A periodic meromorphic pseudodifferential operator A(r, IL) is called elliptic if
its principal symbol (l(0) does not vanish anywhere. Theorems 7.3 and 7.4 allow
130 Chapter 3.

us to prove the following statement (cf. theorem 3.8).

The 0 rem 7.6. Let A(r, /L) be an elliptic periodic merom orphic pseudodijJerential
operator of order a o. Then there is a periodic meromorphic pseudodijJerential opera-
tor R(r, /L) of order - a 0 such that the symbols of the operators A(r, /L)R(r, /L) and
R(r, /L)A(r, /L) are equal to one.

§S. Change of variables in meromorphic pseudodifferential operators

Let g be a diffeomorphism of the sphere Sn -I, g(a) = (gl (a), ... ,gn(a»,
a E IR n, Ia I = 1, Ig (0) I = 1. We will assume that the functions gj are defined
on IR n \ 0 and are homogeneous of degree one, that g'(a) = Ilcrgj1crakllj,k=l, and
that Ig'(a) I is the modulus of the determinant detg'{a).

The 0 rem 8.1. Let U(A) = E{A+ia)-ICP(cp,w)E(A). Then

U1jH9(A)u(~) = Ig'(a)-l IU,.->o(A) Ig'{7") Iu(7"), cp = g{a),
where ~ = g{7"),
u(7") = u(g(7"», and U,.->o is the meromorphic pseudodijJerentia/ operator with sym-
bol

~ ~ cr~eih(W,O''')1 D~CP(g{a),(g'(a)-I)*w),
lyl =0 y. 1"=0

where, moreover, h(w,a,7") = (w,7"-g'(a)-lg(7"» = 0(10-7"12).

Proof. Recall that the kernel of lly-><I>(A) is given by

= f t -i(A+ia +in/2)-IG{q,,tq,-~)dt,
00
G(q,,~;A) (8.1)
o
where G(q"tg) = t -n -aG(q"g), t > 0 (§1.3). For simplicity of notation the depen-
dence on cp of G(cp,g) will not be indicated. Since the functions gj are homogene-
ous we have g(a) = g'(a)a. Hence the equation g(r) = g(a)+g'(a)(7"-a)+ R(a,7")
can be rewritten as g(7") = g'(a)7"+R(a,7"), R(a,7") = 0(10-7"12). This implies
that

G{tq,-~) = G(tg(a)-g(7"» = G(g'(a)(ta-7")-R(a,7"» = (8.2)

N (-l)IYI
~ t cr YG(g'(a)(ta-7"»R(a,7")Y+TN + 1(a,7";t).
lyl =0 y.
§8. Change of variables in meromorphic i'DOs 131

The term TN + I is a sufficiently smooth function of its arguments (for N large),

having only singularity of power type at infinity. Whatever given numbers p,s and
an admissible set 'J, we can choose N large such that the inequality

IITN +1,x;Hs(A,sn -I) ~Hs +p0,sn -1)11.,;;; c(p,s, §) (8.3)

holds; here TN + I,X is the operator on Sn -I with kernel

00
TN+I (a "'f"A) = jt- i (X+inI2+ia)-ITN+I (a "T·t)dt.
o
Apply the Mellin transform to both sides of (8.2). Using (8.1) we obtain

(-l)iYi
G(cp,l[;;A) =
N
~ t 00
j t -i(X+inIHia)-1 X (8.4)
iyi=o y. 0

X ay G (g'(a)(ta- T»dtR (a, T)Y + TN + I (a, T;A).

Since

j G(cp,l[;;A)u(l[;)dl[; = j G(g(a),g(T);A)U(g(T» Ig'(T) IdT,

relation (8.4) between kernels corresponds to the following equation for the opera-
tors:

EW~<I>(A + ia) -11f>(w)E~w(A)u(l[;) = (8.5)

N (-l)iYi
~ t Ew-+,,(A-i Iy I +i(a + Iy I»-llf>y(W) X
iyi =0 y.

where U(T) = and If>y(w) is the Fourier transform Fz-+w of

U(g(T»,
zl-+aYG(g'(a)z). In view of the estimate (8.3) for TN+I,x it suffices to consider the
other terms at the righthand side of (8.5). Put ~ = g'(a)z. Then

If> (w) = 1 je-iWZaYG(g'(a)z)dz =

Y (27T)nl2
132 Chapter 3.

Hence the righthand side of (8.5) can be rewritten as

N
Ig'(o)-11 ~ E",-+o(A+ia)-IcI>«(g'(o)-I)*w) X (8.6)
Iyl =0

Note that for vectors ~,1/ E R n the formula

(8.7)

holds. Thus,

~ ~[(g'(o)-I)*w]Y R(o,'r)Y =
lyl=s'Y·

= ~«(g'(o)-I)*wR(o,'r)Y = ~(wg'(o)-lR(o,'r)Y.
s. s.
This and formulas (8.5), (8.6) imply

E"'-+<j>(A + ia) -1cI>(w)F"H",(A)u(1/I) = (8.8)

= Ig'(o)-11 ~
N i::J1...
-: E",-+o(A+ia)-IcI>«(g'(o)-I)*w) X
s=o s.

By proposition 3.2.1,

E",-+o(A + ia) -lcI>«g'(o)-I)* w)E..-+", (A - is)(wg'(O) -1 R (o,'r)Y X

X Ig'(-r)lu('I")- ~ ~E",-+o(A+i(a+s-I'YI))-IX
lyl =0 'Y.

X oHwg'(o)-l R (o,'I")Y I.-=oD~ cI>«(g'(o)-I)* W)E..-+",(A) Ig'('I") Iu('I").

Combining the last relation and (8.8), the proof is finished. In the exponential
series expansion for the expression o~eih(""O,T) 1'-=0 for fixed 'Y, only finitely many
nonzero terms occur. •
 Chapter 4

Pseudodifferential operators with discontinuous symbols on

manifolds with conical singularities

In this Chapter we consider various algebras of pseudodifferential operators. The

first paragraph is devoted to the study of pseudodifferential operators on the
spaces Hp(Rn). Symbols of 'canonical' pseudodifferential operators of order a
will be functions (Rn \ 0) X (Rn \ 0) 3 (x,e)~If>(x,~, homogeneous of degree a
both in x and in g. A 'general' pseudodifferential operator
A: Hp (Rn) ~ Hp - Rea (Rn) of order a can be expressed in terms of a meromorphic
pseudodifferential operator U by formula (1.1) (compare with (2.3.6) and (2.3.7».
We derive from the results of Chapter 3 concerning meromorphic
pseudodifferential operators a symbolic calculus for pseudodifferential operators
on IR n (modulo operators acting from Hp(Rn) to Hp+~ea +N(Rn) for arbitrary N).
Similar questions are considered for pseudodifferential operators with 'periodic'
(in the sense of §7, Chapter 3) symbols.
In §2 we introduce and study pseudodifferential operators on a cone. This
allows us to define, in §3, pseudodifferential operators on a manifold with a finite
set of conical points (including, in particular, the points at which the symbols of
the operators have a discontinuity 'of the first kind'). Necessary and sufficient con-
ditions for a pseudodifferential operator of arbitrary order to be Fredholm are
elucidated.
Algebras generated by pseudodifferential operators of order zero will be con-
sidered in §4. The symbols of the operators are allowed to have isolated singulari-
ties. The set of singular points of a symbol depends on the operator, and by clos-
ing the algebras, symbols with an everywhere dense set of discontinuity points
arise. We describe algebras of 'operator symbols', which are isomorphic to quo-
tient algebras of pseudodifferential operators by the ideal of compact operators.
Operator symbols are functions whose values are meromorphic pseudodifferential
operators or periodic meromorphic pseudodifferential operators. We consider, in
particular, an algebra of singular integral operators on a composite contour.
133
134 Chapter 4.

§1. Pseudodifferential operators on JRn

1. Definition of the operators. Let U be the meromorphic pseudodifferential
operator of order a = ao with symbol '2.j=o <Piq"O) (cf. definition 3.3.1). For
u E Co(JR n \ 0) we put

(Au)(x) = (1.1)

_1_ f ri(A+ia+inI2)U",~(A)u(A+inI2,I/J)dA,
&ImA='-

u
where, as before, r = Ix I, If> = x 1 Ix I, is the Mellin transform of the function
u, and the number T is such that the line ImA = T does not contain poles of the
operator-function U.

Pro p 0 sit ion 1.1. For T = /3-s the operator (1.1) realizes a continuous map
A: Hft (JR n ) ~ Hft - Rea (JR n ).

Proof. Multiply both sides of (1.1) by r a an~en apply the Mellin transform
As the result we obtain (Au)(A+ia +inl2,lf» = U"'--7cp(A) X
M r --7A+inI2.
XU(A+inl2,I/J). Definition 3.3.1 and proposition 3.l.3 imply that the estimate

11(A0(A+ia +inI2,·);H S - Rea (A,Sn -1)11 2 .;;;

.;;; c,-llu(A + in 12,);H s (A,sn - I )11 2
holds on the line 1m A = T. After integration of this inequality we have

f 11(A;)(A + in 12,);HS - Rea(A,sn -1 )ledA';;;

ImA=f3-s+Rea

';;;c f lIu(A+inI2,·);H s (A,Sn-I)1I 2dA.

ImA=f3-s

D e fin i t ion 1.2. The operator (1.1) will be called a pseudodifferential opera-
tor of order a on Hp(JRn). The formal series
00
o(x,O) = r -a ~ <Pix, 0) (1.2)
j =0
§l. 'I!DOs on R n 135

is called the (complete) symbol of A, and the term with index j = ° is called the
principal symbol; recall that the function (x, U) I-> cPix, U) is homogeneous of degree
aj both in x and in 0, Reao > Real ~ Reaz ~ ... , Reaj ~ - 00. If a canonical
meromorphic pseudodifferential operator of order a is taken in (1.1) instead of U,
the operator (1.1) is called a canonical pseudodifferential operator on ~p(lRn).

E x amp I e. An expression of the form ~j =0 r -j Liq"D x ), where Liq"Dx ) is a

homogeneous differential operator of order k - j with coefficients depending on q"
is a pseudodifferential operator of order k on any space Hp(lRn) (cf. (3.1.15».
Extend the function (q"U)I->Liq"O) as a homogeneous function of degree k - j
both in q, and o. The symbol of the operator ~j =0 r -j Li$,Dx) equals
r-k~j=oLix,U) = ~j=or-jLiq"U).
Associate to the series (1.2) the formal operator series
00
~ uj-uA· (1.3)
"'" r J'
j=o

where a = ao and the operator Aj is defined by (1.1) with a = aj'

U(X) = UiX) = E(X+ia)-ICPiq"U)E(X) (i.e. the Aj are canonical
pseudodifferential operators of orders aJ>. Proposition 1.2 implies that if there are
no poles of Uj on the line ImX = T = /3-s, the term in (1.3) with index j is a
bounded operator from Hp(lRn) to HP~~_a/lRn) C HP -ao (IR n), {Xj = Reaj.
Assume that there are no poles of the operators U, Uj , j = 0, ... ,N, on the line
ImX = /3-s. Then, using (3.3.1) and by a reasoning as in the proof of proposi-
tion 1.1 we obtain the estimate

(1.4)

here SN denotes the N-th partial sum of the series (1.3).

E.g., if the {Xj (j =0, 1, ... ) are integers, then there are no poles of the opera-
tors U, Uj , j = 0,1, ... , on the line ImX = T, for all T E IR except possibly for
some set of isolated values. Hence (1.4) holds for any N = 0,1, ....
Theorem 3.3 immediately implies

The 0 rem 1.3. For any given series (1.2) there is a pseudodifferential operator
A of order a on Hp(lRn) having it as symbol. If a merom orphic pseudodifferential
136 Chapter 4.

operator U is fixed, then A is a pseudodifferential operator on Hp(Rn) for all

1" = /3 - s, except possibly for a set of isolated values.

2. The algebra of pseudodifferential operators.

The 0 rem 1.4. Let A:Hp(Rn)~Hp-Rea(Rn), B:Hp+Reb(Rn)~Hp(\Rn) be
two pseudodifferential operators, of orders a and b respectively, and let
00 00
(]A(X,D) = r- a ~ «Pix,D), (]B(X,D) = r- b ~ '¥k(X,D)
j =0 k =0

be their symbols. (Here a = ao,

= b o, «Pj and '¥k are homogeneous functions of
b
degrees aj and bk (in both
arguments x and (J).) Then the operator
AB:Hp+Reb(Rn)~Hp-Rea(Rn) is a pseudodifferential operator of order a +b (in
the sense of definition 1.2), and its symbol is defined by

where the series (]A and (]B are to be differentiated and multiplied termwise.

Proof. By definition 1.2,

Au = _1_ f r i (Hia+inI2)U(A)u(A+inI2,·)dA,
y;j;;Iffi>"=P-s

Bu = _1_ f r i (}L+ib+inI2)'13{J.L);'{J.L+inI2,·)dp"
y;j;;IffiJ.l= P-s - Reb

where U and 18 are meromorphic pseudodifferential operators of orders a and b.

After multiplication of the first equation by r a and application of the Mellin
transform M r ->A+inI2 we obtain the relation
(A0(A+ia +inl2,.) = U(A)u(A+inI2,.) on the line ImA = /3-s. Similarly,
(Bv){J.L+ib +inl2,·) = '13{J.L)v{J.L+inl2,.) on the line Imp, = /3-s -Reb. Hence

ABv = _1_ f r i(}L+i(a+b)+inI2) X

y;j;;IffiJ.l=P-s -Reb

X U{J.L + ib )'13{J.L)v{J.L + in 12, -)dp,.

It remains to prove that U{J.L + ib )18{J.L) is a meromorphic pseudodifferential

§1. i!DOs on R n 137

operator of order a + b whose symbol equals

ru +b ~ J,DOO'A(X, O)·aIO'B(X, 0). (1.5)

Irl;;"o y.

By proposition 3.2.2 the symbol of the meromorphic pseudodifferential opera-

tor U(p. + ib) of order a equals

rb ~ J,aIr-bDo(l/X,O).
Irl.Po y.

Theorem 3.3.4 implies that U(,u+ib)~(,u) is a meromorphic pseudodifferential

operator of order a + b, and that its symbol has the form
rb ~ -+,aIr- bD o+6(l/x,O)aIi!k(k,O) =
1rl.181.}.k;;"0 y.fI.

= rb ~ J,D8(llx,O)a~(r-bi!k(x,0))
1KI;;"o /C.

(we have used Leibniz' formula). Obviously, the righthand side coincides with
(1.5) . •

The 0 rem 1.5. Let A:Hp (Rn) ~ H S - Reu(lRn) be a pseudodifJerential operator

of order a and let A*:H-s+Reu(Rn)~H=:8(Rn) be the operator adjoint to A with
respect to the extension of the scalar product of H8(R n) = L 2 (R n). Then A* is a
pseudodifJerential operator of order ii and its symbol 0'A' can be expressed in terms
of the symbol 0'A of A by the formula

O'A'(X,O) = ~ J,aIDoO'A(X,O),
Irl;;;'O y.

Proof. Note that for w E Hp-Reu(Rn), V E H=p+Reu(lR n) = (Hp-Reu(lRn))*

the equality

(w,v) = f <w(A+inl2,), v(}.+inI2,»dA (1.6)

ImA=fJ-s +Reu

holds, where (w, v) and <f,g > are the extensions of the scalar products in
L 2 (R n) and L 2 (Sn -1), respectively. Put w = Au. Then
w(A+ia +inl2,) = U(A)U(A+inI2,') on the line Im"A. = j1-s. Here U is the
138 Chapter 4.

meromorphic pseudodifferential operator of order a whose symbol is

raaA(x,U) = ~j=ofPlx,U). This and (1.6) imply

(Av) =
f <U(A-ia)u(A + in/2- ta, .), v(A +in/2,»dA=
Im>.=p-sRea

f U(A)u(A+in/2,),v(X+i(n/2-a),»dA =
Im>.=p-s

f <u(A+inI2,), U*(X)v(X+i(nI2-a),»dA.
Im>.=p-s

Put h(~+inI2 .. ) = U*(~)v(X+i(nI2-a) .. ) on the line ImA = P-s. Rewrite

this as h(p.+ia+inI2,) = U*(p.+ia)ii(p.+inI2,), where Imp. = s-p-Rea.
Thus,

h(r,·) = _1_ f r i Cp.+Ia+in/2)u*(p. + iQ)ii(p.+ in 12,)dp..

y!2;Imp.=s -p-Rea

For arbitrary functions u,v E C~(Rn\o) we have (Au,v) = (u,h). Hence

h = A*v. It remains to prove that the symbol of the meromorphic
pseudodifferential operator U* (p. + iii) equals

(1.7)

Theorem 3.3.5 implies the asymptotic expansion

U*(p.+iQ),..., ~ ~E(p.+i(Q+aj-I)'I))-la~D8fPlcp,U)E(p.+ia).
j, I"fi ;;;.0 )'.

Apply proposition 3.2.2 to each term. Then

U*(p.+iQ),..., ~ 1~16~r-allxl=1 X
j, 11' I, 181 ;;;.0 y. .

X E(p.+i(Qj-lyl-161))-laIDo+8fPlx,u)llxl=1E(p.) =

= ~ ~E(p.+i(Q-I/CI))-la~D8(r-afPlx,u))IIXI=lE(p.).
I" 1;;;.0 /C.

This means that the symbol of U*(p.+iQ) coincides with (1.7) . •

§l. 'i!DOs on IR n 139

We now give some information necessary in the proof of theorem 1.7. Let
HI ,H 2 be Hilbert spaces. A bounded linear operator B:H 1 ~ H 2 is called
Fredholm if its range, 1mB, is closed and if the spaces kerB = {xEHI:Bx=O}
and coker B = H 211m B are finite-dimensional. It is well-known (e.g. see [40))
that if K:H 1 ~ H 2 is a compact operator and B is a Fredholm operator, then
B + K is also a Fredholm operator. The following assertion is a particular instance
of theorem 3.3 given in [21].

Pro p 0 sit ion 1.6. Let>. 1-+ B(>'):H 1 ~ H 2 be an operatorJunction that is

holomorphic in a domain !2 and whose values are Fredholm operators. If at some
point Ao E !2 the operator B(Ao):H 1~H 2 is an isomorphism, then all the B(>.) are
isomorphisms, except possibly for a countable set whose limit points can lie on the
boundary of!2 only.
Let A be an operator of the form (l.l). A number 'T is called admissible for A
if for all u E Coo (S n - I) the inequality

clllu;Hs(A,sn-I)11 ~ 11U(>')u;Hs-Reu(A,sn-I)11 ~

~ c21Iu;H s(A,Sn -1)11

holds on the line 1m>. = 'T. In other words, for admissible 'T the line 1m>. = 'T
does not contain poles of the operator-function >'1-+ U(>.), the inverse operator
U(>.) -I exists and the estimate

IIU(A)-I;HS -Reu(A,sn -1) ~Hs(A,Sn -1)11 ~ cT

holds. It is clear that only for admissible 'T = /J-s the operator (l.l) realizes an
isomorphism A:Hp(lRn) ~Hp -Reu(lR n).
A pseudodifferential operator A is called elliptic if its principal symbol (10 does
not vanish on Sn -I X Sn -I.

The 0 rem 1.7. Let A be an elliptic pseudodifferential operator of order a of the

form (1.1) and suppose there are no poles of the meromorphic pseudodifferential
operator U in the strip 'TI ~ 1m>. ~ 'T2. Then every number 'T E ['Tj,'T2] (with the pos-
sible exception of a finite set of values) is admissible for A.

Proof. The pseudodifferential operator A and the meromorphic

140 Chapter 4.

pseudodifferential operator U in (1.1) are simultaneously elliptic. Hence the condi-

tions and theorem 3.3.8 imply the existence of a meromorphic pseudodifferential
operator m of order -a such that the difference ~ = mu - I is a meromorphic
pseudodifferential operator of order - 00. This means, in particular, that for arbi-
trary h > 0 the following inequality holds on the set {A: lImA I :0;;;;;; h,
IReA I ;.: c(h)}:
11Z5(A):Hs(A,sn-I)-')H s(A,sn-I)11 :0;;;;;; cp(l + IAD-P,

whatever p will be. Hence, for sufficiently large IAI there exists on this set an
inverse operator (I + Z5(A)) -I , and with it also the operator
U(A)-I = (I+~(A))-lm(A). In view of proposition 3.2.2 and the compactness of
the imbedding Hs-Reu+I(A,sn-l) C Hs-Reu(A,sn-I), the operator
U(A)-U(p.):Hs(A,sn-I)-,)Hs-Reu(A,sn-l) is compact for any IL. Together with
the invertibility of U(A) for certain A this leads to the conclusion that the values of
the function A~ U(A) are Fredholm operators. By applying proposition 1.6 we find
that U(A) is an invertible operator on the whole A-plane, except at isolated points
which can only condense at poles of U. Hence a strip 'TI :0;;;;;; 1m A :0;;;;;; 'T2 contains at
most finitely many points at which the operator U(A) does not have a bounded
inverse. •

Rem ark 1.8. In theorem 3.3 of [21] proposition 1.6 is extended to finite-
meromorphic functions; in this case the points at which the inverse operator
U(A) -I does not exist does not condense at poles. However, canonical mero-
morphic pseudodifferential operators are, in general, not finite-meromorphic (cf.
§l of Chapter 3). Hence the question of whether the singularities of U- I can con-
dense at points at which the operator-function A~ U(A) is not finite-meromorphic
remains open.

Rem ark 1.9. The isolated singular points at which the operator U(A)-I does
not exist are poles of the operator-function A~ U(A)-I and are 'normal eigen-
values' of the function A~ U(A) (here A plays the role of spectral parameter); cf.
[21]. The proof of theorem 1.7 implies that the poles of the functions A~ U(A)±I
which are located in a strip I1m AI < h cannot condense at infinity. Hence a line
ImA = 'T is admissible for the operator A if and only if it does not contain poles
of the operators U ± I.
§l. 'l!DOs on R n 141

Rem ark 1.10. Let 9t be the meromorphic pseudodifferential operator con-

sidered in the proof of theorem 1.7. Clearly,
9t(A) = 9t(A)U(A)U(A) -I = (I + ~(A»U(A) -I, i.e. U -I differs from 9t by a term of
order - 00 (it is, however, not excluded that the singularities of this term may
condense to poles of U).

3. The quotient nonn of a pseudodifferential operator. We introduce in H:e(R n )

the equivalent norm

where 0 = ReA if I ReA I ;;;;. 1, and 0 = 1 if I ReA I < 1 (cf. Remark 3.6.3). Since
for ImA = °
the operator E(A):HO(sn-I)~HO(sn-l) is unitary (corollary
1.5.6), we have I u;H~(Rn) Ie = Ilu;H~(Rn)ll.

Pro p 0 sit ion 1.11. Let A be a canonical pseudodijferential operator of the

form (1.1), where U(A) = E(A + ia) - 1cp(</I, 8)E(A)' a is a real number, T = /3 - a, and
/3 - a =1= k + n 12, fJ =1= - k - n 12, k = 0, 1, . . . . Then there is for any positive £ a
continuous operator T:Hp (Rn) ~ H 1+ 1 (Rn) such that the inequality

(1.8)

holds, where K = Ilcp;c(sn-1 XSn-I)11 is an exact constant.

Proof. By theorem 3.6.5 there is an operator-function A1-+ ~(A) on the line

ImA = /3-a such that

11~(A)v(A + in 12, ·);HI(A,sn -I )11 ...;;

..;;; cllv(A+ in 12, ·);Ha(A,sn -1)11

and

II(U(A) - ~(A»v(A + in 12, ·);HO(sn -I )11 ..;;; (1.9)

..;;; (K +£)IIE(o- ia)v(A + inl2, ·);HO(sn - I )11.

142 Chapter 4.

Introduce an operator T, coinciding with the righthand side of (1.1) after replac-
ing U(A) by ~(A). It is easy to see that the map T:Hp(lRn)~Hhl (IRn) is con-
tinuous. The estimate (1.9) implies (1.8).
It remains to prove that K is exact. Let {Uj}, {Aj } be the sequences of func-
tions and numbers used in proposition 3.6.4, let Uj be intervals on ImA = /3-a
with centers at the points Aj such that the length of ~ tends to zero as j ~ 00.
Let also r j be a nonnegative smooth functions with supports in ~, and equal to
one near Aj . Put vJ (A+inI2,cp) = r/A)u/cp). We have
ITvj;~(lRn)le';;;;;8jlvj;Hp(lRn)le, where 8j~0. It is obvious that
IIU(A)ii/A + in 12, ')11 ,;;;;; II(U(A)- U(A.i»iij(A + inl2, ')11 + IIU(Aj)ii/A + in 12, ')11 (the
norms are taken in HO(Sn - I ». The first term does not exceed
Y)lv/A+inI2,');H a(A,Sn-I)II, where Yj~O, while the second term can, by pro-
pOSItion 3.6.4, be written as coIIE(oj-ia)v/A+inI2,');
HO(sn-I)II+tjllvj(A+inI2,'); HO(sn-I)II, with Co = 1(1($0,80 )1, (j~O. This
means that 1 (A-T)vj; ~(lRn)le = ColVj; Hp(lRn)le+fjlvj; Hp(lRn)le, fj~O.
This implies that the constant K is exact (since the point (CPo, 8o) E Sn - I X Sn-I
can be chosen arbitrarily) . •

4. Periodic pseudodifferential operators on IRn. In 1, starting with meromorphic

pseudodifferential operators, we arrived at pseudodifferential operators on IR n on
the spaces Hp(lRn). Similarly, the periodic meromorphic pseudodifferential opera-
tors obtained in §7, Chapter 3, lead to periodic pseudodifferential operators on
Rn.
A canonical periodic pseudodifferential operator has the form (3.7.3). The
representation (3.7.5) is valid for such an operator. Let A be the general periodic
meromorphic pseudodifferential operator of order a defined in Chapter 3, §3.3.
For U E C(f (IRn \ 0) we put

Au = (Gr->(r,J.I+ia +inI2) -I A(r ;p.)Gr->(r,J.I+ia +inI2)u, (1.10)

where Imp. = T and T is such that the interval L,. = {P.EC: 0 ~ Rep. ~ 27T,
Imp. = T} does not contain poles of A. The norm Ilu;H:HlRn)11 is equivalent to
the norm
§l. i'DOs on R n 143

(This can be verified using (3.7.3).) Since A(·,,u):Hs(II)~Hs-Rea(II) is bounded,

uniformly with respect to ,u E I,., for 7" = /3-s the operator (1.10) realizes a con-
tinuous map A:Hp(Rn)~Hp-Rea(Rn).
For 7" = /3-s the operator (1.10) is called a (general) periodic
pseudodifferential operator of order a in Hp(Rn). If ~<P/r,q"U) is the symbol of
A, then r -a~<P/r,x, U) is called the symbol of A; recall that the function
(r,x, U) t-+ <P/r,x, U) is regarded as being homogeneous of degree aj both in x and
in (), and that <P/re,·) = <P/r, .), r = 'x ,. The operator (1.10) has an asymptotic
series expansion ~raJ -a A j , where the Aj are canonical periodic pseudodifferential
operators of orders aj (compare with (1.4». Theorem 3.7.3 implies that there is a
periodic pseudodifferential operator with arbitrary, pre-given, symbol.

The 0 rem 1.12. Let A:Hp(Rn)~Kp-Reao(Rn), B:Kp+Rebo(lRn)~Hp(Rn)

be periodic pseudodifJerential operators, and let 0A(r,q"U) = ~j=oraj-a°<Plr,q,,(),
00 b -b .
0B(r,q,,() = ~k=Or k °i'k(r,q"U) be their symbols. Then the operator
AB:Hfi + Reb o (IJ~n) ~ Kp - Reao (Rn) is a periodic pseudodifJerential operator, and its
symbol is
~ raj+bk-Iyl-ao-bo X
(1.11)
j,k,y

X { _1
y! r Iyl +bo-bkDyn.
IJ'!!j (r,q" U)a xY( r -b O•TTk
> (
r,x, U»} .

The expression between brackets, Sjky, is periodic in r: Sjky(re,') = Sjky(r,). The

principal symbol of AB is, according to (1.11), equal to <Po(r, q"U)i'o(r, q,,().

Proof. The asymptotic series for A and B can be written as

A -- (G r -4(r,p.+ iao+inI2)-1 ~Air, /L)Gr->(r,p.+inI2),

j

B -- (G r ->(r,p.+inI2) - I ~Bk(r, /L- ibo)Gr->(r,p.-ib o+inI2),

k

where Aj , Bk are periodic meromorphic pseudodifferential operators with symbols

and i'(k), Im/L = /3-s. Hence
<p(j)
144 Chapter 4.

moreover, 1m v = f3 - s - Re b o. Further,

+00
Aj(r,v +ibo)U = ~ r27rikUj(r,v +ib o +2hr)Uk ·
k=-oo

By proposition 3.2.2, the symbol of the meromorphic pseudodifferential operator

Ulr, v + ib o) is equal to rbo}; h'l ~o(l/y!)a~ r -b oDb <fJlr, x, 0). By theorem 3.7.4,
Alr, v + ibo)Bk(r, v) is a periodic meromorphic pseudodifferential operator, with
symbol

So, the operator AB has the asymptotic series expansion

(Gr->(r,/.L+ i (ao +b o)+ nI2) -I ~ Ajkir , p.)Gr->(r,l'+inI2),

j,k,r

where Ajkr is a periodic meromorphic pseudodifferential operator with symbol

-1r Irl +bo-bkDyn.. (

(I'f.'jr,tp, O)ar(
xr -bO,T, (
Tkr,x, 0) .
y!
This series can be rewritten as
~ raj+bk-Irl-ao-bo X
j,k,r

X (Gr->(r,l'+i(a) +bk- Ir I)+inl2) - I AjkrG r->(r,l'+inI2).

This immediately implies the assertion of the theorem. •

In order to describe the adjoint operator we need

Pro p 0 sit ion 1.13. Suppose a junction (p.,r,tp)1-+ V(p.,r,tp) is 'periodic' in

r:V(p.,re,tp) = V(p.,r,tp), p. E C,tp E sn-I. Suppose also that u E Coo(lRn\o).
Then

j(Gr..... (r,l'+inI2)-1 V'udx =

R"

e dr
= j dp. j - j V(Gr .....(r,/i+ in 12) u)dtp.
IT 1 r S"-I
§1. 'l!DOs on R n 145

Proof. By the inversion formula (3.7.2) we have

Using the periodicity of V, the last expression can be rewritten as

+00 1 e. . ~
~- - j dq,ju(reJ,q,)(reJ)n- X
j= -00 ...;2;S"-1 1 r

X j (re j Yf.p.+inf2) V{p.,r, q,)dp. =

I.,

e dr 1 + 00
= jdp.j-
---------
j V{p.,r,q,) . r,:;- ~ (re j )-i(ji+inf2)u(rej ,q,)dq,. •
I., 1 r s" -1 V 2." j = - 00

The verification of the two following theorems is left to the reader.

The 0 rem 1.14. Let A:H:8 (Rn) ~ Hft - Reao (Rn) be a periodic
pseudodifferential operator of order ao, and let }:ra,-ao¢.llr,q"O) be its symbol. Then
the operator A*:H~e;O-S(Rn)~H=:8(1Rn) adjoint to A with respect to the scalar
product in L 2(R n) is also a periodic pseudodifferential operator. The order of A* is
a0, and its symbol is
~ ra,-Iyl-ao {J,r Iyl +ao-a'aID8(r -aO¢.llr,x,O»}. (1.12)
j, Iyl =0 y.

The expression between brackets, Sjir, q,,0), satisfies the condition

Sjire, .) = Sjir,). The principal symbol of A * is, by (1.12), equal to ¢.lo(r, <p,O).

The 0 rem 1.15. Let A be an elliptic periodic pseudodifferential operator of order

a 0 (i. e. the principal symbol ¢.lo (r, <p,0) of this operator does not vanish on
[1, e 1X S n - 1 X S n - 1). Then there is a periodic pseudodifferential operator Q of
order -ao such that the symbols if AQ and QA are the identity function.
146 Chapter 4.

§2. Pseudodifferential operators on a conic manifold

1. The conic manifold X. A set KeRn \ 0 is called conic if x E K implies
Ix E K for all I > O.
Let X be a Hausdorff space and 19 a point of it. Suppose that the following con-
ditions hold: 1) there is a finite cover {Uj } of the subspace X \ 19 by open sets for
each of which there is defined a homeomorphism ICj onto a conic subset of R n \ 0;
and 2) if l'J n U i =1= 0, then the map ICjOIC i- 1: ICi( U i n U j ) ~ IC/ U i n Uj ) is homo-
geneous of degree one and infinitely differentiable. In this situation we can define,
as usual, a differentiable structure, which will be called a conic structure of class
v
Coo on X. The space X endowed with this structure will be called an n-
v
dimensional conic manifold (a cone) of class Coo with vertex at 19.
Multiplication of a point x of X by a positive number I is defined by
Ix = ICT 1(tIC/X», if x belongs to the coordinate neighborhood Uj. The fact that
this is a well-defined definition follows from the homogeneity of the transforma-
tion ICjOICi-1. By the same token, IC/tX) = tIC/X), i.e. the homeomorphism ICj is a
homogeneous map of the first degree.
We denote by HS(A,R n) the space of homogeneous functions on R n \ 0 of
degree iA.-nI2, with norm lIu;Hs(A.,Rn)11 = Ilu!s·-I;H s(A.,sn-1)11, where
u! S·-I is the trace of u on Sn -1. Let also {O be a partition of unity of X\ 19
subordinate to the conic atlas {U, IC} and consisting of infinitely differentiable
homogeneous functions of degree zero (for simplicity, the indices are sometimes
omitted). Introduce the space HS(A,X) of homogeneous functions on X\ 19 of
degree i A. - n 12, endowed with the norm

Ilu ;Hs(A,X)11 = 2:11~KU ;Hs(A.,Rn)ll,

U

where ~KU = (tu)oIC- 1 on IC(U) and tKU = 0 outside IC(U). A distinct equivalent
atlas and a distinct partition of unity lead to an equivalent norm.

2. Meromorphic pseudodifferential operators on X. We will assume that canonical

meromorphic pseudodifferential operators, written using the exterior form (cf.
Chapter 3, § 1.4), act on homogeneous functions on R n \ 0 of degree iA. - n I 2.
Taking this understanding into account, definition 3.3.1. for general meromorphic
pseudodifferential operators remains valid; we only need replace the HS(A.,sn-l)
§2. itDOs on a conic manifold 147

by the spaces Hs(X,lR n) of homogeneous functions. Such operators will be called

meromorphic pseudodijJerential operators on IRn.
We further consider operator-functions At-> U(A) whose values are operators
given on HS(A,X). We may free ourselves from the dependence on A of the
domains of definition of these operators. For this we fix a sub manifold S of X \ (9
intersecting every ray {yEX:y=txO,XOEX\(9,t>O} at a single point, and
replace the homogeneous functions by their traces on S. Thus, we may speak of
an operator-function At-> U(A) being meromorphic, and use general theorems con-
cerning meromorphic operator-functions when applicable.

D e fin i t ion 2.1. A meromorphic operator-function At-> U(A) is called a

meromorphic pseudodijJerential operator of order a on X if the following conditions
hold:
a) every strip lImA I < h contains only finitely many poles of U;
b)for two arbitrary infinitely differentiable functions rand 1'/, homogeneous of
degree zero and with intersecting supports (on X\ (9), the following inequality
holds:

where 'J is an arbitrary admissible set, A E 'J. P and N are arbitrary positive
integers, and c('J. N,p) = const < 00;
c) if the supports of rand 1'/ belong to one coordinate neighborhood U, then

where U u is a meromorphic pseudodifferential operator of order a on IR n,

rKv = r(VOIC) on U and rKv = 0 outside U, while 1'/KU is to be understood as in
1.

The fact that this definition is well-defined follows from the rule of change of
variables in a meromorphic pseudodifferential operator on IR n (cf. §8, Chapter 3).
This rule allows us to bring every meromorphic pseudodifferential operator in
correspondence with a (principal) symbol q, - a function on the bundle To (X) of
nonzero cotangent vectors over X\ (9; the principal symbol q,u of the mero-
morphic pseudodifferential operator U u on IR n figuring in formula (2.1) is the
representation of q, in the coordinate neighborhood U, i.e. q, u = q,O(IC*) - 1, where
148 Chapter 4.

K*: TO (X) I u-,>(Bi n \ 0) X (Bin \ 0), where the part of To (X) above Uis regarded as
the direct product U X (Bin \ 0), IC*(X,'T) == (IC(X),IC'(X)'T), (x,'T) U X (Bin \ 0).
E
Recall that the function (y,z) ...... 4:l u (y,z) is homogeneous of degree a both iny and
in z. Therefore we have that if'Tx is a cotangent vector in the fiber To
(x)x above
a point x, then for t > 0,

(2.2)

Pro p 0 sit ion 2.2. Whatever a function 4:1, smooth on To

(X) and satisfYing
(2.2), there is a meromorphic pseudodifferential operator U of order a on X whose
(principal) symbol coincides with 4:1.

Proof. Let {~, "}} be an atlas on X, and {f}} a partition of unity subordinate
to it and consisting of smooth homogeneous functions of degree zero. Let {TJ}} be
a set of smooth homogeneous functions of degree zero such that SUPPTJ} C U},
TJ}f} = fl·
Put

U(A) == ~(f})"JULj(A)(TJ}),,), (2.3)

}

where UU(A)
J
= E(A+ia)-14:l u} (y,z)E(A) are canonical meromorphic
pseudodifferentia1 operators on Bin and the notations f", '11" have the same meaning
as in (2.1).
We will show that U is the required meromorphic pseudodifferential operator
on X. It is obvious that condition a) of definition 2.1 is fulfilled. Equation (2.3)
and proposition 3.1.6 imply that U also satisfies b). Let, finally, the supports of '11
and f lie in one coordinate neighborhood Uk. In view of the fact that

fU(A)TJ = ~(ff}tJULj(A)(rJ71}),,), (2.4)

}

the sum will only contain those term for which supp f} n Uk =1= 0. Write the
righthand side of (2.4) in the local coordinates of Uk. By the rule for change of
variables, (2.4) takes the form

fU(A)1J = ~(ff}tk'B(A)(1JTJ})"k'
)

where 'B is a meromorphic pseudodifferential operator on Bin with principal

§2. 'l!DOs on a conic manifold 149

symbol (> Uk. By applying proposition 3.2.1. we obtain that the righthand side
equals

2.(tt/rJj)/(k~(A)lI/(k = 2.(tt)/(k~(A>1I/(k = rk~(A)lI/(k'

j j

where ~ is a meromorphic pseudodifferential operator on IR n with the same prin-

cipal symbol (> Uk. •
Using theorem 3.3.4 the following assertion is easily verified.

Pro p 0 sit ion 2.3. Let U and ~ be meromorphic pseudodifJerential operators

on X of orders a and b. Then their composite U~ is a meromorphic pseudodifJerential
operator of order a + b, and its symbol is the product of the symbols of the factors.

3. Pseudodifferential operators on X. For a function u E Co (X\ 0) we define the

Mellin transform by

U(A,X) = _1_ jt-i"A-lu(tx)dt, (2.5)

...n;;; 0
where A is a complex parameter. It is clear that x ~ u(A-,x) is a homogeneous func-
tion of degree iA. The inversion formula

u(x) = _1:- !
V2'1Tlm"A==T
u(A-,x)dA (2.6)

and Parseval's equality

!
00

IU(A-,x)1 2dA = !t2Plu(tx)12dt, 'T = ,8+112,

Im"A==T 0

hold.
For arbitrary real sand ,8 we introduce the space Hp(X) as the completion of
Co (X\ 0) with respect to the norm
(2.7)

Let S be a 'directing' submanifold of the

150 Chapter 4.

manifold X\(9 intersecting every ray {yEX:y=txO,XOEX\0,t>0} at a single

point. For every point x E X\ (9 we denote by Ps(x) the positive number for
whichps(x)-lx E S.Clearly,ps(tx) = tps(x)fort>O.

D e fin i t ion 2.4. A pseudodifferential operator of order a on a space HjJ (X) is

an operator of the form

(Au)(x) = 2~ps(X)-U f Uy4 AA)u(A+inI2,y)d,\, (2.8)

ImA='T

where U is a meromorphic pseudodifferential operator of order a on X (cf.

definition 2.1) without poles on the line 1m>.. = 7" = /3-s.

Pro p 0 sit ion 2.5.

l)An operator (2.8) realizes a continuous map A:Hp(X) --i>Hp -Reu(X).
2) For two arbitrary infinitely differentiable functions rand 1/ on X \ (9, homogeneous
of degree zero and with intersecting supports, the following inequality holds:

IlrA1/;Hp(X) --i>Hp+~eu +N(X)II :;;;; CN,

where N is an arbitrary number, CN = const < 00.

3) If the supports of rand 1/ belong to a coordinate neighborhood U on X, then

rA1/u = rIC Au1/"u, where Au is a pseudodifferential operator of order a on IR n and
the notations r", 1/" have the same meaning as in (2.1).
Proof. Put v = Au. Then

~(A+inI2,x) = Uy4x (A)u(A+inI2,y). (2.9)

On the other hand,

__ 00

(p~v)(>"+inI2,x) = f t- i (A+inl2)-lps(tx)u v (tx)dt

o

f t -i(A+iu + in/2)-J
00
V (tx)dt Ps(x )u.

o
Hence for XES we have
§2. i'DOs on a conic manifold 151

~(A+in/2,x) = v(X+ia +in!2,x). (2.10)

Further,

Ilv ;Hp -Rea(X)11

= lf
ffiA=fJ-S
Ilv(A+ia +in/2,·);HS-Rea(X+ia,X)112dX l1l2

This, equation (2.10) and the definition of the norm in Hi(P,X) imply that

Ilv;Hp-Rea(X)11 ",;;;

",;;;) f 11~(A+in/2,.);HS -Rea(A,x)112dXlll2.

~ffiA=fJ-S
By (2.9) this means that

Ilv ;Hp -Rea(X)11 ",;;;

.; c LAlft~,"W;;-}(A + in / 2, -);H' ~ """0. Xlll2dAll!2 '"

",;;; cllu ;Hp(X)II.

Assertions 2) and 3) readily follow from definitions 2.4 and 2.1. •

A meromorphic pseudodifferential operator U on X is called elliptic if its sym-
bol To(X) :3 'l"x ..... 4»('I"x) does not take the value zero on To (X). An operator A
of the form (2.8) will be called elliptic only if the corresponding meromorphic
pseudodifferential operator U is elliptic.
As in the case of a Euclidean space, a number 'I" = f3 - s will be called admissi-
ble for an operator A:Hp (X) ~ Hp - Rea (X) if this operator realizes an isomor-
phism.
The following statement is a variant for conic manifolds of theorem 1.7.

The 0 rem 2.6. Let A be an elliptic pseudodifferential operator of order a on X

of the form (2.8), and suppose there are no poles of the meromorphic
pseudodifferential operator U in a strip '1"1"';;; ImX",;;; '1"2' Then every number
'I" E ['1"1,'1"2] (except for an at most finite set) is admissible for A.
152 Chapter 4.

The proof is completely analogous to that of theorem 1.7. If cP is the symbol

of U, then for lR we must take a meromorphic pseudodifferential operator of order
- a whose symbol is cP - J; the existence of this operator is guaranteed by proposi-
tion 2.2 . •
Remark 1.9 remains true for operators on the cone X.

4. The one-dimensional case (singular integral operators on a system of rays). Let

Rj = {ZEC:z=peilJ,} be a ray in the complex plane emanating from the coordi-
nate origin, j = 1, ... ,N, 0 .;;;;(h < ... < ON < 2'7T, 0.;;;; p < 00, and let
X = U f= JRj . Consider the singular integral

(%4)(x) = ~f~4Y,
'7TI xy-x
x EX,

where u E Co (X \ 0).

Pro p 0 sit ion 2.7. The following representation holds for the operator (2.11):

f
+00
(%4)(r, OJ) = _1_ r i (h+in/2) X (2.12)
& -00

N
X ~ ~k0)u0+il2,Ok)dA, j = 1, ... ,N;
k=J

here u(r,Ok) = u(re ilJk ), U(-,0k) is the Mellin transform of the function rt-+u(r,Ok),
and ~k is defined by

~k0) = (2.13)
exp{A(Ok - OJ - '7T) + i (Ok - OJ) 12}
if k > j,
cosh'7TA
_ exp{A(Ok- Oj+'7T)+i(8k -Oj )/2}
if k <j,
cosh'7TA
-tanh'7TA if k = j,

Proof. Apply the Melling transform MHh+il2 to equation (2.11):

-(%4)0+ il2,O) = (2.14)
§2. i'DOs on a conic manifold 153

First we consider the terms for which j =1= k. Changing the order of integration
we obtain

fo r - iA -112 dr 0f
00 00 ( () i(Jk d
u p, k e p-
(2.15)
'(J '(J-
pe' k - re' j

-
-e
i«(Jk-(J) f upe
00
( i(Jk) p -iA-1/2dp f 00
t
-iA- 1I2 d
i«(Jk-(J)
t
,
o 0 e } -I

where the new variable I equals the quotient rip. If a is a nonnegative number,

jt -iA-l 12 dl = 2'ITiexp{ -(iA + 112)lna} . (2.16)

o a-I exp{-2'ITi(iA+1/2)}-1

11ris and (2.15) imply that for k > j the k-th term at the righthand side of (2.14)
equals exp{A«()k-()j-'lT)+i«()k-()j)/2}/cosh'ITA, while for k > j it equals
- exp{f..(()k -()j - 'IT) + i «()k - ()) I 2} I cosh 'ITA.
Nc ',V consider the j-th term in (2.14). The inner integral is to be understood as
a principal value integral. By the well-known formula of Sokhotskii (cf., e.g.,
[41]),

J.... foo u(p,()j)dp 1 foo u(p,()j)dp _

__
-. '0 u(r, OJ).
'1Tl 0 p-r '1Tl 0 p-r-l

11ris, taking into account (2.16), implies that the j-th term in (2.14) equals
- u(A,O)tanh 'ITA. It remains to use the inversion formula (1.2.6) for the Mellin
transform. •

Rem ark 2.8. For the sake of being specific we have assumed that all rays are
emanating. If the q-th ray is incoming*), then '5jq must be replaced by -'5jq. In
particular, the entries of the matrix II'5jk IIJ,k = 1 corresponding to the operator

~ j u{y)dy
'1Tl_ oo y-x
*)i.e. if - 00 < p ~ 0 for the q-th ray. (Translator's note.)
154 Chapter 4.

are defined by '511 (A) = -'522 (A) = -tanh'ITA., '512 (A) = -'52 ](A) = ilcosh'ITA
(which is in accordance with Chapter 3, §1.2).
Let a and b be complex-valued functions on X \ 19 which are constant on each
ray Rj , and let aj, bj be the values of a,b on Rj . A singular integral operator on X
is an operator A = aI +b%, where % is the integral (2.11). Applying proposition
2.7 we rewrite A as

= _1_ f
+00
(Au)(r,·) r i (Hi/2)U(A)u(A+il2,)dA, (2.17)
Y"2; -00
where U(A) =
IIUjk(A)llfk =], Ujk(A) = a/)jk + b/5jk (A), U(A + il2,) is the vector
(u(A+iI2,O]), ... ,u(A+iI2,ON)), (Au)(r,·) is a similar vector, and 6jk are the
Kronecker symbols.
Denote by L 2,p(X) the completion of C(f (X\ (9) with respect to the norm

lIu;L 2,p(X)1I = [jo p2P.i 1u(p,Oj) 12dP] 112.

..I =]

If 2/1 E (-1,1), then the operator A = aI +b%,

defined on C(f (X\ (9), is
bounded on L 2,p(X). For the closure A, defined on the whole space L 2,p(X), we
have the formula

(Au)(r,·) = _1_ f r i (Hil2)U(A)u(A+il2,)dA (2.18)

Y"2;IID>'=P
In order to convince ourselves of this, we must replace in (2.17) the line of
integration ImA = 0 by the line ImA = /1 and take into account that the func-
tions Ujk are bounded on the line ImA = /1. It remains to use Parseval's equality
(1.2.7) for the Mellin transform. Note that the operator A is continuous on
L 2,p(X) for arbitrary fJ satisfying 2/1 =1= 2k + 1, k = 0, -+-1, .... However, for
12/1 1> 1 it is necessary to first define the operator on a subset of C(f (X \ (9) that
is dense in L 2,p(X) (compare with theorem 2.3.5). Formula (2.18) remains valid
for A.

5. Pseudodifferential operators with variable symbol on a conic manifold. Let S

and Ps be the directing manifold and function introduced above definition 2.4. To
a point x E X \ 19 we assign 'polar coordinates' (p,O), where 0 = Ps(x) -] x and
§2. i!DOs on a conic manifold 155

p = Ps(x) > O. Replacing homogeneous functions on X\ (9 by their traces on S,

we will further assume that meromorphic pseudodifferential operators are defined
on functions given on S. With this understanding the Mellin transform is defined
by

u(A,O) = _~
v2'1T
7
0
p-iA-Iu(P,O)dp.

The notation Ilu(A, ·);Hs(,u,X)11 has a meaning also for A*- J.L.
Let U(p,A) be a meromorphic pseudodifferential operator of order a on X,
depending on a positive parameter p. We will assume that the representation
U(p,A) = U(A) + U I (p,A) holds, where Uo is a meromorphic pseudodifferential
operator not depending on p, while for U I we have the estimate

uniformly in p > 0, for all nonnegative integers k and q and for real s, on a line
ImA = 'T which does not contain poles of Uo. This implies that on ImA = 'T,

(2.20)

for all s and all positive N; here

Assume that for two arbitrary infinitely differentiable functions f and 1/, homo-
geneous of degree zero and with disjoint supports, the inequality

Ilml (v, A)TJ;HS(lI.,x) -"; H S+P(A,X)II :;;:;; (2.21)

:;;:;;c(l+ Iv I + IAi)-N

holds, where Nand p are arbitrary positive numbers, and 1m;\. = 'T.

In the proof of proposition 3.2.2. we have derived, in essence, the inequality

(2.22)

where 1m;\. = ImlL = 'T. Assume that for U I the estimate

156 Chapter 4.

(2.23)

holds, where s, v,N are arbitrary real numbers, and 1mX = 1mI' = T.

D e fin i t ion 2.9. If condition (2.19) holds (hence also conditions (2.20),
(2.21) and (2.23) hold), the operator U(p,X) = UO(X)+UI(p,X) will be called a
meromorphic pseudodifferential operator of order a with variable symbol on X.

E x amp I e 2.10. Suppose a function (r,cp,O) ..... W(r,cp,O) (r>O,cp,OES n - l ) has a

representation W(r,cp,O) = Wo(CP,O)+WI(r,cp,O), where Wo E c~(sn -I Xsn -I),
while the function WI is infinitely differentiable and satisfies the conditions

for all nonnegative integers p, q and k. Then the operator

U(r,X) = Es.....q,(A + ia) -IW(r, cp, O)~ .....s(A) is a meromorphic pseudodifferential
operator of order a with variable symbol on Rn. Starting with such operators we
can construct meromorphic pseudodifferential operators with variable symbol on
X (compare with the proof of proposition 2.2).

D e fin i t ion 2.11. A pseudodifferential operator of order a with variable .rym-

bol on a space Hp (X) is an operator given on C~ (X \ 0) by

(Au)(x) = (2.24)

where Uq, .....s(p,X) is a meromorphic pseudodifferential operator of order a with

variable symbol on X, the line ImX = T = /3-s does not contain poles of the
operator-function U, and (p,O) are the polar coordinates of a point x.

Pro p 0 sit ion 2.12. The operator (2.24) for T = /3 - s realizes a continuous
map A:Hp (X) ~ Hp - Rea (X).

Proof. By definition we have U(P,X) = Uo(X)+UI(p,X), and, correspondingly,

§2. i'DOs on a conic manifold 157

A = Ao + AI. The operator Ao is continuous (cf. proposition 2.5), hence it

suffices to establish the continuity of the operator

(2.25)

Multiplying (2.25) by pa and applying the Mellin transform we obtain

v{jL+ia +in!2,) = f UI (jL-A,A)u(A+in/2,)dA,

Im>.=p-s

where l' = paA1u, ImlL = fJ-s. The inequality

(2.26)

implies the estimate

Ilv{jL+ia +in/2,);Hs - Rea {jL+ia,X)II.;;;;;

.;;;;; c f (1 + IIL-ADls -Rea I X

Im>.=p-s

Taking into account formula (3.1) we have

Ilv(jL+ia +in/2);Hs - Rea (jL+ia,X)II.;;;;; (2.27)

.;;;;; c f (1 + IIL-AD- N llu(A.+in/2,·);H s (A.,X)lldA,

Im>.=p-s

where N is a sufficiently large number. Put k(v) = (1 + Iv I)-N,

U(A.) = Ilu(A.+in/2,); HS(A,X)II, and denote the righthand integral in (2.27) by
J (jL). By Minkowski's inequality,
158 Chapter 4.

This, (2.27) and (2.7) imply /lAu;H:8 -Rea(X)11 ~ cllu;H:8(X)II. _

6. Composition of pseudodifferential operators with variable symbols. Let

A:Hp(X) --'? Hp -Rea (X) and B:HVReb(X) --'?Hp(X) be pseudodifferential opera-
tors of orders a and b, respectively, with variable symbols on X, where A is given
by (2.24) for'T = /3-s, and B has the form

Bv = _1_ f pi(Hib+inI2)m(p,}..)v(A+inI2,)dA,
-J2;ImA=o

(J = /3 - s - Re b.

Put

Cv = _1_ f /(A+i(a +b)+inI2) X

-J2;ImA=o

X U(p,}.. + ib )m(p,}..)v(}.. + in 12,)dA.

The 0 rem 2.l3. The operator AB-C:HVReb(X)--,?Hp-Rea(X) is compact.

The proof of the theorem is preceded by several propositions. Represent each
operator U(p,}..), m(p,}..), A, and B as a sum, U(p,A) = UO(A)+UI(p,}..), etc. Intro-
duce the operators

COlv = _1_ f pi(A+i(a+b)+inI2) X

-J2;ImA=o

CII v = _1_ f pi(A+i(a +b)+inl2) X

yl2;ImA=o

X U I (p,A + ib)'.B 1(p,A)V(A + in / 2, ·)dA.

Pro p 0 sit ion 2.14. Each of the operators To I, T 11 realizes a continuous map
from H:8 +Reb(X) to H:8+.fea +1 (X).

Proof. We have

TOI V = p
-a-b
f pi(p+inI2) X
yf2; ImjL=o
§2. i'DOs on a conic manifold 159

X Uo{J.t+ib) lm!=o~l{J.t-A,l\)V(l\+in12,.)dl\ +
- \B 1(p,p.)v (J.t+ in 1 2,.)] dp..

Multiply this equation by pU + b, and then apply the Mellin transform. Then
-
(TOlv){J.t+i(a +b)+inI2,·) =

= f (Uo{J.t + ib) - Uo(A + ib »~l {J.t - ,\,l\)v(l\ + in 12, ·)dA.

ImA=o

Inequality (2.22) implies

11~{J.t+i(a +b)+in12,·);HS-Reu+I{J.t,X)II..;; (2.28)

";;C f 1p.-l\I(II~I{J.t-A,l\)v(A+inI2,·);HS(l\,X)II+
ImA=o

+ 11~1 {J.t -l\,l\)v(l\ + in 1 2, ·);Hs{J.t,X)II)dA.

Since, by (2.27),

Ilw;Hs{J.t,X)II";;c(l+ 1p.-l\I)lslllw;HS('\'X)II, (2.29)

and

11~1 {J.t-A,l\)v(A + inl2, ·);Hs(A,X)11 ..;; (2.30)

..;; c{l + Ip.- l\ I)-Nllv(A + inl 2,.);H S +Reb(A,X)II,

(2.28) implies that Tal is a continuous operator (compare with the proof of propo-
sition 2.12).
We now consider the operator

f
-u -b
Tllv = p pi{jt+inI2)U l (p,p.+ib) X
yf2; ImA=o

X lmLo~l {J.t-l\,l\)v(l\ + in 12, ·)dl\ - \B l (p,p.)v (J.t+ in 12,.)] dp..

After Mellin transformation and change of integration variable we find
-
(Tllv)(v +i(a +b)+inI2,.) =
160 Chapter 4.

f dp,
- -
f (U](v-p"p,+ib)-U](v -p"X+ib)) X
Iml'=o ImA=o

X \.B(p.-A,A)V(X + in 12, ·)dA

Inequalities (2.23), (2.20) and (2.30) imply the estimate

11~(v +i(a +b)+inI2,·);H S - Rea +](v,X)11 ~

~c f (1+lv-p,I)-Ndp,X
Iml'=o

X f (1+ 1p,-XU-Nllv(A+in12,·);Hs+Reb(A,X)lldA
ImA=o

Pro po sit ion 2.15. Let T be any of the operators AoB], CO], A]B2, or C ll .
Then for all functions u- from the set
{u: u(X + i(a +b) + in 12,.) = (Tv)(X + i (a +b) + inI2,.), Ilv ;Hp + Rea (X)II ~ I),
defined on the line 1m X = /3 - s the following estimate holds:

r f Ilu(X+i(a +b)+inI2+h, .)+ (2.31)

lImA=p-s

where X is a continuous functiOn, x(O) = 0, independent, moreover, of the choice of

an u from the set indicated

Proof. Suppose, e.g., T = AoB] and u = Tv. For 0 = /3-s - Reb,

u(p.+i(a +b)+inI2+h, ·)-u(p.+i(a +b)+inI2,·) =

f [Uo(p.+ib +h)-Uo(p.+ib)] X
ImA=o

X ~](p.+h -A,X)v(X+inI2,)dX+

+ f Uo(p.+ib)(~](p.+h -A,A)-~](p.-A,A))v(A+inI2,)dA.
ImA=o

The first of these integrals can be estimated using inequalities (2.22) and (2.30),
§2. itDOs on a conic manifold 161

and the second - by using the inequality

IllBI (p,+h -A,A)-lBI(p,-A,A);H S + Reb(A,X) ~ HS(A,X)II ~

~ oN(h)(l + IIL-A I)-N,

where ON is a continuous function, ON(O) = O. As the result we find that

Ilu(p,+i(a +b)+inI2+h, .)+

-u(p,+i(a +b)+in!2,); Hs-Rea(p"X)11 ~

~coN(h) f (1+ IIL-AI)-Nllv(A+in!2,);Hs+Reb(A,X)lldA.

ImA=a

Using Minkowski's inequality (as in the proof of proposition 2.12) we are led
to an estimate of the type (2.31). The other operators COl> A I B I> and C 11 are
considered analogously . •

Proof 0 f the 0 rem 2.l3. In view of the fact that AB-C = TOl +T II , it
suffices to verify that each operator TObTII is compact from Hp+Reb(X) to
HjJ-Rea(X). Let v run over the unit sphere in HjJ+Reb(X) and let u = Tv, where
T is any of the operators TOb Til. Let also {U, /(} be an atlas on X \ (9, and let
{n be a partition of unity subordinate to it. We convince ourselves of the fact
that for any local chart the set {rku:u=Tv} is compact in HjJ-Rea(R n ). For sim-
plicity reasons we replace the notation rkU by u. Proposition 2.12 and the
definition of the norms in the spaces H~(Rn) implies that to an arbitrary positive
number ( correspond numbers M and N such that

km N+iT
~ f (1+m 2 + IAI 2)s-Rea X
k=1 -N+iT

for'T = /J-s + Rea (uniformly in v). Applying proposition 2.15 and a well-known
compactness criterion in the space L2 of square-integrable functions on a
bounded interval (cf., e.g., [30]) we find that for any ( > 0 there is a finite (-net for
the set {rkU: u = Tv} . •
162 Chapter 4.

7. The adjoint operator. We denote by do the volume element on the directing

sub manifold SeX \ e and introduce a volume element on the conic manifold X
by dv = pn-1dpdo, where p = Ps. Let A:H/3(X)~H/3-Rea(X) be a
pseudodifferential operator of order a with variable symbol on X of the form
(2.24), with, moreover, 'T = p-s. We denote by A*:H=/3+Rea(X)~H=/3(X) the
openlLVl <1ujomt to A with respect to the extension of the scalar product in
L 2 (X,dv), ::.~d by U(p).y the operator adjoint to U(p,A) with respect to the exten-
sion of the scalar product in L2(S,do).

The 0 rem 2.16. The following equation holds:

A*v = _1_ f pi(lo.+ia+in/2) X (2.32)

-J2;Imlo.=T·

x U*(p,A+ia)v(A+inI2,·)dA+Tv,
where'T* = s-Rea-p, U*(p,A) = U(p,Xy, andT:H=VRea(X)~H=/3(X) is a
compact operator.

Proof. We have

(Au, v) = (2.33)

f <(A0(p+inl2,),v(ji+inI2,»dp, =
ImJl=fJ-s +Rea

f <(A;;)(P + ia + in I 2,), v(ji + i (n 12 - a),) > dp"

ImJl=fJ-s

where (w, v) and <f,g> are the extensions of the scalar products in L 2(X,dv)
and L 2 (S,do), respectively. Since

(A;;)(P + ia + in 12,) = Uo(P)u(P + in 12,) +

+ f U1(P-A,A)u(A+inI2,)dA,
ImJl=fJ-s

(2.25) can be rewritten as

(Au,v) = (2.34)

f <u(A+inI2,), Uo(A)*v(~+i(nI2-a),»dA+
Imlo.=T
§2. i'DOs on a conic manifold 163

+ f <u(A+inI2,·), f (U I (}L-A,A))*v(jL+i(n12-a),·)d/L>dA,
ImA=7 Imj1=7

where T = /3-s. Note that _

(UI(}L-A,A))* = (Ui(~-i:i,~))

(the operator at the righthand side is obtained from U(p,A) by transition to the
adjoint and subsequently applying the Mellin transform with respect to the first
argument). Since Uo(Ar = Uo(~), (2.34) implies (Au, v) = (u,h), where

iz(~+inI2,·) = Uo0)v(~+i(nI2-a),·)+
+ f Uj0-i:i,~)v(jL+i(nI2-a),·)d/L.
Imj1=7

The latter equation can be rewritten as

iz(~+inI2,·) = U o(A+ia)v(A+inI2,·)+

+ f UI (A -/L,A + ia)v(}L+ in 12), ·)d/L.

Imj1=T'

Taking into account that h = A * v leads to (2.32), in which

Tv = _1_ f pi(Hia)dA X (2.35)

y'2;ImA=T'

X f Uj (A -/L,A + ia)v(}L+ in 12, ·)d/L+

Imj1=T'

- _1_ f
pi (A+;{i)Uj (p,A + iQ)V(A + inl2, ·)dA.
y'2;ImA=7'

It remains to verify that the operator T:H=P+ Rea (X) ~ H=fi (X) defined by
(2.35) is compact. This can be done by a reasoning similar to the one given in the
proof of theorem 2.13 . •

§3. Pseudodifferential operators on manifolds with conical points

1. Manifolds with conical points. Let, as in §2, X be a conic manifold with vertex

°
0, and let S be a directing sub manifold of it (having only one point in common
with each ray). A neighborhood of is a set on X containing a subset of the form
{x =(p,O):P<C,OES}, where (p,O) are the polar coordinates on X defined by using
164 Chapter 4.

S (cf. §2.5), and ( is an arbitrary positive number. If X,X' are conic manifolds,
and K:X ~ X' is a homogeneous map of degree one, infinitely differentiable,
together with its inverse, everywhere except at the vertices, then K is called a
diffeomorphism from X to X'.
We denote by ~ a Hausdorff space having the following properties: 1) there is
a finite subset Q = {x(l), ... ,x(l)} such that ~ \ Q is an n-dimensional Coo
manifold; 2) for every point xv) E Q there are given an open set ~ C ~ con-
taining xv) and a homeomorphism Kj of this set onto a neighborhood of the ver-
tex OJ of an n-dimensional conic manifold Xj' K/X(j) = OJ; and 3) there is a
family VI + I, ... , Vp of coordinate neighborhoods on ~ \ Q, Vj C ~ \ Q, form-
ing together with the sets U I, . . . , UI a finite open cover of the manifold 01L
Let x E ~ and let (p,(J) be the polar coordinates of its image Klx) in Xj.
Coordinates of the form (p, (JI, ... , (In -I) where (Jj, . . . , (In - I are certain local
coordinates of the point (J on a directing submanifold Sj of the cone Xj will be
called local coordinates of x in Uj.
We will assume that the local charts {~,Kj}j=1 and {Vk,Kd~=I+1 form a
v
Coo atlas, i.e. in intersections of neighborhoods the local coordinates on one chart
are infinitely differentiable functions of the local coordinates on the other chart.
The notion of equivalent atlases is naturally introduced (in particular, if two
charts (Uj,Kj) and (U;,K;) belong to equivalent atlases, then the map
KjoKTI:K/Ujn Uj)~Kj(Ujn uj) can be extended to a diffeomorphism of Xj to
v
Xj). The space ~ endowed with this structure is called a Coo manifold with coni-
cal points.
We introduce the space Hp(~)

of functions on 0ll, where s is a real number and f3 = (f31, ... ,f3a is a vector
with real components. Let {rj}J = I be a partition of unity subordinate to the
cover U I , ... , UJ, V1 + I , ... , Vp. Put
1
Ilu;Hp(~)11 = .~ II(fju)OKTI;Hp/Aj)II+
} = I

p
+ ~ II(rjU)OKj-l ;Hs(Rn)11
j =1 + I

(as usual, the function (rju)oKT I is extended by zero onto Xj (j = 1, ... ,1) or
§ 3. it DOs on manifolds with conical points 165

onto R n (j=l + 1, ... ,p)). The space Hp(0fL) is defined as the completion of the
set COO (0fL \ Q) with respect to the norm 11·;H:e(0fL)II.
We will say that a function u given on 0fL \ Q belongs to the class (is of class)
v
Coo (0fL) if it is infinitely differentiable outside Q, while near every conical point
x(f) E Q it is such that the functions Pt-+(pDp)q(uoICT1)(p,) E Coo(Cj ) are con-

tinuous on an interval [0,8]. Here 8> 0, Sj is a directing sub manifold of the cone
Xl' and (p,fJ) are polar coordinates on Xj.

2. Pseudodifferential operators on a manifold with conical points. Let

(x,~t-+tP(x,~ be a function of class Coo(RnX(Rn\o)), positively homogeneous
of degree a in ~, and independent of x outside a ball Ix I < roo Let also
X E Coo(R n), with X(~ = I if I~I > I, and X(~ = 0 if I~I < 112.
The operator

(Au)(x) = feiX~x(~tP(x,~u(~d~, (3.1)

Rn

where U E COO (Rn) and u is the Fourier transform of u, is called a standard

pseudodifferential operator of order a, and tP is called its symbol. It is well-known
(cf., e.g., [32]) that the operator (3.1) realizes a continuous map
A:Hs(Rn) ~Hs -m(Rn).
We now define pseudodifferential operators on a manifold 0fL with conical
points.

D e fin i t ion 3.1. A linear operator A:H:e(0fL)~H:e-Rea(0fL) is called a

pseudodifJerential operator of order a on the space H:e (0fL) if the following condi-
tions are fulfilled.
v
A) For functions ~,1/ from Coo (0fL) with supports in a neighborhood U of a con-
ical point x (f) we have

where Au:Hp/Xj) ~ H pj- Rea (Xj) is a pseudodifferential operator of order a

with variable symbol on the cone Xj' IC: U ~ Xj is the 'coordinate' homeomor-
phism, and T:H:e(0fL) ~ H:e - Rea (0fL) is a compact operator.
v
B) For functions 1/,~ E C oo (0fL) with supports in a coordinate neighborhood
166 Chapter 4.

v, V c ~\Q,

where Av is a standard pseudodifferential operator of order a on IR n (of the

form (3.1)) and T 1S, as before, a compact operator from Hfi(~) to
Hfi - Rea (~).
v
C) If 1I,t E COO(~) and suppr n SUPPll = 0, then the operator
TJAr:Hfi (~) ~ Hfi - Rea (~) is compact.

We denote by ~Q the (C l ) manifold with boundary (a compactification of

~ \ Q) obtained by gluing to ~ \ Q the 'origin' of all generators of the tangent

cones Xj (distinct generators having distinct origins), and let To (~) be the bun-
dle of nonzero cotangent vectors over ~. The boundary a~Q consists of 1 com-
ponents, corresponding to the conical points x(l>, ... ,xU). A generator of Xj
with direction cp corresponds to the tangent (cotangent) space containing the vec-
tor cp (the dual vector cp*).
For a pseudodifferential operator A on ~ we can define its symbol «I> as a
function on To (~Q). In a coordinate neighborhood V such that V C ~ \ Q the
representative of «I> is the symbol «I>v of the standard pseudodifferential operator
Avon IRn. If U is a neighborhood of a conical point x(j), then the operator Au
(cf. A) of definition 3.1) has a representation (2.24). The corresponding mero-
morphic pseudodifferential operator U(J)(p,A) with variable symbol is equal to a
sum of operators U~)0)+ UY)(p,A) (definition 2.9). In the fiber above a point in
the boundary a~Q the symbol «I> of A coincides with the principal symbol of the
meromorphic pseudodifferential operator U~)0). So, the symbol «I> is a homogene-
ous function on the fiber of TO(~Q) above each point x E ~Q.

Pro p 0 s i t ion 3.2. For a given such function there is on Hfi (~) a
pseudodifJerential operator of order a having this function as symbol.
The proof is similar to the verification of proposition 2.2. We only indicate
the necessary changes. Let U I, . . . ,Uz, VI + I, . . . ,Vp be the cover of ~ and let
{tj}j = I be a partition of unity subordinate to it. Denote by lIj' j = I, ... ,p,
smooth functions for which rjllj = tj , SUPPllj C Uj ' j = I, ... ,I, and
SUPPllj C Vj for j = I + I, ... ,p. Suppose that UV)(p,A) are meromorphic
pseudodifferential operators on Xj, j = I, ... ,1, whose symbols coincide with the
§ 3. i'DOs on manifolds with conical paints 167

given functions (cf. proposition 2.2 and example 2.10). Define the operator AUj by
(2.24), in which U is replaced by U(j). Let also AVj be standard pseudodifferential
operators on R n with the given symbols. Put

In order to prove that A is the operator looked for, we may reason as in the
proof of proposition 2.2. Instead of proposition 3.2.2 we must use theorem 2.13
in this situation, and take into account, moreover, the following fact. By (2.3.4),
for arbitrary functions t 1) E CIf (Rn \ 0) we have

X CP(r, cp,w)Eo-><j>(A)(1)u)(A + in / 2,O)dl\. + Ku,

where 'T = f3 - s, K is an operator acting from every space H-p (Rn) to (R n), Co
and the lefthand side is to be understood as the analytic extension with respect to
the parameter a - the degree of homogeneity of the function ~ 1-+ CP(·,g). (Clearly,
~F-l CP(x, ~)F1) = ~F-l CP(x, g)X(g)F1) + T, where T maps Hp (Rn) to CIf (Rn \ 0).)

• A pseudodifferential operator A will be called elliptic if its symbol does not

vanish on To (V1LQ). Thus, if A is elliptic, then the meromorphic
pseudodifferential operator U(j) is elliptic on the cone Xi' j = 1, ... ,I.

The 0 rem 3.3. A pseudodifJerential operator A:Hp(':)TL) ~ Hp - Rea (':)TL) is

Fredholm (i.e. is an operator with closed range and finite-dimensional kernel and co-
kernel) if and only if A is elliptic and the line 1m I\. = f3 - s does not contain poles of
the operatorjunction 1\.1-+ U~)(A)-l, j = 1, ... ,I *).
The study of properties of pseudodifferential operators given in the preceding
paragraphs allows us to use for the proof of this theorem the traditional scheme,
related to the construction of regularizers (parametrices), which invert the
')Every strip T\ .;;; Imi\..;;; T2 not containing poles of the operator-function i\.~ UW1(i\.) can
contain at most finitely many poles of the function i\. ~ U~)(i\.) - \ (cf. remarks 1.8 and 1.9).
168 Chapter 4.

behavior of A up to compact summands (cf., e.g., [1]). By the way, the proof can
be performed by repeating, with obvious changes, the proof of proposition 4.4 to
be given in the sequel.

§4. Algebras generated by pseudodifferential operators of order zero

Everywhere in this paragraph we will consider pseudodifferential operator of order
zero only. In the definition of a standard pseudodifferential operator, we can do
without the cut-off function X, i.e. we can write a standard pseudodifferential
operator on IR n as

1. The algebra generated by pseudodifferential operators on smooth manifolds with

discontinuities 'of the first kind' in the symbols. We assume that the set of conical
points is empty, and that 01L is a compact Coo manifold without boundary. By
proposition 3.1, the operator Av in B) of definition 3.1 can be rewritten as

(Avu)(x) = _1_ f r i (A+inI2) X (4.1)

-)2;ImA=D

X E()-4q,(A) -1f1>v(r, q,,0)4-40(A)u(A + inl2,l/J)dA,

where (r, q,) are local spherical coordinates with origin at a point xeD) E K(V),
o , ; :; r < 00, q, E IR n, Iq, I = 1. Suppose that in every small coordinate neighbor-
hood V, except perhaps at xeD>, the symbol fl>v is a smooth function of the local
coordinates x = (r, q,) and (J E Sn -1 . Suppose also that
fl>y(r,q"O) = fI>~(q"O)+fI>~P(r,q"O), and let for arbitrary nonnegative integersp,q,
andk,

(4.2)

Then (4.1) is a pseudodifferential operator of order zero with variable symbol on

H8(lR n ) = L 2 (lR n ) (in the sense of definition 2.11).
We will assume that a Riemannian metric is given on '!)]t, as well as a positive
measure J1- induced by it. By L2(01L) we denote the space of scalar-valued func-
tions on '!)]t, endowed with the norm
§4. Algebras generated by it'DOs of order zero 169

Let S(~) and S*(~) be the bundles of tangent and cotangent unit vectors.
The symbol eI» of a pseudodifferential operator A on L2(~) is a complex-valued
function on the Whitney sum S(~)EBS*(~). The conditions to which the func-
tion el»v is subjected imply that everywhere on ~ except possibly at a finite point
set singel», the symbol eI» is a constant on a fiber S(~)x. (Near a point x E singel»
the symbol satisfies inequalities (4.2).) Under these conditions we will say that at
the points of sing eI» the symbol has a discontinuity of the first kind. We stress that
the set sing eI» depends on the operator in question.
A pseudodifferential operator A:L2(~) ~L2(~) is put in correspondence with
a family of maps

(4.3)

parametrized by points X E R and x E ~. The operator U(x, X) is defined by

U(x,X) = E6~tjl(A)-lel»(x,tp,8)1;p~6(A),
where tp,1/I E S(~)x, () E S*(~)x. The family (4.3) is called the operator symbol
of the pseudodifferential operator A. Thus, pseudodifferential operators differing
by compact terms only have the same operator symbol.
We denote by @ the involutive Banach algebra*) of operator-functions given on
~ X R, with pointwise multiplication, generated by the operator symbols and
endowed with the norm

Elements U and U* adjoint in @ take values U(x, X) and U(x, X)* adjoint with
respect to the scalar product in L 2 (S (~)x). It is clear that @S is a C* -algebra*).
In this paragraph, an algebra will always mean a C* -algebra.
Let ce be the algebra generated by the compact operators and the
pseudodifferential operators of order zero on L2(~)' and let ~JfLd~) be the ideal
of compact operators. This section is dedicated to the proof of the following
-)Cf. the definitions in §1, Chapter 5.
170 Chapter 4.

assertion.

The 0 rem 4.1. The map A 1-+ U, assigning to a pseudodijJerential operator A its
operator symbol U EO @5, defines an isomorphism between the algebras @/9{L2(0TL)
and @5. In particular*), the norm of a residue class [A] in the quotient algebra
@/:J[L2(0TL) is equal to the norm IIU;@5/1 of its operator symbol in the algebra @5.
The proof of the theorem is preceded by several propositions.

Pro p 0 sit ion 4.2. Let A (I.m) be a pseudodijJerential operator or an operator

adjoint to a pseudodijJerential operator, let u(I,m) be its operator symbol,
I = 1, ... ,1., m = 1, ... ,M. Then to the operators A = ~/IImA (/,m) and A*
correspond the operator symbols U = ~/IImU(l,m) and U*.
The proof of this proposition reduces to the verification of it for operators
on IR n , and hence follows from (4.1) and theorems 2.12 and 2.16 . •

Rem ark 4.3. Let A and B be pseudodifferential operators of order zero on

L2(0TL) (in the sense of definition 3.1, in which A) is omitted), and let CP, i' be
their symbols (ordinary, not operator symbols). If the sets singCP and singi' of
singular points are empty, then AB and A * are, clearly, pseudodifferential opera-
tors of order zero; in the opposite case this need not hold, in general. Theorem 1.5
implies that for an x EO sing cP the operator symbol U(x, .\)* equals
E('\) - 1cp(x, cp, fJ)E('\) + T('\), with, moreover,
IIT('\);L2(sn-l)~H\.\,sn-l)11 ~const on the line Im.\ = O. If x EO singi',
theorem 1.4 gives that U(x, .\)~(x,.\) = E(.\) -lcp(x, cp,fJ)i'(x, CP,fJ)E(.\) + S(.\), while
for S(.\) the same estimate holds as for T('\). If the points of the sets singCP and
sing i' are regarded as being conical, then AB and A * are pseudodifferential
operators of order zero in the sense of definition 3.1 (but now taken into account
requirement A).

Pro po sit ion 4.4. Let A(l,m) be pseudodijJerential operators, let u(/,m), A and
U be as proposition 4.2. Let also cp(l,m) be the ordinary (scalar) symbols of the A(/,m).
The operator A:L 2(0TL) ~ L2(0TL) is Fredholm if and only if the following conditions

')An isomorphism of C'-algebras is always an isometry, cf. §l, Chapter 5.

§4. Algebras generated by itDOs of order zero 171

hold:
1) the function () = ~IIIm()(l,m) does not vanish on S(~)EI1S*(~);
2)at each point x E U/,msing()(I,m) the operator{unction AI-+U(X,A)-l does not
have poles on the line 1m A = 0.

Rem ark 4.5. At the points of the set ~ \ U I,m sing ()(I,m) condition 2) is
satisfied in view of condition 1). Indeed, if x f£ U sing()(I,m) then ()(/,m)(x,cp,O) is
independent of cp, hence

U(X,A) = ~IIE(A)-l()(I,m)(x,O)E(A) =
I m

= E(A) -1 ~II ()(/,m)(x, O)E(A).

I m

By condition 1) the inverse operator U(X,A)-l = E(A)()(x,O)-lE(A) exists on the

line 1m A = 0, which was required.

Proof 0 f pro p 0 sit ion 4.4. Sufficiency of the conditions 1) and 2) is

verified by constructing a left and a right regularizer (parametrix). Write A as

A = ~t" A v'I'/" + T,
where {nis a partition of unity subordinate to the atlas {V,le}, TI E COO(~),
sUPPTl C V, and tTl = t; T E :J{L2(~); Av is the operator on L 2(lR n ) given by

Avu = _1_ f r i (Hin/2)Uy(r,A)u(A+inl2,·)dA,

y'2;Iml\=o

and Uy(r, A) denotes the operator ~IIImE(A) -l()~m)(r, cp,O)E(A).

Suppose that the neighborhood V contains a point of U I,m sing ()(/,m) . Then in
the local spherical coordinates with origin at this point we can represent ()~m) as
()~m)(r,cp,O) = ()U,m)(cp,O)+()~I,m)(r,cp,O). Put Uy(r,A) = Uo(A)+U 1(r,A), where
Uo(A) = ~IIImE(A)-l ()U,m)(cp,O)E(A). Condition (4.2) implies that for a fine par-
tition of unity the norm of U1(r,A):L2(sn-1)~L2(sn-1) can be taken arbi-
trarily small. By condition 2) the operator Uo(A) has an inverse on the line
ImA = 0, and on this line IIUo(A)-1;L2(sn-1)~L2(sn-1)11 o;;;;const (a strip
limA I < 8, with 8 a small positive number, contains only finitely many poles of
the operator-function A1-+ Uo(A) -1; cf. remark 1.9). Hence the operator
172 Chapter 4.

Uy(r,A)-1 :L 2(sn -I) ~L2(sn -I) exists on the line ImA = O. Put

Rv! = _1_ f r i (HinI2)Uy(r,A)-lj(A+inI2,·)dA. (4.4)

-J2;"ImA=O

If the neighborhood V does not contain points of U/,msing«P(/,m), the existence

of U y(r, A) -I is guaranteed by condition 1) already (cf. remark 4.5). The operator
Rv is also in this case defined by (4.4).
Introduce the operator

We show that RA = I+T, where T E %L2(~)' Restoring indices (the subscript

of the local chart) and applying theorem 2.13 we have

RA = ~(rj)"jRv/rJj),,/rktkAvk(l1k)"k +T =
j,k

= ~(rjtjRV;(l1j)"j(rk)") A V/l1k)"j +T =
j,k

where the letter T denotes distinct compact operators. The equality AR = 1+ T is

verified similarly. It is well-known (cf. [1], [3], [17]) that the operator A is
Fredholm if and only if a left and a right regularizer exist. This means that condi-
tions 1), 2) are sufficient.
We convince ourselves of necessity of these conditions. First suppose that con-
dition 1) is violated at a point x(O) fl U/,msing«P(/,m). Let (V,K) be a local chart,
x (0) E V, K(X(O» = O. For functions r,l1 E COO(~) with supports in Vwe have

(4.5)

where T E %L2(~) and the symbol «Pv of the standard operator Avon IR n
satisfies «Py(O,Oo) = 0 for some 00 E Sn -I. Denote by {Vj} a sequence of func-
tions in L 2 (lR n ) such that Ilv)1 = 1, while supp Vj contract towards the coordinate
origin. Clearly, the sequence {Vj} converges weakly to zero in L2(lRn). The
sequence Wj(y) = e -i P/llo'Y)Vj(Y) has the same properties, as Pj ~ + 00. Since
«Py(O,Oo) = 0, we have Avwj ~ 0 in L2(lRn), and the weak convergence to zero of
the Wj implies that IIT(wjOK)11 ~ 0 for any compact operator T. If A were
§4. Algebras generated by i'DOs of order zero 173

Fredholm, the estimate lIull .;;;; (IiAull + IITull) would hold for some T E 'XL2(~)
(norms in L2(~»' This estimate is proved false on the sequence {WjOK} by using
(4.5).
Suppose now that condition 1) is violated at a point x(O) E UI,msingCP(l,m), i.e.
CP(x(O),«Po,Oo) = 0 for certain 1/>0 E S(~)x(O), 00 E S*(~)x(O). Put, as before,
Uo(A) = ~lnmE(A)-Jcp(l,m)(x(O),I/>,O)E(A). Proposition 3.6.4 implies that there are
a sequence {Uj} C Coo(Sn-J), lIu;L 2(sn-J)11 = 1, and a sequence of numbers
{Aj } IR, Ar~OO, such that IlUo(Aj)uj;L2(sn-J)II~0. Hence the sequence
C
v/A+inI2,-.[;) = rj(A)u/-.[;), where IIvj;L2(lR n)1I = 1 and the support of rj lies in
an interval on the real axis with center at Aj and of length dj ~ 0, satisfies

(4.6)

here

Aov = _1_ f r i (Hinl2)U o(A)ii(A+inI2,)dA. (4.7)

y'2;ImA=O

The functions Vj in (4.6) may be replaced by the functions

w/r,l/» = .~r;n/2 f [;.]i(Hinl2)Vj(A+inI2,I/»dA,

V 2'1T ImA=O }

where rj ~ O. The supports of the Vj 'contract' towards the coordinate origin;

more precisely, there is a sequence {8j }, 8j ~ 0, such that JIx I >6j Iw/x) 12 dx ~ O.
The sequence Wj converges weakly to zero in L2(JR n ).
Let Uo(A) and U J(r, A) have the same meaning as in the first part of the proof,
and let the operator A 1 be the righthand side of (4.7) after replacing U(A) by
UJ(r,A). Let also X E coo(Rn), with X(x) = 1 if Ix I < £/2, and x(x) = 0 if
Ix I >(. We have

(4.8)

where [A],X] = A]X-XA]. Since the commutator [A J,X] is compact, the

sequence [A],X]wj tends to zero in L2(Rn). For small (the term XAJwj remains
small, and for large j the last term in (4.8) is small. The estimate
Ilull';;;;c(IIAull+IITull) is in this case not satisfied by the sequence {Wj}'
In order to prove necessity of condition 2) we can use a similar reasoning.
174 Chapter 4.

Recall that a pole Ao of the operator-function AI-> UO(A) - I is an eigenvalue of Uo

(cf. remark 1.9). In other words, there is a nonzero solution u E L 2 (sn -I) of the
equation Uo(Ao)u = 0. For Uj we may now take an eigenvector u of Uo
corresponding to an eigenvalue Ao on the line ImA = 0, and for {~j} we may take
a sequence of functions with supports contracting towards Ao .•
Theorem 2.16 and the proof of proposition 4.4 imply that proposition 4.4
remains valid if there are adjoints to pseudodifferential operators among the
A (I, m). Therefore the following assertion is true.

Pro p 0 sit ion 4.6. Let A, U be as in proposition 4.2, and let, as before,
tP = };/IImtP(/,m). An operator A* A ~ ,u1:L 2(cryR.) ~ L 2(r:)]L) is Fredholm if and only
if the following conditions hold: 1) the function tPtP ~,u does not take the value zero
on S(r:)]L)EBS*(r:)]L); and 2) at each point x E U/,msingtP(/·m) the operator
U(x, A)*U(X, A) ~,ul does not have eigenvalues on the line ImA = 0.

Proof 0 f the 0 rem 4.1. By proposition 4.2 it suffices to prove that

(4.9)

where T runs over the ideal :JCL 2(r:)]L) of compact operators. Equation (4.9) fol-
lows from proposition 4.6 and the fact that for any bounded operator the relation
infTIIB+TI1 2 = infTIIB*B+TII = sup{,u:,u E oess(B*B)} holds; here oess(B*B)
denotes the spectrum of the residue class [B*B] in the quotient algebra
el:JCL2(r:)]L), where e is:he algebra of bounded operators in L2(r:)]L) (cf. [24]). •
2. The algebra generated by pseudodifferential operators with smooth symbols. Let
ifo be the subalgebra of the algebra 6t generated by the pseudodifferential opera-
tors with smooth (scalar) symbols (i.e. sing tP = 0 for the symbol of every
pseudodifferential operator in ifo). Let also A(I,m\ u(I·m), A, and U be as in propo-
sition 4.2. Since the symbols tP(/,m) of the operators A (I.m) do not depend on a
point cP E S (r:)]L)x, we have
L M
U(x,A) = ~ II EO~<f>(A)-ltP(I,m)(x,O)E(A) =
1=1 m=1
§4. Algebras generated by i'DOs of order zero 175

where «P(x,O) = ~IIIm«P(l,m)(x,O). The operator E(A):L2(sn-l)-4L2(sn-l) is

unitary on the line ImA = 0 (corollary 1.5.6). Hence

IIU(x,A);L 2(sn-1)-4L 2(sn-1)11 = max I«P(x, 0) I.

o

Equation (4.9) for A E CPo takes the form

inf IIA+T;L2(01L)-4L 2(01L)11 = !\lax I«PI

T s (~1L)
= 1I«P;C(S*(01L)II.
Together with theorem 4.1 this leads to the following well-known assertion.

The 0 rem 4.7. Let CPo be the algebra generated by the pseudodifjerential opera-
tors with smooth (scalar) symbols on a smooth compact manifold 01L without boun-
dary. Then the quotient algebra CPo 1%L 2 (01L) is isomorphic to the algebra C(S*(01L»
of continuous functions on the bundle S * (01L) of cotangeni unit vectors. In particular,
the norm of a residue class [AJ is equal to the norm 1I«P;C(S*(01L»11 of its symbol «P.
Theorems 4.1 and 4.7 can, in an obvious way, be generalized to
pseudodifferential operators acting on sections of vector bundles. In this case, the
norm 1I«P;C(S*(01L»11 figuring in theorem 4.7 must be understood to mean
SUPj,s'(~1L)Sj, where S] are the eigenvalues of the matrix «P*«P, Sj ;;;. O. In the sequel
we will not give indications as to the possibility of similar obvious generalizations.

3. The algebra of pseudodifferential operators on a manifold with boundary. Let 01L

be a Coo manifold with boundary a0l4 and let ce be the algebra of operators on
L2(01L) generated by the pseudodifferential operators whose symbols satisfy the
same conditions as in section 1 (i.e. they are allowed to have singularities of the
first kind).
Denote by S + (01L)x, x E a0l4 the hemisphere corresponding to the tangent
vectors directed to the side of the manifold, and by p+ :L 2 (S (01L)x) -4 Lt (S(~)x)
the operator of multiplication by the characteristic function of S + (01L)x.
Let A be a pseudodifferential operator with symbol «P, and lel
U(x, A) = EO--><I> (A) -lcI>(x, CP,O)E.r-->o(A), where CP,l/J E Put
S(01L)x, () E S* (01L)x.
~(x, A) = p+ U(x, A):Lt (S (01L)x) -4 Lt (S (01L)x) E a~
if and
x
~(x, A) = U(x, A) if x ft: a01L. Introduce the algebra @ generated by the operator-
functions (x, A) t-> ~(x, A), with pointwise multiplication, involution, and norm
176 Chapter 4.

111S;<r11 = sup 111S(x,A)II.

Cx, A) E":mtx R

The 0 rem 4.8. The map A t-+ IS defines an isomorphism of the algebras
ttl%L 2(sn-l) and@.

The proof is a repetition, with obvious changes, of the proof of theorem 4.1 .

•
4. The algebra of pseudodifferential operators on a manifold with conical points.
v
Let 0TL be a Coo manifold with conical points (cf. §3.1). Denote by tt the algebra
generated by the pseudodifferential operators on ~ (0TL) _ H p(0TL) (here /3 is a
vector with real components (f3J, ... ,/31) and I is the number of conical points
xCI), ... ,x(l)). We will assume that discontinuities of the first kind are allowed in
the symbols of the pseudodifferential operators. For a pseudodifferential operator
A we introduce the operator symbol U, defined for x =F x(j), j = 1, ... ,1, as in
section 1, and for x = x(j) as the operator function At-+ U(x(j\A) given on the
line IX(j) = {AEC: ImA=/3j}; here U(x(j),,) denotes a meromorphic
pseudodifferential operator on the cone Xj (cf. definition 2.1). Put also
Ix = {AEC: ImA=O} for x =F x(j), j = 1, ... ,1.
Let @ be the algebra generated by the operator symbols and endowed with the
norm

IIU:@II = su~ IIU(X,A);L2(S(0TLU ~L2(S(0TL)xll

x E"_lt
AE"lx

(if x = x(j), then S(0TL)x denotes a directing manifold of Xj)'

The 0 rem 4.9. The map A t-+ U defines an isomorphism of the algebras
tt/:JfH p(0TL) and @.
This assertion can be proved similarly as theorem 4.1; the presence of conical
points does not present additional difficulties. Theorem 4.9 can, in an obvious
way, be generalized to manifolds with boundary (cf. theorem 4.8). Of course,
theorems 4.1, 4.7, and 4.8 follow from the assertions formulated in this section.

5. The algebra of one-dimensional singular integral operators with discontinuous

§4. Algebras generated by i'DOs of order zero 177

coefficients on a composite contour. Let {'tu} be a finite set of smooth (Lyapunov)

closed directed arcs in the complex plane. Suppose that two arcs 'til and 'tv can
intersect at finitely many points only; moreover, either 'til U 'tv is a smooth arc in
a neighborhood of a common point x, or 'til and 'tv do not have a common
tangent at x. Under these conditions the union f = U Il'tll will be called a piece-
wise smooth contour. Those points of f at which there is no neighborhood on f
that is a smooth open arc will be called singular points, or cusps, and the other
points will be called regular. The singular points will be denoted by z], ... ,Zm.
A function h defined on f \ U ~=] {zv} is called piecewise continuous if: 1) h is
continuous everywhere except at finitely many points of discontinuity of the first
kind; and 2) when passing along an arc to a cusp h has a finite limit, which
depends, in general, on the arc. For the sake of being specific we will assume that
h is 'left' -continuous.
Put p(z) = II~=]lz-zol.B" where 2f3v E(-l,l), and introduce the space
L 2 (f,p) with norm

A singular integral operator with piecewise continuous coefficients a and b is an

operator of the form Au = au + b'Xru, where

('Xru)(z) = _1 f ~~) dr.

'fTi r.ll z

The operator A is continuous in L 2 {f,p) (cf., e.g., [17]).

"0
Let tj = {z EC: z =Zv +rve' i'}, where 0".;; r < 00, 0".;; Ihv < ... < ()N,v < 2'fT,
be the tangent rays to the arcs containing the cusps Zv. For a piecewise continu-
ous function h, let hlzv) denote the limit of h (z) as z ~ Zv along the arc tangent
to the ray tj, and let, at a regular point z, h±{z) denote the limit values of h 'from
the right' and 'from the left'.
At each regular point Z E f we define a (2 X 2)-matrix-function AH U{z, A) on
the line ImA=O by

U(Z,A) = (4.10)

= [a+{Z)+b+{Z)tanh'fTA. ib+{z)/cosh'fTA. 1
-ih_{z)/cosh'fTA a_{z)-b_(Z)tanh'fTA·
178 Chapter 4.

At a cusp Zv we define a matrix-function "A ~ U(zv,"A) of dimension No X Noon

the line Im"A = Pv' If all arcs containing Zv 'leave' zv, then we put
U(zv,"A) = /lUjk(zv,"A)llfk = 1, where Ujk(zv,"A) = aj(zv)~jk + bj (zvY5jl!("A), with
~I!("A) coinciding with the righthand side of (2.13) after replacing N by N v, and ()j
by ()jV, j = I, ... ,Nv ' If, however, the k-th arc enters, then the k-th column of
the matrix U must be defined by Ujk(zv,"A) = a/zv,"A)~jk - b/zv)~I!("A).
It is obvious that the off-diagonal entries of U(z,"A) decrease exponentially for
Z E f fixed and "A tending to 00. The diagonal entries have finite limits as "A tends
to an arbitrary 'end' at infinity along a corresponding line. We will assume that
the matrix U is defined at these 'ends' at infinity, and equals its limit values.
The matrix-function U will be called the (operator) symbol of the singular
integral operator A = al +b%r.
Denote by ce the algebra generated by the singular integral operators on
L 2 (f,p). We also introduce the algebra ® generated by the symbols of the singu-
lar integral operators. The norm of an element U E ® is defined by
/lU;®/I = sUPA,Z,js/z, "A), where s/z, "A)2 are the eigenvalues of the matrix
U(z,"A)*U(z,"A), Sj;;;;' 0, z runs over f, and the parameter "A runs over the line
Im"A = °if z is a regular point, and over the line Im"A = fiv if z = Zv'
The 0 rem 4.10. The set %L 2(r,p) of compact operators on L 2(r,p) is con-
tained in the algebra Ci'. The map 'singular integral operator -+ symbol' defines an
isomorphism of the algebras ce/%L 2(f,p) and ®; the norm of a class in ce/~{L2(r,p)
equals the norm of its symbol.
The proof of this theorem is simpler than that of its multidimensional version
theorem 4.9. It is given in [56].

6. The algebra of pseudodifferential operators with oscillating symbols. Let ~ be a

Coo manifold without boundary, and let A be a pseudodifferential operator of
order zero on~. We will assume that the symbol «I> of A is a smooth function
everywhere on S * ('!)It) except at a set S* ('JR.) I sing «1>, where sing «I> is a finite subset
of '!)It*). Suppose that in a neighborhood of a point x E sing «I> the symbol «I> has
')S'(~)lsing41 denotes the part of S'(~) above sing4l, ct. §2, Chapter 5. (Translator's
note.)
§4. Algebras generated by i'DOs of order zero 179

in local spherical coordinates a representation ~o(r, cp,O) + ~ 1(r, cp,O), where ~o is a

'periodic' function of r, i.e. ~o(re,·) = ~o(r, .), ~o E coo(dXsn -1 XSn -1) (d
denotes the interval [l,e] with end points identified), while for ~1 the estimates
II(rDr)k~l(r, ·);CP(Sn -1 XSn -1)11 = O(rB) hold, for arbitrary k,p, and a certain
lj > O. The set of functions ~ satisfying these requirements is denoted by 91. We
introduce the algebra ee generated by the pseudodifferential operators of order
zero on L2('!JlL) whose symbols belong to 91. For a pseudodifferential operator A
with symbol ~ E 91 we define its operator symbol A(x, X), x E ~ 0.;;;; X.;;;; 27T,
acting on functions v E Coo(dXS('!JlL)x), by
+00
A(x,X)v = l: r 27Tij U(X, X+ 27Tj)Vj;
j=-oo

e
vicp) = jv(r,cp)r-27Tij-ldr.
1

By () we denote the algebra generated by the functions (x, X) ~ A(x, X), with point-
wise operations and norm

IIA;()II = ~~.r IIA(x,X); L 2 (II) ~L2(II)II,

n=dXS n - 1•

The 0 rem 4.11. The map A ~ A defines an isomorphism of the algebras

ee/:KL 2 ('!JlL) and ().
This theorem can be proved along the same lines as theorem 4.1.
 Chapter 5

The spectrum of a C* -algebra of pseudodifferential

operators with discontinuous symbols on a closed manifold

The spectrum of an algebra is the set of equivalence classes of irreducible

representations of this algebra, endowed with a special topology. The algebra a
generated by the pseudodifferential operators with smooth symbols on a smooth
closed manifold 'JIL becomes commutative after factorization by the ideal %L 2('JIL)
of compact operators. Hence all irreducible representations of the quotient algebra
a/%L 2('JIL) are one-dimensional, and its spectrum can be identified with the maxi-
mal ideal space, which is the bundle S*('JIL) of cotangent unit vectors. If A E a,
[A] E a/%L 2('JIL) is its residue class, and 'TT runs over the spectrum of a/%L 2('JIL),
then the map 'TTI-+'TT[A] coincides with the function S*('JIL) 3 'TTI-+«P('TT) E C, where
«P is the symbol of A.
In this Chapter we study the spectrum of an algebra of pseudodifferential
operators with singularities either in the symbols or on the manifold. The quotient
algebra a/9{L2('JIL) remains commutative, among its irreducible representations
there are infinite-dimensional ones. We compute all equivalence classes of irredu-
cible representations, indicate a realization of these representations (i.e. describe a
map 'TT 1-+ 'IT[ A D, and elucidate a topology on the spectrum.
§1 contains an outline of the results used from the theory of C* -algebras. It is
supplemented by references to [24], where the corresponding proofs can be found.
§2 is devoted to the statement of a theorem about the spectrum of the algebra a
generated by the pseudodifferential operators with discontinuities 'of the first
kind' in the symbols; variants and consequences are discussed also. In §3, §4 we
prepare the proof of this theorem, and study 'local' algebras (of values of operator
symbols at a point or restrictions of them onto a line). We finish the proof of the
theorem in §5. In it, necessary and sufficient conditions for an arbitrary element
of a to be Fredholm are given also. A description of certain important ideals in CP
is contained in §6. In §7 we study the spectrum of the algebra of
181
182 Chapter 5.

pseudodifferential operators on a manifold with conical points, and in §8 we study

the spectrum of the algebra of pseudodifferential operators with oscillating sym-
bols.

§1. Results from the theory of C* -algebras

1. C*-algebras. Let Ci be an algebra over the field of complex numbers. We say
that a map x t->x* of Ci into itself is an involution if the following conditions hold:
(x*)* = x, (x +yr = x* +y*, (;\x)* = ;\x*, and (xyr = y*x*. An algebra
endowed with an involution is called involutive *) .
An involutive normed algebra is a normed algebra Ci endowed with an involution
such that Ilx*11 = Ilxll for all x E Ci. A complete involutive normed algebra is
called an involutive Banach algebra.
If the involution has the property Ilx*112 = Ilx*xll for any element x E (£, then
the involutive Banach algebra Ci is called a C* -algebra.
Let Ci be an involutive algebra. A subalgebra of Ci that is mapped into itself
under the involution is, by definition, an involutive subalgebra. Every closed invo-
lutive subalgebra of a C* -algebra is a C* -algebra.
Denote by e(H) the algebra of all bounded linear operators on a Hilbert space
H. Involution in e(H) means transition to the adjoint operator. Every closed
involutive subalgebra of e(H) is a C* -algebra.

2. Morphisms of C* -algebras. Let Ci and '51J be involutive algebras. A morphism

f:Ci~'51J is a linear map satisfying j(xy) = j(x)j(y), j(x*) = j(x)* for all
x,y E Ci. A bijective morphism is called an isomorphism.
It turns out that every morphism of C* -algebras Ci and '51J (simply regarded as a
morphism of involutive algebras) is continuous. If j:Ci~ '51J is an injective mor-
phism, then Ilj(x)1I = Ilxll for all x E Ci.
Every C* -algebra is isomorphic to a closed involutive sub algebra of an algebra
e(H).

*J An involutive algebra is also called an algebra with involution. (Translator'S note.)

§l. Results from the theory of c* -algebras 183

3. Quotient algebras. Let. te be a C* -algebra and J a closed two-sided ideal in teo

Then J is self-adjoint (i.e. is preserved as a set under involution) and the quotient
algebra ell J, endowed with the quotient norm and corresponding involution, is a
C* -algebra.

Pro p 0sit ion 1.1. Let a and ~ be C* -algebras, cp:a~q>, a morphism, and I
the kernel of cpo Then I is a closed ideal in CP, and the image cp(a) is closed in ~.

Proof. The map cp is continuous, since every morphism of C* -algebras is con-

tinuous. Hence I is a closed ideal and tel I is a C* -algebra. The morphism
tel I ~q>, obtained by factoring through cp is injective, hence an isometry. Thus,
the image cp(te) is complete and closed in~ . •

4. Representations. A representation of a C* -algebra a in a Hilbert space H is a

morphism 'IT:te~ e(H). The space H is called the representation space of 'IT, and the
(Hilbert) dimension of H is called the dimension of the representation.
Two representations 'IT,'1T' of an algebra te in spaces H,H' are equivalent, by
definition, if there is a Hilbert space isomorphism U:H ~ H' such that
U'1T(x) = 'IT'(x)U for all x E a.
We say that a vector g E H is cyclic (or totalizing) for a representation 'IT of an
algebra te in H if the closure of the set 'IT(te)g coincides with H.
A representation 'IT of an algebra te in H is called irreducible if 'IT satisfies either
one of the following (equivalent) conditions:

1) the only closed vector subspaces in H that are invariant under 'IT(te) are 0 and
H;
2) the commutator of 'IT(a) in e(H) consists of scalars;
3) either every nonzero vector g E H is cyclic for 'IT, or 'IT is the null representation
of dimension 1*) .

A two-sided primitive ideal in a C* -algebra a is the kernel of an irreducible

representation of teo Every closed two-sided ideal in a is the intersection of the
primitive ideals containing it.
')This is called the nontrivial representation in [24]. (Translator'S note.)
184 Chapter 5.

In the sequel, an algebra will always mean a C* -algebra.

Pro p 0 sit ion 1.2. Let 7T be an irreducible representation of an algebra t£ in a

space H. The following assertions are true:
1) if I is a two-sided ideal in t£ and 7T(1) =1= 0, then the restriction 7T II of 7T to I is
irreducible;
2) if I 1, 12 are two-sided ideals in t£ and 7T(I 1) =1= 0, 7T(I2) =1= 0, then 7T(I 11 2) =1= 0.

Proof. 1) Put E = {xEH:7T(I)x=O}. Since 7T(t£)E C E, and by requirement

°
E =1= H, we have E = 0. Hence 7T(I)~ =1= whatever the nonzero vector ~ E H.
The subspace 7T(I)~ is invariant under 7T(t£). Since 7T is irreducible we find
'TT(I~ = H. Thus, every nonzero vector ~ E H is cyclic for 7T II.

2) As was proved in 1), 7T(lz)H = H, 7T(11)7T(12)H = H. Hence

7T(I 112) =1= 0. •

Pro p 0 sit ion 1.3. Let 1 1,12 be two-sided ideals in an algebra (f, and let I be
a primitive ideal. If I 112 C I (in particular, if I I n 12 C I), then either I I C I or
12 C 1.

Proof. Assume the contrary, i.e. I I )t I and 12 )l I. Applying the second

assertion of proposition 1.1. to an irreducible representation 7T of t£ with kernel I,
we find 7T(I 11 2) =1= 0, whence I I 12 )l I . •

Pro p 0 sit ion 1.4. Let t£ = 'X{H) be the algebra of compact operators on a
Hilbert space H. Then every nonnull irreducible representation of t£ is equivalent to
the identity representation.

Proof. See [24), 4.1.5 . •

Pro p 0 sit ion 1.5. ([24),4.1.10). Let 7T be an irreducible representation of an

algebra t£ in H. If 7T(t£) n 'X{H) =1= 0, then 'X{H) C 7T(t£) and every irreducible
representation of t£ with the same kernel as 7T is equivalent to 'TT.

Pro p 0 sit ion 1.6 ([24), 2.10.2) (extension of a representation). Let 0?J be a
§1. Results from the theory of c* -algebras 185

suhalgebra of ce. and p a representation of Iffi in a Hilbert space % Then there is a

Hilbert space H, containing K as a subspace, and a representation 'TT of a in H, such
that p(x) = 7T(x) IK for all x E Iffi. If p is irreducible, 'TT can be chosen to be irreduci-
ble.

5. The spectrum of a C* -algebra. Let Prim a be the set of two-sided primitive

ideals in ce, and suppose T C Prima. Denote by I (T) the intersection of the
ideals in T. The set I (T) is a two-sided ideal in a. Let T be the set of primitive
ideals containing I(T). It turns out that there is a unique topology on Prima such
that for all T C Prima the set T is the closure of T in this topology. This topol-
ogy is called the Jacobson topology on Prim a.
Introduce the set a of equivalence classes of nonnull irreducible representations
of a. The map 'TT ...... ker'TT defines a canonical surjection a ~ Prim a.
A

The spectrum of a is the set a endowed with the topology that is the pre-image
A

of the Jacobson topology under the canonical map a ~ Prim a.

We say that a topological space is a To-space if, given two points of this space,
there is a neighborhood of one point not containing the other point.
The following three conditions are equivalent ([24],3.1.6):
A

l)ais a To-space;
2) two irreducible representations of a with the same kernel are equivalent;
A

3) the canonical map a ~ Prim a is a homeomorphism.

A

Finally we given another description of the topology on the spectrum a of an

algebra a. Let {Xj} be a family of elements of a that is everywhere dense in (f and
let Zj = {'TTEa: 1I'TT(xj)11 > I}. Then the Zj form a base for the topology in a.
A A

Thus, if ce is separable, then a has a countable base.

Pro p 0 sit ion 1.7. Let I be a closed two-sided ideal of an algebra (f and let
&! = {'TT E &:'TT 1/:;eo}. Then the map 'TT ...... 'TT II is a bijection of &! onto 1.

The proof is contained in [24], 2.11.2.

Let, as before, 'X(H) be the algebra of compact operators on a Hilbert space H,
let X be a locally compact space, and let :J{{j(X,H) be the algebra of continuous
functions on X, with values in 'X(H), and tending to zero (in the operator norm)
186 Chapter 5.

at infinity. The norm in 'Xo(X,H) is defined by

Ilvll = suollv(x);H ~HII.

XEX

The following proposition is an instance of assertion 10.4.4 from [24].

Pro p 0 sit ion 1.8. For every x E X we define a representation ?T(x) of the

- --
algebra 'Xo(X,H) bY?T(x)v = v (x). Then x .... v(x) is a homeomorphism of X onto
the spectrum 'Xo(X,H) of'Xo(X,H).

6. Rich subalgebras. A subalgebra <!B of an algebra ce is called rich if the following

conditions hold:

1) for every irreducible representation ?T of ce the representation ?T 1 <83 is irreducible;

2) if ?Tl,?T2 are inequivalent irreducible representations of ce, then the restrictions
?Tl 1 <83, ?T21 <83 are inequivalent.

Pro po sit ion 1.9 (cf. 11.1.4 in [24]). If<!B is a rich subalgebra of the algebra
'Xo(X,H), then <!B = 'Xo(X, H).

7. A criterion for invertibility of an element of an algebra.

Pro p 0 sit ion 1.10. Let ce be an algebra with identity. An element a E (f. is
invertible if and only if for every irreducible representation ?T of ce the operator ?T(a) is
invertible (on the representation space H .,).

Proof. Necessity of the condition for a E ce to be invertible is obvious. We

convince ourselves of the sufficiency. Suppose a does not have a right (or left)
inverse. Then the element b = b' = aa' (respectively, a' a) cannot be inverted.
Introduce the commutative algebra <!B generated by b and the identity of ceo It is
well-known ([24], 1.5) that ~ is isomorphic to the algebra of continuous functions
on the spectrum Sp b. Since b is noninvertible, 0 belongs to Sp b, and there is a
one-dimensional representation p of <!B annihilating b. By proposition 1.6 there is a
representation ?T E ce extending p in the sense that H p is a one-dimensional sub-
space of H., and ?T(x) 1 Hp = p(x) for all x E ~. Since ?T(b) 1 Hp = 0, the operator
§l. Results from the theory of c* -algebras 187

7T(b) (hence also 7T(a» is noninvertible. •

Pro p 0 sit ion l.ll ([24], 1.3.10). Let ~ be a subalgebra of an algebra a

which has an identity which, moreover, belongs to~. Then for any x E ~ we have
SpctX = Sp'iSX, where Spcy denotes the spectrum of an element y in the algebra C.
Thus, if an element x E ~ is invertible in a, then it is invertible in ~ also.

§2. The spectrum of a C· -algebra of pseudodifferential operators with discontinui-

ties of the first kind in the symbols on a smooth closed manifold (statement of the
main theorem)
Let we be a closed Riemannian manifold of class Coo, and let S (~) and S' (~)
be the bundles of tangent and cotangent unit vectors. Let we be the set of func-
tions ~ given and everywhere smooth on S'(~) except at a set S*(~) I sing~,
where sing~ is a finite subset of ~ depending on the function ~, and
S * (~) I sing ~ is the part of S' (~) above sing ~. At the points of sing ~ the
function ~ is allowed to have discontinuities of the first kind (for a description of
such discontinuities see Chapter 4, §4.I). A function ~ E we can be regarded as
being given on the Whitney sum S (~)$ S* (~), where, moreover, the map
(x,~,O)~~(x,~,8) (x E ~ ~ E S(~)x, 0 E S'(~)x) does not depend on ~ if
x (/. sing~.

We introduce the space @, which, as will be proved in the sequel, is

homeomorphic to the spectrum of the quotient algebra a/:J{L2(~), where a is the
algebra generated by the pseudodifferential operators of order zero and with sym-
bols from we (here and in the sequel, an algebra means a C· -algebra).
Put @ = (S (1)]'(.)$ S' (~» u e, where e is the disjoint union of a set of lines
{Ix} indexed by the points of the manifold ~. A topology on @ is defined as fol-
lows. The part of S(~)$S'(~) lying above a neighborhood U of a point Xo is
regarded as the set of triples (x,~, 0), where x E U, and ~,O are tangent and
cotangent unit vectors. Let V(~o) and W(Oo) be neighborhoods of points ~o, 00 on
the unit spheres, and let K(~o) be a subset of U whose image under some coordi-
nate map Ie U ~ IR n is the intersection of an open cone with vertex at a point
IC(XO) and containing all directions from IC'(Xo)(V(~o» and an open n-dimensional
ball with center at IC(XO). A neighborhood in @ of a point (xo,~,Oo) in case
188 Chapter 5.

cf.JoOo = 0 is the union of sets {xo} X V(cf.Jo) X W(Oo), K(cf.Jo) X Sn-I X W(Oo), of
the set of lines lx, x E K(cf.Jo), and Ixo. If cf.JoOo~O, a neighborhood is the same
such set, but with Ixo replaced by a set {AElxo:A~N} (every line Ix is regarded as
the real axis). Finally, a neighborhood of a point A E Ix is an open interval on Ix
containing A. These neighborhoods form a base for a topology on ®. This topol-
ogy is not separable: any two neighborhoods of distinct points (XO,tpl,OI) and
(XO,tp2,lJ 2) have nonempty intersection. Oearly, @ is a To-space, i.e. for each pair
of points there is a neighborhood not containing the other point. The topology
5(S,S*) induced on S(~R.)$S*(~) is the weakest in which all functions on IJR are
continuous.
Let a be the algebra obtained by
closing, with respect to the operator norm in
L2(~)' the set of compact operators and pseudodifferential operators of order

zero whose symbols belong to IJR. The following fundamental theorem holds.

The 0 rem 2.1. Let A be a pseudodifferential operator of order zero, tp E ~ its

symbol, and U the operator symbol (ef. (4.4.3) and further). The following assertions
hold:

1) The maps and

'7T(X, A):A 1-+ U(x, A), A E lx, induce pairwise inequivalent irreducible representations
of the quotient algebra a/:J{L2(~); '7T(x,tp,O) is a one-dimensional representation,
and '7T(x, A) is a representation in the algebra of bounded operators on L2(S(~)x).

2) Every irreducible representation of the quotient algebra a/:J{L2(~) is equivalent to

one of the representations mentioned in 1).
3) The correspondence '7T(x,tp,O)I-+(x,tp,O), '7T(X, A) 1-+ AElx' defines a bijection of the
~ ~

spectrum (a/~J{L2(~» onto @. The topology on @ coincides with the Jacobson

topology.

Using theorem 2.1 we can obtain a description of the spectrum of the algebra
a. In fact, the following holds.

Cor 0 II a r y 2.2. Every irreducible representation of a is equivalent to one of the

representations mentioned in 1) of theorem 2.1, or to the iden~ity representation
(which is irreducible). The spectrum is the disjoint union of @ and a point e
(corresponding to the identity representation). In the Jacobson topology {e} is an
§2. The spectrum. Discontinuities 189

open set whose closure coincides with @ U {e}.

Proof 0 f cor 0 11 a r y 2.2. We first note that, since the algebra ce contains
all compact operators on L2(~)' its identity representation is irreducible. Let 'IT
be an irreducible representation of ce such that 'TT(9{1.d~» =!= O. Then the restric-
tion 'IT I%l.d~) is also irreducible (the first part of proposition 1.2). By proposition
1.4, 'IT I%l..2(~) is equivalent to the identity representation. This implies that the ini-
tial representation is also equivalent to the identity representation. So, in order to
obtain the complete list of representations of ce we must add to the representations
given in theorem 2.1 the identity representation. •
We now consider the subalgebra eo
of ce generated by the pseudodifferential
operators whose symbols do not have a discontinuity on a fixed set ~ C ~ i.e.
sing iP n ~ =!= 0 for the symbol iP of an arbitrary A E eo.
Discard from @ the
lines Ix corresponding to points of ~. Identify points (x, q" fJ) and (x, t/J, fJ), where
x E ~, cp,t/J E S(~)x, (J E S*(~)x' The new space, endowed with the quotient
A

topology, will be denoted by @o.

rem 2.3. All assertions of theorem 2.1 remain valid if ~ @ are replaced
A

The 0

by eo., @o, a representation 'IT(x, q" fJ) for x E ~ by the representation

A

'IT(x, fJ):A l-+iP(x, fJ), and the map 'IT(x,q"fJ)I-+(x,cp,fJ) (for x E~) by the map
'IT(x, fJ) 1-+ (x, fJ).
In particular, for ~ = ~ we are led to the following assertion (which, by the
way, follows immediately from theorem 4.4.7).

Cor 0 11 a r y 2.4. If ~ = ~ then the algebra eo I XLo(~) is commutative;

all its irreducible representations are one-dimensional. The maximal ideal space coin-
cides with the bundle S*('!)1t) of cotangent unit vectors. A residue class [A] is mapped
by the Geljand homomorphism to its symbol.
Note that the topology of @o does not have a countable base if ~ \ ~ is
uncountable. This and the assertions at the end of § 1 imply that for ~ \ ~
uncountable the algebra teo IXL2(~) (and also eo)
is not separable. It is easy to
check that this algebra is separable if ~ \ ~ is countable.
Finally, theorem 2.1 and corollary 2.2 imply that for any irreducible represen-
tation'IT of ce in a space H the image 'IT(ce) contains the ideal 'J(H). This need be
190 Chapter 5.

verified only for the representation w(x, A) in L2(S(~R.)x). The set w(x, A)(ce) con-
tains, e.g., the commutator U(x, A)f - jU(x, A), where
U(X,A) = E(A)-l<p(x,4>,O)E(A), f E COO(S('!)]t)x). This commutator is a compact
operator on L 2(S('!)]t)x) (cf. proposition 3.2.1). But then, by proposition 1.4, con-
tinuity of the representation implies the inclusion %ldS (~)x) c w(x, A)(a). Thus
a is a type I algebra (c!. [24], 9.1).
Finalizing this paragraph we give a one-dimensional version of theorem 2.1.
Let '!)]t be a union of finitely many smooth simply closed oriented contours in the
complex plane. Consider a singular integral operator

(Au)(z) = a(z)u(z)+~ j-fill-dr, (2.1)

?Tl ~Jt,~-z

where a,b are functions with, possibly, finitely many singularities of the first kind
and left - continuous. Introduce the space @. The bundle S('!)]t)$S*('!)]t) can be
represented as the disjoint union of four copies '!)]t+ +, '!)It+ _, '!)It_ +, '!)It_ _ of
'!)R., each oriented as '!)]to Let zE ~ let z/ and Zr be arbitrary points of '!)It lying
'left' and 'right' of z (near z), and let < be the order relation on an arc. Points of,
e.g., '!)It+ _ will be denoted by z + -, etc. We define a topology 5(S,S*) on
S('!)]t)E9S*('!)]t). Neighborhoods forming a fundamental system of neighborhoods
(an fsn) in the topology 5(S,S*) are: for a point z + +, the sets of the form
{x + +: z + + =s;;; x < z:- +} U {x - + : z - + < x - + < z; + }; for a point z - + ,
the sets {x - + : z - + ;;;;. x - + > z I +} U {x + + : z + + > x + + > z / + }; etc. The
space @ is the union of the four contours ~± and the set of lines {Ix}, x E '!)]to
Let p :S('!)It)E9 S* ('!)It) ~ '!)]t be projection, and V a neighborhood of z + + (z - -)
on S(~)El)S*('!)]t) in '5"(S,S*). Unions of sets V, {Ix} (X Ep(V) \z) and
{A E Iz : A> N} form an fsn on & of z + + (z - -). Similar unions with
{A E Iz : A> N} replaced by {A E Iz : A < N} form a fundamental system of
neighborhoods of z + (z -). Finally, for an arbitrary point A E Iz an fsn is formed
by the open intervals on Iz containing z. Note that, in distinction to the one-
dimensional case, for dim'!)]t;;;;.: 2 there are points (x, 4>,0) in S ('!)]t) E9 S *('!)]t) every
neighborhood of which contains the line Ix (in case cpO = 0). The topology on @
is not separable.

The 0 rem 2.5. Let A be a Singular integral operator (2.1), let U(Z,A) be the
matrix (4.4.1 0), and let if be the algebra generated by the operators (2.1). Then
§2. The spectrum. Discontinuities 191

ee :J ~JU.d~), and the following assertions hold:

1) The maps nez, -+-, +): AI-+a::t:{z)+b::t:{z), n{z, -+-, -): AI-+a::t:{z)-b::t:{z),
n(z, A): A 1-+ U{z, A), A E i z, generate pairwise inequivalent irreducible representa-
tions of the algebra ee/:xL2(~)·

2) Every irreducible representation of ee/:xL 2 (~) is equivalent to one such represen-

tation.
3) The correspondences n(z,-+-,-+-}I-+z::t:::t:, n(A,z)I-+AAE lz define a bijection of the
spectrum of ee/:J{L2(~) onto ®. The topology on ® coincides with the Jacobson
topology.

The study of the spectrum of ee/:J{L2(~) reduces, by theorem 4.4.1, to the

study of the spectrum of the algebra ® of operator symbols. The proof of theorem
2.1 is prepared in §3, §4 in which the representations of the 'local' algebras ®(A)
and ®(U are described. The algebra ®(A) is generated by the values U(x, A) of
operator symbols at A E lx, and ®(lx) is generated by the restrictions U{x,·) of the
symbols U E ® to the lines Ix. The proof of theorem 2.1 is finished in §5. The
verification of the other theorems stated in this paragraph does not present any
difficulties then. (Note that in view of theorem 4.4.8 the first assertion in theorem
2.5 is obvious; in particular, in order to convince ourselves of the inequivalence of
representations nez, A) with distinct A it suffices to compare the traces of the
matrices n(z, A)A.)

§3. Representations of the algebra ®(A) generated by the operators

E(A)-l~~,~)E(A)

Pro p 0 s i t ion 3.1. The algebra ®(A) generated by the operators

Ew-><p(A)-Iq,(4>,~)E.r->wCA) in L 2 (Sn -I) is irreducible, for every A E R.

Proof. The algebra ®(A) contains all operators of multiplication by smooth

functions (which correspond to symbols that are independent of w). Hence, every
invariant subspace of this algebra is contained in a subspace of the form
xoL 2 (S n - 1), where XO is the characteristic function of a set Q,
0< mes Q < mes Sn -I. In fact, let X be an invariant subspace and let
p:L2{Sn -I) -7 X be orthogonal projection. Then P commutes with all operators
in ®(A), in particular with operators of multiplication by functions f from
192 Chapter 5.

coo(sn -1). Therefore, P(j) = jP(1), i.e. P acts as the operator of multiplication
by P(1). Furthermore, p2(j) = P(fP(1» = jP(P(l» = jP(1)2. On the other hand,
p 2 (j) = P(j) = jP(1). Hence p(1)2 = P(I). Thus, P(1) takes only two values, 0
and 1, and thus is the characteristic function of a certain set D.
Suppose there is a nontrivial invariant subspace; let u be an element of it. We
may assume that the support of u lies in an open hemisphere. Choose a point Wo
such that the set {I/; E S n - I : II/;wo 1 < £}, where £ is a sufficiently small number,
does not intersect supp u. Denote by {Gm } a sequence of smooth averaging ker-
nels such that

f Gm(fJ,I/;)u(l/;)dl/; ~ u(O) in L 2(sn -1).

s· -)
Then

f
S·-I
Gm(O,I/;)( -wol/;+ io)-iA-nI2 u (l/;)dl/; ~ (3.1)

for almost all O. Fix a 0 = 00 such that the limit (3.1) exists and is distinct from
zero. For sufficiently large m,

f Gm(00,1/;)(-wol/;+iO)-iA-n/2 u (l/;)dl/; =1= o. (3.2)

S·-I

Obviously, the function v (I/;) = Gm(0 0,-.J!)u (I/;) belongs to the invariant subspace.
Denote by {cI>k} a ~-sequence of smooth functions such that the sets supPcI>k
contract towards a point woo Also, let a be a Coo(sn -1) function equal to zero in
a small neighborhood of the equator {cp:CPWo = o} and equal to one outside another
small neighborhood of this equator.
Put Uk(A) = E(A)-l a (cp)cI>k(W)E(A). Then

Uk(A)V ~ C(A)a(cp)(cpwo + ioiA- n/2 X (3.3)

X f (-wo-.J!+io)-iA-nI2 v (-.J!)dl/;,
s. -I

where C(A) = (2'17r ne i 'ITn/2r(-iA+nI2)r(iA+nl2) (cf. (1.2.1) and (1.4.1». For-

mula (3.2) and the equation

(cpwo + iO) -iA-nl2 =

§3. Representations of ®(A) 193

imply that the righthand side of (3.3) cannot vanish almost everywhere on
Sn -I \ U. We have obtained a contradiction. •

Pro p 0 sit ion 3.2. Every compact operator on L2(sn -I) belongs to the alge-
bra ®(A).

Proof. Proposition 3.2.1 implies that ®(A) contains certain nonzero compact
operators (commutators). Since ®(A) is irreducible (proposition 3.1), by proposi-
tion 1.5 every compact operators is an element of it. •
Denote by V(n, 2) the manifold of pairs (<p,w) of mutually orthogonal unit vec-
tors of IRn.

L e m m a 3.3. Let (~,wo) be an arbitrary point of V(n,2), and let

U(A) = E(A)-IcI>(cp,w)E(A), where cI>(cp,w) = ~ak(cp)cI>k(W) (the sum is finite, ak,cI>k
are smooth functions). Then there is a sequence {w q} ofcoo(sn-I)functions such
that Ilwq ;L 2 (sn-I)11 = 1, {w q} convergesweaklytozeroinL 2(Sn-I), and

11U(A)wq;L 2(sn -1)11 ~ I cI>(~,wo) I as q ~ 00.

Proof. Let v be a coo(sn-l) function with support concentrated in a small

neighborhood of ~, and with Ilv ;L 2 (sn -I )11 = 1. Denote by g a smooth function
on the sphere, extended onto IR n \ 0 as a homogeneous function of degree zero.
We will assume that v g vanishes nowhere on the support of v and that
vg(~) = Woo (Since (cpo,wo) E V(n,2) the latter is possible.) Theorem 3.5.1

implies

e -il-'g(</»U>¥~</>(A)(eil-'g(>/I)v (1/1» - cI>(CP,ILV g(cp) + ocp)v(cp) = (3.4)

= O«IILI + 1( 1)-1) as IILI + 101 ~OO,

where IL is a real number and 0 = ReA. Using this relation we can prove that for
any null sequence {(q} of positive numbers there are sequences of smooth func-
tions, {v q }, and numbers, {/Lq}, with the following properties: /Lq -,'> + 00,
Ilvq;L2(sn -1)11 = 1, the supports of the Vq contract towards the point c/Jo, and

I~~4>(A)(eil-'qg(>/I)Vq(l/I) - cI>(cp,/Lq vg(cp) + ocp)e il-'qg(4))v q(cp) I < (q.

194 Chapter 5.

Cor 0 11 a r y 3.4. The following inequality holds:

inf I IIU('\) + Til ~ max 1cI>(<p,w) I· (3.5)

TEXL 2(Sn - ) (q"W)E V(n, 2)

L e m m a 3.5. Let T = E(,\)-I'l'(<p,w)E('\), where the symbol 'l' is of class

Coo (S n - 1 X S n - I) and vanishes in a neighborhood of the manifold V (n, 2). Then
the following inequality holds for any s: IITu;Hs(sn -I )11 ,,;;;;; Cs Ilu;L 2(sn -I )11.

The proof follows from the fact that the kernel of the integral operator
v~Qv, where (Qv)(<P) = Ew--></>(,\)-I'l'(<p,w)v(w), is a function of class
c oo (sn-1 X sn-l).
Denote by oess(UU*) the essential spectrum of the operator UU*, i.e. the spec-
trum of the residue class [UU* +T] in the algebra e(L2(Sn -1 »/:J{L2(sn -1).

L e m m a 3.6. The inequality

sup Oess(U('\)U('\)*)::;;;; max 1cI> 12 (3.6)

V(n, 2)

holds, where U(,\) = E('\) - 1cI>(<p, w)E('\).

Proof. It suffices to prove that if p. > maxV(n, 2) 1 cI> 1 2 , then the operator
p.- U('\)U(A)* is Fredholm. The principal symbol of U('\)U('\r is 1cI> 12 (cf.
theorems 3.3.4 and 3.3.5). Hence the symbol p.- 1cI> 12 of p. - U(A)U('\)* is positive
in a neighborhood of V(n, 2) C Sn -1 X Sn -1. Let T be an operator with vanish-
ing symbol near V(n, 2). By lemma 3.5, T is compact. We may clearly assume that
the symbol X of ft- U('\)U(A)* +T is positive everywhere on Sn -1 X Sn -1. Now
theorem 3.3.4 and the compactness of the imbedding Hl(sn -1) C L 2(sn -1)
imply that fJT('\) = E('\) - I X(<P,w)E(,\) is a two-sided regularizer for
ft - U('\)U(A)* + T (i.e.
the product of these operators is zero, up to a compact
term). This, as is well-known (cf. [1], [17], [40]), implies that ft- U('\)U(,\)* is a
Fredholm operator. •

Cor 0 11 a r y 3.7. The following estimate holds:

inf I 11U('\)+TII ::;;;; max 1cI> I· (3.7)

TEXL 2 (Sn - ) V(n, 2)
§3. Representations of@(A) 195

Proof. Recall that

inf I IIU+TI1 2 = inf IIUU· +TII = supoess(UU*)

TE:Ki. 2(Sn ~ ) T

(cf. [24]). This and (3.6) imply (3.7) . •

Let Ujk(A) = E(A)-l<Il(j,k)(.p,w)E(A), U(A) = ~r=IIIf=IUjk' M,N < 00.

Theorems 3.3.4 and 3.3.5 imply that U(A) = E(A)-I<IlE(A)+T],

U(A)· = E(A)-1~E(A)+T2' where <II = ~jIIk<Il(j,k), T 1,T2 E su'2(sn-I). By
corollaries 3.4 and 3.7 the following relation is valid:

inf n~1 IIU(A)+TII = max I <III·

TE:Ki. 2 (S ) V(n, 2)

Thus we have:

The 0 rem 3.8. The algebra @(A)/su'2(sn-l) is isomorphic to the algebra

C(V(n, 2» of continuous functions on V(n, 2).
Denote by @(A) the union of V(n, 2) and the point A. Let A belong to an open
set, and let a fundamental system of neighborhoods of a point (.p,w) E V(n, 2)
consist of sets GJJl(.p,w) U {A}, where GJJl(.p,w) is a neighborhood of (.p,w) in V(n, 2).

The 0 rem 3.9. Every irreducible representation of the algebra @(A) is either
one-dimensional or the identity representation e(A). A one-dimensional representation
can be realized as a map 'IT(.p,w):U 1-+ <II(.p,w), where <II is the symbol of the class [Ul
in @(A)/su'2(sn-I), (.p,w) E V(n, 2). The correspondence 'IT(.p,w) 1-+ (.p,w), e(A)I-+A
~ ~

defines a bijection of the spectrum of @(A) onto @(A). The topology on @(A) coincides
with the Jacobson topology.

Proof. Let 'IT be an irreducible representation of @(A). If 'IT(su'2(sn -I» = 0,

then 'IT induces an irreducible representation of the quotient algebra
@(A)/':XL 2(sn-I). Theorem 3.8 implies that 'IT can be realized as a map
U(A) 1-+ <II(.p,w), (.p,w) E V(n, 2). Every irreducible representation 'IT such that
'IT(su'2(sn-I» =1= 0 is equivalent to the identity representation (cf. the proof of
corollary 2.2). In order to finish the proof it suffices to recall the definition of the
Jacobson topology on the spectrum of an algebra (at the end of §1.5) . •
196 Chapter 5.

§4. Representations of an algebra @(/x)

Let x E 0Jt and let @(/x) be the algebra of restrictions to the line Ix of the ele-
ments of @. The norm in @(/x) of the restriction U{x,·) to Ix of an element U E @
is defined by

Pro p 0 sit ion 4.1. The algebra @(/x) is closed, hence a C*-algebra.

Proof. Let @(/x) be the C* -algebra generated by the restrictions U(x, .),
U E @. The image of the isomorphism @ :3 U t-> U(x,·) E @(/x) is closed (propo-
sition 1.1). Hence @(/x) = @(/x) . •

Pro p s i t ion 4.2. Let >'l,A2 E lx, Al =1= A2. The representations
0

'1T(X,Aj):U(X,)t-> U(X,A), j = 1,2, oj@(Ax) in L 2 (S(0Jt)x) are not eqUivalent.

Proof. Let a be a function defined and smooth on 0Jt \ x, and having a finite
limit f3(tp) when approaching x in the direction tp, f3 E C OO (S(0Jt)x). Clearly, the
operator of multiplication by a belongs to @. Hence we may assume that the
operators of multiplication by tpj' where tpj are the coordinates of the vector
tp E S (0Jt)x (in a fixed local system), belong to @(/x).
Let ~ EC OO (S(0Jt)x X S*(0Jt)x) and U(A) = E(A)-I~(<p,"')E{A). Obviously,
U E @(/x). The formula

U(A)"'j = tpj U(A + i) + E(A) - I D "'I ~(tp, '")E(A + i) (4.1)

holds (d. proposition 3.2.1). Replace in (4.1) ~(tp,,,,) by tpw~(tp,,,,). Taking into
account (1.4.1), which defines E(A)-I, and the relation
(tpw+iOYJl.tpw = (tpw+ioiCp-i), we find that U(A+i) =
= E(A + i)-ltpw~(tp,,,,)E(A + i) = (A + inI2)E(A)-I~(tp,"')E(A + i). Now (4.1) takes
the form

E"'-4t/>(A)-ltpw~(<p'''')~-4W'''j = (4.2)

= tpj(inl2 +A)E"'-4<j>(A)-I~(tp'''')~-4",(A + i) +

+ E"'-4<j> (A) -I D WI (tpw~(<P''''»~-4w(A + i).

§4. Representations of ®(/x) 197

Definition (1.2.1), defining E(A), and the relation

(- w1/J+ i 0) - il'WI/I = - (- WI/I+ iO) -i(p.+i) imply that

E(A) -I tI>(cp, w)E(A)WI/I = (4.3)

= -i(iA-I +nl2)E(A)-ItI>(<t>,w)E(A+i).

Multiply (4.3) by <t>j and subtract the result from (4.2). Then

E(A)-Iq,wtI>(<t>,w)E(ANj -<t>jE(A)-ItI>(<t>,w)E(A)WI/I = (4.4)

= E(A) - 1 '1'(<t>, w)E(A + i),

where i'(cp,w) = i (n -1)cpjtI>(cp,w)+ D w/q,wtI>(cp,w». It is obvious that the lefthand

sides of (4.3) and (4.4) are elements of the algebra ®(lx). Replace in (4.3) the
function tI> by '1'. We find that the operator-functions
A~ (iA -1 + n 12)E(A) -li'(<t>,w)E(A +;), A~ E(A) -li'(<t>,w)E(A + i) belong to O>(lx).
Denote these functions by ~ and 91.
Suppose that for some distinct AhA2 the representations 7T(X, AI) and 7T(x,A2)
are equivalent. Then the norms of both the operators ~(AI) and ~(A2) and those
of 91(A I) and 91(A2) must coincide. If AI =1= - A2 this is impossible, since
IiAI -1-nI21 =1= IiA2 -l-n 121. If however AI = -1.2, then the values of the
operator-function ~(A)- (il.l - 1+ n 12)91(A) = i(A -A)91(A) have distinct norms
at A) and 1.2 • •
Let ~(lx) = ~(lx,L2(S(~)x» be the algebra of continuous functions on lx,
with values in %L2(S(~)x), which tend (in norm) to zero at infinity. In 5to(lx) we
introduce the norm

L e m m a 4.3. The following inclusion holds: 5to(lx) C ®(lx).

Proof. Since representations U(x,.) ~ U(x, Ad and U(x,.) ~ U(x, 1.2 ) of ®(lx) are
irreducible, and inequivalent for distinct Aj,A2 (propositions 3.1 and 4.2), their
restrictions to ~(lx) n O>(lx) are also irreducible and inequivalent (proposition
1.7). Therefore 5to(lx) n ®(lx) is a rich sub algebra of ~(lx). This and proposition
1.9 imply ~(lx) n O>(lx) = ~{{)(lx) . •
198 Chapter 5.

Pro p 0 sit ion 4.4. The quotient algebra @(lx)/St'D(/x) and the algebra
C(S(~)x X S*(~)x) are isomorphic.

Proof. Put U(x,A) == ~jIIkE(A)-IcI>jkE(A) (the subscripts j,k run over finite
sets). By theorem 3.3.4, U(x,A) == E(A)-I~jIIkcI>jk(CP,w)E(A)+T(A), where
T E :JtD(/x)' This and corollary 3.4 imply

inf supIIU(x, A) + T(A) II ~

TE'Xo(l.Y) Ad'y

The reverse inequality is contained in theorem 3.6.5. Thus,

== 117I}cI>jk;C(S(~)x X S*(~)x)ll·
From this we find that the map ~O . . . i", where ~(A) == E(A) - I i"(cp, w)E(A),
defines an isomorphism between the quotient algebra @(/x)/'Xo(lx) and the alge-
bra C(S(~)x X S*(~)x) of continuous functions . •

Cor 0 1 1 a r y 4.5. Let 7T be an irreducible representation of @(lx) such that

7T(Xo(1x» == 0. Then 7T is equivalent to a representation of the form
U(x,) ...... cI>(<<Po,wo), where cI> is the symbol of U (i.e. the symbol of the corresponding
class in @(lx)/%:!(lx») and «Po E S(~)x, Wo E S*(~)x'

Denote by @(/x) the union of the product S(~)x X S*(~)x and the line
Ix == R. We introduce a topology on ®(lx). If (CP,w) E S(~)x X S*('J1L)x, cpw ~ 0,
then a fundamental system of neighborhoods of (CP, w) consists of sets
CV(CP,w) U {AElx:A~N}, where CV(CP,w) is a neighborhood of (CP,w) in
S(~)x X S*(~)x, N E R. In case cpw == 0 a fundamental system of neighbor-
hoods is formed by the sets "\~cP,w) U Ix. A neighborhood of A E Ix is an interval
on Ix.

The 0 rem 4.6. Let U(x;) be an element of the algebra @(/x) and let
cI> E C(S('J1L)x XS*('J1L)x) be its symbol. Then:
§4. Representations of ®(ix) 199

l)the maps '/T(</>,,,,):U(x,)I-+Ii'(</>,"') and '/T(x,A):U(x,') 1-+ U(x,A) are pairwise ine-
quivalent irreducible representations of ®(ix);
2) every irreducible representation of ®(lx) is equivalent to one of the representations
in 1);
3) the correspondence ";.(</>, "') 1-+ (</>, "'), '/T(x, A) ': A E Ix defines a bijection of the spec-
trum of®(/x) onto ®(/x). The topology on ®(/x) coincides with the Jacobson topol-
ogy.

The proof of this theorem is preceded by two lemmas.

L e m m a 4.7. Let

= 1 e -A'lT1 2 f(n 12 - ill.)

(2'/T)n12
f
S.-I
(<P"-')i~-nI2v (",)d""

where d", is the volume element on the unit sphere. On the line ImA = 0 we have the
estimates

11&±(A);L2(sn -I) ~L2(sn -1)11....; const,

11&±(A);L2(sn-I)~L2(sn-I)11 = O(e-'lTIAI) as A~+oo.

Proof. Expand a function v in a series in spherical harmonics:

co km
V (</» = l: l: Vmk Y mk(</».
m =0 k =1

Consider, e.g., the operator &+. Using a reasoning similar to the proof of proposi-
tion 1.3.1, we are led to

&+(A)v = l:l'm(A)VmkYmk>
m,k

where

(A) =c e A'lT12 f(nI2-iA+m)f«l+iA-m -nI2)/2)

P-m n 2mf«nI2+iA+m)/2) '
200 Chapter 5.

while, here and below, the Cn denote constants whose absolute values depend on
the dimension of the space only.
In view of the formula r(l- z )r(z) = 'IT 1 sin 'lTZ, for
z = (-iA+m+1+nI2)/2wehave

r«(l +iA-m -nI2)/2) =

r(l-(nI2-iA+m + J)/2)r«nI2-iA+m + 1)/2)

T«nI2-iA+m + 1)/2)

T«nI2-iA +m + 1)/2)sin {'IT(nI2-iA+m + 1)/2} .

Therefore

x __________________~r~(cn~/~2_-~iA~+~m~)____________________
2mr«n 12- iA + m + 1) 1 2)T«n 12+ ill. + m)!2)sin {77{n!2 - ill. + m + 1)/2}
By the duplication formula T(2z) = 22z -I 'IT -1/2 T(z )T(z + 112), for
z = (- ill. +m +nI2)/2 we obtain
r(nI2- ill. +m)
r«nI2-iA+ m)/2+ 1/2)
= 2nl2-iHm y';T«n!2-iA+m)/2).

Thus,

x r«nI2-iA+m)!2)
r{{nI2+iA+m)!2)sin{'IT{nI2-iA+m + 1)/2}
In order to derive from this the estimate for the norms of the operator 0+ (A) it
remains to remark that for ImA = 0 the modulus of a quotient of r-functions is
one. •

L e m m a 4.8. Let U(A) = E(A) -lcp(tp,w)E0.), where CP(tp,w) = 0 for q,w,,;;;;;; 0

(respectively, q,w;;,O). Then IIU(A)II = O(e-wl>"l) as A--i>-oo (respectively,
A --i> + 00) on the line 1m A = O.
§4. Representations oj@(/x) 201

Proof. Clearly, U(A) can be written as U(A) = 0+ (A)«PE(A) (respectively,

U(A) = 0_ (A)«PE(A». Now it suffices to expand the symbol «P in a series:

«P(CP,w) = ~ amk(cp) Y mk(W)

m,k

and use the preceding lemma (compare with the proof of proposition 1.3). •

Proof 0 f the 0 rem 4.6. I) The truth of this assertion follows from pro-
positions 3.1 and 4.2.
2) Let 'TT be an irreducible representation of @(/x) such that 'TT('Xo(/x» = o. By
Corollary 4.5, 'TT is equivalent to a representation of the form U(x,') I-> «P(CP,w).
If 'TT('Xo(lx) =1= 0, then the restriction 'TT I~(lx) is an irreducible representation of
'Xo(/x) (proposition 1.2). As is well known (proposition 1.8), every irreducible
representation of 'Xo(lx) is equivalent to a representation Ti-> T(A), where
T E 'Xo(/x), A E Ix. In view of proposition 1.7 this means that the correspond-
ing representation 'TT of @(/x) is equivalent to a representation U(x,·) I-> U(x, A).
3) It suffices to verify that the topology on @(/x) coincides with the Jacobson
topology, since the first part of the assertion follows from 1) and 2).
By §1.5 a base of the Jacobson topology on the spectrum of an algebra @(/x)
is formed by the set of (equivalence classes of) representations 'TT for which
11'TT(U(x, '»11 > 1, where U(x, .) runs over a set that is dense in @(/x)'

Let Ao be an arbitrary point of Ix and let U(x,') be an element of @(/x) for

which IIU(x, Ao)11 > 1. The set {'TT: II'TTU(x, ')11 > 1} is a neighborhood of Ao in the
Jacobson topology. Since the operator-function AI-> U(x, A) is norm-continuous,
this set is also a neighborhood of Ao in the usual topology on the real line. In view
of the fact that an operator-function AI->C(A)(-,f)g, j,g E L2(S(~)x), where C is
an arbitrary continuous function on the real line that vanishes at infinity, belongs
to @(/x), every open interval on the real line is open also in the Jacobson topol-
ogy.
Now we will consider neighborhoods of a point (<I'o,wo) E S(~)x X S*(~)x,
CPowo > O. Let «P(CPo,wo) > I and U(X,A) = E(A) -l«p(CP,w)E(A). Since
'TT(CP,w)U = «P(CP,w), the inequality 11'TT(cp,w)UII > I holds for all (CP,w) in a neighbor-
hood of (<I'o,wo) on S(~)x X S*(~)x' By (3.4) the inequality II'TT(X,A)UII > I holds
for all A> N, with N a large number. Hence a neighborhood of (CPo,wo) in the
202 Chapter 5.

Jacobson topology is also neighborhood in the topology on @(lx).

Let '\I(</>o,wo) U {AElx:A>N} be an arbitrary neighborhood of (</>o,wo) in
@(/x). We show that it is a neighborhood of this point in the Jacobson topology.
Choose a function cI> E C OO (S(0R)x X S*(0R)x) such that cI>(</>o,wo) > land
cI> = 0 outside '\~</>o,wo) U {(CP,w):CPW>O}. Put U(A) = E(A)-IcI>E(A). Then
{(cp,w): 11'7T(CP,w)UII > I} c "'~</>o,wo). Now note that by lemma 4.8, '7T(x, A)U ~ 0 as
A~ - 00. Moreover, lemma 3.5 implies that U(x, A) is, for every A, a compact
operator. Hence, by lemma 4.3, an operator-function A1-+ C(A)U(X, A), where
C E Co (Ix), belongs to @(/x)'
Choose the function e such that 0.;;;; e';;;; 1 and e(A) = 1 for A E [ - M,M],
where M is a large number. Replacing, if necessary, U by U - e U, we find an
operator-function for which {'7T: 11'7T(U)II>1} c ("\~</>o,wo) U {AElx: A>N}.
So, it has been proved that the systems of neighborhoods of a point (CPo,wo),
A

CPowo > 0, in the topology of @(/x) and in the Jacobson topology coincide. This
assertion is similarly proved for points (</>o,wo) satisfying <Powo < O.
Let </>owo = 0, cI> E C OO (S(0R)x X S*(0R)x), cI>(</>o,wo) > 1, and
U(A) = E(A)-IcI>E(A). In (3.4), g is understood to be an arbitrary homogeneous
function of degree zero. Choose g such that the equation Wo = Y'g(CPo) is fulfilled.
Letting p. tend to infinity leads to the inequality IIU(A)II = 11'7T(X,A)UII > 1 for all
A E Ix. Hence a neighborhood of (CPo,wo) in the Jacobson topology is a neighbor-
hood also in the topology of @(/x). The converse assertion is obvious. •

§5. Proof of theorem 2.1

Recall that @ denotes the algebra generated by the operator symbols defined just
above theorem 4.4.1.
Let U be an arbitrary element of @, and let {Un} be an approximating
sequence for U every term of which is a sum of finite product of generators:

Un(X,A) = ~IIE(A)-IcI»Z)E(A),
j k

where cI>jZ)(x,') E C OO (S(0R)x X S*(~M. For all x E 0R., except for a finite set,
the maps (CP,w) 1-+ cI>jZ) (x, CP,w), (CP,w) E s(,!m.)x X S*(~)x, do not depend on cpo
For such x,
§5. Proof of theorem 2.1 203

Un(X,A) = E(A)-l~IIIf»k)(x, ·)E(A)

j k

and

(5.1)

where If>n = ~jIIklf»k)' We will denote :he space S(~)$S-em.) ~th the topol-
ogy ':r(S,S-) induced by the topology on @ (cf. §2) by IJR, while C(IJR) will denote
the corresponding space of continuous functions. Formula (5.1) implies that
Illf>n;C(IJR)11 ~ IIUn;@II. Hence also for a limit element U E @ there is defined its
symbol If> E C (IJR). It is obvious that to the product of two elements U, IS in @
corresponds the product If>i' of their symbols.

Pro p 0 sit ion 5.1. Let U be an element of @, with corresponding symbol

If> E C (IJR). The correspondence '71{x, cp, 0): U 1-+ If>(x, CP,O), cP E S (lJR)x, 0 E S - ('::m.)x'
defines a one-dimensional representation of the algebra @, while the correspondence
'/T(x, A):U 1-+ U(x, A), x E ~ A E Ix> defines an irreducible representation of @ in
L 2 (S (~)x)' These representations are pairwise inequivalent.

Proof. The fact that '/T(x, cp,O) is a one-dimensional representation of @ follows

from the assertions made before stating the proposition, while the fact that '/T(x, A)
is a representation immediately follows from the definition of ®. Irreducibility of
'/T(x, A) is verified using proposition 3.1, while the proof of the inequivalence of dis-
tinct representations reduces to an application of proposition 4.2 . •
Our nearest goal is to prove that every irreducible representation of @ is
equivalent to a representation listed in proposition 5.1. It is reached in proposition
5.5, which is preceded by some lemmas.
We introduce the function

where U E ®, x E ~.

L e m m a 5.2. The function 9Lu is upper semicontinuous.

204 Chapter 5.

Proof. First suppose that U can be represented as

U(X,A) = ~IIE(A)-I«Pjk(x, ·)E0), (5.2)

j k

where «Pjk(X,·) E COO(S(WC)x X S*(')JL)x), «Pjk E WC, while the subscripts j,k run
over finite sets. Fix an arbitrary point x. By theorem 3.3.4, the operator U(x, A)
can be rewritten as

U(x, A) = E(A) -I ~II«Pjk(X' ·)E0) + T0), (5.3)

j k

moreover, IIT(A);L2(S(')JL)x)~L2(S(')JL)x)11 = O(IAI-I) as IAI ~ -00. In

order to estimate the first term on the righthand side in (5.3) we use (3.4). Note
that in (3.4) for vg(CP) we may take any vector orthogonal to cp, while the
numbers fL and 0 may be chosen arbitrarily. Hence we may assume that
IfL I + I01 ~ 00 and that the vector fLvg(CP) + ocp remains on a given ray. This and
(3.4) imply

~u(x) ;:.117IJ«Pjk(X, .); C(S (')JL)x X S*(')JL)x)ll·

Recall that functions from WC depend everywhere, with the exception of a set of
singular points x E 0lL (which depends on the function under consideration) only
on the cotangent vector w. If x is not singular for any function «Pjk, then

~IIE(A) -1«pjk(X, ·)E(A) = E0)-1 ~II«Pjk(x, -)E0),

j k j k

and, since E0) is unitary for any A E Ix = IR,

~u(x) = IIE(A)-1 7 IJ«Pjk(X, .)E(A)II =

= 7
11 IJ«P jk (x, .); C(S * (0lL)x) 1I·

If the nonsingular point x tends to a point Xo that is singular for ~jIIk«Pjk along
a direction cp E S (0lL)xo' then, uniformly in cp,

lim IIE(A)-I
X-HO
~II«Pjk(x, .)E0)11 =
j k
§5. Proof of theorem 2.1 205

So, the function iJLu is continuous at every point x that is nonsingular for the
symbol ~jIIk(ljk' while at singular points the relation limx->xo iJLu(x).;::;; 'VLtt(xo)
holds. Thus, the lemma has been proved for the operators (5.2). In order to com-
plete the proof it remains to use the denseness of the set of operators (5.2) in the
algebra @. •

Let I be a closed ideal in @. Denote by Ix, x E 01L, the set of operator-

functions on Ix defined by Ix = {U(x, .): UEI}.

L e m m a 5.3. U E I if and only ifU(x,·) E Ix for all x E ~

Proof. It must be verified that U(x,·) E Ix for all x implies U E I.

For any point Xo E ~ there is a function ~ E I such that U(xo,) = ~(xo').
Then, by lemma 5.2, the inequality

holds in a sufficiently small neighborhood of Xo. Hence, for an arbitrary f > 0

there are a finite cover {Uj } of ~ and a set of operator-functions {~j}' ~j E I,
satisfying SUPhElx IIU(x, A) - ~iX,A)11 < f for x E Uj . Denote by {7Jj} a partition
of unity subordinate to the cover {~}, and put ~(x,·) = ~j7Jix)~ix, .). Clearly
~ E I. We have

f~fIIU(x, A) - ~(x, A);L2(S (~)J ~ L 2(S (~)x)11 ~

~ ~7Jiix)supllU(x,A)-~ix,A);L2(S(~)x)~L2(S(~)x)11 < (.
j hE1x

Hence IIU-~;®II < f. The assertion of the lemma now follows from the fact
that I is a closed ideal. •
Denote, as before, by ®(/x) the algebra of restrictions {U(x, -): U E®} to Ix of
the elements of ®, and by Y the set of points x E ~ for which Ix =F ®(lx).

L e m m a 5.4. If I is a primitive ideal, then Y consists of a single point only.

206 Chapter 5.

Proof. Assume the opposite holds. Let Xl,X2 be two distinct points in Yand
let U I, U 2 be disjoint neighborhoods of them on '!)]t. Denote by ~J{i the closed
two-sided ideal in ® formed by the functions which vanish outside U;, i = 1,2.
Clearly, :JCI:JC2 = °
c I. Then, by proposition 1.3, at least one of 'J{j is contained
in I. If 'J{I C I, then (:JCI)x, C I x I ' which is impossible since, on the one hand,
('J{I)xl = ®(lx,), and on the other hand XI E Y . •
Let '!T be an irreducible representation of ® whose kernel is I, and let X 0 E '!)]t
be a point for which Ixo *-
®(/xo) (by lemma 5.4 there is at least one point with
this property). If U, IB are elements of ® such that U(x 0,·) = lB(x 0, .), then
'!T(U) = '!T(IB), by lemmas 5.3 and 5.4. This allows us to consider '!T as a representa-
tion of the algebra ®(/xo) of restrictions to lxo of the elements of ®. This and
theorem 4.6 imply

Pro p 0 sit ion 5.5. Every irreducible representation of the algebra 0) is

equivalent to a representation listed in proposition 5.l.
By theorem 4.4.1 the algebras tU'J{L2('!)]t) and ® are isometrically isomorphic.
Hence the first assertion of theorem 2.1 is that of proposition 5.1, and the second
assertion that of proposition 5.5. Thus, it remains to prove the third assertion of
the theorem only.
The description of the representations implies that the set of residue classes of
equivalent irreducible representations can be parametrized by the points of 0). It
is obvious that inequivalent representations have distinct kernels. Hence the spec-
trum of the algebra 6U:JCL 2('!)]t) can be identified with the set of primitive ideals of
this algebra. We verify that the topology on ® coincides with the Jacobson topol-
ogy. For this we need

Pro p 0 sit ion 5.6. Let U be an operatorJunction defined by

O,
{
U(x,'\) = P(A) , (5.4)

where x E '!)]t and P is a continuous function with compact range, given on lxo and
tending to zero at infinity (in the operator norm in L 2(S ('!)]t)xo)j. Then U E 0).

Proof. By lemma 4.3, 'J{{)(lxo) C 0)(lxo)' hence there is a ~ E 0) coinciding on

§5. Proof of theorem 2.1 207

lxo with P. Let {rn} be a sequence of C OO ('!J1t) functions, 0:0;;;; rn :0;;;; 1, rn(XO) = 1,
with, moreover, the support of rn contracting towards Xo. We will show that the
sequence {rn'3'} converges in ®, and that limrn'3' = U. Let
~(X,A) = ~jIIkE(A)-l()jk(X, ·)E(A), where ()jk E IDe, be a combination of genera-
tors of ® such that 11'3'-~;®11 <f. The operator '3', as does every element of ®,
has a symbol, ()<:J> E C(IDe). Clearly, ()<:J>(xo,·) = 0, since in the opposite case we
cannot have P E 9\o(lxo). This and the inequality
« «,
A

imply
11()<:J>-~jIIk()jk;C(IDe)II:O;;;; 11'3'-~;®11 l~jIIk()jk(X,·)1 which holds
for all x in a neighborhood U of x o. We have

IIrn'3'-rm'3';®II:O;;;; (5.5)

:0;;;; IIrm('3'-~);®11 + lI(rm -rn)~;®11 + IIrn(~-'3');®11 <

< 2(+ lI(rm -rn)~;®II.
We may assume that supprn u supprm C U and that the ()jk have in U no
points of discontinuity other than x o. Then,

II(rm-rn)~;®11 = sWI(rn-rm)7I}()jkl <f.

Together with (5.5) this leads to IIrn'3'-rm'3';®11 < 3(, for sufficiently large m,n.
Thus, {rn'3'} is a Cauchy sequence in ®. The relation limrn'3' = U clearly
holds. •
We return to the description of the Jacobson topology on the spectrum of the
algebra cU:J{1d'!J1t). Recall that a base for this topology is formed by the sets
{"7T: 11"7T(~)11 > I}, where ~ runs over some dense subset of ®. We must prove that
the topology of ®, described in §2, coincides with the Jacobson topology.
Ao
A

Let be an arbitrary point of lxo. In ®, a fundamental system of neighbor-

hoods of this points consists of the open intervals on lxo containing Ao. The con-
tinuity of the functions A1-+ U(XO,A), U E ®, implies that every neighborhood of
Ao in the Jacobson topology is a neighborhood in the topology of ® also. Proposi-
tion 5.6 means that, in particular, every open interval on lxo containing Ao is a
neighborhood in the Jacobson topology also.
Now consider neighborhoods of a point (X 0, <Po,(}o), Xo E '!)R., <Po E S('!J1t)xo'
()o E S* ('!J1t)xo. Let U E ® and let
208 Chapter 5.

U(X,A) = ~IIE(A)-1cfljk(x,.)E(A),
j k

where cfljk E ~. Put cfl = };j1hcfljk; we will assume that I cfl(xo,</>o,Oo) I > 1.
We verify that every neighborhood of (xo,</>o,Oo) in the Jacobson topology is a
neighborhood of this point in the space @. As has been shown in the proof of 3)
of theorem 4.6, a set {AElxo: IIU(xO,A)11 > I} either coincides with Ixo (if
<PaOo = 0) or coincides with an interval {AElxo: A ~ N} (if <PaBo ~ 0).
The set {(xo,</>,O): I cfl(xo,</>,O) I > I} contains a product V(</>o) X G[Jf, where
V(</>o), GlJf are neighborhoods of </>0, 00 on the spheres S(~)xo' S*(~)xo' respec-
tively.
The part of the bundle S(~)$S*(~) above a neighborhood U of a point Xo
can be represented as the product U X Sn -1 X Sn -1. Since cfl is a continuous
function in the topology '!r(S,S*), the inclusion
{(x, </>,0): Icfl(x,</>,O) I > I} ::> {(x, </>,0): xEK(</>o), </> E sn-l, 0 E G[Jf(00)} holds.
Here K(</>o) is a subset of U whose image under some coordinate map IC: U ~ IR n is
the intersection of an open cone, with vertex at K(XO) and containing all directions
from IC'(XO)(V(xo», and an open n-dimensional ball with center at IC(XO)'
Finally, if x E K(</>o), 0 E G[Jf(00), we can choose </> E Sn -1 such that <PO = o.
Since I cfl(x,</>,O) I > 1, the inequality IIU(X,A);L2(S(~)x)~L2(S(~)x)11 > 1
holds on all of Ix (cf. the proof of 3) the theorem 4.6).
All this implies that a neighborhood {w: Ilw(U)II> I} of a point (xo,</>o,Oo) in
the Jacobson topology contains a neighborhood of this point in the topology of @.
~

We will now show that every neighborhood ~ of a point (xo,</>o,Bo) in @ con-

tains a neighborhood of this point in the Jacobson topology.
We will assume that ~ is a neighborhood from the fundamental system, and
that in case </>000 ~ 0 the intersection ~ n (S(~)$S*(~»xo belongs to the set
<PO ~ O.
The intersection ~ n
(S(~)$S*(~» is a neighborhood of
(x 0,</>0, 0o) E S(~)$S*(~) in the topology '!r(S,S*). There clearly exists a func-
tion cfl E C (WC) such that
(xo,</>o,Oo) E {(x, </>,0): Icfl(x,</>,O) I > I} C ~ n (S(~)$S'(~» and cfl = 0 out-
side ~ n (S(~)$S*(~». If <PoOo *- 0, we will assume that also cfl(xo,</>,O) = 0
for <PO = O. Since WC is dense in C(WC), we may assume that a function cfl with
§5. Proof of theorem 2.1 209

such properties exists in IDl. Put ll(x,X) = E(X) -1 cfI(x, ·)E(X). In view of the ine-
quality II'/T(x, </>, O)U II = IcfI(x, </>,0) I an element (x, </>,8) which belongs to a Jacob-
son neighborhood of (xo,</>0,80) belongs to ~ also.
If </>080 = 0, then ~ contains the whole line lxo and the lines lx,
x E p(~) n '!)TL (p: S ('!)TL)EB S" (~1L) ~ '!)TL is projection). Clearly, in this case ~ con-
tains also a neighborhood {'/T: II'/T(ll) II > l} of (xo,</>0,80) in the Jacobson topology.
If, however, </>080 ~ 0, a part {X Elxo:X ~ N} of lxo belongs to ~. By lemma 4.8,
IIU(xo,X);L 2(S ('!)TL)x) ~ L 2(S ('!)TL)x) II ~ 0 as X~ -+-00. Therefore, by adding to II
the operator (5.4) from proposition 5.6 (with a suitable P) we obtain that the ine-
quality IIU(xO,A)11 < 1 holds on the part of lxo not belonging to~ . •

Together with the description in theorem 2.1 of the irreducible representations,

the following theorem gives a criterion in order that an arbitrary operator from te
be Fredholm.

The 0 rem 5.7. Let te and ® be the same algebras as in theorem 2.1. An opera-
tor A E te is Fredholm if and only if for every irreducible representation '/T of ® the
operator '/T(A) has a bounded inverse on the representation space H'IT'

Proof. An operator A is Fredholm if and only if its residue class [AJ is inverti-
ble in the Calkin algebra e(L 2('!)TL)/:JUd'!)TL). (Here e(L 2('!)TL» is the algebra of
bounded operators on L2('!)TL).) The element [AJ is invertible simultaneously in the
algebra e(L 2('!)TL»IXL 2('!)TL) and in the subalgebra cUXL 2 ('!)TL) ~ ® (proposition
1.11). It remains to use proposition 1.10. •

§6. Ideals in the algebra of pseudodifferential operators with discontinuous symbols

Every closed two-sided ideal in a C" -algebra is the intersection of the primitive
ideals containing it. This and the theorems in §2, containing a description of the
spectra of algebras of pseudodifferential operators, allow us to clarify the structure
of all closed two-sided ideals in such algebras. In this paragraph we consider cer-
tain ideals of the algebra te from theorem 2.1.
Let e be the disjoint union of straight lines lx, enumerated by the points of a
manifo~d '!)TL. On e we introduce the topology induced by the topology of the
space ® (the spectrum of the algebra ® of operator symbols ® ~ teIXL2('!JlL»).
210 Chapter 5.

Clearly, e becomes a locally compact space, and every line Ix (with the ordinary
topology) is an open subset of e.
Let Qx be a compact subset of lx, and let Q = U xQx, where x runs over a
finite point set in 0lL. Sets e \ Q form a fundamental system of neighborhoods of
the point at infinity, 00, in e. The point 00 clearly does not have a countable fun-
damental system of neighborhoods, hence e is not metrizable.
We introduce the algebra :Jfo(e) of operator-valued functions
At-+ U(x, A) E :J{L2(S (0lL)x), A E lx, given on e, continuous (in the operator norm),
and tending to zero at infinity.

Pro p 0 sit ion 6.1. Let U E ~(f), and let f be an arbitrary positive number.
Then everywhere on e, except possibly on a finite set of lines, the following inequality
holds:

(6.1)

Proof. Since U tends to zero at infinity, (6.1) holds in a neighborhood U of 00.

The complement e\ U can intersect at most finitely many lines Ix . •

Pro p 0 sit ion 6.2. 'Ji{)(f) is a closed two-sided ideal of the algebra @ of opera-
tor symbols.

Proof. Let U be an arbitrary element of 'Jfo(e). It suffices to prove that U is the

limit of a sequence of elements of @ (in the norm of @). Denote by {fn } a
decreasing null sequence of positive numbers. Proposition 6.1 implies that there is
an at most finite set ~ of lines Ix on which (6.1) does not hold. Introduce the
operator-valued function Un equal to U on ~ and equal to zero everywhere out-
side ~. By proposition 5.6, Un E @. It is clear that U is the limit of the sequence
{Un}· •

Pro p 0 sit ion 6.3. The following isomorphism holds:

A

@/:Jfo(e) ~ C(UR).

Proof. Let jx:U ~ U(x,') be restriction to Ix of an element U E @. Define a

map h:@/Xo(e)~C(UR) by h([U])(m) = ?T(m)ojAU), where [U) is the residue
class of U E @ in @/:JfoO::), m E (S(0lL)$S*(0lL»x, and ?T(m) is a one-
§6. Ideals 211

dimensional representation of ®(/x) (the definition is correct, which follows from

theorem 4.6; cf. also proposition 4.4). We will show that h is an isomorphism.
We verify that h is monomorphic. If h([UJ) = 0, then, as follows from theorem
4.6, U(x, -) E Xo(lx) for all x E 0lL. It remains to convince ourselves of the fact
that U tends to zero when passing to 00 along e. Let cI>jk E 1))(,

Ujk(X,A) = E(A)-IcI>jk(X,cp,8)E(A), cp E S(0lL)x, () E S*(0lL)x, and

IIU-7lJUjk;®11 <£/2, (6.2)

where j,k run over finite sets. (The operator symbols of the form Ujk ~enerate the
algebra ®.) Then for anyone-dimensional representation 7T(m), mEl))(, of ® the
following inequality holds:

(6.3)

The symbols cI>jk have at most finitely many points of discontinuity. At every
point x E 0lL at which all cI>jk are continuous,

~~r 117lJUjk(X,A);L2(S(0lL)X) ~L2(S(0lL)x)11

= max I~IIcI>jk(x,cp,()I.
<p,e j k

Taking into account (6.2) and (6.3), we thus obtain that SUPAEI, IIU(x, A)II <
°
£

everywhere except possibly on a finite set of lines Ix. Thus, for any £ > there is
in e a neighborhood of 00 in which (6.1) holds. Therefore U E :J{{j(e), and h is a
monomorphism.
If cI> E I))( and U(x, A) = E(A) -1cI>(x, cp,8)E(A), then h([U]) = cI>. The set I))(
generates C (I))(). This and the fact that h is isometric gives that h is an epimor-
phism . •
Denote by com 1.8 the closed two-sided ideal of an algebra 1.8 that is generated
by the commutators of the elements of 1.8.

Cor 0 1 1 a r y 6.4. com ® = Xo(e).

Proof. The ideal com @ is the intersection of the kernels of all irreducible
one-dimensional representations of @. The algebra :J{{)(e) does not have nonnull
212 Chapter 5.

one-dimensional representations. Hence :Ko(e) c com@. Since @/:Ko(e) is a com-

mutative algebra, the opposite inclusion com @ C :Ko(e) holds . •
Let p:ti!~cU%L2('JJt) be projection and let i:ti!1%L 2 ('JJt) ~ @ be the isomor-
phism mapping a pseudodifferential operator A to its operator symbol U.

Proof. Anyone-dimensional representation of ti! can be regarded as a represen-

tation of the quotient algebra ti!1% ~ @. Since :Ko(e) belongs to the kernel of all
one-dimensional representations, p-1oi-1(:Ko(e)) C comti!. We verify the oppo-
site inclusion. If A E comti!, then 7T(m)A = 0 for all m E WC. This implies (as
has been shown in the proof of proposition 6.3) that iop(A) E :Ko(e) . •
All results stated in this paragraph can be combined into the following commu-
tative diagram
0 0
~ t
0 ~ %L 2('JJt) ~ comti! ~ :Ko(e) ~ 0

O~ %L 2('JJt)
~ti!
~ ~ @
t
~ 0

~ t
C(WC)

t'" 0
0
All straight paths ('without comers') in this diagram are made up by exact
sequences.
Finally we note that all values of an operator symbol U E @ are compact
operators if and only if U belongs to the kernel of every representation 7T(m),
where m is a point of siR of the form (x,cp,O), cp E S('JJt)x, 0 E S*('JJt)x, and
cpO = O.
§7. Spectra. Manifold with conical points 213

§7. Spectra of C* -algebras of pseudodi8"erential operators on a manifold with coni-

cal points
In this paragraph we study the spectra of algebras generated by pseudodifferential
operators with discontinuities of the first kind in the symbols, on manifolds with
conical points. The operators act on the weighted normed spaces H .B{~) (cf.

°
§4.4). In particular, we consider the case when at a nonsingular point x(O) E ~
the weight takes the value or 00, while the symbols of the operators remain
smooth. It turns out that also in this case infinite-dimensional representations
appear - the functions space 'is to blame' - but they are 'twice as small' in com-
A

parison with the discontinuous situation (instead of the line Ix(O) the space ® con-
tains a half-line, cf. theorem 2.1).
The presence of conical points does not lead to new substantial problems.
Therefore we consider in detail only the above-mentioned effect of the appearance
of infinite-dimensional representations because of singularities of the weight. (In
particular, we consider the local algebras ®c(A) for complex A on the line
{AEC: ImA=,8o}, where,8o is the exponent of the weight at x (0).)
We begin with studying the local algebras. Then we give a description of the
spectrum of the algebra of pseudodifferential operators on a manifold with conical
points. Finally we investigate the one-dimensional case: singular operators on a
composite contour.

1. The algebra ®c{l\.). This algebra is generated by the operators

U.y-->4>(A) = EW-->4>{A)-1 (>(w)F~IHW(A) on L 2 {Sn-I); here, A is a complex number,
A =1= -+-i(k+nI2), k = 0,1,···, and (> E coo{sn-I). The operator U(A)*
adjoint to U(A) has the form U(A)"" = E(~)-l(>(w)~). If A is real, the algebra
®c(A) is, clearly, commutative. Moreover, for real A the operator E(A) is unitary,
hence IIU(A);L 2(sn -I) ~ L 2(sn -I )11 = II(>;C (sn -I )11. Thus, in this case ®c(A)
is isomorphic to the ring C (S n - 1).

Pro p 0 sit ion 7.1. IfImA =1= 0, then ®c(A) is an irreducible algebra.
The proof proceeds in several steps.
I. We show that every nontrivial invariant subspace Xc L 2{sn-l) for ®c{A)
contains a Coo (Sn - 1) function f that is not identically equal to zero.
The operator K = i(F.(A)-l(>(w)E(A)-E(~)-l(>(w)E(~», where (> is a real-
214 Chapter 5.

valued function, is selfadjoint and belongs to @Se(A). By proposition 3.2.2 its order
does not exceed - 1. Denote by JL an arbitrary nonzero eigenvalue of K and by
5<;. the corresponding eigenspace. The projection operator p:L 2(sn -I) ~ 5<;. is an
element of @e(>"). Since K is a smoothing operator, all functions from ~ belong
to coo(sn-I). Since PSC ex!' n X, any nonzero element of PSCcan be taken as
the function f In case PSC = 0 the subspace X must be replaced by its orthogonal
complement Xl.., which is also invariant under @e(>").
II. In this step we show that there is a function U E coo(sn -I) in X for which
(E(>..)u)(c/lo) =1= 0, where c/lo is a given point on the sphere.
Letf EX n coo(sn-I),f;;s O. Put u = E(~)-l4l(w)E(~)f, where the smooth
function 4l is chosen to satisfy: 1) E(A)f =1= 0 on supp 4l; and 2) CPo fl supp 4l. (In
view of the fact that all operators E(~)-14lE0) belong to @Se(>"), the subspace
E(>..)'JC is invariant under multiplication by smooth functions.)
We have E(A)U = E(A)E(~) - 14l E(~)f The operator E(A)E(~) -I can be written
as

E(>")E0)-lv = f G(CP,l/I;>..)v(l/I)dl/l,
s" -1

where

t -i;\+nI2-2'T-I
f
00

G('" ""A) - dt
't','/', - 0 ItCP-l/II n - 2'T '

7' = 1m>.., CP,I/; E Sn -I, while the integral is understood in the sense of analytic
extension in >.. (cf. Chapter 3, §1.3).
So,

(E(>..)u)(c/lo) = f
s" -1
G(CPo,w;A)4l(w)(E(~)f)(w)dw. (7.1)

The function (CP,I/;),.....G(CP,I/;;>") is analytic on (sn-l XSn-1)\d (d the diago-

nal). If for all 4l satisfying 1) and 2) the integral (7.1) would vanish, then the func-
tion w,.....G(CPo,w;A) would be identically zero on the set (supp(4lE(>")f)\ c/lo,
hence, by analyticity, would be equal to zero outside c/lo. On the other hand, the
kernel G cannot have support concentrated on the diagonal, since the operator
E(>")E(X) - 1smoothing
IS (E0) acts as a continuous operator
L2(sn-I)~H'T(sn-I), and 7' may be taken, without loss of generality, a positive
§7. Spectra. Manifold with conical points 215

number).
III. The previous step implies that the subspace E(A)X contains a smooth func-
tion that is distinct from zero everywhere on the sphere. Thus,
Coo(sn -I) C E(A)X. This implies that X = L2(Sn -I). We have obtained a con-
tradiction, which proves the proposition. •

Cor 0 11 a r y 7.2. Every compact operator on L 2(sn -1) belongs to ®c(A).

P r o o f . The algebra ®c(A) contains the nonzero compact operator

E(A)-lcp(w)E(A) - E(X)-lcp(w)E0), where cP is a real-valued function. Since ®c(A)
is irreducible, by proposition 1.5 every compact operator is an element of it. •

Pro p 0 sit ion 7.3. Let ImA,#O, A'# -+-i(k+nI2), k = 0,1,···, and
U(A) = E(A)-lcp(w)E(A). The map U(A)I-+CP induces an isomorphism between the
algebras ®c(A)/:J{L2(sn -I) and c(sn -I). In particular, the norm of a residue class
[Ul is equal to IICP;c(sn-I)II.

Proof. By proposition 3.2.2,

U(A)* = E0)-lcp(w)E(X) = E(A)-lcp(w)E(A)+T, (7.2)

T E :J{L2(Sn -I).

Formulas (3.5) and (3.7) are also true in the case considered; their righthand sides
must be replaced by IICP;c(sn-I)11 (in the proofs obvious changes must be intro-
duced). Hence

(7.3)

It remains to compare (7.2) and (7.3) and to use the relation

U(A»)B(A) = E(A)-ICP'1'E(A) for the operators U(A) = E(A)-lcp(w)E(A) and
)B(A) = E(A)-I'1'(w)E(A) . •
Denote by ®c(A) the disjoint union of the sphere sn - I and the point A. Intro-
duce a topology on ®c(A) in which the point A forms an open set, while a funda-
mental system of neighborhoods of a point w E Sn -1 is formed by sets
V(w) U {A}, where V(w) is a neighborhood of won Sn -I.
216 Chapter 5.

The 0 rem 7.4. Every irreducible representation of the algebra (~\(A) is either
one-dimensional or the identity representation e(A). A one-dimensional representation
can be realized as a map 7T(W):UI->~(w), where ~ E coo(sn-I) is the function
corresponding to the residue class [U]. The correspondence '1/"(W) I-> W, e{A) I-> A defines
A A

a bijection of the spectrum of fJJc(A) onto @\(A). The topology on @\(A) coincides with
the Jacobson topology.
The proof is obtained from proposition 7.3 by standard reasoning. •

*
2. The algebra @\(mp). Let mp = {AEC: ImA=,8} and @c(mp) the algebra gen-
erated by the functions AI->U{A) = E(A)-I~(w)E(A) on m{3, ,8 +(k +nI2),
k = 0,1, .... The norm in @c{mp) is defined by

IIU;@c(m{3)11 = sup IIU{A);L 2 (sn -I) ~ L 2 (sn -1 )11.

hEmp

Denote by 7T(A) the representation U I-> U{A) of @c{m{3) in L 2 (sn -1). For real

*
A all representations 7T(A) are equivalent. They are reducible, since for A real @cCA)
is a commutative algebra. If, however, ImA 0, then 7T{A) is an irreducible
representation in L 2 {sn -I) (cf. proposition 7.1).

P r o p o s i t i o n 7.5. LetA+ =o+i,8,A- - 0 + i ,8, ,8 * 0. Then the

representations 7T(A+) and 7T(A_) are equivalent.

Proof. We show that the operator U = E{A+)-IE(A_) is a unitary intertwin-

ing operator for 7T{A+) and 7T(A_), i.e. 7T{A_)U = U7T{A+ )UU for all U E @c{m{3).
Clearly, for the operators U(A±) = E{A±)-I~{w)E{A±) the equation
U(A_) = U-1U(A+)U holds. Hence it suffices to show that U is unitary.
Formula (1.3.3) for an arbitrary harmonics Ymk implies

_ -2" r[ mH\; +nl2]r [m-0,; +nI2]

U Y mk - 2 r [ m -fA; +n12 ] r [ m +iA; +nl2 ] Y mk.
§7. Spectra. Manifold with conical points 217

Note that

r [ m+i~; +nl2] m +n/2]

r [ -i~;
1,
r[m+i~;+nl2] r[m-i~;+nl2]

since f(Z) = f(z ). It remains to take into account that the spherical harmonics
{Ymd form an orthonormal basis in L 2 (Sn + I). •

Pro p 0 sit ion 7.6. Let p" v be two arbitrary, but distinct, points on m p,
f3 =1= 0, nonsymmetrically located about the imaginary axis. Then the representations
'1T(p.) and'1T(v) are inequivalent.

Proof. Suppose the assertion is false, and that there exists a unitary operator
V:L2(sn-I)~L2(sn-l) such that '1T(p.)U = V- I'1T(v)UV for all U E @c(mp).

We show that every intertwining operator must have the form E(v)-IX(w)E(p.),
where X is a smooth function that does not vanish and is of constant modulus.
Put

K(X) = (U(X)SB(A)* -SB(X)*U(X» (U(A)SB(X)* -SB(X)*U(X»*,

where A E mp, and U(A) = E(A)-Iq,(w)E(A), SB(X) = E(A)-I'lr(w)E(A) are

operator-functions from @c(mp). Theorems 3.3.4 and 3.3.5 imply that
IIK(A.)u;H 2(A,sn - I )11 .;:;; c pllu;L 2(sn - I )11 on mp. Clearly, K(A.) is selfadjoint for
all A E m p, and V-I K(v) V = K(p.). If f is an eigenfunction of K(p.) correspond-
ing to a nonzero eigenvalue, then Vfis an eigenfunction for K(v) corresponding to
the same eigenvalue; moreover,f E coo(sn -I) and Vf E coo(sn-I).
Formula (1.2.1) for E(A) implies that the operators E(A):±:I commute with rota-
tions of the sphere. Therefore, if g is an arbitrary rotation and
u(cp) = Ew--></>(A)-I:=:(w)4..... w(X)v(I/;), then we have
u(gcp) = Ew--></>(A)-I:=:(gw)E</>-->w(A)V(gI/;). Put

(7.4)

Then 1/; ...... f (gl/;) is an eigenfunction of the operator

218 Chapter 5.

where lEg is defined as in (7.4).

The above-said implies that there is a nontrivial subspace X of
coo(sn -I) n ucoo(sn -I) which is invariant under rotations of the sphere.
Let j be a nonzero element in x..
Applying the operator E(v) to both sides of
the equation U(v)Uj = UU{J1.)j and replacing j by E{J1.)-lh, h E coo(sn-I),
leads to

4>E(v)UE{J1.)-lh = E(v)UE{J1.)-I4>h (7.5)

(recall that E{J1.)'!::I(C oo (Sn -I» c Coo (Sn -I ».

We show that we may set h _ 1. For this we must show that E(jL)-ll E X.

'*
Let {gi} be a finite set of rotations of the sphere Sn -I such that the sets
{I/! E Sn -I: h (gil/!) O} cover the sphere. Denote by {fd a partition of unity
subordinate to this cover. Replacing in (7.5) 4> by fil(g7h) and h by g7h, where
(g7h)(I/!) = h(gil/!), and adding the equations obtained, leads to

(7.6)

Since g7h E g7E{J1.)X = E{J1.)g7X = E{J1.)5(, we have E{J1.)-lg7h EX. Hence

the lefthand side is an element of coo(Sn -I), therefore the righthand side belongs
to this space. Applying to both sides of (7.6) the operator E(v)-I we find that
UE{J1.)-II E coo(sn-I). It remains to note that E{J1.)-ll = const (in view of
(1.3.3». So, E{J1.)1 E x..
Substituting in (7.5) h = 1 gives that E(v)UE{J1.)-1 coincides with the operator
of multiplication by the smooth function X = E(v) UE{J1.) - 1 I, whence
U = E(v)-IXE{J1.).
We now verify that the function X is nowhere vanishing. Interchanging the
roles of p, and v, we obtain, as before, that U-I = E{J1.)-lgE(v), g E coo(sn-I).
Hence I = UU- I = E(v)-IXgE(v), hence xg = 1. So IX I > O.
The fact that the operators E(A):±:I commute with rotations implies that
together with U = E(v)-I XE{J1.), the operator U g = E(v)-lg*xE{J1.) is intertwin-
ing. Consequently, the operator UUi 1 commutes with an arbitrary operator
i
from @c(v). Since this algebra is irreducible, the operator UU I is scalar (Chapter
§7. Spectra. Manifold with conical points 219

5, §1.4), i.e. UUi l = c(g)-II, Ic(g)1 = 1. Therefore X(gw)/X(w) _ c(g). The

map SO(n) 3 g t-+ C (g) defines a one-dimensional representation of the rotation
group SO(n). For n ;;;. 3, SO(n) has only the trivial one-dimensional representa-
tion, i.e. c(g) I, while for n = 2 we are led to c(g) = e ima , where m is an
integer and g is identified with rotation over an angle a (cf. [10]). So X(w) = const
for n ;;;. 3, and X(w) = c e imw for n = 2.
Consider the case n ;;;. 3. We compute the value of the operator U on a spheri-
cal function Y mk. By formula (1.3.3), UYmk = I-'m{J1.)l-'m(v) -I Ymk, as

Pm(~) = (-i)m2"r [ m +i~+nI2]r [ m -;A+n ]-1

If we would show that the modulus of the quotient I-'m{J1.)l-'m(v)-1 depends on m,
we find a contradiction with the fact that U is unitary. We have

+ (J1.)l-'m(v) I
II-'m2
I-'m +2(V)l-'m{J1.) -
(7.7)

= I(m +nI2+io-T)(m +nI2-iol +1') I

(m +nI2-io+T)(m +nI2+iol -1') ,

where I-' = 0+ iT, v = 01 + iT. The expression in (7.7) can be equal to one only if

(m +nl2)(O-OI)+T(O+OI) = (7.8)

= -+-«m +nI2)(01 -0)+1'(0+01)),

which implies that 0 = 01 or 1'(0+01) = O. Since by assumption l' *- 0 and

o *--+-01, (7.8) does not hold. Hence U cannot be unitary.

The case n = 2 is considered in a similar manner. •

Cor 0 II a r y 7.7. Let @c(mt ) be the algebra generated by the restrictions of the
elements U E @c(m,8) onto mt = {AEm,8: ReA;;;'O}, with norm

Then the restriction morphism @c(m,8) ~ @c(mt ) is an isomorphism.

Proof. It suffices to note that, by proposition 7.5, IIU(A+)II = IIU(A_)II for

220 Chapter 5.

any element U E @c(mp). Therefore, IIU;@c(mp)11 = IIU;@c(mp)11 (the element

U E @c(mp) and its restriction onto mp are denoted by the same letter) . •
Denote by 'J{{)(mp) the algebra of continuous functions on the half-line
mp = {AEmp: ReA;:';;'O}, with values in 'J{L2(sn-I), and tending to zero (in
norm) at infinity. The norm in ~(mp) is introduced analogously to the norm in
(S)c(mp ).

L e m m a 7.8. The inclusion ~(mp) C ®c(mp) holds, where f3 =1= 0 and

@c(mp) is the restriction of the algebra @c(mp) onto mp.

The proof is the same as the proof of lemma 4.3. We must only replace @(/x)
by @c(mp), and use propositions 7.1 and 7.6 instead of 3.1 and 4.2 . •

Pro p 0 sit ion 7.9. Suppose the operatorjunction At-+ U(A) = E(A)-Iq,(w)
E(A) belongs to the algebra @c(mp). Then the map U t-+ q, induces an isomorphism of
the quotient algebra @c(mp ) 1'Xo(mp ) onto c(sn -I).

Proof. Put IB(A) = }';jIIk IBjk (A), where each of the operators IBjk (A) is either
the operator E(A)-Iq,jk(W)E(A) or the operator
E(X)-I()jk(W)E(~) = (E(A)-Iq,jk(W)E(A))*. Using the relation
E(X)-Iq,jk(W)E(~) = E(A)-I«pjk(W)E(A)+T(A) with T(A) E ~(mp) (proposition
3.2.2), we find

IB(A) = E(A)-I}:IIq,jk(w)E(A)+S(A), S E 'J1:o(mp).

j k

This and (7.3) imply

inf + SUP+IIIB(A)+T(A)II;:.;;.II}:II«Pjk;C(sn-I)II.
TE:KD(m,B) AEm,B j k

The opposite inequality is contained in theorem 3.6.5 . •

Put ®cCm,B) = Sn -I U mp. We introduce a topology on @c(m,B). A funda-
mental system of neighborhoods of a point w E Sn -I is formed by sets
"'{w) U mp, where 'Ww) is a neighborhood of won Sn -I; a fundamental system
of neighborhoods of a point A E mp is formed by open intervals on mp contain-
ing A.
§7. Spectra. Manifold with conical points 221

The 0 rem 7.10. Let i\t-+U(i\) = E(i\)-Iq,(w)E(i\) be an operatorfunctionfrom

the algebra @c(mf3). The following assertions hold:
l)the maps 'I1{w):Ut-+q,(w), wE sn-l, and ?T(i\):Ut-+U(i\), i\ E ml, induce pair-
wise inequivalent irreducible representations of@c(mf3);
2) every irreducible representation of @Jmf3) is equivalent to a representation listed
in 1);
3) the correspondence.. ?T(w)t-+w, ?T(i\)t-+i\, i\ E mr defines a bijection of the spectrum
of @c(mf3) onto @c(mf3). The topology on @c(mf3) coincides with the Jacobson
topology.

The proof of this assertion differs little from the proof of theorem 4.6 (we
do not need lemmas 4.7 and 4.8 in this case), and is left to the reader. •

3. The algebras ~(A),~(mlJ) related with a conical point. These algebras play the
same role for a conical point as the algebras @(i\), @(/x) in §3 and §4 for a non-
singular point x E ~. Irrespective of the fact that the real line Ix is replaced by a
line mf3 = {i\EC: Imi\=fi} (the case fi = 0 is not excluded), the dependence of
symbols on two vectors - the tangent and the cotangent (and not just the
cotangent) vector - makes the algebras ~(i\), ~(m f3) resemble more the algebras
@(i\), @(/x) mentioned than the algebras from the previous sections 1 or 2 of this
paragraph. If ~(i\), ~(mf3) are considered for a nonsingular point x E ~ (they
differ from @(i\), @(/x) by the replacement of Ix by m f3 only), the results of §3, §4
can be literally transferred to ~(i\), ~(m f3). The study of ~(i\), ~(m f3) for a coni-
cal point reduce also to this situation (in view of definition 4.2.1); it requires
almost no additions to the assertions in §3, §4. Therefore we confine ourselves
here to statements only (for a more detailed account see [57]).
Let X be a conical manifold, and Y a fixed, smooth, directing submanifold with
measure v. Denote by ~(i\) the algebra generated by all operators U(i\) (U a mero-
morphic pseudodifferential operator of order zero) on L 2(Y,dv). For every i\, the
identity representation of ~(i\) is irreducible. Every compact operator on L2(Y,dv)
belongs to ~(i\).
Let cp be a unit vector in the directing cone, T; the cotangent space containing
the dual vector </:, V the set of all pairs (cp,O), where () is a unit vector in T; for
which, moreover, cpO = O. The following relation holds for the operators in CI(i\):
222 Chapter 5.

inf IIU(A)+TII = II(I;C(V)II,

TE:JCL 2 (Y,dv)

where (I is the symbol of U. Therefore the map <I(A)/:J\L 2(Y,dv) ~ C(V), assign-
ing to a residue class the symbol of an element representing this class, is an iso-
morphism of the algebra C£(A)/:J\L 2(Y,dv) onto the algebra C(V) of continuous
functions.
Put <I(A) = V U {A}. We will assume that {A} is an open set, while a funda-
mental system of neighborhoods of a point (cp,8) E V is constituted by sets of the
form Gl.n(cp,O) U {A}, where (111)(cp,8) is a neighborhood of (cp,O) on V.

The 0 rem 7.11. Every irreducible representation of the algebra C£0) is either
one-dimensional or the identity representation e(A). Anyone-dimensional representa-
tion can be represented as a map 7T(cp, 0): U ~ (I(cp, 0), where (I is the symbol of the
class [Ul E C£(A)/SIL2(Y,dv), (cp,8) E V. The correspondence 7T(cp,O) 1-+ (cp,8),
~ ~

e 0) 1-+ A defines a bijection of the spectrum of I£(A) onto C£(A). The topology on I£(A)
coincides with the Jacobson topology.
Introduce the algebra <I(mp) of operator-functions on the line
mp = {AEC: ImA=,B}, generated by the meromorphic pseudodifferential opera-
tors of order zero, with norm IIU;<I(mp)11 = sup"IIU(A);L2(Y'dv) ~L2(Y,dv)ll.
Let S"(X) be the cotangent sphere bundle over X\ 0, and let S"I Y be the part of
S"(X) above t~e directrix Y of the cone X. Put <I(mp) = S"I y U mp. We define
a topology on <I(m p). A fundamental system of neighborhoods of a point A E m p
is constituted by the open intervals containing A. Let cp E Y and w a unit vector
r;" i.e. (cp,w) E S" I y. It is obvious that cp may be regarded to be a tangent
°
in
vector to X. If cpw > (respectively, cpw < 0), a fundamental system of neighbor-
hoods of (cp,w) is constituted by sets G(Jf U {AEmp: ReA>N} (respectively,
G(Jf U {AEmp:ReA<N}), where G(Jf is a neighborhood of (cp,w) on S"I y and N is
a real number. If cpw = 0, a fundamental system of neighborhoods is constituted
by sets G(Jf U mp.

The 0 rem 7.12. Let U be an element of the algebra <I(mp), and let (I be its
symbol. The follOWing assertions hold.
1) The maps 7T(cp,W):U ~ (I(cp,w) and 7T(A):U ~ U(A) induce irreducible representa-
tions of C£(m p). Any two of them are inequivalent.
§7. Spectra. Manifold with conical points 223

2) Every irreducible representation of ~(m p) is equivalent to a representation listed in

1).
3) The correspondence 7T(cp, ) ...... (CP, w), 7T(A) ...... A defines a bijection of the spectrum of
A A

~(mp) onto ~(mp). The topology on ~(mp) coincides with the Jacobson topology.

v
4. 100 main theorem. In this section we state the main theorem. Let ~ be a Coo
manifold with finite set Q = {x(l), ... ,x(l)} of conical points (cf. Chapter 4,
§3.1). For symplicity of description we will assume that in a neighborhood of a
point x(j) E Q the manifold ~ coincides with a cone Xj , j = 1, ... ,1. Introduce
the set IDl of functions q, that are given, and are smooth, on S*(~ \ Q) except at
a set S*(~ \ Q) Isingq,; here S*(~ \ Q) is the bundle of cotangent unit vectors
over ~ \ Q, and sing q, is a finite subset of ~ \ Q depending on q,. At the points
x E sing q, the function q, is allowed to have discontinuities of the first kind (cf.
Chapter 4, §4.1). In a neighborhood of a conical point x(f) E Q a function in we
is a smooth map (cp,O) ...... q,(cp,O), where cp is a unit vector in the directrix of the
cone 10, 0 is a unit vector in T;, and T; is the cotangent space containing the
vector cp* (dual for CP). We will also assume that singq, n P = 0 for a fixed
finite set P = {y(l), ... ,y(k)} c ~\Q and any q, E IDl. We may assume that
functions in IDl are given on the disjoint union we of the Whitney sum
A

S(~\(PUQ»$S*(~\(PUQ», the cotangent spheres S*(~)yv),

j = 1, ... ,k, and the bundles S*(10) I Yj , j = 1, ... ,I, where Yj is a directing
submanif01d of Xj •

we.
A

We introduce a topology on Neighborhoods constituting a fundamental sys-

tem of neighborhoods will be called typical. If x 0 f£ P U Q, a fundamental sys-
tem of neighborhoods of (xo,CPo,Oo), where <Po E S(~)xo'00 E S*(~)xo' is
defined as in the topology '5"(S,S*) (§2). A typical neighborhood of a point
(y(f),00) (wherey(f) E P, 00 E S*(~)y<iJ) is the union of sets {y(f)} X 61Jf(OO) and
(611 \ {y (f)}) X S n - 1 X 6(jf(00 ) (where 611 is a neighborhood of y (j) on ~; the part
of Wl above 611 \ {y (f)} is regarded as the product (U \ {y (j)}) X S n - 1 X S n - 1,
and 6(jf(00) is a neighborhood of 00 on the 'cotangent' sphere). Let now x(f) E Q,
<Po E Yj , and 00 be a point of the cotangent sphere S:(/),</>o containing CPo. Denote
by 'I/(<Po) a neighborhood of <Po on Yj , and by K(<Po) the set of points x E Xj
belonging to the directrix of Xj with directrices from 'i,{c/Jo) and such that
0< dist(x,x(f)< £. By 6(jf(00) we mean a neighborhood of 00 on S:vl,</>o. A
224 Chapter 5.

typical neighborhood of a point (x v), cflo, (0) is the union of sets

{xv)} X '"V(cflo) X "1Il{fJo) and K(cflo) X Sn -1 X iillf(Oo). This topology on
A
9R is the
weakest in which all functions in IDl are continuous. Moreover, IDl is the maximal
ideal space of the uniform algebra generated by the functions in IDl.
Let fJ = (fJ}, ... ,fJl) and 8 = (8}, ... ,8k ) be vectors with real coordinates,
8j =1= +(q +n/2), q = 0, I, .... Denote by H p,8(GJrl.,) the space of functions on
~ with norm defined by a partition of unity as follows. For a function u with
support in a neighborhood of a conical point xv) the norm in H p,8(GJrl.,) coincides
with the norm in Hpj (Xj ); for a u with support in a neighborhood of yV) E P,

lIu;H p,.('llt)II = V-l x 123, I(u·.-')(x) 12dxf'2;

finally, for a function u with support in a coordinate neighborhood,
supp(u) n (PUQ) = 0, the norm in Hp,8(GJrl.,) is equivalent to the norm of
U O J(-1 in L 2 (R n ) (here, as before, J( is a coordinate homeomorphism). Let & be the

algebra generated on H p,8(GJrl.,) by the pseudodifferential operators of order zero

whose symbols belong to IDl.
We introduce the space ® (which is homeomorphic to the spectrum of the quo-
tient algebra &/:KH p,8(GJrl.,». It is the union of the set IDl and sets Ix enumerated
by the points x E GJrl.,. Here Ix = {AEC: ImA=fJj} for x = xv),
Ix = {AEC: ImA=8j ,ReA;;;a:O} for x = yV), and, fin~y, Ix is identified with R if
x E GJrl., \ (P U Q). We describe a topology on ®. Let Xo E GJrl., \ (P U Q),
(cflo,Oo) E S(GJrl.,)xo X S*(GJrl.,)xo' with, moreover, cfloOo = O. A typical neighborhood
A A

of (xo,cflo,Oo) in ® is the union of a neighborhood U of this point in IDl and the

set of lines lx, x E P (U), where p (U) is the projection of U on GJrl.,. If 4>000 > 0
(respectively, </JoOo < 0), Ixo is replaced by a set {AE/xo: A>N} (respectively,
{AE/xo: A>N}). Neighborhoods of a point (x v),4>o,Oo) with xv) E Q are
described similarly. A typical neighborhood of a point (yv),Oo) (here yV) E P,
00 E S*(GJrl.,)yV» is the union of a neighborhood U of (yv),Oo) in 9R, the set of
lines lx, x Ep(U)\ (yV)}, and the ray Iyv}. On the lines Ix and rays Iyv} the ordi-
nary topology is taken.

The 0 rem 7.13. Let A be a pseudodifferential operator with symbol cP E IDl,

and let U be the operator symbol of A (if. Chapter 4, §4.4). The following assertions
§7. Spectra. Manifold with conical points 225

hold.
A

1) The maps 7T(m):AI-+W(m), m E ffi1, 7T(X,h):AI-+ U(X, h), h E l;v x E '!)]t. induce
pairwise inequivalent irreducible representations of the algebra (lI'JfH .B,B(~).
2) Every irreducible representation of if / 'JfH .B,B(~) is equivalent to a representation
listed in 1).
3) The correspondence 7T(m)l-+m. 7T(X,h)l-+h E Ix is a bijection of the spectrum of
A A

(£/ 'JfH .B,B(~) onto ®. The topology on ® is the Jacobson topology.

The considerations concerning local algebras, given in 1 - 3, allow us to use in
order to prove this theorem the reasoning given in the proof of theorem 2.1 (§5),
with minor modifications. This is left to the reader.

5. The spectrum of an algebra of singular integral operators on a composite con-

tour. We retain the notations from Chapter 4, §4.5, and that of theorem 2.5. Let
Yp. be an arc of a contour r. Two copies of it with initial point deleted are
denoted by Y;; - and y;; +, and two other copies with end point deleted are
denoted by y + - and y: + . We also introduce the set {iz} of lines, parametrized
by the points of r. For every regular z, the line lz is represented by the line
Imh = 0 in the complex plane, and for Z = ZV, Iz is understood to be the line
Imh = f3v. AThe disjoint union AUP.{Y;- Uy;+ Uy:- Uy:+} U {UzEriZ} is
denoted by ®. The topology on ® is defined as in §2 (above theorem 2.5); more-
over, if Z is the initial (end) point of an arc, we can only speak of the points z + +
and z + - (respectively, z - + and z - -).

The 0 rem 7.14. Let A be a singular integral operator on L2(T,P) with

coefficients a and b, let U be its (operator-) symbol and let YV,j be arcs intersecting at
cusps zv, j = 1, ... ,N v, v = 1, ... ,M. The follOWing assertions hold.
1) The maps 7T(Z, +,-+-):AI-+a + (z)-+-b + (z), 7T(Z, -,-+-):AI-+a_(z)-+-b_(z),
7T(Z, h):A 1-+ U(z, h), h E Iz, where z is a regular point of the contour r, and the
maps 7T(Zv,j, -+-):A 1-+ aizv)-+-bj(zv), j = 1, ... ,Nv, 7T(ZV ,h): A 1-+ U(zv, h), h E iz"
where Zv is a cusp, v =
1, ... ,M, induce pairwise inequivalent irreducible
representations of the quotient algebra (U9U.. 2 (r,p). (The representations
7T(Z, -+-, -+-) are one-dimensional the 7T(Z,h) are two-dimensional 7T(Zv,h) is N v-
dimensional.}
226 Chapter 5.

2)Every irreducible representation of ce/%L 2(I',p) is equivalent to a representation

listed in 1).
3) The correspondences 7T(Z, -+-, -+-)I-+Z±±, 7T(Z, A) 1-+ A E lz' where Z is a regular
point, and, in case of a cusp zv, the correspondences 7T(Zv,j, +) 1-+ z:/'
7T(ZV,j,-)l-+z:j-' ifYv,j is an emanating arc (by z:? we understood the initial
point of the arc Y:?), or 7T(Zv,j' +) 1-+ z~:/, 7T(Zv,j, -) 1-+ Z;'j-, if Yv,j is an incom-
ing arc (by z;,? we understand the endpoint of the arc Y;'?), 7T(Zv,A)I-+A E Iz,
A A

define a bijection of the spectrum 0fce/%L2(I',p) onto ®. The topology on ® is the

Jacobson topology.

We will now assume that the coefficients of all singular integral operators con-
sidered are continuous at a certain regular point Zo E r. Put
Po(z) = IZ - Z 0 I 0 p(z), where 2.80 E ( - 1, 1). Introduce the algebra cfo generated
6

by such singular integral operators on L2(r,PO)' We identify the points Z6 + and

Zo +, and denote the new point by Z6. The point Zo is obtained by identifying
zit - and Zo - . If ~o =1= 0, we identify points on 1'0 that lie symmetric about the
A

imaginary axis, while if ~o = 0 we discard the line Izo' The set obtained from ®
by these operations will be denoted by ®o; it is endowed with the quotient topol-
ogy.

The 0 rem 7.15. All assertions of theorem 7.14 remain valid if ce, ® are replaced
by cfo, ®o, the representations 7T(Z 0, + , -+-), 7T(Z 0, - , -+-) in 1) by the representations
7T(z6 ):AI-+a(zo)+b(zo), 7T(ZO ):AI-+a(zo)-b(zo), and the correspondences
7T(Z 0, -+-, -+-) 1-+ z6=± in 3) by the correspondences 7T(Z6=) 1-+ z6=.
Some remarks concerning the proof. The first assertions in theorems 7.14
and 7.15 must be proved without recourse to propositions 3.1, 4.2, etc. Moreover,
the first assertion of theorem 7.14 is almost trivial, the inequivalence of represen-
tations 7T(Zv,A) for distinct A can be established by comparing traces. This manner
is not suitable for the representations 7T(ZO,A) in theorem 7.15 (for every singular
integral operator A, the traces of the matrices 7T(Z o,A)A for distinct A coincide).
We verify the part of the first assertion of theorem 7.15 that is related to the
representations 7T(Z O,A).
We first prove that a representation 7T(ZO,A) is irreducible if and only if
{J = 1m A =1= 0 (it is also assumed that I1m AI < 1). The algebra 7T(Z 0, A)ce is gen-
erated by the matrices
§7. Spectra. Manifold with conical points 227

= [ - tanh 7TA - i I cosh 7TA]

U(A) if cosh 7TA tanh 7TA .

If 7T(ZO,A) is irreducible, then there is a one-dimensional subspace that is invariant

under the operators U(A) and U(A)*. Hence the representation is irreducible only
if the operator has two orthogonal eigenvalues. The normalized vectors

-112 [exp {i7T(1 12-iA)/2} ]

fl(A) = (2cosh7To) exp {-i7T(1/2-iA)/2} ,

-112 [exp {-i7T(1/2-iA)/2}]

!2{A) = (2cosh7To) exp {i7T(1 12-iA)/2} ,

where 0 = ReA, are eigenvectors for U{A) and correspond to the eigenvalues
ILl = 1 and IL2 = - 1. The scalar product of these vectors equals; for p = 0 these
vectors are clearly orthogonal.
We now prove that for P =F 0 two representations 7T(ZO,AI) and 7T{ZO,A2) are
equivalent if and only if ReAl = - ReA2. Suppose that the representations are
equivalent, i.e. 7T{AI)A = U- I7T(A2)(A)U, where U is a unitary operator. Then
the vectors U/J(AI) and U!2(Al) are eigenvectors for U(A2) and correspond to the
eigenvalues 1 and - 1. Hence they can differ from the vectors fl (A2) and !2(A2)
by factors whose moduli are 1 only. Thus, I(f1(A2),!2{A2)) I = I(f1{AI),!2(AI)) I;
however, this is possible only if ReAl = - ReA2. If the last equation is true, then
the linear operator U mapping jj(AI) to jj(A2), j = 1,2, is unitary. Since the
jj(A) are eigenvectors for U{A), U is intertwining for the representations 7T{AI) and
7T(A2)·
Note that similar considerations may be used also in the higher-dimensional
situation in order to prove proposition 7.6. (The eigenvector of the operator
E",~tj>{A)-I«p(w)~~",{A) corresponding to an eigenvalue «P{wo) has the form
~~",{A)-I~{l/I,wO)' where ~(l/I,wo) is the Dirac-function on sn -I concentrated at
Wo .)
It is left to the reader to become convinced of the fact that for P = 0 all
representations 7T{Z O,A) are equivalent (and split in a sum of two one-dimensional
representations).
The proof of assertions 2) and 3) in theorems 7.14 and 7.15 can be given along
the general scheme (ef. §5).
228 Chapter 5.

§S. The spectrum of a C* -algebra of pseudodifferential operators with oscillating

symbols
1. The algebra 19(~). Let IT = d X Sn -I, d the interval [I,e] with identified ends,
and let iP E Coo(dXS n - 1 XS n - I ). Denote by 19(A), X E R, the C*-algebra gen-
erated on L 2 (IT) by the operators
+00
(A(A)u)(r,q,) = ~ r2wiq~~(r,X+2'1Tq)uq(I/I), (8.1)
q=-oo

where
I
uq(l/I) = jr-2wiq-Iu(r,l/I)dr.
1

Note that the operator A(A)*, adjoint to A(X) with respect to the duality
Idr __
<u,v> = j - j u(r,q,)v(r,q,)dq"
1 r S·-I

differs from the operator u 1-+ }:qr 2wiq E(A + 2'1Tq)iP(r, q"Co»E(X + 2'1Tq)uq by a compact
term only (theorem 3.7.5).

Pro p 0 sit ion s 8.1. For every X E iii the algebra 19(X) is irreducible.

Proof. Irreducibility of the algebra is equivalent to the fact that every nonzero
vector u E L 2(IT) is totaIizing (§1.4). Let uqo E L 2(Sn -I) be a nonzero Fourier
coefficient of a vector u. Denote, as before, by ~(IIi) the algebra of functions on
iii with as values compact operators in L2 (sn -I) and tending to zero at infinity.
Take for U in (S.l) the element of ~(IIi) such that U(A+2'1Tqo) = ("uqo)uqo and
U(X+2'1Tq) = 0 for q =f= qo. By lemma 4.3, the corresponding operator A(X)
belongs to 19(X). Clearly, A(A)u = r2wiqOuqo' The algebra generated on L 2(sn -I)
by the operators E(X) - I iP(q" Co))E(A) is irreducible, for every X E iii (proposition
3.1). The operator of multiplication by r2wiq belongs to 19(X). This and all that has
been said above imply that U is totalizing for 19(X). •

Cor 0 11 a r y S.2. The following inclusion holds: 9fL 2 (IT) C 19(X).

Proof. 19(A) contains nonzero compact operators (commutators; cf. theorem

§8. The spectrum. Oscillating symbols 229

3.7.4). It remains to apply proposition 1.5. •

Let Ajk(A.) be the operator (8.1) in which Ujk(r,A.) = E(A)-Ic{ljk(r,q"w)E(A)
plays the role of U(r, A). Put A = ~jIIkAjk' c{l = ~jIIkc{ljk (j and k run over
finite sets).

Pro p 0 sit ion 8.3. The following equation holds:

inf IIA+TII = IlcIl;C(dXS n - 1 XSn-I)II. (8.2)

TE%L 2(II)

Hence the map A t-+ c{l establishes an isomorphism between the quotient algebra
(9(A) / %L 2(II) and C (d X Sn -I X Sn -I).

Proof. We first show that the following inequality holds:

infIIA+TII;:;;;, 11c{l;C(dXS n - 1 XSn-I)II. (8.3)

T

Let (ro,<[>o,wo) be an arbitrary fixed point. Denote by {u(n)} a sequence of

smooth functions on the sphere for which Ilu(n);L 2(sn - I )11 = 1, and for which
the supports supp u(n) contract towards the point <[>0. It is obvious that {u(n)}
weakly converges to zero in L 2(sn -1). Let also {v(n)} be a sequence of functions

n
v(n)(r) = L ckn)r27Tik,
k=-n

with 1v(n)(r) 12 / r ~ 8(r - ro), where 8 is the Dirac 8-function. (The Fejer kernel
can be used to construct a sequence {v(n)}.)
We will assume that <t>owo =1= o. If 4'owo > 0 (respectively, <t>owo < 0), we
choose a function g, smooth on IR n \ 0 and homogeneous of degree zero, and a
sequence {/Lq} of numbers /L q ~ + 00, such that the vector /Lq Vg(<Po)+(A+27Tq)q,O
is parallel to Wo for all q = 1,2, . .. (respectively, q = - 1, - 2, .. '). (Note
that since g is homogeneous, vectors Vg(q,) and q, are orthogonal.) Theorem 3.5.1
implies that for u Ecoo(sn-I),

(eil'qgU(r, A+ 27Tq)e i l'qg u)(q,) +

- cIl(r, <P,/Lq V g(<P) + (A + 27Tq)<P)U(q,) = 0 (1)

as 1q 1 ~ 00, uniformly in q, E Sn -1, rEd. Let {(n} be an arbitrary positive

230 Chapter 5.

null sequence, and let w(n)(r,l/;) = u(n)(l/;)v(n)(r)eiJlqg(>/;)r-2'1Tiq, where q = q(n) is

chosen such that the inequality

Ik :t~nr2'1TikUV""'4>(r,;\ + 27Tk)w~n)(l/;) - <P(r, cp, T(;\,q, cp»w(n)(r, cp)1 < (n

holds for all rand cp; here, T("A,q,cp) = ILq'Vg(cp) + (;\+ 27Tq)cp. This inequality can
be rewritten as

Since

we have II(A-1)w(n);L2(II)II~ 1<P(ro,<PQ,wo)l. The sequence {wen)} is weakly

convergent to zero in L2(II), hence {Tw(n)} is norm convergent to zero, for every
T E %L 2 (II). This implies inequality (8.3).
Now note that if inf I<P I > 0, then the bounded operator R(;\):L 2 (II) ~ L 2(II),
R(;\)u = "2.qr2'1Tiq~r,;\ + 27Tq )uq, where ~r,;\) = E(;\) - 1<P(r, cp, w) -I E(;\), is an
inverse for A(;\), up to a compact term. This implies that
suppoess(AA*) < 11<P;C(dXsn -1 XSn -1)11 (compare with lemma 3.6). Together
with (8.3) this gives (8.2) . •

2. The algebra 0(S). Let S be a circle, which can be represented by the interval
o..; ;\ ..; 27T with ends identified. Introduce the algebra 0(S) generated by the
operator-functions S :3 ;\I-+A(A) of the form (8.1), with norm
IIA;0(S)11 = sUPAIIA(;\);L2(II) ~L2(II)II. By 7T(A) we denote the representation
AI-+A(A) of 0(S) in L2(II).

Pro p 0 sit ion 8.4. If ;\,Jl are distinct points on S, then 7T(;\) and 7T(p,) are ine-
quivalent.

Proof. Suppose there is a unitary operator V:L 2 (II) ~ L 2(II) such that
V-I A(p,)V = A(A) for all A E 0(S). Let also Pq be projection of L 2 (II) onto the
subspace Hq of functions (r, cp) 1-+ r 2'1Tiq u (cp), U E L 2 (Sn -I). Consider the subalge-
bra of 0(S) generated by the operator-functions (8.1) in which
§8. The spectrum. Oscillating symbols 231

U(A) = E(A)-I(I(cp,w)E(A) «(I independent of r). For the elements of this subalge-
bra we have V-I A(,u.)VPo = U(X)Po, hence A(,u.)VPo = VPoU(X)Po. Letting Pq
act on both sides gives

U(,u.+2'1Tq)PqVPO = PqVPoU(X)Po. (8.4)

Replace A by the adjoint A*,

A(v)*v = ~r2'ITiqU(v+2'1Tq)*vq.
q

Then (8.4) becomes

(8.5)

Consider the representations U 1-+ U(X), U 1-+ U(,u.+ 2'1Tq) in L 2(sn -I) of the alge-
bra generated by the operator-functions R 3 v 1-+ U(v). Equations (8.4), (8.5) imply
that either these representations are equivalent, or PqVPo = O. Equivalence of
these representations means that p.+ 2'1Tq = A (proposition 4.2), and since both A
and p. belong to S, we find q = 0 and A = p.. But, by requirement, A =f= p., hence
P qVPo = 0 for all q. The last equation contradicts the unitarity of the operator
V. •
The verification of the following assertion is similar to the proof of lemma 4.3.

Pro p 0 sit ion 8.5. The following inclusion holds: 9{(S) C 0(S), where 9{(S) is
the algebra of continuous functions on S with compact values (in :J(I.dII»).

Pro p 0 sit ion 8.6. Let A(X) be an operator of the form (8.1). The map A 1-+ cI>
establishes an isomorphism between the algebras 0(S) I9{(S) and
C(dXS n - 1 XS n - 1 ).

Proof. Put B(A) = ~jIIkAjk(A), where every operator Ajk is either an operator
of the form (8.1) or the adjoint of such an operator. By theorems 3.7.4, 3.7.5,
B(A)V = ~qr2'ITiql8(r,A+2'1Tq)vq+T(A)V, where l8(r,A) = E(A)-I~II(ljk(r,cp,w)E(A)
and T E 9{(S). This and proposition 8.3 imply the inequality

inf suoIIB(X)+T(X)II;;;;. suPI~II(ljkl. (8.6)

TEg{'(S) AE:S j k

For a fixed A E S we have, by (8.2),

232 Chapter 5.

in! IIB(A)+T(A)II
TEX(S)
= sup I~II«Pjkl.
j k

Hence, for any £ > 0 we can indicate a finite covering of S by open intervals
Aa and a family of operator-functions Ta EO :J{(S), a = 1, ... ,m, such that
IIB{A)+Ta{A)II';:;; sup I~jrrk«pjk I +£ for all A EO Aa. Let gal be a partition of
unity subordinate to the cover {Aa}. Put T{A) = ~ta(A)Ta(A). Then T EO :J{(S)
and IIB(A)+T{A)II.;:;; sup I~jrrk«pjk I +£. This means that equality holds in (8.6) .

•
3. The algebra of pseudodifferential operators with oscillating symbols. Let if be
the algebra of pseudodifferential operators introduced in Chapter 4, §4.6. (In the
sequel we will use the notations from that section.) Let also T o{~) be the bundle
of nonzero cotangent vectors over ~ and let G be the multiplicative group of the
numbers en, n an integer, acting as an automorphism group of T o{~) preserving
fibers. Denote by T o{~) / G the quotient space obtained from T o(~) by factoriza-
tion with respect to the action of G, and by ~ the Whitney sum
(To(~)/G)$S*(~). To every function «P EO W a map «P:t_c is associated.

°
Locally this map can be regarded as a map (x, r, 1>,0) t-+ «P(x, r, 1>,0), where x EO ~
1> EO S(~)x, EO S*{~)x, and r is a point of the interval d. If x tf. sing«P this
map is independent of «P and r.
e
Put (9 = {(T 0 (~) / G) E9 S * (~)} u e, where is the disjoint union of a set of
circles Cx enumerated by the points of ~. Every circle can be regarded as the
interval 0.;:;; A';:;; 2'17 with ends identified. We define a topology on (9. The part of
(T o(~) / G) $ S * (~) lying above a neighborhood G(L of a point x 0 EO ~ will be
represented as a set of quadruples (x,r,1>,O). Let CV{cf>o) and 611)(00 ) be neighbor-
hoods of the points cf>o and 00 on the unit spheres, let K(1)o) be a subset of G(L
whose image under some coordinate map K:G(L_lR n is the intersection of an open
cone with vertex at K(XO) and containing all directions from K'(XO)(CV{1>o», and the
open n-dimensional ball with center at K(XO). Let also Q(ro) be an £-neighborhood
of a point ro on d. We denote by K(cf>o,ro,£) the subset of points of K(cf>o) satisfy-
ing one of the membership relations emdist(xo,x) EO Q(ro), m = mo,
mo + 1, .... A typical neighborhood in (9 of a point (xo,ro,1>o,Oo) is the union of
the sets {xo} X Q(ro) X '\'{cf>o) X 611)(00 ), K(1)o,ro,£) Xd X Sn -] X 611)(00 ), and the
circles c Xo and cx, x EO K(cf>o,ro,£). A neighborhood of a point A EO Cx is an open
§8. The spectrum. Oscillating symbols 233

interval containing A. All these neighborhoods form a base for a topology.

The space (To('!)lt)I G) e S*('!)lt) with the topology induced by the topology of
e will be denoted by ~. This topology is the weakest topology in which all func-
tions from 91 are continuous. These functions generate a uniform algebra, for
A

which 91 is the maximal ideal space.

Let A be a pseudodifferential operator of order zero, acting on L2('!)lt), let
() E 91 be its symbol, and A its operator symbol.

The 0 rem 8.7. The maps 'IT{v):A~()(v), v E 91, and 'IT(x,A):A~A(x,A.),

A E cx , induce pairwise inequivalent irreducible representations of the quotient alge-
bra tel%L 2('!)lt). The correspondence 'IT{v)~v, 'IT{X,A.)~A E Cx defines a bijection of
the spectrum (tel%L 2('!)lt») ontoe. e
The topology on coincides with the Jacobson
topology. tel %L 2 ('!)lt) is a type I algebra.
Using the results of sections 1 and 2, the proof of this theorem can be given
along the lines of the proof of theorem 2.1. •
 Chapter 6

The spectrum of a C* -algebra of pseudodifferential

operators on a manifold with boundary

This chapter is devoted to the study of the spectra of algebras generated by

pseudodifferential operators on a smooth manifold with boundary. Discontinuities
of the first kind are allowed in the symbols of these pseudodifferential operators.
Dependence of the spectrum on the discontinuities of the symbols, as well as on
the choice of a (weighted) function space, is elucidated.
In §l - §4 local algebras are studied. In the last §5, theorems concerning the
spectrum of an algebra of pseudodifferential operators on '!)1L are given.

§1. The algebras @:(A)

1. Integral transforms. We introduce the operators

(Il+ f)(s) = -+-_i j'~ f?)dt (1.1)

2'TT -00
S -+-10-(

_ i.J!l- _,_0 +joo f (s )ds

- 2 -+- 2'" p.v. .
" -00
s-t

Denote by x::!:: the characteristic function of the semi-axis R::!::, and put
(P::!::u)(x) = X::!::(xn)u(x',x n), x = (x',x n) ERn. The following formulas hold:

(1.2)

If w is a function on the sphere sn-l , then, by definition,

(P::!::w)(O) = X::!::«()n)w«()',()n), I()I = 1.

Pro p 0 sit ion 1.1. For u E Co (Rn \ {x:x'=O}) we have

235
236 Chapter 6.

P±E(A)u(in 12+A,') = E(A)M lx l-..in12H,IIj.'-..x. u(x',Yn), (1.3)

p±E(A)-lu(inI2-A,') = E(A)-lM lx l-+in/2-AII~-+x.u(x"Yn)'

Moreover, in the first equation A. =1= i (k + n 1 2), and in the second
A. =1= -i(k +nI2), where k = 0,1, ....

Proof. We verify, e.g., the first formula in (1.3). It suffices to do this for A. real
(the formula is generalized to other A. by analytic extension). By (1.2.8),

P±E(A)u(inI2+A,') =

= P±MI~I-->in/2-AFx~(M-l)in/2H,-->lx I u(inI2+A,') =

= M 1~I-->in/2-AP±Fx~u(x).
Since by (1.2), P±F = FII±, we have

P±E(A)u(inI2+A,') = (1.4)

Note that the support of the function x ..... v(x',xn ) = IIj.'-..x.u(x',Yn) lies in a
cylinder containing supp u; the directrix of this cylinder is parallel to the xn-axis.
It is obvious that every ray emanating from the coordinate origin has compact
intersection with the support of v. Hence the Mellin transform Mix I-..A+in12V is
defined for a ray. Insert the identity operator MA"Jin/2-"lx IMix I-..Hin/2 in
between F and n± in (1.4). It remains to use (1.2.8) . •

Pro p 0 sit ion 1.2. Let u EC(f(lRn\{x:x'=O}) and

v(x',xn) = IIj.'-..x. u(x',Yn)' Then, for A. E R,

v(inI2+l\.,IC("',"l/» = (1 +TJ2)<n/2-iA)/2 X (1.5)

X IIt.'1(1 +r2iiA-n/2)/2u(inI2+A.,IC(""m.

Here, as in Chapter 1, §6 Ihe map IC:S n - 2 X lR~sn-l \ {(O,-+-)} is defined by

sn-2 X IR :3 (w,/) ..... IC(w,t) = (w(1+/ 2)-I12, 1(1+1 2)-112) = W E sn-l.

Proof. We have
§1. The algebras ®:- (A) 237

v(jnI2+A,I/I) = _1_ Jr- i (i,,/2-A)-ldr X (1.6)

&0
dv
-+--'-
.] +00
X [
2'TT
-00"
f :,>'"
x ±iO-y
"
X

where z = (x',y,,), r = 1x I. Put ~ = y" 1 r 11/1'1, "f/ = 1/1" 1 11/1'1. Now (1.6) can be
rewritten as

-
v(inI2+A,IC(I{I,"f/» =
1
_r,;:-
00. [.]
-+-dt
f0 r-1A-1dr -+--;f'TT _f "f/-'
+ 00
'O-t X
V 2'TT 00

X _b-
v2'TT
f
+00 [

-00
I+~2
2]Up.-"/2)/2
1 +"f/
,ip.u(inI2+p.,IC(I{IJ»dp..

By changing the order of integration with respect to the variables t and p, we

obtain (1.5) . •
For smooth bounded functions on R we introduce the operators

and denote by IC(Q:!::(A» the following operators, acting on functions

U E Coo(S,,-l):

The following assertion can be verified by combining propositions 1.1 and 1.2.

Proposition 1.3. Forallu E Coo(S,,-l) we have

Let x±(R) be a subspace of L 2 (R), x±(R) = {h EL 2 (R):h = Ff,

supp f C R::;:::}, i.e. x:±: (R) consists of the functions that are holomorphic in the
upper (lower) halfpIane. Let also L 2(S,,-2,x±(R» be the space of square-
integrable functions on S,,-2 with values in x±(R), and let X:!::(-iA-nI2) be
the space of functions homogeneous of degree - iA - n 12 and analytic in the
238 Chapter 6.

halfplane Imgn ~ 0, which was introduced above theorem 1.6.5.

Pro p 0 SIt ion 104. For every "A E iii the map
J(A):::lC:±:( - i"A - nl 2) ~ L 2(sn -1, r-(IIi» defined by

(J("A)U)(tt-,71) = (1 +712)-UHnl2)/2(u o IC)(tt-,71)

is unitary.
The proof follows from the definitions of the spaces and the relation
dl/; = (1 +712)-nI2d7ldtt-, where I/; = IC(tt-,71) and dl/;, dtt- are the volume elements
on sn-l and sn-2, respectively . •

2. The algebra @;-(A) for ImA=O. Let II>Ecco(sn-l) and let

be the Toeplitz operator composed of the opera-
rr~lIl1>(IC(tt-,m:SC (iii) ~ SC (iii)
r
tor of multiplication by the function ~ II>(IC(tt-J» and the projection operator
rr- . Denote by Z5 the algebra of operator-functions Sn -2 3 tt-~ II~lIl1>(IC(tt-,m
with norm

Let Li (sn -I) be the subspace of L 2 (sn -1) consisting of the functions that
vanish on the lower hemisphere of Sn -1; the elements of Li (Sn -1) are assumed
to vanish on the upper hemisphere of Sn -I. Let also @;-
("A) be the algebra gen-
erated by the operators p+ E("A)-lll>(w)E("A) on Li (sn -1).

Pro p 0 sit ion 1.5. The algebras @;-

("A) and Z5 are isomorphic for any "A E iii.
Thus, @;- ("A) and @;-
(p.) are isomorphic for all real "A and /-t.

Proof. By proposition 1.3,

P+E("A)-lll>(w)E("A)u = E("A)-IIC(Q-(-"A»II>E("A)u = (1.7)

= E("A)-l ([J("A)-l rr~lIl1>(IC(W,mJ(A)(E("A)U)(IC(W,m]OIC-l(w)},

where W = IC(W,n Theorem 1.6.5 and proposition 1.4 imply that the operator
u(A):Li(sn-l)~L2(sn-2,SC(IIi» defined by U("A)u = J("A)[E("A)u)oIC] is uni-
tary. It is obvious that U("A)-lv = E("A)-1[(J("A)-lv)oIC- 1]. This and (1.7) imply
that U("A) is the required intertwining operator for the identity representations of
§l. The algebras ®-: (A) 239

®-: (A) and ;so •

Introduce the algebra ;s(~) generated by the operators
rr- W(IC(~, .»:X- (R) ~ ~JC- (R) (with ~Esn -I fixed). The following theorem
describes the spectrum of ;s(~). The proof of this theorem can be found in [19]
(see also [25]).
The meridian m(~) = {III=(I/I,I/In) E Sn-I: 1/1'/11/1'1 =~} of Sn-I passing
through the point ~ on the equator is suplemented by the diameter
d = {x ERn: x = (0, - t), - 1 :;;;;;. t :;;;;;. I} joining the north and south points. On the
closed curve y(w) = m(w) U d thus obtained we define the function 04l(·,w) by
04l(z, w) = w(w) if z = 1/1 E m (w), and 04l(z, w) =
= (1-t)w(O, -1)/2+(1 +t)w(O, 1)/2 if z = (O,t) Ed.
The 0 rem 1.6. The algebra ;s(w) is irreducible and contains the ideal :JGC- (R)
of compact operators on x- (R). The quotient algebra ~(w)/::J{3C- (R) and the alge-
bra C(y(w» of continuous functions on y(w) are isomorphic. The isomorphism is real-
ized by mapping a function 04l(-,w) E C(y(w» to the residue class containing the
operator rr-1W(IC(w,» E ;s(w).
So, the maps ?T(z):rr-1w(IC(w,·»I-+04l(z,w), z E y(w), define pairwise ine-
quivalent one-dimensional representations of ;s(w). Together with the identity
representation of ;S(w) these exhaust, up to equivalence, all representations of
;s(w). Hence the spectrum ;s(w) can be bijectively mapped onto the disjoint union
of y(w) and a point e(w) corresponding to the identity representation. In the
Jacobson topology, {e(w)} is an open set, while a fundamental system of neigh-
borhoods of a point Z E y(w) is formed by sets "lIf(z) U {e(w)} , where "lIf(z) is an
ordinary neighborhood of z on y(w).
We now turn to a description of the spectrum of the algebra ®-: (A) for
ImA = 0. Let &-: be the disjoint union of the sphere Sn -I, a diameter d joining
the north and south poles, and a second, individual, copy of the equatorial sphere,
denoted by as'!t-- I. The ends of the diameter are identified with the respective
poles. Endow Sn -I U d with the topology inherited from Rn. Denote by p the
projection of the sphere punctured at the poles onto the equator as'!t-- 1; the
points 1/1 and p (1/1) belong to the same meridian.
We introduce a topology on &-:. A subset on as~-I is regarded as being
open in &-: if and only if it is open in the ordinary topology of the sphere
240 Chapter 6.

as'!t--I. For a point -.[; E Sn -I distinct from a pole, sets GlJl,(-.[;) U P (GUJ(-.[;», where
GUJ(-.[;) is a neighborhood of -.[; on Sn -I, constitute a fundamental system of neigh-
borhoods of -.[; (we may assume that GUJ(-.[;) does not contain poles). A neighbor-
hood of a point ZEd (including poles) is a set GUJ(z) U as n -I (where GlJf(z) is a
neighborhood of z on Sn -I U d). The topology on @; is not separable. It is
clear that @; is a To-space.

The 0 rem 1.7. Let U(A) = E,;,--+<I>(A)-I~(-.[;)E{/ .....tf(A). Then P+U(A) E @;(A),
and the following assertions hold.
I) the maps

17(-.[;):P+U(A)~~(-.[;), -.[; E sn-I, -.[;=t=(O,+l);

17(t):P+ U(A.) ~ (l ~ t) ~(o, -1) + (1; t) ~(o, 1), -1 ~t ~ 1;

7"(W): p+ U(A) ~ rr ~(IC(W,. », W E as'!t--I,

induce pairwise inequivalent irreducible representations of @;(A). (Both 17(t) and
17(-.[;), -.[; E sn-I, are one-dimensional 7"(w), WE S'!t-- I , is a representation in
SC- (Ill).)
2) Every irreducible representation of@; (A) is equivalent to a representation listed in
1).
3) The correspondence 17(-.[;) ~ -.[; E Sn -I, 17(t) ~ (O,t), 7"(w) ~ W E as'!t-- I defines a
bijection of the spectrum of@;(A) onto the space @;. The topology on @; coin-
cides with the Jacobson topology.

Proof. The first assertion follows by combining proposition 1.5 and theorem
1.6. We verify the second assertion.
Let i' be a function given, and smooth, on the cylinder Sn -2 X m, where
Sn - 2 is the equator and m is a meridian of the sphere Sn -I. In other words, i' is
smooth on sn -I outside the poles, and has discontinuities 'of the first kind' at the
poles. Introduce the algebra 0' generated by the operator-functions
Sn -2 3 w~IIr::...'1 i'(IC(W,m. It is obvious that 0 is a subalgebra of 0'.
Denote (for a fixed W E Sn -2) by 0'(W) the algebra generated by the operators
II-'.V(J(w,·»:SC-(IIl)~SC-(IIl). For each WE Sn-I and any irreducible
§1. The algebras @:- (A) 241

representations'TT of 0'(W) we define a representation Pw of 0' by Pw(x) = 'TT(x(w»,

x E 0'. Our nearest aim is to convince ourselves of the fact that the map 'TTl-> Pw
is a bijection of the union of spectra U~'(w), wE Sn -I, onto the spectrum ~'.
Let p, E ~'(WI)' v E ~'(W2)' with WI =l=w2. Let also j E coo(sn-I),
j(wI) = 1, j(W2) = O. Choose an element x E 0' for which P,(X(w2» =1= O. It is
clear that jx E 0' and that PJJ(jx) = p,(X(WI» =1= 0, Po(jx) = v(O) = 0, i.e. PJJ
and p" are inequivalent representations. Hence the map 'TT I-> Pw is injective.
Now assume that P E 0' and that J is the kernel of p. Put
J (w) = {x (w): x EI}. Just as in the proof of lemma 5.5.3 it can be established
that xEJ if and only if X(W)EJ(W) for all WES n - 2. The set
Y = {w: J(w) =1= 0'(W)} consists of a single point, 0 (cf. lemma 5.5.4). This
implies that if two elements x,y of 0' coincide on 0, then p(x) = p(y). This allows
us to regard P as a representation of 0'(0). In other words, p(x) = 7T(X(O) for a
A

certain 'TT E 0'(0), i.e. the map 'TT I-> Pw is surjective.

So, the map indicated is bijective. Together with theorem 1.6 this implies that
every irreducible representation of 0' is equivalent to one of the following
representations: T(W):rr '1'("0)1-> rr it(,,(w,»,w E sn -2; '1T(\jI):II- '1'("0)1-> it(\jI),
\jI E Sn -I, #(0, -+ 1); 'TT(t, w):II- '1'(,,(.» I-> (1- t)it(m _ (w» 12 + (1 + t)it(m + (w»/2,
-1 ..;;; t ..;;; 1, where m + (w) (respectively, m _ (w» is the north (respectively, south)
end of the meridian m (w). The truth of assertion 2) now follows from proposition
5.1.6.
The third assertion can be verified by directly comparing the topology on @ t
with the Jacobson topology . •

3. The algebra @t (A) for ImA =1= O. In this section it is assumed that A. is an
arbitrary complex number subject to the requirements A. =1= i (k + n 12),
k = 0,1, ... ,ImA. =1= O.

Pro p s i t ion 1.8. The algebra @t (A.) generated by the operators

0

p+E(A.)-I~(w)E(A.) on L 2 (sn-l) is irreducible.

Proof. Let X C Lt (Sn -I) be an arbitrary invariant subspace of @t (A.)

(hence its orthogonal complement X 1. in Lt (Sn -I) is also an invariant subspace
for @t(A.». We must convince ourselves that either X = 0 or X = Lt (Sn -I).
242 Chapter 6.

I. In this step we prove that there is in at least one of the spaces X and X -L a
°
function U ;ijI!; that is sufficiently smooth on Sn -I.
Put i'(x) = Ix'IV(xn-i Ix'I)-V!(x'/lx'j), where! E c oo (sn-2) and p>o
is large. For every x' =1= 0, the function i' has an analytic extension to the half-
plane 1m Xn < 0, which remains to be homogeneous of degree zero. By increasing
p we may assume that the restriction of i' onto Sn -I has any given (finite)

smoothness. By proposition 3.1.1, for p large the operator p+ E(A) -I i'E(A)

belongs to @\:(A) (as does the operator p+ E(~)-li'E(X) = (p+ E(A)-I~E(A))·).
For v E Co (S~ -I), the function x t-+ 'l'(x )(E(A)V )(x), given and homogeneous
of degree - fA - n 12 on iii n \ 0, can be analytically extended to the halfplane
Imxn < 0, the support of E(A) - I i'E(A)v lies on S~- 1 (theorem 1.6.5). Thus, the
operator E(A) -I i'E(A) coincides on Lt (sn -I) with the operator
p+E(A)-li'E(A), and belongs to ®:(A). Similarly, E(X)-li'E(X) E ®:(A). This
implies that the selfadjoint operator

K = (E(A)-li'E(A)-E(X)-li'E(X)i X (1.8)

X {p+ (E(X)-I~E0)- E(A)-I~E(A»)}/,

where I is an arbitrary natural number, also belongs to ®: (A). If the number p

participating in the definition of i' is large (i.e. i' is sufficiently smooth), then the
operator E(A)-li'E(A)- E(X)-I'l'E(X) acts continuously from Hs(Sn -I) into
Hs+1(sn-l) for o~s ~N, where N can be taken arbitrarily large by increasing
p (proposition 3.2.2). Hence the map K:HO(sn -I) ~ HI(sn -I) is continuous for

p large. Consequently, every eigenfunction of K:Lt (sn -I) ~ Lt (sn -I)

corresponding to a nonzero eigenvalue is sufficiently smooth on Sn -I.
Let J.L be a nonzero eigenvalue of K, let ~ be the subspace spanned on all
eigenvectors corresponding to this eigenvalue, and let Pjl:Lt(sn-I)~~ be
orthogonal projection. Since Pjl E ®:
(A), we have PjlX C ~ n X. If PjlX =1= 0,
then X contains a nonzero smooth function. If, however, PflX = 0, then
°
PflX-L =1= and there is a smooth function in PflX-L C ~ n X-L.
II. We prove the existence of a set Q C Sn - I of positive measure and satisfy-
ing the requirements: a) Q contains, for every point w =1= (0, +1) of it, the meri-
dian passing through wand the poles (0, +1 ); b) p+ E(X) -I hEX for all
h E L2(Sn -I) with support in Q.
§1. The algebras ®: 0) 243

By I above, we may assume that the invariant subspace X contains a

sufficiently smooth function u, u ;;E 0, on Sn -1. Put v = E0)u. By theorem 1.6.5,
the function v (which is smooth and homogeneous of degree - iA. - n / 2) has, for
x' =1= 0, an analytic extension to the halfp1ane Imxn < O. Hence, if for a certain
x' the function Xn ~ v (x',x n) is not identically equal to zero, then the zero set of
this function on Imxn = 0 has measure zero.
Denote by Z (v) the zero set of v on S n - 1, and by Q the subset of the sphere
equal to the union of the meridians passing through the poles (0, + 1) and not
completely contained in Z(v). It is obvious that Z(v) is a closed subset of sn-1,
and that the set obtained from Q by delecting the poles (0, +1) is open; moreover,
mesn-l(QnZ(V» = O. This implies that the set {«Pv: «P E c oo (sn-1)} is dense
in the space XOL2(sn -1), where xo is the characteristic function of Q.
Since U EX, we have p+ E(X) -1 «P E(X)u E
X, i.e. p+ E(X) - 1«Pv E
X for all
«P E Coo (Sn -1). Without loss of generality we may assume that ImA. > 0 (in the
opposite case we interchange the roles of A. and A. from the very beginning). The
continuity of the map E(X)-1:L2(sn-1)~HlmA(sn-1) C L 2(sn-1) (proposi-
tion 1.5.5) and the denseness of the set {«Pv } given above imply that
P+E(X)-lh E Xforallh E XOL2(sn-l).

III. We prove that if P,..X = x;. ,

or, equivalently, x;.
C X, then
X = Lt (Sn -1) (cf. the notation in the first part of the proof). Introduce the
operator Kq obtained from K by replacing it by g*it = itog, where g is an ele-
ment of the orthogonal group SO(n -1), regarded as a subgroup of
SO(n): g(x) = (g(x'),x n). The operators E0), E(X), and p+ commute with g*,
g E SO(n -1), hence g*XI' = {fog: JEx;.}
is an eigenspace of Kg for the same
eigenvalue,u. Let P,..,g: Lt (sn -1) ~ g+ x;.
be orthogonal projection, and let U be
the set of elements g E SO(n -1) for which PI',gX = g*x;.. By assumption, the
identity element e E SO(n -1) belongs to U, hence U =1= 0. The inclusion
g*x;. c X (true for g E U) implies that U is closed. Now assume that go E U
and let h 1, ... ,hm be elements of X such that {P,..,gOhj}j= 1 is a basis in go x;.. It
is obvious that for any g E SO(n - 1) near go the system {P,..,ghj } j= 1 is linearly
independent. In other words, a neighborhood of go belongs to U, i.e. U is open.
Consequently, U = SO(n -1) (under the assumption that PI'X = x;. ,
which is
assumed to hold in this step of the proof).
Let u, v, and Q be as in the first step, and let g 1, ... ,g[ be a set of elements of
244 Chapter 6.

SO(n -1) such that the sets gjrl, j = 1, .. ',1, cover the sphere. The set gjrl
corresponds to the function gj U in the same sense in which rl corresponds to u. In
particular, gjv = E(~)gju, and p+ E~)-Ih E X for all h E Xgj rl L 2(sn -I).
Denote by {7]j} a partition of unity on the equator as,+- I subordinate to the
cover {gjrl n as,+- I }. Extend the 7]j onto Sn -I by taking them independent of
Xn· For any f E L 2(sn-l) we have P+E(~)-If = ~jP+E(~)-I7]Jf E X, since
every term belongs to X. Since the set {h: h = P+E(~)-If, f E L 2 (sn-I)} is
dense in Li (sn -I), we have X = Li (sn -I).
IV. Replacing, if necessary, X by Xi., we will assume that PI'X =#= O. The
equality dimX = 1 implies that the condition PI'X = 5Cj, holds. If K has an
eigenvalue p, such that dim 5Cj, = 1, the proof of the proposition would be
finished. Suppose there is no such eigenvalue; we will show how to modify K such
that an eigenvalue as required appears.
Let, as before, 5Cj, be an eigenspace of the operator (1.8), p, =f= 0, and
dim 5Cj, ;;;. 2. Put Q = PI'\8PI' , where \8 is a selfadjoint operator from ®: (A).
Clearly, Q is finite-dimensional and selfadjoint, and Q E ®;(A). All eigenvectors
of Q corresponding to nonzero eigenvalues lie in 5Cj,. We show that by choosing \8
suitably, we may also assume Q to have at least two distinct eigenvalues. Assume
the opposite holds. Then for any \8 = \8* E ®: (A) we have Qu = JlU, where U is
any vector from 5Cj, and JI is a number. The real and imaginary parts of every
operator in ®: (A) belong to ®: (A). Hence an equality Qu = JlU holds if and
only if \8 is an arbitrary operator in ®: (A); in particular, \8 = p+ E(A)-Iq,E(A).
This implies that the quantity c = (E(A)-Iq,E(A)U,U) = (q,E(A)U,E(~)u) is
independent of U E 5Cj" Ilull = l. Take orthonormal vectors u,v in 5Cj, and com-
plex numbers a,{3, Ia 12 + I{31 2 = l. We have

c = (q,E(A)(au + {3J1), E(A)(au + {3J1» =

- - -
= c +a{3(q,E(A)U, E(A)JI)+ii{3(q,E(A)JI, E(A)U).

Suppose that q, is real-valued. Then the last formula implies that

ReaJj(q,E(A)u, E(~)v) = 0 for all a,{3 E C, Ia 12 + I{31 =
2 1. This is possible
only if (q,E(A)U, E(A)v) = O. Since q, is an arbitrary smooth function,

E(A)uE(A)JI = O. (1.9)
§l. The algebras @: (A) 245

- -
Moreover, the equations c = (cfJE(A)u, E(A)U) = (cfJE(A)v, E(A)v) imply
-
E(A)uE(A)U = E(A)vE(A)V. (1.10)

r
Combining (1.9) and (1.10) gives

[E(A)UE(X)u) = E(A)uE(X)uE(A)vE(X)v = O.

Thus, 0 = (E(A)u, E(A)u) = (u, u) = 1. This contradiction shows that there is an

operator Q with the properties listed above.
Let Vb v2 be distinct eigenvalues of Q. Then the operator K + fQ has at least
two distinct eigenvalues ILj = IL+fVj, j = 1,2. For f small these numbers are dis-
tinct from zero, and the sum of their multiplicities does not exceed dim ~ ([ 16],
theorem 1.3.1 about stability of root multiplicities). Denote by ~, the eigenspace
of K+(Q corresponding to ILl. Clearly, dim~, <dim~, ~, c~. If
dim~, > 1, by repeating this reasoning we find an operator K+fQ+f1Q1 which
will have an eigenspace of dimension less than dim ~, and contained in %Pl' In a
finite number of steps we have constructed an operator with a one-dimensional
eigenspace. •

Cor 0 II a r y 1.9. The algebra @: (A) contains the ideal 'XLi (sn -1) of com-
pact operators.

Proof. @:(A) contains the nonzero compact operator

p+E(A)-lcfJ(w)E(A)-P+E(X)-lcfJ(w)E(X), where cfJ is a real-valued function. The
irreducibilityof @: (A) and proposition 5.1.5 guarantee that
%L 2(sn-1) C @:(A) . •

Pro p 0 sit ion 1.10. The map P+E(A)-lcp(w)E(A)~IrcfJ(KO) induces an

isomorphism between the algebras (A) I @: 'XLi
(S n - 1) and ~.

Proof. Let Ujk(A) be an operator of the form E(A)-lcpjk(W)E(A) or

E(X)-lcpjk(W)E(X), and let 0 = ReA. By proposItion 3.2.2,
P+Ujk(A) = P+Ujk(o)+Tj , where Tjk E %Li(sn-1). This and proposition 1.5
imply
246 Chapter 6.

where the indices j, k run over finite sets.

We now prove the opposite inequality. Let
<!J(w) = ~jnkn- (ljk(K(W, '))::JC (~) ~ X- (~), W E as'!t.-I. Let also U E X- (~),
Ilull = 1, and II<!J(wo)u; X- (~)II ;;;.11'8'(wo); X- (~) ~ X- (~)II-£, for given
Wo E as'!t.- I and £>0. Put v(w',wn) = (l+(wn/Iw'li)(i(J+n/2)/2
x(w'/lw'l)u(wn/Iw'I); here X E Coo(as'!t.- I ), the support of X is concentrated
in a neighborhood of wo, and Ilx;L2(aS'!t.-I )11 = 1. By proposition 1.4,
v E X_(-ia-n!2), and Ilv;L2(sn-I)11 = 1. By theorem 1.6.5, W = E(a)-Iv
belongs to L{ (sn-I), and Ilw;L2(sn-I)11 = 1 (since E(a) is a unitary operator).
By formula (1.7),

11:rI}p+ Ujk (a)w;L 2(sn -I )11 = II<!JOxu ;L2(aS'!t.-I, X- (~))II.

Taking into account that the support of X is small, as well as the choice of u, we
find

-£ < 117I}p+ Ujk(a)w ;L 2(Sn -1)11 = 11'8'(wo);X- (~) ~ X- (R))II < £.

Let {£n} be a null sequence of positive numbers, and let {w n } C L{ (sn -I) be a
sequence of functions corresponding to it (i.e. satisfying the inequalities above);
we may moreover assume that the supports of the functions Xn contract towards
the point Wo E Sn -2. It is obvious that {wn} has the following properties:
Ilw n ;L2(Sn-I)11 = 1; {wn } is weakly convergent to zero in L{(sn-I); and, as
n ~OO,

Recall that ~jnkP+Ujk(A) = LjIIkP+Ujk(a)+T, where T E 'XL{(Sn-I). This

leads to the inequality opposite to (1.11) . •
§l. The algebras @-: (X) 247

Cor 0 11 a r y 1.11. Let X E C, X =1= +i(k +nI2), k = 0,1, .. " and

ImX =1= 0. There is an isomorphism @-: (X) I 'XLi (sn -I) ::::::: @-: (JL), where p. is an
arbitrary real number.

The 0 rem l.l2. Theorem 1.7 remains valid if X is an arbitrary complex number,
X =1= +i(k +nI2), k = 0,1, ... ,1mX =1= 0, if the representations listed in 1) are
supplemented by the identity representation of the algebra @-: (X), and if the space
®-: is replaced by the union ®-: U {e(A)}, where e(X) is the pOint corresponding to
the identity representation. This point constitutes an open set, whose closure coincides
with ®-: U {e(X)}.

In order to verify this theorem it suffices to combiI!e proposition 1.8, corollary

1.11, and proposition 5.1.4.

§2. The algebras @ + (.\)

1. The algebras @*(.\) and jf3. Let @+ (A) be the algebra generated on Li (sn -I)
by the operators P+U(A), where U(X) = E(X)-I«II{«p,w)E{X),
«II E coo{sn-IXsn-I), X E C, X =1= +i{k+nI2), k = 0,1,···. The following
assertion can be proved similarly to proposition 5.3.1.

Pro p 0 sit ion 2.1. The algebra @+ (A) is irreducible.

It is obvious that operators of the form fP+ U(X)1), with f,1) E (S'+- 1 ), Co
belong to @+ (X). Such operators with supp f n supp 1) = 0 are compact on
Li (Sn -I). This and propositions 2.1 and 5.1.5 imply

Cor 0 11 a r y 2.2. The following inclusion holds: 'XLi (Sn -I) C ® + (;\).

Pro p 0 s t ion 2.3. Let U{;\) = E{;\)-I«II{«p,w)E{;\),

~(A) = E(X)-li'{«p,w)E(A). Then

p+ U(A)P+ ~(X)u - T(A)u =

= EW---><j>{a)-I{[J{a)-lrr;--->r«ll{«p,IC{w,x)) X

X rr;--->x i'(<<p,IC{W, 1)))J(a){E(;\)u)(IC(w, 1)))]OIC- \w)},

248 Chapter 6.

where o=ReA, W=K(W,n, UECo(sn-I), T(A)E'XLi(sn-I). If

ImA = const and A ~ 00, then IIT(A)II ~ o. Thus, up to a compact term, to the
composite operator p+ U(A)P+ \8(A) corresponds the composite function
n- q,(K(-)n- i'(K(-).

Proof. Put TI(A) = P+U(A)-P+U(o), T 2(A) = P+\8(A)-P+\8(o). L.~, ini-

tially, i'(O,w) = a (O)Y(w). We have

p+U o.....<j>(o)p+\8"".....o(o) =

= p+ Ew .....<j>(o)-Iq,(q"w)Eo .....w(o) X

X p+ Ew.....o(o) -1i'(q"w)E"r, .....w(o) + T 3(0),

where T3(0) = P+(U(o)a-aU(o»P+E(o)-IYE(o). Note that each of the opera-

tors Tj , j = 1,2,3, has the properties stipulated for T. (For T 1 ,T2 this follows
from proposition 3.2.2, and for T3 from proposition 3.2.1.) Writing each of the
operators p+ Ew .....<j>(o)-Iq,(q"w)Eo-->w(o) and p+ Ew-->O(o)-li'(q"w)Ev-->jo) using
formula (1.7), our assertion is proved in this particular case. The general case
reduces to this one by expansion of the function in a series in spherical harmon-
ics: i'(O,w) = ~amk(O)Ymk(W) . •

Pro po sit ion 2.4. Let U(A) = EW--><j>(A)-Iq,(q"w)E"r,-->w(A). Then thefollow-

ing formula holds for the operator (p+ U(A»* adjoint to p+ U(A) with respect to the
scalar product in Li (Sn -I):

(P+U(A»*U -T(A)U = Ew--></>(o)-I X

X {[J(o)-I n~'1q,(q"K(w,mJ(o)(E(o)U)OK(W,nloK-I(w)};

here 0 = ReA, T(A) E 'XLi (Sn -I), and IIT(A)II ~ 0 as A ~ 00, ImA = const

Proof. Put TI(A) = P+U(A)-P+U(o), T2(O) =

(p+ U(o»* - E(o)-Iq,(q"w)E(o). The operators TJ, T2 have the properties stipulated
for T (propositions 3.2.2 and 3.2.5). It remains to use (1.7) . •

Rem ark 2.5. The operator p+ (of multiplication by the characteristic func-
tion of the sphere S~-I) is continuous on Hs(sn-l) for Is 1< 112 (cf., e.g.,
§2. The algebras ® + (X) 249

[74]). This and propositions 3.2.1 and 3.2.2 imply that on a line ImX = const the
following estimate holds for the operators T participating in the statements of pro-
positions 2.3 and 2.4:

IIT(X);H -6(A,sn -I) ~ Li (sn -1)11.;;; c(6);

here 6 < 1/2 is arbitrary positive and c(6) is a constant depending on l) but not
onX.
Our nearest aim is to give a description of the algebra \l5, which, as will be
shown in the sequel, is isomorphic to the quotient algebra ®+(X)I:xLi{sn-I).
Let </J E Sn-I and sn-2{</J) = {WES n - l : q,w=O}. For </J E S,+--I we denote by
qJ!(</J) the algebra C{Sn -2{</J» of continuous functions. For </J E as~-I, by qJ!{</J) we
mean the algebra of continuous operator-functions WH'5{W) on the (n -3)-
dimensional sphere {w: WES n -2(</J) n as~-I}; here '5(w) is a Toeplitz operator on
SC (R) of the form ('5(w)u)('1) = II~->'l :=: (wJ)um, where :=: is a continuous
complex-valued function on {(w,f): (W,nE(sn-2(</J)naS~-I)x[-oo,+oo]). By
definition, 11'5";qJ!{</J)11 = sup", 11'5"(w):SC (R) ~:JC (R)II.
Let V+ (n, 2) be the manifold of pairs (</J,w) of orthogonal unit vectors in R n
with </J E S'+--I. Every function </J E C{V+ en, 2» induces a vector field, i.e. a map
S~-I 3 </JHB~{</J) E qJ!{</J). For </J E S~-I this field is defined by
B ~(</J) = fIJ{</J, .), while for </J E as,+--I by B ~(</J) we mean the operator-function
sn-2(</J) n as~-I 3 wHII-flJ(</J,J(w,». Finally, we introduce the algebra ~
generated by such vector fields; it is an algebra under 'pointwise' multiplication
and involution, endowed with the norm

IIB~O;\l511 = sup IIB~(</J);qJ!(</J)II.

2. The isomorphism between ®+(A)I:xLi(sn-l) and ~. Let

U(X) = E(A)-lfIJ{</J,w)E{X), and let B~ be the element of ~ determined by the res-
triction of fIJ to V+ en,
2). Denote by [p+ U(A)] the residue class in the quotient
algebra ® + (X) l:xLi (Sn -I) containing P + U(X).

Pro p 0 sit ion 2.6. The map [P+U(X)]HBcIl induces an isomorphism between
® + {X)I:xLi (sn -I) and ~.
250 Chapter 6.

Proof. Preservation by this map of the algebraic operations is implied, in par-

ticular, by propositions 2.3 and 2.4. It remains to verify that it is an isometry.
This we do in several steps.
1. Put Ujk(A) = E(A)-lcfJjk(tp,w)E(A); we prove the inequality

i¥f Il~1 k~t+ Ujk(A)+T;Li (Sn -1)~Li (Sn -1)11 ~ (2.1)

~ Il~1 k~/~Jk;~II'
where cfJjk are arbitrary coo(sn-I XS n - l ) functions, and T runs through
'XLi (sn-I).
Let (<Po,wo) E V+ (n, 2) with <Po E S'+-I. We show the existence of a sequence
{v q } C cOO (S,+-I) such that Ilv q;L 2(sn-I)11 = 1, {v q} is weakly convergent to
zero in Li (Sn -I), and II~jIIkP+ Ujk(A)Vqll ~ 1~jIIkflljk(tpO,WO) I. Denote by rk>
k = 1, ... ,N + 1, COO (S'+-1) functions satisfying rkrk + 1 = rk + I,
k = 1, ... ,N, and rN + 1 = 1 in a neighborhood of <Po. Let also v E COO (S'+-I)
with VrN + 1 = v. Recall that ru
j k'l1 E 'XLi (sn -I) for any smooth functions r
and '11 with disjoint supports (proposition 3.1.6). We have

where T E 'XLi (sn -I). By applying theorem 3.3.4 we find that

[7If 1
P + Ujk v = rl E(A)-1
7
If fllj k(tp,W)E(A)rN + 1 v +T 1v;

where also TI E 'Jui (sn-I). Now the existence of a sequence {v q } is proved

literally as in lemma 5.3.3. An obvious consequence is the inequality

i¥f 117IfP+Ujk(A)+T;Li (Sn -I) ~Li (Sn -1)11 ~ (2.2)

~ (4).w)EV
suo l~rrflljk(tp'W)I.
(n,2) j k
T

This inequality gives the necessary estimate for II~jIIkB~}k(tp); 0l(tp) II for
tp E S'+-1 (cf. (2.1)). It remains to estimate this quantity for tp E as,+-1.
§2. The algebras @+ 0) 251

Introduce the operator qf(</>,w) = };jIIklr fl>jk(</>,K(W, .»:

3C (IR) ~ 'JC (IR).
Let (</>o,wo) E V+ (n, 2) and </>o,wo E as~- I (so that Wo = wo). Let also
U E 'JC (IR), Ilull = 1, and IIqf(</>o,wo)u; X- (IR)II > IIqf(</>o,wo);
X- (IR) ~ X- (IR)II- i for a given i > O. Put, as in the proof of proposition 1.10,
v(w',w n) = (1 +(wn I Iw' I )2)Ci(J+nl2)/2 X(w' I IW' I)u(wn I Iw' I); here
X E Coo(sn -2), the support of X is concentrated near wo, and
IIX;L 2(sn-I)II = 1. Then v E X_(-jo-nI2), IIv;L 2(sn-I)II = 1. Denote by g
a smooth real-valued function of a point (w',wn) E Sn - I which is independent of
Wn outside small neighborhoods of the poles (0, -+-1). We will assume that g has
been extended onto IR n \ °
as a homogeneous function of degree zero, and that
\lg(wo) = </>0. We have

(2.3)

- 'I'(</>,JL\l g(</» + o</»v (</» = 0(1)

as I JL I + 1 0 1 ~ 00, JL E IR (cf. (5.3.4»; in the first term of this formula we may

replace E(o)±1 by E(o}+1 and, simultaneously, JL by - JL. Relation (2.3) means
that as JL ~ 00 and 0 is fixed, the support of the function
wet/»~ = Ew .....l/-(o)-le-i/Lg(w)v(w) 'contracts' towards the point </>0 = \lg(wo); more
°
precisely, for ( > and a small neighborhood ua, of <Po given, and for JL large, the
inequality f S·-I va I wet/»~ 12dt/> < i holds.
Since v E X_(-o-nI2), up to a small term belonging to L 2(sn-l) the func-
tion w belongs to Li (sn -I). Therefore we may assume that for JL» 1,

S·-I
f
\("1.lns:- I )
I wet/»~ 12dt/> < 6,

where 6 is an arbitrarily given positive number.

Now take JL sufficiently large and modify the function w constructed to be
equal to zero outside a small neighborhood 'If of </>0 on S~-I. After normalization
in L 2 (Sn -I), the function thus obtained will also be denoted by w.
Let t, TI E Coo (Sn -I), with both functions vanishing outside a small neighbor-
hood of </>0, with t = 1 on '\~ and with tTl =r Check that

(2.4)
252 Chapter 6.

~ [7IJP+ -l<l>jk(of>, E(A) .)E(A) l' +T 1 ~

~ "[7IJ P+ 1,+T+ S,
E(A) - 1<I>jk ( . . ,.)E(A)

where T,T I E 5CLi(sn-l) and S:Li(sn-I)~Li(sn-l) is an operator with

small norm. By applying (2.4) and (1.7) we find that

[fIJP+UjkO+-TW-SW ~
= E(0) -I ([ J(0) -I '5"(</la, w)J( o)(E(o)w )(K(w, '»]OK -I (w)}.

It follows from the way by which w was constructed that the norm in Li (sn -I)
of the righthand side of this equation can be estimated from below by
11'5"(oo,wo);:JC-(IR)~~}C(IR)II-8, for 8>0 small (cf. the proof of proposition
1.10).
These considerations allow us to become convinced of the existence of a
sequence {wn } of normalized functions converging weakly to zero in Li (Sn -\)
and such that 11(~jIIklr Ujk(A»Wn; L 2 (Sn -I )11 ~ 11'5"(cJ>o, wo); (JC (IR) ~ X- (IR)II.
This implies the inequality

i¥f II:rIJP+ Ujk(A)+T;Li (sn -I) ~Li (sn -1)11 ~

~ suo
(</>,w) E V'"' (n, 2)
II~IIII- fPjk(cJ>,K(W,»; X- (IR)~X- (IR)II·
. k
" -1
</>E as +
}

Again using (2.2), we are led to (2.1).

II. We now show that if

(2.5)

then A = ~jIIkP+Ujk(A)-aI is a Fredholm operator. It suffices to construct a

left and a right regularizer, i.e. operators that invert A up to compact terms.
Let {Uq } ~ = I be a finite open cover of the hemisphere S~-I, with
as~-I n U I = 0, while the sets U 2 , ' " ,UQ have small diameters and
§2. The algebras ® + (A) 253

Uq n as~-I =1= 0, q = 2, ... ,Q. Let also {~q} be a partition of unity subordi-
nate to this cover.
Inequality (2.5) implies that I ~jIIk4ljk(CP'w) l-a<O in a neighborhood of
V+ (n, 2). Denote by it a coo(sn -I XSn -I) function which equals
(~jIIk4ljk - a) -I in a neighborhood of V+ (n, 2). Put
RI = p+E(~rlit(cp,w)E(A). We have
RI~IA = p+E(A)-lit(cp,W)E(A)~1 [.7IJUjk(A)-aI] +T,

where T E 'XLi
(Sn -I) (we have used proposition 3.1.6). In the sequel, distinct
compact operators are denoted by the same letter T. Applying theorem 3.3.4 we
find

RI~IA = ~IE(A)-lit(cp,w) [7IJ4ljk(cp,W)-a]E(A)+T.

The function it(~jIIk4ljk - a) equals I in a neighborhood of V + (n, 2). This and
lemma 5.3.5 imply that

(2.6)

Now consider the operator ~qA for q ;;;;.: 2; it can be written as

(.,Au)(~) ~ • ,(~) [7I}P+ E(o) - I",jk(oj>, .)E(0) - 1

oJ u + T u (2.7)

(we have used proposition 3.2.2 in order to replace A by a = ReA, and proposi-
tion 3.2.1 in order to give all functions 4ljk the same argument CP). Choose an
arbitrary point CPq E supp ~q n as~-I, and introduce the set
O(q,£) = {WESn-I:lcpqwl<£} U {WES - : w = (w',w n ), Iwnl >I-£}, where
n l
£ is given and small. Let p:sn -I X sn -I ~ Sn -I be projection on the second
factor. We may assume that the diameter of Uq is small such that
p({CP,W)Esn-1Xs n - l : CPEUq, cPw=O}) C O(q,£). Let Xq,. be the characteristic
function of O(q, f), and let TJq E Cr
(Uq), ~qTJq = ~q. Represent each operator
TJ qE(a)-I4l jk (cp,w)E(a) as a sum

TJq(cp)E(a)-I4ljk (cpq,W)Xq,.(w)E(a) + (2.8)

+TJq(cp)E(a) -14ljk (cpq,w)(l- Xq,.(w»E(a) +

254 Chapter 6.

By lemma 5.3.5, the second term in (2.8) is a compact operator, while the third
term has small norm.
Consider the operator §"(q,q,w) = ~jIIkII- (»jk(q,q,IC(W, .».
If q,qW = 0, then
11§"(q,q,w);:JC (R) -7:JC (R)II < a by requirement. Consequently, this inequality
holds for I<PqW I < £, since £ is small. Modify (if necessary) the (»jk on
Sn-I \O(q,£) such that the inequality I1§"(q,q,w)1 I <a holds for all W E as~-I;
the new functions will be denoted, as before, by i'Djk . It is clear that we have to
add a compact operator to (2.8) as a result of this modification. lbis and (1.7)
implies

(rqAu)(q,) = rq(q,)E(a)-I{[J(a)-IW(q,q,w)-aI)]J(a) X (2.9)

X (E(a)u)(IC(W, .»]ol(-I(w)} +Tu+ Su,

where T E 'XLi (sn -I), S is an operator with small norm, and

11§"(q,q,w);:JC (R) -7:JC (R)II < a for all W E as~-I.
Put E:(<pq,w) = (§"(<pq,w)-aI)-I,

= E(a) -I ([J(a) -I E:(q,q,w)J(a)(E(a)v)(IC(W, '»]OIC-1(W)}

(Rqv)(q,)

and introduce the operator R = ~~=IRqrq: Li(sn-I)-7Li(sn-I). Note that

by using proposition 3.2.1 we may replace in the first righthand term in (2.9) the
function rq by u. (A compact operator should be added to the righthand side of
(2.9) in this case.)
Therefore, for q = 2, ... ,Q,

(2.10)

Combining (2.6) and (2.10) we find that RA = ~~=IRq~qA = I+T+S, where

T E 'XLi(sn -I), and S is an operator with small norm. Thus, (I + S) -I R inverts
A up to a compact term. The existence of a right regularizer is proved similarly.
III. Finally, we will prove the equation

if 117IJP+ Uj k(A)+T;Li (sn-I)-7Li (sn-I)II = (2.11)

§2. The algebras ® + (A.) 255

The preceding part of the proof implies that the inequality

supoess(D*D) 0;;;;

holds for the operator D = ~jIIkP+Ujk().) (oess(G) is the essential spectrum of

an operator G). This means that

i¢" ._IIID+T;Lt(sn-I)~Lt(sn-I)11 0;;;; II~IIB4IJk;$11

TE:KL z (S ) j k

(compare with the proof of lemma 5.3.6). Using (2.l) we are led to (2.11) . •

3. The spectrum of the algebra @+ (A). By proposition 2.6 the description of the
spectrum of ® + ().) reduces to the description of the spectrum of $. Denote by ~
the disjoint union of the manifolds V+(n,2) = {(cp,w): cpES n - l , WES n - l ,
cpw=O}, V(n-I,2) = {(cp,W)EaS~-1 xas~-I: cpw=O} and the set
as~-I X[-I,I]. Identify the sets as~-I X {-+-I} c as~-I X[-I,I] with the
sets as~-I X {(O, -+-I)} c S~-I X Sn -I, respectively. We will regard [- 1,1] as
the diameter of Sn -I with end points at the poles (0, -+-1). We introduce a topol-
A A

ogy on $. A fundamental system of neighborhoods on $ of a point

(c/Io,wo) E V+ (n, 2), c/Io f!. as~-I, is formed by ordinary neighborhoods of this
point on V+ (n, 2). Let c/Io E as~-I, (c/Io,wo) E V+ (n, 2), with, moreover, Wo not
one of the poles (0, -+-1). For such a point (c/Io,wo) a fundamental system of neigh-
borhoods is formed by sets Gh(c/Io,wo) U '"V{c/Io,wo), where Gh(c/Io,wo) is a neighbor-
hood of (cpo,wo) on V+ (n, 2) and '"V{c/>o,wo) is a neighborhood of (c/>o,wo) on
V(n -1,2). (Recall that w denotes the point of intersection of the meridian pass-
ing through w = (w',wn)' w' =1= 0, with the equator as~-I, i.e. w = w'l 1w' I.) If
(ct>o,wo) E V(n -1,2), a neighborhood '"V{c/>o,wo) of it on V(n -1,2) will also be a
neighborhood on $. Further, let p: V(n -1,2) ~ as~-I be the composite of the
inclusion V(n -1,2) C as~-1 X as~-1 and projection
as~-1 X as~-1 ~ as~- I o n the first factor. Neighborhoods of
(c/Io,to) E as~-1 X (-1,1), forming a fundamental system of neighborhoods, are
unions 61l)(c/Io) X ~(to) U p-I(61l)(c/Io», where ~(to) is a neighborhood of to on the
open interval (-1,1), and 61l)(ct>o) is a neighborhood of ct>o on as~-1. Finally, for a
point (c/Io, "'0) where c/Io E as~- 1 and Wo is one of the poles (0, -+-1), sets
61l)(c/Io) X ~("'o) U Gh(c/Io,wo) U p -1(61l)(c/Io» form a fundamental system of
256 Chapter 6.

neighborhoods. Here '5"(wo) is a neighborhood of Wo on the closed interval [-1,1]

(we require in the sequel that the end points of this interval are identified with the
poles).

T h e 0 r e m 2.7. Let U(A) = E"'-+4>(A)-I()(<t»,w)I;;,-..(A). Then

p+ U(A) E @ + (A), and the following assertions hold:
I) the maps

'IT(,p,w):P+ U(A) 1-+ ()(,p,w), (,p,w) E V+ (n, 2);

+ I-t I+t
'IT(,p,t):P U(A) 1-+ -2-()(,p,(O, -I»+-2-()(,p,(O, 1»,

,p E as'!r- 1, t E [-1,1];

induce pairwise inequivalent irreducible representations of the quotient algebra

@+ (A)/CXLt (sn -I) (the representations 'IT(,p,w), 'IT(,p,t) are one-dimensional, and
the T(,p,w) are representations in X- (R»).
2) Every irreducible representation of @i + (A) is equivalent to either a representation
listed in 1) or to the identity representation.
3) The correspondence 'IT(,p, w) 1-+ (,p, w), 'IT(,p, t) 1-+ (,p, t), T(,p, w) 1-+ (,p, w) defines a bijec-
tion of the spectrum if @i + (A) I CXLi (sn -1) onto il. The topology on il is the
Jacobson topology.

Proof. The validity of the first assertion follows from proposition 2.6 and
theorem 1.6. The second assertion can be verified by a reasoning similar to the
corresponding part of the proof of theorem 1.7. Direct comparison of the topology
on \l! and the Jacobson topology establishes the third assertion. •
Note that, for any irreducible representation 'IT, the algebra 'IT(@i + (A» contains
all compact operators on the representation space. The same is true for the alge-
bra @it
(A). Hence @i+(A) and @it
(A) are type I algebras U24], theorem 9.1).
§3. The algebras ®: (1 p) 257

§3. The algebras ®: (/p)

We introduce the algebra ®:
(Ip) of operator-functions on the line
Ip = {XEC: ImX=,B}, ,B =I=-+(k +nI2), k = 0,-1, ... , generated by the func-
tions X~P+U(A) where U(X) = E(X)-ItP(w)E(X), tP E coo(sn -1). The operations
in this algebra are pointwise, and the norm is defined by

lIiB;®:(lp)11 = suplliB(X);Lt (sn-I)~Lt (sn-I)II.

AE1/l

For ,B = 0 this algebra is isomorphic to any of the algebras ®:

(A), ImX = O.
Therefore we will assume that also ,B =1= O. By proposition 1.8 a representation
'/1'(A): p+ U ~ p+ U(X) of ®:
(I p) is irreducible. It turns out that representations
corresponding to distinct X's are inequivalent. The proof of this assertion is pre-
ceded by a lemma.
Define in the halfplane 1m t .;;;; 0 with slit along the interval {t: Re t = 0,
-1.;;;;lmt.;;;;O} the function t ~ fI(t) = exp{[( -iX-nI2)/2]ln(l +t 2)}, where
1n(l +t 2 ) is understood to mean the principal branch on the positive real line.

L e m m a 3.1. There exists for any sufficiently large number k a function

u E Coo (sn -1) n E (A)Lt (sn -1) satisfying: 1) the analytic extension of
t~(l+t2)-(iHnl2)/2(uOIC)(W,t) is zero free in the halfplane Imt.;;;;O (for all
W E Sn -2); and 2) as t ~ 00 in the halfplane Imt .;;;; 0, then

(l+t 2)-(iHnI2)I2(u O IC)(W,t) = /1(t)+O(11 IImA-k-I-nI2).

Proof. Introduce, in the halfplane Iml < 1, the function

12(/) = exp{( - iX - n 12)1n(t - i)} (where the branch of the logarithm is chosen as
before). The quotient 1 ~ 13(/) = fl (t) 112(1) can be extended as a holomorphic
function to the plane with slit along the interval [- i, i 1 of the imaginary axis.
Hence, in the domain 1 1 - i 1 > 2 the function 13 has a Laurent expansion:
00
13(/) = ~ aj(1 -j)-j. (3.1)
j=O

1hls implies that outside the slit in the lower halfplane we have
fI(t) = f(/)+Rk(/), where
k
f(t) = 12(/) ~ a/t -j)-j,
j=o
258 Chapter 6.

1 Rk(t) 1 = O(lt IImA-k-I-n/2) as I~oo, Iml ,,;::;0. The function/2 is zero free
in Iml ,,;::; 0, while a partial sum of the series (3.1) can have at most finitely many
zeros in this halfplane (being a polynomial in (t - i) -I). This means that f is holo-
°
morphic for 1m I ,,;::; and has at most finitely many zeros.
Hf(/) =1= °
for Iml ";::;0, we put u(x) = Ix'l-iA-nl2f(xnllx'j). Outside the
poles (O, -+-1) the function u is infinitely differentiable on Sn -I. The estimate for
Rk and the definitions of f and u imply that in a neighborhood of the north pole
u (x) = 1 + 0 (I x' Ik + I); this ensures that u has continuous derivatives up to
order k inclusive at the north pole. Near the south pole we have
u(x) = exp{ - 'lTi{iA + n 12)} + 0 (I x' 1k + I), and the same conclusion holds.
The function u is clearly homogeneous of degree -iA-nI2, and can be analyt-
ically extended to the halfplane Imxn ,,;::; 0. The formula f = fl - Rk and the esti-
mate for Rk give the inequality necessary in order to conclude
u E :JL(-iA-nl2); thus, u = E{A) v, where v E H+.lmA(Sn-l) (theorem 1.6.5).
Since k may be chosen arbitrarily large, by proposition 1.5.5 we have
v = E{A)-I u E L2(sn-I). So, v E H+.lmA{Sn-l) n L 2(sn-l) c L:{(sn-I),
i.e. u is the function looked for.
Now assume thatfhas, in Iml ";::;0, only the zeros II, ... ,1m with multiplici-
ties KI, ... ,/em. Let q be a fixed natural number. If aj is a negative number that
is large in absolute value, and if Ij = I} - ia}, then the function

has the unique pole I = Ij in 1m I ,,;::; 0, and is zero free there. The function
get) = f (/)p I(ttl ... Pm {t)Km is holomorphic and zero free in 1m I ,,;::; 0. More-
over, plt) = 1 + 0 (I I 1-q) as Therefore, by choosing q large we may
t ~ 00.
define u by u(x) = 1 x' l-iA-nI2g(xn I 1 x' I) . •

Pro p 0 sit ion 3.2. Let 'IT(A.), A E I p, be the representation of ®: (l p) defined

by 'IT(A.)IB = IB(A), 113 E ®: (l p). For distinct numbers /-L and v on I p the representa-
tions 'IT(!-L) and 'IT(v) are inequivalent.

Proof. ®:(lp) contains both functions AI-+P+U(A) = P+E(A)-Iq,(w)E(A) and

A1-+ p+ U(A.)" = p+ E(~) - Iq,( w)E0), hence we may assume, without loss of gen-
erality, that f3 = ImA is positive.
 259

Suppose 'IT{JL) and 'IT(v) are equivalent. Then there is a unitary operator
U:Lt(sn-1)~Lt(sn-1) such that U*p+E(v)-l()E(v)U = p+E{JL)-l()E{JL).
°
Assume that the function IAn \ 3 ~ ~ ()(~, homogeneous of degree zero, is
sufficiently smooth on sn -1 in order that U(A) be continuous (cf. proposition
3.1.1), and that the function ~n ~ ()(~"~n) can be analytically extended, for every
f =1= 0, to the halfplane Im~n < 0, while remaining bounded. Then, by (1.6.5), p+
may be cancelled. Thus, U*E(v)-l()E(v)U = E{JL)-1()E{JL), i.e. ()E(v)UE{JL)-1 =
= E(v)UE{JL-I)() (here () denotes the operator of multiplication by the function
(). This implies

E(v)UE{JL) -I()u = h()u, (3.2)

where u E E{JL)Lt (sn -I) is the function of lemma 3.l,

h = (E(v)UE{JL)-lu)/u E HP(sn-I); the membership is valid, for
E(V):L2(sn-1)~HP(sn-l) is continuous (proposition 1.5.5), while u may be
assumed smooth such that u and u -I are multipliers in Hfi(sn -1).
Consider the set e = {v: v=()u}, where u is the, fixed, function of lemma 3.1
(as A = p,), while () runs through the set of functions homogeneous of degree zero
and having only the properties listed. The set e is dense in the space
E{JL)Lt (sn -1) (with respect to the norm of Hfi(sn -I ». Indeed,
E{JL)Cgo(S~+I) C C<Xl(sn-l) and for a v E E{JL)Cgo(S~-I), the function
t ~(l +t 2)-(iJ.l+nf2)f2(v OIC)(W,t) can be analytically extended to the halfplane
°
Imt ..;; (theorem l.6.5). It follows immediately from the form of the kernel of the
operator E{JL) (cf. (1.2.l» that for It I large, Imt .;;;; 0, we have
(1+t 2)-(iJ.l+nf2)f2(VOIC)(W,t) = t- i J.l-nI2b(w,t), with Ib I ';;;;const. Thus,
t ~«()OIC)(W,t) = (VOIC)(W,t)/(UOIC)(W,t) is a bounded analytic function in the lower
halfplane. In other words, the function () = v / u, which is homogeneous of degree
zero, has the properties listed at the beginning of the proof. Thus, any element
v E E{JL)Cgo (S~-I) can be written as v = ()u, and hence e is dense in
E{JL)Lt (sn -1). Together with (3.2) this leads to: E(v)UE{JL)-l w = hw for every
WE E{JL)Li(sn-I). Whence,

(3.3)

Note that the proposition concerning the existence of an intertwining operator V

implies that (U-1)*p+EG£)-I()EG£)U- 1 = p+E(V)-I()E(V). Now, by a reasoning
similar as in the proof of (3.3) we obtain an analogous expression for V-I:
260 Chapter 6.

(3.4)

where hI = (E(jL)U-IE(p)-luI)lu1 and UI E E(p)Lt(sn-l) IS a function

whose existence is ensured by lemma 3.1.
Using the relation U* = U- I and (3.3), (3.4), we are led to the formula
(E(v) - Ih E(jL»· = E(jL) - Ih I E(p), which can be rewritten as
E(jL)-IQhE(P) = E(jL)-lhIE(p), where Q:H-fi(sn-I)--;.E(jL)Lt(sn-l) is the
projection operator. TIris means that for any function v E o
E(p)C (S'+-I) the ele-
ment Qhv - h I V E E(jL)Lt (Sn - I) is the zero element. Hence for
o
W E E(p.)C (S'+-1) we have

- -
0= (Qhv -hlv,w) = (Qhv-Qhlv,w) = (3.5)

(hv -h1v,Q·w) = (hv -h 1v,w),

since Q* is the projection operator Hfi(sn-I)--;.E(jL)Lt(sn-I). (Here if,g)

denotes the extension of the scalar product in L 2 (Sn -I).) In order to shorten the
notation, for a function w .... v(w) on sn-I we will write v(w,t) instead of
(VOK)(W,t), where w = w' I Iw' I, t = Wn II w' I. Equation (3.5) leads to the relation

f (h -h I)vwdw =
sn-l

+00
= f
sn~2
dw f
-00
(h(w,t)-hl(w,t»v(w,t)w(w,t)(1+t 2)-nI2dt = o.

TIris implies that h(w,t) = hI (w,t) for almost all t E iii and for fixed w E Sn -2.
By the definitions of h, h I (and by theorem 1.6.5), the functions
t .... h (w, t)(l + t 2)(i l1 - iv) 12 and t .... hI (w, t)(1 + t 2)Uv- iii) 12 can be analytically
extended to the halfplane Imt ~ O. Put

h I (w,t)(1 +t 2 i v- iiL)/2 for Imt ~ 0,

J(t) = { (3.6)
h (w,t)(1 +"?-)(ir iv)/2 for Imt ~ O.

Since h,h1 E L2(sn-I), it follows that J is locally integrable on the real line.
Hence J is an entire function.
Recall that h = v I u, where v = E(v)w, wELt (Sn - I), and u is the function
from lemma 3.1 (as A = JL). By applying theorem 1.6.5 we obtain the estimate
§3. The algebras @:- (I fJ) 261

f Ih(w,t)1 2(l+ It 12)-n'2dt =

Imt =T

I Y(w,' t)(I+t2)-(iv+nI2)/212
f Iu(w t)(1 +t 2)-(//-I+nI2) /2 12
(l + It 12) - n12 dt ,,;;;
Imt =T '

,,;;; cI f ( ' )(I+t2)-(iv+nI2)/212 (l + II 12)-n/2dt

I vW,t =
Imt =T
(l + It 12)fi-nI2

= cI f IV(w, 1)(1 +t 2)-(iv+n/2)/2 12 (1+ II 12)- fJ dl";;;C2

Imt =T

for all T";;; O. Here f3 = Imp. = 1m P, and the constant C2 is independent of T. A

similar estimate holds for h 1 too. Hence, for all real T,

f I/(t)1 2(l+It 12)-nI2dt,,;;;c = const<oo.

Imt =T

Now Liouville's theorem implies that lis a polynomial. (In the form necessary for
us, this theorem can be found in, e.g., [74].)
For V-I we can derive

(3.7)

where h' E HfJ(sn -I) (cf. the derivation of (3.3)). The function
t ..... h ' (w,t)(l+t 2)(iv-i/-l)/2 can be analytically extended to the halfplane Imt ,,;;;0.
Formulas (3.3) and (3.7) lead to hh' = 1. Hence the function
t ..... h(w,t)(l+t 2)(i/-l-iv)12 is zero free for Imt <0. A similar conclusion holds for
h I too. Thus, the function I in (3.6) is zero free in the complex plane. (It has no
real zeros since h' = h - I is locally integrable on the real line.) So, I const.
This means that (hOK)(W,t)(1 + t 2)(i/-l-iv)/2 = q(w), i.e. h(w) = Iw' IiV-i/-lq(W). By
(3.3) we hence obtain V = E(p)-llw/liV-i/-lq(w)E(p.).
Introduce the operator Vg = E(p)-lh(gw)E(p.), where g E SO(n -1). The
operator E(A) commutes with the operator (g*u)(w) = u(gw), and
V g = g* V(g - 1)*. Thus, V g is also unitary. Furthermore,

VgP+U(p.) = g*V(g-I)*p+U(p.) = (3.8)

= g*Vp+(g-I)*U(p.) = g*VP+Ug_/(p.)(g-I)*,
where U(p.) = E(p.) -I cfl(w)E(p.), Ug -I (p.) = E(p.) -I cfl(g -I w)E(p.). Since
P+Ug-I(p.) E @:(p.), we have
262 Chapter 6.

g*UP+ Ug -1 (,u)(g -1)* = g* p+ Ug -1 (JI)U(g -1)* = p+ U(JI)Ug. (3.9)

Formulas (3.8), (3.9) imply that Ugp+ U(,u) = p+ U(JI)Ug . The same is true if {L
and JI are replaced by Ii and Ii. Hence Ug is intertwining for the identity represen-
tations of the algebras @;:-(,u) and @:(JI). It is obvious that
U-IUgP+U({L) = U-IP+U(JI)Ug = P+U(,u)U-IUg, i.e. U-IUg commutes with
all elements of @("7- (,u). Since @("7- (,u) is irreducible (proposition 1.8), this implies
that U -I U g = c (g )1, where c (g) is a complex number, Ic (g) I = I (d. Chapter
5, §1.4). So, h(gw)/h(w) = c(g).
The map SO(n - 1) :3 g t-+ C (g) defines a one-dimensional representation of the
group SO(n -1). For n > 3, SO(n -1) has only the trivial one-dimensional
representation, i.e. c(g) =
1, and for n = 2,3 we obtain c(g) = e imw , where m is
an integer and g is identified with rotation over the angle w (if n = 2, then
w = O,'IT).
Suppose n >3. Then h(w) = Iw'liv-il' and U = E(JI)-llw'liV-iI'E(,u). It is
obvious that for Im{L = ImJl = p = 0, U is an intertwining operator. We prove
that for P =t= 0 this operator is not intertwining. Thus a contradiction arises,
finishing the proof.
Define a unitary operator r on L 2 (sn-l) by (ru)(w',wn) = u(w', -wn). Clearly,
r:Vf(sn-I)~Lf(sn-I), and U = r-IUr, where U is regarded as an operator
on L 2 (sn -I). Suppose that p > 0 and that U:Lt (sn -I) ~ Lt (sn -I) is unitary.
Then U:Li (sn -I) ~ Li (sn -I) is also unitary, and so
U:L 2 (sn -I) ~ L 2 (sn -I) is unitary. Applying the operator U* = U- to 1, we
I
find

By formula (1.3.3) this can be rewritten as

(3.l0)

where {Lm(A) = (- i)m2iA r«m + iA + n / 2) / 2)f«m - iA + n / 2) 12) -I. Expand the

function Wt-+ IW' I ip.-iv in a series in spherical harmonics: I w' I ip.-iv = 2.Ymk Ymk ·
This function is not of class coo(sn-I), hence the series is infinite. Relations
(3.10) and (1.3.3) imply that
00 km 00 km
/-Lo(P) ~ ~ Ymk{Lm("ji)Ymk = /-Lo(JI) ~ ~ Ymk{Lm(,u)Ymk·
m =0 k = I m =0 k = I
§3. The algebras @: (I p) 263

Therefore 1Lo(P)l-'-m(ji) = 1Lo(")l-'-mCl-'-) for an infinite set of natural numbers m. How-

ever, as m ~ 00, the lefthand side of the last equations grows as (m 2 + 11-'-1 2 )PI2,
while the righthand side decreases as (m 2 + 11-'-1 2 )-PI2 (cf. (1.5.23)). We have
obtained a contradiction, implying that for f3 =1= 0 the operator cannot be unitary.
In case n = 2 or 3 a similar reasoning leads to the same conclusion. •
Let %it (I p) be the algebra of continuous functions on the line I p with values in
"JILt (sn -1), and tending to zero at infinity, with norm

L e m m a 3.3. The following inclusion holds: %it (I p) C @-: (l p).

This assertion can be proved similarly to lemma 5.4.3; we only have to use pro-
positions 2.1 and 2.2 instead of proposition 5.3.1 and 5.4.2.

Pro p 0 sit ion 3.4. The algebras @:(/p)/'X.tJ+(Jp) and @:(lo) are iso-
morphic.

Proof. By proposItlOn 3.2.2, the operator-function

AI-+T(A.) = P+U(A)-P+U(o), where U(A.) = E(A)-1~(w)E(A), 0 = ReA, belongs
to the ideal 'Xo+ (1 p). Thus,

W sup II~IIP+
j k
TEXu (lp)AEtp
Ujk(A) + T(A)II.;;; sup II~IIP+ Ujk(O)II,
(lEn j k
(3.11)

where Ujk(A.) is an operator of the form E(A)-1~jk(W)E(A) (or E(~)-1<lljk(W)E(~)),

and j,k run through a finite set of values. It remains to convince ourselves that
equality holds in (3.11). For any A E Ip,

TE%L~fs.-l)II:rIJP+Ujk(A)+TII = II:rIJP+Ujk(O)11 (3.12)

(a consequence of (1.11)). But every algebra @-:

(0) is isomorphic to the algebra
@: (/0) (proposition 1.5). Hence the righthand side of (3.12) equals
II~jIIkP+ Uk;@: (/0)11, and the inequality opposite to (3.11) holds. •
A+ A+ A+ .
Let @e be the space of theorem 1.7, and let @e (Ip) = @e U Ip. We mtro-
duce a topology on &-:
(/p). A fundamental system of neighborhoods in (/p)&-:
of a point s E &-: is formed by sets "11 U I p, where "11 is a neighborhood of s in
264 Chapter 6.

@:-, and a fundamental system of neighborhoods of a point A E IP is formed by

open intervals on I p containing A.

The 0 rem 3.5. Let U(A) = E(A)-I()(~)E(A), fJ =t= 0, and let Ip 3 AI-+P+U(A)
be a function from the algebra @;- (I pl. The following assertions hold:
1) Together with the maps '17'(A):P+ U 1-+ p+ U(A), A E I p, the maps '17'(~), '17'(t), T(tf)
defined in theorem 1.7 induce inequivalent irreducible representations of ®;- (I pl.
2) Every irreducible representation of ®;- (I p) is equivalent to a representation listed
in 1).
3) The correspondence '17'(~)I-+~ E sn-l, '17'(t) 1-+ (O,t), T(tf)l-+tf E as,+-l,
'17'(A) 1-+ A E I p is a bijection of the spectrum of ®;- (I p) onto @;- (l pl. The topology
on @;-(Ip) coincides with the Jacobson topology.

Proof. The first assertion follows from theorem 1.12 and proposition 3.2.
We verify the second assertion. If '17' is an irreducible representation of the
quotient algebra ®;-(lp)I:Kt(lp), then by proposition 3.4 and theorem 1.7 it is
equivalent to one of the representations '17'(~), '17'(t), or T(tf) (for any A E 10 = IR,
the algebras ®;- (/0) and ®;- (A) are isomorphic). Suppose that '17' is an irreducible
representation of ®;-(Ip) with nonzero restriction '17'1X;(/p). Then 171X;(/p) is an
irreducible representation of ext (I p) (proposition 5.1.2). Thus, '17' IX; (lp) is
equivalent to a representation of the form 17(A):V 1-+ v (A), v E (I p), A E IP (pro- ext
position 5.1.8). This implies that 17 is equivalent to the representation '17'(A) of
®:-(lp).
The verification of the third assertion is left to the reader. We only show that
every neighborhood in ®;- (Ip) of a point s E @,: contains the complete line Ip.
By the description of the Jacobson topology (given above proposition 5.1.7), it
suffices to prove that if 1117(S)P+UII>1 for some P+UE®;-(lp), then
II'17'(A)P + UII > 1 for all A E I p. (Here 17(S) denotes one of the representations 17(~),
'17'(t), or T(tf).) By corollary 1.11, for all A E I p,

~f
TE9Cl. 2 (S
n-I
)
IIP+U(A)+TII = IIP+U(o)ll, 0 = ReA.
Hence II'IT(A)P+UII = IIP+U(A)II ;;;;.IIP+U(o)11 ;;;;.1117(S)P+UII, i.e.
11'17'(A)P+ UII > 1. •
§3. The algebras @: (I p) 265

§4. The algebras @ + (l p)

1. The algebras @+(/p) and '1). The algebra @+(/p) is generated by the operator-
functions Ip 3 AI->P+U(A), where U(A) = E(A)-)(>(4>,w)E(A), f3 =1= -+(k +nI2),
k = 0,1, .... The operations are pointwise, and the norm is

IIP+U,@+(lp)11 = ~~~IIP+U(]..);Lt(sn-)~Lt(sn-)II.

Thus, the algebra @: (l p) considered in §3 is a subalgebra of @+ (/ p).

Pro p 0 s i t ion 4.1. Let 1-', p be distinct points of I p. Then the maps
7/"(]..): p+ U I-> p+ U(A) for A = 1-', P define inequivalent irreducible representations of
@+(lp).

Proof. Irreducibility of the representations is ensured by proposition 2.1, and

inequivalence is established verbatim as in proposition 5.4.2; moreover, for f3 =1= °
inequivalence follows from proposition 3.2. •
The verification of the following assertion is completely similar to the proof of
lemma 5.4.3.

L e m m a 4.2. The following inclusion holds: cxJ (l p) C @+ (l p). (Here, 'J{(j (l p)

has the same meaning as in §3.J
Now we introduce the algebra '1); in sequel it will be shown to be isomorphic
to the quotient algebra 0)+ (/p)/cxJ(lp). For a 4> E S'+-) we denote by '1)(4)) the
algebra c(sn-I) of continuous functions. For a 4> E as~-), '1)(4)) is understood
to mean the algebra of continuous functions WI-> 'J( w) on as~- ), where 'J(w) is a
Toeplitz operator on 3C (~) of the form ('J(w)u)(s) = TIt-_s:::(w,t)u(t), and::: is a
continuous function on as~-I X [- 00, + 00). Put 11'J;'1)(4))11 = supwll'J(w)ll. Every
function '1' E Coo (sn -) X Sn -) generates a vector field, I.e. a map
S~-I 3 4>I->GDi'(4)) '1)(4)). By definition, for 4> E S~-) we set GDi'(4)) = '1'(4),.),
E

while for 4> E as,+-) , GDi'(4)) denotes the operator-function

as~-I 3 WI->TI-'1'(4),K(W,)). The algebra '1) is generated by all such vector
fields. Multiplication and involution in it are defined pointwise, and the norm is

IIGDi'O;'1)11 = sup 1160'1'(4));'1)(4>)11.

266 Chapter 6.

2. The isomorphism between ® + (lp) 1 ext

(/13) and SD. Let
U(i\) = E(i\)-I«p(</>,w)E(i\) and let 6j)~ be the element of SD determined by «P. Pro-
positlOns 2.3 and 2.4 imply that the correspondence
® + (I p) 1'Xo+ (l p) :3 [p+ Ujl-+ 6j)~ E SD induces a map which preserves the algebraic
operations. This section is devoted to proving that this map is isometric.

Pro p 0 s i t ion 4.3. Let Ujk(i\) = EW --->4>(i\)-I«pjk (cp,w)E,p--->w(i\), and

B(i\)= }';.7=IIIt"=IP+Ujk (i\), where M,N < 00. Then, for any positive number f
and any positive l) < 1/4, there is a constant c such that

IIB(i\)u;L2(sn - I )11,,;;;; f]
[l11lJ6j)~Jk ;SDII + Ilu ;L2(sn - I )11 + (4.1)

+ Ilu ;H- 6(i\,sn -I )11

for all u E Lt (sn -I), i\ E Ip.

Proof. Let U I , U 2 be open covers of the sphere Sn-I such that VI C S~-I,
and such that the boundaries auq , q = 1,2, are located near the equator as~-I.
Let also t E c·,(sn -I), supp t q C Uq , and tI + t~ = 1.
Rewrite the operator t2B(i\) as

t2B(i\)u - T(i\)u =

~ h E.->4>(A) - 1 { [1(A) - 171f4>jk(';' K(w,» X

X J(A)(E,,~ (A)U)(K(W,)ljo. -1 (~+

by remark 2.5 the operator T is subject to the inequality

(4.2)

for (J < 112 and i\ E Ip. Put '5(CP,w) = }';jIIk«Pjk(CP,(K(w,)) and introduce the
operator

where '5(cp,wr is operator adjoint to <J(CP,w) on X- (IR). Denoting by (t2B(i\)r

the operator adjoint to t2B(i\) on Lt (Sn -I), we find that (4.2) holds for the
§4. The algebras @+ (I f3) 267

difference T(A) _ M(A) - (r2B(A))* r2B(A) (the letter T stands for several operators
at the same time). Hence

Let F = II ~J Ilk GD4>Jk ;SD II. Since the set supp r 2 n S'+- 1 lies in a small neigh-
borhood of the equator as,+-I, the operator-function (t[>,w) ~ E(t[>,w)2 =
= r2(t[»z(F 2 -'Y(t[>,w)*§'(t[>,w))+£ is positive for (t[>,w) E S,+-I X SI1-I. Introduce
the operator

The estimate (4.2) holds for the operator T(A) = r~F2-M(A)+£-N(A)*N(A).

(The last assertion, as well as the boundedness of the map
N(A):Li (SI1 -I) ---,> Li (SI1 -I), requires further substantiation; propositions 2.3,
2.4, and remark 2.5 are not immediately applicable in this situation. We will not
dwell into this.) This implies the inequality

F211r2U ;L Z(SI1-I)11 2 +£llu;Lz(SI1-I)ll z - Re (M(A)u, u) ~

~ cllu ;H- o / z(A,SI1-I)ll z + IIN(A)u;L z(SI1-I)ll z ~

~ cIIu ;H- o !2(A,SI1 - I )112.

Together with (4.3) this leads to the estimate

(4.4)

The last inequality is contained in theorem 3.6.2; also,

(4.5)

,,;:;;; F 211rl u;L 2(SI1 -I )11 2 +£llu ;L2(SII -I )11 2+ cllu;H -a/2(A,SI1 -I )112.

Adding (4.4) and (4.5) we obtain

IIB(A)u;L2(SI1 -I )11 2 ,,;:;;;

,,;:;;; (F 2 +£)llu;L 2 (SI1 -1)11 +c(o,£)llu ;H- a12 (A,SI1-I)11 2

for any a < 1/2. •

268 Chapter 6.

Pro p 0 sit ion 4.4. Let B(A) be the operator from proposition 4.3. Then there
is, for every positive number £. an operatorJunction T E 'Xi (/ p) such that

(4.6)

Proof. Denote by qK a C~ (R) function such that qx(/) = 1 for II I < K and
qx(/) = 0 for II I > K + 1. Introduce the operator qx(A):L 2(sn -I) ~ L 2(sn -I)
by
00 kM

qx<A) = ~ ~ qK<.m 2 + IA1 2 )umk Y mk

m=O k=1

and put T(A) = - B(A)P+ qx(A). By (4.1), for u E Lt (Sn -I),

1I(B(A)+T(A»u;L 2(sn -1)11 = (4.7)

= 1IB(A)P+(I-&X<A»u;L 2(Sn-I)11 0;;;;

0;;;; (F +f)II(I -qx(A»u ;L 2 (Sn -1)11 +cll(I -qx(A»u ;H- 8 (A,Sn -1)11,

where, as before, F = II~jIIk6D.jk ;5011. The definitions of the norm in

H- 8 ().,sn-l)(1.5.20) and of the function qx imply the estimate
1I(I-qx<A»u;H- 8 ()"sn-I)llo;;;;K- 8 1Iu;L 2 (sn-I)II. Using (4.7), we obtain for K
large, IIB(A)+ T(A»u;L2(sn -1)11 0;;;; (F+f)lIu;L 2 (sn -1)11 everywhere on /p . •

Pro p 0 s i t ion 4.5. Put Ujk(A) = E",--+</>(A)-lcIljk(cf>,,,,)~--+,,,(A) and

B(A) = ~7=III~=IP+Ujk(A), where M,N < 00. Then

W
TEXu (I~)
liB + T;® + (/ p)11 ;;;;.11~II6D
j k
•.) k ; 50 11. (4.8)

Proof. Let cf> E S'+- 1 and "'0 E Sn - I. Also, let r,." E COO (Sn - I),
o0;;;; r, ." 0;;;; 1, with f = 1 in a neighborhood of cf>o and f." = r. We have
B(A)f = ."E(A)-I~II()jk(cf>,,,,)E(A)f+T(A), (4.9)
j k

where T E 'Xi (lp). Choose a function g, homogeneous of degree zero, such that
"Vg(cf>o) (=:FO) lies in the plane spanned by the vectors cf>o and "'0' Suppose that
sequences {Am} C I p, {Tm} C R are subject to the requirements: Am ~ 00,
Tm ~ 00, and that the direction of the vector (ReAm}cf>o +Tm "Vg(cf>o) tends to the
§4. The algebras @S + (/~) 269

direction of WOo Theorem 3.5.1 implies that for a function u E COO (S'+-}) sup-
ported in a small neighborhood of cPo the following inequality holds for m large:

IUQ.m)eiT.,gu -7ItCtjk(cPo,WO)eiTmgul < (,

where UQ.) = E(A)-lnq,)~jIIkCtjk(q"W)E(A) and ( is an arbitrary pre-given

number. Together with (4.9) this leads to

W IIB+T;@S+(l~)1I ~ max I~IICtjk(q"w)l. (4.10)

Te~ (III) j k
(4),IoI)es"+-1xs·- I
Let now cPo E as,+-I. We may assume that

sup I~ItCtjk(cPo'W)1 < sup II~ItII- Ctjk (4)>o,IC(6'>, ·»11

loIeS.-1 J t:>eS.- l J

(in the opposite case the estimate required already holds). Choose a point
Wo E sn -2 such that the following inequality holds:

sup I~ItClljk(4)>o'W)1 < IIIJ"(cPo,WO);5C (R) -+ 5C (R)II, (4.11)

loIeS.-1 J

where, as before, 1J"(4)>o,wo) = ~jIIkII- Ct(4)>o,IC(WO, .».

Determine a function v just
as in formula (2.3) (orthogonality of 4»0 and Wo is not assumed here). Denote by g
a real-valued smooth function on Sn -I, homogeneous of degree zero, which is
everywhere constant on the meridian passing through WO except in small neighbor-
hoods of the poles. Let also \7g(wo) =1= O. Choose OO,TO E R such that 00 is large
while the direction of the vector OOWo +TO \7g(wo) coincides with the direction of
cPo. Put w(q,) = EIoI~(oo)-le -iTQg(Iol)V(W). Let Ube a domain on sn-I contain-
ing a set V = {(OOW+TO \7g(w»/1 00W+TO \7g(w) I: W E (suppv) \ "Ilf}, where "1J)
is a small neighborhood of the poles (O,± 1) (outside which g is constant on the
meridian indicated above). We may assume that for a given 8 > 0 we have

f Iw(q,) I2dq,<8 (4.12)

S·-I \(uns:- I )

(d. the proof of proposition 2.6). Moreover, the definition of w implies

117ItP+ E( 0)-ICt k(4»o,)E( 0)P+ W;L2(Sn -1)11 ~

0 j 0 (4.13)
270 Chapter 6.

Let t, TI Co (U) with t = 1 in a neighborhood of V and tTl =

E Since r.
(I-Tl)~jnkP+ E(o)-lfIljk(<<Po, ·)E(o)t E :Jet (/
0), by taking into account (4.12) and
(4.13) we obtain

TI~IIP+ E(oO)-I~jk(<Po, ·)E(oo) ;:a: 11§"(<Po,wo)II-c5 (4.14)

j k

(on the left the norm is the operator norm in Li (Sn -I), on the right that in
X-(R».
Suppose that domains U I> U 2 cover the sphere S n - I, U1 c
S'+- I, and that
the boundaries aUq lie in a small neighborhood of the equator as,+- I. Denote by
XI> X2 two coo(sn-I) functions such that suppXq C Uq , XT+X~ = 1. Let also
t q E Co(Uq), tqXq = 1, q = 1,2. Put A(A) = TI~np+E(A)-lfIljk(<Po,.)E(A). We
have
2
IIA(o)ule = ~ IIXjA(o)ule = (4.15)
j=1

2
= ~ Ilt/A(o)+T/o»xjuI12 0;;;;
j =1

~ max{llt/A(o)+T/o»;Li (Sn -1)~Li (sn -1)11 2 }lluI1 2,

J

where T j E :Jet (10). Proposition 4.4 implies that for large 101 and any l) > 0,

Ilt2(A(0) + T2(0»I I < s~p 17IJfIljk(<Po,W)1 +c5,

and (4.11) leads to the estimate

Ilt2(A(0)+T2(o»11 < 1I§"(cf>o,wo)ll. (4.16)

Therefore

(4.17)

(for 101 large). Indeed, in the opposite case we would have, by (4.15), (4.16),
IIA(o)11 < 11<J(<<Po,wo)ll. This contradicts (4.14), in which I 00 I may be regarded
arbitrarily large and l) arbitrarily small. The formula A(X) = A(o)+T(A), where
0= ReA., T E :Jet (Ip), and inequality (4.17) imply
§4. The algebras ~ + (//1) 271

Since r, 'IJ are supported near <Po, this implies that

(Assume this to be not true. Then, for some To EX; (lp) we have

sUP II~IIP+ E(A)-lWjk(tPo, .)E(A)+T(A)II < 11~(tPo,6)o)lI.

AEt/l j k

and thus

Oear1y, in view of the smallness of the support of r1'IJ the latter inequality also
holds if 4> is replaced by tPo. The result contradicts (4.18).) Applying proposition
2.3, formula (4.19) implies the estimate

This, together with (4.10), gives (4.8) . •

Pro p 0 sit ion 4.6. The algebras <M + (//1) IX; (1 p) and SO are isomorphic.
The proof can be obtained by combining propositions 2.3, 2.4, and 4.4,
4.5 . •

3. The spectrum of the algebra <M + (//1). Denote by @+ (I p) the disjoint union of
the sets S~-l X sn-], as~-] X as~-], as~-] X [-1,1], and the line 1/1. Iden-
tify as~-] X {+1} c as~-] X [-1,1] with as~-] X {(O,+I)} C
C S~-] X Sn -), i.e. the interval [-1,1] is represented by the diameter with end
points at the poles (0, +1) of the second factor in the product S~-] X sn -) .
272 Chapter 6.

We introduce a topology on &+(/13). The typical neighborhoods (i.e. neighbor-

hoods forming a fundamental system) of a point (410,"'0) E S'!t-- I X sn-l, under
the assumption that 410 E S'!t--I, 4Iowo = 0, are unions GlL(4Io,"'o) U 113; here
6h(4Io,wo) is a neighborhood of (4Io,wo) on S'!t--I X Sn -I.
Let (4Io,wo) E S'!t--I X Sn -I, with also 410 E as'!t--I, 4Iowo = 0, and suppose
Wo is not one of the poles (0, + I). Typical neighborhoods are in this case unions
6h(4Io,"'o) U 'V'{4Io,wo) u 113 , where 'V'{tPo,wo) is a neighborhood of (tPo,We!) on
as'!t--I X as'!t--I (recall that if w = (w',w n ), w' ¥= 0, then w = w'l Iw' I). For a
point (tPo,wo) E as'!t--I X as'!t--I satisfying tPo"'o = 0, a fundamental system of
neighborhoods is formed by sets'V'{4Io,wo) u /13.
°
If the requirement 4Iowo = is replaced by 4Iowo > 0 (respectively, 4Iowo < 0),
then in the description of typical neighborhoods the line 113 must be replaced by a
set {AElp:ReA> N} (respectively, {AElp:ReA < N}), where N is an arbitrary
real number.
A fundamental system of neighborhoods of a point (</10,"'0) E as'!t--I X (-1,1)
is formed by unions "2tf(tPo) X '5"(to) U "2tf(tPo) X as'!t--I U 113 , where "2tf(tPo) is a
neighborhood of tPo on as'!t--I and '5"(to) is a neighborhood of to on [-I, I].
Typical neighborhoods of a point (410,"'0), where 410 E as'!t--I and "'0 is one of
the poles (0,+1), are sets "2tf(</Io) X '5"(wo) u "2tf(410) X as'!t-- I U GlL(tPo,"'o) U 113 • A
neighborhood in @s + (113) of a point A E 113 is an open interval on 113 containing A.

The 0 rem 4.7. Let U(A) = E(A)-I~(cp,w)E(A), cI» E coo(sn-IXsn-I).

Then the operator{unction 113 :3 A t-+ p+ U(A) belongs to the algebra ® + (/13)' and the
following assertions hold:
1) The maps

'IT(cp,w):P+ U ~ cI»(cp,w), (cp,w) E S'+-1 X sn -1;

+ I-t l+t
'IT(cp,t):P Ut-+ -2-~(CP,(0, -1»+-2-cI»(CP,(O, 1»,

(cp,t) E as,+-I X(-I,I);

T(cp,W):P+Ut-+ rr-cI»(cp,I(w,"», (cp,w) E as,+-I X as,+-I;

'IT{A):P+Ut-+P+U(A), A E 113,
§4. The algebras <M + (I fJ) 273

induce pairwise inequivalent irreducible representations of <M + (I fJ); the representa-

tions 'IT(cp,w), '1T(CP,t) are one-dimensional, the T(Cp,w) are representations in X- (R),
and the '1T(A) are representations in L{ (sn -1).
2) Every irreducible representation of <M + (1 fJ) is equivalent 10 a representation listed
in 1).
3) The correspondence '1T(CP,w) t-+ (CP,w), 'IT(cp,t) t-+ (cp,/) T(cp,w) t-+ (cp,W), '1T(A) t-+~ defines
a bijection of the spectrum of @S + (I fJ) onto ®+ (/ fJ). The topology on ®+ (I fJ) coin-
cides with the Jacobson topology.

Proof. The first assertion follows from propositions 4.1, 4.6, and theorem 2.7.
The second can be verified similarly to the corresponding part of theorem 3.5. The
topology on the spectrum (third assertion) is clarified by a reasoning similar to
that in the proof of theorem 5.4.6. •

§S. The spectrum of an algebra of pseudodifferential operators on a manifold with

boundary
In this paragraph we describe the spectra of algebras generated by
pseudodifferential operators on a smooth manifold with boundary. The transition
from spectra of the 'local' algebras studied in the preceding paragraphs of this
Chapter to the spectrum of an algebra of pseudodifferential operators is per-
formed similarly to the case of a closed manifold. This transition is not discussed
here; we confine ourselves to statements only. We will study three types of alge-
bras: the algebra of pseudodifferential operators in a halfspace (this case is dis-
tinguished for its simplicity of description); the algebra generated by
pseudodifferential operators with continuous symbols; and the algebra of
pseudodifferential operators in whose symbols discontinuities 'of the first kind' are
allowed (the algebra from theorem 5.4.8).

1. Algebras on a balfspace. Let L2(R, Ix 12fJ) (=~(Rn» be the space of func-

tions on R n with finite norm
274 Chapter 6.

let Lt (R, Ix 12/1) be the subspace of functions with support in R,+, and let
P+:L2(Rn,lxI2/1)~Li(Rn,lxI2/1) be the projection operator. Let also
U(A) = E"'-I>(A)-IcI>(If>,w)~-+",(A), cI> E coo(sn -I XSn -I), and

(Au)(x) = (5.1)

= _1_ f ri(H inI2)E"'-I>(A) -1cI>(If>,w)~-+",(A)ii(A + in / 2,t/J)dX,

...f);;1m A= /1
where r = Ix I, If> = x / Ix I, fJ = +(k +n / 2), k = 0,1, . . .. Recall that the
map A:L2(Rn, Ix 12/1)~L2(Rn, Ix 12/1) is continuous, and that on a dense set the
operator (5.1) coincides with an operator of the form Fi=lxcI>(cp,~/ I~I )Fy~,
cp = x / Ix I (cf. theorem 2.3.5). Denote by if the algebra generated on
Lt (Rn, Ix 12/1) by the operators p+ A, and by ifc the subalgebra of if generated
by the operators whose symbols cI> depend on w E Sn -I only. Parseval's equality
(12.7) for the Mellin transform means that If- and If-c are isomorphic, respectively,
to the algebras ® + (//1) and ® -: (//1). Thus, a description of the spectrum of if is
given by theorem 4.7, and of If-c - by theorem 3.5 (if fJ =1= 0). If fJ = 0, then If-c is
isomorphic to any of the algebras ®-:
(A), A E R. The spectrum of ®-:
(A) is eluci-
dated in theorem 1.7.

2. The algebra of pseudodifferential operators with continuous symbols. Let '!JIt be

a Riemannian manifold with Coo boundary (}'!JIt; let S· (~) be the bundle of
cotangent unit vectors. Denote by if the algebra generated on L2('!JIt) by the
pseudodifferential operators of order zero whose symbols are smooth functions on
S·('!JIt). Introduce the space ® (which will tum out to be homeomorphic to the
spectrum of the quotient algebra if/%l.d'!Jlt» as the disjoint union of the sets
S· ('!JIt), S· «(}'!JIt), and 9L(3'!J1t), where 9L(3'!J1t) is the bundle of vectors with length at
most one, cotangent to ~ and normal to ()~ The fiber 9L(3'!J1t)x is the diameter
of the sphere S· ('!JIt)x; vectors +n x E 9L(3'!J1t)x, Inx I = 1, are identified with the
corresponding poles of S· ('!JIt)x (nx is the inner normal vector to (}'!JIt). We define
A A

a topology on ®. Below, neighborhoods of a point of ® are understood to be sets

forming a fundamental system of neighborhoods.
The neighborhoods in ® of a point (xo,Oo) E S·('!JIt), Xo E ~ coincide with
its ordinary neighborhoods in S· ('!JIt).
Let (xo,Oo) E S·('!JIt), Xo E 3~ while 00 does not coincide with any of the
§5. The spectrum. Manifold with boundary 275

poles of S·(~)xo' i.e. end points of the diameter m{a~)xo' Denote by U a neigh-
borhood of x 0 in ~. We will represent the bundles S· (~) I u and S· (a~) I u n ~
as the products U X S·(~)xo and (Una~) X S·(~)xo' Let also W(Oo) be a

neighborhood of 00 on S·(~)xo' and let "Uf(00) be a neighborhood of 80 on

S· (a~)xo (recall that for w = (w',w n ), w' =1= 0, the point wis defined as w' II w'I)·
A

A neighborhood of (xo,Oo) in @ now is a union of sets U X W(Oo) and

o
(Una~) X G(!)(Oo).
Now suppose that (xo,Oo) E S·(~), Xo E a~ while 00 is a pole. A neigh-
borhood of (xo,Oo) ill @ 1S a union of sets U X W(On) and
(U n a~) X (<5"(00 ) u S "(a~)xo)' where <5"(00 ) is a neighborhood of 00 on m{~)xo'

A neighborhood in & of a point (xo,Oo) E s·(a~) is an ordinary neighbor-

hood of (xo,Oo) in s·(a~).
Finally, the neighborhoods of a point (xo,to) E m{a~), Ito I < 1, are sets
(Una~) X (<5"(to)us·(a~», where <5"(to) is a neighborhood of to on m{a~)xo'

The 0 rem 5.1. Let A be a pseudodijferential operator from the algebra &. and
let til be its symbol. The following assertions are true:
1) The maps

T(x,O):Al-+n-tIl(x,,,(O,.»::JC(R)~:J[-(R), x E a~ ° E s·(a~)x,

'1T(x,O):Al-+tIl(x,O), x E ~ ° E S·(~)x,
I-t l+t
'1T(x,t):A 1-+ -2-tIl(x, - nx )+ -2-tIl(x,nx), x E a~ It I < 1,

induce inequivalent irreducible representations of the quotient algebra tU%L2(~)'

2) Every irreducible representation of €I %L 2 (~) is equivalent to a representation

listed in 1).
3) The correspondence 'T(x, 0) 1-+ (x, 0) E S· (a~), 'IT(x, 0) 1-+ (x, 0) E S· (~),
'IT(x,t) 1-+ (x,t) E m{a~) defines a bijection of the spectrum of €1%L2(~) onto @.
A

The topology on @ is the Jacobson topology.

3. The algebra of pseudodifferential operators with discontinuous symbols. Denote

by S + (~) the bundle of tangent unit vectors above ~; the sign + signifies the
276 Chapter 6.

fact that the fiber S + (~)x above a point x E a~ contains only vectors directed
to the side of~; the fiber S + (~)x contains, in addition, also the vectors directed
along a~ Let IDl be the set of functions () with discontinuities of the first kind (it
was introduced at the beginning of Chapter 5, §2; see also Chapter 4, §4.1). A
function () E IDl can be regarded as being given on the Whitney sum
S +(~)6)S*(~).
In this section, ct denotes the algebra generated on L2(~) by the
pseudodifferential operators of order zero whose symbols belong to IDl.
Introduce the space ~ as the disjoint union of the sets S +(~)6)S*(~),
S(aGJlt)6)s*(a~), S(aGJlt)6)~~), and e, where e is the disjoint union of lines Ix
indexed by the points of GJIt. Moreover, S (a~)x 6) { -+- n x } is identified with the
subset of (S + (GJIt)6)S(GJIt»x, and the fiber ~~)x is regarded as the diameter of
S*(GJIt}x. We define a topology on ~ by indicating for every point the neighbor-
hoods forming a fundamental system.
A) Neighborhoods of a point A E Ix. Every line Ix is represented as R. A
A

neighborhood in @ of A E Ix is an open interval on Ix containing A.

B) Neighborhoods of a point (xo,~,80) E S +(~)6)S*(GJIt). Xo E a~ These
are defined as in the case of a closed manifold ~ We recall this definition. The
part of S + (~) (£) S *(GJIt) above a neighborhood U of a point x 0 in ~ is regarded
as a set of triples (x, 4>. f1), where x E U, and 4>,8 are tangent and cotangent unit
vectors. Let V(~) and W(80 ) be neighborhoods of~, respectively 80 , on the unit
spheres, let K(4)o) be a subset of U whose image under some coordinate map
1C: U ~ R n is the intersection of an open cone with vertex at IC(XO) and containing
all directions from 1C'(XO)(V(~», and an open n-dimensional ball with center at
IC(XO). In case ~80 = 0, a neighborhood of (xo,~,80) is a union of sets
{xo} X V(~) X W(80 ), K(4Io) X Sn-l X W(80 ), and of the lines Ix (xEK(~»
and Ixo. If ~80 ~ 0, a neighborhood is also such a union, but with Ixo replaced by
a set {AElxo :A~N}, where N is an arbitrary real number.
C) Neighborhoods of a point (xo,~,80) E S +(~)6)S*(~), Xo E a~ If
~ fi.as + (~)xo' neighborhoods are defined as in case B); of course, IC is under-
stood to mean a map 1C: U ~ R ~, and V (4)0) is a neighborhood of 4>0 in the open
hemisphere S + (GJIt)xo.
Suppose that ~ E as +(~R.)xo and that 80 is not a pole of S*(GJIt)xo' i.e. an end
§5. The spectrum. Manifold with boundary . 277

point of the diameter orthogonal to ~ Let, also, V(cfJo) be a neighborhood of cfJo

in the closed hemisphere S + (~)xo' and let K(cfJo) be defined as in B). Finally,
o 0 *
denote by 'V(cfJo) and 6JJ.\'(80) neighborhoods of cfJo, 80 on S(a~)xo and S (~)xo'
respectively. If cfJo80 = 0, a neighborhood of (x 0, cfJo, 80) is a union of the following
sets: the sets of lines lx, x E K(cfJo), and
Ixo:{({xo}XV(cfJo»U(S+(~)IK(</to»)} X W(80 ); and
o
{({xo) X 'V(cfJo»U(S(a~)IK(</to)na~)} X 6!1)(80). If cfJo80 =1'= 0, then lxo must be
modified as in B).
Let, as before, cfJo E as + (~)xo' but now 80 is a pole of S *(~)xo. A neighbor-
hood of (xo, cfJo, 80) is in this case a union of the sets
lxo: {Ix }xEK(</to); {({xo} X V(cfJo» U(S + (~) XK(cfJo»} X W(80 ); and
{({xo} X'V(cfJo»U(S(a~) IK(</to)n(lG)tt)} X (~(80)US*(a~)xo) (here, ~(80) is a neigh-
borhood of 80 on ~a~)xo).
D) Neighborhoods of a point (xo,cfJo,80) E S(a~)El1s*{a~). A neighborhood
in this case is a union of sets {({xo} X'V(cfJo»U{S{a~) IK(</to)na~)} X6!1){80), the
sets of lines {Ix} x EK(</to) n (lG)tt, and lxo in case cfJo80 = o. If cfJo80 =1'= 0, then lxo is
modified as in B).
E) Neighborhoods of points (xo,cfJo,to) E S{a~)El1~a~), I to I < 1. Let
~(to) be a neighborhood of to on ~a~)xo. A neighborhood of (xo,cfJo,to) is a
union of sets {{{xo} X'V(cfJo» U{S{a~) IK(</to)n(l6)!L)} X(~{to) U S* (a~)xo) and of
the lines lx, x E K(ct>o) n a~ and Ixo.
A A

The topology defined on ® by these neighborhoods is not separable; ® is a

To-space. The topology induced on the subspace S +(~)$S*{~) is the weakest
of all topologies in which all functions from ID'l are continuous.
Let A be a pseudodifferential operator of order zero, and let () E ID'l be its
symbol. Introduce the maps

'IT{x,A):AI-+ U(X,A) = (5.2)

= E(/~4>(A)-l()(x, .p,8)E.t,~(A):L2(S (~)x) ~ L2(S (~)x),

where x E ~ \ a~ .p,l/; E S(~)x, 8 E S*{~)X, A E Ix = R;

'IT(x, A):A 1-+ p+ U{x, A):Li (S(~)x) 1-+ Li (S{~)x), (5.3)

278 Chapter 6.

where x E a~ A E Ix = R, Lt (S(~ll.)x) is the subspace of L2(S(~)x) consist-

ing of the functions with support in the 'upper' halfsphere S + (~)x, and
p+ :L2(S(~)x)~Lt (S(~)x) is the projection operator;

T(x, cp,O):A ..... rr- 4>(x, CP,,,(O, .»:~C- (R) ~ X- (R), (5.4)

where (x,CP,O) E S(a~)$s·(a~);

'1T(x,CP,O):A ..... 4>(x,cp,O), (x,CP,O) E S + (~)EBS·(~), (5.5)

I-t 1+1
'1T(x, cp,t):A ..... -2-4>(x, cp, -n x )+ -2-4>(x, cp,n x ), (5.6)

where (x, cp,t) E S(a~)EB~a~), It I < 1.

The 0 rem 5.2.

1) The maps (5.2) - (5.6) induce pairwise inequivalent irreducible representations of
the quotient algebra eV 9a.d~).
2)Every irreducible representation of eV~2(~) is equivalent to a representation
(5.2) - (5.6).
3) The correspondence

'1T(x, cp,O) ..... (x, cp,O) E

-
S + (~)EDS (~);
.
'1T(X,cp,t)""'(X,cp,t) E s(a~)$~a~)
A A

defines a bijection of the spectrum of (i/~2(~) onto ®. The topology on ® coin-

cides with the Jacobson topology.

This theorem implies that (i is a (non-separable) type I algebra.

 Bibliographical sketch

Chapter 1. 1bis chapter is based on Plamenevskii [44]. All results from the
theory of generalized functions used in it are contained in [11] or in [13]. Formula
(1.3.3) may be derived from (1.4.7) and results in [7]. Expansion in spherical har-
monics from the point of view necessary for us was considered in [1], [39], [40].
The proof of theorem 1.5.1 originates from [39]. Parameter-dependent norms
were introduced, in an appropriate setting, in [2]. A.O. Derviz helped in the
preparation of §6. A sketch of results concerning the operator E(A) on 4 spaces
can be found in [63].

Chapter 2. It seems that V.A. Kondrat'ev [33] was the first who systematically
used the spaces Hp(Rn). The results concerning the Fourier transform given in §2
are, to the best of my knowledge, new, and are published for the first time. The
theorem asserting boundedness of a one-dimensional singular integral operator on
a space L2(R, \ x \2P), if \2fJ \ < 1, was proved by K.I. Babenko [4]. The multidi-
mensional analog of this theorem is due to E.M. Stein, who established that a
singular integral operator is bounded on L2(R, \ x \2P) if \2fJ \ < n, [66]. Results
concerning boundedness of singular integral operators with an exponent 2fJ out-
side the interval [- n,n] were announced in [43] and proved in [44]. The latter also
contains a representation of the form (2.3.6). Another manner of proving an esti-
mate of the form (2.3.8) was given in [70]. In this connection we point to the
recent article [38], which is concerned with the one-dimensional case. The spaces
Hp(Rm,Rm-n) and t9p(Rn) were used in [35], [36] (for nonnegative integers s). In
§4 - §6, results from Plamenevskii [53] are expounded.

Chapter 3. 1bis Chapter contains (in reworked form) results from Plamenevskii
[45], [46], [50], [51]. The symbolic calculus for meromorphic pseudodifferential
operators is an analog of the symbolic calculus of Kohn-Nirenberg [32] for classi-
cal pseudodifferential operators. Theorem 3.5.1 is close to the corresponding
theorem for classical pseudodifferential operators (cf., e.g., [67], Vol. 2).
279
280 Bibliographical sketch

Chapter 4 is based on Plamenevskii [45], [47], [49]. In Chapter 5 results of joint

work of Plamenevskii and Y.N. Senichkin [54] - [57] are studied, and in Chapter 6
- such results from [58] - [60].
For one-dimensional singular integral operators, other operator symbols and
the various reflections based on them see [18] (smooth contour, discontinuous
coefficients) and [20] (compound contour). In the case of a smooth contour the
symbols in [I8] are unitarily equivalent to the symbols in Chapter 4, §4.5 (formula
(4.4.10)). If singularities are allowed on the contour, the symbols from [20] and
those defined in Chapter 4, §4.5 (in accordance with [56]) essentially differ. The
symbols in [20] do not realize, in general, irreducible representations of the alge-
bra of singular integral operators, and the algebra of symbols is not * -isomorphic
to the quotient of the algebra of singular integral operators by the ideal of com-
pact operators. The literature devoted to one-dimensional singular integral opera-
tors with discontinuous coefficients is very extensive. We yet mention the work of
R.Y. Duduchava [26], [27], and the fundamental articles [64], [65] of I.B.
Simonenko. One-dimensional operators play the role of examples in the present
book, and our list of references is not intended to serve as a reliable orientation in
this direction.
 References

[1] M.S. AORANOVICH; Elliptic singular integro-differential operators, Uspekhi

Mat. Nauk SSSR 20, No.5 (1965), 3-120. (Translation: Russian Math. Sur-
veys 20, No.5 (1965), 1-121.)
[2] M.S. AORANOVICH, M.I. VISHIK; Elliptic problems with a parameter and para-
bolic problems oj general type, Uspckhi Mat. Nauk SSSR 19, No.3 (1964),
53-161. (Translation: Russian Math. Surveys 19, No.3 (1964),53-157.)
[3] F.V. ATKINSON; Normal solvability oj linear equations in normed spaces, Mat.
Sb. 28, No.1 (1951), 3-14 (in Russian).
[4] K.I. BABENKO; On dualJunctions, Dokl. AN SSSR 62, No.1 (1948), 157-160
(in Russian).
[5] H. BATEMAN, A. ERDa-Y, ET AL.; Higher transcendental Junctions, Vol. 1,
McGraw-Hill, 1953.
[6] H. BATEMAN, A. ERDa-Y, ET AL.; Higher transcendental Junctions, Vol. 2,
McGraw-Hill, 1953.
[7] S. BOCHNER; Theta relations with spherical harmonics, Proc. Nat. Acad. Sci.
USA 37 (1951),804-808.
[8] N.L. VASlLEVSKII; On the algebra generated by two-dimensional integral opera-
tors with Bergman kernel and piecewise continuous coefficients, Dokl. AN SSSR
271, No.5 (1983), 1041-1044. (Translation: Soviet Math.-Dokl. 28, No.1
(1983), 191-194.)
[9] N .P. VEKUA; Systems oj singular integral equations and certain boundary value
problems, Noordhoff, 1967 (translated from the Russian).
[10] N.YA. VILENKIN; Special Junctions and the theory oj group representations,
Amer. Math. Soc., 1968 (translated from the Russian).
[11] V.S. VLADIMIROV; GeneralizedJunctions in mathematical physics, Nauka, 1976
(in Russian).
[12] F.D. GAKHOV; Boundary value problems, Pergamon, 1966 (translated from the
Russian).
[13] I.M. GEL'FAND, G.E. SHILOV; GeneralizedJunctions, Vol. 1, Acad. Press, 1964
(translated from the Russian).
281
282 References

[14] LM. GEL'FAND, N.YA. VlLENKlN; Generalized functions, Vol. 4, Acad. Press,
1965 (translated from the Russian).
[15] LM. GEL'FAND; Eigenfunction expansions of equations with periodic coefficients,
Dok!. AN SSSR 73, No.6 (1950), 1117-1120 (in Russian).
[16] LTs. GoKHBERG, M.G. KREIN; Introduction to the theory of linear nonselfad-
joint operators, Amer. Math. Soc. 1969 (translated from the Russian).
[17] LTs. GoKHBERG, N.YA. KRUPNIK; EinfUhrung in die Theorie der eindimen-
sionalen singuliiren Integraloperatoren, Birkhauser, 1979 (translated from the
Russian).
[18] LTs. GOKHBERG, N.YA. KRUPNIK; Algebra generated by one-dimensional
singular integral operators with piecewise continuous coefficients, Funkts. Anal. i
Prilozhen. ~,No. 3 (1970), 26-36. (franslation: Funct. Anal. Appl. ~, No.3
(1970), 193-201.)
[19] LTs. GOKHBERG, N.YA. KRUPNIK; On the algebra generated by Toeplitz
matrices, Funkts. Anal. i Prilozhen. 3, No.2 (1969), 46-56. (franslation:
Funct. Anal. Appl. ~,No. 2 (1969), 119-127.)
[20] LTs. GoKHBERG, N.YA. KRUPNIK; On singular integral operators on a com-
pound contour, Soobshch. AN GSSR 64, No.1 (1971), 21-24 (in Russian).
[21] LTs. GoKHBERG, E.I. SIGAL; An operator generalization of the logarithmic resi-
due theorem and the theorem of Rouche, Mat.-Sb. 84, No.4 (1971), 607-629.
(franslation: Math. USSR-Sb. 13 (1971), 603-625.)
[22] V.V. GRUSHlN; Pseudodifferential operators in R n with bounded symbols,
Funkts. Anal. i Prilozhen. 4, No.3 (1970), 37-50. (franslation: Funct. Anal.
Appl. ~,No. 3 (1970), 202-212.)
[23] A.O. DERVIZ; Boundary-value problems for meromorphic pseudodifferential
operators, Izv. Vyzov. Mat. ~ (1985), 84-86 (franslation: Soviet Math.-Izv.
29, No.3 (1985),108-111).
[24] J. DIXMIER; C* -algebras, North-Holland, 1977 (translated from the French).
[25] R.G. DOUGLAS; Banach algebra techniques in the theory of Toeplitz operators,
Amer. Math. Soc., 1973.
[26] R.V. DUDUCHA VA; On integral operators of convolution type with discontinuous
symbols, Math. Nachr. 79 (1977), 75-98 (in Russian).
References 283

[27] R.V. DUOUCHAVA; On bisingular integral operators with discontinuous

coefficients, Mat. Sb. 30, No.4 (1976), 584-609. (franslation: Math. USSR-
Sb. 30, No.4 (1976),515-537.)
[28] V.P. IL'IN; On inequalities between the norms oj partial derivatives of functions
of several variables, Tr. Mat. Inst. Steklov. 84 (1965), 144-173 (franslation:
Proc. Steklov Inst. Math. 84 (1965), 161-193 (edited by V.N. Faddeeva).)
[29] A.P. CALDERON, A. ZYGMUND; Singular integral operators and difJerential
equations, Amer. J. Math. 79, No.4 (1957), 901-921.
[30] L.V. KANTOROVICH, G.P. AKU.oV; Functional analysis in normed spaces, Per-
gamon, 1964 (translated from the Russian).
[31] A.I. KOMECH; Elliptic boundary value problems for pseudodifJerential operators
on manifolds with conical points, Mat. Sb. 86, No.2 (1971), 268-298. (frans-
lation: Math. USSR-Sb. 15, No.2 (1971), 261-297.)
[32] J.J. KOHN, L. NIRENBERG; An algebra of pseudo-difJerential operators, Comm.
Pure Appl. Math. .!!, No. 112 (1965), 269-305.
[33] V.A. KONDRATEV; Boundary problems for elliptic equations in domains with
conical or angular points, Tr. Moskov. Mat. Obshch. 16 (1967), 209-292.
(franslation: Trans. Moscow Math. Soc. 16 (1967), 227-314.)
[34] H.O. CoRDES, DA. WILLIAMS; An algebra of pseudo-difJerential operators with
non-smooth symbols, Pacific J. Math. 78, No.2 (1978), 279-291.
[35] V.G. MAz'yA, BA. PLAMENEVSKlI; Lp-estimates of solutions of elliptic boun-
dary value problems in domains with edges, Tr. Moskov. Mat. Obshch. 37
(1978), 49-93. (franslation: Trans. Moscow Math. Soc. 37 (1978), 49-98.)
[36] V.G. MAz'YA, B.A. PLAMENEvSKII; Elliptic boundary value problems on mani-
folds with singularities, Probl. Mat. Anal. ~ (1977), 85-142 (in Russian).
[37] V.G. MAZ'YA, T.O. SHAPOSHNIKOVA; Theory of multipliers in spaces of
differentiable functions, Pitman, 1985 (translated from the Russian).
[38] B. MUCKENHOUPT, R WHEEDEN, WO-SANG YOUNG; Lrmultipliers with
power weights, Adv. Math. 49, No.2 (1983), 170-216.
[39] N .M. MlKHAILOvA-GUBENKO; Singular integral equations in Lipschitz spaces
II. Vestn. Leningr. Gos. Univ. Ser. Mat. Mekh. Astr. 21, No. 7 (1966),
45-57 (in Russian).
284 References

[40] S.G. MlKHLIN; Multidimensional singular integrals and integral equations,

North-Holland, 1970 (translated from the Russian).
[41] N.I. MUSKHELISHVILI; Singular integral equations, Noordhoff, 1953 (translated
from the Russian).
[42] SA. NAZAROV; Elliptic boundary value problems with periodic coefficients in a
cylinder, Izv. AN SSSR Ser. Mat. 45, No.1 (1981), 101-112. (Translation:
Math. USSR-Izv. 18, No.1 (1982), 89-98.)
[43] B.A. PLAMENEVSKII; Singular integral equatiOns in a cone, Dokl. AN SSSR
179, No.5 (1968), 1057-1059. (Translation: Soviet Math. Dold. 9, No.2
- -
(1968), 509-511.)
[44] B.A. PLAMENEvSKII; On the boundedness of singular integrals in weighted
spaces, Mat. Sb. 76, No.4 (1968), 573-592 (in Russian).
[45] B.A. PLAMENEVSKII; On an algebra of pseudodifJerential operators in spaces
with weighted norms, Mat. Sb. 106, No.2 (1978), 296-320. (Translation:
Math. USSR-Sb. 34 (1978),841-865.)
[46] B.A. PLAMENEVSKII; On meromorphic pseudodifJerential operator{unctions, Izv.
Vyzov. Mat. ~ (1978), 79-90 (in Russian).
[47] B.A. PLAMENEVSKII; On algebras generated by pseudodifJerential operators with
isolated singularities of symbols, Dok!. AN SSSR 248, No.2 (1979), 297-302.
(Translation: Soviet Math.-Dold. 20, No.5 (1979), 1013-1017.)
[48] B.A. PLAMENEVSKII; On boundary value problems for meromorphic
pseudodifJerential operators, Izv. Vyzov. Mat. ~ (1980), 69-78. (Translation:
Soviet Math.-Izv. 24, No.4 (1980), 77-87.)
[49] B.A. PLAMENEVSKII; On algebras generated by pseudodifJerential operators with
isolated singularities of symbols, Probl. Mat. Fiz 10 (1982), 209-241. (Transla-
tion: Se1ecta Math. Sov. 5, No.1 (1986), 77-100.)
[50] B.A. PLAMENEVSKII; On pseudodifJerential operators on H n with discontinuities
of the second kind in the symbols, Izv. Vyzov. Mat. 12 (1982), 30-38 (in Rus-
sian).
[51] B.A. PLAMENEVSKII; On traces ofpseudodifJerential operators, Izv. Vyzov. Mat.
g (1982) 30-38. (Translation: Soviet Math.-Izv. 26, No. 12 (1982),27-37.)
[52] B.A. PLAMENEVSKII; On the index of pseudodifJerential operators with isolated
singularities of symbols in Rn, Dold. AN SSSR 263, No.5 (1982), 1062-1065.
References 285

(franslation: Soviet Math.-Dokl. 25, No.2 (1982), 494-497.)

[53] BA. PLAMENEVSKu; Estimates for a convolution operator in spaces with
weighted norms, Dokl. AN SSSR 286, No. 1 (1986), 36-39. (franslation:
Soviet Math. Dokl. 33, No.1 (1986),25-29.)
[54] B.A. PLAMENEVSKII, V.N. SENICHKIN; On the spectrum of c· -algebras gen-
erated by pseudodijJerential operators with isolated singularities of symbols,
Dokl. AN SSSR 261, No.6 (1981), 1304-1306. (franslation: Soviet Math.-
Dokl. 24, No.3 (1981), 686-689.)
[55] B.A. PLAMENEvSKII, V.N. SENICHKIN; On the spectrum of c· -algebras gen-
erated by pseudodijJerential operators with discontinuous symbols, Izv. AN
SSSR Ser. Mat. 47, No.6 (1983), 1263-1284. (Math. USSR-Izv. 23, No.3
(1984), 525-544.)
[56] B.A. PLAMENEVSKII, V.N. SENICHKIN; On C· -algebras of singular integral
operators with discontinuous coefficients on a complex contour 1, Izv. Vyzov.
Mat. 1. (1984), 25-33; II, Izv. Vyzov. Mat ~ (1984), 37-46. (franslations:
Soviet Math.-Izv. 28, No.1, (1984), 28-37; Soviet Math.-Izv. 28, No. 4
(1984), 47-58).
[57] B.A. PLAMENEVSKII, V.N. SENICHKIN; On the spectrum of a C·-algebra of
pseudodifferential operators with singularities in the symbols, Math. Nachr. 121
(1985),231-268 (in Russian).
[58] B.A. PLAMENEVSKII, V.N. SENICHKIN; The spectrum of an algebra of
pseudodijJerential operators with discontinuous symbols on a manifold with boun-
dary, Dokl. AN SSSR 277, No.6 (1984), 1327-1330. (franslation: Soviet
Math.-Dokl. 30, 1 (1984), 286-289.)
[59] B.A. PLAMENEVSKII, V.N. SENICHKIN; On the spectrum of a C·-algebra of
pseudodifferential operators with discontinuous symbols on a manifold with boun-
dary, Probl. Mat. Fiz. .!.! (1985), 178-209 (in Russian).
[60] BA. PLAMENEVSKII, V.N. SENICHKIN; On the spectrum of an algebra of
pseudodifferential operators in a half·space, Probl. Mat. Anal. 10 (1986),
160-179 (in Russian).
[61] B.A. PLAMENEVSKII, V.N. SENICHKlN; Spectra of C·-algebras of
pseudodijJerential operators with multidimensional singularities in symbols,
Funkts. Anal. i Prilozheo. 20, No.4 (1986), 85-86. (franslation: Funct. Anal.
286 References

Appl. 20, No.4 (1986), 328-329.)

[62] S.G. SAMKO; Hypersingular integrals and their applications, Izd. Rostok. Univ.,
1984 (in Russian).
[63] S.G. SAMKO; Singular integrals over the sphere and construction of the charac-
teristics with respect to the symbols, Izv. Vywv. Mat. ~ (1983), 28-42. (Trans-
lation: Soviet Math.-Izv. 27, No.4 (1983), 35-52.)
[64] LB. SIMONENKO; A new general method for studying linear operator equations
of the type of singular integral equatiOns, I, Izv. AN SSSR Ser. Mat. 29, No.3
(1965), 567-586 (in Russian).
[65] LB. SIMONENKO; On global and local factorability of a measurable matrix func-
tion and the Noetherian property induced by its singular operator, Izv. Vywv.
Mat. ~ (1983), 81-87. (Translation: Soviet Math.-Izv. 27, No.4 (1983), 99-
106.)
[66] E.M. STEIN; Note on singular integrals, Proc. Amer. Math. Soc. ~, No.2
(1957), 250-254.
[67] F. TRiNEs; Introduction to pseudodifferential and Fourier integral operators,
Vol. 1-2, Plenum, 1980.
[68] M. TAYLOR; Pseudo-differential operators, Springer, 1974.
[69] K.O. FRIEDRICHS; Pseudo-differential operators, Courant lnst. Math. Sci. Lec-
ture Notes, 1970.
[70] Yu.E. KHArK.IN; On operators of convolution type in weighted spaces, Vestn.
Leningr. Gos. Univ. Ser. Mat. Mekh. Astr. Q (1969) (in Russian).
[71] B.V. KHvEDEUDZE; The method of Cauchy-type integrals in the discontinuous
boundary-value problems of the theory of holomorphic functiOns of one complex
variable, Itogi Nauk. i Tekhn. Sovr. Probl. Mat. 7.. (1975), 5-162. (Transla-
tion: J. Soviet Math. 7. (1977), 309-415.)
[72] L. HORMANDER; Pseudo-differential operators, Comm. Pure Appl. Math. ~
(1965),501-517.
[73] M.A. SHUBIN; Pseudodifferentialoperators and spectral theory, Springer, 1987
(translated from the Russian).
[74] G.!. EsKIN; Boundary value problems for elliptic pseudodifferential equations,
Amer. Math. Soc., 1981 (translated from the Russian).
 Index

Asymptotic series for a meromorphic pseudodifferential operator 89

C* -algebra, representation of a 183

C* -algebra, spectrum of a 185
C*-algebras, isomorphism of 182
C* -algebras, morphism of 182
Calderon-Zygmund formula 20
Calkin algebra 209
Canonical meromorphic pseudodifferential operator 74
Canonical meromorphic pseudodifferential operator, trace of a 99
Canonical periodic meromorphic pseudodifferential operator 126
Canonical periodic meromorphic pseudodifferential operator, symbol of a 126
Canonical pseudodifferential operator on Hp(lRn) 135
Characteristic of a singular integral operator 20
Complete symbol of a (meromorphic) pseudodifferential operator 89
Complete symbol of a pseudodifferential operator on Hp (IRn) 135
Cone 146
Conic manifold 146
Conic set 146
Conic structure 146
Criterion for invertibility of an element of an algebra 186
Cusp 177
Cyclic vector 183

(Laplace-Beltrami operator)
l) 23
!::.. (Laplace operator)
23
Dimension of a representation 183

t9p(lRn) 47
E(>-)-1 16
~(-i(k +nI2»-1 l7
E(i(k +nI2» 9
E(>-) 8
Elliptic merom orphic pseudodifferential operator 96
287
288 Index

Elliptic periodic meromorphic pseudodifferential operator 129

Elliptic pseudodifferential operator 139
Elliptic pseudodifferential operator on Hpem.,) 167
Elliptic pseudodifferential operator on Hp(X) 150
Equivalence of representations 183

Finite-meromorphic operator-function 76
Fourier transform 9

G (transform) 122
Garding's inequality, generalized 116
Generalized Garding's inequality 116

H/J(~) (0lL a manifold with conical points 164

Hp(lRm,lR m -n) 47
Hp(lRn) 38
Hp(X) (X a cone) 149
HS(~) 23
HS(A,Sn -I) 28
HS(lRn) 22

Involutive Banach algebra 182

Involutive algebra 182
Involutive normed algebra 182
Irreducible representation 183
Isomorphism of C* -algebras 182

Jacobson topology 185

Laplace operator 23
Laplace-Beltrami operator 23

Manifold with conical points 164

Mellin transform 9
Meromorphic pseudodifferential operator (of order ao) 89
Meromorphic pseudodifferential operator of order - 00 96
Meromorphic pseudodifferential operator on IR n 147
Meromorphic pseudodifferential operator on a cone 147
Meromorphic pseudodifferential operator with variable symbol 156
Index 289

Meromorphic pseudodifferential operator, asymptotic series for a 89

Meromorphic pseudodifferential operator, canonical 74
Meromorphic pseudodifferential operator, canonical, trace of a 99
Meromorphic pseudodifferential operator, complete symbol of a 89
Meromorphic pseudodifferential operator, elliptic 96
Meromorphic pseudodifferential operator, periodic 126
Meromorphic pseudodifferential operator, periodic, canonical 126
Meromorphic pseudodifferential operator, principal symbol of a 89
Meromorphic pseudodifferential operator, symbol of a 89
Meromorphic pseudodifferential operator, trace of a 102
Morphism of C*-algebras 182

Nuclear operator 97
Nuclear operator, trace of a 97

One-dimensional singular integral operator 177

Operator ~ Laplace 23
Operator 8, Laplace-Beltrami 23
Operator E(A) - 1 16
Operator ~(-i(k +nI2»-1 17
Operator E(i (k + n 12» 9
Operator EG\) 8
Operator symbol of a pseudodifferential operator 169
Operator symbol of a singular integral operator 178
Operator-function, finite-meromorphic 76

Parseval's equality (for the Mellin transform) 9

Periodic meromorphic pseudodifferential operator 126
Periodic meromorphic pseudodifferential operator, canonical 126 126
Periodic meromorphic pseudodifferential operator, elliptic 129
Piecewise smooth contour 177
Primitive ideal (two-sided) 183
Principal symbol of a (meromorphic) pseudodifferential operator 89
Principal symbol of a pseudodifferential operator on Hp(lRn) 135
Pseudodifferential operator on Hp(':)lL) 165
Pseudodifferential operator on Hp(':)lL), elliptic 167
Pseudodifferential operator on Hp(lRn) 134
Pseudodifferential operator on Hp (IRn), (complete) symbol of a 135
Pseudodifferential operator on Hp{lRn), canonical 135
290 Index

Pseudodifferential operator on Hp (IR n), principal symbol of a 135

Pseudodifferential operator on Hp(X) 150
Pseudodifferential operator on Hp (X), elliptic 150
Pseudodifferential operator on IR , meromorphic
n 147
Pseudodifferential operator on a cone, meromorphic 147
Pseudodifferential operator, asymptotic series for a meromorphic 89
Pseudodifferential operator, canonical meromorphic 74
Pseudodifferential operator, complete symbol of a meromorphic 89
Pseudodifferential operator, elliptic 139
Pseudodifferential operator, meromorphic 89
Pseudodifferential operator, meromorphic, elliptic 96
Pseudodifferential operator, meromorphic, of order - 00 96
Pseudodifferentialoperator, meromorphic, periodic 126
Pseudodifferential operator, meromorphic, periodic, canonical 126
Pseudodifferential operator, meromorphic, trace of a 102
Pseudodifferentialoperator, meromorphic, with variable symbol 156
Pseudodifferential operator, operator symbol of a 169
Pseudodifferential operator, principal symbol of a meromorphic 89
Pseudodifferential operator, symbol of a merom orphic 89

Regular point on a contour 177

Representation (of a C* -algebra) 183
Representation space 183
Representation, dimension of 183
Representation, irreducible 183
Representations, equivalence of 183
Rich sub algebra 186

Singular integral operator on a composite contour 225

Singular integral operator, (operator) symbol of a 178
Singular integral operator, characteristic of a 20
Singular integral operator, one-dimensional 177
Singular integral operator, symbol of a 20
Singular point on a contour 177
Space Hp (GJrL) (GJrL a manifold with conical points) 164
Space Hp (IR m, IR m -n) 47
Space Hp(lRn) 38
Space Hp(X) (X a cone) 149
Space HS(GJrL) 23
Index 291

Space H S (X, sn - J) 28
Space HS(lRn) 22
Space 0p(lRn) 47
Spectrum of a C* -algebra 185
Symbol of a (meromorphic) pseudodifferential operator 89
Symbol of a (meromorphic) pseudodifferential operator, complete 89
Symbol of a (meromorphic) pseudodifferential operator, principal 89
Symbol of a canonical periodic meromorphic pseudodifferential operator 126
Symbol of a pseudodifferential operator on Hp(lRn) 135
n
Symbol of a pseudodifferential operator on Hp{lR ), complete 135
Symbol of a pseudodifferential operator on Hp{lR n), principal 135
Symbol of a singular integral operator 178
Symbol of a singular integral operator 20

To-space 185
Totalizing vector 183
Trace of a canonical meromorphic pseudodifferential operator 99
Trace of a meromorphic pseudodifferential operator 102
Trace of a nuclear operator 97
Transform G 122
Transform, Fourier 9
Transform, Mellin 9
Transversal operator (for an operator) 51
Typical neighborhood 223

Young's inequality 26