Download as pdf or txt
Download as pdf or txt
You are on page 1of 405

ILBERT SPACE METHODS

. . . . . . . . QUANTUM MECHANICS
FuNDAMENTAL SciENCES

ILBERT SPACE METHODS


QUANTUM MECHANICS
Werner 0. An1rein

PFL Press
A Swiss academic publisher distributed by CRC Press
Taylor and Francis Group, LLC
6000 Broken Sound Parkway, NW, Suite 300,
Boca Raton, FL 33487

Distribution and Customer Service


orders@crcpress.con1

www.crcpress.com

Library of Congress Cataloging-in-Publication Data


A catalog record for this book is available fron1 the Library of Congress.

This book is published under the editorial direction of


Professor Philippe-Andre Martin (EPFL, Lausanne).

is an in1print owned by Presses polytechniques et universitaires romandes, a


Swiss acade1nic publishing company whose n1ain purpose is to publish the
teaching and research works of the Ecole polytechnique federale de Lausanne.

Presses polytechniques et universitaires romandes


EPFL- Centre Midi
Post office box 119
CH-1015 Lausanne, Switzerland
E-Mail: ppur@epfl.ch
Phone: 021 I 693 21 30
Fax: 021/693 40 27

www.epflpress.org

© 2009, First edition, EPFL Press


ISBN 978-2-940222-35-3 (EPFL Press)
ISBN 978-1-4200-6681-4 ( CRC Press)

Printed in Spain

All right reserved (including those of translation into other languages). No part
of this book n1ay be reproduced in any fonn- by photoprint, microfilm, or any
other 1neans - nor transn1itted or translated into a n1achine language without
written pennission fro111 the publisher.
Preface

This text is based on lectures given at the University of Geneva during the period
1994- 2005. These courses were intended for advanced undergraduate students who
had completed a first course in quantum mechanics and were thus expected to be fa-
miliar with the physical aspects and the basic 1nathernatical formalism of quantum
theory. Partly due to lack of time, quantum mechanics is often taught with no or little
exposition of the mathematical questions arising through the introduction of infinite-
dimensional Hilbert spaces. I hope that the present volume will prove useful, espe-
cially to somewhat theoretically minded students, for deepening their knowledge and
understanding of the Hilbert space aspects of quantum mechanics, and prepare them
for reading research papers.
Mostly the lectures were organised as one-year courses (80 hours plus 25 hours of
problem sessions) and covered essentially the contents of Chapters 1- 5 and parts of
Chapters 6 or 7. To offer a few more applications, some 1naterial has been added to the
original lecture notes (Sections 5.8, 6.6, 6.7, 7.2 and 7.5). Of course a strict selection
of topics and applications to be treated had to be made from the outset. The emphasis is
placed on a certain number of basic mathematical techniques, usually without striving
for the most general results. For example there is no discussion of quadratic forms,
and we have avoided the use of techniques from stochastic analysis. However we give
essentially complete proofs for all results involving Hilbert space objects. Some of
these proofs are collected in appendices to the various chapters. Son1e acquaintance
with measure theory is required: the essential facts are explained without detailed
proofs.
Chapter 1 gives the basic properties of Hilbert space and a description of the neces-
sary material from measure theory. In Chapter 2 we present various classes of bounded
linear operators and general notions on unbounded operators, including the invariance
of self-adjointness under a class of perturbations. The problem of self-adjointness is
further investigated in Chapter 3 which contains the theory of extensions of symmetric
operators, with applications to Sturm-Liouville and Schrodinger operators. Chapter 4
deals with the spectral theory of self-adjoint operators, in particular with the spectral
theorem and with the various spectral types. In Chapter 5 we prove Stone's Theo-
rem and then discuss the fundamental aspects of scattering theory: scattering states,
asymptotic condition and wave operators, S-matrix, scattering cross sections. Chapter
6 is devoted to the Mourre method for controlling the resolvent of self-adjoint oper-
ators near the real axis and to implications for their spectrum. In the final Chapter 7
PREFACE

we present stationary-state scattering theory and various applications of the results of


Chapter 6: asymptotic completeness, properties of the S-matrix, time delay and the
Flux-Across-Surfaces Theorem.
Each chapter ends with some bibliographical notes and a selection of problems.
The bibliography consists mostly of books. They are referred to for alternative or more
advanced presentations of some material or for certain points not treated in the present
text. A very small number of original and review papers are cited for those interested
a deeper understanding of certain topics treated in the text. In combination with
tnodern electronic means. these papers can also be useful as a basis for searching in
the vast literature.
The majority of the problems have been tested in class-room sessions. Many of
then1 are meant to help students to become familiar with the concepts discussed in the
main body; son1e require a more detailed study of technical aspects of the text. A few
of the more difficult problems are provided with a hint for the solution.
As regards notations, it should be pointed out that some symbols have more than
one 1neaning in the text. In particular: the symbol II · II is used for various norms, the
Greek letter CJ for spectra and for scattering cross sections, the letter P for projections
and for mon1entum, and the letter R for resolvents and for S - I, where S is the
scattering operator. For the convenience of the reader we have included a Notation
Index (page 385). Constants are often denoted generically by c, but in some proofs
different constants are numbered as c1 , c2 , etc.
is a great pleasure to thank all those who helped me in one way or another during
the preparation of the lectures or of this volume. The problems in the text are mostly
due to Marius Mantoiu, Joachim Stubbe and Rafael Tiedra de Aldecoa; they assumed
the responsibility for the problem sessions with much competence and devotion. The
continual interest of my students and their constructive comments on earlier versions
of the text have influenced a considerable number of details. I received precious sup-
port from Philippe Jacquet, Andreas Malaspinas, Peter Wittwer and Luis Zuleta for
coming to grips with TEX -related difficulties. I thank Philippe Martin for proposing
the publication of my lectures, the referee for pointing out some errors and for useful
suggestions, and Fred Fenter from EPFL Pre,ss for advice and his very efficient man-
agement of the publishing process. Finally~' am indebted to the Physics Department
of the University of Geneva for its kind hospitality after my retirement.

Werner Amrein
Geneva, Switzerland
December 2008
Contents

1 Hilbert Spaces 1
1.1 Definition and elementary properties 1
1.2 Vector-valued functions . . . . . . . . 6
1.3 Subsets and dual of a Hilbert space . . 8
1.4 Measures, integrals and LP spaces 14
Problems . . . . . . . . . . . . . 25

2 Linear Operators 27
2.1 The algebra B(H) . . . . . . 27
2.2 Projections and isometries 37
2.3 Compact operators . . . . 41
2.4 Unbounded operators . . . . 47
2.5 Multiplication operators . . . . . . 56
2.6 Resolvent and spectrum of an operator .. 67
2. 7 Perturbations of self-adjoint operators 7
Appendix ... . 80
Problems . . . . . . . . . . . . . . . . . 82

3 Symmetric Operators and their Extensions 87


3.1 The method of the Cay ley transform . . 8 . . . . 87
3.2 Differential operators with constant coefficients 96
3.3 Schrodinger operators . . . . . . . 111
Appendix 126
Problems . . . . . . . 131

4 Spectral Theory of Self-Adjoint Operators 133


4.1 Stieltjes measures . . . . . . . . . . . . . 133
4.2 Spectral measures . . . . . . . . . . . . . 141
4.3 Spectral parts of a self-adjoint operator . 157
4.4 The spectral theorem. The resolvent near the spectrum . . . . 0 • • • 170
Appendix: Proof of the Spectral Theorem .... 179
Problems . . . . . . . . . . . . . . . . . . . . . . . . . 190
Vlll CONTENTS

5 Evolution Groups and Scattering Theory 193


5.1 Evolution groups 193
5.2 Characterisation of the scattering states .. 204
5.3 Asymptotic condition. Wave operators . 213
5.4 Simple scattering systems. Scattering operator . . 218
5.5 Scattering operator and S- matrix . 226
5.6 Scattering cross sections 235
5.7 Bounds on scattering cross sections 243
5.8 Coulomb scattering .. 252
Problems .. . .. 260

6 The Conjugate Operator Method 263


6.1 A simple example . 264
6.2 The method of differential inequalities . 271
6.3 The Mourre inequality 283
6.4 Application to Schrodinger operators .. 286
6.5 Relatively smooth operators 292
6.6 Higher order resolvent estimates 300
6.7 Some comtnutators .. 306
Appendix: Interpolation of operators 310
Problems 315

7 Further Topics in Scattering Theory 317


7.1 Asymptotic completeness . 317
7.2 Flux and scattering into cones 324
7.3 Time-independent scattering theory ... 343
7.4 The scattering matrix 352
7.5 Time delay .. 364
Appendix 374
Problems 379

References I 381

Notation Index 385

Subject Index 389


CHAPTER!

Hilbert Spaces

Hilbert space sets the stage for standard quantum theory: the pure states of a physical
system are identified with the unit rays of a Hilbert space 1{ and observables with self-
adjoint operators acting in H. In this initial chapter we present the essential concepts
and prove the basic results concerning separable Hilbert spaces (Sections 1.1 - 1.3 ). In
Section 1.4 we then introduce £ 2 spaces, which are of special importance for quantum
mechanics. This requires some familiarity with measure theory, and we include a short
description of the necessary concepts from this theory.

1.1 Definition and elementary properties

Throughout this text a Hilbert space 1neans a co1nplex linear vector space,
equipped with a Hermitian scalar product, which is complete and admits a countable
basis. More precisely a (separable) Hilbert space His defined by the four postulates
(HI)- (H4) stated below:
(Hl) H is a linear vector space over the field CC o.f co1nplex nu1nbers:
With each couple {f, g} of elements of H there is associated another element of H,
denoted f + g, and with each couple {a, f}, a E CC, f E H, there is associated an ele-
ment af ofH, and these associations have the following properties (where f, hE 1{
and a, {3 E CC):

f+g==g+f f + (g + h) ==
(f +g) + h (1.1)
a(f+g)==af+ag (a+(3)f-af+(3f (1.2)
a({3f) == (a{3)f If- f. (1.3)

1
Furthermore there exists a unique eletnent 0 E H (called the zero vector) such that

O+f==f Of== 0 \If E H. (1.4)


1
Here 0 denotes the complex number a = 0.
2 HILBERT SPACES

1i is equipped vvith a strictly positive scalar product 2 :


With each couple {j, g} of elements ofH there is associated a complex number (f, g),
and this association has the following properties 3 :

(q, f) == (.f, g) \lj,gEH (1.5)


\f, g + o:h) == \f, g) + a(f, h) vaEC, \lj,g,hEH (1.6)
\J,J)>O except for f == 0. (1.7)

One then defines


(1.8)

(H3) 1i is complete:
Each Cauchy sequence in 1i has a lin1it in H. In other terms, if {fn}nEN is a sequence
elements of 1{ satisfying 4

li1n II fn - f m II
m,n-----+c:x:J
== o, s (1.9)

then there exists an element f of 1i such that lim n-----+oo II f - fn II == 0.


(H4) 1{ is separable:
1i has a countable orthonormal basis, i.e. there exists a sequence { e 1 , e 2 , e 3 , ... } of
elen1ents of 1{ such that
( 1.1 0)
and such that each element f of 1i can be expressed as a linear combination of
' ...
The dimension of a Hilbert space 1i is defined as the number of elements in an or-
thonormal basis. This number 1nay be finite or infinite, and it is independent of the cho-
sen basis. In an infinite-dimensional space the postulate (H4) involves infinite linear
co1nbinations. These are interpreted as follows: if { e 1 , e 2 , e3 , ... } is an orthonormal
basis of 1i and f an arbitrary element of H, there E}Xists a sequence {a1 , a2 , a3 , ... } of
complex numbers such that f == Lkakek and llifll 2 == Lklakl 2 . The coefficient ak
this develop1nent off is given by O'k == (ek, f), and the series Lkakck converges
to f in the sense that II f - L~=l akck II ---+ 0 as N ---+ x.
elements of 1i are called vectors; they will be denoted by f, g or h (sometimes
e vectors satisfying II ell == 1). The letters a and f3 will stand for complex num-
bers. reader should be fatniliar with finite-dimensional Hilbert spaces from Lin-
ear Algebra. However many problems in quantu1n mechanics involve Hilbert spaces
infinite dimension. We shall see that, when suitably interpreted, some properties of
finite-ditnensional spaces have analogues in infinite-din1ensional spaces (for example
spectral properties of compact operators are similar to those of tnatrices ), while
2
Abo called "inner product" by sotne authors.
1
a denotes the complex conjugate of the cmnplex nutnber a.
-+ N { 1, 2. ~3 .... } .
5
i.e. for each E > 0 there exists a number N N(s) > 0 such that llfn - fm II < E \In, m > N.
DEFINITION AND ELEMENTARY PROPERTIES 3

other aspects of infinite-dimensional spaces do not appear in finite-dimensional ones


(for example in finite-dimensional spaces there is no distinction between weak and
strong convergence, and matrices do not have continuous spectrum).
Observe that, contrary to the convention adopted by mathematicians, the scalar
product is linear in the second entry and anti-linear in the first one. The non-negative
number I f I defined by ( 1.8) is called the norm of the vector This norm function
has the properties of a metric. The quantity II!- g\1 can be interpreted as the distance
fromftog, andonehas llafll == \cvlllfll, in particular 11-fll 11(-l)fll == llfll· The
scalar product can be expressed in terms of the norm of 1-i; this is the content of the
polarisation identity:

4\f,g) ==\if+ gl\ 2 -II!- gll 2 - ill!+ ig\\ 2 + illf- igll 2 · (1.11)

This equation is easily checked by using the definition of the norn1 and the linearity of
the scalar product [as in ( 1.16) below].
The following four inequalities are simple consequences of (H1) and (H2) 6 :

Schwarz inequality : 1\f,g)l < llfii·IIDII (1.12)


Triangle inequality : \if+ g\1 < 11!11 +\\gil (1.13)

II!+ gll 2 < 2\1!11 2 + 2llg\\ 2 (1.14)

IIIJII-IIgll\ < llf- g\1. ( 1.15)

PROOF. Iff== g, (1.12) is evident: (J,f) == llfl\ 2 . Iff#- g, assume for example
that g #- 0. Then, for any a E C:

0 <\If +ag\\ 2 == \f +ag,f ag) == 11!11 2 a(f,g) +a(g,f) la\ 2 \lg\\ 2 . (1.16)

By taking a== -(f, g) /llgll 2 one obtains


o < 11!11 2 - : 21(.f,gW,
11 1
which implies (1.12) upon multiplication by llgll 2 . For (1.13) we take a== 1 in (1.16):
\If+ gl\ 2 == llf\1 2 + \J,g) + (g,f) + llg\\ 2 (1.17)
2 2
< llfl\ + I(J,g)l + \(g,f)\ + \\g\\
< llfl\ 2 + 2llfll·llgl\ + \\g\1 2 == (llfll + llg\1) 2 .
For (1.14) we observe that
o < llf- g\\ 2 == llfll 2 - \f,g)- (g,.f) + llgll 2 ,
hence

In ( 1.12), II f II · II g II denotes the product of the norms off and g. In the remainder of the text a product
6

of two norms will be written simply as 11·1111·11· With this convention, (1.12) will read l(f, g)J:::;; llfllllgJJ.
4 HILBERT SPACES

Upon insertion this inequality into ( 1.17) one obtains ( 1.14). Finally, to obtain
(1.15), supposeforexan1plethat Jlgll < 11!11; then, by using (1.13):

11!11- llgll == II!- g +gil llgll < (II!- gil+ II gil)- llgll == II!- gil· o

1.1 . 2. There are two types of convergence in a Hilbert space, called strong conver-
gence and weak convergence, defined in terms of the norm and the scalar product
respectively:

Strong convergence: A sequence of vectors {fn}nEN in a Hilbert space His said to


converge strongly to a vector f E H if llfn - !II converges to 0 when n ~ oo. One
then writes s-limn_,.oo fn == f.
Weak convergence: A sequence of vectors {fn}nEN of H is said to converge weakly
to a vector f E H if, for each vector g E H, the sequence of complex numbers
{ \fn, g) }nEN converges to (f, g) when n ~ oo. One then writes w-limn---7 00 fn ==f.
The postulate (H3) is a Cauchy criterion for strong convergence: a strong Cauchy
sequence has a li1nit (in H). The Cauchy criterion is also true for weak convergence: if
{ fn} n EN is weakly Cauchy (i.e. such that, for each g E H, the sequence { (fn, g)} n EN
is a numerical Cauchy sequence), then there exists a vector f in H such that { f n}
converges weakly to f. By using ( 1. 7) it is easy to prove the uniqueness of the limit of
Cauchy sequences (see Proposition 1.3).

Proposition 1.1. (a) Let {fn }nEN be a sequence of vectors in H. Then

s -liln Jn == f ¢===:;> w - lim f n == f and li1n II fn II == II f II·


n oo
---7 n oo
---7 n oo
---7

(b) ffs-lin171---7oo fn == f and s-lillln---7oo gn == g, then limn---7oo \fn, gn) == (J, g).
PROOF. (a) Assume that s-limn---7oo fn ==f. By using (1.12) one obtains that

1\fn,g)- (J,g)l == 1\fn- f,g)l < llfn- fllllgll ~ 0 ~~


as n ~ oo,

so w-limn__, 00 .fn = f. On the other hand, by using (1.15\ one obtains lllfn 11-11.!111 <
II fn - !II ---+ 0 as n ---+ oo, hence limn---7oo II fnll == II fll·
obtain the implication~ we observe that II fn- !11 2 == II fn 11 2 +II !11 - (fn, f)-
2

\f~fn)· w-linln---7oofn == f andlimn---7oo llfnll == 11!11, the expression on the right-


2 2 2
hand side converges to 11!11 + 11!11 2 - llfll - 11!11 == 0, so that s-lin1n---7oo fn ==f.
(b) This follows from (1.12) and the result of (a):

.gn)- (J.g)l == 1\fn,gn- g) (fn- f,g)l < 1\fn,gn- g)l + 1\fn- f,g)l

< llfniiiiYn gil+ llfn- fllllgll ~ II Jll · 0 0 · llgll == 0. D

Example 1.2. Let { e 77 } nEN be an infinite orthonormal sequence in a Hilbert space


H of infinite dimension, i.e. satisfying (e 1 , ek) == 61k. Then w-limn---7oo en == 0,
because for any vector gin H:
DEFINITION AND ELEMENTARY PROPERTIES 5

00

( 1.18)
n=l
consequently limn-'roo (en, g) == 0 [the sequence {en} can be considered to be part of
an orthonormal basis of H, and for such a basis one has equality in ( 1.18) as explained
afterEq. (1.10)]. Thus:
Every il1finite orthonormal sequence {en} converges weakly to the zero vector 0.
On the other hand such a sequence cannot be strongly convergent (if one assumes
that s-limn-'roo en== f, then w-linln-'roo en== f by Proposition 1.1(a), and by the
uniqueness of a weak limit, one must then have f == 0, i.e. s -li1n 11 -'r()() e 11 == 0; this is
impossible since lien- Oil == lien II == 1 for each n). This example shows in particular
that in an infinite-dimensional Hilbert space the notion of weak convergence is really
weaker than that of strong convergence.
Proposition 1.3. The limit of a strong or weak Cauchy sequence is unique.
PROOF. If {fn} is a strong Cauchy sequence, it is also weakly Cauchy [see Propo-
sition 1.1 (a)]; thus it is enough to consider weak convergence. So let us assume that
w-lilTI 11 -'roo fn == h1 and w-lilnn-'roo fn == h2. Then one has for each g E H:
(g, h1 - h2) == (g, h1) - (g, h2) == lim (g, fn) - lim (g, fn)
n-'roo n-'roo
== n-'roo
lirn [(g, fn) - (g, fn)] == 0.

Upon taking g == h1 - h2, one obtains that II h1 - h2ll 2 == 0. By (1.7) one then has
h1 - h2 == 0, i.e. h1 == h2. D

REMARK. If {fn}nEN is a strong Cauchy sequence, then this sequence is bounded in


H, i.e. supnEN II f n II < 00. This follows immediately from Proposition 1.1 (a) (because
llfnll ---+ 11!11). The same conclusion is true if one assumes only that {fn} is weakly
Cauchy, but the proof is longer. This fact is often useful in the theory of Hilbert spaces,
and we shall see in Chapter 2 other results of this type. All these results can be deduced
from a general theorem (the Uniform Boundedness Principle) which we state below
for completeness (the proof can be found in most textbooks, see for example [K]).

Uniform Roundedness Principle: Let A be a set and, for each element A of A, let
<p;>..: H ---+ [0, oo) be a continuous mapping 7 such that <p;>.. (f g) < <p;>.. (f) + <p;>.. (g)
for all J, g E H. If for each fixed vector g the family { <p;>.. (g)} >-.EA is bounded (i.e.
<p;>.. (g) < M \1 A E A, where M < oo is a constant depending only on g), then the
family { cp >-.} is uniformly bounded on the unit ball of H, in other terms there exists a
constant c < oo such that cp;>..(h) < c for all hE H satisfying llhll < 1 and all A EA.
To deduce for example the boundedness of a weakly convergent sequence { fn} nEN,
take A == N and cp >-.(g) cpn (g) == I(fn, g) I· For each g E H, the numerical sequence
{I (fn, g) I} nEN is bounded because (fn, g) converges to a finite limit; then one uses
the fact that llfn II== suphEH,Jihll=l 1\fn, h) I [see Eq. (2.6)] to arrive at the inequality
II fn II == suphE'H.JJhlJ=l cpn (h) ~ c for all n.
7
i.e. cpA (fn) ---+ cpA (f) when fn ---+ f strongly.
6 HILBERT SPACES

Vector-valued functions

Let A be a set and H a Hilbert space. A vector-valued function on A is a mapping


f: A ~ H. It associates a vector in H with each ele1nent of A. Thus a sequence of
vectors {fn}nEN can be viewed as a function f: N ~ H by setting f(n) == fn for
n EN. The 1nost important case in our context is that where A is a (finite or infinite)
interval ~l; we write f (s) for the value of a function f: J ~ H at the point s E J. So
for each s E J, f (s) is a vector in H. Iff (s) depends continuously on s, the family
{!( s) }sEJ describes a curve in H.
As for nu1nerical functions, one can consider the notions of continuity or differen-
tiability and define integrals of such functions. The derivative at a point of J and the
integral of a vector-valued function will be vectors in H. Since derivatives and inte-
grals are limits, these notions may be defined in the strong sense or in the weak sense
(for example a function f: J ~ H may be weakly differentiable but not strongly
differentiable). Similarly one can define integrals in the sense of Riemann or in a
In ore general sense, and one can integrate with respect to different measures on J (the
concept of a measure will be discussed in Section 1.4). We restrict ourselves here to
definitions in the strong topology of H, and we consider only integrals in the sense of
Riemann (and with respect to Lebesgue measure).
A function f: J ~His strongly continuous if one has for each t E .1:

lirn
s-d,sEJ
I ! (s) - ! (t) I == o. ( 1.19)

A vector-valued function f on an open interval J is strongly differentiable at the point


s E tl if there is a vector g in H such that:

~-To II* [.f(s + T)- .f(s)] - g II= 0. (1.20)

We shall write g == f' (s). More generally, f is strongly differentiable in J iff is dif-
ferentiable at each point s E J, i.e. if there exists a function : J ~ H such that

~-=!~ I *[f (s + T) - .f (s) J - f' (s) I =0\ vs E J. ( 1.21)

The function f' is called the (strong) derivative off, and one also writes
f S)
1
( == dd f (S) = S- lim T- l [ f (S + T) - f (S) J · (1.22)
S T-+0

To define the Riemann integral of a vector-valued function, one proceeds in analogy


with the construction for numerical functions. Let J == (a, b] be a finite interval. A
collection of real numbers II== {so, s1, ... , sN; u1, ... , UN}, with a== s0 < s1 <
· · · < s N == band with Uk E ( s k- 1 , s k], will be called a partition of J (Figure 1.1). We
set IITI == maxk=1, ... ,N isk- Sk-11·
Iff: J ~ H, let
N
L:(IT,f) == L (sk- sk-1)f(uk)· (1.23)
k=1
VECTOR-VALUED FUNCTIONS 7

This is a finite linear combination of vectors in H, hence it defines a vector I:(IT,f)


in H. One chooses a sequence {I1r }rEN of partitions of J such that lin17'-""0C IITr I== 0
and defines

/ c
. f( s) ds
J
.la
6
f( s) ds = s-lim I::(Il,.,f)
r~~
(1.24)

exists and is independent of the sequence {I1r}.

o Bo 8 3 . . . . . . . . . . . . . . . . . . . . . 8 N-l 8N b
( )( ><I I><]
11,~3 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1L N

Figure 1.1 Partition of an interval (o, b].

The integral of a vector-valued function on a finite open interval (a, b) or on a


finite closed interval [a, b] is defined similarly by suitably adjusting the definition of
a partition of J: if J == (a, b), then one must assume that UN b, and if tl == [a, b],
then u 1 == a is also admissible.
If J is an infinite interval, one first defines the integral for each finite subinterval
and then takes a sequence of finite subintervals converging to J in an appropriate
way. For exa1nple, J == (a, oo) for some finite a, one defines (if the limit exists)
00

.l .f(
a
s) ds = s- lim / b .f (8) d s.
b----7~ .fa
( 1.25)

We state below some simple properties of these integrals and sufficient conditions for
their existence. The proofs are analogous to those for numerical (real or co1nplex)
Riemann integrals; it suffices to replace in the latter proofs the absolute value If (s) I
by the norm llf(s)ll (details can be found for example in Section 4.4 of [AJS]).

Proposition 1 . 4 . Let (a, b] and (b, c] be finite or infinite intervals and suppose that all
integrals below exist. Then

( 1.26)

( 1.27)

( 1.28)

PROOF. (i) Assume that a and bare finite. Then (1.28) is obtained by using Proposi-
tion 1.1 (a), the triangle inequality (1.13) and the definition of the numerical Riemann
integral:
8 HILBERT SPACES

(ii) If for exa1nple a is finite and b == oo, one finds from Proposition 1.1 and (i)
above that

Proposition 1.5. (a) ff [a, b] is a finite closed interval and f: [a, b] ---+ H is strongly
continuous, then ,[abf ( s) ds exists.
(b) ff a and b are arbitrary (a < b), f: (a, b) ---+ H is strongly continuous and
fab llf(s)llds < oo, then fabf(s)ds exists.
(c) Iff is strongly differentiable on (a, b) and its derivative f' is strongly continuous
and integrable on [a, b], then
Fundamental Theorem of Calculus
1bf'(s)ds = J(b)- J(a). ( 1.29)

Notice that, iff is strongly continuous, then llf(s)ll depends continuously on s by


Proposition 1.1 (a), so that ,[ab I f (s) I ds can be defined as a numerical Riemann in-
tegral.

Subsets and dual of a Hilbert space

We consider here certain types of subsets of a Hilbert space H. A subset D of


H is dense in H if, given any vector f E H and E > 0, there exists an element g of
D such that II!- gJI < c. Equivalently, Dis dense in H if, given any vector f E H,
is a sequence {fn }nEN in D such that liinn-+oo II/- fnll == 0. An example of
a dense set is given by the collection all finite linear combinations of vectors of
an orthononnal basis {en} of H. In this case D is the set of all vectors of the form
1 CYk , with ak E C and N == 1, 2, 3, ... (N < oo); if one assumes in addition
real part and the imaginary part each a k are rational nu1nbers, one obtains
a dense subset of H that is countable. i

linear manifold£ is a linear subset of H, i.e. such that f, g d and c1: E C ====?
f cxg E £ (in other words each finite linear combination of elements of £ also
belongs to£). An exa1nple is the already indicated set of all finite linear combinations
of of an orthonormal basis {en} of H; more generally, if N is an arbitrary
subset H, the set all finite linear combinations vectors belonging to N is
a linear 1nanifold, called the linear manifold spanned by N. If the linear manifold
spanned by a subset N is dense in H, one says that N is a total set in H or a total
1-l. An exa1nple a non-total subset is given by the vectors e 2 , e 4 , e 6 , ... of
an orthononnal basis {en}.
linear 1nanifold £ in H is clearly a complex linear vector space equipped with a
product (induced by the scalar product in H, i.e. if J, g E £, then their scalar
product (f. g) E £ is sin1ply their scalar product (/,g) in H). In general £ will not
SUBSETS AND DUAL OF HILBERT SPACE 9

be complete in the norm 11·11, i.e. the limit of a Cauchy sequence {fn}, with fn E £,will
not belong to£ but only to H. If£ is cornplete [i.e. if£ satisfies all of the postulates
(Hl)- (H4)], then£ will be called a subspace of H. In what follows we use the syrnbol
M for subspaces.
A linear manifold£ determines uniquely a subspace£, called the closure of£: a
vector f in H belongs to£ if and only if there exists a sequence {fn }nEN in £ that
converges strongly to f as n ---+ oo. If£ is dense H, then£ == H, otherwise£ is a
subspace of H that is strictly smaller than H.
More generally, an arbitrary subset N of H defines uniquely a subspace M.!V,
called the subspace spanned by N: MN is the smallest subspace containing N, in
other terms the closure £ of the linear manifold £ spanned by N.

EXAMPLES OF SUBSPACES
(a) A subspace of dimension 0 is {0}, consisting of only the zero vector.
(b) Each vector f 0 determines a one-dimensional subspace, the set { af \a
E C},
i.e. all multiples of the vector f.
(c) A subspace of infinite dimension of an infinite-dimensional Hilbert space H is
obtained by taking all (finite and infinite) linear combinations belonging to H of the
vectors e2, , e6, ... of an orthonormal basis { en}nEN of H.

REMARK. The terminology concerning subspaces is not uniform; sotne authors say
"subspace" for a linear manifold (they have in mind a subspace of the vector space H)
and "closed subspace" for what we call a subspace.

The scalar product leads to the notion of orthogonality in H. Two vectors f and g
of H are said to be orthogonal if (j, g) == 0, and one then writes f _L g. A sequence
{!1 , f 2, j 3 , ... } of vectors of His called orthonormal if (f~, fk) == 5.Jk·
If N is a subset of H, one defines its orthogonal comple1nent N _L as the set all
vectors in H that are orthogonal to each vector inN:

N_L=={JEH\(J,g)-o vgEN}. (1.30)

We write f _L N iff E N _L. The set N _L is a subspace of H (Problern 1.7). MN is


the subspace spanned by N, one has

( 1.31)

Indeed: (i) the implication <¢== is trivial. (ii) For the implication =? we first observe
that f _L N implies f _L £, where £ is the linear manifold spanned by N (because
each g E £ is of the form g == 'I:~=l akgk with .9k E N for some N < CXJ, hence
(f, g) == 'I:~=l ak (f, gk) == 0). Finally f £ implies that f £ MN (because
each element g of the subspace£ is given as g == s-linln-+oo .9n with .9n E £,hence
(.f,g) == lirnn-+oo (f,gn) == linln-+oo 0 == 0).

Proposition 1.6. A subset N o.f H is total in H ~f and only iff N inlplies that
f == 0.
10 HILBERT SPACES

PROOF. N is total in H and f N, then f is orthogonal to H by ( 1.31 ). In partic-


ular f is then orthogonal to itself, i.e. (f, f)== llfll 2 == 0, hence f == 0 by (1.7).
Conversely, suppose that f N implies that f == 0. By virtue of (1.31) this means
that the orthogonal co1nplement of the subspace MN spanned by N is {0}. Hence
Mjv == {O}_L == H (here one applies the Projection Theorem given below). D

The next proposition generalises a well-known property of finite-dimensional vec-


tor spaces equipped with a scalar product. This is a fundamental result in the theory of
Hilbert spaces, and it also holds in non-separable Hilbert spaces [the postulate (H4)
not used in its proof].

Proposition 1.7 (Projection Theorem). Let M be a subspace of H, M_L its orthog-


onal cornplement. Then each vector f qf H admits a unique decomposition into a
con1ponent in M and a conzponent in M _L:

( 1.32)

PROOF. We fix a vector f E H, and we denote by d the distance from f to M, i.e.


d == infgE.A1 I f gjl. We shall use the following identity: if h 1 and h2 are vectors in
then, by Eq. ( .17) with f == h 1 and g == -::r-h2:

( 1.33)

We first show that there is a unique vector g0 E M such that d== I f - go II·
~~~ Choose a sequence {gk}kEN in M such that 1in1k----+oo ll.f- gkli == d. Let
us take h 1 - (f- g.7) /2 and h 2 == (f - gk) /2 in (1.33). We obtain

Since g1 + 9k EM, we have II f - (g.J gk) /211 2 > d 2 . Hence

~ llrh gkll <~III- gJII +~III- gkll


2 2 2 2
- d -'to asj, k -'too.

{gk} is a Cauchy sequence H. We denote its litnit by 9-D· Since M is a sub-


space, we have go EM. Consequently I f- go II > d. On the oth~r hand

llf- 9oll <II!- 9kll ll9k- go II---+ d 0 == d ask---+ oo.


have thus shown that II f - go II - d.
Uniqueness. Let us assume that there exist two vectors gg) and g~2 ) in M such that
/If- g~1 )11 == II/- g62)11
1
==d. By taking h1 == - g~l))/2 and h2 == (f- g62))/2 in
( 1.33) one obtains as in (i) that

41 II 9o(I)
2 2 2
II f - .Jo
9o() 11 -< _!_
2 .
c (I) 11

Th us II rJo(I ) - go( 2) II - 0, l.e.


. go(1) -- 9o(2) .
SUBSETS AND DlJAL OF A HILBERT SPACE 1

(ii) prove (1.32) we set / 1 == g 0 , where g 0 is the unique vector in M associated


with f according to (i). Then f 2 == f - g0 , and we have to show that / 2 E Mj_, in
other terms that (h, / 2 ) == 0 for each h E M. So let h #- 0 be a fixed vector in M
and set a== (h, /2). Let g == /1 + ah/llhll 2. Clearly gEM, hence II!- .c;ll > d and
consequently

This implies that lal 2 < 0, hence that a == 0.


(iii) Finally let us check the uniqueness of the decomposition ( 1.32). Assume that
g1, g2 E M and h1, h2 E M _i are such that g1 + h1 == 92 h2. We must show that
g1 == .92 and h1 == h2. Now

0 == IIOII 2 == 11.91 + hl- (g2 + h2)11 2


== Jl(gl- .92) + (hl- h2)11 2 == 11.91-.9211 2 2
Jlht- h211 .
because (g1 - g2, h1 - h2) == 0. Thus Jlgl - g21i == II h1 - h2ll == 0, i.e. 91 == .92 and
h1 == h2 by (1.7). D

The projection theorem shows that H may be viewed as the orthogonal (direct) sum
of two mutually orthogonal subspaces M and M _i. We recall that the orthogonal sum
H H1 E8H2 · · ·ffi'HN of N Hilbert spaces H 1, ... , HN is the Hilbert space forn1ed
by all N-tuples {/1 , ... , }, where E Hk, with the following scalar product:

( 1.34)

Each Hk is a subspace of H, and if j #- k then the spaces H.7 and Hk, considered as
subspaces of H, are mutually orthogonal.
Occasionally we shall also use orthogonal sums of an infinite number Hilbert
spaces. The orthogonal sum H == EBZ:
1Hk of an infinite sequence {H1, H2, .. .} of
Hilbert spaces is formed by all infinite sequences {/1 , f 2, ... } , with fk E Hk and such
that ~Z: 1 11 ll~k < oo, the scalar product in H being defined as in (1.34) but with
N == oo. Again each Hk is a subspace of H (the set of all sequences {/1 , , ... } with
f 7 - 0 for all j k), and the subspaces H.7 and Hk are mutually orthogonal if j #- k.
As an example, suppose that each Hk is a one-dimensional space, i.e. Hk == C for
each k. If N is finite, the orthogonal sum H == H 1 E9 H 2 E9 · · · E91iN is (isomorphic
to) the space CN. If N == oo one obtains the space denoted by I! 2 , the Hilbert space of
all infinite sequences { a1, a2, ... } of complex numbers satisfying ~ZO= 1 1ak 12 < oo,
with scalar product
00

\{al,a2, ... },{Pl,P2,···}) == LakPk· (1.35)


k=1
12 HILBERT SPACES

Each (separable) infinite-dimensional Hilbert space His isomorphic toR 2. Indeed H


may be identified in a natural way with f 2 by choosing in H an orthonormal basis
{en} and by identifying the vector f == ~kakek (see Section 1.1) with the element
off 2 given by the sequence { a 1, a 2, ... } . In the terminology that will be introduced
in §2.2.2, the correspondence H 3 f t-7 { a 1, a 2 , ... } defines a unitary operator from
H to f 2 .

1.3.2. We end this section with a characterisation of the dual of a Hilbert space. The
dual H* of H is by definition the set of all bounded linear functionals on H, i.e. the
set of all mappings <p: H ---+ C having the following two properties:

cp(f ag) == cp(f) + acp(g) (linearity)


Jcp(f)J < cJJfll \If E H, (boundedness)

where cis a constant (independent of f). The space H* is a complex linear vector
space (for example the sum of two bounded linear functionals is again a bounded
linear functional), and H* is also normed with the norm

(1.36)

Each vector g of H detennines an element <p 9 of H * by setting

(f?g (f) == (g' f) (where f varies over H). (1.37)


"
Indeed, the scalar product (for fix~d g) is linear in f, and [use (1.12)]

In fact one has


(1.38)

because the supremum is attained for f == g: (g, g) /Jig II== jjgjj 2 /II gil== llgJJ.
We have thus shown that each vector of H determines an element of the dual space
H*. It is an important fact that the converse is also true (so that one may identify H*
and H):

Proposition 1.8 (Riesz Lemma). Let H be a Hilbert space and cp a bounded linear
functional on H. Then there exists a unique vector g in H such that

cp(f) == (g, f) vf E H. (1.39)


In particular: llcpiiH* == jjgjJH jjgjj. (1.40)

PROOF. (i) Define M to be the set of all vectors f E H for which cp(f) == 0. Let us
show that M is a subspace:
( 1) f1· f2 EM and()! E C ====? cp(fi af2) == cp(fi) + acp(f2) == 0 0 == 0,
SUBSETS AND DUAL OF A HILBERT SPACE 13

(2) if {fn} is a sequence of vectors in M such that limm,n-r(X) llfnL- fnll == 0, then
the sequence {fn} has a li1nit fin 7-i, and we must show that this vector f belongs to
M (completeness of M), i.e. that cp(f) == 0. Now

because cp(fn) == 0. As llf- fnll ~ 0 when rt ~ oo and cp is bounded, one must


have cp(f) == 0.
(ii) If cp 0 we take 9 == 0; indeed one then has 0 == cp(f) == (0, f) for each
f E 7-i. If cp ¥::- 0 there is a vector h in 1i such that cp(h) i- 0. By the Projection
Theorem we can write h == h 1 + h 2 with h 1 E M and h 2 E M _L. Then

cp( h2) == cp( h h1) == cp( h) - cp(h1) == cp( h) i- 0,


~
0
in particular h 2 i- 0 (because cp(O) == 0).
Iff is an arbitrary vector in 7-i, let us consider the vector f- [cp(f)/cp(h 2)] h 2 . We
then have

hence the vector f- [cp(f)/cp(h 2)] h 2 belongs toM. Since h 2 j_ M we obtain

Upon setting 9 == [ cp( h2) /II h2ll 2] h2 (which defines a vector in 1i because I h2ll i- 0),
one obtains ( 1.39).
(iii) We finally show that 9 is unique. If 9 1, 92 are two vectors such that cp(f) ==
(91, f) == (92, f) for all f E 7-i, then (91 -92, f) == 0 for all f E 7-i. Hence 91 - 92 == 0
by Proposition 1.6. D

Let us add that the identification of 7-i* with 1i is not linear but anti-linear: if cp 1
and C;J2 are elements of 7-i* such that cp 1(f) == (9 1, f) and cp 2 (f) == (9 2 , f) for all
f E 7-i, then

hence the vector in 1i representing the linear functional cp 1 + (~cp 2 is the vector 9 1 +
a92·
14 HILBERT SPACES

1.4 Measures, integrals and LP spaces

rigourous treatlnent of the theory of L P spaces requires familiarity with measure


theory and integration in the sense of Lebesgue. We shall here present a description
of these spaces and of so1ne aspects of measure and integration theory, without giving
the proofs. The co1nplete theory is treated in numerous books.

1.4. 1. Consider a set 0 and a collection A of subsets of 0 having the structure of a


CT-algebra, i.e. satisfying: ( 1) A contains the empty set 0 and 0 itself, (2) if V E A,
then its complementary set 0\ V also belongs to A, (3) A is stable under the operation
taking countable unions in the sense of set theory: if Vk E A for k == 1, 2, 3, ... , then
UkVk E A. If these conditions are satisfied, then A is also stable under the operation
of taking countable intersections, and the couple (0, A) is called a measurable space.
We use the letter x for the elements of 0 (the "points"). A measure m on (0, A) is
a mapping from A to [0, oo] that is a--additive, i.e. such that rn(Uk Vk) == I:km~(Vk)
for each countable family {Vk} of disjoint elements of A (i.e. satisfying ~in Vk == 0
if j i- k ), and with rn( 0) == 0. A measurable space (0, A) together with a measure m~
defined on it is called a measure space, denoted by (0, A, rn ).
We see that a measure rn on (0, A) assigns a "weight" (which may be infinite)
to each subset V of 0 belonging to A, and this assignment is additive (the weight
of a finite or countable union of disjoint subsets of 0 is the sum of the weights of
these subsets). A set V of 1neasure zero will be called a null set with respect to m. If
rn( 0) < oo, then 1n is called a finite measure. If 0 can be expressed as the union of
a countable collection of elements of A each of which is of finite measure, then rn is
called aCT-finite measure. In particular every finite measure is a--finite. An example
of a a--finite n1easure that is not finite is given by~ Lebesgue measure on JR. (see
§ .4.2). A finite measure with rn( 0) == 1 is a probability measure; in applications
the set 0 then represents the possible outcomes of an experiment, the elements of the
a--algebra A correspond to the events of interest in the experiment, and the measure
rn (V) is the probability that the event V will occur (an example in quantum mechanics
will be discussed on page 168).
A function cp: 0 ~ JR. assigns a real number cp(x) to each point x of 0. Given a
a-algebra A of subsets of 0, such a function cp is said to be a measurable function
if. for each interval J in IR, the set cp- 1 (.J) :== {x E 0 I cp(x) E J} belongs to the a--
algebra A. It is seen that measurability is not simply a property of the function cp but
depends very much on the a- -algebra A. cp may be measurable with respect to certain
a--algebras and non-measurable with respect to other a--algebras of subsets of 0.
The important fact for integration theory in the sense of Lebesgue is that each
n1easurable function is the limit of a sequence of simple measurable functions (one
could take this characterisation as the definition of a measurable function). A simple
function is a function of the form cp == I:~= I akXv" with ak E JR. and N < oo,
where Xv,", is the characteristic function of the subset Vk of 0; a simple function is
MEASURES, INTEGRALS AND L P SPACES 15

measurable if E A for k == 1, ... , N. We recall that the characteristic junction xv


of a set V is defined as

XV (X) == 1 if X E V, XV (X) == 0 if X Et V. ( 1.41)

Thus, if cp a measurable function, there exists a sequence { cpn} of measurable si1nple


functions that cp(x) ==limn-too cpn (x) for each x E 0. Furthermore, if cp > 0, the
sequence {cpn} can be chosen non-decreasing, i.e. such that 0 < cprz (x) :::; cprn (x) <
cp(x) each x E 0 and m > n.
If cp is a measurable simple function, say cp == ~~= 1 akXv", then the integral of cp
with respect to the 1neasure m on (0, A) is defined to be the real number

1 0
cp(x)m(dx) =
N

Lo,.m(Vk)·
k=l
( 1.42)

This has a sense provided that the sets Vk are such that nt(Vk) < CXJ. If this is the case,
the function cp == ~~=l ak Xvk is called a simple m-integrable function.
If cp is a general measurable function (not necessarily simple), one can define its
integral by approximating cp by a sequence of simple functions { 4?n}. One considers
first measurable functions cp > 0 and defines

.!.
0
cp(.T)m(dx) = lim
n-too
1 0
cpn(x)nl(d:r), (1.43)

where { 4Jn} is such that 0 < cpn (x) < cp( x) and lirr1 n -too cpn (x) cp( 1;) for all
x E 0. This makes sense if each of the simple approximating functions 4?n is rn-
integrable and if the limit in (1.43) is finite. If cp is not> 0, one applies the preceding
definition to its positive and its negative part, i.e. one decomposes cp into cp -
with > 0 and cp_ > 0; the integral of and that of 4?- are defined as in (1.43),
and cp is called m-integrable if each of these two integrals is finite. Then one sets

r cp(x)m(d:r:)
Jo
= 1o (.T )m( dx) - ./~ cp_ (:r)m (d:r ). (1

If cp: 0 ~ CC is a complex function, it is rn-integrable if its real part and its imag-
inary part are rn-integrable in the sense of the preceding definition. The integral of cp
is obtained by integrating separately its real part and its itnaginary part.
The Lebesgue type integral introduced above generalises the (proper) Riemann in-
tegral on the real line. As an illustration we shall co1npare in §1.4.2 below the Le-
besgue integral on IR with the Rietnann integral for continuous functions. For measur-
able functions without continuity properties, only the Lebesgue integral has a tnean-
Ing.
In the situations considered in this text, the underlying set 0 will be a subset of IRn
(n) == 1, 2, 3, ... ) , often a finite open interval (a, b) or a semi-infinite interval (a. CXJ),
or IRn itself. In each of these situations the a--algebra A will be the Borel a--algebra
or the Lebesgue a--algebra of the considered set 0. Mostly the measure rn will be the
16 HILBERT SPACES

Lebesgue 111easure m( dx) == dx [m( dx) == dnx if n > 2]. In the next subsection we
explain these tenns in somewhat more detail.

1.4.2. B is an arbitrary collection of subsets of a set 0, there is a smallest a-algebra


A containing B (i.e. such that, if A' is a a-algebra with B C A', then A C A'). This
1ninimal a-algebra A, called the a-algebra generated by B, is simply the intersection
of all a-algebras containing B (this intersection is not empty because the collection of
all subsets of 0 is a a-algebra). In certain situations a measure m, on A is completely
detern1ined if it is given on !3. An important example is the Borel-Lebesgue measure
on the real line which we shall now describe.
So let us consider the case 0 == IR. We take for !3 the set of all half-open intervals
(a, b] with -oo < a < b < +oo. The a-algebra generated by the collection of all
these intervals is called the Borel a-algebra of IR; we denote it by AB, and its ele-
Inents will be called Borel sets. 8 Let us take for the measure of an interval (a, b] its
length R, i.e. 112,( (a, b]) == R( (a, b]) b - a. The thus defined mapping m: B ~
[0, oo) has an extension to a 1neasure on (IR, AB ), often called the Borel measure on
IR. This 1neasure associates with each Borel set of IR a non-negative number (which
can be infinite) in a a-additive way, and the number associated with an interval is its
length. The length of a general Borel set V is obtained in the following manner: one
considers covers of V of the fonn V C Uk Jk, where { Jk} kEN is a countable collection
half-open intervals [Jk == (ak, bk], ak < bk], and one then defines

R(V) ==infLR(Jk) infL(bk -ak), (1.45)


k k

where the infi1num is over all covers of V of the form specified above. This length
function has properties of a measure on (IR, AB ), viz) the Borel measure mB on
the real line: 1nB (V) == R(V) for V E AB·
We 111ention some examples of null sets with respect to Borel measure (Borel sets
length zero). The simplest example is a single point { x 0 }. This set is the intersec-
tion of all intervals (x 0 - k- 1 , x 0 ], k == 1, 2, 3, .... The length of such an interval is
7n~ B ( (;:r 0 - k- 1 , x 0 ]) == 1/ k; since { x 0 } is contained in each of these intervals, its mea-
sure must be less than 1/ k for each k == 1, 2, 3, ... , hence equal to zero. 9 A second ex-
ample is that of a countable set { x 1 , x 2 , x 3 , ... } ; as m B is a -additive and the measure
of each {xk} is zero, one has mB({x1,x2,x3, ... }) == LZ: 1 n~B({xk}) == 0. There
also exist Borel null sets that are not countable, for example the Cantor set <t which
is a subset of the interval [0, 1] obtained by removing from [0, 1] a countable union
8 The smne a-algebra is obtained by starting with the set of all open intervab or with that of all closed
intervals. Thus AB contain~ all open, closed and half-open intervals. In fact each open and each closed
subset of IR is a Borel set (this is a consequence of the fact that each open subset of IR is the union of
countably many disjoint open intervals), but there is no simple characterisation of individual Borel sets
directly in tenns of unions and intersections of open and closed sets. The use of intervals of the forn1 (a, b]
n1ay be ju~tified by the fact that ~uch intervals fit together in a nice way, and this will be crucial for the
definition of the n1ore general Stiel~jes n1easures (see Ren1ark 4.6).
9
A basic property of a measure rn is that, if V, W E A are such that V ~ W, then m (V) :s; m(W),
which is a ~traightforward consequence of the fact that a 111easure is additive.
MEASURES, INTEGRALS AND L P SPACES 17

of disjoint open subintervals of total length 1. This is done recursively in the follow-
ing manner (Figure 1.2): first one removes the open middle third (1/3, 2/3) from the
interval [0, 1], next one removes the middle thirds (1/9, 2/9) and (7 /9~ 8/9) of the
remaining intervals, and so on. The Cantor set is a closed Borel set, it has Lebesgue
measure zero and is uncountable. The reader may find more details for example in
Vol. I of [RS] or in [GO].

0 1

etc.
Figure 1.2 Construction of the Cantor set.

As another illustration, let us compare the construction of the Riemann integral


and of the Lebesgue integral for a continuous non-negative function f on some finite
interval [a, b]. Both are defined as limits of sequences of finite su1ns. The difference
resides in the fact that the Riemann integral is formulated in terms of partitions of the
domain [a, b] off, whereas the Lebesgue integral uses partitions of the range off. For
the Riemann integral, let IT be a partition of [a, b] as on page 6. The contribution to the
Riemann sum~ (II, f) of an individual subinterval J of this partition is just f (u')£ (J),
as indicated in Figure 1.3(a) [f( u') is the value off at some point u' E J]. The in-
tegral is of course given by the area under the graph of f. For the Lebesgue integral
one partitions the ordinate axis into disjoint intervals { ~{!}. The contribution to the
integral of a fixed one of these intervals, say of 6k, is fk£(Vk), where fk is so1ne
number in ~k and Vk - f- 1 (~k) is the Borel set on the x-axis formed by all points
x E [a, b] at which f assumes a value in ~k [Figure 1.3(b)]. It is clear that, upon
adding up the contributions from all intervals ~{! and passing to the limit where their
length tends to zero, the value of the integral is again equal to the area under the graph
of f. Observe that, for a continuous function f, the set Vk f- 1 (~k) is simply a
finite union of intervals, whereas for an arbitrary measurable function on (IR, AB) it
could be any Borel subset of IR.

(a) (b)

J( u')

J b

Figure 1.3 (a) Rien1ann integration; (b) Lebesgue integration.


18 HILBERT SPACES

By prescription ( 1.45) one can assign a length to any subset V of JR (not only
to Borel sets). It seems that, by using this length function on the a--algebra Amax
consisting of all subsets of JR, one might obtain a measure on this largest possible o--
algebra, with the property that the measure of any interval coincides with the length of
that interval. As a 1natter of fact this maximal extension of the Borel measure in terms
the length function is not a measure (it is not a--additive on Amax). It is impossible
to extend the Borel measure to the class of all subsets of JR in such a way that the
extension is still a measure which is invariant under translations. 10
Of course there exist extensions of the Borel measure to a-algebras larger than
AB. One such extension is frequently used in mathematics, namely the extension to
the Lebesgue a-algebra AL, called Lebesgue measure. The elements V of AL have
the fonn V == W U V0 , where W is a Borel set (WE AB) and Vo can be any subset
of some Borel set of measure zero. 11 The Lebesgue measure of V is just the Borel
Ineasure of W, the Lebesgue n1easure of V0 is zero: m~L(V) == rnB(W) == £(W). So
A L contains, in addition to the Borel sets, all subsets of Borel sets of measure zero, in
fact all subsets of JR of length zero. This is not trivial, as shown for example by the fact
that there exist subsets of the Cantor set which are not Borel sets (we recall that the
Cantor set is a Borel set, and it is of measure zero). The measure space (JR, AL, mL)
is complete in the sense that the a--algebra AL contains all subsets of sets of measure
zero, vvhereas the measure space (JR, AB, m~B) is not complete. Some statements in
measure theory depend on the completeness of the underlying measure space.
Fro1n now on we shall use the standard notation dx for the measure determined
by the length function R of Eq. (1.45) and call this measure simply Lebesgue mea-
sure: rnB ( d:r) == 11l)L( dx) == dx. The underlying a--algebra will be either the Borel
o- -algebra A B or the Lebesgue o- -algebra A L. In contexts where it is necessary, we
shall specify which of these two a--algebras is involved.
Above we have introduced the Lebesgue measure on the ~};line. Its analogue on
JRn is defined in much the same way by taking forB the collection of all subsets of the
form { x E JRn j - oc < ak < Xk < bk < +oc, k == 1, ... , n} and for the measure of
such a subset its n-dimensional volume [here we use the notation x == (x1, x2, ... , Xn)
the individual points of JRn]. A Borel set in JRn is an element of the a--algebra gen-
erated by this collection B, and the Lebesgue measurable sets are obtained by adding
to the Borel a--algebra all subsets of Borel sets of measure zero.
Other important measures in this text are Stieltjes measures on (JR, AB ), obtained
by considering weight functions different from the length function on the set B of
half-open intervals. In quantum mechanics, measures of this type occur in the charac-
terisation of the spectral parts of observables; they are introduced in Section 4.1 and
play an important role throughout Chapter 4.

10
One uses the term outer 1neasure for the extension of a 1neasure given on some a-algebra A to the
1naximal a-algebra consisting of all subsets of 0. In general an outer measure does not satisfy all conditions
in1posed on a n1easure, i.e. an outer measure is not a 1neasure.
11
So Vo is a subset of IR such that there exists a Borel set U satisfying P(U) = 0 and Vo C U.
MEASURES, INTEGRALS AND LP SPACES 19

We next consider LP spaces on a general measure space ( 0, A, m,). Two mea-


surable functions (real or complex) cp 1 and cp 2 on 0 are said to be equivalent if they
are equal rn-almost everywhere, i.e. if they differ at most on a null set with respect to
the measure rn. Let W == {x E 0 \ cp 1 ( x) =1- cp 2 (:x;)} be the set of all points x E 0
at which these functions have different values; this set belongs to A, hence rn(W) is
well defined, and cp 1 and cp 2 are equivalent if and only if rn(W) == 0. 12
We now consider equivalence classes of measurable functions. The important fact
is that, for any t > 0, the integral of Icp 1 - cp2l t is zero if cp 1 and cp 2 are equivalent and
that, one of these two functions ism-integrable, then

/' cp1 (:r:)rn(dx) = ;· cp2(x)rn(dx).


Jo . (J

If p E [1, +oo ), one defines the space LP( 0, m) LP( 0 ~A, rn) to be the set of
equivalence classes of measurable functions f : 0 ~ CC satisfying

ll.f II P := l.Jr if (x )iP rn (dx) ]1/p < oo.


r

0
(1.46)

LP (0, m,) is a normed vector space, the norn1 being given by ( 1.46), and this space
is complete with respect to this norm (in other terms it is a Banach space). By con-
sidering equivalence classes of functions rather than individual functions as elements
of LP(O, m)), one ensures that the only element of LP(O, m,) with nonn zero is the
equivalence class of the function f 0 (so the zero vector is unique, as in a Hilbert
space).
When considering functions f: 0 ~ CC, one should each time specify whether f
is to be interpreted as an individual function or as an element of an LP space, i.e. as an
equivalence class of functions. In most situations encountered in this text the precise
meaning of f will be clear from the context and there will be no need to mention it in
detail.
An important result in the context of LP spaces is the Dominated Convergence The-
orein (also called Lebesgue Convergence Theorem) which gives a sufficient condition
for interchanging limits and integrals. This criterion is frequently used in our text.

Proposition 1. 9 (Dominated Convergence Theorem) . Assun~e that


1
(i) fn, hE L (0, rn),
(ii)foreachnEN, one has lfn(x)i < h(x) rn-a.e.,
(iii) limn-+oo fn(x) == f(x) m-a.e.
Then the lin1it function f ism-integrable and one has

lim
n-+OO
/' .fn(x)m(dx) =
lo
r .f(x)m(d:r;).
lo
( 1.47)

12
A statement involving the points of 0 is said to hold m-abnost everywhere (abbreviated 'JTI-a.e.) if the
set of points x E 0 at which the statetnent is not satisfied is a null set with respect to the measure m. One
then also says that the statement holds at almost all point~ of 0.
20 HILBERT SPACES

The following inclusion relations between LP spaces are useful:

( 1.48)
f E L q ( 0, rn) and mj (V) < oo ::::::::? X v f E L r ( 0, m) Vr E [1, q] . ( 1. 4 9)
PROOF. Let us write f == f> + f< with f>(x) == f(x) if lf(x)l > 1 and f>(x) == 0
otherwise (hence f<(x) == f(x) if lf(x)l < 1 and f<(x) == 0 otherwise). Writing LP
for LP( 0, nL) and assuming that s < oo, we have the following implications:

fEL 5 ::::::::? f>ELt VtE[1,s] (since lf>(x)lt < lf(x)l 5 ) (1.50)


fEL 5
::::::::? f<EL 11
VuE[s,oo] (since If< (x) Iu < If (x) Is) . ( 1.51)

To prove (1.48), let f E LP n Lq. If r E [p, q ], then f> E Lr [by (1.50) with s == q and
t == r] and f< E Lr [by (1.51) with s == p and u == r]. Hence f f> + f< E Lr.
To verify (1.49) we observe that, under the conditions of (1.49), one has f> E Lr for
r < q [apply (1.50) as above], hence Xvf> E Lr. On the other hand l(xvf<)(x)l <
xv(;:c), and Xv belongs to each LP if m(V) < oo (one has llxviiP == [mj(V) J l/p). D

Among the spaces LP( 0, rn ), only that for p == 2 is a Hilbert space (Problem 1.14).
The scalar product in L 2 ( 0, m) is given as follows:

(!,g)= l f(x)g(.T)rn(dx). ( 1.52)

By the Schwarz inequality this number is finite if f and g belong to L 2 ( 0, m) (and


the integral in ( 1.52) depends only on the equivalence classes of f and g). We use
the simplified notation II · II for the norm II · 11 2 . In quantum mechanics, when 0 is the
configuration space of a physical system [typically 0 == JR or IR.n or an interval (a, b)],
the elements of L 2 ( 0, 1n) (or the representative functions of an equivalence class) are
usually called wave functions; we refer to Remark 2.32 for further considerations on
this terminology.
Before considering more explicit cases, let us mention the following generalisation
(we restrict ourselves to p == 2): instead of looking at functions defined on 0 with
values in CC, one may consider functions with values in a Hilbert space JC (vector-
valued functions in the sense of Section 1.2). In analogy with the discussion above,
one can introduce a Hilbert space L 2 ( 0; JC, m); its elements are equivalence classes
of such functions, 14 with scalar product and norm given by

(f. g)= l (f(x),g(.T))JCrn(dx) and [.£ II f (x) II k rn (d:r:) J


112
, ( 1.5 3)

where ( ·, ·) JC and II · II JC denote the scalar product and the norm in JC respectively. For
J( == CC one regains the space L 2 ( 0, mj) considered before.
13The space L 00 ( 0, rn) will be defined in Section 2.5. The proof of (1.48) is given here for q < oo.
14A function f: 0 ~ JC is measurable if for each g E JC the numerical function x ~ (g, f(x)) JC is
measurable in the sense specified before.
MEASURES, INTEGRALS AND L P SPACES 21

1 . 4 . 4 . One may introduce £P spaces on the real line II{ in terms of Lebesgue mea-
sure, by taking either AB orAL for the a--algebra. Both definitions lead to the same
space: £P(JR, AB, mB) == £P(JR, AL, mL) £P(JR, dx). This follows from the fact
that null sets are irrelevant in an integral, by remembering that the elements of an
£P space are equivalence classes of measurable functions [of course an equivalence
class with respect to Lebesgue measure on (JR, A L) contains more functions than
the corresponding equivalence class with respect to Lebesgue measure on (JR, AB),
since AB is strictly smaller than AL]. Another way of understanding the relation
£P(JR, AB, mB) == £P(JR, AL, mL) is to use the fact that these spaces can be defined
in terms of Riemann integrals without having recourse to measure theory, as explained
towards the end of the present subsection.
When the measure defining an £P space on a Borel subset V of II{ or of JR 17 is the
Lebesgue measure, we use the simplified notation £P(V) for the space £P(V, rn);
it is understood that the measure in L P (V) is the Lebesgue measure. Also we then
say "almost everywhere" instead of "m-almost everywhere" or of "Lebesgue ahnost
everywhere".
If V is a Borel set in JR, for example an interval, the space £ 2 (V) is a Hilbert
space, consisting of all equivalence classes of measurable functions f : V ----* C that
are square-integrable over V (with respect to the Lebesgue 1neasure on which is the
restriction of the Lebesgue measure on II{ to the collection of Borel sets contained in
V). The space L 2 (V) can also be considered in a natural way as a subspace of £ 2 (JR),
namely the subspace of all (equivalence classes of) functions f in £ 2 (JR) that are zero
outside V, i.e. which satisfy f (x) == 0 if x tJ_ V (except possibly on a Borel set W of
points x tJ_ V with IWI == 0, where IWI denotes the Lebesgue measure of W). In what
follows we do not distinguish between these two meanings of £ 2 (V); it will each time
be clear from the context whether the space £ 2 (V) is considered as a Hilbert space
without reference to L 2 (JR) or rather as a subspace of L 2 (JR).
If V and V' are two disjoint Borel subsets of JR, then £ 2 (V) and £ 2 (V ') are mutu-
ally orthogonal as sub spaces of £ 2 (JR). Also £ 2 (JR \ V) is the orthogonal complement
[in £ 2 (JR)] of the subspace £ 2 (V).
Examples of dense linear manifolds in the Hilbert space £ 2 (JR) are (modulo equiv-
alence) the Schwartz space S (JR), consisting of infinitely differentiable functions of
rapid decay, and the set C0 (JR) of all infinitely differentiable functions of compact
support. A function f: JR ----* C belongs to S(JR) if it is infinitely differentiable and
if f and all its derivatives converge to zero at infinity more rapidly than any inverse
power of IxI, i.e. if

sup lxl kldRf(x)l


R < oo \lk,£ == 0, 1, 2, ... (1.54)
xElR dx
A function f: JR ----* C belongs to C 0 (JR) if it is infinitely differentiable and if there
is a finite interval J (depending on f) such that f (x) == 0 for all .r tJ_ tl. Clearly
C 0 (JR) c S(JR), so that the density of S(JR) in £ 2 (JR) follows from that of C 0 (JR).
The density of the latter set is discussed in §1.4.5.
22 HILBERT SPACES

More generally, if -oo < a < b < +oo, a function f: (a, b) ~ C belongs to
C/) ( (a~ b)) if j is infinitely differentiable and j (X) == 0 in son1e neighbourhood of
0

x ==a and in some neighbourhood of x == b. The set C 0 ((a, b)) is dense in LP( (a, b))
for each p E [1, oo); In ore precisely the set of equivalence classes of the functions in
C0 (IR) is a dense set in £P((a, b)).
Si1nilar definitions and results apply in LP (IRn). C 0 (IR 11 ) is the set of all infinitely
differentiable functions f: IRn ~ C such that f(x) == 0 for lxl > NI for some number
1\I < oo (depending on f); it dense in £P(IR 11 ) for each p E [1, oo ). The Schwartz
space S (IRn) consists of infinitely differentiable functions satisfying

\1 k, £1, ... , fln == 0, 1, 2, ... (1.55)

is clear that C0 (IR 72 ) c S (IRn). In certain considerations the Schwartz space


S (IR 11 ) is more suitable because it is invariant under Fourier transformation: if f be-
longs to S(IRn ), then so does its Fourier transform j. The Fourier transformation will
be introduced in Section 2.5.
The fact that LP(J:Rn) or LP( (a, b)) contains a dense set consisting of continuous
functions (modulo equivalence) allows one to characterise these spaces in terms of
Riemann integrals. For example each continuous function f defines an elen1ent of
L P( (a, b)) if fab If (a;) IP dx < oo, where the integral can be interpreted as a Riemann
integral. One can then define LP((a, b)) as the nor1n closure of C0 ((a, b)), i.e. one
takes as elements of LP((a, b)) the (equivalence classes of) sequences {fn} such that
each f n belongs to Co ( (a, b)) and limrn,n-+oo J~b Ifn (x)- frn (x) JP dx == 0 (Riemann
integral!); two such sequences {fn} and {gn} are equivalent if their difference con-
verges to zero in LP norm.
For con1pleteness we recall that the support of a function f: 0 ~e, where 0
is an open subset of IR 11 ( n == 1, 2, 3, ... ) , is the closure of the set of points E 0 x
at which f (x) =1- 0. So the support of f is a closed subset of IRn, which need not be
contained in 0 because 0 is assumed to be open. 15 If the support off is contained in
0 and bounded, 16 then f is said to have compact support in 0 (see the exa1nple of
the functions in C0 ((a, b)) mentioned above).

1. 4 . 5 . We end with so1ne re1narks concerning the fact that C0 (IR) is dense in L P (IR)
for 1 < p < oo. This can be proved for example as follows: one shows (1) that
J
1neasurable function f satisfying IR If (.:r) JP dx < oo is a limit (in L P norm) of
a sequence of bounded functions of compact support; (2) that a bounded function of
co1npact support is a limit (in LP norm) of a sequence of continuous functions of com-
support; and (3) that a continuous function of compact support is a 1i1nit (in L P
norm) of a sequence of functions belonging to C 0 (IR). Some details on each of these
three steps are given below. Then one concludes the argument as follows. Consider an
15
If for exa1nple f(:r) 1- 0 at all points of 0, then its support is the closure of 0 in 1R 71 •
16
A subset of lR 71 is conzpact if it is closed and bounded.
MEASURES, INTEGRALS AND L P SPACES 23

element P(JR) and let E > 0. Choose a representative f the considered equiva-
lence class. ( ) there is a bounded function of co1npact support 9 1 in P (IR) such
that II f 91!!P < E /3. By (2) there is a continuous function of compact support 92
1n P(JR) such that 119 1 - 92IIP < c/3. By (3) there exists 93 in C 0 (IR) such that
1192 93IIP < c/3. Then llf- 93IIP < E, which proves the denseness of C 0 (IR) in
£P(JR) (it is easily seen that the result is independent of the choice of a representative
in the equivalence class under consideration).
(1) Let f: IR ---+ C satisfy J~!f(x)jPdx < oo. For n/ E N define fn as follows:
fn(x) == f(x) if lf(x)l < nand j:x;l < n, and fn(:x:) == 0 otherwise. Clearly fn is
bounded (lfn(x)l < n for each x) and of compact support (i.e. fn is zero outside
some finite interval). Furthermore the sequence {fn} converges to fin P nor1n: one
has f n E L P(IR) (because Ifn (x) I < If ( x) I for all x) and
1
II!- fnllv = [hlf(.'£)- fn(x)IPdxr v.

As n ---+ oo the integrand converges to zero almost everywhere (for each x E IR for
which If (x) I oo ), and for each n it is bounded by the function h( x) == If ( x) jP
(because f(x)- fn(x) is equal to f(x) or to 0). Since hE L 1 (IR), the result of (1)
follows from the Dominated Convergence Theorem.
(2) Here one uses the fact that the measurable space (0, A) is special: one has
0- IR and A== AB. In this situation one can approximate a measurable function by
a sequence of step functions; these are particular simple functions, namely functions
of the form rp == ~~= 1 CYkXv", where ak are constants and Vk are intervals of the
form (ak,bk], see Figure 1.4(a). If 9 is a measurable function with lg(:r)l <It! for
all x, there exists a sequence { VJn} of step functions such that VJn (x) ---+ 9 (;x;) and
IVJn (x) I < J\;f for all x.
So let 9 be a bounded function (I 9 ( x) I < 1\II) with support in an interval [- N, + N]
for some N E (0, oo). Then there exists a sequence {9n} of step functions, with sup-
port in the interval [- N, N], such that I9n (x) I < A1 and li1n n-'tcx; g77 ( ;r;) == 9 (x) for
all x E JR. By applying the Dominated Convergence Theorem as in (1) (observe that
l9(x)- 9n(x)IP < (2Af)P and that the domain of integration [-N, N] is bounded)~
one sees that 119 - 9n liP ---+ 0. Hence, given E > 0, one can choose a step func-
tion h with support in [-N, N] such that 119 hi!P < c/2. It is clear that one can
approximate h in P norm by a continuous function h 0 with support in the interval
[- N- 1, N 1] and with II h- h0 liP < E /2 [slightly round off the real part and the
imaginary part of the function h at its points of discontinuity, as illustrated in Figure
1.4(b) for a real-valued step function, or else use a regularisation as in (3) below]. By
the triangle inequality one then has 119- hol!p < 119- h!IP llh- ho!IP <E.

l
(a) (b)

I
Figure 1.4 (a) Step function; (b) approximate step function.
l
24 HILBERT SPACES

(3) One now applies a method of regularisation. 17


Choose a non-negative function 0 E C0 (IR) such that
OCr) == 0 if lx I > 1 and J~ 1 0( x )dx - 1, for example
0 (:r) == 0 if j.T I > 1 and 0 (X) == r exp [-1 I (1 - X2)] if
Ixi < 1, where the constant r is such that J~ 1 O(x )dx ==
1. For 0 < E < 1' set 0c (X) == E- 10 (:r; IE). ClearI y -1 -E c 1
Os E C0 (IR), Os(x) > 0, Os(x) == 0 if lxi > c and
J~oo 0E ( x) dx == 1. A function f E L P (IR) is approximated by a family { f s} of smooth
functions defined by

fs (x) is so1ne sort of average of f over a small neighbourhood of x (over the interval
[x - E, .T E]). For each fixed E the function fs is infinitely differentiable (one may
interchange the derivatives with the integral, and the integrand is infinitely differen-
tiable with respect to x). It can be shown that II!- is liP----+ 0 as E----+ 0.
We consider the situation in which f is continuous and has compact support, say
f(x) == 0 if lx:l > N. Then the support of is is contained in [N - E, N E], be-
cause f(y) is zero if IYI > Nand Os(x - y) is zero if lx- Yl > E. Let us show that
J
II f - is liP ---7 0 as E ---7 0. Since ~00 Or:; (X) dx == 1, we have

fc(x)- f(x) =I:ec(x- y)f(y)dy -I:ec(z).f(x)dz

=I: Be (z) [f (x - z) - .f(x)] dz.

One may consider z f---7 f (x- z) - f (x) as a vector-valued function of z with values in
P (IR, d:r). By using the triangle inequality [for the continuous sum, e.g. as in ( 1.28)],
one finds that

ll.f- !cliP <f:ec(z)ll.f(- z)- !OIIpdz


roo roo 1/p
=Leo dz Bc(z) [Leo d.T If (x - z) - .f(x )IP J

+1 JN+1 ] 1/p
=
!_
1
dw8(w) [ -N-ll.f(x- cw)- .f(x)IPdx . (1.56)

To obtain the last expression we have 1nade the change of variables z f---7 w - z IE and
taken into account the hypotheses that O(x) == 0 if lxl > 1 and f(x) - 0 if lxl > N.
To see that the expression ( 1.56) converges to zero when E ----+ 0, one applies twice the
Dominated Convergence Theorem:
(i) Let us first consider the integral with respect to dx (for fixed w). For each x the
integrand If (x - EW) - f (x) IP tends to zero as E ----+ 0, because f is continuous.
17
See for exan1ple Chapter II of [Adl.
PROBLEMS 25

Furthermore the integrand is bounded: since If (x) I < M < oo for all x E ~' one
has lf(x- Ew)- f(x)IP < (2M)P, which is integrable on [-N- l,N 1]. By
Proposition 1.9 with h(x) == (2M)P we obtain

N+1 1/p
lim [ lf(x- Ew)- f(x)IPdx] == 0.
c:-----+0 J -N-1

We notice that

N+1 1/p JN+1 1/p


[ -N-llf(x
J
~ cw) ~ f(x)IPdx] < [ -N- (2M)Pdx] = 2M(2N
1

(ii) One can now apply Proposition 1.9 to the integral with respect to dw in ( 1.56). It
suffices to observe that, for each fixed w, the integrand converges to zero as E ----+ 0 [by
theresultof(i)] and that it is majorised by the function h(w) == 2M(2N 2) 1 /P()(w).

Bibliographical Notes
The books by Akhiezer and Glazman [AG], Riesz and Sz.-Nagy [RN] and Weidmann
[W1] provide excellent presentations of the theory of Hilbert spaces and linear opera-
tors. There are numerous books on measure theory and integration; the following is a
selective list: [B], [CK], [D], [H], [Na] [R], [Ra], [ST], Volume I of [RS]. £P spaces
are also discussed in many texts, see for example [CV] or [R].

Problems
1
1.1. Let H be a Hilbert space. Verify the parallelogram identity

( 1.57)
and the inequality
N N
2
112:fkll < N2: llfkll 2 · (1.58)
k=1 k=1

1.2. Show that, in a finite-dimensional Hilbert space, weak convergence and strong
convergence of sequences of vectors are equivalent.
1.3. Prove that, if {fn} converges weakly to f and llfnll < 11!11 for each n, then {fn}
converges strongly to f.
1.4. Let {fn} be an infinite sequence of mutually orthogonal vectors, i.e. satisfy-
ing (fj,fk) == 0 if j -=J k, and let FN == L:=1 fk· Show that the following three
statements are equivalent: (i) s-lim N-----+(X) FN exists, (ii) w -limN -----+(X) FN exists, (iii)
LZ':1II fk 11 2 < oo.
1.5. In Proposition 1.5(c), prove that the values of f at the endpoints a and b of the
open interval under consideration are well defined.
SPACES

1 a total a Hilbert space Let { } a bounded sequence in


< that linln-+oo (fn ~g) == \f, g) some f E and each
weakly to Show means an example that

1 any subset of a Hilbert space a subspace of 7-{


equal to subspace spanned by
f 2 is a space.
2
an orthonormal basis in f .

) space of all bounded continuous functions f: IRn ----+ CC.


is a normed (even cotnplete) space for norm llflloo == supxEIRn lf(x)j. Show
(JRrl) In (JRTI).
an exatnple a square-integrable continuous function f on IR does
not satisfy lin11.rl-+oo f (x) == 0.
an example of a sequence {f n} of integrable functions on IR (with respect
converges almost everywhere to a function f but such
not converge to f (x) dx.
fol1owing to P (IR 77 ):
- I - c~ , f (:r) == (1 I ) -a , f (x) - 1
, f(x) == lxl, J(x) == 1
< p < oo, n EN. (:t E IR).
57 definition of
a continuous function vanishing outside some finite interval.
fn by fn (x) - f (x - n). using the fact that the set of
compact support is dense (IR), that the sequence
not strongly 0 (IR).
Pis not a space the norm II· lip·
identity.]
the collection A L all of IR
L<.LJU.._,.._,, V == U Vo,
set a subset so1ne set of measure zero. that A L a
CHAPTER2

Linear Operators

Throughout this text we are concerned with linear 1nappings from a Hilbert space
into itself or into another Hilbert space, called linear operators. the present chapter
we introduce and discuss various types of operators that frequently occur quan-
tum mechanics. Section 2.2 we consider projections, iso1netric and unitary opera-
tors. Section 2.3 is devoted to compact and Hilbert-Schmidt operators and Section 2.5
to multiplication operators L 2 spaces. Often quantum-mechanical operators are un-
bounded. Section 2.4 is concerned with some of the subtleties that one meets in
situations, especially with the notion of closedness of an operator. Section 2.6 we
define the resolvent and the spectrum of closed operators.
A very important class of linear operators are the self-adjoint operators, used to
describe observables quantum 1nechanics. unbounded operators the property
of being self-adjoint is rather subtle. Some preliminary explanations, pointing out the
difference between symmetric and self-adjoint operators, are given in Section 2.4,
and Chapter 3 will be entirely devoted to a detailed study relations between
symmetric and self-adjoint operators. present chapter ends with some criteria
the preservation of self-adjointness under perturbations (Section 2.7); particular we
prove the self-adjointness of some Hamiltonians of non-relativistic quantu1n Inechan-
ics as perturbations of the self-adjoint free Hamiltonian P2 /2rn.

The algebra (H)

Let H be a Hilbert space. We denote by B (H) the set of all bounded linear
operators on H. element of B(H) is a mapping that associates with each vector
f E another vector Af belonging to H, and this 1napping is linear and bounded, i.e.
A( af g) == aAf + and there is a finite constant 1\;f (depending on A) such that
IIAfll < 1\IIIfll for each vector f The infimum of all possible numbers M
28 LINEAR OPERATORS

called the norm of the operator A and denoted by II A II :

II All == inf{ ME JR lilA! II < Mllfll Vf E H}


= sup IIAJII = sup IIAgll· (2.1)
0:/::fEH llfll gEH,jjgjj=l

The elements of B(H) will usually just be called bounded operators.


Observe that (2.1) implies that

IIAJII < IIAIIIIJII VfEH. (2.2)

If A E B(H), then the itnage { Afn} of each strong Cauchy sequence {fn} is again a
strong Cauchy sequence:

s-limfn-f
n-----+c::x) ===? s-limAfn==Af.
n-----+c::x)
(2.3)

This follows from (2.2) which implies that IIAJ- Afnll < IIAIIIIJ- !nil·
The following expression for the operator norm is sometimes useful: If V1 and V2
are arbitrary dense linear manifolds in H (in particular if V 1 == V 2 -H), then

II All- sup l(f, Ag)l. (2.4)


!ED1, gED2,ilfll=llgll=l

PROOF. Let g be a vector in H with 11911 == 1. Let {gn} be a sequence in V2 con-


verging strongly to g. One may assume that ll9n II == 1 [if s-limn-----+c::x) hn == h,
with h 0, then s-limn-----+c::x)(hn/llhnll) - h/llhll, which is easy to see because
llhnll -+ llhll by Proposition 1.1(a)]. Then, by (2.3) and Proposition 1.1, IIAgll ==
lirn n-----+c::x) II Agn I . Hence, if we restrict the vectors g in the last expression in (2.1) to
the set {g E V 2 11911 1}, we obtain the same supremum:

IIAII == sup IIAgjj. (2.5)


gED2,IIgll=l

Next lethE H, h -::J 0. Then, if 11!11 == 1, one has l(f, h) I < llfllllhll == lihll, with
equality for f - h/llhll· So llhll == sup!EH,II!II=ll(f,h)l. By using an argument
as above (approximating a vector f E H, 11!11 - 1, by a sequence {fn} in V1, with
II frz II == 1), one sees that
llhll == sup l(f, h) I. (2.6)
fEDl,lifil=l
By taking h == Ag in (2.6) and inserting this expression into (2.5), one obtains the
formula (2.4). D

Let us briefly look at the structure of the set B(H). It is clear that B(H) is a linear
vector space over the field C: if A, B E B(H) and a E C, then aA B is again a
bounded linear operator, given by (a A B) f == aAf B f, and it is bounded:

llaA Bll < laiiiAII IIBII, (2.7)


THE ALGEBRA B(H) 29

because ll(aA B)fll < llaAfll IIBJII < laiiiAIIIIfll + IIBIIIIJII [we have used
(1.13)]. B(H) is an algebra, i.e. a vector space equipped with an operation of multipli-
cation. The product of two bounded operators A and B is simply denoted by AB and
is just the composition of the two mappings: ( AB) f == A( B f) is the vector obtained
from f by acting first with B on f and then with A on the image vector B f. The mul-
tiplication B (H) has the following properties:

(aA B)C- aAC + BC and C(aA B)== aCA CB.

To see that AB is bounded, observe that IIABJII < IIAllllBfll < IIAIIIIBIIIIJII as a
consequence of (2.2), so that
IIABII < IIAIIIIBII· (2.8)

2.1.2. The algebra B(H) is involutive (called a *-algebra). The involution, denoted
by an asterisk, associates with each operator A E B(H) another element A* of B(H),
called the adjoint of A, such that (A*)* == A [one usually writes A** for (A*)*].
Given A, its adjoint A* is defined by the relation

(A *j, g) == (f, Ag) Vf, g E H. (2.9)

This relation determines uniquely an element A* of B(H): for fixed f, the mapping
g ~ (f, Ag) is a bounded linear functional on H (it is linear in g, and I (f, Ag) I <
IIJIIIIAIIIIgll). By virtue of the Riesz Lemma (Proposition 1.8) there exists a unique
vector f *E H such that (f, Ag) == (f *,g) for each g E H. One sets A*f == f *, which
defines A* on each vector f of H. We must check that the thus-defined mapping A*
is linear and bounded:
(i) One has

(A*(af h),g)==(af+h,Ag) a(f,Ag) (h,Ag)


== a(f*,g) + (h*,g) == a(A*j,g) (A*h,g)- (aA*f + A*h,g).
So A* (af + h) a A* f - A* h is orthogonal to each g E H, hence equal to the zero
vector (Proposition 1.6), i.e. A* (af + h) == aA *f + A* h.
(ii) By using (2.4) one finds that

IIA*II == sup l(g,A*f)l == sup I(Ag,f)l == IIAII·


j, gEH, II !II= 11911=1 j, gEH, 11 fll=ll9ll=l

Hence A* is bounded and


IIA* II== IIAII· (2.10)
1
The definition (2.9) also implies that, if A,B E B(H), then

(AB) * == B*A * and A**== A. (2.11)


1 The relation A** = A is obtained from Proposition 1.6 by observing that (A** j, g) = (j, A* g) =
(Aj, g) for all j, g E 'H.
30 LINEAR OPERATORS

Another useful relation is the following:

(2.12)

Indeed one has on the one hand IIA *All < IIA *II II All == IIAII 2 and, on the other hand,

An operator A E B (1-i) is self-adjoint if == A, i.e.


(Af, g) == (J, Ag) \1 j, g E (2.13)

Together with (1.5), Eq. (2.13) implies that (j, Af) is real for each f E 1t if A is self-
adjoint. nonn of a bounded self-adjoint operator A can expressed as follows:

II II == sup I (f, A f) I~ (2.14)


fED,IIJII=l

where V an arbitrary dense linear tnanifold 1-i.

We set a== supJED,IIJII=ll (j, Af) I· Observe that I(h, Ah) I < ailhll for each
2

hE V (because h/JihJJ is a vector of norm 1 if h 0).


The inequality a < II A II follows from (2.4). To obtain the converse inequality we
must that a> I(f, Ag) I for all j, g E V with 11!11 == 11.911 == 1. Now I(J, Ag) I ==
I for each real a, and for fixed f and g one can find a real number a such
(e~ 0
Ag) is real and non-negative. Hence it suffices to show that a > (f, Ag)
f. g E V with llfll == 11.911 - 1 and such that (f, Ag) E JR.
we use the following generalisation of the polarisation identity ( 1.11 ):

4(f,Ag)-(f g,A(f+g)) -g,A(f-g))


- i(j ig, A(f ig)) i(f-ig,A(f-ig)). (2.15)

(h, real each h; E 1-i, the real part of 4(j, Ag) is given the su1n of
first two terms on right-hand side of preceding equation. So, iff and g are
such that (f, E IR, one has [by also using (1 .33)]

(f. = ~ [(f 9,A(.f+9))-(f-9,A(.f-9))]

< ±[l(f+9,A(.f 9))1 l(.f-9,A(.f-9))l]


1
< 4 [allf + .911 2 +all!- .911 2 ]
= ~ [2llfll 2 + 211911 2 ] =a if IIlii = 11911 = 1. D
31

}nEN
II ----t 0 as n ----t x.

are as
is clear convergence of } to i1nplies weak convergence
a consequence of Proposition 1.1 ).
Convergence norm imp lies strong convergence: I II ----1 one
fE :II fll <II llllfll----t
} converges weakly to then
, because f, g) == (f. Ang) ----t
convergence of a sequence }
sequence {A~}
{An} each 1s lS
g) == li1n 17 -joo [(f,

a sequence {An to
lilnnl,n~oo[(f,A1ng)- (f,Ang)] 0 for all f,g E
E B(H) that w-lin1 77 ~CX) ==A. Analogous
are strongly and for sequences are Cauchy nonn. We
proof of these statements (it involves Uniform Principle
end Section 1.1).

Let B, , Bn E (H) be such that n~oo and


== s- n~oo Bn. Then the sequence l.S sat-
isfies s -lirn n~oo Bn ==

PROOF. We use that, as a consequence of Boundedness


ciple (Section 1.1), a strongly (or weakly) convergent sequence of
bounded: there is a constant M < oo such that II II < JJI[ all n E N (in our
applications of Proposition 2.1 the validity of will be evident).
So let f E 1-{. Then
32 LINEAR OPERATORS

IIABJ- AnBnfli II(A An)Bf + An(B- Bn)fli


< II(A An)BJII IIAn(B- Bn)JII
< II(A An)BJII M!I(B- Bn)J!I, (2.16)

which converges to 0 when n ----+ oo. 0

Proposition 2.1 remains true if one replaces everywhere strong convergence by


norm convergence. However it has no analogue for weak convergence (i.e. if {An}
and { Bn} are weakly convergent sequences, then the product sequence {AnBn} may
not be weakly Cauchy).

Proposition 2.2. Let {An}nEN be a sequence in B(H) such that IIAnll < M < oo
.for all n E N. Let M be a subspace of H and suppose that there is a subset N of M
which is total in M and such that, for each f E N, the sequence { Anf} is strongly
Cauchy. Then the limit s-limn-+CXJ Ang exists for each g E M. If furthermore A is
an operator in B(H) such that s-limn-+CXJ Anf Af for each fEN, then one has
s-lim n-+CXJ Ang - Ag for each g E M.

PROOF. Let£ be the linear manifold spanned by N (so that M == £). It is clear that
{ Anf} will be strongly Cauchy for each f E £. So let g E M and E > 0; choose a
vector hE£ such that llg- h!l < E/ (3M). Then by the triangle inequality:

IIAmg- Angll < IIAm(g- h)ll IIAmh- Anhli + I!An(h- g)ll


< 2MIIg- hll IIAmh Anhll < 2c/3 IIAmh- Anhll.

As {Anh} is strongly Cauchy, one has IIAmh- Anhll < E/3 provided that m, n >
N(E) for some N(E) E (0, oo ). So IIAmg Angll < Eprovided that m, n > N(E ), i.e.
{ Ang} is strongly Cauchy. The proof of the second part of the Proposition is similar,
it suffices to replace Arn by A in the arguments given above. 0

2.1.4. We continue with our remarks on the structure of B(H). We have seen that
B(H) is a *-algebra. We have also seen that B(H) is a normed algebra, the norm
being given by (2.1), and that B(H) is complete with respect to this norm [a norm
convergent sequence of elements of B(H) has a limit in B(H)]. A complete normed
algebra is called a Banach algebra. Here we even have a *-algebra, in which case one
uses the term B*-algebra (a complete normed algebra equipped with an involution).
A B*-algebra in which the identity (2.12) (i.e. IIA *All == !lA 11 2 ) is satisfied is called a
C*-algebra. The notion of a C*-algebra is used for abstract algebras; in our context
the algebra B(H) is realised by operators acting in a Hilbert space H, and there are
other topologies on B(H) than just the norm topology (in particular the strong topol-
ogy and the weak topology). A C*-algebra of operators on a Hilbert space H (i.e. a
*- subalgebra of B(H) that is complete in the norm II · II) and which is closed in the
weak topology is called a W*-algebra or a von Neumann algebra. In particular B(H)
itself is a von Neumann algebra.
THE ALGEBRA B(H) 33

2.1 . 5 . In analogy with vector-valued functions discussed in Section 1.2, one may con-
sider functions defined on a set A with values in B (1-t). An important case in what
follows (see Section 2.6) is that in which A is an open subset of the complex plane
CC. Here we consider the case A - J, where J is an interval in JR. We write A( s) for
the value of a function A: J ---+ B (1-t) at the point s E J. Again there are various no-
tions of continuity and differentiability of such functions (uniform, strong or weak)
and different possibilities for defining integrals of such functions. We recall below a
few definitions. For the strong and the weak topologies one refers to the case of vector-
valued functions already considered in Section 1.2 by acting with A ( s) on the vectors
of1i, i.e. by considering for each vector f E 1i the vector-valued functions~ A(s )f.
A function A : J ---+ B (H) is strongly continuous if, for each f E 1i and for each
t E J, one has
lim II A( s )f -A( t)fll == 0. (2.17)
s----+t,sEJ

It is continuous in norm if one has for each t E J:

lim IIA(s)- A(t)ll 0. (2.18)


s----+t,sEJ

A function A: J ---+ B (1-t) on an open interval J is strongly differentiable on J if


there exists a function A': J ---+ B(1i) (called the strong derivative of A) such that for
each f E 1i and for each s E J:

lim
T----+0
II~[A(s
T
+ T)j ~ A(s)f] ~ A'(s)fll = 0. (2.19)

It is differentiable in norm on J if there exists a function A': J ---+ B(1i) such that
for each s E J:
lim II~[A(s + T) ~ A(s)] ~ A'(s)ll = 0. (2.20)
T----+0 T

As regards integrals, we shall use only Riemann integrals in the strong sense. If J
is a finite interval, II== {so, s 1 , ... , sN; u 1 , ... , UN} a partition of J (as Section
1.2) and A: J---+ B(1i) an operator-valued function, we define
N
L:(II,A)- L (sk Bk-l)A(uk)· (2.21)
k=l

L:(II,A) is a finite sum of operators in B(H), so (2.21) defines an operator in B(H).


One then takes a sequence {IIr }rEN of partitions of J with limr----+oo IIIr I 0 and sets
for each f E 1-t:
Nr

[
1
J
A(s) ds] f- s-lim L:(IIr,A)f ==s-lim"" (Sk - Bk-l)A( uk)f,
r----+oo r----+oo~
k=l
(2.22)

if the limit exists and is independent of the sequence {IIr}. This defines the operator
JJA(s)ds [Eq. (2.22) gives its action on each vector f E 1-t].
b

are
we

f(s)

I II< II I
I (s)ll < oc,

j3gE f }.
THE ALGEBRA B(H) 35

mapping A - 1 : R( A) ---+ 1t satisfying A - 1 Af == f for all f E 1t and AA - l g - g


for all g E R( A). In our context an interesting situation is that in which A - 1 also
belongs to B(H); we then say that A is invertible in 13(?-l). For this to be the case it
is necessary that R( A) == 1i and that A - 1 is bounded (in fact the latter condition is
superfluous, see the footnote on page 68).
We denote by I the identity operator defined by If == f for all f E 1i and by 0 the
zero operator (0 f == 0 for all f E 1-l). Clearly I is invertible in B(H), with I- 1 == I.
It is useful to know that operators close to the identity operator are also invertible in
B(H). More precisely, if A E B(H) is such that \\A II < 1, then I - A is invertible in
B(H):

B(H) with \\A\\ < 1. Then I - A is invertible in B(H),


Proposition 2.5 . Let A E
inparticularR(I- A)== 1t and (I- A)- 1 E B(H). Furthennore \I(I- A)- 1 \\ <
[1- !IAIIJ- 1,
and (I- A)- 1 is given by the following series which converges in the
uniform topology (in operator norm), called the Neumann series:
(X)

1
(I- A)- ==I+ A+ A 2
+ ... == L Ak. (2.27)
k=O

PROOF. The series (2.27) is convergent in norm, because by (2.8) and (2.7) we have
forn > m
n n-m, (X)

1
II L Akl\ < \\Am\1 L IIAk\1 < \\AI\m L \\AI\k == 1\A\\m[l-\\A\\]- ,
k=m k=O k=O

which converges to zero as m ---+ oo since I A II < 1. Thus the series (2.27) defines an
operator B in B (1-l), with
(X)

1\BII ==II LAk\1 < [1-I\AII]-l < 00.


k=O
We now observe that

(I - A) (I + A + A 2 + · · · + An) == I - An+ 1 . (2.28)

As 11An+ 1 l < IIA lln+ 1 ---+ 0 as n---+ oo, Eq. (2.28) implies that
(X)

(I- A)B (I- A) (LAk) =I.


k=O

Consequently R(I- A)== H. A similar calculation shows that (LZ:o Ak)(I- A)


B(I- A)== I. Hence I- A is invertible and (I- A)- 1 ==B. In particular (I- A) 1
is given by the series (2.27). D
36 LINEAR OPERATORS

2.1.7. So far we have considered operators mapping 1t into H. Occasionally we


deal with operators between different Hilbert spaces. Let 1t and JC be two Hilbert
spaces. We denote by B(1t, JC) the set of all bounded linear operators from H to /C.
An element of B (1t, /C) is a linear mapping that associates with each vector f E H a
vector Af E /C, with
IIAJIIJC
IIAII - sup
0#-fE'H
11!11 }{ < oo. (2.29)

Here II·IIH denotes the norm in Hand II·IIJC that in JC. The properties of B(1t, JC) are
analogous to those of B(H); the adjoint is defined by

(A *j, g)H == (j, Ag) JC VjEJC and VgE1-l. (2.30)

A* is an operator that maps JC into 1-l, i.e. A* E B(/C, 1-l). Also, if A E B(1t, JC) is
invertible, then A - l is a mapping JC ----+ H [defined on R( A) C JC].

2.1.8. When the Hilbert space 1t is an £ 2 space, some care may be required to make
sure that an operator, often given by its action on individual functions, is indeed a
1napping between equivalence classes of functions. We mention here a few frequently
encountered situations, assuming that H == £ 2 ( 0, m).
An operator A in £ 2 ( 0, m) is said to be an integral operator if there exists a
1neasurable function a: 0 x 0 ----+ C such that for all f in the domain of definition of
A (this latter concept will be discussed in §2.4.1):

(Af)(x) = l a(x, y)f(y)m(dy). (2.31)

The function a is called the kernel of A. Observe that the value of the integral in (2.31)
is the same for all functions f belonging to the same equivalence class, so that A maps
all representatives of an equivalence class onto the same function (A f) (x).
A multiplication operator A in £ 2 ( 0, m) is an operator given as follows:

(Af)(x) - a(x)f(x). (2.32)

The value of the function Af at the point x E 0 is obtained by multiplying the value
of f at the point x by the number a( x), where a: 0 ----+ C is a fixed measurable
function. In other terms the function Af is the product of the two functions a and
f. It is clear from (2.32) that equivalent functions get mapped onto equivalent func-
tions. Properties of 1nultiplication operators are presented in Section 2.5.
Somewhat more delicate are differential operators, i.e. operators involving deriva-
tives of the functions f on which they act. Their definition on equivalence classes may
involve subtleties of differential calculus which will not be discussed here. Usually
such operators will be defined only on some dense subset of the Hilbert space (on equi-
valence classes containing a sufficiently differentiable representative f). We shall be
more explicit when treating specific classes of differential operators.
PROJECTIONS AND ISOMETRIES 37

2.2 Projections and isometries

2.2.1 . The Projection Theorem (Proposition 1. 7) shows that, if M is a subspace of a


Hilbert space H and M j_ its orthogonal complement, then each vector f in H can
be written in a unique way as f == !I + f 2 with fi EM and f 2 E Mj_. The corre-
spondence f t-----7 !I defines a linear mapping PM on H with range R(PM) == M,
called the projection onto M (it would be more precise to use the term orthogonal
projection, because PM annihilates the orthogonal complement of M; as we consider
only orthogonal projections, we usually omit the specification "orthogonal" and sim-
plyusethetermprojection).Since IIJIII 2 < IIJIII 2 +IIf2ll 2 == 11!11 2, weseethatPM is
bounded, with I PM I == 1 (if dim M > 1). Consequently PM E B(H). It is also clear
that P~ == PM. In addition one has PM == PM. Indeed, iff and 9 are decomposed
as f - !I+ !2 and9- 9I + 92 respectively, then (PMf,9) == (fi,9l == (JI,9Il ==
(f,gi) == (f,PMg) (usingtherelationM _L Mj_).
It is interesting to know that the two properties P~ == PM and PM - PM char-
acterise an orthogonal projection. In our context an operator E E B(H) is called a
projection if E* == E == E 2. Given a projection E there exists a (unique) subspace
M of H such that E - PM, i.e. such that E is the orthogonal projection onto M.
Indeed, set
M M(E)=={fEHiEf f}. (2.33)
It is easy to see that M is a subspace (this also follows from Proposition 2.38 proved
in Section 2.6). If 9 M, then one has for each hE H:

(E9,h)- (9,E*h)- (9,Eh). (2.34)

Since E 2 h == Eh, one has Eh EM [set f == Eh in (2.33)]. Consequently (2.34)


implies that (Eg, h) == 0, hence Eg == 0 by Proposition 1.6. So, if a vector f E His
decomposed into f == !I+ !2 with !1 EM and !2 EM j_, one has E f == Ef1 + E !2 ==
fi 0, i.e. Ef == PMf·

RESUME. With each subspace M there is associated a projection PM describing the


orthogonal projection of the vectors of H onto M, and with each projection E is asso-
ciated a subspace M(E) of H such that PM(E) - E. This subspace is just the range
of the operator E: M(E) == R(E). It is evident that PMj__ ==I- PM and PH ==I,
P{o} == 0.

It is clear that geometric relations between subspaces correspond to algebraic re-


lations between the associated projections. In the next proposition we specify some
important examples, without giving the details of the proof.

Proposition 2.6. Let M and N be subspaces of H, and let PM and PN be the


associated projections.
(a) If PMPN == PN PM, then PMPN is a projection, the associated subspace being
M nN, i.e. PMPN PMnN·
(b) If M C N, then PMPN- PNPM- PM.
38 LINEAR OPERATORS

(c) If M _L N, then PMPN == PNPM == 0 and PMtBN ==PM PN.


rr
(d) PMPN == 0, then M N.
{En} is a strongly convergent sequence of projections and E denotes their limit
(i.e. E == s-limn-+oo En), then E is also a projection. Indeed E is self-adjoint [see
point (4) on page 31] and E 2 == s-lim n-+oo E~ == s-lim n-+oo En == E, where the
first equality follows from Proposition 2.1.

2.2.2. An operator 0 belonging to B (1-l) is called an isometry (or an isometric opera-


tor) if
o*o ==I. (2.35)

Proposition 2. 7. Let 0 be an isometry. Then


(a) ~l preserves all scalar products:

(OJ, Og) == (j, g) Vj,gEH. (2.36)


In particular:
IIOJII-11!11 \If E 1-l. (2.37)

(b) diln H -1- 0, then II 0 II == 1.


(c) 00* is a projection, its corresponding subspace is M(OO*) == R(O).
(d) 0 is invertible.
(e) One has 0* g - o- 1 g if g E R(O) and ~1* g == 0 if g _L R(~1).
(f) an operator WE B(H) satisfies IIWJII == llflifor all vectors f E 1-l, then W is
an iso1netry.

PROOF. (a) (OJ, Og) == (f, O*Og) == (j, g).

IIOJII 11111
(b)
IIDII = o~~~J-t IIIII = o~~~J-t ffi = 1.

(c) Setting F == [10* and using (2.35) and (2.11), we obtain

F 2 == 0(0*0)0* == OIO* == f20* == F and F* == (00*)* == 0**0* == 00* ==

is a projection. By (2.33) we have M(F) == {! E H I OO*f - f}. Conse-


quently, iff E M(F), then f E R(O), i.e. M(F) C R(~1). On the other hand, if
g ==Of E R(O), then Fg == OO*g- OO*Oj == OIJ ==Of== g, which shows that
g E M(F). Thus R(~1) c M(F).
(d) If Of == 0, then 11!11 == IIOJII == 0 by (2.37), hence f == 0. This shows that 0
is invertible [but not necessarily invertible in B(H), because o- 1 is defined only on
M(F)].
Let g - Of E R(O). Then o- 1 g - f - O*Of == O*g. On the other hand, if
g _L R(O), then (O*g, f) - (g. Of) - 0 for all f E H, hence O*g is orthogonal to H,
so that O*g == 0 by Proposition 1.6.
==I can ( 1. ):

2 2 2 2
11 II 11 iII 11 iII (f-. 11

+ 911 2 - llf 9!1


2
- illf + 11
2
+ illf- 11
2

- 4(f~ D

gE one
40 LINEAR OPERATORS

consequently U0 == (U*) 0 . This implies that U0U0 == (U*) 0 U0 . But iff EM, one
has (U*) 0 U0 f == U*U f == f, so that U0U0 == ! 0 the identity operator in the Hil-
bert space M. Similarly one shows that U0 U0 - ! 0 . This establishes the unitarity of
U0 (as an operator in M).
(b) This is easily obtained by taking into account ( 1.31 ). D

If Hand}( are two Hilbert spaces, an operator V E B(H, JC) is called unitary if it
is isometric (i.e. IIVJIIK == llflln for each f E 1-l) and surjective (i.e. R(V) == JC).

2.2.3 . A generalisation of the notion of isometry is that of a partial isometry. An oper-


ator 0 belonging to B (H) is called a partial isometry if

0*0 == E (2.39)
'
where E is a projection. In the next proposition we collect some properties of this type
of operators.

Proposition 2.9. Let 0 be a partial isometry. Then


(a) OE-0. (2.40)
(b) (OJ, Og)- (Ej,Eg) Vj,gE1t. (2.41)
(c) One has 11011 == 1 except when E == 0.
(d) 00* is a projection, and the associated subspace is M(OO*)- R(O).
(e) Let M be a subspace and W: M ---+ H a linear operator satisfying IIW!II - IIJII
for all f E M. If one defines an operator 0 by Of == W f iff E M and Of - 0 if
f E M j_' then 0 is a partial isometry with 0 *0 p M.
PROOF. (a) Let f E H. Then we have with (2.39):

IIOEJ Ofll 2 == (O(Ef- f), O(Ef- f))- (Ef j, O*O(Ef f))


== (Ef j, E(Ef f))- 0,

since E(Ef- f) - E 2 f Ef - Ef Ef - 0. Thus OEJ- Of == 0, hence


OEJ- Of.
The proof of (b) - (e) is similar to that of the corresponding statements in Proposi-
tion 2. 7 and is left as an exercise. D

One sees that a partial isometry 0 maps the subspace M (E) isometrically onto the
subspace M(F), where 00*, whereas Of == 0 for f M(E) [iff M(E),
then by (2.40): Of == OE f == S10 == 0]. In the language of operators between two
Hilbert spaces, 0 would be called a unitary operator from M (E) to M (F). The sub-
spaces M (E) and M (F) are called the initial set and the final set of 0 respectively.
COMPACT OPERATORS 41

2.3 Compact operators

Compact operators in an infinite-dimensional Hilbert space have properties that are


very similar to those of operators in a finite-dimensional Hilbert space (for example
they have no continuous spectrum, see Example 4.24). These operators are very useful
in numerous theoretical developments and appear in various situations in this text.

2.3.1. We first consider operators A defined by equations of the form


N

Af == L (gk, f)hk, (2.42)


k=1
where {gk, hk} are 2N fixed vectors in Hand N < oo. In the notations used by physi-
cists, A is written as A == 2::~= 1 lhk) (gkl· The range R(A) of A is contained in the
finite-dimensional subspace spanned by h 1 , ... , hN, and Af == 0 if .f is orthogonal to
the subspace spanned by g 1 , ... , g N. Thus A may be considered to be an operator act-
ing in the finite-dimensional subspace M spanned by {g 1 , ... , gN, h 1 , ... , h N} in the
sense that it is zero on M _L and its range is also contained in M. A can be described
by a matrix acting on the vectors in M. An operator of the form (2.42) with N < oo is
called a .finite rank operator. One can show that each operator A with dim R(A) < oo
is a finite rank operator. Some simple properties of this type of operators are given in
the following proposition:

Proposition 2.10. Let A be a finite rank operator. Then:


(a) A belongs to B(H).
1
(b) A* is a finite rank operator.
(c) lfC E B(H), then CA and AC are finite rank operators.
(d) If B is a second finite rank operator and a E C, then A + aB is also a finite rank
operator.

PROOF. (a) This follows from the estimate IIAJII < 2::~= 1 ll9kllllflllihk\l, which im-
plies that II A II < 2:::~= 1 \lgk II II hk 11. For (b) we observe that fore, f E H:

N N
(A*e,f) == (e,Af) == L\gk,f)(e,hk) == ((L\hk,e)gk),J),
k=1 k=1
which shows that A*e == I:f:=
1 (hk, e)gk. Thus A* has the same form as A but with 9k
and hk interchanged; one has A* == L:f:= 1 !gk) (hkl· For (c) it suffices to observe that
CAJ == I:f:=
1 (gk,f)Chk, which shows that CA == L:f:= 1 !Chk)(gkl· Also ACJ ==
I:f:=1 (C*gk,f)hk, i.e. AC == 2:~= 1 !hk)(C*gkl· Finally (d) is evident. D

An operator A E B(H) is compact if there exists a sequence {AN} of finite rank


operators such that II A AN II ---+ 0 as N ---+ oo, i.e. if A is the norm limit of a sequence
E

1V
I II< I I I liS + ,cpn)lll II·
k=l

no EN I I < c/
n >no. II II< E n>
n
COMPACT OPERATORS 43

(b) Set Cn == Bn B, so that s-limn-7ex; Cn == 0. There exists a real number M


such that IICnll < M for each n. Let {AN} be a sequence of finite rank operators
satisfying IIA- AN I - 7 0 as N - 7 oo. Then [assuming AN to be of the form (2.42)]:

<MilA- ANIIIIJII + L l(gk,f)IIICnhkll


k=l
N

< [MilA- ANI!+ L ll9kiiiiCnhk11] ll.fll·


k=l

Given E > 0 we first choose N such that IIA- ANll < E/(2NI) and then a number
no such that ll9klliiCnhkll < E/(2N) for all n > no and each k == 1, ... , N. One
obtains IICnAII < E n > n 0 , which proves the first assertion in (b). The second one
follows from the first one by observing that IIAB~- AB*II == IIBnA*- BA*II and
by remembering that A* is also compact. D

Example 2 . 13 . A projection P == PM is compact if and only if dim M < oo. Indeed:


(i) if dim M < oo, PM is a finite rank operator, given by PM == L~:tt Iek) (ek I for
any orthonormal basis {ek} of M. (ii) dimM == oo, one can choose an infinite
orthonormal sequence {en} in M (for example an orthonormal basis of M). This
sequence converges weakly to 0 but is not strongly convergent (see Example 1.2); so,
by Proposition 2.12(a), PM cannot be compact.

2 . 3 . 2 . There are various special classes of compact operators that are of consider-
able use in the mathematical study of quantum-mechanical problems. We present here
the most important one for our purposes, namely the class of Hilbert-Schmidt opera-
tors. These operators are useful for proving compactness of certain operators and play
a role in scattering theory for the description of scattering cross sections.
The Hilbert-Schmidt norm IIA IIHs of an operator A E B(H) is defined as follows:

(2.43)

where { ek} is an orthonormal basis of H. A priori this number depends on the chosen
orthonormal basis, and it may be infinite. Let us first show that IIA IIHs depends only
on A, i.e. that the sum in (2.43) is the same in each orthonormal basis of H. So
{gk} be an arbitrary orthonormal basis of H. Then, by developing Agk in the basis
{ek} (see Section 1.1), one has

k k j k j J

(the interchange of the order of summation is permitted because all tenns are non-
negative). By considering the basis { ek} as being fixed, one sees that Lk llAgk 11 2 is
44 LINEAR OPERATORS

identical to the same number I:j IIA *ej 11 2 for each orthonormal basis {gk} and hence
independent of this basis. The equation (2.44) also shows that

IIAIIHs == IIA*IIHs· (2.45)

An operator A EB(H) is called a Hilbert-Schmidt operator if IIA IIHs < oo. The set
of all Hilbert-Schmidt operators in H will be denoted by B2 (H) or simply by B2 .

Proposition 2. 14. (a) One has A E B2(H) if and only if A* E B2 (H).


(b) Each Hilbert-Sch1nidt operator is compact, i.e. B2 (H) C JC(H).
(c) If A1 EB2(H) and A2 EB2(H), then A1 + aA2 EB2(H) (a EC).
(d) One has
IIAII < IIAIIHS· (2.46)
(e) If A EB2(H) and BE B(H), then ABE B2(H) and BA EB2(H), with
and (2.47)

PROOF. The result of (a) follows from (2.45) and that of (c) from the definition (2.43)
together with the inequality ( 1.14):

IIA1 + aA2II~s == L II(A1 + aA2)ekll 2 < 2 L IIA1ekll 2 + 2 L lai 2IIA2ekll 2


k k k
== 2IIA1II~s + 2lai IIA211~s·
2

For (d), let f f:. 0. Choose an orthonormal basis { ek} of H such that e 1 == f /llfll.
Then
11
;~1 1f = IIAe1ll 2
< L IIAekll = IIAII~s·
2

Since this inequality holds for each f f:. 0, one obtains (2.46). For (e) one has

IIBAII~s- L 2
IIBAekll < IIBII
2
L 2
IIAekll == IIBII IIAII~s'
2

k k

which gives the second inequality in (2.4 7). The first one then also follows: II AB I HS ==
IIB*A*IIHs < IIB*IIIIA*IIHs == IIBIIIIAIIHs [by(2.45)and(2.10)].
It remains to prove (b). If dim H < oo, this is trivial. So assume that dim H == oo
and fix an orthonormal basis { ek} of H. Let A be a Hilbert-Schmidt operator. Define
AN as follows by its action on f == 2:~ 1 akek: AN [ 2:~ 1 akek] == 2:~= 1 akAek;
so ANen == Aen if rn < N whereas ANen == 0 if n > N. AN is a finite rank operator:
ANf == 2:~= 1 \ek,f)Aek, and IIA- ANII~s == I:~N+ 1 11Aekii 2 -r 0 as N -roo
because A E B2. This implies with (2.46) that IIA - AN II -r 0 as N -r oo, hence
A E JC(H). D
In an L 2 space the Hilbert-Schmidt operators have a simple and very useful char-
acterisation as integral operators. Let H == L 2 ( 0, m). We recall that an operator A is
said to be an integral operator if there exists a measurable function a: 0 x 0 -r C,
COMPACT OPERATORS 45

called the kernel of A, such that for all fin the domain of definition of A (see §2.4.1
for this concept):
(Af)(x) = l a(x, y)f(y)m(dy). (2.48)

A measurable function a: 0 x 0 ~ C is called a Hilbert-Schmidt kernel if

Ma :=fa fa la(x,yWm(dx)m(dy) < oo, (2.49)

in other terms if the function a belongs to L 2 ( 0 x 0, m x m).


Proposition 2 . 15 . Let H == £ 2 ( 0, m ). Then A E B 2 (H) if and only if A is an integral
operator with Hilbert-Schmidt kernel. Furthermore

(2.50)

REMARK. The preceding proposition, combined with (2.46), illustrates the usefulness
of Hilbert-Schmidt operators: in certain cases it is possible to estimate the norm of an
operator by calculating or estimating an integral. Another result of this type is given
in §2.5.5.
PROOF OF PROPOSITION 2.15. To simplify the notations we do not specify the mea-
sures in the occurring £ 2 spaces (this concerns either the measure rn or the product
measure m x m ). We use the fact that £ 2 ( 0 x 0) == £ 2 ( 0) 0 £ 2 ( 0), the tensor prod-
uct of £ 2 ( 0) with itself. More precisely we use the following fact: if the functions
{ek(x)}kEN form an orthonormal basis of £ 2 (0) and ek denotes the complex conju-
gate of ek, then the collection { ek ( x)} kEN is also an orthonormal basis of £ 2 ( 0), and
the functions { ej ( x )ek (y)} j,kEN give an orthonormal basis of £ 2 ( 0 x 0).
(i) Let A be an integral operator with Hilbert-Schmidt kernel a(x, y), and let V
{x E 0 I J0 !a(x, y) 2 m(dy) == oo }. Then m(V) - 0 because a(·,·) E £ 2 ( 0 x 0).
1

Now let f E £ 2 ( 0) and x ¢:. V. If the integral in Eq. (2.48) is interpreted as a scalar
product, one obtains by applying the Schwarz inequality that

I(Af)(xW <fa la(x, vWm(dy) ·fa lf(v'W m(dy').


Integration of this inequality with respect to x leads to IIAJII 2 < Ma llf\1 2 . This shows
that A associates with each vector f of 1i an element Af of 1i and that A E B(H)
with IIAII <]}fa.
Let {ek} be an orthonormal basis of £ 2 (0). By setting o:1 k == (e,1,Aek) we have
Aek == ~jO:]kej and

(2.51)
k j,k

Let us also consider the orthonormal basis { e.7ek} of £ 2 ( 0 x 0). One can develop
a(·,·) in this basis: a == ~j,k Pjkejek, and one then has Ma == ~j,klf3jkl 2 . Now
observe that
LINEAR OPERATORS

a1 k = (c 1 ,Aek) = .Ia fa ej(x)a(x, y)ek(y)m(dx)m(dy)


== (ejek, a) £2(0xO) == /3jk· (2.52)

Eqs. (2.51) and (2.52) imply that IIAII~s == I: 1·' klajkl 2 ==I:.J, klf3jkl 2 == Ma. In par-
ticular A belongs to B2 .
(ii)For AEB2 , wedefineajk asaboveandsetaN(x,y) == I:·.k<Na]ke J, - 1 (x)ek(y),
where N == 1, 2, .... Notice that aN is a finite linear combination of functions be-
longing to L 2 (0 x 0), hence aN E L 2 (0 x 0). If M > N we have

N !vi M M

fo1aM(:r:, y) ~aN (x, y)l 2


m(dx)m(dy) = 2:::: 2:::: laJkl 2
+ L L laJkl 2

J=l k=N+l J=N+l k=l


(X) (X) (X) (X)

j=l k=N+l j=N+l k=l

Since A E B2 , the double sum in (2.51) is convergent. Thus the preceding inequality
implies that {aN} is a strong Cauchy sequence in L 2 ( 0 x 0). We denote its limit by
a 00 • By (i) above this function a 00 is the kernel of a Hilbert-Schmidt operator which
we denote by A 00 • To know that A is an integral operator with Hilbert-Schmidt kernel,
we still must prove that Aoo == A. Since each of these two operators belongs to B(1i),
it suffices to verify that Aoo e k - Aek for each vector ek of the basis { e.J}. We know
already that Aek == I: .7 a 1 kej. If we write Aooek == I:..7 rjkej, we have by (2.52):
~l.7k == (cJ, Aooek) == (e 1 ek, aoo) £2(0xO) == J~ (ejek, aN) £2(0 xO) - O:jk. D

Remark 2. 16 . We mention some further interesting properties of the set B 2 (1i) (for
more details and proofs, see e.g. Chapter 3 of [A] or Section VI.6 of [RS]). For A, BE
B2 (1i), set
(A,B)- L
(Aek,Bek) == TrA*B, (2.53)
k

where } is an orthonormal basis of1i and Tr stands for trace. From the polarisation
identity ( 1. 11) one finds that

(A, B) = ~(II A+ Bll~s- IIA- Bll~rs +iliA- iBII~rs ~iliA+ iBII~s), (2.54)

shows that the sum in (2.53) is finite and independent of the chosen orthonor-
basis.
(a) Eq. (2.53) defines a scalar product on the set B 2 (H) of all Hilbert-Schmidt opera-
tors in H, and B2 (1i) is a Hilbert space with this scalar product (and clearly the norm
this space is the Hilbert-Schmidt norm I · I Hs). It should be noted that the
vectors (i.e. the ele1nents) of this Hilbert space B 2 (1i) are linear operators acting in
another Hilbert space, namely in 1i. As a Hilbert space B 2 ( 1i) is isomorphic to the
tensor product 1i ®H. The fact that B2 is complete with respect to the norm II · IIHs
IIHs -t 0 ass -t

f }
-- Qlj·

2
one says that A is an
48 LINEAR OPERATORS

B is defined as follows: D(A +B) == D(A) n D(B) and (A+ B)f == Af + Bf for
f E D(A +B). For the product AB one has: D(AB) == {f E D(B) I B f E D(A)}
and (AB)f == A(Bf) for f E D(AB). D(A +B) and D(AB) are linear manifolds,
but it may happen that they consist of only the zero vector, even if D(A) and D(B)
are dense in H (see e.g. Problem 2.26). If A E B(H), then clearly D(A +B) == D(B)
and V(AB) == D(B).
In quantum mechanics one often uses commutation relations. When unbounded op-
erators are involved some caution is required. We recall that if A,B E B(H), then their
commutator [A,B] is defined as [A,B] == AB- EA. If at least one of the operators
A and B is unbounded, the products in the preceding definition are not defined on all
vectors (possibly only on the zero vector 0). The commutator is a priori defined on the
domain D([A,B]) == {f E D(A) nD(B) l B f E D(A) and Af E D(B) }. Nevertheless,
in certain situations it is possible by different rnethods to give a meaning to [A,B] on
a larger domain. For example, if A and B are self-adjoint, the number (f, [A, B]g)
makes sense for f,gED(A) nD(B) ifitisinterpretedas (Aj,Bg)- (Bf,Ag) (the
self-adjointness of unbounded operators will be discussed in §2.4.4 ).
If the operator B belongs to B(H), then D([A,B]) == {f E D(A) I Bf E D(A)}. In
this case A and B com1nute if AB f == BAf for each f in this domain; in operator lan-
guage the commutativity of A and B would be written as [A, B] C 0. If D(A) f H,
then the zero operator 0 is a strict extension of [A,B] because D( 0) == H whereas
D([A,B]) f- H.
An operator A is bounded if there exists a real number M < oo such that I Af I <
1\!IIIfll for all f E D(A). In this case one can define the (finite) norm of A in analogy
with (2.1):

A bounded operator with domain D(A) has a natural bounded extension to the sub-
space D(A) spanned by D(A). This extension is denoted A and called the closure of
A. We explain this in the particular case in which the domain D(A) is dense in H
(hence D(A) ==H); the operator A then belongs to B(H).
So let us assume that A is bounded and that D(A) is dense in H. Given a vector
f E H, choose a sequence {fn} in D(A) such that f == s-limn-7CXJ fn· Since the op-
erator A is bounded, the sequence {Afn} is also strongly Cauchy: IIAJm - Afn) II <
IIAIIv(A) llfm- fnll ---t 0 as m, n ---t oo. Let h == s-limn-7(X)Afn and set Af == h.
Notice that A is well defined, i.e. the vector h is independent of the sequence {fn}
converging to f: if {gn} is another sequence in D(A) satisfying s-lim n-7CXJ 9n == f,
then the sequence {Ag 11 } also converges to h:

because llfn- 9nll < llfn- fll + llf- 9nll- So A is well defined on each vector
f E H, and it is an extension of A: iff E D(A), one may take fn == f for each n,
UNBOUNDED OPERATORS 49

so that Af == Af. Finally let us show that the norm of A [as an operator in B(H)] is
identical with the norm of A [as an operator defined on V(A)], i.e. that

- - IIAJII - IIAg\l
IIAII = 0=/::fEH
sup IIIII = IIAIIvcAl = sup
0=/::gED(A)
II g II ·

For this we observe that, iff and {fn} are as above, one has (apply twice Proposition
1.1)
\IAJ\1 == n-+oo lim llfnll == \\A\\vcA)I\f\1,
lim I\Afn\1 < \IAI\vcA) n-+oo

hence \\A II < \\A \\v(A)· The opposite inequality is evident.


One can try to apply the preceding construction also to an unbounded operator A
in order to obtain a natural extension of A. Assume again that V(A) is dense in H.
Consider a vector fin Hand a sequence {fn} of vectors in V(A) converging strongly
to f. Here the sequence { Afn} will not be Cauchy in general. But assume that {Afn}
is (strongly) Cauchy and denote its limit again by h. Then it is natural to set Af == h
as before. However, in order for this to make sense, one has to know that each se-
quence {fn} with the indicated properties leads to the same vector h, in other words:
if {gn} is any other sequence in V(A) converging strongly to f and such that { Agn}
is Cauchy, then the limit of { Agn} must be the same as that of {Afn} (i.e. the vector
h). If this condition is satisfied, then Af is well defined.
These considerations lead to the introduction of two important classes of operators,
each of them containing B(H), namely the closable operators and the closed opera-
tors. We now describe these operators.
An operator A is called closable if the condition on Cauchy sequences mentioned
above is always satisfied: whenever {fn} and {gn} are Cauchy sequences in V(A)
converging to the same limit and the sequences {Afn} and {Agn} are also Cauchy,
then s-limn-+ooAfn == s-limn-+ooA9n· By observing that the sequence {fn- 9n}
converges strongly to the zero vector 0, this condition then means that the limit of the
sequence {Afn- Agn}- which is assumed to be Cauchy- must also be the vector
0. Thus the definition of a closable operator can be written as follows:
An operator A is closable if the three properties
(i) fn E V(A), (ii) s-limn-+oofn == 0, and (iii) {Afn} is strongly Cauchy
together imply that s-limn-+oo Afn == 0.
If A is closable then, as illustrated above, one can associate with it a natural exten-
sion A, called its closure: the elements of V( A) are the vectors f in H for which there
exists a sequence {fn} in V(A) which converges strongly to f and which is such that
the sequence {Afn} is also strongly Cauchy, and then Af == s-limn-+ooAfn· To see
that A is an extension of A, it suffices to take fn == f for each rt iff E D(A).
A special class of closable operators are the closed operators, defined as follows:
An operator A is closed if the three conditions
(i) fn E D(A), (ii) s-limn-+oofn == f, and (iii) {Afn} is strongly Cauchy
together imply that f E D(A) and s-limn-+oo Afn == Af.
s-limn-+oo fn - 0
sequence

- S lirn n--+oo Afn -

an {f,Af}
r (strong) Cauchy sequence in r (A) a
1
A~~un1e that A is closable, that fn E that s -lin1 n ______, oo f n = f and that { Afn} is Cauchy,
say s -lirn ex h. To see that the i~ we 1nu~t show that f E V(A_) and that
= Ft. Now, by the definition of the closure of there exists for each n a sequence {g i
71
) in V( A)
such that llfn -gi,n) II--+ 0 and 1\Agkn)- II____, 0 ask--+ oo. For each n, choose an integer k(n) such
II - I < \\Ag~(~)- II < 1/n. Then {gi(~)}nEN is a sequence in V(A) such
7
that gt('} )
7
l/n and
.f and
' ~uch
~ that {A gk(n)
(n) } -. . 1·IIDn----+oo A gk(n)
(n) -- l1. Agam
·
11
that t:, linln ( )
gk(n) -
- ·'
IS '
Cauchy. namely t:;-
the definition of the closure of we have f E V(A) and Af = s-lirrln ---+oo Agk(~) =h.
4
Ob~erve the analogy with the graph of a function r.p: lR --+ lR in the plane JR 2 , given by the set of points
.cp(x)}.
UNBOUNDED OPERATORS 51

sequence {fn,Afn} with fn E 'D(A) such that both {fn} and {Afn} are Cauchy (in
H). This shows that
(i) an operator is closed if and only if its graph r(A) is closed in 1-i 1-i (i.e. if and
only if f(A) a subspace of 1-i EB 1-i);
(ii) A is closable, then the graph r (A) of its closure coincides with the closure in
H of graph of
(iii) an is closable if and only if the closure of its graph contains no element
of the h} with h =f- 0.
use properties (i) and (ii) in the proof of Proposition 2.22.
expression given above for the norm of an element {f, Af} ofr(A) [remember
that f E V( A)] suggests that one define a scalar product on by

(j, g) A == (f, g) (Af, if j, g E 'D(A), (2.58)

where ( ·, ·) is the given scalar product in H. A closed, the linear manifold V


equipped the product (2.58), is a Hilbert space, and norm associated
with this scalar product, called the graph norm of A, is given by
112
llfl\A == [llf\1 2 I\Af\1 2 ] . (2.59)

2. 4. 3 . We now turn to another very important question, of adjoint


of an unbounded operator. We recall that, for an E B(H), the adjoint A*
is another operator in B(H) defined by the condition that (A *j, g) == (f,
.f, g E 1-i. A is not bounded, it is clearly necessary to restrict g to
this restriction one can no longer use the Riesz Lemma to
it seems that will not be defined for vectors f to specify
not only the mapping · V(A *). The
obtained by applying
of of the domain of
on the vectors in this don1ain:

be an operator in H don1ain
is the following
domain: a vector f E H belongs to V(A *) ~{there exists a vector f*

(f*, g) == (f, \::lgEV

ntapping: f E V(A *), one sets == f *.


COMMENTS. (i) condition V(A) be to ensure
defined, that the f* it
fi and f~ are satisfying (fi, g) == (f,
f i - f ~ is orthogonal to , hence equal to
(provided that V(A) is 1-i).
(ii) It clear f, h E V(A *) f *, h * are to
them (2.60), and a E CC, (f * , g) == (f nh, g E V(A).
52 LINEAR OPERATORS

So f + ah E V(A *), i.e. V( A*) is a linear manifold, and A* (f + ah) - (f + ah) * ==


f * + ah * == A* f + a A* h, i.e. the mapping A* is linear.
(iii) It may happen that V( A*) is rather small, even that it consists of only the zero
vector. One can show that, if A is closable (and densely defined), then V(A *)is dense
in H.
(iv) Finally we observe that, once A* has been defined, the relation (2.60) can be
written as
(A*j,g) == (j,Ag) VjEV(A*), VgEV(A). (2.61)
Proposition 2 . 19 . Let A be an operator in 1i with dense domain V(A). Then
(a) A* is closed.
(b)OnehasN(A*) ==R(A)j_.
(c) If B is an extension of A, then A* is an extension of B * : A C B ====? B * C A*.
PROOF. (a) Assume that fn E V(A *), s-limn---+oo fn == f and s-limn---+oo A *fn == h.
Then, for each g E V(A):

(j, Ag) == n---+


lim (fn, Ag) == lim (A *fn, g) == (h, g).
oo n---+ oo

Hence f E V(A *) and A *j == h, which proves that A* is closed.


(b) Let f E N(A *), i.e. f E V(A *) and A *j == 0. Then 0 == (A *j, g) == (j, A g)
for all g E V(A), hence f j_ R(A). For the opposite inclusion let f E R(A)j_. Then
(f, A g) == 0 == (0, g) for all g E V(A), hence f E V(A *) and A *j == 0.
(c) Let f E V(B*). Then (B*j, g) == (f, Bg) for all g E V(B). In particular, since
B is an extension of A, one has (B*j, g) == (j, A g) for all g E V(A). This means that
f E V(A *)and that A *j- B*j. Thus A* is defined on V(B*) and coincides with B*
on this domain, i.e. A* is an extension of B *. D

Proposition 2 . 20 . Let A be a closable operator with V( A) dense in 1i. Then one has
A*_ (A)*== A*, i.e. A and its closure A have the same adjoint.
PROOF. Since A C A, one has A* A* by Proposition 2.19(c). To verify the opposite
inclusion, let g E V(A) and choose a sequence {gn} in V(A) having the properties
s-limn---+oo gn == g and s-limn---+oo Agn == Ag. Now let f E V(A *).Then

(A *j, g) == n---+
lim (A *j, gn) == lim (j, Agn) == (j, Ag).
oo n---+ oo

Since this holds for each g E V(A), we see that f belongs to the domain of A* and
that A *j == A *j. Hence A* is an extension of A*. D
The following properties of adjoint operators are obtained as simple applications
of Definition 2.18; we leave the details as an exercise.
Proposition 2 . 21.. Let A and B be operators with dense domains in a Hilbert space
1i.
(a)IfR(B) c V(A), thenB*A* c (AB)*.
(b) If BE B(H) and a f= 0, then one has V( (aA +B)*) == V(A *)and (aA +B)* ==
aA* +B*.
UNBOUNDED OPERATORS 53

If an operator A is such that D(A) and D(A *) are dense in the Hilbert space, then
A* and its adjoint (A*)* _ A** are well defined. We know that A** == A if A belongs
to B(H). In the general case A** is an extension of A [if D(A) and D(A *) are dense].
We shall see that this extension is identical with the closure of A; hence, if A is a
closed operator, one again has A** == A.

Proposition 2. 22. Let A be an operator such that V(A) and V(A *) are dense in H.
Then A C A**. Furthermore A is closable and A == A**. In particular, if A is closed,
one has A**== A.

PROOF. (i) Fix a vector hE D(A). Then (A*j, h) == (f,Ah) for all f E D(A*). By
virtue of Definition 2.18 (applied to A* instead of A) this implies that h belongs to the
domain of the adjoint of A*, i.e. h E V (A**), and that A** h == A h. This shows that
A** is an extension of A.
(ii) Since A** is an adjoint, it is a closed operator (Proposition 2.19). So A has a
closed extension. By Proposition 2.17, A is closable.
(iii) To see that A== A** one shows that the graph f(A) of A is identical with the
graph r(A **) of A** (we know that each of these two graphs is a closed subset of
H EB H). Let us introduce the following operator U in H EB H: U{f, g} == {g,- f}. It
is evident that U is isometric: I U {f, g }II~EBH == 11911 2 llf\1 2 == I {f, g} II~EBH· Also
R(U) == H EB H. Consequently U is a unitary operator in H EB H.
If B is a linear operator in H with dense domain and if {g,Bg} E f(B), then
U {g,Bg} == { Bg, -g }. Thus, iff, j* E H:

(U{g,Bg},{j,j*})HEBH == (Bg,f)H- (g,j*)H. (2.62)

Comparison with (2.60) shows that f belongs to D(B*) if and only if there exists a
vector j* E H such that the expression (2.62) is zero for each element of f(B), and
then f * == B *j. Thus the graph r (B *) of B * is the orthogonal complement in H EB H
of the subspace Uf(B), i.e.

(2.63)

[take into account Proposition 2. 8(b) for the second equality].


By taking B == A* orB == A, one obtains from (2.63) that

f(A **) == [Uf(A *)]j_ and

Since U 2 == -I, with I the identity in H EB H, and since a subspace M is invariant


if each of its elements is multiplied by -1 (i.e. - M == M), one obtains from the
preceding two relations that
D

2.4.4. An operator A is self-adjoint if A* ==A, i.e. if D(A *) == D(A) and A *j == Af


for all f E D(A). One must assume that A has dense domain (so that A* exists). A self-
adjoint operator is always closed (see Proposition 2.19).
LINEAR OPERATORS

For E B(H) the condition of self-adjointness is that

(Af,g) == (j,Ag) \/j,gEH. (2.64)

is unbounded, this equation makes sense only if both f and g belong to V(A). An
operator satisfying (2.64) for all f and g in its domain is called syn1metric. So IS a
sy1nmetric operator if V(A) is dense in Hand

(Aj, g) == (j, A g) \/j,gEV(A). (2.65)

In particular, if A is sy1nn1etric, one has (j, A f) E JR. for each f E V(A).


In applications the validity of the condition (2.65) is often easy to verify, and some-
physicists are not aware of the distinction between a symmetric and a self-ad-
joint operator. Eq (2.65) means only that C A*. Indeed, if f E V(A) and (2.65) is
satisfied, then there exists a vector f * (namely f * == A f) such that (fj A g) - (f *,g)
all g in V( A), and since A *j == f * by definition, one then has A *j == Af. So
(2.65) in1plies that A* is an extension of A. However the condition that V(A *) be
exactly the sa1ne as V(A) is very strong and is in general not satisfied for a symmet-
operator. relation between symtnetric and self-adjoint operators is illustrated
below and studied detail in Chapter 3.
operator closable (because a closed extension of -'4, see
Proposition 17). closure is also sy1nmetric: If f, g E V(A), then there are
{fn} and {gn} in V(A) such that s-lilnn-fcx:; fn == f, s-lilnn-fCX) gn == g,
- Af and s-lilnn-fCX) Ag. Then [see Proposition l.l(b)]

general will not self-adjoint. closure A of a symmetric operator A is


self-adjoint (i.e. if one says that is essentially self-adjoint, and V( A) is
a core for this case the symmetric operator A has exactly one self-adjoint
closure A. (Indeed, let B == B* be a self-adjoint extension of A;
B is closed, it also an extension of A, see Proposition 2.17; thus A C B == B*.
in1plies that B* C *, and * by the assumption of essential self-
of A. one will have C B == B* C i.e. one must have ==B. 6 )
each self-adjoint operator is also essentially self-adjoint.
is a symn1etric operator that essentially self-adjoint, it has exactly one
closed extension, and this extension is self-adjoint. 7 In more general cases
5 More generally, if is a closed operator with dmnain D( C), a linear submanifold D of D( C) is
called a core for if the clo~ure of the restriction of C to D is equal to Thus, if Co is defined by
= D ~ D(C) and Cof = Cf for f E ), then D i~ a core for the operator if Co ==
D i~ chosen too sn1all, then one will only have ~ and D may not be a core for C. We refer to
3 for examples.
The san1e rea~oning allows one to show that the closure of an essentially self-adjoint operator A is
its closed ~yn1n1etric extension: one replaces the hypothesis that B == by the condition that B be
and closed, i.e. B = ~ B*.
7 Thi~ the terminology "essentially self-adjoint": except for passing to its closure, A is self-
adjoint.
a many closed
of is self-adjoint, are
3
of these subtleties we consider the following
B is an of then * is the
) [Proposition C
general). Thus
extension of
operators
B

the 1nore one


adjoint
meet exactly, i.e. possible or

then selj~adjoint.
(b) se~f-adjoint, *

· essentially self-adjoint and only


facts that sym1netric (Propositions
(a) Suppose that symmetric, that C *
seen before, is also C *
(b) Now assume is essentially self-adjoint, i.e. that **
* , it follows that :t- i'

[for1nally
Let us by operator (d / dx)
the operator -i( d/ dx) with the domain
on 1]. f and g are continuously differentiable functions on ' 1],
(-if', g) - (f, == i[f(1)g(1) (0) J.

Observe the right-hand side is zero when g belongs to the dense set ( , 1)).
shows that the operator is symmetric, and is contained domain
of the adjoint of , so that C . Obviously one can find functions
which the right-hand side of (2.66) does not so that hence also are
not symmetric. view of Proposition we conclude that but
essentially self-adjoint. Another way of arriving at this conclusion as follows:
8
In order to have f E there must exist a vector f"' in H such that (J. = (J )<.g) for all
g E D(B). Thus, when one increases the nu1nber of conditions in1posed on f increases, conse-
quently D(B-r) will shrink.
56 LINEAR OPERATORS

assumption that Po == Po implies, together with the relation Po c P 1 c P 0 obtained


above, that P0 == P 1 , hence that P 1 is also self-adjoint. This is not possible: the dif-
ferential equation -if' == if has the solution f(x) == e-x which is square-integrable
on the interval (0, 1) and hence belongs to the domain of P 1 . So the number +i is
an eigenvalue of the operator P 1 , and since a self-adjoint operator can have only real
eigenvalues, P 1 is not self-adjoint. A detailed description of the self-adjoint extensions
of the operator Po is given in §3 .2.3.
We finally mention a result that will be discussed and proved at the beginning of
the next chapter (Proposition 3.3). This result represents the basic self-adjointness
criterion for symmetric operators and is used already in the next section to establish
self-adjointness properties of multiplication operators. It involves knowledge of the
range of A i and of A- i:

Proposition 2.24 . Let A be a symmetric operator.


(a) A is self-adjoint if and only if R( A i) == H and R( A - i) == H.
(b) A is essentially self-adjoint if and only ifR(A i) and R(A- i) are dense in H.

Remark 2.25. If A is an operator in H with domain V(A) and U is a unitary oper-


ator in H, one can define an operator Au by V(Au) == UV(A) [the linear manifold
obtained by acting with U on all vectors of'D(A)] and Auf== UAU*f iff E V(Au ).
This makes sense because, iff E V(Au ), then there exists a (unique) vector g E V(A)
such that f == Ug (in fact g - U*f), hence Auf == UAU*(U g) == UAig == UAg,
and Ag is well defined since g E V( A). It is easy to check (exercise) that a unitary
transformation preserves most properties of an operator; for example, if A is closable
or closed or symmetric or self-adjoint or essentially self-adjoint, then the operator Au
also has the respective property. If A is closable, then the closure Au of Au is given
by Au == UAU* _ (A)u. The resolvent set, the spectrum (defined in Section 2.6), the
eigenvalues and the norm of an operator are invariant under a unitary transformation.

2.5 Multiplication operators

2 . 5 . 1. A multiplication operator A in H == L 2 ( 0, m) is an operator given as follows:

(Af)(x) == a(x)f(x), (2.67)


i.e. the value of the function Af at the point x E 0 is obtained by multiplying the
value off at the point x by the number a(x ), where a: 0 --+ Cis a fixed measurable
function. In other terms the function Af is the product of the two functions a and f.
The maximal domain on which this makes sense is clearly as follows:

D(A) = {f E L 2 (CJ, m) /foia(x)[ 2 [f(xWm(dx) < oo }. (2.68)

This domain is a linear manifold, and normally we consider a multiplication operator


to be defined on its maximal domain.
MULTIPLICATION OPERATORS 57

The properties of such an operator depend on the function a. To characterise the


multiplication operators belonging to B (H), it is useful to introduce the Banach space
LP(O,m) for p - oo. A measurable function f: 0--+ CC is said to be essentially
bounded if there exists a constant M < oo such that lf(x)l < 1\1 m-almost every-
where (in other terms if, for some M < oo, the set W == {x E 0 II f (x) I > M}
belongs to the considered a--algebra A and satisfies m(W) == 0). The infimum of all
numbers having this property is denoted by llflloo:

llflloo == inf {ME IR llf(x) I < M m,-a.e.}. (2.69)

L 00 ( 0, m) is the set of equivalence classes (see page 19) of essentially bounded func-
tions. A function f representing an element of L 00 ( 0, m) satisfies If ( x) I < II f II oo
m-a.e. 9 For this reason one also writes llflloo = esssupxEolf(x)l (essential supre-
mum, i.e. the supremum is taken modulo null sets). If for example 0 == (a, b) is
an interval, m is the Lebesgue measure and f is continuous on (a, b), then llflloo -
sup xE (a,b) If (x) I [this supremum will be infinite if f (x) diverges as x --+ a or x --+ b,
in which case f will not define an element of L 00 ( (a, b))].

Proposition 2.26 . Let A be a multiplication operator in 1-i == L 2 ( 0, m ). Then A


belongs to B(H) if and only if a E L 00 (0, m), and in this case IIAII == llalloo·
PROOF. (i) One has

IIAJII 2 = fo la(x)l lf(x)l m(dx) < llall~ Llf(xWm(dx)- llall~llfll ·


2 2 2

Thus, if llalloo < oo, one has D(A) == 1-i and IIAII < llalloo·
(ii) Now assume that a tJ_ L 00 (0, m). For n EN set Wn == {x E 0 jla(x)l > n}.
Then Wn belongs to A (because a is a measurable function) and m(Wn) > 0 because
a~ L 00 (0, m). Let Vn be a subset of Wn with Vn E A and 0 < m(Vn) < oo [if
m(Wn) < oo, one can take Vn == Wn]. Let fn == Xvn' the characteristic function of
Vn [see Eq. (1.41)]. One has llfnll 2 rn(Vn) < oo, hence fn E 1-i- L 2 (0, m), and

IIAfnll 2 = J1a(x)l 2 lxvJx)l 2 m(dx) =


0
Jla(x)l m(dx)
Vn
2
> n
2
m(Vn)- n
2
llfnll 2 .
This shows that IIAfnll/lfnll > n. Since n is arbitrary, A cannot be bounded.
(iii) It remains to show that II All > llalloo if a E L 00 (0, m). For this one proceeds
as in (ii). If c > 0 and Ws - {x E 0 jla(x)l > llalloo- c}, then Ws belongs to A
and m(Ws) > 0. Iff E £ 2 (0, m) is such that 11!11 =f 0 and f(x)- 0 for all x ~ Ws,
then IIAJII/11!1 > llalloo- c, which implies the desired inequality. D

Example 2 . 27 . Let A E B(H) be the operator of multiplication in 1-i == L 2 ( 0, m) by


a function a E L 00 ( 0, m). Observe that

(f,Ag) = fo f(x)a(x)g(x)m(dx) fo a(x)f(x)g(x)m(dx).


= (2.70)

9 It can be shown that the infimum in (2.69) is attained, see e.g. [HS].
58 LINEAR OPERATORS

shows that is the operator of multiplication by the complex conjugate function


a of a: (A ~f) (x) == a(x ).f (x ). In particular:
A is self-adjoint and only if a is a real function.
is a projection if and only if a(x) == 0 or a(x) == 1 m-a.e. If one defines V ==

jection onto the subspace M(A) = v


{x E 0 I a( x) == 1}, then a is the characteristic function of the set V and A is the pro-
E £ 2 (0, m) I f(x) = 0 x tJ. V} = L 2 (V, m)
fconsidered as a subspace of H == L (0, mj)].
(c) ~4 an iso1netry if and only if ia(x) I == 1 m-a.e., i.e. if a(x) == ezo:(x) 1n-a.e., with
a: 0 ---t JR. Then A* is the operator of multiplication by e-ia(x), and one sees that
in fact unitary: Each isonzetric multiplication operator is unitary.
(d) is a partial isometry if a(x) == eza(x)Xv(x), where a: 0 ---t JR. and Vis a sub-
set of 0 belonging to A. The initial set and the final set of A are identical, given by
(V, m) [considered as a subspace of L 2 (0, m)].

regards unbounded multiplication operators, we shall meet only a few simple


examples. All the same we mention in Remark 2.31 some general results that are
rather easy to obtain.
a is a real function, one sees from (2.70) that (f,Ag) - (Af, g) for all f and
g in the maximal domain D( A) given in (2.68). If this domain is dense in L 2 ( 0, m),
it follows that the operator A is symmetric. We first show by applying the self-
adjointness criterion stated in Proposition 2.24 that in fact self-adjoint.

Proposition 2.28.. Let be the multiplication operator in H == L 2 ( 0, 'm) by a real


function, defined on its maxilnal domain (2.68). If this domain is dense in H, then
is self-adjoint.

PROOF. We have seen above that A is symmetric. By virtue of Proposition 2.24(a)


it then suffices to show that R(A + i) == Hand R(A i) - H. So let g E H. We
must find a vector fin V(A) such that (A+ i)f - g. For this we define f: 0 ---t CC
by f(x) == [a(x) + i]- 1 g(x), where x E 0. Since ![a(x) + i]- 1 < 1, we have 1

llfll < 11.911 < oo, so that the function f defines an element of L 2 ( 0, m ). Furthermore,
since la(x)[a(x) + iJ- 1 < 1, we have
1

Thus f even belongs to D( A). Finally it is clear that (A+ i) f g. Hence R( A+ i) ==


1
H. Similarly, taking f(x) == [a(x)- i] g(x), one sees that R(A- i)-H. D

Proposition 2.29. (a) Let a: JRn ---t JR. be a continuous function. Then the multiplica-
tion operator by a in H == L 2 (JRn) is essentially self-adjoint on Cc)(JRn ).
(b) Let a be a real nleasurable function satisfying Ia (x) I < /1; ( 1 + Ixi) m for some con-
stants /1;, m E ( 0, oo). Then the operator o.f multiplication by a in L 2 (JR. n) is essentially
se{f-adjoint on C0 (JRn) and on S(JRn ).
MULTIPLICATION OPERATORS 59

PROOF. We consider the case n 1. The proof for n > 1 is essentially the same. It
==
is seen from the proof that the assumption of continuity of a made in part (a) could be
weakened to the condition that a be locally bounded (i.e. bounded on each finite ball
{xEIRn llxl < R}, 0 < R < oo).
(a) Let us denote by A the multiplication operator by a on the maximal domain
D(A) given in (2.68) and by A 0 the restriction of A to V(Ao) == C 0 (JR) [observe
that C0 (IR) is contained in D(A) because the function a is continuous, hence locally
bounded]. We shall show that R(Ao + i) is dense in H. An analogous argument gives
the denseness of R(Ao - i), and the result of (a) then follows from Proposition 2.24.
Each vector hE L 2 (JR) can be approximated (in L 2 norm) by a sequence {hJ} of
functions of compact support. By taking hj(x) == h(x) lxl < j and hj(x) == 0 if
lxl > j, one has
II h 2
hj 11 ==
lxi>J
1 2
lh(x) 1 dx,

which converges to zero as j --+ oo because his square-integrable. Thus it suffices to


show that each square-integrable function g of compact support can be approximated
in L 2 norm by a sequence of vectors {gk} of the form 9k (x) == [a(x) + i]fk(x ), with
fk E D(Ao) == Co(IR) [i.e. 9k E R(Ao + i)].
So let g E L 2 (JR) be such that g(x) == 0 for lxl > M (M > 0 is such that the
support of g is contained in the interval [-M, M]). We set f(x) == [a(x) + i]- 1 g(x).
The function f has compact support (identical with the support of g), and it is square-
integrable (see the argument given in the preceding proof). One can choose a sequence
{!k} in C 0 (JR) such that s-limk--7oo fk == f and such that fk(x) == 0 if lxl > M + 1
for each k == 1, 2, 3, ... (see §1.4.5). The continuous function a is bounded on the
finite interval [- M - 1, M + 1]. If we set C == maxlxl :s;M + 1 1a( x) + i I < oo, then

II (Ao + i)fk - 911


2
==.!lxl:s;M+l [a(x) + i][fk(x) -
I f(x )] dx
1
2

2 2 2
<C ./ l!k(x)- f(x)l dx == C ll!k- !11 2 ,
lxl:s;M+l
which converges to zero as k --+ oo.
(b) The function a considered in part (b) is locally bounded. By (a) and the obser-
vation made at the beginning of the proof, the operator of multiplication by a is essen-
tially self-adjoint on C 0 (JR). Since la(x) I < ~(1 + lxlrn, we have S(JR) c V(A), so
A is also essentially self-adjoint on the larger domain S (IR). D

Example 2 . 30. We illustrate the preceding propositions by some simple operators


from quantum mechanics.
(1) Let (a, b) be an interval and m E N. Let A be the multiplication operator by
a(x) == xm in H == L 2 ((a, b)). If the interval (a, b) is finite, then A E B(H) and
IIAII max{lalm, lblm}. If b- a == oo, then A is unbounded. For (a, b) == JR and
m - 1 one obtains the position operator Q of quantum mechanics in L 2 (JR):

(Qf)(x)- xf(x), D(Q) = {!: lR ~ C I j_:(l+x 2 )l.f(x)l 2 dx < oo}. (2.71)


60 LINEAR OPERATORS

The operator Q is self-adjoint on this domain, and it is essentially self-adjoint on


C0 (JR) and on S(JR). The operator conjugate to Q in L 2 (JR) is the momentum op-
erator P which satisfies [P, Q] C -if. P is a differential operator, given formally
by (P f) (x) == -if' (x) - -i df (x) / dx. More is said about this operator in the next
subsection and in §3.2.3 (page 102). In the next chapter we also study differential
operators on intervals (a, b) f=. JR.
(2) More generally, in dimension n > 1, let a(x) == Xj, where x == (xi, ... , xn)
and j E {1, ... , n }. The associated multiplication operator Qj represents the j-th
component of the n-dimensional position operator Q == ( Q I' ... ' Q n) in L 2 (JR n)'
i.e. (Q j f) (x) == Xj f (x). The operator Q j is self-adjoint on its maximal domain and
essentially self-adjoint on C0 (JRn) and on S(JRn). The operator conjugate to Qj is
the j-th component Pj of then-dimensional momentum operator P == (PI, ... , Pn),
given formally by (P1 f)(x) == -i8f(x)j8xj.
To end this subsection we state some further properties of multiplication operators.
Remark 2 . 31 . (1) If a E £ 2 (0, m), then V(A) contains all functions in £ 2 (0, m)
that also belong to L 00 ( 0, m ), i.e. V(A) :=> [L 2 ( 0, m) n L 00 ( 0, m )]. For example in
H == L 2 (JRn) one has S(JRn) C V(A) if a E L 2 (JRn), hence V(A) is dense in H.
(2) If A is the multiplication operator by a function a, let us denote by A the multipli-
cation operator by a (the complex conjugate of a). Clearly the maximal domain of A
is the same as that of A: V(A) == V(A). Iff and g belong to this domain, then by Eq.
(2.70): (J, Ag) == (Af, g). Thus, if V(A) is dense in H == L 2 ( 0, m ), then the adjoint
A* of A exists and is an extension of A : A C A*. In this situation the domain of A*
is dense in H, hence A** is well defined. Since A is closed by (3) below, we see from
Proposition 2.22 that A** == A.
(3) A multiplication operator is closed [on its maximal domain (2.68)]. This can be
seen by using the following fact:
Iffn, f E LP( 0, m) and llfn - fliP ~ 0 as n ~ oo, then there exists a subsequence
{fnk} of {fn} such that f(x) == limk-7oo fnk(x) m-almosteverywhere.
So let us assume that {fn} is a sequence in V(A) such that llfn -!II ~ 0 (we take
p = 2) and IIAJn - gjj ~ 0 as n ~ oo for certain vectors j, g E L 2 ( 0, m ). Let {fnk}
be a subsequence of {fn} converging m-almost everywhere to f. Then IIAfnk -gil
converges to 0 as k ~ oo, hence there exists a subsequence of{!nk }, say {!nk 1 } ,
such that a(x)fnk J (x) ~ g(x) m-a.e. as j ~ oo. Since fnk J ~ f(x) m-a.e., one
then has a(x)f(x) == g(x) m-a.e. Since g E £ 2 (0, m), this shows that f E V(A) and
that Af ==g. Thus A is closed.

2. 5.2. It is well known that a differential operator with constant coefficients be-
comes a multiplication operator under Fourier transformation. For example the mo-
mentum operator P can be diagonalised by a Fourier transformation. We recall that
the Fourier transform F f f of a function f : JR ~ C is defined as
(Ff)(k) /(k) = ~ Joo e-ikxf(x)dx. (2.72)
v 21f -00
MULTIPLICATION OPERATORS 61

The integral in (2.72) is an (improper) Riemann integral if f is continuous and a


Lebesgue integral in general, and k varies on JR. For fixed k, the integral in (2.72) has
a sense e-ikxf(x) is an integrable function of x, i.e. iff belongs to L 1 (JR). 10 Under
this condition one has
~
lf(k)l <
1
~
joo lf(x)ldx ==
1
~llfll1 \fk E JR, (2.73)
v 21f -()() v 21f

henc~f E L 00 (JR), and by applying the Dominated Convergence The<?_rem one finds
that f is a continuous function. If f belongs to S (JR), then so does f, and F is an
isometric mapping on S (JR), i.e.

(!,g) i: f(x)g(x)dx = i: ](k)j](k)dk =(},g) if j,gES(JR). (2.74)

If we set f == g E S(JR) we obtain I !II == I ]II· This shows that F is bounded on S(JR),
consequently its closure belongs to B(H) and is an operator of norm 1. It is usual to
denote this closure also by F (rather than by F) and to denote by the vector Ff for J
each f E L 2 (JR). Iff E L 2 (JR) also belongs to L 1 (JR), then ](k) can be defined by
J
the formula (2.72) for each fixed k, and one has E L 00 (JR) n L 2 (JR). Iff E L 2 (JR)
but f tJ_ £ 1 (JR) 11 , one must apply the definition of the closure of an operator, in other
terms one must approximate f by a sequence {fj} with fj E £ 2 (JR) n £ 1 (JR), and one
J
then obtains its Fourier transform as the strong limit of the sequence {!1 }. Usually
one takes fj as follows: fj(x) == f(x) for lxl < j and fj(x) == 0 for lx! > j. Then,
by (1.49), fj belongs to L 1 (JR), hence jj(k) is given by the formula (2.72), and is J
the strong limit of the sequence { Jj}. One then writes

1 1
](k) == J27f l_.i.m.j e-ikxf(x) dx (2.75)
21f J-+00
-J
.

[l.i.m. stands for "limit in the mean", i.e. the limit is not defined point-wise- for each
k - but in the L 2 sense, its existence is guaranteed by the fact that L 2 (JR) is complete].
Equation (2. 7 4 ), which is obtained first on S (JR), holds for all f, g E 1-i (by a lim-
iting argument):

(f,g) == (Ff,Fg) == (f,F*Fg) (2.76)

So F*F == I, i.e. F is an isometric operator in the sense of Section 2.2. In fact F


is unitary, since F* is also isometric; F* has properties analogous to those ofF be-
cause, forgE S(JR), F*g is given by a formula of the same type as (2.72):

(F*g)(z) == - -
1
J21f -()()
joo e+izxg(x)dx !]( -z). (2.77)

1°F is an integral operator as defined in §2.1.8, all representatives of an equivalence class in £ 1


(IR) get
mapped onto the same function (Ff)(k).
11 There is no inclusion relation between L 2 (IR) and L 1 (IR), simply L 2 (IR) n L 1 (IR) is dense in L 2 (IR).

If J =(a, b) is a finite interval, then one has L 2 (J) C L 1 (J) because fablf(x)l dx
1 2
=
fabllf(x)l dx :S
llliiL2((a,b)) llfiiL2((a,b)) = (b a) 1 llfiiL2((a,b)) by the Schwarz inequality [see also (1.49)].
62 LINEAR OPERATORS

Thus F* - :r:- 1 [see Proposition 2.7(e)], i.e. if ](k) is given in terms of f(x) by
(2.72), then one can recover f(x) by the formula

f(x)- _1_ Joo e+ikxj(k)dk (2.78)


J21f -oo

[provided that f and j belong to £ 2 (JR.) n L 1 (JR.)].


Iff E S(JR.) and m, EN, then the Fourier transform of ( -id/ dx )m f(x) is km j (k ),
which is easily checked by integrating m times by parts. For example

[F( -if')](k) == ~ Joo e-ikxj'(x)dx = _t_· Joo f(x)_!!:_e-ikxdx


J21f - oo J21f - oo dx

= ~ Joo f(x)e-ikxdx = k](k).


v 21f -00

This may be written as FPF* == Q on the domain S(JR.), i.e. after Fourier transfor-
mation the differential operator P becomes the multiplication operator by the variable
in L 2 (JR.). More generally one has F pm F * == Qm on S (JR.).
It is clear that ( -id/ dx )m is symmetric on S(JR.):

oo_ d rn Joo
J
-oof(x)(- i d) g(x)dx = -oo [( i dx }:J (X)] g (X) dx if f, g E S (JR.) .

Since the operator Qm is essentially self-adjoint on S (JR.) and since F maps S (JR.) onto
S(JR.), one sees that the differential operator ( -id/ dx )m is essentially self-adjoint on
S(JR.) (apply Remark 2.25 with U - F).
Similar results hold in n dimensions (n > 2). The n-dimensional Fourier trans-
fonnation and its inverse are given by expressions that are analogous to (2.72) and
(2.77):
(2.79)

(2.80)

where for example x- x


(x 1 , ... ,xn) E JR.n and k · == ~7= 1 kjXj. After Fourier
transformation, the differential operator P1 - -i8 / 8x j becomes multiplication by
k.7: [:F( -i8j8x1 f)](k) == kj](k) iff belongs to the Schwartz space S(JR.n) defined
in ( .55). So, by Proposition 2.29, ?_7 is essentially self-adjoint on S(JR.n ), and we
denote its closure also by Pj. Similar statements hold for other differential operators,
in particular for the operator - ~ which is discussed in §2.5 .3.

Remark 2. 32. From the mathematical point of view the set of the (equivalence classes
of) Fourier transforms j of the elements f of L 2 (JR.) is naturally identified with L 2 (JR.)
[the Fourier transformation F is interpreted as a unitary operator in L 2 (JR.)]. From the
point of view of quantum mechanics, however, it is useful to distinguish between the
MULTIPLICATION OPERATORS 63

two representations of this Hilbert space, because the variable x in f (x) is interpreted
as and variable k j (k) as 1non1entun1 of a physical system (see Ex-
ample We refer configuration space when variable of the functions in
L (~) is interpreted as position, and to momentum space when this variable is mo-
2

mentum. latter case (in the momentum representation of Hilbert space), the
momentu1n P diagonalised, i.e. the vectors are identified with the Fourier
transforms j(k ). We sometimes write L2 (~) to specify that we consider the momen-
tum of the Hilbert space L 2 (~) (and we adopt analogous conventions
for n-dimensional systems). Note that, if one makes the distinction between the two
mentioned representations of the space L 2 (~), the Fourier transformation should
be interpreted as a mapping between these two representations: F maps wave func-
tions (viewed as defined on configuration space) to elements of L2 (~), and F * y-l
maps back momentum representation to the position representation.

2.5 . 3. We consider here differential operator -.6. - L7=t8 2 /8xJ L 2 (~n).


The free Hamiltonian H 0 of a non-relativistic quantum-mechanical particle in ~n
is formally given by this differential operator: 12 one has (Hof)(:l) == -Llf(x) if
f E L 2 (~n) is twice differentiable and Llf is square-integrable, particular iff be-
longs to Schwartz spaceS(~ n). momentum space (after Fourier transforn1ation)
this operator becomes multiplication by k2 == L~=l k}, i.e. [F(- Llf)] (k) == k2 j (k).
Thus H 0 is essentially self-adjoint on S(~n ), and in sequel we use the notation
H 0 for unique self-adjoint extension (the closure) of the operator -.6. defined on
S(~n); so H 0 -closure of Llls(JRn). fact, as shown the Appendix to the present
chapter (see page 81 ), differential operator - Ll is even essentially self-adjoint on
Co(~n).
The domain H 0 is easy to specify after Fourier transformation of wave func-
tions. Since, in representation L (~ n) of H (in momentum space),
2

-.6. becomes multiplication by square of the variable, its domain is just maxi-
mal domain of this multiplication operator [see (2. 68)]:

D(Ho)- {f E L 2 (~n) I k2 .f(k) E i 2 (~n)}


{jEL 2 (JRn) I
JrJRrl iP.f(kWdnk < oo}.
= (2.81)

-+2 -
We see H0 - P == L.7= 1 P12 .
n

Let us consider operator of multiplication in £ 2 (~n) by the function k ~


(k2 - z) - 1 , where z is a fixed complex number. It is natural to denote this opera-
tor by (Ho- z)- 1 . If z tJ_ [0, oo), preceding function belongs L 00 (~n);
(Ho- z)- 1 E B(H) its norm is given as follows (see Proposition 2.26):

II(Ho- z) 1
11- ll(k 2 z)- 1 lloo ==sup lA- zl- 1 == [dist(z, [0, oo))] 1
. (2.82)
-\~0

12
We systematically set li = 1 for Planck's constant and m = 1/2 for the mass.
64 LINEAR OPERATORS

Since function k f----CT k 2 ( k 2 - z) - 1 is bounded if z tf_ [0, oo), each vector in the
range of (H0 - z)- 1 belongs to D(H0 ) and, iff E D(H0 ), then f E R((Ho- z)- 1 )
[one has f - (H0 - z )- 1g for g - (Ho- z )f]. The operator (H0 - z )- 1 is called the
resolvent of H 0 , and we refer to Section 2.6 for a more general theory of resolvents.
Section 2. 7 it will be useful to know properties in configuration space of the
functions f belonging to the domain of H 0 . In (2.81) this domain is given in the rep-
resentation L 2 (JRn), and to find properties of f (x) one must make an inverse Fourier
transformation. Since D(Ho) == R((H0 - z)- 1 ), one must study properties of the
inverse Fourier transform of functions of the form 17(k) == (k 2 - z)- 1 g(k), with g
square-integrable [i.e. g E L2 (JRn)]. It turns out that the case of dimension n - 1, 2
and 3 is more elementary than the case n > 4. For n == 1, a detailed description of
V(H0 ) in configuration space is obtained in §3.2.4.
We first observe that, if z tf_ [0, oo ), then the function ez(k) :== (k 2 z)- 1 belongs
to LP(JRn) for each p > n/2 but not for p < n/2; indeed ez is locally bounded, hence
belongs locally to L P for each p E [1, oo], and it suffices to consider its behaviour at
infinity. By choosing spherical polar co-ordinates one sees that ez belongs to LP(JRn)
if and only if p satisfies
r
Jl
CX) kn-1
k2p dk < oo,
i.e. if and only if 2p- (n- 1) > 1 or p > n/2. In particular Bz is square-integrable if
n == 1, 2 or 3, but if n > 4 one has ez E LP(JRn) only for p > n/2.
So, if n- 1, 2 or 3, the function 17(k) == (k 2 - z)- 1 g(k) belongs to L 1 (JRn), and
one has

I[(Ho ~ z)-lg](X)I- (27r)-n/21ln eik-x(p z)-lg(k)dnk I

< (21r)-n/ 2 r i(P


}ffi.n
z)- 1 jj(k)ldnk < IIBzii2II?JII2,

. . . . . . . . . . . . ,., . . . . (H 0 - z) - 1 g is a bounded function of x E JRn (by a similar argument, using


Proposition 1.9, one finds that this function is uniformly continuous). If n > 4, one
can apply the Holder inequality (2.121) to obtain that, as a function of k, 17 belongs to
Lr(JRn) for r satisfying 1/r- 1/p 1/2 with p > n/2 [hence for r > 2nj(n 4)].
one can conclude by the Hausdorff-Young inequality (2.122) that, as a function
of x,
(Ho- z)- 1 g belongs to Lr' (JRn) with 1/r' == 1- 1/r == 1/2- 1/p (still with
p > n/2). Summing up one has: 13
g E D(Ho) and n- 1, 2 or 3 ===.::} g E LCX)(JRn)
(2.83)
gED(H0 )andn>4 ===.::} gEL 8 (JRn) \fsE[2,2n/(n-4)).

2 .. 5.4. Let us consider the behaviour of multiplication operators under Fourier trans-
formation. Let A be the operator of multiplication in L 2 (JR) by a function a. If a is a
13
The restrictions 2:: 2 on the second line is due to the fact that, in the argument involving the Hausdorff-
Young inequality, one must require that r E [1, 2] (see the Appendix to Chapter 2), hence r' s 2:: 2. =
MULTIPLICATION OPERATORS 65

polynomial, then :FA:F* is a differential operator. Indeed the identity :F pm:F* Qm


implies that F*Qm:F == pm, and, since the passage from F to :F* simply amounts
to changing ito -i, one obtains FQm:F* == ( -P)m == ( +id/dx)m on S(JR). If a is
a bounded function, then formally :FAF* is the operator of convolution by a, more
precisely
(:FAf)(k) ~ =
v 21f -00
joo
ii(k- k')/(k')dk', (2.84)

where a is Fourier transform of the function a. This makes sense for example if
one assumes that a E L 2 (JR). Indeed one then has

(:FAf)(k) = ~joo e-ikxa(x)f(x)dx = ~(eik·a(·),f)


v 21f -00 v 21f
=v~(a.k,f),
21f
(2.85)
2
i.e. the integral can be interpreted as scalar product in L (JR) of the two functions
ak(x) eikxa(x) and f(x). Now if a denotes the Fourier transform of a, then the
value of the Fourier transform ak of ak at the point k' is as follows:

ak(k') == _1_
v'2K
joo e-"k'x [eikxa(x)] dx
-oo

= =1- joo e-i(k-k')xa(x)dx a(k- k').


v'2K -oo

Consequently, by virtue of (2.76), Eq. (2.85) may be written as

(:FAf)(k) = ~(a.k,J) = ~ joo ii(k- k')](k')dk',


V 21f V 21f -oo

so that one obtains (2.84) in this case. In particular (2.84) shows that F AF* is an
integral operator in L2 (JR).
Since the resolvent (Ho z)- 1 E C\[0, oo)] is multiplication by (k 2 - z)- 1 in
momentum space, it will become an integral operator in the position variables, namely
[(H0 -z)- 1 f](x)- Jffi.nG~0 )(x-iJ)f(iJ)dny. Its kernel G~o) is called free Green's
function.

2.5 . 5 . To end this section we present some compactness criteria for operators in the
space £ 2 (JRn ), n EN. These results will be important in later chapters. If cp: -+ C
is a measurable function, we denote by cp( Q) the multiplication operator in £ (JRn) 2

by cp( x), and we use notation cp(P) for the multiplication operator by cp( k) acting
on the Fourier transforms of the functions in £ 2 (JRn), i.e. multiplication operator
by cp( k) L (JRn) ~ each of these operators is considered on its maximal domain as
2

specified in (2.68). Thus


[cp( Q) JJ (x) cp( x) J (x) , [F cp( P) JJ (f) == cp( f) J(f). (2.86)
The symbols Q and J3 stand for the n-dimensional position and momentum operator
respectively, as defined in Example 2.30, with (Q j f) (x) - X J f (x) and (F PJ f) (k) ==
kj ](k).
LINEAR OPERATORS

notations cp( Q) cp(P) are in agreement with functional calculus for the
self-adjoint operators Qand P that is developed Chapter 4. What is in1portant in
context is the fact that one often encounters operators of form cp( Q)1jJ(P),
it is useful to know that such a product is a co1npact operator under suitable
on functions cp et 'ljJ. Before proving a result one must be able to
a meaning to the product cp( Q)1jJ(P) as an operator B(H). This is possible for
the functions cp et 1/J belong to same space (JR.11 ) for so1ne p > 2.
proof is sin1ple if p == 2 or p == oo; these two cases suffice for most applications in
one considers physical systems dimension n == 1 2 j

or order to treat problems configuration spaces of arbitrary it is


to know the of cp( Q)1jJ(P) also for values of p (2, oo ). We
result following proposition indicate the proof p 1- 2, oo in the
Appendix this chapter.

- L 2 (JR.n), p E [2, oo] and let cp,'ljJ: JR. 11 ~ CC be two June-


to P(JR. 77 ). Thentheoperatorscp(Q)tjJ(P) and1jJ(P)cp(Q) are densely
to B(H),
II ~) II < (21r) 1 II cp II p 111/J II p ~
n P 111/J cp (Q) II < (21r) - n 1P II cp II p 111/J II p . c2. 87)

case cp( Q)tjJ proof 1/J cp( Q)is analogous (or one
[~J(P)cp(Q)]* == cp(Q)tjJ(P), for example denotes
of VJ ).
both cp( Q) 'ljJ are 1nultiplication operators by essentially
belong to B(H), llcp(Q)II == llcplloo 111/J(P)II ==
II lloo (see Proposition 2.26).
do1nain of 1/J(P) the Schwartz space S(JR. 11 ) (see
1). f belongs to S(JR.n ), there a finite constant c such
f] (x) I < c for all E JR.n [see (2.88) below], hence one can again
(1) of 1 to conclude tjJ(P)f E V(cp(Q)). Consequently
on S (JR.n). To prove (2. 87) for p == 2 it suffices to show
f II < (27T) --n 1211 cpjj2II1/J ll2ll f II f E S(JR.n ). we apply the
as follows:

ll ...... _. .. J,Jl"'"""-" inequality easily obtained fro1n (2.88):

case p E , oo) is treated the Appendix to this chapter. D


RESOLVENT AND SPECTRUM OF AN OPERATOR 67

Proposition 2 . 34 . Let 1-{ == £ 2 (JRn) and cp, 1/J: JRn -+ C measurable functions.
(a) cp(Q)1j;(P) and 1/;(P)cp(Q) are Hilbert-Schmidt operators if and only ifbothfunc-
tions cp and 1/J belong to L 2 (JRn) (or if one of these functions is zero alnlost everywhere),
and
llcp(Q)w(P)IIHs == llw(P)cp(Q)\IHs == (2n)-n 12 \lcplbllwlb· (2.89)
(b) If cp and <jJ belong to LP(JRn) for some p E [2, oo ), then cp( Q)1j;(P) and 1/;(P)cp( Q)
are compact operators. 14

PROOF. (a) Assume that cp,1/; E L 2 (1Rn). Then, by the results of §2.5.4, cp(Q)w(P) is
an integral operator in i 2 (1Rn) with kernel a(k, k') == (2n)-n1 2 ip(k- k')1j;(k'):

The Hilbert-Schmidt norm of this operator is easily calculated by using Proposition


2.15 (settingp== k- k'):

which gives (2.89) for cp( Q)1j;(P). The operator 1/;(P)cp(Q) is treated by considering
its adjoint (see the beginning of the preceding proof). The "only if" part of the proof
of (a) is given in the Appendix to this chapter (page 81).
(b) We denote by XR the characteristic function of the ball { x E JRn j\xl < R}, i.e.
XR(x) == 1 if \i\ <Rand XR(x) == 0 otherwise. By (1.49) the functions XRCf and
XR</J belong to L 2 (1Rn). Hence, if we set CR == XR(Q)cp(Q) and DR== XR(P)1jJ(P),
then CRDR E B2 (H) JC(H) by the result of (a). So, to conclude that the operator
cp( Q)1j;(P) is compact, it suffices to show that \\cp( Q)1j;(P) -CRDR\1 -+ 0 as R-+ oo
and to apply Proposition 2.11(d). Now (2.87) implies that

l\cp(Q)w(P)- cRDR\1 < 1\[cp(Q)- cR]?/J(P)\1 I\CR[1/J(P)- nRJII


< 1\[1- XR(-)]cp(·)llpii?/J\\p + \\cpl\pll[1- XR(·)]rl)J(·)IIp,
which converges to zero as R -+ oo (for example for the first term one uses Proposi-
tion 1.9 with h(x) == \cp(i) \P). D

2.6 Resolvent and spectrum of an operator

Let A be a closed operator. The resolvent set of A is defined as the set of all complex
numbers z for which the operator A- z A- zi is invertible with (A- z )- 1 E B(H).
14
The value of p must be the same for <p and 1/J, and p = (X) is excluded.
68 LINEAR OPERATORS

This set is denoted by p(A). Thus a complex number z belongs to p(A) if and only if
(i)N(A- z) == {0}, (ii) R(A- z) ==Hand (iii) (A- z)- 1 is bounded. 15 It is clear
that, if z E p(A), then (A- z)- 1 maps H onto D(A) and gives a bijection between H
and V(A). If z E p(A) one has

A(A- z)- 1 ==(A- z)(A- z)- 1 z(A- z)- 1 ==I+ z(A- z)- 1 (2.90)

and, for f E D(A):

(A- z)- 1Af ==(A- z)- 1(A- z)f z(A- z)- 1! == f z(A- z)- 1j, (2.91)

particular
iff E D(A). (2.92)

z E p(A) the operator (A- z)- 1 is called the resolvent of A at the point z. The
term "resolvent" is also used for the collection of operators { (A z) - 1 j z E p( A)},
i.e. for the function z f-t (A- z)- 1 from p(A) to B(H).

Proposition 2.35 . Let A be a closed operator and p( A) its resolvent set.


(a) One has the first resolvent equation: if z 1 , z 2 E p(A), then

(2.93)

(b) If' z1, z2 E p(A), then (A- z1)- 1 and (A- z2)- 1 commute.
(c) The resolvent set p( A) is an open subset of CC (which may be empty).
(d) each connected component of p(A) the function z f-t (A- z )- 1 is holomorphic,
i.e. at each point z E p(A) this function has a Taylor expansion converging in the norm
o.f B(H).

PROOF. (a,b) If z 1 ,z2 E p(A), then by (2.90):

z1(A-z1)- 1 - - I A(A-z1)- 1 and z2(A-z2)- 1 ==-I A(A-z2)- 1,

(z1- z2)(A- z1)- 1(A- z2)- 1 == [-(A- z2)- 1 + A(A- z1)- 1(A- z2)- 1]
- [-(A- z1)- 1 (A- z1)- 1A(A- z2)- 1]
(A- z2) 1 (A- z1)- 1 [A(A- z1)- 1 - (A z1)- 1A] (A- z2)- 1,

and the last term on bottom line is zero by (2.92). This proves (2.93). The result of
(b) is a simple consequence of (2. 93 ).
(c) Let z E p(A) and let ( E CC be such that lz- (I < II(A- z)- 1 11- 1 . Then, by
15
One can show that (iii) is a consequence of (i) and (ii) if A is closed. In fact, by Problem 2.19(a),
(i) and (ii) imply that (A - z) 1 is closed and has domain H, and it suffices to apply the Closed Graph
Theoren1 (we on1it its proof; the reader may consult a book, e.g. §51 of [AG] or §III.5.4 of [K]):
Closed Graph Theorem: ff B is a closed operator with D(B) == H, then B is bounded, i.e. BE B(H).
RESOLVENT AND SPECTRUM OF AN OPERATOR 69

Proposition 2.5, operator I+ (z- ()(A- z)-- 1 has an inverse belonging to B(H),
given by
00

[I (z- ()(A- z)-Irl = 2)z- ()k[(A z)-l]k. (2.94)


k=O

We claim that (belongs to p(A). To see this we observe that

(A- ()f == [I+ (z- ()(A- z)- 1 ] (A- z)f iff E V(A). (2.95)

Since each of the two operators on the right-hand side [i.e. I (z - () (A - z) 1 and
A- z] is invertible, so is ·A - (. To see that R( A- () == 11, we fix a vector g E 1i and
set f ==(A- z)- 1 [I (z- ()(A- z)- 1 ]- 1 g. Clearly f E V(A) and (A- ()f == g
by (2.95). This shows that R(A- () == 11, and since f == (A- ()- 1 g, it also follows
that
(A- ()- 1 ==(A z)- 1 [I+ (z- ()(A- z)- 1 ] - l E B(H) (2.96)
as a product of two operators each of which belongs to B(H). We have thus shown
each z E p(A) has a complex neighbourhood belonging also to p(A); consequently
p(A) is an open subset of C.
(d) The Taylor expansion in a neighbourhood of z E p(A) is obtained by inserting
(2.94) into (2.96):
00

(A ()-1 = 2.:)(-z)k[(A-z)-l]k+I. (2.97)


k=O
D
The function z f----+ (A - z) 1 is infinitely differentiable, its derivatives exist in
the uniform topology [in the norm of B(H)]. They can be calculated from (2.97)
[differentiate (2.97) n times with respect to (and then set ( == z] or from (2.93):

(n > 1) (2.98)

with the convention that (A - z)- k == [(A - z) - 1 ] k. In particular one

(2.99)

The complementary set of p(A) in the complex plane Cis called the spectrum of
A and denoted by u (A):
u(A) == C\p(A). (2.100)

o-(A) is a closed subset of C. A complex number z belongs to u(A) if A - z is not


invertible or if R( A - z) 1i (see the footnote on page 68). The spectrum of a
closed operator A contains in particular all eigenvalues of A. A complex number z is
an eigenvalue of A if there exists a vector f 0 in V(A) such that Af == zf, other
terms if A- z is not invertible. Apart from the eigenvalues (if any) the spectrum of A
contains complex numbers z for which A - z is invertible but with unbounded or not
densely defined inverse.
70 LINEAR OPERATORS

If A is closed with dense domain in 1i, then its adjoint A* exists and is closed (see
Section 2.4). One then has a very simple relation between the spectrum of A and that
of A*: the latter is obtained from the former by reflection at the real axis, i.e.

z E a(A *) {:=:::::? z E a(A) and z E p(A *) {:=:::::? z E p(A). (2.101)

We do not give the proof in the general case (see Problem 2.20). In this text we are
particularly interested in self-adjoint operators; as shown below, the spectrum of such
an operator is a subset of JR, i.e. each z E CC with ~ z # 0 belongs to p( A). The
spectrum of symmetric operators will be discussed at the end of Section 3 .1. A simple
property of their eigenvalues is given in the next proposition.

Proposition 2 . 36. (a) A symmetric operator has no non-real eigenvalues. In particu-


lar all eigenvalues of a self-adjoint operator are real.
(b) Eigenvectors associated with different eigenvalues of a sy1nmetric operator are
orthogonal to one another.
PROOF. (a) Assume that A C A* and that f E V(A) is such that llfll 0 and Af ==
zf for some z E CC. Then

zllfll 2 == (f, zf) == (f, A f) == (Af, f) == (zf, f) == zllfll 2 ·

Since llfll # 0, one must have z == z.


(b) Assume that Af == Aj and Ag == J-Lg with f, g E V(A) and with A # f-L· Then
A(f, g) == (Af, g) == (f, Ag) == J-L(f, g) (remember that A and f-L are real). Since
A f-L, we must have (f, g) 0. 0

The following identity will be important. Let A be a symmetric operator and z ==


A+ i{L a complex number (A, f-L E JR). Then one has for each f E V(A):

II (A- z)fll 2 -II (A- A)fll 2 J-L 2 1lfll 2 - itL((A- A)j,f) + itL(f, (A- A)f)
== II(A- A)fll 2 J-L 2 ilfll 2 . (2.102)

Proposition 2 . 37 . The spectrum of a self-adjoint operator A is real, i.e. a( A) C JR.


If' z- A+ i{L with A, f-L E JR and f-L # 0, then z E p(A) and II (A- z)- 1 11 < 1/IJ-LI.
PROOF. Assume that z == A+ i{L E CC\JR. By (2.102) we have I (A- z)fll > 1~-LIIIfll
for each f E V(A). Since f-L # 0, this implies that A - z is invertible. Setting f ==
(A- z)- 1 g in (2.102), with g E R(A- z), we get I (A- z)- 1 gll < IJ-LI- 1 IIgll- This
shows that (A- z)- 1 is bounded, with norm< 1/IJ-LI. Finally we see from Proposition
2.19(b)andProposition2.36(a)thatR(A-z)_L ==N(A*-z) N(A-z) == {O}.So
R( A- z) is dense H, and since (A- z) - 1 is bounded, we must haveR( A- z) == 1i
(see the proof of Lemma 3.1 (c) for more details about this last property). 0

Proposition 2.38.. If z is an eigenvalue of a closed operator A, denote by M (z) ==


{f E V( A) I Af == z f} the set of all eigenvectors of A associated with this eigenvalue.
Then M (z) is a subspace of H. In particular the null space N(A) is a subspace.
PERTURBATIONS OF SELF-ADJOINT OPERATORS 71

{fn} be a (strong) Cauchy sequence in M(z). We denote limit by


We f E M(z), i.e. that f E D(A) Af zf. this we
observe llfn- fn~,JJ --+ 0 as rn,n--+ oc consequently llAfn- Afrnli ==
llzfn - zfmli == \ziilfn- fmll --+ 0 as m~, n --+ oc. Thus the sequence {.fn} and
sequence { Afn} are Cauchy. Since A is closed, it follows that limit of
sequence {fn} belongs to D(A), i.e. f E D(A), Af == s-lin1n---+oo Afn ==
S- n-+ 00 Z f n == Z · S - li1ll n-+ 00 f n == Z j.
z == 0 is an eigenvalue of A, N(A) == M(z == 0), so N(A) is a subspace
result. z == 0 not an eigenvalue of A, N (A) == , which
D
The spectrum of an operator A E B(H) contained the disk
{z E CC liz\ < II All}.

z E CC be such \zl > II II· may write A- z == -z - A/z).


IIA/zll == IIAII/Izl < 1, Proposition 2.5 implies that operator - z
vertible, with

(A-z) _ 1 == [ A J -1 ==-I--:-
-z(I--) -1 ( ) 1
EB(H).
z z ~

z belongs to p(A).

we point out that, if the operator is self-adjoint and z E p( A),


[(A- z)- 1 ]* ==(A- z)- 1 (A== A*). (2.1

iff, g E H,
-'LLJl.._..,,..,.,..,'Ul. exist vectors fo, go E D(A) such f == (A z)fo
g == (A- z)go.
(f, [(A - z) 1
] *g) == ( (A - z) -r f ~ g) == ( f o, g) == ( fo , (A - z) go)
==((A- z)fo,go) == (f, (A z)- 1g).

An ilnportant aspect of theory of self-adjoint operators concerns 1nvar1ance


of self-adjointness under perturbations. If A is a self-adjoint operator
belongs to B(H), it is easy to see that A is also self-adjoint on
D(A +B) == D(A). Thus property of being self-adjoint invariant . . . ~~ A A ......

addition of bounded self-adjoint operators. 1nany quantum-mechanical situations


one considers perturbation of a self-adjoint A operator by an unbounded operator
it important to know criteria implying the self-adjointness of B.
section is devoted to this question.

2 . 7 . 1. If A are symmetric operators and if D(A) n D(B) dense H,


their sum A is also sy1nmetric: iff, g E D(A B) D(A) n D(B),
((A+ B)f, g) == (Af, g) + (B f, g) == (f, Ag) (f, Bg} == (f, (A B)g).
72 LINEAR OPERATORS

However, even if both A and are self-adjoint, the operator A + B will not be self-
adjoint general (it will only be symmetric). In certain situations one can consider
B as a "perturbation" of A; roughly this means that, in a certain sense, B is small
compared to A, and under a condition of this type it is possible to prove that the sum
A is self-adjoint on a suitable domain, for example on D(A) for a certain type of
perturbations
The simplest situation is that in which E B(H) (B == B*). In this case the sum
A is self-adjoint on D(A). To see this, it suffices to show that D( (A +B)*) C
D(A B) D(A). So suppose that f E D((A B)*), and let j* E H be such that
(f, (A+ B)g) == (j*, g) for all g E D(A). Then
(j,(A B)g) == (f,Ag) + (f,Bg) == (j,Ag) + (Bj,g) == (f*,g).
Thus one has (f,Ag) == (j*- Bf, g) for each g E D(A). Since j* Bf E H, this
means that f E D(A*) _ D(A).
The preceding argument shows that the addition of a bounded self-adjoint operator
B to a self-adjoint operator A does not change the self-adjointness. The same con-
clusion can be obtained when B is unbounded but "small compared to A modulo a
bounded operator". Mathematically concept "small compared to A" could for ex-
ample be expressed as follows: I B f I < E I Af I for some sufficiently small E > 0 and
for all f E D(A) n D(B). More precisely one says that is A-bounded (or relatively
bounded with respect to A 16 ) if D(A) D(B) and there exist constants a > 0 and
(3 > 0 such that

I I <aliA! II+ !311!11 VjED(A). (2.104)


The infimum of all numbers a for which such an inequality is true (for some (3 which
can depend on the value of a) is called A-bound of (or the relative bound of B
with respect to A). If this number is < 1, one may consider that B is smaller than A
modulo a bounded term.
If BE B(H), then (2.104) holds with a== 0 and (3 == liB II, hence B is A-bounded
with relative bound 0. We shall see further on examples of unbounded operators B
having also A-bound 0, i.e. which are such that, for each E > 0, there exists a number
/3 == (J(E) E (0, oo) liB!II < ellA! I (J(E) 11!11 for all f E D(A) [when B is
unbounded one must have (J(c) ----+ oo as E----+ 0]. The important fact is the following:
if A is self-adjoint and is symmetric with A-bound< 1, then A+ B is self-adjoint on
D(A) D(A) n D(B). Before proving this we give a reformulation of the condition
(2.1 04).
D(A) _ D(B) and z E p(A) (the resolvent set of A), the operator B(A- z)- 1
is on each vector g E 1-{ because (A- z)- 1 maps H onto D(A). We shall see,
for a self-adjoint operator A, that condition (2.1 04) is equivalent to the condition
1
that B(A- z)- E B(H). To obtain this result, it useful to consider an inequality
is si1nilar to (2.1 04) but contains the squares of the involved norms, viz.
VjED(A), (2.1 05)
a 2 0 and b 2 0 are constants.
16 This definition holds for an arbitrary operator A (A does not have to be self-adjoint or closed).
PERTURBATIONS OF SELF-ADJOINT OPERATORS 73

Lemma 2 . 40 . Let A be self-adjoint. For a, b > 0, the condition (2.105) is equivalent


to IIB(A ± i£)- 1 11 <a. 17

PROOF. By using (2.1 02) one sees that (2.1 05) is equivalent to

VjED(A). (2.106)

Since ±ibI a belong to the resolvent set of A, each of the operators (A± ib I a) 1 gives
a bijection between Hand D(A), so that the preceding inequality is equivalent to 18

IIB(A VgEH. (2.107)


D

Lemma 2 . 41 . Let A be a self-adjoint operator and D(A) C D(B).


(a) If (2.104) holds, then (2.105) is satisfied for each a > a (the value ofb depends
on a).
(b) If (2.104) holds, there exists a number /<l; > 0 such that B(A + i/<l;)- 1 E B(H) and
B(A- i/<l;)- 1 E B(H).
(c) If B(A- z)- 1 E B(H) for some z E p(A) [in particular if (2.105) is satisfied},
then (2.104) holds with a== IIB(A- z)- 1 1\.

PROOF. (a) If(, rJ E JR and c: > 0 one has 0 < (c: 112 r- c:- 1 12 rJ) 2 == c(
2
- 2rrJ +
c:- 1 rJ 2 . This implies that

(r + rJ) 2 - '2+ 2rrJ + T/2 < '2 + (c(2 + c -1 T/2) + T/2


== (1 + c:)r 2 + (1 + c:- 1 )rJ 2 . (2.108)

Assume now that (2.104) holds, and choose in (2.108) r == aiiAfll and rJ == /311!11·
Then

By taking c: sufficiently small one obtains the validity of (2.1 05) for any a > a.
(b) This is immediate from (a) and Lemma 2.40.
(c) We set a== liB( A- z)- 1 11 and write for f E D(A):

IIBfll == IIB(A- z)- 1 ·(A- z)f\1 < IIB(A- z)- 1 1\I\(A- z)fll
< a[IIAJII + lzl\1!11]. D

17
If an equation or an expression contains double signs (for example±), it is understood that this 1neans
two equations or expressions, one by taking everywhere the upper sign, the other one by taking everywhere
the lower sign.
18
To see this equivalence, set f == (A ± i ~) - l gin (2.106) or g == (A± i% )fin (2.1 07).
LINEAR OPERATORS

be a selj'-adjoint operator and B an operator such that

for so1ne z E p(A), then B is A-bounded.


then - z) - 1 E for each z E p(A) and, if b denotes
o.f B, then

1/B
follows fro1n Lem1na 2.41 (c).
By Len1ma 1 there exists /"{; > 0 B if'{;) 1 E B(H).
E one resolvent equation (2.93):
(A- iK)- 1] (z- i/"{;)[B(A- if'{;) 1J(A- z) 1 (2.110)
belongs to , hence B - z)- 1 E B(H).
Now let b be A-bound of B. By l(c), z E p(A), then (2.104)
cv == IIB(A- z)- 1/, hence b < infzEp(A) IIB(A- z)- 1 11- To obtain
1

to show c) > there exists a real number /"{; #- 0


such II ~iK) 1
< b
11 (2.1 04) satisfied for any a > b,
hence for a == b then 1(a) implies (2.1 05) satisfied with
a==o~ b S so1ne b > virtue of this means
II fiK) ll<b+S
1
/"{; == b/(b + 6)]. D

5'eif-adjoint. Then:
A-bound b < 1 ¢:=? there a nu;nber z E p(A) with

A-bound b < 1, exists a /"{; > 0 such that


liB IIB(A + i/"{;)- 1 11 < 1.
im1nediate (2.109). follows the arguments part of
proof. D
next proposition we use the notion of semi bounded self-adjoint operators. A
called lower or below there
A the ( -x, A) belongs to resolvent set
semibounded if there exists A E JR

be a self-adjoint and
operator A-bound b < 1. Then:
B is se(f-adjoint on
B-bounded.
sen1ibounded, then B is also sen1ibounded.
1
Since b < 1, there exists K > 0 that IIB(A ±if'{;) 1/ < 1 [Corollary
. The operator A+ B is symmetric on V (A). Let us R(A+B±i/"{;) ==
write for f E V(A) V(A +B):
(A+ i/"{;)f == [I B (2.111)
PERTURBATIONS OF SELF-ADJOINT OPERATORS 75

As ±i/<l; E p(~4), the operators (A±i/<l;) map V(A) onto H. Since IIB(A±i/<l;)- 1 11 < 1,
the operators I+ B(A ± i/<l;)- 1 are invertible, with inverses in B(H) (by Proposition
2.5). Hence I+ B(A ± i/<l;)- 1 map H onto H. Consequently Eq. (2.111) implies
that R(A + B i/<l;) == H. SoH == R(A + B ± i/<l;) == R(/<l;- 1 (A + B ± i/<l;)) ==
R(/<l;- 1 (A -+-B)± i). By Proposition 2.24, /<l;- 1 (A +B) (hence also A+ B) is self-
adjoint on V(A).
(b) For f E V(A) V(A +B) we may write I!Afll == II(A + B)f- Bfll <
\\(A B)f\1 + \IBJ\\. We insert this inequality into (2.104) to obtain

1\Bf\1 < a\IAJI\ +;3\lfl\ < a\I(A +B)f\\ + aiiBf\1 + ;3\lf\1. (2.112)

Since b < 1, this inequality holds for some a < 1. Assuming that a < 1, (2.112)
becomes
(1- a)I\Bfll < a\\(A + B)f\1 + ;31\JII·
Hence B is (A +B)-bounded with relative bound< a/ (1- a) [observe that in general
the (A+ B)-bound of B will not be smaller than 1].
(c) We shall use the following fact that is proved in Chapter 4 [see (4.51)]: if
rp: JR --7 Cis a continuous function [here rp(-A) == (-A- z)- 1 or rp(-A) == -A(-A- z)- 1
with z E p(A); see also the footnote on page 155], then the norm of the operator rp(A)
is equal to the supremum of lrp(-A) I as A varies over the spectrum of A (if A is a multi-
plication operator, this result is contained in Proposition 2.26).
(i) Assume first that A is a positive operator, i.e. such that ( -oo, 0) c p(A). In this
case the spectrum of A is contained in [0, +oo ), so that one has for z < 0:

II(A-z)- 1 \l < sup(-A-z)- 1 == ( -z)- 1 and IIA(A-z)- 1 11 <sup -X(-A-z)- 1 == 1.


A~O A~O
(2.113)
We must show that there exists ,;\ 0 E JR such that the interval (- oo, Ao) belongs to
p(A +B). For this we write f == (A- z)- 1 g in (2.104) (assuming that z < 0):

VgEH.
Combined with (2.113) this leads to

Since one can choose a < 1, it is possible to find a number ,;\ 0 E (- oo, 0) such that
\IB(A- z )- 1 1\ < 1 for each z < -A 0 . For these values of z the operator I+ B(A- z )- 1
is invertible with inverse in B(H), i.e. I+ B(A- z )- 1 maps H onto H if z E ( -oo, -X 0 ).
Let us show that ( -oo, ,;\ 0 ) C p(A +B). For this let z < -X 0 . One has on V(A):

A+B-z== [I+B(A-z)- 1 ](A-z). (2.114)

(1) Iff 1- 0 one has (A- z)f 1- 0 and [I+ B(A- z)- 1 ](A- z)f 1- 0, because
A- z and I+ B(A- z)- 1 are invertible. Hence A+ B- z is invertible.
(2) A- z maps V(A) onto H, and I+ B(A- z)- 1 maps H onto H. Hence A+B- z
maps V(A) onto H, i.e. R(A + B- z) ==H.
76 LINEAR OPERATORS

(3) One has (A+ B- z)- 1 == (A- z)- 1 [! + B(A- z)- 1 J- 1 E B(H).
(ii) The case of a general lower semibounded operator is easily reduced to the
preceding one: if ( -oo, -A) C p(A) for some A < 0, then ( -oo, 0) C p(A - -X), so
that it suffices to replace the operator A by A - A in the preceding considerations (and
A+ B by A+ B- -A). Upper semibounded operators are treated similarly [if A < 0,
then -A > 0, hence there is a number Ao < 0 such that ( -oo, Ao) C p(- A - B),
because - B will be relatively bounded with respect to -A with relative bound < 1.
So(j-Aol,+oo)cp(A+B)]. D

Corollary 2.45. Let A be a self-adjoint operator and Ban A-bounded symmetric op-
erator with A-bound b < 1. Then one has the second resolvent equation which holds
for all z E p(A) n p(A +B):

(A- z)- 1 - (A+ B- z)- 1 ==(A- z)- 1 B(A + B- z)- 1


==(A+ B- z)- 1 B(A- z)- 1 . (2.115)

PROOF. If z E p(A) n p(A +B), then the operators B(A +B- z)- 1 and B(A- z)- 1
belong to B(H), hence each term in (2.115) is well defined in B(H). As (A+ B- z )- 1
maps H onto D(A) one may write

(A+ B- z)- 1 ==(A+ B- z)(A + B- z)- 1 (A- z)(A + B- z)- 1


==1-(A z)(A+B-z)- 1 .

Upon multiplying this identity by (A- z)- 1 , one obtains

This proves the first equation (2.115). The second one can be obtained in a similar
w~ D

Remark 2.46. (a) The condition that the A-bound of B be strictly smaller than 1 is
essential in the Rellich-Kato Theorem. In the case in which (2.1 04) is satisfied with
a == 1, it is possible to prove that A+ B is essentially self-adjoint on D(A) (Theorem
ofWiist). The following simple example shows that one cannot expect more. Let A be
self-adjoint and unbounded, and take B == -""(A with r > 0. Then IIBJII < ri!Afll
for each f E D(A), hence B is A-bounded with relative bound r· If r < 1, then
+ B == (1 - r)A, which is a self-adjoint operator on D(A). If r == 1, then A+ B
is the restriction of the zero operator 0 to V( A); this operator [the restriction of 0 to
D( A)] is essentially self-adjoint but not self-adjoint, since its adjoint is the zero op-
erator with domain H, and V(A) 1- H.
(b) A simple adjustment of the proof of Proposition 2.44 shows that, if A is essentially
self-adjoint and B is symmetric and A-bounded with A-bound b < 1 , then A+ B is
essentially self-adjoint on V(A).
PERTURBATIONS OF SELF-ADJOINT OPERATORS 77

2. 7.2. A particular class of relatively bounded operators are the relatively compact
operators. Let A be a closed operator with domain D(A). An operator B is said to be
relatively compact with respect to A (or A-compact) if V(A) C D(B) and if there
exists a number z E p(A) such that B(A- z )- 1 is a compact operator. The A-compact
operators are A -bounded with relative bound 0, as shown in the following proposition.

Proposition 2. 47 . Let A be a self-adjoint operator and B a symnletric operator that


is A-compact. Then
(a) One has B(A- z)- 1 E JC(H) for each z E p(A).
(b) B is A-bounded with A-bound b == 0; in particular A+ B is self-adjoint on V(A).
(c) If C is symmetric and A-bounded with A-bound b < 1, then B is also (A +C)-
compact.

PROOF. (a) By hypothesis there is a number z0 E p(A) with B(A- z 0 )- 1 E JC(H).


For z E p(A) one has by the first resolvent equation (2.93):

B (A - z) - 1 == [B (A - z0 ) - 1 J [I + (z z 0 ) (A - z) - 1 J .

The first factor on the right-hand side is compact and the second one belongs to B(H).
By Proposition 2.11 (b) the operator B (A - z) - 1 is compact.
(b) We must show that, given E > 0, there exists a number z in the resolvent set of
A such that IIB(A- z)- 1 11 < E [see (2.109)]. For this we take z == i{L with fL > 0 and
show that IIB(A- ifL)- 1 11 - 7 0 as fL - 7 oo. We write

(2.116)

and observe that

Thus (A- i)(A- i{L)- 1 belongs to B(H) and one has (see Proposition 2.37):

(2.118)

We show below that [(A- i)(A- i{L)- 1 ]* converges strongly to zero as fL - 7 oo.
Since B(A- i)- 1 E JC(H) by the result of (a), it then follows from (2.116), by using
Proposition 2.12(b), that lim~--t(X) IIB(A- i{L)- 1 11 == 0.
Let us set T~ == [(A- i)(A- i{L)- 1 ]* == (A+ i)(A + i{L)- 1 [the second equality
is formally evident; it can be established by taking the adjoint of the right-hand side
of (2.117), using (2.1 03 ), and one sees that this amounts to replacing i by - i]. If
f E V(A), one has for p, > 0 [using (2.92) and then again Proposition 2.37]:

which converges to zero as fL - 7 oo. Since IIT~II < 2 by (2.118), it follows from
Proposition 2.2 that s-lim ~--t(X) T ~ - 0.
78 LINEAR OPERATORS

(c) By the second resolvent equation (2.115) one has

B(A + C- i)- 1 == B(A- i)- 1 - B(A- i)- 1C(A + C- i)- 1


== B(A- i)- 1 [I- C(A + C- i)- 1 ].
We have B(A- i)- 1 E JC(H) by (a) and C(A + C- i)- 1 E B(H) by Propositions
2.44(b) and 2.42(b). As in (a), one obtains that B(A + C- i)- 1 E JC(H). 0

2. 7 .3. The results obtained above can be applied to prove the self-adjointness of quan-
tum-mechanical Hamiltonians. We have seen in §2.5.3 that the free Hamiltonian Ho
of non-relativistic quantum mechanics is essentially self-adjoint on C0 (IRn), and we
have specified its domain of self-adjointness in (2.81). We note that, in the sense ex-
plained in §2.5.5, Ho is the operator cp(P) for cp(x) == x 2 .
Let us consider Hamiltonians of the form H == H 0 + V, where V - V (Q) denotes
the operator of multiplication by a function V (x) called a potential. The following
terminology will be used. We say that
( 1) v E Lfoc (IRn) J
if w Iv (x) jP dnx < 00 for each bounded Borel subset w of IRn'
(2) V E L P(IRn) + L (X) (IRn) if V can be written in the form V (x) == V1 (x) + V2 (x)
with vl E LP (IRn) and V2 E L (X) (IRn)'
(3) V E LP(JRn) + L~(IRn) if, given any r5 > 0, one can decompose V into V ==
vl
vl + v2 with E LP(JRn), v2 E L(X)(IRn) and I/V211(X) < 6 (the decomposition may
depend on the value of 6).

Example 2.48.. An important example in quantum mechanics is the Coulomb poten-


tial V(x) == rlxi- 1 with r E IR and n == 3. This potential is contained in none of the
spaces £P(JR 3) (unless r == 0). Indeed one has (for p-::/- oo)

(2.119)

If p > 3, the integral is convergent at infinity but divergent at zero, if p == 3 it diverges


both at zero and at infinity, and if p E [1, 3) it is convergent at zero but divergent at
infinity. However the Coulomb potential (for n == 3) belongs to LP(IR 3 ) + L~(IR 3 )
for each p E [1, 3): take V1 (x) == rlxl- 1 for jxj < Rand V1 (x) == 0 for !xl > R,
hence v2 ( v2 (
x) == 0 for Ixl < R and x) == r Ixl-l for Ixi > R, where R is a positive
number. It is clear that, for any choice of R, one has V1 E £P(JR 3) if 1 < p < 3
[apart from a factor I1 I, I V1 l P is given by (2.119) but with the upper limit of the
integral replaced by R]. v2
is a bounded function, and IIV21l(X) == SUP!xi>R lrlxl-ll ==
I;IR- , which can be made arbitrarily small by choosing R large enough.
1

Let us now discuss the properties that a potential should have so that it is a pertur-
bation of H 0 in the sense considered here.
order for V( Q) to be symmetric (and self-adjoint on its maximal domain), one
must assume that V is a real function.
PERTURBATIONS SELF-ADJOINT 79

condition V(H0 ) C V(11) we can use the characterisation of


) obtained in (2.83), since is a multiplication operator
) contains S(IR.n ), must be square-integrable, one must
E (IRn): a set IRn, choose a function f (IRn)
f(x) == 1 for all x E W. Since S(IRn) belongs to V(H 0 ), one
(IRn)'

, 2 or 3, so I (x)f(.i) I <
E (IRn) also
(JRn ).
f E V(Ho), f E (IRn) s
2
E L (1Rn ), assume
12l)one I!
1/ s. scan be , 2n/(n- ), one
q> n/2 [s == 2 ====? q == x; s == 2n/(n- 4) ====? q ==
+ be a

(a) 1l == 1 2 or E L2(1Rn) +
' bound b ==
(IRn ), then (Q) -co1npact.
n > P(ffi.n) +
bounded with b ==
V E P(JRn) + (IRn) for so1ne p E ( n/2, x ), then
We have
V E Lfoc (JRn);

II
(2.1

a (surface area of
converges zero as fL -----1- x (if n == write
/L2)1 + fL2)-E < k2 /[(k4 /12)1 . P~-2E
is H 0 -bounded with relative 0 [see (2.1
operator, is H 0 -bounded with relative b == 0.
(ii) V E (IRn) (IRn) 6 > 0 we choose a decomposition ,6 .6
with '6 E (IR. n)' E LX) (IR n) II v2' 611 < 6. The (X) ,6(Ho- i)-
11 < II v2,6ll - i) 111 <
1
is compact (even Hilbert-Schmidt) and II v2.6 - i) II (X)

1
6. Consequently - i)- is a compact operator as the of
80 LINEAR OPERATORS

of COinpact operators { v1 ,6 ( H 0 - i) - 1 } as b - 7 0 [Proposition 2.11 (d)].


(b) The proof is similar. Let V == V1 + V2 with V1 E L P (JR. n) and V2 E L oo (JR. n).
this case V1 (Ho - iJL)- 1 is compact [Proposition 2.34(b)] and the norm of this
operator can be estimated by using (2.87):

Again, since p > nj2, the integral is finite and converges to zero as f-L -7 oo. We omit
the remaining details of the proof. 0
Remark 2.50. In view of the results of Example 2.48, part (a) of the preceding propo-
sition implies in particular that the Coulomb Hamiltonian H c in L 2 (JR 3 ) is self-adjoint
on V(H0 ). Since H 0 is essentially self-adjoint on C0 (JR 3 ) (see page 81), He is also
essentially self-adjoint on C0 (JR 3 ) [see Remark 2.46(b)].

Appendix to Chapter 2

A.2.1. Proof of Proposition 2 . 33 for 2 < p < oo. The proof involves the following
two results:
(1) The Holder inequality: Let (0, A, m,) be a measure space. Iff E LP(O, m),
g E ( 0, m,) and 1/r == 1/p+ 1/ q, then the product function fg belongs to Lr (0, m)
and
(2.121)
The proof is simple, see e.g. page 113 of [R].
(2) The Fourier transformation :F is bounded from Lq(Rn) to Lq' (Rn) if 1 < q < 2
and q' is defined by 1/ q' == 1 - 1/ q:

II illq' [l, li(kW' dnkrN < (27r)n(l- 2q_, l/ 2 11 !llq

- (27r)n(l-2q-1)/2 [l,IJ(XWdnx] 1/q. (2.122)

This is called the Hausdorff ... Young inequality. For a proof see for example page 11
Volume II of [RS]. If q - 2 one has q' == 2, and (2.122) expresses the unitarity of
:F as an operator in L 2 (JRn ). If q == 1, one has q' == oo, and (2.122) coincides with the
evident inequality llflloo < (21T)-nl 2 llflh [see (2.73)]. If q E (1, 2), then q' E (2, oo)
(2.1 is obtained by interpolating between the cases q == 1 and q == 2.
(i) We first show that C 0 (JR.n) C V( 1/J(P)cp( Q) ). Iff E C0 (JRn) one can choose
a bounded open set in JRn such that f(x) == 0 for all x tf_ V. Clearly llcp( Q)JIIP <
llcpiiPIIflloo, hence cp(Q)f E LP(JRn). Then, by (1.49), one has cp(Q)f E Lr(JRn) for
each r E [1, p]. Since p > 2, this holds in particular for r == 2, which shows that f
APPENDIX TO CHAPTER 2 81

belongs to D(cp(Q)). To see that cp(Q)f E D(~(P)), we use the Hausdorff-Young


inequality to find that :Fcp(Q)f E L 8 (JR.n) for each s E [2, oo], in particular for s ===
so=== [112-llp]- 1 ; since 112 == 1lp+1ls 0 , onehas~(·)[:Fcp(Q)f](-) E L 2 (JR.n) by
the Holder inequality.
(ii) To estimate the norm of ~(P)cp(Q), let g E L 2 (JRn) and define q E , 2) by
1lq ===lip+ 112. Then (2.122) and the Holder inequality imply that
II:Fcp(Q)gl\q' < (27r)n(l- 2q-l)/ 2 llcp(Q)gl\q < (27r)n(l- 2q-l)/ 2 llcpllpl\g\\2
(27r) -n/p II cp\lp 119112·
Now 1I p + 1I q' == 1I p + 1 - 1I q == 1I 2. So, by using again the Holder inequality,
one arrives at

II~(P)cp(Q)gll ==II~(·) [:Fcp(Q)g] (·)112 < ll~llpii:Fcp(Q)gllq'


< (27r)-n/pll~\\p\\cp\\P\\g\\2· 0

A.2 . 2. End of the proof of Proposition 2.34(a). Let us assume that~ E L 2 (JR.n)
tt
that cp tf_ L 2 (JRn ). We must show that cp( Q)~(P) B2 (H) except when~ == 0 a. e. [in
which case cp( Q)~(P) == 0]. Fork EN we define cpk by

cp(x) if lxl < k and jcp(x)l < k


'Pk(X) = ~ if lxl < k and jcp(x)l > k
{
if lxl > k.
Clearly cpk E L 2(JRn) for each k, and \cpk(x)l < jcp(x)l for each x E JRn. The last in-
equality implies that llcp(Q)gll > 1\cpk(Q)gll for each vector g. By taking into account
the definition (2.43) of the Hilbert-Schmidt norm and Eq. (2.89), one then obtains
l\cp(Q)~(P)IIHs > l\cpk(Q)~(P)IIHs == (27r)-n/ 2 llcpkii2111PII2· Since cp tf_ L 2(JRn)
and cp k ( x) ---t cp (x) for almost all X E JR n' one has II cp k 112 ---t 00 as k ---t 00 and hence
llcp(Q)~(P)I!Hs === oo (unless 111)]1\2 == 0). Thus cp(Q)~(P) tf_ B2(H). The other situ-
ations in Proposition 2.34(a) are treated similarly. D

A.2 . 3. -~is essentially self-adjoint on Cg<>(JRn). Let A 0 == ~lco(IRn) and A 1 ==


.6.)s(IRn). We have seen that A 1 is essentially self-adjoint, so the essential self-adjoint-
ness of A 0 will follow if we can show that the closure of A 0 is an extension of A 1 . To
do this we fix a function () E Co (JR n) such that () (x) == 1 for all X satisfying Ixl < 1'
and we define ek by ()k(x) == e(xjk), k == 1, 2, ... (so ek(x) === 1 for all lxl < k).
For f E V(A1) S(JRn) we set fk == Bkf· Then we have fk E Cc)(JRn) V(Ao)
and llfk -!II ---t 0 ask ---t oo. Now

[D.fk](X) = fh(X) [D.f] (X)+ I [VB](~). [V !] (X)+ k\ [D.B] (~)f(X)


and consequently
LINEAR OPERATORS

in1plies lirn k-+oo II~~ - ~fk II == Since is closable, we conclude


f belongs to V(Ao) Aof == ~~ f, which shows that A 1 C Ao. D

Notes
alternative reading we recommend the texts [AG], [RN] and [WI] already cited
Bibliographical Notes to Chapter 1, also Volume I of [RS], [KJ and Part I [W2].

Show that an operator E B(H) is completely (and uniquely) determined by its


expectation values k == \cJ ~ Ac k) orthonormal basis {en} of the underlying
Hilbert space H.
Let {en} be an orthonormal basis of an infinite-dimensional Hilbert space. For
== 1, 2, 3, ... let AN be the operator defined as follows:
(X)

AN ( L akek) == L ak+Nek.
k=l k=l

(a) Show that E B(H) and determine its nor1n II II·


Prove that } converges strongly to the zero operator 0 as ~ x.
(c) A1\r (AN)*.
One knows * ~ 0 weakly. Does '~- converge strongly to 0 ?
{ e 77 } be an orthonormal basis of an infinite-dimensional Hilbert space. For
== 1, 3 .... let be operator defined as follows: B 1v == eN and ENek ==
0 fork> 1. liEN II== 1, thatw-lirnN--+oo EN== 0 thats-limN-+oo EN
does not exist.
two sequences of bounded operators {An} {En} in a Hilbert space H
that w -lirn n-+oo == 0 and w -lirn n-+oo En == 0 but such that {An En} does
converge weakly to 0.
2.5. { An}nEN a sequence of operators satisfying II II < M < x for all
n E N and lirn n-+oo (f, Ang) == 0 for all f and g belonging to some dense set V.
Prove that w-lirnn-+oo == 0.
2.6. Let E B(H) be such that \f, Af) is real for each vector f H. By using the
polarisation identity, show that is self-adjoint.
Let E be a projection and M(E) == {! E H j Ef == f}. Prove that M(E) is a
subspace.
2 . 8. Prove Proposition 2.6.
Let {en} be an orthonormal basis of an infinite-dimensional Hilbert space. For
== 1, 2, 3, ... let PN be the projection onto the subspace spanned by the vectors
c N. 1, , ... Determine which of the following properties are satisfied by the
sequence of operators {PN }: (i) u-limN-+ooPN == 0, (ii) s-limN-+ooPN == 0,
(iii) w-linlN-+oo PN == 0.
2 . 10 . Prove Proposition 2.8(b ).
PROBLEMS 83

2.11. Prove Proposition 2.9(b)- (e).


2.12. Show that a bounded operator A in a Hilbert space H is a finite rank operator if
and only if its range R( A) is finite-dimensional.
2.13. Let [J be a unitary operator in Hand A E B(H). Show that \\UA\\ == \\AU\\ ==
\lUAU*\! == IIA\1 and 1\UA\\Hs == \\AU\\Hs == \\UAU*\\Hs == \IAIIHs·
2.14. Let {en} be an orthonormal basis of an infinite-dimensional Hilbert space. Let
A be the operator defined as follows:

A(L akek) == L PkO~kek,


k=1 k=1

where Pk are complex numbers. Specify the conditions that have to be imposed on
these numbers Pk such that A is: (i) bounded, (ii) compact, (iii) Hilbert-Schmidt, (iv)
of finite rank, (v) an isometry, (vi) unitary, (vii) a partial isometry, (viii) a projection,
(ix) self-adjoint.
2.15. Prove the following assertions.
(a) If A is closed, then it is closable, and one has A == A.
(b) The null space of a closed operator is a subspace.
(c) If A is closed and BE B(H), then A+ B is closed.
(d) If A is closable and BE B(H), then A+ B is closable.
2.16. Let {en} be an orthonormal basis of an infinite-dimensional Hilbert space. De-
fine a linear operator A as follows: D(A) is the set of all finite linear combinations of
vectors in {en} and Aek == ke1 fork == 1, 2, 3, ....
(a) Show that A is not closable.
(b) Show that D(A *)is not dense in H.
(c) Find a sequence {An} of operators in B(H) converging strongly to on D(A).
2.17. Let A be an unbounded closed operator, and let g and h be two vectors with g tJ_
V(A). Define an operator B by D(B) == {f+agl f E D(A), a E C} and B(f+ag) ==
Af + ah. Show that B is closed. [By varying h, one obtains an uncountable number
of closed extensions of A].
[Hint: Consider a Cauchy sequence {fn + ang} in the do1nain of B such that the
sequence { B (fn + ang)} is also Cauchy. Show that the sequence {I an I} must be
bounded; hence there is a subsequence for which the ntunbers (Y n converge to a limit
a.] If necessary, consult page 71 of [P] for the complete solution.
2.18. Prove Proposition 2.21.
2.19. Prove the following statements.
1
(a) If A is invertible and closed, then is closed.
1
(b) If in addition A - E B(H) and B is a closed operator, then AB is closed.
(c) If A is symmetric and invertible and its range R( A) is dense in H, then A - l is
symmetric.
2.20. If A has dense domain in H, is invertible and A - 1 E B(H), then A* is also in-
vertible and A *- 1 == (A - l )*.
84 LINEAR OPERATORS

2.21 . Let {A, D(A)} be a linear operator and U a unitary operator in a Hilbert space
7-l. Define an operator Au by D(Au) == UD(A) and Au == UAU* (see Remark 2.25).
(a) Show that if A is closable, then so is Au.
(b) Show that if A is self-adjoint, then so is Au.
(c) Assume that A is closed. Show that Au and A have the same spectrum and the
same eigenvalues.
2.22. Show that a bounded symmetric operator is essentially self-adjoint.
2.23. (a) Let A be essentially self-adjoint and let B be a self-adjoint operator belonging
to B(H). Show that A+ B is essentially self-adjoint on D(A).
(b) Let V: JRn -+ 1R be a bounded measurable function and V(Q) the associated
multiplication operator in £ 2 (1Rn). Show that the operator P 2 + V (Q) is essentially
self-adjoint on C0 (1Rn ).
2.24. Consider the differential operator d 2 j dx 2 in 1-{ == £ 2 ( (0, 1) ). Find dense linear
manifolds Din 1-{ such that d 2/ dx 2 is symmetric on D.
2.25. Show that the canonical commutation relation [A,B] == if cannot be satisfied if
both A and B belong to B(H).
[Hint: By assuming that [A,B] ==if with A,B E B('H) and by calculating [A,Bn] for
n == 2~ 3, ... , one can arrive at a contradiction.]
2.26. Find two dense linear manifolds D 1 and D 2 in £ 2 (JR) such that the restrictions
of the position operator Q to D 1 and to D 2 are essentially self-adjoint and such that
D1 n D2 == {0}.
2.27. Consider the infinite-dimensional matrix Aj k given as follows:

0 if j == k
A j k == i I (j + k) 2
if j >k
{ 2
-ij(j+k) ifj<k

with j, k == 1, 2, 3, ... Let {en} be an orthonormal basis of an infinite-dimensional


Hilbert space 1-{ and let A be the linear operator defined by setting Aek == l:jA]kej.
Show that A belongs to B(H) and is self-adjoint.
2.28. Determine the spectrum of the operator A in L 2 (1R) defined by (Af)(x) ==
f(x + xo), where xo is a fixed real number.
2.29. Consider the shift operator Df == 2: kakek+l iff== 2: kakek, where { ek}kEN
is an orthonormal basis of an infinite-dimensional Hilbert space 1-{ (page 39).
(a) Show that f2 is isometric but not unitary and determine its adjoint D*.
(b) Find the eigenvalues (if any) of f2 and of D*.
(c) Determine the spectrum of D.
[Hint for (c): Use Proposition 2.39 and (2.101).]
2.30. Let A be a self-adjoint operator and B an A-bounded operator with relative
bound b. Furthermore let {Bk} be a sequence of A-bounded operators such that
ll(Bk- B)(A + i)- 1 !1 -+ 0 ask -+ oo. If bk denotes the A-bound of Bk, show
that b == limk-+oo bk.
PROBLEMS 85

2.31. Let H == H 0 + V(Q) be a Schrodinger operator in L 2 (IRn).


(a) Show that, if Vis continuous and V(x) -+ 0 as lxl -+ oo, then V(Q) is a Ho-
compact operator.
(b) Let n == 1 and let V(x) - I:kEZ k 0 · 999 xJk (x), where XJk denotes the characteris-
tic function of the interval Jk == (k- k- 3 , k + k- 3 ). Show that V( Q) is a H 0 -compact
operator.
2.32. Let - H 0 + 2A(Q) · P- idivA(Q) + IA(Q)I 2 + V(Q) be a Schrodinger
operator in L 2 (IR 3 ), with Ak, V: IR 3 -+ IR.
(a) If Ak E £P(IR 3 ) for some p > 3, then A( Q) ·Pis H 0 -bounded with relative bound
b- 0.
(b) If Ak E L 4 (IR 3 ) + L 00 (IR 3 ) and V, divA E L 2 (IR 3 ) + L 00 (IR 3 ), then the operator
2A( Q) · P i divA( Q) + lA( Q) 2 + V( Q) is H 0 -bounded with relative bound 0.
1

REMARK. The operator His obtained by developing H == [P + A(Q)] + V(Q),


2

which (in convenient units) is the Hamiltonian of a non-relativistic particle in the po-
tential V and the magnetic field B == rotA.
2.33. Prove the result given in Remark 2.46(b ).
CHAPTER3

Operators and their

When discussing unbounded operators in Section 2.4, we pointed out the difference
between symmetric and self-adjoint operators. For the purposes of quantum mechan-
ics it is very important to know the self-adjoint extensions, if any, a given sytnmetric
operator. In this chapter we present a beautiful method, due to J. von Neumann, for
obtaining a co1nplete characterisation of the closed symmetric extensions of general
symmetric operators (Section 3.1 ). We then apply the results to first and second order
differential operators, including Schrodinger operators with spherically syn1metric po-
tentials, and give criteria for such operators to be essentially self-adjoint.
We recall that A is a symmetric operator if its domain V(A) is dense in Hand A*
is an extension of i.e. if

g) == (f, Ag) Vj,gEV(A). (3.1)

Since A and its closure have the same closed extensions (Proposition 2.17), it will
be enough to consider symmetric operators A that are closed.

3.1 The method the Cayley transform

3.1.1. We begin with some preliminary results.

Lemma 3.1. Let B be a linear operator in a Hilbert space H. Assume that there exists
a constant f);> 0 such that liB !II > f);llfll for all f E V(B). Then:
(a) B is invertible,
(b) the inverse B- 1 of B is bounded, with IIB- 1 1 < 1/ f);,
(c) the rangeR( B) of B is closed (hence a subspace of H) ~f and only ~f B is a closed
operator. 1
1 Itis important to note that even if B is closed but R( B) -::/- 1-l, then B - l is not defined everywhere
and so does not belong to B(H) but only to B(R(B), 1-l). The norm II B- 1 11 appearing in (b) of the lemma
1neans the nonn IIB- 1 iln(B) in the notation introduced in (2.57).
88 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

PROOF. (a) B f == 0, then 11!11 < f);- 1 IIB !II == 0, hence f == 0. SoB is invertible.
(b) If g E R(B), then g == Bf for some f E D(B). Then IIB- 1 gll == 11!11 <
/<); -
1
II B f II - /<); -- 1 I g II· Thus II B I n (B) < /<); - 1 ·
(c) (i) Assume B to be closed. Let {gn} be a strong Cauchy sequence in R(B) and
denote its limit by g. To see that R(B) is closed, we must find a vector f E V(B) such
that g- Bf. Since gn E R(B), there exists for each n a vector fn E V(B) such that
gn == Bfn· We then have llfn- fmll < f);- 1 11Bfn- Bfmll == f);- 1 llgn- gmll, hence
the hypothesis that {gn} be strongly Cauchy implies that {fn} is strongly Cauchy. Set
f - s-lin1n-too fn· We are now in the following situation: we have a sequence {fn}
in D( B) which is strongly Cauchy and such that the sequence {B f n} is also strongly
Cauchy; since B is closed, this implies that fED( B) and B f == g (see page 49).
(ii) Assume that R(B) is closed. Consider a sequence of vectors {fn} in V(B)
such that s-lim n-too f n = f and s-lim n-too B f n g exist. To see that B is closed,
we must show that f E V(B) and Bf == g. Since R(B) is a subspace, one has g E
1
R( B); if we set h - g, we must show that f - h, i.e. that s-lim n-too f n - h. By
writinggn == Bfn as before, we have llfn-hll- IIB- 1 (gn-g)ll < f);- 1 llgn gjj ~ 0
as n ~ oo. D

We recall the identity (2.1 02 ): If A is a symmetric operator and z - A + i /); a com-


plex number (A,/); E JR), one has for each f E D(A):
(3.2)

This implies, together with Lemma 3.1, that if /J; -I 0, then (A- z)- 1 exists and
is bounded. However in general the closure of (A - z) - 1 does not belong to B (7-i)
because R(A - z) need not be dense in 7-i. From Proposition 2.19(b) we see that
R( A - z) j_ - N (A* - z). So, if A is a closed symmetric operator, we have:
(A- z)- 1 E B('H) -<::==? N(A *- z) == {0} -<::==? z is not an eigenvalue of A*.
(3.3)
Thus we have:
Proposition 3.2. If A is a closed symmetric operator and z a non-real complex num-
ber, then z belongs to the spectrum of A if and only if z is an eigenvalue of A*.
If is self-adjoint and z - A + ij); with /); -I 0, then z cannot be an eigenvalue
of A (Proposition 2.36), hence (A z) - 1 exists, is everywhere defined and
bounded. Consequently (as already seen in Proposition 2.37), the complex number z
belongs to the resolvent set of A, i.e. C\JR C p(A).
The preceding statement has a converse: if A is a symmetric operator such that
(A - z) - 1 belongs to B (H) for each non-real z, then A is self-adjoint. In fact it
suffices to know that (A- z)- 1 and (A z)- 1 belong to B(H) for a single non-real
z, which we may take to be for example the number i (see Remark 3.10 for further
comments on this point). This is the fundamental self-adjointness criterion:
Proposition 3.3. Let A be a symmetric operator in a Hilbert space 1-i.
(a) is self-adjoint if and only ifR(A + i) == 1-i and R(A- i) == 1-i.
(b) A is essentially self-adjoint if and only ifR(A + i) and R(A- i) are dense in 1-i.
THE METHOD OF THE CAYLEY TRANSFORM 89

PROOF. (a) The implication =? has been noted above. For the implication {=: sup-
pose that R(A i) == H. To conclude that A is self-adjoint, it suffices to show that
V(A*) C V(A) (because we have A _ A* for A symmetric). So let g E V(A*).
Since R(A i) == H, there is a vector hE V(A) such that (A*- i)g == (A- i)h. But
(A- i)h == (A*- i)h because A is symmetric, hence (A*- i)(g- h) == 0. Thus
g- hE N(A * - i) - R(A + i)j_ [see Proposition 2.19(b)]. Since R(A + i) - H,
this implies that g- h == 0. Consequently the vector g belongs to V(A) (as it is equal
to h).
(b) This proof is left as an exercise [use the result of (a) and the fact that R(B)j_-
R(B)j_ B is a closable operator]. D

Proposition 3.4. Let A be a sy1nmetric operator in 1i such that R(A + A) - H for


some A E JR. Then A is self-adjoint.

PROOF. The proof is a simple adaptation of the preceding one [forgE V(A *),choose
hE V(A) such that (A*+ A)g == (A+ A)h to obtain that (A*+ A)(g h) - 0]. D

3.1.2. If A is a closed symmetric operator that is not self-adjoint (or a symmetric


operator that is not essentially self-adjoint) then, by (3.3) and Proposition 3.3, at least
one of the numbers -i and +i is an eigenvalue of its adjoint A*. The multiplicities of
these eigenvalues, i.e. the dimensions of the two subspaces R(A i)j_ and R(A+i)j_,
should be a measure for the degree of non-self-adjointness of A. These multiplicities
are called the deficiency indices v+ and v_ of A:

v+ - dimN(A * + i) dim R(A i)j_, v_ == dimN(A *- i) dim R(A + i)j_.


(3.4)
In terms of the deficiency indices, Proposition 3.3 can be rephrased as follows.

Corollary 3.5. (a) A closed symmetric operator is self-adjoint if and only if v+ -


v_- 0.
(b) A symmetric operator is essentially self-adjoint if and only (f v+ == v_ == 0.

To simplify the notations we set M+ == R(A -i) and M == R(A+i). We assume


A to be closed so that M+ and M_ are subspaces. The operator A -i gives a bijection
between V(A) and M+, whereas A+ i is a bijection between V(A) and M_. If
f E V(A), one has I (A+ i)fll == I (A i)fll as a consequence of (3.2). It follows
that the mapping W: (A+ i)f f-----7 (A i)f [f E V(A)] defines a bijection between
M_ and M+ and that one has IIWgil == 11911 for each gEM_. The mapping W is
given by
W ==(A- i)(A + i)- 1 (3.5)

and is called the Cayley transform of the symmetric operator A. W is a unitary op-
erator when viewed as an operator from M_ toM+ or, if one defines n E B(H) by
Og == Wg if gEM_ and Og == 0 if g _L M_, then n is a partial isometry with initial
set M_ and final set M+ [see Proposition 2.9(e)]. By virtue of Proposition 3.3, n is
90 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

unitary if and only if A is self-adjoint.


recover from its Cayley transform W, observe that for f E V(A):

(I+ W)(A + i)f ==(A+ i)f +(A- i)f == 2Aj (3.6)

and
(I- lV)(A + i)f ==(A+ i)f- (A- i)f- 2if. (3.7)
Eq. (3.7) implies that V(A) == (I- W)M . Let us show that the operator (I W)
is invertible on M : If g E M_ there exists f E V(A) such that g - (A + i)f. If
(I-lV)g == 0, then(3.7)impliesthatf == Oandconsequentlyg == (A+i)f- 0. This
proves that (I - W) is invertible on M _. Its inverse (I - W) - l is defined on V (A)
and n1aps V(A) onto ./Vi , and by (3.7) one has
1
(A+i)f == 2i(I- W) j VjEV(A). (3.8)

By inserting this relation into (3.6) one obtains terms ofW:

= ~(I+ W)(A + i)f = i(I + W)(I- W)- 1 f iff E V(A). (3.9)

For each closed sy1n1netric extension of A one can repeat the preceding considera-
tions. us denote by A such an extension and by W == (A- i)(A + i) 1 its Cayley
transform, and let us write M+ == R(A - i) and M_ == R(A + i). Since A is an
extension of A [hence V(A) C V(A) and Af - Af for f E V(A)], it is clear that
R(A- i) C R(A- i), i.e. M+ C M+. Thus M+ is a subspace of H contained in
M+, and we can write M+ == M+ E9 N+, where N+ is the orthogonal complement
of M+ M+ (i.e. N+ == M+ n [M+ ]j_ ). Then each f EM+ has a unique decompo-
sition into f == f 1 + f2 with f 1 EM+ and f 2 EN+· In the same way one finds that
C M_, and one 1nay write M_ - M_ E9 N_.
Let us have a closer look at the structure of W relative to the above orthogo-
nal sum decompositions. operator W: M_ - t M+ is unitary. If f E M ,
then (A+ i)- f == (A+ i) 1f E D(A), so that Wf == (A- i)(A + i)- 1f ==
1

- ~i) + i) 1 f == lV f: On M_ the operators Wand l¥ coincide. The difference


between them resides in the fact that W is defined on a larger subspace than W. The ac-
tion of Won N_ is as follows: the restriction of W toN_ is a unitary operator U from
onto ; indeed, if f1 EM_ and !2 EN_, then (Wj1, Wj2) == (j1,f2) == 0,
hence W f 2 must be orthogonal to the subspace W M_ W M_ - M+, i.e. W
InapsN_ intoN+. As R(W) == M+ == M+ EBN+, WmustmapN_ antoN+.
We summarise the situation (see also Figure 3.1). The property that A is a closed
symmetric extension of is reflected in the Cayley transforms W of A and W of A by
thefactthatD(W)- V(W)ffiN_ == M ffiN_, whereN_ is a subspace orthogonal
to _,and the restriction of W to gives a unitary mapping U from N_ onto a
subspace N + orthogonal to M + (in particular dim N _ = dim N +):

on M ffiN_. (3.10)
THE METHOD OF THE CAYLEY TRANSFORM 91

V(A)
W = (A- i)(A + i)- 1
W (A i)(A + i)- 1

Figure 3.1 Cayley transform of a symmetric operator A and of a sym1netric extension A of A.

Evidently, once a closed symmetric extension A of A has been chosen, the three
quantities N_, N+, and U are uniquely deter1nined. What is important is the fact that
given A, then by an arbitrary choice of (i) a subspace N_ in [M ]_L, (ii) a subspace
N+ in [M+ ]_L with dim N+ = dimN_, and (iii) a unitary mapping U from N_ onto
N+, one determines a unique closed symmetric extension A of having the property
that its Cayley transform is W == W E9 U as in (3.10). The set of all closed symmetric
extensions of A can thus be parametrised by all possible choices of the three quantities
N_, N+ and U, as indicated. We now verify this claim and then point out some im-
portant consequences. Examples are considered in Sections 3.2 and 3.3.
Verification. Let A be a closed symmetric operator and W its Cayley transform. Let
N_ andN+ be subspaces of H withN_ _L M_, N+ M+, dimN_ = di1nN+, and
let U: N_ -+ N+ be a unitary operator. Define M_ == M ffiN_, M+ == M+ EBN+
and let W: M_ - t M+ be given by (3.1 0).
(i) Let us show that I - W is invertible on M_. Suppose that g E M_ is such that
(I- W)g == 0. Iff EM_, one obtains by taking into account the isometry of W that
(g,(I- W)f) == (g,f)- (g,Wf) == (Wg,Wf) (g,Wf) ==-((I W)g,Wf) ==
-(O,Wf) - 0. Thus g is orthogonal to the set {(I- W)f f EM_}, hence in j

particular to the set {(I- W)f If EM_} == D(A) [see (3.7)]. Since D(A) is dense
in H, one obtains from Proposition 1.6 that g - 0.
(ii) Define A == i(I + W)(I- W)- 1 , with domain D(A) == (I- W)M_. It is
evident that A is an extension of A. Let us then check that A is a closed symmetric
operator. Iff, g belong to D(A), there exist fo and g0 in M_ such that f - (I- W)fo
92 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

and g == (I- W)g 0 . Clearly one has Af- i(I + W)fo and Ag ==- i(I + W)go, and
therefore (by using the isometry of W):

(f, Ag) == i( (I- W) fo, (I+ W)go) == i [(fo, W go) - (Wfo, go)]
((I+ W)fo, (I- W)go) - (Af, g).

This shows that A is symmetric. To see that it is closed, we apply Lemma 3.1 (c) with
B ==A+ i; it suffices to show that R(A + i) is closed (i.e. a subspace). Now iff and
fo are as above, one has (A+ i)f i(I + W)fo + i(I W)fo == 2ifo. Since fo may
be any vector in M_, this shows that R(A + i) - M_, which is a subspace. Let us
also observe that (A i)f - i(I + W)fo - i(I W)fo 2iWfo, and since W is
unitary as an operator from M_ toM+, it follows that R(A- i) == M+.
(iii) Let again A - i(I + W)(I- w)- 1 . Denote by W' == (A- i)(A + i)-I
its Cayley transform. Let us show that W' W. Remember that W' is the mapping
(A+ i) f r-----7 (A i) f for f ED( A). With the expressions obtained above for (A+ i) f
and (A- i)f, it follows that W' is the mapping 2ifo r-----7 2iWfo, i.e. fo r-----7 Wfo for
fo EM_. This shows that W' is identical with W. D
We now give some important consequences of the preceding results, namely ( 1) the
von Neumann Formulas characterising the adjoint and the symmetric extensions of a
sy1nmetric operator A, and (2) a description of the different types of symmetric ex-
tensions of A in terms of its deficiency indices.

Proposition 3.6 (First von Neumann Formula). Let A be a closed symmetric oper-
ator. Then fED( A*) if and only if

f-g+h++h- (3.11)

with g E D(A), h+ EN(A* + i) and h_ EN(A* i), and

A *f - Ag- ih+ + ih_. (3.12)

Furthennore the deco1nposition (3.11) is unique (g, h+ and h_ are uniquely deter-
lnined by f).

PROOF. (i) It is clear that each vector of the form (3.11) belongs to D(A *),because
D(A), N(A * + i) and N(A *- i) are all contained in D(A *).
(ii) Let f E D(A *). By the Projection Theorem one can write (A* + i)f - !I +
f 2 with !I E M_ and f 2 j_ M_. Then there exists a vector g E D(A) such that
!I == (A+ i)g. We seth_ == -if2/2. One has A*g == Ag and A*J2 == if2 [since
(M-)j_ _ R(A+i)j_ ==N(A* -i), seeProposition2.19(b)],henceA*h_ == !2/2.
So

A*(f-g-h_)== [!I+!2-ij] -Ag-!2/2


- [(A+i)g+f2 -if] -Ag !2/2
== -i(f- g)+ !2/2 == -i(f- g- h_).
THE METHOD OF THE CAYLEY TRANSFORM 93

Thus (f g h_) E N(A * + i) and we have obtained (3.11) with h+ f- g- h_.


Eq. (3.12) is now evident.
(iii) To obtain the uniqueness of the decomposition (3.11) we n1ust show that, if
g + h+ + h_ 0 with g E D(A), h+ E N(A * + i) and h_ E N(A *- i), then g ==
h+- h_ == Now if g+h++h- == 0, thenA*(g+h++h-) Ag-ih++ih_ == 0,
and by combining these two equations one obtains

(A+i)g == (ih+ -ih_) -i(h+ +h-)- -2ih_ and (A-i)g- 2ih+. (3.13)

The first equation in (3.13) shows that h_ E R(A + i) - N(A*- i)j_. Since h_ E
N(A* i) by assumption, one hash_ E N(A*- i) n N(A*- i)j_ == {0}. So
h_ == 0. In the same way one obtains that h+ == 0 by using the second equation in
(3.13). Finally g == -h+- h_ == 0. D

Proposition 3.7 (Second von Neumann Formula). Let A be a closed symmetric op-
erator and let A be the closed symmetric extension o.f A determined by three quantities
N_, N+ and U as before. Then f E D(A) if and only if
f == g +h--Uh (3.14)

with g E D(A) and hEN_, and then

Af == Ag + ih + iUh. (3.15)

Furthermore the decomposition (3.14) is unique.

PROOF. Iff E V(A), there is a unique fo EM_ such that f == (I- W)fo, and
fo has a unique decomposition into fo !1 + !2 with !1 EM_ and !2 EN_. So
f == (I- W)JI +(I- U)j2. We take g- (I- W)f 1, which belongs to V(A), and
h - !2. Then Af == i(I + W)fo == i(I + W)f1 + i(I + U)J2 == Ag + ih + iUh.
The uniqueness of the decomposition (3 .14) follows from Proposition 3.6 since A* is
an extension of A. D

Deficiency indices and maximal symmetric extensions


(a) If A is a closed symmetric operator with M+ == 1{ or M_ == 1{, i.e. if one of
the deficiency indices of A is zero, then A has no proper symmetric extension (if A
is a symmetric extension of A, then A is equal to A); indeed one then necessarily has
N+ - {0} or N_ == {0}. Such an operator is called a maximal symmetric operator.
The following two possibilities may occur:
(i) v+ == v_ == 0. In this case A is self-adjoint,
(ii) v+ == 0, v_ #- 0 or v+ #- 0, v_ == 0. The operator A is symmetric but not self-
adjoint and has no self-adjoint extension.
(b) If M+ #- 1{ and M_ #- 1{ but dim [M+]j_ =dim [M-Jj_, the deficiency in-
dices of A are equal: v+ == v_ #- 0. Again one must consider two possibilities:
(i) The deficiency indices of A are finite (0 #- v+ == v_ < oo). In this case all closed
maximal symmetric extensions of A are self-adjoint. The number of self-adjoint exten-
sions of A is infinite (and uncountable). Indeed, a maximal symmetric extension A is
94 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

obtained by taking N_ == [M_ Jj_, N+ == [M+ ]j_ [so that R(A+i) == R(A-i) == 1{]
and by choosing an arbitrary isometric operator U from [M_ ]j_ to [M+ ]j_ (U is auto-
matically unitary because dim [M+ ]j_ =dim [M_ ]j_ < oo). Already if 7/+ == v_ == 1
the number of such operators U is uncountable: Fix a vector e_ in [M_ ]j_ and avec-
tor e+ in [M+]j_ with lle-11 - lie+ II == 1 and set Ue_ == eu~e+, with a E [0, 21r). For
each value of a one obtains a self-adjoint extension of A, and these extensions are
all different [because their Cayley transforms are different, see for example (3.10)]. If
v+ == l/_ v > 1, the self-adjoint extensions of A are in one-to-one correspondence
with the elements U of the group of all unitary v x v matrices; this group is described
by v 2 continuous parameters.
(ii) The deficiency indices of A are infinite: v + - v _ - oo. In this case A has self-
adjoint extensions as well as non-self-adjoint maxin1al symmetric extensions (the num-
ber of self-adjoint extensions is again uncountably infinite). To obtain a self-adjoint
extension one takes N_ == [M-]j_, N+ == [M+]j_ and chooses for U an arbitrary
unitary operator from N_ onto N+ (remember that dim [M+ ]j_ ==dim [M- ]j_ == oo
by assumption). On the other hand one may takeN_ == [M-]j_, choose for N+
an arbitrary infinite-dimensional strict subspace of [M+]j_ (thus N+ -/= [M+]j_,
di1nN+ == oo) and take for U an arbitrary unitary operator from N_ onto N+. The
associated extension is symmetric but not self-adjoint; if one denotes its deficiency
indices by n+ and n_, then one has n_ == 0, n+ -1- 0 (n+ ==dim [M+ EB N+]j_,
and by choosing N+ appropriately one may obtain any value n+ E {1, 2, 3, ... , oo} ).
It is clear that one can in a similar way obtain maximal symmetric extensions with
n+ == 0 and arbitrary n_ > 0.
(c) If the deficiency indices of A are different (v + -/= v _ ), then at least one of them
is finite. In this case A has no selj~adjoint extension at all. A special case has already
been discussed in (a,ii) (one deficiency index == 0, the other one -/= 0). If v_ -/= 0,
v + -1- 0 but v_ -1- v+, consider for example the case in which v+ < v_ (one then nec-
essarily has 7/+ < oo). The maximal symmetric extensions of A are obtained by taking
N+ == [M+]j_ and N_ a subspace of [M-Jj_ with dimN_ == dimN+ == v+. It is
clear thatN_ is strictly smaller than [M- ]j_, hence the deficiency indices n+ and n_
of the obtained extension A are n+ == 0, n_ == v_ - v+. Since v_ - v+ > 0, the
operator A is not self-adjoint.

3 . 1.3. We now describe a situation in which a symmetric operator always has self-
adjoint extensions. A mapping K: 1{ ---+ 1{ is called a conjugation if it has the fol-
lowing properties:
(1) K is anti-linear, i.e. K(af +g)== aKf +Kg iff~ g E 1-i and a E C,
(2) f{ 2 == I,
(3) \Kf.Kg) == (g,J)- (f,g) Vj,gEH.

Example 3 . 8 . Let 1-i- £ 2 (0, m;). Then the mapping f ~ f, i.e. f(x) ~ f(x) for
each 1; E 0, is a conjugation.
METHOD OF CAYLEY TRANSFORM 95

Let A be a symmetric operator. Assume that there exists a conjuga-


that maps V(A) into V(A) and commutes with A, i.e. such that KV(A) C
D(A) == KAJ for each f E D(A). Then the deficiency indices of A are
equal; particular A has se~f-adjoint extensions.

setN+ == R(A- i)_L andN_ == R(A + i)_L.


E D(A). By the relations K 2 == I and KV(A) C V(A) we have
h - h == (Kh) E KV(A). This shows that D(A) _ KD(A). Hence we have
KD(A) ==
fEN_ and g E D(A). Then [using and (1), then (3)]:

\Kf, (A- i)Kg) == \Kf, (A+ i)g) ==\(A+ i)g,f) == 0.


Since {Kg I g E D(A)} == V(A) by (i), this shows that Kf is orthogonal toM+,
i.e. Kf EN+. Thus mapsN_ intoN+. Similarly one shows that K mapsN+
N_.
(iii) Iff EN_, then f == K 2 f == K(K f). Since Kf EN+ by (ii), this shows
that is surjective as a mapping from N+ toN_ (K maps N+ onto N_). By (3)
preserves the orthogonality of vectors and their norm; so if {e 17 } is an orthonormal
basis of N +, then {Ken} is an orthonormal basis of N _. This shows that dim N _ =
dimN+. D

Remark 3.10. If A is a linear operator, one defines its domain of regularity G(A) as
the set of all complex numbers z for which there exists a number rc( z) > 0 such that
II(A z).fll > rc(z)\1!11 for all f E (A). One shows that G(A) is an open subset of
CC. Then one associates with each z E G(A) a deficiency subspace of A at the point
z, namely R(A- z)_L, and shows that the dimension of this subspace is constant (i.e.
independent of z) in each connected subset of G(A) (see for example Section 8.1 of
[Wll for the proof). If A is symmetric, one sees from (3.2) that G(A) contains all
complex numbers z == A + iJ-L with 11 #- 0; hence the open upper half-plane (J-L > 0)
and the open lower half-plane (p; < 0) belong to G (A), and in each of these half-
planes the dimension of R( A - z) _L is independent of z and equal to the deficiency
index v + and v _ respectively. This has the following consequences:
(a) Our choice of z == ±i for defining the deficiency indices has no particular
significance, one could take any pair of complex numbers {z+, z_} with ~z+ > 0
and ~z- < 0.
(b) In the self-adjointness criterion of Proposition 3.3 one may replace R( A - i)
by R(A - z ), where z is any complex number with ~z #- 0, provided that R(A + i)
is simultaneously replaced by R(A- z).
(c) Assume that the domain of regularity G (A) of a symmetric operator A contains
a real point. Then G(A) is connected, hence v+ == v_. Such an operator does have
self-adjoint extensions [see point (b) in the discussion at the end of §3.1.2].
(d) Let A be a closed symmetric operator. If v+ == 0, then R(A - z) == IH for
all z with ~z > 0, hence the upper half-plane belongs to the resolvent set p(A). On
the other hand, if v+ > 0, the entire upper half-plane belongs to the spectrum of
A. Similar results hold for v_ and points z in the lower half-plane. This leads to the
96 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

following characterisation of the spectrum of a closed sy1nmetric operator A (recalling


that the spectru1n is closed as a subset of the complex plane C):
(1) if v+ == v_ - 0: O"(A) is contained in JR. (A is self-adjoint),
(2) if v+ == 0, v_ > 0: O"(A) is the closed lower half-plane {z E C I ~z < 0},
(3) if v+ > 0, v_ == 0: O"(A) is the closed upper half-plane,
(4) if v+ > 0 and v_ > 0: p(A) is empty and O"(A) is the whole plane.

3.2 Differential operators with constant coefficients

3.2.1. We begin with some preliminary results. Let us consider a differential expres-
sion [, of the form

(3.16)

where p 0 , ... , p N are functions of the real variable x, called the coefficients of the
differential expression £. By applying this expression to functions f (x) one can as-
sociate to it linear differential operators in function spaces, in particular in L 2 ( J),
where J C JR. is an interval. One can choose various domains for these differential op-
erators, but the elements of such domains should contain in their equivalence class a
function that is at least N times differentiable. One can similarly consider differential
expressions involving partial derivatives (n variables, n > 2) and associate with them
differential operators in L 2 (W), where W is an open set in JR.n.
Suppose that one has associated with the expression (3 .16) a differential operator
A in L 2 ( J), with domain V(A), on some interval J == (a, b). By partial integration
one obtains for f, g E 'D(A) (assuming that the coefficients Pk are sufficiently differ-
entiable):

where the expression PN (b) - PN (a) contains the boundary terms at x == b and at
x - a that appear in the course of the integrations by parts. In order for A to be a
symmetric operator, this expression must vanish and one must have
N k N k
2)-l)k~
dx
[Pk(x)f(x)] = L_pk(x)~J(x)
dx
\lfE'D(A). (3.17)
k=O k=O

It is easy to avoid the boundary terms by taking for the domain of A only functions
f: (a, b) - t C such that j, f', f", ... ,j(N-l) are zero at x ==a and at x == b, or even
more simply V(A) == C0 ((a, b)) [then each f E 'D(A) vanishes in some neighbour-
hood of a and in some neighbourhood of b]. The condition (3 .17) is a condition on
DIFFERENTIAL OPERATORS WITH CONSTANT COEFFICIENTS 97

the functions p 0 , ... , p N that we shall not specify here in more detail. If this condition
is satisfied, one speaks of a self-adjoint differential expression; this does not imply
that A is a self-adjoint or essentially self-adjoint operator if V( A) is chosen such that
PN (b) - PN (a) - 0. The operator A will then be symmetric and may possibly have
self-adjoint extensions, as we shall see. In order to be cautious we use the terminol-
ogy formally self-adjoint differential expression rather than "self-adjoint differential
expression".
In the case of a differential expression in several variables, one proceeds analo-
gously; expression will be formally self-adjoint if the coefficients satisfy certain
conditions (example: an expression of the form - ~ V (x) if V is a real function),
and if one chooses as domain of for example V(A) == C 0 (W) if 1{ == L 2 (W)
[then each function f in V(A) vanishes in some neighbourhood of the boundary of
the set W C ffi. 71 ] .
If [, is a formally self-adjoint differential expression, defined for x in an open
subset W of ffi. or ffi. 71 , the differential operator in £ 2 (W) with domain C0 (W) is
often called the minimal operator associated with £. We shall denote it by A 0 . 2
Sometimes one uses the term maximal operator associated with £ for the adjoint
of the minimal operator, and we denote this operator by A+, i.e. A+ == A 0. If£ ==
L~=O p k(x) (dI dx) k as in (3 .16), the maximal operator can be interpreted as follows:
an element f of L 2 (W) belongs to the domain of A+ if and only if the distribu-
tion L~=O (dl dx )k [Pk (x) f (x)] is a function and this function belongs to L 2 (W), and
then A+f == L~= 0 (dldx)k [Pk(x)f(x)], where the derivatives are taken in the sense
of distributions. We recall that a distribution is a continuous linear mapping from a
function space [here C0 (W)] to C. If T is a distribution, one denotes by T (f) its
value for the function f. The derivative T' of T is again a distribution defined by
T' (f) - - T(f'), and by iteration one has T(k) (f) == ( 1)kT(f(k) ). The condition
that f belongs to the domain of A 0 is (f, A 0 g) == (f*, g) for so1ne f* E L 2 (W) and
all g E C 0 (W), and one has A 0f == f*; so by definition f* is the distribution associ-
ated with L~=O (dl dx )k [Pk (x) f (x)] [interpreted the sense of distributions, i.e.
derivatives are taken in the sense of distributions; the signs ( -1 )k appearing in (3.17)
are incorporated into the definition of the distributional derivatives].
In what follows we restrict ourselves to the cases N == 1 and N == 2 which are
by far the most important ones in quantum mechanics (in particular the mo1nentum
operator for N == 1, non-relativistic Hamiltonians-~+ V(x) or [-tV- A(x)] 2
V(x) for N == 2, called Schrodinger operators). The case N > 2 can treated
similarly. Let us write the condition of formal self-adjointness for N == 1 N == 2:
N == 1: The differential expression£== PI (x)(dldx) +p0 (x) is formally self-adjoint
if PI is purely imaginary, i.e. PI (X) == i¢( X) with ¢(X) real, and if Po (X) + riq/ (X) ==
p 0 (x), i.e. ~p 0 (x) == q/(x)l2. A special case is obtained by taking for¢ a constant
function, i.e. ¢( x) == r E ffi. for each x; then p 0 ( x) is real. Thus: if £ has the
£ - ir( dI dx) + p( x), where p is a real function and r a real constant, then £ is a

2 Some authors use the term "minimal operator" for the closure of Ao or for the restriction of the
1naximal operator to functions having compact support in W.
SYMMETRIC OPERATORS AND THEIR EXTENSIONS

formally self-adjoint differential expression.


Here it is convenient to write £ in a more symmetrical form:

£ == (dldx)q2(x)(dldx) q1 (:E)(dldx) + qo(:r).


is easy to check (integration by parts) that this differential expression is formally
self-adjoint if q2 (x) is real, q1 ( x) == i¢( x) for a real function ¢ and :Sq0 ( x) ==
q/ (x) particular, if we consider again a constant function ¢, then £ is a for-
Inally self-adjoint differential expression if it is of the form£ == (dI dx) q(x) (dI dx)
ir(dldx) + V(1;), where q and V are real functions and r is a real constant.
Below we shall study the above two types of operators, restricting ourselves to
p(x) == 0 the case N == 1) and to q(x) == 1 (in the case N == 2). The latter class
== 2) contains non-relativistic Schrodinger Hamiltonians in one dimension and
Hamiltonians occurring in the radial Schrodinger equation for three-dimensional quan-
tuin systems with spherically symmetric potentials. We treat here the simplest case
which 1 == 0 and V == 0. The case in which V #- 0 and Hamiltonians in dimension
n > with spherically symmetric potentials will be discussed in Section 3.3. Suf-
ficient conditions for self-adjointness with potentials without spherical symmetry in
dimension n > 1 have been obtained in §2.7.3.

3 . 2 . 2 . We begin with some preliminary considerations on differentiability properties


of functions defined on an open interval J == (a, b). function g : J ---+ C is said to be
absolutely continuous it is an indefinite integral (in the sense of Lebesgue). Thus g
is absolutely continuous if it can be expressed the fonn g( x) == J~ro h( x )dx + const.,
with fixed xo E J and with h: J ---+ C a locally integrable function [which means
that hE £ 1 (-J) for each bounded closed subinterval -J of J]. is clear that absolutely
continuous functions g are continuous on J. Moreover it can be shown that they are
differentiable almost everywhere in J, with g' == h (ahnost everywhere). Of course
each function of class C 1 on J is absolutely continuous on this interval. On the other
exist continuous functions that are differentiable almost everywhere but
which are not absolutely continuous, i.e. not equal to the indefinite integral of their
derivative (an exa1nple given by the Cantor function described in Example 4.5).
An absolutely continuous function g may be diverging at the endpoints of the in-
because its derivative g' is assumed to be only locally integrable on J. Such
divergence cannot occur if g' is globally integrable on J, i.e. if g is an absolutely
continuous function with g' h E £ 1 ( J). Indeed, if I hi is integrable over J, then
g has boundary values at the end points of this interval: g (a) lim x ~a g (x) and
3
g (b) _ linl :r /b g (:r) exist. We shall use the fact that, for functions of this type, the in-
tegration by parts formula is valid: Iff and g are absolutely continuous on J - (a, b),
and g' in L 1 ( J), then

1 b.f(x )g' (:r) d.r + /bf' (x )g(x) dx = .f(b )g(b) - f( a)g( a). (3.18)

3
For exan1ple, if b is finite, then g(b - c) - g(b - 5) = Jbb~; h(y )dy --+ 0 as 5, c --+ +0, so that
linlx/b g(2~) exists.
DIFFERENTIAL OPERATORS WITH CONSTANT COEFFICIENTS 99

Details on the preceding results are given in numerous textbooks, see for example
§23- §26 of [RN] or Section 18 of [HS].
In the theory of ordinary differential operators one often encounters continuous
functions g with square-integrable derivative g' _ h, i.e. satisfying h E L 2 ( J). Re-
calling from (1.49) that h E L 2 ( J) implies that h E L 1 (J) for each finite subinterval
J of J, it follows that such a function g is absolutely continuous on J. In this context
the following fact will be important:
Proposition Let g be an absolutely continuous function on J == (a, b) such that
2
g' E L (J).
(a) If a is finite, then the boundary value g( a) exists, and similarly for the endpoint b.
(b) If J is an infinite interval and g itself is also square-integrable on J, then g( x) --t 0
as x tends to an infinite endpoint of J (e.g. as x --t +oo if b == +oo ).

PROOF. (a) If a is finite and c E (a, b), then the assumption made on g' implies that
g1 E L 2 ((a, c)) C L 1 ((a, c)), see (1.49). Thus g(a) exists.
(b) We treat the case b == +oo. One has (d/dx)!g(x)j 2 == g(x)g'(x) g'(x)g(x).
Hence, for any c E (a, b):

The integrals on the right-hand side have limits as x --t oo, because g and g' belong to
£ 2 ( [c, oo)) and these integrals can be interpreted as scalar products. This shows that
the limit of jg(x)l 2 as .r:c --t oo exists. Since lg(x)j 2 is integrable at infinity, this limit
must be zero. D

In our discussion of ordinary differential operators we use the following terminol-


ogy and convention: An element f of L 2 ( J) is said to be absolutely continuous if
its equivalence class contains an absolutely continuous function, and in our equations
involving f it is understood that we work with this continuous representative. For
example, if f and g are absolutely continuous elements of £ 2 ( J) such that f' and g'
belong to £ 2 ( J), then the integrals in (3 .18) may be interpreted as scalar products
between elements of £ 2 ( J) (they are independent of the chosen representatives in the
equivalence classes off and g), but for a large class of representatives the right-hand
side will not make sense.
With the preceding convention in mind we introduce an operator B in £ 2 ( J) as
follows:

V(B) = {g E L 2 (J) 13 hE L 2 (J) such that g(x) =fox h(y)dy + g(O)}, (3.19)

and B g == -ih == -ig' [we assume that the point x == 0 belongs to J or is an endpoint
of J, so that g(O) is defined]. The elements of V(B) are the equivalence classes of the
(absolutely continuous) functions specified on the right-hand side of (3.19), and by
Proposition 3.11 their boundary values g( a) and g(b) are well defined (with g( a) == 0
if a == - oo, g (b) == 0 if b == + oo).
100 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

3. 2.3. We now discuss the differential expression £ - -i ( d / dx) which formally cor-
responds to the momentum operator. We consider this expression on IR, on a semi-
infinite interval [which we take to be IR+ == (0, oo)] and on a finite interval (a, b) (we
shall set a == 0 and b == 1). So let J == IR, IR+ or (0, 1) and denote by C 0 (J) the set
of all complex functions of class CCX) on J each of which vanishes in some neighbour-
hood of each endpoint of J. The minimal operator Po is given by V(Po) == C 0 (J)
and P0 f == -if' for f E C 0 (J). We first show that the adjoint of P0 (i.e. the max-
operator) coincides with the operator B introduced above, which gives a useful
description of the domain of P 0 [namely the domain introduced in (3.19)].

Proposition 3.12. The operator B is the adjoint of the minimal operator: B == P 0.


Hence B coincides with the maximal operator P+, and the closure of the minimal
operator is the adjoint of B, i.e. Po == B*.
PROOF. The last assertion follows from the first one because the closure of an opera-
tor Cis equal to C** (Proposition 2.22). So let us determine the adjoint P 0 of P0 • An
element g of £ 2 ( J) belongs to V( P 0) if and only if there exists a vector g* E £ 2 ( J)
such that (g, -if') == (g*, f) for all f E C 0 (J). By interpreting g as a distribution
on the space of test functions C0 (~!), this means that the derivative (in the distribu-
tional sense) of g is equal to ig*, hence that this derivative is a function. 4 One can
then define an absolutely continuous function G on J by G(x) ==fox g*(y)dy. The
function g* is locally of class £I, hence G is bounded on each finite subinterval J of
J and thus belongs to £ 2 (-J). For f E Co(J) one obtains upon integrating by parts
that (g*, f) - \G', f) == -\ G, f') (the boundary terms are zero because f vanishes in
some neighbourhood of each endpoint of J). It follows that ig + G is orthogonal to the
set of functions {f' I f E Co(J) }. We shall see below that this implies that ig + G
is (equivalent to) a constant function, i.e. g ( x) == iG (x) + con st. a. e. in J. Since J is
arbitrary, this shows that the elements of V( P 0) are absolutely continuous functions
with square-integrable derivative, and a square-integrable absolutely continuous func-
tion on J belongs to V (P 0) if its derivative (which will be equal to g * in the preceding
notations) belongs to L 2 (J). Then one will have P 0g == g* == -ig'. Thus P 0 ==B.
To see that ig is constant on each finite subinterval J == (c, d) of J, we de-
note by g0 the restriction of i g + G to J, more precisely g0 ( x) == i g (x) + G (x)
if x E J and g0 (x) == 0 otherwise. g0 belongs to L 2 (J) and we may assume that
go 0 (otherwise there is nothing to prove). Observe now that if cp E C 0 (J) is such
that fed cp(x )dx == 0, then the function {) defined as ?3(x) _ fcx cp(y )dy belongs to
C 0 (J) and satisfies 19'(x) == cp(x). Hence we have (cp,g0 ) == 0. Thus, by setting
VI== {hE Co(J) fcdh(x)dx == 0}, we have (h,g 0 ) == 0 Vh E VI. Let us fix a
j

function e E C 0 (J) such that fd e(x) - 1 and observe that, for any f E Co(J), the
d c
function f- [ ,[c f(x)dx] e belongs to VI. Hence

(.[,go)- ([1~f(x)dx]e,go)- 0 Vf E Co(J).

4
The function g determines a distribution T by T(f) = (g, f); its distributional derivative T' is given
by T' (f) = -(g, f') := (ig, -if') = (ig*, f).
DIFFERENTIAL OPERATORS WITH CONSTANT COEFFICIENTS 101

This equation may be rewritten as

ld f(x) [go(x) - (e, go)] dx = 0 VjECo(J).

Since C 0 (J) is dense in L 2 (J), one must have g0 - (e, g 0 ) == 0 (see Proposition
1.6). Thus g0 (x) == (e, g 0 ) for almost all x E J. D

Knowing the domain of the n1aximal operator, we can now determine the deficiency
indices of minimal operator P 0 . If g E R(Po =t= i)_L == N(B ± i) C V(B), then g is
absolutely continuous and satisfies -ig' ± ig == 0, i.e. g' == ±g. Since g is absolutely
continuous, this implies that g' has the same property. If g'' denotes the derivative of
g', we have g'' == ±g'. Thus g 11 is absolutely continuous, and by iterating this argu-
ment one finds that g of class CCXJ. It is easy to calculate g: the general solution of
g' == g is given by .9+ (x) == c+ and that of g' == -g by .9- (x) == c_ , where
c+ and c_ are arbitrary complex constants. One sees that the deficiency indices can
assume only the values 1 or 0 [depending on whether g+ and g_ belong to L 2 (J) or
not]. Let us consider separately the cases J == IR, J == IR+ and J == (0, 1).
(1) J == IR: One has g+ ~ L 2 (IR) and g_ ~ L 2 (IR) (assuming that C± # 0). Thus the
deficiency indices are v+ == v _ == 0. The minimal operator [the operator - i ( dJdx)
with domain C 0 (IR)] is essentially self-adjoint [see Corollary 3.5(b)]. Recall that we
have already established the essential self-adjointness of -i( d/ dx) on S(IR) in §2.5.2
by a different method.
(2) J == IR+ == (0, oo): One has g+ ~ L 2 (J) but g_ E L 2 (J), and the deficiency in-
dices are v + == 0, v _ == 1. The closure of the minimal operator is a maximal symmet-
ric operator that is not self-adjoint. The differential expression -i( d/ dx) determines
no self-adjoint operator in £ 2 (IR+).
(3) J == (0, 1): Here .9+ E L 2 (J) and .9- E L 2 (J). The deficiency indices of P0 are
v+ == v_ == 1. Every closed maximal symmetric extension of Po is self-adjoint. The
differential expression -i (d / dx) determines an uncountable family of self-adjoint op-
erators in £ 2 ( (0, 1)).
To characterise the self-adjoint extensions of P0 that exist in case ( 1) and in case
(3), it is useful to know the domain of the closure Po == B* of the minimal operator
Po.
Proposition 3.13. The domain of the closure P 0 of the minimal operator is as follows.
An element f of £ 2 ( J) belongs to V( P 0 ) if and only if it has the .following properties:
( 1) f is absolutely continuous,
(2) f' E £ 2 ( J),
(3) if J # IR, then: (3a) if J ~ (0, oo): f(O) ~ 0
(3b) if J ~ (0, 1): f(O) ~ j(1) == 0.
Iff E V(Po), then Pof == -if'.
PROOF. (i) If J == IR, the closure of Po is equal to B because Po is essentially self-
adjoint. Thus the elements of V( P0 ) are absolutely continuous functions with deriva-
tive in L 2 (IR) and converging to zero as x-+ ±oo [see Proposition 3.ll(b)].
102 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

(ii) In the other two cases let us denote by E 0 the restriction of the operator E to the
domain V(Eo) consisting of all f E L 2 (J) having the properties (1), (2) and (3) of the
proposition [remember that the elements of V(E) are characterised by (1) and (2) and
that we use the convention made at the end of §3.2.2]. We must show that E 0 == Po.
We use the following identity which is a special case of (3.18) (integration by parts):
If j,g E V(E) and J ==(a, b), then

\Bf,g)- \J,Bg) ==\-if', g)- \f, -ig') == i[f(b)g(b)- f(a)g(a)]. (3.20)

If J- (0, oo) or J == (0, 1) we have a== 0 and b == oo orb== 1. Iff E V(E 0 ), the
condition (3) ensures that f(a) f(O) == 0; if b == oo we have f(b) f(oo) == 0 by
Proposition 3.11, and if b == 1 we have f(b) == 0 by the assumption (3). Thus (3.20)
implies that

\Eof, g) - \J, Po g) == o \ljEV(Eo), VgEV(Po) V(E).

This shows that the adjoint P 0* _ Po of P 0 is an extension of E 0 : B 0 C P0 .


To obtain the inclusion P0 C E 0 , let us fix f E V(Po) - V(P0*). Since E is an
extension of P 0* one has

\Bf,g)- \f,Eg)- \P()*j,g)- \f,Pog) == 0 \lg E D(P0). (3.21)

Let us choose a particular g E D(P0) V(E): we take an infinitely differentiable


function g: [0, oo) ---+ JR.+ such that g(O) == 1 and g( x) == 0 for x > 1. As g(b) == 0
(forb == 1 as well as forb == oo), Eq. (3.20) becomes [for this particular g and for
f E V(Po)]:
\Bf, g) - \f, Eg) == -if(O)g(O) == -if(O).
Comparison with (3.21) shows that one must have f(O) == 0. Thus the elements of
V(P0 ) satisfy f(O) == 0. This proves the inclusion P0 C E 0 if J == (0, oo).
In the case J == (0, 1) we still have to show that f(1) == 0 iff E D(Po). This is
done by repeating the argument given above, by choosing the particular function g
such that g(O) == 0 and g(1) == 1. D

We can now specify the self-adjoint extensions of the minimal operator.


J == JR.: We have already seen that the minimal operator is essentially self-adjoint. Its
unique self-adjoint extension is the momentum operator P. In configuration space, i.e.
in L 2 (JR.), one has (P f)(x) == -if'(x) and the domain of Pis the set of absolutely
continuous f: JR. ---+ CC such that f and f' belong to L 2 (JR.). In momentum space, i.e.
in l 2 (JR.), P is the operator of multiplication by the variable: (:FP f) (k) == k j (k),
and its don1ain in this representation is the set of (equivalence classes of) measurable
functions g j: JR. ---+ CC satisfying J 1 + k 2 ) g (k) 2 dk < oo.
-CX)CX) ( I 1

J == (0, 1): Let P be a self-adjoint extension of P0 (hence of P0 ). The operator P is the


restriction of P 0 to a linear submanifold ofD(P0) D(E). If we take in (3.20) f ==
g E V(P), the expression on the right-hand side (which is i[lf(1) 2 -lf(O) 2 ]) must be
1 1

zero because Pis assumed to be self-adjoint. Iff E D(P0 ), one has f(O) == f(1) == 0,
DIFFERENTIAL OPERATORS WITH CONSTANT COEFFICIENTS 103

hence condition lf(1)j 2 - !f(O)j 2 == 0 is automatically satisfied. Iff E D(IP) but


f ~ V(Po), one have f(O) -I 0 or f(1) -I 0; since !!(1)1 2 -lf(O)I 2 == 0, one then
has in fact f(O) 0 and f(1) -I 0, and there exists a number a E [0, 27r) such that
f(1) == e""af(O). priori a depends on f: a== a(f). By taking in (3.20) two vectors
f-Ig belonging to D(IP) \ V(P0 ), one 1nust have
f(1)g( - f(O)g(O) == [e-""a(f)eia(g)- 1] j(O)g(O) == 0,

consequently a(f) == a(g). Thus the nun1ber a is a constant depending only on the
self-adjoint extension IP. If we denote by IP a the restriction of P 0 to the domain

V (IP a) == { f E V (Po) \ f ( 1) == e 1af (0)}, (3.22)

then we have shown that IP a is an extension of IP: IP IP a· Now IP a is symmetric [the


last expression in (3.20) is zero if j, g E D(IP a)L hence IP a is a symmetric extension
of the self-adjoint operator IP. This is possible only if IP a == P.
These considerations show that any self-adjoint extension of the minimal operator
Po must be equal to one of the operators IP a' a E [0, 2Jr ). In fact each IP a is a self-
adjoint extension of Po because each P a is a self-adjoint operator: if g E D(IP~ ), then
(3 .20) implies that one tnust have

f(1)g(l)- j(O)g(O) == [e-iag(1)- g(O)] f(O) == 0


By taking f such that f(O) 0, one sees that one must have g ) == e""a g(O), hence
g E D(P a)· An interpretation of the operators Pa is given below, and sotne additional
properties of these operators can be found in §3 .2.5.
We have thus found all self-adjoint extensions of the operator Po in L 2 ( ( 0, 1)).
Let us indicate how the same result follows from von Neumann's formulas (3.14) and
(3.15). If IP is a self-adjoint extension of P0 , one has f E 'D(IP) if and only iff is of
the form f == g + h- Uh with g E V(Po), hE N(P0 - i) and U: N(P0 - i) ---+
N(P0 + i) unitary. Now N(P0 - i) is the one-dimensional subspace spanned by g_,
hence h(x) == f;;C-x with /<l; varying over C. Sin1ilarly N(P0 + i) == { I /<l; E C};
so Uh U(f;;e-x) is a multiple of , say U(f;;e-x) == f;;+ , and, since U must be
unitary, one must have I f;;C-x I == I /<i;+ex II· By calculating these norms [in L 2 ( ( 0, 1))]
one sees that this condition is equivalent to 1/<i;+ I == 1/<i;l/ e. So, if we set /<l;+ == {3/<l;/ e
with 1!31 == 1, we have
f E D(IP) <====? f(x) == g(x) + /<i;C-x- {3/<i;e:c-l a.e., with g E V(Po). (3.23)

(3.15) shows that [Pf](x) == [P0 g+ih+iUh](x) == -ig'(x) +i/<i;e-x +if3/<i;ex-I,


so that, by (3.23), IP f == -if'. Hence IP is given by -i( d/ dx ). Finally let us consider
the relation between f (0) and f ( 1) if f is of the form (3 .23 ). By using the fact that
g(O) == g(1) == 0 one obtains
/{; j3 /{; /<i;
!(1) == e- (3/<l; == -e(1- {3e) '
/{;
f(O) == /<i;--
e
== -(e- {3).
e

Hence f(1) == (1-;3e)(e-{3)- 1f(O) == -{3({3-e)({3-e)- 1 f(O) eu~f(O), because


I - {3 ({3 - e) (;3 - e) - l I == 1 (the value of a can be determined in terms of that of {3).
104 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

The spectrum of the operators JP a is quite easy to determine (see Problem 3.3). It
is an interesting fact that each real number ,\ is an eigenvalue of some JP a. Let us also
point out an interpretation of these operators. Since each JP a acts in the same way as
Po on functions in L 2 ( ( 0, 1)) having support away from 0 and 1, it is clear that one
must consider what happens at the endpoints of the interval J == (0, 1). Let us recall
from quantum mechanics that, by exponentiation, the momentum operator generates
translations in configuration space: in £ 2 (JR), the operators {exp( -iP0 a)} aEJR give
a unitary representation of the translation group, i.e. [exp( -iP0 a)f](x) == f(x- a)
(this will be discussed in detail in Section 5.1; see in particular Example 5.4). So one
expects that, iff has co1npact support in (0, 1) and if a is sufficiently small [so that the
set {x a}xEsuppf is still contained in (0, 1)], one will have [exp( -i1Paa)f](x) ==
f (x - a). To understand the situation when these conditions are not satisfied, let us
take for fan eigenvector of JP a, i.e. f(x) == exp(iAnx) with An == 21rn +a (n E Z).
Then [exp( -i1Paa)f](x) == [exp( -iAna)f](x) == exp(iAn(x- a)), with x E [0, 1]. It
follows that, for 0 < a < 1:

f(x-a) if x- a> 0
[exp( -iJP a a) f] (x) == { .
e-""af(x- a+ 1) if x- a< 0

Thus, for 0 < a < 1, the action of exp( -iJP a a) corresponds to a translation by a
to the right in such a way that the part of the translated function falling above the
endpoint 1 gets re-injected into J at the other endpoint 0, by acquiring an additional
phase factor . Similarly, for -1 < a < 0, exp( -iJP a a) acts as translation to the
left with re-injection at the right endpoint 1 with an additional phase factor e+ia. The
action of exp( -iJP a a) for Ia I > 1 is then obtained by using the group property of the
fa1nily {exp( -iJP aa) }aEJR·
The above considerations also shed some light on the fact that on the interval (0, oo)
the sy1nmetric minimal operator has no self-adjoint extension. If we consider the case
where a < 0 (translation to the left), then on a finite interval the part of a function
f that gets translated outside Jon the left is re-injected (with a change of phase) at
the right endpoint, thus preserving the norm of function (unitarity); on the interval
(Oi oo) re-injection at the right endpoint, i.e. at x == is clearly not possible.

~J We now consider differential operators of the second order and begin with the
......... -. . . . .

simplest case£ == -d 2 / dx 2 . For1nally this differential expression corresponds to the


Ha1niltonian of a free particle (the kinetic energy operator); as for the momentum
operator one encounters problems when one tries to associate with it a self-adjoint
operator in £ 2 ( J) when J is a finite or semi-infinite interval. We denote the minimal
operator by Ko: V(Ko) == C 0 (J), (Kof)(x) == - f"(x), and we shall again take
J == ( 0, 1), J == ( 0, oo) or J == JR. The minus sign in £ is clearly irrelevant for the
question of self-adjointness, it has been inserted so as to work with the usual expres-
sion of quantum mechanics.
The determination of the extensions of K 0 is carried out in a very similar way
to that e1nployed for the 1nomentun1 operator. We shall first characterise the maxi-
mal operator (the adjoint of the minimal operator K 0 ) by proceeding in analogy with
DIFFERENTIAL OPERATORS WITH CONSTANT COEFFICIENTS 105

Proposition 3.12. Then we indicate the self-adjoint extensions of K 0 and some of their
properties.

Proposition The domain D(K0) of the maximal operator K 0 is given as follows


(modulo equivalence ojfunctions):
g E D(K0) ~ g is continuously d(fferentiable, its derivative g' is an absolutely
continuous function and g" belongs to L 2 (J). If g E D(K0), then (K 0g) == -g".

PROOF. proof is similar to that of Proposition 3 .12. An elernent g of £ 2 ( (J) be-


longs to D(K0) if and only if there exists g* E L 2 (~J) such that (g,- f") == (g*, f)
for all f E C 0 (J). Thus the second derivative (in the sense of distributions) of g is
equal to-g*, so this second derivative is a function (which is locally of class £ 1 ). One
can then define a function G of class C 1 with absolutely continuous derivative G' by
G (x) == fox dy J~ g * ( z) dz. This function G is bounded on each finite subinterval J of
J and thus belongs to L 2 (J). For f E C0 (J) one obtains after integration by parts
(observe that the boundary terms are zero): (g*, f) == (G", f) == (G, f"). Hence
g + G is orthogonal to the set of functions {!" If E C 0 (J) }. We shall show below
that this implies that (g + G) (x) == c 1 + c2 x for certain constants c 1 and c2 and
for x E J, so that g(x) == -G(x) + c1 + c2 x a.e. in J. Since J is arbitrary, this
shows that the elements of D(K0) are functions of class C 1 with absolutely contin-
uous derivatives, and a function on J with these properties belongs to V(I<o) if its
second derivative belongs to L 2 ( J). It is clear that one will have (using the above no-
tations): K 0g == g* == G" == -g".
To see that g + G is a linear function on each finite subinterval J == (c, d) of J, we
denote by g0 the restriction of g + G to J, more precisely: g0 ( x) == g ( x) + G ( x)
x E J and g0 ( x) == 0 otherwise. g0 belongs to L 2 ( J), and we know that (g 0 , f' ') == 0
for all f E Co(J). Let us define V2 by

For cp E D2 and x E (c, d) we set ¢(x) == fcx dyfcy cp(z)dz. Clearly¢" == cp, and¢
belongs to C0 (J): (i) ¢(x) == 0 in some neighbourhood of x == c because cp has this
property, (ii) upon integrating by parts (with respect toy) one obtains

¢(x)- j c xdy jy
c tp(z)dz = x
jx jx
c tp(z)dz- c ytp(y)dy,

and for x so close to d that cp(y) == 0 for y > x, each term on the right-hand side is
zero because cp E V 2 . Thus we have (g 0 , cp) _ (g 0 , ¢") == 0. So (g 0 ,h) == 0 for all
hE V 2 . We now fix two functions e 1 , e 2 E C0 (J) such that (see the example given at
the end of the proof)
SYMMETRIC OPERATORS AND THEIR EXTENSIONS

that, for any f E Cc)(J), f- [J~d j(x)dx]e1 - [fcdxf(x)dx]e2 belongs to

(go,J)- [1 d f(:r:)dx] (go,el)- [


id xf(x)dx] (go,e2)
c = 0 VjECo(J).

equation can be rewritten as follows:

VjECo(J).

Since (J) denseinL 2(J), onemusthaveg0 (x) == (e 1,g0 )+(e 2,g0 )xforalmost
all .r E J.
We finally indicate how to obtain and e2 0 < c <
d. Choose a function f) as indicated the figure, i.e. such
fJ > 0 on (c, ru), f) < 0 on(~'{;, d) for some/'{; E (c, d)
dfJ (x) dx == Clearly a :== Jdc xfJ(x) < 0 and
/3 ·- d x- 1fJ( :c) d;r; > 0. One can then take c2 (x)
a- 1f) (X) and e l (~r) == {3 - l ~c - l e(X) . D

characterise the self-adjoint extensions of K 0 , the following remark is useful: If


gE 0 ) in addition to g", its first derivative g' also belongs to L 2 ( J). This
will be discussed in relation with Proposition 3.16. By taking into account Propo-
3. 1, one then finds that, g E V(K0), then boundary values of g and of g'
at endpoints of J exist, and that these boundary values are zero at - oo and at +oo
J extends to -oo or to
us now deter1nine the deficiency indices of the operator I<0 . functions in
0 i) satisfy the differential equation - f'' if == 0, i.e. f'' == ±if. Two
linearly independent solutions are as follows:
f" ==if: !1(x) == exp[(1 + i)x/v/2], !2(x)- exp[-(1 + i)xjv/2]
!"==-if: j3(x) == exp[(1- i)x/v/2], j4(x) == exp[-(1- i)xjv/2].
of these four functions is square-integrable on (0, 1), and none of them belongs
(JR). For J - (0, oo) one has f 2, f 4 E L 2( (0, oo)) and f 1, f 3 ~ L 2( (0, oo) ). So
deficiency indices of the minimal operator are as follows:
(1) u+ == u_ == 0, i.e. K 0 is essentially self-adjoint on C 0 (JR).
J == (0, oo ): u+ == 7/_ == 1. K 0 has a one-paratneter family of self-adjoint exten-
Sions.
J == (0~ 1): u+ == u_ == 2. K 0 has a four-parameter family of self-adjoint exten-

case (1) the do1nain of the unique self-adjoint extension of K 0 , i.e. the domain of
the closure J{0 of K 0 , is equal to V (K 0) given by Proposition 3.14 with J == JR. the
other two cases one obtains the domains of different self-adjoint extensions by restrict-
ing the functions in V(I< 0) to suitable linear submanifolds. The following identity is
DIFFERENTIAL OPERATORS WITH CONSTANT COEFFICIENTS 107

easily verified by twice integrating by parts: if j, g E V(K 0) and J == (a, b), then
(K0j,g)- (j,K 0g) -(j",g) + (j,g")
== f(b)g'(b)- f'(b)g(b)- f(a)g'(a) + f'(a)g(a). (3.24)

To obtain a self-adjoint extension of K 0 , one must choose a linear submanifold of


V(K 0) in such a way that the right-hand side of (3.24) is zero for f and g in this
submanifold; this expression is given in terms of the values of j, f', g and g' at the
endpoints a and b of J, so the choice of a suitable submanifold is reduced to imposing
boundary conditions at x ==a and at x == b. We have mentioned that, if a== -CXJ, then
each function in V(K0) vanishes at x == a; more precisely f(x) --+ 0 as x --+ -CXJ.
Similarly, if b == +x and f E V(K0), then f(x) --+ 0 as x --+ +x. This explains
the fact that, if J == IR, then the right-hand side of (3.24) is always zero, that
K 0 is self-adjoint. If J == (0, CXJ ), then the right-hand side of (3.24) is - f(O)g' (0) +
f'(O)g(O), and this expression is zero one imposes the boundary condition h' (0) ==
j3h(O) for some f3 E IR [i.e. the right-hand side of (3.24) is zero if f and g satisfy
this boundary condition at x == 0]. The case ;3 == CXJ is also possible [rewriting the
preceding boundary condition in the form h(O) == (3- 1 h' (0)]. Upon identifying ,B with
cot a we obtain:
If J - (0, CXJ ), then, for each a E [0, n), the operator K(a) defined by I<(a)g == -g",
with domain
V(K(a)) == {g E £ 2 ((0, x)) I g is continuously differentiable, g' is absolutely
continuous, g" E L 2 ( (0, CXJ)) and g' (0) sin a == g(O) cos a}
is a self-adjoint extension of K 0 , and K 0 has no other self-adjoint extensions. The
self-adjoint extension for a == 0, characterised by the boundary condition g(O) == 0,
is the most important one in quantum mechanics (see §3.3.3).
If J == (0, 1), one must impose two boundary conditions, involving the values of the
functions in V(K 0) at x == 0 and at x == 1. One can impose conditions analogous
to those discussed above at each of the points x - 0 and x == 1 (one then speaks
of separated boundary conditions), or one can impose mixed boundary conditions
involving at the same time the values of these functions at x == 0 and at x - 1. This
leads to the following self-adjoint extensions of K 0 :
Wtih separated boundary conditions: K(;3, r)g == -g" with domain
V(K((3, r)) == {g E L 2 ( (0, 1)) I g is continuously differentiable, g' is absolutely con-
tinuous, g" E L 2 ( (0, ), g' (0) sin f3 == g(O) cos (3 and g' (1) sin r == g(l) cos r}.
Here f3, r E [0, 1r). One speaks of Dirichlet conditions at :r == 0 if ;3 == 0 [this is the
condition g(O) - 0] and of Neumann conditions at x == 0 if ;3 == 1r /2 [this is the
condition g' (0) == 0]. The same terminology applies at x == 1. If f3 == r == 0 one
speaks of the Dirichlet Hamiltonian.
With mixed boundary conditions: K(P) g == -g'' with domain
V(K(P)) == {g E £ 2 ((0, 1)) I g is continuously differentiable, g' is absolutely
continuous, g" E £ 2 ((0, 1)), g(l) == eiPg(O) and g'(1) == eiPg'(O) },
where p E [0, 2n). Of special interest is the case of periodic boundary conditions ob-
tained by taking p == 0 : g (1) == g (0) and g' (1) == g' (0).
108 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

The general self-adjoint extension of K 0 in L 2 ( (0, 1)) can be characterised as fol-


lows. 5 To each vector f E V(K 0) we associate two elements f and f 1 of C 2 by setting
f == {f( ),f(O)} and f 1 == {f 1 (1),- f'(O)}. Then the right-hand side of (3.24), for
the case J == (0, 1), is just
(3.25)
Each self-adjoint extension of K 0 is characterised by the following restriction on the
boundary values f and f 1 of the elements f E D(K0):

(cosT) f- (sin T) f 1 == 0, (3.26)

where T is a self-adjoint operator C 2 (given by a Hermitian 2 x 2 matrix; the


matrices cosT and sin T are defined by the usual power series for these functions).
other terms, if the self-adjoint extension of K 0 associated with such an operator T is
denoted by KT, then V(KT) == {f E V(I<o) I the couple (f, f 1 ) satisfies (3.26) }. It
is easy to see that the expression (3.25) does vanish iff, g E V(KT ). Indeed, using
the relation Ic2 == (cos T) 2 + (sinT) 2 along with the self-adjointness of cosT and of
sin T and with the fact that these operators commute, one finds that (all scalar products
being in CC 2 )

(f,g
1
) == ((cosT)f, (cosT)g 1) + ((sinT)f, (sinT)g 1)
== ((sinT)f ~(cosT)g ) + ((sinT)f,(cosT)g)
1 1

== ((cosT)f 1 ,(sinT)g 1) + ((cosT)f,(sinT)g).


Similarly one obtains
1
(f ,fl) == ((cosT)f 1,(sinT)g 1) + ((cosT)f,(sinT)g).
The operator J{ ((3, r) defined above by separated boundary conditions corresponds to
the operator T given by the matrix ( 6 ° ).
Remark 3 . 15. In order to arrive at an interpretation of a boundary condition of the
type g' (a) sin (3 == g (a) cos (3, let us consider a time-dependent wave function in
L 2 ( (0, 1)) and write its development in the basis { exp( iknx) }nEZ, where kn == 21rn:
f(x, t) == L anei(knx-k~t).
nEZ

Let us assume that the time evolution of f is governed by the operator K ((3, r). Then,
if the derivative !, t a at t
f 1 off with respect to is square-integrable, !, t win belong
to the domain of K((3, r) for each t (see Proposition 5.1). Taking t == 0, we have

5
We refer to Appendix A.l and Appendix A.2 of [RK] or to §125 of [AG] for a detailed derivation.
DIFFERENTIAL OPERATORS WITH CONSTANT COEFFICIENTS 109

and the boundary condition of,t(x, t)jox lt=O ==(cot f3)!,t(x, t) at x == 0 reads

L (ikn)k;an == cot (3 L k;an. (3.27)


nEZ\ {0} nEZ\ {0}

A single mode (an == 0 except for one value of n) cannot satisfy this equation. How-
ever, since k_n == -kn, (3.27) can be satisfied if an and a_n are related as follows:

ikn +cot (3
an == .k (3 a_n
z n - cot

[then one has iknan + ik_na-n == cot {3( an + a_n), and of course k'?:_n == k~]. So
the boundary condition g' (0) sin (3 == g (0) cos (3 selects wave functions for which the
amplitude of each incoming mode at x == 0 (n < 0) differs from that of the corre-
sponding reflected mode (n > 0) by a phase factor (ikn +cot (3)(ikn- cot (3)- 1 .

3.2.5. symmetric operator K 0 == -d 2 jdx 2 is the square of P0 == (djdx), both


operators having the domain C0 (J). Consequently there must exist certain relations
between self-adjoint extensions of Po and those of K 0 . We specify some of these
relations below.
If J == lR, both Po and K 0 are essentially self-adjoint, and the self-adjoint closure
K o _ H 0 is the square of the momentum operator P Po.
Let us discuss the case tl == (0, 1 ). We first consider the operator JP~ JPa JP;,
where JP a is the self-adjoint extension of Po defined in (3.22). A function f belongs
to D(JP;) if and only iff E D(JP a) and f' E D(JP a), i.e. if and only iff and f' are
absolutely continuous, f' and f" belong to L 2 ((0, 1)) and satisfy f(1) == eiaf(O) and
f'(1) == eiaf'(O). Then clearly f is of class C 1 , and it is seen that V(JPc~) == D(J<(a)),
where K(a) is the operator K(P) with p == a introduced in §3.2.4. This shows that

1l1l*ITD
lLal[a- lla
== 1l1l2 == K(a) · (3.28)

Next let us consider the operator P 0Po in L 2 ((0, 1)). As above, a function f be-
longs to D(P0Po) if and only if we have f E D(P0 ) and f 1 E D(P0), i.e. if and only
iff and f' are absolutely continuous, f' and f 1' belong to L 2 ((0, 1)) and f(O) ==
f (1) == 0. This domain coincides with the domain of the operator K (0, 0) [the opera-
tor K(f3,!) ofpage 107with(3 == r == 0]. Hence 6

Po Po == K(O, 0). (3.29)

It follows from Eqs. (3.28) and (3.29) that the operators K(a) and K(O, 0) are
positive. Other self-adjoint extensions of I<0 need not have this property. However,
a self-adjoint extension of K 0 can have at most two negative eigenvalues. We refer
6
The operator P 0Po is self-adjoint. This is a particular case of a very general theore1n stating that if A
is closed and densely defined, then the operator A *A is self-adjoint (see for example §51 in [AG] or §V.3.7
in [K]). For the self-adjointness of the square of a self-adjoint operator (e.g. of JID a), see also Problem 3.12
or Chapter 4.
110 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

to Example 4.25 for a proof of this fact and for other results on the relation between
the spectra of different self-adjoint extensions of symmetric operators with finite defi-
ciency indices.
We should point out that the familiar commutation relations between the operators
P, H 0 and Q in L 2 (JR) have no analogue in L 2 ((0, 1)). Of course the operator IfDa
con11nutes with its square K(a) but not with other self-adjoint extensions of K 0 . Also
none of the operators JP a is canonically conjugate to the position operator Q (multi-
plication by the variable x) in L 2 ((0, 1)). Even though [Pa,Q]g == [P0 ,Q]g == -ig' if
g E C0 ((0, ), there are functions f and gin 'D(JP a) such that (JP af, Qg)- (Q f, IfD a9)
is different from -i(f, g) (Problem 3.11).

3.2.6. To end these considerations, let us show that the functions in 'D(K0), and also
their first derivatives, vanish at infinity if J is an infinite interval. The argument is
based on the following result which is a particular case of the fact that, iff is a function
of class em (m times continuously differentiable), then the intermediate derivatives
j(k), 1 < k < m- can be estimated in terms off and f(m).

Proposition 3.16.. Let {J == (a, b) be an interval. Let £ (J) be the set of functions
f: J ----+ C that are continuously differentiable and such that f' is absolutely contin-
uous. Then there exist constants C1 > 0 and C2 > 0 such that for all f E E ( J):

(3.30)

vvhere the value +oo is admitted for these integrals (if the .function in the integrand is
not square-integrable on J). In particular: iff E E ( J) is such that f and f" belong
to L 2 ( J), then one also has f' E L 2 ( J).

We give the proof of this proposition in the Appendix to this chapter. As an applica-
tion, let K 0 be the minimal operator associated with -d 2/ dx 2 in L 2(J) and observe
that 'D(J{0) ~ E(J). g E 'D(K0), then g and g" belong to L 2 (J), hence so does g'
by the preceding proposition. So g and g' are absolutely continuous, belong to L 2 ( J),
and their derivatives also belong to L 2 ( J); the vanishing of their boundary values at
infinity then follows from Proposition 3.11 (b).

Remark 3 . 17. If one equips the linear manifolds 'D(P0) and D(K0) with the graph
norm of the respective operator [see (2.59)], they become Hilbert spaces. These spaces
are particular cases of Sobolev spaces, denoted 1{ 1 ( J) and 1{ 2 ( J) respectively:
H 1 ( J) == {f: J ----+ C I f is absolutely continuous, f E L 2 ( J) and f' E L 2 ( J)},
with norm llfiiHl == [11!112 + 11!'112]1/2.
H 2 ( J) == {f: J ----+ C j f is continuously differentiable, f' is absolutely continuous,
f E L 2 ( J) and f" E L 2 ( J)},
2
with norm llfiiH 2 == [11111 + 11!'11 + 111''11 2]112 .
2

More generally one considers the following Sobolev spaces Hm,p form == 0, 1, 2, ...
and p E [1, oo ):
SCHRODINGER OPERATORS 111

Hm,p the set of (equivalence classes of) functions f : .J ----+ C such that f is rn - 1
times continuously differentiable, f(nt,-l) is absolutely continuous and f, f', . .. , f(ni)
belong to P( J), norm II f II Hm P == [L~~o ,[J If(k) ( x) IP dx PIP.
These spaces are Banach spaces (complete normed linear vector spaces), even Hilbert
spaces p == , the notation Hm (J) is frequently used for 7-{ 17~,, 2 ( .J). theory of
Sobolev spaces developed for example in [Ad].

3.3 operators

3. 3.1.. We now consider differential operators corresponding to Hamiltonians of non-


relativistic quantum mechanics, i.e. operators given for1nally by£- -d 2 / d.r 2 + (x)
in L 2 ( J), where J == (0, 1), (0, oo) or JR and V is a real function assumed to be
continuous on the open interval ~! [so V could diverge at the endpoints of
example at x == 0 and/or at x == oo if J == (0, oo )]. We denote by A 0 the
operator associated with£: D(Ao) == C 0 (J), [Aof](x) ==- f"(x) + (;r)f(x).
Iff E C0 (J), f is zero in some neighbourhood of each endpoint of J and V
continuous on the support of f. Hence V f is a bounded function which is zero out-
side some bounded subinterval of J (outside the support of f which is contained
some finite interval), in particular E £ 2 ( J). This shows that A 0 is well defined on
C0 (J). Since V is assumed to be real, A 0 is a symmetric operator. Co1nplex conjuga-
tion (Example 3.8) maps C 0 (J) into C 0 (J) and commutes with A 0 ; by Proposition
3.9 the deficiency indices of A 0 are equal: v+ == v_. We shall see that pos-
sible values are 0, 1 and 2; thus each closed n1aximal symmetric extension A0
self-adjoint. We begin by specifying domain of the 1naximal operator A~:

Proposition 3 . 18 . The domain V(A 0) o.f the 1naxilnal operator A~ is as .follows


(modulo equivalence o.ffunctions):
(a) g E D(A~) -¢:==:=} g is continuously differentiable, the derivative g' is an absolutely
continuous function, and the function -g" + V g belongs to ( J).
(b) ff g E V(A~), then g" E L (J) .for each subinterval J o.f Jon which the .function
2

V is bounded, in particular for each subinterval o.f the fonn J == (a, b) with a and b
in the interior of J (hence d(fferentfro1n the endpoints of J). ForgE D(A~) one has
A~g == -g" + Vg.
PROOF. Let g E D( A~) and let J be an open subinterval of J such that I (x) I < l'vf <
oo for all x E J. Denote by g0 the restriction of g to J, more precisely g0 ( ;r) == g (x)
if x E J and g0 ( x) == 0 otherwise. operator of multiplication by in £ 2 ( J) is
bounded, hence Vg 0 E £ 2 (J). In particular one has for each f E C (J): 0
(f, A~g) == (Aof, g) == ( - !" + V f, g) == ( - f", go) + (f, V go),
I.e.

(- f", go) (Kof, go) = (!, AQg- V go/ - / f(x) [(AQg )(x) - V (x )g(:r:)] dx.
JJ
(3.31)
112 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

Since (A 0g)(x) - V(x)g(x) is square-integrable on J, (3.31) shows that g0 belongs


to the domain of the adjoint K 0 of the minimal operator I<o associated with -d 2 / dx 2
in £ 2 ( J). By Proposition 3.14 (applied in J) this implies that g is continuously differ-
entiable on J, g' is absolutely continuous on J, g" belongs to L 2 ( J) and ( A 0g) (x) -
V(x)g(.T) == -g"(x) for x E J. In particular the restriction to J of -g" + Vg belongs
to L 2 (J).
Since these considerations apply in each subinterval J == (a, b) of J with a and
b different from the endpoints of J, we have (A 0g)(x) == -g"(x) + V(x)g(x) al-
lnost everywhere in J. Since A 0g belongs by assumption to L 2 ( J), we conclude that
-g" + Vg E L 2 (J). D
We have seen that, if g E V(A 0), then -g" + Vg belongs to L 2 (J) and both -g 11
and V g belong to L 2 ( J) if J lies in the interior of J (so - g11 and V g are square-
integrable except possibly in a neighbourhood of the endpoints of J). If Vis bounded
on the entire interval J, then Vg E L 2 (J) and consequently also -g 11 E L 2 (J).
V is unbounded near at least one endpoint of J, then in general - g 11 and V g will
not belong separately to L 2 ( J), only their sum will be in L 2 ( J). V is bounded in
a neighbourhood of one endpoint of J, then V g E L 2 (J) for each subinterval J of
J not containing the other endpoint. Some further local properties of the functions
in the domain of A 0 are described in the next proposition; the term "local" means
that one considers these functions in a subinterval of J or that one localises them by
multiplying them by a smooth function f) which vanishes outside some subinterval of
J (a result of this type is given in part (c); this result will be used later, and we refrain
from presenting a general discussion of the method of localisation).
Proposition 3.19 . Let£ == -d 2 jdx 2 + V(x) be a.formally self-adjoint differential
expression on an open interval J. Let A 0 be the minimal operator associated with £
zn ( J), hence V(A 0 ) == C0 (J), and let A 0 be the adjoint of A 0 in L 2 ( J).
(a) Let J be an open subinterval of J. Denote by B 0 the minilnal operator associated
vvith in L 2 (J), V(Bo) == C 0 (J), and denote by B 0 the adjoint of B 0 in L 2 (J). If
g E V(A{)), then the restriction of g to J belongs to V(B 0).
(b) ~f J is an open subinterval o.f J on which the function V is bounded and if
g E V(A 0), then the restriction of g to J belongs to V(K0), where K 0 is the ad-
JOlnt L 2 (J) o.fthe minin1al operator K 0 associatedvvith -d 2 /dx 2 in L 2 (J) [hence
V(Ko) == Co(J) and Kof ==- !"].
Let e: J --+ c be a .function o.f class C 2 such that fJ( X) == 1 in some neighbour-
hood o.f one of the endpoints of J, and fJ(x) == 0 in some neighbourhood of the other
endpoint. If g E V(A 0), then Bg also belongs to V(A 0).
PROOF. (a) is an immediate consequence of Proposition 3.18(a) (applied in J and in
(b) has been obtained in the proof of Proposition 3.18.
(c) Suppose for example that f) ( x) == 0 in some neighbourhood of b and f) ( x) == 1
near a. One has (Bg Y == fJg 1 + f)' g and (Bg )' 1 == ()n g + 2() 1g1 + fJg 11 • Since () is of class
and g belongs to V(A 0), one sees that Bg is continuously differentiable and that
(Bg)' is absolutely continuous. Also -(fJg)' 1 + VfJg == f) [-g 11 + Vg]- f) 11g- 2fJ 1g1 •
Clearly f) [ -g' 1 + V g] and fJ 11g belong to £ 2 ( J) (because g E V(A 0), see Proposition
SCHRODINGER OPERATORS 113

3.18(a), and because e and e" are bounded). e' is bounded, e' (x) == 0 near Q, and
near b, hence e' is non-zero only on a finite subinterval in the interior of tJ. Since g'
is absolutely continuous, it is bounded on the support of e'' hence e' g' is a bounded
e'
function with support in a finite interval, so that g' E L 2 ( J). Summing up, one has
-(eg)" + Veg E L 2 (J), hence eg E D(A 0) by Proposition 3.18(a). 0

For j, g E D(A 0) and x E J we set

[j, .9] x == j 1 (X) g (X) -- j (X) g' (X) · (3.32)

By Proposition 3.18(a) this expression is well defined for each x E J, and [f, g ]x
is continuous in x. So if J == (a, b) with -oo < a < b < +oo, then [f, g ]x is a
continuous function on (a, b). Let us show that [f, g ]a:== limx'\.a [f, g ]x and [f, g ]b :==
limx/b[f, g ]x exist. For this observe that, if a < A < {L < b, then (using Proposition
3.18(b) and the hypothesis that V is real, and integrating by parts):

1/J- [f(x) (AO g) (x) ~ (A0f)(x )g(x)] dx = 1/J- [~ f(x )g" (x) f" (x )g(x)] dx

= [ ~ f (X) g' (X) + f I (X) g (X)] I: = [f' g l!J- ~ [f' g l A .

The integrand of the first integral belongs to ( (a, b)) [scalar product between func-
2
tions of L ( (a, b))]. In fact, as A - t a and M - t b, the left-hand side converges to
(f,A 0g)- (A 0j, g) [scalar product in L 2 (J)!]. Thus the right-hand side must have
limits as A - t a and/or M - t b, and one has

\f, A ag) - \Aof, g) == [f, g ]b - [f, g ]a V j, g E D(A 0). (3.33)

We observe that (3.24) is a special case of this identity.

Remark 3.20. Below we shall use the following fact concerning solutions of differ-
ential equations (see e.g. Section 2.1 of [Br] or page 150 of Volume II of [RS]): Let
(-X, M) be an interval and W: (-X, M) - t C a continuous function. Denote by x the
set of solutions h of the differential equation -h"(x) + W(x)h(x) == 0, x E (-X, M).
Then x is a vector space of dimension 2 (each solution is a linear combination of two
fixed linearly independent solutions) and the functions h E x are of class C 2 [twice
continuously differentiable] on (-X, M).

We return to the situation considered in this section: J == (a, b) is a (finite or infi-


nite) interval and V: J - t JR. a continuous function which could diverge at a and/or at
b. Let c E (a, b) be an interior point of J, and consider the subinterval (a, c] [or simi-
larly [c, b)] of J. V is continuous on (a, c) and bounded on (a', c] for each a' E (a, c].
Let x be the two-dimensional vector space formed by the solutions of the differential
equation -h"(x) + V(x)h(x) == ih(x) on (a, c) [take W(x) == V(x)- i in Remark
3.20]. Each element of x is continuous on (a, c) and has a finite limit as x - t c [in
fact h is of class C 2 on (a, c] and even has a continuation of class C 2 to (a, c + 8) for
sufficient! y small 8 > 0]. Thus the elements of x are square-integrable on (a', c) for
114 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

a' E (a, c). As a, the norm in £ 2 ( (a/, c)) of an element h of x will either
---+
re1nain bounded or diverge~ in the first case h belongs to L 2 ( (a, c)), in the second case
one has h ~ L 2 ( (a, c)). is a very important fact that there exists at least one function
h 0 in x that is square-integrable on (a, c):
Alternative). Let (a, b) be an interval, V: (a, b) ---+ JR
continuous and c E (a, b). Then either all solutions of -h"(x) + V(x)h(x) == ih(x)
belong to L 2 ((a. c)) or there exists exactly one (up to a nun1ericaljactor) solution
h 0 in L 2 ( (a, c)).
Following Weyl, one says that Vis in the limit circle case at a if all solutions of
(x) (x) h ( x) == i h( x) are in £ 2 ( (a, c)) and that Vis in the limit point case at
a if only one (up to a numerical factor) solution of this differential equation belongs
to ( (a, c)). One applies the same terminology at the other endpoint b, replacing
L ( (a, c)) by L 2 ( ( c, b)) [Weyl's alternative also holds in £ 2 ( ( c, b)), the proof is com-
2

pletely analogous]. We also observe that any solution h will have the property that h"
a continuous function on (a, b) because we have assumed V to be continuous on
interval. comment on the origin of the limit point-limit circle terminology can
be found Prob le1n 3. 1

OF PROPOSITION 3.21. It suffices to show that there exists at least one hE


((a, c)) (not identically zero) satisfying -h"(x) + (x)h(x) == ih(x). For this
we denote by the minimal operator associated with the differential expression
2
+ V (x) in ( (a, c)) and by B 0its adjoint in £ 2 ( (a, c)). We first show
that * is not sym1netric. For this we choose two functions j, g: (a, c] ---+ C of class
in son1e neighbourhood x == a and satisfy f (c) == g' (c) == 0 and
(c) == g(r) == 1. Clearly f and g belong to D(B 0) [see Proposition 3.18(a)]. Hence
applied in L 2 ((a. r)), gives
(f, Bog) - (Bof, g) == [f, g ]c- [f, g ]a == 1,
since [f. g ]a == 0. This shows that B 0 is not symmetric.
Since B 0 not symmetric, B 0 is not essentially self-adjoint (Proposition 2.23).
by Proposition 3.3, either R(B 0 + i) or R(Bo - i) is not dense in H, or
equivalentlyeitherN(B 0-i) {0} or N(B 0+i) #- {0}. IfN(B 0-i) #- {0}, then
any h 0 in N(B 0- i) satisfies -h" (x) + V(x )h(x) == ih(x ). If N(B 0+ i) #- {0 },
let g 0 be any element of N(B 0+i); then g satisfies -g" (x )+ V (x )g(x) == -ig(x ),
hence its complex conjugate h g is a non-zero solution of -h"(x) + V(x)h(x) ==
ih(x). 0
Proposition Let A 0 be the n1inimal operator associated with£ == -d 2 / dx 2 +
(,r) on an interval J == (a, b), with V: J - t JR continuous. Denote the pair of de-
ficiency indices o_f Ao (v +, v _ ). Then
The deficiency indices o_f Ao are (2, 2) if and only if V is in the limit circle case at
a and at b.
deficiency indices qf Ao are ( 1, 1) if and only if V is in the limit circle case at
one endpoint o_f J (at a or at b) and in the linzit point case at the other endpoint.
is essentially self-adjoint if and only if V is in the lin~it point case at a and at b.
SCHRODINGER OPERATORS 115

PROOF. We use the fact that A 0 has equal deficiency indices. In (i) we prove the "if"
part for all three situations and in (ii) the "only if" part.
(i) V is in the limit circle case at a and at b, then each solution of A 0h == i h
belongs to L 2 (J). Thus N(A 0 - i) is two-dimensional, i.e. the deficiency indices of
Ao are v + == v _ == 2. If we assume for example that V is in the limit point case at a
and in the limit circle case at b, then there exists exactly one (up to a numerical factor)
non-zero solution of A 0h == ih satisfying h E £ 2 ( (a, c)) if c < b. Since V is in the
limit circle case at b, this solution belongs to L 2 ( ( c, b)), hence also to L 2 ( (a, b)). Thus
dimN(A 0- i) == 1, i.e. v+ == v_ == 1. Finally assume that is in the limit point case
at both endpoints of J. This case is rather subtle. At each endpoint of J there exists
exactly one (up to a numerical factor) solution h ¢. 0 of A 0h == ih that is square-
integrable near the respective endpoint. If one denotes by ha the solution which is
square-integrable near a [so ha ¢- 0 and ha E L 2 ((a,c)) for each c E (a, b)] and if
one continues this solution to all of J, it could a priori coincide with a multiple of the
solution which is square-integrable in a neighbourhood of b. Part (c) of the proposition
asserts that this is never the case (under the assumption that is continuous on .J). To
see this one uses the fact (proved in the following lemma) that [j, g ]o == [f, g ]b == 0
for all j, g E V(A 0) if Vis in the limit point case at a and at b. Eq. (3.33) then implies
that the maximal operator A 0 is symmetric, and the essential self-adjointness of A 0
follows from Proposition 2.23.
(ii) If A 0 has deficiency indices (2, 2), then each solution of -h"(x)+ V(x)h(x) ==
ih(x) belongs to L 2 ((a, b)), hence to L 2 ((a, c)) as well as to L 2 ((c, b)) for any c E
(a, b). Thus Vis in the limit circle case at both endpoints of ~1, and (a) is completely
proved. It follows that if the deficiency indices of A 0 are ( 1, 1) or (0, 0), then must
be in the limit point case at least at one endpoint of J, say at a. If the deficiency indices
are (0, 0) then we must have the limit point case also at b [otherwise, by the "if" part
of (b), A 0 would have deficiency indices ( 1, 1), i.e. a contradiction]. Similarly, if the
deficiency indices are (1, 1), one must have the limit circle case at b [otherwise, by the
"if" part of (c), A 0 would have deficiency indices (0, 0), i.e. a contradiction]. 0

Lemma 3 . 23 . Let J == (a, b). Let£== -d 2 jdx 2 + V(:r) be a.formally selj~adjoint


differential expression on J, and let A 0 be the minilnal operator associated with £
in L 2 (J).
(a) Let z E C and let f, g be two solutions o.f Lcp == zcp, and let W[f, g] == f (x )g' (x)-
f' (x )g(x) be their Wronskian at x E J. Then W[f, g] is a constant [i.e. independent o.f
x E (a, b)]. Furthennore W[f, g] # 0 if and only ~f f and g are linearly independent.
(b) ~f f E V(Ao) and g E D(Ao ), then [f, g ]a == [f, g ]b == 0.
(c) IfV is in the limit point case at a (or at b), then [f, g]a == 0 (or [f, g]b == 0) for
all j, g E D(A 0).

PROOF. (a) We first verify that f(x)g'(x)- f'(x)g(x) does not depend on :r:

d~ [f(x)g'(x)- f'(x)g(x)] = j(:1:)g"(x)- .f"(x)g(x)


== j (X) [(V (X) - Z) g (X)] - [(V (X) - Z) j (X)] g (:r) == 0.
16 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

Next assume that W[f, g] == 0 and for example that g is not identically zero. Choose
a number c E (a, b) such that g(c) # 0 and set p == f(c)/ g(c) so that f(c) == pg(c).
Since W[f, g] == f(c)g'(c)- f'(c)g(c) == 0 one also has f'(c) == pg'(c). Since the so-
lution <p of an ordinary differential equation of second order is uniquely determined by
prescribing the values of <p and cp' at one point, and since pg is a solution of L<p == z<p
with initial conditions cp( c) == f (c) == pg( c), <p 1 (c) == f' (c) == pg' (c), one must have
f(x) == pg(x) on (a, b). Thus, if W[f, g] == 0, then f and g are linearly dependent.
The converse is easy to obtain.
(b) Let e: J --7 JR be of class C 2 and such that e( X) == 1 in some neighbour-
hood of a and B(x) == 0 near b, and set g0 == Bg. Then g0 E V(A 0) by Proposition
3.19(c). Since go vanishes in some neighbourhood of b, we have [f, go]b == 0. On the
other hand go (x) == g (x) near a, hence [f, g] a == [f, go] a. Thus, by using also the hy-
pothesis that f E V(A 0 ), the identity (3.33) (applied to the couple {f, g 0 }) becomes
[f, g]a == [f, go]a == (A'Qf, go)- (f, A() g) == (Aof, go) - (J, A() g) == 0, because Ao
is the adjoint of A 0 (Proposition 2.20). Hence [f, g ]a == 0. By interchanging the roles
of a and bin this argument one finds that [f, g ]b == 0.
(c) Let us treat the situation in which is in the limit point case at a. Fix a number
c E (a, b), set J == (a, c) and let B 0 be the minimal operator associated with the dif-
ferential expression£== -d 2 jdx 2 + V(x) in L 2 (J), as in Proposition 3.19(a). The
deficiency indices of B 0 are ( 1, 1); so, by the First von Neumann Formula (3 .11 ), the
domain V(B 0) of the adjoint of B 0 is the linear manifold spanned by V(Bo) and two
linearly independent functions belonging to V(B 0)\D(B0 ) [in (3.11) these functions
are the solutions of B 0g == ±ig, but clearly one obtains the same linear manifold if
one chooses any two linearly independent functions in V(B 0)\ D(B0 )].
Let <p1, <p2 be linearly independent solutions of L<p == i<p on (a, c). cp 1 and cp 2 are of
class C 2 ' in particular square-integrable, on each interval of the form (a'' c) with a' >
e e
a. Let be a function of class CCX) on (a, c) such that (X) == 0 in some neighbourhood
of a and B(x) == 1 near c, and set h 1 == B<p 1 and h 2 == B<p 2. Then h 1 and h 2 are linearly
independent and belong to V( B 0) [use Proposition 3 .18( a) in £ 2 ( (a, c)) by observing
that h% E L 2 ((a, c)) and that Vis bounded on the support of B]. We show below that
h 1 and h 2 do not belong to D(B 0 ), which implies that V(B 0) is the linear manifold
spanned by V(Bo) and the two functions h 1 and h 2. Thus, iff E D(B 0), there exist
a vector hE V(Bo) and constants a, j3 such that f == h + ah 1 + j3h 2, in particular
f (x:) == h (x) in so1ne neighbourhood of a. If g is another element of V ( B 0), one
will have [f, g ]a == [h, g ]a, and the last expression is 0 by the result of (b). Hence
[f, g]a == 0 for all J, g E V(B 0). Since the restriction to (a, c) of a function in V(A 0)
belongs to V(Bo) by Proposition 3.19(a), we have [f, g]a == 0 for all f, g E V(A 0).
It remains to show that h 1 , h 2 tf- V(B 0 ). Let c > 0 be such that B(x) == 1 for all
x E (c- E, c). Then £h 1 == ih 1 and £h 2 == ih 2 on (c- c, c). Assume for example
that h 2 E V(B 0 ), and denote by h 1 the complex conjugate of the function h 1 . It is
clear that h 1 E V(B 0), hence one must have [h 1 , h 2 ]c == 0 by the result of (b). But
[h1, h2]c == h~(c)h2(c)- h1(c)h;(c), hence W[h1, h2] == 0. This contradicts the fact
that W[h1, h 2] # 0 since h 1 and h 2 are linearly independent [cf. (a)]. D
SCHRODINGER OPERATORS 117

We briefly indicate how one obtains the self-adjoint extensions of A 0 in the situa-
tion in which this operator is not essentially self-adjoint. Recall Eq. (3.33):
\f,A'Qg)- \A'OJ,g) == [f,g]b- [f,g]a V f, g E V(A'Q). (3.34)
As in the case of the operator -d 2 / dx 2 discussed in §3.2.4, one has to restrict A 0 to
a linear submanifold of its domain, by imposing boundary conditions at x == a and/or
at x == b, so that the right-hand side of (3 .34) is zero.
We first consider the case in which the deficiency indices of A 0 are ( 1, 1). Assume
for example that V is in the limit circle case at a and in the limit point case at b. Then
[f, g]6 == 0 for all j, g E V(Ao ), hence it suffices to impose a boundary condition
at x - a. In §3 .2.4 the number a was finite, g (a) and g' (a) were well defined for
each g E V(K 0), and one imposed a condition of the type g'(a) == ag(a). Here it
is more convenient to formulate a condition of the form [fo, g] a == 0, where fo is a
fixed function belonging to V(A 0) but not to V(A 0 ), because [f, g]a is well defined
for any J, g E V(A 0) [we recall that, by the Second von Neumann Formula (3.14), the
domain of a self-adjoint extension of A 0 is determined by exactly one function fo of
the above type if the deficiency indices of A 0 are ( 1, 1), because dim N _ == 1 in this
case]. 7
If the deficiency indices of A 0 are (2, 2), one can impose boundary conditions of the
preceding type at a and at b, i.e. [fo, g ]a == 0 and [! 1 , g ]b == 0 with f 0 , f 1 two functions
in V(A 0)\V(A0 ), or one can impose mixed boundary conditions. We omit the details
and refer to Proposition 8.29 of [WI] for a precise formulation. The following result
is interesting: If the deficiency indices of A 0 are (2, 2), then the spectrum of each self-
adjoint extension of A 0 consists of only eigenvalues (no continuous spectrum!); each
eigenvalue is of multiplicity 1 (the subspace spanned by the eigenvectors associated
with each fixed eigenvalue is one-dimensional), and the eigenvalues form a discrete
set, i.e. their only possible accumulation points are ±oo.

3.3.2. It remains to see under which conditions a potential V is in the limit point case
or in the limit circle case at the endpoints of an interval J. The answer depends on the
behaviour of V near these endpoints, and one must distinguish between the case of a
finite endpoint and that of an infinite one. To simplify the notations we consider the
situation of an endpoint at x == 0 and that of an endpoint at x == oo (the other situations
are entirely analogous). We give for each of these two situations a simple criterion
permitting one to decide, for a class of relatively well behaved potentials, whether
they are in the limit point case or in the limit circle case. A considerable number of
finer criteria are presented in Chapter XIII of [DS]. We begin with an auxiliary result
which is useful in that it leads to a simplification when determining the deficiency
indices, in the sense that it will suffice to study the equation A 0f == 0 rather than
Aof ==if.
Lemma 3.24. Let (a, b) be an interval, V: (a, b) --+ JR. continuous and c E (a, b).
Assume that there exists a number z 0 E C such that all solutions of the differential
For example the condition g'(O) == ag(O) used in §3.2.4 is equivalent to [fo,g]o == 0 for fo(x) ==
7

exp[(l+i)x/vi2)+K:exp[(l-i)x/vi2) with K: == (1-i+ivl2a)/(l+i-ivl2a), or for fo(x) == ax+l.


118 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

equation -h"(x) + V(x)h(x) == z 0 h(x) belong to L 2 ((c, b)). Then for each z E CC,
all solutions o.l- h" (x) + V ( x) h( x) == z h( x) belong to L 2 ( ( c, b)) [and silnilarly with
£ 2 ( ( c, b)) replaced by L 2 ( (a, c))].

The proof of this letnma is given in the Appendix to this chapter. We point out the
especially useful particular case mentioned before the lemma:

Corollary 3.25. (a) Vis in the limit circle case at x == a if and only if all solutions of
-h"(x) + V(x)h(x) == 0 belong to L 2 ((a, c)) [with cE (a, b)].
(b) \l is in the limit point case at x == a if and only if there exists a solution of
-h"(x) + (x)h(x) == 0 that does not belong to L 2 ((a, c)).

We first study the deficiency indices at infinity. The next result shows that in prac-
tically all cases of physical interest one has the limit point case at infinity, hence that
it is not necessary to impose boundary conditions at infinity to define a self-adjoint
Hamiltonian.

Proposition 3.26. If (x) > -~x 2 in sonte neighbourhood of x == oo, where K, is a


real constant, then is in the limit point case at x == oo. Similarly, if V (x) > - K,X 2
in so1ne neighbourhood of x == -oo, then V is in the limit point case at x == -oo.

PROOF. Without loss of generality we can assume that r\; > 0. Let c > 0 be such that
V(x) > -Kx 2 forJ~ >c.
(i) Let us first prove the following result: If g is a real solution of J:g == 0 with
g E L 2 ((c, oo)), then :r- 1 g' belongs to L 2 ((c, oo)). For this, let us denote by 11.911 the
nonn of g in £ 2 ( ( c, oo)). Since g" == V g, we have for each y > c:

(3.35)

On the other hand, upon integrating by parts, one obtains the following expression for
the left-hand side of this inequality:

Yg"(:r)g(x) dx -_
- -- -
g'(y)g(y) _ g'(c)g(c) _ .!Y g'(x) 2dx, + 2.!Y g'(x)g(x) dx.
2
X y2 C
2
c X
2
c X
3

(3.36)
We set \II (y) == [ Icy x- 2
g' ( x) 2 dx J
112
. By the Schwarz inequality one has (by using
1/ :r 2 < 1/ c2):
1./Y g'(x;,;(x) dx I < :2[[g[[W(y).
Together with (3.36) and (3.35) this implies that

(3.37)

Suppose that x- 1 g' tf_ £ 2 ( ( c, oo)). Then \II (y) ~ oo as y ~ oo, so that there exists
Yo E ( c, oo) such that the right-hand side of (3.37) is positive for y > y 0 . Then
SCHRODINGER OPERATORS 119

g'(y)g(y) > 0 y > y0 , i.e. g' and g have the same sign on (y 0 , oo). This implies
that lg(y) I > jg(y0 ) I for y > y0 , which contradicts the hypothesis that g E Ij 2 ( ( c, oo)).
Therefore one must have x - 1 g' E £ 2 ( ( c, oo)).
(ii) us now consider two linearly independent solutions f 1 , f 2 of£ f == 0. We
may assume and f 2 to be real (the differential equation £f == 0 has two linearly
independent solutions). Let W[f 1 , f 2 ] be their Wronskian [see Lemma 3.23(a)].
One ,J2J =1- 0 and !1(x)j~(x)- f{(x)f2(x) == W[J1,J2] for each x E
(c, oo). that W[j1,f2]x- 1 t/ L 1 ((c, oo)) because ,[cCX)x- 1 d:c == oo. Now

Lf1 , f 2] x - 1 == f 1 ( x) [x - l f~ (x)] - [x- 1 f { (x)] f 2 ( :r) .


If both f 2 belong to £ 2 ( ( c, oo) ), then the right-hand side the preceding
equation belongs to £ 1 ( ( c, oo)) [by the result of (i) and the Schwarz inequality], a
contradiction. Thus at least one of these two functions is not square-integrable at in-
finity, in the limit point case at x == oo.

Example Consider the Hamiltonian of a harmonic oscillator which corresponds


to the differential expression£ == -d 2 I dx 2+ rX 2on J == JR, with r > 0. The po-
tential V (x) == rX 2 is in the limit point case at x == - oo and at :x; == +oo (for any
value of r E IR). So the Hamiltonian of a harmonic oscillator (in one dimension) is
essentially self-adjoint on C 0 (JR)!
Remark 3 . 28.. The result of the above proposition shows that a potential V is in the
limit point case at infinity if it is bounded below near infinity particular it con-
verges to zero at infinity) and even if it tends to -oo but not too rapidly. One can show
that V in the limit case at x == +oo if V(x) ~ -oo and J~CX) I (x) l- 112 dx <
8
oo for some c < oo, for example if (x) == a with 1 > 0 and a > 2. Observe
that in classical mechanics a potential of the last type [V (:r:) == a ~~ > 0
and o: > 2] corresponds to a repulsive force increasing in strength as x approaches
±oo; a classical particle submitted to such a force (a > 2) would reach .rr: == oo or
x == - oo after a finite time. For not too bizarre potentials there is a correspondence
between the classical behaviour and Weyl's Alternative: Vis in the limit point case at
infinity and only if a classical particle, with arbitrary initial condition, would require
an infinite time to arrive at infinity, and V is the limit circle case at infinity if and
only if there exist initial conditions under which a classical particle reaches infinity
after a finite time. 9
Let us now discuss the deficiency indices at x == 0. We consider the case J == (0, a)
for some a E ( 0, oo]. The next proposition concerns potentials are in a
neighbourhood of x == 0. This result can be applied to purely repulsive potentials, i.e.
satisfying V' (x) < 0 for x E (0, b) for some b > 0; the proposition shows that strongly
repulsive potentials are in the limit point case. the other hand potentials that are
only moderately repulsive (for example Coulomb potentials) are in the lin1it circle
case (Proposition 3.30). The case of attractive potentials [for example (x) > 0
x E (0, b), V(x) ~ -oo as x ~ 0] is considered in Proposition 3.30(b).
8 For a proof see [DS], Theorem XIII.6.20.
9 For a more detailed di~cussion see [RS J, Volume II, Appendix to Section X.l .
120 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

Proposition 3.29. Assu1ne that V(x) > ~x- 2 in a neighbourhood ofx 0 [i.e. for
each x E (0, b), for some b > 0]. Then Vis in the limit point case at x == 0.

PROOF. Consider first the differential equation- f"(x) + (3/4)x- 2 f(x) - 0. Two
linearly independent solutions are fi(x) == x-I/ 2 and f 2 (x) == x 3 12 . Let us fix a
number 1 > 0 and denote by fo the unique solution satisfying the initial conditions
fo(b) == 0, f6(b) == -~. fo is a linear combination of !I and !2, given by fo(x) ==
(1 /2) (b 3 12 x-II 2 - b-II 2 x 3 12 ). Observe that fo tf- L 2 ( (0, b)). Furthermore fo (x) > 0
and !6 < 0 on the interval (0, b).
To prove the proposition it suffices to exhibit a solution g of - g" (x) + V (x) g (x) ==
0 satisfying g(x) > f 0 (x) on (0, b) [then g will not belong to £ 2 ((0, b))]. We fix g by
the initial conditions g(b) == 0, g'(b) == -21. Then g will be real and we have

g(x)- fo(x) = -.ib bg'(z)dz +.ib f~(z)dz = .ib [f~(z)- g'(z)]dz, (3.38)

so that it suffices to know that f6 (x) > g' (x) for all x E ( 0, b). This inequality is satis-
fied in a neighbourhood of x == b as a consequence of the initial conditions at x == b
ilnposed on the functions fo and g. So let y be the largest number in (0, b) such that
f6(Y) == g'(y) (assuming that such a number exists). Then g(x) > f 0 (x) on (y, b] by
(3.38). By using also the hypothesis that V(x) > ~x- 2 , we obtain:

f~(y) - .f~(b) -1 bf~' (x) dx = ry -1b ~ x~ 2 .fo( x) d:x;

> -ry-1b ~x~ 2 g(:x;)dx > ry-1b V(x)g(x)dx

= -ry-1bg"(x)dx = -ry- g'(b) + g'(y) = ry + g'(y) > g'(y),

in contradiction with the hypothesis that f6 (y) == g' (y). So there is no number y
having the stated properties, hence f6(x) > g'(x) on (0, b). D

Proposition 3.30. (a) Assume that lx 2 V(x)l < {3 < 3/4 in some neighbourhood of
x == 0 [i.e. for each x E (0, b), for some b > OJ. Then V is in the limit circle case at
X== 0.

(b) If V is monotone increasing in an interval (0, b) with b > 0 [i.e. V (y) > V ( x) if
0 < x < y < b], then V is in the limit circle case at x == 0.

PROOF. We give the proof of (a) and refer to Theorem XIII.6.24 of [DS] for that of
(b). The arguments for proving (a) are similar to those of the preceding proof.
(i) One considers first the differential equation - f" (x) {3x- 2 f (x) - 0 (we can
assume that {3 =1- 0). Making the ansatz f (x) - xP one obtains the following pair of
linearlyindependentsolutions: fi(x)- xP 1 andf2 (x) == xP 2 , where pi andp 2 are
given by PI == [1 + (1 4{3)II 2 ]/2 and p 2 == [1 - (1 4{3)II 2 ]/2. One has PI > 1
and P2 > -1/2, because {3 < 3/4. Thus !I and f 2 , and consequently all solutions f
SCHRODINGER OPERATORS 121

of- f"(x) + {3x- 2 f(x) == 0, are square-integrable on (0, b).


(ii) Now consider a solution g of -g"(x) V(x)g(x) 0. Let fo be the solution
of- f"(x) + (Jx- f(x) == 0 satisfying the initial conditions fo(b) == lg(b)l + 1,
2

f~(b) == -lg' (b) I - 1. The function fo is real, and fo E L 2 ( (0, b)) as seen in (i).
Clearly there is a neighbourhood of x == bin which fo(x) > lg(x)l and f6(x) <
-lg' (x) I· Let y be the largest number in (0, b) for which one of the preceding two
inequalities is violated, i.e. the largest number in (0, b) such that either lg(y) I - fo (y)
or f~ (y) == -lg' (y) I· We shall show that such a y cannot exist. Hence one will have
lg(x)l < fo(x) for all x E (0, b), which shows that g E £ 2 ((0, b)). Since g is an
arbitrary solution of -g"(x) + V(x)g(x) == 0, Vis in the limit circle case at x == 0.
(o:) Suppose that y E (0, b), fo(Y) == lg(y)i and f6(x) < lg'(x)l for x E [y, b].
Then

fo(y) = fo(b) -1b f~(x)dx = lg(b)l + 1 -1b f~(x)dx

> lg(b)l + 1 1blg'(:r)ldx > lg(b)l + 1 + 11bg'(x)dxl

- lg(b)l 1 lg(b)- g(y)l > lg(b)l + 1 + [lg(y)l- lg(b)l] > jg(y)j,

where the second inequality from the end is obtained by applying ( 1.15) in the Hilbert
space C. We have obtained a contradiction with the hypothesis that fo(Y) - lg(y) I·
({3) Suppose that y E (0, b), f6(y) == -lg'(y)l and fo(x) > lg(x)l for x E [y, b].
Then

!MY)= fMb) -1b f~'(x)dx = -lg'(b)l 1 -1b f~'(x)dx


lg'(b)l- 1 1b (3x- 2 fo(x)dx < -lg'(b)l- 1 -1biV(x)llg(x)ldx

= -lg'(b)l- 1-
1 Y
b
lg"(x)ldx < lg'(b)l- 1 -11 b
g"(x)dxl

== -lg' (b) I - 1 Ig' (b) - g' (y) I < lg' (y) I'

where the last inequality is obtained as in (o:). We have obtained a contradiction with
the hypothesis that f6 (y) == -lg' (y) I· D

3.3 . 3. The above considerations have implications for Schrodinger operators with
spherically symmetric potentials inn dimensions (n > 2). We discuss the most im-
portant case n - 3. The Hilbert space is 1-{ == (JR 3 ). We first recall the partial wave
decomposition of L 2 (IR 3 ). The wave functions f(x) in L 2 (IR 3 ) can be considered, in
spherical polar co-ordinates, as functions of r' e and ¢, where
X3 X2
r-lxl, cose--
r
and tan¢==-'
x1
so
X1 == r sin 8 COS¢, X2 - r sin 8 sin c/J and X3 == r COS 8.
122 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

Let us write f (r, e, ¢) for the wave functions in spherical polar co-ordinates. One has

(3.39)

So, for almost all r, j (r, e, ¢) must be square-integrable over the angular variables e
and ¢ (otherwise the integral with respect to dr would be infinite, i.e. one would have
lifii == oo). Thus, for almost all r, f(r, e, ¢)belongs to the Hilbert space L 2 (S 2 ), the
set of (equivalence classes of) functions g: S 2 --+ CC that are square-integrable with
respect to the n1easure sine dB d¢. Here S 2 == {x E JR 3 IIxl == 1} denotes the unit
sphere in JR 3 , and the scalar product in £ 2 ( S 2 ) is given by

(g, h) L2(S2) =
r 7r r2n
Jo sin 0 dO .fo d¢ g( 0,¢ )h( 0,¢). (3.40)

We see that L 2 (JR 3 ) can be identified with £ 2 ((0, oo); (S 2 ),r 2 dr), the Hilbert
space of square-integrable functions defined on (0, oo) with values in L 2 (S 2 ), with
tneasure r 2 dr on (0, oo ). 10
An orthonormal basis of L 2 ( S 2 ) is given by the surface spherical harmonics Yf,
where fJ assumes the values 0, 1, 2, 3, ... and m - £,£- 1,£- 2, ... , -£. 11 Iff is a
wave function, one can develop f(r, e, ¢),for each fixed r, in this basis {Yf}:
()() £
f(r, e, ¢) == L L f-em(r)Yf(B, ¢) (3.41)
£=0 m=-£

[more precisely, the function f is given, this development makes sense for almost
all r > 0, namely for those values of r for which f (r, e, ¢) belongs to (S 2 ) as a
function of e and ¢ J. The coefficient of Yf depends on r; for this reason it has been
denoted f-ent (r) (for fixed £ and m it is a function of the radial variable r ). Since the
functions {Ye} form an orthonormal basis of L 2 (S 2 ), one has for fixed r:

~ m~e lfem(r)
2
2 2
iif(r, 0, ¢) iil2(S2) 11r sin 0 dO 1 1rd¢lf(r, 0,¢) 1 = 1 ,

(3.42)
and consequently

(3.43)

The wave functions of the form f (r, B, ¢) == f-ern (r) Y f (e, ¢), for fixed £ and m, form
a subspace of L 2 (JR 3 ) which we denote by H-em· One has H-em Hem' if£ #- f' or
rn #- rn'. We set H-e == EB~=-£ H-em. Thus the wave functions in H_e are of the for1n
10
This is a special case of the situation considered in relation with (1.53): 0 = (0, oo), JC = £ 2 (8 2 )
and m(dr) = r 2 dr.
11
See for example Section 2.1 of [Ne].
SCHRODINGER OPERATORS 123

f(r, 8, ¢) == I:~=-£ f£m (r )Y£(8, ¢ ), with f£,m E £ 2 ( (0, oo ), r 2 dr) [so each f£m is
a complex-valued function defined on (0, oo) which is square-integrable with respect
to the measure r 2 dr]. The wave functions in 7-{£ are called wave functions in the
partial wave of angular momentum £, 7-{£ is the associated partial wave subspace.
Since £ 2 (IR 3 ) == EB~o H£, each wave function g admits a unique decomposition
g == I:~o with 9£ E 7-{£ (this represents the partial wave decomposition of g).
Let us now consider the free Hamiltonian H 0 == P2 introduced in §2.5.3. We use
the fact that each H£m (hence also each 7-{£) reduces the operator H 0 , i.e. that the
restriction of Ho to H£rn defines a self-adjoint operator Ho,Rm in H£m, the domain
of Ho,Rm being H£m n D(Ho) (some details on this are given in the Appendix to
this chapter). Ho,Rm is an ordinary differential operator. To see this we recall that, on
smooth functions, H 0 acts as - ~. In spherical polar co-ordinates the Laplacian has
the following form:
82 2 a 1
~ ==- 2 + - - + -~s
(3.44)
8r 2
r or r '

where ~sis the spherical Laplacian (acting on functions of e and¢):


a2 a 1 a2
~s ==-+cote-+----
oe2 ae sin2 e 8¢ 2 . (3.45)

The spherical harmonics are the eigenfunctions of ~s, more precisely: ~s Y£ ( e, ¢) ==


-£(£ + l)Y£(8, ¢).Thus, iff E H£m is twice differentiable, then
-~f(r, 8, ¢) == -~ [fRm(r)Y£(8, ¢)]
2
d 2 d R(R + 1) ] m,
== [ - (dr2 +-:;: dr)f£m(r) + r2 fRrn(r) Y£ (8, ¢),

i.e. Ho,Rm is a self-adjoint extension in £ 2 ( ( 0, oo), r 2 dr) of the ordinary differential


operator
_ (_!!__2 + 3_ ~) + R(R + 1).
dr r dr r2
In particular Ho,£m, is independent of rn, it depends only on the value of£. For this
reason we simply write Ho,£ for these operators. In quantum mechanics Ho,£ is called
the free radial Hamiltonian of angular momentum£.
In order to be able to apply the results obtained earlier in this section, it is useful to
map I1 2 ( (0, oo), r 2 dr) in a unitary way onto £ 2 ( (0, oo), dr) £ 2 (IR+ ). The unitary
mapping U is simple: iff E £ 2 ( (0, oo), r 2 dr ), then Uf defined by (Uf) (r) == r f (r)
belongs to L 2 (IR+) and

2 2
II U.f II iz(IR+) = foool.f (r) 1 r dr = ll.f II iz((o,oo) ,TzdT) ·

Also, if g E £ 2 (IR+) is twice differentiable and f _ u- 1 g is the associated element


of £ 2 ( (0, oo), r 2 dr ), i.e. f (r) == r- 1 g(r ), then a short calculation (using the Leibniz
rule) shows that
2 2
2 d )f( r ) ==- ( -d + --
- ( -d 2 + -- 2 d) r -1 g ( r ) == -r -1 g"( r ) .
2
dr r dr dr r dr
124 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

This implies that

[UHo.Rf](r) == r[-r- 1 g"(r) + £(£ + 1)r- 2 f(r)] == -g"(r) + £(£ + 1)r- 2 g(r).

This shows that the operator Ho,£ is unitarily equivalent to the differential operator
-d 2 I dr 2 + £(£ + 1)r- 2 in £ 2 ( (0, oo) ). More precisely:
UHo,R u- 1 is a self-adjoint extension of the minimal operator associated with the dif-
ferential expression -d 2 I dr 2 + £( £ + 1)r- 2 in £ 2 ( (0, oo)).
If£ -1- 0 this minimal operator is essentially self-adjoint [the deficiency indices are
(0, 0) by Propositions 3.26 and 3.29], so that in this case UHo,£ u- 1 is the closure
of the minimal operator. If £ == 0, however, the minimal operator has deficiency in-
dices ( 1, 1) because the potential V 0 is in the limit circle case at T == 0. Hence
1
UH0 .0 U- coincides with one of the self-adjoint extensions of the minimal operator
associated with -d 2 I dr 2 in £ 2 ( (0, oo) ), i.e. with one of the operators K( a) described
on page 107. To know the value of a, we recall from (2.83) that the wave functions in
V( H 0 ) are bounded on IR 3 • Since the spherical harmonic Y3 is a constant (i.e. inde-
pendent of e and¢), the elements in Ho,o are just functions f(r ), and iff E V(Ho,o),
then f(r) 1nust be bounded and continuous, in particular f(O) < oo. Hence the func-
tion g == Uf which is associated with f in £ 2 (IR+), i.e. g (r) == r f (r), must satisfy
g(O) == 0. By comparing with the definition of K(a) given on page 107, one sees that
UH0 .0 U- 1 is the operator K(a) for a== 0.
Note that the non-self-adjointness in the partial wave of angular momentum£ == 0
appears in a somewhat artificial way through the introduction of spherical polar co-
ordinates and the separation of variables (the point x == 0 is singular for spherical
polar co-ordinates). The operator -~ is essentially self-adjoint on C 0 (IR 3 ) but not
on C 0 (IR 3 \ {0}) [C0 (IR 3 \ {0}) is the set of functions in C 0 (IR 3 ) that vanish near
x == 0], and the fact of considering the minimal operators UHo,£ u- 1 on C 0 ((0, oo))
amounts in £ 2 (IR 3 ) to restricting - ~ to C 0 (IR 3 \ { 0}) [one avoids the singular point
x == 0 of the co-ordinate system].
Let us next consider Hamiltonians including a potential V, and assume that V is
spherically symmetric (and real, of course). In other words let us consider differential
operators of the type - ~ + V (r) in three dimensions, assuming that V is a real func-
tion that is continuous on (0, oo). If Vis square-integrable in a neighbourhood of the
origin (as a function defined on IR 3 ), i.e. if J0 IV(r)j 2 r 2 dr- < oo, then-~+ V(r)
1

is defined on C 0 (IR 3 ). On the other hand, if V(r) is too singular at r == 0, then the
preceding integral will not be finite and one must consider the operator - ~ + V (r) on
C 0 (JR 3 \ {0} ). In this situation one is naturally led to avoid the singular point x == 0
of the spherical co-ordinate system (this point then coincides with the singular point
of the potential).
The component in 1i£m of a function f belonging to C 0 (IR 3 \ { 0}) will have the
forn1 !Rm(r)Y£(8, ¢)with f£m E Co((O, oo)), and the function 9£m Uf£m corre-
2
sponding in £ (IR+) to f£m will also belong to C 0 ((0, oo)). The deficiency indices
of the operator - ~ + V (r) on C 0 (IR 3 \ { 0}) are equal to the sum of the deficiency
indices of the minimal operators associated with -d 2 I dr 2 + £(£ + 1)r- 2 + V(r-) in
SCHRODINGER OPERATORS 125

L 2 (JR+), each of the partial deficiency indices being multiplied by 2R + 1 (because for
fixed£, the parameter m assumes 2£ + 1 values). We give below a somewhat more
precise description of this situation but leave the verification as an exercise. We de-
note by VR the value of the deficiency indices of -d 2 / dr 2 + £(£ 1)r- 2 + V(r) in
£ 2 ((0, oo)) (remember that the two deficiency indices are identical for each£).
Let us denote by A the symmetric operator [- ~ + V] I c~(IR3 \ { 0 }), i.e. the differen-
tial operator ~ + V (r) in the Hilbert space L 2 (JR 3 ) with domain C ff (JR 3 \ { 0}). The
deficiency indices of A are equal (Proposition 3.9 applies with I< the complex con-
jugation). In the decomposition 7-{ == E9 Rm HRm of the Hilbert space, the operator A
has the form == EBRmARm, where ARm is the following differential operator in the
2
space £ ((0, oo), r 2 dr):

Aem - _:f_2 + 3_ .!!._ + £(£ + l) + V(r) V(ARm) == Co((O,oo)).


dr r dr r2 '

Let BR == UARmU- 1 be the differential operator associated with ARm in L 2 (JR+),


also with domain V(BR) == C0 ((0, oo)) [U: L 2 ( (0, oo), r 2 dr) ~ L 2 (JR+) is given
as above by (Uf) (r) - r f(r ), and UARmu- 1 is independent of m]. One has

Thus the deficiency indices of ARm are equal to those of BR, i.e. to VR (for each mE
{£,£- 1, ... , -£}).Moreover

N(A*- i)- EBN(A£11~- i)


R,m
and consequently
()() R ()()
v dim N(A*- i) == L L dim N(A£m- i) == 2:(2£ + 1)vR.
R=O m=-R
If v < oo, each closed maximal symmetric extension of A is self-adjoint. If v == oo,
A has self-adjoint extensions and also non-self-adjoint maximal symmetric exten-
sions. To determine these extensions one applies the theory of the Cayley transform
in the subspace Hdef :== EBR,vR#O HR (the subspace spanned by those partial wave
subspaces 7-{R for which vR # 0). Let us mention some possibilities for self-adjoint
extensions.
Spherically symmetric self-adjoint extensions
In the representation 7-{ - E9 Rm HRm of the Hilbert space, one considers self-adjoint
extensions of A of the form H == EBRm HRm (i.e. the restriction of H to the subspace
HRm is HRm). The operators HRm are chosen as follows [after identification of HRm
with L 2 ( (0, oo), r 2 dr )]: (i) if VR == 0, HRm is the unique self-adjoint extension of
e
ARm, (ii) for the values of for which VR # 0, one takes HRm == u- 1 IffiRmU, where
IffiRm is one of the self-adjoint extensions of BR (which may depend on m). The oper-
ator H leaves each HRm invariant ( H commutes with the usual representation of the
rotation group in L 2 (JR 3 ); it is clear that each maximal symmetric extension of A that
126 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

commutes with this representation is self-adjoint, even if the deficiency indices of A


are oo, because such an extension must leave each Hem invariant and is therefore of
the form indicated here).
Self-adjoint extensions without spherical symmetry
Suppose for example that there are at least two values of£ for which V£ # 0, say £1
and £2 . Consider the subspace He 1 He 2 of L 2 (JR 3 ), and denote by A(o) the restriction
of A to this subspace. The dimension of [R(A(o) i)]j_ n He 1 is non-zero for j == 1
and j == 2. Let A 0 be a self-adjoint extension of A(o) (in He 1 He 2 ). If the operator
U in the Second von Neumann Formula (3.14) maps certain vectors belonging to the
subspace [R(A(o) + i)]j_ n 1ie 1 into [R(A(o)- i)]j_ n He 2 , then Ao mixes 1ie 1 and
Hp 2 ; so any self-adjoint extension of A coinciding on He 1 7iR 2 with Ao will not
commute with the representation of the rotation group.
EXAMPLES. (a) If V(r) > ~r- 2 near r - 0 and V(r) > -~r 2 in a neighbourhood
of infinity, then -6. + V is essentially self-adjoint on C0 (JR 3 \ { 0}).
(b) If r 2 V(r) ~ 0 as r ~ 0 and V(r) > -~r 2 in a neighbourhood of infinity, the
1ninimal operator Bf! in £ 2 ( (0, oo)) is essentially self-adjoint for £ > 0, whereas
for f - 0 its deficiency indices are ( 1, 1). For example for a Coulomb potential
V(r) == [r- 1 the deficiency indices of -6. [r- 1 on C0 (JR 3 \ {0}) are (1, 1).
We have seen in Remark 2.50 that the Coulomb Hamiltonian He == -~ + [ r - 1 is
essentially self-adjoint on C0 (JR 3 ): its restriction to C0 (IR 3 \ {0}) has deficiency in-
dices ( 1, 1), it is non-self-adjoint in the partial wave of angular momentum £ == 0.
The restriction of He [defined on its maximal domain, which is equal to D(H0 )] to
the partial wave subspace 1{0 of angular momentum£ == 0 corresponds again to the
boundary condition g' (0) sin a == g(O) cos a with a == 0 [hence to g(O) == 0] in
£ 2 ((0. oo)), as for the free Hamiltonian H 0 .
(c) If V(r) == -[r-P near r - 0, with r > 0 and p > 2, then for each£ the func-
tion -[r-P + £(£ + 1 )r- 2 is monotone increasing near r == 0. Thus, by Proposition
3.30(b), the deficiency indices of [-6. + V]jc~(IR3\{0}) are (oo, oo).

Appendix to Chapter 3

A.3.1. Proof of Proposition 3.16. One divides the interval J == (a, b) into a finite (if
b - a < oo) or infinite number of disjoint subintervals of length 3c, where E > 0 is
fixed. It suffices to show that there exist constants C1 and C 2 such that one has for
each of these subintervals J and for f E £ ( J):

h lf'(xWdx < c1 h lf(xWdx + c2 h lf"(xWdx

[then, by summing this inequality over all subintervals J introduced above, one obtains
(3.30)].
So let J == (a, J)] be an interval of length 3c-, and write J == J 1 U J 2 U J 3 with
J 1 ==(a, o:+c-], J 2 == (a+c-, a+2c-], J 3 (a+2c-, J)]; the length of Jk (k == 1, 2, 3)
is thus c. Let x 1 E J 1 and x 3 E J 3; since f is of class C 1, there exist y E [x 1 , x3] such
APPENDIX TO CHAPTER 3 127

that j(x3)- f(xi) == J'(y)(x3- xi) (the pointy depends on XI and We thus
have each x E (a, JJ]

lf'(x)l = IJ'(y) + 1x f"(z)dzl < IJ'(y)l + 1(31f"(z)ldz


< i [lf(xl) + If(x3) I] +1(31 f" (z) Idz
i (3.46)

(because - XI > c). This inequality holds for all E J 1 , x 3 E J 3 and x E (a, j3]
(y does not occur any more in the last expression). We integrate (3.46) with respect to
dx 1 over , which leads to

We now integrate this last inequality with respect to dx 3 over ,!3 :

(3.47)

By applying the Schwarz inequality [with g(z) 1] to the integrals on the right-hand
side of (3.47), one arrives at

c
2
1f' (x) I < [3c Jn
{(3 If (z) I dz 2 ]1/2
+ c2
[
3c J,
{{3 If" (z) I dz 2 ]1/2
.

Consequently [using the inequality (a + b) 2 < 2 (a 2 + b2 )]:


2
IJ'(x)\ <
2
4 3c
c
1(3 \f(z)i
a
2
dz 2 [3c .lp f" ( I
2
z) 1 dz J. (3.48)

We now integrate (3.48) with respect to dx over (a, JJ] (observe that the right-hand
side does not depend on x):

{3 IJ'(x)l 2
dx
6
< 3 3c
1(3 lf(z)l 2
dz + 6c·3E
1{3 IJ"(z)l 2
dz
1 a c ex a

181{3 If (z) I dz + 18c 1{3 If" (z) I dz.


== 2
2 2 2

E a a

Hence (3.30) is true [and it is possible to choose in (3.30) a constant C 2 that is arbi-
trarily small !] . D

A.3.2. Proof of Lemma 3.24. Let f 1 , f 2 be two linearly independent solutions of the
equation £f == z 0 f and g a solution of £g == zg. We have f 1 ,f2 E £ 2 ((c, b)) by
128 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

hypothesis. We assume that W[f 1 , f 2] == 1 (if W[fi, !2] # 1 it suffices to replace !1


by !1 [fi, !2]). For x E (c, b) we set

h(x)- h(x) 1x h(y)g(y)dy- h(x) 1x h(y)g(y)dy, (3.49)

where /'1; is a fixed number in (c, b). One has

h'(x) = f{(x) 1x h(y)g(y)dy- f~(x) 1x h(y)g(y)dy,


because the terms coming from the derivatives of the integrals in (3 .49) cancel. Since
J;' == (V- zo)fj, we have
h"(x) == [V(x)- zo]h(x) + [f{(x)f2(x)- J~(x)j1(x)]g(x),
I.e.

It follows that

(£ - z 0 ) [ g - (z - z 0 ) hJ - [(£ z) g + (z - zo) gJ - ( z - zo) (£ - zo) h


== [(£ - z) g + (z - z0 ) gJ - ( z - z0 ) g == (£ - z) g == 0.

Consequently g- (z- z 0 )h is a linear combination of f 1 and !2: there exist constants


a1 and a 2 (depending on /'1;) such that

Thus [by using the inequality (1.58)]:

lg(x)l 2 < 3laii 21!1(x)l 2 + 3la21 21!2(x)l 2 + 3lz- zol 2lh(x)l 2· (3.50)

Next, applying the Schwarz inequality to the integrals in (3.49), one gets for x E ( /'1;, b):

lh(x)l 2 < 2{lh(xW 1xlh(Y)I 2dy + lh(xW 1xlh(Y)I 2dy} 1xlg(zWdz

< 2{1h(xW 1\h(Y)I 2dy + lh(xW 1blh(Y)I 2dy} 1xlg(z)l 2dz.


Upon inserting this inequality into (3.50) and then integrating over an interval ( /'1;, v)
with /'1; < v < b, one obtains

1vlg(:c)l 2dx < 3laii 2 1blh(x)l 2 dx + 3la21 21blh(xWdx

6lz- zol 2{1blh(xWdx 1blh(Y)I 2dy + 1blh(:cWdx 1blh(Y)I 2dy}

X ivlg(zWdz. (3.51)
APPENDIX TO CI-IAPTER 3 129

As f 1 , .f2 E £ 2 ( ( c, b)), one can choose~ sufficiently close to b so that

For such a choice of~ and for v E ( ~, b), (3 .51) gives

Since the right-hand side of this inequality is finite and independent of v, one obtains
by letting v ~ b that J:
!g(x) l 2 dx < oo, i.e. that g is square-integrable in so1ne neigh-
bourhood of b and hence on (c, b).
The preceding argument applies to each solution g of £g == zg. Thus the lemma is
~~. D
A.3.3. The restriction of H 0 to a partial wave subspace is self-adjoint.
(i) We first show that each of the sub spaces H£m, is invariant under Fourier transfor-
mation. For this we introduce the analogue of the partial wave decomposition (3 .4 ) ~

in momentum space. One can introduce spherical polar co-ordinates for the variable k
in L(JR 3 ) and decompose a function j (k) into partial wave components in L(JR 3 ): if
we write (k, ek, ¢k) for the vector kin spherical polar co-ordinates, then, as in (3.41),
each j E £ 2 (JR 3 ) can be written as

()() £
](k,ek,¢k) == L L l£m(k)Y£(ek,¢k)· (3.52)
£=0 1n=-£

The important fact is that iff E 11£m, i.e . .f(r, e, ¢) == .f£m (r)Yf(e, ¢) then the
Fourier transform j off has the form ](k, ek, ¢k) == f£rn(k)Y£(ek, ¢k) tvith the
same values of £ and m []Rm ( k) is obtained from f£ 1n (r) by a certain formula which
will become evident in the calculations below]. To see this one uses the formula for
the development of a plane wave exp{ ik. x} in spherical harmonics (for fixed k and
r, exp{ ik ·x} is a function of the polar angles e and ¢ of x, hence it can be developed
in spherical harmonics; see for example Chapter 2 of [Ne] for the details):
()() £
ik·x 41T """' """' ·£ ~ (k )Yrn (e ~ )Ym(e ~) (3.53)
e -== kr L...t L...t ~ J £ r R k ' '+' k R ' '+' '
£=0 m=-£

where f£ denotes the Riccati-Bessel functions:

~ ( ) -_
]£X 1 -
( - l)£ X £+1 ( - sinx
d)£ - - (if X > 0).
X dx X

By using the complex conjugate of (3.53) to express e-ik-x, one obtains for the Fourier
transform of .f£m(r)Y£(e, ¢):
130 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

(ii) We denote by P£m the projection in L 2 (JR 3 ) onto the subspace H£m and show
that f E D(Ho) ===? Pgmf ~ V(Ho) and HoP£mf E H£m· By using the decomposi-
tion (3.52) for the function f and the hypothesis that f E D(H0 ), one finds

(3.55)

Since the Fourier transform of P£m,f is f£m(k)Yf (Bk, ¢k) by (3.54), one sees from
2
(3.55) that PRmf belongs to V(Ho) because IIHoP£mfll 2 - f 0 Jk 2 f£m(k)J k 2 dk.
00

Moreover the Fourier transform of HoP£m f is k 2 f£m (k )Yf( ek, ¢k), consequently
Pfirnf E 7-iRrn·
(iii) Let Ho.£m be the restriction of Ho to Vgm :== PgmV(Ho) (considered as an
operator in the Hilbert space 1i£rn). Ho,£m is a symmetric operator in Hfm· To see that
it is self-adjoint, it remains to be shown that R(Ho,Rm ± i) == Hem (Proposition 3.3).
Now for f E V(Ho):

00 £ 00 £
f == L L PRmf and (Ho ± i)f == L L (Ho,bn i)Pemf·
€=0 m=-£ £=0 m=-£
Consider a non-zero vector g E Heomo· Since H 0 ± i are bijections between V(Ho)
and H, there exist unique vectors f± E D(Ho) such that g == (H0 i)f±:

00 £
g- L L (Ilo,Rm ± i)PRmf±·
£=0 m=-£
Since g E He 07n 0 , only the term with £ == £0 and 'm == rn 0 is non-zero in this sum.
Consequently g == (Ho,R 01n0 ± i)Pe 0 m 0 , with Peomof± E De 0 rno· This shows that
R(Ho,Rorno ± i) - Hfomo · D
REMARK. In the language of group theory, the situation can be described as follows.
(t) == exp(-iHot), the family {U(t)}tEJR defines a strongly continuous unitary
representation of the additive group JR in L 2 (IR 3 ). The operators U (t) leave each Hfm
PROBLEMS 131

invariant, hence the restriction ll£m of this representation U to H£m defines a strongly
continuous unitary representation of IR in 11£m. The infinitesimal generator A£m of
this representation U£m is a self-adjoint operator in H£rn (see Section 5.1 ), and this
operator coincides with the restriction Ho,£m of H 0 to H£m·

Bibliographical Notes
The method of the Cayley transform is presented in many texts, e.g. in [AG], [W1]. For
further results on ordinary differential operators the reader may consult [AG], [N] and
also [DS]. In particular, second order operators (called Sturm-Liouville operators)
and Dirac operators are treated in Part II of [W2]. Details of the limit point-limit circle
theory may be found in Section 6.2 of [P] or in Chapter 9 of [CL]. For distribution
theory we recommend [FJ] or [GS].

Problems
3.1. Prove Proposition 3.3(b ).
3.2. Consider the operator A in the space £ 2 (IR) EB£ 2 (IR) defined as follows: A { f, g} ==
{g',- f'}, with {f, g} E D(A) == Co(IR) Co(IR).
(a) Show that A is symmetric.
(b) Does A have self-adjoint extensions?
3.3. Let P0 be the minimal operator associated with (d I dx) in £ 2 ( ( 0, 1)).
(a) Determine the eigenvalues, the eigenvectors and the spectrum of the self-adjoint
extensions JP a of P0 , a E [0, 27T). Show that each A E IR belongs to the spectrum of
some JPa.
(b) Determine the spectrum of the maximal operator P ~ .
(c) Determine the spectrum of P0 .
3.4. Let P0 be the minimal operator associated with -i(dldx) in £ 2 ((0, oo)). Deter-
mine the spectrum and the eigenvalues of its closure Po.
3.5. Consider the differential expression£ == -i( dl dx) on the following domains in
£ 2 ([0, 1]): (i) D1 == C 1 ([0, 1]), (ii) D2 - {f E C 1 ([0, 1]) I f(O) == f(1)} and (iii)
D 3 == {f E C 1 ([0, 1]) I f(O) == f(l) == 0}. Determine in each case whether the corre-
sponding differential operator is symmetric, essentially self-adjoint or self-adjoint.
3.6. In £ 2 ((0, 1)), let Po be the minimal operator associated with -i(dldx). Then
one has P0 P~ \cgo((O,l)) == K 0 . Show that the operator P0 P~ (defined as an operator
product on its maximal domain) coincides with one of the self-adjoint extensions
the operator K 0 given on page 107.
3.7. Let K 0 be the minimal operator associated with -d 2 I dx 2 in £ 2 ( (0, 1) ). Deter-
mine the eigenvalues of the following self-adjoint extensions of K 0 :
K (0, 0), K (0, 7T I 2), K (7T I 2, 7T I 2) and K (p) , p E [0, 27T).
3.8. In £ 2 ( ( 0, oo)), let P0 be the minimal operator associated with -i ( d I dx). Find
the domains of the self-adjoint operators P~ P 0 and P0 P~ and relate these operators
to one of the operators K (a) described on page 107.
See also Problem 4. 7 for spectral properties of K (a).
132 SYMMETRIC OPERATORS AND THEIR EXTENSIONS

3.9. (a) Show that the operator K (0, 0) in L 2 ( (0, 1)) is invertible and that its inverse
is an integral operator the kernel of which is given by

if X< y
G(x, y) == { (1 y)x
(1 - X )y if X> y,

where x, y E (0, 1).


(b) Show that G is a Hilbert-Schmidt kernel.
(c) Show that, for each z E p(K(O, 0) ), the resolvent [K (0, 0)- z] - 1 is also a Hilbert-
Schmidt operator.
[Hint: use the first resolvent equation].
REMARK. Since each Hilbert-Schmidt operator is an integral operator (see Proposi-
tion 2.15), [K (0, 0) - z] - 1 is an integral operator. Its kernel is called the Green's
function of J{ (0, 0) at the point z.
3.10. The limit point-limit circle terminology comes from attempts to study Schrodin-
ger operators on an infinite interval (0, oo) as limits of such operators on finite inter-
vals (0, b) as b --7 oo. Let f and g be solutions of the differential equation -cp" (x) +
V(x)cp(x) == icp(x) on (0, oo) satisfying f(O) == g'(O) == 0 and- f'(O) == g(O) == 1.
For fixed b > 0, consider the set Cb- {z E CCI h(x) :== f(x) + zg(x) satisfies
h(b) cos r + h' (b) sin r == 0 for some r E [0, 1r)}. Each r E [0, 1r) determines a point
z == z (r) in CC, and as r varies, the points z (r) describe a closed curve Cb in the
complex plane. One can show that Cb is a circle. As b --7 oo these circles Cb either
converge to a limit circle Coo or to a single point (a circle of radius 0). In the first case
each solution of -cp"(x) + V(x)cp(x) == icp(x) belongs to L 2 ((0, oo)), while in the
second case only one solution (up to a numerical factor) has this property.
(a) Prove the statement made above that, for given b > 0, each r E [0, 1r) determines a
unique point z == z (r).
(b) For V == 0, calculate f and g, show that Cb is a circle, that Cb' is contained in the
interior ofcb if b' > b, and that the circles { cb} shrink to a point when b --7 00.
3 . 11 . Show by means of examples that there are functions J, g E V(JP a) c L 2 ((0, 1))
such that (JP af, Qg)- (Qf, JP ag) -i(J, g). Similarly show that there are functions
f, g E V(JPo) n V(K(j3, r)) such that (JP af, K(j3, r )g) - (K (/3, r )J, JP a g) 0.
3.12. By using results from Sections 2.4 and 2.6, verify that the square A 2 of an un-
bounded self-adjoint operator A is self-adjoint, i.e. (A 2 ) * - A 2 .
CHAPTER4

Spectral Theory of Self-Adjoint Operators

We have seen in Section 2.6 that the spectrum of a self-adjoint operator A is a subset
of the real axis, and it is well known from quantum mechanics that such a spectrum
may involve different spectral types (eigenvalues, discrete spectrum, continuous spec-
trum). One of the topics of the present chapter concerns the definition and charac-
terisation of these different spectral parts of A (Sections 4.3 and 4.4). Before that, in
Section 4.2, we study the concept of a spectral measure; this means a measure defined
on (JR., AB) the values of which are projection operators in a Hilbert space H rather
than just real numbers. The fundamental result here will be the Spectral Theorem es-
tablishing a one-to-one correspondence between self-adjoint operators and spectral
measures. This theorem somehow expresses the fact that each self-adjoint operator
can be diagonalised, and it also furnishes a simple means for defining functions of
self-adjoint operators.
Given a spectral measure, one can associate a numerical measure mf with each vec-
tor f of the underlying Hilbert space H. These measures rnf are of the Stieltjes type
(Section 4.1) and are the basis for obtaining the spectral parts of self-adjoint operators
in Section 4.3.

4.1 Stieltjes measures

4.1.1. Let F: lR ---t lR be a bounded monotone function. Since each non-decreasing


or non-increasing bounded sequence of real numbers has a limit, all of the following
limits exist (A E JR.):

F(A 0) :== lim F(A- c), F(A + 0) :- lirn F(A +c),


c:___,.+O c:___,.+O

F( -oo) :== lim F(A), F ( + oo) :== liln F (A) .


A---7--oo ).___,.+oo

IfF is continuous at the point A, then F(A - 0) == F(A + 0). IfF is discontinuous
at the point A, then F(A - 0) -1- F(A + 0). We make the convention that, if A is a
point of discontinuity of F, then the value of F at the point A is chosen such that
134 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

its jump at this point is already included in this value F (A), in other terms such that
F(A) == F(A + 0):

p ~------------------------

~ F(>..) == F(>.. + 0)
I
I
~ F(>..- 0)
F I
I

One could choose other conventions, for example F(A) - [F(A- 0) + F(A + 0)]/2
or F(A) F(A - 0). If a monotone function has more than a finite number of dis-
continuities, then the set of all points of discontinuity is a countable subset of JR. The
reader should verify this important simple fact (see Problem 4.1).
In this section we consider functions F: lR ---t lR having the following properties:
( 1) F is monotone non-decreasing, i.e. A > tL ====? F (A) > F (tL),
(2) F is right continuous, i.e. F(A) F(A + 0) for all A E JR.,
(3) F( -oo)- 0 and p :== F(+oo) < oo.
A function F with these properties generates a Borel measure m F on the real line,
called the Borel-Stieltjes measure (or simply the Stieltjes 1neasure) associated with F.
It is obtained in the same way as the Borel measure discussed in §1.4.2, by replacing
the length of an interval (a, b] by its Stieltjes measure defined as

mF((a, b]) == F(b)- F(a) ==the variation ofF on (a, b]. (4.1)

The value of mF( (a, b]) is a number in [0, p ], and the condition of a-additivity of mF
on the collection of all half-open intervals is satisfied; for example if (a, b] and (b, c]
are adjoining intervals, then m F( (a, c]) == m F( (a, b]) + m F( (b, c]). The measure of
a general Borel set V is determined, in analogy with ( 1.45), by

mF(V) == inf L mF(Jk), (4.2)


k

where again { J k} denotes a sequence of half-open intervals covering the set V, and
the infimum is taken over all such coverings of V. Since p < oo, the Stieltjes measure
m~F is a finite measure, i.e. mF (JR.) < oo:

00

mF(JR)- mF(U~_ 00 (k,k+ 1]) == L mF((k,k+ 1])


k=-oo
00

== L [F(k+1) F(k)] ==F(+oo)-F(-oo)==p.


k=-oo
STIELTJES MEASURES 135

is interesting to know that there is a converse of the preceding result: if 1n is a


finite measure on the Borel a-algebra AB of JR, then there exists a (unique) function
F on lR satisfying (1)-(3) such that the associated Stieltjes measure 1nF coincides with
m/, i.e. such that mF(V) == rn(V) for each V E AB. Indeed, it is easy to verify that the
following function F has all the required properties:

F(A) == m(( -oo, A]). (4.3)

We continue by discussing some examples that will be useful in the sequel. We say
that a Stieltjes measure m F is SU]Jported on a Borel set V if rn F (JR \ V) - 0.

Example Let F be a piece-wise constant function, for example

if A< 1
F(A)- { O
k/(k 1) if k <A< k 1, k == 1, 2, ...

1 ~--------------------------

3/4 ------- • ' -....!·~......s·~-


2/3 ---- -··--
1/2 --··--

0 1 2 3 4 5 6

== 0 if A tJ_ N {1, 2, 3, ... }, whereas 1nF( {n}) ==


In this case one has mF( {A})
n/(n + 1)- (n- 1)/n == 1/[n(n + 1)] if n == 1, 2, 3, ... , i.e. rnF( {n}) is equal to
the jump ofF at the point n. For an arbitrary Borel set V one has:

m
F
(V)= "'"""
L... 1
n(n+l)'
nEVnN

in particular mJ F (V) == 0 if V n N == 0. A measure m F of this type is called a discrete


measure, it is supported on a discrete subset of lR (a set consisting of isolated points).

Example 4 . 2.. If the function F assumes only the values 0 and 1, one obtains a Dirac
measure. If the jump of F is at the point A == 0, the associated Dirac measure 5 is as
follows:

6(V) = { ~ if 0 tJ_ v
if 0 E V

In this case F is the Heavisidefunction, viz.


136 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

F(A) = { ~ if A< 0
if A 2::: 0
le------

Example 4 . 3.. A generalisation of the situation considered in Example 4.1 is that of


a function F which is continuous except for a countable number of jumps, say at the
points of a sequence {A 1 }. These points need not form a discrete set (they need not
be isolated from one another). For example, let { r 1 } be an enumeration of the rational
numbers in the interval (0, 1) and set

F(A) == L 2-J.
rJ :=;A.

This is ajumpfunction: F(A) == 0 for A< 0, F(A) == 1 for A> 1, and on (0, 1) the
function F is continuous except at the rational points, with a jump of 2-1 at r 1 . The
associated measure is given by

This measure is supported on the set of all rational numbers in (0, 1), the measure of
the rational point r1 being 2-1. A measure of this type is called a pure point measure
or an atomic measure. 1

Example 4 . 4 . Let F be a continuous function. Then each point of lR has measure zero:
if A E JR., then A E (A- E, A] for each E > 0, hence mF( {A}) < [F(A) F(A- c:)]
for each E > 0; since F is continuous, one must have mF( {A}) - 0. An example is
given by the function F(A) 1/2 + 7r- 1 arctan(A):

1 1----------

The explicit function F(A) == 1/2 7r- 1 arctan(A) given in the preceding example
is absolutely continuous, i.e. an indefinite integral: F(A) == J~ 00 [7r(1 + tL 2 )]- 1 dfL. In

1 Son1e authors use the term "discrete measure" also in such a situation.
STIELTJES 1\1EASURES 137

this case the measure m F (V) of a Borel set V is obtained by integrating the derivative
ofF over V: 2

(V) == j F (JL dfL =! jv


00

m F 1 )
F 1 (JL)Xv(JL)dJL ==-1 dfL · (4.4)
v -00 7r 1 + fL 2

This is an example of an absolutely continuous measure on JR. There also exist con-
tinuous functions that cannot be expressed as an indefinite integral, as explained in the
next example.

Example 4 . 5.. Without entering into explicit mathematical derivations, we describe



here the Cantor function. This is a non-decreasing continuous function which is dif-
ferentiable at each point in 1R not belonging to the Cantor set <t (see page 16), hence
almost everywhere with respect to Lebesgue measure, and the derivative is zero at all
these points: F 1 (A) == 0 if A ¢:. <t. Thus F is not the integral of its derivative, since
F( -oo) == 0 and F( +oo) == 1. This function is illustrated in Figure 4.1:

1 -----------

3/4

1/2 --------------

1/4 ---
1/8 ......

0 1/3 2/3 1

Figure 4.1 The Cantor function.

The Cantor function is constant on each interval disjoint from <t: F(A) == 0 for A < 0,
F(A) == 1 for A > F(A) == 1/2 on (1/3, 2/3); F(A) == 1/4 on (1/9, 2/9),
F(A) == 3/4 on (7 /9, 8/9): F(A) == 1/8 on (1/27, 2/27) etc., and its values at the
points A of <t are defined by continuity. A more precise definition of the Cantor set
and the Cantor function can be obtained in terms of the expression in base three of the
numbers A E (0, 1) (for details see for example [R]).
The measure m F associated with the Cantor function has the property that rrn F (V) ==
0 if the intersection of V with the Cantor set is empty; thus this measure is entirely
supported on the Cantor set, which is a Borel set of Lebesgue measure zero. A measure
m on 1R having these properties [i.e. a measure satisfying (i) F().) - m(( -oo, A]) is
continuous in A, (ii) there exists a Borel set W of Lebesgue measure zero such that
m(V) == 0 whenever V n W == 0] is called a singular continuous measure.
The Cantor function is an example of a singular continuous function, i.e. a func-
tion <I> having the following properties: (i) <I> is continuous, (ii) <I> is differentiable
2
xv denotes the characteristic function of the Borel set V, see (1.41 ).
138 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

except possibly on a null set W (with respect to Lebesgue measure), (iii) its derivative
zero at each point of JR \ W.
Remark 4 . 6.. To construct Stieltjes measures m F it is important to start by defining
the rneasure for half-open intervals (a, b]. Indeed, contrarily to Lebesgue measure, the
Stieltjes measure of an open interval (a, b) can be different from that of (a, b]; for
instance in Example 4.1 one has mF((O, 2)) == 1/2 but m~F((O, 2]) == 2/3. It is easy
to give expressions in terms of the function F for the measure of intervals that are not
of the type (a, b]:

mF((a,b)) ==F(b-0) F(a), rnF([a,b]) -F(b) -F(a-0). (4.5)

..
~ Jil.~·-·
The preceding examples exhibit all typical features of Stieltjes measures on the
real line: any Stieltjes rneasure m on JR admits a unique decomposition into a sum of
three measures:
(4.6)
where mp is a pure point 1neasure, m~sc a singular continuous measure and m~ac an
absolutely continuous measure. Each of the four measures occurring in this equation is
associated with a unique function having the properties (1)-(3) stated at the beginning
the preceding subsection. Let us denote these functions by F, Fp~ Fsc and Fac, so
rn- 1n~F,rnp == rnFP,msc == mFsc and mac== . In view ofEq. (4.3), one
has
(4.7)
other terms a unique decomposition of the function F associated with the rneasure
1nj into a sum of a jump function, a singular continuous function and an absolutely
continuous function.
is easy to construct examples of functions F such that each of the three measures
, rnt; and mjrc the decomposition (4.6) of mJF is non-trivial. One can take for
example for Fp the Heaviside function (see Example 4.2), for Fsc the Cantor function
and Fac(A.) == 1/2 1r- 1 arctan(A.), and set F == Fp Fsc + Fac·
The remainder of this subsection is devoted to some explanations concerning the
above-mentioned decomposition of a general Stieltjes measure. This decomposition
can be obtained either by decomposing the function F, as in Eq. (4.7), or by decom-
posing the rneasure m~ itself, as in Eq. (4.6). The former method is based on special
properties of monotone functions, the latter on general results from measure theory.
We first comn1ent on the decomposition (4. 7). Let F be a function having the prop-
erties (1)- (3) of §4.1.1. One first collects all discontinuities ofF into a function Fp.
has no discontinuities, take Fp == 0. Otherwise let { A.f} be an enumeration of the
discontinuity ofF (remernber that this set is at most countable), and define

(4.8)

Fp is the the unique jump function having discontinuities of the same size and at the
same points as the function F. Setting Fe == F - Fp, it is easy to see that Fe is
STIELTJES MEASURES 139

a continuous function also satisfying (1) and (3) of §4.1.1 (Problem 4.1). Now each
continuous monotone function ¢ is differentiable almost everywhere 3 : there exists a
Borel set V of Lebesgue measure zero such that ¢' (A) exists and is finite at all points
.\ E lR \ V. Applying this result to the function Fe, one finds that its derivative F~ is
equivalent to a non-negative integrable function on JR. One may thus set

(4.9)

It is then clear that Fsc :==Fe- Fac is continuous and that F:c(.\) == 0 for A E lR\ V.
One can also check that Fsc is monotone non-decreasing.
To obtain the decomposition (4.6) of a Stieltjes measure m, one can appeal to
an abstract theorem on comparison of measures which will now be described. Let
(0, A) be a measurable space and m 0 , m two measures on (0, A). One says that rn
is absolutely continuous with respect to rn 0 if m(V) == 0 for each V E A such that
mo (V) == 0, and that m and mo are mutually singular if there exist wl' w2 E A such
that W 1 nW2 == 0, W 1 UW2 == 0 and m(W1) == mo(W2) == 0. Thus m is absolutely
continuous with respect to m 0 if all null sets with respect to m 0 are also null sets with
respect tom, and m and m 0 are mutually singular if m is supported on an m 0 -null set
and m 0 is supported on an m-null set. The following theorem shows that, given two
measures m 0 and m on the same measurable space ( 0, A), one can express m as a
sum of two new measures one of which is absolutely continuous with respect to mjo
and the other one is singular with respect to m 0 (this result holds provided that both
measures m and m 0 are a-finite, see page 14).

Proposition 4 . 7 (Lebesgue Decomposition Theorem) . Let m, m 0 be a-finite mea-


sures on a measurable space (0, A). Then there exist two unique measures rrt1, m2
on (0, A) such that m1 is absolutely continuous with respect to mo, m2 and mo are
mutually singular and m == m1 + m2 [i.e. m(V) == m1(V) + mj2(V)jor each V E A].

By taking form a Stieltjes measure and for rn 0 the Lebesgue measure on (JR, AB),
one obtains from Proposition 4.7 a decomposition of minto m == mac + rn 8 , where
mac is absolutely continuous with respect to Lebesgue measure and rns is singular
with respect to Lebesgue measure 4 . The pure point part of the Ineasure m is of course
contained in m 8 , and it is easily isolated by setting for each Borel set W:

rnp(W) == L m({A}) L r11s({A}). (4.1 0)


.AEW .AEW

There are at most a countable number of contributions to the sum in Eq. (4.10), aris-
ing from the points of discontinuity ofF (the points having non-zero measure). The
difference ms - mp is a measure without atoms (i.e. such that mj( {A}) - 0 for each
3
See the Bibliographical Notes at the end of this chapter.
4In this case [measurable space (JR, AB)] it is usual to speak simply of an absolutely continuous
measure and of a singular measure; it is understood that these notions are defined with respect to Lebesgue
measure.
140 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

A E JR) and supported on a Borel set of Lebesgue measure zero, hence it is a singular
continuous measure.
Let us finally mention the Radon-Nikodym Theorem which gives a characterisation
of the absolutely continuous part of a measure with respect to some other measure. In
the terminology of Proposition 4. 7, the absolutely continuous part m 1 of the measure
rn with respect to m 0 has the following property: there is a measurable function f
such that the measure m 1 (W) of any Borel set W can be obtained by integrating this
function with respect to rn 0 over W. In the statement below we write m in place of
m 1 and assume that m is absolutely continuous with respect to m 0 .

Proposition 4 . 8 (Radon-Nikodym Theorem) . Let m and m 0 be a-finite measures on


a measurable space (0, A) and assume that m is absolutely continuous with respect
to m;o. Then there exists a non-negative measurable function f: 0 ~ [0, oo] such
that for each V E A:
m(V) = fv f(x)mo(dx). (4.11)

f is unique in the sense that if g is another function having the above properties, then
g ( x) == f (~r) 1TLo- a. e.
The function f appearing in (4.11) is called the Radon-Nikodym derivative of m
with respect to m~o and sometimes denoted by f == dm / dmo.
For a Stieltjes measure rnF on the measurable space (JR, AB) we know from Eq.
(4.9) that its absolutely continuous part mrc can be written as an integral of the deriva-
tive of (or equivalently of Fe) with respect to Lebesgue measure: mrc (W) ==
fw F~c (!L) dfL == fw F; (tL) dfL. In this situation the derivative of Fac coincides with
the Radon-Nikodyrn derivative of the absolutely continuous part mrc of mF with re-
spect to Lebesgue measure.

4 . 1. 3 . Let us briefly consider integration with respect to a finite Stieltjes measure


1YL F. One can apply the general definition of an integral given in §1.4.1 by simply re-
placing in ( 1.44) the measure m by the Stieltjes measure m F under consideration. This
amounts to approximating the function cp that is to be integrated by a sequence { cp k}
of simple functions [observe that each measurable simple function is m F_ integrable
because m F (V) < oo for each Borel set V, since we assume that m F is a finite mea-
sure]. The integral obtained in this way is called the Lebesgue-Stieltjes integral.
For a continuous function cp one can also define the integral J cp (A) m F (dA) in
-CX)CX)

the sense of Riemann (one then speaks of a Riemann-Stieltjes integral). If (a, b] is a


finite half-open interval and IT a partition of this interval (see Section 1.2), one sets
N N
L:(IT, cp) == L cp(uk) [F(sk)- F(sk-1)] == L cp(uk) mF((sk-1, sk]) (4.12)
k=1 k=1
and
(4.13)
SPECTRAL MEASURES 141

where {IIr} is a sequence of partitions of the interval (a, b] with limr~oo \IIr \ == 0.
The second notation in (4.13) for this integral is the more usual one, and the limit
in (4.13) exists for each function cp that is continuous on [a, b]. The integral over lR
of a continuous function is defined as an improper integral (if this improper integral
00
J_
exists). cp : lR ---t C is continuous and bounded, then the integral 00 cp (A) m F ( d:A)
exists (we assume that mF is a finite measure), and one has

(4.14)

Finally, if the function F is absolutely continuous, the following identity is inter-


esting: if cp is a bounded measurable function, then

1 a
b
fP(A) dF(A) -1 b
fP(A)F'(A)dA. (4.15)

4.2 Spectral measures

4 . 2 . 1 . A spectral measure is some kind of a generalisation of the Stieltjes measures


discussed in the preceding section. Instead of considering non-decreasing functions
defined on JR. with values in JR., one takes non-decreasing functions defined on lR
whose values are projections in a Hilbert space H. Consider for each A E JR. a sub-
space M>.. and assume that this family of subspaces {M>..} is non-decreasing in the
sense that M >.. M f-L if A < fL· Furthermore assume that the intersection of all sub-
spaces M >.. consists of only the zero vector 0 and that the union of all M >.. is dense
in H. If we denote the projection onto M>.. byE>.., i.e. E>.. - PM>-, we can express
the properties of the family { M>..} in terms of the family of projections { E>..} (see in
particular Proposition 2.6):

(a) the family {E>..} is non-decreasing, i.e. E>..Ef-L- Ernin{>..,J-t}~ (4.16)


(b) s-limE>..- 0 and s-limE>..- I. (4.17)
>..~-oo >..~+oo

REMARKS. (i) In view of (b) it is natural to set E( -oo) == 0 and E( +oo) I.


(ii) Equation (4.16) implies that

VA., fL E JR., (4.18)

i.e. the family of projections { E>..} is a family of commuting operators. Equation


(4.16) can also be written as E>.. < Ef-L if A < fL, i.e. Ef-L - E>.. is a positive oper-
ator if A < fL· We recall that A is a positive operator if (J, A f) > 0 for each vector f
in D(A).
A family of projections { E >..} >..EIR satisfying (4.16) and (4.17) is called a spectral
family or a resolution of the identity. It is easy to show by using (4.16) that the limits
s-lin1c:~+o E>..±c: exist for each A E JR. (Problem 4.2). We use the notation E>..±o for
142 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

these li1nits and observe that E>..-o < E>.. < E>..+O· As in the case of Stieltjes measures
we consider only spectral families that are right continuous, i.e. such that

(4.19)

and we use the simplified terminology "spectral family" in place of "right continuous
spectral family".
If -oo <a< b < +oo one can define in analogy with (4.1)

E((a, b]) == Eb Ea. (4.20)

This associates with each interval of the form (a, b] a positive operator E ( (a, b]).
Clearly if (a, b] and (b, c] are two adjoining intervals, then E( (a, c ]) - E( (a, b]) +
E( (b, c] ). If (a, b] and (c, d] are two arbitrary intervals, then (4.16) implies that

E((a, b])E((c, d]) == (Eb- Ea)(Ed- Ec)


== Emin{b,d}- Emin{b,c}- Emin{a,d} + Emin{a,c}·
• If (a~ b] n (c, d] == 0, for example if a < b < c < d, then
E((a, b])E((c, d]) == Eb- Eb- Ea + Ea- 0. (4.21)

• (a, b] C ( c, d], i.e. if c < a < b < d, then

E((a, b])E((c, d]) == Eb- Ec- Ea + Ec == Eb- Ea- E((a, b]). (4.22)

• If (a, b] and (c, d] overlap without one of them being a subset of the other one, e.g.
if a < c < b < d, then

E((a,b])E((c,d]) == Eb- Ec- Ea + Ea == Eb- Ec == E((c,b]). (4.23)

These three formulas can be combined into the following one (with the convention
that E(0) - 0):
E ( (a, b]) E ( (c, d]) == E ( (a, b] n (c, d]). (4.24)

Consequences: (i)IfE((c,d]) ==Oand(a,b] C (c,d],thenE((a,b]) ==0.


(ii) E( (a, b]) is a projection: E( (a, b]) is self-adjoint because Ea and Eb are self-
adjoint, and the relationE( (a, b]) 2 == E( (a, b]) is obtained by taking a == c, b == din
(4.22).
As for numerical measures, one can extend the family of operators defined by
(4.20) to the a-algebra generated by the family of half-open intervals (a, b], i.e. to
the Borel a-algebra AB of JR. In this way one associates a projection E(V) with each
Borel set V in AB, and this association is a-additive:

E(Uk vk) == L E(Vk) (4.25)


k

if {Vk} is a countable family of disjoint elements of AB. Furthermore

E(0) == 0 and E(JR) == I. (4.26)


SPECTRAL 1\1EASURES 143

The collection projections { E (V)} is called the spectral measure associated with
the spectral fa1nily { E>..} (the term "spectral" will become clear later). Thus a spectral
measure is a a--additive mapping from AB into the set of projections a Hilbert space
1-{ satisfying also (4.26). The following property generalises (4.24 ):

E(V)E(W) == E(V n W) (4.27)

Thus, given a spectral family { E>..}, one can associate with each Borel set V of JR a
subspace (V) of 7-i, viz. M(V) == {f E 1-{ I E(V)f == f}, and one has

M(V) _ M(W) if V _ W and M(V) j_ M(W) if V n W == 0. (4.28)

In analogy with (4.5) one has for example for the projection associated with an
open or closed interval:

E((a, b))== Eb-o- Ea, E([a,b)) == Eb- Ea-o· (4.29)

The following consequence of (4.28) is useful. Let {Vj }~1 = 1 be a collection of dis-
joint Borel subsets JR (i.e. Vj n Vk == 0 if j -# k), { aj }~= 1 a collection of complex
numbers and f E 7-i. Then, since E(Vj )f j_ E(Vk)f if j -# k:
n n
2 2 2
II LajE(Vj)Jii == L laJI \\E(Vj)f\l · (4.30)
.J=l j=l

One may take in (4.20) a == b - E and let E ~ +0, which defines the projection
E ({ b}) associated with the Borel set V == { b} consisting the single point b:

E({b})==s-limE((b- b])==Eb-Eb-o (4.31)


E---++0

[equivalently: set a== bin the second equation in (4.29)]. E( {b}) is a projection (since
it is the strong limit a sequence projections, see the end §2.2.1 ). One has:
(i) E( {b}) == 0 (spectral measure the point b equal to zero) ~ the spectral
family { E>..} is (strongly) continuous at the point A == b,
(ii) E( {b}) -# 0 (spectral measure the point b not equal to zero) ~ the spectral
family { E>..} is discontinuous at the point A == b.
In case (ii) one can associate with the point b a subspace M ({b}) non-zero dimen-
sion, namely

M({b}) == {fEH I E({b})f == !} == R(E({b})). (4.32)

The interpretation of this subspace will be discussed in §4.3 .1. Let us 1nention here
that the discontinuities of a spectral family form a countable subset of JR. Indeed, if
b1 ¥ b2 are two points of discontinuity, then E ( { b1 }) E ( { b2 }) == 0 by (4.27), hence
M ({b1 }) is orthogonal to M ({b2 }). If one chooses at each point discontinuity b
a vector fb ¥ 0, one obtains a family {fb} of non-zero vectors that are pair-wise
orthogonal; this family is countable since the Hilbert space is assumed to be separable
[postulate (H4) of §1.1 .1].
144 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

The support of a spectral family {E>.} is the following subset of JR:

(4.33)

other terms supp { E:>..} is the set of points of non-constancy (or points of increase)
of the spectral family { E>.}. The complementary set in JR is the set of points of con-
stancy of the family { E>.}, i.e. the set of points fl such that EJ-L+E == EJ-L-E for some
E > 0; in each sufficiently small neighbourhood of such a point the spectral fam-
ily does not vary. It is clear that the set of points of constancy is an open set, hence
supp { E :>..} is closed.
An important case is that in which the support of { E:>..} is bounded, hence contained
in a finite interval [M_, M+] for some real numbers M_ and M+ (M_ < M+)· In this
case one has E>. == 0 for each A < M_, E:>.. ==I for each A > M+ and E(V) == 0 if
v n [1\1_, M+] == 0.
One often encounters spectral families that are sentibounded, i.e. bounded from
below or bounded from above. One says for example that a spectral family { E:>..}
is bounded from below if there is a number M_ E JR such that E:>.. == 0 for each
A < M_. Such a number M_ is a lower bound for {E>.}, and the supremum of all
nu1nbers 1\1_ having this property is called the lower bound of { E :>..}.
If the support of { E:>..} is bounded, then E( [M _, M+]) ==I, i.e. E( [M _, M+]) f ==
f for each vector f E H. In the general case there exists no bounded interval J such
that E( J) ==I. However E( J) converges strongly to the identity operator I when J
tends to JR. By taking for example J == [-1\1, M], with M > 0, one obtains as a conse-
quence of (4.17) that s-limJ\;f-too E([-M, M]) ==I, i.e. II!- E([-M, M])JII ~ 0
as 1}1 ~ oo for each f E H. This shows that the set of vectors f of co1npact sup-
port with respect to the spectral.family { E :>..} is dense in H (a vector f is said to have
compact support with respect to { E.x} if there exists a number ME (0, oo) such that
E([-1Vf, M])f ==f).
Example 4.9. Let H == L 2 (1R) and let E:>.. be the operator of multiplication by the
characteristic function X( -oo,:>..] of the interval ( -oo, -A]:

0 if X> A
[E:>..f](x) == { (4.34)
f(x) if X< A

[see ( 1.41)]. The associated spectral measure is easy to write: if V is a Borel set in JR,
then E(V) is the operator of multiplication by the characteristic function Xv of V:

if X tj_ V
[E(V)f] (x) - { .fo(x) (4.35)
if X E V.

One has E(V) == 0 if the Lebesgue measure of V is zero and E(V) #- 0 if the
Lebesgue measure of Vis non-zero [the range of E(V) is the subspace L 2 (V) which
is infinite-dimensional if the Lebesgue measure of V is non-zero]. In particular the
support of this spectral family is the full real line JR, hence it is not bounded [if f-L E JR
SPECTRAL MEASURES 145

and E > 0, then Ep,+c- Ep,-E is the projection onto the subspace L 2 ((J-L- E, J-L c))
which is infinite-dimensional]. The spectral family defined by (4.34) is continuous,
one has E( { -\}) == 0 for each A E JR.

4.2.2 . Let {E,\} be a spectral family and fa fixed vector in a Hilbert space H. Con-
sider the following function defined on JR: 5

(4.36)

This function satisfies the conditions (1)- (3) stated at the beginning of Section 4.1~
with p == 11!11 2 . Indeed: (1) if f-L < A, then Ft(-\) == (f,E,\f) > (J,Ep,f) == Ft(J-L)
because E,\- Ep, is a positive operator, (2) Ft is right continuous as a consequence of
the right continuity of the spectral family { E,\}, (3) Ft( -oo) == lim.\---7-(X) (J,E,\f) ==
(J,Of) == 0 and Ft(+oo) == lim,\--t+(X)\f,E>-f) == (J,If) == 11!11 2 by (4.17). Thus
each vector f E H determines a finite Stieltjes measure on JR, viz. the Stieltjes mea-
sure associated with the non-decreasing function Ft. In what follows we denote this
measure by mf. One has mt((a, b]) == IIE((a, b])fll 2 == (f,E((a, b])f), and more
generally
mt(V) == IIE(V)fll 2 == (f,E(V)f) forVEAB· (4.37)
For J-L E JR the Stieltjes measure mE{Lf determined by the vector Ep,f is obtained from
the Stieltjes measure mf determined by f in the following way:
2 2
mE{Lt(V) == IIE(V)Ep,fll == IIE(V)E(( -oo, J-L])fll
== IIE(V n (-oo, J-L])fll 2 == mt(V n (-oo, J-L]). (4.38)

One can classify the vectors f of the Hilbert space H with respect to the spectral
family { E>.} by having recourse to properties of the associated Stieltjes measures mf
(absolutely continuous or singular continuous or pure point measures). This will be
discussed in detail in Section 4.3; for the moment we consider integrals with respect
to these measures, and more generally with respect to a spectral measure (one then
speaks of spectral integrals).

4 . 2.3. Let A ~ A(A), A ~ B(-\) be functions defined on JR with values in B(H).


One can consider integrals with respect to a spectral measure { E ( ·)} of the type
I A(A)E(dA) or E(dA)B(A) or I A(-\)E(d-\)B(A) [observe that in general the op-
J
erators A(-\) and B(-\) do not commute with the spectral projections E(V)]. If A
and B are sufficiently smooth as functions of A, these integrals will exist in the sense
of Riemann-Stieltjes (as weak or strong or uniform limits of sequences of sums, by a
construction that is analogous to that used in Section 4.1 to define numerical Riemann-
Stieltjes integrals). Such integrals do occur in quantum mechanics, for example in scat-
tering theory, as explained in Section 7 .3. We refrain from giving sufficient conditions
on the functions A ( ·) and B (·) that guarantee the existence of integrals of the type
2
5
If Eisa projection, then \IEfl\ 2 == (Ef,Ef) == (f,E*Ef) = (f,E f) == (f,Ef).
SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

1A(A)E(dA)B(A). We rather discuss the simple situation in which these functions are
multiples of the identity operator, hence of the form cp( ·)I, where cp is a real or com-
plex function defined on JR. So we consider operators the form cp(A.)E(dA.); we 1
assume cp to be continuous (the considerations are easily extended to the case in which
cp is piece-wise continuous, and this will be sufficient for our applications). Under
these conditions one can define 1cp(A)E(dA) in the sense of Riemann-Stieltjes (limit
of sums if the domain of integration is bounded).
(a. b] is a finite interval and II a partition of this interval (as in Section 1.2), we
set
N
~(II; cp)- L cp(ruk)E((sk-1, Sk]). (4.39)
k=1
Clearly ~(II; cp) is a well defined operator in B(H). The fundamental identity is as
follows: let 1/J be another continuous function on JR and cp?/J the product of the functions
cp and 1/J, i.e. (cp1/J)(A.) == cp(A.)1/J(A); then

~(II;cp) · ~(II;ri{;) == ~(II;cp?/J). (4.40)

Indeed one obtains by using (4.24):


N N
~(II;cp) · ~(II;1jJ) == L L cp(u )1/J(uk)E((sj-1, sj])E((sk-1, sk])
1
J=lk=1
N N
== L L cp(u )1jJ(uk)6jkE((sk-1, sk])
1
J=1 k=1
N

== L cp(Uk )1/J(Uk )E( (Sk-1' Sk]) == ~(II; cp?/J).


k=1
Proposition 4.10. Let { E:>..} be a spectral .family in a Hilbert space 7-i, oo < a <
b < +oo and cp: [a, b] ~ C a continuous function.
(a) Let {IIr} be a sequence of' partitions of the interval (a, b] with IIIr I ~ 0 as
r ~ oo. Then the sequence o.f operators {~(IIr; cp) }~ 1 is strongly convergent; its
limit, denoted lab cp(A. )E( dA.), belongs to B(H) and is independent o.f the sequence
{IIr }. Moreover the lilnit operator co1n1nutes with each E:>.. and satisfies

lllbcp(A)E(dA) I
a
= sup
p,E[a,b]nsupp{E).J
IY(tt)J. (4.41)

(b) If cp denotes the co1nplex conjugate of the function cp, then lab cp(A)E( dA) is the
adjoint o.ffab cp(A.)E( dA).
(c) One has for each vector f E H:

I l b
cp(A)E(dA)JII
a
lb 2
= a Jy(AWmt(dA). (4.42)
SPECTRAL MEASURES 147

(d) If 1/J: [a, b] ~ C is a second continuous function, then

(4.43)

(the product of two spectral integrals is equal to a single spectral integral, that of the
product function) .
PROOF. (a) Let II== {so, ... , sn; u1, ... , un} and fr == {so, ... , Sm; fi1, ... , Un~} be
two partitions of the interval (a, b]. By arranging the numbers s 0 == 80 == a, s 1 , ... ,
Sn- 1, sn == sm, == b, 31, ... , S1n- 1 in increasing order, one obtains a subdivision of
the interval (a, b] into N disjoint half-open subintervals ~11 , ... , J N, with N satisfying
max{rn, n} < N < mJ n- 1 (see Figure 4.2). We then have J1 n Jk == 0 if j #- k,
(a, b] == U~J Jt: and E((a, b]) == ~~ 1 E(J£). Hence
n N
~(IT;cp) == Lcp(uj)E((sj- 1 ,sj]) == Lcp(11Jt:)E(Jp),
J=l R=1
m N
~(fr;cp) == L cp(uk)E((sk-1~ .sk]) == L cp(wt:)E(Jt:),
k=1
where W£ coincides with one of the numbers 1L 1 , ... , Un ( VJp == Uj where j is such
that J£ _ ( s j 1 , s j] ), w£ coincides with one of the numbers u1 , ... , fim, and one has
!wR- W£1 < IITI +liT!. Consequently, using (4.30) in the first and in the last step:
N
II[~(IT;cp)- ~(fr;cp)]fll == 2
L \cp(u;t:)- cp(wt:)I 2
IIE(Jp)fll 2
£=1 N
< sup
P=1, .... N
jcp(wp)- cp(1ilt:)l 2
L IIE(Jp)fll
£=1
2

2 2
sup lcp(vJp)- cp(1vp)I IIE((a, b])fll .
£=1, ... ,N

Since cp is continuous, it is uniformly continuous on the closed interval [a, b], hence
the supremum in the preceding expression can be made arbitrarily small by taking the

a= so 83 · · · · · · · · · · · · · · · · Sn =b

a= so s1 83 · · · · · · · · · · sm = b
( x I X I ]

a b
J6 · ............ JN

Figure 4.2
148 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

partitions IT and fr sufficiently fine (recall that lw£ - W£1 < IITI + jiTj). In particular,
if IT and fr belong to a sequence {ITr} satisfying IITr I ~ 0 as r -7 oo, and if E > 0,
one will have II [~(ITr 1 ; cp) - ~(ITr 2 ; cp )]fll 2 < E if r1, r2 > R == R(c ). This implies
the existence of the limit fab cp(A)E( dA) and its independence on the sequence {ITr }.
Since E>- commutes with each ~(ITr; cp ), it also commutes with the limit operator
fabcp(A)E(dA). The remaining statements of (a) will be established at the end of the
proof.
(b) It is clear that [~(IT; cp)] * == ~(IT; cp ). Thus one has for all J, g E 7-i:

(!, [1 b~.p(A)E( dA) ]*g) = « b~.p( A)E( dA)j, g) = }~~ (~(IIr; ~.p )j, g)
== liln (J, ~(ITr; cp)g) == (J,jb ~.p(A)E(dA)g).
T---+oo a

(c) If IT is a partition (a, b] as above, then [using (4.40)]:

II~(IT;cp)fll 2 == (~(IT; cp)f, ~(IT; cp)f) == (J, ~(IT; cp)~(IT; cp)f)


N
== (f, ~(IT; jcpj )f) ==
2
L jcp(uk)l 2 (f,E((sk-1, sk])f)
k=l
N
== L lcp(uk)l 2 mJ((sk-1, sk]). (4.44)
k=l

Consider a sequence {ITr} of partitions of (a, b] with limr---+oo IITrl- 0. Write (4.44)
for each ITr and let r ~ oo. The left-hand side converges to 11fabcp(A)E(dA)fll 2
[use the implication =? of Proposition 1.1 (a)], and the right-hand side converges to
fab jcp(A)j 2 mJ(dA) by (4.13). This proves (4.42).
(d) This result is obtained by using Proposition 2.1 and (4.40):

1b~.p( A) E (d).) ·1 b1j;( A)E (d).) = Sr~;;; ~ (IIr; !.p) · ~.-_!!;;; ~ (IIr; 1/J)
b
==s-lim
T---+ 00
~(ITr;cp)~(ITr;~) ==s-lim
T---+ 00
~(ITr;cp~) ==1 (cp~)(A)E(dA).
a

We finally complete the proof of (a). We set w == sup f-tE [a,b] n supp{E>-} jcp(M) j.
Eq. (4.42) implies that

(see also the Remark following this proof). It follows that llfab cp(A)E(dA) II < w. If
w -1-0, let8 E (O,w). There exists Ao E (a, b) n supp{E>-} with jcp(Ao)l > w-8. Since
the function cp is continuous, there exists c > 0 such that Icp (M) I > w - 28 for M E
[Ao -E, .Ao+c] and such that [Ao c, Ao+c] c (a, b). Clearly E( (.-\ 0 -E, Ao+c]) 1- 0.
SPECTRAL MEASURES 149

For f belonging to the subspace M((Ao - c, Ao c]), one obtains from Eq. (4.42)
6 6
that \1fa cp(A)E(dA)fl\ 2 > (w- 26) 2 \\!11 2 , i.e. \1fa cp(A)E(dA)I\ > (w- 26). As 6 is
arbitrary, we have proved (4.41 ). D

REMARK. Ao is a point in (a, b] which does not belong to supp { E :>..}, then there
existsc > OsuchthatE((A 0 -c,Ao+c)) == O,sothat(Ao-c,Ao+c)nsupp{E>.} ==
0. Then, if in a sequence of partitions {llr }, r is sufficiently large so that IITrl < c,
and if J >.a denotes the interval ( s k-1, s k] of this partition I1r that contains Ao, one has
E( J A.o) == 0; thus the contribution of this interval J >.a to the sum in (4.39) is zero, in
other terms the value of cp at the point Ao (even in some neighbourhood of Ao) has no
6
influence on the integral fa cp (A) E (dA). This integral, in particular its norm as given
by (4.41), involves only the values of cp(J-L) for J-L E supp {E>.}.

Proposition 4.10 gives the essential properties of spectral integrals on a finite in-
terval. We next consider the case in which the domain of integration is JR. We define
J~00 cp(A)E(dA) as the strong limit of J~Mcp(A)E(dA) as JVI ~ oo. One has to dis-
tinguish between different situations. We again assume cp: IR ~ CC to be continuous.
(1) Suppose that the support of the spectral family {EA.} is bounded. Then the limit
M ~ oo is not necessary, i.e. the operator J~4 cp(A)E(dA) is independent of M
forM > Mo if supp {EA.} C [-M0 , Mo]. In this situation f~oo cp(A)E(dA) exists
for each continuous cp: IR ~ CC, belongs to B(H) and has the properties stated in
Proposition 4.10.
(2) Suppose that the function cp is bounded, i.e. \cp (A) I < II cp II 00 < oo for all A E IR.
Again the strong limit of J~M cp(A)E(dA) exists in B(H); to see this it suffices to
show that for each vector f E 7-i one has \\Jab cp(A)E(dA)fll 2 ~ 0 as a, b ~ -oo and
as a, b ~ +oo (always with a< b). By (4.42) one has

rb rb
II'PII~}a I 'PI I~ mf ( (a, b]).
2
II}a cp(A)E( dA)f II < mf ( dA) =

and mf ( (a, b]) ~ 0 as a, b ~ -oo and as a, b ~ +oo since mf is a finite measure on


JR. In this case it is easy to convince oneself that the statements (b) - (d) of Proposition
4.10 remain true for the integral over IR [in other terms one may replace fa by J_~
6

in Proposition 4.10(b)- (d), by assuming also that \\1/J\\oo < oo in (d)].


(3) Finally we consider the situation in which cp is unbounded (a continuous function
cp on IR is necessarily bounded on each finite interval [- M, A1] but could diverge at
infinity, as does for example cp(A) == A2 ). We shall see that in this case (if supp { E>.} is
not bounded) the limit ofJ~ cp (A) E (dA) f exists only for a dense set ( f= 7-i) of vectors
f. Thus J_0000 cp(A)E(dA) will be an unbounded operator, with dense domain Dcp. To
determine this domain we first observe that each vector f having compact support
with respect to {EA.} belongs to Dcp; indeed in this situation there exists a number
Mo E IR+ such that E(( -Mo, Mo])f == f, hence the integral J~1 cp(A)E(dA)j is
150 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

independent of M forM > l'v10 . This shows that Dcp is dense in 7-i. Next suppose that
f belongs to Vcp. Then one must have [use==> of Proposition l.l(a) and (4.42)]:

00 > III: cp(A)E(dA).f 112 = J~oo III: cp(A)E(dA)f 112

= J~oo I:!cp(>.WmJ(dA) = I:!cp(AWmJ(dA).


Consequently f E Vcp ====? J_O: lcp(-A) 2
mt (d-A) < oo. On the other hand, iff is such
l

that this integral is finite, one will have II Jb cp(-A)E(d-A)fll 2 ~ 0 as a, b ~ -oo and
IV! a
as a, b ~ +oo, hence J_M cp(-A)E(d-A)f will be strongly Cauchy as M ~ x. Sum-
1n1ng up:

I: cp(A)E(dA) exists on Dcp = {.f E H. I I: !cp(AWmJ (dA) < oo }. (4.45)

Lemma 4 . 11 . Assu1ne cp: IR ~ C to be continuous. Then f~oo cp(-A)E(d-A) is a closed


operator [on Vcp defined in (4.45)] and its adjoint is given by J_ 00 cp(-A)E(d-A) (with
00

== Vcp)· In particular _{_0: cp(-A)E(d:A) is self-adjoint ~lcp is a reaf.f'unction.

PROOF. We set 1> == J_oc00 cp(-A)E( d:A), with V( 1>) == Vcp.


(i) Let f and g belong to Vcp Vcp. Then [use Proposition 4.10(b)]:

This shows that Vcp C V( 1>*) and that ci>*f == f~oo cp(-A)E( d-A)f iff E V( cp ). Hence
J_
<1?* is an extension of f~oo cp(-A)E( d-A). To see that 1>* == 00 cp(-A)E( d:A), it now suf-
00

fices to show that V( <I>*) _ Vcp. Iff E V( 1>*) there exists f * E 7-i such that

(f,I:cp(A)E(dA)g) = (f*,g)

in particular each g having compact support with respect to { EA}· By taking into
account (2.6), we then have for each ME (0, oo ):
SPECTRAL MEASURES 151

1\E((-M, J\;f])f*ll == sup \ (E(( -M, M])f*, g)\


gEH.JJgJJ=l

= sup
gEH.JJgJJ=l
IU*,E(( -M, M])g) I = sup
gEH,JJgJJ=l
Ju,jM <p(A)E(dA)g) I
-!vi

== sup
gEH,JJgJJ=l
l\j·M <p(A)E(dA)f,g)l 11/M <p(A)E(dA)JIII
-M
==
-M

It follows that sup M>O J:~\cp(-A)I 2 mt(d-A) < 11!*11 2 < oo. Consequently we have
f~oolcp(-A)I 2 mt(d-A) < oo. This shows that f E D(<I>*) ====? f E Vcp == Dcp.
(ii) Since f~oo cp(A.)E(d>..) is the adjoint of J_
00

00
cp(A.)E(d>..), it is a closed operator
[see Proposition 2.19(a)]. D

The following consequence of (4.24) will occasionally be used further on: if f is a


vector such that E((b, c])f == f for some numbers b < c, then f~oo cp(A.) (dA.)f ==
Jbc cp( A.)E( dA.) J.

4.2.4. We proceed by discussing the significance of the preceding developments for


the theory self-adjoint operators. Let us consider the function cp(A.) == >... {EA}
is a spectral family, then
00

00
J_
A.E (dA.) defines a self-adjoint operator A, called the
self-adjoint operator associated with { E.A}, with domain

(4.46)

and one has 6

(g,A.f) = 1: A(g,E(dA)f) VgEH, VfED(A). (4.47)

The operator A belongs to B(H) if and only if the support of { EA} is bounded,
which case II A II == rnax {I M -I, [1\!I+I}, where and M + are the lower bound and
the upper bound respectively of { EA} [one has \\A II ==sup {LEsupp{EA} IMI by (4.41)].
So each spectral family determines canonically a self-adjoint operator by the pre-
ceding description. The Spectral Theorent asserts the converse: given a self-adjoint
operator in a Hilbert space H, there exists a unique (right continuous) spectral fam-
ily {EA}, called the spectra/family of A, such that A==
00

00
J_
AE(d>..). Thus there
is a one-to-one correspondence between self-adjoint operators and right continuous
spectral families! We shall occasionally write {Ef} (rather than simply { EA}) for the
spectral family of a self-adjoint operator A.
6
If g #- f, the integral in (4.47) is with respect to a cmnplex measure (g, E (·)f). One can interpret
(4.47) as the scalar product between the vectors g and J~oo :AE( d:A)f, or one may decompose (4.47) into
a sum of four Stieltjes integrals by taking into account the polarisation identity (2.15).
152 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

The Spectral Theorem is a fundamental result in the theory of operators in a com-


plex Hilbert space (it generalises the fact that a Hermitian n x n matrix is diagonal-
isable ). One finds in the literature various different proofs of this result; they are all
rather long, and we shall discuss one of them in Section 4.4 and present another one
in the Appendix of this chapter. For the moment we assume that this result is known
and consider some aspects of the functional calculus of a self-adjoint operator. A
comment on the quantum-mechanical meaning of the spectral family of a self-adjoint
operator A is made at the end of §4.3.4.
So let A be a self-adjoint operator and { E>.} its spectral family. Then, as seen
before, one can associate with each continuous function cp defined on JR. an opera-
tor f~oo cp(>..)E(d>..). Henceforth this operator will be denoted cp(A), and we now
try to justify this notation. In order to avoid domain questions, we assume A to be
bounded, so that cp(A) E B(H) for each cp. Since A== J~00 >..E(d>..), one has by Eq.
(4.43): A 2 == [ J~00 >..E(d>..)] == J~00 ).. 2 E(d>..), which does coincide with cp(A) for
2

00
J_
cp(>..) == >.. 2 . Similarly one will have Ak == 00 )..k E(d>..) for cp(>..) == )..k fork ==
1, 2, 3, .... We shall also discuss other functions cp that justify the notation cp(A) for
f~oo cp(>..)E(d>..) [for example cp(>..) - (>..- z)- 1 for the resolvent of A: we shall see
J_ 00
that (A z) - 1 == 00 (>..- z) - 1 E (d)..)]; for more general functions cp one should take
this notation as a definition.
If the number J-L belongs to the support of the spectral family of A and if c > 0,
then E((M- c, J-L + c]) # 0, and iff E M((M- c, f-L + c]), then

if c is sufficiently small. (4.48)

Thus A is approximately multiplication by J-L in M ( (M - c, J-L c]) if c is sufficiently


s1nall. A similar argu1nent shows that f~oo cp(>..)E(d>..) is approximately multiplica-
tion by cp (M) in M ( (tt - c, J-L + c]), thus justifying the notation cp (A) for this operator.
One should also note the analogy with multiplication operators discussed in Section
2.5; the Spectral Theorem is a more precise statement of the fact, often cited by physi-
cists, that a self-adjoint operator can be diagonalised. 7

7 Since the operator A is essentially multiplication by JJ on subs paces of the form M ( (JJ - c, JJ + c]),
it can be approxitnately diagonalised by covering its spectrum by a disjoint union of intervals of length 2s,
with very sn1all c, and writing }-{ as the direct sun1 of the subs paces M ( J) associated with the intervals
J in this partition. To obtain an exact diagonalisation one would have to consider the limit where c ----7 0.
Now, if JJ is a point in the continuous spectrmn of A, then M ({JJ}) will not be a subspace of J-i, so that the
tnentioned litnit has to be interpreted carefully. This can be done by using the notion of a direct integral of
Hilbert spaces which generalises that of direct sums. Without entering into details we mention that each self-
adjoint operator A is diagonalisable in the sense that it is unitarily equivalent to the multiplication operator
by the (real) variable in s01ne direct integral, called a spectral representation of A. As a rather simple
example we shall treat in Chapter 5 the diagonalisation of the free Hamiltonian H o, see the discussion
under point (2) on page 229 and the ensuing fonnulas for the spectral representation of H o. In this example
the direct integral reduces to a space of the form £ 2 ( 0; JC, m) [see the end of§ 1.4.3], with 0 = [0, oo) =
the spectrum of H o. A formulation of the diagonalisation of a general self-adjoint operator is given for
example in Volume I of [RS].
SPECTRAL MEASURES 153

We return to functional calculus by also admitting unbounded operators A. We use


the fact (which will be proved in Example 4.13) that the support of the spectral family
of A coincides with its spectrum: supp { E-\} - a(A), and we denote by BC(IR) the
set of bounded continuous complex-valued functions defined on ffi.. The set BC(IR)
has the structure of an algebra, i.e. it is a linear vector space over the field CC equipped
with an operation of multiplication. We have the following results on the bounded
functional calculus of A:
(1) With each bounded continuous function cp is associated an operator cp(A) in B(H),
given by

c,o(A) -I: c,o(A)E(d>.). (4.49)

cp(A) commutes with each E-\, its adjoint is given by

cp(A)* =I: cp(A)E(dA) = cp(A), (4.50)

and one has


llcp(A)II- sup lcp(A)i. (4.51)
-\Eo-( A)

(2) The mapping cp cp(A) defines a homomorphism from the algebra BC(ffi.) into
f-t

B(H), in other terms: if cp, 1/J E BC(IR) and a E CC, then

(acp + 1/;) (A) - acp(A) + 1/;(A) and (cp1/J) (A) == cp (A) 1/J (A) (4.52)

[the operations on the left of the equality signs are in BC(IR), those on the right of
these signs are in B(H)]. In particular one has

cp(A)1j;(A) - 1/;(A)cp(A) \1 cp, 1/J E BC (IR). (4.53)

(3) cp(A) is self-adjoint if and only if cp(A) is real for A E o-(A).


(4) cp(A) is unitary if and only if lcp(A) I == 1 for A E a( A).
The result of (4) is obtained from (4.50) and (4.52): cp(A)*cp(A) == cp(A)cp(A)*
(cpcp) (A), which is equal to the identity operator I if and only if rpcp 1 on the
spectrum of A.
If cp is continuous but unbounded, then Eq. (4.49) defines a closed operator cp(A)
with domain

V(cp(A)) = {f E H II:Ic,o(AWmt(dA) < oo }. (4.54)

The relation (4.50) remains true for unbounded cp, however (4.52) is replaced by
acp(A) + 'ljJ(A) C (acp + 1/;)(A) and cp(A)1j;(A) C (cp'ljJ)(A) (it may happen for
example that 1/J is unbounded but cp1/J is bounded).
154 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

Example 4.12. Continuous unitary group associated with A


We consider the function <pt (A) == e-i>-.t, with t E JR, and introduce the notation Ut for
the operator <pt (A): Ut - e- iAt. By (4) above, Ut is a unitary operator. The family
{Ut} forms a group, more precisely a representation of the additive group JR, i.e.

UsUt =I: e-iAsE(dA) I : e-il"tE(dJL) =I: e-iAse-tAtE(dA)


= I:e-tA(s+tlE(dA) = Us+t·

particular Ut == [Ut] 1 == U-t· Also the family {Ut} is strongly continuous in t.


To see this one can use the Do1ninated Convergence Theorem (Proposition 1.9): since
je-i>-.(t+c) - e-i>-.t 2 < 4 one obtains
1

One can avoid the application of the Dominated Convergence Theorem by arguing as
follows. For }vf > 0 set El\ 1 == E(( -M, M]) and write
2
IJUt+cf- Utfll == II[Ut+c- Ut]E1V1fll
2
+ II[Ut+c- Ut][J- E1V1]fll 2 ·
Let 6 > 0. Since El\1 converges strongly to I as 1\II ---7 oo and II U7 II == 1 for each T,
one can choose 1\;f sufficiently large so that II [Ut+c - Ut][I- E1V1 Jfll 2 < 6/2 for all
and all E. With this fixed M one has

JJ[Ut+c- Ut]EMfll 2 = I:Je-tA(t+c)- e-"A 1 J2 mt(dA)


2
=I:)e-i>-.c -lJ rnt(dA) <I:JAcJ 2 rnt(dA) < c2 M 2 JI!IJ 2 ,
which is less than 6/2 for all E satisfying lsi < Eo for a suitable Eo > 0. It follows that
IIUt+cf- Utfli 2 < 6 for each lsi <Eo, thus proving the strong continuity of {Us} at
s == t.
The function YJt(A) == e-i>-.t can be written as a series <pt(A) == I:~ 0 ( -iAt)kjk!
One can ask if Ut is given by the corresponding series of operators, i.e. if

When A belongs to B(H), the preceding series converges in the sense of uniform
convergence, and Ut is given by this series. If A is unbounded, the terms in the series
make sense only on vectors f belonging to D(Ak) for each k 1, 2, 3, ... The set
of these vectors is dense in H, as it contains all vectors having compact support with
respect to the spectral family { E>-.} of A. Iff is a vector of compact support with
respect to {E>-.}, then the series I:~ 0 [( -it)kjk!]Akf is strongly convergent and its
sum equals Ut f. The verification of these statements is left as an exercise.
SPECTRAL MEASURES 155

Example 4 . 13 . Resolvent of A
If z E CC with 8z =/- 0 or if z E IR\supp{E-\}, then the function 1/Jz(A) == (A z)- 1
is continuous and bounded on the support of {E-\} (we recall that IR \supp{E-\} is
an open set). Thus 1/Jz (A) E B(H), with 111/Jz (A) II == sup fLEsupp{E>-} IJL - zl 1 -
[dist(z, supp{E-\} )]- 1. 8 One should use the notation (A- z)- 1 for 1/Jz(A). For this
to be justified we show that 1/Jz (A) is the resolvent of A at the point z.
(i) Let f E D(A). Then

11'1/Jz (A)fll
2
= i: I (/1-
2
z) l mt (df1) > [dist(z, supp { E>J)]
2
2
/ _ : mt( df1)

== [dist(z,supp{E-\})] 11!11 2 .
Thus: if 1/Jz(A)f == 0, then 11!11 == 0, hence f == 0, i.e.1/Jz(A) is invertible.
(ii) It remains to be seen that (A- z )1/Jz (A)g == g for each g E H [which implies
that R(A z) == H] and that ?f;z(A)(A- z)f == f for each f E D(A). Let us check
for example the second relation. Let f be a fixed vector in D(A). One has 1/Jz(A) -
s-lhnM-+oo f-~;f(A- z)- 1 E(dA) and (A- z)f == s-limM-+oo f-~;f(A- z)E(dA)j.
By using first Proposition 2.1 [or (2.16) ifA is unbounded] and then (4.43) one finds
that
!1\;f
J
M
1/Jz(A)(A- z)f == s-lirn [ (A- z)- 1 E(dA) · (JL- z)E(dtt)J f
NI -+oo -M -1\;f
NI
== s-li1n
M-+oo J-M
(A- z)- 1 (A- z)E(dA)j == s-limE((-M,M])f- f.
M-+oo

Wehavethusjustifiedtheidentity (A z)- 1 - 00
J_00
(A z)- 1E(dA) ifSSz =J Oorif
z E IR\supp{ E-\ }. The set of these values of z coincides with the resolvent set p(A) of
the operator A. Indeed, the preceding considerations show that IR\supp{ E-\} C p(A);
on the other hand, if Ao is a real point in p( A), then (A - Ao - is) - 1 has a limit in
B(H) as E ---7 0; however, if A E supp{ E-\}, one has II (A- A- is) - 1 11 == 1/lsi, which
shows that supp{ E-\} %p(A). In conclusion, at each z E p(A) the resolvent of A is
given by

(4.55)

and satisfies
(4.56)
If A is bounded and if }vf_ and M + denote the lower and upper bound respectively
of { E-\}, it is clear that o-(A) C [1t1_, M+] C [-II A II, I A II]. If A is semibounded, for
example bounded below, one has a(A) C [M_, oo ), where Af_ E IRis the supremum
of all real numbers M such that EA == 0 for each A < M.
8
For a real z the function 1/Jz is not continuous at,\ = z. However, if z E lR \ supp{ E,A. }, then the
point z is at positive distance from supp{ E)..}, and one obtains the same operator 1jJ z (A) if one modifies
the function ?jJ z in a sufficiently s1nall neighbourhood of z in such a way that the modified function is
continuous and bounded.
156 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

Remark 4.14. (a) So far we have considered integrals Jcp(.\)E( d.\) for continuous
functions <p, which permits their definition in the sense of Riemann-Stieltjes. One can
extend this functional calculus to functions that are not necessarily continuous by us-
ing (numerical) integrals of the Lebesgue-Stieltjes type. If for example <p: JR ---+ CC
is a bounded measurable function, then the integral J~CX) cp(.\)mJ (d.\) exists for each
vector f E Hand defines the quantity (f, cp(A)f). By applying the polarisation iden-
tity one obtains an expression for (g, cp(A)f) for all j, g E H, and in this way one
defines a bounded linear operator cp(A) with llcp(A)II < sup-\Eo-(A) lcp(.\)1. If <pis the
characteristic function Xv of a Borel set V E AB, then cp(A) xv(A) == E(V) is the
spectral projection of A associated with V. For certain locally unbounded functions
[e.g. cp(.\) == (.\- .\ 0 )- 1 for some .\ 0 E o-(A)], one can also give a meaning to cp(A)
as an unbounded operator by conveniently defining its domain.
(b) If the Fourier transform(/) of <p belongs to L 1 (JR), then cp(A) admits the following
representation in terms of the unitary group { e-iAt} associated with A:

cp(A) _1_ Joo ijJ(t)eiAtdt. (4.57)


V21T -CX)

Indeed

(.f, cp(A)f) = ( cp(A)mt(dA) = ( mt(dA) ~ ( ei>-tij)(t)dt.


JJR JJR y J
21f JR

By interchanging the order of integration (Fubini's Theorem) one arrives at

(f,cp(A)f) = ~ ( dtij)(t) ( eiAtmt(dA) = ~ ( dtij)(t)(f,eiAtf),


Y 21r JIR JIR Y 21r JIR
and (4.57) follows by applying the polarisation identity.
Example 4 . 15.. Spectral family of Q
The spectral family of the position operator Q in £ 2 (JR) has already been indicated
in (4.34). In this case cp(Q) is the operator of multiplication by cp(x). To illustrate
this fact, let us denote this multiplication operator by <I> and consider for example an
element f of £ 2 (JR) such that f (x) 0 almost everywhere outside some finite interval
[a~ b], and assume <p to be continuous. If II is a partition of (a, b] as in Section 1.2, one
has [<I> f- ~(II; cp)f] (x) == [cp(x) cp(uk)] f(x) if x E (sk-1, sk], and consequently

For given E > 0, one can choose the partition II so fine that lcp(x) - cp( uk) I < E for
all x E ( s k-1, s k] and all k - 1, ... , N, and then one will have II <I> f - ~(II; <p) f 11 2 <
6
c 2 ll f 11 2 . This implies that <I> f == fa cp( .\)E( d.\) f cp( Q) f [here { E( ·)} is the spec-
tral family of the operator Q given in (4.35)].
Example 4.16. Spectral family of Q2
The operator Q 2 in L 2 (JR) is multiplication by x 2 • The projection E-\ will be a mul-
tiplication operator; one must have (E-\f)(x) - f(x) if x 2 < .\and (E-\f)(x) == 0
SPECTRAL PARTS OF A SELF-ADJOINT OPERATOR 157

otherwise. Thus EA. == 0 if A < 0, whereas for A > 0, EA. is the operator of multipli-
cation by the characteristic function of the interval [-viA, viA]:

(4.58)

One can also consider the analogue in £ 2 (JR 11 ). As seen in Example 2.30, the position
operator Q is then a multi -component operator: Q == (Q 1 , ... , Qn), where Qj is the
multiplication operator by x j:

(4.59)

Q2 Q2 is the operator of multiplication by x 2 = xi+ ·· · x;. Again one has


EA - 0 if A < 0 and, for A > 0: (EA.f)(x) - f(x) if x < A and (EA.f)(x) - 0
2

if x 2 > A. Thus, for A > 0, EA. is multiplication by X[o,A.lf2](1xl), in other terms:


-+ -+2
EA.== X[o,A. 1 12J(IQI)- X[o,A.](Q ).
Example 4.17. Spectral family of P 2
We have seen in §2.5.2 that the momentum operator Pin £ 2 (JR) is given by :F*Q:F, in
other terms P is the multiplication operator by the variable k on the Fourier transforms
1 1(
of the functions f E L 2 (JR): [:F P f] (k) == k k). Therefore the spectral family of
P 2 has the same form as that of Q 2 given in Example 4.16, except for the replacement
of the wave functions by their Fourier transforms:

(4.60)

In £ 2 (JRn) the momentum operator Pis an n-componentoperator: P == (P1 , ... , Pn),


where Pj is given by

(4.61)

and where 1(k) denotes the (n-dimensional) Fourier transform off [see (2.79)]. The
operator P 2 P 2 is multiplication by k 2 ki + · · · + k; on ](k): [:FP 2f](k) ==
k 2 ](k), and its spectral family is given as follows: EA.- 0 if A< 0 and, for A> 0:
[FEA.f](k) == ](k) if k 2 < A and [FEA.f](k) == 0 if k 2 > A. Thus, for A > 0:
2
EA. X[o,fi](IPI)- X[o,A.](P ).

4.3 Spectral parts of a self-adjoint operator

Let A be a self-adjoint operator, {EA.} its spectral family and o-(A) its spectrum. In
quantum mechanics one generally distinguishes between the point spectrum (often
called "discrete spectrum"), composed essentially of the eigenvalues of A, and the
continuous spectrum. We give here definitions and precise characterisations of the
spectral parts of self-adjoint operators. We should mention that one can use different
methods for subdividing the spectrum, and the spectral parts obtained by means of
158 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

different definitions are not necessarily identical. We present here a subdivision of


the spectrum based on a decomposition of the Hilbert space. Other definitions use for
example properties of the operator A- ,\for,\ E o-(A) [A- ,\not invertible, A - A
invertible but with unbounded inverse and with V( (A - ,\) -l) dense or non-dense in
7-i].
We have seen in §4.2.2 that each vector f E H determines a Stieltjes measure mf
by the formula mt(V) == (f,E(V)f) == IIE(V)fll 2 for V E AB, and by (4.6) this
measure has a unique decomposition into a sum of three measures mf,p, mf,sc and
m'f, ac. It is very interesting to know that each of these three measures coincides with
the Stieltjes measure associated with some vector in 7-i, in other words that there exist
three vectors f p, fsc and f ac in H such that f == f p + fsc + f ac and such that the part
mf,p of the Stieltjes measure associated with the vector f coincides with the Stieltjes
measure associated with the vector fp (i.e. mf,p == mfp ), etc.
We shall see that the decomposition of the vectors f E H into f - fP + fsc + !ac
induces an orthogonal decomposition of the Hilbert space H into three subspaces
that one denotes by Hp(A), Hsc(A) and Hac(A). The notation indicates that this
deco1nposition depends on the considered spectral family { E-\}, or equivalently on
the self-adjoint operator A associated with this spectral family. Thus each self-adjoint
operator A in H induces a decomposition of H into

(4.62)

in such a way that mg is a pure point measure (a measure that is supported on a count-
able set of points) if g E Hp (A), a singular continuous measure if g E Hsc (A) and an
absolutely continuous measure if g E Hac(A) (assuming g -1- 0).

4.3.1. We begin by the characterisation of Hp (A). Let us first show that (with 11) E IR)

If EM( {~L}) <¢==? f E V(A) and A.f = JLf.l (4.63)

To see this we use the following formula which is obtained from (4.42) with (a, b) -
(- x ~ oo), f E V (A) and ~ (,\) == ,\ - f-L:

IIA.f- PJ!I 2 = L:!A -~LI 2 rnt(dA). (4.64)

Since the integrand is strictly positive at each,\ -1- JL, one can have IIAJ p,fll 2 - 0
if and only if m'i (V) == 0 for each Borel set V with f-L tf_ V, i.e. the measure mf is
supported on {J_L }. 9
The relation (4.63) expresses the fact that a number ,\ E IR is an eigenvalue of A if
and only if dim M ( { ,\}) -1- 0, and in this case M ( {A}) is the subspace of H spanned
by the set of eigenvectors of A associated with the eigenvalue ,\. Since the set of points
of discontinuity of a spectral family { E-\} is countable [as observed after (4.32)], the
9
If f E M ({p,}), then the measure m f is supported on the set {tL} consisting of the single point /L:
rr~f is proportional to the Dirac measure at the point fL, formally m 1 (dA.) = ll.fii 2 5(A.- tL)dA..
SPECTRAL PARTS OF A SELF-ADJOINT OPERATOR 159

eigenvalues of a self-adjoint operator form a countable subset of JR. 10 In the sequel we


denote the eigenvalues by { )q, A2 , ... } . The set of all eigenvalues of A is called the
spectrum of A.
One then defines Hp (A) as the subspace of H spanned by the set of all eigenvectors
of other terms
(4.65)

where sum extends over all eigenvalues of A. Thus a vector f belongs to Hp(A)
if and iff- ~jfJ with Afj == AJjJ and with ~jllfjll 2 < oo. It is clear that
f E Hp (A) if and only if there exists a countable set W of real nu1nbers such that
mf (JR\ W) == 0, in other terms such that E(W)f == f. We write Ep for the projection
onto subspace Hp(A):

Ep == PHp(A) == LE({AJ}). (4.66)


j

4 . 3 . 2. We continue by defining the other subspaces occurring in (4.62). We denote by


lVI the Lebesgue measure of a Borel set V. One sets
• Hae(A) == {f E H mt is an absolutely continuous measure}
{f E H the function A r--+ IIEAfll 2 is absolutely continuous}
• Hse(A) == {f E H mf is a singular continuous measure}
{f E H the function A r--+ IIE-\!11 2 is continuous and 3 V E AB with
lVI == 0 and E(V)f == f}
• He(A) == {f E H mt is a continuous measure}
{f E H mf ( {A}) == 0 vA E JR}
{f E H the function A r--+ IIEAJII 2 is continuous}
{f E H A r--+ E-\f is strongly continuous} 11
• Hs (A)== {f E H the measure m f and the Lebesgue measure are mutually singular}
J

{f E H J3 V E AB with lVI == 0 and E(V)f == f}.


We next discuss the properties of these subsets of H. The following inclusions are
evident:
Hae(A) c He(A), Hse(A) c He(A), Hp(A) c Hs(A), Hse(A) c Hs(A).
(4.67)
Let us verify first that He (A) == Hp (A) _L [which also shows that He (A) is a
subspace]. Iff E He (A), then clearly E( {A} )f - 0 for each A E JR, hence Epf == 0
and consequently f _L Hp(A). Conversely, iff E Hp(A)j_, then Epf == 0, hence
E( {A}) f == 0 for every eigenvalue A of A and consequently for each A E JR [by
(4.63)]; thus mt ({A})== 0 for each A E JR, i.e. f E He (A).
10 This also follows from Proposition 2.36(b) (orthogonality of the eigenvectors associated with different
eigenvalues) because we assutne that each basis of H is countable [condition (H4) for separability of H in
§1.1.1].
11 See Problem 4.2(b) for the last identity.
160 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

Let us now show that Hse(A) is a subspace. If j, g E Hse(A) and a E CC, then
clearly E>.. (f +ag) == E>.. f +aE>..g is strongly continuous~ also there exist V, WE AB
with lVI == IWI == 0 and E(V)f- f, E(W)g == g, and then one has IVUWI - 0 and
E(VUW) (f +ag) == f +a g. This shows that f +agE Hse(A), hence that Hse(A) is
a linear manifold. To see that it is a subspace, consider a strong Cauchy sequence {fk}
belonging to Hse (A) and denote its limit by f; we have to show that f E Hse (A). We
know already that f E He (A) because He (A) is a subspace. For each k E N there exists
a Borel set vk with IVkl == 0 and E(Vk)!k- fk· Then v :- uk
vk is a Borel set with
lVI < Lk IVkl 0, andonehas E(V)f s-limk-+ooE(V)fk == s-limk-+oofk- f.
Thus we have shown that f E Hse(A).
We next show that, iff E He(A) and V E AB with lVI == 0, then one has E(V)f E
Hse (A). By the assumption made on f, we have mf ( {A}) - liE( {A} )f 11 2 == 0 for
all A E JR. Hence mE(V)J({A})- IIE({A})E(V)fll 2 == IIE(V)E({A})fll 2 - 0 for
all A E JR, i.e. E(V)f E He(A). Now E(V)[E(V)f] == [E(V)] 2 f == E(V)f. Since
lVI - 0, the claim is proved.
We now assume that f E He(A) and f _L Hse(A) and show that then f E Hae(A)
[which implies that Hac (A) is a subspace, the orthogonal complement in He (A) of
Hse(A), i.e. one has He(A) == Hse(A) E8 Hae(A)]. We must prove that E(V)f == 0
for each Borel set V with lVI == 0. We set E(V)f - g. One has g E Hse(A) by the
result of the preceding paragraph. So it is enough to prove that g _L Hse (A). For this
lethE Hse(A). Then (again by the preceding paragraph) E(V)h E Hse(A), hence
(h, g) == (h, E(V)f) == (E(V)h, f) == 0 because E(V)h E Hse(A) and f _L Hse(A).
The results obtained so far prove completely Eq. (4.62). We mention the identity
Hs(A) == Hp(A) E8 Hsc(A) the proof of which we omit (Problem 4.3). These results
will be collected in the first part of Proposition 4.19.

Proposition 4. 18. Let A E JR. Then:

jEHp(A) ====? E>..fEHp(A)


jEHae(A) ====? E>..fEHae(A) (4.68)
{
f E Hse(A) ====? E>..f E Hse(A).

PROOF. (i) For the first relation it is enough to see that, if f is an eigenvector of A
(say Af - J-Lf for some J-L E JR), then E>..f is also an eigenvector of A. Now the
equation Af == JLf implies that E( {J-L}) f - f [see (4.63)]; thus E( {JL}) [E>..f] ==
E>..E( {p,} )f - EA.f, so [again by (4.63)] the vector E>..f is indeed an eigenvector of
A (associated with the same eigenvalue J-L), or else E>..f - 0. 12
(ii) Let us show for example that, iff E Hse(A), then E>..f E Hse(A). For this
let g E Hp(A)~ then one has (E>..f, g) == (J, E>..g) == 0 because E>..g E Hp(A) by
(i) and f Hp(A). This shows that E>..f E Hp(A)_i == He(A). Since f E Hse(A)
there exists a Borel set V with IV I - 0 and E (V) f == f. As in (i) one obtains that
E(V)[E>..f]- E>..E(V)f == E>..f, hence E>..f belongs to Hse(A). D
12
Evidently E>.f can be the zero vector; indeed this happens if and only if 1-L > A because E>.E( {!-L}) =
0 if 1-L > A, whereas E>.E( {!-L}) = E( {!-L}) if 1-L :::; A.
SPECTRAL PARTS OF A SELF-ADJOINT OPERATOR 161

The preceding proposition states that each of the subspaces Hp (A), Hsc (A) and
Hac (A) is invariant under the spectral family (hence also under the spectral measure)
of A. Iff is a vector in H and if one writes f == f P + fsc + f ac for its decomposition
in the representation (4.62), then the measure mf is the sum of three measures mtp'
m !sc and m f ac determined by the three components of f:

mf (V) == mfp (V) +mise (V) + 1Y~fac (V) (4.69)


Indeed

rnt(V) == (j,E(V)f) (fp !sc + !ac,E(V)fp + E(V)fsc + E(V)fac)


==
== (jp,E(V)jp) + (fsc,E(V)fsc) + (fac,E(V)fac)
== mtp(V) + m!sc(V) + mfa,(V),
because the cross terms in the last scalar product of the first line are zero [they involve
vectors that are orthogonal to each other by Proposition 4.18: E(V)fp E Hp(A),
E(V)fsc E Hsc(A) and E(V)fac E Hac(A)].

4.3.3. We now study the action of the operator A on the vectors belonging to the dif-
ferent subspaces introduced above. We shall see that A leaves each of these subspaces
invariant. If M is a subspace of H we denote by AM the restriction of A to M:

V(AM) == M n V(A) and (4.70)

Proposition 4.19. Let A be a self-adjoint operator in a Hilbert space H.


(a) H can be decomposed as follows:
H == Hp(A) Hsc(A) Hac(A) == Hp(A) E8 Hc(A) == Hs(A) Hac(A). (4.71)
The restriction of A to one of the sub spaces appearing in (4. 71) defines a se(f-adjoint
operator in this subspace. If these restrictions are denoted by Ap, , Aac, Ac and
As, then
A== Ap Asc Aac Ap Ac As Aac· (4.72)
(b) If rp is a bounded continuous function on JR, then each of the subspaces in (4. 71)
is invariant under rp(A):
(4.73)
Moreover one has
a( A) == a( Ap) U a(Asc) U a(Aac). (4.74)
In what follows we denote the projections onto the subspaces appearing in (4. 71)
by Ep, Esc, Eac, Ec and Es respectively (all of these operators depend on A Each of
the five operators Ap, Asc, , Ac and As introduced in (a) of the preceding proposi-
tion is self-adjoint as an operator in the respective subspace [for exatnple Asc is a self-
adjoint operator in Hsc(A), defined by V(Asc) == Hsc(A) n V(A) and Asci == Af
for f E V(Asc)]. Eq. (4.74) expresses the spectrum of A (a self-adjoint operator
H) in terms of the spectra of Ap, Asc and Aac (viewed as self-adjoint operators in the
respective subspace).
SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

PROOF OF PROPOSITION 4.19. Equation (4.71) has already been established.


Let f E D(A) and write f == JP + fsc -f- fac == Epf +Esc!+ Eacf in the first
decomposition of H in (4. 71 ). By (4.69) one has

2
: A mJ"(dA) + £: 2
A mfsc(dA) I: 2
A mfa,(dA) =I: 2
A mt(dA) <
(4.75)
00.

This implies that each of the vectors JP, !sc and fac belongs to D(A). We claim that
Afp E Hp(A), Afsc E Hsc(A) and Afac E Hac(A). Let us verify for example the
last inclusion: for each 1\1 > 0, let {II~1 } be a sequence of partitions of the inter-
val ( 1\1] with III~1 1 ~ 0 as T ~ oo. Let~~ be the operator associated with
the partition {II~} [defined as in (4.39) with II == II;r
and cp( ,\) == ,\] and ob-
serve that this operator maps the subspace Hac(A) into itself [because each E(V) has
this property by (4.68)]. Thus ~;1fac E Hac(A). Since Hac(A) is a subspace, hence
closed, one has
Afac ==s-lim [s-lim ~~1fac] E 'Hac(A).
Jvf -+(X) T-+(X)

In a similar way one sees for example that cp (A) f ac E Hac (A).
(ii) The preceding considerations show that each f E D(A) has a unique decom-
position f == JP + fsc + fac with fp E D(Ap), fsc E D(Asc) and facE D(Aac), and

AJ - Apjp + Af:>cfsc Aacfac·


confirms the decon1position (4. 72) of A and also implies that, for z E CC:

(A- z)f == (Ap- z)fp + (Asc- z)fsc + (Aac- z)fac· (4.76)

(iii) It is clear that the operators Ap, and Aac are symmetric [for example: if
f. g E D(Ap) then (f,Apg) == (f,Ag) == (Af, g) == (Apf, g)]. To see that they
are self-adjoint, it suffices to show that R(Ap - i) == R(Ap + i) == Hp(A), etc.
(see Proposition 3.3). Now, since A is self-adjoint, we have R(A i) == H, i.e.
i)D(A) == H. By (4.76) this is possible only if (Ap i)D(Ap) == Hp(A),
i)D(Asc) == 'Hsc(A) and (Aac i)D(Aac) == Hac(A), which establishes the
self-adjointness of , Asc and Aac.
(iv) Eq. (4.76) shows that A- z is invertible if and only if each of the three operators
on its right-hand side invertible, and if this is the case then one has (A - z) - 1 ==
-z)- 1 (Asc-z) 1 (Aac-z) 1 .ThisimpliesthatzEp(A)ifandonlyif
z E p(Ap) n p(Asc) n p(Aac ). It follows that

a-( A) == CC\p(A) == CC\ {p(Ap) n p(Asc) n p(Aac)}


== [CC\p(Ap)] U [CC\p(Asc)] u [CC\p(Aac)]
== 0" (A p ) U 0" ( Asc ) U 0" ( Aac) · D

Let us now introduce the spectral parts of a self-adjoint operator A. Except for the
point spectru1n, they are defined in terms of the spectra of the restrictions of A to the
subs paces occurring in the decomposition (4.62) of the Hilbert space:
SPECTRAL PARTS OF A SELF-ADJOINT OPERATOR 163

• point spectrum: C5 P (A) == set of all eigenvalues of A,


• continuous spectrum: C5 e (A) == C5 (Ae) == spectrum of the self-adjoint operator Ae
in He [if dim He(A) -/= 0],
• singular continuous spectrum: C5se(A) - CY(Ase) [if dim Hse(A) -/= 0],
• absolutely continuous spectrum: C5ae(A) - CY(Aae) [if dim Hae(A) -/= 0],
• singular spectrum: C5 8 (A) == CY(As) [if dim Hs(A) -/= 0].
In the case in which the dimension of the subspace involved is 0, one says that the
respective part of the spectrum is empty. For example, if Hse (A) == {0}, one sets
C58 e(-4) == 0.
The last four parts of the spectrum [C5e(A), C5se(A), C5ae(A) and C58 (A)] are closed
subsets ofJR because each of them is the spectrum of a self-adjoint operator in a certain
Hilbert space. The point spectrum however is not closed in general (see the examples
given in §4.3.4); the spectrum CY(Ap) of the self-adjoint operator Ap in Hp(A) is equal
to the closure of the set of eigenvalues of A:

(4.77)

Thus CY( Ap) is composed of the eigenvalues of A and of their accumulation points [if
A is a real number such that dist( A, C5P (A)) > 0, then A E p( Ap); the verification is
left as an exercise, see Problem 4.12)].
By (4. 74) the spectrum of A is the union of the different spectral parts, more pre-
cisely

(5 (A) == (5 P (A) U (5 se (A) U (5 ae (A) == (5 P (A) U (5 e(A) == (5s (A) U (5 ae (A) . (4. 78)

We say that A has pure point spectrum if Hp(A) == H, that it has purely continu-
ous spectrum (or that its spectrum is purely continuous) if He (A) == H and that the
spectrum of A is purely absolutely continuous if Hae(A) == H. One says that the
continuous spectrum of A is absolutely continuous if Hse (A) == {0}.
If U is a unitary operator and Au - UAU* as in Remark 2.25, then the spectral
parts of Au and of A (point spectrum, absolutely continuous spectrum, etc.) are iden-
tical [so CY(Au) == CY(A), C5p(Au) == C5p(A), etc.]. This follows from the fact that
mut(V) == mJ(V) for each vector f and for each V E AB [observe that {UEA.U~}
is the spectral family of the operator Au and that F(J1u(A) IIUEA.U*(Uf)ii 2 ==
IIUEA.fll 2 == IIEA.fll 2 Ff(A)]. One has for example Hp(Au) == UHp(A) (if Af ==
JLf, then AuUf == UAf - fL Uf; so iff is an eigenvector of A, then Uf is an eigen-
vector of Au associated with the same eigenvalue).

4.3.4. The spectral parts of a self-adjoint operator introduced above are defined in
terms of a decomposition of the underlying Hilbert space. In general the different
spectral parts will not be disjoint. For example one may have C5 P (A) n C5 e (A) 0
(A may have eigenvalues embedded in its continuous spectrum). We now discuss a
subdivision of the spectrum which is frequently used and which divides the spectrum
into two disjoint parts.
164 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

The discrete spectrum C5d (A) is the set of all eigenvalues of A of that are of finite
multiplicity and isolated from the rest of the spectrum. So A E C5 d (A) if and only if
0 < dim J\1[ ( { ,\}) < oo and if there exists c > 0 such that the interval (A - c, A + c) is
disjoint fro1n the rest of the spectrum of A, i.e. such that (A c, ,\+c) n CY( A) == {A}.
The essential spectrum C5ess (A) is the complementary set of C5 d (A) in the spectrum of
the operator A:
(4.79)
A point A E CY(A) belongs to C5ess(A) for example if A E C5c(A), or if A is an eigen-
value of infinite multiplicity (isolated or non-isolated from the rest of the spectrum),
or if ,\ is an eigenvalue of finite multiplicity embedded in the continuous spectrum [in
which case ,\ belongs to both C5P (A) and C5 c (A)], or if ,\ is an accumulation point of
eigenvalues of A. We refer to the examples given at the end of this subsection.
One can characterise the points in the discrete spectrum and those in the essential
spectrum as follows in terms of the family of subspaces { M (V)} associated with the
spectral family of A:
AEC5ci(A)-¢==} 3c 0 > OsuchthatO < dimM((.A-c,.A+c)) < oo'icE(O,c 0 ),
A E C5ess(A) -¢==} dimM((.A- c, A+ c))== oo for each c > 0.
The essential spectrum is closed: if {An} is a sequence of points in C5ess (A) con-
verging to a point A, then M ((.A- c, A+ c))~ M ((.An- c/2, An+ c/2)) if I.A- An I <
s/2, hence dim M( (.A E, A+ c)) - oo for each c > 0.
Part (a) of the next proposition gives a useful characterisation of the essential spec-
trum; part (b) shows that the notion of the spectrum of a self-adjoint operator in an
infinite-dimensional Hilbert space is a natural generalisation of the notion of an eigen-
value of Hermitian n x n matrices: A E CY(A) -¢==} for each c > 0 there exists a vector
f E 'D(A), with llfll == 1, such that Af ~ .Af in the sense that IIAf- .Afll <c.
Proposition 4.20 (WeyPs Criterion). Let A be a self-adjoint operator in a Hilbert
space H.
(a) real number A belongs to the essential spectrum C5ess (A) of A if and only if there
exists a sequence {fn} o.fvectors in D(A) having the following properties: II f n II == 1
for each n, w-limn~(X) fn == 0 and s-limn~(X)(A- .A)fn == 0.
(b) real nunlber A belongs to spectrum CY(A) of A if and only if there exists a se-
quence {fn} of vectors in D(A) satisfying llfnll - 1 and s-limn~(X)(A .A)fn == 0.

PROOF. Let us fix A E JR and set ln == [.A- 1/n, A+ 1/n] for n 1, 2, 3, ...
(a) Suppose that A E C5ess(A), so that dimM(Jn) == oo for each n. First take
f1 E M(J1) arbitrarily, with llf111 == 1. Next choose a vector f2 E M(J2) such that
f2 j_ E(J2)f1 and llf2ll == 1; then \f1,f2) - \f1,E(J2)f2)- (E(J2)f1,f2) == 0.
Continue recursively in this way: fn E M(Jn) is chosen such that llfnll - 1 and
such that fn is orthogonal to E(Jn)fl, ... , E(Jn)fn-l· One obtains in this way an
orthonormal sequence {fn} [i.e. (fn, frr~) == bnmJ satisfying E(Jn)fn- fn· One has
\v-lin1n~(X) fn == 0 (see Example 1.2) and
SPECTRAL PARTS OF A SELF-ADJOINT OPERATOR 165

Thus the sequence {jn} has the properties stated in the proposition.
Conversely, suppose that there exists a sequence {fn} having these three proper-
ties, and assume that A tf:. C5 ess (A), i.e. that there exists a number N E N such that
dim M (J N) < oo. We must obtain a contradiction. For this let us consider an interval
(a, b) with a< .A< b. Then

IJ(A A)fnll 2 = j_aoolA- Ml 2 mt,(dp,) + loolA- p,J 2 mt,(dp,)

+ L rb
2 2
lA- Ml mt,(dp,) >(A- a) IIE((-oo,a))fnll
2
+ (b- 2
A) IIE((b,oo))fnll
2
.

Since \\(A- .A)fn\\---+ 0 as n---+ oo, this i1nplies that \\E((-oo,a))fn\1---+ 0 and
\\E((b, oo))fn\\ ---+ 0; consequently \\E([a, b])fn\\ ---+ 1 [because 1 == llfnll 2 ==
\\E((-oo,a))fn\1 2 + l!E((b,oo))fn\\ 2 + !IE([a,bl)fnll 2 ]. In particular, taking a==
A - 1 IN and b == A + 1 IN, we have II E ( J N) f n II ---+ 1 as n ---+ oo. If { e 1, ... , e1\lf}
is an orthonormal basis of M(JN), we have IIE(JN)fnll 2 == ~~~ 1 \(ek,fn)\ 2 . Now
(ek, fn) ---+ 0 as n---+ oo for each fixed k because fn converges weakly to zero, and the
sumisfinitebytheassumptionthatM dimM(JN) < oo. Hence \\E(JN)fn\\---+ 0
as n---+ oo, in contradiction with the relation \\E(JN )in\\ ---+ 1.
(b) =? In view of the result of (a), it suffices to consider the case .A E C5d (A). In
this case one can take f n == f for each n, where f is a normalised eigenvector of A
associated with the eigenvalue .A: Af == .Af, 1\fll == 1.
~ Let a < .A < b. The argument given in the second part of the proof of (a) shows
that \\E([a, b])fnl\ ---+ 1 as n---+ oo, which implies that dimM([a, b]) #- 0. Hence A
is a point of non-constancy of the spectral family { E-A}, i.e . .A E CY(A). D

We can illustrate the different types of spectra by means of examples of self-adjoint


operators. Most of the examples given below are of a purely mathematical nature, but
we cite some operators of interest in physics without, however, proving the stated
spectral properties (these proofs are often quite long, we shall return to this question
and some further examples in Chapter 6). We also refer to Section 14.6 of [P] for a
collection of examples of potentials in JR or JR 3 so1ne of which exhibit unusual spectral
or scattering properties.

Example 4.21. Operators with pure point spectrum


(a) We begin with some well-known examples.
(i) Projections: if is a projection, then CY(P) == {0, 1} if P #- 0, I. Indeed if P
is the projection onto the subspace M and if M #- {0} and M #- H, then M -
M ( { 1}) is the subspace of all eigenvectors of with eigenvalue .A == 1, whereas
M j_ == M ({0}) is the subspace of all eigenvectors of P with eigenvalue A == 0.
(ii) The Hamiltonian of the harmonic oscillator H == P 2 +Q 2 in L 2 (JR). Its spectrum
is discrete, the eigenvalues (each of them of multiplicity 1) are the numbers An ==
2n + 1 with n - 0, 1, 2, ...
(b) Let { e 1 , e 2 , ... } be an orthonormal basis of an infinite-dimensional Hilbert space
H. An operator in B(H) is entirely determined by its action on each ek.
166 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

(i) Let A be defined by Aek == k- 1 ek. Each ek is an eigenvector of A, so A has


pure point spectrum. The point A == 0 is not an eigenvalue of A, but it belongs to the
spectrum of A because it is an accumulation point of eigenvalues. Thus G(A) == {0} U
{ 1I k I k == 1, 2, 3 ... } , (} d (A) == (} p (A) == { 1I k I k == 1, 2, 3 ... } and (}ess (A) == { 0}.
(ii) Let A be determined by Ae 1 == 0 and Aek == (k -1)- 1 ek fork> 2. Again each
ek is an eigenvector of A, so A has pure point spectrum. In contrast to the operator
considered in (i), the point A == 0 is now also an eigenvalue. Hence G(A) == Gp (A) ==
{ 0} U { 1I k I k == 1, 2, 3 ... } , (} d (A) - { 1I k I k == 1, 2, 3 ... } and (} ess (A) == { 0}.
(iii) Let {r 1 , r 2 , ... } be an enumeration of the rational numbers in (0, 1) and Aek ==
rk ek. Again A has pure point spectrum. Its point spectrum GP (A) - { rk} is composed
of all rational numbers in (0, 1). This set is dense in [0, 1]. A has no discrete spectrum,
even though each of its eigenvalues is of multiplicity 1 (none of them is isolated from
the rest of the spectrum). One has G(A) == Gess(A) == [0, 1].

Example 4.22. Spectrum of multiplication operators


(a) The operators Q, Q2 , P, P 2 in L 2 (JR) as well as the operators Q2
and P2 in
L 2 (JRn) (n > 2) have purely absolutely continuous spectrum. Indeed the spectral
families of these operators have been given in Examples 4.9 and 4.15 - 4.17, and
in each case it is easy to see that the measures mf are absolutely continuous for all
f E H. For example for Q one has 'm,t(V) == fv lf(x)l 2 dx, its Radon-Nikodym
derivative is therefore lf(x)l 2 .
(b) More generally, if A is the operator of multiplication in L 2 (JRn) by a function
a: 1R11 -+ JR, then the spectrum of A is purely absolutely continuous if the function
a is continuously differentiable and grad[a(x)] -::/= 0 almost everywhere. We omit the
proof.
(c) Consider the multiplication operator A by a real function a in H == L 2 ( [0, 1]).
(i) For_\ E JR, let WA == { x E [0, 1] I a(x) == A}. The number A is an eigenvalue of
A if and only if the Lebesgue measure of WA is non-zero. In this case the subspace
M ({A}) of all eigenvectors associated with this eigenvalue is identical with L 2 (WA),
the set of functions f in L 2 ( [0, 1]) that vanish outside WA. Each eigenvalue is of in-
finite multiplicity and thus belongs to the essential spectrum of A. By conveniently
choosing the function a, one can again construct examples of operators with purely
discrete spectrum or of operators with pure point spectrum that is dense in an interval
or in JR.
(ii) If a is the Cantor function, then the spectrum of A is purely singular continuous:
G(A) - Gsc (A) == Gess (A) == <t (the Cantor set).
(iii) It is easy to give multiplication operators with mixed spectrum. For example let
a(x) == 4.:r if 0 < x < 112 and a(x) == f-L for 112 < x < 1, where fL is a fixed real num-
ber. Then Gac (A) - [0, 2], Gp (A) - {J-L} and Gsc (A) Gd (A) - 0. If f-L E [0, 2],
then G(A) == Gess (A) == [0, 2]. If f-L ~ [0, 2], then G(A) == Gess (A) [0, 2] U {J-L}.

Example 4.23. Coulomb Hamiltonian


An important operator in quantum mechanics is the Coulomb Hamiltonian He
P2 + ! IIQI in L 2 (JR 3 ), where ! -::/= 0 is a real constant. Its spectrum contains an
absolutely continuous part equal to [0, oo) and, if r < 0, an infinite number of negative
SPECTRAL PARTS OF A SELF-ADJOINT OPERATOR 167

eigenvalues AJ - -1 2 /(4j 2 ), where j - 1, 2, 3, ... These eigenvalues accumulate


at A == 0, and each of them is of finite multiplicity. Thus one has: (i) if r > 0, then
0" (A) == 0" ac (A) == 0" ess (A) == [0, 00) and 0" p (A) == 0" sc (A) == 0" d (A) - 0, ( ii)
r < 0' (} (A) - [0 ~ 00) u {A J I j == 1 ' 2' 3' . .. } ' (} ac (A) == (} E'SS (A) == [0' 00)'
o-P (A) - o-d (A) == {A .7 I j == 1, 2, 3, ... } and o-sc (A) - 0.

Spectrum of compact operators


a compact self-adjoint operator in a Hilbert space H. Let E > 0. We claim
that ( , oo)) and M ( ( -oo, -E]) are finite-di1nensional subspaces. Indeed suppose
for example that dirn M ( [E, oo)) - oo. Then there exists an infinite orthonormal se-
quence vectors {en} in M([s, oo) ). One has w-limn---+oo en == 0 (see Example 1.2),
hence s-lim n---+oo A en== 0 by Proposition 2.12( a). But, as in the proof of Proposition
4.20(a), one has IIAenll 2 == J~oo A2 mjen(dA) > ilenll 2 == E 2 > 0, in contradiction to
the fact that I!Aen 11 2 -+ 0.
Thus A has pure point spectrum, each non-zero eigenvalue is of finite multiplicity,
and for each E > 0 there are at most a finite number of eigenvalues A with !AI > E.
The point A == 0 may or may not be an eigenvalue. However, if dirn H == oo, then
A == 0 must belong to the essential spectrum of A. The following three situations may
occur:
(i) A == 0 is an accumulation point of eigenvalues but not itself an eigenvalue,
(ii) A == 0 is an accumulation point of eigenvalues and itself an eigenvalue (of finite
or infinite multiplicity),
(iii) ,\ == 0 is an eigenvalue of infinite multiplicity and isolated from the other eigen-
values of A. Hence, in an infinite-dimensional Hilbert space one has o-ess (A) == {0}
and o-(A) == O"ess(A) U o-d(A), where o-d(A) n {( -oo, -s] U [s, oo)} is a finite set for
each E > 0.

Example Spectra of various self-adjoint extensions of a symmetric oper-


ator. Let A be a closed symmetric operator with equal and finite deficiency indices
(v_ == v+ v < oo). Then:
(a) All self-adjoint extensions of A have the same essential spectru1n. In particular, if
some self-adjoint extension of A has purely discrete spectrum, then each self-adjoint
extension of A has this property.
(b) If o-( A) n J == 0 for some self-adjoint extension A of A and some open interval J,
then the spectrum in J of any self-adjoint extension A of A, if not empty, consists of
a finite number of eigenvalues of total multiplicity < z;; in other terms the dimension
of the range M A( J) the projection EA ( J) is not greater than v.
(c) If A is bounded from below, i.e. if (j, Aj) > rnl!fli 2 for some rn E JR and all
f E V(A), then the part of the spectru1n in ( -oo, m) of any self-adjoint extension A
of A is discrete, and total multiplicity of the eigenvalues of A in (- oo, m) is at
most z;, i.e. dimMA(( -oo, m)) < u.
The preceding results hold in particular for the self-adjoint extensions of the symmet-
ric ordinary differential operators considered in Chapter 3. Below we give the proof
of (b); the proof of (c) is similar, that of (a) is given for example in §8.3 of [Wl].
168 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

PROOF OF (b). (i) We first consider the situation in which the interval J is bounded,
say J == (a, b). Assume that the stated result is incorrect, i.e. that dim M A( J) > v. Let
.f1 , .f2 , ... , fl/+ 1 be v + 1linearly independent vectors in MA( J). By Proposition 3.7,
each .fk has the form .fk == gk + (I - U)hk, with gk E D(A) and hk E N(A * - i)
(U is some partial isometry). Since dimN(A * - i) == v, the v + 1 vectors hk must
be linearly dependent. So there exist constants a 1, ... , al/+l, not all of them equal to
zero, such that L~!~ akhk - 0. Then g L~!~ akgk == L~!~ akfk is a vector
in MA(tl) belonging to D( ..4), and g -/::- 0 because the vectors .fk are linearly inde-
pendent. Applying again Proposition 3.7, we see that II (A - c)gll == II (A c)gll ==
II (A- c)gll for any constant c. To arrive at a contradiction, we now show that the sec-
ond equality cannot be satisfied. We let c == (a + b)/ 2 be the midpoint of the interval
J and observe that [since (}(A) n (a, b) == 0]:

II(A-c)gll 2
= r
}1>--cl?_(b-a)/2
IA-cl 2
m:(dA) > C; 0
)
2
llgll 2 ,
whereas
II(A- c)gll 2 = r
}1>--cl<(b-a)/2
2
lA- cl m:(dA) < c; arllgll 2 ·
(ii) If J is unbounded, we have dimMA(J) < v for each bounded open subinter-
val J of J, which i1nplies that dimMA(J) < v. D
Let us comment on the 1neaning of spectral projections in quantum mechanics.
Observables are represented in standard quantum mechanics by self-adjoint operators.
A be a self-adjoint operator (for example the position operator or the Hamiltonian
of a quantum-mechanical system) and let {E>.} be its spectral family. If Vis a Borel
set in JR., then the projection E(V), which is also a self-adjoint operator, represents
observable "the value of A lies in V ". More precisely, if f is a vector in the un-
derlying Hilbert space (a state vector for the system under consideration), the number
IIE(V).fll 2 /llfll 2 gives the probability of obtaining a value in V if one measures the
observable A when the system is in the state described by f. If II f II == 1, then mf
is a probability measure on (JR., A B). For example if A is the position operator Q in
L 2 (JR.) and f is a vector with IIlii == 1, then the number fvlf(x)l 2 dx represents the
probability that the system is localised in V when it is in the state described by f. This
interpretation says nothing about how the measurement is performed, it refers to some
kind ofideal1neasure1nent, and the fact that IIE(V)fll 2 /ll.fll 2 is a probability implies
that the measurement must be repeated many times (each time on a system in the same
state f, i.e. prepared in the sa1ne way). Since E(V) == 0 if V n (}(A) == 0, i.e. if V
is disjoint from the spectrum of A, one sees that the only possible values of a mea-
surement of the observable A (the only values that one would obtain with non-zero
probability) are the values in the spectrum of A. More precisely: if A is an eigenvalue
of A, then E ( {A}) 0, hence the probability of obtaining the value A in the course
of a measurement is non-zero for certain states of the system. If A E (}(A)\(} P (A), then
E ( {A}) == 0, so the probability of obtaining exactly the value A in any measurement
is zero; however the probability of obtaining a value in an interval (A c, A + c)
is non-zero for each c > 0 and certain states the system.
SPECTRAL PARTS OF A SELF-ADJOINT OPERATOR 169

4 . 3 . 5 . Finally we consider the behaviour of the spectrum under a perturbation. Let A


be a given self-adjoint operator and let B be a second self-adjoint operator that can
be considered some sense as a perturbation of A, in particular such that + B is
self-adjoint. What relations can be established between the spectral parts of A and
those of A B? It is difficult to give answers in this generality, the spectrum of
A + and its nature can be quite different from the corresponding quantities for
A. Assu1ne for example that B is a multiple of the identity operator: B == a! with
a E JR. Then o-(A +B) == o-(A) +a, o-c(A +B) == o-c(A) +a, etc. (so for exanl-
ple .A E o-(A +B) {:=::? (.A- a) E o-(A), i.e. o-(A +B) is obtained by translating
o-(A) by a on the real line). In general, as subsets of JR, o-(A +B) and o-(A) can be
entirely different, and similarly for the spectral parts of these operators, even if IS
very "gentle".
Let us mention another example that is well-known in quantum mechanics. Let A
be the operator P 2 in £ 2 (JR) and B the potential of a square well, i.e. the multiplica-
tion operator by V (x), with V (x) == - V0 < 0 if x E [0, 1] and V (x) == 0 otherwise
(V0 being a constant). The spectrum of A is absolutely continuous and equal to [0, oo ):
o-( A) == o- ac (A) == o-ess (A) == [0, oo). The addition of does not change the essential
spectrum but adds a finite number of negative eigenvalues (each of multiplicity 1):
O"ac(A +B)- O"ess(A +B)== [O,oo), a(A +B)- [O,oo)Uad(A +B), where
ad (A+ B) is a finite set contained in (- V0 , 0). This example shows that there must
exist situations in which the spectrum or the spectral type of A + B is identical with
or very similar to the corresponding quantities for A.
The important general result is the invariance of the essential spectrum under a
relatively compact perturbation (the preceding example is a special case of this situ-
ation). We have seen in Section 2.7 that, if A is self-adjoint and B is symmetric and
relatively compact with respect to A, then A+ is self-adjoint on D(A). We recall
that the condition that be A -compact means that there exists a number z E p( A)
1
such that (A- z )- E JC(H).

Proposition 4.26 . Let A be a se(f-adjoint operator in a Hilbert space H. If B is


a symn~etric operator which is A -compact, then the essential spectrum o.f the self-
adjoint operator A+ B coincides with that of A: O"ess(A +B) == O"ess(A).

PROOF. (i) Let A E O"ess (A). We choose a sequence {fn} in D(A) satisfying Ilin II ==
for each n, w-limn--+(X) fn == 0 and s-limn--+(X)(A- .A)fn == 0 [the existence of such
a sequence is guaranteed by Proposition 4.20(a)]. By taking into account once more
Proposition 4.20(a), we may conclude that A E O"ess (A+ B) as soon as we have shown
thats-limn--+(X)(A+B- .A)fn- 0, which here is equivalent to s-linln--+(X)Bfn == 0.
For this we fix a number z E p(A) such that B(A z) 1 E JC(H) and write

Each term on the right-hand side converges strongly to zero: for the first tenn this
lows from the hypothesis that s-limn--+(X) (A- .A)fn == 0, for second one from
hypothesis that w -lim n--+(X) f n == 0 combined with the fact that the compact operator
B (A z) - 1 maps each weakly convergent sequence onto a strongly convergent one
SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

see Proposition 12(a)]. We have thus shown that o-ess(A) C o-ess(A +B).
The converse inclusion is obtained in same way by interchanging the roles of
A and A the preceding argument (more precisely by the substitutions
r---+ (A +B) ~---7 -B). The only additional input needed is to know that
is also (A+ B) -compact, i.e. that there exists a number z' E p(A +B) such that
(A+ - z')- 1 E JC(H); this has been established in Proposition 2.47(c). 0

Example 4 . 27.. As an application of Proposition 4.26, let us consider Schrodinger


operators of the type studied in Proposition 2.49. So let H == Ho+ V( Q) in Hilbert
space H == L (JR ), with H 0 == P . Assume that V E L (JR.n) + L~(JR.n) if n == 1, 2
2 11 2 2

or 3 and that V E LP(JR.n) + L~(JR 11 ) for somep E (n/2,oo) ifn > 4. Then V(Q)
is Ho-co1npact by Proposition 2.49, hence His lower semibounded and o-ess(H) ==
( H 0 ) == [0. oo). A particular case is that of the Coulomb Hamiltonian introduced
in Example 4.23.

Proposition 4.26 expresses the invariance of essential spectrum under a rela-


tively compact perturbation. However nothing is said about the invariance of spec-
tral type. In fact the nature of the essential spectrum A+ B can be entirely different
from that essential spectrum of A. For example A could have purely absolutely
continuous spectrum and A+ B pure point spectrum 13 , and this even if B is a compact
operator close to the zero operator. This statement is a special case of the following
theoren1 (see §X.2 of [K] for the proof).

Proposition (WeyJ . . von Neumann) . Let A be an arbitrary self-adjoint operator


E > 0. Then there exists a Hilbert-Schmidt operator B with I IIHs < E such that
has pure point spectrun~.

so1ne situations that are important in quantum mechanics, the spectral type is at
least partially preserved under a relatively co1npact perturbation. As an example, let
== P 2 L 2 (JR 77 ) and let == V (Q) be the multiplication operator by a real func-
tion V converging to zero at infinity. The spectrum of P 2 is purely absolutely con-
tinuous, equal to [0, oo ), and for a large class of potentials V of the indicated type
(e.g. the potentials considered in Example 4.27) one can show that the essential spec-
trum of the operator P 2 + V( Q) is also purely absolutely continuous (the proof is
quite technical, however; see Proposition 4.31 and Chapter 6 for some results in this
direction).

The spectral theorem. The resolvent near the spectrum

{E,\} a spectral family and A == J_CX)00 AE( dA) the associated self-adjoint op-
We first prove an important for1nula which expresses the spectral projection
13 In this case the eigenvalues of A +B wilJ be dense in a ac (A); thus A +B could be an operator of the
type given in Exmnple 4.21 (b,iii).
THE SPECTRAL THEOREM. THE RESOLVENT NEAR THE SPECTRUM 171

E( (a, b]) Eb - Ea associated with a finite interval (a, b] in terms of the resolvent
of In order to present a somewhat formal explanation of this formula, recall that

1 1
A . --+ 2Jri5(x- .A) as E--+ +0,
x- .A- iE x- +zc
the convergence being in the sense of distributions. If we formally replace x by a self-
adjoint operator A, hence replacing (x- ,\ =f ic)- 1 by the resolvent (A- z)- 1 for
z == A and then integrate with respect to d.A, we arrive at

ic)- 1
) dA--+ 1 6
J(A- A)dA

== X(a,b)(A) == E((a,b)).

The following proposition shows that this formal argument is essentially correct and
gives a precise meaning to the preceding formula (this formula is entirely correct if
the endpoints a and b of the considered interval are not eigenvalues of A).

Proposition 4 . 29 (Stone's Formula). Let A== J~(X).AE(d.A) be the self-adjoint oper-


ator associated with a spectral fan~ily { E>.. }. For -oo < a < b < oo one has
b
1 1 1
- . s-liml [(A-.A-ic)- 1 -(A-.A+ic)- 1 ]d.A == E((a,b))+- E({a}) - E({b})
2JrZ c--1-+0 a 2 2
(4.80)
and
b+6
1
E((a, b]) == - . s-lim s-lim
2 JrZ 6--1-+0 c--1-+0 1 a+<5
[(A- A- ic)- 1 - (A- A+ ic)- 1 ] dA (4.81)

(the order of the two limits in (4.81) is important).

PROOF. (i) For c > 0 and A E 1R set

b 2'ZE
=
1
l
21ri a (x _ A) 2
1
+ E 2 dx = :;r [arctan - c -
b- A
-arctan
a- A
c J.

Clearly Ws is a continuous function of A and IWs (A) I < 1 for all A E JR. If J is an
interval, we denote by XJ its characteristic function [see (1.41)]. By using the fact that
lims--1-+0 arctan[ ( x- A)/ c] == +1r /2 if A < x and== -1r /2 if A > x, one easily sees
that
1
wo(A) :== lim ws (A) == - [x( b) (.A) + X[ b] (.A)] .
c--1-+0 2 a,
(4.82)
a,

(ii) For each E > 0, Ws (A) J


Ws (A) E (dA) is a well defined operator in B (1{).
-(X)(X)

Let us show that { Ws} is strongly convergent as E --+ 0 and calculate its limit [the result
172 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

is already suggested by (4.82)]. Iff E H, we can write f == E( {a} )f + E( {b} )f fo,


where f 0 E [ M ({a}) M ({ b}) Jj_. One has

1 b-a
wc(A)E( {a} )J == wc(a)E( {a} )J == Jr [arctan -c-J E( {a} )J,

which converges strongly to (1/2)E( {a} )f as E --+ +0. In the same way one finds
that s-limc-r+O Wc(A)E({b})f == (1/2)E({b})f. Moreover, since E((a,b))fo
E((a, b])fo == fab E(A)fo, one has

E((a,b))fo- Wc(A)fo = 1b[l- Wc(A)]E(dA)fo- I~ Wc(A)E(dA)fo

-1= Wc(A)E(dA)fo·

By using the Dominated Convergence Theorem (Proposition 1.9) one finds that each
term on the right-hand side converges strongly to zero; for example [see (4.42)]:

and one has ll- wc(A)i 2 < 4 and 1- wc(A)--+ o m 10 -a.e. on (a, b] [the point b is of
measure zero: rr7J0 ( { b}) == 0]. We have thus shown that

1
wo(A) :== sc-r+O
-li1n wc (A) == E ( (a, b)) - E ({a})
2
+ -21 E ( { b}).
(iii) To obtain the validity of (4.80) it now suffices to verify that, forE > 0:

By virtue of the polarisation identity (2.15) this holds provided that one knows that

To prove (4.83) we use the first resolvent equation (2.93) to write

It follows that
THE SPECTRAL THEOREM. THE RESOLVENT NEAR THE SPECTRUM 173

and consequently

~ r (f, [(A -
n~Ja
A - ic) - l - (A . -1] f)dA=Jrc
A+~c)
1b
a
dA
joo (
-oo f-L
mj(df-L)
A) 2
+ c2 ·
Since the integrand is positive, one may interchange the order of integration (Fubini's
Theorem):
b
1
(J, [(A- A- ic)- 1 - (A- A+ ic)- 1 ]f)dA
2ni

which proves (4.83).


(iv) Finally we deduce (4.81) from (4.80). One has as a consequence of (4.80):
·b+6
~s-lim
2n~
j
c:---++0 a+b
[(A- A- ic)- 1 - (A- A+ ic)- 1 ] dA E((a + 5, b + 5))
1 1 1 1
+ 2E( {a+ 5}) + 2 E( {b + 5}) Eb+6- Ea+6 + 2E( {a+ 5}) - 2E( {b + 5} ).
(4.84)

Observe that if f-L E JR and 5 > 0, then liE( {t-L + 5} )!11 2 < liE( (f-L, f-11 + 5])!11 2 ==
II (Ep,+6 - Ep,)fll 2 ---7 0 as 5 ---7 0 (by the right continuity of the spectral family), i.e.
s-lim6---++0 E( {t-L + 5}) == 0. Thus, by taking into account again the right continuity
of { EA}, one sees that the strong limit as 5 ---7 +0 of the right-hand side of (4.84) is
Eb-Ea E((a,b]),whichproves(4.81). D
Proposition 4.30 (Spectral Theorem). If A is a self-adjoint operator in a Hilbert
space H, there exists a unique spectral family {EA} such that A== f~oo AE(dA).

REMARKS ON THE PROOF.


(a) Uniqueness: Let { EA} and { E\} be two spectral families defining the same self-
adjoint operator A, i.e. such that A == J~00 AE(dA) == J~00 AE(dA). Since the right-
hand side of Stone's Formula (4.81) depends only on A (it does not involve the spectral
family in terms of which A has been defined), it follows that E( (a, b]) == E( (a, b])
for all a, bE JR (a< b). Hence

E A == s-lim E ( (a, A]) s-lim E((a, A]) == EA.


== a---+
a---+-oo -oo

(b) Existence: There exist a fair number of different proofs of the Spectral Theorem.
All of them are rather long. We give an essentially complete proof in the Appendix of
this chapter, and we indicate here the strategy of another one of these proofs. If { EA}
exists, then E((a, b]) must be given by (4.81). The integral appearing in this formula
is well defined for each 5 > 0 and each c > 0, because (A - A ± ic) 1 are operators
in B(H) if A is self-adjoint [one has C \ JR E p(A)] and depend continuously on A
174 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

[the functions A ~ (A - A ± ic) - 1 are even infinitely differentiable in norm, see


Proposition 2.35(d)]. One must prove that the limits in (4.81) exist (at least weakly)
and define for each a < b a projection; that the collection of all these projections has
the properties of a spectral measure; and that A is the associated self-adjoint operator.
The limits (in the weak sense) in Stone's Formula are treated as follows. By taking
into account the polarisation identity (2.15), one sees that it suffices to consider
b+6

1a+6
(!,(A- A+ ic)- 1 f)dA]

(4.85)

[because (f,(A A+ ic)- 1f) ==([A- A+ ic)- 1 ]*f,f) ==((A- A- ic) 1f,f),
since [(A- z)- 1 ]* ==(A- z)- 1 if A is self-adjoint, see (2.103)]. Now z ~ (_[)(z) :==
(f, (A z)- 1 f) is a C-valued holomorphic function in the upper half complex plane
{z E C I ~ z > 0}, and one has to control its behaviour at the boundary of the domain
of analyticity (here for ~z ~ +0). For this one can use a theorem from the theory
of analytic functions of a complex variable: If (_[) is holomorphic in the upper half
plane, has a non-negative imaginary part and satisfies the estimate j<_p(z) I < C /~z
for some constant C, then there exists a finite Stieltjes measure m on JR such that
(_[)(z) == J~00 (A- z)- 1 rn(dA). Furthermore rn((a, b]) is then given by the expression
J:tf
lim5__,.+0 limc:--++0 7r- 1 [~<_p(A+ic )]dA. The application of this theorem gives for
each f E H the measure m f; the remainder of the proof reduces essentially to direct
verifications. The details of the complete proof are presented for example in §§65, 66
and 59 of lAG] or in §7 .3 (Theorem 7 .17) and Appendix B of [Wl].

Stone's Formula (Proposition 4.29) suggests that the spectral properties of a self-
adjoint operator should be determined by the behaviour of its resolvent (A- z) - 1 near
the real axis. We illustrate this by some examples.
(a) If A E IR belongs to the spectrum of A, then (A A± ic )- 1 ~ oo as c ~ 0.
II 11

Indeed one then has E( (A - c, A c)) -1- 0 for each c > 0, so that for each c > 0
there exists a vector f such that llfll - 1 and E((A- E, A+ c))f ==f. Then

==
1A.+c:l 1
A.-c: f-L- A± ic
12
rn (dtr)
.f r--v
> -2
-
1 1.:\+c:
2c A.-c: r--v 2c
1
m 1 (dtr) == -llfll
2
2
==-2
2c ·
1

Hence li(A A± ic)- 1 11 > (1/V2)c- 1 .


In Stone's Formula one finds the difference (A - A - ic) - 1 (A - A + ic) - 1
which may have better behaviour as c ~ 0 than each individual term, i.e. it could be
less rapidly divergent than each term separately.
(b) If IL is a real number in the resolvent set p( A) of A, then (A- f-L) - 1 E B (H) and
(A- tL±iE) - 1 ~ (A- f-L )- 1 in norm as c ~ 0. Since p(A) is an open set, there exists
{{; > 0 such that the interval (t-L - 2{1;, f-L 2/'1;) also belongs to p( A). If one considers
SPECTRAL THEOREM. RESOLVENT NEAR SPECTRUM 175

Stone's Formula (4.81) with a== fJ;- /'{;and b == fL + K, one may take the limit c -+ +0
under the integral sign, and one obtains that E((a, b]) _ E((JL f);, fJ; + K,))- 0.
(c) ,\ E a( A), then (A- ,\ - is )- 1! and (A - ,\+is )- 1! n1ay be convergent
as c -+ 0 certain vectors f. In particular, this is the case if the measure ntf is
supported from the point .\, more precisely if there exists /'{; > 0 such that
E( -f);,,\ /'{;]) f == 0. Indeed, if ( -f);, A+/'{;]) f == 0, then one example

~ ± ~c
2
I -.\ is) 1
!11 ==j>..-K
2
-oo
I

fJ; -
. 1 mr(df.h)
.

+
1 00 I
A+K fJ; -
1
A
.~E 12 mf(dtJJ) < 2 1
/'{;
00
m,J(dp)
1 . 2
== 2llfll ·
K

This inequality holds for each c > 0 and even for c == 0. In particular the integral
J~00 (fl;- A)- 1E(dfL)f defines a vector in H [let us denote it by - A)- 1!1 and a
calculation similar to the preceding one, using the Dominated Convergence Theorem
(Proposition 1.9), shows that I - .\ is)- 1f- (A- .\)- 1fll 2 -+ 0 as c-+ 0.
If ,\ tf:- a P (A), the set of vectors f of the type considered above (m f supported
from A) is dense in H, because s-lim K----tO ( [,\ - f);,,\+ /'{;)) == 0 in this case. On the
other hand, if,\ is an eigenvalue of A, the set of these vectors is not dense in H because
it contains only vectors that are orthogonal to the associated subspace M ({A}). To see
this, let g be an eigenvector of A associated with the eigenvalue A: ==- Ag, 11911 == 1.
Then -.\±is )- g == (±is )- g and [(A- A- is )- - (A- A +is )- 1 )g - 2is- 1
1 1 1

which are divergent as c-+ 0. We 1nust show that, if hE His such that (g, h) i- 0,
[(A-,\- is )- 1 - -,\+is) 1 ]h diverges as c -+ 0. Let us write h == (g~ h)g + ho
with (g, h 0 ) == 0. We observe that

[(A- A-iE) - 1 - (A- A+ic) - 1 ] h = ~i (g, h)g [(A- A-iE) - 1 - - A+ic) -t] h 0 .

Thus, since (g, (A - ,\ is) - 1 h 0 ) == ((A - ,\ is) - 1 g, h 0 ) == (±is) - 1 (g h0 ) == 0,


we have

II [ (A - A - is) - 1 - (A - A is) - 1 J h 11 2 == ; (g, h) I 1


2
c
II [(A- A- is)- 1 - (A- A is) 1
] ho\1 2 -+ oo as c-+ 0.
(d) We have seen that if A tf:- a P (A), then - ,\ is) 1 f converges as c -+ 0 for
a dense set VA of vectors f. This set VA depends on .\. A better situation would be
that in which these limits exist for each ,\ in some interval ~! and for all f in a dense
set independent of In such a situation one can conclude that spectru1n of A tJ
is absolutely continuous, i.e. that as(A) n J == 0. the next proposition we give a
precise description of such a situation that is frequently met in applications. order
to obtain a somewhat better understanding of these results, we recall the following
formula from the theory of distributions. If¢: JR -+ Cis a test function,

lim f oo ¢ (f.h) . dfl· = in¢ (A) cjJ(fL) i ]


-p--A (fl; .
c----t+O.} _ 00
p- A- v::
(4.86)
176 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

Let f Hac (A). Then there exists a non-negative function ()f E L 1 (IR) such that
E
mJ(V) == fv BJ(J-L)df1 if Vis a Borel set in IR. Assume that ()f is smooth (for example
that it is of class C 1 ). Then (4. 86) implies that

as E ~ +0. The expression on the last line is well defined (the limit 6 ~ +0 exists if
()f is smooth and is called the principal value of the integral). Thus there exists a dense
set of vectors of Hac(A) such that (f, (A-,\- ic)- 1f) has a limit as E ~ +0 for
each ,\ E IR. Some sort of converse is as follows: if the limit of (j, (A - ,\ - iE) - 1f)
as E ~ +0 exists for each ,\ in some interval (a, b) and each f in a dense subset of H,
unifor1nly in ,\ for each fixed f, then the spectrum of A in (a, b) is purely absolutely
continuous. This is the content of the next proposition.

Proposition 4.31. Let A be a se(f-adjoint operator and Jan open interval.


(a) Let f E H be a vector such that,,{or each,\ E J, (J, (A-,\- ic)- 1f) converges
as E ~ +0. Assu1ne in addition that the convergence is uniform in ,\ on each finite
closed subinterval J == [a, b] of J (i.e., if <!!(A.) :== limc:~+o (j, (A-,\- ic)- 1f),
then sup AEJ Iif!( A.)- (f, (A- A- iE )- 1!) I ~ 0 as E ~ +0). Then E( J)f E Hac(A).
(b) If there exists a dense set V of vectors in H such that the hypotheses of (a) are
satisfied for each f E V, then the spectru1n of A in J is absolutely continuous, in other
tern1s (J)H c Hac(A) and o-8 (A) n J == 0.

The continuous spectrum of the Schrodinger operators that one usually encounters
in quantu1n mechanics is absolutely continuous (this question will be treated in Chap-
ter 6 where we present a method of verifying the hypotheses of Proposition 4.31). Nev-
ertheless, many potentials also give rise to singular continuous spectrum 14 • Typically
one can create singular continuous spectrum by an infinite sequence of potential bar-
riers, and the spectrum of an absolutely continuous Hamiltonian can be transformed
into a singular continuous spectrum by the addition of a Uudiciously chosen) potential
with arbitrarily small L(X) norm.

PROOF OF PROPOSITION 4.31. (a) (i) Consider a finite closed subinterval J of J.


Define D by D == { z == ,\ + iE I ,\ E J, 0 < E < 1}, and let D == { z - ,\ + iE ( ,\ E J,
0 < E < 1} be its closure. Let¢: D ~ C be as follows: ¢(z) == (J, (A- z)- 1!)
for z E D and ¢(z) ¢(,\) == <!!(,\) if z == ,\ E J, where<!!(,\) is defined in the
statement of the proposition. ¢ is continuous on D [continuity of the resolvent in
p(A)], and ¢(A.+ ic) converges [uniformly in,\] to if!( A.) ¢(A.) as E ~ 0. Since the
unifonn limit of a sequence of continuous functions is continuous, it follows that if! is
a continuous function of ,\ E J. Thus ¢ is a continuous function on D. Since D is a
compact set, it follows that ¢ is bounded and uniformly continuous.
14
This question has been investigated for example by D. Pearson, see [1 3].
SPECTRAL THEOREM. THE RESOLVENT NEAR THE SPECTRUM 177

(ii) Assume that J == (a, j3), where -oo < a < j3 < +oo. Observe that E( J) ==
s-lima~as-limb/,BE((a,b]) (here a< a< b < j3 and a --7 a, b --7 j3). Hence,
to conclude that E (J) f E Hac (A), it suffices to know that E ( (a, b]) f E Hac (A)
all a, b satisfying a < a < b < j3. In other terms it suffices to know [see (4.36)] that,
for a < a < b < j3, the function FE((a,b])J(A) :== (E((a, b ])f,EAE((a, b])f) is
absolutely continuous, i.e. of the form J~(X) g(f-L)df-L with g E £ 1 (IR). Now

0 if A< a
(E((a, b])f,EAE((a, b])f) == (f, E((a, :A])f) if a< A< b
{ 2
II E ( (a, b]) f 11 if A > b.

Thus it suffices to show that, for a< A< b, (J,E((a, :\])!)is of the form fa:\ g(f-L)df-L
with g E £ 1 ([a, b]). By virtue of Stone's Formula and (4.85) one has

11A+6
(f,E((a, :A])f) == lim lim - [~(f, (A- f-L- ic)- 1f) J dfL. (4.87)
6---++0 c:---++0 1f a+ 6
If 5 is sufficiently small, the interval J [a + 5, b 5] is contained in J. Then, by
the result of (i), the integrand in (4.87) is bounded and uniformly continuous for fL E J
and c E [0, 1], and the Dominated Convergence Theorem (Proposition 1.9) allows one
to take the limit c --7 +0 under the integral sign:

11A+6 11A
(J,E((a, :A])f) == lim - [~<P(!-L)] df-L ==- [~<P(tL)] df-L.
6---++0 1f a+6 1f a

Since <I> is continuous on [a, b], ~<I> belongs to L 1 ([a, b]). Thus the proof of (a) is
complete.
(b) One has E(J)f E Hac(A) for each fED. Since Vis assumed to be dense in
H, the set {E(J)f fED} is dense in the subspace M(J)
j E(J)H (as is easily
15
verified ). Hence M(J) C Hac(A). D

The assumption of convergence of (f, (A- A- ic)- 1!) as c --7 0 in Proposition


4.31 is somewhat too strong but sufficient for our applications. It would be enough to
impose the boundedness of ~(f, (A - A - ic) - l f) near the real axis (see the proof of
the next proposition). In this context it is very interesting to know that the bounded-
ness of ~ (f, (A - A - ic) - 1f) (and hence absolute continuity) can be obtained as a
consequence of the positivity of a commutator, as shown by the next proposition.
Proposition 4.32 (Putnam's Theorem). Let A and H be bounded self-adjoint oper-
ators such that [H, iA] > CC* for some operator C E B(H). Then one has for all
A E IR, all c -1- 0 and each f E H:
(4.88)
andR(C) C Hac(H). Iffurthermore N(C*) == {0}, then the spectrum of His purely
absolutely continuous.
15
If Dis a dense subset of H and M a subspace of H, then the projection of D into M, i.e. the set
PMD, is a dense subset of M.
178 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

Observe that [H, iA] is self-adjoint and that CC* > 0. Thus the assumption of
positivity of the commutator [H, iA] entails absolute continuity of at least part of
the spectrum of If C* is invertible, i.e. if N(C*) == {0}, then R(C) is dense
the Hilbert space [see Proposition 2.19(b) l, and then Proposition 4.32 leads to the
conclusion Hac (H) == H. Of course, since the hypotheses of the proposition are
completely symmetric in A and H, the same conclusions hold for the operator A.
concrete situations one is interested in knowing the spectral properties of a given
operator H (often a quantum-mechanical Hamiltonian, whence the notation), and the
preceding proposition shows that a useful idea is to look for an auxiliary self-adjoint
operator A such that the commutator [H, iA] is positive in some sense (the operator
A is then said to be conjugate to H). Chapter 6 we present a method based on this
idea that leads to absolute continuity also for unbounded operators H and A. 16 We
refer to the introduction of Chapter 6 for further comments.

PROOF OF PROPOSITION 4.32. We assume throughout the proof that c > 0 and use
the notation R(z) - (H-z) 1 if z E p(H).
(i) For f E H we have [use (2.103) and the first resolvent equation (2.93)]:

1
'25( C j, R(>. +is )Cf) = i ( C j, [R(>. +is) - R( >. -is )]Cf)
2
== c(Cf, R(A + ic)R(A- ic)Cf)
== cliR(A- ic)Cf// 2 < ciiR(A- ic)Cii 2 llfll 2 . (4.89)

Now (use I/ // 2 == II II~ Eqs. (2.14) and (2.103) and the hypothesis that CC* <
):

IIR(A- ~ic)C// 2 == IIR(A- ic)CC* R(A + ic)l/


sup (R(A ic)j,CC*R(A+ic)f)
!EH,IIJII=l
< sup (R(A + ic)j, [H, iA]R(A ic)f)
!EH,IIfll=l
== //R(A- ic)[H, iA]R(A + ic)ll
== 1/R(A- ic)[H- A+ ic~A]R(A ic)/1
< //AR(A + ic)ll + IIR(A- ic)AII + ll2icR(A- ic)AR(A + ic)ll.
1 2 1
Since IIR(A ic)ll < , this implies that IIR(A- ic)CII < 4c- IIAII. Upon in-
setting this estimate for IIR(A- ic )C/1 2 into (4.89), one obtains the inequality (4.88).
(ii) Let { EA} be the spectral family of H. If J == (a, b] is a bounded half-open
interval, one has by Stone's formula (4.81 ):

1 1b+6
(Cf,E((a,b])Cj) ==-lim lim c;s(Cj,(H-A-ic)- 1 Cf)dA.
1f 6--++0 c:--++0 a+b
16
It can be seen from the proof of Proposition 4.32 that the assumption of boundedness of H in this
proposition is not in1portant, whereas the boundedness of A is essential.
APPENDIX TO CHAPTER 4 179

Now

hence
(Cf,E((a,b])f} mcJ(J) < 4IIAIIIIJII 2 IJI,
where I · I means Lebesgue measure. This implies that mcf (V) < 41\A IIIIJII 2 IVI for
each Borel subset V of JR (take into account the definitions (4.2) and ( 1.45) of these
measures). Hence the n1easure mcf is absolutely continuous with respect to Lebesgue
measure, Cf E Hac(H). Finally, if N(C*) == {0}, the set {Cf If E H} is dense
in H, as already pointed out before the proof. D

Appendix to Chapter 4: Proof of the Spectral Theorem

A.4.1 . Bounded self-adjoint operators. We construct the spectral family { E:>..} for a
bounded self-adjoint operator A. To motivate this construction, let us observe that the
subspace M:>.. (the range of the projection E:>..) is the subspace on which the opera-
tor A:>.. :== A - AI is negative. In more detail: one can write A:>.. == Af + A_\ with
At== (A- A)(I- E:>..) and A_\ ==(A- -X)E:>.., which expresses A- -XI as a sum of a
positive operator and a negative operator (the positive and negative parts of - -XI).
The positive part At annihilates M:>.., because one clearly has AtE:>.. == 0. One can
thus characterise the vectors f in M :>. by the property that At f == 0. Consequently, in
order to define M :>. in terms of the operator A, it suffices to find a means of exhibiting
the positive part r+ and the negative part r- of a self-adjoint operator T, i.e. to write
T == T+ r- with > 0 and r- < 0. Then one can set (by taking forT the
operator A- AI): M:>.. == {f EH I (A- AI)+ f == 0}.
To define r+ and r-' we first construct an operator ITI > 0 representing the
absolute value ofT, and then we set == (T + ITI)/2 and r- == (T- 1)/2. To
obtain IT\, we proceed in analogy with the numerical case in which lzl is the positive
square root of zz. Here one must define the positive square root of a self-adjoint pos-
itive operator T*T. We begin with some preliminary results. We observe in particular
that (4.90) below gives a generalisation of the Schwarz inequality (1.12).
Lemma 4.33. Let A be a bounded positive self-adjoint operator (hence (hiAh} > 0
for all hE H). Then
(a) for each k == 1, 2, 3, ... , the operator A k is positive,
(b) if S E B(H), then S~AS is positive,
(c) one has for all f, g E H:
\(g,Af)l 2 < (j,Af)(g,Ag}, (4.90)
(d) ~{furthermore A< I, then IIAII < 1.
PROOF. (a) If k is even, say k ==2m with m~ EN, then (f,Akf} == IIAmfll 2 > 0. If
k is odd (k == 2rn + 1, mEN), then (f,Akf} == (Amj,AAmf} > 0 because A is
positive.
l80 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

(b) One has(/, S*ASJ) ==(Sf, ASf) > 0.


(c) The inequality (4.90) is trivial if (g, A f) == 0. So let us assume that (g, A f) -f. 0.
As A is positive and self-adjoint, one has for each a E C:

0 < (f ag,A(f ag)) == (f,Af) lal 2 (g,Ag) + a(g,Af) + a(Af,g).

By setting a== -A(g,Af) with A E IR, this inequality becomes

(4.91)

This implies that (g,Ag) -f. 0 [otherwise (4.91) would be violated for A --7 +oo]. We
may thus take A == 1/ (g, Ag); for this value of A and after multiplication by (g, Ag),
the inequality (4.91) becomes:
2
0 < (f,Af)(g,Ag) -l(g,Af)l ·

(d)IfO <A< I,onehasby(2.14):

II All == sup (f,Af) < sup (/,If) == sup 11/11 2 == 1. D


fErLIIJII=l fEH,IIfll=l !EH,IIJII=l

Proposition 4.34. Let A be a bounded positive self-adjoint operator. Then


(a) there exists a unique bounded positive self-adjoint operator B such that B 2 == A
(it is usual to denote this operator B by A 1 12 ),
(b) ifC E B(H) commutes with A, then it also commutes with B.

A < I and set T ==I- A. By writing B ==I- X,


PROOF. (i) We first assume that
the equation B == A becomes X == (T + X 2 ) /2. We try to solve this equation re-
2

cursively, by setting

1
x1 == -2 T and (4.92)

Let us determine the properties of this family {Xn}·


(1) It is clear from (4.92) that each Xn is a polynomial in T of order 2n-l, and all
coefficients of these polynomials are > 0.
(2) The operators Xn commute with each other.
(3) As 0 < A < I, one has 0 < T < I, hence liT II < 1 [Lemma 4.33(d)]. It follows
that IIX1II < 1/2 < 1 andbyrecurrencethat IIXnll < (IITII + IIXn-111 2 )/2 < 1.
(4) Let us check that 0 < Xn < I. One has

and

(f,Xnf) = ~[(f,Tf) + 11Xn-dll 2 ] < ~[IIJII 2 11Xn-lii 2 IIJII 2 ]


< ~ [11!11 2 11!11 2 ] = 11!11 2 = (!,If).
APPENDIX TO CHAPTER 4 181

(5) Finally we show that Xn+ 1 - Xn > 0. We have the following recurrence relation
[using (2) in the last step]:

Xn+l- Xn = 4[(T +X~)- (T + X~_ 1 )] = 4(X~- X~_ 1 )


1
== 2(Xn + Xn-1)(Xn- Xn-1)· (4.93)

As X 2 - X 1 == T 2 /8, one deduces from Eq. (4.93) recursively that Xn+ 1 - Xn is a


polynomial in T (of order 2n ), also with all coefficients > 0. Since T is positive, so
is Tk for each kEN [Lemma 4.33(a)]. Thus each term of this polynomial is positive,
hence Xn+1 Xn > 0.
(ii) Let us now show that the sequence { Xn} is strongly convergent. We fix f E H.
By (5) and (3) the numerical sequence { (f,Xnf}} is monotone non-decreasing and
bounded above by II fll 2 . Therefore this sequence is convergent. Now let n > m. Then
Xn - Xm > 0 by (5), and by using the inequality (4.90) (with A == Xn - Xm and
g == Xnf - Xmf) one sees that
2
IIXnf- Xmfll 4 == !(Xnf- Xmf,Xnf- Xmf)l
< \f,Xnf- Xmf)\Xnf- Xmf, (Xn- Xm)(Xnf- Xmf)).

The first scalar product on the last line converges to zero as m, n ~ oo (see above).
By (3) the second scalar product is majorised by IIXn- Xmii 3 IIJII 2 < 8llfll 2 . This
shows that { Xnf} is strongly Cauchy for each f E H.
(iii) Let X be the strong limit of the sequence of operators {Xn}· Since (f,X f) ==
limn-too (f, Xnf}, one sees by taking into account (4) that 0 < (f, X f) < II fll 2 ,
which proves that 0 < X < I. Hence B I - X is a positive operator. B is self-
adjoint, because X is self-adjoint as the limit of a sequence of bounded self-adjoint
operators. Finally, by letting n ~ oo in (4.92), one obtains X == (T + X 2 ) /2 (observe
that X~ converges strongly to X 2 by Proposition 2.1). The equation X == (T + X 2 ) /2
is equivalent to B 2 == A.
As Xn is a polynomial in (I - A), Xn commutes with each bounded operator C
satisfying [C, A] == 0. By passing to the limit, one concludes that X (and hence also
B) commutes with these operators C. This proves part (b) of the proposition.
(iv) If the assumption A < I made so far is not satisfied, let A 0 - A/ II A 11. Then
Ao is self-adjoint and satisfies 0 < A 0 < I, hence there exists a positive self-adjoint
operator Eo with B5 == Ao. Then clearly B :== IIA 11 1 / 2 Eo is self-adjoint and positive,
and B 2 ==A.
(v) Let us finally prove the uniqueness of B. Assume that B' is a second positive
2 3
self-adjoint operators satisfying B' == A. Since B'A == B' == AB', the operator B'
cornmutes with B [by part (b) of the proposition, applied with C == B']. Thus

2
(B- B')B(B- B') + (B B')B'(B- B') == (B- B')(B 2 - B' )

== (B- B')(A- A)== 0. (4.94)


182 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

Each of the two terms on the left-hand side of (4.94) is a positive operator [Lemma
4.33(b)]. Hence each of these terms is zero, and consequently so is their difference:

0 == (B- B')B(B -- B')- (B B')B'(B- B') == (B B') 3 .

It follows that ( B - B') 4 == 0, and by applying twice the identity (2.12 ):

whence B == B'. D

Corollary 4.35. LetS, and Z be bounded self-adjoint operators satisfying S > T,


Z > 0 and [Z,S] == [Z,T] == 0. Then ZS- SZ > ZT == TZ.

PROOF. Z 112 commutes with S and with T. By virtue of Lemma 4.33(b) one has
SZ- ZT == Z 1 12 (S- T)Z 1 12 > 0 if S > T. D

We can now define the positive and negative parts of a bounded self-adjoint opera-
torT. One sets ITI == (T*T) 112 (T 2 ) 112 , the positive square root of the self-adjoint
operator T 2 (observe that (f, T 2 f) == 11Tfll 2 > 0). Let

~(T
]
y+ = ITI) and == 2 (T -ITJ). (4.95)

Let also M(T) == {! E H j r+j == 0}. This is the closed subspace N(T+) of all
eigenvectors of the self-adjoint operator r+ associated with the eigenvalue 0. We
write E for the projection PM(T) onto M(T), i.e. == PM(T)· In (A)-(J) below we
specify some properties of these quantities .
• (A) The operator ITI (and hence also T+ and r-) commutes with
each operator E B(H) satisfying [C, T] == 0.
This follows from Proposition 4.34(b ) .
• (B) The operators ITI' T+ and r- commute among theTnselves.
• (C) One has 1- Tl ==
ITI [because (- T) 2 == T 2 ] .
• (D) (a) If T > 0, then ITI == T, hence == T and r- == 0.
(b) If T < 0, then IT!== -T, hence == 0 and r- == T.
• (E) One has

r+E == ET+ == 0 and r- E == ET- == ET- E == T-. (4.96)

Indeed: (i) The equation == 0 is simply a reformulation of the definition of


M (T) as the set of eigenvectors of with eigenvalue 0. Then it follows that ET+ ==
* == 0* == 0.
(ii) Iff == T- g, then (because [T, ITIJ == 0):

r+t = r+r-g = ±(T ITI)(T -ITI)g = J(T 2 -ITI 2 )g = o,


APPENDIX TO CHAPTER 4 183

hence f E M(T). other terms g == ET- g for each g E H, i.e. . It


follows that - (E7l-)* _ and then
• (F) The operator commutes with and with , hence also with T +
and with ITI . Furthennore

(4.97)

and
(I- E)T == (4.98)
These state1nents are simple consequences of (4.96).
• (G) has > and < 0.
Indeed: (i) == E(T+- ) !TIE > 0 by Le1nma 4.33(b)
(because I I > 0).
(ii) (4.96) implies that (I- E)T+(I- E) == and (I- E)T-(I- E) == 0.
Hence ==(I- E)(T+- )(I- E)== (I- E)\T!(I- E) > 0.
• (H) One hasT+> T+ + (because < 0).
• (I) If Sis bounded and commutes with then [S,E] == 0.
Indeed: (i) Let f E M(T). Then ST+j == 0. ButS commutes with IT!, hence with
T+; thus T+Sj == 0, i.e. Sf E M(T) or Sf == ESJ. In other terms one has SE ==
ESE. By taking the adjoint of this relation one gets ES* == ES*E. If we assume for
the moment that Sis self-adjoint, this becomes ES == ESE. Since ESE == SEas
already seen, one then has E S == S E.
(ii) We have shown that [S, T] == 0 implies that [S,E] == 0 if Sis self-adjoint.
For a general S, observe that [S, T] - 0 implies that [S*, T] == 0. Since S + S*
and i ( S - S*) are self-adjoint, the result of (i) implies that [S + Si', E] == 0 and
i[S- S*,E] == 0. By combining these two relations one obtains [S,E] == 0.
• (J) LetS be a bounded self-adjoint operator satis_fying S > 0, S > T
and [S,T] == 0. Then S > T+.
Indeed: since I - > 0 and I - commutes with S and with T by (I) and (F),
Corollary 4.35 implies that (I- E)S > (I- E)T. On the other hand, since ES ==
ESE and S is positive, one has ES > 0 [Lemma 4.33(b)]. Consequently S ==
(I E)S + ES > (I- E)T ==
Now let A be a bounded self-adjoint operator. We set, for,\ E IR: AA == A - .\I,
At== (AA + IAAI)/2, A~ == (AA- IAAI)/2 and

MA == M(AA)- {f E HI At f == 0}.

Furthermore let EA be the projection onto the subspace MA. In the next two lemmas
we check that { E,\} is a spectral family (Lemma 4.36) and that A is the self-adjoint
operator associated with it (Lemma 4.37).

Lemma 4 . 36.. The family of projections { E A} defined above is a spectral fan1ily, i.e.
it satisfies (4.16) and (4.17).
184 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

PROOF. (i) By the property (F), AA commutes with EA. Since A AA + A.I, it
follows that A commutes with each EA.
(ii) We now show that MA C MJ-L if A < M; this implies that EAEJ-L == Emin{A,J-L}
for all A., ME JR.
Since [A,\, AM] == 0, one obtains from (A) [with T ==A- A., C ==AM] that AM
commutes with lA AI and with A~. By using again the property (A) [with T == AM and
C == lA,\ I or A~], one concludes that lAM I and A~ commute with lA,\ I and with A~.
In particular we have [At, At] == 0.
Property (H) shows that, for M > A.:

At > AA ==A A.I >A- Mf ==AM.

Since At > 0, we can apply (J) ([with S == At, T- AM] to conclude that At > At.
By virtue of Corollary 4.35 this implies that At At > (At) 2 . 17 Consequently we
have (J,AtAt f) > IIAtJII 2 for all f E 1-i. Iff EM,\, i.e. if At f == 0, we ob-
tain from this inequality that At f == 0, i.e. f E MM. This establishes the inclusion
MAc M{L.
(iii) Next we show that MA == {0} for all A. < IIAII· So let A. < -IIAII and set
c ==-II All- A.. We have c > 0 and, by using (2.14):
2 2 2
(f,AAJ) == (f,Af)- A.llfll > ( IIAII- A.)IIJII == cllfll == (j,c;If),
i.e. AA > c;J > 0. Consequently At== A_\. Iff EM,\, i.e. if Atf- 0, then

cllfll 2 < (j,AAJ) == (J,At f)== 0, hence f - 0.

(iv) Finally we show that MA == 1-i if A.> IIAII· Here we set c ==A. -IIAII· Then
2 2 2
(f,AAJ) == (f,Af)- A.llfll < (II All- A.) 11!11 == -cllfll ·

Hence AA < -c;J < 0 and consequently At == 0. Thus At f == 0 for all f E 1-i, i.e.
MA -1-i. D

We have thus established that the projections {EA} define a spectral family. We
shall see in the following lemma that this spectral family is also right continuous. As
a preliminary step we prove a useful auxiliary inequality: If A. < M, let E( (A., M]) ==
EM- E_\. 18 Then
A.E((A.,M]) < AE((A.,M]) < ME((A.,M]). (4.99)

Indeed: by using the relations [A, E( (A., M])] == 0 [see (i) of the preceding proof],
E((A., M]) 2 == E((A., MD == E((A., M])EJ-L and as well as (4.97), one obtains
(MI- A)E((A., M])- E((A., M])(MJ- A)E((A., MD - -E((A., MDA{LE((A., M])
== -E((A., M])EJ-LAJ-LE((A., M]) == -E((A., M])A~ E((A., M]) > 0
17
TakeS= At and T = Z =At in Corollary 4.35.
18 Re1nen1ber that E((A., p,]) is a positive operator, in fact a projection (see page 142).
APPENDIX TO CHAPTER 4 185

(the final inequality is obtained from Lemma 4.33(b ), because- A~ > 0). This proves
the second inequality in (4.99). The first one is obtained by a similar argument, by
using the relation E((.A, JL]) == E((.A, JL])(I- E>J and (4.98):
(A .A)E((.A, JL]) == E((.A, JL])(A- .AI)E((.A, tL]) == E((.A~ JL])(I- EA)AAE((.A, JL])
((..\, JL])At E((.A, JL]) > 0.
Lemma 4 . 37.. The spectral family { E A} in Lemma (4.36) is right continuous, and A
is the self-adjoint operator associated with it.
PROOF. (i) We first prove the right continuity of the family of projections { EA}. We
fix A E JR. The number IIEA+cfll 2 is the square of the norm of the projection off
onto the subspace MA+E· As MA+5 C MA+E if 0 < 5 < c, one has IIEA+5fll 2 <
IIEA+cfll 2 . So the numerical sequence {IIEA+cfll 2 } is monotone non-increasing as
c ~ +0, i.e. it has a limit. Hence (since EA+EEA+5 - E,\+5 if 0 < 5 <c):
2 2 2
IIEA+Ef EA+5fll == IIEA+cfll IIEA+5fli - (EA+Ef, EA+5f)
2 2 2
- (EA+5f, EA+Ef) == IIEA+cfll IIEA+5fll - 2IIEA+5fll
== IIEA+cfll 2 IIE,\+5!11 ~ 0 as 5, c ~ +0.
2

This shows that { EA+E} is strongly Cauchy as c ~ +0. We define a projection FA


by FA == s-lilnc-++O(EA+c EA) s-limc-++0 E((.A, A+ c]). We must show that
FA - 0, in other terms that the equation FAJ == f implies that f - 0. For this we
use (4.99) with JL- A+ c:
..\E((.A, ..\ + c]) < AE((.A, ..\ + c]) < (..\ c) ((..\, ..\ c]).
Upon taking the limit c ~ +0, one obtains that for each f E 7-i:
(f, ..\FA f) < (f, AFAf) < (f, ..\FA f).
So (f, .AFAJ) - (f, AFAJ) for all f E 1-i and hence (f, ..\FAg) - (f, AFAg) for all
j, g E 1-i [by polarisation, see (2.15)]. This shows that AFA == ..\FA. Consequently,
iff satisfies FAf - f, one has AAJ == 0. By taking into account Eq. (4.98) (with
T- A,\), one then obtains that At f == (I EA)AAJ == 0. Thus f E MA C MA+E
for each c > 0, hence EAJ- EA+Ef ==f. It follows that EA+cf- EAf == 0 for each
c > 0, hence FAJ 0. Since FAJ- f, we have f == 0.
(ii) We now check that A is the self-adjoint operator associated with { EA}· Let
(a, b] be an interval containing [-IIAII, IIAIIJ. Let us denote by A' the self-adjoint
operator associated with { E A}. Then, by (4.39):
N
A'== s-lim
IIII-+0
L ukE((sk-1, sk]),
k=1
(4.100)

where II== {s 0 , s 1 , ... , sN; u 1 , ... , UN} is a partition of (a, b]. From (4.99) we de-
duce that
N N N
L Sk-1E((sk-1, sk]) <A L E((sk-1, sk]) < L skE((sk-1, sk]). (4.101)
k=1 k=1 k=1
186 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

We observe that A L~= 1 E( (Sk-1, sk]) - AE( (so, s N ]) == A. Thus the second in-
equality in (4.101) implies that A< A' [we have taken ruk ==Skin (4.100) to express
. Next, applying (4.100) with Uk - sk- 1 + IIII/2, we obtain from the first inequal-
ity in (4.101) that A'< A. Thus (f,Af)- (.f,A' f) for all f E 7-i, and by polarisation
(f,Ag) == (j,A' g) for all j, g E 7-i. So we have shown that A== A'. 0

IAI f~oo J.AJE(d.A) J~CX).\E(d.\) + J000


Remark 4.38. (a) One has == == - A (d.\),
hence

i.e. A+== AE( (0, oo) ), and similarly


·O
A- = j
-oo AE(dA) = AE( -oo, 0)) - AE(( -oo, 0]),

because AE( {0}) == 0 [since E( {0} )7-i == N(A)]. It is clear that A+j- 0 for each
vector f belonging to subspace E(( -oo, 0])1-i, and A+j -1- 0 for each non-zero
f in the subspace E( (0, oo) )7-i. Consequently, if M(A) == {f E 1-i I A+ f == 0}, we
have M(A) == E((-oo,O])H, i.e. the projection PM(A) associated with the
self-adjoint operator A is just the spectral projection ( ( -oo, 0]). This observation
will be useful in the proof of Proposition 4.39.
(b) If A > 0 and A < 0, then A- .AI > 0, hence At == A>.. > 0. follows that
M:>.. N(At) - {0}. So E:>.. == 0 for all A< 0. Similarly one finds that if A< I,
then M:>.. == 1-i and hence E:>.. - I for all A > 1.

A . 4.2 . Semibounded self-adjoint operators. We have proved the Spectral Theorem


under the assumption of boundedness of the operator A. The case of an unbounded
self-adjoint operator can be reduced to that of a bounded operator. The simplest situ-
ation is that of a semibounded operator A. The case of a lower semi bounded or upper
semibounded operator is easily reduced to that of a positive operator; so we now dis-
cuss this latter situation.
Let ~4- A* > 0. Then one has for each f E V(A):

II(A + I)fll 2 == IIAJII 2 + 2(f, Af) 11!11 2 > IIAJII 2 + 11!11 2· (4.102)

Hence I is invertible and I (A I) 1


11 < 1 (see Lemma 3.1). Also R(A
I)j__ ==
N(A +I) == {0} [Proposition 2.19(b )]. Thus, again by Lemma 3.1, Y :== (A+ I)- 1
belongs to B(H), and -1 E p(A). The operator Y is self-adjoint, and 0 < Y < I:
the second inequality is satisfied because I Y I < 1 and the first one follows from the
positivity of A: if g -1- 0, then

(g.17 g)== ((A+ I)(A + I)- 1 g, (A+ I)- 1 g) > ((A+ I) 1


g, (A+ I) 1
g)
- II(A + I)-1gll2 > 0
APPENDIX TO CHAPTER 4 187

[for the last inequality observe that (A+ I)- 1 g == 0 only if g == 0, because (A+ I)- 1
is a bijective mapping].
The correspondence ,\ ~----+ ( ,\ + 1) - 1 defines a bijection between the intervals [0, oo)
and (0, 1], which will allow us to establish a one-to-one relation between the spectral
projections E:>.. of A and those of the bounded operator Y. We denote the spectral
measure of Y by {JE( ·)} and observe that, by Remark 4.38(b ), its support is contained
in the interval [0, 1], so that E((O, 1]) ==I. We define E:>.. as follows:

EA == 0 if A< 0 EA = lE([A, 1J) if A> 0. (4.103)

The projections E:>.. defined by (4.103) do form a spectral family, because


(i) IfO < ,\ < JL one has

EAE~t = lE ([A~ 1 , 1)) lE ([ !L ~ 1 , 1)) - lE ([A~ 1 , 1] n [!L ~ 1 , 1))


= lE([A ~ 1 , 1]) = EA Emin{A,p}·

(ii) { E:>..} is right continuous: if,\ > 0, then

EA+E- EA = lE([A + ~ + E' A~ 1)) -> 0 as c -7 +0.

(iii) One has s-lim:>..-++oo E:>.. == E((O, 1]), and E((O, 1]) == JE([O, 1]) == I (because
Y > 0, hence 0 is not an eigenvalue of Y).
Let us denote by A' - f 0 .\E( d.\) the (positive) self-adjoint operator associated
00

with the spectral family { E:>.. }. By a simple calculation [consider partitions of an in-
terval (a, b] and make the change of variables ,\ ~----+ JL - ( ,\ + 1) - 1 ] one finds that 19

(A'+ !)~ 1 = (oo A


1
E(dA) = tfLlE(dJL) = Y- (A+ 1)~ 1 . (4.104)
Jo +1 Jo
This relation implies that D(A') == D(A) and that (A+ I)(A' + I)- 1 == I. Since
(A' + I)(A' + I)- 1 == I, we conclude that Af == A' f for f E D(A') == V(A),
whence A' == A: { E:>..} is the spectral family of A!
REMARK. One can see that, if A is assumed to be only symmetric, the preceding proof
is not applicable (in fact the Spectral Theorem is not true without the assumption that
A be self-adjoint, or at least normal). Let us consider for exa1nple the symmetric (non-
self-adjoint) operator K 0 == -d 2 /dx 2 in £ 2 ((0, oo)), with domain C 0 ((0, oo)). Its
adjoint K 0 has -1 as eigenvalue: the equation !{0g == -g has the square-integrable
solution g(x) == e-x (see Proposition 3.14). Since R(Ko +I) _i N(K0 +I), this
solution is orthogonal to R(Ko +I) and thus (Ko + I)- 1 cannot belong to B(H).
19 The spectral integrals are defined by using partitions with subintervals ( s k _ 1 , s k) that are open on the
left and closed on the right. Now E( ( c, d)) = IE([1/ d, 1/ c)), which is equal to IE( (1/ d, 1/ c]) if and only
if c and d are not eigenvalues of A. Thus one should consider here only partitions of (a, b] for which the
endpoints s k of the involved subintervals are points of continuity of the spectral family {EA.}. We recall
that there are at most a countable number of points of discontinuity, see page 143.
188 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

A.4 . 3 . Non-semibounded self-adjoint operators . If a self-adjoint operator A has a


spectral gap (i.e. if o-(A) -1- JR), then its resolvent set p(A) contains a real number
,\ 0 , and one can still apply the method used in §A.4.2 above to obtain the spectral
family of A: the operator Y :- (A- .\ 0 )- 1 belongs to B(H), and the correspondence
,\ ~ (,\ - .\ 0 )- 1 defines a bijection between the spectrum of A and that of Y. In
the general case (without the hypothesis that o-(A) -1- JR), one relies on the following
result which shows that it is possible to express A as a sum of a positive and a negative
operator. We refer to (4. 70) for the concept of the restriction AM of A to a subspace
M.
Proposition 4 . 39 . If A is a self-adjoint operator in a Hilbert space 1-i, there exist two
mutually orthogonal subspaces M_ and M+ of 1i such that 1i == M_ EB M+ and
such that the restrictions A- :== AM_ and A+ :== AM+ of A to these subspaces are
self-adjoint operators (in M_ and M+ respectively), with A-< 0 and A+> 0.

PROOF. Let us set Z == (A+ i)- 1 +(A- i)- 1 . This operator is self-adjoint 20 ,
with liZ II < II (A+ i) 1 11 II (A i)- 1 11 < 2 (see Proposition 2.37). Formally one
has Z == 2A/ (A 2 1). Thus A will be positive or negative on the same sub spaces
as Z. Consequently, if {lEp} denotes the spectral family of the bounded self-adjoint
operator Z, we take M_ - JE( ( -oo, 0])1-i and M+ == JE( (0, oo) )H. We then define
A_- AM_ by

V(A-) M_ n D(A), A-f==Af forjEV(A-). (4.105)

(i) We first show that D(A-) == (A+ i)- 1 M_ == (A i)- 1 M_. We shall write
lE_ - JE( (-oo, OJ), so that M_ == lE_H.
Since (A ± i) - 1 commute with Z, property (I) of §A.4.1 (page 183) allows us
to infer that [(A i)- 1 , JE( (-oo, OJ) J - 0 (recall from Remark 4.38(a) that JE_
JE(( -oo, OJ) is the projection == PM(Z) associated with bounded the self-adjoint
operator Z). Hence
[(A i)- 1 , lE_] == 0. (4.106)
Now let f E D(A- ). Then, using (4.106):

f -lE_f == lE_(A i)- 1 (A ± i)f- (A i)- 1 1E_(A i)f,

which shows that f ==(A i)- 1 9± with 9± EM-. Thus D(A-) C (A i)- 1 M_.
For the converse inclusion we observe that the ranges of (A i) - 1 are contained
in D(A) and that (A i)- 1 map M_ into itself: if hEM_, then (A i)- 1 h -
(A±i) 1 lE_h==lE-(A i)- 1 hEM_.
(ii) Next we show that A_ is a self-adjoint operator in the Hilbert space M_.
Iff E V(A-), then one has f ==(A+ i)- 1 9 with 9 EM_, as seen above. Thus
A-f == Af - 9- if EM_. This shows that A- is indeed an operator in M_. Its
domain is dense in M_: If hEM_ and c > 0, there exists a vector 9c E 1i such
that llh-(A+i)- 1 9cll <c. Then llh-(A+i)- 1 1E-9cll- II1E-[h-(A+I)- 1 9c]ll <c.

20 Because (A- i) 1 =(A+ i)- 1 *, see (2.103).


APPENDIX TO CHAPTER 4 189

Thus each vector h E M _ is the strong limit of a sequence of vectors belonging


to(A i)- 1 M_==D(A-).
Since A == A*, it is clear that A_ is a symmetric operator in M _. Furthermore
(A_± i)D(A-) - (A i)D(A-) - (A i)(A ±i)- 1 M_ - M_. By Proposition
3.3 we conclude that A- is self-adjoint.
(iii) Let us check that A_ is a negative operator. The identity 2A == (A _ri)+ (A +i),
valid on D(A), implies that 2(A- i)- 1 A(A + i)- 1 == (A+ i)- 1 (A- i)- 1 == z.
So let f E D(A- ). Writing f- (A+ i)- 1 g with gEM_, we then have
1
(f, A-f)- (f,Af) == (g, (A- i)- 1 A(A + i)- 1 g) == 2 (g, Zg) < 0.

(iv) By reasoning as above, one finds that the restriction A+ of A to M+ is a


positive self-adjoint operator in M+ [here one has (f, A+ f) > 0 for each f -1- 0 in
M+, with strict inequality, because Z > 0 in M+ - JE( (0, oo) )H]. D

Since A- and A+ are semi bounded, one can obtain the spectral family { E:>..} of A
from the spectral families of A_ and A+. If {JE~} denotes the spectral family of A-
in M_ and {JEt} that of A+ in M+, one has in the decomposition H == M_ E9 M+
of the Hilbert space H:

if .A< 0
if .A> 0

Alternatively one could follow the proof of the Spectral Theorem given for bounded
operators in §A.4.1: by applying Proposition 4.39 to the operator A>.. A - .AI, one
obtains a decomposition of this operator into A>.. == A>..- EBA:>..+ in M:>..- EBM:>..+, with
A>..- < 0 and A>..+ > 0. The spectral projection E:>.. of A is the projection onto the
subspace M:>..-. By taking into account the arising domain questions, one can repeat
the verifications presented in §A.4.1 (the details are given for example in §VI.5.3 of
[K]).

REMARK. It is clear that the result of Proposition 4.39 is true in particular if A is


a bounded self-adjoint operator. For such an operator we have already obtained in
§A.4.1 a decomposition into a sum of a positive operator A+ and a negative opera-
tor A-. The following difference should be noticed: the two operators A+ and A-
given in §A.4.1 are operators acting in H, whereas the operators A+ and A- con-
structed in Proposition 4.39 are operators acting in some subspace of H (in M+ and
M_ respectively). The relation between these different operators is as follows: in the
decomposition H - M_ E9 M+, A+ acts in M+, A_ acts in M_, and one has
A+ - 0 E9 A+ and A- == A- E9 0 (see also Remark 4.38).
190 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

Bibliographical Notes
As regards measure theory, we refer to the Bibliographical Notes for Chapter 1.
Stieltjes measures and integrals are also discussed in [CV]. Properties of monotone
functions (in particular their differentiability almost everywhere) are treated in [RN]
(§ 1- §3 of Chapter I and§ 1- §2 of Chapter VIII) and in [Na] (§6 of Chapter IX). The
spectral theory of self-adjoint operators is presented for example in the texts cited in
the Bibliographical Notes for Chapter 2.

Problems
4 . 1 . Let F: JR -7 [0, p] be a function satisfying the conditions (1), (2) and (3) given at
the beginning of Section 4.1.
(a) Show that the set of points of discontinuity ofF, if not empty, is at most countable.
(b) Let Fp be the unique jump function having the same discontinuities as F [see Eq.
(4.8)]. Show that F- Fp is continuous and monotone non-decreasing.
[Hint for (a): For n EN, consider the set of points A E JR at which F(A)- F(A- 0) >
1/n.]
4 . 2 . Let {E.\} -AEIR be a spectral family.
(a) Prove that the strong limits s-lims----7+0 E 11 ±s exist for each f-L E JR and that these
limits define projections E 1L±O satisfying EtL-o < ElL < EtL+O·
(b) Verify that for A, f-L E JR:

(4.107)

Thus A~ E-Af is strongly continuous if and only if A~ IIE-Afll 2 is continuous.


4.3 . Let A be a self-adjoint operator. Show that Hs(A) == Hp(A) EB Hsc(A).
4.4 . For A E JR denote by C -A (JR) the set of continuous functions cp: JR - 7 CC having
compact support and satisfying cp( A) -1- 0. Let A be a self-adjoint operator. Show that
the spectrum and the essential spectrum of A are given by the following expressions:

a-(A) == {A E JR cp E c,\ (JR)


1 ===} cp(A) -1- o}

o-ess (A) - {A E IR I cp E C.\ (JR) ===} cp(A) is not compact}.


4 . 5 . Let A be a bounded self-adjoint operator. Show that the continuous unitary one-
parameter group associated with A (Example 4.12) is given by the Taylor series for
e-iAt, with convergence in the operator norm. If A is unbounded, this series converges
strongly when applied to vectors of compact support with respect to the spectral fam-
ily of A.
4.6 . Let A be the multiplication operator in £ 2 ( 0, m) by a real function a E L 0, m).
(X) (

Determine the spectral projections of this operator and discuss its spectrum.
PROBLEMS 191

Consider the self-adjoint extensions K (a) of the minimal operator -d 2 / dx 2 in


((0, oo )), a E [0, 1r) (see page 107).
(a) Determine the eigenvalues (if any) of K(a).
(b) Show that each A > 0 belongs to the spectrum of I< (a).
(c) What can be said about the nature of the spectrum of K (a) on the interval [0, oo)
and on the interval ( -oo, 0)?
[Hints. For (b): Use Proposition 4.20 by considering functions of the form f(;r;) -
fJ(x)eivfXx with 1J E C0 ((0, oo)). For (c): Example 4.25.]
00
4. 8 . Let be a convolution operator in L 2 (JR): (Bf) (x) == ,[_00 b( x- y) f (y )dy, where
b is a fixed function belonging to L 1 (JR) n (JR). Find the spectrum of B.
Consider a differential operator H == -d 2 / dx 2 + V(x) in (JR), where V
J
a bounded real-valued function of compact support with JR V (x) < 0. Show that
a ess (H) == [0, oo) and that H has at least one negative eigenvalue.
[Hint: Show that one can find vectors f E L 2 (JR) for which (f, H f) < 0 by consider-
ing functions of the form gn(x) == g(x/r~) with g E C0 (JR) satisfying g(x) - 1 for
lxl < 1.]
4 . 10. Let { E:>..} be the spectral family of a self-adjoint operator A. Let ~1 be an interval
and set N dimR(E(J)) _ dimM(J).
(a) If N < oo, then the spectrum of A in J consists of only isolated eigenvalues of
finite multiplicity. The sum of the multiplicities of these eigenvalues is equal toN.
(b) If N == oo and J is bounded and closed, then a ess (A) n J -1- 0.
(c) If dim R( E 1L) < oo for some IL E JR, then A is lower se1nibounded.
4 . 11 . Let A be a self-adjoint operator in a Hilbert space 1-i and M (V) == E(V)H,
where { E( ·)} denotes the spectral measure of A. Show that
(a) For each Borel set V the subspace M (V) is invariant under the operator A [i.e.
f E M(V) n D(A) ====? Af E M(V)].
(b) If Vis a bounded set, then the restriction AM(V) of A to the subspace M (V) [see
(4.70)] is a bounded operator.
(c) If Vis a closed set, then a(AM(V)) C V.
4 . 12 . Let A be a self-adjoint operator and let A E JR be such that dist( A, a P (A)) > 0.
Show that A belongs to the resolvent set of the operator Ap.
4 . 13 . Let A be a bounded self-adjoint operator in an infinite-dimensional Hilbert
space. Show that the following three statements are equivalent: (i) a ess (A) is not
empty and A is compact, (ii) a ess (A) == {0}, (iii) A is compact. What can be said
if the condition of boundedness of A is dropped?
4 . 14. Let A be a compact self-adjoint operator.
(a) Show that IIA I or -II A I must be an eigenvalue of A.
(b) Let {An} be the eigenvalues of A and mn the multiplicity of An. Prove that A is a
Hilbert-Schmidt operator if and only if LnEN mnA~ < oo.
4 . 15 Let { E:>..} be a spectral family and V the set of vectors in 1-i having compact
support with respect to {E:>..}, i.e. V == {f E 1-i E(J)f == f for some bounded
j

interval J}. Let A be the self-adjoint operator associated with { E :>..}. Show that A is
essentially self-adjoint on V.
192 SPECTRAL THEORY OF SELF-ADJOINT OPERATORS

4 . 16. An operator A E B(H) is called a normal operator if it commutes with its ad-
joint, i.e. if A *A == AA *.
(a)AisnormalifandonlyifiiA*fll == IIAJII foreachjEH.
(b) Denote by RA == (A+ A*) /2 the real part of A and by ~A - (A - A*)/ (2i) its
imaginary part. Then A is normal if and only if RA and ~A commute.
(c) A bounded self-adjoint operator is normal.
(d) Let A be a self-adjoint operator and cp: IR ---7 C a bounded continuous function.
Then cp (A) is a normal operator.
(e) The resolvent (A - z) - 1 of a self-adjoint operator (for z outside the spectrum of
A) is a normal operator.
REMARK. The Spectral Theorem has an extension to normal operators.
4. 17. Polar decomposition of a bounded operator
Let A E B(H) be a bounded operator in a Hilbert space H.
(a) Show that the operator A *A is self-adjoint and positive. We denote by IAI its pos-
itive square root: IAI == (A *A) 112 , i.e. IAI - cp(A *A) with cp(.\) - +.\ 112 (,\ > 0).
(b) Show that there exists a partial isometry U with initial set R( IA I) and final set
R(A) such that A- UIAI (see Section 2.2 for the notion of a partial isometry).
(c) Show that, if A is invertible in B(H), then U is unitary.
(d) Let Q be the operator of multiplication by the variable in £ 2 ( [-1, 1]): (Q f) (x) ==
xf(x), x E [-1, 1]. Give the polar decomposition of Q.
(e) Let A be the multiplication operator in £ 2 (0, m) by a function a E £ 00 (0, m).
Find the polar decomposition of A.
CHAPTERS

Evolution Groups and Scattering Theory

In this chapter we develop the principal aspects of quantum scattering theory in a


Hilbert space framework. If H is the Hamiltonian of a quantum-mechanical system,
and if H does not depend on time, then the evolution of a state vector f E 1{ during a
time interval of length tis obtained by acting on f with the unitary operator e-iHt:
f is the state vector of the considered system at time t 0 , then the state vector at time
t 0 +tis given by e-iHtf (in quantum mechanics the time evolution off is given by
e-iHtfnJ, and in this text we set n == 1). Thus, for time-independent Ha1niltonians,
the evolution of all possible states can be described in terms of the continuous unitary
group { e-iHt}tElR associated with the Hamiltonian H, as described in Example 4.12.
Scattering theory deals with the behaviour of scattering states (normally the states
associated with the continuous spectrum of H, as we shall see) at very large times,
viz. for t - 7 ±oo, by comparing their evolution at very large times with a "simpler"
evolution (called the free evolution, i.e. the evolution in the absence of interactions or
external forces).
The family of operators { e-iHt }tElR forms a strongly continuous one-parameter
group of unitary operators. In Section 5.1 we derive some abstract mathematical prop-
erties of such groups, and in the following sections we present the basic aspects of
scattering theory, in particular the definition of scattering states, wave operators, the
scattering operator and scattering cross sections.

5.1 Evolution groups

5.1 . 1. If is a self-adjoint operator in a Hilbert space H, then for each t E IR the


operator Ut :== e-iAt is unitary, and the collection {Ut}tElR is a unitary one-parameter
group that is strongly continuous (see Example 4.12). To simplify the notations we
shall say that it is an evolution group (we think of the situation that will be of interest
in this chapter, in which will be the Hamiltonian of a quantum system). Thus by
definition an evolution group is a family { Ut }t E lR of unitary operators, indexed by a
194 EVOLUTION GROUPS AND SCATTERING THEORY

real parameter t, such that

(5.1)

We observe the following consequences of this definition:

Uo ==I and (5.2)

The first identity in (5.2) follows from the first formula in (5.1 ); if one takes s == t == 0
in the first identity in (5.1) one obtains U0 U0 == [!0 , and upon multiplying this relation
by * and using the isometry of U0 (i.e. U0U0 == I), one arrives at U0 == I. For the
last identity in (5.2) we observe that the unitarity of Ut implies that Ut == [Ut] - 1 (see
Section 2.2) and that U1 U-t == U-tUt == Uo == I, hence [Ut] - 1 ==
According to Exa1nple 4.12 each self-adjoint operator A determines an evolution
group by the formula Ut == e -?At (the exponential being defined by the functional
calculus or as a series, as explained in Example 4.12). It is an interesting fact that each
evolution group is necessarily of this form, i.e. that, given an evolution group {Ut}tEJR,
there exists a (unique) self-adjoint operator A such that Ut == e-~At. This is the con-
tent of Stone's Theoren1 which we prove below. To motivate this theorem, consider
the evolution group e-~At associated with a self-adjoint operator A and let us try to
calculate A fron1 the group. Formally it is clear that (d/ dt)e-iAt == -iAe-iAt, which
suggests the relation A i(d/ dt)e-~At lt=o· A is unbounded one has to consider
this relation on the domain D(A) of A. Indeed, iff E D(A), then the vector-valued
function t f----7 e-~Atf is strongly differentiable and its strong derivative, evaluated at
t == 0, is equal to -iAf, i.e. one has li1nt-+O llt- 1 [e-iAt- I]f- ( -iAf) I - 0 for each
f E D(A). This is easily obtained by applying the Dominated Convergence Theorem:

2
and j 1 [e- 1.At - 1] + i A j converges to zero for each fixed A as t ------+ 0 and can be
1najorised as follows:
-~ A.t 1 2 - ? A.t 1 2
+ iA + 2liAI 2 ;
I

e t- <2 e t-
I 1 1

the right-hand side is integrable with respect to the measure mf iff E D(A) since

because X -11 < 1 vxEIR.


Proposition 5.1 (Stone's Theorem). Let { Ut} be an evolution group in a Hilbert
space H. Define a linear operator A in Has follows:

V(A) = {f E 7-i I ==J s-lim~


t-+0 t
[Ut - I]!}, (5.3)

iff E D(A). (5.4)


EVOLUTION GROUPS 195

Then D( A) is dense in H and A is a self-adjoint operator (called the infinitesimal


generator of the group {Ut} ). One has Ut == e-iAt (i.e. {Ut} is the strongly contin-
uous unitary group associated with A), the domain D(A) is invariant under each Ut
[i.e. f E D(A) ===> Utf E D(A)], and

i! Utf = AUtf ~ UtA! Vf E D(A). (5.5)

J
PROOF. (i) Let z == x + iy with x, y E JR andy > 0. We set Rz - i 0 e~z 8 U3 ds. This
CX)

integral exists by Proposition 2.3( c) because the integrand is strongly continuous and

By also taking into account (2.25) one sees that Rz E B(H) and IIRz II < y- 1 . Let
us also show that the range of Rz is dense in H. We must show that if g E H is such
that (g,Rzf) == 0 for all f E H, i.e. if g is orthogonal to R(Rz), then g == 0. For
this we observe that RzUt - UtRz for all t E JR, which implies that (g, UtRzf) -
(g, Rz Ut f) - 0 for all f E H and all t E JR. By inserting the definition of Rz into this
formula one obtains (setting T - t + s)

Hence ft eizT (g, UTf) dT - 0. Differentiating this identity with respect to t leads to
CX)

Hence (g, Utf) - 0 for each t E JR, in particular fort - 0. We have thus shown that
(g, f) == 0 for each f E H, which implies that g == 0.
(ii) We now show that Rzf E D(A) for each f E Hand also calculate the action of
A on the vector Rzf. We have

1 -izt.! eizTU fd T--i R f - -i ( e -izt


-_ --e CX)

t T t z t
t

As t ~ 0, the right-hand side converges strongly to zRzf +f. Thus we have:


(1) Rzf E D(A) for all f E Hand (2) ARzf == zRzf + f, i.e.

(A z)Rzf- f Vf E H. (5.6)

Since the set Rz His dense in H, (1) implies that D(A) is dense in H.
(iii) By taking z == i in (5.6) one sees that R(A - i) H. One can repeat the
arguments given in (i) and (ii) for z - x + iy with y < 0, by setting in this case
196 EVOLUTION GROUPS AND SCATTERING THEORY

Rz -i~[~CX)e"'zsusds. One finds that (5.6) holds also in this case, in particular (for
z == -i) that R( A + i) == H. Thus, to conclude that A is self-adjoint, it only remains
to show that it is symmetric (see Proposition 3.3). So let f, g E D(A). Then (using
ut == u_t):
\Af,g) -liln \it- 1 (Ut- I)f,g) ==lim \f, -it- 1 (U-t- I)g)
t~o t~o

1
lirn \f, is- (Us -I) g) -
s~o
\J, Ag).

(iv) We now show that D(A) is invariant under the group. Assume that f E D(A)
and s E JR. We must check that Usf E D(A). We observe that Us (it- 1 [Ut - I] f) -
it- 1 [Ut - I]Usf· Since f E D(A), the left-hand side converges strongly to UsAf
as t ~ 0. Consequently the right-hand side is strongly convergent as t ~ 0, which
means that Usf E D(A). In addition the strong limit of the right-hand side is equal to
AUsf, which establishes (5.5).
(v) We finally show that Ut - e-iAt. Since Ut and e-iAt belong to B(H), it suf-
fices to prove that \g, Utf) == \g, e-iAtf) for each f in some dense subset D and
all g in H. We take forD the set of vectors of compact support with respect to the
spectral family of A (see page 144). If f has this property, then f E D(Ak) for
each k E N. Consequently, by (5.5), the mapping t ~ Utf is strongly of class CCX)
and (d/ dt)kUtf == ( -i)kUtAkf. From the argument given before the statement of
Stone's Theorem, one sees that t ~ e-iAtf is also of class CCX), with derivatives
(djdt)ke-?,Atf == ( -i)ke-iAt Akf. Thus one has for each g E HandfED:

fork == 0, 1, 2, ...

This shows that the functions t ~ \g, Utf) and t ~ \g, e-iAtf) have the same Taylor
expansion at t == 0, hence they are equal. Consequently Utf - e-iAtf for fED. D

Corollary 5 . 2. Let Ut and A be as in Proposition 5.1. Then the resolvent of A is given


as.follows in terms of the group {Ut}:

if ~z > 0: (A- z)- 1 = i.LXJ eiztUtdt (5.7)

if ~z < 0: (A- z) 1
= -iL~eiztUtdt. (5.8)

PROOF. SinceAisself-adjoint,onehaszEp(A) if~z # O,hence(A-z)- 1 EB(H).


Upon multiplying (5.6) by (A- z )- 1 one obtains Rzf == (A- z) 1f for each f E H,
i.e. Rz is identical with the resolvent of A at the point z. By taking into account the
definition of Rz in the preceding proof one obtains (5.7) and (5.8). D

For explicitly given groups it is often easy to calculate the derivative of Ut f, hence
the infinitesimal generator, for a set of particularly simple vectors f [for example for
f E S (lR 71 ) in quantum mechanics of a particle]. If this set D of vectors is a core for the
EVOLUTION GROUPS 197

infinitesimal generator, i.e. if the restriction A 0 of the infinitesimal generator to this


set of vectors is essentially self-adjoint, this determines the infinitesimal generator
completely (it is simply the closure of A 0 , both deficiency indices of A 0 being zero).
The next proposition gives a criterion for this to hold.

Proposition 5 . 3 (Nelson's Criterion) . Let {Ut} be an evolution group and A its in-
finitesimal generator. Let D be a linear submanifold of D(A) having the following
properties: (1) Vis dense in 1{, (2) Dis invariant under the group, i.e. Utf ED for
each f E D and each t E JR. Then the restriction of A to D is essentially self-adjoint.

PROOF. Let us denote the restriction of A to D by A 0 . Then Ao is a symmetric oper-


ator. By Proposition 3.3(b) we must show that R(Ao - i)_L N(A 0 i) == {0} and
R(A 0 + i)_L N(A 0 - i) == {0}. Let us show for example thatN(A 0 - i) == {0}.
Assume that h EN(A 0 - i), i.e. hE D(A 0) and A 0h- ih. Let fED. Then

~ (Utf, h) = ( - iAUtf, h) = i(AoUtf, h)


== i(Utf, A() h) - i(Utf, ih) == - (Utf, h)
[using Utf E D(A) and Utf ED in the first and second equality respectively]. Thus
the function ¢ defined as ¢( t) == (Ut f, h) satisfies the differential equation ¢' (t) -
-¢(t); hence ¢(t) - ¢(0)e-t. If ¢(0) i- 0 then l¢(t)i -+ oo as t -+ -oo. which
contradicts the fact that I(Utf, h) I < I f 1111 hll. Consequently ¢(0) == 0, i.e. (f, h) == 0.
This shows that his orthogonal to the dense set D, hence h - 0. D

Example 5.4.. Interesting examples of evolution groups are obtained by considering


unitary representations of one-parameter transformation groups of a set 0. We con-
sider here the translation group acting on JR: each value of the parameter t induces
a mapping JR -+ JR given by JR 3 x r-+ x t, and the collection of these mappings,
as t varies over JR, forms a one-parameter group of transformations of 0 JR. If one
2
defines operators Ut in 1{ == L (JR) by (Utf)(x)- f(x-t), one obtains a continuous
unitary one-parameter group in this Hilbert space. Clearly Us Ut == U ( s + t), and by a
simple change of variables one finds that

(g, Utf) = L:g(x)f(x- t)dx = L:g(y + t)f(y)dy = W-tg,J),

and similarly that 11Utfll 2 == 11!11 2 . These relations imply that Ut E B(H) and that
Ut == U-t· Then UtUt == U_tUt - Uo == I and also UtUt == I, which gives the
unitarity of Ut. To verify the strong continuity of { Ut}, we observe that

If f is continuous and bounded, the integrand converges to zero as E -+ 0 and is ma-


jorised by the constant 411 f II~. If furthermore f vanishes outside some finite interval
[a, b], then forE E [-1, 1] the domain of integration in (5.9) reduces to [a- 1, b + 1],
198 EVOLUTION GROUPS AND SCATTERING THEORY

and an application of the Dominated Convergence Theorem (Proposition 1.9) shows


that this integral converges to zero as E ~ 0. This gives in particular the strong con-
tinuity of Utf for each f E C 0 (IR). Since C 0 (IR) is dense in L 2 (IR), it follows by
virtue of Proposition 2.2 that Utf is strongly continuous for each f E H.
Let us determine the infinitesimal generator of { Ut}. Differentiation with respect to
t of (Utf) (x ), for fixed x, shows that (8 I 8t) (Utf) (x) lt=O == f' (x). This suggests
that the infinitesimal generator should be - i ( dI dx). To make this precise one has to
show for suitable vectors f that llt- 1 [Ut- I]f +!'II ~ 0 as t ~ 0. So consider

(5.10)

Let us take again f E C 0 (IR) and apply the Dominated Convergence Theorem as
above. The integrand converges to zero as t ~ 0, and is bounded by 411!'11~ (since
lf(x-t)-f(x)i == IJ;-tf'(y)dyj < itillf'lloo).ThusCc)(IR)belongstothedomain
of the infinitesimal generator, and on C 0 (IR) this operator is given by -i( dl dx ).
We know that -i (dI dx) is essentially self-adjoint on C 0 (IR) and that its closure is
the momentum operator P. So the infinitesimal generator of the translation group in
L 2 (IR) is the momentu1n operator P. One has Ut == e-iPt.
A shorter way of arriving at the preceding result is to make a Fourier transformation
and express the group {Ut} in momentum space: [FUtf](k) == e-iktj(k). In the
momentum representation of the Hilbert space, {Ut} is multiplication by e-ikt and
thus coincides with the operator cp(P) for cp(,\) - e-i>.t, see Exa1nple 4.12. By the
uniqueness of the infinitesimal generator we conclude that it must be the operator P.
In this context another useful example concerns the dilation group in IRn, given
by IRn 3 x ~ e-t;t (where again t varies over IR). The unitary representation of this
group in L 2 (IRn) and its infinitesimal generator are discussed in Problem 5.4.

On various occasions we shall meet expressions involving products of evolution


groups. To calculate the derivative of such an expression, one uses the Leibniz rule.
However, some care is necessary when the infinitesimal generators of the occurring
evolution groups are unbounded. The following result describes situations in which
the formal application of the Leibniz rule is justified.
Proposition 5 . 5 . Let A and B be self-adjoint operators.
(a) Let C E B(H) be such that CD( B) CD( A). Then, for each fED( B), the vector-
valued function t ~ eiAtce-iBtj is strongly differentiable, and

(5.11)

(b) Let C be B-bounded and such that CD(B 2 ) C D(A). Then, for f E D(B 2 ), the
vector-valued function t ~ eiAtce-iBtf is strongly differentiable and its derivative
is given by (5.11).
PROOF. (a) Setting Ut == e-iAt and Wt == e-iBt, we have for f E D(B):

~ [Ut+rCWt+r f ~ Ut CWtf] = [~ (Ut\r ~ Utj CWtf + Ut+rC [~ (Wt+r ~ Wt) f].


EVOLUTION GROUPS 199

Since CWtf E D(A), the first term on the right-hand side is strongly convergent to
iUt ACWtf as T - 7 0. For the second term we have Ut+TC - 7 C strongly and
1
T- )f - 7 -iBWtf strongly as T - 7 0. Then, by a straightforward adap-
tation of the argument in (2.16) (which requires the boundedness of C), one obtains
that second term on the right-hand side is strongly convergent to -iUtCBWtf·
(b) f E D(B 2 ), then (B + i)f E D(B). Since the operator Co :== C(B + i)- 1
belongs B(H) and maps V(B) into D(A), we obtain by applying the result of (a)
e Atce-~Btf == eiAtC0 e-iBt(B + i)f is strongly differentiable, with derivative
2

ie At (AGo- C 0 B)e-'&Bt (B + i) f. The last expression is identical with the right-hand


1

side .11). D

We continue by describing some results on the asymptotic behaviour (t -7 ±oo)


of Utf· We begin with a classical result from Fourier analysis:

Proposition 5.6 (Riemann-Lebesgue Lemma). Iff E £ 1 (JRn ), then its Fourier trans-
fonn J is a continuous junction converging to zero at infinity: f(k) - 7 0 as !kl - 7 oo.

PROOF. To simplify the n~tations we consider the case n - 1. Iff belongs t_? S (JR),
then its Fourier transform f also belongs to S (JR); in particular the function f is con-
tinuous and converges to zero at infinity. Iff E £ 1 (JR), there is a sequence {f7 } of
functions in S(JR) such that II!- fj!h - 7 0 as j - 7 oo, see §1.4.5. Then

1/(k)- };(k)l = vk II: e-ikx [f(x)- fi(x)] dx I


00

< 1 /
rn= jJ(x)-f1 (x)jdx- r.LIIf-fjlh·
1 (5.12)
y 21T . - 0 0 v 21f
Since the final expression is independent of k and converges to zero as j - 7 oo, one
~ ~ -
concludes that fj ( k) coEverges to f ( k) as j - 7 oo, ~niformly in k E JR. Hence f is
continuous (since each f 1 is continuous). To see that f converges ~o zero~ infinity,
us fix c > 0. By (5.12) we ~an choose a number j EN such that If( k) f 1 (k) I < c /2
for all k E JR. Then, sine~ fj (k) tends to zero at infinity, we can choose a real number
k0 E (0, oo) such that lfj (k) I < c/2 for each k E JR with lkl > ko. Thus one has for
lkl > ko:
j](k)j < j](k)- h(k)j ih(k)j < c/2 + c/2- E~
~

which shows that f converges to zero at infinity. D

In the next proposition we consider the asymptotic behaviour of vectors in the


absolutely continuous subspace of the infinitesimal generator of an evolution group.

Proposition 5.7. Let Ut be an evolution group and A its infinitesimal generator. Let
f E Hac(A). Then:
(a) Utf converges weakly to zero as t - 7 ±oo.
(b) Let BE B(H) be an A-conlpact operator. Then I!BVt.f!l - 7 0 as t - 7 ±oo.
200 EVOLUTION GROUPS AND SCATTERING THEORY

(c) Let B E B(H) and assunze that there exists an operator C E B(H) with dense
range (i.e. R( C) j_ == { 0}) and such that BC E JC (H) and CUt == Ut C for all t E YR..
Then IIBUtfll-+ 0 as t-+ ±oo.
PROOF. (a) (i) If hE Hac(A), then mh [see the definition (4.37)] is an absolutely con-
tinuous measure, there is a non-negative function() E Lfoc (YR.) such that mh (V) ==
fve(A)dA for all V E AB. In particular m,h(TR.) == J_~ ()(A)dA == llhll 2 < oo, hence
() dmh/ dA belongs to £ 1 (YR.). Now observe that

(h,Uth) = L: e~i>.tmh(dA) = L: e~i>.te(A)dA.


So cp(t) :== (h, Uth) is the Fourier transform of the function() E L 1 (TR.), consequently
cp(t) -+ 0 as It I -+ c:xJ by the Riemann-LebesgueLemma. Thus limt~±oo (h, Uth) 0
for each hE Hac(A).
(ii) We now show that (g, Utf) -+ 0 as t -+ ±oo for each g E H and each f E Hac (A).
f E Hac(A), then Utf E Hac(A) for all t E JR, hence (g,Utf) == 0 for all t E lR
if g E Hs (A) Hac (A) j_. So it suffices to consider the case in which g E Hac (A).
Then, by virtue of the polarisation identity (2.15), (g, Utf) is a linear combination of
four terms of the form (g + af, Ut(g + a f)), and each of these terms converges to
zero as t-+ ±oo by the result of (i), because g + af E Hac(A) for each a E CC.
(b) By Proposition 2.2 it is enough to show that IIBUtfll -+ 0 as ltl -+ oo for
all vectors f in some dense subset of Hac (A). We take for this dense set the linear
manifold Hac (A) nV(A). So let f E Hac(A) nV(A); we set g == (A +i)f and observe
thatgbelongstoHac(A) andthatf == (A+i)- 1 g. ThusBUtf == BUt(A+i)- 1 g-
B(A i)- 1 Utg. Since Utg converges weakly to zero by (a) and B(A + i)- 1 is a
co1npact operator by hypothesis, we obtain by applying Proposition 2.12(a) that BUtf
converges strongly to zero: limt~±oo IIBUtfll == 0.
(c) The argument here is similar to that in (b), replacing (A+ i)- 1 by C and
Hac (A) n V( A) by Hac (A) n R( C). It suffices to know that Hac (A) n R( C) is dense
in Hac (A) and that C maps the subspace Hac (A) into itself. The last property is
equivalent to [C, E~-J == 0, and this commutation relation can be deduced from the
hypothesis that C commutes with the group {Ut}. Then, since R( C) is dense in H
by assumption, one sees that C must map Hac(A) onto a dense subset ofHac(A). We
o1nit the details since the result of (c) will not be used further on. D

us now consider vectors in He (A) that do not belong to Hac (A). In general,
when f E Hc(A), then Utf is not weakly convergent to zero, as can be seen in explicit
exa1nples. 1 Nevertheless results similar to those of the preceding proposition are true
one introduces an average with respect to the parameter t. Since in many of our
applications t is interpreted as time, we shall then speak of a temporal mean. For
example one can consider the average of Ut f over an interval of length T, and this
average converges weakly to zero as the length T of the interval tends to infinity:

Tl.iT0
Utfdt and T1 !0
-T
Utfdt and
2
1
TJT
-T
Utfdt (5.13)

1 See for example Problem 149 in Chapter XI, Volume III of [RS].
EVOLUTION GROUPS 201

J
converge weakly to zero as T -+ oo, and for example r- 1 0T II BUt! l\ 2 dt converges to
zero as T -+ oo if f E He (A) and B is A -compact. We give below a proof of this last
result. As regards the first one, it will not be needed, but let us mention that there exists
a much stronger result: iff is orthogonal to the subspace M ( { 0}) == N (A) consisting
of all eigenvectors of A associated with the eigenvalue 0, then the expressions in (5.13)
converge strongly to zero. This is a consequence of the following proposition (its proof
is simple, see e.g. Problem 5.6 here or Proposition 4.20 in [A]):

Proposition 5 . 8 (Mean Ergodic Theorem). Let {Ut} be an evolution group, A its in-
finitesimal generator and E( {0}) the projection onto N(A). Then one has for each
jEH:

s-lim T
T-+±CXJ
11T0
Utfdt == E({O})f.

In particular, iff j_ N(A), then the preceding li1nits are 0.

Results of the type of those given in the next proposition and in Proposition 5. 12
are often cited under the name of RAGE Theorem (e.g. Volume III of [RS] or Volume
II of [W2]).

Proposition 5 . 9. Let { Ut} be an evolution group and A its infinitesimal generator.


Let f E Hc(A). Then
(a) One has for each vector hE H:

1
li1n T
T-+CXJ
j
-T
0
J(h,Utf)J
2
dt=O. (5.14)

(b) If BE B (H) is an A-compact operator, one has

and lim Tl
T-+CXJ
jo IIBUtfll
-T
2
dt = 0. (5.15)

(c) More generally, let BE B(H) and suppose that there exists an operator C E B(H)
with dense range (i.e. R(C)j_ - {0}) and such that CUt == UtC for all t E lR and
BC E JC(H). Then (5.15) is true.

PROOF. We shall prove the first relations in (5.14) and (5.15), and we shall o1nit the
proof of (c). Before proving (a) we show that the result of (b) follows from that of (a).
(b) We may assume that B # 0. Given f E Hc(A) and E > 0, we fix a vector fc: # 0
in Hc(A) nD(A) such that II!- fc:l\ 2 < c/(9\IBI\ 2 ), and we set g == (A+i)fc:· Hence
BUtfc: == B(A + i)- 1 Utg, and g E Hc(A). We choose a finite rank operatorS such
that IIB(A i)- 1 S\\ 2 < c/(9\\g\\ 2 ), and we writeS in the form (2.42): Sh ==
~~= 1 (gk, h)hk (where h varies overHand gk, hk are fixed vectors). Then, by using
(1.58) with N- 3 and with general N, we may write:
EVOLUTION GROUPS AND SCATTERING THEORY

Sll ll9ll + 3jj L


2 2 2 2 2
< 3IIBII IIf- fc:ll + 3IIB(A + i) l_ (gk, Utg)hkll
k=l
N
<-
E
3
-
3
3NL llhkll 2
j(gk,Utg)j
2
.
k=l

follows that

Each term in the sum on the right-hand side converges to zero as T ----+ oo,
by (5.14 ). Thus there exists T 0 such that the right-hand side is less than c for each
> T0 , which proves (5.15).
(a) h E Hp(A), then (h, Utf) == 0 for all t E IR [assuming that f E He(A)],
hence .14) is trivially true. Thus it suffices to prove (5.14) under the assumption that
f and h belong to He (A). By using polarisation identity (2. 15) we can express
(h. f) as a linear co1nbination of four terms of the form (h + af, Ut(h +a f)). As
h a f E He (A) if f, h E He (A), it then suffices to prove that

}~ ~ iT/(g,Utg)/ 2
dt = 0 vg E He(A). (5.16)

this we denote { E -A} the spectral family of A, and we fix g E He (A), g i- 0.


> 0 we set

clear that this multiple integral is absolutely convergent:

Thus we may change the order of integration in (5.17):

(5.18)

A fL the integrand jei(.A-JL)T -1]/[i(A- tL)T] j converges to zero as T----+ oo, and it
can be majorised by jei(-A-p)T- 1]/[i(A- tL)T] < 1 for all A, fL (A i- tL) and all T >
j

0. We show below that the subset of IR x IR defined as the set of points { (A, fL) j A == fL}
EVOLUTION GROUPS 203

is a null set (with respect to the measure m 9 (dA.)m 9 (dJ-1) on JR x JR). Assuming this
result, we can apply the Dominated Convergence Theorem [Proposition 1.9] in (5 .18)
to conclude that w(T) ~ 0 as T ~ oo, which proves (5.16).
It remains to show that the diagonal { (A., 11) I A. == 11} of JR x JR is a null set for
the measure m 9 (dA.)m 9 (dJ-1). The argument runs as follows. Since g E Hc(A), the
function 11 r---+ (g, EJ-Lg) is continuous on JR. It is even uniformly continuous: given
s > 0, one can choose M E (0, oo) such that 1\g,EJ-Lg)l < s/2 if 11 < -M and
0 < 11911 2 - \g,EJ-Lg) < s/2 if f-1 > M (hence l(g,EJ-Lg)- (g,EJ-L'g)l < s if J-1,J-1 <
1

-lVI and I (g, EJ-Lg) - (g, EJ-L' g) I < s if J-1, J-1 1 > M). On the other hand, the function
f-L r---+ (g,EJ-Lg), being continuous, is uniformly continuous on each finite interval, in
particular on [-1\!1 - 1, M + 1].
Now let E > 0 be given. By the uniform continuity of 11 r---+ (g,EJ-Lg), there exists
6 > 0 such that l\g,EJ\+69)- (g,E>.-69)1 < E/llgll 2 for all A. E JR. Thus
oo rJ\+6 joo
j -= m 9 (d>..)} >-.-o m 9 (dp,) = -= m 9 (d>..) [ (g,E>-.+Og) - (g,E>-.-8g)]
E joo
< ~ -=m 9 (d>..) =E. (5.19)

Let 06 - {(A., M) II A.- Ml < 6} C JR x JR. The set 06 is a neighbourhood of the


diagonal {A. == J-1} in JR x JR. The inequality (5.19) shows that the measure of 06
is less than E; since E is an arbitrary positive number, the measure of the diagonal
{(A., M) j A== M} must be zero. D
We have shown that iff E Hac(A), then Utf converges weakly to zero as t ~ ±oo,
whereas for f E Hsc this is not necessarily the case. Nevertheless, iff E Hsc(A),
then Utf converges weakly (even strongly by Proposition 5.8) to zero in temporal
mean. Also Proposition 5.9(a) asserts that then limy--7 00 r- 1J0TI (g,Utf) l 2 dt == 0 for
each vector g E H. This implies that I (g, Utf) I must be small for sufficiently many
values oft, in particular that there must exist a sequence of real numbers {tk} tending
to infinity such that (g, Utk f) ~ 0 ask ~ oo. This is a consequence of the following
fact:
Let cp: [0, oo) ~ [0, oo) be a bounded non-negative function. Then:
(i) if c.p(t) ~ 0 as t ~ oo, one has limr-7oo r- 1J0T cp(t)dt == 0,
J
(ii) if limr-7oo r- 1 0T c.p( t) dt == 0, there exists a sequence {tk} such that
tk ~ oo andlimk-7ooc.p(tk) == 0.
PROOF. To check these properties, let 0 < s < T and write

T
i0
Tcp (t) dt == T lis0
cp (t) dt + T11Tcp (t) dt.
s
(5.20)

For each fixed s the first term on the right-hand side converges to zero as T ~ oc.
Thus one has for each fixed s > 0:

lim TliTcp (t) dt == lim Tl1Tcp (t) dt. (5.21)


T-7oo o T-7oo 3
204 EVOLUTION GROUPS AND SCATTERING THEORY

y(t) ~ 0 as t ~ oo and if s > 0 is given, we chooses> 0 such that y(t) < s for all
t > s. Then r- 1
fsT cp(t)dt < s for each T > s, which by (5.21) proves (i).
The result of (ii) can be obtained by reductio ad absurdum. Let us assume that
J
linlr-7(X) r- 1 0T y(t)dt == 0 and that for each s > 0 there is a number 6 > 0 such
that y(t) > 6 for all t > s. Then r- 1fsT cp(t)dt > 6(T- s)IT > 612 for each
> 2s. By taking into account (5.21) one obtains a contradiction with the assumption
that lin1r-7(X) r- 1._f0T cp(t)dt == 0. Hence, for each kEN there exists a number tk > k
such that cp( tk) < 1I k [if one sets s == k above, one cannot have y( t) > 1I k 6 for
all t > k]. D

In the following proposition we indicate a somewhat stronger result: it is possible


to find a sequence { t k} that is independent off and g, i.e. a sequence {t k} converging
to +oo such that (g,Utkf) ~ 0 for all f E He(A) and for all g E H. We omit the
proof, because the result will not be needed in this text (see the Corollary on page 343
of Volume III of [RS]).

Proposition 5.10 . Let { Ut} be an evolution group and A its infinitesimal generator.
Then there exists a sequence {tk} of real nu1nbers such that tk ~ oo and Utk f con-
verges weakly to zero as k ~ oo for each f E He (A). In particular, if B E B (H) is an
A-contpact operator, one has limk-7(X) iiBUtkfll == 0 for each f E He(A).

5@2 Characterisation of the scattering states

In a scattering experiment particles are impinging from very far (from a great dis-
tance compared to the size of the scatterer) on a scatterer (a centre of force) and then
observed again far away from the scatterer. The mathematical description of such situ-
ations involves states having the following property: under the evolution group charac-
terising their temporal evolution they are localised very far from the scattering centre
at very large times (positive or negative). State vectors having this property corre-
spond to scattering states. One can also consider bound states; a particle is in a bound
state if it remains localised near the scattering centre at all times. It is convenient (and
useful if the interaction has spherical symmetry) to identify the scattering centre with
the co-ordinate origin.
In quantum mechanics these notions of localisation can only be approximate as a
consequence of the uncertainty relations. We present below a reasonable mathematical
definition and study some of its consequences. Physicists often use other definitions
which do not have a direct interpretation in terms of the temporal behaviour of states in
configuration space: bound states are identified with eigenvectors of the Hamiltonian
H and scattering states with vectors in its continuous subspace He (H). This has its
origin in the fact that, for most atomic systems, there is no difference between the
two ways of defining bound states and scattering states (that in terms of the temporal
behaviour in configuration space and that in terms of the spectral subspaces of the
Ha1niltonian, i.e. of the infinitesimal generator of the relevant evolution group).
CHARACTERISATION OF THE SCATTERING STATES 205

Consider a quantum mechanical system that is described by the states in the Hilbert
space H == L 2 (1Rn). Let H be its Hamiltonian (a self-adjoint operator in H) and let
{Ut == e-iHt} be the evolution group describing the time evolution of the system (here
the parameter t is interpreted as the time). To describe the localisation in configuration
space one uses the operators xv( Q), where V is a Borel set in JRn (specifying the
region of localisation that one considers) and xv( Q) is the operator of multiplication
by the characteristic function Xv of V, i.e. the projection onto the subspace £ 2 (V) of
L 2 (JRn) consisting of all wave functions f(x) that vanish outside V. We have in mind
the situation in which the scattering centre is close to the origin of the co-ordinate
system, and we shall use only very particular localisation regions V, namely balls B R
of radius R centred at the origin: B R : { x E JRn jlxl < R}. We use the notation x R
for XBn' so that

[xR(Q)f](x) == f(x) if !xi < R, [xR(Q)f](x) - o if !xi > R. (5.22)

A vector f E H will be called a scattering state at t- -oo or at t == +oo respec-


tively if

lim
t --'t - CX)
llxR(Q)Utfll == 0 resp. lim
t --'t + CX)
llxR(Q)Utfll == 0 for each R E (0, oo).
(5.23)
This means that at very large (negative or positive) times the probability of finding
such a state in a finite part of configuration space is very small, or more precisely that
such a state disappears from each given finite region of configuration space as time
tends to infinity: for each bounded subset V of JR.n one has

A vector f E H will be called a bound state if

(5.24)

or equivalently if, given E > 0, there exists a finite ball B R such that at all tinzes the
system in the state f has probability less than E to be localised outside this ball:

- 1
11!11 2
j lxi2::R
I f] (x) 12 dnx < c V t E lR (if R is sufficiently large).

More generally, one could consider states that are bound until some finite time (for
example for all t < 0 but not necessarily as t ~ +oo) or states that are localised after
some finite time (for all times t > 0 but not necessarily as t ~ -oo). This amounts
to replacing in (5.24) suptEIR by supt:s;o or by supt2::0 respectively.
It is interesting to know that, under a certain compactness condition [for example
if the operator XR( Q) is H -compact for each R E (0, oo )], the set of bound states
coincides with the subspace Hp (H) spanned by the eigenvectors of the Hamiltonian
H and the set of scattering states with the subspace He (H) if H has no singular
206 EVOLUTION GROUPS AND SCATTERING THEORY

continuous spectrum. To treat Hamiltonians with singular continuous spectrum one


has to slightly generalise the definition of the scattering states by introducing a time
average, for example by requiring instead of the second condition in (5.23) that

lim T
T ---+CXJ
llT
0
llxR(Q)Utfll 2 dt == 0
---+
VRE(O,oo). (5.25)

This means that the state f disappears from each finite region of configuration space in
time average (roughly speaking: most of the time a system in such a state is localised
far from the scattering centre, but from time to time it could return into a neighbour-
hood of the co-ordinate origin).
We give an abstract theorem that implies the above results. The only relevant prop-
erties of the localisation operators XR( Q) are the fact that they are projections and
that s-lim R---+CXJ XR ( Q) == I. Hence we now consider a general Hilbert space H, an
evolution group { } in 1i and a family { FR} R>O of localising operators having the
following properties 2 :

(a) each FR is a projection: FA== FR == FR (5.26)


(b) s-lim FR ==I. (5.27)
R---+CXJ

For each sign of time we introduce subsets of 1i corresponding to the bound states, the
scattering states and the scattering states in temporal mean. We denote these subsets
by Mt(H), M 0(H), Mto(H), M~(H), Mto(H) andM~(H) respectively (the
sign + or - corresponds to the sign of time, the subscript 0 indicates that it concerns
states that are "close" to the origin, the subscript oo refers to states moving to infinity,
and the over-bar on M indicates an average over time; it is clear that each of the
preceding subsets depends on the Hamiltonian H involved). Hence 3

jEM~(H) ~lim sup II(I-FR)Utfii==O, (5.28)


R---+CXJ tE [0, ±CXJ)

VR < oo, (5.29)

(5.30)

Furthermore we define
Mo(H) == Mt(H) n M 0(H) (5.31)
M(X)(H) == Mto(H) n M~(H) (5.32)
M(X)(H) == Mto(H) n M~(H). (5.33)

We begin by establishing some simple properties of these sets and some relations
between them.
2
In fact condition (a) could be weakened to: (a') F'R == FR E B(H).
3 We recall from the footnote on page 73 that, if an equation or an expression contains double signs, it
is understood that this means two equations or expressions, one by taking everywhere the upper sign, the
other one by taking everywhere the lower sign. This convention will be frequently applied in the present
chapter.
CHARACTERISATION OF THE SCATTERING STATES 207

Proposition (a) Each of the subsets ofH defined in (5.28)- (5.33) is a subspace
of 1-l. Each of these subspaces is invariant under Ut, t E JR.
(b) has (X)(H) c Mto(H) and (X)(H) c M~(H).
(c) One has t(H) j_ Mto(H), 0(H) j_ M~(H) and (H) j_ M!(H).
(d) 1-lp(H) Mo(H) and M!(H) c 1-lc(H).
These results are rather easy to obtain by using the triangle inequality,
the \If+ g\\ 2 < 2\\f\1 2 + 2\\g\\ 2 and the fact that \\FR\\ == \\Ut\l == 1. We
discuss some typical situations.
us show that to(H) a subspace. j, g E Mto(H) and a E C, then

liln \\FRUt(f + ag)ll < t-7CXJ


t-7CXJ lim \IFRUtf\1 + la\lin1
f-7(X) 1\FRUtg\l] == 0.
Thus f + ag E (X)(H), which implies that (X)(H) is a linear manifold. To see
that this linear manifold is a subspace, let us consider a strong Cauchy sequence {fk}
in to(H) and denote its limit by f: \\fk- f\1 ~ 0 ask-* oo. We must show that
f E Mto(H). One has
IIFRUtf\\ == \\FRUt(f- fk) + FRUtfki\
< IIFRUt(f- fk)\1 + 1\FRUtfk\\ <II!- fkll + IIFRUtfkl\.
Given E > 0, first choose an integer k such that \If - fk \\ < E /2, then a nu1nber to
such that 1\FRUtfk\\ < s/2 fort> to. One then has 1\FRUtf\\ < E fort> to, which
shows that f E Mto(H).
(ii) Invariance under Us. It is evident that to (H) is invariant under each Us.
a somewhat less trivial example, let us show that f E 0 (H) ====? f E 6(H).
One has

sup 1\(I- FR)UtUsf\\ == sup \\(I- FR)Ut+sf\\ == sup \\(I- )UTJ\\.


tE[O,CXJ) tE[O.CXJ) TE[s,CXJ)
If s > 0 one has sup 7 E[s,CXJ) 1\(I- UTJ\\ < sup 7 E[O,CXJ) \\(I- FR)U7 fl\, which
converges to zero as R ~ oo by the assumption made on f. If s < 0 we write

sup 1\(I- FR)UTfll < sup II(I- FR)UTJ\\ +sup I - FR)UTJ\\.


TE[s,CXJ) TE[O,CXJ) TE[s,OJ

Again the first term on the right-hand side converges to zero as R ~ oo. For
second term, let E > 0 be given~ choose a number b > 0 such that I f- !II < s/2
all t E [ -b, b], which is possible by the strong continuity of the group { } . Then one
has I f- U(Jf\\ < s/2 if \K:- a\ <b. Since [s, 0] is a finite interval, we can choose
a finite number of points ao, ... , aN such that a 0 == s < a 1 < · · · < aN-l <aN == 0
and \ak- ak-l\ < 2b for each k == 1, ... , N. Then choose for each T E , 0] one of
these points, say a 1, such that \T- aj I < b. With this choice one has

\\(I- FR)UTJ\\ < \I(I- FR) ]1\ +\\(I- FR)(UTJ- Ua


7 7
1\
<\\(I- FR)Uajf\\ + c/2.
208 EVOLUTION GROUPS AND SCATTERING THEORY

Thus
sup II(I- FR)Urfll < sup II(I- FR)Ua 1 ]//-+ s/2,
rE[s,O] j=O, ... ,N

which is less than s if R > R 0 (for some R 0 depending on s and on s) by the hypoth-
esis that s-lim R-7CX) FR == I.
(b) This is evident [take into account the observation (i) made before (5.20) with
cp(t) == IIFRUt.fll 2 ].
(c) Since (.f, g) == (Ut.f, Utg) for all .f, g E Hand all t E JR, we get by using the in-
equality /a+ bj 2 < 2ja/ 2 + 2/b/ 2 and the Schwarz inequality that

l(f,g/1 2 = ~ 1Ti(Utf,Utg/J 2 dt
= ~ foTi(Utf,FRUtg/+((I- FR) 2
J,Utg/l dt

< ~ iifii foTifFRUtgJJ


2 2
dt + ~ JJgJJ foTii(J- FR)Utfli
2 2
dt. (5.34)

Now assume that f E M6(H) and g E M~(H), and let 6 > 0 be given. Since f E
M6(H), onecanchooseanumber R < oo such that II(I -FR) fll 2 llg// 2 < 6/4for
each t > 0. Then last term in the inequality (5.34) is < 6/2 for each T > Since
g E M~(H) the first term on the bottom line of (5.34) converges to zero as T -too,
hence it is < 5/2 by a suitable choice ofT. Thus I (J, g) /2 < 6 for each 5 > 0, which
shows that f j_ g.
(d) Let f be an eigenvector of H: H f == A.f for some A E JR. Then f == e-i)..t.f
and consequently I (I- FR)Utfll == II (I- FR).fll, which does not depend on t. Since
s-linlR____.CX)(I- FR) == 0, one has
li1n sup 1/(I- FR)Utfl/ == lim 1/(I- FR)fl/ == 0,
R-7CX) tElR R-7CXJ
hence f E M 0 (H). Thus the set of all eigenvectors of H belongs to M 0 (H). Since
Mo(H) and Hp(H) are subspaces, it follows that Hp(H) C M 0 (H). Then, since
M!(H) are orthogonal to M 0 (H), one has M!(H) M 0 (H)j_ C Hp(H)j_ ==
Hc(H). D

In the next proposition we give sufficient conditions for having



and MCX)(H) == Hc(H).

Proposition 5. 12. Assu1ne that for each < oo the operator FR is H -compact, i.e.
that FR(H- i)- 1 E JC(H) for each R < oo. Then
-± -
(a) One has (H)== Hp(H) and MCX)(H) == MCX)(H) == Hc(H).
(b) If H has no singular continuous spectru1n, then M~ (H) == (H) == Hac(H). (X)

REMARK. The conclusions of this proposition are true in various other situations. For
example it suffices that one of the following two conditions is satisfied:
(i) FRE([ -1\!I, l\!f]) E JC(H) for each R < oo and each 0 < M < oo [{ E(·)} denotes
the spectral measure of H],
CHARACTERISATION OF THE SCATTERING STATES 209

(ii) there exists an operator C E B(H) with dense range such that CUt == UtC for all
t E lR and such that FRC E JC(H) for each R < oo.
PROOF OF PROPOSITION 5 .12. (a) By taking B - FR in Proposition 5.9(b) one ob-
J
tains that for all vectors f E He (H) and for all R E (0, oo ): r- 1 0±TIIFRUtfll 2 dt ~ 0
as ~ oo. Consequently one has Hc(H) C Mt(H) nM~(H) M(X)(H). The
opposite inclusion M!(H) C Hc(H) has already been obtained in Proposition
5.11(d). Hence one has M~(H) == Mt(H) == M(X)(H) - Hc(H). Then, since
M 0 (H) l_ M!(H) [see Proposition 5.11(c)], it follows that M 0 (H) C Hp(H). The
opposite inclusion is contained in Proposition 5.11(d); thus M 0 (H) == Hp(H).
(b) The argument is similar to that used in (a). By Proposition 5.7(b) we have
limt-7±(X) IIFRUtfll == 0 for each f E Hac(H) and each R < oo. Hence Hac(H) C
M~(H) nM~(H) = M(X)(H). Thus Hac(H) c M(X)(H) c M(X)(H) == Hc(H).
Under the assumption that Hac (H) == He (H) all these inclusions must be equalities,
hence one has in particular that Hac (H) == M (X) (H). D
DISCUSSION. We return to the particularly important situation introduced at the be-
ginning of this section in which H == L 2 (1Rn) and FR xR( Q), with xR( Q) given by
(5.22). To indicate the different situations without distinguishing too many cases we
set n == 3; very similar results hold for the other values of n (the conditions on the
potential depend on n, as seen for example in Proposition 2.49).
(A) We first give an example showing that the conclusions of Proposition 5.12 are not
automatically true. We take H == Q2 in L 2 (JR 3 ). Then XR(Q) commutes with Hand
hence with Ut == e-iHt. Consequently IIXR(Q)Utfll == IIUtXR(Q)fll - IIXR(Q)fll,
which is independent of the parameter t for each f E H (the states do not propagate
at all). Similarly II [I- XR( Q)]Utfll - I [I- XR( Q)]fll. So each f is a bound state
in the sense of (5.24): H == M 0 (H). Since the spectrum of Q2 is purely absolutely
continuous, one hasH== Hac(H) == Mo(H).
Evidently the operator H considered in this example does not correspond to a phys-
ical Hamiltonian. A physical Hamiltonian contains a kinetic part, which is usually a
function of the momentum operator P. If the kinetic energy becomes infinite as the
momentum tends to infinity, then, as shown below, the conclusions of Proposition 5.12
are true.
(B) Next let H cp(P), where cp: JR 3 ~ lR is a continuous function assumed to satis-
fy limlki-7(X)Icp(k)l == oo. In momentum space L2 (1R 3 ), the resolvent (H i)- 1 is
the multiplication operator by 1 I [cp (k) - i], and the function 1 I [cp (k) i] belongs to
L 2 (JR 3 ) + L~(JR 3 ) (by the same reasoning as that applied in Example 2.48). Let us
write

XR(Q)(H- i)- 1 - XR(Q)xR(P)(H- i)- 1 + XR(Q)[cp(P) i]- 1 [I- XR(P)].

The compactness of XR( Q) (H -i)- 1 follows since XR( Q)xR(P) is a Hilbert-Schmidt


operator (by Proposition 2.34) and XR( Q) [cp(P) - i] - 1 is compact [this can be seen
by following the arguments in part (ii) of the proof of Proposition 2.49(a), replacing
V by [cp (P) - i] - 1 and (H o - i) - 1 by X R ( Q)].
210 EVOLUTION GROUPS AND SCATTERING THEORY

As an example, let us take cp(k) -- k 2 , so that fl == flo == P 2 is the free Hamil-


tonian (the kinetic energy operator) of non-relativistic quantum mechanics. Since the
functions k r--+ 1/ (I kl 2 - ~i) and XR belong to L 2 (1R 3 ) one obtains directly from Propo-
sition 2.34(a) that XR( Q) (llo - i) - 1 is a compact operator. So II xR( Q)e-iHot f II ~ 0
for each vector f E £ 2 (1R 3 ), i.e. all state vectors are scattering states (without ti1ne
average)_for the free Ha1niltonian fl 0 . Evidently the case of a general cp covers vari-
ous other situations. Also one can easily adjust the preceding arguments to show for
exa1nple that, for the free Dirac Hamiltonian, all vectors represent scattering states.
(C) Let us now consider Hamiltonians involving a potential, i.e. self-adjoint operators
of the form == H 0 + V (Q), where V is a real function defined on lR 3 . In general
will be a self-adjoint extension of the differential operator - ~ + v (x) defined on
C0 (1R 3 ), but let us consider the simple case in which V (Q) is fl0 -bounded with flo-
bound< 1, so that fl is self-adjoint on the domain V(fl0 ). In this case we can write
[see also the second resolvent equation (2.115)]:

xR(Q)(fl- i)- 1 == XR(Q)(flo- i) 1


· (Ho- i)(H- i)- 1
== [xR(Q)(flo- i) 1
] [I- V(Q)(fl- i) 1
]. (5.35)

operator xn( Q) (Ho-i) - 1 is compact as seen in (B), and [I-V( Q) (H- i)- 1 ] E
B(H). Hence XR(Q)(H i) 1 is compact, and theconclusionsofProposition5.12are
true. This covers a fair number of Schrodinger Hamiltonians in L 2 (1R 3 ), in particular
the Coulomb Hamiltonian: it suffices to have V E L 2 (1R 3 ) + L 00 (lR 3 ), see Proposition

(D) Without giving the proofs we mention some further results that concern the case
in which V(Q) is not H 0 -bounded. The details can be found for example in §2.5 and
§5.1 of [A] (for the question of absorbed states see also §13.1 of [P]). When V(Q) is
not H 0 -bounded, the domain of H will in general be different from that of H 0 . First
all one has to define what is meant by H. We assume that there exists a closed set
Z 1R 3 of Lebesgue measure zero such that potential is only moderately singular
away from Z (more precisely such that Vis square-integrable on each bounded closed
subset of 1R 3 \ Z). It is interesting to know that the results that follow are true for each
self-adjoint extension H of the minimal operator-~+ V(x) defined on C 0 (1R 3 \Z)!
This is due to the that in regions of configuration space where the potential is
only moderately singular (bounded or square-integrable), V(H) and V(Ho) coincide
[which is a local generalisation of Proposition 2.49(a) stating that the domains of
and of H 0 are identical if the potential belongs to L 2 (1R 3 )+L 00 (lR 3 )]. More precisely:
be a self-adjoint extension of the n1inimal operator. Let W be an open subset
3
lR \ Z on which the potential Vis only 1noderately singular, i.e. such that V Xw E
(JR 3 ) + L 00 (lR 3 ). Then, for each 0 E C0 (1R 3 ) having support in W, one has:
f E D(Ho) ===} O(Q)f E D(fl), and f E D(H) ===} O(Q)f E D(Ho).

As D(H) D(Ho) in general, the operator (Ho i)(H- i)- 1 will not be defined
on all vectors ofH but only on those f E H for which (H -i)- 1f E D(fl0 ) [remember
CHARACTERISATION OF THE SCATTERING STATES 211

that (H- i)- 1 maps 1{ onto the domain of H]. Thus the operator (Ho- i)(H- i)- 1
will not belong to B(H) in general, and the argument given in (5.35) to show that
XR(Q)(H- i)- 1 E JC(H) cannot be applied any more. However a generalisation of
this argument can be obtained by using the fact that V(H) and V(H0 ) coincide lo-
cally in regions where the potential is only moderately singular. We just consider two
particular situations, that in which the potential is moderately singular everywhere ex-
cept at infinity (for example a potential of the type of the harmonic oscillator) and that
where the potential is strongly singular only at the origin x == 0. In the first instance
we have Z == 0, in the second one Z == {0}.
(1) Suppose that V Lfoc(lR 3 ), i.e. that .{jxi<P!V(x)\ 2 d 3 x < oo for each p < oo; the
E
behaviour of the potential at infinity can be entirely arbitrary. For fixed > 0 choose
a function e E Cc)(lR 3 ) such that B(x) == 1 if \x\ < Rand B(x) == 0 for \x\ > + 1.
Since XR(x)B(x) == XR(x), one has XR( Q) == xn( Q)B( Q) and hence

XR(Q)(H- i)- 1 == Xn(Q)B(Q)(H- i)- 1


== XR(Q)(Ho- i)- 1 · (Ho- i)B(Q)(H- i) 1
(5.36)

Since B(Q)(H- i)- 1 maps H into V(H0 ), the operator (Ho- i)B(Q)(H- i)- 1 is
defined on each vector of H. This operator belongs to B(H) by virtue of the Closed
Graph Theorem (page 68). Thus one can deduce from (5.36) that XR( Q) (H - i) 1
is a compact operator for each R < oo and hence that the conclusions of Proposition
5.12 are true.
(2) Suppose now that V E Lfoc(lR 3 \{0} ), in other terms that j~<lxl<r1: \V(x) \2 d 3 x < oo
for all 0 < p < /-); < oo. Thus we admit that the potential could have bad behaviour at
the origin (for example V (x) - 1\ x\- 405 for some real! #- 0) and also at infinity. In
this case the argument of ( 1) does not apply any more. Here the range of the operator
B( Q)(H- i)- 1 belongs to V(Ho) if e E C 0 (1R 3 \ {0} ), i.e. if e vanishes not only in a
neighbourhood of infinity but also near the origin x == 0, whereas ( 1) the function
e had to be equal to 1 in the ball B R· However one can still obtain interesting results
by modifying in (5.36) the projection FR XR( Q). We replace in the preceding
definition of FR the function XR by the characteristic function of the domain O(R) :==
{x E lR 3 j R- 1 < \x\ < R}. Then one may take fore a function in C 0 (1R 3 \ {0}) which
is equal to 1 on this domain [and for example B(x) == 0 for \xi> R + 1 as well as for
lxl < 1/(2R)]. With FR == Xo(R)(Q) and eas indicated, one has as in (5.36):

FR(H- i)- 1 == FRe(Q)(H- i)- 1 == FR(Ho- i)- 1 · (Ho- i)B(Q)(H- i)- 1 ,


which implies that FR(H- i)- 1 E JC(H).
To interpret the result of (2), assume (in order to avoid having to consider temporal
means) that the continuous spectrum of is purely absolutely continuous (i.e. that
O"sc(H) == 0). Then one has \\FRUtfii ~ 0 as t ~ ±oo for each f E Hc(H) -
7-lac(H), see Proposition 5.7(b). In physical terms, remembering the definition of FR
in (2), a state in the continuous subspace of 1{ will disappear at large times fro1n each
annular region {x E 1R3 I E < \xi < R} withE> 0 and R < oo. For such states there
212 EVOLUTION GROUPS AND SCATTERING THEORY

are two possibilities: they may propagate to infinity, as do scattering states, or they
may move closer and closer to the singularity of the potential (here the origin). In the
second case the state will be absorbed at the singularity. It is therefore appropriate
to introduce, besides the bound states and the scattering states, the absorbed states
M;bs(H):
jEM;bs(H) ~ lim jj[I-xR(Q)]Utfii==O VR>O, (5.37)
t-7±CX)

in other terms

fEM~bs(H) ¢==? lim {


2
!(Utf)(X)! d 3 x=O \Is> 0.
t-7±CX) Jlxl ?_c

It can be shown again that M ;bs (H) are sub spaces and invariant under each Ut. Their
intersection M~bs (H) nM;;bs (H) is contained in M 0 (H): iff is a vector belonging
to M ~bs (H) n M ~bs (H), then it describes a state that is localised very close to the
origin It I > t 0 for a certain t 0 , and during a finite time interval (here the interval
[-t 0 , to]) each vector remains essentially confined to a finite part of configuration
space because of the continuity of the evolution [see the argument given in part (ii) of
the proof of Proposition 5.11(a)]. Thus the states in M~bs (H) n M~bs (H) are bound
states. such a situation, if M ~bs (H) n M ~bs (H) -1- { 0}, the set of bound states is
not equal to the subspace spanned by the eigenvectors of the Hamiltonian, there exist
vectors in the absolutely continuous subspace Hac (H) that represent bound states!
Under the assumptions that V E Lfoc (JR 3 \ { 0}) and a sc (H) == 0 one can show that 4
Hc(H) - Hac(H) == Mto(H) ffi M~bs(H) == M~(H) ffi M~bs(H). (5.38)
So, if example f E M~ (H) is a scattering state at t == -oo (f describes a situ-
ation in which a particle arrives on the scattering centre from a large distance), then
one can decompose f into a part J! describing the scattered part after the interaction
has taken place and a part f:!bs that will be absorbed as t ~ +oo [f == J! + f!s
with f~ E Mto(H) and f!s E M~bs(H)]. A potential for which this situation is
realised has been constructed by D. Pearson; in this example the potential is highly
x
oscillatory and unbounded above and below near the point == 0 (see e.g. Section
14.6 of [P] or Volume III of [RS], pp. 70-71). It can be shown that no absorption
occurs if the potential has only positive singularities, e.g. if V is bounded below in
some neighbourhood of the set Z containing its singularities (we refer to [ 1] which
also contains various other applications of Proposition 5.12).
Since an absorbed state is essentially localised (for sufficiently large t) in a very
small neighbourhood of the origin, it should have a very large kinetic energy at large
times (as a consequence of the uncertainty relations). Indeed it is easy to show that,
for each 0 < M < oo, one has limt-7±oo IIEo([O, M])Utfll == 0 iff E M;bs(H);
here {E 0 ( ·)} denotes the spectral measure of the free Hamiltonian H 0 == P2 . Thus,
for each M < oo, the probability that the kinetic energy of the state described by f
remains less than M tends to zero as t ~ ±oo iff E M;bs(H).
4
A 1nore detailed decotnposition of the Hilbert space 1i in terms of bound states, scattering states and
absorbed states for a class of short range potentials is described in the Asymptotic Decomposition Theorem
in Chapter 13 of [P].
ASYMPTOTIC CONDITION. WAVE OPERATORS 213

5.3 Asymptotic condition. Wave operators

5 . 3 . 1 . We consider again the situation in which H is the Hamiltonian of a simple non-


relativistic scattering system in H- L 2 (JRn), i.e. H- H 0 V(Q), where H 0 == P2
is the kinetic energy operator and the potential V is small at infinity in some sense [for
example v (x) --7 o as xl --7 oo, so that at large distances from the scattering centre
1

there is essentially no force on the system, the system will be essentially free]. Here
H 0 is the free Hamiltonian and His called the total Hamiltonian. Iff is a scattering
state of H, for example iff E M~(H), then for very large positive t the particle de-
scribed by f is localised far from the scattering centre, i.e. in a region where the poten-
tial has very little effect, it should therefore behave essentially like a free particle. A
mathematical theory of scattering phenomena must express this by means of a for-
mula. One speaks of the asymptotic condition. In physical terms this condition means
that the scattering states of the total Hamiltonian H become free as time tends to in-
finity. There are various possibilities for formulating this condition in mathematical
terms; we give two examples, setting Ut == e-iHt.
(1) One could require that, for each f E M~(H) and sufficiently many free observ-
ables A, the expectation values \Utf, AUt f) have limits as t - t +oo (and similarly
for negative times). By a free observable, one understands a self-adjoint operator that
is constant under the free evolution, i.e. such that eiHotAe-ii-Iot - A for all t (thus
an operator A that commutes with the free Hamiltonian); important examples are the
components Pj of the momentum operator P. The idea here is that during a scattering
experiment one measures observables of this type, and the measured values should
be independent of the moment at which the measurement is done (provided that it is
made sufficiently far from the scattering centre, i.e. at sufficiently large times). The
effect of the scattering will then be given by the relations between the values of these
observables at t == -oo and their values at t +oo. This type of theory (called al-
gebraic scattering theory, since one considers a collection of free observables having
the structure of a C*-algebra or of a von Neumann algebra) requires rather refined
mathematical techniques.
(2) More simply (but less generally) one can require that each scattering state of H be-
haves asymptotically like a free state, in other terms that, given a vector f E M~ (H)
for example, there should exist a vector f+ E MCX)(Ho) - H such that, as t - t +oo,
Utf approaches the vector e-iHotf+' i.e. II Utf - e-iHotf+II - t 0 as t - t +oo (and
similarly for negative times). Since II Utf- e-iHotf+II == II eiHot [Utf- e-iHotf+]II ==
lleiHotUtf - f +II, the preceding condition means that s-limt-++CX) eiHotUtf exists
(and is equal to f +)·Thus, by setting u? == e-iHot, one could formulate the asymp-
totic condition by requiring that, for each f E M~(H), the strong limits of u?*Utf
as t - t ±oo exist. Observe that one also has
IIUtf- e-iHotf+ll == IIUt[Utf- e-iHotf+JII II! u;e-iHotf+ll, (5.39)
hence f - s-limt-++CX) u;u?f+·
In what follows we study the second formulation of the asymptotic condition (it
covers potentials V that tend to zero at infinity more rapidly than lxl- 1 ; on the other
214 EVOLUTION GROUPS AND SCATTERING THEORY

hand this condition is not satisfied for the Coulomb potential). We first consider an
abstract setting: given two evolution groups in a Hilbert space H, we study properties
of limits of the type of those considered above (see §5.3.2). Then, in Section 5.4, we
apply the obtained results to the case of Schrodinger Hamiltonians and we discuss
scattering theory for such Hamiltonians.

5 . 3.2. Let us consider two self-adjoint operators A and B in a Hilbert space H and
denote by {Ut} and {Wt} the associated evolution groups: Ut == e-iAt, Wt- e-iBt.
The asymptotic condition can be formulated by requiring that the strong limits of
UtWt (or of TiVtUt, which is equivalent from the mathematical point of view) exist
as t ---* ±oo. Thus we shall here make the assumption that these limits exist on a cer-
tain set of vectors and establish their mathematical properties. By taking into account
Proposition 2.2 we can assume without loss of generality that the set of vectors for
which these lin1its exist form a subspace of H. This subspace may be different for
the limit t ---* +oo and for the limit t ---* -oo. In this subsection we formulate our
results only for the limit t ---* +oo; it is clear that analogous statements can be made
fort ---* -oo.
So let M be a subspace of H. We assume that M is invariant under the group
{ }, i.e. that Wtf _ e-iBtf EM iff EM, for all t E JR. We denote byE the
projection onto M (i.e. E - PM). We recall that the letter E is also used for the
projections of a spectral family; one should not confound these notations. The use of
the letter E for the projection onto M is justified by the fact that in many applications
E is in fact a spectral projection of the operator B. We denote the spectral measure of
the operator A by { EA ( ·)} and that of B by { EB ( ·)}.
The condition that M be invariant under Wt can be expressed in purely operatorial
form, it is equivalent to
VtEJR. (5.40)
Verification. (i) We first observe that f E M if and only if E f == f. Thus the condition
T¥1 f E M if f E M can be written in operatorial terms as EWt E == Wt E. If this is
satisfied, one finds by taking the adjoint that ETil_tE == EW_t· Upon replacing t by
(which is justified because we assume M to be invariant under all Wt) this leads
to EWtE == EWt. Consequently (by comparing with the identity EWtE == WtE
given before), one has ETiflt == TiVtE.
(ii) Assume that EWt == WtE for each t E JR. Then, iff EM: EWtf == MltE f -
f, which shows that Wtf EM. D
For the remainder of this section we n1ake the hypothesis that the strong limit of
UtWt as t---* +oo exists on M. We denote this limit by D(A,B;E). So
D(A,B;E) == s-lin1 UtWtE s-lirn eiAte-iBtE. (5.41)
t---+ +(X) t---+ +(X;

The operator ~l(A,B;E) is called a wave operator (at t == +oo). Later, it will be
denoted by D+(A,B;E), since there will also be a wave operator at t--oo.
We refer to §2.2.2 and §2.2.3 for the definition of the notions used in the following
proposition.
ASYMPTOTIC CONDITION. WAVE OPERATORS 215

D(A,B;E) is a partial ison~etry with initial set M satis-


5 . 13 . (a) D
fying DE == D and fl*fl == E. 1f E == I, then D is an isometry.
(b) The wave operator D intertwines A and B, i.e. one has the following relations:

Utfl == flWt (5.42)


(V)fl == DEB(V) (5.43)
cp(A)D == flcp(B) if cp: JR -7 CC is continuous and bounded. (5.44)

Moreover, iff E V(B), then Of E V(A) and flf ==DB f.

PROOF. (a) f j_ M one has Of== 0; iff EM, then IIDfll == lilTit-too IIUtWtfll ==
limt-too llfll == llfll [using Proposition l.l(a)]. By Proposition 2.9(e) [or Proposition
2.7(f) if E == I], Dis a partial isometry with initial set M (and an isometry E == I).
The fact that DE == fl is obvious view of the definition (5.41 ), and fl * f2 == 1s
contained in Proposition 2.9( e).
(b) One has

Ut D == Ut [ s-lim u; WsE] == s -liin UtU; VVsE == s -l~m u,;-t lVsE.


s-t+oo s-t+oo s-tToo

By setting T == s-tand using the hypothesis that lVt commutes with this leads to

Utfl == s-lim [J;W7 +tE == s-liin


T-t+oo T-t+oo
u; WtE == s-lim U;M17 EWt == flWt.
T-t+oo

This proves (5.42). Upon multiplying (5.42) by ieizt and then integrating with respect
to dt over [0, oo) or over ( -oo, OJ (if z E CC is such that SSz > 0 or SSz < 0), one
obtains by taking into account (5.7) and (5.8):

(A- z) 1
0 == fl(B- z)- 1 vz E CC with SSz #- 0. (5.45)

Consequently, for u, v E JR:

ju·v[(A-A-ic:) 1
-(A-A+ic)- 1 )DdA = D r[(B-A-ic)
Ju
1
-(B-A+ic) 1
)dA.

By setting u == a+ 6, v == b + 6 and then taking the li1nits c - 7 +0 and 6 - 7 +0, one


obtains (5.43) for V == (a, b] by taking into account Stone's Formula (4.81 ). By taking
the limit a - 7 -oo one finds that Etn == DEf: for each bE JR, which implies (5.43)
for the spectral measures. Eq. (5.44) is a consequence of (5.43) and the definition of
functions of an operator given in Section 4.2.
Finally, iff E V(B), then DWtf is strongly differentiable, its derivative at t == 0
is -iDBf. Now DWtf == Utrlf. Hence Utrlf is strongly differentiable; by Stone's
Theorem (Proposition 5.1), Of belongs to the domain of A and one has Aflf ==
~i(djdt)Utrlflt=O- i(djdt)DWtfit=O- i[-iDBf] == DBf. D
Remark 5 . 14. The argument used to obtain (5.43) from (5.42) (passage from the
groups to the resolvents and then from the resolvents to the spectral projections) can
216 EVOLUTION GROUPS AND SCATTERING THEORY

also be applied to obtain the following result (one has twice the same evolution group
and, in place of an intertwining operator 0, a bounded operator commuting with this
group):
Let { Ut} be an evolution group and { E ( ·)} the spectral family of its infinitesimal
generator. Let C be an operator in B(H) such that UtC CUt for each t E JR. Then
CE(V) == E(V)C for each V E AB.

Proposition 5 . 15 . (a) Let N == R(fl(A,B;E)) be the final set of 0 == O(A,B;E).


The projection F PN onto N is given by

F ==on*. (5.46)
The limit
s-lim WtUtF _s-lim eiBte-iAt F O(B,A;F) (5.47)
+ CXJ
t---+ t---+ + CXJ

exists, and one has


O(B,A;F)- [D(A,B;E)]*. (5.48)
(b) The subspace N is invariant under the group {Ut}:

'v'tEJR. (5.49)

(c) {f cp: JR. -7 CC is continuous and bounded, then

cp(A)F == Fcp(A). (5.50)

(d) One has the following relations:

M c Hac(B) :::::::? N c Hac(A), M c Hc(B) :::::::? N c 1-ic(A),


M c M~(B) :::::::? N c M~(A), M c M~(B) :::::::? N c M~(A).
PROOF. (a) We refer to Propositions 2.7(c) and 2.9(d) for the fact that the projection
onto R(O) is F == 00*. If g E R(O), there exists f EM such that g == Of. Then

which converges to zero as t - 7 +oo. So the limit s-limt---++CXJ WtUtg [hence the
operator O(B,A;F)] exists. Eq. (5.51) also shows that s-limt---++CXJ WtUtfl == E, so
that

fl(B,A;F) == s-lin1 WtUtF- s-lim WtUtflfl* ==EO*- (DE)*== 0*.


t---+ + CXJ t---+ + CXJ

(b) Upon taking the adjoint of (5.42) and replacing in itt by -t (one has for exam-
ple ut == u-t) one finds

'v'tEJR. (5.52)

By using first (5.42) and then (5.52) one finds that UtF == Utflfl* == OWtfl* ==
flfl*Ut == FUt.
ASYMPTOTIC CONDITION. WAVE OPERATORS 217

(c) The argument is similar to that in (b). By (5.44) we have O*~(A) == ~(B)O* if
~is continuous and bounded [take the adjoint of (5.44) and take into account (4.50)].
Then ~(A)F- ~(A)OO* == O~(B)O* - OO*~(A) == F~(A).
(d) Let gEN. There exists f EM such that g ==Of. Then, by using (5.43) in the
second step:

(g,Efg) == (OJ,EfOf) - (Of, OEf f) - (O*Oj,Ef f) == (f,Ef f).


Thus, f E 'Hac(B), i.e. if the measure determined by (f,Ef f) is absolutely con-
tinuous, then g E Hac (A), i.e. the measure determined by (g, Ef g) is also absolutely
continuous (because these two measures are identical). This gives the first implication
in (d). The second one is obtained in the same way. For the third one, assume that
f EM nM~(B). Observe that
IIFRUtgll == IIFR[Utg- Wtf] + FRWtfll < IIFR[Utg- Wtf]ii + liFRWtfll
< I!Utg- Wtfll + IIFRWtfll- llg- UtWtfll + IIFRWtf!l- (5.53)

Now llg UtWtfll converges to zero as t --7 +oo; since f E Mto(B), IIFR Wtfll also
converges to zero, and then the inequality (5.53) implies that IIFRUtgll also converges
to zero as t --7 +oo, i.e. that g E Mto (A). D

The existence of the limits defining the wave operators is sometimes easy to obtain
by using the result of the following proposition (see §5.4.3 for examples and Remark
5.22 for comments).
Proposition 5 . 16 (Cook Criterion) . Assume that there exists a subset D of' the sub-
space M having the following properties:
(1) Vis total in M,
(2) Wtf E V(A) n V(B) for each f E V and each t E JR,
(3) Jto I (A B)Wtfll dt < oofor each fED.
Then O(A,B;E) exists (where ==PM)·
PROOF. (i) Let fED. Then, by using the hypothesis (2) and the Leibniz rule:

~UtWtf= [~Ut]Wtf Ut[~Wtf]


== UtiAWtf + Ut( -iBWtf) == iUt(A B)liVtf·
By virtue of Proposition 1.5(c) one then has (assuming t > s > 1)

UtWtf- U;Wsf -1tdd U;Wr.fdT = iltu;(A B)Wr.fdT. (5.54)


s T s

Thus, by taking into account ( 1.28):

IIUtWtf- u;Ws.fll = ll.l·t u;(A B)Wr.fdTII

< lt IIU;(A- B)Wr.flidT = lt II(A- B)Wr.flldT.


218 EVOLUTION GROUPS AND SCATTERING THEORY

Combined with the hypothesis (3) this implies that {UtWt f} is strongly Cauchy as
t - t oo (the right-hand side of the preceding inequality is arbitrarily small provided
that sand tare sufficiently large).
(ii) We have shown that {UtWtf} is strongly Cauchy for all fin a subset of M
which, by the hypothesis (1), is total in M. As JIUtWtll == 1 for each t, we can
conclude that {UtWtf} converges for each f EM by applying Proposition 2.2. D

5.4 Simple scattering systems. Scattering operator

We discuss here the simplest situation in which, in physical terms, there is only elas-
tic scattering (for example the scattering of a particle by a potential converging to
zero at infinity, or equivalently the scattering between two particles described in their
centre-of-mass frame). The description of scattering systems with inelastic scattering
similar but mathematically more complex; in this case there are several scattering
channels (sometimes an infinite number of them) and consequently a family of wave
operators (for each sign of time), in fact, for each sign of time, one wave operator for
each channel. A simple example of a two-channel system is given by a particle on the
real line subject to an infinitely long potential barrier, i.e. to a potential V (x) which
tends to different li1nits as x - t -oo and as x - t +oo. Here reflection at the barrier
an elastic process whereas transmission is inelastic (non-conservation of the kinetic
energy). Another simple example is that of scattering of a deuteron by a fixed centre
of force. A deuteron can either be scattered elastically or it can be broken up into its
constituent parts (a proton and a neutron) through the interaction, so again there are
two scattering channels. More generally, in atomic collisions one can have creation of
various new systems or particles, for example a particles.

5 . 4 . 1. We consider again the situation described in §5.3.1: H == L 2 (1Rn), B == H 0 ==


and A == H == H 0 + V(Q), where V: 1R 11 - t JR. is a potential assumed to be
"small" at infinity. We write Ut == e-iHt for the evolution group associated with the
total Hamiltonian H. This group is often called the total evolution group. We use the
notation u? for thefree evolution group, i.e. for the evolution group determined by
the free Hamiltonian Ho: u? == e-iHot.
Let us first consider the physical situation before the interaction takes place, i.e. the
asymptotic condition at t == -oo. A free particle 5 is directed towards the scattering
centre. When the particle approaches the scattering centre the interaction will begin
to become effective. Thus initially (as t is very close to -oo) the evolution of
state of the particle will be described (at least to a very good approximation) by
free evolution group {uP}' but at later times by the total evolution group { Ut}. Let
us first describe the free initial state. It is determined at all times if it is given at one
ti1ne t 0 . The preceding description suggests taking to == oo, but this does not make
5 In practice it is not a single particle but rather a beam of particles. We describe here the scattering of
a single particle, and in Section 5.6 we treat the tnore realistic case of an initial beatn so as to arrive at
expressions for the scattering cross section.
SIMPLE SCATTERING SYSTEMS. SCATTERING OPERATOR 219

sense mathematically. One therefore adopts the convention that the state vectors are
specified at time t == 0. Thus, if a vector f _ E H is taken as an initial state 6 , it is
understood that the prepared state at large negative time t is upf _. So the particle will
be Cto a very good approximation) in the state uP
J_ if tis close to -oo, but in princi-
ple it will never be, in the course of its evolution, exactly in the state UPJ _ (because
its evolution is not governed by the group { u?} but rather by the group { Ut}). To de-
scribe the real evolution of the particle one must find a vector g which, under the total
evolution [lt, approaches u?f _ as t - 7 -oo. So g describes the real state (i.e. taking
into account the interaction with the scatterer) of the particle at time t == 0 under the
assumption that the initial state (taken at time t == 0 according to the above conven-
tion) is f _. mathematical terms, given f _, the vector g must satisfy the condition
limt--+-00 IIUtg- u?f-11 == 0.
We see that g is determined from f- by the relation g == fl_(H,Ho)f_, where
fl_(H,H0 ) == s-lin1t-+-oo Utu? is the wave operator at t == -oo associated with
the couple {H, H 0 }. We have set E ==I which, in the considered physical situation,
is reasonable because each vector in £ 2 (JR.n) represents in principle a possible initial
state. In other situations it would be natural to take for E the projection onto the
subspace M~(Ho) formed by the scattering states of the free Ha1niltonian H 0 at
t == -oo; in the present case, where H 0 == P2 , we know that M~(H0 ) ==H.
Let us consider the situation after the interaction has taken place (in a reasonable
case in which there is no absorption), i.e. for t - 7 +oo. The state leaving the scat-
tering region will be given by Utg, with t - 7 +oo. Again it will behave to a very
good approximation like a freely evolving state: there should exist a vector f + such
that Utg approaches the free evolution U?f+ off+ as t - 7 +oo. f+ is interpreted
as the final state at time t == 0 (given the initial state f _). Thus f+ must obey the
condition limt-++oo IIUtg- u?f+ll == 0, i.e. f+ == n+(Ho,H;E~(H))g. Here we
have inserted the projection E~ (H) onto the subspace Moo (H) in the definition of
the wave operator, because g belongs to M~(H) by Proposition 5.15(d) (for negative
tilnes).
In the preceding description of the scattering of a particle, the effect of the inter-
action is contained in the association f _ ~---t f +: with each possible initial state f _
(here with each vector f _ E 7-i) there is associated another vector f + interpreted as
the final state corresponding to f _, all states being given at time t == 0. In principle,
if the correspondence f _ f--7 f+ is known, one must be able to calculate all physical
observations that can be made after the scattering has taken place. The correspondence
f _ ~---t f + defines a linear operator called the scattering operator S (sometimes also
the S-matrix in the physics literature). Thus one writes f+ == Sf_, and the operator
S should determine the physical quantities measured after the scattering if the initial
conditions are known.
The preceding description 1nakes sense provided that the wave operators fl_ (H,H0 )
and D+(H0 ,H;E~(H)) exist. It is usual to apply a more symmetrical treatment to
the sign of time and to define
fl_ fl_(H,H0 ;I) ==s-lim u;up (5.55)
t--+-00

6 In the physics literature the state f _ is frequently called fin or 'l/Jin.


220 EVOLUTION GROUPS AND SCATTERING THEORY

and
(5.56)

Then, the above description, the relation between g and f + can be written as g -
O+f+ or f + - O+g [see Proposition 2.7(e)]. Thus the scattering operatorS is given
as follows: Sf_== f+ == O+g == O+O-f_. Hence

(5.57)

This reformulation makes sense provided that one knows that the vector g == [2_ f _
belongs to the final set of [2+ [i.e. that g E R(O+ )], otherwise the relation g - O+f+
has no sense. So one must know that R(O_) C R(O+ ). On the other hand, if this
condition is not satisfied (when one has to use the first formulation to clearly depict
the physical situation), one can still define an operatorS by the formula (5.57) because
n+ and [2_ are well defined operators in B (H) (in fact of norm 1), assuming of course
that the lilnits in (5.55) and (5.56) exist.
Figure 5.1 symbolically represents the situation. In this representation the points in
the plane correspond to vectors of the Hilbert space, a straight line to the free evolution
of a state vector and a curved line to the total evolution of a vector.

J_(t~o)
t---t-(X)

Figure 5.1

5 . 4 . 2. Before establishing the existence of the wave operators for a certain class of
potentials V and discussing the properties of the scattering operator S, we give some
useful results concerning the free evolution group. These explicit characterisations of
UP UP
are possible because the operator e-iHot assumes a simple form in lllOmen-

tum space, namely u? is the multiplication operator by e-ik t:


2

(5.58)
SIMPLE SCATTERING SYSTEMS. SCATTERING OPERATOR 221

Indeed, for f belonging to the dense subset C0 (JR.n) of infinitely differentiable func-
tions of compact support in momentum space [i.e. ](k) == 0 for k outside some
bounded subset of JR.n], e-iHotf is given as follows (see Examples 4.12 and 4.17):

[Fe-zHotf](k) = f (-~~)j
. 0 J.
[F(fi2)jf] (f)= f (-~~)j
. 0 J.
(f2)Jj(k)- e-iPtj(k) .
J= J=

(5.58) expresses the free evolution in the momentum representation. By an in-


verse Fourier transformation one can obtain a formula for in configuration space: uP
5 . 17 . In L 2 (JRn ), uP is an integral operator given by
(U?f) (X) - (4ni~)n/2 .l~n exp { il:f ~iN} .f(fl) dny. (5.59)

The square root in (5.59) is defined such that

(
4Jrit )-n/2 == { exp[ -inK I 4] if t > 0
I4Kitl exp[+inK I 4] if t < 0.

REMARK. The integral in (5.59) has a meaning iff E L 1 (JRn). So (U2f)(x) is given
by (5.59) iff E L 2 (1Rn) n L 1 (1Rn). For vectors f E L 2 (IR 11 ) not belonging to L 1 (1Rn),
this integral must be interpreted as a limit in the mean, as in §2.5 .2.
PROOF OF PROPOSITION 5.17. In order to simplify the notations we consider the
case n == 1.
(i) One first establishes (5.59) for particular functions f, namely for Gaussian func-
tions fa (x) == exp[- ( x a ) 2 ], a E JR., by using the following explicit expression for
Gaussian integrals: If p > 0, 1 E IR and~ E IR, then

_1_
~
joo eizf;e-i"(z2-pz2dz- J2(p + ir) exp[- --~-2-J
-(X) 4(p + ir) '
1 (5.60)

where the determination of the square root is such that R-J p + ir > 0. In particular
2
the Fourier transform of fa is ia(k) == 2- 112 e-ikae-k 14 , and

(UPfa)(x) == _1_
~
joo eixke-itk2fa(k)dk
-(X)
= _1_
V41T
JCX) ei(x-a)ke-itk2 e-k2/4dk
-(X)

2
1 [ (x-a) ]
- v1 + 4it exp 1 + 4it .
The same expression is obtained by using (5.60) to calculate the right-hand side of
(5.59) for f == fa:
2 2
1 JCX) {ilx Yl }f ( )d 1 [i(x-a) ]
(41Tit) 1 12 -(X) exp 4t a y y == (2it) 1 12 exp 4t

x-
~
1
-j 00

-(X)
exp[-i(y-a)(x
2t
a)Jexp[-(y-a) 2 (1 il4t)Jdy

2
1 (x-a) [ ]
== v1 + 4it exp - 1 + 4it .
EVOLUTION GROUPS AND SCATTERING THEORY

(ii) The set of functions N :== {fa}aEffi. is total in L 2 (JR). Indeed, suppose that g j_
N, i.e. that (fa, g) == 0 for each a E JR., so that J~00 eikae-k / 4 g(k)dk == 0 for each
2

a E JR. This means that the inverse Fourier transform of the function exp [- k 2 /4] g(k)
at the point a is zero, and this for each a E JR. Thus the inverse Fourier transform of
this function is the zero vector 0, and by the unitarity of the Fourier transformation
exp[-k 2 /4]g(k) == 0 almost everywhere, hence g(k) == 0 a.e., i.e. g == 0. By Propo-
sition 1.6 the set N is total in L 2 (JR).
(iii) f is an arbitrary vector in £ 2 (JR.) n £ 1 (JR.), it can be approximated by a se-
quence {g1 } 1 EN belonging to the dense linear manifold V spanned by N. (UP gj) (x)
is given by (5.59), and one obtains the validity of (5.59) for f by a limiting argument.
We omit the details because the result is not essential in what follows (it will suffice
to know (5.59) for a dense set of vectors f, for example for the set V used here). D

Proposition 5.18. Let H == £ 2 (JRn ). Fort =1- 0 let Zt be the following unitary operator:

or (Ztf) (x) = exp [ i:t J.f(x).


2
(5.61)

(5.62)

where cp( Q) and cp(2tf3) denote the n1ultiplication operators by cp( x) and cp(2tk) in
(JRn) and L2 (JRn) respectively.
PROOF. One has
II [Zt - Ilfll
2
= lJ i:t
exp [
2
J-
2
11 [.f( X) [2 dnx.
By applying the Dominated Convergence Theorem (Proposition 1.9) one finds that
this expression converges to zero as t - 7 ±oo. Next, since all operators in (5.62)
belong to B(H), it suffices to obtain this equation on a total set of vectors f. We take
f E N 77 , where Nn is the n-dimensional analogue of the set N introduced in the
preceding proof: Nn :== {fa}aEffi.n, where !a(x) == exp[-(x- a) 2 ]. So it suffices
show that (g,UP*(p(Q)UPJ) == (g,Ztcp(2tP)Ztf) for all j,g ENn. This is easily
obtained by using (5.59) and making the change of variables f---t k- z/2t: z

Clearly the last expression is equal to (Ztg, cp(2tP)Ztf). D


SIMPLE SCATTERING SYSTEMS. SCATTERING OPERATOR

Since Zt ~ I at large values of ltl, Eq. (5.62) shows that, under the
free evolution and in the Heisenberg picture, the operator rp( Q) is essentially the same
as rp(2tP) at large It!, hence Q(t) ~ 2tP. This is a quantum-mechanical analogue of
the fact that x(t) == Xo + ptjrn for a freely moving classical particle (note that, with
our choice H 0 == , one has 2rn == 1, hence rn == 1 for the n1ass).

Let us consider some applications of Eqs. (5.59) and (5.62).


Scattering states of H 0 . The formula (5.59) in1plies that

(5.63)

follows that, if e is a function in L 2 (~ 17')' then

(5.64)

Let us take fore the characteristic function XR of the ball BR [see (5.22)]. Then (5.64)
becomes
2
IIFRUP!II < (41Tftlt' IBRIII!IIi-
where IB R I denotes the volume of this ball. Thus, for f E L 2 (~ n) n L 1 (~ 17 ) ' the pro-
bability of presence in BR tends to zero like ltl-n as ltl ----+ oo. So each vector in
L 2 (~n) n£ 1 (~n) represents a scattering state of H 0 . Since L 2 (~n) n£ 1 (~n) is dense
in L 2 (~n), this gives a new proof of the fact that M~(Ho) ==H. 7
(B) Existence of the wave operators, I. Let == H 0 + V (Q) + V ( Q). Assume
that vis a (real) function belonging to L (~ ) and take e ==
2 17
v in (5.64):
(5.65)

n > 3, this function of the variable t is integrable at t == ±oo. One can then apply
the Cook Criterion (Proposition 5 .16) to obtain the existence of the wave operators
0±. Let us take == H, B == H 0 (hence Wt == U2) and V == Nn in the Cook
Criterion and consider the conditions ( 1) - (3) that have to be satisfied in this criterion:
(1) V Nn is total in L 2 (~n).
(2) Observe that Nn C S(~n ). f E S(~ 17 ), then UPJ E S(~n) for each t [this is
evident in the momentum representation, see (5.58)], hence U?f E V(Ho). One must
also know that U?J E V(H), which holds in particular V(H) == V(Ho). We know
from Proposition 2.49 that this latter condition is satisfied V E L 2 (~ 17 ) the case
n == 3 and V E LP(~n) for some p > n/2 if n > 3.
(3) We have seen that this integrability condition is satisfied V E L 2 (~ 17 ) and n > 3.
7 We have verified (5.59) only for f E Nn. By replacing £ 1 (JRn) in the preceding argument by Nn
one arrives at the smne conclusion, because Nn spans a dense linear manifold in £ 2 (JRn ).
224 EVOLUTION GROUPS AND SCATTERING THEORY

conclusion:
• lfn == 3 and V E £ 2 (:[~. 3 ), then the wave operators D± exist.
• lfn > 4 and V E L 2 (JR 3 ) n LP(JR 77 )for some p > n/2, then the wave operators
D± exist.
(C) Existence of the wave operators, II. The above proof of the existence of the wave
operators in dimension n > 3 involves the assumption that V E L 2 (JRn). Let us see
what this assumption implies for the asymptotic behaviour of the potential V (x). Sup-
pose that Iv (x) I - Ixl- j3 (/3 > 0) in a neighbourhood of infinity. The function <p( x)
lxl-/3 is square-integrable in a neighbourhood of infinity in JRn if JCX) r- 2 !3rn- 1 dr <
x, i.e. if {3 > n/2. This is not an optimal decay assumption for the existence of the
~ 0 ~

wave operators. Indeed, since IIIQI-JJUt ill == III2Ptl-/3 Ztill, which behaves approx-
imately like (21ti)-!3111PI-/3 ill at large times (see Remark 5.19), it can be expected
that IIVUPill is integrable near t == ±x if /3 > 1 but not so if /3 < 1. We verify below
that the wave operators do exist if /3 > 1 and more generally for potentials converging
to zero at infinity at least as fast as Ixl-/3 for some /3 > 1 (called short range poten-
tials). On the other hand, if /3 < 1 (in particular for the Coulomb potential), the wave
operators do not exist; one then speaks of long range potentials. For such potentials
it is not possible to approximate asymptotically the evolution e-iHti of a scattering
state i by a free evolution [in the sense indicated in (5.39)]. We refer to Section 5.8
for some details on time-dependent scattering theory for long range potentials.
So let us now consider the case /3 > 1. If we assume V to be bounded on ]Rn and
IV (x) I < cjxl-/3 in some neighbourhood of infinity, with /3 > 1, we have V E Lq (JRn)
for some q n - c with c > 0. If /3 is close to 1, c will be very small; in any case
we may assume that c < 1, so that n c > 2 if n > 3. We now prove the existence
of the wave operators if V E LCX) (JRn) n Lq (JRn) for some q E (2, n) (with n > 3), by
invoking the Cook Criterion. Let i E S(JRn). Then (I+ Q2 )ni E S(JRn). By taking
into account (5.62) one sees that

IIV(Q)UPill - IIUP*v(Q)UPill IIZtV(2tP)Ztill


IIV(2tP)(I + Q2 )-n Zt(I + Q2 )nill < I!V(2tP)(I + Q2 )-nllll (I+ Q2 )nill·

Thus it suffices to show that IIV(2tP)(I + Q2 )-nll is integrable with respect to dt


near t == ±x. This norm can be estimated by using (2.87):

Since njq > 1, the function t ~ IIV(2tP)(I + Q2 )-nll is integrable at t == ±x.


By combining the results of I and II we have obtained:
SIMPLE SCATTERING SYSTEMS. SCATTERING OPERATOR 225

Proposition 5 . 20 . Let - H 0 + V(Q) in L 2 (1R 17 ), with V real, and H 0 == P 2 .


(a) ff n == 3, let V == V1 + V2 with V1 E L 2(JR 3 ) and V2 E L CX) (JR 3 ) n (JR 3 ) for
SOJne c E (0, 1). Then the wave operators n± n± Ho;I) exist.
(b) n > 3, assume that v == vl + v2 with vl E L 2(1Rn) n £P(JR 17 ) for sonle
p > n/2 and v2 E LCX)(JRn) n Ln-E(JRn) for some c E (0, 1). Then the wave opera-
tors fl± = D±(H,Ho; I) exist.
. ( 1) In the preceding proposition, the part V1 of the potential characterises
the local behaviour of V, and the part V2 the behaviour of V in a neighbourhood of
infinity: may have local singularities of class Lq with q == 2 if n - 3 and q > if
n > 3, and it should be bounded and of class Ln-E in some neighbourhood of infinity
(0<s<1).
(2) The proof of the existence of the wave operators given above cannot be applied
dimension n == 1 and in dimension n == 2. In these cases different methods are nec-
essary. We present one such method in the following chapters and refer to Proposition
7.3 for results on wave operators in dimension 1 and 2.
(3) Let us mention an extension of the preceding considerations. The Cook method
requires an estimate of II vuP f II at large times. If f is any state vector, then upf
essentially localised in a neighbourhood of infinity if It I is very large (each f E 1{
is a scattering state of H 0 ). Therefore only the behaviour of V near infinity should
be important in the estimate (5.65). Indeed one can prove the existence of the wave
operators by imposing the conditions of Proposition 5.20 only in a neighbourhood of
infinity (V can be quite arbitrary in a finite part configuration space!). Let us ex-
plain this Theorem of Kupsch and Sandhas in the case n == 3. Let R be a po-
sitive number and assume that, for lxl > R, one may write V(x) == V1 (x) (x)
2 3 3 3
with V1 E L (1R \BR) and V2 E LCX)(1R \BR) n Lq(1R \BR) for some q E , 3). We
impose no condition on V in the ball BR [except that V(Ho) n V(V(Q)) must be
dense in L 2 (JR 3 )]. One must first give a meaning to the Hamiltonian We admit
that is an arbitrary self-adjoint extension of the symmetric operator H 0 + V
[with H 0 + V( Q) defined on V(Ho) n V(V (Q) )], and we shall show that under these
conditions the wave operators D± (I-I, Ho; I) exist.
Let us fix a function ¢ E C0 (JR 3) satisfying ¢( x) == 1 on B R and set 1./J == 1 - ¢.
Then
utuP == ut¢(Q)u? + Ut'l/J(Q)UP.
As ltl ---7 CXJ, the first term on the right-hand side converges strongly to zero [use
for example (5.64) with e == ¢ ]. Thus n± == s-lilnt-7±CX) Ut1/J( Q)UP. To obtain the
existence of these lilnits one proceeds in analogy with the Cook method. operator
1/J(Q) maps V(Ho) into V(H) (since V(H) and V(Ho) coincide locally in regions
of configuration space where the potential is only 1noderately singular, see page 21
observe that 1/J(x) -1- 0 only for lxl > R). By applying the Leibniz rule [Proposition
5.5(a)] one obtains for f E D(Ho):
226 EVOLUTION GROUPS AND SCATTERING THEORY

Now 1/J -1/J(Q)Ho == V?jJ(Q) + [Ho,1/J(Q)], hence

II! Ut1/J(Q)U2JII < IIV¢(Q)U?JII + II[Ho,7/J(Q)JU2JII·


Since v 1/J is a function satisfying the conditions of Proposition 5 .20, II v 1/J (Q) uP f II
is integrable at t - ±oo if f E S (~n) by the arguments given in the derivation of
Proposition 5.20. By using the Leibniz rule one finds that, forgE S(~ 3 ):

([Ho, 'tP( Q) ]g) (x) == ( ~, 1/J( Q) ]g) (x) == - ~{ 7/J( x)g(x)} + 1/J( x)~g( x)
== - { ~1/J(x) }g(x) - 2[V7/J(x)J . [V g(x)],

or in operatoriallanguage: [H0 ,1jJ(Q)] ==- (Q)-2i[V7/J](Q) · P. Here for exam-


ple [~w] (Q) denotes the multiplication operator by the function ~7/J. This function
belongs to C0 (~ 3 ) because 1/J(x) == 1 in some neighbourhood of infinity. Hence
II (Q)UPfll behaves like itl- 3 12 at infinity [using (5.64) withe - ~1/J] and is
therefore integrable with respect to dt at t == ±oo. Similarly each component of V1/J
belongs to Co(~ 3 ), so that II [V1/J] (Q) · PUPfli < I:~= 1 ll [V1/J]k (Q)UP Pkfll also be-
ltl-312 infinity.

operator and JS . . matrix

We again consider the simple situation of §5 .4.1. We recall the definition of the
scattering operator: S == n.+-~1-. Let us first indicate some important simple proper-
ties of this operator.

(a) The scattering operator commutes with the free Hamiltonian


More precisely one has

[S,UP] == o (5.67)

and, ~f f E V(Ho), then Sf E V(Ho) and SHof == HoSf.


S an isonzetric operator if and only if R(fl_) C R(O+ ).
(c) S a unitary operator ~f and only ~f R(fl_) == R(O+ ).

PROOF. (a) Eq. (5.67) is an immediate consequence of the intertwining relation (5.42)
and its adjoint (5.52): in the situation considered here (Wt UP) one has

suP== n.+n_up == r2.+-utn- == u?n+n- ==ups.


Next let f E V(Ho). By (5.67) one has then it- 1 S[UP - I]f == it- 1 [U2 - I]Sf.
As t ---0 0 the left-hand side converges strongly to SH0 f by Stone's Theorem. Conse-
quently the right-hand side is strongly convergent as t ---0 0, so that (again by Stone's
Theoretn) Sf belongs to V(Ho) and one has SHof- HoSf.
(b) We denote by F + the projection onto the subspace R( 0+) and by F _ that onto
SCATTERING OPERATOR AND S-MATRIX 227

R(O_). We recall that F± == 0±0± [see (5.46)]. The hypothesis R(O_) _ R(O+)
implies that F+O- == 0_. Under this hypothesis one will have

s*s == o:_o+o-f-o_ == o:_p+o- - o:_o_ == 1,


i.e. S will be isometric. If the inclusion R( 0_) R( 0+) is not satisfied, there is a
vector g E R(O_) such thatg tf:. R(O+) R(F+)· Iff== o:_g (so thatg == O_f),
then IIF+gll < llgll == 110-fll - 11!11 (because 0_ is isometric) and consequently
IISJII == IIO-f-0-fll == 110-f-gll == 110+0-f-gjj IIF+gll < 11!11 (using the isometry of
0+ for the third equality). In this caseS cannot be isometric [see (2.37)].
(c) S is unitary if and only if S and S* are isometric. By (b) S is isometric and
only R(O_) c R(O+)· Since S* o:_o+ one obtains [by exchanging in (b) the
roles of the signs+ and-] that S* is isometric if and only if R(O+) C R(O_ ). D

In relation with parts (b) and (c) of the preceding proposition, let us briefly dis-
cuss the notion of asymptotic completeness. It follows from Proposition 5.15( d) that
R(O±) c Hac(H). One says that the wave operator 0+ (or 0_) is complete if
R(O+) == Hac (H) [respectively if R(O_) - Hac (H)], and that one has asymptotic
completeness if R( 0+) R( 0_) == Hac (H), i.e. if both 0+ and 0_ are complete.
This terminology is justified by the fact that, under the assumption of asymptotic com-
pleteness, one has a description of the large time behaviour of all scattering states of
H; more precisely, if one assumes that the set of scattering states M (H) coincides
(X)

with Hac (H), and if g is an arbitrary vector in Hac (H), there exist vectors f + and f _
in H such that Utg behaves asymptotically (i.e. as t ----+ ±oo) as U?f± (if - fl± g,
then II Utg - U?f ±II ----+ 0 as t ----+ ±oo ).
One can consider other notions of asymptotic completeness, for example:
Strong asymptotic completeness: 0+ and 0_ are complete and has no singular
continuous spectrum, i.e.

Asymptotic completeness in the geometric sense: we have seen that one always has
R(O±) C M~ (H) [Proposition 5.15(d)]. One speaks of asymptotic completeness in
the geometric sense if R(fl+) == Mto (H) and R(O_) - MCX) (H) [without assum-
ing a relation between M~(H) and Hac(H)].
Generalised asymptotic completeness: this notion has been used in the particular con-
text where absorption of states by local singularities of a potential may occur, as ex-
plained at the end of Section 5.2. In such a situation one would say that generalised
asymptotic completeness holds if R(O+) == Mto(H), R(O_) == M~(H) and
(5.38) is satisfied. As explained after (5.38), one again has a complete description
of the large time behaviour of states in He (H) (decomposition into an absorbed part
and a part propagating to infinity and becoming free).
Asymptotic completeness (and a fortiori strong asymptotic completeness) implies
the unitarity of the scattering operator S. In various physical theories the unitarity
of S is assumed essentially in an axiomatic way. Observe that the isometry of S im-
plies the "conservation of probability": if f _ is an initial state and S is isometric, then
228 EVOLUTION GROUPS AND SCATTERING THEORY

f + == Sf_ has the same nor1n as f _, i.e. I f + I == I f -II; in the simple situation con-
sidered here, if the S -operator is isometric, then the scattering does not lead to a loss
of probability. Evidently, in cases where the potential is absorptive, a part of the inci-
dent state could be absorbed at its singularities and the norm of the out-coming state
f + could be less than I f -II; nevertheless one should not speak of non-conservation
of probability because the evolution is unitary and one knows [for example if (5.38)
is satisfied] what happens with the part of the state that is not leaving the scattering
region f _ is the initial state, then the associated scattering state of H, i.e. the vector
[2_ f, has a unique decomposition into a part that will be absorbed and a part that will
be scattered as t ---+ +oo ); so it is only for an experimentalist (situated far from the
scattering region) that an apparent loss of probability occurs.

Remark 5 . 22 . We have seen that in the case in which H 0 is the operator P2 in di-
mension n > 3 and the interaction is given by a potential, the existence of the wave
operators 0± - fl± (H, H 0 ; I) is easy to prove by using the Cook Criterion. To obtain
asymptotic completeness one has to show that n± (H 0 , Eac (H)) exist 8 . For this
the Cook Criterion not useful because one would have to know for example that
J1ociiV(Q)Utflldt < CXJ for sufficiently many vectors f; but the total evolution group
{ U,} cannot be written explicitly (as was the case for the free evolution group), hence
one has no simple estimate of the asymptotic behaviour of I V (Q) Ut f I as a function
of parameter t. prove asymptotic completeness one has to have recourse to more
elaborate mathematical methods. One such method, involving the theory of relatively
smooth operators, is presented in Section 7 .1.

Since R( ~l±) C Hac (H) it is clear that, in all cases in which the spec-
trum of the Hamiltonian is not purely absolutely continuous, the wave operators
are not unitary (they are only isometric). In particular, if H has eigenvalues, then n+
and [2_ are not unitary, the eigenvectors being orthogonal to R(O+) and to R(O_ ).

Let us now discuss part (a) of Proposition 5.21, i.e. the fact that S commutes with
the free Hamiltonian H 0 . In physical terms this roughly expresses the conservation
of the kinetic energy by the scattering process. Indeed Eq. (5.67) implies that S com-
mutes with the spectral projections Eo (V) of the free Hamiltonian H 0 (as explained in
Remark 5.14), in particular one has E 0 (J)S == SE0 (J) for each interval J. Assume
that the initial state f _ satisfies Eo ( J) f _ == f _ for some interval J, in other terms
the kinetic energy of this state is localised with certainty in J. Then, however
small the interval J, the outgoing state f + - Sf_ has the same property, i.e. one has
Eo(J)f+ == f+; indeed

Eo(tl)f+ == Eo(J)SJ_ == SEo(J)f- ==Sf_==

The following two observations are interesting in this context:


( 1) S is not isometric, the amount of kinetic energy is not conserved for an observer

For exatnple if g E R(D+), so that g = D+f for SOlne f E H, then II!- u?*utgll - 7 0 as
8

t +oo, i.e. one has f = rl+(Ho,H;Eac(H))g [see also Proposition 5.15(a)]. Thus, in order to have
-7

R(D+) = Hac(H), one must know the existence ofs-limt---++oo u?*ut on Hac(H).
SCATTERING OPERATOR AND S-MATRIX 229

far from the scattering centre. The amount of kinetic energy of the initial state f _ is
given by IIHof-II == {f0 A?(j_,Eo (dA)j _) }1 / 2 . The associated amount kinetic
CX)

energy of out-corning state f+ ==Sf_ will be IIHof+ll- JIHoSf-11- IJSHof-11


and this number can be less that JJHof-II if S is not isometric (this is by no means
surprising: for example if a part of the scattering state remains absorbed in the poten-
tial, then part will retain a certain amount of the total energy).
(2) The relation Eo ( J) S == S Eo ( J) implies that the subspace M (J) R( Eo (J))
gets mapped into itself by the operator S. Thus the restriction of S to this subspace
defines a bounded operator in M (J), and it is natural to denote this operator by S (tJ).
If for example S is unitary (as an operator in H), then S ( J) is unitary as an operator
in M (J). One may take for J an arbitrarily small interval, and one may be tempted
to consider the limit in which J consists of a single point A. One would then have an
operator S (A) (unitary if S is unitary) acting in the space M ( {A}). However, since
the spectrum of H 0 is purely continuous, the functions having support in a single point
A with respect to the spectral measure of H 0 are equivalent to the zero vector 0 (one
has Eo ( {A}) == 0), so the notation M ( {A}) does not correspond to a subspace of H.
Nevertheless, since the operator H 0 is very explicitly known, one can give a meaning
to this notation by introducing the spectral representation of H 0 . This means that one
chooses a representation of the Hilbert space H == L 2 (~n) in which H 0 is a multipli-
cation operator, in other terms one "diagonalises" H 0 by mapping H by means of a
unitary operator U 0 onto the space L 2 ((0,oo);L 2 (sn- 1 ),dA) 9 in such a way that,
after transformation by ~~ 0 , the operator H 0 becomes multiplication by the variable A
in L 2 ( ( 0, CXJ); L 2 ( S n- 1 ), dA) [the interval (0, CXJ) appearing here corresponds to the
spectrum of H 0 , with the exception of the point A - 0 which we omit because spher-
ical polar co-ordinates are singular at this point, and of course {A == 0} is a null set].
Let us first give an explicit formula for U 0 . The construction is similar to that in-
dicated at the beginning of §3.3.3, but it is carried through in the momentum repre-
sentation i 2 (~n) of the Hilbert space. One denotes by sn- 1 the unit sphere in ~n,
i.e. the set of vectors w E ~n satisfying JwJ == 1, and one writes the wave vectors
k E ~n in spherical polar co-ordinates: k == vf.\w (so A == k2 ). L 2 (sn- 1 ) is the set
of (equivalence classes of) complex functions defined on sn- 1 that are measurable
and square-integrable with respect to Lebesgue measure dw on sn-l [for example for
n == 3 with respect to dw ==sin BdBd¢ as in (3.40)]. For f E L 2 (~n) we set

(5.68)
For each f, Eq. (5.68) defines for almost every A > 0 an element of L 2 (sn- 1 ) (as a
function of w). By varying A, one associates by Eq. (5.68) with each f E L 2 (~n) an
element of L 2 ( ( 0, CXJ); L 2 ( S n- 1 ), dA) 10 • The operator U 0 is isometric as a mapping
9 See (1.53) and below for the notations. Further comments on spectral representations can be found in
the footnote on page 152.
10
In order to be in agreetnent with the notations of Chapter 1, we should write (Uof)(:A)(w) rather than
(Uof)>..(w): one considers a vector-valued function [with values in L 2 (S 77 1 )l defined on (0, oo), and in
Chapter 1 we used the notation (Uof)(:A) for the value of such a function at the point A E (0, oo ). Since
(Uof)(:A) belongs to L 2(sn-l ), it is a function of the angular variable w, and it is 1nore convenient to
write (Uof)>.. rather than (Uof)(:A).
230 EVOLUTION GROUPS AND SCATTERING THEORY

from L 2 (1Rn) to L 2 ((0, oo);L 2 (sn- 1 ), dA) [see (1.53); observe that if one sets k ==
jkj, then dA == 2kdk and consequently dnk == kn- 1 dkdw == (1/2)A(n- 2 )1 2 dAdw]:

It is also clear that U 0 is surjective, hence a unitary operator. Its inverse is given as
follows: if g =={g.\} E L 2 ((0, oo);L 2 (sn-l ), dA), then 11

with w== k/lkl. (5.69)

Iff E V(Ho), then (UoHof).\(w)- A(Uof).\(w), because in momentum space Ho


is the operator of multiplication by k2 . More generally, if <p: JRn - t Cis a bounded
continuous function, then 12

[Uorp(Ho)f].\(w) == rp(A)(Uof).\(w) (5.70)

Since S commutes with H 0 , it cannot mix the different values of A in the spectral
representation of H 0 . This is expressed in the following manner: for each A > 0 there
exists a bounded operator S (A) acting in L 2 ( S n-l) such that the component at A of
the vector Sf [i.e. the element (U 0 Sf).\ of L 2 (sn- 1 )] is obtained by acting with
S (A) on the component at A of the vector f:

(UoSf).\ == S(A)(Uof).\· (5.71)

Equation (5.71) is (for each A > 0) an equation in L 2 (sn-l ). If S is isometric or


unitary as an operator in L 2 (1Rn), then each S(A) is an isometric or unitary operator,
respectively [in L 2 (sn-l )]. A priori S(A) is defined only for almost all A > 0 (with
respect to Lebesgue measure). In most situations one can show that one may choose
S (A) to be continuous as a function of A, and with such a choice S (A) is defined
for all A > 0 (physically this seems plausible, since physical quantities measured in a
scattering experiment should depend continuously on the kinetic energy). The operator
S (A) is frequently called the S-matrix at energy A. The space L 2 ( S n-I) associated
by (5.68) with a fixed value of A is sometimes called an energy shell.
Above we have given a description of the mathematical context in relation with
concept of the S-matrix. We still have to verify these assertions, i.e. to prove the
existence of S (A) and its properties. This will be done in §5 .5 .2 in an abstract setting
(the situation considered above is a particular case of that treated in §5.5.2). Before this
we indicate briefly how the scattering phase shifts, introduced in quantum mechanics
courses, arise in the present formalism (in the case n == 3).
11
We denote by 9>.. E £ 2 (Sn 1 ) the value of gat the point A.
12
(5.70) will hold for almost all A E (0, CX)) and almost all wE sn- 1 .
SCATTERING OPERATOR AND S-MATRIX 231

Below we use the letter R to denote rotations in IR 3 around axes passing through
the origin x == 0. The set of all such rotations forms a group, the rotation group. A
unitary representation of this group in L 2 (IR 3 ) given by the following formula:
[ (n) f J( x) == J (n - 1 x), (5.72)
1
where is the inverse of the rotation R. This formula associates with each rotation
a operator U(R) in L 2 (IR 3 ), and the collection of these unitary operators
gives a strongly continuous representation of the rotation group. is easy to calcu-
the Fourier transform of (5.72) to obtain the action of U(R) in the momentum
representation, i.e. on the Fourier transforms of the wave functions: [FU (R) f] (k) -
j (R - 1 k). Thus U (R) acts in £ 2 (IR 3 ) exactly in the same way as in L 2 (IR 3 ). is clear
each U(R) commutes with the free Hamiltonian Ho, because

[FU(R)Hof](k) == [FHof](R- 1 k) == IR- 1 kl 2 f(R- 1 k)


== jkj 2 f(R- 1 k)- [FHoU(R)f](k)
(the length of a vector of:IR 3 remains unchanged under a rotation: IR- 1 kl == lkl). If
H == H 0 V( Q) and if the potential Vis spherically symmetric (i.e. if (x) depends
only on T -ixi), then (R) also commutes with (Q) and hence with 13
. In such
a case (R) commutes with u? - e-iHot as well as with == e- Ht, hence also
1

with f2+ and with fl_. The relation U(R)O+ - O+U(R) implies that ~2'+U(R)* ==
U(R)*O'+; since (R)* == U(R- 1 ), this shows that ~1+ commutes with U(R- 1 ) for
each rotation R, hence with each U(R) (when R varies over the rotation group, n- 1
also varies over this group, each rotation being the inverse of another rotation). Thus
each U(R) commutes with ~1- and with ~1+ and hence with S n+n_.
us now consider the subgroup of rotations around one of the three Cartesian co-
ordinate axes. This subgroup is a one-parameter group (the parameter being the angle
of rotation), and the representation of this subgroup (given by the operators U (R)
for R belonging to the subgroup) is a strongly continuous unitary one-parameter
group (an evolution group in our terminology). Its infinitesimal generator is the an-
gular momentum operator relative to the considered co-ordinate axis. By choosing
successively the co-ordinate axis of x 1 , then that of x 2 and finally that of x 3 , one
obtains the three operators L 1 , and as infinitesimal generators these three
subgroups. Since the scattering operator S commutes with the representation of each
of these subgroups, it also commutes with their infinitesitnal generators, i.e. with
L 1 , L 2 and L 3 , and consequently also with L2 :== Li L~ L~. Thus S leaves
invariant the subspace associated with an eigenvalue of any one of these operators
(the subspace spanned by all eigenvectors associated with an eigenvalue of one of
these operators). In particular S leaves invariant the subspace Hern formed the
common eigenvectors of L with eigenvalue f( £ + 1) and of L 3 with eigenvalue m
2
-+2
(£ == 0, 1, 2, ... ; m- £,£- 1, ... , -£),because L and L 3 commute.
13 One assumes here that V is only moderately singular, so that the operator II o + V (Q) is self-adjoint
on D(Ho) n D(V(Q)). In more general situations H could be a self-adjoint extension of Ho + V(Q)
which does not commute with U(R) (as indicated on page 126), and then the m·gu1nent presented below
does not apply.
232 EVOLUTION GROUPS AND SCATTERING THEORY

Now in the representation (5.68) of L 2 (JR 3 ), i.e. in L 2 ((0, oo);L 2 (S 2 ), dA.), the
common eigenvectors of i 2 with eigenvalue £(£ + 1) and of L 3 with eigenvalue m
have the form h (A.) Yen (w), where h is a complex function defined on (0, oo) satisfying
J0 h (A.) 2 dA. < CXJ (the dependence on the angular variables is fixed, of the form
CX) I 1

Y£ (w), the dependence on the radial variable is free; the situation is essentially the
same as that discussed-+ in §3.3.3, but here the decomposition into partial waves is
written in the variable k rather than in the variable x, as discussed in Appendix A.3.3).
For fixed A. [i.e. in L 2 (S 2 )], the subspace formed by the common eigenvectors of i 2
with eigenvalue £(£ + 1) and of L 3 with eigenvalue m, is one-dimensional, given by
the scalar multiples Y£. The operatorS (A.) (for fixed A.) maps this one-dimensional
subspace into itself, so S(A.) simply multiplies Y£ by a number bem; this number may
depend on A.:

Sis unitary (which is the case for moderately singular potentials), then so is S (A.). In
this situation the absolute value of bem(A.) must be 1: lbbn(A.)I == 1. One may always
write bbn (A.) - exp[2i6£(A.)J, which defines 6£(A.) (mod n). Sis unitary, 6£(A.)
1nust be real. We shall see below that 6£n(A.) does not depend on m: 6£(A.) == 6R(A.).
Thus, in each partial wave (of angular momentum £), the operator S is given by a
function 6R(A.) of the kinetic energy A.. This function 6R is real if S is unitary, and
it is just the scattering phase shift at angular momentum £ introduced in quantum
mechanics texts for scattering by a spherically symmetric potential.
Finally let us verify that 6;n 1 (A.) == 6;n 2 (A.) for each m 1 , m 2 - £,£ 1, ... , -£.
For one uses the fact that S also commutes with L 1 and L 2 , hence with ·-
'and that for example Y£ == C£my;n+l for some constant C£m (C£7n #- 0
rn #- f). One then has form #- £:

Consider a Hilbert space H == L 2 (1R; JC, m) as in (1.53), where m is a a-finite


measure on (JR, AB). We use the notation x for the variable in JR, and we denote by X
the operator of 1nultiplication by the variable x in H (defined on the maximal domain
for a multiplication operator): (Xf)(x) == xf(x) iff == {f(x)} is an element of
(JR; JC, m) [thus f(x) E JC for each x E JR] 14 .

14
In the application considered in §5.5.1, one has JC = L 2(sn-l) and X= UoHoU 01, and m
is identically zero on (-oo,O] and coincides with the Lebesgue measure on (O,oo): m(dx) = dx on
(0, 00 ).
SCATTERING OPERATOR AND S-MATRIX 233

An operator A in H is said to be decomposable if it has the following form: for


each x E JR there exists an operator A (x) in !3 (JC) such that 15

D(A) ~ {f E 1t I i IIA(x)f(x) ll'fcm(dx) < oo} and (Af)(x) ~ A(x)f(x).


(5.73)
A is a decomposable operator we write A - {A (x)}, and we denote by II A (x) II JC
the norm of the operator A( x). The following properties of decomposable operators
are easy verify by mimicking the argurr1ents given for multiplication operators in
Section 2.5:

Proposition 5 . 24 . (a) Let A be a decomposable operator. Then A is bounded, with


'D(A) - H, if and only if the function 1/J(x) :== IIA(x)IIJC belongs to L 00 (1R, m), and
one has II A II == ll?t011 oo·
(b) If A== {A(x)} and B == {B(x)} are decomposable and belong to B(H), and if
a E CC, then A + aB, AB and A* are also decomposable and given by

A+ aB == {A(x) + aB(x)}, (5.74)


AB- {A(x)B(x)}, (5.75)
A*== {A(x)*}. (5.76)

Corollary 5 . 25 . Let A be a decomposable operator. A is unitary if and only (f A(x)


is unitary m-almost everywhere. A is isometric if and only if A(x) is isometric m-a.e.

A particular class of decomposable operators are the diagonalisable operators


which have the form A == { cp(x )I}, where cp is a complex measurable function de-
fined on JR and I denotes the identity operator in JC. In this case A is the multiplication
operator by cp(x) in L 2 (JR; JC, m,), which is just the operator cp(X). If cp == Xv is the
characteristic function of a Borel set V, then cp(X) == xv(X) is the spectral projection
E(V) of X associated with V. It is clear that each decomposable operator commutes
with all diagonalisable operators, in particular with X. the next proposition we
2
prove the converse: a bounded operator in L (JR; JC, rn) that commutes with X is de-
composable:

Proposition 5 . 26 . Let H == L 2 (JR; JC, m). If A E B(H) is such that [A, e-zXt] == 0
for each t E JR, then A is decomposable.

PROOF. The hypothesis [A, e-iXt] == 0 for all t E JR implies that AE(V) - E(V)A
for each Borel set V (see Remark 5.14), hence E(V)A* == A*E(V) for all V E AB.
(i) We fix an orthonormal basis { ek} of J( and a function e: JR ---7 JR such that
f~ool8(x)l 2 m(dx) < oo and B(x) -1- 0 for each x (for example B(:r) == l:rl if rn is
the Lebesgue measure). We set ek(x) == B(x)ek. For each fixed x, ek(x) is a non-zero
J_ J_
vector in JC. Furthermore, since 00 II ek(x) life m( dx) == 00 IB(:r;) 12 11 ck 11Jc7n( dx) -
00 00

f~ool8(x)l 2 m(dx) < oo, the family {ek(x)}xEX (for fixed k) defines an element
of the Hilbert space H. We write ek for this element of H, i.e. ek == {ek (x)}. We
15 To be precise, the fan1ily {A(x)} must also be measurable, i.e. for all f, g E K the function x 1---7

(f, A(x)g) must be Borel measurable.


EVOLUTION GROUPS AND SCATTERING THEORY

denote Do ~ JC the set of finite linear combinations with rational coefficients of


the vectors { ek} and by D C H the set of finite linear combinations, also with rational
coefficients, of the vectors { ek}. We observe that if f == {f (;r)}, then f E D if and
f(x) == B(x)fo some foE Do (modulo equivalence).
each x E JR we define an operator B (x) in JC, with domain Do, by
B(x)e~ = B(~) [A *ek](x) (5.77)

(here [A* e~,:] (x) denotes the component at x of the vector A* ek of H, i.e. A* e k ==
{[A*ek](x)}). Eq. (5.77) specifies the operator B(x) on each vector ef, and its ac-
tion on the vectors in Do is defined from (5.77) by linearity. (5.77) we have cho-
sen for each k a representative in the equivalence class of the vector-valued function
(;r); a different choice would change the definition of B (x) at most on a count-
(over k E N) of null sets r k, hence on a set of measure zero. Thus B (x) is
defined 771~-almost everywhere.
Let f - {B(x)fo} ED. By using the relations B(x)ek(x) == [A*ek](x) and
(V) ==A *E(V) for V E AB, one finds that

II (x)f(x)llkm(d:r) = .fv II[A*f](x)llkm(dx) = IIE(V)A*fll 2

<II
2 2
II IIE(V)fll =II l1
2
fv llf(x)llkm(dx).

fv [II ll 2 llf(x)llk- IIB(x)f(:r)llk]m(dx) > 0


implies that IIB(x)f(x)IIJC < IIAIIIIJ(x)IIJC m-almost everywhere. Conse-
quently, each f E D there exists a Borel set f(j) of measure zero such that
II (x)f(x)IIJC < II llllf(x)IIJC for each x ~ f(j). Since Dis a countable set, the
union r of all r (f), as f varies over D, is a Borel set of measure zero, and for each
x~ (x) is a bounded operator (on Do). Thus the closure of B (x) belongs to !3 (JC)
(see page 48), the adjoint of the closure of B (x) is equal the adjoint of (x)
2.20). Thus, we set A(x) == B(x)* for x ~ r, we have A(x) E B(JC)
and IIA(x)ll < I I each x ~f). By finally defining A(x) == 0 if x E r we have
constructed a fa1nily {A(x;)}xE~ of operators B(JC) satisfying IJA(x)ll < IIAII for
each :r E JR. This fa1nily {A( x)} defines a decomposable operator belonging to !3 (H)
[Proposition 5.24(a)]. We denote this operator by A' (so A' == {A(x)} ), and it suffices
to show that == A'.
Let us fix a number k E N. Let h == {h( x)} E H. By using the hypothesis that
(V) for each Borel set V, one sees that

.fv (A(x )h(x ), ek (x) h:: m(dx) = fv (h(x), B(x )ek (x)) K rn( dx)

= fv (h(x). [A*ek](:r))Krn(dx) = (h, (V)A*ek)

([
= (h, A*E (V) ek) = (Ah, E (V) ek) = j~ Ah]( :r:) , ek(x)) K rn (dx) .
SCATTERING CROSS SECTIONS 235

This implies that (A(x)h(x),ek(x))JC == ([Ah](x),ek(x))JC m-a.e. 16 Since (x) ==


B(x)ek: and B(x) -1- 0, it follows that (A(x)h(x),e%)JC == ([Ah](x),ek)JC m-a.e.,
i.e. for all x E JR not belonging to a certain Borel set ~k (h) of measure zero. If one
denotes by ~ (h) the union of all sets ~ k(h), then ~ (h) is a Borel set of measure zero,
and one has (A(x)h(x)~ e%)JC == ([Ah](x), e%)JC for all x tf_ ~(h) and for all k. Since
the family { ek} is total in JC, it follows (Proposition 1.6) that [Ah] (x) == A( x )h( x) ==
[A'h](x) for x tf_ ~(h). Thus one has Ah- A'h for each hE H, i.e. A- A'. D

5.6 Scattering cross sections

We begin with a mathematical model of a scattering experiment in which a beam


of independent particles is scattered by a fixed scattering centre. We shall study the
temporal behaviour of a single particle of this beam and then deduce (by using the
scattering operator describing the scattering of this particle by the scattering centre) a
formula for the probability that this particle will be observed, after the scattering has
taken place (t ---7 +oo ), in a cone C with vertex at the co-ordinate origin. By adding
the probabilities for all particles of the beam, one obtains the scattering cross section
for the cone C (here one uses the hypothesis that the particles of the beam are scattered
independently, i.e. one after another, because one adds the probabilities rather than the
probability amplitudes- no super-positions of the states of the different particles). Let
us recall the usual definition of the scattering cross section for a cone C:

C _ number of particles scattered into C per unit time


a( ) number of particles incident per unit time and unit surface area ·

where the surface area is determined in the plane orthogonal to the direction of prop-
agation of the beam.
We assume that 1{ == L 2 (JRn) with n > 2 and that the free evolution is described by
the group U2
which was discussed in §5 .4.2. We fix a direction w0 (a vector in JRn of
length 1) which will give (approxin1ately) the direction of the velocity of the parti-
cles of the incident beam, and we denote by IT 0 the hyperplane of dimension r1 - 1
orthogonal to w0 and containing the origin. The beam will be described by an en-
semble of vectors {fz;}z;En in H, obtained from a fixed vector f (with 11!11 == 1) by
-+ 0
translating f by b in the hyperplane IT 0 [more precisely by acting with the operator
T(b) on f, where {T( ·)} is the usual unitary representation of the translation group
of JRn in the Hilbert space L 2 (JRn)], and f is such that the support of Fourier trans-
form j is contained in a (small) neighbourhood of vx;-
w0 • Here .-\ 0 is a fixed positive
16
If~ is a function in L 1 (1R,m) satisfying fv~(x)m(dx) = 0 for each Borel set V, then ~(x) = 0
m-a.e. This is clear if one knows that ~(x) 2:: 0 m-abnost everywhere. If~ is real, one is reduced to
the preceding situation by setting W + = {x I ~ ( x) 2:: 0} and W _ = {x ~ x) < 0}; by observing
J (

that fv~(x)xw+(x)m(dx) = fvnw+ ~(x)m(dx) = 0, one obtains ~(x)xw+(x) = 0 m-a.e.


Similarly one finds that ~ ( x) x w _ (x) = 0 rn-a. e. If ~ is complex, one applies the preceding argument
to the real part and to the itnaginary part of ~ (observe that the relation Jv 1/J (x) m ( dx) = 0 implies that
fvat[~(x)]m(dx) = atfv~(x)m(dx) = 0, and similarly for the imaginary part).
236 EVOLUTION GROUPS AND SCATTERING THEORY

number, corresponding to the kinetic energy of a particle of the beam. Thus, to a good
approximation, the momentum of each particle of the beam is k0 _ v:x;
w0 .
So ~et f be a vector in L (1Rn) such that llfll == 1. We assume that its Fourier trans-
2

form f vanishes outside a compact subset (bounded, and also small) of the half-space
{k E IR 77 k · wo > 0}, i.e. that ~suppj C {k E IRn [ k · wo > 0}, where suppj stands
j

for the support of the function f. We denote by Cin the cone in IRn (with vertex at the
origin) that is spanned by supp j, i.e. Cin == { k E IRn! c:f E supp j for some a-> 0},
and we assume that woE Cin· Finally we assume that f E L 00 (1Rn). For each bE ITo
we define the vector fr; by

fr;(x) == f(x- b) or (5.78)

ITo

suppf
(neighbourhood of ko - ~wo)

Figure 5.2

If Vis a Borel set in IRn and hE L 2 (1Rn), with llhll == 1, we denote by P(h;V) the
probability that the scattering state [2_ h associated with the initial state h be localised
in the region V of configuration space at time t == +oo, i.e. (assuming that the limit
exists):
P(h;V) = lim
t---++oo} V
r
I(Ut~Lh)(X)I 2 dnx. (5.79)

one adds the probabilities P(fr;; V) of all states in the ensemble {.fr;}, one obtains
a quantity representing the number n(f; V) of scattered particles localised asymptoti-
cally in V:
n(f;V) = { P(f-;;;V)dn- 1 b. (5.80)
Jn 0
SCATTERING CROSS SECTIONS 237

We are mostly interested in the case in which V is a truncated cone Cp, defined as
CP == {x E C II xi > p}, where C is a fixed cone with vertex at the origin and p is a
positive number (the observation of the scattered particles will be made at distances
greater than p).
should be mentioned here that in most texts on quantum mechanics, the scattering
cross section is introduced in terms of probability currents across distant surfaces. The
current density associated with a (differentiable) state vector g E £ 2 (IRn) is given as
v
; (x) == 2 [g (x) g (x) J, and the flux 1? o (g) across a hypersurface o is then

<I>o(g) = 2 'S L g(X) fi(X) · Vg(X)dJ(X), (5.81)

where n(x) E IRn is a unit vector orthogonal to 0 at the point x E 0 and do-(x) denotes
the surface element at X. Let 0 == sd n C, where sd == {X E IRn II == d}, and let
n(x) be the outward unit vector at the point x E 0. Then the integrated flux (counted
positively in the outward direction) associated with Utfl_h, after some initial timeT,
across this hypersurface is given by

is of considerable interest to know that, for very large d (and any finite T and p ),
this quantity is the same as the probability P ( h; Cp) that the scattering state associated
with h is localised in the truncated cone C P at t == +CX). More precisely one has, under
suitable conditions:

(5.82)

A proof of Eq. (5.82) is presented Section 7.2. The quantity on its right-hand side
seems closer to an experimental situation, where one counts the particles passing
(at some random time) through a detector located far from the scattering centre. Of
course, in order to interpret the expression (5.82) as a scattering cross section, one
would have to know that at large positive times the probability current through sdn c
is essentially outgoing. This point will be discussed at the end of Section 7 .2. Here we
work with the expression (5. 79) for P ( h; CP) in terms of the probability of presence in
the truncated cone Cp.
The scattering cross section o-(f; C) for scattering into the cone C, for a beam
defined in terms of a vector f as described above, is then simply o-(f; C)- n(f; Cp),
because the number of incident particles per unit surface in IT 0 is equal to 1 [it is given
by J~ dn-lb, where :E a subset of ITo of measure the measure being the (n- 1)-
dimensional Lebesgue measure] 17 . We shall see that, under reasonable assu1nptions,
17 Until now we have considered the scattering of a beam of particles by a single scattering centre. Of-
ten the target consists of a large number N of scattering centres (N of the order of 10 20 ). If the spatial
distribution of these scattering centres is rand01n (and if multiple scattering is neglected), then the scatter-
ing cross section for the cone C will simply be N n(f; Cp), because the probabilities for scattering into C
238 EVOLUTION GROUPS AND SCATTERING THEORY

the quantity n(f; Cp) is independent of p. We begin with a fundamental result, the
Scattering-Into-Cones Formula; this formula expresses a fact that is intuitively ev-
ident: if his an initial state, then the probability P(h;Cp) of finding the out-coming
particle in the truncated cone CP is just the probability that the momentum of the cor-
responding final state Sh lies in the cone C, i.e. the probability that the out-coming
particle propagates in a direction in which it will eventually penetrate into the cone C.

Proposition 5.27. Assume that the wave operators 0± == O±(H,H0 ) exist and that
R(fl_) C R(O+)· Then one has for each hE L 2 (JR.n) and for each p > 0:

(5.83)

PROOF. V is a Borel set in JR.n, we denote by Fv the projection onto the subspace
of vectors with support in V, i.e. Fv == Xv( Q), where Xv is the characteristic function
of V. As before, F + is the projection onto the subspace R( 0+).
(i) We first show that, under the assumptions of the proposition, one has for each
Borel set V (again assuming that the limit exists):

(5.84)

Indeed: P(h;V) == lirrlt-++oo IIFvUtfl_hll 2 , and it suffices to see that

(5.85)

Now [using (1.15)]:


2 0 2
I11FvUtfl_hll - IIFvUt Shll 1

- [IIFvUtfl-hil IIFvU2Shll] ·IIIFvUtfl_hll- IIFvU2Shlll


< 2JJhiiiiFvUtfl_h- FvU?Shll < 2llhii11Utfl_h- U?Shjj. (5.86)

As t ---7 +oo one has

Under the assumption that R(fl_) R(O+) one has F +n- h == n_ h. Thus (5.84) is
proved.
(ii) We now show that P(h;Cp) does not depend on p, more precisely that one has
P(h;Cp) == P(h;C) for each p > 0. For this we decompose the cone C into two
disjoint parts C ==CpU [C\Cp], which implies that

P(h;C) == P(h;Cp) + P(h;C\Cp),


fron1 the different scattering centres are added. On the other hand if the spatial distribution of the scattering
centres presents smne regularity (for example if the target is a crystal), the theory discussed here cannot be
directly applied (one must superpose the wave functions scattered by the different centres rather than add
the probabilities).
SCATTERING CROSS SECTIONS 239

and it suffices show that P(h;C \ Cp) == 0. Now (5.84) implies that (we use the
notations of page 205)

2
(h;C\Cp) = lim
t---7+(X)
IIFc\C P
u2Shll 2 - lim
t---7+(X)
/
JC\CP I(U2Sh)(X)I

< lim /
t -----++(X) JB
I(Ut~Sh)(X)I 2
d"x = lim
t---7+(X)
llxp(Q) Shll 2 •
p

and is zero because M(X)(H0 ) ==H.


us finally calculate P(h;C) limt---7+(X) IIFcU?Shll 2 . We denote by Xc
the characteristic function of the cone C [so that Fe == x c( Q)], we use (5.62):

IIFcU?Shll ==II *FcU?Shll == IIZtxc(2tP)ZtShll == llxc(2tP)ZtShll


== llxc(2tP)[Zt- Sh + Xc(2tP)Shll·

Now llxc(2tP)[Zt- I]Shll < II[Zt- Shll ~ 0 as t ~ +oo (Proposition 5.18).


2 2
Hence (h;C) == limt---7+oc IIFcU?Shll == lin1t---7+(X) llxc(2tP)Shll . t > 0 one
has Xc(2tk) == Xc(k) for each k E IRn, because fort > 0: 2tk E C only
k E C. Thus, writing the last norm in the momentum representation £ 2 (IRn):
2
(h;C) = lim
t---7+(X)
llxc(2tft)Shll 2 = lim
t---7+(X)
llxc(P)Shll 2 =
Jc I(Sh)(k)l
/ k.

is useful to introduce an operator R as follows

R==S- (5.87)

Clearly R is deco1nposable in the spectral representation

(UoRh)A == R(A)(Uoh)A, with R(A) == S(A)- IJC, (5.88)

where IJC denotes the identity operator the Hilbert space JC == L 2(sn-l ). ~ f is the
state vector defining incident beam (as described on page 236, with supp f Cin\
and if the cone Cis disjoint from Cin, one can express the formula (J; Cp) given
(5.83) ter1ns of the operator R (assuming the assumptions of Proposition 5.27 to
be satisfied):

if C n Cin == 0. (5

Indeed, by writingS== R +I and setting h == fin (5.83), one has

and the last two terms are zero because .1 == 0 for k E C if supp .1 c Cin
0.
240 EVOLUTION GROUPS AND SCATTERING THEORY

To obtain an expression for (]"(!;C) having a simple interpretation, one must sup-
pose that the R-matrix R(A) is an integral operator in L 2(sn-l) (i.e. an integral
erator with respect to the angular variables in momentum space, A being fixed). We
know a criterion for this to be the case: it suffices to know that R(A) is a Hilbert-
Schmidt operator (see Proposition 2.15). Further on we specify conditions implying
that R(A) E B2 (L 2 (sn- 1 )). We denote by IIR(A)IIHs the Hilbert-Schmidt norm [in
L 2 (srz--l )] of the operator R(A) and by r(A; w, w') its kerneL Thus one has

IIR(A)II~s = fs,_ dW lsn-tdw'jr(A;W,W'W.


1
(5.90)

Proposition 5. 28. Assume that R(A) E B2 (L 2 (sn- 1)) for each A> 0. With the no-
tations introduced above and under the hypotheses made (in particular C n Cin == 0 ),
the scattering cross section (J"(f; C) for scattering into the cone C, for a beam defined
in ter1ns o.fa vector f (11!11 == 1), is given by the following expression:

(5.91)
PROOF. (i) We know from the remarks preceding Proposition 5.27 that

Now the support of the Fourier transform of fr; is identical with that off [see (5.78)],
hence contained in Since C n Cin == 0, we can use (5.89) to find an expression
for P(fr;;Cp)· By using the spectral representation of H 0 , introduced in §5.5.1, one
obtains (see also Figure 5.2):

We interchange the order of integration over dn-lb with those over dA and dw. To
arrive at (5.91) it then is enough to verify that
2
d 71 - 1 b I
112
r(A;W,W')c-i>. w'.b(U 0 .f);..(W')dw'l
-ln clll

== [ 2 ;,_Jn-lj 2
!r'(A;W,W')j j(Uo.fh(W')j
2~ 1
~, dw'. (5.92)
v A. S 11 1 wo · w
SCATTERING CROSS SECTIONS 241

(ii) To prove (5.92), let us set cp-A,w(w 1 ) - r(A; w, w 1 )(Uof)-A(w 1 )Xcmnsn 1(w 1),
where Xcmnsn-1 denotes the characteristic function of Cin n sn-l. By virtue of the
hypotheses made [R(A) E !3 2 (L 2 (sn-l)), j E L 00 (1Rn)], the function cp.A,w(·) be-
longs to L 2(S n -l) for each A and almost all w. Its support is contained in the hemi-
sphere {w 1 I w0 · w 1 > 0}. This allows us to parametrise the points w 1 in the support of
cp.A,w by their orthogonal projection onto the hyperplane II 0 ; in other terms we write
z z
w 1 -- + (w0 · w 1 ) w0 with E II 0 , and we make the change of variables w1 f----+ We z.
thenhavedw 1 == [wo·w 1 ]- 1 dn-lzandw 1 ·b- z·bforbEI1 0 . Ifwesetb 1 == ~b
and use the unitarity of the Fourier transformation in L 2 (JRn-l ), we find that

Upon inserting the definition of cp.A,w (w 1 ), one obtains (5.92). D

DISCUSSION OF EQUATION (5.91). Let us recall this equation:

(5.93)
We have assumed that the support of f is contained in a small neighbourhood of a
vector k0 == ~ w0 . Thus the integral with respect to dw 1 in (5.93) extends only over
a small neighbourhood of w0 , i.e. one has cos(w 0 · w 1 ) ~ 1 for each w 1 belonging to
this set. Moreover, if (for the vectors w in C), r( A; 1
w, w
) is a continuous function of

w and of A as w varies over Cin n sn-l and ~varies over a neighbourhood of


1 1

lko I (more precisely as A varies over the support of the vector f with respect to the
spectral family of H 0 ), we may replace in Eq. (5.93) the quantity jr(A; w, w1 ) 2 by 1

Ir (Ao; w, wo) 2 , where Ao - ko 2 . Thus, if the support of j is sufficiently small, one


1 I 1

will have to a very good approximation 18

(5.94)

18
A rigourous justification can be obtained for example by considering a sequence {f1 } of initial states
such that I(Uof1 )-\(w')l 2 is a Dirac sequence, i.e. such that I(Uof7 )-\(w')l 2 -----7 6(.\- .\o)8(w'- wo)
as j -----t CX>, see page 287 of [AJS].
EVOLUTION GROUPS AND SCATTERING

side ofEq. (5.94) can be interpreted as the scattering cross section


o-( Ao ~ w0 ; C), for an initial beam of particles momentum k0 == ~ w0 , for scatter-
into a cone C satisfying C n Cin == 0 for some open cone Cin with wo E Cin· One
sees that a- (A. 0 , w0 ; ·) defines a measure on S n - 1\ { w0 } which is absolutely continuous
with respect to Lebesgue measure dw. The Radon-Nikodym derivative of this measure
is called the differential scattering cross section do- I dw:

da ( /\O,
-d ' wo;
-+
w ==
-+ ) [ 21f
r'\
] n- 1/r (/\O;
' w,-+
w-+0 ) /2 , (5.95)
,w yAo
and the quantity
• -+ -+ [ -21fi] (n 1)/2 -+ -+
.f (Ao; wo ---+ w) := JX;;- r(Ao; w, wo) (5.96)

1nay be interpreted as the scattering amplitude for scattering at energy A.o from the ini-
tial direction 0 intow w
direction (the factor -i has been inserted in order to obtain
usual result for scattering amplitude calculated by considering the Schrodinger
equation n == 3 din1ensions; we recall that, by definition, the differential scattering
cross section is the square of the absolute value of the scattering amplitude). One sees
the of the operator R (A) == S (A) - I JC coincides, apart from a
with the scattering amplitude!
quantity is the scattering cross section O"tot (A.o, w0 ), de-
the differential scattering cross section (do-/ dw) (A. 0 , w0 ; w)
(w -1- wo):

da (A.o,wo;w)dw. (5.97)
1 dw

is also useful introduce the averaged total scattering cross section a-( A. 0 ) obtained
by averaging quantity (5.97) over all initial directions w0 . If we denote by 8 n the
surface area of the sn- 1 [8n == 41T if n == 3], then (we write A. place of A. 0 ):
1
o-(A) = r_; j Gtot(A., wo)dwo == / dw j dwo dda (A., wo; w)
'CJn sn-1 8n sn l sn-1 w

[ v2;Jn-
1

IIR(A)II~s·
1
==
1
8n
[
2
;,_]n-
vA

s?l l
dwj
sn - 1
dwojT(A.;w,wo)J
2
==
1
8n A
(5.98)

example for n == 3: o-(A) == [1r I A.] IIR(A.) ll~s·


One notices that the total scattering cross section O"tot (A., w0 ) at energy A is finite for
initial directions w0 when R(A.) E B2(L 2(sn-L)). This question will be fur-
discussed Section 5. 7. If the potential V is spherically symmetric, O"tot (A., wo)
independent of the initial direction w0 , and then one has o-(A.) == O"tot(A., w0 ). is
instructive to verify that in this case (with n == 3) one obtains the usual expression for
the total scattering cross section in terms of the scattering phase shifts. We recall (see
§5 .1) that S(A.)Y}17 == e 2 z 6 R(-\)y£~. Hence
R(A.)Y£17 == [e 216t(A) -l]Yf == 2iei 6£(-\)sin[6£(A.)]Yf
BOUNDS ON SCATTERING CROSS SECTIONS 243

and ~ e
O"tot(A,Wo) = a(A) = : IIR(A)II~s = : L L IIR(A)Y£11
2

£=0 rn=-£
~ £ ~

= ~ L L 2
sin [Je(A)] = ~ L (2Ji + 1) sin [Jg(A)].
2

£=0 m=-£ £=0

5.7 Bounds scattering cross sections

5.7.1. We have seen above that the total scattering cross section at energy A is finite
when R(A) is a Hilbert-Schmidt operator in £ 2(sn-l ). We describe below the type
of potentials for which this condition is satisfied, but we begin with some general
remarks.
One may be surprised that the scattering cross section could be infinite. In fact this
possibility is inherent in the definition of the cross section in which one considers the
idealisation of an infinitely extended beam (in the directions orthogonal to its direction
of propagation, i.e. in the hyperplane II 0 ). This idealisation is equivalent to consid-
ering an incident plane wave (which is infinitely extended) in the usual description
of scattering by means of the stationary Schrodinger equation. Let us recall the situ-
ation in classical mechanics (in n == 3 dimensions). If the potential is spherically
symmetric, then the total scattering cross section at each energy A is equal to 1r R 2 ,
where R denotes the range of the potential [i.e. the smallest positive nu1nber such that
V (r) == 0 for r > R]. In this case the total scattering cross section is identical with
the (two-dimensional) area of the obstacle encountered by the particles of the incident
beam [if a classical particle is incident with an impact parameter bsatisfying Ibl > R,
it will not penetrate into the region where (r) -1- 0, hence it will not be scattered].
This explains the terminology 'scattering cross section'. If R == oo, i.e. if the potential
is non-zero in each neighbourhood of infinity, then the total scattering cross section is
infinite.
In quantum 1nechanics the situation is somewhat different because of the uncer-
tainty relations. If (r) is not of finite range but converges to zero sufficiently rapidly
as r ---+ oo, then the particles with large impact parameter will be only very weakly
scattered (in some sense they are subject to very small forces), and the total scattering
cross section is finite. On the other hand, if V (r) converges to zero very slowly, the
particles with large impact parameter will be 1nore strongly scattered, which leads to
an infinite total scattering cross section (see further on for details). We shall also see
at the end of this section that the finiteness or infiniteness of the total scattering cross
section depends only on the behaviour of the potential near infinity (the potential may
be essentially arbitrary in a finite part of configuration space).
The fact that the scattering cross section is determined by R( A) rather than directly
by S (A) is interesting. If the interaction is zero (V == 0), then there is no scattering
and S- I. Thus the effect of scattering is described by R == S- I (one has R == 0
in the absence of interaction). In the formalism of the time-independent Schrodinger
equation, this corresponds to splitting the eigenfunction of the Hamiltonian (the out-
244 EVOLUTION GROUPS AND SCATTERING THEORY

going wave function in the stationary formalism) into two parts, the incoming plane
wave (the analogue of the operator I) and the scattered wave (the analogue of the op-
erator R).
To discuss the scattering cross section at a fixed energy A, one thus has to study
the operator R(A). Now R(A) acts in the Hilbert space L 2 (sn- 1 ) rather than in the
Hilbert space £ 2 (IRn) in which the theory is initially defined. It is possible to prove
for1nulas expressing R(A) in terms of the potential, of the resolvent (H- A- iE)- 1
for E - 7 0 [these operators act in £ 2 (IRn)] and of certain operators from £ 2 (lRn) to
L 2 (sn- 1 ) which we call transition operators. These formulas allow one to obtain cer-
tain mathematical properties of R(A) [for example continuity or differentiability in
A, Hilbert-Schmidt property, behaviour of its kernel r(A; w, w') as a function of the
angular variables] by using properties of the transition operators and the behaviour
of the resolvent close to the real axis. An expression for R( A) of the indicated type
will be derived in Section 7 .4. In the present section we present a relatively simple
method for estimating total scattering cross sections in terms of only operators act-
ing in £ 2 (IRn). More precisely, we consider averages of the scattering cross section
over an interval of energies and obtain bounds for these averages directly in terms of
the potential. Let rp: (0, oo) - 7 [0, oo) be a non-negative weight function, typically a
smooth function vanishing outside some finite interval. We consider integrals of the
J
00
type 0 rp(A)o-(A)dA, and we shall obtain upper bounds for such integrals in the form
of a product of a certain norm of the weight function rp and of a certain norm of the
potential In particular, if each of these two norms is finite, one will have a-( A) < oo
for almost all A in the support of rp, hence also R( A) E B 2 ( £ 2 ( S n- 1 )) for almost all
of these A.

5.7.2. We consider the spectral representation £ 2 ((0, oo);L 2 (sn- 1 ), dA) of the free
Hamiltonian H 0 given by means of the mapping U 0 defined in (5.68). If TJ: (0, oo) - 7
1R a continuous function, we denote by P( TJ) the following operator in £ 2 (lRn)
(defined by its action in the spectral representation of H 0 ):

(5.99)

The integral in (5.99) is an integral of a vector-valued function in L 2 (sn- 1 ); one may


assume [Uof]p, to be continuous, because it suffices to define the bounded operator
P( TJ) on a dense subset of the underlying Hilbert space.
Lemma 5.29. Assume that J0 ITJ(A)j 2 dA == 1. Then
00

(a) P(TJ) is a projection, more precisely the projection onto the subspace

(b) If A E B(L 2 (1Rn)) is a decomposable operator in the spectral representation of


Ho, i.e. such that [U 0 Af]A == A(A)(U 0 f)A with A( A) E B(L 2 (Sn- 1 )), then
19
In spherical polar co-ordinates, the Fourier transforms of the elements f of M (r;) have the form
](k) = V2ifiC 2 -n)/ 2 r;(k 2 )g(w), see (5.69).
BOUNDS ON SCATTERING CROSS SECTIONS 245

(5.101)

where the first Hilbert-Schmidt norm is in L 2 (1Rn) and the second one in L 2 (sn-l ).
REMARK. The importance of this result can be seen by taking for A the operator R
S- Indeed: IIR(J\)II~s is proportional to the total scattering cross section o-(A) at
energy A averaged over the angular variables, and in this case Eq. (5.101 ), co1nbined
with (5.98), allows one to obtain bounds on the scattering cross section by estimating
an operator in L 2 (lRn ), i.e. without having to consider directly the S-matrix (without
having to consider restrictions to energy shells). Of course these bounds are not point-
wise bounds, i.e. they do not refer to a- (A) at a fixed energy A, but rather some sort of
average of a- (A) over an interval of energy (depending on the choice of r7):

II RP( 'l]) II ~s (2;r) 1 -nenJ; 00

A(n- 1 ) / 2 lry( A Wa( A) dA, (5.1 02)

where 8n is the surface area of the unit sphere sn-l in lRn.

PROOF OF LEMMA 5.29. We set lC == L 2 (sn-l ).


(a) (i) Let us first check that P(rJ) E B(H). By using the continuous triangle in-
equality (1.28) and then the Schwarz inequality in L 2 (IR), we obtain:

'lJ(M) (U of)~ df.t II~ 'lJ(M) (U of) f1 df.t II~


00 00

II P( 'l]) f 11
2
= 1oodA ry( A)
2
111 = lifo
< [ fa 1 M) Ill (uof)~ Ik dM
00

'l] ( r < [ fa
00

1'l] ( MWdM] [ ;; II (u o
00

.n ~ II kdM]
= fa 11 (Uo.f)~ Ilk df.t = llfll
00
2
·

Hence IIP(rJ)II < 1.


(ii) Let us show that P( rJ ) 2 == P( rJ):

[u oP( ry ) 2 fL = ry( A) [ 1oo ry(f.t) [U oP( ry) f] ~ dJL]

~ ry( A) [fa fa of)~, J


00 00

df.t 'l] (M) · ry(p.) df.t' 'll(!t') (U

= ry( A) [fa oo df.t' ry(f.t') (U of)~~] ~ [UoP( 'l7) fL,


where we have used the relation J0 00
I rJ(J-L) 2 dJ-L == 1. Let us also see that P( rJ) * ==
1

P(rJ):
(g, P( 'l7 )f) = (U og, UoP( 'l7 )f) = laoodA( (Uog )A, (UoP( 'l7 )f)>..) K

=fa fa of)~) K.
00 00

dA ry( A) df.t ry(f.t) ((U og) >.., (U

After interchanging the order of integration and by using the assumption that rJ is real,
one finds that the last expression is equal to (P (rJ) g, f).
246 EVOLUTION GROUPS AND SCATTERING THEORY

(iii) Clearly, in the spectral representation of H 0 , the vectors in the range of P(r7)
have the following form: (U 0 j)).. == r;(A.)g for some g E L 2(sn-l ). Consequently
R(P(17)) _ M(17). On the other hand, iff E M(r;), then (5.99) implies that

[UoP (77) fL. = 77( A) fa=77(/L) [Uof] fL dtt 77( A) fa =77(/L )[ry(fL) g] dfL
= faoc177(tL)I 2 dfL · ry(A)g = ry(A)g = (Uof);_,
which shows that M(r;) C R(P(r;)).
(b) One has IIAP(r;)ll~s == Lk IIAekll 2 , where {ek} is an arbitrary orthonormal
basis of M(rJ). The vectors of such a basis are necessarily of the form (Uoek)).. ==
r;(A.)e/:, with { ek} an orthonormal basis of JC L 2(sn-l ). Hence

IIAP(rl)ll~s- L
00

k
IIAekll 2
= L
k
IIUoAekll 2
= L Jork
11(UoAek)>-llkd>.

= L rooiiA(A) . ry(A)ek Ilk d).. = roodAiry( A) 1


2
L IIA(A)ek Ilk
k Jo Jo k

= fa oodAI77( A) I IIA(>.) ll~s'


2

which proves Eq. (5.101). D


By taking into account Eq. (5.1 02), we now have to estimate the Hilbert-Schmidt
nonn of the operator RP(r;). For this we observe that

n+)- -n+ roo !!:_utu?dt


J
L: L:
-00 dt

- -n+ ut(iH iHo)UPdt =-in+ utvuPdt (5.103)

[this equation holds on V(H0 ) if one assu1nes that V(H) == V(H0 )]. Consequently

u?
is seen that one can estimate II RP (r;) II HS in terms of II V P (TJ) II HS, and, as in the
proof of the existence of the wave operators, this last expression involves only the free
evolution group, which is known rather explicitly.
It turns out that, the operator - H 0 is a potential, i.e. if V == V( Q), then
one can explicitly calculate integrals of the type occurring in the last expression of
. 104) if one replaces the integrand by its square. In the next proposition we denote
by II · II the norm in £ 2 (lRn) (if the argument is a function defined on JRn) as well
as the norm in £ 2 ( ( 0, oo)) [if the argument is a function defined on (0, oo)]. r7
is a function defined on (0, oo) and a E JR, we denote by rJcx the function given as
7Ja (A.) == )\ex/ 4 r;( )\) and by r;~ its derivative. So we have for example:
BOUNDS ON SCATTERING CROSS SECTIONS 247

1177~ 112 = fa 00 I~~ [X' I 4T/ ( ,\)] I 2

2 2
111Qiwll 2 = f
}IJKn
IXI Iw(X)I dnx.

Let W == W (Q) be the n1ultiplication operator in L 2 (JRn) by a


(x). Let r]: (0, x) -7 1R be continuously differentiable, 'IJ(.A) == 0 in
a of 0 and in a neighbourhood of x, and llrlll == 1. E (1R 17 ),
one
c:x::l

IIWU2P(7J)II~sdt = ~(21f)J~n8ni\7Jn~211 2 ll 11 2. (5.105)


c:x::l

/f(l +I·\)~¥(·) belongs to L 2 (JRn), then

£:t 2 IIWU2P(7J)II~s dt = ~(21f) 1 ~n8ni\7J~~211 2 IIWII 2


+ S~t (21f ) 1 ~nen llrln~4II 2 IIIQI 2
11 . (5.1 06)

REMARK. It is interesting to observe that the tenns on the right-hand sides of


equations .1 05) and (5.1 06) have the form of a product of a factor depending only on
rJ and a factor involving only W; apart from these factors there are different constants
depending on rL
PROOF OF PROPOSITION 5.30. (i) By the relation [WU?P(rJ)]* == P(rJ)U~t one
has IIWU2P('t]) IIHs == llP(rJ)U~t W!!HS· To calculate this Hilbert-Schmidt norm, we
express the operator P(rJ)Uf_}_t as an integral operator in momentum space and
apply Proposition 2.15. We observe that liV is a convolution operator in £2 (JRn ), more
precisely the integral operator with kernel (21f)-ni 2 W(k- k/), see (2.84). By setting
k == vf:Xw and using first (5.69), then the definition (5.99) of P( rJ) and finally .68),
one obtains

[F P( 77 )U~t W f] (k) == 2 1 / 2 .A (2 -n)/Ll [U oP('l] )U~t W f] A (w)

= 21/2 A(2~n) I47J( A) [ .Ia oo7J(JL) [U oU~t W.f] 11 dJL] (W)

= 2112,\ (2~n) I47J( ,\) [ ;: oo7J(JL )e""t (U 0W.f)'" dJL] (W)

= 2112 A(2~n) I 4 7J( A) fa oo dtLrl(l~ )e'llt [2~ 112 JL (n 2) 14 (FW f)] (foW)

(2~n) I4 1!( ,\) ;: dJL7](JL )e'id JL (n~2) I 4


00
= ,\

X (21f)~nl 2 r dnk'W(foW- k')i(k').


} TJKn
248 EVOLUTION GROUPS AND SCATTERING THEORY

This shows that P(17)U~t W is indeed an integral operator in L2 (JRn) and that its
kernel Nt (k, k') is given by

Nt(k, f') = (21f) -n/2 A(2-n)/477(A) laoodp,p,(n-2)/417(/L )eil-'t W( VJiW- f'),

with k == vf\w. Thus, by observing that dnk == (1/2)A (n- 2 )1 2 dAdw, one obtains
from Proposition 2.15 that

IIWU2P(77)II~s = r dnk r dnk'INt(k, k')l


}TJKn }TJKn
2

==
1
-2 (27r)-nii7711
2
Jsn-1
dw r dnk' Ilo(
}YKn
00

dP,77n-2(JL)eil-'tW( VfiW- k')l


2
· (5.107)

We recall that 111711 2 == 1. The integration with respect to df-L may be considered [apart
from a factor (27r) 1 12 ] as a one-dimensional Fourier transformation (extend 17 to 1R by
setting 17(.A) == 0 for A< 0). Thus, upon applying J1Rdt to the expression (5.107), we
can use the unitarity of the Fourier transformation to evaluate the integral with respect
to dt:

J:11WU2P(77)il~sdt = ~(21r)-n+l./ dw ./dnk'la


00
2
dP,hn-2(JL)W(VJiW- f')l
1
== -2 (27r)-n+lj dw r=dJLI77n-2(~LWIIWII 2 ,
sn-l lo
which identical with (5.1 05).
(ii) To prove (5.106) we set Wj(x) - x 1 W(x), j == 1, ... , n. The kernel of the
integral operator tP( 17 )U~t W is given as follows [writing teif-Lt == -i( d/ df-L )ei~-Lt]:

-i(27r )-n/2 A(2-n)/417( A) .Ia oodp,p,(n-2)/417(/L) W( VJiW-k') d~ eil-'t' with k- -J:\W.


In this expression one first integrates by parts with respect to the variable f-L, by ob-
serving that the boundary terms vanish because 17(0) == 17( oo) == 0. Then, as in (i)
above, one can use the unitarity of the Fourier transformation to obtain that

2 0 1
J:t IIWUt P(77)il~sdt = ~(21r)-n+ ./ dw ./dnk'la dJLI11~-2(JL)W(VJiW- k')
00

a ----- 2
+ 17n-2 (p,) O{L W( VJiW- k') I .
--+

(5.108)

If one denotes by w1 (j 1, ... , n) the components of the vector w, one has


!!_ w( Vfiw- k') - ]:_ J-L- 1 ; 2 _!!_ W( Vfiw- k')
atL 2 aVJI
1 2
= };_p,-1/2_!!_ (21r)-nj2 ( e-i,_, 1 W·Xeik'·XW(X)dnx
2 aVJI }1Rn
. n

= -~p,-1/2 LwJWJ(VJiW- k').


j=l
BOUNDS ON SCATTERING CROSS SECTIONS 249

We insert this expression into (5.108) in which we make the change of variables k' ~
fJ =: Vfiw- k':

J:t 2 IIWU2P(7])11~sdt = ~(21r)-n+llsn-ldw laoo dfllh:,_2(fl)W(-)


(5.1 09)

where the norm is in L2 (JRn) with respect to the variable · == p. Next one calculates
the integral with respect to dw by writing the norm occurring in (5.109) as a sum of
scalar products. One has to know the values of the following integrals 20 :

Jrsn-1
WjdW = 0,

One finds that the right-hand side of Eq. (5.109) is equal to


n
2 2
~ (21f) -n+l en [1177~-211 IIWII + ~II ~T)n-411
2
L IIWj 11
j=l
2
]'

which is identical with the right-hand side of Eq. (5.1 06), since

t
n

lnxJIW(XWdnx = j~niXI 2 IW(XWdnx = IIIQIWII 2. D


J=l

Let us apply the preceding proposition to the question of the finiteness the total
scattering cross section. Let TJ be as in Proposition 5.30. By using Eq. (5.102), then
the inequality (5.104) and the Schwarz inequality (for the integral with respect to dt),
and finally (5.105) and (5.106) with l/VT- V, one finds that

fooo)..(n-l)/
2177(AWa(A)dA = ( 2 ~:- IIRP(77)11~s 1

( ~: [J:IIVU2P(77)11Hsdtr
2 1
<

( ~:- [J:(l + t2)- 112· (1 + t 2)1 12IIVU2P(7])11Hsdtr


2 1

( ~:- [J:(l + t 2)- 1 dt] · [J:(l + t 2 )IIVU2P(77)11~sdt]


2 1

<

- ; [1177n-2II 2IIVII 2+ 1177:,-2II 2IIVII 2+ 4~ ll77n-4II 2IIIQI 11 2]. (5.110)

Since TJ is differentiable and vanishes in a neighbourhood of 0 and of oo, all norms in


(5.11 0) containing the function TJ are finite. Thus the right-hand side of (5.11 0) is finite
°For the last integral with j
2
= k, observe that its value is independent of j and that 2::?= 1 w; = 1.
250 EVOLUTION GROUPS AND SCATTERING THEORY

[and consequently the total scattering cross section O"tot (A, w) is finite for almost all
A > 0 and w E sn- 1 ] provided that (1 + lxl) (x) is a function belonging to £ 2 (JR.n ).
Assume for instance that is locally square-integrable and that I ( x) I - c Ixl- p in
a neighbourhood of infinity. Then the preceding condition is satisfied if and only
(3 > (n + 2)/2.
Consider the case n == 3. The condition {3 > (n + 2)/2 becomes {3 > 5/2. This
result is not optimal, it is known that the total scattering cross section is finite if (3 > 2
and infinite if (3 < 2 (for n == 3).
In order to improve the preceding result [i.e. the condition (3 > (n + 2) /2], we ob-
J
serve that the integral ~oo (1 + t 2 ) - 1 dt in (5.11 0) converges too rapidly at infinity: for
large t the integrand behaves like t- 2 , whereas a decay like t- 1 -s, withE > 0, would
be sufficient; in other terms it would suffice to insert in the calculation in (5.11 0),
before applying the Schwarz inequality, the factor (1 + ltl)- 1 12 -v · (1 + 1tl) 112 +v
in place of (1 + t 2 )- 1 12 · (1 + t 2 ) 1 12 , with v > 0. One can see from (5.105) and
(5.106) that the introduction of a factor t in II wu?P( 17) IIHs [more precisely there-
placement of IIWU?P(77)11Hs in (5.105) by 11tWUPP(17)11Hs] is essentially equivalent
to the replacement of W(x) by lxiW(x) in the norm appearing on the right-hand side
of (5.105).
One can therefore expect that the replacement in Eq. (5.105) of II wu?P( 17) IIHs by
0
11t WUPP(77)11Hs should amount to replacing Won the right-hand side by IQia
Assuming this and taking IW(x)l - lxl-11, one has lxlaiW(x)l - ixla-11, and this
latter function is square-integrable in a neighbourhood of infinity if (3 a > n/2, i.e.
(3 > (n + 1)/2 if a is sufficiently close to 1/2 (recall that one needs a == 1/2 v,
with v > 0, for the integral J_ 00

00
(1 + 1tl)- 20 dt to be finite). For n == 3, the condition
(3 > (n 1) /2 is equivalent to {3 > 2, so one can expect to arrive in this way at an
essentially optimal result on the finiteness of total scattering cross sections.
The fact that the replacement of IIWUPP(17)11Hs by lltawu?P(77)11Hs in the left-
hand side of (5.1 05) is essentially equivalent to the replacement of the operator W (Q)
by IQiaW( Q) in the norm on the right-hand side is not really surprising. In a classical
approximation one has Q ~ 2Pt at large times (see Remark 5.19). Also, by (5.62):
llcp(Q)[l?P(17)i!Hs == llcp(2Pt)ZtP(77)11Hs, hence

11taW(Q)U2P(77)11Hs == lltaW(2Pt)ZtP(77)11Hs
== 0
II [(21Pit) W(2Pt)] (2IPI)-a ZtP(17) IIHS·

Since 77 is assumed to be zero near the origin, there exists a number b > 0 such that
17 (A) == 0 if 0 < A < b, and hence state vectors in the range of P (17) have a minimal
momentum 61 12 . So, for large It I (Zt ~I), the last expression should be less than

This suggests that one has indeed f~oo iltawu?P( 17) ll~s dt < Cn (17) IIIQiaW( Q) 11 2
for some constant Cn (17) depending on nand 17· The next proposition gives a rigourous
meaning to this.
BOUNDS ON SCATTERING CROSS SECTIONS 251

Proposition 5.31. Let rJ be as in Proposition 5.30, a E (0, 1) and V: JRn ---+ JR such
that (1 + l·l)aV(·) E L 2 (JRn). Then there exists a finite constant C- C(TJ) such that

(5.111)

PROOF. Denote the kernel of the Hilbert-Schmidt operator V( Q)U?P( TJ) by Mt (x, if).
Then one has

For fixed t, one divides the domain of integration with respect to dnx into two parts
01 and 02, with 01 == {xE JRn llxi > ltl} and 02 == {xE JRn lixl <It!}. In 01 we
majorise (1 + 1tl) 2 a by (1 + lxl) 2 a, then we extend the integral over 0 1 again to all
of JRn. In 0 2 we use the following estimate:

(1 + 1ti)2a = (1 + 1tl)2<> . (1 + IXI)2a < { (1 + JtJ) }2 . (1 + IXI)2a


(1 + Jxl) 2 a (1 lxl)

< ~1(~ ~~~~ . (1 + IXI)2a < 2(1 + IXI)2a + 2t2(1 + IXI)2a-2.


Again we extend the integral over 0 2 to JRn. We obtain

i:(l + 1ti) "IIV(Q)U2P(ry)ll~sdt


2

<3 i: dtlndnx ln dny(1 + IXI) 2"1Mt(X, YW

+ 2! 00

-CX)
t 2 dt
}
r dnx r dny(1 + IXI) 2"- 21Mt(X, Y)l 2
JRn } JRn

= 3/_:11(1 + IQI)"V(Q)U2P(ry)ll~sdt

+2 i: 2
t II(J + IQI)a-l (Q)UP P(ry)ll~sdt.

Hence, by using Proposition 5 .30, one arrives at

/_:(1 + 1ti) 2"11 (Q)u2P(ry)ll~sdt < 3Cl('I7)11(J IQI)" 11 2

+ 2C2(TJ)!I(J + IQI)a-l 11
2 + 2C3(TJ)IIIQI(J + IQia- 1 ) 11
2

< C(TJ)II(J + IQI)aVII 2,

because II(I + IQI)- 1 1 == IIIQI(I + IQI)- 1 11 == 1. D


252 EVOLUTION GROUPS AND SCATTERING THEORY

is very interesting to observe that the local behaviour of the potential has no
influence on the finiteness of the total scattering cross section. More precisely, if
flxi;?:Ri (1 + lxl)a (x) 2 dnx < oo for some R E [0, oo ), then o-(-X) < oo for almost
l

all values of A [the Hamiltonian H can be an arbitrary self-adjoint extension of the


symmetric operator H 0 + (Q), as in point (3) on page 225]. To see this one uses the
fact, proved on page 225, that 0± == s-limt-t±oo Ut'l/J( Q)U?, where 1/J is a function
of class coo such that 'lj;(x) - 0 if lxl < Rand 'lj;(x) == 1 if lxl > 2R. In place of
(5.1 03) one can then write (by using the calculations and notations of page 226):

R- S I== n+(n_- ft+) == -n+1oo !!:_Ut'l/J(Q)UPdt


L:
-00 dt

= -m+ ut{v(Q)'l/J(Q)- [Ll'l/JJ(Q)- 2i[V'l/JJ(Q). P}uPdt.

Consequently [compare with (5.104)]:

IIRP( 77) I!Hs < L: !IV( Q)'lj;( Q)UPP(ry) I!Hs dt + L: II [Ll'l/;]( Q)UPP( 77) IIHs dt

+2 f= 1oo II [V'l/;]J(
J=l -00
0
Q)Ut PJP(ry) IIH 8 dt. (5.112)

The first integral is finite under the specified condition on the potential V, and the
remaining integrals are finite because the derivatives of 1/J have compact support. One
can use Proposition 5.30 to estimate these integrals and obtain bounds in terms of cer-
tain norms on the function TJ, of the parameter Rand of ~xi;?:R I(1 + lxl)aV(x) 2 dnx. l

We refer to [A] for details [see also Problem 5.17 for the last integral in (5.112)].

5.8 Coulomb scattering

In §5 .4.3 it was pointed out that for a Coulomb potential, and more generally for
long range potentials, the wave operators do not exist, namely that Ut u?
will not be
strongly convergent as t -+ ±oo, and that this property is due to the slow decay of
the potential at large distances from the scattering centre. It turns out that one can still
introduce some kind of wave operators generalising the notion considered before. We
shall first describe the definition and some properties of such modified wave opera-
tors for long range potentials, and then we shall prove their existence for Coulomb
potentials.

5.8.1. The modified wave operators are given as strong limits, as t -+ ±oo, of opera-
tors of the form UtU? e-iXt, where {Xt }tEIR is a time-dependent family of self-adjoint
operators commuting with the free Hamiltonian H 0 (for spherically symmetric poten-
tials the operators Xt are functions of H 0 , and more generally they are functions of
the momentum operator P). Thus the free evolution u? - e-iHot occurring in the
COULOMB SCATTERING 253

standard definition of wave operators is replaced here by unitary operators of the form
U? e-iXt _ e-i[Hot+Xt ]:

0±- s-lim UtU?e-iXt ==s-lim eiHte-i[Hot+Xt]. (5.113)


t~±CX) t~±CX)

As a function oft, the operators e-i[Hot+Xt] will not form a group: the dependence of
Xt on the parameter tis not linear but rather such that Xt is small compared to H 0 tat
large values oft, viz. such that Xt/t converges to zero as t ~ ±oo [see Proposition
5.32(a) for an alternative and more precise formulation of this property]. In this sense
the modified free evolution used to define the (modified) wave operators for long range
potentials is close to the free evolution e-iHot at large times.
A justification for the introduction of these modified wave operators can be found
through the algebraic scattering theory mentioned on page 213. Let us consider first
a situation in which the usual wave operators 0± - s-lim t~±CX) Ut U? exist. Let
A be a free observable, i.e. a bounded self-adjoint operator commuting with the free
Hamiltonian H 0 (i.e. with each UP). If f is a vector in the range of 0+, then [see
Proposition 5.15(a)]

as t ~ +oo. (5.114)

One sees that the asymptotic condition is satisfied in the algebraic sense, i.e. the expec-
tation values (Utf, AUt f) have limits as t ~ +oo for each observable A commuting
with H 0 and for each state vector fin the range of 0+. In many situations the range of
0+ coincides with the set of all scattering states of the total Hamiltonian H, and then
the asymptotic values of (Utf, AUt f) at t == +oo exist for each free observable A and
each scattering state of H. Analogous statements can be made for the limit t ~ -oo.
A similar reasoning applies for the modified wave operators. If A is an observable
commuting with each u?
and also with each Xt, then we have as in (5.114), with
21
jER(0+):

as t ~ +oo.
(5.115)
Again the ranges of 0± frequently coincide with the set of all scattering states of H,
and in many situations (in particular for Coulomb potentials), the operators Xt are
functions of the free Hamiltonian H 0 . Then the asymptotic condition in the algebraic
sense is again satisfied for all scattering states of H and all observables A that com-
mute with the free Hamiltonian (if the operators Xt -which commute with H 0 -are
not functions of H 0 , then asymptotic convergence of (Utf,AUtf) will hold only for
a more restricted class of free observables A than that indicated above).
The equation (5.113) means that, in a long range potential, the evolution of a scat-
tering state of H will not be asymptotically comparable to a free evolution but rather to
a somewhat distorted free evolution. Iff is a scattering state of H then, as t ~ +oo,
the state vector e-iHtf will not be close to e-iHotf+ (for some scattering state f+
21 As in the proof of Proposition 5.15(a) one finds that the operators e1 Xtu2*Ut converge strongly to
O.t- on the subspace R(O+ ).
254 EVOLUTION GROUPS AND SCATTERING THEORY

of H 0 ) as in Eq. (5.39) but rather to e-i[Hot+Xt]j+· The distortion of the free evolu-
tion (i.e. the operator e-iXt) takes into account the effect of the potential at very large
distances, in particular it depends on the potential.
It should be said that the operators Xt are highly non-unique. Only their asymptotic
behaviour (as t ~ ±oo) is essential for obtaining convergence of UtU?e-iXt. For
example the addition of a fixed self-adjoint operator Z to each Xt will lead to new
wave operators O±e-iZ if 0± == s-limt--+±oo ut u? e-iXt. However, provided that z
is a function of H 0 , the asymptotic values of (Utf, AUt f) in (5.115) will not depend
on Z. More generally, without changing the asymptotic values of (Utf,AUtf), one
has the freedom of adding to each Xt an operator Zt depending on Ho and on the
value oft such that the family { Zt}tEIR satisfies s-limt--++oo e-iZt - eiZ == 0 for
some fixed self-adjoint operator Z.
The determination of a suitable family {Xt} for a given potential V may be moti-
vated by the use of an adaptation of the Cook Criterion (Proposition 5 .16) for proving
asymptotic convergence. By applying the Leibniz rule one finds, as in the proof of
Proposition 5 .16, that

-d U*Uo
t t e -1Xtj - ~·u*t [H - rr -
110 d X t J Uo
- t e -iXtj -- ~·u*t [V - -d X t J Uo
t e -iXtj
dt dt dt '
(5.116)

and the existence of S -lin1 t--++oo Ut U? e-iXt is guaranteed if one knows that, for a
dense set of vectors f' the function t r----t II [V - (dI dt) X t] u? e- iX t f II belongs to
L 1 ([1\t, oo)) for so1ne ME [0, oo ). For a short range potential [when Xt - 0, hence
(dl dt)Xt == 0], this integrability condition is satisfied as a consequence of the suf-
ficiently rapid decay of the function V at infinity, see the discussion under point (C)
on page 224. For a long range potential the function t r----t II vu? f II is not integrable
at infinity, but the subtraction of a suitably chosen term of the form (dl dt)Xt from
V will improve the decay of the occurring L 2 norm at large values of t. By taking
into account that Q(t) ~ 2Pt at large times, one would take Xt _ Xt(P) such that
(dldt)Xt(P) - V(2Pt). It turns out that, with this choice of Xt, the modified wave
operators exist for long range potentials decaying like xi- (3, with 112 < f3 < 1 (in
1

particular for Coulomb potentials). 22 For more slowly decaying potentials a suitable
choice for Xt is given by the solution of the non-linear equation

or by some approximation of this solution. In §5.8.2 below we discuss the modified


wave operators for Coulomb potentials. More comprehensive treatments of long range
scattering and references to the literature can be found for example in Chapter 13 of
[AJS], in Section XI.9, Volume III of [RS] and in [P].
22 Of course this construction of a modified free evolution can also be applied to short range potentials,
for example for potentials decaying like Ixl- ,L3 with (3 > 1. In this case both the modified wave operators
s-lirnt---f±(X) UtU?e- 1 Xt and the usual wave operators s-lirnt---f±(X) UtU2 will exist, and they will be
identical (for each sign of ti1ne) except for the non-uniqueness pointed out before.
COULOMB SCATTERING 255

To end this subsection, let us point out some general properties of the n1odified
wave operators.

Proposition 5.32 . Assume that there are constants c > 0, R > 0 and (3 > 0 such that
IV(x)l < clxl-/3 for lxl > R. Let Xt == Xt(P), where the function Xt: JRn ~ JR
satisfies (8/ot)Xt(k) V(2kt). Then
(a) The family of unitary operators { e-iXt} is feebly oscillating in the following sense:
s -limt--+±CXJ [e-iXt+T- e-iXt] - 0 for each fixed T E JR.
(b) Assume that the wave operators 0± == s -lin1 t--+±CXJ Ut U2 e-iXt exist. Then 0±
are isometric operators which intertwine Hand H 0 , i.e. one has UtO± == O±U2 for
each t E JR.

PROOF. (a) We have for example fort> 0 and T > 0:

If k f- 0 and t > 0 is so large that 21 kl t > R, we obtain

CT
<---
[2iklt]!3

This shows that, as t ~ +oo, e-iXt T(k) - e-iXt(k) converges to zero for almost
all k E JRn. Since ie-iXt+T(k)- e-iXt(k)l < 2, one finds by applying the Dominated
Convergence Theorem in L2 (JRn) that II [e-iXt+T- e-iXt]fil 2 converges to zero for
each vector f.
(b) The operators 0± are strong limits of a family of unitary operators, hence
they are isometric [as in the proof of Proposition 5.13(a)]. To verify the intertwining
property we write for T E JR:

n
UT~~± == 1·
S- lffi
U*t-T uote -iXt == 1.
S- lffi
U*Uo
S Se -iX5 ·e 2X8 e -iXS + T uoT' (5.117)
t--+±CXJ s--+±CXJ

Since s-lim s--+±CXJ eiXs e-iXs + 7 == I as a consequence of the result of (a), (5.117)
shows together with Proposition 2.1 that UTO± == n±u$. D

5. 8.2. Now we consider Coulomb potentials V(x) == r lxl- 1 in L 2 (JRn ), with n > 3. It
is of course assumed that r E JR and r f- 0. Fork E JRn we set k == lkl. A solution of
the equation (8j8t)Xt(k) == V(2kt) == r(2kt)- 1 is given by Xt(k) == r(2k) 1 logt
fort> 0 and by Xt(k) == -r(2k) 1 log ltl fort< 0. We treat the limit t ~ +oo. To
ascertain the existence of the wave operator 0+ == s-limt--++CXJ UtU?e-iXt, we show
for a dense set D of vectors f that II[V (djdt)Xt]U?e- 2x 1fll is integrable as a
function oft in some neighbourhood oft == +oo. Here Ut == e-iHct, where He ==
Ho rlcJI- 1 is the Coulomb Hamiltonian, and Xt(k)- r(2k)- 1 logt.
We take V == {f E L 2 (JRn) If
E C 0 (JRn\{O} )}. Clearly V S(JRn), and further-
more V E V(H0 ) for any m E JR because of the support properties of the Fourier
256 EVOLUTION GROUPS AND SCATTERING THEORY

transforms of the functions in V. Also, iff E V, then FUPe-iXtj E C 0 (1Rn \ {0} ),


hence u?e-iXtj E V, and t ~ e-iXtj is strongly differentiable.
From now on we consider a fixed vector f belonging to V. Let a > 0 be such that
](k) == 0 for all k E ]Rn satisfying lkl <a. We choose a function¢: ]Rn ~ [0, 1] of
class coo satisfying ¢(k) == 0 if lkl < a/2 and ¢(k)- 1 for lkl >a. We set V(x) -
V(x)¢(x) == rixi 1 ¢(x) and write V == V W, with W(x) == V(x)[1 ¢(x)]. This
gives a splitting of the Coulomb potential into a short range part W (the support of
the function l¥ is contained in the ball { x E lRn llxl < a}) and a bounded long range
potential V satisfying

IVV(x)l < c(1 + lxl)- 2 and I~V(x)l < c(1 + lxl)- 3 (5.118)

for some constant c and all x E ]Rn (notice that this decomposition of V and the con-
stant c depend on the vector f).
We have

IIW !xt]U?e-ixtfll = II[V(Q)- V(2Pt)]U?e-ix fll 1

< II[V(Q)- V(2Pt)]UPe-iXtfll + IIW(Q)UPe-iXtfil IIW(2Pt)fll· (5.119)

We now show that each of the three terms on the last line belongs L 1 ( [ e, oo)) as a
function oft. The lower endpoint of the involved interval is taken to bee so that we
can use the inequality 1 < log t, and from now on we assume that t > e.
2
We first discuss the long range part in (5 .119). With Zt - e iQ I (4 t) we have by
(5.62):

II [V(Q) V(2Pt)]U?e-ixtfll - I [u?*v(Q)UP V(2Pt)]e-ixtfll


- II [ZtV(2Pt)Zt- V(2Pt)]e-iXtfll

- I t dE~e-icQ2/4tV(2Pt)eic:Q2/4te-iXtfll
Jo 8c;
4
r1 dEZt;c:[Q-2 ,V(2Pt)]Zt;c:e-•
- 1t I Jo - .x
tf I

< :t j~ 1 dE I [Q2' V(2h)]Zt;c:e-iXtf II·

The commutator is easily calculated by remembering that Q2 acts as - ~ in momen-


tum space: [Q 2 , V(2Pt)] == (2t) 2 (~V)(2Pt)- 4it(VV)(2Pt) · Q. By using also
zt
the equation 1srJ(2Pt)Zt;s - u?1:rJ(c;Q)U21s, which is a simple consequence of
(5.62), one finds

II [V( Q)- V(2Pt)] uP e-iXtj II


<fa dE I [t( ~V) (EQ)U21c:e-iX f- i('\lV) (EQ)U2;c: · Qe-iX f] II·
1
1 1
COULOMB SCATTERING 257

It is useful to introduce the notation (cQ) == (I + c 2 Q2 ) 1 / 2 . Then the preceding


equality implies together with (5 .118) that

3 2
So it suffices to estimate II (cQ) - u?1se-iXtfll and II (cQ) - U?;sQ 1 e- 1 x 1 fll as func-
tions of E and t. The results of the following lemma show that the integrand in the
preceding inequality is bounded, uniformly in E E (0, 1], by c1 (log t) 3 jt 2 for some
finite constant c1 depending on f (using the assumption that log t > 1). Consequently
II[V(Q)- V(2Pt)]UPe-iXtfll belongs to L 1 ([e, oo)).
Lemma 5.33. Let f E L 2(JRn) be such that j E C~(JRn
\ {0} ), with n > 3, and let
k E {0, 1, 2, 3, ... }. Then there exist finite constants ck(f) and ck(f) such that.for all
t > e and all E E (0,
II (cQ) -ku?;se-ixtfll < ck(f)(logt)kt-k (5.120)
and
ll(cQ)-kU?;sQje-iXtfll < ck(f)(logt)k+ 1 t-k. (5.121)

Before proving the lemma, let us consider the short range terms in (5.119). By the
support properties of j and W we have W(2kt)](k) - 0 for all k E JRn if t > /2,
so that I!W(2Pt)fll == 0 on [e, oo). For the term I!W(Q)UPe-iXtfll the argument is
similar to that applied in the proof of existence of short range wave operators given
in (C) on page 224. For hE S(JRn) we have W(Q)h == s-lin1N-+CXJ WN(Q)h with
WN(x)- X[ 1 /N,CXJ)(Ixi)W(x). Since WN E LCXJ(JRn), we obtain by using (5.62):

IIW(Q)UPe-iXtfll - li1n IIWN(Q)UPc-~x 1 fll == lim I!VVN(2Pt)Zte-~xtfll


N-+CXJ N-+CXJ
< lim I!WN(2Pt)(Q) - nllll (Q) 2 ne-~x 1 fl!.
2
(5.122)
N-+CXJ

Now [see (5.66)]

< IIWN(2kt)llqll(1 x 2)-nllq < IIW(2kt)llqll(1 x2)-nllq


217
IIWN(2Pt)(Q)- 11

< t-nfqi!WIIqll(1 + x2)-nllq· (5.123)

2
By choosing q == n- /4, so that I!WIIq < oo, we have I!VVN(2Pt)(Q)- nll <
cnt_ 1 _ 6 forsomeconstantc 17 andallNEN, with6 == 1/(4n- ) > 0. Furthermore,
since Q1 acts as 8/okj in momentum space and logt > 1, one has II (Q) ne-~x 1 fll <
2

CJ (log t) 2n for some constant CJ. So (5.122) and (5.123) imply that II (Q) e-iXtfJI
1
belongs to L ([e, oo) ).
PROOF OF LEMMA 5 .33. We shall use the fact that, for any rn E JR, the vectors H0 n~f,
PjH0 mf, Qjf and P1 H0 mQjf all belong to V.
258 EVOLUTION GROUPS AND SCATTERING THEORY

(i) We first prove (5.120) by induction. Clearly (5.120) is satisfied fork == 0, vvith
co (f) == II f II· As a preparation for the induction step we observe that, by the commu-
tationrelation [Qj,Us0 ] == 2P1 sU~, one has for g ES(JRn):

EQjU?;sPJg- u?;s(EQyPj + 2Pft)g.


After summing over j and then setting g == (2P 2 t)- 1 e-iXtf, one obtains the follow-
ing identity:

~uo Q-. PR-1 -ixtf (5.124)


2t t/ s o e .

Now assume that (5.120) holds fork == N, and consider II (cQ) -N - 1 U2;se-iXtfil·
1
By using (5.124) and observing that II (cQ) - 11 1, one finds forE E (0, 1]:

;t L ll(cQ)
n
1 1
ll(cQ)-N-lUt/ce-iX,fll < CQJIIII(cQ)-NUt/ce-iX1pjHo- fll
j=1

(5.125)

1
Since II (cQ) EQ 1 II == , the first term on the right-hand side of this inequality is
bounded by [cN(f)/2](logt)Nt-N- 1, which is less than [cN(f)/2](logt)N+ 1t-N- 1
if > e. To handle the second term on the right-hand side we commute the operators
Q 1 through P1 H0- 1 e-~xt; remembering that Qj acts as 8/okj in momentum space,
this gives
n

Q· PH0 1e-'x'f = e-iX1 [(n- 2)iH0- 1


f + L P1H0- 1QJf- ~ (logt)H0- 312f.
J=l
(5.126)
By inserting this expression into the last term in (5 .125) and by using the induction
hypothesis, one finds that this term is bounded by f);N (f) (log t)N + 1t-N - 1 with

1
KN(f) == (n- 2)cN(Ho- f)+ CN(Ho- p · Qf)
1 - - r CN(Ho-3/2 f).
2
We conclude that (5.120)holds fork- N + , with CN+1(f)- cN(f)/2 + f);N(f).
(ii) The inequality (5.121) follows easily from (5.120) by observing that Qje-iXtf
Q 1 f- (r/2)(logt)e-iXtpjH~ 312 f. D
REMARKS. (a) By an adjustment of the arguments presented before the proof of
Lemma 5.33 one easily sees that one may add a short range potential Vs to rlxl- 1
and still get strong convergence of eiHtu?e-~xt, where H == H 0 rlxl- 1 Vs and
Xt is as before.
(b) It is rather clear that essentially the same proof as that presented for Coulomb
potentials can show the existence of modified wave operators for other long range po-
tentials [for details see the references mentioned before Proposition 5. 32].
COULOMB SCATTERING 259

(c) Unlike the asymptotic expectation values of observables, the scattering operator
for long range potentials will not be unique. More precisely, if one requires the dis-
tortion e-iXt of the free evolution to be a function of Ho, then limt-+±oo (Utf, AUt f)
for a free observable A are uniquely determined, whereas the scattering operator S ==
0~0- is determined only up to a unitary factor eiZ, where Z is a self-adjoint function
of H 0 . So the S-matrices S(A) at energy A for different admissible distortions of the
free evolution may differ by a factor eia(-\), with a(A) E JR. However the probability
of scattering into a cone C is not effected by this indeterminacy: the Scattering-Into-
Cones Formula (5.83) is valid for Coulomb potentials (Problem 5.18), and the value
of the integral in (5.83) does not change if Sis multiplied by a unitary operator of the
form eiZ(Ho).

5 . 8.3 . To complete our discussion of Coulomb scattering we show that the usual wave
operators do not exist for a Coulomb potential, i.e. that Ut u? cannot have strong limits
as t -----7 ±oo if Ut == e-iHct. Specifically we shall prove that UtU? converge weakly
to zero as t -----7 ±oo. Then, assuming the existence of W ± :== s-limt-+±oo UtU?, one
would have 0 == w -lim t-+±oo ut u? == w±' a contradiction because the limit oper-
ators W ± must be isometries as the strong limits of a family of unitary operators (see
Proposition 5.13).
We begin by establishing an extension of the result of Lemma 5.33. Let f be as
in Lemma 5.33, and let Ck (f) and Ck (f) be the constants constructed in the proof of
Lemma 5.33, k E {0, 1, 2, 3, ... }. Then one has for 1 < s < t:

I (c:Q) -kU2;se-iXteiXs!II < ck(f)(log t)kt-k (5.127)

and I (c:Q) -kU2;sQ]e-iXteiXs!II < ck(f)(logt)k+lt-k. (5.128)


PROOF. As in the proof of Lemma 5.33 one can proceed by induction. It is clear that
(5 .127) is satisfied for k == 0. So assume that it holds for k == N and consider the
quantity I (c:Q) -N - l U2;se-iXt eiXs f II· One can repeat the arguments given after the
equation (5.124) by replacing everywhere e-iXt by e-i(Xt- Xs). After this replacement
all the expressions in the proof of Lemma 5.33 remain valid except that the last term in
(5.126) now becomes (r /2) (log t logs )H~ 312f. By observing that I log t logs I <
log t if 1 < s < t, one finds as before that (5 .127) holds for k == N 1, with the same
constant CN+ 1 (f). Finally (5.128) follows from (5.127) as in the proof of Lemma
5.33. D
Next let f belong to the dense D set introduced at the beginning of §5.8.2. Then

IIO+eiXsj- u;u~JII- 11[0+- u;u~e-iXs]eiXsfll

< 1 II!
(X) I
(Ut uf e-iXt) eiXsf dt = 1I
(X) [V(<;J) - V(2Pt)]U2 e-iXt eiXsf I dt.

By following the proof of the existence of the modified wave operators given in §5.8.2,
using (5.127) and (5.128) in the place of (5.120) and (5.121), one finds that the inte-
grand in the last expression is majorized by an s- independent function oft belonging
to £ 1 ([s, oo)) if s >e. It follows that s-lims-+oo[O+eiXs- U;U~] == 0.
260 EVOLUTION GROUPS AND SCATTERING THEORY

Now, taking again fED, we have for any g E 7-{:

Ass ~ oo, the second term on the right-hand side converges to zero by the preceding
result, and so does the first one because, as shown below, eiXs converges weakly to
zero ass ~ oo. Hence w-lims-+CX) u;u2 f == 0 for each fED and hence for each
f E H (see Problem 2.5).
It remains to prove that w-lims-+CX) eiXs == 0, or equivalently (again by Problem
2.5) that lin1s-+CXJ (g, eiXs f) == 0 for all j, g E D. So let j, g E D. Then, setting
h(k) == g(k)](k) and B(k) == kn-lj~n 1 h(kw)dw, with k ikl E (0, oo), we have

By making the change of variables k f--7 u == I (2k) this becomes

(5.129)

e(
Since h E D' the function u f--7 u- 2 1I (2u)) is bounded and has compact support in
(0, oo ), hence it belongs to L 1 ( (0, oo) ). By the Riemann-Lebesgue Lemma (Proposi-
tion 5.6) the integral in (5.129) converges to zero ass ~ oo.

Bibliographical Notes

We have essentially followed [A] for the discussion of evolution groups and scattering
cross sections. For scattering theory we cite [AJS], [Ne], [P], Volume III of [RS] and
[BW]. Extensive bibliographies concerning spectral and scattering theory are given in
[BW] and in Chapter 14 of [P]. Alternative proofs of Proposition 5.9(a), based on the
Mean Ergodic Theorem, can be found for example in [A] and in [14].

Problems

5 . 1. Show that a one-parameter group { Ut} of unitary operators is strongly continuous


and only if it is weakly continuous.
5 . 2. Prove the uniqueness of the infinitesimal generator of an evolution group: If A
and B are self-adjoint operators such that Ut == e-iAt == e-iBt for all t E JR, then
A== B.
5 . 3. Let { Ut} be an evolution group and A its infinitesimal generator. Define Ut by
. Verify that the family { Ut} is also an evolution group and determine its
infinitesimal generator.
PROBLEMS 261

5 . 4 . Let {Ut} be the following family of operators L 2(JRn):

(5.130)

(a) Show that t ~---t Ut defines a strongly continuous unitary one-parameter group (an
evolution group), which gives a unitary representation of the dilation group mentioned
in Example 5.4.
(b) Determine the infinitesimal generator of this group.
(c) The space C~(JRn \ {0}) (the set of functions in C~(JR 71 ) that vanish near x- 0)
2
is dense in L (JRn). By using Nelson's Criterion (Proposition 5.3), show that there-
striction of the infinitesimal generator to C 0 (JRn \ {0}) is an essentially self-adjoint
operator.
5 . 5 . Let { Ut} be an evolution group satisfying U (27T) == I. Show that its infinitesi-
mal generator has no continuous spectrum and that all its eigenvalues are integers, i.e.
o-(A) == o-p(A) C Z. What can be said about o-d(A)?
5 . 6 . Prove the Mean Ergodic Theorem (Proposition 5.8).
5 . 7 . Verify that the statement of Proposition 5.9(a) has a converse: iff is a vector in
J
H such that limT-HXJ r- 1 0T I(h, Utf) 2dt == 0 for each hE H, then f E Hc(A).
1

-+ - - -+ -
5 . 8 . Prove that Moc(H), M~(II) and MCXJ(H) _ Moo(H) n M~(H), defined in
(5.30), are subspaces and that each of these subspaces is invariant under the operators
Ut.
5 . 9 . (a) Let A be a self-adjoint operator in a Hilbert space H, ;\ a real number and
f E H. Show that e-i(A-.A)t f is strongly convergent as t ~ +oo if and only iff is
an eigenvector of A associated with the eigenvalue :\.
(b) Let B be a second self-adjoint operator. Assume that s-lirnt-++oo eiAte- 1 Btg ex-
ists and that B g - :\g for some ;\ E JR. Then Ag == :\g.
5 . 10. LethE L 1 (JR 3 ) n L 2(JR 3 ) such that llhll llhll2 == 1. Let Vh be the operator
defined by

or in Dirac notation Vh == Ih) (hI· Prove the existence of the wave operators for
H == P2 + r Vh, where r is a real constant.
REMARK. An interaction of the form r V/1 is called a separable potential and is used
for example in models of nuclear physics. Mathematically such an operator is an op-
erator of rank 1.
5 . 11 . In 1-{ == L 2 (JR), let P be the self-adjoint momentum operator [formally given by
-i( d/ dx )]. Let A - + V, where Vis the multiplication operator by a bounded real
function belonging to L 1 (JR).
(a) Prove that the wave operators n± == s-lin1t-+±CXJ eiAte-iPt exist and are unitary
operators.
(b) Give explicit expressions for 0± and for the scattering operator S == * [2_ in
terms of the potential V.
[Hint: Show first that [e -iPt f] (x) == f (x - t) and that there is a function W such that
[e-iAtf](x) e-iW(x)+iW(x-t) j(x _ t).]
262 EVOLUTION GROUPS AND SCATTERING THEORY

5.12. Let H be an infinite-dimensional Hilbert space and { ej,k} an orthonormal basis


of H, where j == 1, 2, ... and k == 0, ±1, ±2, .... Let X and Y be the following
operators defined by their action on each e j, k:

X ej,k - ej,k+l
Y ej,k == ej,k+l if k I- - j' - j + 1
Y ej,-j - ej+l,-j+l
Yej,-j+l == ej-1,-]+2 if j I- 1
Ye1 o - e1 1
' '
(a) Make a pictorial representation of the action of X andY by representing each e1 ,k
by its co-ordinates (j, k) in the plane IR 2 and verify that X andY are unitary opera-
tors.
(b) For each integer n, let U~ == xn and Un == yn. Show that the wave operators
0± - s-limn--+±oo u~u~ exist.
(c) Determine O±ej,k and show that R(O+) ==Hand R(O_) I- H. Verify explicitly
that S- 0~0- is isometric but not unitary and that w-limn--+-oo u~*Unf == 0 for
each f orthogonal to R(O_ ).
5.13. Let A 1 , A 2 and A 3 be self-adjoint operators in a Hilbert space H. For j == 1, 2, 3
denote by. Ej the projection onto the absolutely continuous subspace Hac ( Aj) of Aj
and set 01k ==s-lim t--+±oo eiA1te-iAkt Ek if the limit exists.
(a) Assume that 0~ and 0~ exist. Show that then Oi3 also exist and that Oi3 ==
Oi2 0i3 (chain rule for wave operators).
(b) Assume that Oi2 and Oi1 exist. Show that R( Oi1 ) == E 2 H. In particular, if the
spectrum of A 1 is purely absolutely continuous, the scattering operator (0~1 )* 0~1 for
the couple { A 2 , A 1 } is unitary.
5.14 . In §5.3.1 the asymptotic condition was introduced on the basis of strong conver-
gence, for example it was required that, for given f E M~(H), there should exist a
vector f+ such that s-lirnt--++oo[Utf- e-iHotf+]- 0. Show that if this hypothesis is
weakened to the requirement that w-limt--++oo [Utf e-iHotf+] - 0, then the vector
f + will in general not be unique.
5.15 . Let { S (t) }t>o be a strongly continuous family of bounded self-adjoint operators
satisfying S(t)S(T)- S(t + T), S(O) ==I and IIS(t)ll < emt for some real constant
m and all t, T > 0. Prove that there is a self-adjoint operator A such that S(t) -e-At.
[Hint: Follow the proof of Stone's Theorem.]
5.16. Prove Proposition 5.24 and Corollary 5.25.
5.17. Let W, r; and TJa be as in Proposition 5 .30. Show that, for j 1, ... , n:

5.18. Prove the validity of the Scattering-Into-Cones Formula (5.83) for scattering by
a Coulomb potential.
CHAPTER6

The Conjugate Operator Method

We have seen in Section 4.4 that, if H is a self-adjoint operator and A E JR. a point
in its spectrum, then (H A- iJ-L)- 1 cannot converge [in B('H)] asp, --7 0, but that
(f, (H- A- iJ-L)- 1 f) may have a limit or remain bounded for certain vectors f. We
have also seen that these properties of convergence or boundedness are linked to de-
tails of the spectrum of H in a neighbourhood of A. In particular, the positivity of the
commutator of iH with some other self-adjoint operator A leads to absolute continu-
ity of part of the spectrum of H (see Proposition 4.32 and the remarks following its
statement). We discuss here a method that was introduced by E. Mourre for proving
the existence of the limits of (f, (H- A- iJ-L)- 1 f) under certain conditions on H, A
and f. It exploits in a subtle way the idea of obtaining absolute continuity from the
positivity of a commutator. Important consequences are the absence of singular con-
tinuous spectrum of H (see Corollary 6.9) and asymptotic completeness in scattering
theory (see Section 7.1 ). The Mourre method is particularly interesting through the
fact that it also allows the treatment of N -body Hamiltonians, which have a much
more complex structure.
Let us give a brief description of the idea of this method. It relies on the introduction
of two auxiliary quantities: a parameter c and a self-adjoint operator A having suitable
commutation properties with H. The function (f, (H- A- iJ-L)- 1f) is extended to an
c-dependentfunction <I>c(A+iJ-L), 0 < c < 1, with <I>o(A+ip,) == (f, (H A-iJ-L)- 1f).
The definition of <I>c(A iJ-L) involves the commutator between iH and the auxiliary
operator A, and it is essential that this commutator be positive in some sense. One then
establishes a differential inequality for <I>c (as a function of c), i.e. an inequality of the
form

for some function h of two variables (not depending explicitly on z). By integrat-
jng this differential inequality, one rather easily arrives at the existence of the limit
limp,-*+O (f, (H- A- iJ-L)- 1 f). The existence oflimP-*+O (f, (H- A itL)- 1f) can
be obtained in a similar way. In general these two limits will be different.
As in the case of the Spectral Theorem (Appendix to Chapter 4), the arguments are
simpler for self-adjoint operators H that are bounded. For this reason we elaborate
264 THE CONJUGATE OPERATOR METHOD

in Sections 6.2 and 6.3 the essentials of the Mourre method, as well as its conse-
quences for the spectral properties of H, by assuming H to be bounded. In Section
6.4 we extend the results to a large class of unbounded operators H and apply them
to Schrodinger operators in L 2 (JRn). As a motivating introduction, we discuss in Sec-
tion 6.1 a simple exa1nple that already exhibits many of the aspects of the method. In
Section 6.5 we introduce the notion of smooth operators relative to H; this concept is
related to the behaviour of (H A. iJ-L) - l for small values of J-L and has applications
in particular in scattering theory. Section 6.6 is devoted to some resolvent estimates
involving higher order commutators of H with the auxiliary operator A; these results
will be needed for some developments in Chapter 7.
Evaluation or estimation of commutators is often an important tool in proving
mathematical results in quantum mechanics. Commutator calculations play a very im-
portant role in the present context as well as further on, especially in Section 7 .2.
Interesting examples can be found in §6.4.1 and, further on, in the proof of Lemma
7.5 and in the estimates of §7 .2.5. These calculations are complicated through the fact
that they involve unbounded operators. In order not to lengthen the arguments, we
often deter1nine the occurring commutators by formal manipulations. In each instance
one would have to give a rigourous meaning to these manipulations by showing that
they can be justified on a suitable domain of vectors in the Hilbert space. The final
Section 6. 7 of the present chapter is devoted to some comments on this and to the
derivation of some commutator estimates that are used in the text.

simple example

In this introductory section we calculate certain commutators between unbounded op-


erators without taking into account domain questions. The aim is rather to illustrate
the Mourre method, as outlined above, by means of a simple example. A more careful
treatment will be presented in the subsequent sections of this chapter.

We begin by considering a very simple differential inequality involving two


variables E and f-L· We fix some number Eo in (0, 1) and let D be the following subset
of JR 2 : D == { (c, J-L) E JR 2 I 0 < E < Eo, 0 < J-L < 1, E J-L > 0}, Thus Dis a closed
rectangle with one vertex [the point (0, 0)] removed. Let F: D ~ C be a continuous
function assumed to be everywhere non-zero and continuously differentiable with re-
spect to the first variable E on Do == { ( E, /L) ED I 0 < E < Eo} 1 • Assume that the
following differential inequality is satisfied on D 0 :

(6.1)

where f) is a non-negative function on (0, co] satisfying e :== foco fJ( E )de < 00. We
shall show that, under these conditions, F can be extended to a continuous function
1
In the context explained at the beginning of this chapter, one would take F(s, J-L) = <I>c:(A + iJ-L), with
a fixed value of A.
A SIMPLE EXAMPLE 265

on the closed rectangle D, i.e. that there exists a complex number F (0, 0) such that
F( E, JL) -+ F(O, 0) as the point (E, JL) ED tends to (0, 0) in an arbitrary way. In par-
ticular F(O, JL) will converge as JL -+ +0, which is the consequence that is important
in our context.
To obtain the claimed result, we rewrite (6.1) as follows:

(6.2)

For 0 < E <Eo we define a number cp(c) by cp(E) == sup 0 ::;p::; 1 l (E, JL)I and observe
that cp( E) is finite as a consequence of the continuity ofF on D. It then follows from
(6.2) that, for all (E, JL) ED:

1/2
IF (E) p,) I = IF (E) p,) 1 I 21 = IF (E 0 ) p,) 1 I 2 -
leof: a
DT F (T) p,) 1 I 2 dT I

~ + fo
00

< ·13(T)dT. (6.3)

By inserting this estimate into (6.1), one obtains the inequality

(6.4)

with the constant/"\;- ~ 8. This inequality holds on D 0 .


The inequality (6.4) implies that, for 0 < 1J < E <Eo and all JL E [0, 1]:

1.~ f:! F( T, p,) dT I< j~ '13( T) dT.


0

[F(E, p,) F( 7], p,) I = K, (6.5)

Since J0c:o ?9( T )dT < oo, this shows that, for each JL E [ 0, 1], (E, JL) is convergent as
E-+ 0. By taking JL == 0, one can define F(O, 0) as F(O, 0) == limc:---7o F(E, 0).
We have thus extended the function F to the set D, and it remains to verify its
continuity at the point (0, 0). We must show that, given c5 > 0, there exist E1 E (0, co)
and J-Lo E (0, 1) such that IF(c, JL)- F(O, O)l < c5 for all E E [0, E1] and all fL E [0, J-Lo].
For this we write

IF(E, JL)-F(O, O)l < IF(E, JL)-F(EI, JL)I+IF(ci, JL)-F(c:I, O)I+IF(EI, 0)-F(O, O)!.
(6.6)
The inequality (6.5) implies that we can choose E1 E (0, Eo) such that the first and the
third term on the right-hand side of (6.6) are less than c5 /3 for all E E [0, c 1] and any
value of JL. Then, by the continuity on [0, 1] of JL ~---t F (E1, JL), there exists J-Lo > 0
such that IF( E1, JL) - F( E1, 0) I < c5 /3 for all JL E [0, J-Lo]. This completes the proof of
the continuity ofF on D.

6 . 1. 2 . We now discuss an example illustrating the usefulness of a differential inequal-


ity in studying boundary values of a resolvent. Let H be a self-adjoint operator in a
Hilbert space H. We denote by R(z) (H- z)- 1 its resolvent at the point z E p(H)
266 THE CONJUGATE OPERATOR METHOD

(see Section 2.6). We recall from (2.103) that R(z)* == R(z). Thus (j,R(z)f) is the
complex conjugate of (f, R( z) f). Hence, by also using the first resolvent equation
(2.93), we have

IIR(z)fll 2 = (R(z)f,R(z)f)- (f,R(Z)R(z)f) = ~[(f,R(z)f)-


z-z
(f,R(Z)f)]
1 1
== ~ [2i~(f,R(z)f)] - z;-~(f,R(z)f). (6.7)
2z~z ~z

Since ~z == ~z and ~(f,R(z)f) - ~(R(z)f,f) == -~(j,R(z)f), the preceding


equation implies that one also has 2

IIR(Z)fll 2 - IIR(z)*fll 2 - c~ ~(f,R(z)f).


~z
(6.8)

We know from (2.99) that (d/dz)R(z) == R(z) 2 . Furthermore, if A is an arbitrary


operator, one has
R(z)[H,A]R(z) ==R(z)[H- z,A]R(z)
- R(z)(H- z)AR(z)- R(z)A(H- z)R(z)
AR(z)- R(z)A,
l.e.
[A,R(z)] == R(z)[H,A]R(z). (6.9)
In particular, if A is a self-adjoint operator such that [H, iA] == H, then one obtains
from Eq. (6.9) that
d
[·iA,R(z)]- R(z)HR(z) == R(z)(H- z)R(z) zR(z) 2 == R(z) z dzR(z).
(6.10)
Let f be a vector in H and set <I>(z) - (f, R(z)f). One would like to know if
<I>(A ± itL) converges as fL --7 +0. As already pointed out, the Mourre method consists
in so1newhat modifying the resolvent. One considers a family {G c: (z)} of operators
depending on a parameter E (0 < E < Eo) such that G 0 (z) == R(z) and such that, for
c 1- 0, the quantity <I>c:(z) :- (J, Gc:(z)f) has a limit as z tends to the real axis and
satisfies a suitable differential inequality.
To illustrate this, we assume as above that [H, iA] -Hand choose fin the domain
D(A) of A. We take
Gc: (z) == ( e- zc: H - z) - 1 == [e-ic: (H - z eic:)] - 1 == eic: R (z eic:) . (6.11 )
The following property of these operators follows from (6.7) and (6.8):

) ~(!, R(ze" )f) . ) ~ [e-ic (f, Co (z )f)]


2 1 1
liCe (z )fll 2 = liCe (z )*JII = 0.<( 10
= 0.<(
~ ze?c: ~ ze""c:

= I ~(zleic) ~ [e-ic (f, Cc(z )f) Jl < l~(zleic) II(!, Cc(z )f) I· (6.12)

2
The identity IIR(z)fll 2 = IIR(z)fll 2 also follows from (4.42) by taking <p(A) = (A- z)- 1 or
<p( A) = (A - z) - l in that equation.
A SIMPLE EXAMPLE 267

Let us now calculate the derivative (8l8c)G 6 (z) _ G~(z) of G 6 (z) (in the sequel a
prime will denote derivatives with respect to the parameter c). Let us set ( == zeis and
observe that, by using (6.10):

:E R(zei i( ~R(()L:=ze"' = i{- R(() + [iA,R(()J} l(=ze'"


0
) =

- -iR(zei 6 ) - [A, R(zei 6 )].

Thus, by applying the Leibniz rule:

~Gs(z) == ~{ei 6 R(zei 6 )} == iei R(zei + ei 6 ~R(zei 6 )


6 6
)
& & &
== iG (z)- iei R(zei
6 ei [A,R(zei
6 6
)-
6 6
)]

== -eis [A, R(zei == -[A, Gs (z )] .


6
)] (6.13)

It is clear that, if z == A iJ-L with either A > 0 and J-L > 0 or A < 0 and J-L < 0,
and if E is strictly positive and sufficiently small, then zeis belongs to the resolvent set
of H, hence, by (6.11), G 6 (A ± iJ-L) is a well-defined operator in B(H) (in particular
p, == 0 is admitted). It remains to see how to treat the limitE ~ 0. Let us set <I> 6 ( z) ==
(j, G 6 (z)f) and use Eq. (6.13) to express its derivative with respect to c:

I<I>~(z) I = I(Gc(z)*j, Af)- (Aj, Gc(z)f) I < IIAJII [IIGc(z)*JII IIGc(z)fll] ·


By using (6.12) one finds that

I<I>~ (z) I < 211 A f


1
II [ I'S (z eic) 1(f, G c ( z) f) I
]1/2 = 211 A f II [I'S (z1eis) ]1/2 ·
1<I> c ( z) I
1 1

Thus, for z == A+ iJ-L [hence ~(zeis) == A sin c + Mcos c], one has the following in-
equality:

(6.14)

If A > 0, then for J-L > 0 and E E ( 0, 1r I 2]: A sin E J-L cos E > A sin E > A · E ( 2I 1r) ==
21r- 1 Ac, hence for these values of A, J-L and c:

(6.15)

We have arrived at a differential inequality for the function E ~ <I> 6 (A + i M). This
differential inequality is of the form (6.1), with F(c, JL) == <l> 6 (A iM) (for a fixed
value of A) and with 1J(c) == ~IIAJIIc- 1 1 2 . By the results of §6.1.1 we have the
following conclusion 3 :
3 It is easy to see from the definitions of G c: ( z) and <I? c: ( z) that F has all the properties required in
§6.1.1.
268 THE CONJUGATE OPERATOR METHOD

If there exists a self-adjoint operator A such that [H, iA] == H, then .for each f in
the dense linear manifold D(A) and each A > 0, (f, (H A- iJL)- 1f) converges as
f-1 ~ +0.
A simple adjustment of the above arguments shows that (f, (H A - iJ-L) 1f)
converges as JL ~ iff E V(A) and A < 0. One can also prove the existence of
(f, (H- A- iJ-L)- f) for A > 0 and /L ~ -0 and for A < 0 and JL ~ +0 by taking
1

for G E(z) the operator (e+iE H - z) - 1 in place of the choice made in (6.11 ).
COMMENTS. (a) The method leads to a control on (f,R(A.±iO)f) for A 1- 0 but gives
no result for A == 0.
(b) It can be shown that the limits lim fi--++O (f, (H - A - iJL) - 1f) exist uniformly in
A on any interval [A. 0 , oo) with A. 0 > 0 (Problem 6.1).
(c) It is clear that the given arguments involve only small values of the parameter c.
For small E one may write e-iE ~ 1- iE, hence e-iEH ~ H- iEH _ H- ic[H, iA]
and GE (A +itL) ~ { H- A-i(J-L+c[H, iA])} - 1 . Thus the replacement of R(A. iJL) _
(H- A.- iJL)- 1 by GE(A. + iJL) essentially amounts to adding to JL a term c[H, iA].
For this to be useful, it is necessary that the operator [H, iA] be positive (we assume
here that JL > 0); this increases the imaginary part of the denominator (for p, < 0 one
would add a term -E[H, iA]).
Mourre realised that the choice GE (A.+ iJ-L) == {H - A - i(J-L + c[H, iA])} - 1
allows one to obtain the existence of the limits of R( A. iJL) as f-L ~ +0 if [H, iA] is
a positive operator (in a neighbourhood A.) 4 without assuming that [H, iA] == H.
This will be discussed in the following sections.
(d) In the literature one says that a limiting absorption principle is satisfied at A. if the
limit of (f, (H- A.- iJL)- 1f) as JL ~ +0 exists for each vector fin some dense sub-
set of H.
(e) The validity of the stated results can be extended in several ways by introducing
the parameter E also for other quantities (this complicates the calculations and the dif-
ferential inequality but does not require new techniques):
(1) One can approximate H by a family {HE}O<E<Eo such that HE ~ Has E ~ 0
and such that HE has better properties relative to A than the operator H. One then
takes G E == {HE - A - i (JL E[HE, iA])} - 1 . This allows one to treat a larger class of
operators H.
(2) One can obtain the limits (f, R( A. iO) f) for a set of vectors f that is larger than
V( A) by approximating a vector f in this larger set in an appropriate way by a family
{fE }o<E:::;Eo, with fEE D(A).
We shall not take into account the possibility (1) but introduce (2), because this has
itnplications for scattering theory.

4
The precise definition will be given in §6.2.2. In the exan1ple treated here, the operator [H, iA] = H is
positive on a subspace, the spectral subspace EH ( (0, oo) )H. Indeed we have been able to control the limit
of R(A + it-L) as J-L---+ +0 for the spectral values A> 0. On EH(( -oo, 0))1t, the operator [H, iA] = H
is negative, and for the value~ A < 0 we have obtained the existence of the limit of R( A+ it-L) for 1-L - t -0.
A SIMPLE EXAMPLE 269

6 . 1. 3 . We introduce here the dilation gro!!:_p in the Hilbert space £ 2 (JRn) and its in-
finitesimal ge~rator which we denote by D. For numerous Schrodinger Hamiltonians
this operator D can assume the role played by the operator A in §6.1.2, and the dila-
tion group will be an important tool in settling certain technical points in this chapter.
For each B E ffi. we define an operator U e in £ 2 (JRn) by

(6.16)

It is easy to check (Problem 5.4) that the family {Ue}eEJR forms a strongly continu-
ous unitary group of operators (an evolution group in the terminology of Chapter 5).
The action of U e in momentum space is also easy to calculate (using the change of
variables x ~ y == eex, dnx == e-nedny):

[FUef](f)- (2n)-n/2 { e-zk·Xen&/2f(eeX)dnx


}JRn
= ( 2 n) -nf2e-ne /2 { e-ie-e k-Yj(iJ) dny = e-ne /2 ]( e-e f). (6.l?)
JJR11
To determine the infinitesimal generator D of the group {Ue }eEJR, we observe that
S(JRn) is invariant under Ue: f E S(JR_:) ===? Uef E S(JRn ). It is rather evident
that S(JRn) belongs to the domain of D [for each fixed the expression (6.16) is x
differentiable with respect to e, and the derivative still belongs to S(JRn) and th~s is
square-integrable]. By Nelson's Criterion (Proposition 5.3), the restriction of D to
S(JRn) is essentially self-adjoint. So let f E S(JRn ). Then 5

Hence
n

[15f] (X) = i! [Uef] (X) &=o i; f(X) iI>J [:.~ J(X)


I =
j=l
. n

= [(t;J- LQJPi)f](X) .
.1=1

By writing
1 1 1
QJPj - 2QjfJ.J + 2JJ.7QJ -
2

one finds that, for f E S (IR n):


n

[l5J](X) = -~[L(QJPJ +PJQ.J)f](X).


j=l

5 A s01newhat more careful way of arriving at the expression for Df, for f E S(JR'n ), would involve the
use of the Dominated Convergence Theore1n in analogy with what was done in the derivation of (5.1 0).
270 THE CONJUGATE OPERATOR METHOD

/"'-.

Thus: the infinitesimal generator D of the dilation group is the closure of the following
s
operator on (JR n):
n
1
-~ 2)QJPJ
-+ -+
-(Q .p
2
j=l
1 -+ -+ • 1 -+ -+ •
== --(2Q · P- zni) == --(2P · Q + zni). (6.18)
2 2

For later applications it is useful to defineD == - D/2, hence formally


1 -+ -+ -+ -+ 1 -+ -+ • 1 -+ -+ •
D==-(Q·P P · Q) == -(2Q · P- zni)- -(2P · Q + zni). (6.19)
4 4 4

Let H 0 - P2 . Since [Qj ,Pk] == ibjkl [on S(lRn )], one has [Qj ,P 2 ] == 2iPj, hence
[for example on S(JRn )];

Thus the example treated in §6.1.2 covers the case H == P2 (take A == D).
A similar calculation gives [1/Q 2 , iD] == 1/Q 2 . So, if H == P2 + 1 jQ 2 (with
1 E JR), then [H, iD] == H. However, here one should be careful if one wants to apply
the results of §6.1.2, since for example if n == 3 and 1 < 3/4, one has to impose
boundary conditions at the co-ordinate origin to define H as a self-adjoint operator
(see Proposition 3.30), and the domain defined by these boundary conditions is not
necessarily invariant under U e.
It is clear from Eq. (6.17) that the operators Ue map V(Ho) into V(Ho) [apply
twice Eq. (6.17)]:

[FUeHoD-ef](k)- e-nef 2[FHoU-ef](e-ek)- e-ne; 2e- 2ek 2[FU_ef](e-ek)


== e-nef2e-2ef2enef2](k) == e-2ef2](k),

hence
(6.21)

By differentiating this equation one obtains again Eq. (6.20) (recall that Ue == e-iBe ==
e2iDe ):

and

In the next chapter it will be useful to know how the operator D acts in the spectral
representation of H 0 . Since in momentum space the operators Ue correspond to radial
THE METHOD OF DIFFERENTIAL INEQUALITIES 271

dilations, it is rather evident that their generator D 1nust be a first order differential
operator in the variable A. To find its explicit form, observe that [F f · Qf) (k) ==
ik. \7 ](k) == iko ](k) I ok, where k == lkl. Now, with A == k 2 ' we have k (d/ dk) ==
2A(d/dA), and by (5.68): ](~w) == 2 112 A-(n- 2 )1 4 (U 0 f)A· Thus

(UoP · Qf)A(w) == 2- 1 / 2 A(n- 2 )/ 4 2iA~ j(~w)


8A
=-in~ (U 0 j)>..(W) + 2iA
2
:A (Uof)>.(W). (6.22)

It follows that
1 --7 --7 1 d
(UoD f) A == 4 [2(UoP · Qf)A in(Uof)A] == 4 [2i(Uof)A + 4iA dA (Uof)A].
This shows that, in the spectral representation of H 0 , D is the following differential
operator:
d
dA A). (6.23)

6.2 The method of differential inequalities

In this and the next section we develop the essential technical aspects of the Mourre
method. The proof of Theorem 6.3 contains the principal estimates leading to the
existence of the boundary values of (f, (H A iJ-L) 1 f) as J-L ---+ +0 for suitable
vectors f, under the assumption of positivity of the commutator [H, iA] between the
self-adjoint operator H under consideration and an auxiliary self-adjoint operator A.
Implications for the spectral properties of H are presented in Theorem 6.4 and in
Corollary 6.9.

6.2 . 1. We first discuss the differential inequality that one encounters the general set-
ting indicated in Comment (c) on page 268. It is similar to that given in the preceding
section but involves two additional terms on the right-hand side. Also it is useful to
admit operator-valued functions rather than just numerical functions. We let D and
Do be as in §6.1.1 and consider a norm-continuous function F: D ---+ B(H) that is
continuously differentiable with respect to the first variable c on (0, co) and satisfies
the following differential inequality on Do:

(6.24)

where the functions 19k: (0, co] ---+ [0, 00) satisfy ek :== foco19k(c)dc < 00 (k == 1, 2)
and where r > 0 is a constant. The aim is again to prove that F (0, p) converges
as fL -+ +0. We show below that the differential inequality (6.24) also leads to the
estimate (6.4 ), namely that
(6.25)
272 THE CONJUGATE OPERATOR METHOD

for some function fJ with foe: a fJ( c )de < oo. Once this estimate is established, the
existence of the limit u -lim fi-++O F (0, J-L) is obtained as in §6.1.1 [see the reasoning
after (6.4 ), in which I · I should be replaced by I · II]. For the applications it will be
ilnportant to have an explicit upper bound for IIF(c,f-L)II in terms of 81,82,/ and
cp(co), where cp(c) is again defined as cp(c) == sup 0 :;1-l:; 1IIF(c, f-L)II· Hence we must
carefully treat these constants in the derivation of the inequality (6.25).
REMARK. is important that the coefficients of the powers of II F (c, J-L) II in (6.24)
1
belong to L ( ( 0, E 0 )) (as functions of c) and also that they do not depend on the
variable f-L· The last property could be weakened, but this is not necessary for our
applications. Also one could replace the constant 1 in (6.24) by a function fJ 3 of finite
integral and require continuity of only on the subset Do of D.

PROOF OF (6.25). The inequality (6.24) implies that one has on D 0 :

(6.26)

We set /"\;1 == cp( co) + 81 and

We notice that 1/J(c, p) > 0 and that 1/;(co, J-L) == /"\; 1 for all f-L· We may now write (6.26)
in the short form
(6.27)
Finally we set TJ(c, J-L) == 1/J(c, J-L)e'(c:-c:o). Denoting again by a prime the differentia-
tion with respect to c and omitting the arguments c and f-L, we have for 0 < c < c0 :

77' == (1/Y' !1/J)e'(c:-c:a) > (1// riiFII)e'(c:-c:o)


== ( _rB2IIFII 112 - rll II+,,, ll)e'(c:-c:o) -fJ211FII 112 e'(c:-c:o)

follows that 6
! r1(c, 11) 1
/
2> -~192(c) > -172(c). (6.28)

Consequently, for 0 < c < co and all fL E [0, 1]:

6 Since 77( c, 11) > 0, division by 17( c, 11) 1 12 causes no problem.


THE METHOD OF DIFFERENTIAL INEQUALITIES 273

This leads to the following estimate for II F (E, f-L) \I:

(6.29)

This bound valid on D, it is independent of E and f-L. Insertion of this estimate into
(6.24) gives the required inequality (6.25), with

Remark The functions F appearing in our applications will also depend on an


additional argument A. For each A in some interval J 0 , let p(:\): D - 7 B(H) be norm
continuous and continuously differentiable with respect to E for 0 < E < Eo (with
Eo < 1). Assume that each p(:\) satisfies the differential inequality (6.24), with 79 1 , V2

and r independent of A and J-L. Then the norm limit of p(:\) (0, J-L) as J-L - 7 +0 exists
for each fixed A E Jo. Moreover, setting ¢(c):== sup:\EJo sup ME[O,l] IIF(:\) (c, J-L) \J, the
following bound for p(:\) ( E, J-L) is satisfied for all A E J 0 , E E [0, co] and J-L E [0, 1]:

with ~1 -¢(co) 81. (6.30)

6.2.2. We now turn to the question of resolvent estimates and begin with a simple pre-
liminary result which will be very useful in what follows.

Lemma 6.2. Let A be a self-adjoint operator in a Hilbert space H and C E B(H).


Assume that there exists an operator Z E B(H) such that (C*j,Ag)- \Af,Cg)-
\f, Zg) for all j, g E V(A). Then C maps V(A) into V(A); hence CA- AC is
defined on V(A) and has an extension to B(H) (given by Z, which is identical with
the closure of the operator CA- AC).

PROOF. Letj,gEV(A).Byassumptionwehave(Aj,Cg) == (f,CAg) (j,Zg)-


\f, g*), with g* == GAg Zg E H. So Cg E V(A *) V(A) and A *Cg ACg - =
g*-CAg-Zg. D

Now let H and A be self-adjoint operators in a Hilbert space H. We denote by


{ E ( ·)} the spectral measure of H. One says that a strict Mourre inequality is satisfied
on an open interval J if there exists a number a > 0 such that

E(J)[H, iA]E(J) > aE(J). (6.31)

This means that on the subspace M(J) E(J)H, the operator [H, iA] is strictly
7
positive, namely larger that a.
In order for (6.31) to make sense, it is necessary to define a scalar product, corre-
sponding to the commutator [H, iA], for vectors in the range of E(J). We first treat
the situation where His a bounded operator which we denote by T, whereas the case
7 Recall that S ~ T means that (f, Sf) ~ (f, T f) for each f E 7i [we assu1ne here that S and T
belong to B(7i)].
274 THE CONJUGATE OPERATOR METHOD

of unbounded H will be discussed further on (see page 286ff.). So letT E B(H). For
j, g E V(A) the number (j, [T, iA]g) is then well defined if one sets

(J, [T,iA]g) == i(Tj,Ag)- i(Aj,Tg).


We make the assumption that there exists an operator B E B(H) such that

(j, [T, iA ]g) == (j, Bg) Vj,gEV(A), (6.32)

and we then say that the closure of [T, iA] belongs to B(H). Indeed, by Lemma 6.2,
Eq. (6.32) implies that T maps V(A) into V(A). Then [T, iA]g == iTAg- iATg is
well defined for each g E V(A): [T, iA] defines a symmetric operator on the dense
domain V(A), and by virtue of (6.32) this symmetric operator has an extension B (its
closure) to B(H). It is clear that B is self-adjoint (closure of a bounded symmetric
operator).
We shall also need the second commutator ofT with A, more precisely the operator
[[T, iA], iA] [B, iA]. By analogy with the preceding developments, we assume that
there exists an operator Z E B(H) such that

(j, [B,iA]g) i(Bj,Ag)- i(Aj,Bg) == (f,Zg) Vj, g E V(A), (6.33)

hence that [B~ iA] has an extension Z to B(H). However we prefer to use the notation
[B, , rather than Z, for this extension [so we use the same notation [B, iA J for the
symmetric operator on V(A) defined by the left-hand side of (6.33) and for its closure
in B(H)].
The next theorem involves the operator IAiv for v > 1/2. This operator is de-
fined by the functional calculus: /Aiv == f~co /Aiv EA(dA), where {EA(-)} denotes
the spectral measure of A. Let us observe that IIAJII == IIIAIJII iff E V(A), because
each of these norms is equal to [ J~=,A?mj( dA) J
112
.

Theorem 6.3 . Let T E B(H) be self-adjoint. Assume that there exists a self-adjoint
operator A such that the closure B of [T, iA] and the closure of [B, iA] i [[T, iA], A]
belong to B(H) and such that a strict Mourre inequality is satisfied on some open in-
terval J, i.e.
E(J)BE(J) > aE(J) with a> 0, (6.34)
where {E( ·)} is the spectral measure ofT. Let v > 1/2. Then the norm limits
u-lim 1L-++o(I /AI)-u(T- A iJ-L)- 1 (1 IAI)-v exist for each A E J. In par-
1
ticular, (j, (T- A irt)- g) converges as IL----+ +0 for all j, g E V(/Aiv) and each
A E J. Moreover, if Jo is a bounded closed subinterval of J, then

sup //(I /AI)-v(T- A iJ-L)- 1 (1 /A/)-vll < 00. (6.35)


A.EJo,JL>O

REMARKS. (a) We shall see later that the limits in Theorem 6.3 exist in fact uni-
formly in A E J 0 if ~lo is any bounded closed subinterval of J.
(b) If A belongs to the spectrum ofT, then the limits of (j, (T- A iJ-L)- 1!) and
THE METHOD OF DIFFERENTIAL INEQUALITIES 275

(f, (T- A-itL )- 1f) as fL -+ +0, denoted somewhat formally by (f, (T- A i0)- 1f),
are not the same.
(c) The hypotheses of boundedness of [T, A] and [[T, A], A] are not trivial. Indeed it
is not difficult to show that, if CT(T) n J # 0 and A is bounded, then the inequality
(6.32) can be satisfied only for a < 0. 8 Thus the validity of (6.32) for some a > 0 (and
J non-disjoint from the spectrum of T) implies that A is an unbounded operator [if
CT(T) n J == 0, then E( J) == 0 and (6.32) is trivially satisfied for each a E IR~ but this
is uninteresting in the context of the Mourre method, since then (T- A)- 1 E B(H)
forAEJ].
(d) It is possible to obtain the conclusions of the theorem with the assumption of
boundedness of [B, iA] replaced by a weaker condition.

PROOF OF THEOREM 6.3. The proof is carried out in three parts: (1) definition and
properties of GE, (2) establishment of a differential inequality of the form (6.24), (3)
boundary values of the resolvent as an application of the results of §6.2.1. We shall
only treat the limit of (T- A- itL)- 1 as fL -+ +0. The limit of (T- A+ itL)- 1
(/L-+ +0) is obtained in an analogous way. We fix a closed subinterval Jo == [a, p] of
J, and we assume throughout the proof that z - A+ ilL with A E J 0 and 0 < fL < 1. In
calculations in which z is fixed, we sometimes omit the argument z.
We shall denote by Ck ( k == 1, 2, 3, ... ) various constants occurring in the proof.
Their value is of no importance for arriving at the conclusions of the theorem. What
matters is the fact that all these constants will be independent of A and fL, of the
parameter c: and of the considered vectors f, g, h. For the moment we define c 1 by

(6.36)

Finally, without loss of generality, we can assume in the proof that v < 1 (because
(I IAI)-v < (I+ IAI)- 1 if v > 1). The hypothesis that v > 1/2 is necessary in
order for I<I>~ I to be integrable at c: == 0, see (6.56).
(1) Definition and properties of GE
As pointed out in Comment (c) on page 268, we wish to define G E ( z), for 0 < c: < 1
and z == A + ifL (A E Jo), by

(6.37)

We must show that the operator (T- A- itL- ic:B) is invertible in B(H).
(i) For c: E [0, 1] we define Nj:- E B(H) by Nj:- == T- -A i(tL c:B). Clearly
II N(= II < 4c 1 . We fix a number 5 > 0 chosen sufficiently small so that the interval
[a - 5, p 5] is contained in J. We write I == E Ej_ withE == E([a - 5, p + 5])
8 Let A E a(T)nJ. By Proposition 4.20(b) there exists a sequence of vectors {fk} in E( J)7i satisfying
llfk II = 1 and II (T- A)fk II ----7 0. Assume A to be bounded. Then (6.32) in1plies that
a= alifkll 2 :S (fk, [T, iA]fk) = (fk, [T- A, iA]fk) = i((T- A)jk,Afk)- i(Afk, (T- A)fk)
:S 211 (T- A)fk llliAfk II :S 2IIA II II (T - .A)fk II 0 as k
----7 oo,
----7

hence a :S 0.
276 THE CONJUGATE OPERATOR METHOD

and Ej_ == E(IR \[a- 5, {3 + 5]). Let us observe immediately that, if A E J 0 == [a, {3]
and fL > 0 [using (4.51)]:

Consequently one has for each g E H:

11Ej_gll 2 ==II(T- A iJ-L)- 1 Ej_(N~ =f ic:B)gll 2


< I (T- A iJ-L)- 1 Ej_ 11 2 11 (N~ =f ic:B)gll 2
2
<
02
[11N;=gll 2 + E 2 c1llgll 2 ] · (6.39)

(ii) The assumption (6.34) implies that aE < EBE 9


. Thus, for each g E H:
allgll 2 == aiiEgll 2 ai1Ej_gll 2 == a(g,Eg) + a11Ej_gll 2 < (g,EBEg) ai1Ej_gll 2 .
(6.40)
By using the inequality I(f, h) (h, f) I < 11!11 2 + llhll 2 (see page 3) one finds that

(g,EBEg) == (g,Bg) - (g,BEj_g) - (g,Ej_Bg) + (Ej_g,BEj_g)


= (g Bg)- (.;a g _'!:_BEj_g)- ( _'!:_BEj_g fog) (Ej_g,BEj_g)
' 2 'Fa Fa ' 2
< (g,Bg) + [ ~ llgll 2 ~ IIBII 2 11Ej_gll 2 ] + IIBIIIIEj_gll 2
(g,Bg) + ~llgll + c1IIEj_gll ·
2 2
<

Insertion of this inequality into (6.40) leads to (using a+ c 1 < 2c1 )

Let us choose co E (0, 1) such that 4c:5cr/ 5 2 < aj 4. We find that, for 0 < c: < Eo,
A E Jo and f-L > 0:

VgEH. (6.41)

This is the basic inequality for the proof.


(iii) Let us give a first application of the basic inequality (6.41). Since 1-L > 0, this
inequality implies that

ac: + J-L 4c1E ±


2
llgll 2 < (g, (c:B + tL)g) +52 liNe gll 2 · (6.42)

9 One has E = EE(J) = E(J)E; hence, if hE 7-i:


a(h, Eh) = a(Eh, E(J)Eh) :S (Eh, E(J)BE(J)Eh) = (h, EBEh).
THE METHOD OF DIFFERENTIAL INEQUALITIES 277

Because (g, (T :\)g) isreal, one has (g, (c:B+J1)g) == ±2s(g,Nt-g) < l(g,Nt-g)l <
llgiiii1Vt-gll· Consequently (using the bound liNt- II < 4c1)
16cic: J
()2 . (6.43)

Assume that c: I 0 or p, I 0. After division by I! gil the inequality (6.43) has the form
liNt-gil > wllgll for some w > 0 and for each g E H. Hence Nt-g == 0 implies that
g == 0, i.e. the operators Nt- are invertible. Moreover R( Nt-) are closed and one has
II (Nt-)- 11 < w- (see Lemma 3.1).
1 1

Now by Proposition 2.19(b): R(Nt-)j_ == N((Nt-)*) == N(N!) == {0}. Since


R(Nt-) are closed, this implies that R(N1=) -H. Thus GE == (N;) 1 is defined on
Hand hence belongs to B(H), with

1 2
i11) II < -w - as + M[1

One sees that Gc(:X iM) E B(H) if E I 0 or if fL I 0. If one assumes in addition


that Eo is chosen so small that 16cic:o/ 5 2 < 1, the preceding estimate is simplified as
follows:
(for 0 < c: < Eo, J1 > 0). (6.44)

In the remainder of the proof it is always understood that c: < c: 0 .


(iv) It will be useful to know that, if E > 0, then Gc(:X iM) depends continuously
on A and 11· Setting z == A ijj and z0 == :\ 0 + ijjo, one has

GE (z) - GE (zo) == GE (z) [N; (zo) - N; (z)]GE (zo) == GE (z) [z- zo]GE (zo).
Consequently, by virtue of (6.44):

(6.45)

(v) Finally we deduce a useful estimate on GE. Let h E H and set g == GEh in
(6.42) (remember that 11 > 0):

c:a 2 4c1E
2IIGchll < (Gch, (c:B + !L)Gch) + 52llhll 2
-;i (Gch, (N: - Nc-)Gch)
4
~~C llhll 2 • (6.46)

Now observe that N;GE == I and (G; N:)* == N:*GE - N;GE I, hence
c;N: - I and consequently c;
(N: N; )GE == GE - Thus c;.

and (6.46) implies that


278 THE CONJUGATE OPERATOR METHOD

Thus
(6.47)

Let Y E B(H) be a self-adjoint operator satisfying IIYII < 1. By taking h - Yf in


(6.47) and setting c2 == ~ + 5- 1 jSCJa, one gets the following inequality:

(6.48)

The same inequality can be obtained for IIG;YII by setting g == G;h in (6.42) and
observing that G; == (N:) - 1 . So

(6.49)

(2) The differential inequality


(i) We first calculate the derivative G~(z) == (8/8c)Gc;(z). If X andY are invert-
ible operators in B(H), with x- 1 , y- 1 E B(H), then one has the identity x- 1 y- 1
== x- 1 (Y- X)Y- 1 . Hence
GT Gc; == (T- A- iJ-L- iTB)- 1 - (T- A iJ-L- icB)- 1
==i(T-E)(T A-iJ-L-iTB)- 1 B(T-A iJ-L-icB)- 1 . (6.50)

Since II (T- A- iJ-L- iTB)- 1 11 < 4/(aT) by (6.44), one deduces from Eq. (6.50) that
IIGT- Gc;ll -7 0 as T - 7 E (c # 0). Thus Ef--t Gc; is continuous in norm. By dividing
(6.50) byT-E and then taking the limit T - 7 E (which exists in norm), one finds that
(8/ 8E )Gc; == iGc;BGc;. Hence E f--t Gc; is of class C 1 in the norm of B(H). The same
reasoning as that leading to (6.9) gives the following result (use the formal identity
[A,X- 1 ] == x- 1 [X,A]X- 1 ):

[A.Gc;] == Gc;[T- A- iJ-L- icB,A]Gc; == Gc;[T,A]Gc;- icGc;[B,A]Gc;


- -iGc;BGc;- icGc;[B,A]Gc; == -G~- icGc;[B,A]Gc;. (6.51)

As in §6.1.2 this argument is somewhat formal, we have not considered the domain
questions. The right-hand side of (6.51) belongs to B(H) as a consequence of the
hypotheses of the theorem. The left-hand side has a meaning between vectors f and
g belonging to V(A) (i.e. (j, [A, Gc;]g) is well defined for such vectors). Thus (6.51)
has a meaning between j, g E V(A), and it is in this form that it will be used [see Eq.
(6.55)]. We indicate a more careful proof of (6.51) in §6.2.3. Here we rewrite (6.51)
in the following form [which has a meaning between j, g E V(A)]:

G~(A + iJ-L) == [Gc;(A iJ-L),A]- icGc;(A + ip,)[B,A]Gc;(A iJ-L). (6.52)

(ii) For c > 0 we set Xc; == (ciAI + I)- 1 . Clearly R(Xc;) - V(IAI) _ V(A).
Moreover IIXcll < 1, s-limc;--tO Xc; ==I and, ifO < {3 < 1:

IIIAI 13 Xcll =sup .riJ =sup (y/c:)/3 = c:~/3 sup L < c:~ 13 • (6.53)
x 2:0 EX +1 y 2:0 y +1 y 2:0 y +1
THE METHOD OF DIFFERENTIAL INEQUALITIES 279

10
We also set Z == (I+ IAI)-z/ and define

<Pc(A + itL) ==(I \AI)-v XcGc(A i{L)Xc(I IAI)-v- XcZGcZXc. (6.54)


ByusingtheLeibnizrule, the relation X~== -IAI(ciAI I)- 2 == -IAIX; and finally
(6.52), one obtains that

(_p~ == - IAIX; ZGCZXC- xczcczx;IAI + XCZG~ZXC


== -I A IX; zcczxc- xczcczx;IAI + xczccAzxc- xcAzcczxc
- icXCZGc[B,A]GcZXC. (6.55)

To estimate the norms of the individual terms in the last expression for <P ~, we write
IAIZXC as IAiv Z · IAI 1-v XC. For the first term we then obtain:
1
v~YcllliGcZXcll < Ev- IIGcZXcll~
1
IIIAIX;zcczxcll < IIXciiii1AIVZIIIIIAI
where the last inequality is obtained by using II XC II < 1, IIIAiu Z\1 < 1 and (6.53).
Similarly the norm of the second term is bounded by cv- 1 \IXcZGcll· For the third
term one obtains in the same way:
1 1
\\XcZGcAZXcll < IIXcZGcii!IIAiv ZIIIIIAI -v Xc I! < cu- \IXcZGc\\,
and the norm of the fourth term is bounded by cv- 1\IGcZXcll· So (6.55) leads to the
inequality

From (6.48) and (6.49), with Y == ZXC XCZ, we see that each of the norms
IIGcZXcll and IIXcZGcll is bounded by c2(l c- 112II<Pcll 112). So, by taking into
account the inequality II [B, A] II < c1 and remembering that v < 1, hence 1 < cu- 1,
one finally obtains the estimate
(6.56)

with c3 == 4c2 + 2c1 c~.


(3) Boundary values of the resolvent
The differential inequality (6.56) is of the form (6.24), with p(>.) (c, M) == <Pc(A +ilL),
191 (c) == C3Cv- 1' 192 (c) == C3Cv- 3 12 and r == C3 (see also Remark 6.1 ). Since v > 1/2,
f
the integrals 8k _ 0c0 19k(c)dc- are finite. So, by the results of §6.2.1, <Po(A + itL)
(I+ IAI)-v(T A- itL)- 1(I + IAI)-v converges as fL---+ +0. Furthermore
¢(co) sup sup II<Pc 0 (A+itL)II
>.EJo o::;p,::;1
== sup sup II (I \AI) u X co Gco (A+ i{L )Xco (I+ IAI)-u II
AE.lo o::;p,::;1
<sup sup IIGc 0 (A+'iM)il < ~.
>. E J o o::; p,::; 1 ac:: o
10 If v = 1 the arguments leading to (6.56) can be somewhat simplified: in this case one 1nay set Xc = I
and use IIAZII = IIIAIZII ::; 1.
280 THE CONJUGATE OPERATOR METHOD

So (6.30) shows that IIF(.\)(c,fL)II II<I>c:(A it-£)11 < c4 for some constant c4
which we do not write down explicitly [we recall that A E J 0 , 0 < tL < 1 and
E E [0, co)]. Since one has II(I IAI)-v(T- A- itL)- 1 (I + IAI)-vll < 1 if IMI > 1,
the proof of (6.35) is complete. The convergence of (f, (T- A- itL)- 1 g) as tL --7 +0,
for f, g E V(IAiv), is now obtained by writing f- (I+ IAI)-v }, g == (I IAI)-v g
withj,gEH. D
Theorem 6.4 . Assume that the hypotheses ojTheore1n 6.3 are satisfied. Then
(a) The spectrum ofT in J is purely absolutely continuous. In other terms as (T) n J ==
0, or equivalently E(J)H C Hac(T).
(b) The convergence of(I IAI)-v(T- A itL)(I + IAI)-v as tL --7 +0 is uniform in
A on each bounded closed subinterval J 0 of J, and the limit defines a norm continuous
function of A: if one sets X(A) == u -limM-t+O(I IAI)-v (T- A- itL)- 1 (I IAI)-v,
then IIX(A)- X(Ao) II --7 0 as A --7 Ao (with A, Ao E J).
PROOF. (a) In view of Theorem 6.3, the result of (a) follows from Proposition 4.3l(b)
by taking V- V(IAiv) in that proposition; the required condition of uniformity in A
of lirn 11 ---t+O (f, (T - A if];) - 1 f) is satisfied by part (b) of the present theorem.
(b) Let J 0 be a bounded closed subinterval of J. We let Eo > 0 be as in the preced-
ing proof and use some estimates obtained there, valid for 0 < E < Eo. Let A, Ao E J 0
and fL, tLo E [0, 1]. We have

/I <I> o(A + i tL) - <I> o(Ao + i tLo ) II < II <I> o(A i tL) - <I> c: (A + i tL) II
II<I> c: (A + i tL) - <I> c: ( Ao + i fLo) II + II<I> o(Ao i {Lo) - <I> c: ( Ao + i tLo) II· (6. 57)
Now
II<I>o(A if.L) <I>c(A i~L)II <lac II~ <I>p(A + i~L)II dp.
Since II<I>c:(A + itL)II < c4 , we see from (6.56) that the integrand is majorised by an
integrable function which does not depend on A and M· Thus, given 6 > 0, we can
choose a number E > 0 such that the first and the third term on the right-hand side of
(6.57) are less than 6/3 for all A, Ao E Jo and all fL, tLo E [0, 1]. For the second term
we have by (6.45):

For A == Ao this is less than 6/3 if fL, tLo < a 2 c 2 6/48. Then the expression on the left-
hand side of (6.57) is less than 6 for all t-£, tLo satisfying the preceding condition and
for all A == Ao E Jo, which proves that the convergence of <I>o (A i tL) to <I>o (A iO)
is uniform in A on Jo.
Next, taking tL == /Lo == 0 in (6.57), we see that its left-hand side is less than 6 if
A- Ao < a 2 c2 6/48, which establishes the norm continuity of X(A) <I>o(A + iO)
on J0 , hence on J since each A E J belongs to some bounded closed subinterval J 0 of
the interval J. D
THE METHOD OF DIFFERENTIAL INEQUALITIES 281

6. 2 . 3 . We now present an extension of Lemma 6.2 and some applications of this result.

Proposition 6. 5.. Let A be a self-adjoint operator in a Hilbert space H and let C E


B(H). ForT E JR, set C(r) - eiATce-iAT. Assume that there exists an operator
Z E B(H) such that (C*j,Ag)- (Aj,Cg) (j,Zg) for all vectors j,g E V(A).
1
Then T f--t C(r) is strongly of class C , and (d/dr)C(r)IT=O == -iZ. lffurtherlnore
Cis invertible with c- 1 E B(H), then

(cy- 1 *f,Ag)- (Af, c- 1 g) == -(J, c- 1zc- 1 g) \lj,gEV(A), (6.58)

and c- 1 maps V(A) into V(A).

REMARK. Formally these results are evident. For example:

j__C(r)j - !!:_eiATce-iATI - eiAT[iAC- iCA]e-iATI


dT T=O dT T=O T=O
- i[A, C] == -iZ.

The purpose of the proposition is to give a rigourous meaning to this formal calculation.
PROOF OF PROPOSITION 6.5. (i) Let g E V(A). Let us write

I _ e- iAs e iAs I )
!_ ( eiAs Ce-iAs C)g- ( eiAsc . g - - - .- Cg.
c zc: zc:
Since g and Cg belong to V(A) (see Lemma 6.2), the right-hand side converges
strongly as c: ---+ 0, the limit being GAg ACg. Thus

. d ( e iAsc e -iAs
z- C)gls=O- GAg- ACg == Zg forgE V(A)
dE
(the last identity follows also from Lemma 6.2). Now observe that

~ [eiA(r+c:)ce-iA( r+c:) - ( eiAT ce-iAT] g = eiAT [~ ( eiAc:ce-iAc: C)] e-"Arg.


(6.59)
From Proposition 5.1 we have e-iATg E V(A). By the preceding result the right-hand
side of (6.59) is strongly convergent as c: ---+ 0, and

Thus we have for all g E V(A) and all C5 E JR:

eiAa C e -iAag (6.60)

Since all operators occurring in this identity belong to B(H) and since V(A) is dense
in H, Eq. (6.60) can be extended to each g E H [approximate g by a sequence {gk} E
V(A), write (6.60) for gk and then take the limit k ---+ oo]. Now the right-hand side of
282 THE CONJUGATE OPERATOR METHOD

(6.60) is strongly differentiable with respect to a (for each g E 7-i), consequently this
is also the case for the left-hand side, and

(ii) Now assume in addition that cis invertible with c- 1 E B(H). Set c- 1(T) ==
eiATc- 1e-iAT. Then c- 1(T)C( T) - I, and the following identity is evident:

(6.61)

Now s-limT--tO c- 1(7) == c- 1 and s-limT--tO T- 1[C(T)- CJ == -iZ. Thus (6.61)


implies that c- 1(T) is strongly differentiable at T == 0, and

(6.62)

Eq. (6.58) follows easily by calculating the derivative, using the Leibniz rule: iff and
g belong to V(A), then

!i_(fC-1(7)
dT ' g
)I T=O
== !i_(e-iATf c-1e-iAT
dT ' g
)I T=O

== i(Af, c- 1g)- i(f, c- 1Ag) == i(Af, c- 1g)- i(C- 1*j, Ag).


The fact that c- 1 maps V(A) into itself follows from Lemma 6.2. D

As a first application of Proposition 6.5, we give a rigourous derivation of the for-


mula (6.51). We take c - N;, hence c- 1 - GE GE(A iJ-L). To obtain (6.51)
from Proposition 6.5, one has to know that

((T- A- if-L icB)*f,Ag)- (Af, (T A- itL icB)g) == (f,Zg)

for some operator Z E B(H) and all f, g E V(A). The terms involving the constants A
and fL on the left-hand side cancel, whereas for T and B this condition is satisfied as
a consequence of the hypotheses (6.32) and (6.33) of Theorem 6.3. We thus have by
Eq. (6.58):

\lf,gEV(A),
where

Z -- z.!i_
dT [e iATN-
E e
-iATJ I _ -- z.!i_
dT [e iATT e -iAT
T- 0

The derivatives can be obtained by using Proposition 6.5 (with C == T and with
C ==B):

dT e iATr e -iAT IT_- 0


z. !i_ __ z·n ,
- (6.63)
THE MOURRE INEQUALITY 283

Thus one has for all f, g E D(A):

which is identical with Eq. (6.51). Proposition 6.5 also implies that Gc maps D(A)
into itself. D
As a second application of Proposition 6.5 we give a result that will be of great
importance in the next section.

Proposition 6.6 (Virial Theorem). Let T and A be self-adjoint operators, with T E


B(H), such that the closure B of [T, iA] also belongs to B(H). Let f and g be eigen-
vectors ofT with the same eigenvalue A: T f == Aj, Tg == Ag. Then (f,Bg) - 0.

PROOF. Formally the conclusion is clear, it suffices to open the commutator:

(f,Bg) == i(Tf,Ag)- i(Af,Tg) == iA[(f,Ag)- (Af,g)] == 0.

This argument is only formal because one does not know that f and g belong to
D(A). A rigourous argument is as follows. We set T(T) == eiArre-iAr. Then T(T)
is strongly of class C 1 by Proposition 6.5, and (d/dT)T(T)Ir=O ==-B. Writing f as
f == s-lim r-+O eiAr f, we obtain by virtue of Proposition 1.1 (b):

(f,Bg) ==-lim l_(f, {T(T)- T}g) ==-lim l_(eiArf, {T(T)- T}g)


T-+OT T-+OT
-lim ~{(TJ,e-iArg) (eiArf,Tg)}
T-+OT

Since (f,e-iArg)- (ezArf,g), we have (f,Bg) == 0. D

6.3 The Mourre inequality

The results obtained so far were consequences of the hypothesis of the validity of a
strict Mourre inequality E(J)[H, iA]E(J) > aE(J) with a> 0, where J is an open
interval. A weaker condition, which is more easy to verify in concrete situations, is
the validity of such an inequality modulo a compact operator. More precisely, if H is a
self-adjoint operator and J an open interval, one says that a self-adjoint operator A is
conjugate to H on J if there exist a number a > 0 and a self-adjoint compact operator
K such that
E(J)[H, iA]E(J) > aE(J) + K. (6.64)
This is called aMourre inequality. We have seen that a strict Mourre inequality (6.31)
implies that the part of the spectrum of H in J is absolutely continuous. We shall
show that the validity of a Mourre inequality (6.64) leads to almost the same con-
clusion: the continuous spectrum of H in J is absolutely continuous, i.e. H has no
284 THE CONJUGATE OPERATOR METHOD

singular continuous spectrum in J~ however H may have eigenvalues in J, but each


of these eigenvalues is of finite multiplicity and their number is finite. These state-
ments will be proved in §6.3.1, again in the case in which His bounded (we again use
the letter T for such an operator).

6.3.1. LetT E B(H) be a self-adjoint operator and A an operator that is conjugate to


Ton some open interval J. So A is self-adjoint and there exist a number a > 0 and a
cornpact operator K such that
E(J)BE(J) > aE(J) K, (6.65)
where B [T, iA] is assumed to belong to B (H) and { E ( ·)} denotes again the
spectral measure ofT.
Proposition 6.7. Assume that the closure B of [T, iA] belongs to B(H) and that A is
conjugate to T on an open interval J. Then the number of eigenvalues of T (if any)
in J is finite, and each o.f these eigenvalues is offinite multiplicity.
PROOF. We denote by Mp(J) the subspace of H spanned by all eigenvectors ofT
associated with the eigenvalues in J. We must show that dimMp(J) < oo. Assume
that this is not the case, i.e. that dim Mp( J) == oo. Then we can choose an infinite
orthonormal sequence {fj} of vectors in Mp( J) such that I fj I == 1 and T fj == Aj fj,
where A.7 is an eigenvalue ofT in J. The inequality (6.65) implies that
(f.J,BfjJ > aiif.711 2 (fj,KfjJ -a (fj,KfjJ·
But (fj, B f.J) == 0 by the Virial Theorem, hence
0 >a (fi,KfjJ Vj EN. (6.66)
Now I(fj, Kfj) I < llfj 1111Kf.711 == IIKfj II, and IIKfj I ----+ 0 as j ----+ oo [the sequence
{f1 } converges weakly to zero by Example 1.2, hence { K fj} converges strongly to
zero by Proposition 2.12(a) since K is a compact operator]. By taking the limit j ----+ oo
in (6.66), one obtains 0 > a, which contradicts the hypothesis that a > 0. D

Theorem 6.8. LetT E B(H) be self-adjoint. Assume that there exists a self-adjoint
operator A such that the closure B of [T,iA] and [B,A] i[[T,A],A] belong to
B(H) and such that the Mourre inequality is satisfied on an open interval J. i.e.
E(J)BE(J) > aE(J) K with a > 0, K E JC(H). (6.67)
Let v > 1/2. Then
(a) If A E J is not an eigenvalue ofT, then (I IAI)-v(T A iJ-L)- 1 (1 IAI)-z/
converge in norm as J-L----+ +0. The convergence is uniform in A E Jo on each bounded
closed subinterval Jo of J\ (J"p(T), and the limit operators depend continuously (in
nornz) on A in J\(J"p(T). Furthermore, for J 0 as above:
sup II(I IAI)-v(T- A iJ-L)- 1 (1 IAI)-vll < 00. (6.68)
A.EJo ,f-L>O

(b) If Jo is as in (a), one has E( Jo )H C Hac (T).


THE MOURRE INEQUALITY 285

PROOF. (i) We shall prove in (ii) below the following fact: If A belongs to J\ ap (T),
there exists a number b,\ > 0 such that a strict Mourre inequality is satisfied on the
interval (A - b,\, A b,\), more precisely such that

where a is the constant in (6.67). So the first convergence result stated in (a) fol-
lows from Theorem 6.3. Then, by Theorems 6.3 and 6.4, the statements involving a
bounded closed subinterval J 0 of J\ ap(T) are true on each interval J 0 of the form
[A- c,\, A c,\] if 0 < c,\ < b,\, hence also on each open interval (A- c,\, A c,\)
with c,\ E (0, b,\). For an arbitrary subinterval J 0 == [a, ;3] of J\ ap(T) these results
follow easily by observing that J 0 can be covered by a finite number of intervals of
the form (A c,\, A+ c,\) with A E [a, ;3] and c,\ as above.
(ii) We now prove (6.69). We use the following fact: If C, ME B(H) and C > 0,
then M*CM > 0 [see Lemma 4.33(b)].
Let A E J\ ap (T). We choose E > 0 such that (A- E, A E) C J. Upon multiplying
(6.65) on the left and on the right byE( (A- b, A+ b)), where 0 < b < E, one obtains
by observing that E( J)E( (A b, A b)) - E( J n (A b, A b)) == E( (A b, A b)):

E ( (A b, A + <5)) BE ( (A - b, A <5))
>aE((A-b,A+b)) E((A-b,A+b))KE((A b,A b)). (6.70)

Now s-lims-+o E((A- b, A b)) == E( {A}) == 0 (A is a point of continuity of the


spectral family {EM}). Consequently liKE( (A-b, A+b)) II ----+ 0 as b----+ 0 [Proposition
2.12(b)]. Hence there exists a number b,\ E ( 0, E) such that 11

Upon multiplying the second of these inequalities from the left and from the right by
E((A- b,\, A b,\)), one obtains

E( (A- 5>.., A 5>-.))KE((A 5>.., A 5>..)) > -~ E((A- 5>.., A 5>..)). (6.71)

By taking b- b,\ in (6.70) and taking into account (6.71), one obtains (6.69). D

The following consequences of Theorem 6.8 and Proposition 6.7 are easy to verify
(Problem 6.5):
Corollary 6.9. Under the hypotheses of Theorem 6.8, the operator T has no singular
continuous spectru1n in J, i.e. asc(T) n J- 0, the number of eigenvalues ofT in J
is finite (or zero), and each of these eigenvalues is offinite multiplicity.
11
If C == C* and IICII :s; ~'and if {E 0 (·)} denotes the spectral measure of C, one has E 0 (V) ==
E0 (V n [-~,~])(the spectrum of Cis contained in the interval [-~,~]).Hence, by setting mj(W) ==
IIE 0 (W)fll 2 :
r;,(f,f) = r;, l: rnf ( d>..) 2'. /_: Arnf ( d>..) = (!, C.f) 2'. -r;, /_: rnf ( d>..) = -r;,(f, f).
286 THE CONJUGATE OPERATOR METHOD

6.3 . 2. The principal results of Mourre theory obtained so far are valid if the operator
T is replaced by a general self-adjoint operator which will be denoted by H. The idea
of their proof is the same, but for unbounded H one has to take into account domain
properties, which lengthens the arguments. For example one has to give a meaning to
the operators [H, iA] and [[H, A], A] which are not bounded (this is seen already in the
simple case in which H is the free Hamiltonian P2 in £ 2 (JRn); a conjugate operator
to P2 is the operator D introduced in (6.19), and we have seen that [P2 , iD] == P2 , so
this commutator is unbounded). Thus one will have to impose conditions relating the
domains of H and A. For example, it is required in many approaches that the group
{ e-iAs}sEIR: should leave the domain D(H) of H invariant, i.e. that e-iAsj E D(H)
for each f E D(H) and each s E JR.
If the spectru1n of H does not cover the entire real axis, i.e. if p( H) n JR is non-
empty, one can make use of the results obtained so far by introducing the bounded self-
adjoint operator 7 1
:-H- Ao) - 1 , where Ao is a fixed real number in the resolvent set
(

p(H) of H. The mapping A ~ 1/ (A- Ao) gives a one-to-one correspondence between


the spectrum of H and that ofT, which allows one to relate the spectral properties of
H to those of T and to transform a Mourre inequality for H into a Mourre inequality
forT. One then arrives rather easily at the following theorem:
Theorem 6.10. Let H be a self-adjoint operator in a Hilbert space H such that Ao E
p(H) for some Ao E JR. Assume that there exists a self-adjoint operator A such that
the closures of [(H- Ao) 1 , iA] and [[(H- Ao)- 1 , A], A] belong to B(H) and such
that, .for a bounded open interval J:

E(~l)[H, iA]E(J) > aE(J) K with a> 0, K E JC(H), (6.72)

where { E( ·)} denotes the spectral 1neasure of H. Then all conclusions of Theorem
6.8 (with T replaced by H) are true, and the spectrum of H in J is as described in
Corollary 6.9.
In §6.4.2 we give the proof of this theorem when H is a Schrodinger operator.
Except for some domain considerations 12 , the proof of Theorem 6.10 in the general
case is exactly the same.

6.4 Application to Schrodinger operators

The 1nain purpose of this section is to apply the general theory to two-body Schro-
dinger Hamiltonians in the centre-of-mass frame with scattering-type potentials. Such
Ha1niltonians are unbounded self-adjoint operators. However they are bounded from
below and therefore covered by Theorem 6.1 0. §6.4.1 we discuss the necessary
commutators and state sufficient conditions on the potential V so that the the Mourre
method is applicable. Certain more technical aspects are presented in §6.4.2 and §6. 7 .2.

12
The necessary domain properties, all of which are consequences of the assumption that the commutator
[(H- .Ao) 1 , iA] belongs to B(H), are discussed in Theorem 6.2.10 of [ABG].
APPLICATION TO SCHRODINGER OPERATORS 287

6 . 4 . 1. In order to make use of the results of Sections 6.2 and 6.3, one has to know that
[(H ;\ 0 )- 1 ,iA] and [[(H ;\ 0 )- 1 ,A],A] belong to B(H) (in the sense specified
at the beginning of §6.2.2) for some ;\ 0 E p(H). We make a few comments about
this point, remaining on a formal level for the moment. By using (6.9) and the formal
identity [XY, Z] - [X, Z] + X[Y, Z], one sees that for any z E p(H):

[(H- z)- 1, A] - (H- z)- 1[A, H] (H- z)- 1 (6.73)


and
1
[[(H-z)-\A],A]- [(H-z) [A,H](H- z)-I,A]
~ [(H- z)-I,A][A,H](H- z) 1
(H- z)- 1 [[A,H],A](H- z)- 1
(H- z)- 1 [A,H][(H- z)- 1 ,A]. (6.74)

This shows that the conditions [(H- z)- 1 , iA] E B(H) and [[(H- z)- 1 , A], A] E
B(H) can be expressed in terms of certain conditions on the first and second commu-
tator of H with
Let us consider Hamiltonians H in L 2 (JRn) of the form H ~ H 0 + V(Q), with
H 0 - P2 . We denote by { E( ·)} the spectral measure of H, and we assume that V ( Q)
is Ho-compact, so that aess(H) - aess(Ho) ~ [0, oo) by Proposition 4.26. Thus the
spectrum of H in ( -oo, 0) is discrete and H is bounded from below, and we shall
apply the Mourre method for intervals J contained in (0, oo) to obtain details of the
nature of the essential spectrum of H. We know that a conjugate operator in the case
V - 0 is the operator D ~ - D12 - (Q · J3 J3 ·Q) I4 given in (6.19), where Dis the
infinitesimal generator of the dilation group in L 2 (JRn). We shall show that the same
operator D can serve as a conjugate operator to H if the potential V is sufficiently
well behaved. We observe that the domain of H is invariant under the dilation group;
indeed D(H)- D(H0 ) by the Rellich-Kato Theorem (Proposition 2.44), and we have
seen in §6.1.3 that D(Ho) is invariant under the dilation group.
To calculate [H, iD], we again proceed somewhat formally. Recall from (6.20) that
[H0 , iD] - H 0 [at least on S(JRn )]. It follows that

[H, iD] ~ [H0 , iD] + [V(Q), iD] ~ H0 [V(Q), iD] ~ H- V(Q) [V(Q), ~iD].

The commutator [V (Q), iD] is easily calculated by using the following expression for
D in L 2 (JR n) [see (6.19)] :

D- 1 (2Q- ·P-in!)-
- - i - -
nl).
4 4(2Q · \7
It follows that
- zD]
[V(Q) . ~ 1 - · VV)(Q)
--(Q - - ~ DV) -
--1 ( r - (Q). (6.75)
' 2 2 Dr
The last relation follows from the fact that the projection of the gradient operator in the
direction X is the radial derivative (in spherical polar co-ordinates): I:Z=1 Xk (a I axk)
r(818r).
288 THE CONJUGATE OPERATOR METHOD

Eq. (6.75), [r(oVIor)](Q) denotes the multiplication operator in L 2 (JRn) by


[r( oVI or)] (x) lxl (oVI or) (x). One must assume that the radial derivative avI ar
of is defined in some sense. The simplest situation is that where the function V
(expressed in spherical polar co-ordinates) is differentiable in the usual sense, except
possibly at the origin r - 0. Another possibility is to define x · VV in the sense of
distributions. If the function belongs to Lfoc (JRn), then f TRn V (x) ¢( x) dnx is well
defined for each ¢ E C0 (JRn), so that V defines a distribution on the set of test func-
tions C()(JRn ), and its partial derivatives oVI OXk exist as distributions. Assume that
these distributions are given by functions, i.e. that there exist functions ()k E Lfoc (JRn)
such that

We shall use the natural notation (oVI ax k) for these functions () k' and we write
[r( oVI 8r)] (x) for L~=l Xk( 8VI axk) (x). Under the stated integrability condition on
() k, the scalar products

are well defined for all j, g E C0 (JRn), and one has [in the sense of scalar products
between vectors from C0 (JRn)]

[H, iD] == H ---


V(Q) av) (Q)- -
"21 ( ra;: H- Z(Q), - (6.76)

with Z(x) - V(x) _!_


2
(r av)
ar (x). (6.77)

If the functions ()k are locally square-integrable, then the operator of multiplication
by the function r(8VIor) is defined on C0 (JRn), and Eq. (6.76) holds as an operator
equation on C 0 (JRn) (provided that Vis also locally square-integrable).
Proceeding in the same way one finds [using (6.75) for Z in place of V] that

[[H, iD],iD] == [H, iD]- [Z(Q), iD]


= H ~ Z(Q) + ~ (r ~~) (Q) ~ (r! r ~~) (Q) H ~ Z 1 (Q), (6.78)

with Z 1 (X) = V (X) ~ ~ (r ~~)(X) ~ ~ (r 2 ~:~)(X). (6.79)

We shall consider a rather simple class of (real-valued) potentials defined by the


condition (C) stated below. Further on, when discussing properties of Schrodinger op-
erators that rely on results obtained by the Mourre method, we shall usually assume
that the potentials satisfy this condition.
(C) V and r(8VI8r) belong to LP(JRn) L~(JRn), and r 2 (8 2 VI8r 2 ) belongs to
LP(JRn) L 00 (JRn ), with p == 2 if n == 1, 2, 3 and p E ( nl2, oo) if n > 4.
APPLICATION TO SCHRODINGER OPERATORS 289

If (C) is satisfied, then V ( Q), Z ( Q) and Z 1 ( Q) are defined not only on C 0 (JR n)
but on V(H0 ), hence also on V(H). The operators V(Q) and Z(Q) are H-compact
(see Propositions 2.49 and 2.47); Z 1 ( Q) is H 0 -bounded with relative bound 0, hence
also H- bounded with relative bound 0 [see Proposition 2.49 and the proof of Proposi-
tion 2.44(b)]. Hence, if z E p(H), the operators V(Q)(H- z)- 1 and Z(Q)(H- z) 1
are defined on H and are in fact compact, whereas Z 1 ( Q) (H - z) -J E B (H). In par-
ticular:
(1) ](H- z)- 1 and [[H,D],D](H- z)- 1 belong to B(H) [this is evident by
taking into account (6.76) and (6.78); see also §6.7.2 for a more careful calculation of
the commutators in (6.76) and (6.78)];
(2) if J is a finite interval, then HE(J) E B(H) and V(Q)E(J) == V(Q)(H- z)- 1 ·

(H z)E(J) E JC(H); similarly Z(Q)E(J) E JC(H) and Z 1 (Q)E(J) E B(H).


Assume now that J is a bounded interval in (E, oo) for some E > 0. Then, for each
f E H: (E(J)j,HE(J)j) == JJAmJ(dA) > cJJm~.r(dA) - s(f,E(J)f), hence
E(J)HE(J) > sE(J). By taking into account (6.76) we see that

E(J)[H, iD]E(J) - E(J)H E(J)- E(J)Z( Q)E(J) > cE( J)- (J)Z( Q)E( J).

Since E( J) Z (Q)E( J) is a compact operator, this shows that D is conjugate to H on


each bounded interval J in (0, oo) provided that 0 does not belong to the closure of
J. 13 By the extension of the Mourre method to unbounded operators (see Theorem
6.1 0), one has the following results:
(a) H has no singular continuous spectrum: o-sc(H) - 0.
(b) o-ac(H) - [0, 00 ). 14
(c) In each interval J == [a, {3] with 0 < a < {3 < oo, H has at most a finite number
of eigenvalues, each of them of finite multiplicity. 15
(d) If A E (0, oo) is not an eigenvalue ofT, then (I IAI) l/(T-A i!I)- 1 I 1)-v
converge in norm as 11 ~ +0, for any v > 1/2. The convergence is uniform in A on
each bounded closed subinterval J of (0, oo) disjoint from the eigenvalues of H, and
if J is such an interval, then

sup II(I IAI)-v(H- A i!I)- 1 (1 I I)-vii< oo. (6.80)


.\EJ, t-L>O

13
It is clear that the condition (C) is not optimal for aniving at this conclusion (we have not used at all
the first factor E( J) occurring in the Mourre inequality~ for exatnple E( J) W ( Q)E( J) 1nay be c01npact
without W(Q)E(J) being compact).
Simple examples of potentials satisfying (C) for n = 3 are the Yukawa potential [P- P r I r and the Coulmnb
potential r I r [ r E JR, fL > 0].
14 We know that O"css (H) = [0, oo ), and the assertion of (b) follow~ by taking into account (a) and (c).
15 One can show that if V satisfies the condition (C), then H has no positive eigenvalues. On the other

hand one knows spherically sym1netric bounded potentials converging to zero at infinity for which the
associated Hamiltonian has strictly positive eigenvalues (see Section 4.4 of [CFKSl and Section XIII.l3,
Volume IV of [RS]).
290 THE CONJUGATE OPERATOR METHOD

6 . 4.2 . As explained in §6.3.2, we give here a proof of Theorem 6.10 when H is a


Schrodinger Han1iltonian with a potential V satisfying the condition (C) of page 288.
The idea of the proof is to obtain the results of the theorem from the corresponding
statements for bounded operators. We fix a real point Ao in the resolvent set of (so
Ao < 0) and denote by {E(·)} the spectral measure of H. By the results of §6.4.1 the
boundedness assumptions of Theorem 6.10 are satisfied (with A - D), and one has
a Mourre estimate for H on each bounded open interval J in (0, oo) the closure of
which does not contain the point A == 0. the sequel we consider a fixed interval J
having these properties.
Since p(H) is open, there exists a number b > 0 such that the interval (A 0 -b, A0 +b)
belongs to p(H), and we can assume without loss of generality that Ao b < 0. One
then has dist(Ao, J) > b > 0.
(i) We set T- -(H Ao)- 1 == (Ao - H)- 1 . Clearly Tis a self-adjoint operator
belonging to B (H), and we denote its spectral measure by { ET (-)} [as we also use
the spectral measure { E(-)} of H]. These two spectral measures are not independent,
because Tis a function of T == cp(H), with cp(A) == (Ao - A)- 1 . We know (see
Ren1ark 4.14) that ET(V) == Xv(T) == [xv o cp] (H). Let us calculate the function
Xv 0 cp:
A _ { 1 if cp (A) E V
[x v 0 'Pl ( )- o if cp (A) rf_ v.
Let us denote by W the pre-image of V under cp, i.e. W =={A E JR I cp(A) E V}. Thus
[xv o cp](A) == xw(A). other terms ET(V) == E(W): the spectral projection ofT
for Vis identical with the spectral projection of H for W - cp- 1 (V)! In particular, if
W == J, let us set J - cp(J) == {a E JR I a- (Ao A)- 1 for some A E J}. Then
ET(J) == E(J).
(ii) One has [T, iA] == -[(H- Ao)- 1 , iA] == (H- Ao)- 1 [H, iA](H- Ao)- 1 ==
T iA]T, hence (from the Mourre inequality on J) 16 :

ET(J)[T, iA ]ET(J) == E(J)T[H, iA]T E(J)


- TE(J)[H, iA]E(J)T > aTE(J)T TKT.
~ ~

We set I< == T KT. Clearly K is a compact operator. Also there exists a number rc > 0
such that T E( J)T > rcE( J): indeed, if [a, J)] denotes the closure of the open interval
J (with 0 < a < j)) and if one sets rc == (jJ Ao) - 2 , then

(!, TE(J)Tf) = r ( 1
}J A- Ao
)2 Tnj(dA) > K r rnt(dA)- K(j, E(J)f).
}J
16
We observe that for the Schrodinger Hamiltonians considered here:

E(J)T[H,iA]TE(J) = E(J)(H- .\ 0 )- 1 [H- Z(Q)](H- .\o) 1


E(J)
= (H- .\o)- 1 E(J)[H- Z(Q)]E(J)(H- .\ 0 )- 1 =: TE(J)[H,iA]E(J)T,

since [H- Z(Q)]E(J) belongs to B(H). For a more general self-adjoint operator H the establishment of
this identity n1ay require some care (see the footnote on page 286).
APPLICATION TO SCHRODINGER OPERATORS 291

As a consequence we have

with a~ > 0 and k E JC(H): By Theorem 6.8 and Corollary 6.9, T has a~ most a finite
number of eigenvalues in J, each of finite multiplicity. Moreover, if J 1 is a closed
subinterval of J \ ap(T), then the limits of (I+ IAI)-v(T- A iJ-L)- 1 (I IAI)-v
as f-L - t +0 exist uniformly in A E ] 1 for any v > 1/2 [in fact the B(H)-valued
function <I> 0 (z) - (I+ IAI)-v(T- z)- 1 (I + IAI)-v is uniformly norm continuous
on 01 :== { z ==A iJ-L I AE 11 , 0 < f-L < 1} ]. Furthermore one has

sup II(I + IAI)-v(T- A± iJ-L)- 1 (I + IAI)-vll < 00. (6.81)


A.EJI,J-t>O

(iii) The relation E ( {A}) == ET ({ cp (A)}) shows that the eigenvectors of H and
ofT are the same. A real number A is an eigenvalue of H if and only if cp(A) is an
eigenvalue of T, with the same multiplicity [one can see explicitly that if H f == Aj,
then T f == (Ao - H)- 1 f _== (Ao - A)- 1 f == cp(A)j]. Since T has at most a finite
number of eigenvalues in J, each of them of finite multiplicity, it follows that H has
at most a finite number of eigenvalues in J, all of them of finite multiplicity.
(iv) Now let J 0 be a closed subinterval of J\ap(H). Define io by ] 0 == cp(J0 ). ] 0
is a closed subinterval of]\ ap (T). Observe that one has the following identity (it is
evident if one replaces H by A):
1
(H- z)- 1 == (Ao- z) 1
[(Ao H)- 1 - (Ao- z)- 1 ] (Ao H)- 1
1
== ( Ao z) - 1 [ ( T - (Ao - z) - 1 J- T.

Consequently

(I+ IAI)-v (H- A- iJ-L)- 1 (I+ IAI)-v == (Ao- A- iJ-L)- 1


1
x (I + IA I) - v [T - ( Ao - A - i f-L) - 1 J- (I + IA I) - v x (I IA I)vT (I + IA I)- v .
(6.82)

Consider the set 0 == {z - A iJ-L I A E Jo, 0 < fL < J-Lo}, where f-Lo E (0, 1]
is fixed. The mapping z ~ (Ao - z) - 1_ maps the int~rval J 0 onto ] 0 and 0 in a
continuous manner onto a complex set 0 containing J 0 ; if f-Lo is taken to be small
enough, 6 is contained in the rectangle { z == a iE I 0 < E < 1, a E ] 1 } , where ] 1
is a closed subinterval of J disjoint from a P (T) and containing ] 0 in its interior. As
f-L - t +0, (Ao -A- iJ-L)- 1 converges [uniformly in A E J] to (Ao - A)- 1 . So, since
(I+ IAI)-v(T- z)- 1 (I IAI)-v is (uniformly) continuous on the set {z- a+ iE I
0 < E < 1, a E ] 1 } (assuming again that v > 1/2), we find from Eq. (6.82) that
(I+ IAI)-v(H- A- iJ-L)- 1 (I + IAI)-v has limits in norm as tL - t +0, uniformly
in A E Jo, provided that one knows that (I IAI)vT(I IAI)-v belongs to B(H). If
v == 1 we know that this condition is satisfied, because T _ - (H Ao) - 1 maps V (A)
292 THE CONJUGATE OPERATOR METHOD

into V(A) under the hypotheses of the theorem (see Lemma 6.2) 17 . If v < 1 one can
show by an interpolation method that this is also the case [see the Appendix to this
chapter, in particular Example 6.31].
(v) The results obtained so far imply the spectral properties of H (the validity of
Corollary 6.9 for H). The bound (6.80) for H follows from (6.81) and (6.82). With
this the proof of Theorem 6.10 is complete.

6.5 Relatively smooth operators

The concept of relatively smooth operators with respect to a self-adjoint operator H


fits naturally into the context of the present chapter, since it can be defined in terms of
the resolvent of H close to the real axis. In §6.5 .1 we give the definition and some re-
sults that are important for our subsequent considerations, in particular a theorem that
will permit us in Chapter 7 to prove asymptotic completeness in scattering theory.
§6.5 .2 we indicate a class of smooth operators relative to the Hamiltonians that were
considered in §6.4.1.

6.5.1. Let H be a self-adjoint operator in a Hilbert space H and { E( ·)} its spectral
measure. Let C be a closed operator in H. One says that Cis smooth relative to H (or
simply H-smooth) if V( C) ~ V(H) and if there exists a constant c 0 E (0, oo) such
that for each fL E (0. 1) and each f E H:

(6.83)

We immediately give a result whose relevance for scattering theory is rather evident
(see Chapter 7 for details):

Proposition 6.11. Let H be self-adjoint and let C be a closed operator which is H-


s1nooth, i.e. satisfying (6.83). Then one has

VjEV(H). (6.84)

PROOF. We set R(z)- (H-z) 1 , and we shall always assutne that f E V(H). If
z E p(H), the operator C(H- z)- 1 - CR(z) is defined on each vector g E Hand
is closed, hence it belongs to B(H) by virtue of the Closed Graph Theorem (see page
17
II(J + IAI)vT(I + IAI)-v fll ~ cllfll with c < oo. Indeed:
Observe that, for v = 1, one does have

II + IAI)T(I + IAI)- 1 !II ~ IIT(J + IAI)- 1 !II+ IIAT(I + IAI)- 1 !II


~ II T (I + IA I) - f II + II T A (I + IA I) f II + II [A' T]( I + IA I) f II
1 1 1

~{liT(!+ IAI)- 11 + IITIIIIA(J + IAI)


1 1
+ II[A, T]ll}llfll = cllfll·
11

For v = 0 one has (I+ IAI)vT(I + IAI)-v = T E !3(7-l), and for 0 < v < 1 one must interpolate
between these two cases, as explained in the Appendix.
RELATIVELY SMOOTH OPERATORS 293

68; in our applications it is not necessary to invoke this theorem, because the operators
C that will be considered are H -bounded as a consequence of results proved in earlier
chapters). We then have IICe-iHtfll < IICR(i)IIII(H- i)fll < oo iff E V(H), i.e.
the integrand in (6.84) is bounded on JR. As a consequence, if J-L > 0, then the function
rJ(t) :== e-p,ltiiiCe-iHt !II belongs to L 2 (IR), in other terms the vector-valued function
t r-----7 e-p,ltl ce-iHtj belongs to £ 2 (IR; H) (see page 20 for this type of spaces).
By Eqs. (5.7) and (5.8) one has for J-L > 0:

R(A + i fL) - R(A - ilL) i .L: ei>.t e- 0ltl e -iHt dt. (6.85)

So A r-----7 C[R(A + iJ-L) - R(A- iJ-L)]f is the inverse Fourier transform of the vector-
valued function t r-----7 iyl21[-e-p,ltice-iHtf. By using the unitarity of the Fourier trans-
formation in £ 2 (IR; H) (i.e. for vector-valued functions defined on IR), one obtains

2~ I:IIC[R(A+if-L) ~~ 11!11 2 -
2
I :e- 20 1tii1Ce-iHtfll 2 dt = R(A-i!L)]fll dA <

In particular one has for each M E N and each IL > 0:

.L: e- 20 1t111Ce-iHtfll 2 dt < ~~11!11 2 ·


This inequality is still true for J-L == 0 (use for example the Dominated Convergence
Theorem by taking into account the fact that IICe-iHt fll 2 is bounded as a function of
t, hence in £ 1 ([- M, M]), or a more explicit argument):

As M - 7 oo, the left-hand side (a non-decreasing and bounded sequence of real


numbers) converges, the limit being J~ IICe-iHtfll 2 dt, and this limit is still bounded
by the same number (27r)- 1 collfll 2 . D

We next introduce a local version of the concept of relatively smooth operators. Let
again C be a closed operator with V(C) =:) V(H), and let J be an interval (or more
generally a Borel subset of IR). One says that isH-smooth on 1 if CE(J) isH-
smooth in the sense used before, i.e. if there exists a constant c 0 E (0, oo) such that
for each J-L E (0, 1) and each f E H:

.L: II C[(H- A- if-L)- 1 - (H- A+ i!L)-


1
]E(J)fll
2
d)..< coll.fll 2 - (6.86)

Proposition 6.12. Let H be self-adjoint and let C be a closed operator with V( C) =:)
V(H). Let J be a closed interval such that
sup IIC(H- A- i{L)- 1 C*II < oo. (6.87)
>..EJ, O<p,<l

Then (6.86) is satisfied with co== 20Jrsup>..EJ,O<p<IIIC(H- A- iJ-L)- 1 C*Ij, and C


isH-smooth on J.
294 THE CONJUGATE OPERATOR METHOD

REMARK. The operator C(H- A-it-L)- 1 C* is well defined on a dense domain: indeed
V( C*) is dense [see the Comment (iii) on page 52; in our applications C will even be
self-adjoint], and C(H- A- it-L)- 1 belongs to B(H). Thus in condition (6.87) it is
assumed in particular that the closure of C(H- A- iJ-L)- 1 C* belongs to B(H).
PROOF OF PROPOSITION 6.12. Let again R(z) == (H-z) - 1 . We assume throughout
the proof that 1-L > 0. We can then use the following identity [take into account (4.42)]:

(6.88)

By the first resolvent equation (2.93) we have

hence [taking into account (6.88) in the last step]

1: liCE(.!) [R(A + ip,) ~ R(A ~ ip,)] Jll 2 dA


2
- 4p, l:IICR(A + ip,)E(.J)R(A ~ ip,)fll 2 dA
< 4p,
2
~~~ IICR(o- + ip,)E(.J)II 2 l:IIR(A- ip,)fll 2 dA
== 47rJ-LSUp IICR(o- + i;L)E(J)II 2 IIJII 2 . (6.89)
o-EIR

To estimate the supremum occurring in this formula, we consider separately the case
in which o- E J and that in which o- tJ_ J.
(i) If o- E J, we write (using IIT*TII == IITII 2 == IIT*II 2 )
2
IICR(o- + iJ-L)E(J)II < IICR(o- + iJ-L)II 2 <sup IICR(A + it-L)II 2
A.EJ

-sup II [CR(A + iJ-L)][CR(A + iJ-L)]*II -sup IICR(A + it-L)R(A iJ-L)C*II


A.EJ A.EJ

-sup -
A.EJ
~
2Zf-L
IIIC[R(A + it-L)- R(A iJ-L)] C*ll
1

< -2 sup [IICR(A + ip,)C*II + IICR(A ~ ip,)C*II]


1
/1 A.EJ

-!_sup IICR(A + iJ-L)C*II, (6.90)


/1 A.EJ

because IICR(A- iJ-L)C*II == II [CR(A- iJ-L)C*]*II - IICR(A + iJ-L)C*II·


RELATIVELY SMOOTH OPERATORS 295

(ii) If a-t/:. J, let a-o E J be such that dist( a-, J) == Ia-0 - a- I (i.e. a-0 is the endpoint of
J on the side of a-). One has

IICR(a- + iM)E(J)II < IICR(a-o + iM)IIII(H a-o- 'itL)R(a- + i{L)E(J)II


== IICR(a-o + iM)IIII[I +(a-- a-o)R(a- + iM)]E(J)II·

As a-o E J, IICR(lJo + iM)II 2 is majorised by (6.90). On the other hand

II +(a-- a-o)R(a- + iM)]E(J)II ==sup


>..EJ
[1 + IAI(J- (Jol
- CY - ZJ-L 1
] < 2.
Sum1ning up we have
2
sup IICR(a- + iM)E(J)II
o-EIR
2 2
<sup JJCR(a- + iM)E(J)II +sup IICR(o- + i{L)E(J)II
o-EJ o-~J

< _!_sup IICR(A + itt)C*II (1 + 22 ) =~sup IICR(A + ip.)C*II·


/-L >..EJ /-L >..EJ

Insertion into (6.89) gives (note that the factor 201r is not optimal):

j oo
-CX)
IIC E(.J) [R(A + iJL) - R(A- itt)] f 11
2
dA < 201f sup IICR(A + itt)C* llllf 1 2 ·
AEJ

By taking into account the hypothesis (6.87), one obtains the validity of (6.86). D

6 . 5 . 2 . Proposition 6.12, combined with Theorem 6.1 0, gives the following result:
Proposition 6. 13 . If H and A are two self-adjoint operators satisfying the hypotheses
of Theorem 6.10 on a bounded open interval J, then for each v > 1/2 the operator
(I+ IAI)-v isH-smooth on each bounded closed subinterval J 0 of J\o-p(H).

If H == H 0 + V ( Q) is a Hamiltonian of the class considered in §6.4.1, we find


that (I+ IDI)-v is H-smooth on each interval [a, /3] c (0, oo) \ a-p(H). This can
be used to show that certain functions of the position operator are also H- smooth
[for example (I+ IQI)-v if v > 1/2], a fact that will be very useful in scattering
theory. Again the case v == 1 is simple, whereas for v < 1 one needs an interpolation
result. In the following lemma we present the general idea permitting the replacement
of (I+ IAI)-v by another operator C.
296 THE CONJUGATE OPERATOR METHOD

Lemma 6 . 14. Let be a lower semibounded self-adjoint operator. Let v > 1/2
and let J be a bounded interval. Let A be a self-adjoint operator and let C1, C2 be
closed operators such that, .for some real number ( with -( < inf a-( H), one has
Ck(H + ()- 112 E B(H) 18 and such that the closures ofCk(H + ()- 1(1 + [AI)v
belong to B(H) (k == 1, 2).
(a) Assu1ne that
sup II(I + IAI)-u(H- A± iM)- 1 (1 + IAI)-vll < oo. (6.91)
>..EJ,O<p,<1

Then one has


sup IIC1(H- A iM)- 1C211 < oo. (6.92)
>..EJ,O<~-t<1

In particular the operators ck are H -Slnooth on J.


(b) Assume that u-limf-t-7+o(I + jAj)-v(H- A- itL)- 1(I + IAI)-v exists for each
A E J, the convergence being unifonn in A. Then u-lim~-t---7+0 C1 (H- A- iM)- 1 C~
exists for each A E J; the convergence is uniform in A and the limit operators depend
continuously on A (in norm).

PROOF. (a) One has-( E p(H) and H + ( H- (-() > 0. Hence (H + ()- 112 is
a well defined bounded and positive operator. By iterating the first resolvent equation
(2.93) one obtains, for z E p(H):

- z)- 1 - (H + () 1 + (z + ()(H + ()- 1(H- z)- 1


== (H + ()- 1 + (z + ()(H + ()- 2 + (z + () 2(H + ()- 1(H- z)- 1 (H + ()- 1.

C1 (H- A iM)- 1C2 == C1 (H + ()- 1 C~ +(A =f iM + ()C1 (H + ()- 2 C~


+(A iM+() 2C1(H+()- 1(H-A iM)- 1(H+()- 1C2. (6.93)

Now C 1(H + ()- 1C2 == C 1(H + () - 112 · [C2(H + ()- 112]* E B(H), and similarly
C 1(H + () - 2C2 == C 1 (H ()- 112 · (H + () - 1 · [C2(H + ()- 112]* E B(H). Conse-
quently insertion of the first two terms of (6.93) into (6.92) gives a finite contribution
to the supremum, because J is bounded so that jA =f iM + ([ < c < oo for A E J and
all IMI < 1. Thus it suffices to show that

sup IICI(H+()- 1(H A iM)- 1(H+()- 1C211<oo.


AEJ.O<tL<1

us write

C1 (H + ()- 1 (H- A± itL) 1


(H + ()- 1 C2
== C1(H + () 1
(I + IAI)l/ X (I+ IAI)-l/(H- A itL)- 1 (I + IAI)-1/
x (I+ IAI)l/ (H + ()- 1 c:;. (6.94)
18It suffices to know that V(Ck) :2 V(JHJ 1 12 ) [by the Closed Graph Theorem]. One has inf o-(H) =
inf{,\ I A E o-(H)} > -oo.
RELATIVELY SMOOTH OPERATORS 297

The first and the third factor on the right-hand side are bounded by assumption [the
third one is the adjoint of C 2 (H ()- 1 (I+ IAI)u because (is real]. Hence

sup I\C1 (H + ()- 1(H- A itL)-


1
(H ()- 1C2\\
>..EJ,O<~-t<l

< I\Cl(H ()- 1(I + IAI)l/IIIIC2(H + ()- 1(I + IAI)l/11


x sup 1\(I + !A\)-u(H- A itL)- 1 (I + \A\)-u\1 < oo.
>..EJ,~-t>O

(b) This is easily seen by taking into account Eqs. (6.93) and (6.94). D

Remark 6 . 15 . The preceding proof shows that one has the following inequality:

(H- A 1C211 < c!IC1(H + ()- 112 IIIIC2(H + ()- 112 11


iM)-

+ c2\\Cl(H + ()- 1(I + IAI)l/IIIIC2(H ()- 1(I + IAir/11


x sup \I(I + IAI)-u(H- A± ift)- 1(I + IA\)-u\1, (6.95)
>..EJ,~-t>O

where cis a constant depending only on J and(.

We now apply the preceding lemma to the case in which H == H 0 + V(Q), where
the function V satisfies the condition (C) of page 288. We shall take C 1 == C 2 ==
W (Q), where W is a function defined on ffi.n, and A == D == (2Q · P - in I) I 4.

Proposition 6 . 16. Let H- H 0 + V(Q), where V satisfies condition (C) of page 288.
(a) Let W: ffi.n - 7 C be a function of the form W(x) == [1v1 (x) + 1v2 (x)] (1 + lxl)-u
with w1 E L 00 (IR.n) and w2 E L P (IR.n) for some p satisfying p E ( n, oo) and p > 2, and
with v > 1 I 2. Then liV (Q) is H -smooth on each bounded closed interval J == [a, {3]
in (0, oo) \ o-p(H). Moreover, for each such interval there exists a constant CJ such
that for all f E H:

I:IIW(Q) [(H-).- i!J)- 1 - (H-).+ i!J) 1


]E(J)fll 2 d).
2 2
< CJ[\\wl\\oo + l\w2\\p] llf\l · (6.96)

(b) Let U: ffi.n - 7 C be a second function satisfying the hypotheses imposed on W in


(a), and let J be a bounded closed interval as in (a). Then, for each A E J, the limit
u -lin1 ~-t-++O W (Q) (H- A- itL) - l U (Q) *exists; the convergence is unifonn in A E
and the limit operators depend continuously on A (in nonn ).

PROOF. We treat the case v == 1. This covers all cases in which u > 1, whereas for
u E (112, 1) one has to have recourse to interpolation theory (this will be explained
after the present proof and in the Appendix to this chapter). We fix ( > 1 so large that
the interval ( -oo, -(]belongs to p(H) (recall that His lower semibounded).
(a) We use the fact that [w 1 (Q) + w2 (Q)](H0 + ()- 1 12 belongs to B(H). For
the term with w 1 this is evident; for that with w 2 one applies Proposition 2.33 by
298 THE CONJUGATE OPERATOR METHOD

observing that the function k r----t (k 2 + ()- 112 belongs to £P(JRn) if p > n. Thus, by
observing that II (H0 + ()- 112 11 < 1 because ( > 1, we have for a certain constant c 1:

(6.97)

(i) The inequality (6.97) implies that

ll[w1(Q) + w2(Q)](H + ()- 112 11 < ll[w1(Q) + w2(Q)](Ho + ()- 112 11


112 112
X II (Ho + () (H + ()- 112 11 < [llw1lloo + c1Jiw2IIP] II (Ho + () 112 (H + () 11·

By using the fact that (Ho +() 112 (H +()- 112 E B(H) (see below), one deduces from
the preceding inequality that (I+ IQI)vW( Q) (H + () - 112 E B(H) and that, for some
constant c2:

The inclusion (Ho+() 112 (H +()- 112 E B(H) [orD(IHJ 112) C D(H~ 12 )] is obtained
by interpolation from the fact that V(H) == V(H0 ), see Example 6.30. If w 2 == 0, the
validity of (6.98) is evident; in this case interpolation theory is not necessary.
(ii) Fron1 now on we assume that v == 1. Let us first show that the closure of
the operator W(Q)(H + ()- 1D belongs to B(H). In fact the unbounded factors Qj
occurring in D can be compensated by the decay of W, and the unbounded factors P1
by the resolvent of H. To make this precise we observe that, by (6.19), we have on
S(JRn ):
n .

2W(Q)(H + ()- D 1
=I: W(Q)(H + ()- 1 2
QjPj- ;'w(Q)(H + ()- 1 . (6.99)
J=1

By virtue of (6.98) one has

So the last term in (6.99) is bounded. The same conclusion holds for each term in the
su1n on the right-hand side of (6.99). Indeed we may write [using (6.117)]

l11 (Q)(H + ()- 1 Q1 P1 - Qj W(Q)(H + ()- 1 Pj


+2iW(Q)(H +()- 1 Pj(H +()- 1 P1 . (6.100)

The operator W(Q)(H + ()- 1 belongs to B(H) because W(Q) isH-compact, and
the closure of ~i (H + () - l Pj - [Pj (H + () - 1 12 ][Pj (H + ()- 112]*belongs to B(H)
because 11~7 (H + ()- 112 11 < II (Ho + () 112 (H + ()- 112 11 < oo. This establishes
the boundedness of the last term in (6.1 00). That of the first term on the right-hand
side of (6.100) is obtained by noticing that Q 1 W(Q)(H + ()- 112 E B(H) [by (6.98)
with v == 1] and by remembering that Pj(H + ()- 112 E B(H). By using the bound
RELATIVELY SMOOTH OPERATORS 299

(6.98) for IIQjW(Q)(H + ()- 1 12 11 and 1\W(Q)(H + ()- 1 12 \\, we get the following
inequality from (6.100): II W( Q) (H + ()- 1QjPj II < c4 [llw1lloo + llw2IIP]. We have
thus shown that

for some constant c 5 .

(iii) Let f E V(IDI) V(D) and write f == f+ + f- with f+ == ED([O, oo))f


and f- -ED(( -oo, O))f. Then f± E V(D), IDIJ == D f+- D f- and
IIW(Q)(H + ()- 1(1 + IDI)fll
< llW(Q)(H +() 1!11 + llW(Q)(H +()- 1Df+ll + IIW(Q)(H +()- 1Df-ll
< [IIW(Q)(H + ()- 1 11 + 2IIW(Q)(H + ()- 1 Dll] llfll·
Hence the closure of W( Q)(H + ()- 1(I IDI) belongs to B(H), with
IIW(Q)(H + ()- 1(1 \D\)11 < c6[\\w1\loo + llw2IIP]. (6.101)

(iv) The above results show that the operator C == W (Q) satisfies the hypotheses
of Lemma 6.14(a) [with v == 1; take C 1 == C 2 == W(Q) in Lemma 6.14]. Thus W(Q)
is H- smooth on each bounded closed subinterval of (0, oo) \a-P (H) [because (6. 91) is
satisfied on each interval of this type, with A == D]. To obtain (6.96) one introduces
the estimates (6.98) and (6.101) into (6.95), which gives

sup IIW(Q)(H- A iM)-


1W(Q)*II < c7[llw1lloo + llw2llp] 2.
>..EJ,O<f.-L<1
By Proposition 6.12 the inequality (6.96) is satisfied with CJ == 201rc 7 .
(b) This follows from Lemma 6.14(b) by taking into account the observations made
above [recall that the hypothesis of Lemma 6.14(b ), with A == D, is satisfied for each
bounded closed subinterval of (0, oo) \a-P (H)]. D

We have thus proved Proposition 6.16 for v == 1, with the exception of the fact
that (Ho + () 1 12 (H + ()- 1 12 E B(H). The verification of this fact and the proof of
the proposition for v < 1 involve an interesting interpolation theorem that we now
state; its proof and various applications (Examples 6.30- 6.33) are presented in the
Appendix to this chapter.

Proposition 6 . 17 . Let R and S be positive invertible self-adjoint operators with R E


B(H), and let X E B(H). Assume that the closure of RXS belongs to B(H). Then, for
each v E [0, 1], the operator X* maps R(Rv) into V(S 1/ ) . One has svX*Rv E B(H),
and the closure of the operator Rvxsv also belongs to B(H). Moreover

(6.1 02)

We use this proposition to prove the result of Proposition 6.16 for v E ( 1/2, 1). So
we assume that W(x) == [w 1 (x) w 2 (x)](1 + jxl)-v with v E (1/2, 1). We apply
300 THE CONJUGATE OPERATOR METHOD

Proposition 6.17 with X == [w 1 (Q) + w2 (Q)](H + ()- 1 , R - (I+ IQI)- 1 and


S == I+ jDj. We have seen in the proof of Proposition 6.16 that X E TJ(H) and
(I+ IQI)- 1X(I + jDI) E B(H), and also that IIXII + II(I + IQI)- 1X(I + jDI)II <
c[llu;llloo + llw2!1p]. Thus (6.102) shows that

This inequality generalises ( 6.101) for values of v belonging to ( 1/2, 1), and the rea-
soning that follows (6.101) is again applicable.

6.6 Higher order resolvent estimates

We present here a generalisation of Theorem 6.3 and of its consequences for powers
of the resolvent (T- A- iJ-L)- 1 . This has important applications in scattering theory,
as seen for example in Section 7.2. To obtain a bound on (J, (T - A - it-L)-Nf)
for suitable vectors f, one assumes that the multiple commutators (up to order N)
ofT with iA belong to B(H) (in the sense explained before Theorem 6.3). We set
Bk == [[ · · · [T, iA], iA], ... , iA] (k commutators). Observe that B 1 coincides with the
operator Bused in Section 6.2 and that each Bk is formally self-adjoint. We then have
the following generalisation of Theorem 6.3:

Theorem 6.18. LetT and A be se(f-adjoint operators with T E TJ(H), and let N E
{2, 3, ... }. Assume that each o.f the operators Bk, k - 1, ... , N + 1, belongs to
B(H) and that a strict Mourre inequality (6.34) is satisfied on so1ne open interval J.
Let Jo be a bounded closed subinterval of J and v > N- (1/2). Then the norm limits
u-linlp,-++O(I +I 1)-u(T- A iJ-L)-N (I+ IAI)-u exist uniformly in A E Jo, and

sup I (I+ IAI)-u (T- A it-L)-N (I+ IAI)-u II < oo. (6.1 03)
>..EJo.J1>0

PROOF. The proof follows the lines of that of Theorem 6.3. We just indicate the ad-
justments that are necessary. First the term II [B, A] II in the constant c 1 in (6.36) should
I:~= 2 + [ I:~= 1 IIBk II]
2
be replaced by IIBk I .

( 1) Definition and properties of G s


• Points (i) and (ii). Here the appropriate definition of N(- is as follows:
N
Nt- == T A it-L + z±(c) with z±(c)- L [(±ic)k Bk/k!].
k=1

Hence GE- Gs(A + it-L) == [T A- it-L + z-(c)]- 1, withEE [0, 1]. The basic in-
equality (6.41) remains true, the only change in its verification occurs in the derivation
of (6.39) where ~icE must be replaced by _z±(c).
HIGHER ORDER RESOLVENT ESTIMATES 301

• Point (iii). Here we have


N k
(g, (sB + J-L)g) == ±~(g, Nt-g) - L (-l)(k+l)/2 ~! (g,Bkg)
k=3,5,7, ...

(only odd values of k contribute to the preceding sum, because the terms in z± (E)
containing even powers of E are self-adjoint and thus do not occur in ~Nt-). If N > 2
this leads to an additional term on the right-hand side of (6.43) of the form cs 3 ll9ll 2
for some constant c < c 1 . By absorbing this additional term in the left-hand side, it
follows that, for sufficiently small E, the inequality (6.43) remains true provided that
its left-hand side is replaced by (as+ M)llgll 2 /4. Then the bound on IIGc(A + iM)II in
(6.44) obtains an additional factor 2 (and this unimportant factor should be taken into
account whenever this bound is used further on). Also the value of Eo may be smaller
than in the earlier proof.
• Point (iv). This remains unchanged, in particular (6.45) holds.
• Point (v). This is handled somewhat differently. We set G~ ) == T-A-itt- isB 1
1

and then have [using the equation x- 1 - y- 1 == x- 1 (Y- X)Y- 1 ]:


N . k N . k
G == Q(1) - " ' ( -~s) G B Q(1) == Q(1) - " ' ( -~s) Q(1) B ~G (6.1 04)
E E ~ k! E k E E ~ k! E k E•
k=2 k=2

Now G~ ) is just the operator GE used in Section 6.2, so that by (6.48), (6.49) and
1

(6.35): IIG~ ) (I+ IAI)- 1 11


1
II (I IAI)- 1 G~ 1 ) II < cs- 112 for some constant c and
all E E (0, co) (with c independent of A and J-L). Furthermore Ek liCE II < 8/ a if k > 1
and 0 < E < Eo [see (6.44)], and IIBk II < c 1 . By using these inequalities one obtains
from (6.104) that

(6.1 05)

for some constant c, all E E (0, s 0 ), A E J 0 and /-IE (0, 1 ).


(2) The differential inequality
• Point (i). Using the identity ak- bk ==
(a-b) (ak- 1 + ak- 2 b + ... + abk- 2 + bk- 1 ),
with a == -iT, b == -is, a calculation as in (6.50) shows that
N · k-1
G E' :=
- de
dG
E :=
·"' (-~s)
1} ~ ( k - 1)!
GE B k GE • (6.1 06)
k=1

19
On the other hand, the analogue of Eq. (6.51) is here as follows:
N . k N k
i'L: (-~~) -iL (-~~)
.r-

[A,Gc] = -iGc:BlGc:- Gc:Bk+lGc: = Gc:Bk+lGc:.


k=1 k=O

Comparison with (6.106) leads to

G~ (>.. + i!J) = [Gc(>.. + i!J), A] - i (-;:( Gc:(A + i!J )BN +I Gc (>.. + ifJ). (6.107)
19 We observe that, as in §6.3.2, the operators Gs map V(A) into V(A).
302 THE CONJUGATE OPERATOR METHOD

We also need a formula for (d/ de) [Gc:]N. By the Leibniz rule this is as follows:
N-1
d~ (Gc)"'v = L (GEfc;(Gc)N-k-1
k=O
N-1 N-1
= L (Gc)k[Gc, A](Gc)N-k- 1- i (-;:( L (Gc)k+l BN+1(Gc)N-k
k=O k=O
N-1
== [(G )N A] -i(-ic)N "'(G )k+1B (G )N-k (6.108)
c ' N! ~ c N+1 c .
k=O
• Point (ii). We replace Xc: simply by I and set Z == (I+ IAI)- 1 . We deduce a
differential inequality for the B(H)-valued function E r---+ <Pc: :== zv (Gc: )Nzv. For this
we observe that

N-1
_zv[(G )N A]zu_i(-ic)N '"""'zu(G )k+1B (G )N-kzu (6.109)
c ' N! ~ c N+1 c .
k=O
Now, since IIAZII < 1:
IIZl/[(Gc:)N,A]Zl/11 < IIZU(Gc:)Nzu- 1 . AZII + IIAZ. zu- 1(Gc:)Nzull
< IIZu(Gc:)Nzu- 111 + llzu- 1(Gc:)Nzull· (6.110)
The norms on the last line are treated as follows by an interpolation argument. In
Example 6.33 set B == z- 1 andY == (Gc:)N. Then (6.134), for~== v, leads to the
following bound on the first norm in (6.11 0):
IIZU(Gc:)N zu-111 < IIZl/(Gc:)NZl/111-(1/v) 11Zl/(Gc:)NII1/u.
Now, by (6.105) and (6.44), we have for j > 1:
IIZl/ (Gc: )j II < IIZl/ Gc: IIIIGc: llj- 1 < CjE- 1/ 2 E-(j- 1) - Cj E-(j-( 1/ 2 )]
for some constant Cj. Consequently, taking j == N:
IIZu(Gc:)Nzu- 111 < CNE-(N-( 1/ 2 )]/ui1Zu(Gc:)Nzull 1-( 1/u)- CNE- 13 II<Pc:ll 8 ,
with j3 == [N- (1/2)]/v and 8 == 1 - (1/v); observe that /3,8 E (0, 1) because
v > N- (1/2). The same bound holds for the second norm on the last line of (6.110)
(use the same argument with Gc: replaced by G;).
We next estimate the last term in (6.109). Its norm is majorised by
N N-1
~! L IIZuCciiiiCcllkiiBN+1IIIICciiN-k- 1IICcZvll
k=O
HIGHER ORDER RESOLVENT ESTIMATES 303

for some constant c. Thus (6.109) leads to the following differential inequality:

(6.111)

for some constant c, with p E (0, 1) and 6 E (0, 1).


(3) Boundary values
The preceding differential inequality is of the form (6.24 ), except that the exponent
6 of the fractional power of II <I>c: II may be different from 1/2 (but less than 1). By a
straightforward adjustment of the proof of (6.25) given in §6.2.1 one finds that (Prob-
lem 6.3)

for some constant c that is independent of A and p. As before, the preceding inequality
implies the existence of u -lim,u-t+O zv (T- A- ip,) -N zv, and one obtains a bound
of the form IIZv(T- A- i{l)-Nzvll <canst. for all A E J 0 and all p E [0, 1]. 0

Corollary 6.19. Under the hypothesis of Proposition 6.18 [in particular v > N -
(1/2)], the operator-valuedfunction A ~ (I+ !AI)-v(T- A- iO) 1 (I + IA!)-v
is N - 1 times continuously differentiable in nann on the interval ~l, and for k
1, 2, ... , N- 1 one has

(d~)k(J + IAI)-v(T- A- i0)- 1 ( / + IAI)-v


== k! (J + IAI)-v(T- A- iO)-(k+l)(I + IAI)-v. (6.112)

Here (I+ IAI)-v(T- A- iO)-r;,(I + IAI)-v is aformal notation for the operator
u-limc:~+o(I + IAI) vcc:(A + iO)K,(I + IAI)-v, with Gc:(z) defined in the preceding
proof

PROOF. For c 0, the operators zv G c: (A + iO) zv are infinitely differentiable (in


-1
norm) with respect to A, with (d/ dA)k zvcE (A+ iO)zv == k! zvcE (A+ iO)k+l zv.
As c---+ +0, the derivative kzvcc(A + iO)k+lzv of zvcc(A + iO)kzz/ converges
in norm to kzv(T- A- iO)-Ck+l)zv on each closed bounded subinterval J 0 of J,
provided that k E {1, 2, ... , N -1 }. By a standard result from analysis 20 , this implies
that zv(T- A- iO)-kzv is differentiable with respect to A (in the interior of J 0 ) and
that its derivative is kzv(T- A- iO)-Ck+l) zv. Since J 0 can be any bounded closed
subinterval of J, one obtains in particular (6.112). D

By proceeding as in Section 6.4, one can extend the results of Theoretn 6.18 and
Corollary 6.19 to unbounded operators H satisfying p( H) n JR. -1 0. As in §6.4.2,
this is done by setting T == (Ao - H) - l , with Ao E p( H). One has to know that the
multiple commutators of (Ao- H)- 1 with A (up to order N + 1) belong to B(H). For
H of the form H 0 + V( Q), this is ensured by suitably generalising the condition (C)
of page 288. A convenient condition, sufficient for our applications in Chapter 7, is
to assume that vis of class cN+l and that Tk(akv;ark) belongs to L~(IRn) for
20
Interchange of limits with derivatives; see for example §9 in Chapter V of [L].
304 THE CONJUGATE OPERATOR METHOD

k == 0, 1, 2, ... , N + 1. When repeating the arguments of §6.4.2, one arrives at the


following generalisation of (6.82):
(I+ IAI)-v(H- z)-k(I + IAI)-v
== (Ao- z)-k(I + IAI)-v [(T- (Ao - z)- 1 ] -k (I+ IAI)-v
X (I+ jAj)vTk(I + jAj)-v.
Here one has to know that (I+ IAI)vTk(I + IAI)-v E B(H) fork < Nand u >
N - (1/2). This is easily obtained by interpolation (Example 6.32) by using the fact
that T maps V(AN) into itself (see Problem 6.4).
Under suitable assumptions it is again possible to replace (I + IA I)- v on the left-
hand side of (6.112) by some other operator. In particular, if His a Schrodinger op-
erator, one can control the derivatives of (I + !Q!)-v (H - A i0)- 1 (I+ IQI)-v
and similar expressions if the potential decays sufficiently rapidly, as explained in the
following two propositions.
Proposition 6.20. Let k E N. Let U, W: JRn --+ C be two functions of the form
[w1 (x) + w2(x)](1 + ixi)-u with w1 E L 00 (1Rn) and w2 E LP(JRn)for some p satisfy-
ing p E (n, x) and p > 2, and with u > k + 1/2. Then the operator-valued function
A r-t W(Q)(Ho -A- i0)- 1 U(Q)* is k times continuously norm d(fferentiable on
(O,x).
PROOF. Given an interval J == (a, !3) with 0 < a < f3 < x, we choose a function
¢ E C 0 (JR) such that ¢(1-I) == 1 for all II in (a- c5, f3 + c5) for some c5 > 0. If S<z =1- 0,
we have [setting U == U(Q) and W == W(Q)]

W(Ho- z)- 1 U* == W¢(Ho)(Ho- z)- 1 ¢(Ho)U* + W[I- ¢(Ho) 2 ](Ho- z)- 1 U*.
(6.113)
The second term on the right-hand side may be written as

+ 1)- 1/ 2 ·[I- ¢(Ho) 2 ][I + (1 + z)(H0 - z)- 1] · (Ho + 1)- 112U*.


TiV(Ho
Since W(Ho + 1)-J/ 2 and U(Ho + 1)- 1 12 are bounded operators (see the proof of
Proposition 6.16) and since 1 -¢(II) 2 == 0 for II E (a- c5, (3 + c5), this term is infinitely
differentiable in norm in the domain { z E C I Rz E J}.
To treat the first term on the right-hand side of (6.113), we express it in the form
TiV¢(Ho)(I +D 2)u/ 2 ·(I +D 2 )-vi 2 (Ho -z)- 1 (I +D 2)-u/ 2 ·(I +D 2 )v1 2 ¢(Ho)U*.
The function Ar-t (I+ D 2 )-vi 2 (H0 - A-io)- 1 (I+ D 2 )-u/ 2 is k times continuously
norm differentiable on J by Corollary 6.19, and the closures of (I+ D 2 )v1 2 ¢(H0 )U*
and W ¢(H0 )(I + D 2 )vl 2 are bounded. 21 0
21
We have (I+ D 2 )1/1 2 ¢(Ho)U* =(I+ D 2 )1/1 2 cp(Ho)(I + IQI)-1/ ·(I+ IQI)l/(Ho + 1)- 1 U*,
with <p given by cp(p,) = (fJ, + 1)¢(p,). Now (I+ IQI)l/(Ho + 1)- 1 U* is bounded (see Exmnple 6.23).
For N EN, the operator DNcp(Ho)(I + IQI)-N is bounded, which is easily seen by commuting all of
the operators QJ occurring in DN through cp(Ho) and noticing that prn'tjJ(Ho) E B(7-l) if 1/J E C0 (I~)
(Problen1 6. 9). By interpolation one finds that (I+ D 2 ) u 12 cp( H o) (I+ IQ I) - 1 / is bounded for each v > 0
[takeR= (I+ IQI)- 2 N, X= cp(Ho) and S =(I+ D 2 )N in Proposition 6.17, with N > v/2].
HIGHER ORDER RESOLVENT ESTIMATES 305

Proposition 6.21 . Let k E N. Assume that V: JR.n --+ JR. satisfies condition (C) of
page 288 and may be written in the form V(x) == [v 1 (x) + v 2(x)](1 + jxj)-K with
VI E L
00
(JR.n ), v2 E Lq (JR.n) for some q satisfying q E ( n/2, oo) and q > 1, and with
f{; > k + 1. Let == Ho + V( Q) and let U, W: JR.n --+ C be as in Proposition
6.20. Then the operator-valuedfunction A~ W(Q)(H- A- i0)- 1 U(Q)* is k times
continuously differentiable in norm on (0, oo) \ O"p(H).

PROOF. The proof can be made inductively in k by using the result of the preceding
proposition. We discuss the cases k == 1 and k == 2. We put Rz == (H z)- 1 and
R~ == (Ho- z)- 1 .
(i) Let k - 1. By differentiating the second resolvent equation Rz == R~ - R~VRz
we get [' == d / dz]:

Thus (I+ R~V)R~ == R~'(I- VRz). Since (I- RzV)(I + R~V) ==I (by the
second resolvent equation), we then obtain

R~ == (I- Rz V)(I + R~V)R~ == (I- Rz V)R~' (I-V Rz)· (6.114)

We choose f\;1 > 1/2, f\;2 > 3/2 with f\;1 + f\;2 - f{;, set v(x) == v1(i) + v 2 (x)
and factor the function V into V V 1V2, with V 1(x) == jv(x) j 1 12 (1 + lxi)-K: 1 and
V2(i) == jv(x)l 1 12 [signv(x)](1 + lxi)-K: 2 • Then we have by (6.114):

WR~U* == WR~'U*- WR~'V2 · V1RzU* WRz V1 · V2R~'U*


+ WRz V1 · V2R~'V2 · V1RzU*. (6.115)

Let J be a bounded closed interval in (0, oo )\O"p (H) and let z == A+ ic in Eq. (6.115),
with A E J and c > 0. By taking into account Proposition 6.16(b) [for the operators
of the form X 1 (Q)R-\+ic:X2 (Q)] as well as Proposition 6.20 [for those of the form
X 1 ( Q)R~~ic: X 2( Q)], one sees that the right-hand-side of (6.115), for z == A+ is, is
norm convergent as c --+ +0, and that the convergence is uniform in A on J. So both
WR.A+ic:U* and its derivative with respect to A are convergent as c--+ +0, uniformly
in A E J. This implies that the limit WR.A+ioU* is differentiable in norm, and that its
derivative is given by the right-hand side of (6.115) with z - A+ iO (see for example
Theorem 12 in §9, Chapter V of [L]). So A ~ WR.A+ioU* is continuously norm
differentiable on J.
(ii) Let k - 2. By differentiating (6.115) one gets

WR"u*-
z WR z0 "u*- WR z0 "VR z u*- WR z0 'VR'u*
z
0 0 0
- WR'VR
z z 'U*- WR z VR z "U* WR'VR
z z 'VR z U*
+ WRz VR~"VRzU* + WRz VR~'VR~U*. (6.116)

Each operator V occurring on the right-hand side is factored into V1 V2, with V1 (x) and
v2 (x) of the same form as in (i) above but with the exponents f\;1 andf\;2 depending on
the position of V. If V stands next to an operator R~', we take f\; 1 == f\; 2 == f\;/2 > 3/2
306 THE CONJUGATE OPERATOR METHOD

(so that for example WR~'V1 and V2 R~U* converge as z ---+ J). If V stands on the left
of an operator RzO", we take ~1 > 1/2 and ~2 > 5/2 (so that for example V2RzO" U *
has boundary values on J), and if V stands on the right of an operator R~' ', we take
f1: 1 > 5/2 and ~ 2 > 1/2. Then, as in (i), the right-hand side of (6.116), for z == A+ ic,
is norm convergent as c ---+ +0, uniformly in A on each bounded closed subinterval J
of (0, oo )\o-P (H). Hence A ~ WR{+iO U* is differentiable in norm, and its derivative
is given by the right-hand side of (6.116) with z == A+ iO. So A ~ WR.A+iO U* is
twice continuously norm differentiable on each J. D

6. 7 Some commutators

6 . 7 .1 . As mentioned in the introduction to this chapter, the evaluation of commutators


involving unbounded operators requires care. In the present section we illustrate this
by means of examples the results of which are quoted in the text. The Hilbert space is
1-{ == L 2(JRn ), and we shall use the notations (P) == (I+ P2)1 12, (Q) == (I+ Q2 ) 1 12
a
and o1 == 1ox) (j == 1, ... , n).

Example 6 . 22. In various situations one can proceed by using scalar products with
vectors belonging to a suitable dense set V, chosen such that the commutator makes
sense on V, and then taking into account domain properties to transfer the operators
to the other argument in the scalar product. As a simple example, let us prove that
[P1 ~ <p(Q)] == -i(8.7<p)(Q) on V(P1 ) if <pis of class C 1 with <p and its first order
derivatives Oj4J bounded. Note that we must show in particular that iff E V(Pj ), then
<p( Q)f E D(P_7 ).
(i) We first let g, hE S(JRn ). Then

(P1 h, <p(Q)g) - i(o1 h, <pg) == -i(h, 4Ja.79 )- i(h, (oj<p)g)


== (h' <p (Q) pj g) - i ( h' (8 j <p) (Q) g) '

where the second equality is obtained upon integrating by parts. Since <p (Q) Pj g and
(8.7 <p) (Q)g belong to the Hilbert space, this implies that <p( Q)g E V(Pj) [remem-
ber that Pj is essentially self-adjoint on S(JRn)] and that Pj<p(Q)g == <p(Q)Pjg-
i(o1 <p)(Q)g.
(ii) Now let f E V(P1 ) and take V == S(JRn). ForgE S(JRn) we find by using the
final result of (i) (with <p in place of <p) that

(Pj g, 4J (Q) f) == (4J (Q) P.i g, f) == (Pj 4J (Q) g, f) - i ( (8j 4J) (Q) g, f)
== (g, <p(Q)Pj.f) i(g, (oj<p)(Q).f).

As in (i), this implies that <p(Q)f E V(Pj) and Pj<p(Q).f == <p(Q)Pjf- i(oj<p)(Q)f,
which completes the proof.
By setting f == (I + IPj I) - 1 h in the last relation above, one finds that the operator
P.7 <p (Q) (I + IP1 I) 1 belongs to B (H). In the same way as above one can obtain the
SOME COMMUTATORS 307

result that (Pj )N cp( Q)(I + IPI)-N E B(H) if rp is a bounded function of class eN
such that all its partial derivatives up to order N are bounded (N E N). This implies
that (P) N cp( Q)
-
(P)- N E B(H) for even N, hence also (P) v cp( Q) -
(P)- v E B(H) for
each v E [0, N], by interpolation (applying Proposition 6.17 as in Example 6.31).
Let us point out a different way of getting the result of (ii) above. For J, g E D(Pj)
v;e have

with cpt(Q) == e-iP7trp(Q)eiPJt. Now, since {e-iP1t} is just the translation group
along the j-th co-ordinate axis (see Example 5.4), rpt (Q) is the multiplication operator
by cp(x1, ... ,Xj-1,Xj- t,xj+1, ... ,xn)· So (8/8t)cpt(i)lt=O == -81 rp(x) and we
get (by using the Dominated Convergence Theorem):

(P.i f, rp( Q)g) - (f, cp( Q)P.ig) - i(f, (8.7cp) (Q)g).

This again implies that rp(Q)g E D(P.i) and that Pjrp(Q)g == rp(Q)P1 g-i(8Jrp)(Q)g.

Example 6.23. By interchanging the roles of P and Q in the arguments of the pre-
ceding example, one sees that (Q)vcp(P)(Q)-v E B(H) for each v E [O,N] if rp is
a bounded function of class eN such that all its partial derivatives up to order N are
bounded. Particularly important examples in our context are the following functions:
rp(k)- (k 2 - z)- 1 and cp(k) == k1 (k 2 - z)- 1 [j == 1, ... , n; z E CC\JR]. These func-
tions are of class coo and their partial derivatives of any order are bounded. Therefore
(Q)v (Ho- z )- 1 ( Q) -v E B(H) and (Q)v Pj (Ho- z )- 1 ( Q) -v E B(H) for each v > 0
and each z E p(Ho).
Furthermore, by observing that ( Q.i )N (Ho z )- 1 is given by a sum of terms of
the form ek(P)(Ho- z)- 1(Qj)k with ekE L 00 (JR.n) and k == 0, ... 'N, one finds
that the closure of (Q)N (H0 - z)- 1 W(Q)(Q) -N belongs to B(H) if W: IR?.n - 7 CC is
such that W ( Q) is a H 0 -bounded operator. By interpolation one finds that the closure
of (Q)v (Ho - z)- 1 W (Q) (Q) -v belongs to B(H) for all v > 0.

Example 6.24. Consider a Hamiltonian of the form == H 0 + V (Q), assumed to be


self-adjoint on D(H0 ). Let us show that, for z E p(H), the resolvent (H- z)- 1 maps
V (Q .i) into V (Q .i) and that for each g E V (Q 1 ) :
Q.7(H- z)- 1g == (H- z)- 1Q.ig- 2i(H z)- 1 P.7(H- z)- 1g. (6.117)

For the proof, let J, g E V( Q1 ). Then we have:


(Qjf, (H- z)-lg)- (!, (H- z)-lQjg) = i! (eiQJtf, (H- z)-le'QJtg)jt=o·

We set (t) == e-iQJtHeiQJt, which is well defined on V(H) since this domain is
invariant under eiQJt. 22 Then [H(t)- z ]- 1 == e-iQ 7 t(H- z)- 1eiQ]t and hence

(eiQ,tf, (H- z)-leiQJtg) == (f, [H(t)- z]-1g).


22 We have D(H) = D(Ho), and etQJtD(Ho) ~ D(Ho) (Problem 6.7).
308 THE CONJUGATE OPERATOR METHOD

To calculate (d/ dt) [fl (t) - z] - 11t=O we observe that

1
- { [fl (t) - z] - 1 - ( fl - z) - 1} == - [fl (t) - z] - 1 __;___t- (H - z) - 1.
t

Now fl(t) == flo(t) + V(Q), with Ho(t) == e-iQ tP 2 eiQ t == f~= 1 Pl; + (PJ + t) 2 ,
1 1

where f~= 1 means that the term with k == j has been omitted in the sum 2:~= 1 . So
(d/dt)fl(t)jt=O == (d/dt)flo(t)jt=O == 2Pj [on V(flo)]. Hence we have

(QJJ, (H- z)- 1g) == (J, (fl- z)- 1 Qjg)- 2i(f, (H- z)- 1Pj(fl z)- 1g),

from which the claimed results follow.


Example 6.25. Consider again a Hamiltonian of the form fl == H 0 + V(Q), where
V: JRn - t JR is such that V (Q) is fl0 -compact and (Q) V (Q) is flo-bounded. Then
is self-adjoint on D(fl0 ). Let us show that the operator (Q)v(fl -z)- 1(Q)-v belongs
to B(H) for every z E p(H) and each v > 0. By the second resolvent equation (2.115)
we have

(Q)v(H z)-1(Q)-v _ [(Q)v(flo _ z)-1(Q)-v]


- [(Q)v(flo- z)- 1(Q)V(Q)(Q)-v] [(Q)v- 1(H- z)- 1(Q)-v]. (6.118)

By Example 6.23, the first term on the right-hand side belongs to B(H) for any v >
0. To handle the second term on the right-hand side, we first observe that its first
factor also belongs to B(H) for each v > 0, again by Example 6.23. Its second factor
[i.e. (Q)v- 1 (fl- z)- 1(Q)-v] trivially belongs to B(H) ifO < v < 1. Consequently
(Q) v ( fl- z) - 1 ( Q) -u E B (H) for v E [ 0, 1]. Now let v E ( 1, 2] in (6.118). Then, since
(Q) v-l (H - z) - 1(Q) -u E B(H) by the preceding result, we find from (6.118) that
(Q) v (fl - z) - l ( Q) -u E B(H) for v E (1, 2]. By continuing this argument (taking
v E (2.3], then v E (3,4] etc.), one obtains that (Q)u(fl- z)- 1(Q)-ubelongs to
B(H) for all v > 0.
Example 6.26. Here we give two boundedness results that will be needed in Section
7 .2. We consider the class of Hamiltonians fl of Example 6.25 and assume in addition
that Vis a bounded function. By the second resolvent equation we have for any v > 0:

(Q)u(H- z)- 1 Pj(Q)-v ==[I- (Q)v(H- z)- 1(Q)-uVJ


X [ (Q)u (flo- z)- 1Pj (Q) -u], (6.119)

and by Examples 6.23 and 6.25 above, each factor on the right-hand side belongs to
B(H). Thus, taking v == 2, we have for J, g E S(IRn ):

I(f' (Q) -2 pj (H - i) -1 ( Q) 2 g) I == I( (Q) 2 ( fl + i) -1 pj (Q) -2 f' g) I


2 2
< II filii (Q) (H + i) - 1 Pj (Q) - 1111911'
2 2
and (2.4) shows that the closure of (Q) - Pj (H- i)- 1 ( Q) belongs to B(H).
SOME COMMUTATORS 309

Now let W: JR.n ---+ C be such that W (Q) is a H 0 -bounded operator. Upon re-
placing in (6.119) the operator PJ by W (Q), one finds as above that the closure of
(Q) -vW(Q)(H i)- 1 (Q)v belongs to B(H) for each v > 0.

6.7.2. We add some remarks concerning (6.73) and (6.74), in the framework consid-
ered in §6.4.1, hence with A == D and by assuming condition (C) on page 288 to
be satisfied. To be able to use the conventions made at the beginning of §6.2.2, we
would have to show that the identities (6.73) and (6.74) hold between arbitrary vec-
tors j, g belonging to V(A) V(D), for example that, for some z E p(H) n JR. and
all j, g E D(D):

(f, (H- z)- 1 Dg)- (Df, (H z)- 1 g)- i(f, (H-z) 1


[H- Z(Q)](H- z)- 1 g).
(6.120)
To obtain this equation, we follow the method of Example 6.24. We observe that, if
j,g E D(D):
1
(!, (H- z)- Dg)- (DJ, (H z)- 1 g) = -i d~ (Uo;2f, (H- z)- 1 Uo;2g)io=o'

where {UK;}K:EIR is the dilation group defined in (6.16) [we had UK;== e-iDK e 2 iDK].
Let us set H(B) - Ue; 2HU 8 ; 2, which is well defined on V(H) since this domain is
invariant under the dilation group. Then [H (B)- z] - 1 == Ue; 2(H-z) -J Ue; 2 and con-
sequently

(Ue;2f, (H- z)- 1 Ue;29)- (f, Ue; 2(H- z)- 1 Ue;29)- (f, [H(B) z]- 1 g).
It remains to calculate (d /dB) [H (B) - z] -ll e=o and to check that this operator is
identical with (H-z) - 1 [H-Z( Q)] (H-z )- 1 , where Z (x) is given by (6.77). Now:

By taking the strong limite ---+ 0 we get

e
+ Ve,
-f -f

By Eq. (6.21) we have H(B) == e Ho with Ve Ve(Q) - Ue; 2V(Q)Ue;2


V (e-e 12 Q). Since (d/ dB)e 8 H 0 le=o == H 0 - H V( Q), it remains to show that

! (H- z)- Vo(H- 1


z)- 1 0 = 0
1 = ~(H- z)- 1 [r ~~] (Q)(H- z)- 1 . (6.121)

To obtain this last identity, let us set V == (H - z )C0 (JR.n). Then V is a dense
linear manifold in H, 23 and one has (H- z)- 1 D == C0 (JR.n). Iff~ g ED, we obtain
23
This follows because His essentially self-adjoint on C 0 (IR.n ). To see this, remen1ber that Ho is es-
sentially self-adjoint on C 0 (IR.n) and apply the following generalisation of the Rellich-Kato Theore1n:
Proposition 2.44 remains true if one replaces everywhere in its statement the term "self-adjoint" by "es-
sentially self-adjoint". The proof of this variant is practically identical with that of Proposition 2.44.
310 THE CONJUGATE OPERATOR METHOD

by using Eq. (6.75) and the invariance of C0 (1Rn) under U 8 ; 2 :

d~ (f,(H- z)- 1V(B)(H- z)- 1g)


-i(f, (H- z)- 00; [r ~~] (Q)Ue; (H- z)- g).
1
2 2
1
(6.122)

In particular (sete- 0): Eq. (6.121) is satisfied between vectors j, g E V. Let us now
integrate the identity (6.122) on [0, c-]:

(f, (H-z)- 1 [V(c-)- V](H- z)- 1g)

--i 1s (!, (H z)- 1 00; 2 [r ~~] (Q)Ue; 2 (H- z)- 1 g)dB. (6.123)

Now

av] (Q)Ue;2(H-
[ra;: - z) -1
- [ av] -
ra;: (Q)(Ho- z) -1 · (Ho- z)Ue; 2(H- z) -1

and
(Ho- z)De;2(H- z)- 1 == Ue; 2(e 8H 0 - z)(H z)- 1 .

Since [r(8V/8r)] (Q)(Ho - z)- 1 E B(H) by assumption, this shows that the cor-
e
respondence ~ [r(8V/8r)] (Q)Ue; 2 (H- z)- 1 determines a strongly continuous
B(H)-valued function. Thus the expression

1°(H z)- 1 U0; 2 [r ~~] ({j)Ue;2(H- z)- 1 d(}

defines an operator in B (7-i). Since (H - z) - 1 [V (c) - V] (H - z) - l also belongs to


B(H), Eq. (6.123) is satisfied for all j, g E 1-{ (approximate these vectors by vectors
belonging to V, as in the proof of Proposition 6.5). By differentiating (6.123) with
respect to c (at c == 0, and with arbitrary j, g E 7-i), one obtains the validity of (6.121).
Above we have shown how to obtain (6.73) for arbitrary j, g in V(A). An anal-
ogous argument [using also the assumption made on r 2(8 2V /or 2) in the condition
(C)] allows one to obtain the validity of (6.74) between arbitrary vectors f, g E V(A).

Appendix to Chapter 6: Interpolation of operators

The theory of interpolation for Banach spaces and for operators is an important topic
in functional analysis (see for example [BL], [BS]). One distinguishes between the
real method of interpolation and the con1plex method of interpolation. We give here
a simple result in the framework of complex interpolation. This result is based on the
following lemma characterising certain holomorphic functions in a strip of the form

0 == { z E CC I 0 < Rz < 1}. (6.124)


APPENDIX TO CHAPTER 6 311

Lemma 6.27 (Hadamard Three Line Theorem) . Let F: 0 - 7 C be a holomorphic


function. Assume that F has a continuous bounded extension to the closed strip 0 ==
{z E C I 0 < ~z < 1}. Let Mo == supyEIR IF(iy)l and M1 == supyEIR IF(1 + iy)i.
Then

Vx E [0, 1], Vy E JR. (6.125)


2
PROOF. For z E 0 and c > 0 we set Fe(z) == eez +K:z F(z), where/'\; is a real constant
that will be fixed further on. One has (z - x + iy with 0 < x < 1):

(6.126)

hence IFe ( z) I - 7 0 as y = <ZS z -7 ± x. Let us also observe that

Since Fe is holomorphic in 0 and continuous on 0, the function IFe I attains its


maximum on the boundary of 0 (by the maximum principle for analytic functions),
hence in our situation on one of the two lines { z == i y I y E IR} and { z == 1+iy I y E IR}
(because !Fe (z) I == 0 at infinity):

IFe (z) I < max {sup !Fe( iy) I, sup IFe (1 + iy) I}
yEIR yEIR

In other terms IFe(z)l < n1ax{M0 , ee+K:l\J1} for each z E 0. By taking into account
the first equation in (6.126), this can be rewritten as
2 2
IF(x + iy)l < e-e(x -y )e-K;xmax{Mo, ee+K;Ml}
2 2
< ee e-e(x -y ) e-K:x max{ Mo, eK NI1 }.

This inequality holds for each c > 0, hence also for c - 0:

IF(x + iy)l < e-K;x max{Mo, eK;M1}.


Let us set/'\;== log[M0 /M1 ] (one may assume that M 0 , A11 > 0, otherwise F == 0),
so that eK; M 1 - Mo. Then

IF(x + iy)l < Moe-log[MofM,]x = Mo(:~r- Mfj-xMf. D

Lemma 6.28. Let C be a positive invertible self-adjoint operator and let f E V( Cf).
Then the vector-valued function z ~ czj is strongly d(fferentiable in the strip 0
defined in (6.124) (hence this function is strongly holomorphic in 0) and strongly
continuous on 0.

PROOF. We denote by { E( ·)} the spectral measure of C. The operator cz is given by


the functional calculus: cz - J
0
00
AzE( d:\). Formally one has (d/ dz )Cz == cz log C.
312 THE CONJUGATE OPERATOR METHOD

Since Rz E (0, 1) and C > 0, the vector Czlog Cf - f 0CX)A.x+iY(log A.)E(dA)j is


well defined for f E V (C). We must verify that

(hECC). (6.127)

Let us show that this follows from the Dominated Convergence Theorem (Proposition
1.9). For each fixed A> 0, the integrand converges to zero as Ihi ---+ 0. So it remains
to find an h-independent upper bound of the integrand by a function belonging to
L 1 ((0,x),mt). For this we use Taylor's formula in second order (which is easily
verified by partial integration):

cp(z +h)= cp(z) + hcp'(z) + h2 fo\1- r)cp"(z + rh)dr ( '== _!}_)


- dz ·

\ >..z+hh_ Az - AzlogA\ < lhifo\1- r)(logA) !Az+rh!dr2

< !h!fo\logA?AiR(z+rh)dr.

We assume that !hi < min{x/2, (1 - x)/2}. Since x E (0, 1), we then have 0 <
x/2 < R(z + Th) < (1 + x) /2 < 1 for all T E [0, 1], so that jhj(log A.) 2 A~(z+rh) <
Cx n1ax { 1, A} for some constant Cx depending on x and for all T E [0, 1]. The function

A ~ max{1, A} belongs to L 1 ( (0, x ), m f) for f E V( C).


We have obtained the complex differentiability of Czf in 0. Its strong continuity
on () is proved in the same way. 0

In what follows we use several times the formula (2.4 ): If V 1 and V 2 are dense
linear manifolds in Hand Z is an operator with V 2 C V(Z), then

//Z/1 == sup sup /(f,Zg)/ (6.128)


fED1, 11!11=1 gED2, 11911=1

[the right-hand side is finite if and only if the closure of Z belongs to B(H)].

Lemma 6.29. Let C1, C2 be self-adjoint operators with C 1 E B(H), and let X be an
operator in B(H). Then the following two statements are equivalent:
(a) X* 1naps R(C1) into V(C2) and C2X*C1 E B(H).
(b) The closure ojC1XC2 [defined on V(C2)] belongs to B(H).
If one of these two conditions is satisfied, one has C2X*C1 == [C1XC2] *, in particu-
lar I/C2X*C11/ == I/C1XC21l·

PROOF. For f E 1{ and g E V( C2) one has

(6.129)
APPENDIX TO CHAPTER 6 313

(i) Under the hypothesis of (b), (6.129) implies that

(6.130)

For fixed f, (6.130) shows that the scalar product in (6.129) defines a bounded linear
functional rpf on V( C 2). One can extend rpf to a bounded linear functional on all of
1t [if g E 1t and {gk} is a sequence in D(C2) such that llg- gkll ~ 0, set rpJ(g) ==
limk-too rpJ(gk); the limit exists because rpJ(gj) - rpJ(gk) ~ 0 as j, k ~ oo, by
(6.130)]. By the Riesz Lemma (Proposition 1.8) there exists a vector hE 1t such that
rp1 (g) == (h, g) for each g E 7-t. ForgE D(C2) this means that

(X*C1J, C2g) == (h, g).

Hence X*C 1f belongs to the domain of C2 == C2, and C2X*C1f - h. Thus the
operator C 2X*C1 is defined on all of 1-t; in particular X* maps R(C1) into V(C2).
(6.130) shows [combined with (6.128)] that IIC2X*C1II < IIC1XC2II·
(ii) Under the hypotheses of (a) one finds from (6.129) that

for all f E 1t and all g E V( C2). Hence


IIC1XC2IIv(C2 ) == sup sup I(J,C1XC2g)l < IIC2X*C1II·
!E'H,IIJII=l gEV(C2),II9II=l

This shows that the closure of the operator C 1XC2 [defined on V(C2)J belongs to
B(Jt). D
Proof of Proposition 6.17. We use the notations and the result of Lemma 6.27. Let
g E V(S). Then the vector-valued function z f-----7 RzX szg is strongly holomorphic in
0 and strongly continuous on 0 [Leibniz rule, by taking into account Lemma 6.28
and the boundedness of R, see (2.16)]. Let f E 1t and define

(6.131)

One has
Mo ==sup IF(iy)l < IIXIIIIRiYJIIIISiYgjj IIXIIII!IIIIgll
yEffi.

(because RiY and SiY are a unitary operators), and

M1 ==sup IF(l +iy)j ==sup 1\RiYJ,RXSSiYg)l < IIRXSIIII!IIIIgll.


yEffi. yEffi.

Thus Lemma 6.27 implies that

(6.132)

Hence, taking x == v and y - 0:

IIRvXSvll- sup sup 1\f,Rvxsvg)l < Cv.


fE'H, IIJII=l gEV(S), llgll=l
314 THE CONJUGATE OPERATOR METHOD

This shows that the closure of the operator Rvxsv [defined on V(S)] belongs to
B(H), and all assertions follow from Lemma 6.29 (take C1 == Rv and C2 == sv in
Lemma 6.29). D
We present some applications of Proposition 6.17 that are used in the text. Below
the operators R, S and X refer to Proposition 6.17, identified in each example with
operators of some application in the text.
Example 6.30. In the situation of Proposition 6.16, let us take R - (H + () - 1 ,
S == H 0 + (and X == I, where ( > 0 is chosen such that H + ( > 0. The operator
(Ho + ()(H + ()- 1 I- V(H + ()- 1 belongs to B(H). Hence the closure of
(H + ()- 1 (Ho + () belongs to B(H) (use for example Lemma 6.29). So Proposition
6.17 implies that the closure of (H + ()-v(Ho + ()v belongs to B(H) for each v E
[0, 1J. Applying again Lemma 6.29, we then also have (Ho + ()v (H + () -v E B(H).
This also i1nplies that (Ho + ()v(H + ()- 1 (Ho + ()-v E B(H) for v E [0, 1].
Example 6.31. In the situation of Theorem 6.1 0, let us take R - (I + IA I) - 1 , S -
I+ !AI and X == T. One has (I+ IAI)T(I + IAI)- 1 E B(H) by the footnote on
page 292. Since T maps V(IAI) into itself (page 274), this implies that the closure of
RXS (I+ IAI)- 1 T(I + lA I) belongs to B(H) (using Lemma 6.29). Hence for
each v E [0, 1]~ the closure of (I+ lAI)-vT(I + IAI)v belongs to B(H), and T maps
D(IAiv) into D(IAiv). In particular (I+ lAl)vT(I + IAI)-v E B(H).
Example 6.32. Let B and C be positive invertible self-adjoint operators in a Hilbert
space H, with B- 1 E B(H), and let X E B(H). In Proposition 6.17, takeR== B-~
and S == c-~ for some/'\;> 0. Then, for v E (0, 1):
IIB-vKxc-v~ll < IIXII 1 -viiB-~xc-~llv.

Setting e == VI'\;, this reads as follows:


IIB-exc-ell < IIXII 1 -(e;~)IIB-~xc-~ll 81 ~, o < e < /'\;. (6.133)
Example 6.33. Let B be a positive self-adjoint operator in a Hilbert space H, with
inverse in B(H). Let Y be some operator in B(H) and X - B-~yB-~ for some
/'\; > 0. In Proposition 6.17, takeR I and S == B~. Then, for v E (0, 1 ):
IIXBv~~~ IIB-~YB-( 1 -v)~~~ < IIB-~YB-~11 1 -viiB-~YIIv·

In particular, for v 1/ /'\; (assuming that /'\; > 1):


IIB-~Y B-~+ 1 11 < IIB-~Y B-~ 1 1 -(1/ ~) IIB-~YII 11 ~. (6.134)

Bibliographical Notes
A very extensive exposition of Mourre theory is given in the monograph [ABG]. Var-
ious other methods for determining spectra of quantum-mechanical operators are dis-
cussed in [D K]. A recent review of spectral results for Schrodinger operators is given
in [6], and a presentation of spectral and scattering theory for non-relativistic N-body
Hamiltonians can be found in [9]. For a more complete discussion of the theory of
relatively smooth operators we refer to Volume IV of [RS].
PROBLEMS 315

Problems

6 . 1. In the example considered in §6.1.2 ([H, iA] - H, f E D(A)) it was shown that
(f, (H A- iM)- 1 f) converges for each A > 0 as{); ~ +0. Show that this conver-
gence is uniform in A on any interval [Ao, oo) with Ao > 0.
6.2. Let H and A be bounded self-adjoint operators such that [H, iA] > C > 0
for some bounded self-adjoint operator C. Show that C 112 is H -smooth and that
J~ llC 1 12 e-iHtfll 2 dt is finite for each f E 7-i.
[Hint: Putnam's Theorem.]
6.3. Let D and Do be as in §6.1.1 (page 264). Let F: D ~ B(H) be continuous in
norm and continuously differentiable with respect to the first variable E on D 0 . As-
sume that F satisfies the following differential inequality on Do: ll(8l8c)F(c, M)ll <
c [c-~IIF(c, M)ll 6 + 1], with (3, c5 E [0, 1). Show that u-limt.L-t+O F(O, M) exists.
6.4. LetT E B(H) and let A be self-adjoint.
(a) If [T, A] and [[T, A], A] belong to B(H) (in the sense of §6.2.2), then T maps
D(A 2 ) into D(A 2 ).
(b) If [T, A], [[T, A], A] and [[[T, A], A], A] belong to B(H), then T maps D(A 3 ) into
D(A 3 ).
6 . 5. Prove Corollary 6.9.
6.6. Let H be self-adjoint and let C be a closed operator that is H -smooth on some
interval J.
(a) Verify that if J C J, then Cis also H -smooth on J.
(b) Show that all eigenvectors of H in J belong to the null space of C: if f satisfies
Hf == Aj with A E J, then Cf == 0.
6.7. In L 2 (IR?.n), show that f E D(H0 ) ===? eiQ.1tj E D(Ho) for all real t.
6.8. In L 2 (IR?.), consider the Stark Hamiltonian H - P 2 + EQ, where E > 0 is a
constant.
(a) Verify that His essentially self-adjoint on S(IR?.).
(b) By using the results of Section 6.2 (for unbounded T), show that the spectrum of
H is purely absolutely continuous: o-( H) == o-ac (H).
(c) Verify that His unitarily equivalent to the differential operator fi == iE( dl dx) +
Q2 in £2 (IR?.).
(d) By using Weyl's criterion, show that o-(H) o-(H) - o-ess(H) - IR?. and conclude
that a-( H) == o-ac (H) == IR?..
(e) Show that the operator fi is unitarily equivalent to iE (dI dx) in L 2 (IR?.). This again
proves that o-(H)- o-ac(H) ==JR.
(f) Establish similar results for then-dimensional Stark Hamiltonian H == P2 + EQ 1
in £ 2 (IR?.n ), n > 2.
[Hints. For (b): the operator Pis conjugate to H. For (d): use functions fn of the form
3
f n(X) - Cn e-ix I 3 e2 :Ax f (X In), A E IR?., f E L 2 (IR?.) .]
6.9 Let N EN,~ E C 0 (IR?.) and D == (Q · P + P · Q)l4 in H == L 2 (IR?.n). Verify that
the operator DNcp(Ho)(Q) -N belongs to B(H).
316 THE CONJUGATE OPERATOR METHOD

6.10. In L 2 (1Rn), let H == H 0 + V(Q), where V: JRn---+ JR is a potential of class Woo


(see page 332).
(a) Show that V(Hk) == V((Ho)k) for each kEN.
(b) For z E p(H) and j, k, £, m == 1, ... , n, show that the following are bounded
2
operators: P1 Qk(H- z )- 1 (Q) - 1 , QjQk (H-z )- 1 ( Q) - , PjQkQe(H- z )- 1 ( Q) - 2
andPjPkQRQm(H- z)- 1 (Q)- 2 .
(c) For N > 1, let p(N) be a polynomial in P 1 , ... , Pn of degree N, and let Q(N) be
a polynomial in Q1 , ... , Qn of degree N. Let M be an integer> N /2. Sho\X; that the
operator p(N) Q(N) (H- z)-M (Q) -N is bounded [z E p(H)].
(d) For N, M and z as in (c), show that the operator DN (H-z)- M ( Q)- N is bounded.
6.11. Let V, U, W: JRn ---+ JR satisfy the conditions of Proposition 6.16 and assume in
addition that V(x) == U(x)W(x). Prove the following statements.
(a) For z E CC\lR, the operator Xz :- W(Q)(H0 - z)- 1 U(Q) is compact, I+ Xz is
invertible in B(H) with (I+ Xz)- 1 - I- W(Q)(H z)- 1 U(Q).
(b) For z E CC \ JR, the operator W( Q) (H - z)- 1 U( Q) is compact.
(c) For A E (0, oo) \ ap (H), the operators X-\+iO and W( Q) (H- A- io)- 1 U( Q) are
compact, I+ X-\+iO is invertible in B(H), and

(d) If A > 0 is such that I+ X-\+iO is not invertible, then A is an eigenvalue of H.


(e) A > 0 is an eigenvalue of H, then I + X -\+iO is not invertible: if H g - Ag with
g E D(H), then f :== W(Q)g satisfies (I+ X-\+io)f == 0, and f -=J 0.
CHAPTER 7

Further Topics in Scattering Theory

This chapter contains some applications to scattering theory of the results obtained in
Chapter 6. In the first section we obtain asymptotic completeness by using properties
of relatively smooth operators. In Section 7.2 we prove the relation between scattering
into cones and the probability current across distant surfaces. Section 7.3 is devoted to
a derivation of the principal equations of stationary state scattering theory as a conse-
quence of the asy1nptotic condition. These equations allow us to obtain an important
stationary expression for the S-matrix S(-A) in Section 7.4. In the final Section 7.5 we
present some aspects of the concept of time delay in scattering theory.

7.1 Asymptotic completeness

7.1.1. Let us first consider the general situation of §5.3.2: A and Bare two self-adjoint
operators and Ut == e-iAt, Wt - e-iBt. We denote by { EA (·)} the spectral measure
of A, by { EB (·)} that of B, by E~ the projection onto Hac (A) and by E~ that
onto Hac(B). Assume that 0 O(A,B;Ef!cJ == s-limt--++oo UtWtEf!c exists. In
this general setting one would say that 0 is complete if R( 0) == Hac (A) [we know
from Proposition 5 .15( d) that R( 0) c Hac (A)]. We have the following completeness
criterion:

Proposition 7.1. Assume that 0 _ O(A,B;Ef!c) == s-limt--++oo UtWtEf!c exists.


Then R(O) == Hac(A) if and only if O(B,A;E~) == s-limt--++oo WtUtE~ exists.
PROOF. (i) Assume that R(O) == Hac(A). Then, given g E Hac(A), there exists a
vector f E Hac(B) such that g- Of, so that

as t---+ +oo.

This shows that WtUtg is strongly convergent. Hence O(B,A;E~) exists.


(ii) Assume that O(B,A;E~) exists. Let g E Hac(A) and f- O(B,A;E~)g. One
has f E Hac(B) [see Proposition 5.15(d)] and II!- WtUtgll ---+ 0 as t---+ oo, hence
llg- UtWtfll ---+ 0. In other terms, g == O(A,B;Ef!c)f Of, hence g E R(O). D
318 FURTHER TOPICS IN SCATTERING THEORY

The next theorem gives a criterion for the existence of wave operators. It is entirely
symmetric in A and B (except for a sign, which is of no importance since, if Cis an
H -smooth operator, then so is -C). Thus, in applications this theorem should allow
one to prove at the same time the existence and the completeness of wave operators
(let us recall that, for a class of non-relativistic Hamiltonians in dimension n > 3, the
existence of the wave operators has already been obtained in Section 5 .4, so in this
case the theorem can be used to obtain their completeness).

Proposition 7.2 (Lavine's Theorem). Let A and B be self-adjoint operators in a


Hilbert space H. Assume that A - B can be factored into A - B == C{ C2 in such a
way that C 1 is A-bounded, C 2 is B-bounded, C 1 is A-smooth on some open interval J
and C 2 is B-smooth on J. More explicitly, assume that there exists an operatorC1 that
is A-smooth on J and an operator C2 that is B-smooth on J such that C 1 (A+ i)- 1 E
B(H), C2(B + i)- 1 E B(H) and

(Aj,g)- (j,Bg) == (C1f,C2g) \1 f E D(A), \1 g E D(B). (7.1)

PROOF. If Cis H-smooth, then so is -C. Thus, due to the symmetry of the hypothe-
ses in A and B (except for a sign), it is enough to prove the existence of the limits
s-limt---+±oo e1,Ate-1,BtEB ( J). We treat the case t ~ +oo. By virtue of Proposition
2.2 it suffices to show that s-lim t---++oo e 1Ate -iBt EB ( J) f exists for a set V of vec-
tors f which is dense in R( EB ( J)). We take for V the set of vectors having compact
support in J with respect to the spectral measure of B; so f E V if and only if there
exists a bounded closed subinterval J == [a, f)] of J with EB(J)f ==f. We shall show
that, for such vectors f,

s-lim EA(J)eiAte-iBtf exists (7.2)


t---++oo
and
(7.3)

which implies the existence of O(A,B; EB( J) ). Throughout the proof we assume that
f is a fixed vector in V.
(i) Let us first verify (7.2). We set X(t) - eiAte-iBt. For 0 < s < t < oo and
g E D(A) one has:

(g. EA ( J) [X (t) -X ( s )]f) - (EA ( J)g, [X (t) -X ( s )]EB ( J)f)

= 1td~ (EA(J)g, X( T)EB (J)f) dT = 1td~ (EA(J)e~iATg, EB (J)e~iBTf) dT

(AEA( J)e~>ATg, EB (J)e~iBTf) _ (EA( J)e~iATg, BEE (J)e~iBTf)] dT


1
= i1 [

= i 1t(C1EA(J)e~'ATg, C2EB (J)e~iBTf) dT.


ASYMPTOTIC COMPLETENESS 319

By using the Schwarz inequality in 1-{ and then in £ 2 ( ( s, t) ), one obtains

where the last inequality follows from Proposition 6.11 since C 1 EA(J) is A-smooth.
Thus

IIEA(J)[X(t)- X(s)Jfll == sup j (g, EA(J)[X(t)- X(s)]f) I


gED(A),IIgll=l

(7.4)

Since c2 is B-smooth on J, one has f~oo IIC2EB(J)e-iBTJII 2dT < oo, hence (7.4)
implies that { EA ( J)X (t)f} is strongly Cauchy as t ~ +oo. This proves (7.2).
(ii) The proof of (7 .3) is based on the fact that, for each z E C \ ffi.:

(fED). (7.5)

In our applications, Eq. (7.5) will be satisfied since f will belong to Hac(B), hence
e-iBtf ~ 0 weakly as t ~ +oo [see Proposition 5.7(a)], and because (A- z)- 1 -
(B -z) - 1 == (B- z )- 1 (B- A) (A- z )- 1 will be co1npact (we usually assume B- A to
be A-compact). However, these special assumptions are not necessary, because (7.5)
is satisfied under the hypotheses of the proposition; this will be seen in part (iii) of the
proof, but we first show that (7 .5) implies (7 .3).
We use the following observation. Let J - [a, (3] be a closed subinterval of J
such that EB(J)f == f. Let Y be a closed contour in the complex plane, oriented
as in Figure 7.1 (a), containing J in its interior and intersecting the real axis in two
points belonging to J (say at distance c5 > 0 from a and from (3). Then the operator
(B- z)- 1 EB(J) belongs to B(H) for each z on Y, with IJ(B- z)- 1 EB(J)II
[dist (Y, J)] - l . In particular the integral

(7.6)

exists in B(H). Thus, for each g E H, the following integral exists:

(7.7)
320 FURTHER TOPICS IN SCATTERING THEORY

[remember that f - EB(J)f], and the integrand is a holomorphic function in the


domain C\[ a, j3]. The contour of integration Y can be deformed (by keeping its points
of intersection with the real axis fixed) without changing the value of the integrals in
(7 .6) and (7. 7). In particular

-f j{ (g, (B- z)- f)dz


7rZ y
1
= -
1
{
21rZ } y c:
(g, (B- z)- 1 f)dz

if Y E (c > 0) is as indicated in Figure 7.1(b). By letting c tend to 0, one finds by taking


into account Stone's Formula (4.80) that

1
- { (g, (B- z)- 1!) dz- (g, E 8 ((a- 6, (3 + 6))!) = (g,J).
27rZ }y

Since this identity holds for each g E H, one has

(7.8)

If his a vector such that EB(J)h == 0, then the function z ~ (g, (B- z)- 1 h) is
holomorphic in [C \ JR] U J, in particular in the interior of Y, and one obtains

(7.9)

The condition EB (J) h == 0 is equivalent to [I- EB ( J)] h - h; hence (7. 9) is satisfied


for each h E R( I - EB (J)), in other terms one has

(7.10)

By using (7.8) we may write

[I (7 .11)

By using (7.10) (for A in place of B) we have

(7.12)

By combining (7.11) and (7.12) one finds that

hence

(A z)- 1 ]e-iBtfll dz.


(7 .13)
ASYMPTOTIC COMPLETENESS 321

Yc:

a a b
(
a
a
~ ~
(3
I )b tc
J=(a,b)
(a) (b)
Figure 7.1

By (7.5), the integrand in (7.13) converges to zero as t---+ +oo for almost all z on
Y. It is majorised by

{sup II(B- z)- 1EB(J)II +sup II[I- EA(J)](A z)- 1 ll}llfll


zEY zEY

== {[dist(Y,J)]- 1 + [dist(Y,IR\J]- 1 }IIfll,

which is a finite constant. The Dominated Convergence Theorem (Proposition 1.9),


applied to the integral in (7.13), implies that II [I- EA(J)]W(t)fll ---+ 0 as t---+ +oo~
which proves (7.3).
(iii) Finally let us prove (7 .5) from the hypotheses of the proposition. For any g E H
we may write by using the assumption (7 .1 ):

l\g, [(A- z)-1- (B- z)-1]e-iBtf!l


== I((A- z)-1g, (B z)(B- z)-1e-zBtf)
-((A- z)(A- z)-1g, (B- z)-1e-iBtf!l
I((A z)-1g, B(B- z)-1e-iBt f)- (A( A- z) 1g, (B z)-1e-?Btf) /
== I (C1 (A z)- 1g, C2(B- z)- 1e-zBtf) I
< IIC1 (A- z)- 1 llllgiiiiC2(B- z)- 1e-iBtfli·

Since c1 is assumed to be A-bounded, one has IIC1(A- z-)- 1 11 C1 < 00. Hence,
remembering that f == EB ( J) f:

II[(A- z)-1- (B- z)-1]e-iBtfll


== sup l\g, [(A- z)-1- (B- z)-1]e-iBtf)/
gE'H, !1911=1

< c1IIC2(B- z)- 1e-iBtfii == c1IIC2EB(J)e-iBt(B- z) 1.fll.

Thus it suffices to show that limt--++oo IIC2EB(J)e-iBt(B- z)- 1.fll == 0. We have


for 0 < s < t < oo:

IIC2EB(J)e-iBt(B- z) 1fll2- IIC2EB(J)e-iBs(B- z)-l.fll2

-Jsr dd IIC2EB(J)e-iBT(B- z)-lfll2dT,


T
322 FURTHER TOPICS IN SCATTERING THEORY

hence it is enough to show that (djdT)IIC2EB(J)e-iBT(B- z)- 1!11 2 belongs to


L 1 ( (0, oo) ). Now

d~ JJC2EB(J)e~iBT(B- z)~lfJJ2
== dd (C2EB(J)e-~BT(B- z)-1!, C2EB(J)e-iBT(B- z)-1!)
JT

- i(C2EB(~l)e-iBTB(B- z)- 1 f, C2EB(J)e-~B 7 (B z)- 1!)


- i(C2EB(J)e-iBT(B z)-1!, C2EB(J)e-iBTB(B- z)-1 f).

Thus

Id~ JJC2EB(J)e~iBT(B- z)~l.fJJ2j


< 211C2EB(J)e-iBT(B- z)- 1fiiiiC2EB(J)e-iBTB(B z)- 1 !II· (7.14)

Since C 2 is B-smooth on J, each of the norms on the right-hand side of (7.14) belongs
to L 2(JR) (as functions ofT), so by the Schwarz inequality the right-hand side of (7.14)
belongs to L 1 (JR). D

We now give an application of Lavine's Theorem to non-relativistic Hamil-


tonians. We again consider the simple situation of §5 .4.1, i.e. H 0 == P2 is the free
Ha1niltonian in L 2 (1Rn) and H == H 0 + V(Q) the total Hamiltonian. We have seen in
Section 5.4 that, if n > 3, it is rather easy to prove the existence of the wave operators
== s-liint---+±(X) eiHte~Hot. Sufficient conditions for the existence of these limits
have been given in Proposition 5 .20.
The concept of asymptotic completeness was introduced in §5 .5 .1. One speaks of
strong asymptotic completeness if R(O_) == R(O+) == Hac(H) and O"sc (H) - 0.
In what follows we denote by { E( ·)}the spectral measure of Hand by { E 0 ( ·)} that
of H 0 . We shall consider only potentials satisfying the condition (C) of §6.4.1 (page
288). Then the Hamiltonian has no singular continuous spectrum, o- ac (H) == [0, oo)
and the eigenvalues of H in (0, oo) form a discrete set. Hence (0, oo) \ o-p(H) is the
union of at most a countable number of open intervals { Jk} 1, and E( (0, oo) \ o-p(H))
_ ~kE(Jk) is the projection onto Hac(H). Let V be the union of all subspaces of
the form E ( ~!) H, where J runs over all open subintervals of (0, oo) \ o-P (H) such that
J (the closure of J) still belongs to (0, oo )\o-p(H); then Vis a total subset in Hac(H).
Similarly the union of all subspaces of the form Eo ( J)H, where J runs over the same
set of open intervals, is a total set in Hac ( Ho) H.
To obtain the existence and the completeness of the wave operators, it suffices to
show that s-lirnt---+±(X) eiHte-iHotE0 (J) and s-limt---+±(X) eiHote-iHt E(J) exist for
each open subinterval J of (0, oo) \o-p(H) such that J also belongs to (0, oo) \o-p(H).
For this we apply Lavine's Theorem (Proposition 7.2). We must factor the operator
V(Q) into V(Q) == C 1 C2 , where C 1 must be H-bounded and H-smooth on J, and
C2 must be H 0 -bounded and Ho-smooth on J (with J an interval of the indicated
1
We do not take into account the fact (not proved in this text) that ap(H) n (0, oo) is empty.
ASYMPTOTIC COMPLETENESS 323

class). The simplest way is to use a symmetrical factorisation of V. Let us define


ul,U2: JRn ~ JR by Ul(x) == IV(x)i 1/ 2 and U2(x) == IV(x)l 1/ 2 signV(x). Then
V(Q) == U1(Q)U2(Q). Let us take C 1 == U1(Q) and C2 == U2(Q). We apply Propo-
sition 6.16 (first with W == U1 and V equal to the considered potential, then with
W U2 and V equal to zero); it follows that U1 isH-smooth on J (hence also on
2
J) and that U 2 is H 0 -smooth on J (hence also on J) if U1 is of the following form:

with v > 1/2, w 1 E L 00 (JRn) and w2 E LP(JRn) for some p > nand> 2. One ob-
tains in this way the following (non-optimal, of course) result on the existence and
completeness of wave operators:
Proposition 7.3. Let H 0 == P2 and H- H 0 V(Q). Assume that Vis a real func-
tion oftheform V(x) == (1 lxi)- 1 - 6 [voo(x) vq(x)] with 6 > 0, V00 E L 00 (JRn)
and Vq E Lq(JRn)for some q > n/2 (and q > 1) and that
(i) ifn == 1, 2 or3: V andr(8Vj8r) belong to L 2(JRn) L~(JRn), andr 2(8 2Vj8r 2)
belongs to L 2(JRn) L 00 (JRn),
(ii) ifn > 4: V and r(8V/8r) belong to LP(JRn) L~(JRn) and r 2(8 2Vj8r 2) E
LP(JRn) L 00 (JRn) for some p E ( n/2, oo ).
Then the wave operators 0± == s-limt-t±oo eiHte-iHot exist and satisfy strong asymp-
totic completeness.
PROOF. The conditions (i) and (ii) guarantee that (C) of §6.4.1 is satisfied. The as-
sumption V(x)- (1 lxl)-(l-o) [v 00 (x) vq(x)] (q > 1, q > n/2) implies that

IU1(x)l = IV(x)l 1; 2 < (1 lxl)-(l-o); 2[w1(x) 1v2(x)],

wherew 1(x) -lvoo(x)j 112 E L 00 (JRn) andw2(x) == lvq(x)l 1 12 E LP(JRn) withp == 2q


(hence p > n and p > 2). The conclusions of the proposition follow from the remarks
made before its statement; the assumptions of Proposition 6.16 are satisfied with v ==
1/2 6/2 > 1/2. D
EXAMPLES. We give some very simple examples of spherically symmetric potentials
for which the hypotheses of Proposition 7.3 are satisfied, in the case n == 3 (it is clear
that the spherical symmetry is not essential).
(a) The Yukawa potential V(r) == re-J-Lrjr with r E JR, M > 0 (take for example
6 == 1, V00 (r) == 0 and Vq(r) == r(1 r) 2e-J-Lrjr with q == 2).
(b) The potentials V(r) - r(1 r)-f3 with r E JR and (3 > 1 (take for example
6 - (3 - 1, vq ( r) - 0 and v oo (r) == r).
(c) The potentials rr-!3 with r E JR and 1 < (3 < 3/2 [take for example b - (3 1,
V00 (r) == [(1 r)f3r-(3 for r > 1, V00 (r) == 0 for r < 1, Vq(r) == 0 for r > 1,
vq(r) == r(1 r)f3r-f3 for r < 1, with q E (3/2, 3/ (3)].
(d) Let us briefly consider the Coulomb potential V(r) - [r- 1 . This is a limiting
case of the potentials considered in (c). The condition (C), and hence the assumption
2 If (6.86) is satisfied for an interval J, it is also satisfied if J is replaced by any subset of J.
324 FURTHER TOPICS IN SCATTERING THEORY

(i) of Proposition 7.3, is satisfied for the Coulomb potential (see page 78). On the other
hand one knows that the wave operators do not exist for this potential. In fact it is not
possible to write this potential in the form (1 r)- 2 v[v 00 (r) + vq(r)] with v > 1/2
(only with v == 1/2), because V (r) does not decay sufficiently fast at infinity.

7.2 Flux and scattering into cones

When discussing scattering cross sections at the beginning of Section 5 .6, we pointed
out an important relation [Eq. (5.82)] between the probability of scattering into a cone
C (with vertex at the origin) and the integrated flux over a distant spherical surface in
this cone, viz. 3
(7.15)

Here Sd denotes the sphere of radius din JRn centred at the origin, and Ut - e-iHt.
The nun1ber P ( h; C) is the probability that the scattering state Ut [2_ h associated with
an initial state his localised in the cone Cat timet == +oo:

(7.16)

The flux <I> o (f) of a function f across a hypersurface 0 is given by the following
expression, in which n( x) E JRn is a - suitably oriented- unit vector orthogonal to 0
at the point x E 0 and dO" (x) denotes the surface element at x:

CI>a(f) = 2 ~ fo.t(X) it( X)· "\l.f(X)drJ(X). (7.17)

In the present section we intend to give a proof of the relation (7 .15) for some of
the simple scattering systems considered in Section 5.4, under suitable assumptions
on the cone C, the state vector g _ fl_h and the potential V. We are not aware of a
sufficiently simple proof in the literature. We shall use here results obtained in Chapter
6 by the conjugate operator method to establish propagation properties for certain
observables. A propagation estimate for an observable, represented by a self-adjoint
operator A, is an estimate on the large time behaviour (in the Heisenberg picture) of
this operator in the subspace of scattering states of the Hamiltonian H. A scattering
state will at large times be localised in a part of configuration space where the potential
is very s1nall. So for example, thinking in terms of classical mechanics, its position
x( t) at timet will be given as x( t) ~ x0 + (p/ rn) (t- t 0 ) for some constants x0 and t 0
(tf is the momentum and m the mass). Hence, as t ~ +oo, one will have jx(t) l-1-1: ~
[(IP1/m)t]-K: for any /'1; > 0. The quantum-mechanical analogue is an estimate of the
form I!Ut IQI-K:Utfll < cjtK: for large t and for some constant c; such an estimate
3
We assun1e here that the probability of scattering into a truncated cone Cp is the same as that of
scattering into C, i.e. that P(h;Cp) = P(h;C) for any p > 0. This is frequently the case, see for example
Proposition 5.27.
FLUX AND SCATTERING INTO CONES 325

is expected to be (approximately) satisfied provided that the scattering state f has a


non-zero minimal velocity. In Lemma 7.6 we shall prove a precise inequality of this
type for certain scattering states f.
In what follows we write g for the vector fl_ h occurring in (7 .15) and (7 .16) [with
g E Hac(H)]. In fact the existence of the wave operators will not be used explicitly
and is thus not required to hold.

7 . 2 . 1. We begin by fixing convenient notations (Figure 7.2) and then restate in these
notations the result to be proved. For n > 2, let C be a fixed closed cone in JR.n with
vertex at the origin (C -1 JR.n). We assume that C is convex and with smooth lateral
surface. 4 If 0 < p < d < oo we use the notations Cp,d for the doubly truncated cone
Cp,d - {x E C j p < Jxl < d} and Cp for the simply truncated cone corresponding
to d - oo, i.e. CP - {x E C llxl > p }. The boundary 8Cp,d of the doubly truncated
cone Cp,d is the union of two spherical surfaces :Ep == { x E C llxl == p} and :Ed ==
{x E C llxl - d} and of the lateral surface Ap,d- {x E 8C I p < Jxl < d}. We also
set Ap - { x E 8C llxl > p}. For x E 8Cp,d, we let fi( x) be the outward unit vector
at the point x. Finally we use the following notation for the integrated flux through a
hypersurface 0:
Io(g, T) = loo if>o(Utg)dt. (7 .18)

...

. . . .. ....
0
.:: .... .
..... . . . .
......

Figure 7.2

In the preceding notations the result to be established (sometimes called the Flux-
Across-Surfaces Theorem in the literature) is as follows: if g E Mto
(H), then one
has for each T E JR.:

(7.19)

where Xc( Q) is the multiplication operator in L 2 (JR.n) by the characteristic function


Xc of the cone C.
4 Our considerations can be applied under considerably weaker assumptions. For exan1ple in n = 3
dimensions, it suffices to assume that C is convex and such that its lateral surface can be divided by N
straight lines, each passing through the vertex at x = 0, into N smooth disjoint surfaces (N < CX> ). By
considering disjoint unions or differences of such cones, the final result of this section - the Flux-Across-
Surfaces Theorem- is easily seen to hold for a quite general class of cones.
326 FURTHER TOPICS IN SCATTERING THEORY

Conditions on the potential V and on the state vector g that are suitable for our
proof of (7.19) will be given further on. For the moment we just assume that H ==
H o + V (Q) is self-adjoint on D (H 0 ) and that all states in Hac (H) are scattering
states under Ut, i.e. that Hac(H) M~(H).
Let us give an outline of the proof of (7.19). Assume that g E D(H) n Hac(H).
Then, since g is a scattering state, we have for any p > 0:

Furthermore

Now
d - 2 d -
dT llxcp.d(Q)UTgll - dT (UTg, Xcp,d(Q)UTg)
== i[(HUTg,Xc p.d (Q)UTg)- (UTg,Xc p,d (Q)HUTg)]
= i[(HoUTg,XcP)Q)UTg)- (UTg,XcP)Q)HoUTg)],
(7.21)

since Xc (Q) commutes with the potential. The expression on the last line has the
p.d - -
form i[(Hof, Xc p.d (Q)f)- (f, Xc p,d (Q)Hof)], with f == UTg. By writing Ho == -~
and then using the divergence theorem to transform the resulting volume integral into
an integral over the boundary of Cp,d, one arrives at

i [\Hof, Xcp.d (Q)f) - (f, Xcp,d (Q)Hof)]


= -i JCp,d
[f(X)L:lf(X) f(X)L:lf(X)] dnx

= -i J V[f(X)V
Cp,d
f(X)- f(X)Vf(X)] dnx

-i { [f(X)-9 f(X)- f(X)V.t(X)] ·it(X)dO"(X) = -if>acp,d(f). (7.22)


Jacp,d
The application of the divergence theorem above is justified under the condition that
f is a function of class C 2 . However, by an approximation argument, the validity of
(7.22) can be shown to hold for all f E D(H0 ). 5 Then (7.21) leads to the relation

(7.23)
5
Iff E V(Ho), choose a sequence {gk} in S(IRn) such that s-limk-+oo 9k == (Ho + l)f. Then
fk :== (Ho + 1)- 1 gk belongs to S(IRn) and s-limk-+oo fk == f, hence also s-limk-+oo Hofk == Hof.
By Eq. (7.22) we have i[ (Hofk, Xc p,d (Q)fk)- (fk, Xc p,d (Q)Hofk)] == -<I>ac p, d (fk) for each k, and
the validity of (7 .22) for f follows upon letting k -----7 oo in this equation, using the fact that <I> ac p, d (fk)
converges to <I> ac p, d (f) by (7 .37) with g == fk.
FLUX AND SCATTERING INTO CONES 327

Assuming that, as a function ofT, each of the fluxes occurring on the right-hand side
is integrable on [t, oo ), insertion of (7 .23) into (7 .20) shows that, for any p > 0:

(7.24)

We shall prove in §7 .2.5 that under convenient conditions on V and g:

<I> A p, d(UTg) E £ 1 ([1, oo)), d-tCXJ


lim IA p .d(g, t) exists, and lim lim IA p, d(g, t)
t-t+(X) d-t(X)
== 0.
(7.25)
It then follows from Eq. (7.24) (since its left-hand side does not depend on d) that
lin1 d-tCXJ I ~d (g, t) exists, and we get

To arrive at (7 .19), it remains to know the following results (they will be established
in §7.2.4):

and lim I~b(g, t)


t-t+(X)
== 0 for each b > 0, (7.26)

lim I ~d (g, t) is independent of t. (7 .27)


d-t(X)

It is instructive to consider the interpretation of (7 .25)- (7 .27) in terms of propaga-


tion properties of scattering states. Here (7 .25) essentially means that, at large times
t, the flux through the lateral surface A of a cone C is small (sufficiently small that
its integral up tot == oo is finite). One can expect this to be the case because at large
times the particle current is essentially radial, so that the probability of crossing the
conical surface A should be small. For (7 .26) we observe that at large times the scat-
tering states are localised at very large distances from the co-ordinate origin, which
should imply that the flux through any bounded hypersurface will be small at very
large times. Finally, during any finite time interval [T1 , T2 ], a scattering state will
have a very small probability of being localised in a neighbourhood of lxl == oo, so
that the flux through a very distant spherical surface, integrated over [T1 , T 2 ], should
be very small. This is expressed by (7 .27).

7 . 2 . 2 . The flux <I>o (f) is given in terms of the values of the function f and its first
order derivatives on a hypersurface of dimension n - 1, which is a subset of 1R 11 of
Lebesgue measure zero. Thus <I> 0 (f) does not have an immediate Hilbert space mean-
ing: given a vector in £ 2 (JR. n), the various representative functions of its equivalence
class will give different values for the flux, and for most (if not all) representatives,
the quantity <I>o (f) in Eq. (7.17) will not even make sense. The expression for <I>o (f)
has a well defined meaning for functions f of class C 1 . Thus it is possible to asso-
ciate a flux <I> 0 to those elements of L 2 (JR.n) that contain a representative of class C 1
[this condition on f is satisfied for example if the Fourier transforms off and of P1 f,
j - 1, ... , n, belong to L 1 (1Rn), see Proposition 5.6].
328 FURTHER TOPICS IN SCATTERING THEORY

A convenient description of the flux in terms of linear operators between Hilbert


spaces can be obtained from the following result characterising the restriction of
smooth functions to hyper-surfaces. We denote by ro the linear mapping f r-+ f [0
defined on the set of continuous functions f: JRn ~ C, where f I0 stands for the
restriction of a function f to the hypersurface 0. We shall write (P) for the operator
(I+ P 2 ) 112 . Clearly V( (P)) == V(IPI) and (P)- 1 E B(H).
Proposition 7 . 4. Let 0 be one of the sets ~b, Ap,d or Ap, with b E (0, oo) and with
0 < p < d < oo. Then there is a constant co < oo such that the following inequality
holds for all functions j, rp: JRn ~ c of class C 1 satisfying
( i) f defines a vector in L 2 (JRn) belonging to the domain D ( (P)) of the operator (P),
(ii) rp and its first order derivatives are bounded, and rp(x) == 1 for all x E 0: 6

(7.28)

Furthennore there is a finite constant /'1;, depending on n, such that

Vb > 1. (7.29)

The conditions imposed on f and rp in this proposition guarantee that the left-
hand side of (7.28) is well defined and that the right-hand side is finite. A proof of
the proposition can be found in the Appendix to the present chapter (page 374). We
1nention that more general results can be obtained by using the theory of relatively
smooth operators (see Proposition 7.16 and the comments preceding it). For example
ro can be defined for more general hyper-surfaces 0, and the condition f E V( (P))
can be weakened to f E V( (P)v) for any 7/ > 1/2 [the operator (P) can be replaced
by (P)u on the right-hand side of (7.28)]. An illustration is presented in Problem 7.3.
We now consider some consequences of Proposition 7 .4. First it is clear that CAp,d
can be chosen such that
CAp,d < CAp \f d > p. (7 .30)
1
Next, if hE L 2 (JRn) is such that the equivalence class of (P) - h contains a function
f of class C 1 , then (7.28) with rp 1 shows that

llro!IIL 2 (0) < coii(P)fll == collhll, (7.31)

1
where 11·11 denotes the norm in L 2 (JRn). This means that ro(P) - may be interpreted
as a bounded operator from the domain V :== { h E £ 2 (JRn) I h == (P) f for some
f E L 2 (JRn) n C 1 (JRn)} to £ 2 (0). It is easy to see that this operator is closable,
and we denote its closure by fo. It turns out that foE B(L 2 (JRn), £ 2 (0)), i.e. fo
is defined on all of £ 2 (JRn). To see this, we must know (cf. §2.4.1) that the above
domain Vis dense in L 2 (JRn), i.e. that g _l V implies g == 0. Now, if g _l V, we
have in particular (g, (P) f) - 0 for all f E S(JRn ). This shows that g belongs to the
do1nain of the adjoint of (P) ls(JRn) and that [(P) ls(JRn )] *g 0. Now the operator (P)
6
We write £ 2 (0) for the L 2 space on 0 with respect to the measure detennined by do-(-).
FLUX AND SCATTERING INTO CONES 329

is essentially self-adjoint on S (JRn) [by Proposition 2.29 applied in L 2 (IRn) ], hence


g E D ( (P)) and (P) g == 0. Since ( P) is invertible, it follows that g == 0. From (7 .31)
we obtain the following estimate for the norm of r 0 :

(7.32)

view of the preceding result, we may express the flux [Eq. (7 .17)] as follows,
where rp is any function satisfying the conditions of Proposition 7.4 [observe that
rof == rorpf rorp(Q)f]:

<I?o(f) = 2~~ l[rotpf](X)nk(X)[roPkf](X)do-(X) (7.33)

n
= 2~L (ryotp(C})f,nk(·)roPkf)£2(0)
k=l
n

= 2~ L (ro(P)tp(C})f, nk(·)roPk(P)f) PC or (7.34)


k=l

The last expression for if?o(f) is meaningful for all fin D(H0 ), and by the Schwarz
inequality one has 7
n
2
ji!!o(f)\ < 2llfoii II(P)rp(Q)fll L IIPk(P)fll (7 .35)
k=l
< 2ncbii(I +Ho)rp(Q)fiiii(I Ho)fll- (7.36)

Another useful relation concerns the difference of fluxes: Iff and g belong to D(H0 ),
then
n

if!o(f)- <I>o(g) == 2~ L (ro(P)j, nk(·)foPk(P)(.f- g))


k=l
n

+ 2~ L (ro(P)(.f- g), n~,-(·)foPk(P)g).


k=l
As above, this leads to the following inequality:

Iif?o(.f) - if!o(g) \ < 2ncb [II (I Ho)/11 + II (I Ho)gll] I (I Ho)(.f- g) II·


(7.37)

The following property of ri:d will be useful:

s-lim ri;d == 0. (7 .38)


d-HXJ

7
For the second inequality, use (7.32) and observe that
2
II (P)hll :S II (P) 2 hli = II (I + Ho )hli and
IIPk(P)hll :S II(P) hll = II(I + Ho)hll.
330 FURTHER TOPICS IN SCATTERING THEORY

Indeed, let us choose a function rp on ]Rn of the form rp( x) == r;( XI d)' where r; is a
smooth function such that r;(x) == 0 for lxl < 114 and r;(x) == 1 for lxl > 112. By
using (7.28) and (7.29), then (7.148) and commuting Pk through rp(Q), one gets for
any hEH:
n

II hll < ~II(P)rp(Q)(P)- hll < ~llrp(Q)(P)- hll


1 1
~ L 11Pkrp(Q)(P)- hll
1

k=l
n

== ~llrp(Q)(P)- 1 hll ~ L lirp(Q)Pk(P)- h 1


i[(818xk)rp](Q)(P)- hli
1

k=l
n -+ -+

K L 1
IITJ(~)Pk(P)- h- ~ [(8/8xk)TJ](~)(P)- 1 hll·
k=l

Now (7 .38) follows because r;( QI d) converges strongly to zero as d ---7 oo and be-
cause 11[(8l8xk)r;](Qid)ll < ll(818xk)77lloo < oo.
We finally note the following simple consequences of (7 .28):

and S -lim
d---+00
rA p,
d == r A p
. (7 .39)

7 . 2 . 3. We present here so1ne preliminary results and some simple commutation rela-
tions that will be needed in the proof of the Flux-Across-Surfaces Theorem; some
of these are also of interest by then1selves. The following notations will be used:
(P) == (I+P 2 ) 112 , (Q) == (I+Q 2 ) 1 12 , (x) == (I-~Ixl 2 ) 1 1 2 (xEJRn) and81 - 818x1
(j == 1 , . . . , n).
For v > 0 we set

Dv(H) == {gEH lg -'lj;(H)(Q)-vh forsomehE7-iandsome1j;EC0 (JR)}.

Clearly Dv(H) C V(H) == V(Ho), and Dv(H) is dense in 1-i - L 2 (JRn). Observe
also that Dv (H) ~ Da (H) if v > C5. The subset of functions in Dv (H) having support
in (0, oo) (with respect to the spectral measure of H) will be denoted by D;!; (H):

Dt(H) == {g E 1-i I g == 1/J(H) (Q) -vh for some hE 1-i and some 1/J E C 0 ((0, oo))}.

For the usual Schrodinger Hamiltonians used in scattering theory, D;!; (H) is a dense
subset of Hac (H). 8 We shall prove the Flux-Across-Surfaces Theorem forgE D;!;(H)
with 7/ > 2.

Lemma 7 . 5 . Assume that the potential V satisfies IV(x) I < c(x) -land 18j V(x) I < c
for SOTne constant c and j - 1, ... 'n. Let g E vt(H). Then:
8
To see this, one uses the fact that CY ac (H) = [0, oo) and that H has no singular continuous spectrun1
and no positive eigenvalues (see page 289). Without using the absence of positive eigenvalues, one may
replace the condition 'ljJ E C 0 ((0, oo)) in the definition of D;t;(H) by 1/J E C 0 ((0, oo) \CYp(H)). See
also Problem 7 .4.
FLUX AND SCATTERING INTO CONES 331

(a) One has Utg E V(Ho) n V(Qj) n V(QjPk) n V(PkQj) for each t E JR., and

PjUtg = UtPjg- utfot u_T[(aj V)(Q)JUTgdT, (7.40)

QjUtg = UtQjg + 2utfot u_TPjUTgdT. (7.41)

(b) For j == 1, ... , n let ~(j) E !3(H) be given by ~(.i)- J~TU_ 7 [(8j V)(Q)]U7 dT.
Then
Qj 1 (j)
(Pj-2t)Utg==- tUt[Qj+2~
2 Jg. (7.42)

PROOF. (i) For fixed t JR. and 1/J E C 0 ((0, oo)), we set 1/Jt(A) == e-i>..t1jJ(A). Then
E
1
1/Jt E Co((O, oo)), hence Utg 1/Jt(H)(Q)- 1 h E Vi(H) if g == 1/J(H)(Q) - h. So,
1
to know that Utg E V( Qj ), it is enough to show that Qjr;(H) (Q) - belongs to B(H)
for each r; E C 0 ((0, oo) ). This follows from the fact (shown below) that the operators
Qjr;(Ho)(Q) - 1 and Qj[r;(H)- r;(Ho)] belong to B(H).
1
We have seen in Example 6.23 that Qjr;(Ho)(Q) E B(H). Next let us set L ==
(H +w )- , L 0 == (Ho + w )- and 19().) == r;(A - - w ), where w is a positive number
1 1 1

such that H w > I [so the spectrum of H is contained in [-w + 1, oo)]. We observe
that L 0 - L == L 0 V L (by the second resolvent equation) and that 19 E C0 (JR.) with
support in (0, oo). Then, by using (4.57):

TJ(H) - TJ(Ho) = rJ(L) -1'J(Lo) = ~ joo J(t) [eiLt - eiLot] dt


v 21r -()()
== _1_ Joo J(t)eiLot [e-iLoteiLt _I] dt
v2i -()()
== ~. Joo dtiJ(t)eiLot1t e-iLosLoVLeiLsds. (7.43)
v 21r -()() 0

Now QjeiLo(t-s) == eiLo(t-s) [Qj+2(t-s)PjL5], with PjL5 E !3(H) [for arigourous


1
argument, consult Example 6.22]. By using the assumption that \V(x) I < c(x) - , one
obtains

The right-hand side is finite because iJ E S(JR) and QjLo(Q) -IE !3(H) (see Example
6.23). This completes the proof of the inclusion Utg E V( Qj) in (a).
The inclusion Utg E V(QjPk) is obtained similarly by showing that the operator
QjPkr;(H) (Q) - 1 belongs to B(H) for each r; E C 0 ((0, oo) ). First we get by simple
commutator calculations that

QjPkr;(Ho) (Q) - 1 == PkQjr;(Ho) (Q) - 1 + i6jkr;(Ho) (Q) - 1


== Pkr;(Ho)Qj(Q)- 1 + 2iPkPj[(8jr;)(Ho)](Q)- 1 ibjkr;(Ho)(Q)- 1 ,
332 FURTHER TOPICS IN SCATTERING THEORY

which clearly belongs to B(H). To see that QjPk[r;(H)- r;(H0 ] E B(H), we use
(7.43) to write

Pk[rJ(H)- r;(Ho)] == -=i_loo


~ -oo
dtiJ(t)eiLofe-iLos PkLo VLeiLsds.

As before, on sees that QjPk[r;(H) -r;(Ho)] E B(H). The verification of the inclusion
Ut g E V (Pk Q1 ) is left as an exercise.
(ii) We next indicate a formal proof of (7 .40) and (7 .41 ). These equations are par-
ticular cases of the following formal identity:

It suffices to observe that, for A== Pj, one has [H,A] == i(8jV)(Q), whereas with
A == Qj: [H, A] - -2iP1 . We note that each term on the right-hand sides of (7.40)
and (7 .4 ) is well defined as a vector in 7-i, and we present a careful derivation of these
two equations in the Appendix to this chapter (page 375).
(iii) By (7.40) and (7.41) we have

(7.44)

Now, commuting P1 through U7 in the last integral [using (7.40)] and then integrating
by parts, one obtains

Upon inserting this expression into (7 .44 ), one arrives at (7 .42). D

We say that a function V: JR.n - t JR. is of class Woo if it is infinitely differentiable,


satisfies IV (x) I < c0 ( x) - and if, for each integer £ > 1, all partial derivatives of V
1

of order£ are bounded by cg(x) 1 -£ for some constant C£ (so for example if£ - 2:
181 8~.; V(:r) I < c2 (x) - 3 for j, k == 1, ... , n). In §7.2.4 and §7.2.5 we always assume
V to belong to this class. In §7 .2.6 we also consider potentials with local singularities,
in particular the Coulomb potential.
FLUX AND SCATTERING INTO CONES 333

Lemma 7 . 6 . Assume that the potential Vis of class W00 • Let~ E C 0 ((0, oo) ), v > 0
and E > 0. Then there is a constant c such that the following inequalities are satisfied
for all t E IR: 9
II(Q)-v~(H)Ut(Q)-vll < c(1 + ltl)-v+c, (7.45)

II(Q)-v~(H)Ut(I + D 2 )--v/ 2 ll < c(1 + itl)-v+c. (7.46)

In particular, if g- ~(H)(Q) -vh E Dt(H) and jL > v, then


II (Q) -JLUtgli < c(1 + ltl)-v+c llhll· (7.47)

PROOF. We denote by { E ( ·)} the spectral measure of H and set Rz - (H z) 1


and Z == (I+ D 2 )- 1 12 . If~ > 1/2 and J is a bounded closed interval in (0, oo ), then
by Stone's Formula (4.80):

Z~E(J)Z~- - . s-lim { Z~<[R.\.+ic-


1
27r~ c---++0 } J
R>.-ic]Z~<d>..
= ~ { [ z~<(H - A- i0)- 1 z~<- z~<(H- A+ i0)- 1 z~<] dA. (7.48)
27r~ JJ

Here z~(H- A± i0)- 1 z~ denote the norm limits of z~R>..~iEz~ as c ~ +0. These
limits exist, see point (d) on page 289, and the interchange of the limits with the in-
tegral can easily be justified since the preceding limits exist uniformly in A E J. If
~ > N + (1/2), then z~(H - A ± io)- 1 z~ are N times continuously differen-
tiable on J with respect to A, the derivative of order f < N being the strong limit of
(
f! z~ H - A iE)- (£+ 1 ) z~ as c ~ +0, see Corollary 6.19 and the ensuing remarks.
Eq. (7.48) may be formally written as

Z~E(dA)Z~ == ~ [z~R>..+ioz~- Z~R>..-ioz~J dA == ((A)dA,


27r~

with the obvious definition of ( (A). Now let ¢ E C0 ((0, oo)) and denote its support
by J. Then 10

tNz~< ¢(H)UtZ" = tN 1 e -iAt ¢(>..)((A) d)... =iN1[:). : e-"At] ¢(>..)((A) dA.

Assume that~ > N + (1/2), so that (is N times continuously differentiable on J and
one may integrate N times by parts in the last expression. The boundary terms arising
from the integrations by parts are all zero because ¢ and all its derivatives vanish at the
endpoints of J. In this way one finds, by taking norms, that (1 +!tiN) II z~¢(H)UtZ~ II
is majorised by a finite sum of terms of the following form

9 See (6.19) for the definition of the operator D.


10
A more careful derivation of this equation is achieved by introducing a sequence {IIr} of partitions of
J, by writing ¢(H)Ut as a limit of finite sums as in Proposition 4.10 and using (7.48) on each subinterval
(sk-l, sk] appearing in these partitions.
334 FURTHER TOPICS IN SCATTERING THEORY

where cis some constant (depending on N) and k, f E {0, 1, 2, ... , N}. Since J is
bounded, these integrals are finite. Hence, by taking f); == N + 1 and by using the
inequality (1 +a )N < 2N (1 +aN) for a > 0, one has

IIZN+ 1¢(H)UtzN+ 1 II < c1(1 + ltiN)- 1 < 2Nc1(1 + !ti)-N


for some constant c1.
Next let ?jJ E C0 ((0, oo)) and set ¢(A)== (A 2 + 1)N+ 1?jJ(A). Then
(Q) -(N+1)?/J(H)Ut(Q) -(N+1) == xjyzN+1¢(H)UtzN+1xN

with XN == (I+ D 2 )(N+ 1)1 2 (H + i)-(N+ 1)(Q)-(N+ 1). Since XN is a bounded


operator (see Problem 6.10), it follows that

Now (7.45) is obtained by interpolation: in Example 6.32 take B - C == (Q) and


X== ?jJ(H)Ut. Taking N > v, Eq. (6.133), with f); N + 1 and e == v, gives

II (Q) -v ~J(H)Ut(Q) -vii < c311 (Q) -(N+ 1)?/J(H)Ut(Q) -(N+ 1) llv/(N+ 1)
< C4(1 + It!) -Nv /(N+1) == C4(1 + It!) -v+[v /(N+1)].
Choosing N so large that v / (N + 1) < E, one obtains (7 .45). The estimate (7 .46) is
obtained by the satne interpolation argument but with C == (I + D 2 ) 112 instead of
c == (Q). 0
For later reference we add the following simple facts. (i) If h E D( H), then one has
for each t E JR.:

II(I + Ho)Uthll < No(H;h) :== II(I + Ho)(H i)- 1 IIII(H i)hll· (7.49)

Note that N 0 (H; h) is finite since (I+ H 0 )(H- i)- 1 E B(H).


(ii) If cp is a bounded function of class C 2 , with bounded first and second order deriva-
tives, then the following identities hold on V (H 0 ):

Hocp(Q) == cp(Q)Ho- (~cp)(Q)- 2i(Vcp)(Q) · P (7.50)

== cp(Q)Ho + (~cp)(Q)- 2iP · (Vcp)(Q). (7.51)

Formally this is easily verified by writing H 0 as -~ [acting on functions f(x)] and


using the Leibniz rule. See also Example 6.22.
(iii) As a consequence of (7 .50) and (7 .51) we have for z in the resolvent set of H:

[cp(Q), (H- z)- 1 ] == (H- z)- 1 [H, cp(Q)](H z)- 1


== -(H z)- 1 { (~cp)(Q) + 2i(Vcp)(Q) · P}(H- z)- 1
== (H- z)- 1 {(~cp)(Q)- 2iP · (Vcp)(Q)}(H z)- 1 . (7.52)
FLUX AND SCATTERING INTO CONES 335

7 . 2 . 4 . We now prove (7.26) and (7.27) for V of class Woo and g E Vi(H). We first
show that I Eb ( Ut g) exists for each b > 0 and each t E ffi.. For this we use (7 .36) with
cp E C 0 (Rn) such that cp(x) - 1 if lxl < b + 1 and cp(x) == 0 if lxl > b + 2. Then,
taking into account (7 .49):

By (7 .50) we have

(I Ho)cp(Q)UTg == cp(Q)UTg + cp(Q)[H- V(Q)]UTg


- [(~cp)(Q)]UTg- 2i[(Vcp)(Q)]. PUTg,
and it suffices to show that the norm of each term on the right-hand side admits a
bound of the form canst· (1 + ITI)- 1 -s for some constant that may depend on b, and
for some E > 0. For this we write g == ~(H)(Q)- 2 h and set ~ 1 (A) == A~J(A) and
~ 2 (A) == (A- i)~ (A). From the propagation estimate (7 .45) we then get the following
inequalities:

< II(Q) cp(Q)IIII(Q)- UT~(H)(Q)- IIIIhJI < canst·(1 + ITI)- 312 ,


2 2 2
llcp(Q)UTgll
llcp(Q)V(Q)UTgiJ < IIVIJooJJcp(Q)UTgJJ < canst·(1 + ITJ)- 3 12 ,
2 2 2
llcp( Q)HUTgll == II (Q) cp( Q) II II (Q) - UT1/J1 (H)( Q) - llllhll < canst· (1 + ITI)- 312 ,
II [(~cp )( Q)]UTgll < II (Q) 2 (~cp )( Q) II II (Q) - 2 uT~J(H) (Q) - 2 llllhll
<canst· (1 + ITI)- 3 12 ,
II [ (aj cp) ( Q)] Pj u
T g II < II (Q) 2 (aJ cp) ( Q) II II (Q)- 2 PJ ( H - i) -
1
( Q)
2
11

Q) - UT~2 (H) (Q) -


2 2 3 2
x II ( 1111 hi I < canst· (1 + IT I) - / ,

2 2
since all occurring T-independent norms are finite (for II ( Q) - Pj (H-i) - 1 ( Q) 11 see
Example 6.26). We have thus shown that for each bE (0, oo ), there is a constant c(b)
such that
j<I>Eb(UTg)l < c(b)(1 + ITI)- 3 12 .
This also implies that lim t-+oo I Eb (g, t) == 0 and completes the proof of (7 .26)
We now verify (7 .27). It suffices to show that for any t 1 , t 2 E ffi. and each g E V (H):

This is easily obtained by using the Dominated Convergence Theorem. (i) By (7.34)
and (7.38), <!>Ed (U7 g) ~ 0 as d ~ oo for each fixed t. (ii) Since CEd < ~ < oo for
all d > 1 [see (7 .29)], we obtain from (7 .36), with cp 1, the following d-independent
bound for the integrand:

which is integrable on each finite interval [t 1 , t 2 ].


336 FURTHER TOPICS IN SCATTERING THEORY

2
7 . 2 . 5 . Here we prove (7.25), again for V of class QJ 00 and g == 1/J(H)(Q) - h in
Vi (H). Before producing the necessary decay estimates, we explain the idea of the
proof. \Ve observe that if 0 == Ap,d, then n(x) · x == 0 for all x E 0. Hence, instead
of (7.33), we may use the following expression for the flux across 0 == Ap,d:

L 1bo'Pf](X) nk(X) [ro(Pk- ~k )!] (X)dCY(X),


n

il>o(f) = 2 3t
k=l 0

where sis an arbitrary number. For f == U7 g and s == 2T, the subtraction of QI (2T)
from P should improve the decay rate, since (in classical terms) the momentum p
is expected to behave at large times T approximately like mx
IT and we have taken
rn == 112 for the mass.
Instead of (7.35) we now obtain the following bound for <I> Ap.d (U7 g ):
n

ICI>Apd(UTg)l < 2c,tii(P)tp(Q)UTgll L ii(P)(Pk- ~:)UTgjj. (7.53)


k=l

We shall choose the function rp such that its support belongs to a small neighbourhood
of the cone C. Such a neighbourhood has a finite extent in directions orthogonal to Ap,
so that one can expect that llrp(Q)U7 g/l decays as ITI- 112 due to the spreading effect
for wave packets in one dimension [see for example (5.64) for the case V == 0]. Since
(7 .25) involves only positive times, we assume in the following estimates that t > 1
[in which case 1lt < 21 (1 + t)]. We shall show that, given E > 0, there is a constant
c (depending on g) such that fort > 1 and j, k == 1, ... , n:

(7.54)

and
(7.55)

These inequalities imply that I<I> A p, d (U7 g) I < c' (1 + T) - 3 12 + 2s for some constant
c', all T > 1 and all d E (p, oo] [see also (7.148)]. This gives the first statement in
(7.25). The second one is obtained by means of the Dominated Convergence Theo-
retn: by using the preceding bound on I <I> A p, d (U7 g) I and the fact that <I> A p, d (UTg) -
<I> Ap (UTg) -<I> Ad (U7 g) ~ 0 as d ~ oo for each fixed T [see (7.39)], one finds
that limd--+oo IA p, d (g, t) == IA p (g, t). The third statement in (7.25) then follows since
IAP (g, t) < c" (1 + t)-l/2+2s.
In what follows we us the notation c, Cj and canst for various constants; these
numbers may depend on some of the involved parameters or on a vector in 1-{ but they
are independent oft (they are constant with regard to the variable t).
J
We first verify the estimates in (7 .55). Recalling that ~(j) == 0t TU- T ( Oj V) U7 dT
and that I(OJ V) (x) I < Cl (X)- 2' we obtain from the propagation estimate (7 .4 7) that
FLUX AND SCATTERING INTO CONES 337

Hence [see (7 .42)]

which gives the inequality (7 .55) for the first term on its left-hand side. To treat the
second one we set 9 == ( H + w) g E Vi (H), with w as in the proof of Lemma 7.5, and
then have [using again (7 .42)]

IIPk(PJ - ~: )Utgll < ;t [IIPx:UtQJ(H + w)- 911 + 2IIPkUtl";;(J) (H +


1
w )-
1
911].
(7.56)

To see that the norms on the right-hand side of this inequality are bounded by c(1 +t)c
for some constant c, we commute ( H + w) - 1 through Qj and ~(j) respectively. For
the first term, using (6.117), this leads to

with c 1 < oo because .9 E V( Qj) and g E V(Pj ). For the second norm on the right-
hand side of (7 .56) we get

I!PkUt~(j) (H + w)- 1 911 < IIPk(H + w)- 1 llll~(j) 911 + IIPkUt[~(j)' (H + w)- 1 ]911
<canst· (1 + t)E + !!PkUt(H + w)- 1 [H, ~(J)]g\1,

and it remains to check that IIPkUt(H + w )- 1 [H, ~(j)Jgll < canst· (1 + t)E. For this
we observe that

and, by using (7 .51) with cp == Oj V:

(H + w)- 1 12 [H, Y;;(J)]g = 1t TU_ 7 (H + w)- 1 12 [H, (81 V)(Q)]UTg dT

= 1tTU_ 7 (H +w)- 112 (8j6.V)(Q)U7 gdT

- 2i t
£=1
l t TU_ 7 (H + w)- 1 12 P£(8J8£V)(Q)U7 g dT.
0
(7 .58)

From the hypothesis on the decay of the derivatives of V and the propagation estimate
(7.47), we have li(fJj,6.V)(Q)UTgll + ll(8jO£V)(Q)UTgll < co(1 + T)- 2+c for some
constant c0 [remembering that g E Vi( H)]. Since the closure of (H + w )- 1 12 P£ be-
longs to B (H), one finds that the norm of each of the integrands in the final expression
in (7.58) belongs to £ 1 ( (0, oo) ), so that I (H + w )- 1 / 2 [H, ~(j)Jgll < c1 (1 + t)E for
338 FURTHER TOPICS IN SCATTERING THEORY

all t > 0. Hence IIPkUt(H + w)- 1 [H, ~(j)Jgll < c2(1 + t)E for all t > 0, and (7.55)
is cotnpletely verified.
Now we prove (7.54). Let [be a unit vector pointing into the interior of the cone
C. For a > 0 we denote by C(a) the domain in IRn obtained by translating 8 C along
[by distances less than a, more precisely C(a) x x
{ E IRn I == iJ + JL[ with iJ E 8C
and -a < f1, < a}. Let ebe a bounded smooth function having support in C(a). Then

By using the relation [Qk, U~t] - -2tU~tpk and then (7.55) we get, for some con-
stant c (with t > 1):

I (I+ It QI)U~tUtgll < llgll + 2t t II(~;


k=1
- Pk)Utgll < c(l + t)c. (7.60)

To estimate IIXc(a)(Q)U?(I +I[· Ql)- 111, we choose a Cartesian co-ordinate sys-


tem with the 1-axis is along ~-+' and we introduce the notation x
== ( x 1 , 1 ) with x
J 1 E lR 77 1 . Then { · Q == Q 1 • We use the following factorisation of e-iPit which
follows easily from (5.59) with n == 1: e-iPit == ZtGtZt, where

(Zth)(x) == /4t h(:i!) '

Since the operators Pk, k == 2, ... , n, commute with Q 1 , we then get

llxc(u)(Q)UP(I +It Ql)~ 1 11- IIXc(u)(Q)e~iPft(I +It Ql)~ 1 ll


-+ -1
- llxc(a)(Q)Gt(I + IQ11) II· (7.61)

For fixed x 1 E IRn- 1 , we denote by S (x 1 ) the intersection of C(a) and the line x 1 + s[,
where s varies over IR: S(x 1 ) == {x E IRn I x (x 1 , x') E C(a)}. Since Cis convex,
S(i!') is an interval of length 2a. Then, by the Schwarz inequality in L 2 (IR), we have
for hE S(IRn) and any t E IR:

Thus, by taking into account (7.61), we have fort > 1:


FLUX AND SCATTERING INTO CONES 339

Inserting (7 .62) and (7 .60) into (7 .59) leads to

\\8(Q)Utgll < c(l + t)- 1 / 2 +c, (7.63)

where c is some constant. In particular, for e == cp, with cp as in (7 .54) [choosing a


such that the support of cp belongs to C(a)], this gives the inequality (7 .54) for the first
term on its left-hand side.
Finally let us consider the second term on the left-hand side of (7 .54). The argument
here is similar to that applied in relation with (7 .56). We again write g == (H + w) -I g
and commute (H + w) I through cp( Q) in the term under consideration, using (7 .52):

n
2
< c(l + t)-I/ +c + \\Pk(H + w)-I [(6.cp)(Q)Utg- 2i L P£(0Rcp)(Q)Utg] 1\.
£=1

Since 6.cp and O£cp have support in C(a), we find from (7.63) that \\(6.cp)(Q)Utg\\ <
c(l + t)-I/ 2 +c and \\(8£cp)(Q)Utg\\ < c(l + t)-I/ 2 +c. Hence, by observing that
the closure of the operator Pk(H + w )-Ip£ belongs to B(H), we obtain the estitnate
(7 .54) for the second term on its left-hand side.

7 . 2.6 . So far we have proved the Flux-Across-Surfaces Theorem for smooth potentials
decaying like \xi-I or faster. We now show that this result remains true for potentials
with similar decay but admitting local singularities. This class of potentials includes
the Coulomb potential inn == 3 dimensions. We assume that V == VI+ V2 , where VI
is of class moo and V2(x) == (x) -aw(x) with a > 2 and wE £P(JRn) + L 00 (JRn),
where p == 2 for n == 2, 3 and p > n/2 for n~ > 4. We set HI == H 0 +VI and have
H == H 0 + V == HI V2 , and we shall use results from the preceding subsections for
the operator HI· As before, { Ut} denotes the evolution group associated with H, i.e.
Ut == e-iHt, and we set Ul == e-iH 1 t.
We first discuss the basic decay estimate. Let f E Vt (HI) with v E ( 2, a], and let
E < ( v - 2) /2. Then there is a constant c such that for all T E JR:

IIV2u;JII < c(l + \T\)- 2 -E. (7.64)

This is obtained by writing f == 1jJ (HI) (Q)- v h and


IIV2U;fll < \\W(Q)-a (HI+ i)-I (Q)al\11 (Q) -a 1/Jo(HI)U;(Q) -vh\\, (7.65)

where 1/Jo (A) == (A+ i)~(A). The operator in the first norm on the right-hand side is
bounded: the unboundedness of (Q) a is compensated by (Q) -a and that of lV (if HI
is not bounded) by the resolvent (HI + i)- 1 ; we refer to Example 6.26 for details.
By virtue of (7 .45) in Lem1na 7 .6, the last norm in (7 .65) admits a bound of the form
c(l + jT\)- 2-E forE< (v- 2)/2.
The decay estimate (7 .64) implies that, for f and E as above:

(7.66)
340 FURTHER TOPICS IN SCATTERING THEORY

Thus, by Proposition 5.16, the wave operators 0± == s-limt-'r±OCJ UtUlEac(HI)


exist. The fact that R(O+) == Hac(H) can be obtained as in §7.1.2 from Lavine's
Theorem (cf. Problem 7.5).
We shall use the following consequence of (7.64): Let f E Vt(H 1) with v E (2, u],
and let E < (v - 2) /2. Then there is a constant c such that for all t > 0:

(7.67)

Let us prove this inequality. First we have for t > 0:

IIUtD+f-Ut1fll = IID+f-UtUlfll = li.{XJd~u;u;jdTII <1=11VzU;flldT


<c 1=(1 + T)~Z~odT = c'(l + t)~J~c. (7.68)

Next, setting f 1 == (H 1 + i) f and using the intertwining relation 0+ (H 1 i) -I ==


(H + i)- 1 0+:
IIHo(UtO+f Ut1f)ii == IIHo(H + i)- 1UtO+f1- Ho(Hl + i)- 1Ulf111
< JIHo(H + i)- 1 [UtO+fi- Ulf 1]II+ IIHo[(H + i)- 1 - (H1 + i)- 1]Ulf1]II·
By (7.68) the first term on the bottom line is bounded by c(l + t)-l-c for some
constant c. By the second resolvent equation the second term on this line is equal to
IIHo (H +i) -I V2UlfJ II, which is bounded by c(l +t) - 2 - £ , see (7.64). This completes
the proof of (7.67).
We now verify the validity of the Flux-Across-Surfaces Theorem, Eq. (7 .19), for
the class of Hamiltonians H considered here, with g == O+f, where f E Vt(H 1)
for some v > 2. By using the fact that the distance between Utg and Ulf becomes
very small at large values oft [Eq. (7.67)], we shall find that Utg and Ulf have the
sa1ne probability of being localised in a cone C at t == +oo, and that the associated
integrated fluxes across :Ed become identical as d ~ oo. Assuming these results for
the moment, we then have

lim
t-'r+OCJ
llxc(Q)UtD+fll 2
- lim
t-'r+OCJ
llxc(Q)Ulfll 2
= lim r=ipz:,AUlf)dt
d-'TOCJ } T

= lim r=ipz:,AUtD+f)dt, (7.69)


d-'TOCJ JT
where the second equality is just the Flux-Across-Surfaces Theorem for Ul f obtained
before. Equation (7.69) shows that the Flux-Across-Surfaces Theorem is satisfied for
Utg UtO+f·
It remains for one to check the two asymptotic relations stated before (7.69). First
we have [as in (5.86)]
2 2
lllxc(Q)UtO+fll - llxc(Q)Ulfll 1 < 2llfllillxc(Q)UtO+fll-llxc(Q)Ulflll
< 2llfilllxc(Q)[UtO+f UlfJII ~ 0 as t ~ oo,
FLUX AND SCATTERING INTO CONES 341

which implies the first identity in (7 .69). The third one (asymptotic identity of the
integrated fluxes) is based on the observation that, by (7 .37) and (7 .67):

j<I>Ed(UtO+f) -<I>Ed(Ulf)i < 2n~ 2 [No(H; O+f) + No(H1;f)]


x ii(I+Ho)(UtO+f Ulf)il <c(l+t)- 1 - E ,
(7.70)

with some constant c that is independent of d [see (7 .49) for the definition of N 0 ( ·; ·)
and (7.29) for that of~]. This inequality allows us to use the Dominated Convergence
Theorem to obtain the third identity in (7.69), since by (7.38) both <!>Ed (UtO+f) and
il?Ed (Ul f) converge to zero as d -7 oo.
REMARK. We have proved the Flux-Across-Surfaces Theorem under conditions on
the scattering state g [g E Di (H) if V E 9J oo, and g E 0+ Dt (H 1) in the situation of
the present subsection, with v > 2]. In the short range case 11 one has g == n_ f, and
it is interesting to know conditions on the free initial state f under which this result
is true. For this one has to determine mapping properties of the wave operators 0±
and of the scattering operatorS. We refer to [11] for such results. In the first instance
(V E 93 00 ) the wave operators map Dt(H0 ) into D~,(H) for any v > v' > 0, so that
the Flux-Across-Surfaces Theorem holds for initial states f E Dt(H0 ) with v > 2.
For the potentials considered in the present subsection, with V1 == 0, the operator S
maps Dt(Ho) into Dt(Ho) for v E (0, u - 1), and the wave operators map Dt(Ho)
into Dt(H) for the same values of v. Since here g == O+Sf, with Sf E Dt(H0 )
(v > 2), it is then seen that the Flux-Across-Surfaces Theorem is satisfied for initial
states f in Dt (H 0 ) with v > 2 (this requires that u > 3 in the definition of V2 ).

7 . 2 . 7. To end this section we comment on the condition, stated after (5.82), that at
very large positive times the flux should be purely outgoing. This may be expressed
for example in the following form:

(7.71)

The proof of this relation is easily reduced to showing that

(7.72)

Indeed, we have seen in (7.69) above that the left-hand sides of (7.71) and (7.72) are
identical. The equality of their right-hand sides is obtained by essentially the same
argument after noticing that, by (7. 70):

To discuss (7. 72), it is convenient to express the flux across ~d in a slightly different
form. We notice that on ~done has n(x) == xjd, so that [by using (7.33) and the
11
Also in the long range case with an appropriate definition of the wave operators, see Section 5.8.
342 FURTHER TOPICS IN SCATTERING THEORY

definition (6.19) of the infinitesimal generator D of the dilation group]:

if>z;Ah) = 2 ~ t; hd ['Y-r;dh](X) nk(X)['Y-r;dPkh](X)da-(X)


= ~ ?R ~ hd ['Y-r;dh](X)['yz;dQkPkh](X)do-(X)

- ~ ~ ~ !z;}rz;dh] (X)[rz;d (4D + ini)h] (X) do-( X)


=~~ t{
k=I jEd
['Y-r;dh](X)['Yz;dDh](X)do-(X). (7.73)

This shows that the flux across L:d is determined by the self-adjoint operator D (mo-
mentum in the radial directions).
Let us first consider potentials V with VI - 0, in other terms short range potentials
decaying more rapidly than lxl- 2 . Then Ul == e-iHot == u?, and one can check the
validity of (7.72) (with Ul replaced by UP) by explicit calculations. This was done in
[4] in terms of eigenfunctions of the operator H 0 and in [3] by using the expressions
(7.73) for the flux and (5.59) for the free evolution.
For potentials V with VI 1- 0, such explicit calculations are not possible be-
cause they would involve the evolution group Ul. The fact that the flux of a scat-
tering state g E Hac (HI) is outgoing at very large positive times can be explained
as follows. Assume that the spectral support of g with respect to the spectral mea-
sure of HI is contained in some interval J == [AI, A2 ], with 0 < AI < A2 . Then
(d/dt)(Ulg, DUlg) == (Ulg, [H, iD]Ulg) > allgll 2 for some a > 0, as a conse-
quence of a strict Mourre inequality satisfied on J. This implies that (Ul g, D Ul g)
will be positive after some time t 0 .
The preceding conclusion can be obtained in a somewhat different way by con-
sidering the subspace M ~ corresponding to the negative spectrum of the operator D
(i.e. MD- ED((-oo,O])H). Then, under the evolution group {Ul}, all scattering
states of HI propagate away from this subspace at large positive times, more pre-
cisely liED(( -oo, O])Ulfll ----+ 0 as t ----+ +oo for each f E MOCJ(HI)· The proof of
this result is based on the following propagation estimate (combined with a density
argument): if z;, E > 0 and ~J E C0 ((0, oo)), there is a constant c such that

Vt > 0.

This estimate can be obtained by using the Mourre method as in Chapter 6 to show
that sup,\EJo,fL>O IIED((-oo,O])(H1- A- iJL)- 1v(I + IDI)-ull < oo, where the
notations are as in Theorem 6.18. Details can be found for example in [10].
TIME-INDEPENDENT SCATTERING THEORY 343

7.3 Time-independent scattering theory

In Chapter 5 we have developed the time-dependent scattering theory. This theory was
formulated in terms of the two evolution groups associated with the free Hamiltonian
H 0 and with the total Hamiltonian H; it involves as fundamental objects the wave
operators 0+ and 0_ and the scattering operatorS == n;n_. The time-independent
scattering theory (also called stationary state scattering theory) gives formulas for
these operators in terms of the spectral measures and the resolvents of H 0 and H. To
obtain these equations, one first expresses the wave operators as integrals over the time
variable t (Proposition 7 .8). Then one represents the evolution groups in the integrand
in terms of the corresponding spectral measures (by using the functional calculus);
the expressions obtained in this way contain the resolvents in the form indicated in
Corollary 5 We present here some interesting and useful formulas, and we refer to
Chapter 6 of [AJS] for additional material and more details.
Before entering into mathematical details, let us indicate formally how one ar-
rives at a time-independent formula for the wave operator n+ by following the pro-
cedure indicated above. In the situation considered in §5 .4.1 [H == H 0 + V ( Q),
H 0 == P2 , Ut == e-iHt, up== e-iHot] we haven+== s-limt---++ooUtUP. Ifthis
limit exists, it is also given as the limit of a temporal mean; we saw this in Chapter 5
J
where we considered temporal means of the type r- 1 0T ... dt. Here it is more con-
J 00
venient to take temporal means of the form E 0 e-st ... dt (E > 0). One then has
00
n+ == s-limc---++0 c fo e-stut UPdt. One replaces in this expression the operator ut
by Ut ==fiR eiAt E( dA), where { E(-)} denotes the spectral measure of H, then one
interchanges the order of integration:

D+ =s-limE
c---++0 } 0
roo e~ctu;u?dt =s-limE
c---++0 }
r>O dt }{ E(dA)eiAt~ctu?
JR
0
=s-limE { E(dA)
c---++0 } JR .J 0
rX) dte"At~ctut
By taking into account (5.7) one arrives at the following stationary expression for n+:

D+ =s-lim( -ic) { E(dA)(H0 - A- ic)~ 1 . (7.74)


c---++0 .J JR
The integral in (7. 74) is a spectral integral in the sense of Riemann (strong limit of
approximating sums), see the proof of the following lemma for a precise example.
In (7. 74) the spectral measure is on the left of the integrand ( H 0 - A - iE) - l [the
operators E(·) and (Ho -A- ic )- 1 do not commute]. A different expression for n+
that can be obtained similarly is n+ == s-lims---++O ic:,[IR(H- A- ic)- 1 E 0 (dA), see
Corollary 7.10 (here { E 0 ( ·)} denotes the spectral measure of the free Han1iltonian
Ho == P2 ).
One can see that the stationary expressions will involve spectral integrals with
operator-valued integrands. In Chapter 4 we have not given any criteria for the ex-
istence of such integrals. It turns out that such criteria are not necessary here: each
344 FURTHER TOPICS IN SCATTERING THEORY

of the occurring spectral integrals is equal to an integral with respect to a numeri-


cal measure (in fact Lebesgue measure), and the integrals of the latter type exist as
a consequence of the existence (and sometimes also the completeness) of the wave
operators. This is already well illustrated by the formal derivation of a stationary ex-
pression for n+ given above, and it can be seen in more details from the proofs given
further on in this section.
Lemma 7. 7. Let A be a self-adjoint operator and { EA (·)} its spectral measure.
(a) Let (a, b] be a finite interval and let { CA} AE(a,b] be a family of operators in B(H)
such that IICAII < M < oo for all A E (a, b]. Then J:
CA(A- A)EA(dA) == 0.
(b) Let {CA}AEIR be a family of operators in B(H) such that, for each finite inter-
val J, there exists a constant MJ E (0, oo) with IICA II < MJ for all A E J. Then
fiR CA(A- A)EA(dA) == 0.
PROOF. (a) Since AEA([A, A+ dA]) ~ AEA([A, A+ dA]), hence (A- A)EA(dA) ~
0, the stated result is formally evident. For a rigourous proof, let us consider partitions
IIN == {s 0 , s 1 , ... , sN; u 1 , ... , UN} of the interval (a, b] (see page 6). We set Jk ==
(Sk-l, sk] and we have to show that, for each vector f E H, the sequence of vectors
{ Lr=l Cuk (A- uk)EA ( Jk)f} converges strongly to 0 as N ---7 oo and IIIN I ---7 0.
Let us first observe that [see (4.42)]

II(A- Uk)EA(Jk)fll 2 = r IlL- ukl mJ(dp,) < lsk-


}Jk
2
Sk-11
2
r mJ(dp,)
}Jk
== isk - Sk-li 2 IIEA( Jk)fil 2·
By using this estimate, the triangle inequality (1.13) and the Schwarz inequality for a
sum, we obtain that
N N
II L Cuk (A- uk)EA(Jk)fll < M L II(A Uk)EA(Jk)fll
k=l k=l
N
< M L IIINI 112 (sk- Sk-1) 112 IIEA(Jk)fll
k=l
N N
< MIIINI 112
{ L(sk- sk-1)} 112
{ L IIEA(Jk)fll 2} 112
k=l k=l

This shows that the spectral integral fab C A(A - A) EA (dA) exists and is equal to 0.
(b) The spectral integral over an infinite interval is defined as a strong limit of a
sequence of spectral integrals over finite subintervals. Hence, by taking into account
the result of (a):

j~ C>JA- A)EA(dA) ;:}~~


b.......-:rCXJ
1b C;_(A- A)EA(dA) = ;.:}~~ 0- 0.
b.......-:rCXJ
D
TIME-INDEPENDENT SCATTERING THEORY 345

Proposition 7.8 . Let A and B be self-adjoint operators and Ut == e-iAt, Wt == e-iBt.


Assume that 0± n± (A, B; E{!cJ == s-limt-+±oo UtWtEf!c exist. Then one has
0
D+ =s-lim c: roo e-ctutWtE/!cdt
c---++0 } 0
and D_ =s-lim
c---++0
sf _ 00
ectutWtE/!cdt.
(7.75)
PROOF. If E > 0, the operator-valued function t ~ e-EtUtWt is strongly continu-
J
ous, and t ~ lle-EtUtWtll is integrable on [O,oo). Hence 0 e-EtUtWtdt exists
00

[Proposition 2.3(c)]. Since Ej0 e-Etdt == 1, one has for each f E R(E{!c):
00

II c: faoo e-ctutWtE/!cfdt ~ D+f II = II c: faoo e-ct[UtWt ~ D+ ]fdt II


< c: faoo e-ctii[UtWt ~ D+]flldt
~ s faT e-ctll [UtWt ~ D+] fll dt + sloo e-ct II [UtWt ~ D+]f II dt. (7.76)

Let b > 0 be fixed. One can choose T such that II[UtWt- fl+Jfll < b/2 for each
t > T, so that the second integral on the last line above is less than b/2 for each E > 0.
On the other hand

s faTe-ctii[UtWt ~ D+lflldt < c: faT [IJUtWtfll + IID+fll]dt < 2sTIIfll,


which is less than b/2 for each E E (0, Eo) if Eo == b(4TIIJII)- 1 . So the first identity
in (7.75) follows from (7.76). The second one is obtained similarly. D

Lemma 7.9 . Let A and B be self-adjoint operators and Ut - e-iAt, Wt == e-iBt.


(a) For each E > 0, the integrals below exist and belong to B(H), and one has the
following identities:
00

fa e-ctutWtdt = i l (A~ A+ is)- 1 EB(dA), (7.77)

~i r (A~ A~ ic)- 1 EB(dA).


J o ectutWtdt
-oo
=
}IR
(7.78)

(b) For all E > 0 and b > 0, the integrals below exist and belong to B(H), and one
has the following identities:

~A+ is)- 1 (A
0
roo dte-ctwtj dse 8 su;ws+t- r (B A- ib)- 1 EB(dA),
Jo -oo JIR
(7.79)
346 FURTHER TOPICS IN SCATTERING THEORY

PROOF. (i) The existence of the integrals and double integrals with respect to dt and
ds and the fact that they belong to B(H) is obtained by the arguments used in the
preceding proof. Formally the identities (7.77)- (7.80) are rather evident [follow the
arguments in the derivation of (7.74)]. We give here a detailed proof of (7.77); for
(7.78)- (7.80) one proceeds in the same way, but certain expressions are somewhat
longer to write out.
(ii) We fix a number ME (0, oo ), set EM == EB( (- 1\!I, M]) and consider partitions
TIN {s 0 , s 1 , ... , sN; u 1 , ... , uN} of the interval (- M, 1\1] (see page 6). We set
Jk == (sk-1, sk] and define, for f E H:

By using (1.28) it follows that:

r~ r~ N
IIJo e-ctutWtEMfdt-XrrNU)II < Jo dte-ct IIWtEMf- ~ e-iuktEB(Jk)!ll·
(7 .81)
Consider a sequence of partitions {TIN} with ITIN I -+ 0 as N -+ oo. Then the in-
tegrand on the right-hand side of (7 .81) converges to zero for each fixed t, and it is
majorised for all N by the following integrable function:

N
e-st[IIJII +II Ee-iuktEB(Jk)fii] < 2e-stllfll.l2
k=l

So, by applying the Dominated Convergence Theorem (Proposition 1.9), one obtains

(iii) By making the change of variables t ~ -t, one obtains with (5.8) that
N 0 N
XrrN(f) =~[co dtei(uk-ic)tUtEB(Jk)f = i ~(A-Uk+ ic)- 1 EB(Jk)f.

Hence

1 0
00

e-ctutWtEMfdt- s-lim i t(A


N-+~ k=l
uk + ic)- 1 EB(Jk)f

== ijM (A- A+ ic)- 1 EB(dA)f.


-M
12
= e-~xt, one has II I:f:=l e-~uk t EB ( Jk )!11 2 =
By (4.44), with <p(x)
= I:f:=lle-~ukti 2 (J, EB(Jk)f) = I:f:=l (J, EB(Jk)f) = (J, ENI f)= IIEM fll 2 :=:; IIJII 2 ·
TIME-INDEPENDENT SCATTERING THEORY 347

By letting lvf --? oo in this equation, one obtains

r=e-ctutWtfdt = s-lim ('oe-ctUtWtEMfdt


Jo M~+~Jo

==s-lim i!M (A-A+ic:)- 1 EB(dA)f i { (A-A+ic:)- 1 EB(dA)f. D


1\II~+~ -M }YR.
By combining Proposition 7.8 and Lemma 7.9, one obtains the following stationary
expressions for [2± (similar formulas are valid for f2±J:

Corollary 7.10. Let A and B be self-adjoint operators and Ut - e- 2 At, Wt == e-iBt_


Assume that fl± fl±(A,B;Ef!cJ == s-limt~±~ UtWtEf!c exist. Then

Sl+ =s-limic { (A-A+ic:)- 1 EB(dA)E~ (7.82)


s-----++0 } YR.

Sl_ =s-lim (-ic) {(A- A- ic:)- 1 EB(dA)E~. (7 .83)


c-----++0 } YR.

In what follows we consider the case in which B == H 0 - P 2 , A == H == H 0 + V


with V(H) == V(H0 ), where Vis a self-adjoint operator [for example V == V(Q)].
It is clear that similar formulas are true in the general case. We use the notations of
Section 5.4. In particular the operators u?
== e-iHot and Ut == e-iHt give the free
and the total evolution group respectively, and {Eo (·)} denotes the spectral measure
of Ho.

Proposition 7 . 11 (Lippmann-Schwinger equations) . Let H 0 == P 2 and H == H 0 +


V, where V == V* is H 0 -bounded and H -bounded. If the wave operators fl±
s-limt---+±~ eiHte-iHot exist, then

Sl+- I- s-lim { (H- A+ ic:)- 1 VE0 (dA)


s~+O }YR.

= I - s-lim { (H0 - A+ ic:)- 1 VSl+E0 (dA), (7.84)


s~+O }YR.

Sl_ I- s-lim { (H- A- ic:)- 1 VE0 (dA)


s~+O }YR.

= I - s-lim { (H0 - A- ic:)- 1 VSl_E 0 (dA). (7 .85)


c-----++0 } YR.

PROOF. Letc:- > 0. Since II(H- ,\ ic:-)- 1 11 < 1/c:-, Lemma 7.7 implies that for each
interval (a, b]:

1b(H- A± ic)- 1 (H0 - A)E0 (dA) = 0. (7.86)

On the other hand one has by Corollary 7.10:

Sl+ =s-lim Sl+,c,


s~+O
with n+,c = ic c
}I&
(H- A+ ic:)- 1Eo(dA) E B(H).
348 FURTHER TOPICS IN SCATTERING THEORY

ForME (0, oo) we can write, by using (7.86):

D+,e:Eo((O, M]) =is laM(H- A+ is)- 1 E 0 (dA)

= .laM(H- A+ is)- 1 (Ho A+ is)Eo(dA)-

- .laM(H- A+ is)- 1 (H- A+ is- V)Eo(dA)

-laM [I (H A is)- 1 V]E0 (dA)

{A1
= E 0 ((0, M])- Jo (H- A ic) - 1 V Eo (dA) . (7 .87)

Observe that VE 0 ( J) E B(H) if J is a finite interval, because V is an H 0 -bounded


operator. The assumption made on V has also allowed us to write H 0 E 0 (dA) -
HE0 (dA) VE0 (dA) for A E (O,M]. Since Vis also H-bounded, the closure of
(H -A+ic)- 1 Vis an operator in B(H). As M ~ oo, E 0 ((0, M]) converges strongly
to the identity operator I and O+,cEo ( (0, M]) converges strongly to 0+,£. Thus (7 .87)
J
implies that 01\1 ( H A + ic) - 1 V E 0 ( dA) is strongly convergent as M ~ oo, i.e. the
J
spectral integral 0 H - A + ic) - 1 V E 0 ( dA) exists, and one has
CX) (

[remember that E 0 (V) == 0 if V n (0, oo) == 0]. As 0+ == s-limc---++O O+,c' we have


verified the first identity in (7 .84 ).
The second relation in (7.84) is obtained similarly [using Proposition 5.15(a)]:

I= D'j_D+ =s-lims (oo dte-ctUP*UtD+


c-----++0 } 0

=s-lim is { (H0 - A is)- 1 E(dA)D+


c-----++0 } IR

=s-lim { (H0 - A+ is)- 1 (H- A+ is)E(dA)D+


c-----++0 } IR

= D+ +s-lim { (Ho- A is)- 1 VE(dA)D+,


c-----++0 } IR

and it now suffices to take into account the intertwining relationE (V) 0+ == 0+ Eo (V).
D

In order to represent the operator R S - I as a spectral integral, we write


R- n.+n-- n.::_n_ and use the following expressions for n.+n- and n.::_n_:
TIME-INDEPENDENT SCATTERING THEORY 349

Lemma 7 . 12. Assume that the wave operators exist and are complete. Then

dse 5 su;u~+t
00 0
0-'f-fL- s-lim s-lim (s + J)J { dte-ctuP*j
c-++0 6-++0 } 0 -oo

=s-lim s-lim (s + J)J { (Ho- A+ is)- 1 (H- A- i0)- 1E 0 (dA),


c-++0 6-++0 } lR
(7 .88)

dse 5 su;u~+t
0
n_::_n_ ==s-lim s-lim (c.- 5)5j dtectuP*jo
c-++0 6-++0 _ 00 -oo

=s-lim s-lim (J- s)J { (Ho- A- is)- 1 (H- A- i0)- 1E 0 (dA).


c-++0 6-++0 } lR
(7.89)

PROOF. We give the proof of (7 .88). By taking into account the intertwining relation
Utrl- == n_ up one obtains
00

0-'f-0_ =s-lim UP*Ut0- =s-lims { dte-ctu?*Ut0-


t-++oo c-++0 } 0

(7.90)

For c)> 0, we set n-,6 == 5f~ooe 6 tUtU?dt. Then II0-,611 < 5f~ooe 6 tiiUtUPiidt ==
c5 f~oo e6tdt == 1. One has

111
00

dte-ctUP*(0_- D-,5)UPJII < 1 00

dte-ctll(0_- 0-,5)UNII· (7.91)

For each fixed t the integrand on the right-hand side of (7.91) converges to zero as
c5 ~ 0, and it is majorised by 2e-stll/ll, which is independent of c5 and integrable on
[0, oo ). So the Dominated Convergence Theorem (Propositionl.9) implies that

{ dte-ctu?*(0_-~L,5)U2.f=O
00

s-lim 'VjEH..
6-++0 J0
Hence (7.90) may be written as

0-'f-0_ =s-lim s-lim Js


s-++0 6-++0 }
roo
0
dte-ctur
}_
dse r
5 su;u~+t·
00

The difference between this expression and the first line in (7 .88) resides in the term
00

-s-lim s-lim o f e-ctup* 0-,[JuP dt,


c-++0 6-++0 } 0
and we have to show that this expression is zero. For this we observe that, for fixed
c > 0:

ll
51ooe_stuO*n_
t Uodtjj <
,6 t _ c)j~ooe-stiiUO*n_
t ,6 t - 5looe-stdt == ~E '
Uojjdt <
0 0 0
350 FURTHER TOPICS IN SCATTERING THEORY

which converges to zero as c5 -+ 0. The second expression in (7.88) now follows by


taking into account (7. 79). 0

REMARKS. (a) The existence of the spectral integral in (7.88) (for c -/:- 0 and c5 -/:- 0)
follows from Lemma 7.9. The existence of the limits as c5 -+ +0 and c -+ +0 is a
consequence of the assumption that the wave operators exist and are complete.
(b) The expressions (7 .88) and (7 .89) look rather complicated because of the multiple
limits. The spectral integral on JR itself is already a double strong limit (first on finite
intervals, then as the length of these intervals tends to infinity); this is followed by
the two limits c5 -+ +0 and c -+ +0. It is important to note that the order of these
limits cannot be altered. Nevertheless the expressions (7 .88) and (7 .89), or rather their
consequence (7 .92) proved below, lead to a relatively simple and important expression
for the S-matrix S (A), as we shall see further on.

Proposition 7 . 13 . Let H 0 == P2 and H == H 0 + V, where V(H) == V(Ho) and


the self-adjoint operator Vis H 0 -bounded and H-bounded. For z E C\JR, set R~ ==
( H 0 - z) 1 and Rz == (H - z) - 1 . Assume that the wave operators exist and are
contplete. Then

R =S- I= s-lim s-lim { (R~-ic- R~+ic)(V VRA+ioV)Eo(dA). (7.92)


6---1-+0 }
c---1-+0 IR

PROOF. As in the proof of Proposition 7.11 we use the fact that VE0 (W) E B(1i)
if W is a bounded interval. So the spectral integral in (7 .92) is well defined on
each bounded subinterval of JR. Also, in the special situation considered here, where
H 0 == P 2 , the integral in (7.92) extends effectively only over (0, oo).
By using (7.88) and (7.89) one sees that

(7.93)

with
Xs,6(A) == (c + t5)R~-ic:RA+i6 + (c- t5)R~+ic:RA+i6.
One may re-express X s ,6 (A) by using the second resolvent equation (2.115) and the
first resolvent equation (2.93): if z, ( E C \ JR, one has

((- z)R2Rz == ((- z)R2R~[I VRz] == [R2- R~][I VRz]


== R2 R2VRz- R~[I- VRz] - R2- R2VRz- Rz.

By taking z ==A+ ic5, ( == A- ic or ( == A+ ic, one obtains

Xc:,6(A) == i{ R~-ic - R~-is VRA+i6- RA+i6}


i{ R~+is - R~+is VRA+i6 RA+i6}
- i[R~-is R~+is]- i[R~-is- R~+is]VRA+i6· (7.94)
TIME-INDEPENDENT SCATTERING THEORY 351

We now observe that, by Lemma 7.9:

r RLic:Eo(dA) = -i;·OO e-ctu?*u?dt = -i I,


1~ 0 E
o .

1 ~

This implies that for each


0
RA+ic:Eo

E
(
dA)

> 0:
== z. j_
-00
ec:t Ut0 *Ut0 dt ==
z
E:I.

s-lim 0 { [R~-ic- R~+ic]Eo(dA) 2


=s-lim - iO I= 0.
6---++0 1 ~ 6---++0 E
Thus, by inserting (7.94) into (7.93), one arrives at

R =s-lim s-lim( -iO) {oo [R~-ic- R~+•cJVRA+iOEo(dA)


c:---++0 6---++0 10
00

=s-lim s-lim {
c:---++0 6---++0 10
[Rf-ic- R~+icJVRA+io(Ho- A- iO)Eo(dA), (7.95)

-vvhere the second identity is obtained by taking into account Lemma 7.7 13 • By writing
Ho == H- V, one has VRA+i6(Ho -A- i5) == VRA+i6(H- A- i5- V) ==
V- VRA+i6 V, and (7.95) becomes identical with (7.92). D
Equation (7.92) indicates that the operator R(A) == S(A) - I must be determined
in a certain sense by the operators Tz :== V- VRz Vas the complex number z tends
to the spectral point A of H 0 . A rigourous derivation of this relation is presented in
Section 7 .4. Let us already give a formal argument here. Assu1ne that Tz is an integral
operator in the spectral representation of H 0 [see (5.68)], i.e. that there exists a family
{Tz(A, JL) }, with A, fL E (0, oo ), of operators acting in L 2(sn-l) such that

(UoTzf)>-. = fooo Tz(A, JL)(Uo.n_,dJL.


By observing that (UoR~-ic:f- UoR~+ic:f)f-L == -2ic[(A- tL) 2 + c: 2 ]- 1 (Uof)f-L, one
can write (7.92) in the spectral representation of H 0 as
oo loo
2ic
(f,Rg) =- lim lim
c:---++0 6---++0 o
dJL
l
o
dA (A
- fL
)2
+ E2
X ((Uof)f-L, TA+i6(JL, A)(Uog)A) L2(sn-1)·

By taking first the limitE --? +0 [remembering that limc:---++O c[(A- tL) 2 + c: 2 ]- 1 ==
n5( A - fL )], one obtains
00

(f,Rg) = -2nifo dA ((Uof)>-., TA+io(A, A)(Uog)>-.) L2(S" lJ·

One has I![R~-~c:- R~+~c:]VR>.+~oll = ll[R~-~c:- R~+uJVR~(H- i)R>.+~oll


13

~ 2lsl- 1 l!V(H- i)- 1 11/I(H- i)(H- A- i5)- 1 jj, and jj(H- i)(H- A- i5)- 1 11
= III+ (A+ i5- i)(H- A- i5)- 1 11 ~ 1 +(/AI+ 151 + 1)/51- 1 , which is a bounded
function of A when A varies on a finite interval J.
352 FURTHER TOPICS IN SCATTERING THEORY

J 00
By comparing with the formula (f,Rg) - 0 dA ((Uof)A,R(.A)(Uog)A)£2(sn-1)'
one concludes from this formal argument that R( .A) should be given by the expression
R(.A)- -2niTA+io(.A, .A).

7.4 The scattering matrix

For a simple scattering system the scattering matrix at energy .A is a bounded operator
S(.A) acting in L 2(sn- 1). We have already mentioned on page 244, without proof,
that there exist stationary expressions for this operator S (.A) in terms of the resolvent
(H z)- 1 of the total Hamiltonian H, of the interaction V and of certain transi-
tion operators mapping L 2(JRn) to L 2(sn- 1). We first discuss a useful class of such
transition operators and then derive a stationary formula for S (.A). We assume in this
section that n > 2.

7 . 4. 1. The operator S (.A) may be viewed as acting on the angular variables win the
Fourier transforms of the wave functions in spherical polar co-ordinates, with the ra-
dial variable k - vf\ fixed. Iff E £ 2(JRn) is a wave function and j its Fourier trans-
J
form, the restriction of to the sphere of radius vf\ is not well defined in general, be-
cause this sphere is a set of Lebesgue measure zero in JRn. For certain wave functions
it is possible to give a meaning to this restriction, as already discussed in §7 .2.2. If for
example f E L 1(JR n), then j is (equivalent to) a continuous function (see Proposition
5.6), and its restriction to a sphere in JRn is well defined. One could try to define in
this way an operator with domain L 2(JRn) n £ 1(JRn) and values in L 2(sn- 1); such an
operator would certainly be unbounded, and we are interested in constructing bounded
transition operators to find an expression for the bounded operator S (.A).
A simple method is as follows. We fix a function Win L 2(JRn). For f E L 2(JRn),
we set (Wf)(x) - W(x)f(x). Then Wf belongs to L 1(JRn), so that the restriction
of its Fourier transform to a sphere has a well defined meaning. This holds for each
f E £ 2(JRn). In this way one obtains a linear operator defined on £ 2(JRn) associating
to each f E £ 2(JRn) a continuous function of w (for fixed vf\), hence an element of
L 2(sn- 1). We denote this operator by Mw(.A). Thus

[Mw(.A)f](w)- 2-1/2_A(n-2)/4[FWf](vf\w)
= Tl/2 ,A(n-2)/4(2n)-n/2 { e-i~W·XW(X)f(X)dnx. (7.96)
}IRn
The factor 2- 112.A (n- 2)/ 4 has been introduced for convenience; one then has for all
f E 1J(W( Q)) [see (5.68)]:
(7.97)

The integral in (7.96) depends only on the equivalence class off, hence Eq. (7.96) de-
fines, for each fixed .A, a continuous function of w. The operator Mw(.A) is a bounded
THE SCATTERING MATRIX 353

operator from L 2 (1Rn) to L 2 (sn- 1 ), i.e. Mw(A) E B(L 2 (1Rn), L 2 (sn-l )) [see §2.1.7
for the notation]. Indeed:

I[Mw(A)f](W)I < Tl/2(27r)-n/2,\(n-2)/4 r


}IRn
IW(X)f(X)Idnx

< 2-1/2 (27r) -n/2 A(n-2) /411 WII £2(JRn) II J II £2(JRn).

Hence, denoting by 8n the surface area of the sphere S n- 1 :

j~n- 1 1 [Mw (A)f] (W)


2
11Mw(A)fll'i2(S" -1) - 1 dw

< en
2(21T)n
A(n- 2)12 IITiVII 2
£2(JRn)
11!11 2'

I.e.
(7 .98)
In fact M w (A) is even a Hilbert-Schmidt operator. Here we are considering Hilbert-
Schmidt operators between two different Hilbert spaces. If 1{ and JC are Hilbert spaces,
an everywhere defined operator A: 1{ --+ JC is said to be a Hilbert-Schmidt operator
k
if I: k II Ae k II < oo, where { e k} is an arbitrary orthonormal basis of 1{. If 1{ and JC
are £ 2 spaces, an operator A: 1{ --+ JC is a Hilbert-Schmidt operator if and only A
is an integral operator with square-integrable kernel; the proof consists in an evident
adaptation of that of Proposition 2.15.
In the case of interest here [H == L 2(1Rn), JC- L 2(sn- 1)], Eq. (7.96) shows that
M w (A) is an integral operator with kernel

(7.99)

Hence

IIMw(A)II~s = fs,_ dW j~ndnxia>JW,XW


1

== 1
2(21T)n
A(n-2)/2 r dw }IRnr dnx lW(x)l2 ==
j sn 1
1 A(n-2)/2e IIWII2
2(27T)n n £2(JRn)·
(7.100)

Some important properties of the operators M w (A) are given in the following propo-
sition. The adjoint of an operator between different Hilbert spaces has been described
in §2.1.7.

Proposition 7. 14. Let WE L 2(1Rn). For A> 0, let M 1¥(A): L 2(1Rn) --+ L 2(sn-l)
be the operator defined by (7.97). Then:
(a) Mw (A) is a Hilbert-Schmidt operator. The operator-valuedfunction A f--t Mw (A)
is continuous in Hilbert-Schmidt norm on (0, oo): IIMw(A) - Mw(t-t)IIHs --+ 0 as
tL--+ A.
354 FURTHER TOPICS IN SCAI\TTERING THEORY

(b) The adjoint Mw (A)* of l\1w (A) is a Hilbert-Schmidt operator from L 2(sn- 1 ) to
L 2 (1R 77 ) ; the operator-valuedfunction A r-+ Mw(A)* is continuous in Hilbert-Schmidt
nonn, and one has for each Borel set V in JR:

W(Q)*Eo(V)f = fv Mw(A)*(Uof)>.dA (7.101)

Remark 7. 15 . The integral in (7.101) is a vector-valued integral in L 2 (1Rn) [the range


of J\;fw(A)* belongs to L 2 (1Rn)]. Iff E S(JRn), then A r-+ (Uof);>.. is a continu-
ous L 2(sn- 1 )-valued function, and II(U 0 j);>..ll is of rapid decay as A --+ oo. On
the other hand Mw(A)* is continuous (in Hilbert-Schmidt norm), and IIMw(A)*II <
cA(n- 2 )/ 4 . Thus the integral in (7.101) is well defined with our conventions of §1.2.
PROOF OF PROPOSITION 7 .14. (a) In view of (7.100) it just remains to prove the con-
tinuity in Hilbert-Schmidt norm of A r-+ 1\1w (A). One has

!!Mw(A) ~ Mw(p.)llhs = fs,_ dw l, dnx la>.(W,X) ~ a~'(W,X)I .


1
2
(7.102)

w, x)- a~-t (w, x) --+ 0 as tL --+ A. Also, for tL E ( A/2, 2A]:


It is clear from (7 .99) that a;>.. (
2
Ia;>.. (w, x) -al-L (w, x) 1 < 2{ Ia;>.. (w, x) 12 ia~-t(w, x) 12}
< 2 (2A)(n-2)/21W(x) 12
- (27r)n '

which is an integrable function on sn- 1 X JRn. By virtue of the Dominated Conver-


gence Theorem, the expression (7 .1 02) converges to zero as IL --+ A.
(b) The first two assertions in (b) follow from the corresponding statements in (a)
by passing to the adjoint [taking into account (2.45)]. One could also repeat the argu-
ments given in the proof of (a) by observing that M w (A)* is an integral operator from
s
£ 2 ( n- 1 ) to £ 2 (JRn) with kernel

2-112 (21r) -n/2 A(n-2)/4w (x)eiV-\x·w.

It remains to verify (7.101). Let f E S(JRn) and g E V(W(Q)). Then

(Eo(V)J, W(Q)g) == (UoEo(V)f,UoW(Q)g)

=fv dA((Uof)A,(UoW(Q)g)>.)L (sn-l) fv dA((Uof)A,Mw(A)g)£2(sn-l)


2 =

= fv dA (Mw(A)*(Uof)>.,g) = (fv dAMw(A)*(Uof)>.,g). (7.103)

We have seen in Remark 7.15 that the integral Jv dA M w (A)* (U of);>.. defines a vector
in L 2 (1Rn). Denoting this vector by j*, Eq. (7.103) implies that E 0 (V)f belongs to
V(W(Q)*) and that W(Q)* Eo(V)J == j* == fv dA Mw(A)*(Uof);>... D

The functions W considered so far are not sufficiently general for our later appli-
cations. Indeed, if one assumes that W(x) == jxj-/3 in a neighbourhood of infinity,
THE SCATTERING MATRIX 355

then the condition that WE L 2 (1Rn) implies that {3 > n/2, i.e. a relatively rapid de-
cay of W at infinity. It is possible to associate transition operators MA (A) to a much
larger class of operators A acting in £ 2 (JR.n ). One would define in analogy with (7 .97):
MA(A)j- (U 0 Af);>.. for f E V(A). If the operator A* is locally Ho-smooth, then
MA (A) belongs to B(L 2 (1Rn ), L 2 (sn-l)) for almost all A> 0. We refer to §5 of [12]
or §17 .1.2 of [BW] for details, and we treat here only the case in which A is a mul-
tiplication operator, i.e. A == TV (Q), where W is a function belonging to the class
considered in Proposition 6.16 [hence W (Q) * is H 0 -smooth on each closed subinter-
val of (0, oo )]. For such a function W one can define Mw(A) so that its closure is
defined on each vector of £ 2 (JRn). We also denote this closure by M w (A).
In more detail, let us first recall the condition on W (Proposition 6.16):
(S) W(x) == [w1(x) + w2(x)](l + lxl) v with w1 E L 00 (1Rn) and w2 E LP(JR.n)for
some p satisfying n < p < oo and p > 2, and with v > 1/2.
This allows us to define Mw(A)j for f E S(JR.n) as in (7.96) or (7.97), because the
function W(x)f(x) belongs to L 1 (1Rn) nL 2 (1Rn) iff E S(JR.n) (see the proofbelow).
In the next proposition we state the essential properties of these operators M w (A).
Proposition 7 . 16. Let W be a function satisfying the condition (S). For A > 0, let
Mw (A) be the closure of the operator defined on S(JR.n) by Mw( A)j [Uo W( Q)j];>.,.
Then:
(a) Mw(A) is a compact operator from L 2 (1Rn) to L 2 (sn- 1 )), the operator-valued
function A~ Mw(A) is continuous in norm on (0, oo ), and

(7.104)

where C;>.. is some constant depending on A.


(b) The adjoint Mw (A)* of Mw (A) is a compact operatorfrom L 2 (sn-l) to L 2 (JR.n );
the operator-valuedfunction A~ Mw(A)* is continuous in nor1n, and one has for
each bounded closed interval J (0, oo ):

W(Q)*Eo(J)f = h Mw(A)*(Uof)>-.dA (7.105)

In particular

(7 .106)

Remark 7.17.. In contrast to the case covered by Proposition 7 .14, we do not derive
here a bound on the behaviour of I M w (A) I as a function of A as A --+ oo or A --+ 0. 14
To be sure that the integral in (7 .1 05) exists, we have assumed that V (here called J) is
a bounded set. We take into account that IIMw(A) I is a bounded function of A on each
closed interval [a, {3] E (0, oo) (a continuous function on a compact set is bounded).
In (7.106) we have assumed that (Uof);>.. 0 for A in some neighbourhood of A== oo
and in some neighbourhood of A == 0; thus the integral in (7 .1 06) extends effectively
only over some interval J as in (7 .1 05) [ J is such that Eo ( J) f - f].
14
If v 2:: 1 in the condition (S), we get from (7 .29) that sup A 2:: 1 c A < oo, see the proof that follows.
356 FURTHER TOPICS IN SCATTERING THEORY

PROOF OF PROPOSITION 7.16. We prove the boundedness of Mw(A) and the esti-
mate (7 .1 04) for its norm only under the assumption that v > 1 in the condition (S),
because this allows us to use results from Section 7 .2. The proof for 1/2 < v < 1
uses Stone's Formula (4.81) and the fact that W(Q)* is locally H 0 -smooth (Proposi-
tion 6.16).
(i) Assume that v > 1 in the condition (S). We show here that IIMw(A)fll < c!IJII
for some constant c (depending on A) and all f E S(JR.n). Iff E S(JR.n), then clearly
W1 (Q) f E £ 1(JR.n) n £ 2 (JR.n). By the Holder inequality (2.121 ), W2 ( Q) f also belongs
to £ 1(JRn) n £ 2 (1R 11 ) (because S(JR.n) _ Lq (JR.n) for all q E [1, oo ]). It follows that
W( Q)f E £ 1(JRn) n V( (Q)) for each f E S(JR.n ), so that Mw( A)f (Uo W( Q)f)A
is well defined for A E (0, oo) [we recall the notation (Q) == (I+ Q 2 ) 112 ].
As in §7 .2.2 we denote by ~~b the operation of restricting functions defined on JR.n
to the sphere l:b of radius bin JRn: ~~bf == Jl~. We then obtain from the definition
b
(5.68) ofUo:

(7.107)

Since T¥f belongs to V( ( Q) ), we obtain from Proposition 7.4, with the roles of P and
Q interchanged, that

where rp can be any function in C 0 (1Rn) satisfying rp(k) 1 in some neighbourhood


of the sphere l:vx and c c~vx is a constant depending on A. To estimate the norm of
(Q)cp(P)T¥(Q), we make use of the commutation relation [Qj, rp(P)] == i(ojrp)(P)
to get [see also (7.148) and use (1.14)]:
n

II (Q)cp(P)W(Q)fll
2
== llrp(P)W(Q)fll 2
L 11Qjrp(P)W(Q)fll 2
j=1
n n

< llrp(P)W(Q)fll 2 2 L 2
llrp(P)QjW(Q)fll + 2 L !!(8jrp)(P)W(Q)fll 2.
j=1 1

Each of the norms on the last line has the form II¢(P)(Ho + 1)- 112 W(Q)fll, with
CX) --+ --+ --+ --+

¢ E C0 (JRn) and W(Q) == W(Q) or W(Q) == QjW(Q). From the estimate (6.97)
we get that II(Ho + 1)- 112 W(Q)II < co[llw1!1CX) + !lw21ip] for some constant co. This
implies the boundedness of M w (A) on S (JR.n) and the inequality (7 .1 04) for its norm.
(ii) The compactness of Mw(A) can be obtained (for any v > 1/2) by decom-
posing Vll as w == WR + wA-, where WR(x) == W(x)x[o,R](ixl) and wA-(x) ==
W(:f) - HIR(x), R > 0. Then l\1w(A) == 1\!IwR + Mw_t. By Proposition 7.14
and (1.49), A1wR(A) is a Hilbert-Schmidt operator, and A r---:r MwR(A) is continu-
ous in Hilbert-Schmidt norm. The compactness of Mw(A) and its norm-continuity
as a function A then follow from Proposition 2.11 (d) provided that one knows
that limR-+CX) IIMw_t(A)II == 0, uniformly in A on each closed subinterval [a, f)]
of (0, oo ). For this we choose a number v' satisfying 1/2 < v' < v, and we set
THE SCATTERING MATRIX 357

wj (x) == (1 + lxl)-(v-v')wj (x) (j = 1,2). Then W (x) == (1 lxl)-v' [1n1 (x) +w2 (x)],
and by (7 .1 04) we have:

IIMwJ:t(A)II < A(n-


2)14 cA{ sup IW1(X)I +[r IW2(X)IPdnx]
11
P}. (7.108)
lxi>R llxi>R

Since jw1 (x)j < (1 + ixl)-(v-v') IJw1lloo and W2 E £P(JRn), the inequality (7.108)
implies that 111\1 w} (A) I --+ 0 as R --+ oo. The required uniformity in A follows from
the fact that sup AE[a,,6] C,\ < oo [see (7.29) which is satisfied if v > 1, or else part
(iii) of the proof of Proposition 7.4 given on page 375].
(iii) The first assertions of (b) follow from those of (a) by passing to adjoint.
Equation (7 .1 05) is obtained in the same way as (7 .1 01 ), see the proof of Proposition
7 .14(b): Eq. (7 .1 03) is still valid, then one takes into account the observations made in
Remark 7.17. D

To arrive at the stationary formula for S (A) we shall also need the following results:

Lemma 7. 18. Let W be a function sat;sfying the condition (S).


J
(a) Let A > 0 and assume that f: JRn --+ C is such that E COO (JRn \ { 0} ). Then
W(Q)*(Ho- A- ic)- 1f- W(Q)*(Ho A iE) 1f converges weakly in L 2 (1Rn)
to 21fiMw(A)*(Uof)A as E--+ +0.
(b) Let J ==[a, b] be ajinite closed interval in (0, oo) and C: J--+ B(H) a strongly
continuous operator-valued function. Then one has for each g E S(JR.n ):

{ C(A)Mw(A)*(Uog)AdA = { C(A)W(Q)*E0 (dA)g. (7.109)


JJ JJ
PROOF. (a) Let j E C00(1Rn\ {0}), g E 1{ L 2 (1Rn) and A,E > 0. We observe
that the Fourier transform of [( H 0 - A - iE) 1 - ( H 0 - A iE) 1 ] f also belongs to
COO (JR. n \ { 0}). Hence we have by virtue of (7. 106 ):
(g, W(Q)*[(Ho- A- ic)- 1 - (Ho- A+ ic) 1]f)
-1=(g, Mw(p,)* (Uo[(Ho A- ic) 1
- (Ho- A+ ic)- 1 lf) ~-')dt~

= 1 = [(p.- A ic) - 1 - (p.- A+ ic) - 1 J (g, Mw(l~) *(U of)~") df.L.


(7.110)

Now ~ ~ (g, Mw(~)*(Uof) 1L) is a continuous function compact support (see


Remark 7 .17), and [(~ - A - iE) - - (~ - A iE) - ] is a Dirac sequence as E --+ +0,
1 1

more precisely limc:-++O[(~ A- ic)- 1 - (1-L- A+ ic)- 1] == 21rib(~- A). Thus,


as E--+ +0, the last expression in (7.110) converges to 2ni(g, Mw(A)*(Uof)A)· This
proves the result of (a).
(b) The integrand on the left-hand side of (7 .1 09) is a strongly continuous vector-
valued function [with values in £ 2(JRn )], hence the integral exists by Proposition 1.5.
358 FURTHER TOPICS IN SCATTERING THEORY

Let us consider partitions liN == { s 0 , s 1 , ... , sN; u 1 , ... , UN} of the interval J (see
page 6) and set Jk- (sk- 1 , sk]. We can write
N

i C(A)Mw(A)*(Uog)>..dA- ~ ikC(A)Mw(A)*(U 0 g)>-.dA.

Let E > 0. If the partition liN is sufficiently fine, one has IIC(.A)- C(uk)ll < E for all
.A E Jk and for each k == 1, ... , N. Consequently, setting f(.A)- Mw(.A)*(Uog)A:

This implies that JJC(.A)j(A)dA == s-lim!TINI-+O I:~= 1 C(uk)JJkf(.A)d.\. Now, by


(7.105): JJkf(.A)d.\- JJk Mw(.A)*(Uog)AdA == W(Q)*E0 (Jk)g. So we have

N
r C(A)Mw(A)*(Uog)>-.dA- s-lim Lc(uk)W(Q)*Eo(Jk)g.
JJ ITINI-+0
k=1

The right-hand side is by definition equal to JJ C(.A)W(Q)*E0 (d.A)g. This proves the
existence of the spectral integral as well as Eq. (7 .1 09 ). D

Remark 7.19. Without proof we mention the following interesting result: If W 1 and
W2 are two functions satisfying the condition (S), then one has for each .A > 0:

(7.111)

7.4 . 2 . We now present the stationary formula for R(.A). We have seen at the end of
Section 7.3 that R(.A) should be given by the diagonal part of the operator -2ni TA+iO
in the spectral representation of H 0 . We recall that Tz - V- V(H- z)- 1 V. The idea
is to factor Vas V- UW, hence to write Tz == U[I- W(H z)- 1 U]W, and then
make use of the first factor U and the last factor W in this expression to introduce the
transition operators. One obtains in this way the representation

R(.A) -2niMu(.A) [I- W(H- A- i0)- 1 U]Mw(.A)*. (7.112)

Here Mw(A)* maps L 2 (sn- 1 ) into L 2 (1Rn), the operator I- W(H- .A- i0)- 1 U
acts in L 2 (1Rn), and Mu(.A) maps back from L 2 (1Rn) to L 2 (sn- 1 ). Evidently it is
necessary to impose conditions so that M u (A) and M w (.A) are well defined and that
W(H- .A- ic)- 1 U converges as c--+ +0 [the notation W(H- .A- i0)- 1 U is then
used for the limit]. This is the content of the following proposition. Let us recall that
THE SCATTERING MATRIX 359

a factorisation of V has already been used in the proof of asymptotic completeness in


Section 7 .1. One can take the same factorisation here (with different notations): one
can take for U the operator of multiplication by IV (x) 1 12 and for W the operator of
1

multiplication by v (x) 1 12 sign v (x)' as in Proposition 7.3. Other factorisations will


I 1

also be useful.

Proposition 7.20. Assume that the function V: JRn -+ JR satisfies the assu1nptions of
Proposition 7.3. Let U(x) and W(x) be real such that V(x) == U(x)W(x) and such
that U and W satisfy the condition (S) on page 355. Then:
(a) the wave operators 0± - 0± ( H, H 0 ; I) exist and are complete,
(b) for A E (0, oo) \ CJp(H), the operator R(A) is given by (7.112), with U == U( Q)
and VV == W (Q).

PROOF. The existence of 0± and strong asymptotic completeness have been obtained
in Proposition 7 .3. Moreover we know from Proposition 6.16 that the operators U ==
U(Q) and W == W(Q) are H-smooth. We have also seen that W(H- A- i<5)- 1 U
converges (in norm) as <5 -+ +0 if A tj:_ CJp (H), and that the convergence is uniform
on each bounded closed subinterval J of (0, oo) \ CJ P (H). Since the functions U (·) and
W(·) satisfy the condition (S), the operators Mu(A) and Mw(A) are well defined by
Proposition 7.16. These facts will be used for proving (7.112).
Let J be a bounded closed subinterval of (0, oo) \ CJ P (H). Let f E S (JR n). We use
the notations of Proposition 7.13 and write V- VR,\+i6 V == U(I - W R,\+i6 U) l;\1.
By using first (7.92) and then (7.109) we obtain

REo(J)f ==s-lim s-lim


E--++0 6--++0
j (R~-iE- R~+iE)(V- VRA+i6V)Eo(dA)j
J

==s-lim s-lim
E--++0 6--++0
j (R~-iE- R~+iE)U(I- WRA+i6U)M~v(A)*(Uof),\dA.
J

Hence, for any g E H:

(g,REo(J)f} ==

== lim lim
E--++0 6--++0
j (g,(R~-iE- R~+iE)U(I- ~VRA+i6U)Mw(A)*(Uof)A)dA
J

== lim lim
E--++0 6--++0
j J
\U(R~+iE- R~-iE)g, (I- WRA+i6U)Mw(A)*(Uof)A) dA.
(7.113)

Now assume that g E S(JRn ). Then

which is bounded by a finite constant for all A E J [by Proposition 6.16(b)]. Further-
more IIMw(A)*(Uof),\11 < c < oo for all A E J. As T¥RA+i6U converges in norm
when <5 -+ +0 (uniformly in A E J), the Dominated Convergence Theorem allows us
to take the limit <5 -+ +0 inside the .integral in (7 .113), i.e. to replace lV RA+i6 U in the
360 FURTHER TOPICS IN SCATTERING THEORY

integrand by l;\IRA+ioU. By using the estimate IIWRA+ioUII < c < oo for all A E J
and by writing again U(R~+ic:- R~-ic: )g == U(R~+ic:- R~-ic:) (I+ IQI)- 1 (I+ IQI)g,
one can once more apply the Dominated Convergence Theorem to take the limit
E ---+ +0 inside the integral. One finds in this way that

For each fixed A, one can evaluate the limit under the integral by using Lemma 7.18 (a),
provided that we impose the additional condition that g E C0 (JR.n \ {0}):

(g,REo)f) = 1 (2JriMu(A)*(Uog)>-.,(I WR>-.+ioU)Mw(A)*(Uof)>-.)dA

= -2Jri1 \(Uog)>-.,Mu(A)(I- WR>-.+ioU)Mw(A)*(Uof)>-.)dA.

Since (g,REo(J)f} == JJ((Uog),\,R(A)(Uof)A)dA, we now have

1 \(Uog)>-., R(A)(Uof)>-.)dA

= -2Jri1 \(Uog)>-.,Mu(A)(I- WR>-.+ioU)Mw(A)*(Uof)>-.)dA. (7.114)

We wish to deduce from (7.114) that R(A) == -2nil\!fu(A)(I- WRA+iO U)Mw (A)*
for almost all A E J. For this, let us first observe that, if J == [a, {3], one can repeat the
preceding arguments on each subinterval of the form [a, x], with a < x < {3. This
gives
X

( (Uog )_\,R(A)(Uof)A) dA

= -2Jri lx\ (Uog)>-., Mu(A)(I WR>-.+io U)Mw(A)*(Uof)>-.) dA,

and after differentiation with respect to x:

( (U og ),\, R(A) (Uof)A} == - 2ni( (Uog ),\, Mu (A) (I- W RA+iO U)Mw (A) *(U of),\).
(7.115)
This identity holds for almost all A in J (each side is in principle defined almost
everywhere). In other terms there exists a subset N(f, g) of J of Lebesgue measure
zero (and depending on f and g) such that (7.115) is satisfied for all A E J\N(f, g).
We now choose particular vectors f and g, motivated by the proof of Proposition
5.26. We fix an orthonormal basis { ek} of £ 2 (sn- 1 ) and a function ein C0 ((0, oo))
such that B(A) 0 on J. For each k == 1, 2, ... we define a vector ek in L 2 (JR.n)
by setting (Uoek)A == B(A)ek. Then iik belongs to C0 (JR.n \ {0}) provided that ek is
chosen to be of class coo. If we take in (7 .115) f == ek and g == ej we find that
THE SCATTERING MATRIX 361

or

(ej,R(J\)efJ == -21fi(ej,Mu(A)(I- WR-\+ioU)Mw(J\)*ek)

where Njk is a subset of J of Lebesgue measure zero. Let now N - Uj,k Njk. Then
N is a subset of J Lebesgue measure zero, and one has for all j, kEN:

(ej,R(A)ek) - -21fi(ej,Mu(J\)(I- W R-\+ioU)Mw(J\)*ek) \f)\ E J\N.

This implies that R(A)ek- -21fiMu(A)(I- WR-\+ioU)Mw(A)*ek for)\ E 1\N


and each k [apply Proposition 1.6 by taking into account that the set { e~} is total
in L 2(sn- 1)]. We have thus obtained the validity of (7.112), for almost all A, on
the vectors of an orthonormal basis of L 2 (sn- 1 ). By linearity, (7.112) holds on all
vectors of L 2 (sn-l ), for the same values of A. D

REMARKS AND EXAMPLES. (a) By the results of §5.5.2, the operator R(J\) is defined
only for almost all )\ E (0, oo). The right-hand side of (7 .112) however is defined for
every A > 0 that is not an eigenvalue of H. One can choose to identify R(J\) with the
right-hand side of (7.112) for each)\ E (0, oo )\CYp(H). This choice is compatible with
the scattering operator S.
(b) As a first example, let us assume that V(x) == (1 + lil)-l-c:[voo(i) + vq(x)],
with c > 0, V00 E L 00 (1Rn) and Vq E Lq(IRn) for some q > nl2 (n > 2), and that
V satisfies the assumption (i) or (ii) of Proposition 7.3 (depending on the value of 1~).
In this case we take a symmetric factorisation of V, as in the proof of Proposition
7.3: U(x) == IV(i)l 112 and W(x) == IV(x)l 1 12 sign V(x). The operators Mu(A) and
M w (A)* in (7 .112) are compact and norm-continuous as functions of A (Proposition
7.16). On the other hand we know that W(Q)(H A- i0)- 1 U(Q) is continuous in
norm as a function of A, away from the eigenvalues of H. By assuming the choice
specified in (a) above, this implies that, for each)\ E (0, oo) \ CTp(H), R(J\) is a com-
pact operator in L 2 (sn-l ), and the operator-valued function A r-----7 R(J\) is continuous
in norm on (0, oo) \ O"p(H).
(c) Above the only assumption on c was that c > 0. This covers potentials decay-
ing like lxl-!3 with (3 > 1 and which may have local singularities, hence essentially
the class of potentials for which we have shown the existence (and completeness)
of the wave operators. We know that, if (3 > (n + 1)12, then R(A) is a Hilbert-
Schmidt operator for almost all A (see Section 5.7). So let us assume that V(x) ==
(1 + !x!)-f3[voo(i) + vq(x)] with (3 > (n~ + 1)12 and q > n, q > 2, and again
that V satisfies the condition (i) or (ii) of Proposition 7.3. Let us choose a number
v > 112 such that f3 - v > n 12 and take u (x) == (1 xl ) - v [v 00 ( x) + v q ( x) J
1

and W(x) == (1 jx!)-(!3-v). Both U and W satisfy the condition (S) on page 355.
Mu(A) is a compact operator depending continuously (in norm) on A (Proposition
7.16), and Mw(A)* is a Hilbert-Schmidt operator depending continuously on A in
Hilbert-Schmidt norm. Thus (7.112) implies that R(A) E B 2 (L 2 (sn-I )). Further-
more the operator-valued function A r-----7 R(J\) is continuous in Hilbert-Schmidt norm
362 FURTHER TOPICS IN SCATTERING THEORY

on (0, oo) \ CTp(H) 15 [the averaged total scattering cross section CY(A.) can be consid-
ered to be a continuous function of A if the choice specified in (a) is made].
(d) For later reference we mention the following expression for R(A.). Assume that
V (x) == V1 ( x) + V2 (x) and that 11£ (x) == Ug (x) Wg (x), where Ug and Wg are real
ce
functions satisfying condition (S) == 1, 2). Then (Problem 7.10):
2
R(A.) ==- 21riL A1ut (A.)Mwt (A.)*
R=1
2

+ 27Ti L Mu (A.)[WJ (Q)(H- A.- i0)- Ug( Q)]Mwe (A.)*.


7
1
(7.116)
J,£=1

7. 4.3. We present here a result on the differentiability of the scattering matrix S (A)
R(A.) + I with respect to the variable A. We recall from (7 .112) that
R(A.) == -27TiMu(A.)[I- W(H- A.- i0)- 1 U]Mw(A.)*,
which has the form of a product of three A.-dependent operators. Differentiability prop-
erties of W(H- A.- i0)- 1 U have already been established in Proposition 6.21, so it
is enough to know conditions under which operators of the form 1\1w (A) are differen-
tiable with respect to A. Now Mw(A.) is an integral operator with kernel

aA(w, x) == 2-1/2(27r)-n/2 A(n-2)f4e-iV-\w·xw(x).

Differentiation of aA (w, x) with respect to A. leads (at least formally) to the following
expression for the derivative of M w (A):
n
d n-2 i ""
dA 1\!fw(A.) == ~Mw(A.)- 1\ ~ Njl\IIQ 1 w(A.), (7.117)
2v A j=l

where NJ denotes the multiplication operator by Wj in L 2(sn-l ), with 1


wE
sn- .
Note that, for A > 0, the right-hand side of Eq. (7 .117) represents a well defined
con1pact operator from L 2 (:IRn) to L 2 ( sn-l) if the functions w and X j w satisfy
condition (S) on page 355, and it is norm continuous as a function of A. We refer to
part A.7.3 of the Appendix for a proof of Eq. (7.117) [with differentiability in norm!]
and note the following consequence of this equation:
Assume that (1 + lxl)kW satisfies conditions (S) on page 355 for sonle kEN.
Then l\1w (A.) is k times continuously norm differentiable on (0, oo ).
The next proposition applies to Schrodinger operators for the class of potentials
considered in Proposition 7.3, but with stronger assumptions on the decay of V.
15
Let Af----t C(A) E B(H) be continuous in norm and Af----t Z(A) E B2(H) continuous in Hilbert-
Schn1idt nonn. Then [see (2.47)]:
IIC(A)Z(A)- C(M)Z(M)IIHs = IIC(A)[Z(A)- Z(M)]IIHs + II[C(A)- C(M)]Z(M)IIHs
s; IIC(A)I/1/Z(A)- Z(M)I/Hs + IIC(A)- C(M)IIIIZ(M)i!Hs,
and each term converges to zero as A ~ fL.
THE SCATTERING MATRIX 363

Proposition 7.21. Let H == H 0 + V(Q) in L 2(JRn), where V: JRn ~ JR has the form
V(x) == (1 + !xl)- 1- 6 [voo(x) + vq(x)] with 6 > kfor some kEN, v 00 E L 00 (1Rn),
Vq E Lq(JRn) for some q > n/2 (and q > 1) and
(i) ifn == 1, 2 or3: V andr(oVjor) belong to L 2 (JRn)+L~(JRn), andr 2(8 2Vjor 2)
belongs to L 2(JRn) + L 00 (JRn),
(ii) ifn > 4: V and r(oVjor) belong to £P(JRn) + L~(JRn) and r 2(8 2Vjor 2 ) E
£P(JRn) + L 00 (JRn) for some p E ( n/2, oo ).
Then S(A) is k times continuously differentiable in norm on (0, oo) \ C5p(H).

PROOF. (i) We first prove the differentiability of R( A) under a stronger decay as-
sumption on V, namely we assume that 6 > 2k in the proposition. We denote by V1
and V2 the multiplication operators by voo(x)(1 + lxl)_ 1 _ 6 and vq(x)(1 + lxl)_ 1 _ 6
respectively. Let v - (1 + 6)/2 and observe that v > k + 1/2 if 6 > 2k, and write
V£ - URW£ with U1(x) == lvoo(i)l 1 / 2 (1 + !xl)-v, U2(x) == lvq(x)l 1 / 2 (1 + lxl)-v
and WR(x) == U£(x)signV£(x) (R- 1, 2). The functions (1+lxl)kU£ and (1 !xl)kW£
satisfy condition (S), hence Mue (A) and Mwe (A) are k times continuously norm dif-
ferentiable. By Proposition 6.21 the operators Wj (Q) (H- A-io)- 1 U£( Q) are k times
continuously norm differentiable away from C5p(H). Hence, in view of Eq. (7.116),
R(A) is k times continuously differentiable.
(ii) We note that R( A) in Eq. (7 .116) is a sum of terms each of which has the form
of a product of two or three A-dependent operators. Such a product is differentiable if
each factor is differentiable [as used in (i) above]. However, due to the special form
of these products, one can obtain their differentiability without assuming the individ-
ual factors to be themselves differentiable. This is achieved by shifting operators of
the form (Q)- K between these factors when writing the derivatives, using the identity
MAB(A) == MA(A)B and its adjoint MAB(A)* == B*MA(A)*. In this way the propo-
sition can be proved under the weaker assumption that 6 > k (rather than 6 > 2k).
We leave the detailed verification as an exercise (Problem 7.11) and just explain the
method by considering the derivative of a term of the form M u (A) M w (A)*.
For U and W satisfying the condition (S) and a> 0, (3 > 0, set U(a) == (Q)-aU
and W(JJ) == (Q)-f3W. By virtue of the identities on M-operators given above, we
have for all A, f-L > 0:

This implies that

whenever a + (3 r + 6.

For a+ (3 == 1, we may thus write

Mu(a) (A+c)Mwun (A+ c)*- Mu(a) (A)Mwun (A)*


== [lVfuu) (A+ c)- Mucl) (A)]Mw(A +c)*
+ Mu(A) [Mwcl) (A+ c)*- Mwu) (A)*].
364 FURTHER TOPICS IN SCATTERING THEORY

Upon division byE and then taking the limitE~ 0, one finds that

d~ Mural (A)Mwc 131 (JL) * = [d~ Mu(l/X)] Mw (A)*+ Mu(A) [ d~ Mwc,l (A)*].
(7.118)
The limit E ~ 0 exists in norm and the derivative, given by the right-hand side of
(7.118), is continuous in norm. If the exponent v in condition (S) is very close to 1/2
(and a (3 == 1), then the function U(a) W(JJ) decays like lxi- 2 -E with very small
E > 0. In such a situation at least one of the two operators Mu(a) (:A) and Mltvcp) (J-L)*
is not differentiable, but their product is continuously differentiable in norm! D

7.5 Time delay

Ti1ne delay is one of the fundamental concepts of scattering theory, and it is of much
theoretical importance in various fields of quantum physics. In the simplest situations,
time delay expresses the excess time that scattered particles spend in the scattering
region when compared to free particles subject to the same initial conditions. A posi-
tive time delay means that particles take more time to pass through the region where
they are influenced by the interaction than particles propagating freely through the
same region. A negative time delay means that on average the scattered particles are
accelerated by the effects of the interaction.
In this final section we present a Hilbert space approach to time delay. It will be
shown in particular that the time delay can be expressed in terms of a self-adjoint op-
erator, called the Eisen bud-Wigner time delay operator, which is entirely determined
by the scattering operator S. References to some applications can be found in the bib-
liographical notes to this chapter.

7 . 5 . 1. As in Section 5.2 we consider a physical system described by the states in the


Hilbert space H == £ 2 (JRn). Let H be its Hamiltonian and { Ut }tElR the associated
unitary evolution group, with Ut == e-iHt. We denote by x E £ 00 (JR) the character-
istic function of the interval [0, 1]. Then FR == x(IQI/R) is the projection operator
describing localisation in the ball BR :== { x E JRn jlxl < R} in JRn [see Eq. (5.22)].
To each non-zero g E H and each R > 0 we associate a quantity TR (g) by setting

Since liFRUtgll 2 /11911 2 is the probability that the state represented by the vector g is
localised in the ball B R at time t, the positive number TR (g) may be interpreted as
the total time that the state g spends in the ball B R during its evolution, if the time
evolution is governed by the group { Ut}.
Scattering theory is based on the comparison of the evolution of states with a free
evolution. If H 0 denotes the free Hamiltonian and if up == e-iHot, then the total time
spent in the ball BR by a state f evolving under the free evolution group {U?} will
TIME DELAY 365

be given by the number

(7 .120)

As said in the introduction to this section, the time delay in the ball B R for scat-
tering with initial state f is the difference of the time spent in B R by the associated
scattering state 0_ f and the time spent in B R by the freely evolving state f, i.e. the
quantity TR(Sl-f)- Tfi(f). 16 Omitting the normalisation factors 1/11 · 11 2 occurring
in (7 .119) and (7 .120), we define

(D.T)R(f) -I: IIFRUtfLJII 2 dt- I: IIFRU2fll 2 dt (7 .121)

which, for normalised state vectors Cllfll == 1), is the time delay for the ball BR. In the
framework of quantum mechanics in ffi.n, the effective scattering region is all of ffi. 11 ,
so that one must consider the limit of (llT) R(f) as R ~ oo. This limit (if it exists)
is called the global time delay f:l T (f) for the initial state f: 17

(7 .122)

In most situations both integrals appearing in (7 .121) are finite. Since the integrands
are bounded continuous functions oft, the quantity ( f:l T) R (f) may be interpreted as
the improper Riemann integral of the function t f--7 liFRUtO-fll 2 - IIFRU2JII 2 . As
R ~ oo, each of the two integrals in (7.121) tends to +oo, but, for suitable vectors f,
their difference may be convergent and thus define a finite global time delay f:l T (f).
Before considering these questions we give a result that allows one to express the
global time delay in terms of the scattering operator. We assume that M~(Ho) == H,
and that the wave operators 0± == s-lim t-7±oo Ut UP exist and satisfy asymptotic
completeness. 18 The crucial condition is a sufficiently rapid convergence of Ut UP
to the wave operators for suitable vectors (note that one has IIO±f- UtU?Jll ==
11(0±- I)U2Jll).
Proposition 7.22. Let f be such that, for every fixed R < oo, each of the .functions
t f--7 II FRUt0 fll and t f--7 llFRU2 S fli belongs to L 2 ([0, oo) ). Furthermore assume that
the wave operators exist and are complete, that the function t f--7 I (n_ - ut uP) J II
belongs to £ 1 (( -oo, 0]) and that t f--7 11(0+- UtU2)Sfll belongs to £ 1 ([0, oo)).
Then

(7.123)

16 We assume here that each vector f E 7--l is a possible initial state, i.e. that Moo(Ho) = 7--l, and that
the wave operators exist on 7--l.
17 As pointed out, the interpretation of 6 T(f) as a time delay is justified for vectors satisfying 11!11 = 1.

The advantage of not including the nonnalisation factors 1/11!11 2 in Eq (7.121) is that the quantity 6 T(f)
will be the expectation value of a linear operator.
18 The explicit form of H is not essential in this proposition, and the result could also be formulated in
0
terms of an abstract family of localising operators as in Section 5.2.
366 FURTHER TOPICS IN SCATTERING THEORY

in the sense that the limit in (7 .122) exists if and only if that in (7 .123) exists, with
equality of these two limits.
PROOF. ForgE H we set Jj{(g, t) == IIFRUtSl±g// 2 - IIFRU2g/l 2 . Since Sl+Sf ==
D_f, we have

Hence we may write

(LlT)R(f)
0
I: Iii(!, t)dt =I: Iii(!, t)dt +fa= Iii(!, t)dt

==!-~ Jfi(f,t)dt+ lo{'XJ J}i(Sf,t)dt+ lor=[IIFRUPSJII 2


-IIFRUPJII 2 ]dt.

We shall show that each integral on the last line exists (as an improper Riemann inte-
gral) and that the first two of these integrals converge to zero as R ---t oo. The result
of the proposition then follows by observing that the integrand of the third integral is
identical with the integrand in (7 .123).
So let us consider each of the three integrals on the last line above. It is clear that
the third one exists by the hypotheses on //FRU2SJ/I and I!FRUPJII. To deal with the
other two integrals, we observe [using ( 1.15)] that

jJj{(g,t)j == [IIFRUtSl±gll + IIFRU2gii]IIIFRUtD±gii-IIFRU2glll


< 21/g//IIUtfl±g U2g// == 21/giiii(O±- UtU2)gl/. (7.124)
By taking g == f and g == Sf and using the remaining assumptions of the proposition,
one obtains the existence of the first two integrals under consideration. Finally we
observe that Jj{ (g, t) converge to zero as R ---t oo for each fixed t E JR. As functions
oft, Jfi(j,t) and JJi(Sf,t) admit £ 1 bounds on (-oo,O] and [O,oo) respectively,
uniformly in R, see (7 .124 ). From the Dominated Convergence Theorem, one obtains
the convergence to zero of their integrals as R ---t oo. 0

Proposition 7.22 shows that the time delay (.6.T) R for a finite ball B R can be
expressed in terms of expectation values of the operator f0~U2*FRU2dt. We next
discuss the asymptotic behaviour (R ---t oo) of such expectation values in the case in
which H 0 == P2 . For this we denote by D-:J (v > 0) the following dense subsets of
£ 2 (JRn) [which are just the sets V-:J(H0 ) introduced before Lemma 7.5]:
vt == {f E H I f == 1/;(Ho) (Q) -vh for some hE Hand some 1/J E Co( (0, oo))}.
v-:; is contained in D( (Q)v), see Example 6.23. We also introduce an operator Do as
follows [D being defined in (6.19)]:
1[ 1 -+ -+ -+ -+ 1 ] [ 1 1 ] +
Do=2 1ft12P·Q+Q·Pift12 = 1ft12D+Dift12' withD(Do)=Dl.
(7.125)
It is easy to check that this operator is well defined and symmetric (Problem 7.12). We
recall that FR == x(R- 1 /Q/).
TIME DELAY 367

. [ {CX)I
J~oo Jo \f,Ut X
0* (IQI) R;
R Ut0g) dt- 2J,H -1/2 )] 1; (7.126)
0 g = -
2 J,D0 g).
PROOF. (i) By virtue of the polarisation identity (2.15), it suffices to prove (7 .126)
for g == f. Let us denote the left-hand side of (7.126), with g - f, by Y(f). We
must show that Y(f) exists (as a limit) and is equal to - (f,Dof) /2. We first derive
a convenient expression for Y(f). We set W 7 == exp( iQ 2T) and then get from (5.62)
that u?*x(IQI/ R)U? == w;/4tX(2tiPI/ R)Wl/4t· By making the change ofvariables
t ~ s == (2t)- 1 Rand setting E == (2R)- 1 one finds that

lor I o* ( 1Q1) o )
\f,Ut X R Ut f dt-
CX) / I
lo J, * ( 2t 1PI ) )
wlj4tX ~ wl/4tf dt
CX)

== -41
E
1 (J,CX)

O
* (IF
WssX - 8 I) Wssf)2
ds ·
S

Similarly, setting again E == (2R)- 1 and making the change of variables u ~ s ==


Ikl / u in the second expression below, one obtains

The preceding two equations show that

{CX) 1
Y(f)- cl!!~o Jo 4 Es2 (f,Wcsx(-
[ * IPI
)Wssf)- (f,x(- )f) ds IPI J (7 .127)
8 8

- lim {oo Kc(s)ds,


s~+O Jo
where K s (s) denotes the integrand in (7 .127).
(ii) By using the Dominated Convergence Theorem one can justify the interchange
of the limit and the integral in Eq. (7.127). The details are elaborated in (iii) below.
Assuming this result for the moment, we obtain from (7 .127) that

(7.128)

The integral over dnk in (7.128) is zero for s1nall s, because ](k) == 0 near the origin.
368 FURTHER TOPICS IN SCATTERING THEORY

J
For large values of s this integral is equal to (f, Q2 f)- (Q2 f, f) == 0, because (k) ==
0 near infinity and Q2 is a symmetric operator. It follows that the last expression for
Y(f) is finite, which proves the existence of Y(f). In spherical polar co-ordinates
k == kw, with k == jkj and w == kjk E sn-l, the integral over dnk in (7.128) is as
follows:

Upon inserting this expression into (7 .128) and integrating by parts in the variable s,
one finds (after observing that the boundary terms at s == 0 and s == oo vanish) that

-~1 (logs)sn-lds }~,


8

Y(J) = dw [ ](sW)(:n:}ZJ) (sW)- (.FC) 2J) (sW)](sW) J


1

= -H((loglfti)J,Q 2J)- (Q 2 j,(loglfti)J)].

To complete the proof, it remains to check that (log IPI)Q 2 - Q2 log !PI == -2iD 0
on Vi.This is easily done by observing that [Qj, log IPIJ == iPj/IPI 2 :

(iii) We finally exhibit an E-independent integrable bound for the integrand Kc:(s)
in (7 .127), thus justifying the application of the Dominated Convergence Theorem on
the right-hand side of (7 .127). It will be necessary to consider separately the domains
0 < s < 1 and 1 < s < oo. The following alternative expressions forKs (s) are easily
obtained by simple algebraic manipulations, using the unitarity of We: s and defining
x_L: [0, oo) ~ JR by x_L(u) == 1- x(u):

(7.129)

We shall use the following bound for IIT- 1 (WT- I)fll:

(7.131)

Indeed, as leia- 11 == IJ0aeitdtl < lal for a E JR, we have II(WT- I)fll < IITQ 2 fll,
and it is clear that I (WT - I)fll < 2llfll.
TIME DELAY 369

Now lets E (0, 1]. By (7.129) and (7.131) we have for s > 0:

iiQ 2 fll [llx( I~ I)Wc:s!ll llx( I~ I).til].


1
IKc:(s)l < 43
To arrive at an c--independent integrable bound for Kc:(s) on the interval (0, 1], it is
enough to show that llx(IPI/s)WTfll < cs 8 for all T E JR and some 6 > 0. We take
6 == 1/4. By observing that x(x/ s) - x(x/ s)x(x) if s E (0, 1] and that lx( u) I <
u- 114 for all u > 0, we get for all T E JR:

2
and the required estimate llx(IPI/ s)WT !II < cs 114 follows since IPI- 1 14 x(IPI) (Q) -
belongs to B(H) by Proposition 2.33.
Next let us consider the case s E [1, oo). Since Ix_i (u) I < u for all u > 0, we find
by taking into account (7 .14 7) that

llx_L(I~I)Wr.fll < 1\I~IWr!\\ < ~[i)PJWrfll r


2 12

1 n 1 n
<S L IIWTPjJ + 2TQjWr.fll < S L [11PJJII + 2ITIIIQ.d11] ·
j=l j=l

Hence, by using (7 .130) and (7 .131 ), we obtain

IKc:(s)l <
1
43 [min {IIQ 2.fll, ;s ll.fll}] [llx_L( I~ I)Wc:s!ll + llx_L( I~ I)!11]
n
2
< 4 ~2 [min {IIQ .fll, cs ll.fll}] L [2IIPJ.fll
2
2csiiQ1 .fll]
J·=l
n

< 3~ L [11Q .fiiiiPJfll + 11/IIIIQJfll],


j=l
2

which gives an integrable c--independent bound for K s ( s) on the interval [1, oo). D

We can now prove the existence of the global time delay ~ T (f) for suitable vectors
f in £ 2 (JRn) and a class of self-adjoint Hamiltonians of the form H == Ho + V( Q),
with H 0 == P2 .
Proposition 7.24. Let H == Ho + V(Q) in L 2 (JRn), where V: JRn ~ JR has thefornz
V(x) == (1 + lxl) 1 - 8 [voo(x) + vq(x)] with 6 > 3, v 00 E L 00 (1Rn) and vq E Lq(JRn)
for some q with q > n/2 and q > 2. Also assume that V satisfies the condition (C) on
page 288. Let f EDt for some p > 7/2, and assume that the support of (Uof)A is
disjoint from the point spectrum of H. Then the limit in (7 .122) exists and is given by
the following expression:

!J.T(f) = -~(!, S*[Do, S]f). (7.132)


370 FURTHER TOPICS IN SCATTERING THEORY

PROOF. We know from Proposition 7.3 that the wave operators exist and that the scat-
tering operatorS is unitary. We shall show in (ii)- (iii) that the assumptions of Proposi-
tion 7.22 are satisfied and that the vector Sf belongs to vt.
One can then apply (7.123)
and (7.126). By adding and subtracting (R/2)(f,H~ / f)
1 2
(R/2)(Sf,H~ 112 Sf)
in the first integral below (observe that S commutes with H~ / ), one obtains
1 2

This then proves (7 .132).


(i) We fix a closed interval J in (0, oo) \ o-p(H) such that (U 0 f)>.. [and hence also
(U 0 S f)>..] is zero for A outside J. We first observe that the vector-valued function A f---7
(U 0 Sf)>.. S(.X)(U 0 f)>.. is three times continuously differentiable on J. Indeed,
since V decays as lxi-K: with~ > 4, S(.X) is three times continuously differentiable
in norm on J (Proposition 7.21), and (U 0 f)>.. == M(Q)-p(.X)(Q)Pf is three times con-
tinuously differentiable since p - 3 > 1/2 (see page 362).
It follows in particular that, if D denotes the operator defined in (6.19), then Sf
belongs to D(D 3 ) [in the spectral representation of H 0 , the operator D 3 is a third
order differential operator with coefficients that are bounded on J, see Eq. (6.23)].
(ii) We now show that Sf E vt.Let() be a function in C0 ((0, oo)) satisfying
()(.X) == 1 for all A E J. Then one may write Sf == ()(Ho) (Q) - 2 (Q) 2 Sf, and the inclu-
sion Sf E vt follows once we have shown that Sf E D( Q2 ). For this we express the
operator Q2 in spherical polar co-ordinates in momentum space, viz.

(Fi:}2g)(kW) = -fl.[J(kW) =- [ :k
2

2 n ~1 ! + ~:]g(kW). (7.133)

Here D.s denotes the spherical Laplacian (the Laplace-Beltrami operator on the unit
sphere sn-I, containing the derivatives with respect to the angular variables; an ex-
plicit formula for this operator will not be needed). By transcribing the preceding
expression into the spectral representation of H 0 as in (6.22), one obtains (with k ==
lkl == ~):

(UoQ---+2 g)>..==- [ 4,\ ( dA


d )2

For g == Sf, the first term on the right-hand side defines an element of L 2 (1Rn) because
(U 0 Sf)>.. is three times continuously differentiable with respect to A and vanishes near
.X == 0 and near infinity. So it remains to show that D.s(U 0 Sf)>.. D.s(Uof)>.. +
2
D.s (UoRf)>.. represents a vector in L (1R ).11
TIME DELAY 371

3
To treat the term 6.s(Uof)>.. we write it as 6.s(U 0 f)>.. == 6.sM(q)-3(A.)(Q) f,
and we use Lemma 7.25 in the Appendix which shows that 6.s 1\;f (Q) -3 (A) is a bounded
operator from £ 2 (ffi.n) to £ 2 ( sn- 1 ) that depends continuously on A. on (0, oo). Hence
ll6.s(Uof)>..ll is a bounded function of A. on the interval J, so that A- 1 6.s(Uof)>..
defines an element of £ 2 (ffi.n).
To handle the term 6.s (U oR!) A we write the potential as v == uw with u (x) ==
(1 + !xl)-v lv(X)(x) +vq(x) 11 / 2 , W(x) == (1 lxl)v- 1 - 8 lv(X) (x) +vq(x) 11 / 2 sign V(x),
with v satisfying 5/2 < v < 3. The functions U and W satisfy condition (S), hence
Mu(A.) and Mw(A.) are well defined and depend continuously on A (Proposition
7.16). Then we have by (7.112):

6.s(UoRJ)>.. == -2?ri6.sMu(A.)[I- W(Q)(H- A- i0)- 1 U(Q)]Mw(A.)*(Uof)>..·

Again by Lemma 7.25 in the Appendix, the operator 6.sMu(A.) is bounded from
£ 2 (ffi.n) to £ 2 ( sn- 1 ) and depends continuously on A on (0, oo) (here the condition
v > 5/2 is crucial). It is then seen from the preceding equation that ll6.s(U 0 Rf)>.. II is
a bounded function of A on the interval J, hence A. - 1 6.s(U 0 Rf)>.. defines an ele1nent
of £2 (ffi.n).
(iii) We finally verify that the assumptions of Proposition 7.22 are satisfied. Let J
be as in (i) above. Then Fn is H 0 -smooth on J (see Proposition 6.16), consequently
each of the functions t ~ IIFnUPJII llFREo(~l)UPJll and t ~ llFRU2Sfll belongs
2
to £ (ill:.) (Proposition 6.11 ).
Next we have

ll(fL -u;u?)fll = llj_tood~u;u~jdTII = llj_toou;vu~jdTII


<]_too IIVU~JIIdT.

Now, assuming that f == 1/J(Ho)(Q) -ph and setting ¢(A.) == (A.+ i)r¢(A.):

IIVU~fll < IIV(Ho + i)- 1 (Q)PIIII (Q) -p ¢(Ho)U~ (Q) -pllll (Q)Pfll,

which is less than c ( 1 + IT I) - 3 for some constant c by (7 .45), since p > 3 (the first
norm on the right-hand side is finite by Example 6.26). Thus 11(0_- UtU2)Jii <
c(1 jtl)- 2 for negative t, which is an integrable function oft on ( -oo, 0].
The verification of the integrability of II (0+- UtUt0 )Sfll on [0, oo) is similar. Let
again() be a function in C0 ((0, oo)) satisfying ()(A.) == 1 for all A. E J. Then one may
writeS f == ()( H 0 ) (I+ D 2 ) - 3 12 h, with h == (I+ D 2 ) 3 12 Sf E H because Sf ED( D 3 )
as seen in (i). Setting cp(A.) == (A+ i)()(A), we then have

IIVU~Sfll < II (Ho + i)- 1 (Q)PIIII (Q) -pcp(Ho)U~(I + D 2 )- 312 llllhll,

which is less than cs(1 + ITI)- 3 +s for any E > 0 by (7.46). So II (0+- Ut[J?)Sfll <
ft(X)IlVU~SflldT < Cs(1 + t)- 2 +s fort> 0. 0
372 FURTHER TOPICS IN SCATTERING THEORY

7.5.2. By taking into account the intertwining relation U7 0_ == Q_U~, one sees from
the definition (7 .121) of the local time delay that (.6.T) R (U~ f) == (.6.T) R (f) for each
T E JR. Consequently the global time delay, if it exists, will also be invariant under the
free evolution. In fact this is easily verified by observing that U~*D 0 U~ == D 0 + 2T I
on D{, so that U~*S*[D 0 , S]U~ S*[U~*D 0 U~, S] == S*[D 0 , S]. The last property
also shows that the operator S*[D 0 , S] commutes with each U~, hence with H 0 (on a
suitable domain).
In analogy with (5.71), we conclude from the commutativity of S*[D 0 , S] and
H 0 that there will be a family {T(.A)}.A>O of operators acting in L 2(sn-l) such that
( -1/2)(UoS*[Do, S]f).\ == T(.A)(Uof).\ for A > 0, in other terms that the (un-
bounded) operator S* [Do, S] will be decomposable in the spectral representation of
H 0 (see also Proposition 5.26). The operator T(.A) represents the time delay at en-
ergy .\, and it will be a symmetric operator because S* [Do, S] is symmetric. In fact
it is easy to obtain an explicit expression for T(.A) in terms of the S-matrix: by us-
ing the second expression in (7 .125) for the operator D 0 and the fact that D acts as
i.A(d/d.A) + i/2 in the spectral representation of H 0 [see (6.23)], one finds that the
action of Do in this representation is as follows:

(7.134)

In other terms, in the spectral representation of H 0 , the operator Do is the differential


operator 2i( d/ d.\). We have seen in Section 3.2 that this operator is maximal symmet-
ric but not essentially self-adjoint (page 101). We now get

T(A) = -iS(A)*[d~,S(A)]- -iS(A)*S'(A). (7.135)

Here I (A) denotes the derivative of the B (£ 2(sn-l))- valued function A ~ (.A) at
s s
the point A. Note that the definition of this derivative at,\ involves the S-matrix also
at neighbouring values of A, but, since T (,\) is defined as a decomposable operator,
the action of S'(.A) in (7.135) is as follows: [U 0 S'(.A)f].\ == S'(.A)(Uof).\·
In situations in which,\~ S(.A) is continuously differentiable on (0, oo ), the oper-
ator T(.A) will be bounded for each,\ > 0, hence not only symmetric but self-adjoint,
and A ~ T (,\) will be continuous in norm. The self-adjointness ofT (,\) can also be
inferred from the unitarity of S (A): by differentiating the equation S (A)* S (,\) == I,
one finds that S'(A)*S(.A) S(.A)*S'(.A) == 0, i.e. S'(.A)*S(.A) == -S(.A)*S'(.A).
Thus

T(A)* == [ - iS(A)*S'(.A)J * == iS'(.A)*S(A) == -iS(A)*S'(.A) == T(A).


The family {T(.A)} .\>O determines a self-adjoint operator Tin £ 2 (JRn ), called the
Eisenbud- Wigner time delay operator, given as follows by its action in the spectral
representation of Ho:

(UoTf).\ == T(A)(Uof).\, (7.136)

D(T) = {f E L 2 (~n) /laoo dA IIT(A)[Uofhlli2(sn-1) < oo }. (7.137)


TIME DELAY 373

In general this operator Twill be unbounded, even though each T(A) is bounded. This
is so because IIT(A) II usually diverges as A -4 0 19 . As for the multiplication operators
considered in Section 2.5, the domain (7.137) is the maximal domain on which T
can be defined in a natural way. The global time delay ( ~ T) (f) for an initial state f
CIIJII - 1) is then the expectation value of the observable Tin this state.
In the simple situation considered at the end of §5.5.1 (scattering by a spherically
symmetric potential inn== 3 dimensions), the S-matrix is given in each partial wave
subspace by a number S(A) e 2 ioe(.A). In this situation the time delay operator T
leaves each partial wave subspace invariant, and in the partial wave of angular mo-
mentum R the operator T (A) is determined by the derivative of the phase shift 8.A,
viz. T(A) is just multiplication by 28~(A).

7 .5.3. We mention another quantity of interest, viz. the trace of the operator T (A). Let
us consider potentials of the form V(x) == U(x)W(x) with U and W satisfying the
condition (S). If both functions U and W belong to £ 2 (JRn), then Mu(A) and Mw(A)
are Hilbert-Schmidt operators, and it follows from (7.112) that R(A) is a trace class
operator in £ 2 ( sn-1
). One can similarly find conditions implying that T (A) belongs
to the trace class. If (1+lxi)W(x) belongs to £ 2 (JRn ), then (d/ dA)Mw(A) is a Hilbert-
Schmidt operator, as is easily seen from (7 .117). By using this property in the calcula-
tion of the derivative of R(A) indicated in part (ii) of the proof of Proposition 7.21, one
finds that T(A) belongs to the trace class if one assumes that (1 lxi)V(x) belongs to
L 1 (JR n) (Problem 7.11). This condition requires a decay of v (x) faster than Ixi- (n+ 1 )
[this result is not optimal; it is known that a decay faster than lxl-n is sufficient for
the inclusion T(A) E B 1 (L 2 (sn- 1 ) )].
Similar to the Hilbert-Schmidt norm of R(A), which determines the total scattering
cross section for a beam, averaged over all initial directions, the trace of T (A) has an
interpretation in relation to the scattering of a beam. Consider a family {fs} of vec-
tors obtained as in (5.78) from a fixed vector f by translations in the hyperplane II 0
orthogonal to an initial direction w0 and describing an incoming beam as in Section
5 .6. The total time delay for such a beam will be the quantity JITo ( ~ T) (is). Assuming
that the Hilbert-Schmidt kernels of the trace class operators T (A) a~e continuous func-
tions, and considering wave functions f such that the support of f belongs to a very
small neighbourhood of a vector ko == ~ w0 , one finds that, after averaging over all
initial directions w0 E sn-
1
, the averaged time delay for such a beam is determined
by TrT(Ao):

The verification uses arguments similar to those leading to the total scattering cross
section in Section 5.6. The details can be found in [2].

19 Due to the inverse powers of ~ occurring in (7 .117), S 1


()..) will become unbounded as ).. ~ 0.
374 FURTHER TOPICS IN SCATTERING THEORY

Appendix to Chapter 7

A . 7.1. Proof of Proposition 7.4. (i) We embed 0 in a shell G 0 of thickness 26 > 0:


Let 0 s be the hypersurface obtained from 0 by displacements of length s orthogo-
nal to 0, i.e. Os == {x E ]Rn I X== if sn(iJ), if E 0}, and let Go == U-o::;s::;o Os.
If 0 == Eb, we take 6 == b/2; if 0 == Ap,d, we choose 6 to be sufficiently small so
that each vector in 0 0 has a unique decomposition into x == if sn(iJ) with if E 0
and s E [-6, 6]. Throughout this proof it is understood that x and if are related in the
preceding manner (so, as x varies over G 0 , if and s are functions of X).
We denote by das (if) the surface element on 0 s at the point if sn(iJ) and by
da(iJ) _ da 0 (iJ) the corresponding quantity on 0. There is a constant f3 E (1, oo)
such that

\lyE 0, \Is E (-8, 8]. (7.138)

(ii) Let f be as stated in the proposition and x E G 0 . ForTE [0, 1] and k == 1, ... , n
we set

By Newton's formula one has

(7.139)

With ( 1.58) this implies that

(7.140)

Now, by the triangle inequality:

(7.141)

and, for 0 < 1 < 1 [using (7.138)]:


APPENDIX TO CHAPTER 7 375

das(17)[f~kl(y
5
IIJJk)lli2(G 6
) ==! -c5
ds {
J0
siT(Y))[
2

== j
c5
2
ds { das(Y)[(Pkf)(Y rs'i1(Y))[
-c5 Jo
=~~TO du { danjT(Y)[(Pkf)(ff+u'i1(Y))[
2

-Tc5 J0
~
2
~ 11Pkflll2(G 8 )·
2
2
< JO du { dau(Y)[(Pkf)(Y + uiT(Y))[ =
-c5 lo
Together with (7.140) and (7.141) this implies that

Thus (7.28) holds for rp 1, with co == max{ j73{J, 2/3VnlfS} [take into account
Eq. (7 .148)]. The result for general rp follows immediately because f I0 == (rp f) I 0 .
(iii) We finally verify the inequality (7 .29). Here 0 == L:b, and we have c~b <
max{~' 2/3bVn/Jb(5}, where by (7.138) f3b > 1 is such that

;b bn- 1 < (b S )n-


1
< f3bbn- 1 \IS E [-J, J] (7.142)

(because d(J s (if) == (b s) n-l dw in spherical polar co-ordinates). Since we consider


only values of b that are> 1, we may take b- 1/4 in (i) above (rather than b == b/2).
Then (7.142) is satisfied for all b > 1 and all s E 1/4, 1/4] with f3b == (4/3)n-l. So
we have supb2: 1 c~b < ~n with a finite constant ~n' depending on n, which we do not
determine explicitly. D

A.7.2. Proof of (7.40) and (7.41)


(i) The operator Pj isH-bounded and maps V(H 2 ) into V(H), for P1 (H i)- 2 ==
(H i) - 1 Pj ( H i) - 1 + i (H i) - 1 [ ( 81 V) (Q) J ( H + i)- 2 . If g == 1/J (H) (Q) h E
1

V 1 (H), then U7 g E V(H 2 ) for each T E JR. By Proposition 5.5(b) (Leibniz rule),
t ~ Ut P1 Ut g is strongly differentiable. Hence

UtP1 Utg- P1 g =lot d~u;P 1 UTgdr- ilotu;[H,P1 ]UTgdr


=-lot u;[(8J V)(Q)]UTgdr.

Upon multiplying this equation by Ut one obtains (7 .40).


(ii) For c > 0, let us set Q1 ,E - Qj(I c;Q 2 )- 1 12 . Then one has Q1 ,r:: E B(H)
and s-limr::~o Qj,r::f == Qjf for each f E V(Qj)· As in Example 6.22 one finds that
Qj,r::V(H) C V(H) V(Ho) and that on V(H) [see (7.50)]:

2iB], E • j3- 2i(J + c;Q 2 ) 1 2


/ pJ
376 FURTHER TOPICS IN SCATTERING THEORY

with

Note that llc-Q1 ,c:ll < c- 1 12 , hence IIAj,c:ll < (n + S)c- 112 , and that II[Bj,c:]kll < 1.
We can apply Proposition 5.5(a) to see that T ~ u;Qj,EUTg is strongly differen-
tiable if g E V(H), with derivative iU;[Aj,c: + 2iBj,c: · P- 2i(I c-Q 2 )- 112 Pj]U7 g.
Hence

Q1,cUtg- Q 1,c9- Jort UT*[·zAj,c- 2Bj,c


- · P- + 2(! + EQ-2 ) -1/2 Pj JUTgdT.
(7.143)
1
To obtain (7 .41 ), we assume that g == 1/J (H) (Q) - h E Vi (H) and determine the limit
of (7.143) as c --+ 0. Since Utg E V( Qj) n V(H) [by part (i) of the proof of Lemma
7 .5], the left-hand side of (7 .143) converges strongly toUt Q j Utg- Qj gas c --+ 0. Let
us then consider each of the three terms on the right-hand side of (7.143).
First we have IIJ;u;Aj,c:U7 dTii < ltiiiAj,c:ll· Since IIAj,c:ll --+ 0 as c--+ 0, the
operator i J;u; Aj,c:U7 dT converges to zero in norm as c--+ 0. For the remaining two
terms we apply the Dominated Convergence Theorem. We observe that

The integrand on the right-hand side is bounded by aT-independent constant, since


1
ii[B1 ,c:]kil < 1 and I!PkUTgll - 11Pk?/J(H)UT(Q)- hii < IIPk?/J(H)IIIihll. Further-
more this integrand converges to zero for each fixed T:

and U7 g belongs to V(QkPk) by Lemma 7.5(a). So the second term on the right-hand
side of (7.143) converges strongly to zero. It remains to be shown that the third term
converges strongly to 2J; U;PjU7 gdT, i.e. that

This is obtained as for the second term above by taking into account the fact that
s-limc:--+O [(I c-Q 2 ) - 1 / 2 - I] == 0. 0
APPENDIX TO CHAPTER 7 377

A.7.3. Differentiability of lVlw (.X). Under the assumption that (1 lii)VT! satisfies
condition (S) on page 355, let us prove that Mw(A) is differentiable in norm, with
derivative given by (7 .117). By (7 .1 07) we have for f E S (JR.n)

Differentiation of the factor A(n- 2 )1 4 gives rise to the first term on the right-hand side
of (7 .117). So it suffices to show that the norm derivative of ~~v0: FW with respect
to A exists and is equal to -i(2vf\)- 1 ~]= 1 NjrL-v0:FQj W, or equivalently [setting
viA- fL, hence d/ dA - (2V,\)- 1 d/ dfL] that the norm derivative with respect to fL of
rL-fLFW exists and is equal to -i~]= 1 NjrL-fLFQj W.
With the notation Xc(J-L) c- 1 [rL-,L+c:FW- rL-fLFW] + i~~= 1 NjrL-pFQj 1¥,
we must show that
lim sup IIXc(J-L)f!I£2(Sn 1) == 0. (7.144)
c-+0 fES(IRn)
11!11=1

Now, by Newton's formula (7.139), we have for f E S(JRn):

seW) + ie L WJ ( ~) (pW) Jds


j=1
n
seW)+ i LwJ(~)(pW)] ds
j=1

= -i1 [Lw.ibE,+sc~)(W)- Lwj(rE,~)(W)Jds


0
1 n

j=1 j=1
n

--it 1
j=1 0
1

[(NnE,,+scFQJWf)(W)- (NJ'IE,,FQ.iWJ)(W)]ds.

Since r~fLUf - ViJ-L-(n- 2 )1 2 Mu(J-L 2 ), the integrand of the last expression is a


strongly continuous L 2 ( 1
sn-
)-valued function of s, and by using results from Section
1.2 we then obtain the following inequality, valid for f E S (JR.n):

IIXc(M)JII£2(Sn-1)

<Vi~ la ds\\ (p se~(n- 2 ) 12 MQ1 w((p + se?)- p(n~ 2 ) 12 MQ 1 w(p2 )1\llfll·


1

Since a- ~ a--(n- 2 )/ 2 MQ 1 w(a- 2 ) is norm continuous on (0, oo), this implies the
validity of (7.144) (using for example the Dominated Convergence Theorem). D
378 FURTHER TOPICS IN SCATTERING THEORY

A.7 . 4. An estimate for the spherical Laplacian .6. 8 • We establish here a result that
was used in the proof of Theorem 7 .24.

Lemma 7.25. Let W satisfy the condition (S) with v > 5/2. Then, denoting by N£ the
n~ultiplication operator in B(L 2(sn- 1)) by the function NpJw) == W£ (R == 1, ... , n):

j,£=1
n

(n 1)i~ L NpMQ£w(A). (7.145)


£=1

Theoperator6.sMw(A) isboundedfromL 2(JRn) toL 2(sn- 1)foreachA > 0, and


the n~apping A~ 6.sA1w(A) is continuous in norm.

PROOF. Since Q £ Ti\1 and Q 1 Q £W satisfy the condition (S) [with v > 1/2], the right-
hand side of (7.145) is a nor1n continuous B(L 2(JRn), L 2(sn- 1))-valued function of
A on (0, oo ), see Proposition 7.16. To calculate 6.sMw(A)j, for fixed A and with
f E S(JRn), we extend the function w ~ [Mw(A)j](w) to a function F defined on
some neighbourhood of sn-l by setting, for p E ]Rn \ {0}: F(p) == [Mw (A)j] (pjp ),
where p == IfJI. F is a homogeneous function on ]Rn, independent of the radial variable
p, and as a function of wit coincides with Mw(A)j. As F is independent of p, we
have [see (7.133)]

82 n-1 a !:1;] F(pW) - Li s F(pw),


(LiF) (p W) = [ fJp 2 p2
p 8p p~

in particular (for p == 1)

6.sA1w(A)j == (6.F) Isn-1" (7.146)

Now
F(iJ) = Tl/2 >.(n-2)/4( 21f)-nf2 { e-iv');X·PIPW(X)f(X)dnx
}JRn
and, for if E JRn:

• n-1 -+ -+ J -zy·p
...........1p
z 3 y. p e .
p

Consequently (setting if- ~x)

6.F(p) 'p= 1 == 2-1/2 A(n-2)/4 (21T) -n/2

X r e-iv');x.w{
}JRn
AX 2 +A(X·W) 2 i(n l)~X·W}W(X)f(X)dnx,

which, in view of (7.146), is equivalent to (7.145). D


PROBLEMS 379

Bibliographical Notes
A presentation of various methods that have been developed for proving existence
and completeness of wave operators can be found in Part IV of [BW]. For Schrodinger
operators V. Enss [8] introduced a time-dependent method giving strong asy1nptotic
completeness (see also Section XI.17, Volume III of [RS] for this method, and
for subsequent developments). A review of the literature concerning the Flux -Across-
Surfaces Theorem can be found in [7]. As regards time delay, the reader can find a
presentation of numerous physical aspects and applications of this concept in the re-
view and references to the more mathematical literature in [ 15]; the paper [ 15] also
discusses generalisations of the definition given in Section 7.5.

Problems
7.1 . In L 2 (JRn), show that V(IPI) - nk=1, ... ,nV(Pk) and that the following identities
hold for f E V(IPI): n

IIIPifll ==
2
L 11Pkfll
2
, (7.147)
k=1
n

II\P)JII 2
- 11!11 2
L 11Pkfll
2
· (7.148)
k=1
7 . 2 . Show that s-lim d-+OCJ r Ad == 0.
7 . 3 . Let n > 2 and use the following notation for vectors in JRn: == (x, xn) and x
1
k == (k, kn), where x, k E JR_n- [e.g. x == (x1, ... , Xn-1)]. For v > 1/2 and fL E JR,
let Mv(fJ,) be the following operator from £ 2 (JRn) to £ 2 (JRn- 1 ), with domain S(JRn):

[Mv(tt)J](k) == [F(J + jQI)-v J] (k, p,) (k E JRn- 1 )


[restriction of the Fourier transform of (I+ IQI)-v f to the (n- I)-dimensional hy-
perplane kn == fL].
(a) Show that Mv(rt) is bounded [hence its closure belongs to B(L 2 (JRn ), £ 2 (JRn- 1 ) )].
(b) Show that fL f--7 M v (JL) is continuous in the operator norm.
Hint for (a): by using the Schwarz inequality in L 2 (JR) (variable Xn), prove that

( 2 ~)n I (1 lxn 1)-v lll2(JR)


2
I [F(J + IQI) -v J] (k, JL) 1 <

_ik. x [ 1 + I n I ] 1/J ( 1
2

JR X
1 JRn-1
d Xn
11
1
e X, Xn X .
X
~-~
X
A ) dn -1 A

Show that this implies that IIMv(M)flli2(JRn-l) < (27r)-nll(l + lxnl)-vlli2(JR)IIJII 2 .


7.4. (a) Verify that, for any self-adjoint operator H in L 2 (JRn), the set Vu(H) (see
page 330) is dense in H.
(b) Let H == P2 V in L 2 (JR. n). Assume that the part of the spectrum of H on (0, oo)
is purely absolutely continuous and that a ac (H) - [0, oo). Show that the set Vt (H)
is dense in Hac(H).
380 FURTHER TOPICS IN SCATTERING THEORY

7.5 . In L 2 (lR3 ) consider a Coulomb Hamiltonian He == H 0 + V(IQI), with Ho == P2


and V(r) == 1/r (! E lR). Let W be a real function defined on JR 3 of the form
TV(x) == (1 lfl)_ 1 _ 8 [woo(x) + w2(x)] with 5 > 0, woo E L 00 (1R 3 ), w2 E L 2 (JR 3 ),
r(8W/8r) E L 2 (JR 3 ) L~(JR 3 ) and r 2 (8 2 W/8r 2 ) E L 2 (JR 3 ) + L 00 (lR 3 ), and let
H == H c + W (Q). Show that H c and H have no singular continuous spectrum, and
that the relative wave operators n±- s-limt~±oo eiHte-iHctEac(Hc) exist and sat-
isfy strong asymptotic completeness [i.e. R(O±) == Hac(H) - Hp(H)j_].
7.6 . For V of class QJ 00 , show that the operators ~(j) defined in Lemma 7.5 map D(H)
into itself and satisfy the commutation relation (7 .58).
Under the conditions of Lemma 7.5, verify that one has g E V(PkQj ).
7 . 8 . Let W: ]Rn ~ CC satisfy J~n(l+x2 )1W(x)l 2 dnx < oo. Show that, as a function
of A, the operator Mw(A) is differentiable in Hilbert-Schmidt norm for A E (0, oo ),
and determine its derivative.
7 . 9. In the context of Section 7.4, find conditions on the potential V such that the av-
eraged total scattering cross section o-(A), Eq. (5.98), is continuously differentiable.
7.10 . Prove Eq. (7.116).
7 . 11. (a) Complete the proof of Proposition 7.21 for the case in which it is only as-
sumed that c5 > k.
(b) Show that (djdA)R(A) is a trace class operator if 5 > n in Proposition 7.21.
7.12 . Show that the operator D 0 in Eq. (7.125) is well defined on Vi(Ho) and sym-
metric on this domain.
References

Books and Monographs

[A] W. 0. AMREIN: Non-Relativistic Quantum Dynamics. Reidel, Dordrecht


(1981).

[ABG] W. 0. AMREIN, A. BOUTET DE MONVEL, V. GEORGESCU: C0 -Groups,


Commutator Methods and Spectral Theory of N -Body Hamiltonians. Birk-
hauser, Basel (1996).

[Ad] R. A. ADAMS: Sobolev Spaces. Academic Press, New York (1975).

[AG] N. I. AKHIEZER, I. M. GLAZMAN: Theory of Linear Operators in Hilbert


Space. Pitman, London (1981).

[AJS] W. 0. AMREIN, J. M. JAUCH, K. B. SINHA: Scattering Theory in Quantum


Mechanics. Benjamin, Reading (1977).

[B] H. S. BEAR: A Primer of Lebesgue Integration. Academic Press, San Diego


(1995).

[BL] J. BERGH, J. LOFSTROM: Interpolation Spaces, An Introduction. Springer,


Berlin (1976).

[Br] M. BRAUN: Differential Equations and Their Applications. Springer, New


York (1993).

[BS] C. BENNETT, R. SHARPLEY: Interpolation of Operators. Academic Press,


Boston (1988).

[BW] H. BAUMGARTEL, M. WOLLENBERG: Mathematical Scattering Theory.


Birkhauser, Basel (1983).

[CFKS] H. L. CYCON, R. G. FROESE, W. KIRSCH, B. SIMON: Schrodinger Op-


erators, with Application to Quantum Mechanics and Global Geometry.
Springer, Berlin ( 1987).
382 REFERENCES

[CK] M. CAPINSKI, P. E. KOPP: Measure, Integral and Probability. Springer,


London (2004).

[CL] A. CODDINGTON, N. LEVINSON: Theory of Ordinary Differential Equa-


tions. McGraw-Hill, New York (1955).

[CV] M. CARTER, B. VAN BRUNT: The Lebesgue-Stieltjes Integral. Springer,


New York (2000).

[D] G. BARRA: Introduction to Measure Theory. Van Nostrand, New York


(1974).

[DK] M. DEMUTH, M. KRISHNA: Determining Spectra in Quantum Theory. Birk-


hauser, Boston (2005).

[DS] N. DUNFORD, J. T. SCHWARTZ: Linear Operators, Part II: Spectral Theory.


Interscience Publishers, New York (1963).

[FJ] F. G. FRIEDLANDER, M. JOSHI: Introduction to the Theory of Distributions.


Cambridge Univ. Press (1982, 1998).

[GO] B. R. GELBAUM, J. M. . OLMSTED: Counterexa1nples in Analysis. Hol-


den-Day, San Francisco (1964).

[GS] I. M. GEL'FAND, G. E. SHILOV: Generalized Functions, Vol. I. Academic


Press, New York (1964).

[H] P.R. HALMOS: Measure Theory. Van Nostrand, Princeton (1950).

[HS] E. HEWITT, K. STROMBERG: Real and Abstract Analysis. Springer, New


York (1965).

[K] T. KATO: Perturbation Theory for Linear Operators. Springer, New York
(1976).

[L] S. LANG: Real Analysis. Addison-Wesley, Reading MA (1973).

[N] M.A. NAIMARK: Linear Differential Operators. Harrap, London (1968).

[Na] I. P. NATANSON: Theory of Functions of a Real Variable, Volume I. Ungar,


New York (1964).

[Ne] R. G. NEWTON: Scattering Theory of Waves and Particles. McGraw-Hill,


New York (1966).

[P] D. B. PEARSON: Quantum Scattering and Spectral Theory. Academic Press,


London (1988).

[R] . L. ROYDEN: Real Analysis. Macmillan, New York (1968).


REFERENCES 383

[Ra] RANA: An Introduction to Measure and Integration. American Math.


Society, Providence RI (2002).

S. ROFE-BEKETOV, A.M. KHOLKIN: Spectral Analysis of Differential


Operators. World Scientific, New Jersey (2005).

lRN] RIESZ, B. Sz.-NAGY: Le~ons d'analysefonctionnelle. Gauthier-Villars,


(1955).

[RS] B. SIMON: Methods of Modern Mathematical Physics, Vols. I-IV.


Academic Press, New York (1972-1979).

[ST] M. SAMUELIDES, L. TOUZILLIER: Analyse .fonctionnelle. Cepadues-Edi-


tions, Toulouse (1989).

[W1] WEIDMANN: Linear Operators in Hilbert Space. Springer, New York


(1980).

[W2] WEIDMANN: Lineare Operatoren in Hilbertriiumen, Teil I: Grundlagen,


Teil II: Anwendungen. Teubner, Stuttgart (2000, 2003).

Articles

[1] W. 0. AMREIN, V. GEORGESCU: On the Characterization of Bound States


and Scattering States in Quantum Mechanics, Helv. Phys. Acta 46, 635-658
(1973).

[2] W. 0. AMREIN, K. B. SINHA: Tilne-Delay and Resonances in Potential


Scattering, J. Phys. A: Math. Gen. 39,9231-9254 (2006).

[3] W. 0. AMREIN, L. ZULETA: Flux and Scattering into Cones in Potential


Scattering, Helv. Phys. Acta 70, 1-15 (1997).

[4] M. DAUMER, D. DURR, S. GOLDSTEIN, N. ZANGHI: On theflux-across-


suifaces theorem, Lett. Math. Phys. 38, 103-116 (1996).

[5] C. A. A. DE CARVALHO, H. M. NUSSENZVEIG: Time Delay, Phys. Rep.


364, 83-174 (2002).

[6] S. A. DENISOV, A. KISELEV: Spectral Properties of Schrodinger Operators


with Decaying Potentials, Proceedings of Symposia in Pure Mathematics,
VoL 76, Part 2, 565-589, American Math. Society, Providence (2006).

[7] D. DURR, MOSER, P. PICKL: The Flux-Across-Suifaces Theorem under


conditions on the scattering state, J. Phys. A: Math. Gen. 39, 163-183 (2006).
384 REFERENCES

[8] V. ENSS: Asymptotic Completeness for Quantum Mechanical Potential Scat-


tering, Comm. Math. Phys. 61, 285-291 (1978).

[9] W. HUNZIKER, I. M. SIGAL: The quantum N-body problem, J. Math. Phys.


41, 3448-3510 (2000).

[10] A. JENSEN, E. MOURRE, P. PERRY: Multiple commutator estimates andre-


solvent smoothness in quantum scattering theory, Ann. Inst. Henri Poincare
41,207-225 (1984).

[11 A. JENSEN, S. NAKAMURA: Mapping Properties of Wave and Scattering


Operators for Two-Body Schrodinger Operators, Lett. Math. Phys. 24, 295-
305 (1992).

[12] KATO: Smooth operators and commutators, Studia Mathematica 31, 535-
546 (1968).

[1 D. B. PEARSON: Singular Continuous Measures zn Scattering Theory,


Comm. Math. Phys. 60, 13-36 (1978).

[14] D. W. ROBINSON: Propagation Properties in Scattering Theory, J. Austral.


Math. Soc. 21 (Series B), 474-485 (1980).

[15] R. TIEDRA DE ALDECOA: Tilne delay for dispersive systems in quantum


scattering theory, preprint http://arxiv.org/abs/081 0.1032 (2008), to appear
in Rev. Math. Phys.
Notation Index

A,A*,A**,29,51 EA, 141


A- 1 34 Ef, 151
'
A, 48,49 Eac, 161
A c B, 47 Eo(·), 212
Ao, 97, Ill £, £, 9
AM, 161
A, 14 f,60
AB, 16 [j, g ]x, 113
AL, 18 FR, 206
F,60
b, 74 F*, 61
B(H), 27
B(H, K), 36 Gc(z), 275,300
B2, B2 (H), 44
B1 (H), 47 Ho,63
H' 1
co, 328 H*, 12
Ck,110 Hp(A), 159
Ca((a, b)), 22 He (A), Hs (A), 159
Cf)(JRn), 22 Hac (A), Hsc (A), 159
~' 16 HI!, HRm, 122
dajdw,242 1,35
D, 270 Io(g, t), 325
Do, 366
D(A), 47, 151, 195 Ko, 104
D(A*), 51 K((J, !), 107
Dv(H), 330 K(H), 42
D/;(H), 330 2
D-j;, 366 f ' 11
LP(O, m), 19
ek,2 L 00 (0, m), 57
E(V), 142 L 2(V), 21
386 NOTATION INDEX

L 2 (S 2 ), 122 Rz, R~, 305


L 2 (S 71 - 1 ), 229 R(A), 34, 47
L 2 (0;JC, rn), 20
78 s -lirn, 4, 31
'78 supp {E.A}, 144
LC:, 78 S,220
L2 63'
S(A),230
S(IRS.), 21
rn-a.e., 9 S(Il{n ), 22
mB, 16
rnF, 134 Tr,46
17Lj, 145 Tz,351
1np' m;ac' msc' 138 T(A), 372
1\Iw(A), 352,355
M,MN,9 u -li1n, 31
M(V), 143 U, 39
Mo(H), 206 Ut, 154
Mta(H), M~(H), 206 uP,218
Moo(H), Moo(H), 206 Ue, 269
Uo, 229
No(H;h), 334
N-:, 275, 300 m(X), 332
N(A), 47
w-lim, 4, 31
I' 9
N,2
W[f, g], 115

0, 35 Yf, 122
(0, A), (0~ A, m,), 14 ~(j)' 331
0, 1
Zt, 222
60
' 100
a,a,2
'60
ro, 328
'306
r(A), 50
'37
fo, 328
' 130
oj,306
'244
bg' 232
P(h;C), 236,324
JP 0 , 103 ~' 63
~s, 123
Q, 59 ~T(J), 365
Q, ,60 v+, 89
(Q), 306 v_,89
II, JIIJ, 6
R, 239 p(A), 68
R(A), 239, 358 a(A), 69
NOTATION INDEX 387

0" ac (A), 0"sc (A), 16 3 II. liP, 19


O"c(A), O"s(A), 163 ll·lloo, 57
o-d (A), 164 II· I II· 112,2
0" ess (A), 164 11·11, 28
o-p(A), 163 (-, ·) K' 20
o-(j;C),237 II· llv, 48
O"tot' I · IIHs, 43
o-()t), (-,. ), 2
cp(A), 153 II·IIK, 20
<Po, 237, 324 [·,·]x, 113
Xv, 15 _L, 9
n, 38,40 11
n±, 219, 253 C,47
O±(A,B;E), 214 0,14
0, 1
Subject Index

A-bound, 72 boundary conditions (Continued)


A-bounded operator, 72 Neumann, 107
A -compact, 77 periodic, 107
absolutely continuous separated, 107
function, 98-99 bounded
measure, 137, 139 from below, 74, 144
spectrum, 163 operator, 28
criterion for, 17 6, 177, 280,
284 C*-algebra, 32
subspace, 159 Cantor
absorbed state, 211-212 function, 137
adjoint, 29, 51-53 set, 16
algebraic scattering theory, 213, 253 Cauchy sequence, 2
almost Cayley transform, 89-92
all, 19 chain rule, 262
everywhere, 21 characteristic function, 15
asymptotic closable, 49
completeness, 227-228, 317-324 Closed Graph Theorem, 68
generalised, 22 7 closure
in the geometric sense, 22 7 of a manifold, 9
strong, 227 of an operator, 48-50
condition,213-214,218-220 commutator, 48
compact
Banach algebra, 32 operator, 41-43
Borel spectrum of, 167, 191
a-algebra, 16 support, 22, 144
measure, 16 complete space, 2
null set, 16 configuration space, 63
set, 16 conjugate operator, 283
bound~ate,204-212 conjugation, 94
boundary conditions, 107 conservation of energy, 228-229
Dirichlet, 107 continuity
mixed, 107 in norm, 33
390 SUBJECT INDEX

continuity (Continued) domain


strong, 6, 33 of an operator, 4 7
subspace of, 159 of regularity, 95
continuous spectrum, 163 Dominated Convergence Theorem, 19
convergence dual, 12
in norm, 31
strong, 4, 31-32 eigenvalue, 69, 158
uniform, 31 of symmetric operator, 70
weak, 4, 31 Eisenbud-Wigner time delay operator,
Cook Criterion, 217 372
core, 54 energy shell, 230
Coulo1nb equivalent functions, 19
Hamiltonian, 80, 126, 166, 170, essential
210 spectrum, 164, 190
potential, 78 invariance, 169
scattering, 252-259 supremum, 57
essentially
decomposable operator, 233 bounded, 57
deficiency self-adjoint, 54
index, 89 everywhere defined, 4 7
subspace, 95 evolution group, 193-197, 199-204
dense subset, 8 extension, 4 7
density matrix, 4 7
derivative, 6, 33 final set, 40
diagonalisable operator, 233 finite
differential measure, 14
expression, 96-98 rank operator, 41, 83
formally self-adjoint, 97 first resolvent equation, 68
self-adjoint, 97 First von Neumann Formula, 92
inequalit~263-265,271-273 ftux,237,324
operator, 36, 96 across surface, 237, 327-329
first order, 100-104 integrated, 3 25
maxi1nal, 97 Flux-Across-Surfaces Theorem,
mimimal, 97 325-327,339-342
second order, 104-11 0 Fourier transform, 60-63
dilation group, 269-271 free evolution group, 218, 220-223
dimension, 2 modified, 252-258
Dirac measure, 135 free Hamiltonian, 63-64, 81, 213
Dirichlet domain of, 63-64
conditions, 107 partial wave decomposition,
Ha1niltonian , 107 123-124, 129-131
discrete spectral representation, 229-230
measure, 135 function
spectrum, 164 m-integrable, 15
distribution, 97, 13 1 absolutely continuous, 98-99
SUBJECT INDEX 391

function (Continued) integral (Continued)


characteristic, 15 Rien1ann-Stieltjes, 1
equivalent, 19 spectral, 145-15 , 344
essentially bounded, 57 integral operator, 36
measurable, 14, 20 interpolation, 310-314
sin1ple, 14 intertwining relation, 215
singular continuous, 13 7 inverse operator, 50, 83, 87
step, 23 invertible
functional calculus, 152-157 in B(H), 35
fundamental self-adjointness criterion, operator, 34, 50, 83, 87
88 isometry, 38-39
partial, 40
generator of dilations, 269-271
graph, 50-51 kernel, 36
norm, 51 Kupsch-Sandhas Theorem, 225
Green's function, 65, 132
P(O, m), 19-25
H -smooth, 292
on an interval, 293
(0, m), 57
Lavine's Theorem, 318
Hamiltonian
Lebesgue
free,63-64,81, 123-124,
129-131,213 a-algebra, 18
radial, 123-126 integral, 15-1 7
self-adjointness, 78-80, 124-126 measure, 18
total, 213 Lebesgue Decomposition Theore1n,
Hausdorff-Young inequality, 80 139
Heaviside function, 135 Lebesgue-Stieltjes integral, 140
Hilbert space, 1-13, 25-26 limit circle case, 114-121 132
dimension, 2 limit point case, 114-121, 132
dual, 12-13 limiting absorption principle, 268
Hilbert-Schmidt for Schrodinger operator, 297
kernel, 45 linear
norm, 43 manifold, 8
operator, 43-47, 353 spanned by N, 8
Holder inequality, 80 operator, 27, 4 7
Lippmann-Schwinger equations, 347
identity operator, 35 localising operators, 206
infinitesimal generator, 195, 197 long range potential, 224, 252-259,
initial set, 40 341
integrable function, 15 lower
integral bound, 144
Lebesgue, 15-17 semibounded, 74
Lebesgue-Stieltjes, 140
of vector-valued function, 6-8 m,-almost everywhere (rn-a. e.), 19
Riemann, 6-7, 17 m-integrable function, 15
392 SUBJECT INDEX

maximal operator
operator, 97 adjoint, 29-30, 51-53
symmetric operator, 93 bounded,27,48
Mean Ergodic Theorem, 201 closable, 49-50
measurable compact, 41-43
function, 14, 20 conjugate, 283
space, 14 decomposable,233
measure, 14 diagonalisable, 233
O"- finite, 14 differential, 36, 96
absolutely continuous, 137 essentially self-adjoint, 54
Dirac, 135 everywhere defined, 4 7
discrete, 135 finite rank, 41, 83
finite, 14 Hilbert-Schmidt, 43-47, 353
outer, 18 integral, 36
pure point, 136 inverse, 50, 83, 87
singular continuous, 13 7 invertible, 34, 50, 83, 87
spectral, 143-145 in B(H), 35
Stieltjes, 134-139 isometric, 3 8
measure space, 14 linear, 27, 47
complete, 18 maximal, 97
minimal operator, 97 maximal symmetric, 93
modified wave operators, 252-258 minimal, 97
momentum multiplication, 56-60
operator, 60, 102 norm, 28
space, 63 normal, 191
Mourre inequality, 283 positive, 141
strict, 273 product, 48
Mourre method, 271-292 projection, 37
n1ultiplication operator, 56-60 range of, 34
spectrum of, 166 relatively bounded, 72-78
n1utually singular measures, 139 relatively compact, 77-78
scattering, 219-220
Nelson's Criterion, 197 Schrodinger, 97
Neumann self-adjoint, 30, 53
conditions, 107 sum, 47
series, 35 symmetric, 54-55
norm time delay, 372
Hilbert-Schmidt, 43 trace class, 4 7
operator, 28, 48 unitary, 39-40
vector, 3 operator-valued function, 33
normal operator, 191 continuity, 33
null derivative, 33
set, 14 differentiability, 33
space, 47 integration, 33-34
SUBJECT INDEX 393

orthogonal Rellich-Kato Theorem, 74


complement, 9 resolution of the identity, 141
sum, 11 resolvent, 68-69, 155, 196
vectors, 9 equation
orthonormal first, 68
basis, 2 second, 76
sequence,4,9 estimates, 274-281, 284-286
higher order, 300-306
partial set, 67
isometry, 40 restriction, 47, 161, 191
wave to hypersurface, 328-330,
decomposition, 121-123 374-375
subspace, 123 Riemann integral, 6-7, 17
partition, 6 Riemann-Lebesgue Lemma, 199
perturbation, 72 Riemann-Stieltjes integral, 140
phase shift, 232, 373 Riesz Lemma, 12
point right continuous, 142
of constancy, 144 R-matrix, 239-240
of discontinuity, 134, 143 stationary formula, 358-362
of increase, 144 rotation group, 231
point spectrum, 159, 163
polar decomposition, 192 a--additive, 14
polarisation identity, 3, 30 a--algebra, 14
position operator, 59-60 a--finite measure, 14
positive operator, 141 scalar product, 2
potential, 78 scattering
long range, 224, 341 amplitude, 242
short range, 224 crosssection,235-252
probability measure, 14 averaged total, 242, 245, 249-
projection, 37-38 252
Projection Theorem, 10 differential, 242
propagation estimate, 324, 333, 342 finiteness, 243, 249-252
pure point 1neasure, 136 for a cone, 23 7
Putnam's Theorem, 177 total, 242
into cones, 238-242
radial Hamiltonian, 123-126 operator, 219-220,226-232,239
Radon-Nikodym stationary expression, 350
derivative, 140 unitarity, 227
Theorem, 140 phase shift, 23 2
RAGE Theorem, 201 state, 204-212
range, 34, 47 Schrodinger operator, 97
regularisation, 24 one-dimensional, 111-121, 191
relative bound, 72 spectral properties, 286-292
relatively bounded operator, 72-78 wave operators, 223-226,
relatively compact operator, 77-78 322-324
394 SUBJECT INDEX

Schwartz space, 21 Stark Hamiltonian, 315


Schwarz inequality, 3 step function, 23
second resolvent equation, 76 Stieltjes measure, 134-139
Second von Neumann Formula, 93 Stone's
self-adjoint, 30, 53 Formula, 171
criterion, 88 Theorem, 194
separable space, 2 strict Mourre inequality, 273
short range potential, 224 strong
simple function, 14 asymptotic completeness, 227
singular continuous convergence,4,31-32
function, 13 7 Sturm-Liouville operator, 131
measure, 13 7 subspace, 9
spectrum, 163 absolutely continuous, 159
singular spectrum, 163 continuous, 159
S-matrix, 230, 232 singular continuous, 159
differentiability, 362-363 spanned by N, 9
stationary formula, 35 8-362 support, 22
smoothness compact, 22, 144
local, 293 of spectral family, 144
relative, 292-295 supported on, 135
Sobolev space, 110 symmetric operator, 54-55
sojourn time, 364 spectrum, 88, 95
spectral
family, 141-145 Three Line Theorem, 311
of an operator, 151 time delay, 364-373
support, 144 and S-matrix, 372
integral, 151, 344 Eisenbud-Wigner operator, 372
measure, 143-145 for a beam, 373
representation, 15 2, 229 global, 365, 369-373
Spectral Theorem, 151-152, 173-174, local, 365-366
179-189 partial wave, 373
spectrum, 69-71 , 190 time evolution, 205
absolutely continuous, 163 total
continuous, 163 evolution group, 218
discrete, 164 Hamiltonian, 213
essential, 164, 190 set, 8
of self-adjoint operator, 70 trace, 46
of symmetric operator, 88, 95, class, 47
167 transition operator, 244, 352-358
point, 159, 163 derivative, 362
singular, 163 differentiation, 362, 376-377
singular continuous, 163 translation group, 104, 197-198
spherical triangle inequality, 3
harmonics, 122, 123
Laplacian, 123, 377-378 Uniform Boundedness Principle, 5
SUBJECT INDEX 395

uniform convergence, 31 wave function, 20


unitary operator, 39-40 wave operators, 214-218
upper semibounded, 74 completeness, 227, 322-324
existence, 224, 226, 318
vector, 2 integral expressions, 345
of compact support, 144 modified, 252-258
vector-valued function, 6 spectral integrals, 347-348
continuity, 6 weak convergence, 4, 31
derivative, 6 Weyl's
differentiability, 6 Criterion, 164
integration, 6-8 Alternative, 114
Virial Theorem, 283 Weyl-von Neumann Theorem, 170
von Neumann Wronskian, 115
algebra, 32
Formula zero
First, 92 operator, 35
Second,93 vector, 1

You might also like