Maslov Operational Methods Mir 1976

operational
methods
V. P. MASLOV MIR PUBLISHERS
MOSCOW
B A
1
ee =2 n
= (A+B)
n!
operational
methods
V. P. MASLOV _ _ __
Translated from the Russian by
V. GOLO, N. KULMAN, G. VOROPAEVA
Mir Publishers
Mosco111
First published 1976
Revised from the 1973 Russian edition
This translation has been read and approved by the author

Professor V.P. Maslov, D.Sc. (Math.)
H a ane.n.uuc"oM Habme
© English translation, Mir Publishers, 1976

CONTENTS
Preface . . . . . . . . . . . . . . . . . . . . 7
Introduction to Operational Calculus . . . . . . . 13
Sec. 1. Solution of Ordinary Differential Equations by the Heaviside
Operational Method . . . . . . . . . . . . . . . . . . . 13
Sec. 2. Difference Equations . . . . . . . . . . . . . . . . . . • 20
Sec. 3. Solution of Systems of Differential Equations by the Heaviside
Operational Method . . . . . . . . . . . . . . . . . . . 22
Sec. 4. Algebra of Convergent Power Series of Noncommutative Ope-
rators .............. . 24
Sec. 5. Spectrum of a Pair of Ordered Operators . . . . . ·. 35
Sec. 6. Algebras with ~-t-Structures . . . . . . . . . . . . 40
Sec. 7. An Example of a Solution of a Differential Equation 56
Sec. 8. Passage of the Equation of Oscillations of a Crystal Lattice
into a Wave Equation . . . . . • . . . . . . . . . . . . 58
Sec. 9. The Concept of a Quasi-Inverse Operator and Formulation of
the Main Theorem . . . . . . . . . . . . . . . . . . . . 100
Chapter I
Functions of a Regular Operator 147
Sec. 1. Certain Spaces of Continuous Functions and Related Spaces 149
Sec. 2. Embedding Theorems . . . . . . . . . . . . . 154
Sec. 3. The Algebra of Functions of a Generator . . . . . . . . . 158
Sec. 4. The Extension of the Class of Possible Symbols . . . . . 173
Sec. 5. Homomorphism of Asymptotic Formulas. The Method of Sta-
tionary Phase . . . . . . . 181
Sec. 6. The Spectrum of a Generator . . . . . . . . . . . . . . . 188
Sec. 7. Regular Operators . . . . . . . . . . . . . . . . . . . . 194
Sec. 8. The Generalized Eigenfunctions and Associated Functions 198
Sec. 9. Self-Adjoint Operators as Transformers in the Schmidt Space 205
Chapter II
Calculus of Noncommutative Operators 210
Sec. 1. Preliminary Definitions . . . . . . . . . . • . . . . . . 210
Sec. 2. The Functions of Two Noncommutative Self-Adjoint Operators 224
Sec. 3. The Functions of Noncommutative Operators 228
Sec. 4. The Spectrum of a Vector-Operator 231
Sec. 5. Theorem on Homomorphism . . 239
Sec. 6. Problems . . . . . . . . . . . 242
Sec. 7. Differentiation of the Functions of an Operator Depending
on a Parameter . . . . . 251
Sec. 8. Formulas of Commutation . . . 256
Sec. 9. Growing Symbols . . . . . . 261
Sec. 10. The Factor-Spectrum . . • . • . . 265
6 CONTEN'IB
Sec. 11. The Functions of Components of a Lie Nilpotent Algebra and

Their Representations . . . . . . . . . . . . . . . 266
Chapter III
Asymptotic Methods 273
Sec. 1. Canonical Transformations of Pseudodifferential Operators 273
Sec. 2. The Homomorphism of Asymptotic Formulas 294
Sec. 3. The Geometrical Interpretation of the Method of Stationary
Phase 301
Sec. 4. The Canonical Operator on an Unclosed Curve . . 303
Sec. 5. The Method of Stationary Phase • 312
Sec. 6. The Canonical Operator on the Unclosed Curve Depending on
Parameters Defined Correct to 0 ( !) 315
Sec. 7. V-Objects on the Curve 321
Sec. 8. The Canonical Operator on the Family of Unclosed Curves 327
Sec. 9. The Canonical Operator on the Family of Closed Curves 333
Sec. 10. An Example of Commutation of a Canonical Operator with
a Hamiltonian 339
Sec. 11. Commutation of a Hamiltonian with a Canonical Operator 346
Sec. 12. The General Canonical Transformation of the Pseudodifferential
Operator 348
Chapter IV
Generalized Hamilton-Jacobi Equations 355
Sec. 1. Hamilton-Jacobi Equations with Dissipation 356
Sec. 2. The Lagrangean Manifold with a Complex Germ 360
Sec. 3. y-Atlases and the Dissipativity Inequality 372
Sec. 4. Solution of the Hamilton-Jacobi Equation with Dissipation 378
Sec. 5. Preservation of the Dissipativity Inequality. Bypassing Focuses
Operation 386
Sec. 6. Solution of Transfer Equation with Dissipation 401
Chapter V
Canonical Operator on a Lagrangean Manifold with a Complex
Germ and Proof of the Main Theorem . . . . . . . . . . 419
Sec. 1. Quantum Bypassing Focuses Operation . . . . . . . . . 419
Sec. 2. Commutation Formulas for a Complex Exponential and a
Hamiltonian . . . . . . . . . . . . . . . . • • . 440
Sec. 3. C-Lagrangean Manifolds and the Index of a Complex Germ 452
Sec. 4. Canonical Operator 469
Sec. 5. Proof of the Main Theorem . . . . 482
Appendix to Sec. 5 . . . . . . . 493
Sec. 6. Cauchy Problem for Systems with Complex Characteristics 503
Sec. 7. Quasi-Inverse of Operators with Matrix Symbols 519
Appendix. Spectral Expansion of T-products 545
Index . . . . . . . • • . • . . . . . . . . . . 557
PREFACE
Operational methods are such methods, which make it possible

to reduce differential problems to algebraic problems. This is why
these methods are of particular use to specialists dealing with applied
mathematics.
This book is devoted to one, but sufficiently general operational
method, which absorbs many operational methods known to date
and allows for the uniform solution of both classical problems,
involving differential equations with partial derivatives, and the
absolutely new problems of mathematical physics, including those
connected with non-linear equations in partial derivatives.
We shall proceed to describe this general method after studying
the methods well-known in mathematics for calculating operators
(mainly self-adjoint operators), for which Heaviside's method served
as a source.
The main theorem stated in this book belongs to the theory of
operators and is proved in the last chapter, but its formulation is
given in Introduction as well. This theorem may lead, in particular,
to the existence and uniqueness theorems for hyperbolic, elliptic
and parabolic equations with variable coefficients and allows to
reduce them to integral equations of the second kind with smooth
kernels, i.e., to provide an effective solution for these equations.
We have in mind that by detaching from the solution the non-smooth
(or rapidly oscillating) components, we can thus reduce the problem
to the one that is easily solved on an electronic computer. When
the rapidly oscillating part detached from the solutions was com-
pared with the exact solution by mearis of a numerical experiment
in the one-dimensional example of Sec. 8 of Introduction, their
near coincidence was revealed. It turns out, however, that the com-
puter is unable to perform such an experiment even in a two-dimen-
sional case because of the enormous number of operations. Under
these circumstances the only thing to do is using suitable asymp-
totics and reducing the initial problem to an integral equation.
Thus, the approximate solution constructed in the main theorem,
8 OPERATI'ONAL METHODS
detaching the non-smooth or rapidly oscillating part, serves as

a kind of natural supplement to the electronic computer: together
they provide a numerical answer.
This book on operational methods should be accessible to senior-
course students of mathematics and physics faculties at universities
and departments of applied mathematics. This means that only
a knowledge of classical analysis is required of the reader. The
book provides explanations in sufficient volume of such concepts
as the theory of Banach algebras of distributions (Chapter I), the
theory of linear differential and difference equations (Sees. 1, 2,
and 3 of Introduction), the theory of non-linear equations of the
first order with partial derivatives (Chapter IV).
This material may be also of use to the reader who is already
familiar with these questions, because rather often it is not pre-
sented in traditional style, and adapted for further reference. The
reader who studies the book thoroughly will be equipped to carry
on independent research in the modern theory of linear, non-linear
differential and differential-difference equations with partial derivat-
ives. Besides, a study of the concrete problems presented here may
serve as an excellent springboard for further investigation, although
from a certain viewpoint, of the theory of representations, topology
and the theory of sheafs. An example of the latter is the theory of
V-objects (Chapter III). As to applying the described methods in
physics, quite obviously, it is not limited to the examples given
in the book. Additional mention should be made of the asymp-
totics of solutions in the band solid-state theory, in problems of
molecular collisions, in the theory of laser resonators, in chain-
reaction equations in chemistry, in problems of refraction and
diffraction, in derivation of integral equations of the Lippmann-
Schwinger and Faddeev type, in calculating quasi-classical amend-
ments to the Thomas-Fermi equation, amendments to electronic
plasma equations, asymptotic solutions of Hartree's equations, in
electronic optics, in problems of above-barrier reflection and in
many other problems of modern mathematical physics.
This book has been written in such a way as to serve the widest
possible circle of readers. It is suitable for two methods of study.
The reader, who seeks to avoid fine assessments and passing to the
limit and only wishes to master the practical techniques for obtaining
asymptotic solutions, may omit that part of the book which is
devoted to functional analysis. The "Introduction to Operational
Calculus" has been written with this purpose in view; having studied
it, the reader can tackle operator techniques (omitting Chapters I-II)
learning how to reduce concrete problems to an integral equation,
detaching the non-smooth part of the solution.
In following the second method of learning, which consists in
the gradual and deeper study of operational methods, it is best
PHEFACE 9
to begin with the first chapter and to read Introduction only after
Chapter II.
The most effective way of mastering the subject, however, consists
rather in first reading Introduction and then reading all the book
in succession.
The reader should nevertheless be warned that all these methods
are not at all easy, because the book provides a new operational
calculus-the calculus of ordered operators.
This book is actually a synopsis of the course of lectures delivered
by the author in the duration of three years at the Department of
Applied Mathematics at the Moscow Institute of Electronic Machine-
Building, consecutively to third-, fourth- and fifth-year students.
The material of the last chapters was also given in lectures (in the
duration of three years) to fifth-year (graduating) students specializ-
ing in mathematics at the Physics Department of the Moscow State
University. Besides, this course of lectures (an abridged version) was
also delivered by the author at several mathematical schools (at the
International Mathematics School at Sopot in 1971, in Pushchino
in 1971, in Voronezh in 1972, and other places).
The experience gained in delivering this course of lectures showed
that, despite expectations, operational calculus, with the parallel
examination of comparatively few examples at seminar sessions,
is grasped more easily by students than questions connected with
traditional functional analysis, fine assessments, functional spaces
and passing to the limit.
The students of senior courses and post-graduates can quickly
learn to solve complex problems connected with operational methods.
In this book, for instance, the reader will find important formulas
obtained by M. V. Karasev (Theorems 4.4 and 6.6 of Introduction)
and V. G. Danilov (Theorem 1.1 in Chapter III).
If we should draw an analogy between the exposition of operational
calculus given in the "Introduction to Operational Calculus" and
the hypothetical exposition of differential calculus (see the left
column) the approximate result would consist in the following:
Sees. 1-3 deal with the ring Sees. 1-3 deal with the Heav-
of polynomials. iside method and operator cal-
culus with constant coefficients.
Sees. 4-5. The rules of the Sees. 4-5. The rules of the new
formal differentiation of poly- calculus are derived on the
nomials are given; various for- example of formal power series
mulas of differentiation for po- of ordered operators. Formulas
lynomials are derived: formulas are derived for commutators of
for differentiating the product; series, composite function and
the composite function, ex- expansion in a Newton series.
pansion in a Taylor series.
10 OPERATIONAL METHODS
In Sec. 6 the system of In Sec. 6 the system of axioms

axioms is introduced and the is introduced and the formulas
same formulas are derived for for arbitrary functions in the
arbitrary functions in the form form of theorems are the same,
of theorems, thus determining thus determining all formulas of
all formulas of the techniques the techniques of the calculation
of differentiation. of ordered operators.
Sec. 7 gives a simple Sec. 7 gives a simple
example. example.
Sec. 8 shows how a new Sec. 8 shows how new physical
solution for a classical physics effects are obtained with the
problem is obtained with the use of introduced techniques in
use of introduced techniques. studying the classical problem
of deriving the wave equation
from equations of the oscillat-
ions of a crystal lattice.
Sec. 9 sets a problem on the The main problem is formu-
simplest differential equation. lated in Sec. 9, the fundamental
The fundamental concept of a concept of characteristics for
characteristic polynomial is in- functions of an ordered set of
troduced and the physical operators is introduced, and the
meaning of initial conditions is physical meaning of absorption
discussed. The main theorem of conditions is discussed. The main
existence is formulated and the theorem is formulated and the
formula for the solution of an explicit formula is given for
equation in a particular case is the solution of the main pro-
given. blem in a particular case (the
general formula is given in the
last chapter).
Further (if we are to continue the analogy) the theory of the
derivative as a limit (Chapter I) and partial derivatives as multiple
limits (Chapter II) is conducted in succession.
The theory of functions in Chapter I is built by systematic use
of the conception of "completing with limits", just as real numbers
in Cantor's theory are elements of the completion of a set of sequen-
ces of rational numbers. This point of view is in accord with the
initial conception of physicists with regard to Dirac's delta function
as the limit of bell-shaped functions.
Sees. 3-9 of Chapter I are devoted to calculating the functions of
one operator. These sections introduce the concepts of generating
and regular operators generalizing the concepts of self-adjoint and
normal operators, respectively. The theorem is proved for regular
operators which shows their generality in the case of a discrete
spectrum: the regularity of an operator is a necessary and sufficient
PREFAC.E 11
condition for the completeness of eigen and associated elements.

The known calculation of self-adjoint operators is derived as a con-
sequence.
At the beginning of Chapter II the calculation of functions of
two noncommutative ordered operators, their joint spectrum and
spectral expansion are studied in detail. Then the functions of
several regular operators are examined and formulas for them are
derived, which had been obtained from axioms in Introduction.
All the techniques for calculating noncommutative operators are
thus built, but on a functional basis now. Operational calculus as
such ends with this.
The rest of the book is devoted to a special transformation making
it possible to prove the main theorem. This transformation is called
the canonical operator.
A great deal of preparatory work is conducted in Chapter III.
A canonical operator is introduced in the simplest real one-dimen-
sional case (depending on ordered operators).
Chapter IV calls for an examination in greater detail. This part
of the book is required for the ultimate construction of a complex
canonical operator, but, generally speaking, the chapter is quite
detached and in no way connected with the techniques of ordered
operators. It may even be read all at once. The chapter is devoted
to constructing the solution, in the large, of equations generalizing
Hamilton-Jacobi equations.
We introduce the notion of Lagrangean manifold with a complex
germ which results in a geometric interpretation for solutions of
equations of the Hamilton-Jacobi type.
The following physical analogy may be cited to illustrate the
point. When a stone is dropped into water it causes the waves to
spread over the surface in even circles at first; when the wave is
reflected the picture that is produced can hardly be given a geo-
metric interpretation. The same may be said of the Hamilton-
Jacobi equation (a particular solution of a certain Hamilton-Jacobi
equation will be precisely the one that causes these circles) which
on intervals of time not exceeding a certain t1 has a simple and smooth
solution. Further, when t > t 1 there occurs a similarity of imposition
of waves reflected many times. In order to gain a clear understanding
of multi-valued functions (branching of the solution), it is necessary
to make them uniform in much the same way as this is done by
means of Riemann's sheets for analytical functions with branching.
It is precisely the construction of the object in the phase space of
the Lagrangean manifold with a complex germ that makes it pos-
sible to "unravel" multi-valued solutions of Hamilton-Jacobi type
equations (with absorption). The concept "index"-as a whole
number-introduced here makes it possible to designate these
solutions (sheets) on the Lagrangean manifold with a complex germ.
12 OPERATJ'ONAL METHODS
Besides, the indices of closed paths on the Lagrangean manifold

with a complex germ constitute an important characteristic of this
object (the characteristic class of an object).
After this the complex canonical operator is built in Chapter V
and the main theorem is proved.
The results of Chapters 1-V and Sees. 4-9 of Introduction mainly
belong to the author (with the exception of the theorems in Chap-
ter I concerning self-adjoint operators, and the theorems of Karasev
and Danilov mentioned above).
The starting point of this research was Feinman's remark* to
the effect that if the order of action of the operators is determined
by indices, then the operators become as if commuting.
The theory of the Lagrangean manifold with a complex germ
originated as a result of the study of Leray's works on the Cauchy
problem.
A. A. Kirillov, who edited the book, made a number of valuable
suggestions in principle concerning the structure of the book as
a whole.
I benefited largely from my consultations with A. A. Samarsky,
discussions with D. V. Anosov and V. V. Kucherenko, for which
I am deeply grateful to them. I am also very grateful to P. P. Moso-
lov, G. A. Voropaeva, V. L. Dubnov who made several valuable
remarks pertaining to the author's manuscript.
This book, however, would never have been written as a school
aid if not for my pupils, who made notes and worked at this course
of lectures. Moreover, they did not even have detailed synopses of
Chapters III and V at their disposal. The first two chapters were
jotted down by V. L. Dubnov, the third by M. V. Karasev and
V. L. Dubnov, the fourth by V. G. Danilov, the fifth by V. G. Da-
nilov and M. V. Karasev. The chapters IV and V were edited by
G. A. Voropayeva; A. G. Prudkovsky made all the calculations
on a computer. Besides this, our numerous discussions with
V. L. Dubnov, M. V. Karasev, V. G. Danilov, S. Yu. Dobrokhotov,
A. G. Prudkovsky and G. A. Voropayeva rendered me invaluable
assistance. In preparing the manuscript for the publisher I was
also helped by A. G. Davtyan and G. Yu. Malysheva. My gratitude
to them all is boundless.
V. P. Maslov
* In the article "Operational Calculus Relating to Quantum Electrody-

namics" (Phys. Rev., 84, 1951) in the section "Description of Method of Desig-
nation". By the recommendation of the editor, Feinman's calculus has not
been included in the book, just as the applications to physics enumerated on
page 8.
INTRODUC TION TO OPERATIO NAL CALCULUS
Sec. 1. Solution of Ordinary Differential Equations

hy the Heaviside Operational Method
Let Coo (R) (or C00 ) denote the set of infinitely differentiable
functions <p (x), -oo < x < +
oo. The operation (operator), d/dx,
of differentiation with respect to X iS defined in Coo; this operation
will henceforth be denoted by D:
Dcp (x) = :x <p (x) = cp' (x).
Clearly, the operator

n
Pn (D) : Coo-+ Coo given by Pn (D)=~ a;Di, a;D 0 =a;
i=O
is defined for every polynomial Pn (x), where a 1 are complex num-

bers, i = 0, 1, 2, ... , n. That is, by definition,
def n
P n (D) (x), cp<O> = <p.
Let K 1 [D) denote the set of operators of the form Pn(D); for any
two operators in K 1 [D) their sum and product are defined by the
formulas
n n n
2J a;Di + i=O
'<=O
2J (a;+ b;) Di,
2J b;Di = i=O
n n n m
(.2J a;Di} • (.2.1 bid}=~
t=O J=O t=O 3=0
.2J a;biDi+i.
Let K 1 [x] denote the set of polynomials over the complex number
field
n
Pn EKtfx), Pn (x) = '2] a;xi.
i=O
There is a one-to-one correspondence between the operator Pn(D) E

E K1 [D], Pn (D) = a0 +a1 D + ... +
anDn and the polynomial
Pn E K 1 [xJ. (To prove that this is a one-to-one correspondence, it

N
suffices to show that if ~ anr> (x) = 0 for every f E C 00
, then
n=O
an== 0; but this is obvious.) We shall call the polynomial P (x)
corresponding to the operator P (D) the symbol of the operator P (D).
Under this correspondence the sum of the polynomials goes to the
sum of the two corresponding operators and the product of two poly-
nomials goes to the product of the two corresponding operators.
We shall say that K 1 [x] and K 1 [D] are isomorphic by this con-
struction.
By the main theorem of algebra, any polynomial can be written
in the form
m
~ k;=n, (1.1)
i=i
where a; are the complex roots of the polynomial and k; is the mul-
tiplicity of the corresponding root.
It follows from the isomorphism of K 1 [x] and K 1 [D] that for
every operator Pn (D) we have the formula
m k.
Pn (D)= an II (D- a;)
i=i
', (1.2)
where a; is the operation (operator) of multiplication by the con-

stant a;.
We shall now solve an ordinary differential equation with con-
stant coefficients
n
Pn (D) y (x) = ~ a;y< 1) (x) = f (x), f ECoo. (1.3)
i=O
Suppose for the sake of simplicity that all the roots a; of the
polynomial Pn (x) are simple. Then we can put equation (1.3) into
the form
To solve equation (1.3) we need the following lemma.

Lemma 1.1. Let Pn (D) EK 1 [D], f E Coo. The following formula
is valid:
(1.4)
INTRODl,JCTLON TO ,OPERATIONAL GAIJCULUS 15
We prove this by induction. \Ve compute Deaxt (x):

+
Deaxj (x) = aeaxj (x) eaxj' (x) = eax (aj (x) +
j' (x)) =
= eax (D +a) I (x).
Suppose that formula (1.4) is valid for the operator nn-1 • We
prove that it is valid for the operator Dn as well. \Ve have
Dneaxt (x) = D [Dn- 1eaxt (x)l = DF (x),
where F (x) = vn-1eaxt (x). By the induction assumption

vn-1eaxt (x) = eax (D +
ay<-1 j (x);
hence
DF (x) = peax[(D + a)n-1 f (x)]= aeax !(D + a)"-1 f (x)l +
+eaxn !(D + a)n-1 j (x)]= eax (D + a)n j (x).
Thus for any integer m we have obtained the formula
nmeaxi (x) = eax (D + a)m I (x). (1.5)
Equation (1.4) follows immediately from (1.5), and the lemma is

proved.
Now consider the equation
(D -a) y (x) = f (x), f E C"". (1.6)
Using the lemma we obtain
(D - a) y (x) = (D - a) eaxe-axy (x) = eaxne-axy (x).
Thus equation (1.6) is equivalent to

De-axy (x) = f (x) e-ax. (1.6')
It is well known that the general solution of the equation Dy (x) =
+
= f (x) has the form y (x) = ) f (x) dx C, where C is a constant.
Introduce the notation
I' 1
) l(x)dx+C=nl(x). (1.7)
We note that 1/D maps an element IE C"" into an entire class of

functions 1/D I c C""; if y 0 E ~ j, then z E ~ f if and only if z -
-Yo = const.
We can now write down a solution of equations (1.6) and (1.6')
in the form
1
y = eax 75 e-axj (x). (1.8)
We return to equation (1.3). We have

n n
Pn(D)y(x)=an n (D-ai)y(x)=an(D-al) II (D-ai)y(x)=
i=1 i=2
n
= an/'' 1xDe-alx IJ (D- ai) Y (x) = ... =
i=2
= anealxDe-aix ... eanxDe-anx Y (x) = f (x).
Successively applying ('1.8), we obtain the formula

_ a x 1 -a x a x 1 -a x f (x) dcf 1 j ( )
y - e n De n ... e 1 De 1 • ----a;;-= p n (D) X • (1.9)
We have obtained a formula which gives a solution of equation

(1.3) in terms of integrals of the right-hand side. Thus the existence
of a solution is proved. It is easy to verify that the result is inde-
pendent of the order of the roots a 1 , • • • , an of the polynomial
Pn (x). It is clear from the form of the solution that the general
solution of equation (1.3) given by (1.9) depends on not more than
n constants (in fact on exactly n).
Formula (1.9) defines a mapping from an element f E Coo to a class
of functions. If the equation Pn (x) = 0 has simple roots a 1 , • • • , a,,
then the difference of two elements of the class has the form c1 ea1x +
+ c2 ea2x+ ... + cneanx, where c1 , c2 , • • • , Cn are constants.
The method of proof shows that formula (1.9) is valid for multiple
roots as well; in this case the difference of two elements has the form
pl (x) ealx + p2 (x) ea2x + ... + Pm (x) eamx,
where P 1 , P 2 , • • • , Pm are suitable polynomials.
1
We shall denote this mapping (or operator) by Pn (D)'
We now show that, to obtain uniqueness of the solution, it is suf-
ficient to specify the initial Cauchy data
y (0) = Co, ••• , y<n- 1 > (0) = Cn-1• (1.10)
To prove this, it suffices to show that the problem
Pn (D) y (x) = 0, (1.11)
y (0) = 0, ... 1 y<n-1 > (0) = 0 (1.12)
has the unique solution y (x) = 0. Note that the problem (D -
- a) y (x) = 0, y (0) = 0 has the unique solution y (x) = 0.
We rewrite equation (1.11):
Pn (D) y (x) = lii (D-ai)ki y (x) = (D- a 1) W (x) = 0.
INTRODUCTEON TO OPERATIONAL CAI!CULUS 17
Here W (x) is obviously a linear combination of the derivatives

of y (x) of order~ n - 1. Hence W (0) = 0 by (1.12), and therefore,
W (x) = 0. It is clear that by continuing this process we obtain
y (x) = 0.
We now give another method of calculating the class of functions
Pn~D) /, which uses an extension of one variable x.
Let R (x) = P (x)/Q (x), where P (x), Q (x) are polynomials. We
shall correspond to the function R (x) the mapping R (D), which
maps function f E Coo to the class of functions P (D) Q f c Coo. tD)
The function R (x) will be called the symbol of the mapping R (D).
We verify that the mapping R (D) is independent of the way we
write its symbol R (x) in the form P (x)/Q (x). In fact, let P (x) =
= P (x) S (x) and Q (x) = Q (x) S (x), where S (x) is a polynomial.
We now prove that for any f E Coo we have the equation
1
P (D) Q(D/ = -P (D) Q 1(D) f. (1.13)
Let y E P (D) Q tD) f. Then y (x) may be written in the form y (x) =
= P (D) z (x), where z (x) is a solution of the equation Q (D) z (x) =
= f (x). Let u (x) be a solution of the equation S (D) u (x) = z (x);
then Q (D) u (x) = Q (D) S (D) u (x) = Q (D) z (x) = f (x) so that
u ~_(J ~D/ Hence y (x) = P (D) z (x) = P (D) S (D) u (x) =
= P (D) u (x). Thus
- 1
yEP(D)~f.
Q(D)
Conversely suppose yEP (D) _ 1 f. Then y (x) = P (D) u (x),

Q(D)
where u (x) satisfies the equation Q (D) u (x) = f (x). Let z (x) =
= S (D) u (x). Then z (x) satisfies the equation Q (D) z (x) = f (x)
1 1
and y (x) = P (D) z (x) E P (D) Q(D) f (x); hence yEP (D) Q (D)!.
Q.E.D.
Let F [D) be a set of operators R (D). We define an operation
of a sum of two elements in F [D), induced by the corresponding
operation in the field F [x) of rational functions. By definition,
R (D) = R 1 (D) +
R 2 (D) if, and only if, R (x) = R 1 (x) R 2 (x). +
The following lemma holds.
Lemma 1.2. Let R 1 (x), R 2 (x) be elements of the field F [x], R (x) =
= R1 (x) +
R 2 (x). A junction y belongs to a class R (D) f if, and
only if, it may be expressed in the form y (x) = y1 (x) +
y 2 (x), where
Y1 E R1 (D)/, Y2 E Rz (D) f.
2-01225
Proof. Let P, (x), Q, (x) (i = 1, 2) be polynomials and R, (x) =

= P 1 (x)IO, (x) (i = 1, 2). Then R (x) = R 1 (x) R 2 (x) = +
= P (x)/Q (x), where P (x) = P 1 (x) 0 2 (x) 01 (x) P 2 (x), 0 (x) = +
= Q1 (x) Q2 (x). Denote the class of functions of the form y (x) by
R 1 (D) f (x) +R 2 (D) f (x).

We shall prove that R (D) f c: R 1 (D) f R 2 (D) f. Let y (x) E +
E R (D) f (x). Then y (x) = [P1 (D) Q 2 (D) Q1(D) P 2 (D)l z (x), +
where z (x) satisfies the equation
01 (D) 0 2 (D) z (x) = f (x). (1.14)
Let P 1 (D) 0 2 (D) z (x) = y 1 (x), P 2 (D) Q1 (D) z (x) = y 2 (x). The
function u1 (x) = Q2 (D) z (x) satisfies the equation Q1 (D) u1 (x) ==
= f (x) and the function u 2 (x) = Q1 (D) z (x) satisfies the equation
Q2 (D) u 2 (x) = f (x). Hence
Y1 (x) E R1 (D) f (x), Y2 (x) ER 2 (D) f (x)
and this proves the lemma.

Consider an example of the calculation of 1/Pn (D) by the homo-
morphism obtained R (x)-+ R (D). We express the symbol 1/Pn (x)
as the sum of simple fractions
m m
1 ~ ai ~k,=n,
Pnlx)= ( )ki'
i=i X-CGi i=1
where ai are complex roots of multiplicity ki of the polynomial

Pn (x), ai are complex polynomials. It follows from Lemma 1.2 that
Hence, any solution y (x) of equation (1.3) can be expressed in

the form
m
""' a, (D) ea.x
y (x = L! 1 1 -a.xf()
1 D ... De 1 x, (1.15)
i= 1 '---v--"
ki
which is immediately seen from (1.6), (1.6') and (1.8). In the par-
ticular case of the simple roots, (1.15) has the form
n x
y (x) = ~ [ ai eaix ~ e-a,v f (p) dp + Cieaix J. (1.15')
i=1 0
INTRJODUCTLON TO OPERATIONAL .CAliCULUS 19
It is not difficult to verify*, that the general solution of (1.3) has

exactly n constants. Hence, (1.15) provides the general solution
(i.e., any element of the class P ~D) t}
of equation (1.3).
We can also use formula (1.9) to find the solution of the homo~
geneous equation.
Example. Find the general solution of the equation
(D - a 1 ) (D - cx 2) y (x) = 0, a 1 =F a 2 •
Formula (1.9) implies
. ( ) _ a2 x '1 e -a 2 x ea 1:x: ...!_ -a x0 _
!Y x - e 75 De 1 -
_ ea x 1 -a 2xea 1xc _ ea2x '1 e<a 1-a 2 )xC1=
- 2 De 1- 175
= ea2xC1 ~ e<a1-a2>x dx+ C2eazx = c;ea1x + C2e'?:2x.
Now we shall consider the case of multiple roots.
Example. Find the general solution of the equation
(D - a)3 y (x) = 0.
Equation (1.9) implies:
Y (x) = eax ...!.. e-axeax...!..
D D
e-axeax...!..
D
e-axO -_
1 1 1
= eax 75 7575 0 = e= {C0 +C 1x+C2x2).
Note. Now consider the equation
Pn (D) y (x) = Pk (x), Pn (0) =j=. 0, (1.16)
where Pk (x) is a polynomial of degree k. Let Pn, s (x) be a sum
of s +
1 terms of the Taylor expansion of the function 11Pn (x)
at x = 0. The equation
00
Pn(x)[Pn,s(x)+ ~ (Pn\xJ:k~O ~~]=1

k=s+1
implies
Pn (x) Pn. 8 (x) = 1 + xS+l ·Pn-l (x), (1.17)
* The statement is immediate since a solution of equation (1.3) may be

expressed in the form of a sum of the solution of this equation and a solution
of the homogeneous equation. On the other hand, any term of the expansion
of 1/ Pn (D) into simple fractions applied to zero satisfies a homogeneous equa-
tion Pn (D) y = 0, since Pn (D) may be expressed in the form Pk (D) (D -
- akl'k, where (D - ak)f3h is a denominator of the term.
2*
where Pn_1 (x) is a polynomial of degree equal to, or less than,

n-1.
We shall show that
y (x) = Pn, 8 (D) Pk (x) (1.18)
is a solution of equation (1.16) for s ~ k. Indeed, equation (1.17)
implies
Pn (D) y (x)= Pn(D) Pn, s (D) Pk (x) =
+
= (1 Pn-1 (D) D"+1) ·Pk (x) = Pk (x) +
+
Pn-1 (D) ·Ds+l.pk (x).
But the last term equals zero for s ~ k. Thus formula (1.18) gives
a solution of equation (1.16) for any s ~ k and any solution of
equation (1.16) is expressed by the formula
y (x) = Pn, 8 (D) Pk (x) + cp (x),
1
where cp E Pn(D) 0.
Sec. 2. Difference Equations

Let ehD be a translation operator defined by the formula
def
ehDI (x) =I (x+ h),
where h is a real number.
The definition is justified by the following heuristic argument:
let I (x) be an analytic function defined in the 2h-neighborhood of
a point x, -oo < x < oo. Then
00
I (x+h) = ~ ~~ f<kl (x).

k=O
On the other hand, a formal expansion of ehD in a power series

provides
00 00
hkDk hk
ehDI (x) = ~ ~ f (x) = ~ kf l<kl (x) = f (x +h).
h=O h=O
Thus, the expansion in the Taylor series of f (x + h) is compared

with the formal expansion of the exponent ehD,
Consider an equation
n
2] aky(x+kh)=f(x), IEC'Q, (2.1)
k=O
INTRODUCTEON TO OPERATIONAL CAl.iCULUS 21
where CC: is a space of infinitely differentiable finite functions on R

(i.e., with a compact support). Write it in the form
Pn (ehD) y (x) = f (x),
where
n
Pn (ehD) = ~ a 11 ekhD.
· k=O
We assume for simplicity that all the roots of the equation Pn (x) =
n
= ~ a 11 x 11 are different. Then an equation is true
k=O
n
1 """ Ai (2.2)
Pn (x) = LJ x-ai '
i=i
where A i are constants.

If y (x) is a solution of an equation
(ehD - a) y (x) = f (x) (2.3)
then y (x) is a solution of an equation
_::_Ina -~Ina
(ehD- a) e h e th y (x) = f (x).
It is easy to verify that
..=..Ina -~Ina
(ehD_a)j(x)=aeh (ehD_f)e h f(x).
Hence, (2.3) may be expressed in the form
X X
--Ina 1 --Ina
(ehD_f)e h y(x)=-e h f(x).
a
00
Consider that the series - ~ ekhD is a formal power series of the

k=O
function (ehD - 1)-1 • It is easily seen that
X 00 X
e h
--Ina
y (x) = -a1 LJ
""" --Ina
ekhDe h f (x) + Yt (x),
k=O
1 -=.Ina 00
- x+kh Ina ..=_Ina
y(x)=--eh ~ f(x+kh)e h +yt(x)eh ,
La
k=O
where y1 (x) is a solution of a homogeneous equation (eDh - 1) y1 =
+
= 0; Y1 (x h) = Y1 (x). Hence y1 (x) is a periodic function of
a period h. It can be proved that a solution of equation (2.1) can
be stated in .the form

n
y (x) = "LJ hD '
A· f (x) (2.4)
i=1 e -ai
along the same lines as in the preceding section.
If y (x) is defined by (2.4) we state by definition
def 1
y (x) = Pn (ehD) f (x).
Thus the problem is reduced to the equation (2.3).
Problem. Find a partial solution of a differential~difference
equation
y" (x) +
y (x 1) = xkeax, +
where a satisfies the condition a = ieaf2.
Sec. 3. Solution of Systems of Differential Equations

by the Heaviside Operational Method
A system of differential equations with constant coefficients may
be written in the form
B (D) .y (x) = F (x), (3.1)
where Y (x) = (y1 (x), ... , Yn (x)), F (x) = (/1 (x), ... , In (x)) are
vector functions, and B (D) are n X n matrices having as elements
the operators considered in Sec. 1.
It is easy to verify that the operators B(D) constitute a noncom~
mutative algebra, which we shall denote by M [D). Let M [x] be
m
a matrix algebra having as elements polynomials P (x) = 2] aixi.
i=O
The isomorphism P (x)-+ P (D) of the algebras K 1 [x] and K 1 [D)
defined in Sec. 1 can be extended onto an isomorphism of algebras
M [x] and M [D).
Let
B EM [x], B (x) = II pii (x) 11.
Let Aii (x) be the minor of the matrix B, corresponding to the ith
line and the jth column and consider the matrix
A EM [x], A (x) = II Aii lit = II An II·
The following formula is a corollary of a familiar theorem of linear
algebra
A (x) B (x) = E./). (x), (3.2)
IN'l'RODUCTLON TO OPERATIONAL CAI.iCULUS 23
where E is a unit matrix, 11 (x) = det B (x). Equation (3.2) for

the corresponding operators implies
A (D) B (D) = E/1 (D). (3.3)
Equation (3.1) can be expressed in the form
A (D) B (D) Y (x) = A (D) F (x)
using equation (3.3). Let A (D) F (x) = G (x), then
E/1 (D) Y (x) = G (x) (3.4)
or
11 (D) Yi (x) = gi (x), i = 1, 2, ... , n. (3.4')
Equations of this type have already been considered. Their solutions
have the form
1
Yi (x) = r:.. (D) g; (x), i = 1, 2, ... , n. (3.5)
We find by the argument stated above, that if Y (x) satisfies the

system of equations (3.1), then its components, Yi (x), satisfy equat-
ion (3.4'). Note, nonetheless, that if Yi (x) is a general solution
of equation (3.4'), the function Y (x) = (y1 (x), ... , Yn .(x)) 1 does
not, in general, satisfy the system of equations (3.1). It is-necessary
to impose conditions on the integration constants contained in the
solution of equation (3.5) for the function Y (x) = (y1 (x),
... , Yn (x)) 1 to be a solution of system (3.1). If
1
zi (x) = r:.. (D) fi (x),
then
Y (x) = A (D) Z (x),
where Z (x) = (z1 (x), •.. , Zn (x)) obviously verifies system (3.1):
B (D) Y (x) = B (D) A (D) Z (x) = E/1 (D) Z (x) = F (x).
It is easy to see that we have obtained the general solution of sys-
tem (3.1).
Example. Solve the system of equations
Dy1 (x) + Y1 (x) + Dy 2 (x) = 0,
D 2Y1 (x) - y1 (x) D 2 y 2 (x) +y 2 (x) = 0. +
Put the system in the form
(D + 1)"y1 (x) + Dy (x) = 0, 2
(D 2 - 1) y (x) + (D + 1) y (x) =
1 2 2 0.
24 OPERATIONAL METmODS
We see that in this case 11. (D) = (D + 1) 2 • Hence, the equation

has the form
Yl (x) = ae-x bxe-x, +
y 2 (x) = ce-x dxe-x. +
We obtain, on substituting these functions into the first equation
and dividing by e-x
b- c- dx +d = 0,
i.e., d = 0, b = c. The substitution of y1 (x), y 2 (x) into the second
equation provides the same conditions for the coefficients. Thus,
y1 (x) = ae-x + bxe-x,
Y 2 (x) = be-x.
It has thus been shown that the operational Heaviside method

reduces the problem of ordinary differential and differential-diffe-
rence equations with constant coefficients to the purely algebraic
problem of linear algebra, the familiar methods of which have been
thoroughly investigated. Even if they were not known, it would
he only natural to reduce the analytical problem to an algebraic
one, familiar or not, and then to try and solve it.
Equations with variable coefficients are far more difficult, since
the corresponding algebraic problem has not been studied. In this
case we understand the operational method not only as a kind of
reduction of a differential problem to an algebraic one, but also
as a method for the solution of the latter.
Sec. 4. Algebra of Convergent Power Series of

Noncommutative Operators
Heaviside's operational calculus enables us to solve linear diffe-
rential equations with constant coefficients. We shall now turn to
equations with variable coefficients
n .
"' ai (x) dJ~= I (x).
Ll dx3
j=O
If D is an operator of differentiation, then the equation can be
written in the form
P (x, D) y (x) =I (x),
where P is a polynomial with respect to the second argument
n
P (x, p) = ~ ai (xHpi.
j=O
INTRODUCT]ON TO OPERATIONAL CAIJCULUS 25-
The main difficulty is that the operators of multiplication and diffe-

rentiation with respect to x do not commute. Consider the following
example. Let P (x, p) be a polynomial in x, p
P (x, p) = xp.
It is not clear how to define P (x, D) since different substitutions
of x and D in P (x, p) produce different results
xDy (x) = xy' (x), Dxy (x) = (xy)' = y (x) + xy' (x).
It is obvious that the definition
def .
P~(x, D)=~ ai (x) D',
where P (x, ~) = ~ ai (x) ~i, makes no sense if we do not assume
that the operator D in P (x, D) acts first. Therefore we define the
operator correctly if we assume an order of action of the operators
x and D.
P {:, D) = ~ ai (x)pi.
Thus a natural generalization of the definition of a polynomial in D
in Heaviside's method affords a definition of polynomials in ordered
operators.
We have introduced rules of addition and multiplication fol"'
polynomials in the operator D in the case of constant coefficients.
They turned out to be the same rules as for ordinary polynomials.
In this section we shall introduce similar rules for polynomials.
in ordered operators. The rules will lead us to new algebraic con-
cepts which we shall axiomatize later. .
We shall consider convergent power series of operators as kind
of "infinite polynomials". From the example of Heaviside's method.
we saw that in some cases a solution can be represented as a conver-
gent power series (see Sec. 2). The student unfamiliar with the con-
cept of bounded operators may consider operators as ordinary mat-
rices.
Besides, in Sees. 1-3 we saw that the concept of polynomials in D
is generally speaking not sufficient to solve differential equations.
Hence in this case as well we shall generalize the obtained rules
for more general functions of ordered operators (see Sec. 6).
Let Op be a noncommutative algebra with a unit. Elements
A, B, C, ... of Op will be called operators. Let .:1/; be an algebra
of the formal power series of variables x, y, z, . . . . The set of
variables being infinite, every function is a power series of finite
number of variables.
Let ,ff be a set of formal power series of elements of the algebra Op.
Such series are defined along the same lines as formal power series
26 OPERATIONAL l\1ETfWDS
of commutative variables but now two monomials, for instance,

A 2B and ABA, are equal by definition if they differ only by the
order of multipliers.
An operation of multiplication is defined in the natural way in
the set r!f. Consider an example of the product of two monomials:
2A 2 BC -3CABA = 6 ·A 2BC 2ABA,
The multiplication in rJf is noncommutative.
Now let the letters a, b, c denote some numbers. For the sake
of simplicity we shall introduce the concepts of operational calculus,
for example, only for three noncommuting operators, the generalizat-
ion for any number of operators being obvious.
We shall introduce some definitions and rules of computation
of formal power series of operators of the algebra Op. The rules will
depend on the order of elements of Op.
Put numbers over the operators, for example,
1 2 3
A, B,C.
We shall say that the operator A acts first, B-second, C-third.
Consider the following mapping of the set .JI: in r!f, which, for every
series
00
f (x, y, z) = ~ aukxiyizk,
i, j, k=O
provides
( 1 2 3 ) def 00 •
'tl:f(x,y,z)-+f ,A,B,C = ~ au~tCkB3 Ai.

i, j, k=O
( 1 2 3)
Let U A, B, C be an image of .Jl by the mapping 'tt· Algebraic
operations in .Jl induce by the mapping 'tt algebraic operations in
1 2 3)
U ( A, B, C which provide U
( 1 2 3)
A, B, C. with a structure of com-
mutative algebra.
1 2
Note. The injection U ( A, B, C c rJf induces in U A, B, C
3) (1 2 3)
an operation of multiplication, which differs from multiplication

( 1 2 3)
in U A, B, C defined above. Multiplication in r!f' is not com-
mutative, as distinct from multiplication in U A, B, C . ( 1 2 3)
Example.
( 1 2) 2 1 12 2
A+B =A2 +2AB+B2 =A 2 +2BA+ B 2 •
INTRODUCTION TO OPERATIONAL CAL·CULUS 27
1 2
Note that there is A + B =A +Band, nonetheless, the element
with (A + B) 2 :
r
(A + .§ of the algebra u (A, .§) does not coincide generally
(A + B) 2 = A 2 + AB + BA + B 2 =I= (A_+_§r.
The elements A, B, C have appeared as formal symbols until
now. Let us now take into consideration that there exist relations
in Op: two different polynomials containing elements of Op may be
equal to each other as operators. Then we shall call them equivalent,
which is denoted by the symbol ,.._,between them. Now, let Op be
Banach, and let .!1:' be the subalgebra of A, consisting of series
convergent everywhere. The set of all 1-l:..operations restricted on A'
generates a sub algebra r!f' of r!f. For each series f belonging to r!f'
there is a unique element denoted by [j] which is its sum.
We shall call two elements of r!t' equivalent if and only if their
sums are equal.
If
( 1 2 3 4) ( 1 3 2 4)
j A, B, C, D ,.._, j A, B, C, D ,
then we shall denote each of the equivalent series by
( 1 2 2 3)
j A, B, C, D .
Example. denotes (A+B) Cor AC+BC AC+BC.

The mapping !l has the following properties:
(1) f (1, B, c) ,. _, f (A.~ Jl, c), if n! < n2 < n3. Here
( n1 n2 na) · ·
f A, B, C means (by definition) that iff (x, y, z) = 2j aiikx'y3z\
i, j, k
( nt n2 ns) def k . .
j A, B, C = 2j C 1 ... 1B11 ... 1A'1 ... 1,
i, j, k
where the units take all the places except the n 1th, n 2th and n 3th
one in the right-hand side of each monomial.
(2) If B = 0, then
j (A,1 C3) B"'"'

2 0,
(3) If A and B commute, then
j (A,1 B,2 C3) (2 1 c3)

,.._, j A, B,
( 1 1 2 )
and thus both sides can be denoted by f A, B, C . This definition
makes sense also if B = A
j (A,1 A,1 C2) = g (A,1 C2) ,

where g (x, z) = f (x, x, z).
Let
00
~
f (x, y, z) = L.J aukx i yj zk .
i, j, k=O
1 3
We may substitute operators A and B for x, z and leave y as a for-
mal variable
( 1 3 ) def ~ k . i
j A, y, B = L.J
i, j, k=O
au~tB y1 A.
Consider a substitution of an operational expression acting second

for y into f ( 1
A, y, B
3)
. Let <p
( 1 2
C, D, ... , E
k) be an element of
1 2 k)
U (C, D, ... , E , where some of the operators C, D, ... , E
may coincide with each other and with A and B as well. Take by
definition
j ( 1 2 ( 1
A, [<p C, D, ... , E ], B
2 k) 3) =
~
L.J aiJnBkH3 A',
..
i, j, k=O
where H is an element of cff' equal to <p Hi is (c, D, ... , E),

the jth power of H in the sense of ordinary noncommutative multi-
plication in cff', i.e., indices over the operators in the expression
in the double brackets [ ] do not define the order of action of
operators outside the brackets (and vice versa). That is why we
shall call the brackets [ ] autonomous. In the same way we
define an operation of the substitution of operational expression
of higher order.
Example. (1) Let ex be a convergent power series
00
x "V xi
e = LJ -.-,
].
.
i=O
Then
2 1 00 00
~- LJ
[ eB] . [eA] = ~ "V j1k1-
i=O k=O
= LJ
oo
"' LJ
"
n
1
kl (n-k)l
Bk An-I; /"OoJ "l
oo
LJ
( 1
A+B
nl
2) n 1
= eA+B.
2
n=Ok=O n=O
I:>ITRODUCTEON TO OPERATIONAL CALCULUS 29
00
2 ~ 2nBnAn A2
(2) Take an expression of t he product T = eB Li -n-1- e
n=O
in powers of homogeneous binomials in A and B using the operational
method. We have
21 3 21(?1)
T = eB2 [e2BA] eA2 = [eB2] [eA 2B+A ] =
4 321 4 32 1 2 21 1
= [eB2] e2BA [eA2] = eB2+2BA+A2 = eB2+2BA+A2 =
1 2 2 ( 1 2) 2n oo n
=e (A+B) ~
= Li A+B ~
= Li _!_ ~ ci BiA2n-i
n! n! Li 2n '
n=O i=O
where C~n is the corresponding binomial coefficient.

1 2
(3) [A+B] 2 =
3 4) =A
1 2) (A+B
(A+B 2 +AB+BA+B2 •
12 3 1 ( 1 2) ( 3 4) 5
(4) A [(A+B)] 2 C= A A+ B A+B C.
Now we shall introduce an operation of extraction of the auto-
nomous brackets (operation "prime"). Let
~ .• k
f (x, y, z) = .i..J aiikx'y1z .
i, j, k
Then we shall consider

1 3 (2' 1 2) 4)
I (A, [cp B, C, D ], E (4.1)
as a convergent power series of !!' obtained by the following pro-
cedure. First consider f [cp 1, (.A, (B C, Dh, E)•
where the ope-
rator B1 is supposed to be non-commuting with any of the operators
A, C, D, E. Denote by S the element cp (.Bu C, D) E !/'. Then
1 2 ( 3 C,1 D2) ], E3) =
I (A, [cp Bt, ~
.
a; 1kEkS Ai.1
i, i. k
Now,in every term of the last sum take the operator B 1 (contained
in S 1) from its place to the right so that it might act immediately
after A;. The obtained convergent power series will be denoted by
f (A,1 [3cp (B2' C,1 D2) ], E4) .

17
We need only substitute B for B 1 everywhere in order to define the

series (4.1).
Example.
2(
[ A+BC 2]= A+BC
1') (2 13)(4A+BC
15) .
We shall consider an example where .the operation It immediately
solves the problem on being introduced. Let us take a set of all
homomorphisms of the vector space C': as Op and put
Xcp (6) = 6cp (6), Pep (6) = - i dcp (6)/d6, cp E C':.
Let A be an operator of Op; f (x, y), S (x, y), g (x, y) be poly-
nomials
n m
f(x, y)=
j,
2J1=0 ailxjy 1, g (x, y) =
j,
2J1=0 ci1xjy 1,
e is (x, Y) def eiS (~, Y) S (x y) EC""
' ' .
Then
[/ (X, P)] [ eiS (1, Y) g (X, 1}] =

=
m
2} ailX' [P 1ei8
4.32 (1 ) (2
x, g X,
II ]
1)
A .
i, l=O
Hence, using the identity
pzeis (1, y) = eis(i, y) [P + :i (X, y)J, (4.2)

we obtain
[/ (k, Ph [eis(i, y) g (k, 1}] =

m 4 . ( 4 ) 3 as ( 2 1)
= .2} anX'eiB x, Y [P+ ax (X, y)f g X, A =
j, l=O
4 )
=eis ( x,y 1 ( X,
4 3 as (X, y)] ) g ( 2 1 )
[P+ ax X, A .
Formula (4.2) in the case of the operators P and X can be obtained

similarly to 1.4. Thus we have proved the following theorem.
Theorem 4 .1. Let X, P be the operators introduced above; then
[t (x, Ph [eis(i. y) g (x, .Ph--

4 ) as
1 ( X,[P+-ax(X,
4 3 ) ( 2
,...._,eis ( x,y y)] g X, p1 ) . (4.3)
INTRODUCTION TO OPERATIONAL CAL'CULUS 31
1'
Problem. Substitute P for y into (4.2)
3
ps eiS ( X,
2 1)
""'eiS ( X,
4 1) 2
p [P+
as ( X, p1 • ) ]".
ax
p (4.4)
Obtain an analogue of formula (4.3) from (4.4) using the rules and
the definitions for the ~t-operation.
The operator f (x, P)
in formula (4.3) is called the Hamiltonian.
Formula (4.3) will be called the formula of commutation of Hamil-
tonian and exponential.
1 :2 n+1 n+2)
Theorem 4.2. If f ( A, C1, ••• , Cn, B .- 0, then
1 n+2) ( 1 2 n+ 1 n+2)
cp ( A, B f A, C1, ••• , Cn, B ""'0
for any cp (x, y).
Proof.
( 1 n+2) ( 1 2 n+ 1 n+2)
cp A, B f A, C11 ••• , Cn, B '""'
{ 1 n+4) ( 2 3 n+2 n+3 \
'""'cp \A, B f A, Ct, ... , Cn, B 1'""'
( 1 3 ) 2 ( 1 2 n+ 1 n+ 2 )
'""'cp A, B [f A, C1 , ••• , Cn, B ] '""'0.
Corollary. The relation
[/ (x, Ph [eisCUg (x, Ah'""' o

is equivalent to the relation
f (x, [P+S' (X)]) g (x, j) '""'0.

The proof follows from Theorem 4.2, the formula of commutation
of Hamiltonian and exponential and an obvious formula
e-iS(1)eis(1) f (x, [P+S' (X)]) g (x, A)'""'

""'f ( x. [P + s' (X)]) g (x. A).
We shall introduce two formulas of importance in the theory of
differential equations.
Difference derivatives, being similar to derivatives of the con-
ventiona] calculus, are very important in the calculus of non-com-
mutating operators. We are therefore introducing corresponding
formulas which will be very useful in the sequel.
Let f (x1 , x 2 , • • • , Xn) be a function of n variables. An operator

{)f difference differentiation IS 1 transforms this function to the follow-
ing function of (n +
1) variables
.ll f { , f (xi, xz, ... , xn)- I (x;', x2, ... , x,)
x 1 , x 1 , x 2 , ••• , Xn = ,
n ) def
u1 H ,
xl-x1
the first argument doubling itself. An operator {) i of difference diffe-
rentiation with respect to the jth argument is defined similarly.
Powers of the iterated operator IS i act with respect to any argument
reproduced by the previous differentiations (the reader may him-
:self see that the result does not depend on the choice of the jth argu-
ment).
Example. If f is a function of one variable x, then
{jf (x'. x") = I (x')- I (x")
' x'-x" '
{j2f { '·
X,X,X
"· "')- I (x')
-(x'-x")(x'-x"')
+
I (x") I (x"')
+ (x"-x') (x"-x"') + (x"'-x') (x"'-x")
if f (x, y) is a function of two variables, then
{) {) f(x'· x" y'· y")- l(x', y')-l(x', y")+l(x", y")-l(x", y')
12 ' ' ' - (x"-x')(y"-y') •
We shall use the notation IS/~Sx, li!ISy instead of IS 17 IS 2 •
Theorem 4.3. The following formula of a change of the arguments'
order is true:
[f (A., .Bh-[t (A, ~h [A, B] {j1{)2j (A;~. B; B), rv

3
where [A, B] =3
[ AB - BA] is a commutator of the operators A
and B.
Proof. We obtain by the properties of the f,!-Operation
3 ( 1 2) ( 1 4) ( 5 4) (5 2)
[AB-BA]· I A, B -1 A, B +I A, B -1 A, B rv
U-~) (~-1)
rv (1~-BA). f (~,~)-I(~, 1) +I U, 1) -f (~, 1)
U-~) (~-1)
"'A t u. 1) -1 (1, 1) -A I(~. 1) -t u. 1)
1 5 1 5
A-A A-A
INTROLDUGT]ON TO OPERATIONAL GADGULUS 33
'"'"' A3f (1A, B2) - f (3A, B2) -A

1 f (1A, B2) - t (3A, B2)
1 3 1 3
A-A A-A
'"'"'~
~~[(12)
I A, B -I (32)]
A, B
(21) (12)
'"'"'[/ A, B ]-[/ A, B ].
A-A
Note. Similar formulas are true for the permutation of an order

of action of two operators in the case of two operational arguments.
For example
1 2 3) (1 2 3) [
f (A, B, C - f B, A, C '"'"' B, A 6/12/
3 J (1A; A,5 B;
z 4 6)
B, C .
Theorem 4.4. The following formula is true
2( 1') 2 ( 1 21')
[/ A + BC ] '"'"' [/ A + B C ] +
+[C1[ A,4B J62/ (A+BC,

2 61 [A+BC],
1' 5[A+BC]
1') ].
Proof. It is sufficient to prove that
f(A+B) '"'"'[/ (A_+i1h+[[A,3 s] 62/ (~+B, A+B, A+sh.

We have a relation f (z) "'""'f (x + y) 1+3(z - x - y) 6! (x
2
+
+ y, z). By substituting the operators A, B, A +B for x, y, z,
respectively, we obtain
1 2) ( 2
f(A+B) '"'"'[/ ( A+B ]+[ A+B-A-B 6! A+B, A+B ].
1 3) (1 3 2 )
Using Theorem 4.3 and the note, we obtain
( 1 2)
f(A+B)'"'"' [/ A+B ]+[ A+B-B 6/ A+B, A+B ] -
( 2 3) (1 3 2 )
1 (1 3 2 )
-[A6f A+B, A+B ] ,...., [/ A+B ]+
(1 2)
( 2 3) (1 2 )
+[ A+B-B 6! A+B, A+B ] -
3
(1 4 2 )
3
-[A6f A+B, A+B ]+
3 ( 2 4 1 5)
+[[A,A+B]6 2f A+B,A+B,A+B ].
We have made use of the fact that for the function
g(x y z v)-x6f(z+v· y)-x f(z+v)-f(y)
' ' ' - ' - z+v-y
3-01225
the difference derivative with respect to the first two arguments

is equal to
6162g (xi; Xz, Y1; Yz, z, v) =
[ f(z+v)-f(Yi)
z+v-yi -
j(z+v)-f(yz)
z+v-yi
J(Xi-X2 )
- (xi-xz) (Y1-Y2) -
62f (Y1• Yz, z v).
= +
Now we have only to note that [A, A+ B]"' [A, B] and that
2 3) (1 3 2 )
[ (A+B-B t'\f A+B, A+B ]-[At'\f A+B, A+B ]=
3 (1 4 2 .)
= (.4+.8, A-t-B) -At'\f (A+B, A-tE)=
(A+B-B) t'\f
=(A+ B-Ji-.8) t'\f (.4 + B, A+B) "'0

by virtue of Theorem 4.2 (or, more exactly, its generalization),
Q.E.D. It will be of use to prove the following corollary.
Corollary. An expansion is true
t(A+B)=t(A+B)+ ~ [A,2 B]f"(A+B)+R~, (4.5)

where R 2 is given by a formula, containing difference derivatives
up to the order 4.
2 (1 4 1 4 1 4 3 4)
R 2 = [A, [A, B]] 63 f A+B, A+B, A+B, A+B +
3 (1 42 1 16 5 6)
+[[A, B], B] 63 f A+B, A+B, A+B, A+B +
+[A, 3B][A, 6B]t'l"f (1A+B,A+B,A+B,A+B,A+B

5 1 8 2 1 7 )
+
3 6 (1 2 1 8 4 8 5 7 )
+[A, B][A, B]t'l"f A+B,A+B,A+B, A+B,A+B +
2 5 (1 4 1 6 1 8 3 8 7 8)
+[A, B] [A, B]64f A+B, A+B, A+B, A+B, A+B.
(4.6)
It is easy to write out the next term of expansion (4.5):
f(A+B)=dA+B) +~[A, 2B]f" (A+ B)+! ([A, 2

[A, B]J+
2
+[[A, BJ, B])f"' A+B
(1 2) +s[A,
1 2B]2j< (1A+B3) +R 41 3,
(4.7)
INT'RODUCTDON TO OPERATIONAL CADCULUS 35
where R 3 is expressed by the commutators of the third order just

as in (4.6).
Theorem 4.5. The following formula is valid
2 ll/ ( 4 1
2 1 ) ( 2 1 )
f ( C, A+B =f C, A +B-,- C, A, A
3 )
uxz
+
2 4 1)2j ( 6 1 3 5 )
+BBT2
ux
C, A, A, A+B .
2
Proof. Consider the equality

cp (xu Xz, Xa)= cp (xi' Xz, Ya) ~q>
+ (xa- Ya) uxa (xi' Xz; Xa, Ya) ·
6 1 5 3
We use a substitution xi-+-Cu x 2 -+-C 2 , x 3 -+-A+B, y3 -+-A. Then
we have, for any T E Op,
2 (6 1 5 ) .
Tcp Cu C2 , A +B =
z (6 1 3) 2 llq> ( 6 1 3 5 ) ( 5 3 )
=Tcp Ci,Cz, A +T-,-
ux Ci,C 2 ;A,A+B A+B-A =
3
(4.8)
Hence, in particular, when cp is independent of the second argu-

ment and the function <p (x1 , y 2 , x 2) is equivalent to f (x1 , x 2) and
C1 "'C, T "'1 we have
2 1 ) ( 2 1 ) 2 ll/ ( 6 1 5 )
f ( C, A+B =f C, A +B-,- C; A, A+B .
UX2
We apply formula (4.8) to the last term in the case

Ill
cp (xu Xz, Xa) =-,-(xu Xz, Xa).
uxz
T,...., B, c1,...., c, c2,...., A.
Then the sought-for expansion follows from (4.8), Q.E.D.
Sec. 5. Spectrum of a Pair of Ordered Operators

We shall first consider the spectrum of matrices. Let A be a mat-
rix. Let K be a set of polynomials P with coefficients in C such that
P (A) = 0, where the zero denotes a zero matrix.
Definition. ·The spectrum cr (A) c C of a matrix A is an inter-·
section of the sets of zeros of the polynomials belonging to K: (z Ecr) ~
<=> (P (z) = 0 for any P E K).
3*
Let kp (A.) be a multiplicity of a zero of polynomial P (z) E K

at a point A. E a (A). Define the multiplicity of the point A. of the
spectrum as inf kp (A.).
PEK
Let A.1 , A. 2 , ••• , 1. 8 be the eigenvalues of a matrix A, and Q (z) =
8
= IJ1 (z - A.i)ki is its characteristic polynomial. Linear algebra

provides the equation
Q (A)= 0.
Hence we see that the spectrum of a matrix coincides with the fami-
liar eigenvalues set, the multiplicity of the eigenvectors ki being
A.i. Take for the sake of simplicity s = 3. For any polynomial P (z)
there exists a formula
k!-1
P(z)= ~ -frp<il(A.i)(z-A.i)i+(z-A.i)ltipdz),
i=O
where P 1 (z) is a polynomial with coefficients dependent on Ap
Similarly
ltz-1
P!(z) = ~ -/r
P\i) (A. 2) (z- A.2 )i + (z- A.2)k 2 P 2 (z),
i=O
113-1
Pz (z) = ~
i=O
~.
+ +
p~l (A.3) (z- A.3)i (z- A.3)k3 P 3 (z),
where P 2 (z), P 3 (z) are polynomials, the coefficients of P 2 being

dependent on A.lt A. 2 • Then, taking into account that
(A- A.t)lti (A-A.z)kz (A -A.3)1ta = 0,
we obtain
i=O
kz-1
+(A-A.t) ~ ~ P\il(A.2 )(A-A.z)i+
111
i=O
ka..:.1
l
+(A-A. 1 1 (A-1. 2) 112 ~ -ftp~>p.,z) (A-A.3)i. (5.1)
Hence P (A) is a polynomial of the order k1 + k 2 + k 3 - 1 and

the matrix P (A) depends only on the values of the polynomial and
a finite set of its derivatives at points of the spectrum. For this
INTRODUCT]ON TO OPERATIONAL CAIJCULUS 37
reason formula (5.1) is called the spectral expansion of the operator

p (A).
Let A, B be elements of the algebra of operators. Consider a set K
of polynomials P (z1 , z2 ) with such coefficients in C that P = (.A, B)
= 0.
We wish to extend the given definition of the spectrum and of
the spectral expansion into the case of ordered sets of operators.
For the sake of simplicity we confine ourselves to the spectrum of
1 2
a pair of operators A, B.
Definition. The spectrum a

1 2
(.A, B) c C2 of an ordered pair of
operators A, B is an intersection of sets of all zeroes of polynomials
belonging to K, i.e., (zH z2 E (J (.A, e))¢:>
(P (zH z2) = 0) for
all P (z1 , z2) E K.
Note. Theorem 4.2 states that if Q (z1, z2) E K then Q (z11 z2) X
X P (z11 z2) E K, where P (zH z2) is a polynomial.
Let A, B, C be elements of the algebra of operators. Consider
a set ,JJ' (zH z2) of polynomials P (z1, z2) with such coefficients
in C that
2 ( 1 3 )
CP A, B =0.
Definition. The spectrum ac

2
(.A, B) c C2 of a pair A, B rela-
tive to C is an intersection of sets of zeroes of polynomials belonging
to r!f (z1 , z2); i.e.,
(Z 1, z2 EO'c
'1 B3)) <=> (P (z 1,
(A, z2) = 0)
for any P (z1, z2) E r!f (z1, Z2).
Note, that if Q (z10 z2) E r!f (z11 z2), then Q (z1, z2) P (zH z2) E
E r!f (z1, z2), where P (z11 z2) is a polynomial.
Consider an important example of the spectral expansion of the
2 ( 1 2 )
function Cf A; B . We shall use formula (4.3) of the commutation
of Hamiltonian and exponential. Take in this formula
X=x, P= - i :x, S(X, P)=S(x), g(X, P)=g(x)
(cf. (4.31)). Thus, in the right-hand side we obtain
e;s(~) f (~, [P+ ~~ ]) g U) = eiB(x>[f ( ~. [P+ ~: ]) ] g (x).

38 OPERATIONAL METHIODS
By virtue of Theorem 4.5 we get

2 1 as )
f ( x,[P+ax-] g(x)=
{4 a~) -f (4x -
2!\x- a1 )
=f (x, ~!) g(x)+P , ax 1 3, ax g(x)+
as as
ax-ax
2 4 fJ2f ( 6 a~ a} 5 as )
+PP 6P2 x;Tx, ax' [P+ax ]g(x).
Calculate the middle term of the equation. Consider to the effect
2 ( 1 3)2
i ~ x-x cp(x)=
= i { ~ x 2 cp(x)- 2x :x xcp (x) + x :x cp (x)} = 0.

2
. (1 s)
a
Hence the spectrum x, x with respect to ih ax is located in R2
2
by the formula x 1 = x 2 • Hence
i
2 (1 3)n =0
~ x-x forn::;>2
by Theorem 4.2.
We have the following "spectral expansion" as a corollary:
2 (13) (33) 2 2
i~P x,x "'iP x,x :x+i:xp~ (33)(1 3)
x,x X - X " '
"'iP(x, x) :x-i[P~(x, x)x :x-P~(x,x) :x x]"'
"'iP (x, x) :x + iP~ (x, x).

Hence
. P2 f (~, a~;ax)
1
-! (;,
3
a1;ax) ( )
Cj)X"'
as;ax-aSjiix
~- i [ :; lp=astax :x + ; :~ IP=apas ~:~ Jcp (x).
An operator P such that
p
Ql- iip
_!}_I
~ +_!_ a2S a2j
as ax 2 ax2 ap2 as
cp I
P=ax P-ax
INTROIDUCTllON TO OPERATIONAL CAIJCULUS 39
is called a transfer operator (for the Hamiltonian f (x, p) and the

given S (x)).
Note that we do not use only the concept of spectrum in spectral
expansions, the number of terms in the Taylor expansion being
important as well. This number is connected with a multiplicity
of spectrum in the case of matrices. In the example considered above
it is natural to define the multiplicity of spectrum located on the
bisector of the first coordinate angle as equal to two. The problem
of the multiplicity of spectrum (i.e., the problem of the number of
terms in the Taylor expansion) is very hard in general. This problem
is reduced to the study of a "subsidiary" Banach space Bmld in
Chapters I and II.
We shall point out some analogies between the given example
and the spectral expansion of matrices to clarify the direction taken
in investigating the properties of operators in Chapters I and II;
these are the following:
1
1. The operator x acts in the space of differentiable functions
because the second operator is i :x.
Since the following inequality
is valid
I :x (eixtg (x)) I+ Ieixtg (x) 1~(1 +It I) max (I g' I+ Igj)
the operator of multiplication by eixt increases as the first power
of t in the space of differentiable functions.
2. A matrix A, with a maximal length of adjoint elements equal
to 1, satisfies the condition
I ew g I ~ (1 + It I) I g 1.
where g is a vector, I g I is its modulus.
3. The Taylor formula in both cases is reduced to two terms of
the spectral expansion. In Chapter II we shall see that the number
of terms in spectral expansion is closely connected with the estimates
of growth of the operator eiAt.
Note. In the sequel it is important to bear in mind the following

properties of matrices:
(1) any matrix A may be put in the form
(5.2)
where A 1 , A 2 commute, A 2 and A 1 have a real spectrum.
(2) We can define the matrix spectrum in the following way;
let K be a set of polynomials P (x, y), such that P (A 1 , A 2 ) = 0,
the spectrum of the matrix <J (A) is an intersection of sets of zeroes
of polynomials belonging to K, i.e., x, y E a (A) ¢:;:> P (x, y) = 0

for any P (x, y) E K, the eigenvalues of A in this case are equal
to x +iy, where x, y E a (A).
Thus we may consider only real roots of polynomials of two vari-
ables. Therefore, though formula (5.2) is itself a kind of spectral
expansion, it is worthwhile remembering when studying a number
of non-commuting operators, that each of the operators may be put
in the form of a function of two operators with a real spectrum.
We shall return to the problem in Chapter I.
Sec. 6. Algebras with 1-t -Structures

The introduced calculus is still insufficient to solve differential
equations with variable coefficients. We shall introduce the necessary
algebraic constructions to the effect.
We shall first consider the most simple case and then turn to more
complicated cases (cf. axioms (~-t 4), (~-t 6 ) below). We shall prove
a number of theorems and, on their basis, demonstrate in Sec. 7
how they can be used.
Next in Sec. 8 we shall demonstrate the power of the operational
calculus by the classical example of the deduction of the wave
equation. We shall see that the ·operational calculus provides an
adequate means for studying mathematical and physical effects
proper to the transition from the system of equations of oscillations
of a lattice to the wave equation.
The operational calculus of convergent series of operators is
introduced in the axiomatic way, i.e., the main points are formulated
as axioms, then formulas are derived and the axioms are verified
to be valid for the basic operators necessary for the solution of
differential equations. The method is convenient as well from the
didactic point of view, i.e., it makes easy to become familiar with
the ordered operators calculus techniques. We shall work within
the framework of the basic algebraic structures.
Let .}{; be an algebra with a unit over R, .}{; is generally noncom-
mutative. The elements of .}{; will be called operators.
Let cffoo be a set of infinite differentiable functions I (x), x E Rk
(k is not fixed) growing togethel' with all their derivatives as I x 11
or less at infinity (l is defined for every function I separately). Funct-
ions belonging to cffoo are called symbols. A symbol is called of
1
rank k if the corresponding function depends on k variables. An

algebra .Jl; is provided with a I-t-structure if for any finite set A 1 ,
A 2 , • • • , A k of operators belonging to a set M c: .Jt and any set
of numbers n 1 , n 2 , • • • , nk (so that n 1 =!= ni if A 1 , Ai do not com-
mute), the following operation is defined
11: (x1 -+AI> ... , xk-+ Ak),
INTROcr:JUGT]ON TO OPERATIONAL GAIJGULUS 41
which substitutes an operator A E A written in the form

( ni nk )
A=[! Ail ..• , Ak ]
for the symbol f (x1 , • • • , xk)· We shall cancel out the brackets
[ ], where it will not lead to ambiguity. The operation 11 satisfies
the following axioms:
(~.t 1 ) The homogeneity axiom: if a E R, then
( ni nk ) (ni nk)
[af Ail ... , Ak ] =a[/ A1, ... , Ak, ];
in particular, if f (x1 , ••• , xk) = 0, then
( ni nk )
[/ Ail ... , Ak ]=0.
(~.t 2 ) The shifting of indices axiom. Let n 1 , n 2 , • • • , nk and
m1 , m 2, ..• , mk be such two sets of indices that, if i =t= j and
(ni <n1) ~ (mi <m1), then
( ni nk ) {mi mk)
[/ Ail ... , Ak .] = [/ ,A{, ... , Ak ],
and if ni = nh Ai = A 1 =A, then
1 (n ni ni nk ) 1
[/ Ail ... , Ail ... , Ah ... , Ak ] = [g A1 ,
(n ••• ,
ni nk )
Ail ... , Ak ],
where
g(x 11 x 2 , ••• ,Xj- 1 ,xi+il ... ,xk)=[f(xbx2 , ••• ,xk)]x.=x.·
} I
Example. If f (x, y) = x 2 y, g (x, y) = xy2 , then
. t (A, s) =g (B, A).

Example. Let A, BE M, then
1 2 3 4 1 2
4 sin A cos A sin B cos B = sin 2A sin 2B.
(~.t 3 ) The correspondence axiom: the 11-operation transfers the unit
1
symbol into the unit operator and [A] =A.
(~.t 4) The sum axiom: if ni =I= m1 for any i, j, then
( ni nk) mz)
[/ A 1, ... , Ak +g (mi
B1, ... , Bz ]=
ni nk) (mi mz)
=[/ ( A 1, ... , Ak ]+[g B1, ... , Bz ], Bi EM.
Problem 6.1. If f (x 1, •.• , Xn) = g (xu(1)• •.. , Xa (n)) when a is the
1 n ) ( u(i)
permutation of the set {1, ... , n} then f ( A 1, ••• , An = g Au(i)• •••
42 OPERAT.IONAL METHODS
<J(n) ) [ 1 n
... , Aa(n) hint: apply the operation 11: (xt-+ At, ... , Xn-+ An>
a(i) a(n) )
Yt-+ Aa(i)• Yn-+ Aa(n) to the symbol
••• ,
F(xp · · ., Xn, Yt• • · ., Yn)=l(xt, · · ., Xn)-g(yt, • · ., Yn)J.
Problem 6.2. If I (x 1 , ••• , Xn+t) = g (xt, ... , Xn) then I (1t,

n n+i ) (1 n ) [
... , An, An+1 = g At, ... , An hint: apply the operation
( 1 n+i n 1 n }
11: Xt-+ At, · • •, Xn -+An, Xn+t-+ An+t• Yt-+ At, · · ·' Yn -+An to
the symbol F(xt, .. . , Xn+t• y 1 , ••• , Yn)=l(xt, .. . , Xn+t)-g(yt, ....
.. . , Yn)J.
(~-t 5 )The product axiom: if mi < n1 for any i, j, then
ni nk) (mi mz)
[I ( Ail ... , Ak g Bt. ... , B1 ]=
nt nk ) (mt mz)
[I ( A 1 , ••• , Ak ] [g Btl ... , B1 ].
Example. Under the conditions of the preceding example
1 2
sin 2A sin 2B = [sin 2B] [sin 2A].
(!16 ) The zero axiom: if
ni nk)
[I ( At. ... , Ak ]=0
and p 1 , • • . , pz, r 1 , • • • , rm are such numbers that Pi <nb
ri > n 1 for all i, j, then for any symbol g (x1 , • • • , Xz+m) and ope-
rators B 1 , . • • , Bz, C1 , • • • , Cm, Ci EM,
ni
[I ( At> ... , Ak g B 11
nk) (Pt ••• ,
Pz
Bll C1,
ri
••• ,
rm )
Cm ]=0,
It follows from the stated axioms that the sum axiom is in fact
valid for any ni, m 1• Indeed, the following theorem is true.
Theorem 6.1. (The first sum theorem.) For any ni, m 1 there exists
an equation
Proof. Let n 8 ~ m 8 ~ ns+l ~ ms+t for any s. Note, that if

n8 = m 8 , then B 8 , As commute by the definition of the 11-structure.
INT'RODUCT]ON TO OPERATIONAL CAIJCULUS 43
(6.1)
This sum is equal to the sum (by the shifting of indices axiom)
( n1 mz )
[/ AI> ... , Ak
nit )
+ g (mi
Bb ... , Bz ], (6.2)
where ni, mj are such indices that
mi+t > ni+t > mi > n;.
Besides, there is an equality
nt (nJ.
f ( At, ... , Ak} =! At, ... , Ak ,
nk \ nit )
(6.3)
(mi
g B1o ... , B 1
mz) (m1
= g Bt.
mz )
... , Bz
for any term of the sum as well. Then the sum (6.2) is equal to
n1 (mi mz )
[/ ( A1o ... , Ak ]+[g Bh ... , B 1
nit )
] (6.4)
by the sum axiom since n; =I= mj for any i, j. The theorem follows
from (6.2) and (6.3).
Theorem 6.2. (The second sum theorem.)
ni def n;
where A;+B;= [A; +B;].
Proof. We may assume I n 1 - n; I> 2 for j =I= i without loss
of generality by virtue of the axiom {f1 2). Hence
ni ) (ni ni-i ni+i nk )
[ ( A; +B; f At. ... , A;-t. Ai+t. ... , Ak ] -
n. ( ni ni-1 ni+i nk )
-[Ad Ab ... , A;-t. A;+t. ... , Ak ] -
n. (ni ni-1 ni+t nk )
-[Ed Ab ... , A;-t. Ai+b Ak ]=
ni n;+i n;+2) (n 1 n;_ 1 n;+ 1 nk )
=[A;+B;- A; - B; f At. ... , A;-t. A;+t. ... , Ak ].
We obtain the following formula with the help of the axiom of

sum, the axiom of shifting of indices and the axiom of correspon-
dence:
n; ) n;+1 n;+2 ni n;+1 n;+2
(
[ A;+B; -A; - B; ]=[A;+B;]-[ A;]-[ B; ]=
1 1 1
=[A; +B;]-[A;]-[B;]= A; +B;-A;-B; =0.
Then we apply the zero axiom and the theorem is proved.
Theorem 6.3. (The theorem of product.) Let n1 , n 2 , • • • , nk,
p 1 , • • • , p 1, r 1 , • • . , rm be integers such that p; <nb r;>ni for
n1 nk )
all i, j and [f ( Au ... , Ak ] =F. Then for any symbol g (x1 ,
• • • , Xz, Yr. ... , Ym) there is a relation
(
n1 nk ) ( P1 Pl r1 rm )
[f A11 ... , Ak g B11 •• , , Bz, C1o ••• , Cm ] =
n1 (p 1 Pl r1 rm )
=[Fg Bb .. . , Bz, Cb ... , Cm ].
Proof. We may assume without loss of generality by axiom (~-t 2 },
that there exists such a number n that n =F n 1 , • • • , nk and Pt <
<n <ri for all i, j. Then we have by axiom (~-t 1 } and Theorem 6.1:
( ni nk n)
=[ f ( Ab ... , Ak)-F g Bb ... , Bz, Cb ... , Cm ].
(Pi Pz ri rm)
It is sufficient to apply axiom (~-t 6 ) to complete the proof.
Theorem 6.4. (The formula of the change of indices.) If n; =
= ni - 1 and I n 1 - n; I> 2 when l =F i, j, there exists an equality
INTRODUCTI:ON TO OPERATIONAL CAIJCULUS 45
}>roof. Let k = 2, A 1 =A, A 2 =B. It is necessary to prove

the equality
[t (A, .B h- [t Ct .B h=
=[(A B) f U. ~) -! U. M+! (1 ~) -! u. ~)
3 ].
' U-~) (1-1)
We shall put the right-hand side into this form with the help of
axioms (~-t 1 ), (~-t 2 ), (~-t 4) and Theorem 6.3
[(AB-sA) 1 U. 1) -t U. 1) +t U. 1) -t U. 1) ]=
U-~) (~-1)
=[A
3 I U. 1) -/ U. 1) +I U. 1) -/ (11) ]=
1 5
A-A
3 (12) - / (52)
=[A f A,B A,B ]+[A I
3 U. ~)-/ U. 1) ]=
1 5 1 5
A-A A-A
5 (12) - / (52)
=[At A,B A,B ]+[At A,B
1 (52) - / (12)
A,B ]=
1 5 1 5
A-A A-A
=d~-;t)' U. 1~ -~ UJ) ]=[-t (A, h) +t (~. hh=

A-A
=[t (l sh-[t L~. hh. Q.E.D.

The proof in the general case is the same.
We shall indicate a formula for a composite function.
Theorem 6.5. (K-formula.) Let f, g be symbols of rank 1 and 2,
respectively, A, B operators belonging to M such that [A, B] EM,
[g (A, shE M. Then
1 ([g (A, .Bh) =[/ (g (A . .B) h+[[A~ B] 6~ (A. •.4. s) x

ilg ( 3 4 6 ) 62f
X-,- A, B, B 72
uxz ux
x
(6.6)
46 OPERATr0NAL METHODS
The note made to Theorem 6.4 is also true of this theorem. Indeed~
the following equality is true
(.1 2 k k+1 (1 2) k+2 k+3 s+1\
[/ Cb C 2 , ••• , Ck, [ g A, B ], Ck+t• Ck+z, ... , Cs I]=
1 2 k (k+i k+2) k+3 s+2)
=[/ ( Ct. C2 , ••• , Ck, g A, B , Ck+~> ... , Cs ]+
k+5
+[ [ A, B -6 -
J 6g (k+3 k+7 k+9) bg (k+3 k+t. k+6)
A, A, B -6 - A, B, B -
62f
2 -
X
xt x2 6xk+i
( 1 2 k (k+1 k+9) k+2 ( 1 2)

X ,Ct. Cz, ... , Ck, g A, B , [ g A, B ],
k+B (
[ g A, B ],
1 2) k+10 s+9)
Ck+t. ... , Cs ]. (6.7)
We shall call formula (6.7), as well as a more particular formula
(6.6), K-formula.
Proof of Theorem 6.5. There exists an equality
. 6!
f(z)=f(g(x 1 , x2 ))+(z-g(xt. x 2)) ilx (g(xt. x2 ); z).
(
Applying the operation f.t: z-+ g, x 1 -+ A, x 2 -+ B
2 1 3) to both sides
of the equality, where g = [g
of sum we get
Lt Bh, and using the first theorem
[t(g)]=[t(g()t,.B)h+[(~-g(.4,.8)) ~~ (g()t,.B), ~h.
We further obtain
[t (g)] = [t ( g ( .4, 8) h+ [ (~- g (A, 1) ) g~ (g (A, 8) , ; h=

[t (g (~. 1) h+ [; ~~ (g(~ •.B), ~) -
3 5 ) 6! ( ( 1 5 ) 2)
-:-g ( A, B & g A, B , g ]+
" ) 6g ( 3 5 7 ) 62/ ( ( 1 7 ) 2 6)
+[ ( A, g 6x 1 A, A, B 6x2 g A, B , g, g,]
2 3
with the help of axiom (~-t 2 ) and changing the indices of A and g
by Theorem 6.4.
According to axiom (f.t 2) and Theorem 6.3
3 4 ) 6! ( ( 1 5 )
" 6! ( ( 1 5 ) ,g2) ]=[g ( A,B
[gTx g A,B Tx g A,B ,g2) ].
INTRODUCT]ON TO OPERATIONAL CAIJCULUS 47
By the latter formula and Theorems 6.3 and 6.4 we obtain
The proof is completed by axiom (f1 1) since g (xu x 2) - g (xu x 2 ) =

= 0.
Theorem 6.6. (The Newton e'Xpansion.) Let A, (A +B), B EM,

f be a symbol of rank 1. Then
m-1 ·
2 2k fN ( 1
. 2k+ 1 )
[/(A+B)]=[f(A)]+ ~[B ... B - A, ... , A ]+
6xk
k=1
2 2m fjmf (1 3 2m-1 2m+1)

+[B ... B oxm A, A, ... , A, A +B ]. (6.8)
The note cited in the cases of two preceding theorems is valid

here as well; its statement is left to the reader.
Proof. First obtain the following formula:
2 2r (1 2r-12r+1)
[T 1 ••• T,cp A 1 , ••• , A,, A+B ]=
2 2r ( 1 2r-1 2r+ 1 )
=[T 1 ••• T,cp A1 , ••• , A,, A ]+
2 2r 2r+2 6cp ( 1 2r-1 2r+1 2r+3 )
+[T ... T, B -.,- At. ... , A,, A, A+B ], . (6.9)
uXr+f
where B, A 1 , . • . , An A, A + B, Tu ... , Tr EM, cp is a sym-

bol of rank r + 1. We cancel out T11 • • • , T r; A 11 • • • , A r for
the sake of simplicitry, i.e., we shall prove
[f(A+B)]-[f(A)]=[B g~ (A_, A.f-sh. (6.10)
The proof of general formula (6.9) is different from the proof

of formula (6.10) only by more cumbersome calculations.
There exists an equality
Apply the operation f.t: ( x 2 -+ A -f- B, x -+A); by axiom (~-t 1 ) we get

1
[/ (A-.f-B) -I(.~)- (A+B-~U ~: (A-.f-B, 1h=o.

Hence, by axioms (~-t 1 ), (~-t 2 ) and Theorem 6.2, we get
[/(A +B)]-[/ (A)]=
=dA.f-B) ~~ (A.f-B, 1)-A ~~ (A.f-B, 1h=

{jf ( 2 {jf (
=[ ( A+B Tx A+B, A1 ) ]-[ATx A+B, A1 ) ].
2 ) 3 3
Applying Theorem 6.2 once more, we get the equation
[f(A+B)]-[f(A)] = [B ~: (A-t-B, Ah.

Formula (6.9) is proved. Now applying successively formula (6.9)
to the last term of the right-hand side of (6.8), we obtain the proof
by induction. The theorem is proved.
We shall define the spectrum of pairs of operators of an algebra
with the ~-t-structure along the same lines as we defined it in the
case of convergent power series of ordered operators.
Definition. Let A, BE M. Let K be such a set of symbols f (x11 x 2 )
that
(f (x 1 , x 2) EK) ~([/(.A, B}] =0), 1
i.e., the symbol transformed into zero by the operation
2 1 2
~-t: (x1 -+ A,
x 2 -+ B). The spectrum a of a pair A, B is an intersection of sets of
zeroes of functions belonging to K:
((xu· x 2 ) E a) ~ (f (x11 x 2 ) = 0 for any f E K).
Theorem 6.7. The set K is an ideal of an algebra of symbols of
rank 2.
Proof. Let f (xu x 2 ) E K. Prove that g (xu· x 2 ) f (xlt X 2 ) E K for
any g E r{foo (Rn), i.e., it must be proved that if [/ (1, .Bh = 0,
then [g (J., B) f (.A, .Bh = 0. There exists an equality
[g (1, s) t (A, sh =[g (1, B) t (~ . .Bh (6.11)

by the axiom of shifting of indices. We see that
[/ (~ . .B h=[/ (1, sh =0
INTRODUCTION TO OPERATIONAL CADCULUS 49
by the same axiom. Thus the right-hand side of (6.11) is zero by

the zero axiom, Q.E.D.
We derive a number of corollaries from the last theorem.
We shall consider the two most important corollaries. First recon-
sider the following version of axiom (f.1 4).
(f.L~). Let {fi (x 1, ... , xk)} be such a sequence of symbols that
00
there is only a finite number of non-zero terms of the series LJ fi (xu ...
i=1
... , xk) at any fixed point (x~, ... , x~) and only a finite number
of operators [/i ('A 1, .•. , ;r~h is not equal to zero; then
~ ) ( nl nk ) ~ ( nl nk ')
[ ( LJ li Aj, .. . , Ak ]= LJ [/j AI, ... , Ak ].
j=l j=1
Theorem 6.8. If axiom (f.L~) is true, then the spectrum of any pair
of operators belonging to M is not vacant.
Proof. Assume the contrary: let a = 0. Then for any fixed point
(x~, x~) E R 2 there exists such a function f (x1 , x 2) E K that
f (xu x 2 ) > 0 in neighborhood U of the point (x~, x~). Any com-
pact set in R 2 can be covered by a finite number of such neighbor-
hoods, hence there exists a function belonging to K strictly positive
at points of the set. Cover the plane R 2 by squares qii
qii = {(xu X 2) E R 2 ; -e i ~ x1 ~ i 1 + e, j - e ~ + +
~ x2 ~ j 1 e}, + +
i, j = 0, ±1, +2, ... , e > 0 is a fixed number.
For every square qii consider a function Iii (x11 x 2 ) E K, strictly
positive in qii. Let qJ (x), x E R be an infinitely differentiable non-
negative function such that qJ (x) = 1 when 0 ~ x ~ 1 and qJ (x) =
== 0 when x < - ~ and when x > 1 + ~ . Then a function
00
f(x 1 , x 2)= LJ fii(xp x 2)qJ(x 1-i)qJ(x2-j)

i, i=-00
is everywhere positive and fu (x1 ~ x 2 )/f (x11 x 2) E K by virtue of
Theorem 6.7. Since LJ Iii (x1 , x 2 )/f (xu x 2 ) = 1, we have by axioms
(t-ta), (f.L~)
0=[~ (1
1
2 )/u (1A,B]=[1
2) (1A,BJ]=1,
2)
j A,B
where 1 (x11 x 2 ) is a function equal to 1, and 1 in the right-hand
side is a unit operator. This is a contradiction but we thus obtain
the statement of the theorem.
4-01225
50 OPERATIONAL METH·ODS
Definition. A complement of the spectrum a is a resolvent set of

1 2 ( 1 2)
a pair A, B; it is denoted by p A, B .
We shall now indicate a criterium which will enable us to deter-
mine the spectrum using only the functions of operators of one argu-
ment.
Theorem 6.9. A point (A., !l) belongs to p (.A,

s} if and only if
there exist positive numbers e and 6 and finite junctions p~ (x1 ), p~ (x 2)
positive when xi = A., x 2 = 11 and equal to zero when I x 1 -A. I> e,
I x 2 - 11 I> 6 (respectively) such that p~ (B) p~ (A) = 0.
Proof. The test of sufficiency. We have
p~(B)p~(A)=[p~(B)p~ (A_h
hy axiom (!1 5). Hence p~ (x 2 ) p~ (x1 ) E K and the function does not
vanish at the point (A., !l)· Therefore, (A., !l) ~a by the definition
of the spectrum.
The test of necessity. Let (A., !l) E p (.A, B);
then there exists
a.function qJ E K positive at the point (A., !l) and, therefore, positive
in its. (2e, 26)-neighborhood. Let P~<., p 11 be functions from r!foo (R)
equal to zero outside an e-neighborhood of the point A. and a IS-neigh-
borhood of the point ll• re13pectively.
Since · qJ (A, B) = 0, we have
[fjl (A' B) PJJ. (B) PJ. (A) ] = PJJ. (B) [ fjl (A' B)] FA (A) = 0.
Let 1Jl (x11 x 2 ) E .!foo (R2 ) be a lion-negative function equal to zero
in an (e, 8)-neighborhood of the point (A., !1) and equal to 1 -
- qJ (xu x 2 ) outside a (2e, 28)-neighborhood of the point (A., !l)·
It is obvious that
[1Jl (xu X2) + qJ (xu x2)l P11 (x2) P~<. (xi) = qJ (xu X2) P11 (xl) P~<. (x2)
and
[1Jl (xu X2) + fjl (xu x 2)]-l E goo (R 2).
Thus
[ ["' (A' B) + cp (A' B) JP!l ( B) PJ. (A) ] < 0

and by axioms (!1 2 ), (!1 6)
[[¢(A.. 1) +fjl (A., s) r cfjl (A., s) +

. I
1
.+ljJ (A' B) JPJJ. (B) PJ. ( A)] = 0.

INTRODUCTIJON TO OPERATIONAL CA!JCULUS 51
Therefore
[p~ ( .B) p~ ( Jt h= o
and, finally, by axiom (~-L 5 ), p 11 (B) p~ (A) = 0, Q.E.D.
Define the spectrum a (A) and a resolvent set p (A) of one ope-
rator A in the same way. It is obvious that there are similar theorems
in the case of the spectrum of one operator. Indeed, the following
theorem is a corollary of Theorem 6.9.
Theorem 6.10. There exists an inclusion
(J ( A' B) c (J (A) X (J (B) .

More comprehensive results are obtained in Chapters I and II;
we need the concept of normed algebra to the effect.
Note. We can extend the concepts of the ~-L-algebra and the algebra
of symbols into the domain of topological algebras modifying axiom
(f-1 6 ) conveniently with regard to the emerging topological struc-
tures (cf. Ch. II).
Example of a !J.-structure. Let C~(Rn) be a vector space of infi-
nitely differentiable functions f (xlt ... , Xn) (defined on Rn) equal
to zero outside a ball (defined for every function separately) in Rn.
Let D j and Xj be operators acting in c~ (Rn) by the formulas
. af(x)
Dif (x) =-l---ax:-, ~
xif (x) = xif (x).
}
We shall denote Di by -i 8~. and (;;i by xi; the last notation will
} ~
be used when it is impossible to mistake xi for the independent
variable xi.
Let P (xlt ... , xn; PH .•. , Pn) be a polynomial. Consider
2 2 1 1)
a differential operator P ( x 11 • • • , xn; Dlt ... , Dn . The operator
is written easily by the Fourier transform, which is defined by the
formula
n
f(p) = (2n) - 2 ~ e-ip. xf (x) dx, x = (xh ... , Xn), P= (Pi> . ·., Pn),
Rn
where f is a function, f (x17 • • • , xn)·

There exists an inverse transformation
n
f (x) = (2n) - 2 ~ eip•xj(p) dp.
Rn
;-----' ~
It is easy to prove that Di/ (x) =Pi/ (p). Hence
( 2 2 1 1 ) det ( 2 1 )
P Xt. . .. , Xn, Dt. ... , Dn f(x)=P X, D f(x)=
= (2ntn Jeip·xp (x, p) dp Je-iP'Sf (~) d~. (6.15)

Rn Rn
The right-hand side of (6.15) is also valid when P are continuous

functions (i.e., not necessarily polynomials) of sufficiently slow
growth as I p 1-+ oo, for example as I p I in some power. We can
state the following definition.
Definition. A pseudodifferential operator with a symbol H (x, p)
is an operator H {:, D) defined by the formula
{ 2 1 )'
H\X, D f(x)=(2ntn
det
Jr Jr e'P·(x-6>H(x,
0
p)f(~)d~. (6.16)
R2n
. ( 1 2 )
Note. The same formula defines H x, D for the distributions
(cf. Ch. II).
In the same way we define an operator H (~, D)
( 1 2) det r ('
H x, D f (x) = (2ntn J J eiP·(x-'SJH (~, p) f (£) d~. (6.17)
R2n
The structure of commutative algebra up. (:, D) is now defined

in the set of operators of the form H U, D). Thus
H U, b) L ( ~. D) f (x) =(2ntn JJeip·(x-6>H (x, p) L (x,p) f(~)d~.
R2n
The operators p = ih :x,

x and the noncommutative algebra
generated by them are defined for f (x, h) E C': (Rn) X of""(R)
in the same way.
Now take M consisting of the set of operators having the form
A k =;h_!_+ as~~. B k = X;
• OX ox '
k = 1, 2, ... , S k = S k (x) is a- set of functions belonging to
C': (Rn).
INTRODUCTI!ON '1.10 OPERATIONAL CAIJCULUS 53
Note. The vector-operator A k

. 8
= ~h OX + ax
aSk (x)
has components
Af.= ih.:-
uXj
+ ~Sk
uXj
These components commute with each other,
•
hence the vector-operator Ak may be indexed only by one number.
Denne the f,t-operation for (x) E ,Jtoo (Rn). Let
Consider the case of two variables for the sake of simplicity. Take
by definition, for n 1 < mu cp E C'(;' (Rn),
ni mi) ( 1 2 ) def _..!:._ S1(x)
<D ( Ab B 1 cp(x) =<D A1, B 1 cp (x)=e h X
X([)
(p,1 x2) e.!:._
h
S!(X)
cp (x).
Now verify axioms (f,t1)-(f.t6).

All axioms except the second item of the shifting of indices axiom
are obviously true by definition. The second item of the shifting
of indices axiom is a corollary of the equation F;_.xF x-+p = 1. Con-
cerning (f.t 6) see Ch. II.
We have obtained the formula of commutation of Hamiltonian
with exponential now by definition. The operation "prime" is de-
fined similarly.
We shall indicate another generalization of the f,t-Structures as
well; let symbols be functions defined on a m-dimensional torus
Mm, i.e., a product of m circles. Consider infinitely differentiable
functions on the torus valued in the space ,Jtoo (Rn). We shall denote
the functions by f (xu ... , xk, a), where a E Mm are angular
coordinates on the torus and x1 , • • • , xk are real numbers, k depends
on f.
Consider the algebra .Jl: and its subset M as we did above. Con-
sider, as well, a subset M m consisting of m commuting elements
of the set M; B = (B 1 , . . . , Bm) c: M m· The operation
is defined for any finite number of operators A 1 , . • . , Ak EM

and any number of indices n 11 n 2 , • • • , nk+ 11 satisfying the con-
dition ni =I= ni if Ai and Ai do not commute and ni =1= nk+ 1 if Ai
and B 8 E M m do not commute. The operation substitutes an ope
rator A E .:1/; for a symbol f (x11 • • • , Xn, a) denoted by
and satisfies all the preceding axioms (as if a E Rm) except the
correspondence axiom, which is true only with respect to the vari-
able x and is supplemented by the following condition:
(fA.;) The condition
[t ( It )] = [t* ( iJ )]
is true for any real infinitely differentiable function f (a), a E Mm
and the corresponding periodic function f * (x), x E Rm.
Problem. Consider an algebra M generated by elements x and p
such that [x, p] = ih. Take a function defined on a circle of radius 1
valued in rfFoo as a symbol I (xu a).
Define a f-l.-structure with axiom (fA.;) using as the set M 1 the ope-
rator x, such that the elements of .Jl; might be operators acting on
the functions on R and the following equation might be fulfilled
f (p,1 x2) cp (a)= 2~ =oo

~ 1(nh, a) eina
~
J e-inycp (y) dy.
n=-oo
Symbols of the class :J>oo are primarily important for the definition
of a f-l.-structure. We shall see later that the class ,ffoo is too narrow
for our aims. An extension of the class in the axiomatics of f-l.-struc-
tures can lead to a situation when an algebra containing i and :x x
has no f-l.-structures.
Consider an example to the effect.
Let a fixed class of symbols contain functions of the form
f (xu x 2 ) = e±ix1x2 and let A, B be such operators belonging to M
that [A, B] = - i {for example, A = - i a~, B = Apply the x) .
formula of commutation (6.5) to the left-hand side of the equation
(3 4) 2
e-i A-A B= 1
3
and substitute the operator A into the first place. Thus we get
t 4) 2 ( 1 3 2 4 )
1 =[e-i ( A-A B]-['1' A, A; B; A ],
1 324)
where a symbol of the operator \jJ ( A, A; B; A has the form
f) 13 "( )
\jJ (Pt, P2; x; Y) = ia; l3p e-t P-Yx=
( ) - i(p C y)x ( ) - i(p 2 - y)x

~p~~~-~y~e ________~P~2-~y~e______ _
Pt-Pz
. -i(p 1-y)x -i(p 2 -y)x
=e-t(p 1 -y)x+( _ ) e -e
p2 y Pt-P2
INTRODUCTION TO OPERATIONAL CAliCULUS 55
Hence by the axioms of zero and correspondence, we get
( 1 4) 2 (1 4) 2 ( 3 4) ( 1 4) 2
-i A-A B 1
]=0.
i=[e-iA-A B]-[e-iA-A B]-[ A-A e 1 3 -
A-A
We see that for the given A, B EM and the class of symbols con-
taining e±ix1xz there does not exist any 11-structure.
Therefore it is necessary to be very careful defining 11-structures
for symbols of the class wider than r!/"'\ since this leads to wrong
results even for tlie operators with symbols of the class rff"" (cf.
the details in Ch. II).
Now we shall defi.ne a generalization of 11-structures using an
extended class of symbols.
Consider a two-sided module :£ over an algebra .Jl, i.e., :£ is such
a linear space containing .Jl, that for any pair a E .If, l E :£ a pro-
duct al E :£ and la E :£, linear in l, is defi.ned.
Let X be a set of n +
m elements A 1 , • . . , An, B of algebra .JI;
(n, m are fi.xed) indexed by ki < k;+ 1 (or ki > kH 1 ) and let Cz
be the $""-module generated by the space of functions cp (x, a),
x E Rn, a E Mm of the rate of growth not greater than a power of x
and such that
k1 kn kn+i
A ~-t-structure is defi.ned in :£, if for any set A 11 ••. , An, B,
kn+2 kn+l+2
C11 ••• , C1, where Ci EM, an operation is defi.ned >I'
11:x 1 ~A 11 ••. , Xn~An, y 1 ~C 11 ••• , y 1 ~C 1 , a~B,
which substitutes an element L E :£ for a function f (x, y, a) E c;.

The element L is denoted by
( k1 kn hn+1 kn+2 kn+l+2)

L=[f A11 •. . , An, B, C11 •. • , Cz ].
It must satisfy axioms (~-t 1 )-(~-t 6 ), and be compatible with the I-t-struc-
ture on .J{.
A module :£ with the given operation 11 is defi.ned as a module
over .J!: with a 11-structure.
* Of course, we suppose ki =1= kj whenever the two operators corresponding

to ki and kj, respectively, do not commute.
Sec. 7. An Example of a Solution of a Differential Equation

Consider the differential equation
iJ2u f)2u
fJt2 - Xt, fJx2 = 0. (7.1)
Denote by D = -i :x, D0 = - i :t. Find u(x, t) in the form
u (x, t) = eiD18 (2x, 2) (2

t <p x,
2)t v (x), (7.2)
where
S (x, 0) = 0, <p (x, 0) = 1, v E C'; (R).
Substitute function (7.2) into the left-hand side of (7.1) applying
the formula of commutation of Hamiltonian and exponential:
fj2 fj2 ) 1 (2 2) ( 2 2)
( fii2-x4 iJx2 [eiDS x, t <p x, t ] v (x) =
1 ( 4 4) 3 4, ( 1' as )2
=ews x, t [x4 D+Dax(x, t) -
1 : iJS
-(Do+ DTt (x, t)) ] <p
2 ( 2 2)
x, t v (x) = [eiDS
1 ( 4 '• )
x, t L] v (x).
It should be noted that L is a part of the operational equation, but
it is not an operator.
Let us transform the expression of L. Applying Theorem 6.4 and
taking into account that 62/ is a unit operator in this case and
[. a as
l iJx '
a2s
Tx
J= . l fJx2
we may put L in the form
L= Lx~.
4 ( 3 1 as ( 4 4 ) ) 2
r
D+Dax x, t - ( 3 1 iJS { 4 4 )
D0 +D-at x, t ) 2] 2 2)
cp ( x, t -
. 1 2t, fJ2S . 1 fJ2S •2 2 )
-lDx
2 2
fJx 2 ( x, t) <p ( x,2 2t) + lD fJtZ ( x, t.
Then changing in the first summand the order of action of the
3 3 2 2
operators D, D 0 and x, t by virtue of Theorem 6.4, we obtain
L=[~ 4 (D+D ~! (~. ~) r- (Do+D ~~ (~, ~)rJ cp(~, ~)-

-i [ xt. 2 1 as ( 2 2 )
D+D3 ) +2xilDTx
2 ( 1 acp ( x,
x, t Tt 2 2)
t J +
1
+i [ ( D0 +Do)+2DTt ;'
3 1 as ( 2 2)
t
JTt
acp (
x,
2 2)
t -
~. 2 2 2) (2 2)
- lDx
. 1 2!. azs
fJx 2 ( ""'
2
t) cp ( x, t) + lD 2 . 1 fJ2S
iJtZ ( x, t <p x, t.
INTRODUCTI!ON 'rO OPERATIONAL CAIJCULUS 57
Changing again the order of action of these operators, we obtain
L = [ x~ ( 1 + ~~ ) cp (x, t) D
2 2- [ { D+ D~; (~. ~) t cp ( ~. ~) ]+
0
+ i L~ -2x 4 ~
ax
(1+.2!__
ax
D) +2 ~as
at at
D+2!!£_
at
D-
0
(7.3)
{}2cp {}2cp •
where £ 0 = ot 2 - x 4 ax2 • Assummg that the first two terms do
not contain the derivatives with respect to X "we get an equation
for S:
X~ (1 +.2!__)
OX
2
- (.!!___)
fJt
2
= 0
which is called an equation of D-characteristics of equation (7 .1 ).
The equation splits into two branches:
as
at
=+
- x
2 (1 +!!___)
at ·
We solve the equation with the sign "-" in the right-hand side.
Considering the initial condition S (x, 0) = 0, we obtain
x2t
S (x, t) = - i+xt . (7.4)
Assuming that the last but oLe term in (7 .3) does not contain diffe-
rentiation with respect to x and applying (7 .4) we get an equation
for cp, i.e. the transfer equation
.2 orp ocp 0
- x ax -at+ xcp = .
A solution of the equation with the initial value cp (x, 0) = 1 is

oi the form
cp (x, t) = (1 + xt).
Thus we obtain £ 0 = 0 and
2 2
. 1 x2 t
-tD 2 22 ( 22)
u(x, t)=e 1+xt 1+xt v(x).
Note that
+oo
eil,,(;)Vx= V~n I
-00
eipxeiPf<x>v(p)dp=v(x+f(x)).
Therefore we finally obtain

x2t · ) ( x )
u(x, t)=(1+xt)v ( x - 1 +xt =(1+xt)v i+xt ·
Sec. 8. Passage of the Equation of Oscillations of a Crystal Lattice

into a Wave Equation
We shall apply the developed operational calculus to a physical
problem. Indeed, consider the equation of oscillations of a lattice
and study the question as to which of its solutions turn into the
solutions of the wave equation when the step of the lattice tends
to zero by passing to the limit.
This familiar problem is a classical one. It is hard to expect the
operational method to produce anything new. But, nonetheless,
the operational method permits us find unexpected physical and
mathematical corolaries. We shall find a critical solution of the
lattice oscillation equation by the operational method, which has
no limit, as the step of the lattice tends to zero. These critical solut-
ions are cores of physical phenomena discussed in this section,
some of which had not been known before.
An example under consideration in the section indicates that
the operational calculus is to some extent adequate for the physical
problem. Indeed, we shall see that the equation of characteristics
for a function of ordered operators not only defines the fundamental
phenomena like the effects of the Cherenkov type (a tail of oscillat-
ions behind a shock wave) but indicates the boundary of Cherenkov's
oscillation domain as well. These effects are illustrated fairly well
by computer calculations.
We shall cancel out an elementary study of the obtained formulas
and the computer calculations in the other parts of the book, but
here this simple classical example is studied thoroughly.
( 1) Transformation of the equation of lattice oscillations into
a wave equation. We shall consider a system of ordinary differential
equations of lattice oscillations and derive the wave equation.
Consider 2N atoms of mass m, lying on the circle of radius 1 at
the same distance h = n/N from each other. Assume that the atoms
interact only with neighboring atoms. Let ui (t) be a displacement
of the ith atom at a moment of time t from the equilibrium position:
here i = 0, +1, ... , +N, u_N = uN. The system of Newton's
equations for such a lattice in the linear approximation has the form
••
Un = c
2 Un+1-2un+un-1
kZ ,
0 N
n = , + 1, ... , + , (8.1)
y
where u N = u_N, u N+ 1 =
def
u-N+H u-N-1 =dcf , /
u N-1; c = ·v -;;
is a velocity of sound.
Assume that N is a very great number. Consider a smooth function
u (x, t) which takes the values ui at the lattice knots, i.e., u (jh, t) =
= ui (t). System (8.1) assumes the form of a differential-difference
system on a circle or, equivalently, a differential-difference system
INTRODUCT~ON TO OPERATIONAL CAIJCULUS 59
with periodic coefficients

(]Cu (x, t) c2
at 2 =h2[u(x-t-h, t)-2u(x, h)-t-u(x-h, t)], t>O,
u(x-t-2n, t)=u(x, t). (8.2)
We assume that, generally, the sound velocity c depends on x
and c (x) E C"" is a 2n-periodic function, c (x) =/= 0.
It is natural to consider the smoothest class of functions generated
by the values ui at the knots, since only the values of u (x, t) at
the knots are important. Let M N be a vector space generated by
the vectors eikX, k = 0, +1, ... , +(N - 1), +N. It is easy
to see that for any set {uh j = 0, ... , +N, uN = u_N} there
exists a function u EM N taking on the values ui at the knots of
the lattice. Therefore the initial values for problem (8.2) may be
put in the form
fJu
u (x, 0) = v1 (x) EM N• Tt (x, 0) = v2 (x) EM N· (8.3)
Further, applying the equation
e;( -ih :x )u (x, t) = u (x+h, t),
we rewrite problems (8.2), (8.3) in the form of pseudodifferential

equation
1
def h2 ---at2+
Lhu- fJ2u 4c2( x2) sm2( - ih Tx
2
fJ ) -0
u- , 0
t> 0 , (8 .4)
fJu
u(x, O)=vt(x), Tt(x, 0)=v2 (x), u(x+2n, t)=u(x, t).
Assume
n
II t II= ( ~ It (x) 12 ax
-n
f 12
and take cp E CJ;,> t if the function cp = cp (x, t, h) has s derivatives

with respect to x, t, 2n-periodic in x, and there is
sup 1/fJkq>(x,:; h) ~~~ck<oo, k=O, 1,. .,s. 0
O<h<1 fJx
First consider a problem with a smooth (uniformly in h) right-

hand side
1
Lhu = ()2u
h 2 012 + 4c 2
(
x2) sin ( 2 -
ih
2
fJ )
fJx u = f, f Ec<•)
h, t'
fJu
u(x, 0)=Tt(x,0)=0, u(x+2n, t)=u(x, t). (8.5)
60 OPERATIONAL METH10D.S
Lemma 8.1. Let cp ECh~t 2 >, then

m+1
~
. 2
4 SID ( ih a )
- - - m=
2 ax 'I' -h2 -+
a2<p
ax2
. k -+
ak(-Lh) ak<p
axk
k=4
+hm+2Qmcp,
where Jl Qmcp II~ Cm II ~=:~ II , ak are some fixed numbers.
Proof. We have, by the Taylor formula,
m+1 1
4 sin2 (_.:._) =z2
2
+ L.J
"'
akzk-2zm+2 f (i--r)m+i cos<m+2) (TZ) d1:
J (m+1)! ·
k=4 0
Hence
1
Q 2 (- i)m-2 [ f (1 - ,..)m+t X
mcp = - (m+i)! J •
0
X cos<m+2> ( - ih 1: _!_)
ax
( axm+2
am+2<p ) d'T:] .
The function in the brackets may be obviously put in the form

of a linear combination of the functions
1 a 1
J(
1- 1:)m+1e±i<( -ikax) '¢ (x) d1: = J(
1-T)1n+ 1ljl (x + 1:h) d1:,
0 0
am+2ql
where ljJ = axm+2 .
The following inequality is obvious:
1
II J(1-1:)m+1ljl (x + 1:h) d1: ~~~C ll'i' II·

0
The sought-for inequality for Qmcp follows from this. The lemma is
proved.
The first approximation for the operator L h as h-+ 0 in pro-
blem (8.5) is defined as a wave operator and is denoted by 0 c or L.
Thus we have by Lemma 8.1
·_ ~ a2 2 a2
L - O c - at 2 -c (x) ax 2 ·
The equation 0 cU = f is called the wave equation. Therefore we

shall say that the system of ordinary differential equations of lattice
oscillations (8.1) turns into the wave equation as h-+ 0. But we
still know nothing as to the behaviour of the solutions of the system.
Do they turn into the solutions of the wave equation as h-+ 0?
INTRODUCTI!ON TO OPERATIONAL CAIJCULUS 61
First of all, it is necessary to verify that the solutions of pro-

blem (8.2) and the corresponding wave equation exist.
We shall prove the theorems of existence and uniqueness for the
wave equation and equation (8.4) only for sufficiently small t and,
what is important, indicate formulas of the solutions as well.
(2) The existence and uniqueness of a solution of a wave equation.
We shall prove the following theorem here.
Theorem 8.1. Let f E C):! t (s is sufficiently large). Then for 0 <
< t ~ T, where T is a number, there exists a unique solution u =
= R 0 (f) E c~•.> t of the problem
Dcu=f, (8.6)
u(x+2n, t)=u(x, t), u(x, 0)= ~~ (x, 0)=0 (8.6')
and at the same time,
l
I :~z Ro (f) I~]

k=O
Cz, k II :;t II • l = 0, 1, 2, · .. , s,
and R 0 (f) = R (x, t), where R (x, t) is defined by (8.8).

We shall prove this theorem (as many others of the same type)
with the help of the construction of a regularizer of problem (8.6),
(8.6').
Definition. Let A = -i :x EM, B = x E M 1, M 1 be a circle,
G (t, xu a, s) E Cz,
s = 1, 2, ... be a sequence of one-parameter
families of symbols such that G (0, xu a, s) = 1, Gt (0, x17 a, s) = 0
and the following inequality is satisfied
lfdt, x, a, s)j~C(x~+1)-•1 2 (8.7)
for a symbol t 1 E Cz of an operator
/1
1 2
(t, A, B, s) =det ( 02
ai2 +[c2 (B) A 2]
) 1
G ( t, A, 2
B, )
s . (8. 7')
1 2
Then G (t, A, B, s) is defined as a regularizer of problem (8.6), (8.6')
(we shall cancel out the argument s in the sequel).
Note. The regularizer is obviously not unique. For example, we
1
may add a function finite in: A toG and thus obtain a new regularizer.
This is to be followed henceforth, so that the support of the new
regularizer might be equal to zero in the e-neighborhood of x 1 = 0
and hence we could multiply it by a polynomial in x~ 1 and expand
it by the powers of x~\ defining the symbol x~ 1 as a function x~ 1 cp (x1 ),
where cp (x1) has a support outside an e-neighborhood of the point
x 1 = 0, cp (x1) = 1 when x1 > 2e.
Suppose that such a regularizer is constructed. Now we shall

verify, that the theorem can be proved with such a regularizer.
Substituting function
21
R (x, t) =
t t-'t
j
0
I
0
1
G ( 11• A, B) (x, -r) d11 d-r (8.8)
into equation (8.6) we obtain

fJ2R
Bt2·-c2(x)
fJ2R
fJx 2 =<l>(x, t)+ J
r tt (
J /1 11• A,
1
0 0
(8.8')
Thus, we have reduced problem (8.6), (8.6') to a problem of finding
<l> from the equation
t
<l>(x, t)+ J/2 (t-T, A, B) <l>(x, -r)d-r=f(x, t),
0
1 2
(8.8"}
where
1 2 t
B)= 5tt( T, A,
def 1 2
/ 2( t, A, B) d-r
0
(the initial value conditions and periodicity conditions (8.6') are

obviously fulfilled for the function R (x, t)).
Equation (8.8") is an integral equation of the Volterra type and
is solved by the iteration method. Indeed, we take
00
<D = 21 <Dk,
k=O
where
<D0 (x, t) = f (x, t),
<l>t(x, t) = -J / t
0
2 ( t - T,
1
A, B) f (x,
2
-r) d-r
t 2k-1 2k 't!
<l>k(x~t t)=(-1)k so /2 (t-T, A, B) s/2 (-r~-
0
-T2, A,
2k- 3 2k- 2
B ) ... s 12 (-rk-1-Tk, A, B)
'tk-i
0
1 2
)<
X j (x, Tk) d'tk ... dT 1, k= 2, 3, ... ·

INTRODUCT~ON TO OPERATIONAL CALCULUS 63
We have an inequality for any \jl (x, t) ECh, t• O~T-<:t, t~T, and
sufficiently large s
1 2 t-'t 1 2
ll!z(t-T,A, B)\jl(x, T)il=ll~ /1 (fl,A, B)\jl(x, T)dflJI~
0
t-'t 1 2 '
~ ~ JJ/1 (fl, A, B) \jl(x, T)Jidf.t~M 0 TIJ\jl(x, T)jJ,
0
where M 0 = const, by (8.7).
Hence for 0 ~ t ~ T we have
(MoTZ)k
lieD~< (x, t) IJ ~ ki max II f (x, t) II·
· O""t"'T
00
Hence, the series LJ <Dk

k=O
is convergent. Indeed, it follows from
similar inequalities for the derivatives of <Dk that the series of
derivatives of <Dk converges uniformly (and very rapidly at that),
00
the sum <D = ~ <Dk belongs to Cj,~>t and the inequalities of theo-
h=O
rem 8.1 are true for R 0 (f)= R (x, t).
It is also obvious that equation (8.8") and, hence, problem (8.6),
(8.6'), have a unique solution.
Thus, what we need is to construct the regularizer of problem (8.6),
(8.6') (together with its estimates) to prove the theorem. We shall
proceed by the familiar operational method. Find the regularizer G
in the form
G (t, A, ~) = ei1s(;, t) fll (~, t, A) , (8.9)
where A = - i !,
the functions S and qJ are to be found (here S,
qJ are 2n-periodic in x; S is real).
Substituting operator (8.9) into the wave equation and applying
the commutation formula, we obtain
2
~:~ -cz (~) 0~2 G=eds(~. t) { ( -Az) [ ( ~~ )2 _

-c2 (~) ( ~! + 1 f]fll+ul[ ( a:e~) qJ-c 2 ( ; ) ( ~~) fll+
at •~-
+ 2 .!.§__ at 2c2 ( ~) ( ~
ax + 1 ) ~
ax J+
az<p
+ [ iii2-c 2
( 2 ) az<p
~ axz
J} '. (8.10)
where the arguments (~, t), (~, t, A)

of the functions S and <p are
cancelled out, respectively.
We want the right-hand side of (8.10) (indeed, the expression in
braces) to be equal to zero. It is impossible to have exactly zero
(in the case of the variable c (x)), but it is sufficient to get a smooth
operator / 1 on the right for the construction of the regularizer.
The regularizer G is considered as an operator in a space of 2n-pe-
riodic sufficiently smooth functions. Such functions may be put
in the form of a Fourier series
00
v (x) = a0 + 2J aneinx = a0 +v
n=-oo
1 (x).
n4=0
(1
Note, that f \A, B v (x) =
2) 00
2J anf (n, x) einx by definition. Then

n=-oo
1 2)
G ( t, A, x v=G 0 (x, t) +G ( 1 2)
t, A, x v1o where G0 (x, t) =det
= G(t,A,;) ao· The function Go is a solution of the problem
DcG =/t (t, l, ;) a

0 0,
G0 (x, 0) = a 0 , 8G0 (x, 0) = 0,

L
8 t
G0 (x+2n, t)=G 0 (x, t).
Without changing the regularizer we can obviously take the con-
stant a 0 instead of G0 , which satisfies the written out problem,
but has a zero right-hand side.
Thus it is sufficient to consider the regularizer G only for the
00
functions of the form v1 (x) = 2J aneinx. But for such functions
the operator A -l is defined:
A -tV tdef ~
= LJ
n=-oo
n4=0
and all its powers A -k, k = 1, 2, . . . .
Hence the function <p in (8.9) may be written in the form
<p=~
k=O
(-d- 1 r<pk(~.t).
Substituting this formula into (8.10) and equating to zero all coeffi-
cients of the powers (-iA - 1) \ k = -2, -1, 0, ... , s - 1 in
I"THODl:cTI·ON TO OPERATIONAL CALCULUS 65
the braces (8 .10), we obtain the system of equations
( ~~ )
2
-c 2 (x) ( ~~ + 1 ) 2 =0,
Utero= 0,
+
U1cr1 Occro =0, (8.11)
u tcrs + 0 ccrs-1 = 0,
where
as (:r, t)Dt-2c
U 1 =2Tt a 2 (x) ( a;-(x,
as t)+1 ) ax+
a OcS(x, t).
The operator G satisfies the wave equation with a right-hand side
(cf. (8.10)):
11 (t, A,~) =ei1s(~. t) (Occrs) (~, t) ( -iA- 1 )".
We have by dcfini tion
( 1 2)
It t, A, :r Vt = (- i)s 0 ccrs (x, t) ~
oo a .
n~ etn(x+S(x, t)).
n=-oo
Hence we see that 11 satisfies the estimates (8. 7). Thus we need
only to solve system (8.11) with the initial conditions
( 1 z)
G O,A,x =1,Tt O,A,x =0.
aG ( 1 2) (8.12)
The first equation (8.11) is the Hamilton-Jacobi equation for the

function S (x, t):
as = + c (x) (as
Tt Tx + 1 ) . (8.13)
Substitute the initial condition

S (x, 0) = 0. (8.13')
Consider the corresponding system of bicharacteristics, i.e.,
the Hamiltonian system, for the solution of the Cauchy problem,
(8.13), (8.13')
dx= - C (X±) , x± It=O =
-crt"=+ X0 ,
(8.14)
dP± =+__!!!__(X±) (1
dt - Ox
+ p±)' p± 1-
t-0
= 0'
This system is easy to integrate. Let X± (x 0 , 1:), P± (x 0 , t) be

its solution. Since aX± j = 1 for any t we have
Dxo t=o .'
~-012?5
ax±
axo (x 0 , t) =I= 0 and, therefore, the equation
x =X± (x 0 , t)
has a solution, i.e., there exists a smooth solution x 0 = x;l= (x, t).
Then problem (8.13), (8.13') has the following 2:n-periodic in x
solutions (corresponding to the signs +in (8.13)):
t
s ± (x, t) = ~ [ p± (xo, 't) a;± (xo, T)- H ± (P± (xo, 't),
0
x± (xo, -r))J dT I ± = xt (x, t)- x.

xo=xo (x, t)
as
Note that ax± (X±, t)=P±.
We now turn to the second equation in (8.11). The operator U 1
splits into two operators: Ut, U:; corresponding to two functions S±.
The operator U'f' may obviously be transformed into the form
U±= _ 2H (-a-+_!_ a2H± azs± _ _1_ aH± ali±)
1 ±at± 2 apz .axz 2H± up ax '
(8.15)
where H ± (x, p) def + c (x) (1 + p), the variables (X±, t) (X±, p±)
a
are cancelled out in the functions S±, H±, respectively,
is a derivative along the trajectories (8.14), i.e.,
a a ax± a a _ ± a
at± =at +---;u-a;- = Tt + c (X ) Tx ·
By (8.14)
a
Tt[H±(X±(x0 , t), P±(x0 , t))]=O. (8.16)
Hence, H ±(X±, P ±) = H ± (x 0 , 0) = + c (x 0 ).
Let J~ (x 0 , t) = ax~ (xo, t) ·Differentiating the equation for x± (see
xo
(8.14))
ax±= an± (x± as± (X± t))
at ap ' ax '
with respect to x 0 , we get
.!._Jo =Jo ( azu± azs±
at ± ± apz axz
+ azu±).
ap ax
On comparing this formula with (8.16) for the Jacobian
J ± (x, t) ti.ef 2H ± ( x, ~:x± (x, t)) J~ (xt (x, t), t)
I:\!TRODUCTION TO OPERATIONAL CAlJCULUS 67
we obtain an equation
d 1 1 (
dt± VI!± I=- 2 VI J ± I
Hence, for the operator (8.15) we have
u± ( 'tjJ (x, t) ) =
1 Yll± I
_1_ iJH± iJH±) l/J(x t)
2H ± iJp ax ' '
(8.17)
where the variables ( x, a~: (x, t)) of the function H ± are cancelled
out ('ljl is any).
We have made all calculations making use of the most general
notation, since they will be repeated in the sequel. In our case for-
mula (8.17) for the operator Uf- is of the form:
u± (
1
'tjJ(x,
VI'± I
t)) = - 1 [-d- +
Yl'± I at±
iJc(x)
ax
J
'ljJ (x t).
'
Therefore the second equation of (8.11) has two solutions
+ 1p~ (x~ (x, t)) c (x)
cp- (x t) - --::~~====;=-- (8.18)
o ' - VI J ± (x, t) I '
where 'IJlt' (x) are some initial conditions which ought to be defined.
In the same way all other equations of (8.11) are solved by (8.17)
and 2n:-periodic functions crt' k = 0, 1' 2, ... ' s are defined by
some integrals along the trajectories X±, P±.
Finally the regularizer G is of the form
G(t,A,~)=LJ ±eJs±U.t)crt(~,t) (-iA- r,

± k=O
1 (8.19)
where the functions S ±• cr,:: have been defined above and

2JJ± de! f++f-·
Now we need only to choose the initial conditions for crt such
that (8.12) are verified. Take
cr~ (x, 0) = 1/2. (8.20)
Since J ± (x, 0) = 2H ± (x, 0) = +2c (x), we have the following
equation for 'IJlt::
± ( 1
l/Jo x) = V 2c (x) •
5*
Take the conditions

q>£= (x, 0) = 0, k = 1, 2, ... , s (8.21)
for all other functions q>£=, k ?:: 1.
By (8.19) at the initial moment of time t = 0, and, taking into
account (8.20), we have
G (o, A, ~) = ~ ±
± k·=O
q>t (~. o) ( -iA- 1 )"
= ~ q>[f (x, 0) = ; + ; = 1.
±
The derivative fJG/fJt is of the form (by (8.19))
!
8 (o, )!, ~) = ~
± k=O
±[u! a!± U. o) q>t U. o) +
acp£= ( x,2
+---ar )
0,
J(-zA. 1-1) k
. (8.22)
The first terms in (8.22) are cancelled out by (8.20) and the con-
ditions a!: (x, 0) = -H ± (x, 0) = +c (x) and we have
s k
aG ( 1 2) _ acpt ( 2 ) ( . 1 _ 1)
at 0, A, x - ~ ~ ---;it x, 0 -zA .
± k=O
We find the derivatives

satisfy the condition
:il
a +
by (8.11). It is easy to see that they
iicpk acpk
at (x, 0) = - a t (x, 0)
at the initial moment. Therefore, the operator G ( t, ~) satisfies A,

the second initial condition of (8.12) as well. Thus the reg1:1larizer
for (8.6), (8.6') is constructed, Q.E.D.
(3) Asymptotic solutions of the system of lattice oscillations. We

shall distinguish between the two problems: (1) finding an
asymptotics of an exact solution; (2) finding an asymptotic
solution, i.e., the construction of a function which, substituted
into the right-hand side, is equal asymptotically to the given one.
We shall study problem (1), after we have constructed an asymp-
totic solution (i.e., after problem (2)). There is the following almost
obvious lemma.
DITRQ.DUCTJ:ON TO OPERATIONAL CAlJCULUS 69
Lemma 8.2. Let m > 1, k ~ 2 be integer numbers and j E Ch~ l

be a function, where p = p (m, k) is sufficiently great. Then, there
exist such functions '¢ i E C~~> t, j = 0, . . . , m - 2 that a linear
combination
m-2
1 j
U m = h2 R 0 (f)+ ~ h 'l'i
i=O
is an asymptotic solution of problem (8.5), i.e.,
[h2 ::z +4c2 (~) sin 2 (- i~ fx)] Um=f+hmFmf,

um (x+ 2:rt, t) = um (x, t),
Um(x, 0)= iJ~m (x, 0)=0,
and F mf satisfies an estimate

p {jj f
JJ FmfiJ~~ Cj I TxT II·

i=1
Proof. Substitute the function Um into the equation and apply
Lemma 8.1. On equating to zero the coefficients of the powers of h
we obtain equations for the functions 'i'h j = 0, ... , m - 2. For
example, assuming that the coefficient of h2 equals zero, we obtain
an equation for '¢ 0 :
0 c'i'o + a~,.c2 (x) ( - i :x) 4 R 0 (f)= 0.
Hence we find by Theorem 8.1
\jl 0 = - a~,.R 0 [ c2 (- i a~ r
R 0 (f)] ,
and the function '¢ 0 satisfies the initial conditions with the right-
hand sides equal to zero and is 2:rt-periodic in x.
In the same way we get all other functions '¢i· The estimate of
F mf is obtained from the corresponding estimates of Lemma 8.1
and Theorem 8.1. The lemma is proved.
Thus we see that the solution of the wave equation is an asymp-
totic solution of the system of equations of lattice oscillations. But,
firstly, it is an asymptotic solution for a sufficiently smooth right-
hand side, and, secondly, it is necessary to prove that it is an asymp-
totics of the sought-for solution. We shall apply the method of the
regularizer, studied before in connection with the wave equation,
to find a general asymptotic solution of the equation of lattice
oscillations and to prove that it is an asymptotics of the sought
solution.
In this case the regularizer is a usual operator G ( t, depend- A, B)

ing also on h, the definition is the same as before but in addition
to (8.7) we demand that
I /1 (x11 a, t, s, h) I~ c 8 h 8 +2 (x~ + 1)- 812 • (8.23)
We put the operator
i)2
h2 -+[4c 2 ( x sin 2 2) ( ---
ih 1) ]
~ 2 h
into the right-hand side of (8.7').
If the regularizer is constructed, we may obtain an integral equat-
ion of the Volterra type similarly to item 2, and prove that its
solution exists and obtain as well an estimate for the solution uni-
form in h as h -+ 0.
Note that it is sufficient to demand only condition (8.7) for / 1 •
Indeed, if such a weakened regularizer is constructed, we can
apply the results of this item, i.e. Lemma 8.2, which enables us to
obtain a necessary regularizer correct to kNF, where FE Ch~'tt i.e.
to satisfy (8.23).
Thus we have reduced the problem to the construction of such
a weakened regularizer, that the corresponding right-hand side / 1
satisfies condition (8.7). We shall fulfil this task in the present
item with the help of ordered operators.
Construct the regularizer for the system of equations of lattice
oscillations. Consider problem (8.4) for any arbitrary initial con-
ditions v11 v2 E M N· It is obviously sufficient to consider the case
v2 (x) = 0. Next, put the function v1 in the form
N
Vt (x) = ao + 2} aneinx .
n=-N+1
The constant a 0 obviously satisfies problem (8.2). Therefore it is
sufficient to consider the initial conditions j
= ~
au
u (x, 0) = v 0 (i) aneinx; at (x, 0) =0.
n=±1, ... , ±(N-1), +N
(8.24)
Let A = -i :x.
Define the operator A - 1 (as it was done before)
for the function v 0 by the formula
A-lvo = A-1 ( ~ aneinx) dcf ~ a: einx EM N·
ni=O njoO
The action of A - 1 transforms functions of this type into functions

of the same type, therefore all the powers A-~<, k = 1, 2, ...
are defined for v 0 • Besides, an estimate
II A - 1Vo II ~ II Vo II
I;>~TRODUCTI'ON T·O OPERATIONAL CAIJCULUS 71
is obviously true. Then :x (A - v

1 0) = iv 0 (x, h). Therefore
I ox·'
iJ·' II
(A -k v0 ) <,_C.I!v0 1!, s=O, 1, .. . ,k.
Hence, we see that it is sufficient to solve problem (8.4) correct

within functions of the type (A -kv 0 ).
We shall find the regularizer in the form (h is cancelled out)
G ( t, ~. ~) v0 = ei1s(;, t, ~) cp U, t, ~) v 0, (8.25)
where ffi = hA, h = n!N, S, cp are infinitely differentiable 2n-pe-
riodic in x functions, S is a real function.
It is to be noted that an operator of the form F ( ~. ~} for a function
v E Jv! N is defined by the formula
N +n
F (:r2 1) v~-1(!) ~ eikxp (x, kh) J e-ikyv (y) dy. (8.26)
' 2rt
k=-N+1 -n
The operator (!) is obviously bounded on M N
II (J)V II :::::;, 'J( II v 1!, v EM N· (8.27)
Substitute formula (8.25) into equation (8.4) and make use of
the formula of commutation with exponential. Then we obtain
that G ( t, ~. ~) satisfies equation (8.4), if the equation
azcp 12 ( as ) 2 1 as acp · . azs 1
{ h
2
atz - ffi ---at cp + 2~ffihat
.
Tt + ~ffih at2 cp +
3 ih a 1 1 ' as ( 1
'' 2
+4[c (x)]sin 2 [-zax-+'2(J)Tx x,t,ffi' ) ]cp } v0 =0,
(8.28)
where the arguments ( ~. t, ~) of S, cp are cancelled out, is true.
The last term can be transformed by the K-formula and the com-
mutation formula
. 2 3 ih a 1 1 ' as ( 1• ( 2 1)
sm [ - 2 ax +-z(J) ax x, t, (J))]cp x, t, (!) =
= sin2 ( ~ ( 1+ ~~)) cp + +(- ih) [sin (

m-1
~ ( 1 + ~: } ) ~~ +
++~cos(~(i+ ~~)}cp]+ L (-ih)kx
k=2
X (Rkcp) (~, t, ~) +hmQm (h), (8.29)

where the arguments (~, t, ~) of the functions S, cp are omitted.

Here Rk are some differential operators of order k acting on the
function <p (x, t, w) in the argument x (it is possible to write out
all operators Rk explicitly). The remainder Om and all its commu-
tators 1 8
estimates
=Om
2
\:x- :x ,
(1 3)8
s = 0, 1, 2, ... satisfy by (8.27) the
(8.30)
Note, that hv 0 = A - 1 wv 0 • Hence by (8.28), (8.29) we obtain the
equation
{[ (~ ~: )
2
-4 [c2(x)]sin2 ( ~ (1 + ~~))] cr+
(-L4-t) [2~2~~-2~[c2(x)]
-1- _ at at X
X sin (~ (1 + ~~ ) ) ~; +
+ ~2 ( ~:~ - [ c2 (x)] cos ( ~ ( 1 + :! )) ~:~ )<p J-

m- 1 ( 1 )k1 (2 1)2
-~ -iA- 1 wk(Rkcp) x,t,w [c2 (x)]-
k=2
3 1 1 1
-4[c2 (x)] A-mwwOm} V 0 = 0, (8.31)
where the arguments (~, t, ~) of the functions cp, S are omitted.

We shall solve (8.31) by expanding cp in powers of A -1, i.e.,
<p =
m-2(
2J - iA-1 )k cpk (2x, t, w1) .
1
h=U
Substituting this formula into (8.31) and equaling to zero all coef-
ficients of the powers (-iA - 1) \ k = 0, ... , m - 1, we obtain
a system of equations
(w :~ ) 2 - 4C2 (x) sin 2 ~ ( 1 + ~~ ) = 0,
R;cp 0 = 0, (8.32)
R;cp 1 + R;<p0 = 0,
INTRODUCT:rON TO OPERATIONAL CALCULUS 73
where
R 1' = 2 !.§___!__
iJt iJt
-2 czw(x) Sin
0 (
(J)
(1 + !§.__))
iJx
_!__
iJx
+
+ ( a;t~ _cz (x) cos ( (J) ( 1 + :~ ) ) ~~ ) ,
and the operators Rk, k = 2, 0 0 0' m- 1 are given by the following
formulas:
R/, = -4ffik-2c2 (x) Rko
Assume that S and cpk are found from equation (8032)0 Then we
see that the function
m-2 1 (2 1) ( 2 1 ) ( 1 )k
Vm(x, t) = ~ eiAS x,t,w (jlk x, t, (J) -iA-1 Vo+ao
h=O
sa lisfies equal ion (8 o4) correct within the function
Km (x, t) 1 '(3
= eiAS x, t, w
1) [ 2 1
h2Q;,A-m+z Vo, J (8033)
where the operator Q;, is given by the following formula
3
Q:n = 4 [ c2 (x)] j~ 1 ~
{m-ti-2 (- i)m-z (1)1-2 X
1.
X (Ric:pm-j+l) ( 2
X, t,
1) (
(J) - iA- 1
1 ) l
+ Qmffim-z
2 1 "}
0 (8o34)
We have, as well,
(-i~)Km(x,t)=
= { as (; t ~) eilsG. t, ~) h2Q2' ~-m+3 _
iJx ' ' m
01 (31)
-etAS x, t, (l) h2
[ 2
Q:n,- f) ]1
i ax A-m+z +
+ eiAS
1 (3 1) 2 1
x, t, w h2Q;,A-m+3
}
Voo
We have from (8034) and (8.30)
II-i :x Km (x, t) ~~~c1l1 Vo II•

where c1 = const.
The estimates for all other derivatives are obtained similarly:
II ::s Km(x, t)JI~csl!voll• s=O, 1, 2,. oo,m-2.

Therefore, if sup II v1 II ~ c, then the function V m satisfies

0<h<1
equation (8.4) correct within an element from Cit':'!2 ):
a2 +4 2() . 2 (
[h2 at2 ih a )]vm -K c<m-2)
c X sm -2 a; - m E h, t .
We turn now to the solution of equation (8.32). Firstlyt we have

the equation of Hamilton-Jacobi for the function S (x, t, ro):
~=+2 c(x) . ~(1+~) (8.35)
at - w sm 2 ax ·
The initial condition is
S (x, 0, ro) = 0. (8.35')
Consider the corresponding system of characteristics for the
solution of the Cauchy problem
{
dx± -
dt+ = + C(
x± ) COS [ 2w (1 + p ±) J, x± I t=O = x 0,
(8.3G)
d~t- = + ~ sin [ ~ (1 + p±) ~: (X±), J p± \t=O = 0.
The system is easy to integrate. Let X± (x 0 , ro, t), P± (x 0 , ro, t)

ax±
he its solution (the spectrum a (ro) E [-:rt, :rtl). Since-,-
uXo
I t=O
= 1
the equation
(8.36')
can be solved for sufficiently small t, i.e., there exists a sufficiently
smooth solution x 0 = x(f= (x, t, ro). Then problem (8.35)-(8.35')
has the following 2:rt-periodic in x solution
s± (x, t, ro)
t
=) {p± (xo, ro, 't)
0
d:: (xo, ro, 't) +
+ 2 c (X±(::, w, T)) sin[ ; (1 + p± (x 0, ro, •)) J} X

X d't Ixo=xo± (x, t, w) .
Equations (8.32) are further solved just in the same way as in
item 2 for the wave equation.
Thus the weakened regularizer of problem (8.4) is found in the
form
1 2) m- 2 1 (2 1) ( 2 1) ( 1 ) k
G ( t , ro, x v 0 = ~ ~
.Li .Li eiAS+
- x, t, w ±
'Ph x, t, ro -l
'A-1 v0 .
± h=O
(8.37)
[NTRODUCTI'ON TO OPERATIONAL GAIJCULUS 75
Now, considering the note made at the end of the last item, we
obtain the following theorem.
Theorem 8.2. For sufficiently small t there exists a unique solution
of problem (8.4) which can be put into the form
u (x, t) = G (t, A, B) v 0 - h- 2R 0 (Km) +0 (h),

by condition (8.24).
We shall indicate a few simple corollaries. For this purpose we
must put formula (8.37) into the form of an integral, using a pseudo-
differential operator on a circle, and to study an asymptotics for
h.- 0 of the integral by the method of stationary phase (a simplified
version of the saddle-point method for the real case).
First introduce a new version of the proof of the stationary phase
method.
(4) The stationary phase method. Consider an integral
1 r .<v>
l(h)=v- .I e'_h_cp(y)dy,
2nh •
where cD (y) E C'XJ (R), cp (y) E C0 (R), ~:; =i= 0 for y E supp cp and
the equation 88~ = 0 has a unique solution y = y0 for y E supp cp (y),
i.e. a stationary point.
Lemma 8.3. Under the above conditions for any N~1 there is an
e:cpansion
i
n n --<l>(yo) N
i - i-(1-sgn <l>"(y)) e h
I (h)= e 1' e 4
Vi <ll" (Yo) I
(
cp (y 0 ) + LJ
""' hk'¢k (Yo) ) +
h=l
+ 0 (hN+I), (8.38)
where
th-1
iv
0
J (y, tk)
where y = y (x, t) is a solution of equation (8.49) and the Jacobian

def ~
J (x, t)=J(y(x, t), t)
is defined by formula (8.48).
For the proof we shall apply the asymptotics for h ~ 0 of the

Cauchy problem for the Schrodinger equation of the oscillator
i
. a¢ h2 aztjJ x2
- ~h Tt = --y ax 2 +2 \jl, \jJ lt=O = cp (x) e
-(x)
h • (8.39)
This asymptotics is called quasiclassical. On substituting the follow-
ing function into (8.39)
-i~ -i~(i-sgn t)
e 4 e 4
'" (x, t)
'Y
= (2nh I sin t I) 112
.x
+ 00
Jr
i i
- 2.ht[cos tx2- 2xy+cos ty2] -h(y) ( ) d
X e sm e cp y y (8.40)
-00
we see that it is the solution of Problem (8.39). On taking x = 0,

t = + i
in formula (8.40), we obtain
+itt
I (h)=lJl (o, ~) e-4 , (8.41)
On the other hand, taking x = 0, t = - ~ in (8.40), we get
.3lt
t-
I (h) = e 4 'ljl (0, -n/2). (8.42)
Thus, by virtue of (8.41), (8.42) the expansion of the function I (h)
in a series happens to be a quasiclassical asymptotics of the solution
of the Cauchy problem (8.39). We shall find a solution of (8.39)
in the form of an asymptotic series
i
hS(x, h {
\jl(x, t, h )=e
t)
[cp 0 {x, t)+ cp 1 x, t)+ ...
i
hS(x, (8 43
... l N cpN(x,t
+~
• )
+···l=e def t)
cp(x,t,h), . )
where S (x, t) E Coo, cpt (x, t) E Coo are the sought functions.
On substituting the function '¢ (x, t, h) defined by equation (8.43)
into equation (8.39), we obtain
( ~-
at
(~r+x2) +
2 cp
. ( a<p as a<p 1 a2S ) h2 a2 (x).
II\"TRODUCTI!ON TO OPERATIONAL CALICULUS 77
Consider the Hamiltonian system of the oscillator

q=p, q (0) = y,
a
p=-q, P (0) =-a-
y
(y). (8.46)
Its solutions are obviously the functions

q (t, y) = y cost+ aa~ sin t; p (t, y) = ~: cost- y sin t. (8.47)
The functions q and p are called trajectories of the Hamiltonian
system (8.46). Denote by
~ det aq
J (y, t) =ay (t, y).
By (8.47) we obtain
~ f)2<J> .
J (y, t) =COSt+ ayz Slll t. (8.48)
It follows from this formula that

f)2<J>
(1) if ayz >0 for yEsuppcp(y), then
l(y, t) :F 0 when y E supp cp (y), O~t~ ~ .

az<J>
(2) if ayz <0 for y E supp cp (y), then
l(y, t) :fo 0 when y E supp cp (y), - ~ ~t~O.

a2<J>
Assume in the sequel that ayz > 0. All the arguments are the
same in the case ~:~ < 0, if t is changed for ( -t).
Denote by y = y (x, t) a solution of the equation
q (t, y) = x, (8.49)
which exists by the theorem of the implicit function. By analogy
with item 2 we obtain, that a solution of equation (8.45) is given
by the formula:
t
S(x, t)=<P(y (x, t))++) [p2 (T, y)-q2 (T, y)]dTiy=y(x,t)·
0
(8.50)
Consider the remaining terms in equation (8.43). Introduce a
vector field
a a d
m+ p (t, y (x, t)) ax. =det d-r: •
We want to calculate the derivative :-r: J (y (x, t), t). Note

first, that the vector field d/d-r: corresponds to the differentiation
along the projection on the plane (x, t) of the trajectories of the
Hamiltonian system. We have
..!:_ J
d-r:
(y (x, t) t)
,
It=to-
_ rliD J (y (x, tol. to+ !'J.t) -T(y (x, t 0 ), t 0)
!'J.t
dt-+0
= J( y (x, t)0 ' t)l"

0 liD
dt-+0
f(x,!'J.t)-1
!'J.t '
where J (x, !!..t) corresponds to the Jacobian of the shift transfor-

mation along the trajectories of the following Hamiltonian system
~ def
q=p, q (0) =X=q0,
~ ~ def
p= -q, p (0) = p (t 0 , y (x, t 0))= Po·
~ def
In order to calculate J (x, 1'1t) = J M• we note first that the Hamil-
tonian system provides the equation
q(M) = x + p 0 ·M+ 0 (!'1t2)
and
It is easy to see that the function S (x, t) defined by (8.50) satisfies

the condition :~ = p. Finally,
d ~ op0 o2S ~
dT J (y (x, t), t) lt=to =ax= ox2 (x, to) J (y (x, to) to).
Denote now
J=J(x, t)=J(y(x, t), t).
Hence for any El (x, t) E coo
..!:_ ( 8(x, t)) __1_~_..!__8_ azs
d-r: -v J - vJ d-r: 2 v:; ox2 •
Now put in (8.43)

( t) def eh (x, t)
cph x, = VJ'
and substitute the function 'ljl (x, t, h) from equality (8.43) into
equation (8.39); then equating coefficients of equal powers of h
to each other and applying formulas (8.44), (8.51), we obtain a re-
INTRODUCTI'ON 'DO OPERATIONAL CALICULUS 79
current system
d8o
d-r: = 0, n Jt=O
vo = cp ( Y) ,
Hence it follows thaL 00 (x, t)=cp(y(x, t)) and e", k>i is defi-
ned by the formula
1h-1
J
r v-J :22
uX
cp (y)
V J (th)
dt
k • . •
dt
1>
(8.51)
0
where J (ti) dcf J (x, ti) and the integration is carried along the
trajectories of system (8.46).
Now we shall prove that the function 'ljl (x, t, h) defined by equat-
ion (8.43), where the functions cpk (x, t) and S (x, t) has been just
found by us is indeed an asymptotic expansion of problem (8.39).
Take
l iS(x, t) N
'ljlN (:x:, t, h)= V
J (x, t)
e-h- [ cp (y (x, t)) +~
h=i
h"Ok (x, t) J.
We must prove that the function 'ljl N (x, t, h) is insignificantly
different from the function 'ljl (x, t), i.e., the exact solution of pro-
blem (8.39).
Indeed, we shall prove that
N+2_
max I'ljlN (x, t, h) -ljl (x, t) I= 0 (h 2 ). (8.52)
(x, t)ERnX [ 0, T]
In order to obtain estimate (8.52) we note that 'ljlN (x, t, h) satis-
fies the equation·
-iha'~'N=.!(-hz iJ21pN +xz•" )-hN+2x
ot 2 ox2 'YN
i
X eilS(x, tJ_a_z -'SN'==(x='=t)=
ox2 v
J (x, t) '
.
t-
'i'.v lt=O = cpe h
by definition and therefore the difference ljJ N - ¢ = 1ji is a solution

of the problem
- ih a1jJ = _i_ ( - h2
at 2
a2¢
ax2
+ .x2;h) -
'I'
N+2 h i
S (x, I) a 2 e
N (x, t)
- h e --'::'::::===-
ax2 v J (x, t) ,
(8.53)
It is easy to prove by direct calculation tlJat the solution of pro-

blem (8.53) is given by the formula
. zt . 1t
t _,T -tT(1-sgn(!-'t))
'iJ (:r t) = \ e ' .e X
' ~ V2nhjsin(t-T)I
+oo
Jr
i
X e 2h sm
,· (! ) [cos(t-'t)x2- 2xy+cos(t-'t)y2]
_, X
7 N+2 I~S(y,t) a2 8N(y,t) d d (8.54)

X ~ e -;;--z
uX
V J (y, t) y '·
It follows from (8.51) that the functions ek (x, t), k = 0, ... , and
<p (y) belong to C~ (x, t). Taking into account that 11V sin z is
a function integrable in the neighborhood of the point z = 0, we
obtain estimate (8.52) directly from equality (8.54).
Now substitute x = 0, t = ~ into our formulas. Note that
S ( 0, ~ ) = <D ( y ( 0, ~)) . This equality is easy to obtain
from formula (8.50). We have from (8.47)
q ( 0, ~ ) = 0= aa~ .
By hypothesis, the equality ~~ = 0 is possible only at the point
y = y 0 • Thus y ( 0, ~) = y 0 and we obtain expansion (8.38).
Note 1. The method of stationary phase can be applied to integrals
of the form
Ib
<p (y)
. <D(y)
e'-h- dy.
a
Let the derivative ~~ be different from zero in some neighborhoods
of the points a and b. Consider a partition of unity {e1 (y), e2 (y),
INTRODUCTION 1'0 OPERATIONAL CAliCULUS 81
e3 (y)} on [a, b] under the following conditions: e1 (y), e2 (y), e3 (y) E

acD
ECr:, e1 + e2 + e3 = 1 on [a, b]; e2 (a) = e3 (b) = 1, oy =foO,
when y E supp e2 U supp e3 , supp e1 E (a, b). Then
b . <ll(Y) + oo • <ll(y)
~ {p (y) et-;,- dy = ~ ~ (y) e1 (y) /-h- dy +
a -oo
+J + Je
oo . <ll(Y) b • <lJ(y)
e2 (y) ~ (y) et-h- dy 3 (y) ~ (y) /-h- dy.
a -oo
The first integral has been calculated by us, the other two, as can
easily be proved integrating by parts, are of the order 0 (h).
Note 2. We shall also need sums of the form
N i
1 h ~ eh<ll(nh)~ (nh),
V 2Jth n=-N -1
They can be reduced to the previous case with the help of the equation
.V ..:!_<ll(nh) f i <ll(y)
h ~ eh ~(nh)= J e h ~(y)dy+
n=-N+1 -a
oo +a • <ll(y) ·
+2 ~ J e•-"- ~ (y) cos (2kNy) dy, (8.55)

h=1 -a
which can be proved very easily by an expansion of the function
. <l) ()!)
e' -h-~ (y) on the segment [-:n:, :n:] in a Fourier series and by sub-
sequently using the formula of geometrical progression
N
h ~ cos knh = 2:n:6h, 2 mN, m is an integer.
n=-N+1
(5) The tail of oscillations. Solution (8.37) is a sum of functions
of the form
+a N
2~ .\ Vt (£) L e<ifh)nh[S±(x, t, nh)+x-£]~± (x, t, nh) d£.

-a n=-N+1
Put by formula (8.55)
N
h ~ e(i/h)nh[S±(x, t, nh)+x-£] (p± (x, t, nh) =
n=-N+i
a +a
J ~ (x, t, ro)
oo
= J~ (x, tro) e<ifh)w[S±(x, t, w)+x-£] dro +2 ~ X

-a h=l-a
X e(i/h)w[S±(x, t, w)+x-£] cos 2kroN dro.
6-01225
The stationary point w0 of the integrals is defined by the equa-

tions
(8.56)
For sufficiently small t this equation has a solution only for k = 0,

Since S± (x, t, ffi)lt=O = 0.
We shall estimate the kth (k =1= 0) term of the sum. We shall
obtain the estimate of the form 0 (h/k 2 ) integrating two times by
parts the exponential (the first time putting the cosine under the
differential sign and the second time expanding the cosine by Euler's
formula). Hence it follows that a sum of the series from k = 1 to
infinity is of the order 0 (h).
Now turn to the solution of implicit equation (8.56). It turns
out, that the left-hand side of the equation has a remarkable pro-
perty. Its derivative along the integral curve of system of bicha-
racteristics (8.36) corresponding to a sign plus or minus equals
zero. We leave the proof to the reader; in fact, it is of general nature.
Hence it follows that the left-hand side is constant along the
corresponding bicharacteristic, and since
88
s+ lt=o =
-
" ±
u(t)
J
t=O = o,
it equals xf1= (x, t, w), i.e., a solution of equation (8.36): x =
= X± (x 0 , w, t), for sufficiently small t. Therefore, equation (8.56)
for the stationary point is equivalent to the equation
X±(~, w0 , t)=x.
Thus the equation for the stationary point has a solution only for
those x, which belong to closed domains
Q± = {X±(~, w, t), I w I~ n},
i.e., the domains formed by the ends of the trajectories of the system
of bicharacteristics (8.36) going out of the point ~-
Let x lie inside a domain of the form Q±. On differentiating sys-
tem (8.36) with respect to w, we obtain that fJ~±u(t)
=1= 0 when t =1= 0
and therefore ~
ax±
= - - " - -8
ax±;ax±
=1= 0 when t =1= 0, w =1= 0. Hence
OW u(l) Xo
the derivative of the left-hand side of (8.56) with respect to w is
different from zero when t =I= 0, w =I= 0 and the method of stationary
phase can be applied. Thus, the solution rapidly oscillates inside
the domains Q±, its derivatives are not bounded ash-- 0 and there-
fore they obviously do not turn into the derivatives of a solution
of the wave equation.
INTRODUCTI'ON T<O OPERATIONAL CAL'CULUS 83
Note. If x lies on characteristics of the limit wave equation (i.e.,.

X = x;;= (£, t), where d~;t' = +c (X;t:), x;;: (£, 0) = £), then the
stationary point ffio = 0 is degenerate
Indeed, it means that the main shock wave propagates along

the characteristics of the limit wave equation and the tail of rapid
oscillations follows it. The support of the tail is defined by the
characteristics X±.
We see from this example that characteristics defined by Hamil-
ton's equation dependent on ordered operators define exactly the
support of the oscillation tail. The fact follows from the general
theory for a large body of problems and the discussed equations are
of the general type too.
Now we wish to compare the support of the tail with computer
calculations.
We take l = 10- 5 em for a measure of length, the length of the
inlerval (a circle) 2l, h = 10-s em, the velocity of sound in a crystal
c = 10 5 em is taken for a measure of velocity. We take a periodic
sec
function of the form
ult=o= ~ (.r-2n-1), when 2n~x~2n+2, n=O, +1, ...
with a jump at tbe points x = 2n, n = 0, +1, ... for an initial
condition.
In Fig. 1 there are graphs of the functions
max Jau (y ), .
mill - au (y ),
YE[x-5h,x+5h] X YE[x-5h,x+5h] ax
which correspond to the upper and low boundaries of the domain.
The exact solution is represented by a continuous line and the
asymptotic solution is represented by a dotted line.
(6) The Cherenkov effect. In 1934 the Soviet physicist P. A. Che-
renkov discovered the effect of non-damping radiation of light,
produced in a media by moving charged particles. The effect was
studied theoretically by I. E. Tamm and I. M. Frank in 1937. The
extraordinarily wide scope of the effect, however, became clear
only in the fifties. In 1958 P. A. Cherenkov, I. E. Tamm and
I. M. Frank were awarded the Nobel Prize.
The effect deals with electrons. When an electron moves in a media
with a velocity greater than a certain value depending on the media,
fast oscillating electromagnetic waves, i.e., light waves, radiate
in a cone behind it. These light waves are called the Cherenkov
6*
84 OP8RATIONAL ME-THODS
radiation. The mathematical nature of the effect depends on the

characteristics of ordered operators. The operational method makes
it possible to find a similar (mathematical) effect for a broad class
of pseudodifferential equations and, in particular, difference schemes
(including systems of equations of a crystal). In the latter case it
follows directly from the asymptotic formulas obtained above
max
min
Fig. 1
A function of the 6-type at the point x = vt is introduced into

the right-hand side of the equation of oscillations (of the electro-
magnetic field for the Cherenkov effect, the crystal vibrations in
our case) to define the movement of an electron with a velocity v.
Mathematically, the Cherenkov radiation is a tail of fast oscillations
of the solution, which follows the point x = vt. The effect is related
to the one studied in the previous item. We shall study the tail
theoretically and then compare the results with computer cal-
culations. It will turn out just as in the previous example, that the
tail (i.e., the domain of the Cherenkov radiation) is completely
defined by the bicharacteristics.
Thus consider the problem
def jj2u • ( ih a \
{ Lhu=h2ift2+4c2 sm 2 - 2 ().") U=cp(t)6h(x-vt) (S. 57 )
u(x, O)=ut(x, 0)=0, u(x+2n, t)=u(x, t),
INTRODUCTrON TO OPERATIONAL CAL·CULUS 85
where ()h (x- vt) = 2~ ~ eik <x-vt) and cp (t) is a smooth function.
k=-N+1
We shall solve the problem for sufficiently small t. The solution
def
of problem (8.57) Lhu = f is unique by Theorem 8.2: u =LIN. In
analogy with (8.8')-(8.8") we obtain that
t t-'t
R(x,t)= 1~2 I J G(T', ~. ~) cp(T)()h(x-Tv)dT'dT

0 0
satisfies the equation

t t -'t
I/
2
+I
1
LhR = cp (t) l)h (x, t) 1 ( T', w, x) cp (T) l)h (x, T) dT' dT,
I) I)
where / 1 ~. ~) satisfies estimate (8.7). By Lemma 8.2 and Theo-

(t,
rem 8.1 a solution of an equation
LhRi = .~
0
tr It (
0
T'. ~. ~) (jl (T) l)h (x, T) dT' dT,
I
R 1 t=O -
-aRt
fit
It=O-
-0
belongs to C~'Stn>, where k (m)-+ oo as m-+ oo. Therefore, since

R-+ R 1 is a solution of initial problem (8.57) the whole non-smooth
part of the solution of the problem is contained in R. We have
come to the conclusion, that the main term of the asymptotics of R
as h-+ 0 having been integrated by parts is defined by an integral
of the form (in analogy with the previous item):
1
h2·2nh
ss
+n xjv -
e
iwS+(x, t-1;/v,w)
h
OS+
--
(
X,
't'd {x, t-v,
t - -s,
) W
G
w)
-n 0 _ i)t v
iwS-(x, t-~,ro)
e h (Po ( x, t - VG , w) eiw(x-~) (jl
(vG) dwd'G.
dS_
Ot
(x ' t-lv ' w)
Applying successively the stationary phase method first in w
then in £, we obtain that the common stationary point w0 , £0
(<U 0 =I= 0) is defined by the system of equations

So= x6'= ( x, t- ~ , <U 0 ) ,
(8.58)
<U0 = + c (x) sin [~ ( 1 + as+ ( .x t- so
v _ 2 iJx ' v '
It is easy to verify that if <U =I= 0, I <U 0 I < n, I so I < n, then

the second derivatives are different from zero and the stationary
phase method can be applied.
For the sake of simplicity consider the case c = const. Then
system (8.58) is of the form
<Uo =
+
- v2c .
sm 2 '
f•Jo
- ct cos~
x+ 2 (8.59)
£o= c (i) •
1 -+-cos-
v 2
0
We see on the graph that a solution of (8.59) exists, when <Uo =1=
=I= 0, if
(8.60)
since I <U I ~ n. Inequality (8.60) being satisfied, the solution

oscillates as h -+ 0 in a domain consisting of the ends of all trajec-
~~7'V I
I I \ ~t
0 0.2 0.'1- 0.6 0.8
Fig. 2
tories coming out, when t = 0, of domains
Q~ = { X 0 = ~ ( 1 + ~ cos ~o ) } , (8.G1)
where <Uo is defined by (8.61). Then <U 0 = 0 is a degenerate point

and the solution is of a peculiar character ("resonance").
Compare the width of the oscillation tail for problem (8.57) cal-
culated by formulas (8.61) with an exact solution of problem (8.58)
obtained by means of a computer for various v. Take the length
of a circle (in units of the previous item) equal to 2, c = 1, h = 10-3 •
I~TRODUCTION TO OPERATIONAL CALICULUS 87
The second derivative ~:~ may be approximated by the formula

4 .2i1: a
T=i.J" >< 1 -.
Q4
---::r:rsm 2 at,
The graph of the function IJl (t) is shown in Fig. 2. For the graphs
of the functions
r)u . Du
max -,-, nun -
[x-111/z, x+JIIh] iJx [x-!Oh, x+!Oh] ax
in different moments of time see Figs. 3, 4. The inlerval between
v =O.Bc
0
em)
ll
3867
v=0.99c
100
o~=~!r-.......,.,==
-100
0
-5029
0
v=7.7 c ~
!00
0~~~~~~==~
0
-100 °
Fig. 3 Fig. 4
the graphs of these functions is blackened. The boundary of the

oscillation tail obtained from the asymptotic formulas is indicated
by vertical lines.
88 OPERATIONAL 1\IETHIODS
(7) The focus in a crystal. Consider an exam pie of a fast-oscillating

initial condition
u lt=O = e'iJ (x, h),
-au' 2ic. -(JJ' " (x

at t=o = -e-Stn
h 2 '~" '
h) ' (8.!32)
where 'ljl (x, h) is a periodic extension from the segment [0, 2:rt]
of a function cp(x)cos 8 ix); S, cpECoo, suppcpE(O, 2:rt), e is
a parameter.
The Fourier coefficients of the function 'ljl with a number n, 1 n 1 >
> N are of order 0 (h 00 ) by the stationary phase method, therefore
the formulas obtained in the previous items may be applied in the
case of initial conditions (8.62) as well.
Now we want to compare the results obtained by the formulas
of item 3 and the corresponding computer calculations.
In the computer case we took: the length of a circle 3l, l = 10-5 em,
the moments of time ti = 2i -10- 10 sec (i = 0, 1, ... , 5), in the
units l
S (x) =x2
4x when x E [0; 0.25]
1 when x E [0.25; 2)
4(2-x)+1 whenxE[2; 2.25]
0 when x E [2.25; 3].
A solution of the crystal equation for c = const may be calculated
with the help of formulas (8.37) and (8.38) along the same lines as
it was done in the two previous items. The term corresponding
to the sign "+" has no stationary point and therefore may be can-
celled out. The stationary phase method is obviously true for ct < L
The condition 8 2(J)/8y 2 =1= 0 of Lemma 8.3 is violated at the point
ct = 1, x = ~ . At this point, called focal, the crystal vibrations
have an amplitude bigger than in the neighboring points.
Note that the focal point can be defined directly by the equal ions
of bicharacteristics (similar to (8.36)) for c = consL:
(8.62')
The initial conditions for the system of equations (8.62') are

x (0) = x 0 , p (0) = Po = 2x 0
corresponding to initial conditions (8.62) with S = x 2 • Consider
a phase plane p, x and there the points x = 2:rtn. If we identify
these points we shall obtain a cylinder. The line p = 2x, corres-
INTRODUCTION TO OPERATIONAL CALCULUS 89
ponding to the initial condition p 0 = 2x0 , moves along trajecto-

ries of the system of bicharacteristics: indeed, we obtain a curve
Fig. 5
{x (x 0 , t), p (x 0 , t)} at the fixed moment t. This change of the
curves at the moments of time t = tH . . . , t 5 is shown in Fig. 5.
3
(J·to-5cm)
.:::::
3
(H0-5cm}
Za~ t=O.'N0-'0s
0
-Za~
2ae
2
(2·10-5 cm}
= J
(8-!0-5t:m)
0
-2ae J
(3-!o-5cm)
t=o. B·to-70s
z J
(Z·to-5cm) {J·to-scm)
Fig. 6
The curve is projected on the axis p at the moment t = t5 and when
p = n the tangent of the curve is parallel to the axis p. The
HO OPERATIONAL METHODS
point at which the tangent is parallel to the axis p- is called

a focal point.
The graphs max U and min U (with the domain between them
blackened) for the exact solution of equation (8.4) with. initial data
Fig. 7
(8.62) and t = t 0 , • • • , t4 are shown in Fig. 6 and the corresponding
graphs for the solution obtained with the asymptotic formulas
(a = 1) are shown in Fig. 7.
Zae
o~=====•
1
-2ae (J0-5 cm)
Fig 8
The graphs corresponding to t = t5 are shown in Figs. 8 and 9

(Fig. 8 corresponds to the exact solution). We see that the amplitude
of the solution grows rapidly at the focal point. We shall now con-
sider a nonlinear equation of the lattice vibrations.
INTRODUCTION TO OPERATIONAL CAIJCULUS 91
(8) Nonlinear equation of a crystal. We have confined ourselves

to the harmonic approximation of the system of Newlon's equations
for the atoms of a lattice. In fact, we assumed the amplitude of
the initial displacement to be small and ignored the non-linear
Fig. 9
terms. Now we shall take into consideration the amendment intro-

duced into the equations of lattice oscillations by the quadratic
terms. It means that a term
is added to equation (8.1), where a is a parameter.

On substituting Un + Wn into this new equation, where Un
is a solution of the linear equation, we obtain the following non-
homogeneous equation
i'V n -c2 Wn+t-2Wn+Wn-1

h2
=c2a[(U n+i - Un )2-(U n - U n-1 )2]
for I Wn - Wn-l I ~ I Un - Un-I I· For the sake of simplicity we

consider the case, where c and a are constants. We shall consider
the general case in Appendix.
We assume that the dependence of the initial data Un (0) and
Un (0) on the parameter h satisfies a substantial condition. We shall
introduce the following definition to the effect.
A system of functions Vn (h), -N + 1 ~ n ~ N corresponds to
a limit distribution g (x, p), -n ~ p ~ n, where g is an even
distribution in p, if for any function a E C"" ( -n, n) the following
limit condition is true

N n n
limh ~ a(nh)vn(vn+k!vn-11.)=~ ~ a(x)g(x,p)coskpd:rdp,
h-.O n=-N+1 -n -n
(8.63)
where
The physical sense of this definition is especially clear when

k = 0, since in this case it is necessary that the mean value a (nh)
should converge to a mean value a (x) of distribution.
We shall consider, as we did before, a space of functions Jl!JN
instead of the functions on a lattice; moreover, we shall formulate
the final theorem in terms of the space functions M N· Note, that
the arguments of this item can be extended to a wide class of non-
linear pseudodifferential equations with non-linear terms considered
small. The first approximation of the theory of perturbations in
this case is a solution of equation (8.4) and the second one satisfies
the equation
a
~ ( (sin~
~
Lhu= -8ic2 r:xh2 sin u r), (!) = (!) = - ih -ax '
(8.fi4)
where u satisfies the linear equation Lhu = 0. For the sake of sim-
plicity we take one term of the sum ~ of formula (8.37) corres-
±
ponding to the minus sign as a solution u of the linear equation.
We assume that the initial value u 0 E M n corresponds to a condition
n n n
l~~ J
-n
u 0a(x)(coskw)u0 dx= ~
-n -n
) a(x)g(x,ro)coskrodxdro.
(8.65)
First note that a more natural for the continuous situation and
a more general condition follows from condition (8.65). Denote
(f, rp) =
dcf r
n
-
J f (x) cp (x) dx.
-n
The following lemma is true.
Lemma 8.4. Let f (x, ro) E Coo be an even periodic function ro.
If 'Ph (x) corresponds to the limit distribution g (x, ro), then
2 1 n n
lim ( 1\Jh, f ( x, ro) 1\Jh) = ~ ~ f (x, ro) g (x, ro) dx dro.
h-+0
-n -n
INTRODUCTION TO OPERATIONAL CAIJCULUS 93
Proof. Expand the function I (x, ro) in a Fourier series in ro.

We have
00
f (x, (I))= ~ an (x) e-inw, (8.66)

n=-oo
where
n
an (x) = 2~ J f (x, (I)) einw d(l). (8.67)
-n
By the conditions of the ]emma it follows that for any n there is
an inequality
max., an ()1_,-C~;
x """= - , (8.68)
x nit
k > 0 is any integer, C k = const. The estimate can be easily proved

integrating equation (8.67) by parts.
We prove that
E~ I/(fh (X, (I)) - f (X, (I))) lJln (x)

I 2 1 2 1 II= 0,
where we have denoted by lk (x, ro) a partial sum of series (8.66)
fk (x, (I))= ~ an (x) einw.
In i:S:k
Note that the function lk (x, (I)) uniformly converges to I (x, (I))
as k-+ oo. We have for any smooth function u (x)
2 1
[! (x, (I))- In ( x,
2 1
J
-
(I)) u (x) =
n=oo n
= ;rr ~ [.\ u(y)e-inyayJeinxJf(x,nh)-f~t(x,nh)]=
n=-oo -:rt
oo n
= 2~ ~ [ J u (y) e-iny dy] ~ am (x) ei(mnh+nx).
n=-oo -n !ml>h
Finally, we obtain
-21 21 2 1
[t(x,(l))-!n (x,I(I))Ju(x)=( ~an (x)einw)u(x). (8.69)
JnJ>h
Estimate the nth term in sum (8.69). The inequality is obviously
true
II an ( ~) eino;u (x) I ~ max I an (x) I II u 11, X

94 OPERATJONAL METHODS
or by (8.68)
I an { ~) I
ein~u (x) ~ ~Z \1 u \\, (8. 70)
where k>O is an integer. We have by (8.69), (8.70)
II[!(~. ~)-ht(~.~)Ju(x)ll~ ~ +zcl\\u\\, Vl,

JnJ>k
!~~~~[! (~, ~)-ht (~, ~)Ju(x)II=O, as k-+oo.
Hence for any 8 > 0 there exists such a k0 , that
I{ ¢h, (! {:, ~)- fR { :, ~)) 'i'h ) I< 8 when k';;;-k0,
IJ I [f(x, w)-fk(x, w)]g(x, w)dxdwl<8 when

lt .lt
k';;;-k0 ,
-lt -lt
finally by (8.63) we have

2 1 lt lt
I(¢h, ik ( x, w) 'i'h)- J J!k (x, w) g (x, w) dxd(l) ~~c (k) 6",

-lt -lt
where c (k) = const, 8"-+ 0 as h-+ 0. Fix k > k 0 and let h-+ 0,
then for sufficiently small h the left-hand side of the last inequality
is not greater than 8. The lemma follows from this statement and
the previous inequalities.
Definition. Let for any finite function cp E Coo (Rn) the following
condition be true
lim\ cp(x)F~t(x)dx= f cp(x)r(x)dx, xERn,
h-+0 J J
where F" (x) E Coo is a one-parameter family of functions, h is suffi-
ciently small. Then we shall say that F" (x) tends to r (x) in the sense
of the theory of distributions and write F h (x) -+ r (x).
We find the limit of the right-hand side of equation (8.64) in
the sense of the theory of distributions. Denote F" = -4 (sin ~ u) 2 •
The following lemma is true.
Lemma 8.5.
Jt
Fh -"r 4 J sin 2 ~ g ( x-ct cos ~ , (I)) dw.

-lt
INTRODUCTLON TO OPERATIONAL CA!JCULUS 95
Proof. Denote by F x.... p and F p-+x the Fourier transforms

~ 00
(Fx .... pf)(p)= 2~) f(x)e-iPxdx; (Fp .... x/p)(x)= ~ /peivx.

-R p=-oo
It is obviously true, that

f (w) "'=
Fp .... xf (ph) F x.... p'i'·
We shall obtain the following equality with the help of this
formula
r 'i'tf
R R
~ 'i'd (w) 'i't dx = (w) "'2 dx,

-R -R
where ljl 1 , ljl 2 E C'XJ are periodic functions. Applying this equation,
we obtain the following equation
R ~ ~
) cp(x)Fh(x, t)dx= -4) cp(x) (Tsin ~ u0 rdx=

-R -R
R ~ ~
= 4 ) sin ~ [T*cp (x) T] sin ~ u0u0 dx (8.71)

-R
- 2i . ffi
he t Sill_
f or cp E Coo , u
Tu 0 (T : u 0 -+ u =e
= 2 a 0 ).
Apply Theorem 1.1 of Ch. III to the formula in the square bra-
ckets. We have cp0 = 1, cp;;: = 0, k;;?: 1 in (8.37) when c = canst
and
S 2 . (!)
=- ct sin 2 .
\Ve have
. w' . w
68 2 - Sill 2
Sill
~= -2ct----,,---
uw (J) - ( J )
. w' . w
~ 68 ' Sill 2 - Sill 2
X= X-~=
uW
X+ 2ct ---;-,---
w -w
Hence
J uer ax= 1.
ox
We have by Theorem 1.1 of Ch. III
T*cp (x) T = cp x + 2ct

( 2 Sill
. 2~)
. Z~ - Slll
1
w-w
3
t
•
Hence, by the commutation formula, equation (8. 71) is different

from the integral
4 J. H;+,tcos ~)sin'~ u,) udx 0
only by a term of order 0 (h).

The last integral tends to
Jt :rt
+ 4 ~·
-:rt-:rt
I cp ( x + ct cos ~ ) sin 2 ~ g (x, w) dx dw
as h -+ 0 by Lemma 8.4. Hence follows the statement of the lemma.

Next we shall prove the following theorem.
Theorem 8.3. Let a equal to ~ in equation (8.64). Then the solution
w of problem (8.64), satisfying the zero initial conditions, converges
in the distribution theory sense to a function
t :rt
wdx,t)=-2c J~ sin 2 ~ [g(x-c-rcos ~ +c(t-T),w)-

o -:rt
-g(x-CTCOS ~ -c(t-T),w)]dwdT. (8.72)
Proof. Introduce the following notation:

:rt
r(x,t)=-4 J sin
-:rt
2 ~ g(x--ctcos;, w)dw.
It is easy to prove that Lw1 (x, t) = c 2 r~ (x, t), where L = De

is a wave operator.
2 sin ; ')
fh=Fh-r(x, t); Vh=Lf. 1 ( h ih ;
w
Lh* =
def
h 2 -f)~
[)2
+ 4c 2 '-'in 2 -2c2 (x)·'
w
def [)2 [)2 ( 1 )

L* = fJt2 - fJx2 c2 x .
Denote by (L~)- 1 f and (L*)- 1 f solutions of the problems L'!,u =

= f (x, t) and L*u = f (x, t) satisfying the zero initial conditions
when t = T:
u I
t=T
=au
-
fJt
It=T
=0.
We shall use the following facts for the proof:

1. An analogue of Lemma 8.2 is true for the operators L'f, and L*.
In particular, if 'ljl (x, t) E Coo does not depend on h, then
h2 (L'f,)- 1 'ljl (x, t) - (L*)- 1 'ljl (x, t) = 0 (h).
One can verify this following the proof of Lemma 8.2.
2. The following equality is true
T ~ 2' . (J)
Slll')
J~
" I
h2 \p(x, t)Li/ h ~ fhdxdt=

iJ -n
=h 2
T n
.).)
[ 2i sin ~
h 2
J
(LJ; 1)*'1jl(x, t) fhcl:cdt, (8.73)
o -n
which can be proved with the help of the equality 'ljl = L'f, (LA)- 1 'ljl
and the integration by parts.
3. II F h II ~ c1h - 112, where c1 is a constant not depending on h.
By (1)-(3), the right-hand side of the equation (8.73) is different
from a function
T
J J :x (L*- '\jl(x, t))fh(x, t)clxdt

1
by 0 (h 112) and this function tends to zero as h--+- 0 by Lemma 8.5.

It follows from Lemma 8.2 that
2!. S. l l lw
2 )
lim£;; 1
h- u
(
1L r(x, t)= w;.
c~
The theorem is proved.

Note, that the convergence in Theorem 8.3 in the distribution
sense may be changed for the uniform convergence. But more esti-
mates are necessary to the effect. A more general theorem was
proved in Appendix 1 of the Russian edition.
Note. We have used only the fact, that the function w 1 (x, t) and
not g (x, ffi) is smooth when proving the theorem. The latter function
could be, generally speaking, a distribution and the theorem is true
for those t for which w1 (x, t) remains smooth.
Note that the most interesting results are produced, when g (x, ffi)
is close to the 8-function. (The actual form of u 0 is of no importance.)
We shall consider the data of computer calculations in the case,
when
e21Jl2(x)
g(x, ffi)= 4 l6(ffi+2x)+8(ffi-2x)],
where cp (x) is defined in the previous item.

7-01225
We shall use the most common potential of interaction between

atoms in a lattice leading to a system
~~= ~~ {f ( h- 1 +uhn~Un-1) - / ( h- 1 +u~1+1J--Un)} (8. 74)
f(z) = - ~ { : z!o}, h= 10- tE [0, 1] = [0, 10-

2 -
3, 10 s],
c2 = 1, n=O, + 1, ... , + 3000, uh lt=O = ecp (hn) cos {hn2},
~n lt=O = ceh- 1cp (k (n-;)) cos { h (n-; f}-

- ceh- 1cp ( h ( n + ; ) ) cos { h ( n + ; ) 2
} •
We have taken the same initial conditions as at the beginning

of item 7. The numerical solutions of the linear equation (see the
corresponding graph in Fig. 6) and of equation (8.74) calculated for
e = 10-1 are identical. The deformation of equilibrium points
~at t=0.2·70-10 s
0
-2at 1
(To-scm)
2a~
t =O.lf-·70-10 s
0 1
-2ac (10-5cm)
t =O.S ·70-10s
2ae
2
-2ac (2·70-{jcm}
t =0,8·10'10 s
2
(2·10-5cm)
Fig. 10
INTRODUC'riON TO OPZRATIONAL CALCULUS 99
of the numerical solution of equation (8. 74) and the solution obtained
by (8.72) are shown respectively in Figs. 10 and 11 fore = 2, a = 1.
The moments of time on the graphs are the same as in the linear
-!0
2at: ~ t =0.2·10 s
o~--- ~
}
-2ae {10!5cm) 2,
(2·70-"cm )
2.ae _ t=0J.f.·10-70 s
0~--------L--?~----L---~~~
2 3
-2ae (2·70-5 em) r:uo-5 cm)
2ae
0~------·--~--~------~--------~~
Fig. 11
case. The graph of the exact solution at the moment of generation
of the focal point is indicated in Fig. 12. The amplitude of vibratiolli'l
of the atoms of the lattice is blackened.
The newly found effect may be used to explain a number of phy-
sical phenomena. It is related to a well-known effect of the heat
7*
JOO OPERATIONAL METHODS
expansion of a crystal, i.e., the transformation of heat vibrations

of small amplitude (as if adding them) into a non-oscillating change
of a crystal by a huge, in comparison with the amplitude of the
heat vibrations, value. The obtained formula may be extended
to general equations containing a small quadratic term. A similar
(ro-5cm) (u/2·J0-5cm) (Ho-5 em)

or-------~-----~&~~2------~~J
2 (Z·!IJ-Scm)
-aac
Fig. 12
effect (so-called rectification) of fast-oscillating initial conditions,

is known in some branches of physics. Therefore we can say that
by the combination of these two effects there emerges a mighty
running wave of high energy concentration on the side of which
the crysal is pressed 7 while on the other it is expanding.
Sec. 9. The Concept of a Quasi-Inverse Operator and Formulation

of the Main Theorem
(1) The main problem. We have given examples of the appli-
cation of the method of ordered operators similar to those in Sec. 1,
where we gave examples of the solution of equations by Heaviside's
operational method.
On the other hand, as we have seen in Sec. 3, the solution of
systems of ordinary differential equations with constant coefficients
is reduced to the purely algebraic problem. It turns out that it
is possible to state and solve the problem, which, in essence, embra-
ces both partial differential equations with variable coefficients
and problems on the stability of difference schemes within the frame-
work of algebras with f.t-Structures; i.e., it is possible to include
the solution of all these problems into one algebraic theorem. Though
the proof of the theorem is completed only in the last chapter, we
shall try to elucidate its meaning and importance.
In the preceding examples the reduction of the given equation
to an integral equation of the Volterra type with a smooth non-
oscillating kernel was considered to be the solution of the problem.
IN'TRODUIC'ritON TO OPERATIONAL CALCULUS 101
Such reductions to integral equations (of the Volterra or Fredholm

type) with a smooth kernel are the main problem in the theory of
partial differential equations.
The concept of the quasi-inverse operator introduced below makes
it possible to extend the method of solution of differential equations
to the case of operators.
Definition. Let :£ be a module over an algebra Jl, with the given
rt-structure, A 11 A2 , • • • , An, BE X. An element f E .Jl; is called
quasi-inverse, if there exist such seqnences of elements % k E :£, ~?: k. E :£
that the products f% k and % i.f can be expressed in the form
(9.1)
where the functions Rk (x, a) and Rk_ (x, a) decrease when I x 1-+ oo
faster than I x 1-k; IRk I = 0 2 (! x 1-k), I Ri. I = Oz (I x 1-k).
We shall call the sequences {!l:k} right quasi-inverse and {!t'k_}-
left quasi-inverse.
Note. In the case when the algebra .JI; is an algebra of unbounded
operators (see Chapter I), this definition can be replaced by the
following one. An operator T E .f{; is quasi-inverse if there exist such
sequences {Xk} c .fl, {Xi,} c .JI: that TXk = 1 + St,, Xh_T =
= 1 + SJ;, where operators St,, S!, satisfy the conditions
n .
1/IJ Ajisf,= II :s;;c, '1:/ji: 2J ji = k, c = const.
i=1
In this case the quasi-inverse sequence gives the asymptotic solution

of the problem.
Everywhere below the elements /, % k• %h. themselves are also
(1
represented in the form f = f All ... , An, B
n n+ 1) , % k (1
= 2: k All ...
n n+ 1 ) ( n+ 1 2 1)
... , An, B , !/:h_ = !t'h_ A 1 , • . . , An, B .
Our main problem is to find quasi-inverse sequences for an ele-
1 n n+1)
ment f ( A1 , • • • , An, B . We shall solve it for the operators
A 1 , • . . , An, B and the symbol f (x, a) under some assumptions.
Before formulating these conditions, it is natural to clear up
the question of quasi-inverse elements in the trivial case of com-
mutative algebra. In this case it is possible, for example, to consider
operators xll ... , Xn and a, the symbols coinciding with the ope-
rators satisfying them. If an operator f = f (x1 , . . . , Xn, a) has
102 OPERATIONAL ME·THIQDS
zeroes when a = a0 and I x I---+ oo (for example, f = x~ - x; +

+ ix; sin 2 a), it is not quasi-inverse (since it is not possible to find
such a smooth function .2'(x1 , x 2 , a) that (x; - x; + ix; sin 2 a) X
X .2'(x11 x 2 , a) = 1 + R (x 11 x 2 , a), where R (x 11 x 2 , a)---+ 0 when
x~ +x;---+ oo, because the left-hand side of the equality becomes
zero when x1 = x 2 , a = 0). So it seems natural, at first glance, to
demand that the symbol f (x, a) should not have any zeroes when
I x I-+ oo. But this condition is too rigid: it would deprive us of
the main applications. Indeed, all the difficulties of the construction
of quasi-inverse sequences emerge when the symbol of the operator
f Ct, ... , An, nJ/)becomes zero when I x I-+ oo. In this respect
the case of the commutative operators is a special one. Let us con-
sider the example of a differential equation in order to comprehend
this sudden phenomenon.
In this example, our statement within the framework of com-
mutative algebras seems to be a paradox, since the problem surely
has no solution in the simplest case of a differential equation with
constant coefficients. It is clear that we would like to reduce the
given problem (ignoring certain details and idealizing conditions
somewhat) to an equation with constant coefficients. The crux of
the matter is not only in the fact that these equations have been
studied best of all, but also in the fact that the solution of equations
with constant coefficients serve as a model for equations with variable
coefficients. As we shall see from the main theorem, however, the
case of constant coefficients proves to be very special and degenerate.
For example, consider the symbol / 1 (y 1 , y 2 ) = y~ - y;,
/ 2 (y 11 y 2 , a) = y:- y;- icp (a 2 ) y~, where the function cp (a 2 ) ~ 0
equals zero when a 2 < T and is greater than zero when 2T +
6 ~
~ a 2 ~ 2T. Both symbols for a 2 < T coincide and equal zero
in the infinity when y 1 = y 2 • Let A 1 = x, A 2 = y, B = i :x,
hence
1 2
A 1 and A 2 commute. As we have noticed, the operator A~ -A;
corresponding to the symbol / 1 (y 11 y 2 ) is not quasi-inverse. But
1 2 ( :l ) 2
the operator A~ -A; - icp B 2 A; proves to be quasi-inverse,
which will follow from the main theorem. Suppose A 1 = i ~ ,
A2 = i :x, B = t. The operator corresponding to the symbol
()2 {)2
/1 (y11 y 2 ) will be the wave operator atz- axz (cf. Sec. 8). It is
not quasi-inverse. The addition to it of the term -icp (t2) 0~22 cor-
responds to the symbol f 2 (y 11 y 2 , a) and reflects the phenomenon
of the absorption of high-frequency waves for t2 > 2T.
INTRODUCTWN TO OPERATIONAL CAIJCULUS 103
Note. Let the joint spectrum (taking into consideration axiom

1 n n+1
f.t_~)
of the operators Au ... , An, B belong to a domain D c
c R11 X Mm and let the function P (x, a) E !1 equal 1 for x, a E 00
ED and zero outside some larger domain. Then, by the definition

1 n n+1)
of the joint spectrum, we have P ( Au ... , An, B = 1 and
( 1 n n+1)
consequently in this case we can substitute P A1, •.. , An, B
for the unity in the formulas (9.1), since
1 n n+1
supp(1-P(x, a))na(AP ... ,A 11 , B )=0.
Example. Consider the operators
. 0
A 2=~7iii' A a=X, A 4=y,
and let f (x1 , x 3 , x 4 ) be a polynomial in x1 and x 2 and x 3 , x 4 E

E 111 2 , i.e.,
( .
1
i) .
1 2 2)
i)
t ~ax' x, y
~--:;-----,
uy '
is a differential operator. The construction of the quasi-inverse
operator reduces the problem of the differential equation
f (i fx, i :Y, ~. ~) u(x, y)=F(x, y)

to the problem of the integral equation
[/ ( i }x ' i 0~ ' ~' ~ h gh ( i fx ' i L' ~' ~) fPii (x, Y) =
=[1+Rh (i fx, i :Y, ~. ~)]crh(x,y)=F(x,y),

where Rh (x11 x 2 , x 3 , x 4 ) tends to zero faster than ( I x1 12 +
+ 1 x 2 \2) -h/ 2 • It is not difficult to see (directly from the Fourier
transform of the pseudodifferential operator), that the last equality
is an integral equation of the Fredholm type of the second order
with a smooth rapidly decreasing kernel. In this case the solution
of the given problem can be represented in the form
u(x, y)=gk (i fx, i :Y, ~. ~) cph(x, y),

where cph (x, y) is the solution of the integral equation
fPii + RncPii = F.
104 OPEHATIONAL METHODS
Note, that the first term of the iteration equals F, and the second
one equals RkF. Thus, the second term is smooth by virtue of the
property of the operator Rk· Hence we may conclude that the term
gkF contains the non-smooth part of the solution of the initial
problem, i.e. the difference gkf.Pk - gkF is smooth.
(2) The rule for reducing the main problem. Let A 11 • • • , An•
8 EX, B be a vector operator, B E 111m. Suppose, that the spec-
1 3 2
trum of the pair of the operators A i> A i relative to A i is diagonal
and has a finite multiplicity, i.e., for sufficiently great r
1 3)r2
( A;-A; Ai=O, i, j=O, 1, 2, ... , n, (9.2)
( 1 :; ) 2
where A 0 = B E A!f m· Besides, suppose, that Ai - A i. A 1 =
(
1 3 )k 2
= -iAka. i> for such k that A;- A; A 1 =I= const·1.
We shall call the operators Au ... , An, B generative operators
and the algebra Jf!', generated by them, the Lie nilpotent algebra.
Note. It is easy to verify that conditions (9.2) are equivalent
to the statement that the commutator of order n of A; and Ai equals
zero. Thus we may say that sufficiently high commutators of the
elements Alt ... , An, B equal zero. Besides, for any k, l (1 ~
~ k, l ~ n) and any r (1 ~ r:::;::;, m) there exist the indices j (k, l),
s (k, r) such that
[Ak, Br] = - iAs (h, rl• [Ak, Az] = - iAi (k, !)·
Now we compare every operator A k to a partial differential operator
Lk (:. - i :x' i a~)

acting in the space Cz
(Rn X Jllr') (here a =
= (au ... , am) are local coordinates on Mm, (x11 • • • , Xn) E Rn).
Com,pare the vector operator B to an operator of multiplication
by a.
We shall define the operators Lk, k = 1, 2, ... , n with the
help of the equality
1 n n+ 1) def ( 1 n n+ 1 )
(Lkc.p) ( Ap ... , An, B =Ak [(p AI, ... , An, B ] (!.l.3)
(k = 1, 2, ... , n) for any symbol c.p (x, a) E C';; (Rn X M"'). We

shall call the operator Lk a representation of the operator Ak.
Consider first the construction of the operators Lk. Take operators
A 1 EM, B E M 1 and
A 2 = i [A 1 ,B], As= i [A 11 A 2 ], [A 2 , B] = 0,
[As, Bl = 0, [Au Asl = 0,
INTRODUCnON TO OPERATIONAL CAlJCULUS 105
Besides, let A 2 , A 3 E M. Then A 1 , A 2 , A 3 , B satisfy the conditions

given above.
The operator A 3 commutes with all the remaining operators and
the operator A 2 commutes with B. Therefore their images £ 3 , L 2
are easily constructed
Ls = Xs, L2 = X2.
Find the operator L 1 • Consider Acp {Au A .tfs, B). Expanding cp

1
4 6
2,
in the Taylor series in B - B (see Sec. 5, Sec. 6), we obtain

512 34 512 3 6
Atcp {At, A2 , Aa, B) =AtCf-' (At, A 2 , A 3 , B)+
5 { 4 s acp { 1 2 3 e)
+At B-B) Ta At, A 2 , A 3 , B ,
5 4 6 k
since it is given At { B- B) = 0 for k':;>2.
Thus
12 34 3 12 56
At[cp(At, A2 , A 3 , B)]=Atcp (At, A2 , A 3 , B)+
+ (
2 )
-iA 2 aa
acp ( 1 2
A 1, A 2 , A 3, B .
3 4)
We shall expand the first summand on the right-hand side in the

2 1, 3 ( 2 4 )
Taylor series in A 2 - A 2 • Considering A 1 A 2 - A2 "'= 0 when
k ;?::: 2, we obtain
123 4 3 14 56
At[rp (A 1 , A 2 , A 3 , B)]=Atcp (A 1 , A 2 , A3, B)+
3 ( 3 4 ) arp ( 1 4 5 6 )
+At Az-Az ax2 At,Az,Aa,B,+
acp
+ (- iA2 ) {fry: 1 2
(At, A
2 2,
3
A3, B .
4)
Hence we see that we can obtain equality (9.3) (fork = 1) by getting

L
t = X 1 - lXa
. a
Dxz -
.
lXz
a
aa .
In the same way, for any set Alt A 2 , • • • , An, B of generators of

the Lie nilpotent algebra there exists the representation Lk of the
operator Ak satisfying condition (9.3). Indeed, to substitute A"
n+2 ( 1
for the operator in the kth place in the operator A 2 ·cp Alt ..•
n n+i)
... , An, B we tug Ak with the help of the Taylor expansion
of cp computing a number of commutators [Ak, B], [[Ak, B], B], ...
first with B, then with An, An'-1! ... , etc. In the process many
commutators emerge: Ak with B and with An, An_1 , • • • , Ak_ 1 •
By definition these commutators are some operators from the set
(A 11 A 2 , • • • , An)· At the next stage we shall tug every of the
commutators to its place with the Taylor expansion. Here again
a set of commutators emerges, but they are all commutators from
A 1 , A 2 , • • • , An and we shall tug them to their place. It is clear,
the process results in obtaining large commutators which commute
with all A 1 , A 2 , • • • , An (since A 11 A 2 , • • • , An, B are the gene-
rators of the nilpotent algebra). These large commutators take the
indicated places automatically. At last it is worth noting that every
application of the Taylor expansion to cp produces a sum of several
derivatives of cp. The order of these derivatives cannot exceed n 0 ,
i.e., the maximal order of the commutator of A 1 , • • • , An- Therefore
1 n n+ 1 ) ( 1 n n+ 1 )
A~t[cp ( At, ... , An, B ]=cp~t A1, ... ,An, B ,
where fiJ~t (x, a) is the result of the application to the symbol cp

of a differential operator Lk ( x, i a~ , i of the order not grea- a:)
ter t 11an n 0 • L et T = ....,
L.J Cj, a Aait
H · ... · Aai"
ih . T am
I
j, a
~ (Xjl (Xjh
'A(T)=L.J Cj,a(Lii) · ... ·(Li~t) .
j, (X
If T E J!f', then for any symbol cp E Cz (Rn X Mm)

1 n n+ 1) ( 1 n n+ 1 )
T[cp ( At> ... , An, B ]=('A(T)cp) A1, ... , An, B . (9.4)
Indeed, take for example T = AkAi, then by (9.3) we get

1 n n+1 1 n n+i
A~tAi[cp (A 11 ••• , A,11 B )]=Ak[(Licp) (At, ... , An, B)]=
1 n n+i)
=(L"Licp) (A 1 , •• • , A 11 , B =
1 n n+i
=('A (A"Ai) cp) ( A1, ... , An, B ) .
Equality (9.4) for all elements T of the algebra J/1' is verified in

the similar way. Thus the operators Li are themselves the generators
of the Lie nilpotent algebra, which we shall denote by 11.
The constructed mapping 'A: ufl' -+ IT is called an ordered repre-
sentation of the nilpotent algebra J/('.
It follows from (9.4) that the operators Lk, k = 1, 2, ... , n; a

satisfy the same commutating conditions as the operators Ak, B.
Particularly the operators L 11 L 2 , • • • , Ln, a are generators of
the nilpotent algebra D. Hence if P (x, a) is a polynomial in x
with coefficients depending on a, then
1 n n+ 1 ) ( 1 n n+ 1 )
[P ( A1, .. . , An, B ]·[G A1, .. . , An, B ]=
1 n n+1)
=¢(A 1 , ••• ,An, B , (9.5)
1 n n+1)
where ¢ (x, a) = P ( L 11 • • • , Ln, a G (x, a) for any G (x, a) E
E c:;;. We can ask, whether this equality holds for P (x, a) E r!/ 00
and G(x, a) E Cz.

In all probability the answer is negative if we limit ourselves
to axioms ftdte (without topological axioms). To give a positive
answer to the question we shall formulate two additional algebraic
axioms, which are useful and, in addition, permit to obtain an
explicit formula for any function of non-commuting operators
via a composition of functions of commuting operators.
Consider symbols f (x, a, t) depending on the parameters t E
E Rk (k is not fixed); all the derivatives with respect to the para-
meters belong to r!/ 00
•
Axiom (ft 7) (the parameter axiom). Let f (x, a, t) E Cz, t E Rk

verify the equation
1 n n+1
[/(At, ... , An, B, t)]=O
for all t E Rk. Then for any function <D (t) E (/ 00
(Rk) and symbols
h(x, a, t)=CD (i 8~1 , .. ., i 8~k) f(x, a, t)
the condition is satisfied
1 n n+1
[h(A 1 , ••• ,An, B ,t)]=O.
If the condition of Axiom (ft 7 ) is satisfied for all t =1= 0, it is valid
for t = 0 as well.
Axiom (ft 8) (the uniqueness axiom). Let the following conditions
be verified
1 n n+1
(~t~)[t(At, ... ,An, B, 0)]=0,
1 n n+ 1 n+ 2 1 n n+ 1
(~~) [[/i] ( A1, ... , An, B , t)] = iA f ( A1, ... , An, B , t)]
for A EM and symbols f (x, a, t) E r!foo, t E R.
108 OPERATIONAL :IIETHODS
... ,
for all tER.
Problem. Prove the following formula for an operalor
1
f ( A1, ... , An)
1
n )
f ( A1, · · · , An =
[ a ,
f ( L. 7ft; . a )
. . . , L-~- X
din
X (e-iAnt1e-iAn-1t2 ... e-iA1tn)] •

1=0
We shall prove in Chapter II, that the positive answer to the

question posed above (see (9.5)) follows from these axioms.
The main point is to construct such a function GN (x, a) for
the given function P (x, a) that condition (9.5) is verified when
1jJ (x, a) = 1 RN (x, a), +
where the function R N (x, a) = 0 :e (I x 1-N) for any N. It means
that we should solve the problem (9.5); i.e., the main problem (9.1)
is reduced to the following: find the sequence g .v (x, a) such that
1 n n+1
t(Lt, ... , Ln, a )gN(x, a)=1+R,v(x, a), (Q.G)
where R N = 0 :;e (I x 1-N) decreases faster than Ix 1-N when I x I -+-

oo.
-+-
Suppose we have constructed the function
1/JN (a, YJ, x, t) E coz uwa X R211 ); t E [0, oo)
(for some numbers Pk ~ 1, 1-1 ~ 1, T, rJo and a finite function p (11)
which equals unity when I rJ I ~ 1] 0) decreasing for t = T when
x-+- oo faster than A-N, where
A= ( ~ (xk)2/p"+1)1/2
k=1
and which satisfies the following conditions:
[iA~ :t -f (L1, · · ., Ln, n~ 1 )]lb (a, i a~,~. t) ·.

•
= BN
(
a,
.
L ax~ ' 1
X'
I
tI ' (\L7)
l!JN (a, YJ, x, 0) = p (YJ),

INTRODUCT.IiON TO OPERATIONAL CALCULUS 100
where B N (a, 1], x, t) decreases faster than A - 71• as x ._ oo uni-

formly in t. Here and below F · j denotes F cp (x) = f rp (x),
V cp EC';.
We shall integrate equation (9.7) with respect to t from 0 to T
and apply both sides of equality (9. 7) to the unit function; taking
into consideration that p ( i !)
1 = 1 and denoting
'
¢N(a,x,t)=¢N
(a,~a:;,x,t
.3 1 )
A
1 -J.l,
1,
'
BN(a,x,t)=BN
(a,~a:;,x,t
.3 1 ) 1 -J.l,
A 1, (9.8)
T
gN (x, a)=+ i ~ lJliv (a, x, t) dt,
0
we obtain the following equation:

1 n n+1
I (L 1, .•. , Ln, a ) gN(x, a)=
T
= 1-'ljl]v (a, x, T)-i 1B!v (a, x, t) dt def 1 +RN (a, x),
0
where R N decreases faster than A-N as x ._ oo by the initial con-

ditions and therefore we obtain a solution of the main problem.
Example. Consider the equation
- ~· fJu + ~· ay-
Tt au yu = 0 , u It=O = I (y ) , (9.9)
We shall use this equation to demonstrate the general method of

constructing solutions up to the functions of the class CN in y and
decreasing faster than I y 1-N as I y 1- oo (N > 0 is a given
number). Let A 1 , A 2 , B be the operators
(9.10)
The operators are evidently generators of the Lie nilpotent algebra.

1 2 3) 4
Calculate their representation. Consider an operator cp ( A 17 A 2 , B A 1 •
We must transform in such a way that the operator A 1 should act
first in the obtained operator; to this end we must pass the operator
110 OPERATIONAL METHOD<S
2 3
A 1 through the operators A 2 , B 2 • We have
4 (1 2 3)
Atcp At, A 2 , B = A 1cp At, A2 , B +
4 (1 2 5)
+At
4
(B-B
3 5 ) a<jl
aat
( 1
At, A 2 ,
2
B5 ) =
4 ( 1 2 5) . a<jl ( 1 2 3)
= Atcp A 1 , A 2 , B - ~ aat A11 A 2 , B . (9.11)
4 2
Commutate At and A 2
4 12 54 15 6
Atcp (At, A 2, B)= A cp ( A 1 11 A 2, B)+
4 ( 2 5 ) a<jl ( 1 5 6)
+At A2 -A2 ax2 At, A2, B =
1 ( 1 2 3) . a<jl ( 1 2 3)
=Atcp At,A 2,B -~ax 2 At,A 2,B. (9 .12)
By Eq. (9.11) and (9.12) we have

L . a . a
t=Xt-~--~-.
aal ax2
The representation £ 2 of the operator A 2 evidently has the form
L2 = X2.
Now we shall observe the following general fact. Let us find such
1 n n+1
an operator g (At, ... , An, B, t) that
. a ( 1 n n+1 )
-~Tt[g]+[f][g]=RN At, ... , An, B, t ,[g]Jt=o=i,
(9.13)
where the symbol RN (x, a, t) satisfies an estimate RN = Oz (I i 1-N)
uniformly with respect to a, t. The symbol g (x, a, t) may be deter-
mined by the equation
g(x, ex, t)=[g' (~, -i !x, th1 a, (x), (9.14)
where 1 (x) is the unit function of x, and g' ( ~. -i :x, a, t) is

a solution of the problem (cf. (9.7)):
2
-~Ttg
• a +f £j ,I (
•••
n n+1
,£,11 a, t g -RN, x,
) .!_' ' ( 1
-~
• a
ax'a,t ) , (9.1:•)
1
g' (x, 1'j, a, 0) = p(f!), R1v(x,1'j,cx,t) =Oz(l x 1-N).

INTRODUCT]ON T•O OPERATIONAL CAL·CULUS 111
The proof of the statement follows from (9.5). Hence, to construct

a solution of problem (9.9) we must solve the problem
1 2
-i aa'!' -(L 1 +L 2)'¥ · 0, '~'='~'(x,-i:x, a,t). (9.16)
'¥ lt=o = p ( - i :x) , p ('l'J) ECO' (Rn)

p('l'J)=1 as J'l'JJ~e.
We shall search the solution of (9.16) in the form
ur_ 1A8
.1 . 2 a,t
(1x, -1Bx' {I )
(
1 . a
2 )
r-e cp x, -La;, a, t , (9.17)
where
S (x, 1'], a, 0) = 0, cp (x, 1'], a, 0) = p (1']), A= I x J.
On substituting the function '¥ defined by (9.17) into (9.16) we
obtain by the commutation formulas (we drop the arguments
1 2
x,- i :x ,a, t of cp, S):
Aas
1 . acp
atcp-l--
at
[ 1 as
Xj-A&1] 1
1 as
-+A aa-
a
cp+l-cp-
ax
J . 2
- [
1 1 as
x 2 - A - cp-l--l--=0.
iJ'1]2
J . acp iJ'1]1
. acp
iJ'1]2
Let x 1A - 1 = w 1 , x 2 A- 1 = w2 • Using the symbols and equating to
zero the coefficients at the powers of A we obtain
as as as as
{ -at - ffi1 a11 1
+--- -
aa 1 ffi2 a11 2
= 0' +-
(9.18)
iJcp - i
iJt 1'] 2!p
+ acp +~-O
U'l] 1 a112 - • (9.19)
Equations (9.18), (9.19) have solutions of the form

s= ( + (t)1 ffi2) t
h1 ~
i('ll 2 -t)2
cp= e-2- - - -2-p (1'] 1 -t, 1'] 2 -t),
thus
. 'll~ . (tJ2- 1)2
'f(x,a,'l'],t) =ei(xt+x2lt/2-t-2-p('l'] 1 -t, 1'] 2 -t).
To construct the solution of the initial problem we must calcu-
late the value of '¥ (~. -i!, a, t) 1 (x).

'112 OPERATIONAL METHODS
(
"'¥X,
1
-i:x' 2 ry \
~.t)(1(x))=e2
i ( .
-tax2
{) )2
-2
i ( . {)
-tiJx2 -t
)2
X
x p ( - i -ax8-
1
-t, -i - 8--t) ei <x1+x2>t=
8x 2
it2
= ei (x1+x2) t e-2- = G (x, a, t).
The solution of (9. 9) has the form

1
1 2 3 )
- {)
t- 2 i t2 . t it2
tY+-
(
U (y, t)=G A 1, A2 , B, t f=e 0YeiYte 2 f =e 2 f(y+t)
Thus an application of the operational method enables us to solve

the problem exactly.
Now let a (y) be a positive smooth function uniformly bounded
with its derivatives in the region y E R\ 0 < C1 ~a (y) ~ C 2 •
Consider the equation
azu iJ2u
iJt2 - ayz - yza (y) u= 0 (9.20)
with the initial conditions

U /t=O = U1 (y), u; /t=O = 0, (9.21)
where the function u1 (y) belongs to the space If 8 (R) for some s.
We shall investigate the same problem for (9.20) with initial con-
ditions as for Eq. (9.9).
Let A 1 , A 2 , B be the operators defined by equation (9.10). Their
ordered representation was calculated in the previous example.
Rewrite (9.20) in the form
!~~ -[A~+ a (B 2) A;] u =0.

Similarly to the previous example it is sufficient for the solution of
1 2
the given problem to construct an operator g'rv (x, -i ()/iJx, a, t)
satisfying the conditions
i)g'
g'rv ( x, 'l'], a, 0 ) = p (11), a~ ( x, 'l'], a, 0) = 0. (9.22)
IN'rRODUCTiiON TO OPERATIONAL GAIJCULUS 113
We shall search for the solution of (9.22) in the form
gN
, ( 1
X,
_ .
l 7fX ,
~ a, t - e
) _ ils u. - lx ,a, t) ,
i
)<
1 2 \
X 1fN
(
X, - i :X , a, t j ,
1\ = ( :r 12 +x 2
2 )1/2 • (9.23)
On substituting the function gN defined by (9.23) into (9.22) we
obtain by the commutation formulas
4
,1 (1 '
etAS x, -'
f)
ox , a, t
4 )
{
1 3
A2 [ - (E._
at
- 1'
iA-1 _!_ ) 2+
at
1' 1' 1'
as
+ ( X1 A_1+ - as + - .A_ 1 a- l'A-i -
a )2 ,
. aYJ 1 aa 1- l - aa ax 2 1
+a(~)(x2A- 1 +-:~2 r]wN} · Oz(lxi-N). (9.24)
The arguments x, - i a, t of S and x, - i

1'
a, of '¥ N :x, 2 1
:x ,
2
are omitted.
Apply the K-formula to the term in the braces in (9.24). We
obtain that the symbol of the operator
[ - ( -as
3 ' 1 a)2 1 ( 1 as+ - -
x1A-' 1+ - a
asi A -' 1-.-- 1
at- i A - -at + aYJ aa 1 aa 1
a )2
1' 1' as ) z
- iA-t axz +a (a) x2A-1+ a'l'Jz ] 1fN
(
is
at + (ro 1+~+ as )2 +a (a) (ro 2+ a'as
{ _ (!!__)2 )2 +
a'l'j 1 aa 1 l'J2
+(-iA-t)[-2 as acp_ ~zs +2a(a) (ro2+ as )-a-+
_ at at iJtZ a'l'J2 a'l'J2
azs
+a(a) ,.,-+2'11
( as
2 ro1 +-,-+;:;----
as
as ) +2 ( rot+-,-+
ufjiJ U'l'jj uCXj U'l'jj
+ dcxz
as ) ( a a ) a2s azs azs
aal + O'l'Jt + O'l'Jf + 2 a'lt ocx 1 + ocx[
J+
N
+ ~ (- iA-t)k Rn} '¥ N +A-N+I RN'¥ N• (9.25)
h=2
where ro; = x;A -1, i = 1, 2; Rn are differential operators of the
order k with smooth coefficients, R N is the remainder term in the
commutation formula.
Note that the successive application of the K-formula enables
us to determine an explicit form for all the operators.
8-01225
Equation (9.22) is an easy corollary of the equations
(9.26)
(9.27)
where R 1 is the operator in the square brackets on the right-hand

side of (9.25) and 0 (I x 1-N) is a smooth function decreasing as
I x 1-N with its derivatives as I x 1-+ oo.
Equation (9.26) is a Hamilton-Jacobi equation with the initial
condition S lt=o = 0.
We shall search for the function 'I' N in the form of the expansion
N-1
'I'N= ~ (-iA-t)kqtk·
k=O
On substituting this expansion into (9.27) and equating to zero

the coefficients of the powers A -R, k = 0, ... , N we obtain the
2ystem
R 1W0 =0,
R 1W1 +R 2 W0 =0
Rt ifN-1 + .·. + RN-t'Yo = 0,

lfo lt=O = p (rJ); qfk lt=O = 0, k > 0. (9.28)
Equations (9.26), (9.28) are solved similarly to (8.32) of Sec. 8.

Denote their solutions by s±' W(i=' ... ' if~ -1· Let
2
1 2 N-1 .1 (1 . 0 )
g'rv ( x, - . !_ ) - .2.. "' ""' tAS± X, _, DX' a., t '
~ ax , a, t - 2 L.J .LJ e X
± k=O
2
. 1 -1 k ~± ( 1 . {) )
X (-~A ) qrk X, -~ax' a, t '
then the asymptotic solution of (9 .20)-(9.21) is a functioA
1 2 2 )
where gN ( Alt A 2, B, t is an operator with the symbol
gN(x, a, t)=[giv u. -i a:, thia, (x).

INTRODUICTliON TO OPERATIONAL CALCULUS 115
Sometimes it happens to be useful to reduce the solution of the

main problem to an exact solution of a simpler equation. We have
seen, that the asymptotic of the equations of crystal vibrations might
not be found generally only with the help of bicharacteristics. We
are able to reduce merely the solution of the asymptotic problem to
a simpler one; i.e., an exact solution of the wave equation (see Sec. 8).
This section deals with the construction of a quasi-inverse operator
( 1 · n n+1)
of an operator f A 11 ••• , A,n B .
( 1 n n+1)
Choose a certain operator r A 1 , . . . , An, B E X as a standanl
one.
Definition 9.1'. A right quasi-inverse sequence for an operator
1 n n+1)
f ( A 1, ••• , An, B is such a sequence
I 1 ( 1 n n+1) 2 n+1 n+2)
x11 \[r At. ... , A,, .8 ], At, .. ,, An, B E:t,
that the product fx 11 can be put in the form
2
(~,
n+1 n+2)
fxk=i+Rk At> ... , An, B ,
where the junction R 11 (x0 , x 1 , ••• , x;, x;+t, ... , Xn, a) decreases
faster than [i~l xJ J-h/ 2
as I x 1-+ oo, i.e.,
R11 = 0z ( c~1 xJ) _, ).

12
The left quasi-inverse sequence for the operator f (Au ... , An,
n+1)
B is defined along the same lines (cf. 9.1).
We shall investigate the method of reduction .of the main probl~m
to a solution of the Cauchy problem for a function of differential
operators.
Let Gh ( ~, ~' -i ~, -i ~) be a differential operator with

partial derivatives defined in the space (Rn X Mm). Denote Cz
this operator by Gh. We define G, by the following equation:
(1 n n+
(Ghcp) A11 ... , An, B =[cp A11 ... , A,, .8
1) (1 n n+ 1) ]A~t,
Further, we define Gn+t as follows:
n
. . . , An, B
n+ 1) JB .
r16 OPERATIONAL METHODS
The operators Gk are generators of the Lie nilpotent algebra which

will be denoted by II'. The proof is the same as in Sec. 9. The map-
ping 'A' : J1f'-+- II' is called the (right) ordered representation of the
algebra J1f'.
An analogue of (9.5) is true
( 1 n n+ 1) 1 n n+ 1)
[<p At. ... An, B ][P(A1, ... , An, .B ]=
n+1 2 1 ) ( 1 n n+1)
= P ( G1, ... , G,., Gn+l rp (x, a) Ah ... , An, B , (9.29)
where <p (x, ex) E Cz,

P (x, ex) is a polynomial.
Equation (9.29) is verified for any P (x, a) E r!/ (Rn X llr'), 00
 ... , An, B, r At. ... , An, B EX,
f(x, et)Ec!foo, r(x, a)EcY 00 .
Consider a symbol g (:r 0 , x, ex) Erffoo (Rn+l X Mm) and calculate

the product
1 n n+ 1) ( 1 ( 1 n n+1)
[/ ( A1, . . . An, B ] [g [r AI> ... ' An, B ],
2 n+1 n+1)
A~t ... , An, B ].
On applying (9.5) we get

1 n n+ 1) ( 1 ( 1 n n+1)
[/ ( A 1 , ••• , An, B ][g [r A 1, ... , An . .B ],
n+1 n+2)
... , An, B ]=
1 n n+l)
=f ( Lt. ... , Ln, ex g(x 0 , x, ex)+
1 ( 1 n n+ 1) 2 n+1 n+2)
+[g ( [r At, ... , An .• B ], A1, ... ,An, B ]X
X
1
([r (A~o ... , An, B
n n+ 1) ]-[r (A1 17 ... , n
An, B
n+1) )
] ,
where Xo= 1 ( 1 n n+1)] X= ( 2 n+1) et= n+2.

[r A1, ... , An, B . ' A1, ... , An ' B
IN'TRODUCTLON TO OPERATIONAL CALCULUS 117
On applying (9.29) we get

1 ( 1 n n+1) 2 n+1 n+2)
[g ( [rAt, ... , A,p B ], At, ... ,An, B ]X
I 1 n n+1)
X [r \Ail ... , An, B ]=
= (r(nJt1, •.• , G~, Gn+1) g(xo, x, a)) X

1 n n+ 1 ) 2 n+ 1 n+ 2)
x ( [r ( At, ... , An, B ], At, ... , A 11 , B .
In the same way
1 n n+ 1 ) ( (' 1 · n n+ 1 ) 2
[/ ( At, ... , An, B ] [g [r At, ... , An, B ], At,
n+1 n+2)
... , An, B ]=
1 ( 1
( [r n n+1) 2 n+1 n+2)
= W At, ... , An, B ], At, ... ,A 11 , B ,
where
1 n n+1) (n+1 2 1 )
W(x 0 , x, a)= [ f ( Lp .. . , Ln, a -r Gt, .. . , Gn, Gn+i +
+ xoJg (x 0, x, a).
Thus the main problem is reduced to (9.6) with 1he left-hand

side having the form
[ f ( L1 ,
1 ••• ,
n
Ln, a
n+ 1)
-r
( n+ l
G1 , ••• ,
2 1 )
Gn, Gn+1 +x0 g(x 0 , x, a).
J
The remaining part of the proof of the reduction to the Cauchy
problem is preserved unchanged. The operator
1 n n+ 1) ( n+ 1 2 1 )
f ( L 1, ••• , L,, a , - r G1 , ••• ,G,, Gn+i +xo
1 n n+l)
is called the Hamiltonian of the operator f ( A1, ••• , An, B with
( 1 n n+1 ')
respect to the standard operator r At, .. ·~ An, B .
Example. Let C'h (R") be the space of functions satisfying the
conditions
II (1-7i211)" cp (x, h) IIL2 < Cs,
where s = 0, 1, ... , Cs are constants depending on the function
(p but not on h as h-+ 0.
Let <p (x, h) ECh' (Rn). We shall write

<p (x, h)= 0 c"" (hr,) if <p (x, h) ECh',
h
II h-f,<p (x, h) ll£ 2 ~const for h-+ +0.
Let <p (x, h)= 0 c"" (hr,). Let Q be a subset of the space R 2n of
h
the variables x and p and let Q have the following property:
the equation
f 2
( x, .
-~ax
b ) <p (x, h)=Och' (h
Hi
)
is true for any function f (x, p) E ,ff"" (Rn X Rn), such that
Suppf(x, P)nQ= 0.
The set Q is called a support of singular points of <p (x, h).
Consider an equation of the form
2 1
-ih~-L-H at ~ (y ' -ih _!__)
ay "'¥ = 0 '
{ (9.30)
"'¥ lt=O = 'l'o (y, h),
where H (y, p) E cff"" (Rn X Rn) is a real function, "'¥ 0 (y, h) E

E C'h (Rn). Suppose that for any "'¥ 0 (y, h) E C'h (Rn) a solution of
problem (9.30) does exist and belongs to Ch' (Rn). Let Q 0 be a sup-
port of singular points "'¥ 0 (y, h). Let Qt be a support of singular
points of the solution "'¥ (y, t, h) of problem (9.30).
Theorem. The set Qt is included in an image of the set Q 0 by a cano-

nical transformation gir with the Hamiltonian function H (q, p ).
Proof. We shall use the concept of the standard operator in the
Hamiltonian formalism to prove the theorem.
Take the operators
A1 = - ih -}!.- ,
vYt
... , An= - ih--}--- ,
vYn
B = y.
These operators are evidently generators of the Lie nilpotent

algebra. It is easy to verify that the left-ordered representation of
these operators is constituted by the operators
Li = Xt- ih a!i ' ... ' Ln = X - ih o!n , Lo =a.

IN"l'RODUCT]ON TO OPERATIONAL CAIJCULUS 119
The right-ordered representation has the form

G1=Xb .. . , Gn=Xn,
G(i) "h jj
n = ai- l -;:;--- , i= 1, ... , n.
uXi
Take the operator
r U, - ih :Y ) = H U, - ih :Y ) = r.
as the standard one.

To construct a solution of (9.30) we may construct an operator
1 2 2 3 )
g ( r, A 1, ..• , An, B, t, h satisfying the condition
. 8g ( 1
-zhTt r, Ail ... , An, B, t, h
2
+ 2 3 )
21 1) (1 2
+[H (B, Ail ... , An ][g r, Ail
2 3
... , An, B, t,
1 2 2 3 ) (1 1 2)
g ( r, A~> ... , An, B, 0, h = g0 Ail ... , An, B ,
where g 0 (x, a) E ,ffoo is an arbitrary function.
By the previous construction we need a solution of the Cauchy
problem
. ag ( z 1 1 )
-LhTt(x 0 , x, a, t, h)+[H a, Lil ... ,Ln-
1 2 2
-H (Gn+l! Gil ... , Gn) x 0 ] g (x 0 , x, a, t, h)= 0, +
g (x0 , x, a, 0, h)= g0 (x, a) (9.31)
to construct the symbol g.
If g 0 (x, a)::::::: 1 then a solution of (9.31) is the function
_2_ txo
g(x~,x,a,t,h)=e h
Hence we obtain the following formula for the solution of problem

(9.30):
W(y, t, h)=exp{[- ~ HU, -ih }y h}wo(Y, h).

Now to prove the theorem it is sufficient to prove the following
statement: if F' (q, p) E ,ffoo (Rn X Rn) is a function with
Supp F' (q, p) n Q-r =f= 0 (-r is any) then
F' (~, -ih :y) W(y, t, h)=0 0 'h(hB+ 1),
whence 1¥0 (y, h)= 0 coo (hB). (9.32)

h
120 OPERATIONAL METHOD-S
Consider the canonical transformation g}1 corresponding to the

Hamiltonian function H (q, p). By definition, g-}1 (Qt) == Q 0 • Let
(g}1 )* be an automorphism generated by g}1 of the ring of smooth
function defined in R 2 n.
Let F (q, p) E oY"" (Rn X Rn) be such a function that Supp F (q, p) n
n Q't = 0. Let _f (q, p) = (g}J)* F (q, p). It is obvious that
f (q, p) E oY"" (Rn X Rn), Supp f (q, p) n Qo = 0.
Hence the estimate is true
f (~, -ih a:) 1Y0 (y, h)=Ocr;(hfl+l).
Let G (x 0 , x, a, t, h) be a solution of problem (9.31) with initial

values
g 0 (x, a) = f (x, a). (9.33)
We shall search the function g (x 0 , x, a, t, h) in the form
it:xo i N
GN (x 0 , x, a, t, h)= e --,-+y
' ' S(x, a, t) 'I.' (
..:...~
'h
-l )
h
cph ( x, a, t ) . (9. 3~g)
hc=O
To satisfy the initial value conditions let

S (x, a, 0) = 0, cp 0 (x, a, 0) = f (x, a)
cp" (x, a, 0) = 0, k ~ 1, (9.35)
then substituting the function GN of (9.34) into (9.31) and equating
to zero coefficients of the powers h 0 , hi, ... , hN-i, we obtain a
system of equations
~~· +II(a, x+ :~ )-H(a+ ~~, x)=O (0.36)

Drro) ··-
i!a
arro) = 0 (9.37)
i!x
Rtcpk + R2cp1<-1 + ... + R" rtrro = 0, k= 1, 2, ... , N-1, (H.38)

where
R= :t + <Hp (a, x+ :~ ) ,
1 8~ ) -
- ( Hq (a+ ~~ , x) , 8~ ) ,
R", 1 <k< N are differential operators of order not higher than

k with smooth coefficients.
INTRODUCTl'ON '1"0 OPERATIONAL GAIJCULUS 121
Equation (9 .36) is a Hamilton-Jacobi equation. Solving (9 .36)

and taking into account the initial condition (9.35) we obtain
S (x, a, t) =0
Hence we may rewrite the operator R 1 in the form
R1 = :t +(Hp(a, x), 8~ )-(Hq(a, x), :X)=!,

where ~ is a Hamiltonian vector field corresponding to the one-
parameter family g}r of canonical transformations of phase space.
Therefore by (9.35) the function cp 0 (x, a, t) is a solution of (9.37)
cp 0 (:r, a, t) = (gi{ 1)* f (x, a).
In particular, when t = .-
cp0 (x, a, ..-) = F (x, a).

When k = 2, ... , N- 1 equations (9.38) have the form
{ :t q>lt + Fh (CllJH, ... , cro) = 0,

cph lt=o = 0,
where the alleged function Fh (crh-l• ... , cp 0) depends on the func-
tions Cllh-H .•. , cp 0 and its derivatives. Hence (9.38) is easy to solve.
On substituting the function GN (x 0 , x, a, t, h) into (9.31) the
right-hand side is evidently obtained in the form hNR N (x, a, t, h),
where the function R N (x, a, t, h) is of the space r!/ (Rn X Rn) 00
for any t and hE [0, 1]. Hence the asymptotic solution of problem
(9.31) with initial condition (9.33) has to be of the form
-it
GN (x 0 , x, a, t, h)= eh xo {(gi{ 1)* f (x, a)+ hg 1 (x, a, t, h)},
(0.39)
where the function gi (x, a, t, h) belongs to the space rff oo (Rn X R")
for any t and hE [0, 1].
Since the function GN is a solution of (9.31) with initial condition
(9.33), the following equation is true:
. aGN ( 1
-zhTt r, A 11
2
•• • ,
2 .'l
A,, B, t, h
)
+
( 2 1 1) (1 2 2 3 . )
+[H B, At. ... , A, ][GN r, At. ... , An, B, t, h ]=O(hN),
( 1 2 2 3 ) ( 2 1 1 )
GN r, A 1, ••• , An, B, 0, h =f B, At, ... , A, .
122 OPERATI·ONAL M·ETHODS
Hence the function '¥' (y, t, h)

1 2 2 3 )
'I':V (y, t, h)= [GN ( r, Ail ... , An, B, t, h ] '¥0 (y, h)
is a solution of the problem
1 2 1
- ih awN
at +H ( y, - ih _!____ ) '¥ :V = 0
ay c:;;,
'I'f.rit=o=f (~, -ih ~) '¥ 0 (y, h)=0ch(hP>+ 1 ).
Hence the estimate
'I' f., (y' t, h)= 0 coo (hP,+ 1)
h
is valid for any t.
From the last equation and (9.39) we have
2 1 2 1
((gii 1)* f) (y, -ih :Y) exp { ~it [H (y, -ih :Y )] }X
X 'I'o (y, h)= 0 coo (h!H 1).
Taking t = -r we finally get
2 1
[F ( y, - ih :y )]'I' (y, t, h)= 0 ch (hf3+ 1 ), Q.E.D.
Example. Consider the pseudodifferen tial operator
A= -i~+Hat (y2 ' -i.l_)

ay '
where H (y, p) E coo (Rn X (Rn"'- {0}))
is a real function positively
homogeneous in p of the first order. Let u 1 (y) be a solution of the
problem
Aut= 0, Ut lt=O = U 0, u 0 EHs.
Following Hormander we call the set*
WF (v) c: T*Rn
WF (v) = n char (B)
BuEC 00
a wave front of the function v E H s (Rn), where B = B -i

is a pseudodifferen tial operator with a symbol B (x, p) E S 1 (Rn X Rn).
(~, ~)
In the phase space T*Rn char (B) is a set such that B (y, p) = 0
when (y, p) E char (B).
* For the notations see L. Hiirmander. Acta Math., 127, 1-2 (1971)
INTRODUICT]ON '1.'0 OPERATIONAL CAllCULUS 123
Let gk be the canonical transformation of the phase space T*Rn

with the Hamiltonian function H.
Theorem. There is an inclusion
WF (u 1) c gkWF (u 0 ).
Proof. Following the previous example we use the concept of the
standard operator to prove the theorem.
It is evidently sufficient to prove that the relation
[B (~, -i! hu 0 EC""=*[((gJi 1)*B) u, -i :y hutEC..

(9.40)
is valid for any operator B ( - i ~, fx ) with a symbol of

S 1 (Rn X Rn).
The left representation of 1.he operators
At= - i8-
Yl
° , ... , An= - i -8Yn8 , B=y
has the form
and the right representation of the same operators has the form
Gt=Xt, ... ,Gn=Xn,
G . a i=1, ... , n.
n+t=a·-l--
' OXi ,
Let
r(A,.s)=H(A., s).
Consider the problem (cf. pr. Example)
' 0
[ -i_}_+H(~
at '
x-ij_j\-H(a-il_
aa ax '
~)...Lx ]x
X '¥ (x 0 , x, a, t) = 0,
'I' (x 0 , x, a, 0) = B (a, x), (9.41)
where B (a, x) E S 1 (Rn X Rn), B (a, x) = 0 when I x I < 1.
We shall seek for an asymptotic solution of problem (9.41) in the
form
N
'I'N=e-ixot~ <Jln(a, x, t), (9.42)
h=O
124 OPERATl'ONAL METHODS
where the functions <pk (a, x, t) satisfy the estimates
where '\'• ~ are indices and Caflk are constants. To satisfy the initial
conditions of problem (9.41) let
<p 0 (a, x, 0) = B (a, x),

<pk (a, x, 0) = 0, k ~ 1. (9.43)
On substituting the function '¥ N (x 0 , x, a, t) of form (9.42) into

(9.41) and equating to zero the terms belonging to the spaces
S-" (Rn X Rn), k = 0, -1, ... we obtain
d(j)o =0 (9.44)
dt '
d(j)t i I Q
{2\ 1 · (2 ) Q1 \
iJa 1+
1
(itl iJa , Hpp a, x
~ \}
b ' H xx ( a, 2) ax
+ 2i \I ax I <vo = 0 X
(9.45)
where Rk, 1 ~ k ~ N are differential operators of order not higher

than k with coefficients smooth in the region x =I= 0.
From (9.44) we get
<p 0 (a, x, t) = (g-k)* B (a, x).
It is easy to see that the function <p 0 (a, x, t) belongs to S 1 (Rn X R")
for any t. Hence the solutions of (9.45) <pk belong to the spaces
S-"+ 1 (Rn X Rn) for any t.
By construction the function '¥ N (x 0 , x, a, t) satisfies the condition
[ -i_!_+H(~
iJt
'
'
1
x-i-a- - I f
iJa
) ( 1
a-i-iJ-
ox '
X'¥s(x 0 , x, a, t)=fN(x, a, t),
'¥ N (x 0 , x, a, 0) = B (a, x),
where IN (x, a, t) E s-N (R" X Rn) for any t

( 1 2
We infer from (9.5) and (9.29) that the operator 'V N r, A,
2 3 )
... , An, B, t satisfies the condition
. a'I'N ( 1 2
-l ---;Jt r, A17 ••• ,
2 3
An, B, t
)
+ [H ( B,2 1)
A ] X
( 1 2 2 a ) (1 1 2 )
X ['VN r, A 17 .•• , An, B, t ]=fN A 17 .•• , An, B, t
wN U, A~, ... ,.t, B, o) =B (~, -i }u).

Hence the function
vN(Y, t)=[lfN U, At, ... , An, B, thuo(Y)

-i a;~ u, ! )
+H -i vN=RN(Y, t)EHs+N(R~),
(9.46)
VN't=o=[B(~, -i! huo(y).
By (9.42) the function vN (y, t) has the form
vN(y, t)=((gi/)*B)u, - i !) exp{ -it[H(~, - i ! h}x

xu0 +x (~; -i }
uy
, t) ut(y),
1
(9.47)
x(a, x, t)ES 0 (RnxRn).
Since N >0 is any, we infer from Eqs. (9.46), (9.47) that
((gi/)* B) u, - :y ) i ut(y) Ecoo
if B (~, -i :Y) u 0 (y)EC"",

Q.E.D.
Now we shall consider a still more general case.· Let fi =
= fi
(A1 17 ••• ,
n
An, B
n+ 1) EM. Consider an operator
126 OPERATrGNAL METHODS
In the same way we obtain that the symbol of the product

( 1 2 k ) ( 2 3 n+ 1 n+ 2 1)
[F ft, fz, .. . , fk ][g At. A2, ... , An, B, r ]
is equal to
( 1 ( 1 n n+ 1) ( n+ 1 2 1 \
F [ft L 1, ••• ,Ln, a -r G11 ••• ,Gn, Gn+tl+x 0 ],
2 ( 1 n n+ 1) ( n+ 1 2 1 )
[/2 L 1, ••• , Ln, a - r G 11 ••• , Gn, Gn+t + .r 0],
( 1 n n+ 1) ( n+ 1 2 1 )
k
... ,[fk L 11 ... ,Ln, a - r G1 , ••• ,Gn, Gn+t +
+ .:r 0] ) g (x 11 . . . , Xn, a, x 0 )
We shall call the operator in the right-hand side applied to the
function g the Hamiltonian of the operator P.
Now let r = 0, g (x11 • •• , Xn, a, x 0 ) 1. Then we obtain a formula ==
of the reduction of a composite func.tion to a simple function of the
same ordered operators. Indeed, to define the symbol '¢ (x, a) of the
operator
1P (.41, ... , ln,nB 1

) =F (/11 ... , j,J
we obtain the formula .
'¢ (x, a)= F [/t (1 (L1 1, . . . ,

n
Ln, a
n+ 1) ],
k ( 1 n n+1) )
· · ·, [ fk Lb .. ·, Ln, a ] 1, (9.48)
where the composite function of differential operators in the right-
hand side is applied to the unit element belonging to the space of
symbols.
3. Bicharacteristics and absorption conditions. In the last examples
we came across the concepts of bicharacteristics and characteristics.
We have seen that in those examples a non-smooth and rapidly
oscillating part of the solution is defined by solutions of equations
with partial derivatives of the first order which we have called the
equation of Hamilton-Jacobi (the characteristics equation) and the
system of ordinary differential equations of Hamilton (the bicharac-
teristics equation).
We have considered an example of the equations of vibrations of
a crystal lattice over a circle or, which is just the same, the equations
with periodicity conditions. The corresponding solutions of the
system of equations of bicharacteristics (8.36) X (x0 , t), P (x0 , t)
were such that X (x 0 , t) could be considered as coordinates of a point
moving in a circle and P (x 0 , t)-in a line. The pair X (x0 , t),
INTRODUCTI:ON TO OPERATIONAL CAIJCULUS 127
P (x 0 , t) as a whole defined a trajectory of the point on the cylinder,

i.e., a product of a line by a circle. This cylinder, called a phase
space, is denoted by p, x, where p is a coordinate on a line and x is
a coordinate on a circle (or a q-coordinate). Henceforth, we shall
consider its property more thoroughly and the property of the system
of Hamilton equations defining a family of curves in this phase space.
It is clear that for any n-dimensional torus Ar there exists a phase
space T* Mn of dimension 2n, such that x-coordinates belong to Mn
and p-coordinates belong to Rn.
We have seen that the equation of Hamilton-Jacobi as well as the
system of Hamilton is defined by a function H (p, x), which is called
a Hamiltonian function. This function has the form
H=+2 c~~) sin[~ (1+p)]
in the case of a crystal and depends on the parameter I ffi I ~ n. This
parameter is a symbol of the operator of differentiation and sub-
stantially influences the property of the solution.
In the general case of a function of ordered generators of the Lie
nilpotent algebra the symbol is a function of m + n coordinates,
where m is a dimension of the manifold Mm, i.e., the number of
components of a vector operator B and n is a number of operators
A 1 , A 2 , • • • , An. Thus, the domain of the definition of the symbol is
a manifold Mn X Rm = Mtm.
Let T* Mf+m be the corresponding phase space. We want to dis-
cuss the problem, how to find a quasi-inverse operator: We have seen
from previous examples that the sought quasi-inverse operator is
connected with the system of bicharacteristics and, therefore, with
the Hamiltonian function. Thus it is natural to ask, how to define
the Hamiltonian function corresponding to the symbol f (Ah ...
n n+l)
... , An, B . We want to define bicharacteristics of this system.
We consider the function f. On substituting the representations
L 11 • • • , Ln, a of the operators A 1 , • • • , An, B for its arguments,
1 n n+l)
we obtain the Hamiltonian f ( L 1 , • • • , Ln, a .
Then we make a certain leading terrp. of the obtained formula with
respect to the variables x1 , • • . , Xn, ia: . This leading term defines
the Hamiltonian function.
We shall introduce the definition of the leading term of a function.
(1) Let f (y) E ,J"" (Rs) and let there be such numbers p1 , . • . , Ps ~ 1
and r > 0, that
f (Ap 1Yt. • · •' Ap 8 Ys) = Arf (y), 'VA.> 0;
then the function is a p-quasi-homogeneous function of degree r.
128 OPERATIONAL MgTHtQD>S
(2) The function a (y) is called subordinate to the function f (y), if
__a_lhl_a...:..;:(y-'-)-,--~~c (Vy21P; + 1y-Hhl_

ay h1 hs
1 . . . ay 8 •
N
Let F (y) = 2J
0
f; (y) + a (y), where f; (y) are p-quasi-homogeneous
(p = p1 , • • • , Ps) functions of orders r;, r; ?:: ri+ 1 + 6 for some
6 > 0 and a (y) is subordinate to fo (y). Under these conditions
F (y) is called asymptotically p-quasi-homogeneous (p = p1 , • • . , p.),
N
fo (y) is called the leading term of the function F (y) and 2J /; (y)
()
is called an essential part of the function F (y).

For the time being tak~ X; = Y;, i ~ n, i uah .}!-- = Yn+h• -i uX -%- = 'll·
Let the obtained function be asymptotically p-quasi-homogeneous in y
and Pn+i = 1, 0 < i ~ m (the variables Y), a are taken as parame-
ters; Y) belongs to a small neighborhood of zero in Rn, a E Mm).
The leading term of the function will be called the Hamiltonian
function corresponding to the given Hamiltonian, and will be denoted
by :rt (y, Y), a).
Thus the construction of the Hamiltonian function of an operator
A E .Jl; consists of 4 steps: (1) (f.L- 1) the choice of A 1 , •• • ,An, BE X,
2 n+1 n+2 1
r = r (x, a), f (x, x 0 , a) E r!foo such that f (At. ... , An, B, r) =A;
(2) (A -+ L) the construction of the Hamiltonian by the ordered
representation;
(3) (p) the pointing out of the leading term :rt (y, YJ) of the Hamil-
tonian;
(4) ~ (p, q) the determining of the Hamiltonian arguments in the
function :rt (y, YJ).
Thus
1-t-1 ( 2 n+1 n+2 1) A-+L
A--7 j A1o .. . , An, B, r ---7
-+ p, q
-+ :rt (y, Y), a) ---7 H (p, q, co).
The operations (f.L - 1 ) and (p) are defi.ned not uniquely generally and
do not exist occasionally for a given A E .fl. Now we pass over to the
last step of the indicated construction.
For the sake of definiteness let p; be equal to a unity for the fi.rst s
arguments, s ~ n, Pn+i = 1 for 0 < i ~ k.
INTRODUCTION TO OPERATIONAL CAI.!CULUS 129
Take new variables

Yt=COt-Pm+b ···•Ys=COs-Pm+B; YsH=(OsH• ···• Yn=COn;
1/n+i = Pt• · · · • Yn+k = Pk• Yn+k+i = · · · = Yn+m = 0;
111 = qm+i• ... '11n = qm+n; a1 = qj, ... 'am= qm.
Denote by
n(y, 'Y], a)=Q/t(p, q, co), H=ReQ/8, H=lmQJt.
Besides this, denote by JJ8 0 (p, q, co) the essential part of the symbol
f (LH ... , Ln, a). Assume that Im JJ8 0 ~ 0 in this item.
Let Q e be the manifold in the space of arguments p, q, co defined
by the conditions
n
2] (qmi-i? < 8
i=i
Define the bicharacteristics of the operator f ( A 1,

1
••• ,
n n+i)
An, B
as a solution of the Hamiltonian system
dqi an
1ft= Dpj (p, q, co), qi 1-r=O = q~,
dpi an
([(=- aqi (p, q, co), Pih=o=P~. j=1, 2, ... ,:s+m,
(p 0 , q0 , co) EQe. (9.49)
In the following examples we denote sometimes the arguments of
a symbol by Yu ... , Yn• a to avoid ambiguity.
Examples. Turn to the construction of the Hamiltonian function
and consider an example of a polynomial in the operators - i :x,
x = xh ... , Xn with coefficients dependent on x. In this case the
second commutator is zero and the representation of the operator
Ak = - i-}!-- has the form Yk- i--}!.-
Ur:f.Ji
and Bk = xk = ak. The
uXh
polynomial
az
- "V a-' (x)-
ax~
(9.50)
LI
'
has the form 2] a;
i
(B) ( A; ) 2
• Its representation has the form
~ ah (;) (Yk- i a!h ) 2 • (9.51)

h
9-01225
130 OPERATFONAL METHODS
The Hamiltonian function (for Pi = 1, Vi) has the form

n
H(p, q, ffi}= 2} ah(qh ... , qn}(ffih-Pn+Pn+h)2. (9.52)
1!.=1
Consider also the Hamiltonian system defined by the Hamilto-

nian function
(9.53)
where q= qt, · · ., qn, P = Pt• · · ·, Pn·

A solution of the Hamiltonian system
.
Pn+i =0
.
Pi=Hqi (9.54)
q1=-Hpi j=1, ... ,n

n
satisfying the initial conditions q (0) = q0 , p (0) = 0, 2} ffi~ = 1,
k=1
H (q 0 , ffin) = 0 is defined as bicharacteristics of operator (9.51).
Note, that these bicharacteristics can be defined by a solution of
the system
.
q=Htm
.
p= -Htq
qh (0) = qOht Ph (0) = (J)h·
Thus the definition given above is reduced to the definition of hi-
characteristics of the wave equation, given in Sec. 8.
Now consider an operator
n-1
a2 . a
- ~ ai (x) --;;"'2
ux.
+ lbn (x) -,-.
uXn
(9.55)
i=1 '
In this case the Hamiltonian function equals
n-1
H (p, q, ffi} = 2} ah (qt, ·. ·, q,.) {ffin- P11.
1!.=1
+ PR.+n)-
- bn (qu · · ·' qn} ffin·
Denote
n-1
HJ(p, 'YJ, a)= 2} ak(q)p~+bn(q)ffin.
1!.=1
INTRODUCTilON TO OPERATIONAL CALCULUS 131
The solution of the problem

• iJHt
P= ----aq, P (0) =Po'
• aH
q=--t, q(O)=qo. (9.56)
iJp2
n-1 n-1
Ih=1
~ an (qo) ro:- bn (qo) illn I< B, ~ ro: +I illn I= 1
h=1
is defined as bicharacteristics of operator (9.55).

Now we shall consider the Schrodinger operator
_-ih- 1 :t- 2 ~ a:: +h-2v(x)=Z, m=const, xER. (9.57)
Take
. a
A 1=-~Tt• A · a A 1
B2=X.
2=-~ax' 3=h'
Then
Z= +A 1A3 + 2~ A~+v(B) A~.
The representation of the operators has the form
8 +Yt•
Lt= - i8- 8 +Y2.
L2= - i8- L3=Y3·
a1 a2
Thus, the representation of Z has the form
X= +Y3 ( - i a! 1 +Yt) + 2~ ( -i a: 2 +Y2) 2 +yiv(o:2)·

The Hamiltonian function (for Pi= 1, Vi) has the form
H = {ro3+ P6) (rot+ Pt + Pn) + 2~ (ro2 + P2 + Ps?+
+ (ro3+ P6) 2 v (q2)·
Reduce the equations of bicharacteristics to the system
• p
E=O, t=1, p= -v' (q), q= +-,
m
with the inital conditions and the Hamiltonian function
p2
q (0) = .iQ_
())3
' p (0) = ~
())3
' Ht(q, p) = 2m +v (q)
with the help of a substituiQP of the form
P2+w2 =p, .!J.L=t, ..!f1..=q.
wa w3 w3
We have obtained the equations of the classical mechanical problem,
corresponding to the quantum problem, which is defined by the
Schrodinger operator (9.57).
9*
132 OPERATIONAL ME-THOD<S
Problem. Consider a differential-difference operator, defining the

equation (8.4). Taking
A3=h, ct=X
obtain the solutions of system (8.36) as bicharacteristics.

Definition. We shall say that the absorption conditions are fulfilled
if there exist the constants T > 0, e > 0 and the number ,.;' =
= ,.;' (q 0 , w, p 0 ), 0 < ,.;' < T such that
(1) when 0 :::;;; ,.; ~ ,.;', there exists a solution
q (qo, w, po, ,.;); p (qo, w, po, ,.;)
of problem (9 .49) of the class Coo,
(2) when 0 :::;;; T ~ ,.;' and (q 0, w, p 0 ) E Q e the function
-fi (p (qo, w, po, ,.;); q (qo, w, po, ,.;); w)
is non-negative, and when ,.; = ,.;' the function is strictly positive.
A few words regarding this condition. A condition similar to this
one was first strictly formulated by Zommerfeld for the needs of
electrodynamics and is now called Zommerfeld's radiation conditions
at infinity. It enabled Zommerfeld to solve difficult problems in the
mathematical theory of diffraction. Physicists have come to the
conclusion that this problem (the problem of an absolutely black
body) cannot be solved by putting only boundary conditions without
additional conditions on the coefficients of the equation itself and its
form. This problem is especially often dealt with in modern physical
problems (for example, in the theory of laser resonators). However,
analysis of physical problems shows that the non-smooth part of a
solution is more often absorbed, since the non-smooth part is defined
by higher Fourier harmonics and precisely these higher Fourier
harmonics are absorbed in many problems. For example, the walls
of a house absorb light and let radio waves through. What are the
mathematical devices modelling the phenomenon? It seems likely
that the only way is to introduce a strong absorption potential,
which is equal to zero inside a room and small for low frequences.
The absorption condition introduced above is in some sense a
substitution for the missing condition of the absolutely black body
and in many problems it is quite adequate. From this point of view
the ideal degenerate case is the one when short waves do not fade.
This is the case of constant coefficients.
The main theorem (the simplest form). Let A 11 ••• , An, B be
J n n+1 \
(
generators of a nilpotent Lie algebra. An operator f A 1 , . . • , An, B)
is quasi-inverse, if its Hamiltonian is such that the absorption condition
is satisfied.
INTRODUCT]QN TO OPERATIONAL CAIJCULUS 133
This theorem is proved in the last chapter and at the same time
(and this is the main thing) the construction of quasi-inverse sequence
is given there.
Note 1. Let M 11 = Mk X Rn-k, k ::::;;; n, where 1'lifk is a torus,
then in the simplest form the main theorem remains true, if there
is a function P (B) of a vector operator B with a compact support
equal to 1 in a domain Q in the right-hand side of (9.1) instead of the
unity. Note that when M 11 is compact, the unity is a function of B
with a compact support. In this case it is necessary that q? Esupp P (a)
for i > n in conditions of absorption.
Example 1. Consider the equation
-i~+i~=1.
at ay
Let - i =A~> - i =
:c fx
= A 2 , t = B 10 y = B 2 , B = (B 10 B 2),
then we obtain the equation
(AI - A 2) u = 1.
We shall construct a quasi-inverse sequence for the operator AI- A 2
within the framework of the introduced definitions.
The representations of the operators AI, A 2 are
L . {)
! =x~-~-
aat '
Hence the Hamiltonian of the operator AI- A 2 has the form
f (L 1, Lz) = ( X t - i a! 1 ) - ( Xz- i a! 2 ) •
Thus
f (LI (y), L2 (y)) = (YI + Ya) - (y2 + Y4).
Let PI = P2 = p 3 = p4 = 1, then
n (y, t], a) = n (y) = f (L (y)).
The absorption condition is not satisfied in this case, since
Im n (y) = 0. It looks like the operator AI - A 2 has no quasi-
inverse. Indeed, it has not. The operator AI- A 2 , however, can be
modified in a way that the obtained operator will have a quasi-
inverse and its symbol will coincide with the symbol A 1 - A 2
in a closed domain d c R1 X Rt.
Consider the operator
1 1 2)
f (At, Az, B =A 1 -A 2 -i<p B [V1-A:-A;],
(2)1
where <p ( alt a 2 ) ~0 is a smooth bounded function equal to zero in

a closed domain 1:!. and strictly positive outside !:!.. We shall not dis-
cuss the form of !:!..
We shall prove somewhat later that an addition of the operator
( 2) 1
- i<p B [V1- A~- Ai] to the operator A 1
. - A 2 provides the
fulfilment of the absorption conditions.
The Hamiltonian of the operator I C.t, .t, B) has the form
I(Lt. Lz, ~)=(xt-i~)-(x 2 -i~)-
thus
- ~<p ( ~) rv ( Xt- i a:t )
2
+ (Xz- i a: 2 )
2
+ 1],
I (LI(y), La (y), a)= (Yt + Ya)- (Yz + Y4)-
-i<p (at, a2) V(Yt + Ya) 2 + (Yz+Y4) 2 + 1.
Let Pt = P2 = p3 = Pt.= 1. Then
n (y, a)= (Yt +Ya) -(Yz+ Y4)-
2 +-:--:-(Y-z--:+-Y-4)"'2_,.+--,.1.
- i<p (at, a2)'V (.,...Y-t+-,---Y....,-3)"
Therefore the Hamiltonian function of the operator
1 1 2)
I (At, A B 2, has the form
~(p, q, w)=(Wt-Pt+P3)-(wz-P2+P4)-
- icp (qa, q4) V (rot- Pt + P3) 2 + (w2- Pz + P4) 2 •

The manifold Qe (cf. p. 129) is determined by the equations
w~+w;=1, Pt=Pt=P3=P"'=0, (q3, q4)ER2
q: + q; < e, I&'8 (p, q) I< e. (9.58)
Consider the bicharacteristics of the operator I ( Ah A2 , B

t 1 2)
(9.59)
aRe Q!t
""'Cit = - aq i
The solutions of (9.59) are the functions
Pi=O, j=1, 2, 3, 4;
qi=q~. i=1,2;
qa=q~+T, q~,.=q~-T.
INTRODUCTDON T·O OPERATIONAL CA!.JCULUS 135
Consider the function

ImQ/8(p(qo, 1:), q(qo, T))= -cp(q~+'t, q~-T)(ffi~+ffi;) 112 •
(9.60)
From (9.60) we infer that the absorption conditions will be verified
if the trajectory q3 = q~ +T, q4 = q~ - T of Eqs. (9.59) intersects
the boundary of the domain 11 for any (q~, q~) E 11 and T ~ T', where
T' is a constant not depending on (q~, q~). Therefore, in the space
R~3q. we may take any domain containing no direct lines parallel
to the line q3 = -q4 as the domain. For example we may take the
stripe
11 = {(qu q2) I 0 ::::;;; ql ::::;;; Q, q2 E R1} (9.61)
as the domain /1. Note that in this example a condition stronger
than the absorption condition is verified. -Indeed, the inequality
Im JJ8 (p (q 0 , T'), q (q0 , T')) <0
is valid when (q 0 , p 0) E Qle, where the manifold Q 18 is defined by
the equations (cf. (9.58))
(()~ + ((); = 1, P1 = P2 = P3 = p, = 0, (q 3 , q4 ) E R2
q~ + q~ < e,
i.e., all the bicharacteristics are absorbed.
We shall construct the symbol of a quasi-inverse operator for an
( 1 1 2)
operator j A 11 A 2 , B and, at the same time, explain the meaning
of the absorption conditions. Take the stripe defined by conditions
(9.61) as the set /1. Then the function cp (a 11 a 2) can be chosen not
depending on a 2 • Let cp (al> a 2) = v (a 1 ) and consider the operator
g Ct, A B) =At-A -icp (BJ[Vt-A~-A;].

2, 2 (9.62)
By the reduction rule for the construction of a quasi-inverse

operator we must ftnd a function satisfying the conditions
( [ -i
a
a:r+g (L1 L1, a2) ] 'i'N (1x,
2
11 -i ax~ ,
· BN ( ~. - i L, a, T)
'i'N ( ~. - i I, a, 0) = p (- i a~ ) (9.63)
l BN(x, 'Y), a, T)=Oz(lxi-N).

There is T > 0 such that the estimate

'IJlN (x, 'YJ, a, T) = 0 2 (I x l-111 )
is true when -r = T and M > 0 is any.
Suppose that the function 'ljJ N can be put in the form
'IJlN ( ~. - i !, a,
2
't) =
_
-e
.1 ( 1
- 1AS x. -1
. iJ
iiX" , a, ,; ) 0
UN
{
I1
,X, .~
-lax' a, 't/'l (9.64)
where A = (x~ x~)l/2. +
The function 'ljJ N satisfies the initial conditions of problem (9.63) if
S (x, 'Y], a, 0) = 0 (9.65)
9N (x, 'Y], a, 0) = p ('YJ). (9.66)
On substituting 'ljJ N in (9.63) by the commutation rules we obtain
.1 (1x,
e1AS -1
4
.iJ 4)31'
ax, a, t [A aS _ i ....!!._ + Xt +
f}o; f}o;
as
1 ' as . a ' as 1 ' as 1 1
+A' --A
aat - - l --x 2+A--A - -
81'Jt aat 81'] 2 aa2
- i -8- - iv
fJa2
(~t ) [ (xt +A _!§__X_!§___
oat Ol'Jt
i _!§__) 2 +
oat
' as
+ ( x2 - A -+A
' as1 . a- ) 2 +1 J1;2 ]9N=0.
1
01']2 aa2 - - laa2
On ordering the operators by the K-formula we obtain
' as
1 a ' as ' as a 1 1
[A--i
ao: -+x
ao: 1 +A--A - - i - -
aat 81'Jt aat
- (x2 -A_!§__+A _!§__i - 8 - ) -iv (~t) x (9.67)
01']2 aa2 aa2
' as ' as 1 a )2 ( ' as 1 1
X [ {xt+A--A--i-
aat 81'Jt aat + x2-A-+
81']2
+A' aazas
1 . a )2 J1;2 { as
-l aa2 + 1 ]ElN=A a:r+rotl+ aat-
as 1
as as
----ro2-------w(at) as . [ ( rot+-----
as as ) 2
81'Jt 81']2 aa2 aat a,lt +
::2- ::2 )r + ~ ( -iA-t r+ RJel\.

N-1
+ ( 2+ (!)
2 12 1
X
k=O
2 2
X (
1
x, - i:x , a, t) + RN( x, - i! , x, a, t) , 3 1
RN(y, 11, x, a, t)=Oz((!xl+!yJ)-N).

INTRODUCTI:ON TO OPERATIONAL CAlJCULUS 137
where Rn, k = 0, ... , N - 1 are differential operators of order not

higher than k with coefficients being smooth functions of x, a, 't"
and the derivatives of the functions S, co 1 = x1A -I, co 2 = x 2 A - 1.
Let
S 1 = ReS, S 2 = Im S,
aS aS aS uS DS
---
- - c o 2 - - - -a1']2
- + COj + - - - -a1']1
0T aa 1 aa2
as
- iv (at) [( COt+ aa 1 -
os
a11 1
)2 + (C0 2 + as
aa2 -
as
a'12
)2]1;2 •
(9.68)
Equation (9.68) is the Hamilton-Jacobi equation with dissipation
with the Hamiltonian function r1J8 of the operator f 2, (.A.t A B)=
QJ8 (p, q) = (col- PI + P3) - (co2- P2 + P4) - icp (q3, q4) X
X [(col- P1 + Pa) 2 + (co2- P2 + P4) 2l 112 ·
Applying the formulas of Ch. IV, Eq. (9.68) and equating to zero
the remaining terms in the parenthesis in the right-hand side of
(9.67) we obtain the transport equation with dissipation for the
function eN· This equation can be solved by the method of Sec. 6
of Ch. IV. Let
't
D=- ~ ImH(p(qo, (9.69)

0
We shall prove in Ch. IV that the solution of (9.20) satisfies the
dissipation inequality
Im S (x, a, 1:) ~ yD,
where y is a constant.
When ,; = 2Q + 1 cter T (cf. (9.61)) the inequality D ~ 6 > 0
follows from (9.69) and the absorption conditions. Therefore, the
function
tiJN (x, 'Y], a, ,;) = eiAS(x, '1], a, <leN (x, 'YJ, a, ,;)
satisfies the estimate

lt!JN(x, 'Y], a, T)l~conste-iitlxJ.
Thus the functions tiJN satisfy (i), (ii) and the operator with the
symbol
1 (1x,
. Tf ( [ eiAS
xN (x, a ) = r J
- i:.,2
x
a, 't
)
x
()
1 2
X eN (X, - i :X ' a, 't)] ·1 (x)) d,;
( 1 1 2) by the reduction
is a quasi-inverse of the operator f A 1 , A 2 , B
rule (cf. p. 108).
Note 2. Introduce the construction of the first term of the quasi-
inverse sequence in the simplest case of the main theorem in a par-
ticular case. The general case is not within our scope for the present
time, since it demands a number of topological concepts which will
be introduced only in Chapter IV. In order to orient the reader, how-
ever, we shall provide the construction in a particular case and con-
sider one example. In particular, we shall assume:
(1) The Hamiltonian ii is a homogeneous function in p of degree k
and does not depend on ~ and rJ.
1 n n+1)
(2) The opera tor f ( L 17 • • • , Ln, a does not change if all indices
over the operators (and inside every L; as well) are written in the
inverse order (i.e., n + 1-+ 1, n-+ 2, ... , 1-+ n +
1, i.e., the
( 1 n
operator f L 1, ••• , Ln, a
n+1)
is "formally" self-adjoint.
These conditions are not vital and introduced merely for the sake
of simplicity. Now let us introduce the principal conditions without
which we shall get rather complicated formulas. These two conditions
will invoke the non-existence of the focal points, and we saw on the
-example of Sec. 8 their principal importance for the asymptotics
·of solution.
(3) The Jacobian ddq =I= 0 for 0 ~ t
qo
<
T and the Jacobian
J (q 0 , t) = d (~ + z) =I= 0 in the same segment. Here the function

%
z (a, t) is defined by a linear system of equations (the "germ" equa-
tions)
~=iHp(q, p)+rfJ8pq(q, p)z+rfJ8pp(q, p)w,
(9. 70)
(v= -iHq(g, p)-rfJ8qq(g, p)z-rfJ8qp(g, p)w
and the conditions
zlt=o=Wit=o=0, Z=Z 1, ••• , Zn, W=Wt, .. . , Wnt
where q = g (g 0 , t), p = p (g 0 , t), g = q1 , • • • , gn, P = Pu • • ., Pn
are the solutions of the Hamiltonian system
iJH iJH
g=75P, P=-aq-• glt=o=q 0 , Plt=o=O. (9.71)
Introduce the construction of the symbol g1 (y, a), i.e., of the first
t
term of the quasi-inverse sequence. Denote by D = - I if dt
0
I~TRODUCTDON TO OPERATIONAL CAI!CULUS 139
(all integrals in t are defined as integrals along trajectories), for

example,
~
;
~ H dt =
def t
i _
H (p (qo, t), q (qo, t)) dt;
0 0
w)-
(9. 72)
where the symbol (a, b) denotes the sum 2} aibi. Denote by

t
CD(qo, t)=~t(qo, t)+) Zdt. (9.73)
0
The absorption condition involves an important inequality (the

so-called inequality of dissipation)
Im <1> ;;;;=: c1D - c2D 3fz-e (9.74)
which is true for some c1 > 0, c2 > 0, and e > 0. Let M be the
lower boundary of such constants c2 for which the dissipation ine-
quality is true. Then
gi(Yh ···•Yn, CG)=
T
- i f p (qO) ei{(qO, t)+tMD3, 2-e(qo, t)) lqO=qO(a, yt) dt, (9 •75)
- ~ Vll(qO, t)l
where q0 (a, y, t) is a solution of the equation qi (q 0 , t) = ait m + n ;;;;=:
> n; qi (q 0 , t) = Yi> i ~ n.
;;;::: i
Example. Consider an operator

1
A 1 -icp B A 2
( 2) 1 (9.76)
with the symbol
Y1 - icp (a) Y2•
where cp (a) ;;;;:: 0 equal szero for I a I < ~ and is strictly positive for
a> T/2. If the operators A 1 , A 2 , B commute then the quasi-inverse
element obviously does not exist (cf. p. 102).
Take
140 OPERATIONAL METffiOD<S
Let the spectrum of A 2 lie on the half-axis y 2 > {) > 0. In this case
it is sufficient that the symbol should belong to C';; only for y 2 > 6.
For example, let A 1 = - i :x, B = x, A 2 = k, where k is a
parameter{) < k < oo and let F (x, k) F (x) be a functiOn w1th=
def • •
a support in Q, such that F (x, k) P (x) = F (x, k). The function

P (a) was defined earlier in the remark to the main theorem. Consider
the equation for u (x, k):
(A1 c
1 - icp B
( 2) A1 ) u (x, k)
2 = P (B) F (x, k),
I F (x, k) I < const. (9.77)
We can tell by formula (9.77), that (see Sec. 6 of Ch. V) an asymp-
totic solution of the problem is obtained by a "quasi-inverse" sequence
applied to F (x), which satisfies the equation within the accuracy
to any function, tending to zero ask~ oo faster than k-N for any N.
We shall verify this by the given example, using formula (9.75).
The Hamiltonian of the operator (9.76) is of the form
y1 - i 8~ -icp(ct)IYzl· (9.78)
The Hamiltonian function, corresponding to the Hamiltonian (9.78),

is of the form cf!t = ffi 1 - P1 + P 3 - icp (ql) I ffi2 - Pz + P4 I,
ffi 2 > 0. The systems of Hamilton and of the germ are the following
in this case
. . .
qz =--= q3 = q4 = 0; q (0) = qo,
p=O; p(O)=O.
(9. 79)
Z 1 =Z 3 =0, Zz= Z4 = - i(p (q 1); Z (0) = 0
. . .
Wt = icp' (q 1) ffi 2 , w2 = W3 = w4 = 0; w (0) = 0.
Calculate the values entering formula (9.75). We have
t
<D(q 0 , ffi, t)= -ffi 1t+iffi 2 Jcp(q 10 +•)d't. (9.80)
0
We see that z (q 0 , t) ==
0 by the germ system. Thus 1-t (q 0 , t) = 0.
We shall calculate the dissipation D.
By the definition we have
t t
D (q 0 , t) = - Jfi dt = ffi 2 .\ cp (q 0 + t) dt. (9.81)
0 0
lNTRODUCTIION TO OPERATIONAL CAlJCULUS 141
By (9.80) and (9.81), the inequality of dissipation is satisfied when

c1 = 1, c2 = 0 in our case. We shall find the solution of the equation
q 3 (q 0 , t) = a. We have q3 = q03 - t = a by the Hamiltonian sys-
tem. Therefore q03 = a - t. It is easy to see that J (q 0 , t) = 1.
On substituting the necessary functions in formula (9.75), we
obtain
a
! - iy1t -IY21 ~ fll<s> as
g(y, a)=i ~ e a-t P(a-t)dt=
0
a
00 -iy1t-IYzl ~ CIJ(s)ds
=i) e a-t P(a-t)dt+O(!Y 2 !-""). (9.82)
0
Transform the integral in the right-hand side into the form
-i
-rJ e
-iYI(a-(3)-Y2
a
~ fll<s>ds
13 P(~)d~=
a
a 0
-iy1a-IY2l ~ CIJ(s)ds f iY!I3-IY21) CIJ(s)ds
= + ie 0 J e 13 P (~) d~ (9.83)
-oo
with the help of the substitution a+ t = ~. Considering that

21 f
cixAe-if3AF (x) = F (~), we obtain
X (3
(1 1 2) -k Sfll<s> a; x k S cp<s> as
gAt, A 2 , B F(x)=ie 0 ) e ° F(~)d~+O(k-"")
-00
applying the operator with symbol (9.83) to the function

F (x) = F (x, k).
dcf
It is easy to verify that the first term in the right-hand side produces
the exact solution of equation (9. 77). Thus in this example the first
term of the asymptotic (i.e., of the "quasi-inverse" sequence) happens
1o coincide, correct to 0 (1/kn), for any n, with the exact solution of
the problem.
Note the following circumstance. If F (x, k) = 0 when x < 0
X
and x < T, then the obtained integral equals ) F (x, k) dx.

0
Thus we may take the integral of the function F (x, k) from 0 to

x, i.e., we may solve the problem (Sec. 6 of Ch. V)
i ~~ = F (x, k), u lx=O =0 (9.84)
with the help of the main theorem by the following procedure:
a·
1. Add such a term to the operator ilz , that the absorption con-
ditions might be fulfilled;
2. Take an ordered representation of the obtained operator;
3. Apply the formula of the main theorem (considered above in
the particular case).
Then the above-mentioned formula of the main theorem has merely
stated that the solution of problem (9.84) is given by the integral
X
JF (x, k) dx correct to 0 (k- 1 f 2 +e); indeed,, the solution of problem

0
(9.84) is exactly equal to this integral by a formula indicated in
Chapter V. The same procedure provides a solution of a broad class
of differential equations.
Note. Consider a square of an operator of the form (9. 77):
()2 {)
- iJx2 + 2krp (x) ax- k2rp2 (x).
Here the absorption condition in the form indicated above is

obviously not fulfilled; nonetheless, it is clear that the operator is
quasi-inverse. There is a generalization of the theorem for this case
in the following item. This generalization also contains the example
in Sec. 8, which does not satisfy the condition of this item.
Example. Consider the wave operator (cf. (8.6))
~ {}2
H =fJt2
- - c2 (x) .1 •
The Hamiltonian system corresponding to it has the form
(9.85)
where x = xu x 2 , x 3 ; p = p 1 , p 2 , p 3 • It follows from equations (9.85)

that
(c2 (xo) + 1) I Po 12 = 1
INTRODUCTl!ON TO OPERATIONAL CAI.JCULUS 143
and
p~ = I __1--;-1-
-.
V i+ (xO) c2
Hence
I c (x 0) I
t=+2 1+lc(xO)i 'l:+to.
Add to the wave operator a complex term of Lhe form
iH-:- - icp1 U2) ~2 {P2 (~) , p = i\1,

where cp1 , 2 (t2) = 0, when I t I ~ T, cp1 , 2 (t 2) = a> 0 when t =
= T + 1. Then the absorption condition of each trajectory is ful-
filled, since p 2 =1= 0 is on a trajectory as it follows from equation
(9.85).
Therefore, the operator if + iii is quasi-inverse. It is not diffi-
cult to see that the solution of the problem
[if + ill u = F (x, t),

where F (x, t) is finite, does not depend on the form of cp (t2), when
t 2 < T 2 and if, ii are defined above.
Thus the solution of the problem for sufficiently small t (t < T)
does not change, when we introduce the absorption term.
Note. An addition of the summand -icp (x) p 2 cp (p), where cp (x) =
= 0 when I x I < M and cp (x) = a > 0 when I x I = M + 1
provides the absorption condition for every trajectory as well, since
for sufficiently large 1: I x (x 0 , 1:) I > M + 1 by equations (9.84).
But in this case only the asymptotic of the non-smooth part of the
solution does not depend on the form of the absorption term. None-
theless the addition of the term makes sense (from a physical view-
point) for some problems.
(4) The main theorem. Consider a one-parameter family of symbols
f (x1 , • • . , Xn, a,~) and symbols of order n, r (xlt ... , Xn, a, ~), where·
~ is a parameter, 0 < ~ ~ oo. Let
limf(xh ... , Xn, a, ~)=f0 (x 1 , •• • , Xn, a,)EJ"'"",
~; .... oo
limr(xit ... , Xn, a, ~)=r 0 (x!t ... , Xn, a)EJ"'"",

s-oo
fo =I= 0, r 0 =I= 0.
Let P 0 (x) be a function of the class CO' (Rn) equal to unity in the
domain I x I ~ d, where d is a constant.
'144 OPERATrGNAL METHOD<S
~
Let Alt ... , An, BE X, r=r Alt ... , An, B,
( 1 n n+1
s )
EX and the
n n+1 )
Hamiltonian of the operator f Alt ... , An,
( 1
B, s
with respect
tor is asymptotically p-quasi-homogeneous in the variables Yi =xi,
·O ~ i ~ n; Yn+i = - i 8 ~., j ~ m; £, and the numbers Pi• corres-
ponding to the last (m + 1) arguments, are equal to unity. This
J
Hamiltonian depends also on the parameters a, q> = - i The :x.

leading term of this symbol is denoted by :n: (y, q>, a, s) and the es-
:Sential part is denoted by :Ito (y, q>, a, s), where y = (Yo. YH ..•
· · ., Yn+m), 1P = ('PH · · ., 'Pn), a = (alt · · ., am)·
Let Pi=/= 1 when s < i ~ n, Pi = 1 when i ~ s (in the opposite
-case, we shall change the indices of the arguments y). Introduce the
Hamiltonian coordinates. Denote
Y1 = ffi1- Pm+l• · · ., Ys = Ws- Pm+s; Ys+l = ffis+lt • • .,
Yn = Wn; Yn+l = Pn · · ·• Yn+m = Pm; 'P1 = qm+lt • • •

• • •t 'Pn = qm+n·
al = qlt ••• , am = qm.
Besides this, take the parameter v = s
instead of the parameter S·
Denote by OJ8 (p, q, w, v) the function :n: considered in the new argu-
ments and by OJ8 0 (p, q; w, v) the function :n: 0 •
Define OJ8 (p, q, w, v) as a Hamiltonian function. Denote by H =
= Re JJ8, H = Im ,fJ8 as previously and let Q e be a set in the space
-of the variables q, w, p, v defined by the following system of equations
n n
~ ( roi)21Pt = 1' ~ (qm+i) 2 < e
i=i i=i
(qb ... , qm)EMm, p=O, i&tJ(p, q, w, v)l<e

lvl~d.
Now we shall formulate the absorption conditions, generalizing
the conditions of item 3.
Definition. The absorption conditions are fulfilled, if there exist

constants e > 0, T > 0, a number 1:' = 1:' (q 0 , w, p 0 , v), 0 <•' < T
and a function .It (q, w) E Coo which is not equal to zero when (q, w) E
E {(q, w): (q, w, 0, 0, v) E Q 0 } such that: (1) when 0 ~ 'L ~ •' and
.( q0 , w, p 0 , v) E Q e there exists a solution
q (qo, w, po, v, 1:); P (qo, w, po, v, 1:)
of the Hamiltonian system
q h=o =qo,
Ph=o = P0 ,
belonging to the class C"";
(2) when 0 't < < ,;'
and (q 0 , w, p 0 , v) E Qe the function
.Ji (qo, w) .fl (p (qo, w, po, v, ,;); q (qo, w, po, v, 1:); w; v)
is non-negative and when 1: = ,;' the function is strictly positive.
Theorem 9.1. (The main theorem.) Let the absorption conditions be
satisfied for a Hamiltonian function rJ/8 corresponding to a symbol
f (x, a,~). Then there exists a sequence of symbols gN (x, a, x 0 , ~) E
E Cz depending on the parameter ~. such that
1 •)
(
[f A 17 A 2, .•• ,
X [gN Ct.
( 2 n+1 n+2 1 )
+RN A 1, ••• , An, B, r, £, (9.86)
where the function R N (xi> ••. , xn, a, x 0 , ~) = 0 z (I x 1-N) for any N
uniformly in ~ and a.*
2 n+1 n+2 )
The construction of gN ( A 1 , . . . , An, B, ~ is contained in
Chapter V. In the case ~ = oo we obtain the simplest version of the
main theorem formulated earlier. Though in the right-hand side of
(9 .85) there is the function P 0 instead of 1, which was in the main
problem, this theorem gives a direct answer to the question stated
by the main problem. Indeed, it is possible to take a unity instead
of the function P 0 in the right-hand side of (9.86), if the common
1 n n+1
spectrum of the sequence A 1 , • • • , An, B belongs to the domain
Q; X Q. This condition is not fulfilled as we saw in the problem on
the crystal (there is the spectrum of ih ;: on M N in the segment
* Under the same conditions there is a left quasi-inverse sequence

( n+1 2 1 n+2 )
g}v=g]v A1, ... , An, B, r, ~ :
g]vf-Po=Oz (J x 1-N).
See the Main Theorem in the general form in Ch. V.
10-01225
146 OPERATIONAL l\IETHODS
[-n, n] and ~ = ! ).But it is true for general difference schemes

if a convenient space of smooth functions is taken instead of 111 N·
There is a similar situation in the case, when B is the Hamiltonian
of the oscillator considered in Sec. 8 and therefore has a discrete
spectrum located on a lattice of a step h, where his Planck's constant.
1 n+1n+2)
The construction of gN ( A 1 , . . . , An, B in these problems permits
to obtain the quasi-classical asymptotics of the secondary quantized
equations.
Indeed, all the results obtained for the equations of vibrations of
a crystal lattice can be extended to this general case, moreover, if
the absorption term in the given equation is missing, it must be
introduced in the same way as it was done in the example with the
wave operator.
The expansion in a power series in ~ can be obtained in just the
same way in the general case. Here the parameter v has the function
of w in the crystal equation.
Problem. Find the effect of Cherenkov's type for a difference scheme
approximating, as h ~ 0, the wave equation:
h
"T = d = const, c (x) E Coo, F (n, m) = 0,
when n = -1, -2,
Find an asymptotic of the solution correct to kN.
Hint. The cited Hamiltonian function has the form
sin 2 vp 1 = c2 (x 2 ) d 2 ·sin2 vp 2 •
Use a space of integer functions of the first order of the type ~:
as an analogue of M N (Kotelnikov's theorem).
Note. There is no operation "prime" in the statement of the main
theorem. Indeed, we have used the operation proving the theorem
only for M = { i :x, x} ; the rule of the reduction of the main pro-
blem remains true without the operation "prime". Therefore, the
main theorem is true, if the structures satisfy axioms (f-! 1 )-(f-! 8). If
the axioms (f-! 1)-(f-! 6) are true, then the main theorem can be proved
only in the case, when the symbol f (x, ~. a) is a polynomial in x.
In conclusion, the following statement can be taken as an axiom
instead of f.! (8): (9.5) holds for any P E c!/ (so to say, the f.!-
00
structure agrees with the ordered representation of the nilpotent

Lie algebra).
I. FUNCTIONS OF A REGULAR OPERATOR
In this chapter we shall develop an operational calculus corres-

ponding to a 11-structure (see Introduction) based on the commutative
unbounded operator algebra in a Banach space B. It will be our
purpose to determine how wide an operator class constitutes a set M.
In addition we shall clarify which symbols correspond to bounded
operators on B. This will facilitate the forming of a closed extension
of the C"" -symbol class.
For simplicity our presentation will not depend on the It-structure
notion. The symbol concept introduced in this chapter, however,
proves to be coincident with a similar notion within the 11-structure
framework.
To illustrate the function of an operator concept we shall consider
a function of multiplication operator A by an independent variable x
defined on a Sobolev space W~ (Rn). It is easy to see that if cp E
E Ck (Rn) then one can readily define the function of multiplication
operator A by x, this operator being in turn a multiplication operator
by cp (x). It will be seen that cp (A) is a bounded operator on W~.
If the kth derivative of cp (x) is unbounded then the multiplication
operator by (jl (x) is also unbounded in w~.
On the other hand a multiplication operator by x defined on W~
has the following properties:
(1) it generates a one-parameter group of multiplication operators
by the exp {ixt} family;
(2) the multiplication operator satisfies the condition: for any
g E W~ (Rn), where s ;;;;:: k, the following inequality is valid:
II eixtg (x) llwk:::;;; c (1 + /t lk) II g (x) llws'
2 2
that is,
II eixt 11:~ :::;;;c (1 +It lk),
2
c = const.
10*
148 OPERATIONAL METHJODS
In terms of the corresponding evolution equation we can formulate

the conditions (1), (2) as follows: there exists a unique solution u
of the evolution equation
i d~t(t) =Au (t), (*)
u(O)=g, gEB 2 , B 2 =W~,
u being a continuous function of t with values in B 1 = W~, k ~ l

satisfying the condition
II u (t) I!BJ~c (1 +It J•) ·II u (0) lln 2
for all u (0) E B 2 = W~.

As we shall show, a wide operator class is covered by the relation
between the number s, featured in the estimate on the solution of
evolution equation (*) (where, as a given operator, we have a mul-
tiplication operator by x), and the boundedness of a multiplication
operator by cp (x).
The condition (**)for the growth rate of the solution of the evolu-
tion equation generated by unbounded operator A becomes necessary,
if it is required that the finite k times differentiable function of A
should be a bounded operator. If we skip this requirement and restrict
ourselves only to the infinitely differentiable funcLions of an operator,
then the growth rate condition (**) may be essentially relaxed.
Consider a differentiation operator in a space C (R) of continuous
functions bounded in infinity (or in a space C N of continuous func-
tions increasing less than I x IN). As it is easily seen, the totality of
functions of such an operator form an algebra .9fl 0 (or an algebra$!/ N•
respectively). This example shows the necessity of using the algebra
$!/ 0 (the algebra .'lfJ N is the case of ·unbounded continuous functions)
if we wish to consider an algebra of functions of operators drawn from
a class which includes a differentiation operator in a continuous func-
tion space. With these two examples we wish to emphasize "the neces-
sity" of using a growth rate condition of type (* *) as well as the
algebra $!/ N·
Finally, let us consider a multiplication operator A by x iy +
defined on W~(R 2 ). It is natural to defme a k times differentiable func-
tion of x + iy as a function cp (x, y) of two arguments x and y. It
is evident that each operator, x = ReA and y = Im A, will satisfy
the condition (**), ReA being commutative with Im A.
We shall introduce operators satisfying a condition of type (**)
and call them generators of degree s. Then we shall consider operators
of type A 1 +
iA 2 , where A 1 and A 2 are generators of degrees, A 1
being commutative with A 2 • Such operators will be called regular
operators and functions of them will be considered as functions of
A 1 and A 2 •
CH. I. FUNCTJ10NS OF A REGULAR OPERATOR 149
In Sec. 7 we shall prove the theorem which will show the sufficiency
of a condition of type (**). Specifically, we shall prove that in the
case of a purely discrete operator spectrum the existence of a complete
system of eigenelements and associated elements (the number of
the latter not exceeding N for a given eigenvalue) is equivalent to
the regularity of the operator (of degree N). Thus, for a discrete spect-
rum at least, we shall show how wide the regular operator class is.
When the spectrum is not discrete the concept of a system of eigen-
vectors and associated vectors is introduced in the case of the regular
operator in Sec. 8. The theorem of completeness of such a system will
be proved there. A similar concept in the case of an arbitrary opera-
tor, however, has not yet been involved. Therefore, there is no
inverse theorem as yet.
Sec. 1. Certain Spaces of Continuous Functions and Related Spaces

We shall now consider certain Banach algebras of continuous
functions. First of all, consider the set of continuous bounded comp-
lex-valued functions with the field of definition Q c: R". As usual,
we introduce the structure of a vector space on this set. Define the
norm of function f by the formula
II t II= sup It (x) I· (1.1)
xEQ
Denote such normed space of functions by C (Q). The space C (Q)

is a Banach space.
In analogy with C (Q) we define the Banach space C<k> (Q), k > 0
being an integer. The norm in C<k> (Q) is defined by the formula
II f llck(Q) = max sup I Dif (x) I, (L2)
O<SI j I<Sk xen
n
. ali 1
where i=U11 i2, ... , in), Iii=~ i;, D'= · .
i=1 ax'!i ... a/nn
Let C* (R") be the space conjugated to C (R"). The Dirac 6;-
function is called the functional on C (R"), which gives the value
g (~) at point ~ E R" for every function g E C (R"). The functional
6~ belongs to the space C* (R").
Let N ~ 0. By C N (R") we denote the space of continuous func-
tions in R" with finite norms
· II g llcN<R"> = :~~ / (i: I·l(:)l )N

Thus, by definition, the space C N (R") will include all continuous
functions increasing not faster than 1 x IN.
150 OPERATIONAL METHiODS
In this respect it resembles "a superstructure" on C (Rn) which is

similar to w-~ (R") being "a superstructure" on L 2 (Rn).
Obviously, the Dirac function 8£ can be considered as an element
of the space C'f.r ( Rn).
The functional .o/r" provides another example of an element of
C'N (Rn):
.~ '1' (g)= ~ ffJ (x) g (:r) dx, (1.3)
Hn
where fP is an element of the vector space C'; (R") of infinitely differ-

entiable functions with compact carriers.
It will be convenient to identify the functional .o/q with the fune-
tion fP (it is easy to see that (fll =I= 0) =>- (.~(jl =1= 0), i.e., it is easier
to write down fP (x) E C'N (R") instead of .'1",1 E C"N (Rn)). We shall
denote any functional .!if E C'N (Rn) by fP (x) (though fP (x) may not
be the function in the ordinary sense of the word); the functional .'F
will be denoted as ) fll (x) g (x) dx, though the integral does not
exist in the Riemann sense. In particular, the functional 8£ will be
written ) 8 (x- £) f (x) dx, as accepted in physics literature.-
It will be sufficient for us to consider a space, simpler and easier
to understand, namely, the subspace of C'N (R") which is the linear
core of the association of the Dirac 8-function and CO' (R").
Let us denote by C!v (Rn) the closure in C'N (Rn) of the linear core
of the set C';' (Rn) U {8d£ER""
Let us now define the Fourier transformation in C!v (Rn). As usual,
the Fourier transform of function fll E C';' (Rn) c Cf:v (R") will be
denoted by function q;' defined by the formula
- fief def 1 ,. .
fP (p) = (F4J) (p) = <p (x) e-tP·x dx. ( 1.4)
(2:rt)n/2 •
R"
The function defined by the formula

- <ll'f def 1 ~- . drf 1 .
8-s(p) = (F8~) (p) = 12 6s(x) e'P·x dx = 12 e-tP·£ (1.5)
(2:rt)" • (2:rt)"
R"
will be called the Fourier transform of the 8dunction.

Denote by .9.91v (R") a vector space which consists of linear com-
binations of the functions of type (1.4), (1.5).
The formulas (1.4) and (1.5) define an operator which maps a linear
core of the set CO' (R11 ) U {8 6}[;Enn into .CJ3!~ (R"). The operator F
will be shown to have an inverse operator.
CH. I. FUNC'rl'ONIS OF A REGULAR OP,ERA110R 151
Let us now introduce the norm in space Sfl<]v (Rn):

~ dcf ~
II <p IISS]V<Rnl= II p-i<p llcj\r<Rnr (1.6)
Denote by -Sf} N (Rn) the completion of the space Sfl<]v (Rn) in the
norm (1.6). Then the operator F can be uniquely extended to the
isometric isomorphism of space Ci\r (Rn) on Sf} N (Rn). We shall
denote this isometric isomorphism by F and call it the Fourier
transformation of the elements of space C!v (Rn). We shall later see
that the space S1l N (Rn) is nothing but the space of continuous (smooth)
functions.
In the linear core of the set
there exists a commutative composition which transforms this

variety into the algebra, namely
(<P*'i:)(:r)= 1 )' <p(y)¢(x-y)dy for <p, ¢EC 0 (Rn);
(2rc)n12 ·
Hn
1
<'I~* <'~11 = <'ls+11;
(2:rt)n1
2
(fh*~')(x)= 1
12 ¢(x-E) for ¢ECo(Rn).
" (2rc)n -
Show that there is a constant c, such that

II <r * 1Jl II :::;; c II <r II 111Jl 11.
Let f E C N (Rn). Consider the integral
I = ) [ <p ~ ¢] ( x) f (x) dx. (1.7)

Hn
The integral (1. 7) can be transformed to the form
(2:nt12 I= .l ¢ (t) dt ~ f (y + t) <p (y) dy. (1.8)

Rn Hn
For a proof thereof, it is sufficient to consider three particular cases

since we consider the linear core of the set C 0 (Rn) U {66}: 00
(a) <p, lJl ECo (Rn);

(b) (P ECo (Rn), ~~ = os;
(c) <p=<'ls, ¢=611.
152 OPERATIONAL 2\'IETHODS
In the (a) case

(2n)n 12 I [(P*'i'](x)f(x)dx= I f(x)dx J fP(Y)'iJ(x-y)dy=
Rn Rn Rn
= JfP (y) dy .\ f (x) 'i' (x-y) dx= I (p (y) dy J'i' (t) f(y+ t)dt=
Rn R"
= J\jl(t)dt JfP(Y)f(y+t)dy.
R" Rn
In the (b) case
(2n)" 12 I= .,\ fP (x- £) f (x) dx;
n"
on the other hand,
J 6 (t- s) at Jt (y + t) fP (y) ay =
R" R"
=) f(y+£)fP(y)dy= Jf(x)(p(x-~)dx.
Rn R"
Finally, in the (c) case
(2nt 12 I= I 6 (x-£-YJ) f (x) dx= f (£+11),
Rn
J 6 (t- 11) at Jt (y + t) 6 (y- s) dy =
R" R"
= J6(t-YJ)f(£+t)dt=f(£+rJ).
Rn
IJ(
This completes the proof (1.8). Using (1.8) we have
fP * \jl) (x) f (x) dx I~ (2n) -n/ 2 11 ~J llc)\r<R"> X
X ( su~ I(1 +It I) -N

fER R"
Jf (y + t) fP (y) dy I)~
~ (2n)-" 12 ll 'i' llc)\r<R") II fP llcj\r(R"J X
X sup[(1+1ti)-N sup f(y+t)N J~
tERn YER" (1+1YI)
~ (2n) -n/ 2 11 'i' llcj\r(Rn) II fP llcj\r(R"> X
X sup /
t,yERn
f(y+t)
(1+1ti)N (1+IYI)N
I•
:CH. I. FUNCTJlONIS OF A REGULAR OPERA'])OR 153
we have
IJ(
n"
(jl* ljl) (x) f (x) dx I~ (2n) -n/ 2 lllJl llcN(R"l II cpllcN(Rn) II f llcN(RnJ'
Thus,
II <p * lJl llcN<R"l ~ (2n) -n/ 2 II rp llcN(Rnl lllJl llcNlR"l' ( 1. 9)
Q.E.D.
By virtue of the estimate (1.9) we can extend the operation * to
the whole space Cj~ (R") in a unique fashion, the inequality (1.9)
remaining valid. C[v (R 11) will then become converted into the Banach
algebra (to be more precise, into a fully normalized one). Since the
Fourier transformation is an isometric isomorphism of the Banach
space C!v (R") on !B N (R11), the operation * will induce a binary
operation which transforms !B N (R11) into Banach algebra and is
defined by the formula
cp·~ = F [(F- 1cp) * (F-tljl)].
Let us see how the operation · acts on !BYv(R11). Let cp 0 , ljl 0 E C0 (R11)
and let
(jl = fPo + +
2Jj al)£J., lJl = lJlo 2Jk Pk811k' (1.10)
Then
(jl * lJl = fPo *'Po+ 2J ai8;. * lJlo + fPo * 2J Pk811 +
j J k k
+ 2J aipk (2n) -n;z 8~. + 'YJk,

j, h J
and we thus have

(Ffp·F\jl) (p) =--= (Fcp0 ·F\jl 0) (p) + (2n)-n/Z 2J aieip·~i (F~1 0) (p) +
j
+ (2n) -n/ 2 2J B"eiP·1l" (F cp (p) + 2J (2n)-n aipkeip·

0) ((;i+ 11 k).
h j, h
It is easy to verify that (Fcp 0 ·F\jl 0 ) (p) = cp 0 (p) ljl 0 (p). So (F({J·F\jl) (p)
= cp (p) ljJ (p) for any cp, ljJ of the form (1.10), i.e., in .9?1~ (R") the
operation·is in fact a point-by-point multiplication of func-
tions. For this reason we shall drop the point in the notation of this
operation:
qJ\jl = fP •\jl, ~' ~ EfB N (R11 ).
154 OPERATIONAL ::\iETILODS
Sec. 2. Embedding Theorems

Quite often it is difficult to determine whether functions belong to
the Banach algebra C or, in particular, the Banach algebra !B N·
For this reason it is important for us to establish the salient features
which show that the belonging to a Sobolev space includes the belon-
ging to !B N or C. To be more precise, we have in mind the existence
of the natural embedding of Sobolev spaces in C or .9.9 N·
Theorem 2.1. Fork> N + ; there exists an embedding W~ (Rn) c
C IBN (Rn).
Proof. Let f E S. Then
II f llcJV<R"J = sup _
II QJ llc(R")-1 R"
I·\ cp (x) (1 +I xi)N f (x) dx I=
= sup
llrPIIc(R")=1
I\J f(x)(1-+-lxi)k cp(x)
(1+jxj)k-N
dxi::::::::
'-"'
:;: :
""'II QJ 1/supn =1
C(R)
vr
R"
f
J
n"
(1
I lJl (x) 12 dx X
+I xI )2(k N)
(2.1)
V ('
We have
r
IJ<p/lsu~=1
C(R)
J
n"
I cp (x) 12 dx
(1+lxl)2(k-N) = l/I J
R"
dx
(1+lxl)2(k-N)
<
oo,
(2.2)
since 2 (k- N) > n. Next, (1 +I xI ) 2k~ (1 + x 2 )k 2\ therefore
~ If (x) 12 (1 +I x 1)2k dx~ 2~< II f llw~<n"r (2.3)
R"
By virtue of (2.1), (2.2), (2.3) we obtain II f lie*N ~e-ll f II wk(Rn>
2 J
where the constant c depends on N, n and k, but is independent
off E S. This means that for any cp E Co (R")* the following estimate
is valid:
II (p ll._usNm">~c II Cf! llw~(Rn)' c = const.
It remains to be shown tha l if the sequence {Cf!i} of functions
<p E C0 (Rn) converges to zero in .9.9 N (R") and is fundamental in
* By CQ' (Rn) we denote the space of Fourier iransforms of functions

cp E C0(R").
CH. I. FUNOTFONlS OF A REGULAR OP,ERA110R 155
w: (Rn), then it converges to zero in w:

(Rn) as well. L~ {<ri}
be such a sequence and let (pi= p-!<lli· Then (piE C0 (Rn), <!li--+ 0
in C'; (Rn) and {cpi} is fundamental in
~ -· h
that cpi -+0 in W2 (Rn).
w:
(Rn). It is to be shown
Since {cpi} is fundamental in

vergence of this sequence in L 2 (Rn) ::J
w: w:
(Rn) there follows the con-
(Rn). It is sufficient to
show that II fPi IILz(Rnl--+ 0. Let ¢ EC0 (Rn). Then
IJ cpi (x) ¢ (x) dx I~ II cpi llcN(Rn) II ¢ 1\cN(Rn) _.. 0

for j--+ 0. On the other hand, if cpi--+ 0 in L 2 (Rn) then
E~ ) rpi (x) ¢ (x) dx = (<p, ¢)~(Rn).

R"
Consequently, (<p, ¢)L2 (Rnl = 0 for any ¢ EC0 (Rn), whence it follows
that cp = 0, Q.E.D.
Theorem 2.2. There exists an embedding .tfJ N (Rn) c: c<N> (Rn).
For the proof of Theorem 2.2 we shall need the following simple
lemma.
Lemma 2.1. Let f E.tfJ N0 (Rn ) , f = p-tf. Then
~
Drxf (p) = (2n) -n/ 2 ) (- ixt' eiP· x7 (x) dx.

Rn
Here
... ,
Proof. It is obviously sufficient to verify the statement of the

lemma in two cases: 7EC0 (Rn) and J= 66• In the first case the
statement is obvious. Let = 6s. Then 7
f (p) -_ (2n )-n/2 e -iP·s , so
Daf (p) = (2n) -n/2 ( - i~t eip· ~ =
-=(2n)-n/ 2 ) e-iP·x6(x-~)(-ix)adx,
Rn
this proves the lemma.
Proof of the theorem. Let /E.%'~(Rn), f=F- 1f. Then for I ai~N
I Daf (p) I= (2:rt)-n/ZI J f{x) ( -- ixt e-ip·x dx ~~
~(2:rt)-n/211fllc*(Rn) sup I x~1 . .. x~ne~ip·x ~~

N xERn (1+JxJ)
~ (2:rt)-n12 11 f JI.'J.1N(Rn)"
For this reason operator / 0 of the identical embedding .%' ~ (Rn) c

c.%' N (Rn) is bounded in c<N> (Rn) and can be extended to cover
the homomorphism/: .%' N (Rn)- c<N> (Rn). It remains to be proved
that (/cp = 0) =H cp = 0).
Let /cp = 0 and let {cpi} be a sequence of functions belonging
to .%' J. (Rn) and such that (jlj-+ cp in .%' N (Rn) when j-+ oo. Then
~im II (jlj llc<N><nn) = ~im II I <fi llc<N>(RnJ =II /(p II= 0.

J-+00 J-t-00
Let fPi=F- 1cph cp=F-

1 cp. We have fPr+Cp in C~(Rn), whence,
in particular, for any f EeN (Rn),
r cp (x) f (x) dx = ~im jr Cpj (x) f (x) dx.

J J-+00
Rn Rn
We shall show that for any infinitely differentiable function

f EeN (Rn)
~im
J-+00
JCpj (x) f (x) dx = 0;
Rn
hence it follows that (jl = 0 since C"" (Rn) n c N (Rn) is dense

in C N (Rn).
Thus, let f be a fixed infinitely differentiable function of C N (R")
and e > 0 be a fixed number. Choose a number j 0 so that for j > j 0
the inequality II tVi-
Cpj0 llc'JV<Rn) <e holds. Since CVi 0 is the linear
combination of the function of C'O (Rn) and «'5;-functions, a positive
A will be found such that J
Cpj 0 (x) g (x) dx = 0 for any function
Rn
g E C N (Rn), which becomes zero in a sphere I x I <A. Let f 0 E
E C'O (Rn) be a function which coincides with f in this sphere and
OH. I. FUNCTFONIS OF A REGULAR OP.ERATOR 157
which does not exceed modulo 1. We have

~ ~i (x) I (x) dx = ~ (j)i (x) lo (x) dx +
+ ~ [~i(x)-qJj 0 (x)](f(x)-l 0 (x)]dx, (2.4)

Rn
since
Consider the first term in the right-hand member of (2.4). We have
IRn~ ~i (x) lo (x) dx I= IRn~ IJli (x) [Fiol (x) dx 1:::;;;

:::;;; II IJli IIC(Rn) ) I [Fiol (x) I dx-+ 0
Rn
for j-+ oo. The second term in the right-hand member of (2.4)
may be estimated in the following way:
IJ[{jli (x)- qJj (x)] [/ (x)- lo (x)] dx I<

0
:::;;ell 1- fo llcN(Rn):::;;;2ell f llcN(Rn).
It means such a number j 1 > j 0 will be found that for r;;:- j 1

IJ(j)i (x) I (x) dx I::;;;;e (1 + 2 II f llcN(Rn))·
Since e is arbitrary, it follows that
~im ~ (j)i (x) I (x) dx = 0,
J-+00
nn
Q.E.D.
Thus, from Theorems 2.1 and 2.2 it follows that for
k> ~ + N, W~' (Rn) c $N (Rn) c c<N> (Rn).
'Ve can see that the Banach algebra .911 N (Rn) is "in-between" the
Sobolev space contained in C<N> (Rn) and the Banach algebra C<N> (Rn).
So the .9.9 N (Rn) is a Banach algebra which is closer to the Sobolev
space than C<N> (Rn).
158 OPERATIONAL i\IETHJODS
Problem. Show that for any cp E C"N (Rn) the following formula
is valid:
J
[Fcp](p)=(2:rt)-n/Z cp(x)e-iP·Xdx.
Rn
Consider a subalgebra ffJ N (p) of the Banach algebra fii N (R)
obtained by closure of the set of functions of §J N (R), which are
equal to zero in the neighborhood of set cr (one should note that for
every function there is a corresponding neighborhood). Obviously
§J N (p) is a closed ideal in §J N (R). The factor-algebra §J N (R)/§J N (p)
with the ordinary norm
II {cp} II= cpE{cp}
inf II (P II ct>
.::m N(R)
will be denoted by ffJ N (cr).
Sec. 3. The Algebra of Functions of a Generator

Let E be a vector space given over field C with the norms 11·111
and 11·11 2 , where c ll·lh;;::: 11·11 2 , c = const. Let B 1 be a completion
to space E in the norm 11·11 1 and B 2 be a completion to space E
in the norm 11·11 2 •
In this case let us write B 1 :S B 2 •
Let u (t) be a function with values in E defined on R and satisfying
the equation
. du A
l d t - U=
0, (3.1)
and the initial condition
u (0) = g, gEE. (3.2)
Here A is the linear operatm: A : E-+ E and the derivative ~~l E E
is such that
JJ a~;t)- u(t+l~-u(t) 111-+0 for h-+ 0.
Definition. A homomorphism A : E-+ E is said to be a generator
if, for any gEE, the solution u (t) E E of equations (3.1), (3.2) exists
and is unique for the class of functifms satisfying the inequality
II u (t) il2 :::;;; c (1 + I t I)\
where c > 0, k ;;::: 0; k, c being constants.
We shall assume that A is a generator of degree s ;;::: 0 with the defining
pair of spaces (B 17 B 2 ) if a constant c1 exists such that
II 11 2 :::;;; cl (1 + I t I)" II g lh
u (t) (3.3)
for any t E R and any g E E.
CH. I. FUNCTrONIS OF A REGULAR OPERA'.l10R 159
If B 1 = B 2 = B then we shall assume that A is a generator on the

Banach space B.
Note. By substituting the norms 11·11 1 and 11·11 2 with equivalent
norms whenever necessary we can put c = 1 in the above definition.
We shall do likewise in the text that follows.
Lemma 3.1. The mappings {U (t)}tER• U (t): E._ E defined by
the equality U (t) g = u (t), g and U satisfying the conditions of the
definition, are homomorphisms of the space E and form a one-parameter
group, i.e.,
u (t2). u (t1) = u (t1 + t2).
Proof. Let u (t) = U (t) g, u 1 (t) = U (t) and (t1). By definition
u (t1+ t 2) = U (t1 +
t 2 ) g. This means that one needs only to show
that u (t1 +
t 2) = u 1 (t). Denoting u (t1 +
t) by v (t) we obtain
v(O)="u(t 1), i ~~ =Av.

Since u 1 is a solution of the same Cauchy problem, v (t) = u 1 (t)
due to the uniqueness of the solution, Q.E.D.
Definition. The family {U (t)} will be said to be a group generated
by A.
Lemma 3.2. AU (t) = U (t) A for any t E R.
Proof. For any fixed t ER
lim/li U(t+h)-U(t) g-AU(t)gll =0.
h~O h 1
Presuming t = 0 and taking into account that U (0) g = g we obtain

lim IIi U(h)-1 g-Ag/1 =0.
h~o h 1
We shall apply U (t) to the expression within 11·111 • Then with

h._ 0 (3.3) gives
//U(t) (i U(l~-1 g-Ag) ~~2~(1+1tl)sjji U(hl-1 g-Ag//1--o.
Hence
iU(t) U(t.l-i g=U(t)Ag-ta(L\),
where II 0' (L\) 11 2 . _ 0 for L\ ._ 0. On the other hand,

iU(t) U(t.l- 1 g=i U(t+~-U(t) g=AU(t)g+8(L\),
where II<'\(~) 11 1 ~0 for ~~o. Therefore,

II [U (t) A- AU (t)J g ll2 =II <'l (~) -cr (~) l12 ~ 0
for ~ ~ 0. This completes the proof of the lemma.
Lemma 3.3. Let A be a generator of degree N with the defining pair
of spaces (B1r B 2 ). Let {U (t)} be a group generated by A and let cp E
E !J1! N (R). Then the element I [cp; g] of space B:* defined by the formula
00
I[cp; g](h*)=V12n )cp(t)h*(U(t)g)dt, h*EB:, cp=F- 1<p

-00
belongs to B 2 , where
1
III[cp; g] IIB2~ V 2n II'PI15,JN(RlllgiiB!·
Proof. Let h* E B:. We have
00
I J[cp; gJ (h*) I= V~n j {p (t) h* (U (t) g) dt ~

-00
& 1 II II 1h* (U (t) g) 1

-. -: : V2n cp ·~N(R)"~~~ (1+\ti)N •
By the definition of the generator we have II U (t) g I\B2 ~
~ (1 +It I)N II g I\B1 ; therefore
II [ cp; g) (h*) I~ V 12Jt II cp II5,JN(H) II h* liB~ II gliB I·
Hence
1
II I [cp; g) liB~*~ V 2n II 'P II5,JN(R) II g \\B1· (3.4)
It remains to be proved that I [cp; g] E B 2 •

If cp EC0 (R), then
00
1 (' ~
I [cp; g] = V 2n J cp (t) U (t) gdt.
-00
The above integral, being a Riemann integral of a B 2-valued func-

tion, is an element of B 2 • If iP = 6£, then once again I [cp; g 1 E B 2
since
1
I [cp; gJ = V2n u (s) g.
Hence, if cp E .91l"N (R) we have I [cp; g] E B 2 • From (3.4) it follows
that the norm of the operator on !J1! N (R) to B 2 sending cp into I [cp; g]
CH. I. FUNCTJlONIS OF A REGULAR OPERATOR 161
is bounded by the number v12n- II g II Bl· Since B 2 is closed in B~*

we obtain I [cp; g] E B 2 for any cp E 59 N (R).
In this section A is presumed to be a generator of degree N with
the defming pair of spaces (B 11 B 2 ) while {U (t)} constitutes a group
of homomorphisms generated by the operator A.
Set cp E 59 N (R). Denote by cp* (A) the operator cp* (A) g
def
=
det I [cp; g] on B 1 to B 2 which is defined on region E. Lemma 3.3
gives
1
II cp* (A) II~ V 2n II QJ llseN<R>·
Lemma 3.4. Let cp E 59 N (R) and .!FE 59 N (R)~ where .¥ (x) =

= xcp (x). Then cp* (A) A = ff * (A).
Proof. By definition, for any h* E B~ and q E Ewe have
00
h* [cp* (A) Aq] = V 1n

2 ~ ~ (t) h* [U (t) Aq] dt,
-oo
00
h* [.!F* (A) q] = V12n j

\'
ff (t) h* [U (t) q] dt,
-00
where qJ' and if are the Fourier transforms of the functions cp and .¥.
Let us show that for any function i, which with its derivative
belongs to the space C N (R), the following formula is valid:
00 00
i j' q; (t) g' (t) dt = ~ .i (t) g(t) dt. (3.5)

-00 -00
First of all, note that it is sufficient to prove the formula (3.5)

for a case when i
E coo (R) n N (R) because for c E N (R), g c
i' E CN (R) for any e > 0 there is an infinitely differentiable
function h such· that
II g-h II CN(R) < e, lfg' ~h' llcN(R) <e.
Thus, let gE coo (R) neN (R)' g' EcN(R). Consider the sequence
{gn} c CO' (R) which has the following properties:
(a) gn(t)=g(t) for tE(-n, n),
(b) llgn-'illcN(R)~ch llg~-g'llcN<R)~c2, c11 C2=const.
11-01225
162 OPERATIONAL METILODS
Then, for any element CD E Ci, (R) the following relationships are
valid:
00 00
-oo
J CD (t) g(t) dt = !~ J CD (t) gn (t) dt,
-oo
(3.6)
00 00
-oo
j CD (t) i' (t) dt = !~~
-oo
J<D (t) i~ (t) dt. (3.7)
In fact, let <D =lim <Di, <DiE F- 1.96' N (R). For any e > 0 we shall
~ - - 0
i-+-oo
fmd such a number j 0 that II ~-<:1)11 <e for j'::;?j 0 • Let the function
support <Dj 0 be in the interval ( - n 0 , n 0). Then for n >no
00
.\ d) (t) (g (t)- in (t)) dt =

-oo
00
= J (&(t)-CDJ (t))(g(t)-gn(t))dt0 ~ce, c=const.

-oo
For any JEF- 1 .96'~(R), hECo(R) Parseval's equality is valid:

00 00
JJ'(t) h (t) dt= ~- f (t) h( -t) dt, (3.8)

-00 -00
where f = Ff: h = Fh. From the embedding .96' N (R) c: C (R) and
from the density .96'~ (R) in .96' N (R) it follows that the formula (3.8)
is valid for any f E Ct.; (R). This means that we can rewrite the
formulas (3.6) and (3.7) in the following way:
00 00
JCD(t)g(t)dt=!~~ j <D(x)gn(-x)dx,
-oo -oo
00 00
J CD (t) g' (t) dt

-00
= !~~ J <D (x) Gn ( - x) dx,
-00
(3.10)
where <D=Fd5, gn=Fgn, Gn=Fg'; here Gn(x)=ixgn(x). By

setting d5 = (ji in (3.10) and CD= ff in (3.9) we obtain
00 00
I (jJ (t) g' (t) dt = - i lim I cp (x) xgn (- x) dx =

j
-00
n-+oo J
-00
00 00
= - ilim
n~oo
~· !T (x) gn (- x) dx = - i J ff (t) g (t) dt
-00 -00
and this completes the proof of the formula (3.5).

CH. I. FUNCTl'ONS OF A REGULAR OPERATOR 163
By using formula (3.5) we obtain

00
h* [ff * (A) q]
1 i ~ d
= V2:n: j <p (t) dt h* [U (t) q] dt =
-00
Jr -<p (t) h* [AU (t) q] dt = h* [ <p* (A) Aq].

00
1
= V2:n:
-00
Lemma 3.4 is proved.

Corollary. Let ff 1 E~ N (R), where .!IF 1 (x) =xi ff 0 (x), j = 0, 1, ..
. . . , k. Then § o* (A) A k = !Fk* (A).
The proof consists of successive applications of Lemma 3.4 to
functions <p = .3':: 0 , .!IF u ... , !F k-1 • Now estimate the norm of the
operator (A - z)-1 •
Lemma 3.5. The operator [A - (a + ib)]-1 exists for b =1= 0 ;and
the following estimate is valid:
II[A-(a+ib)r 1 ll~~~m+ lbi~+i t (3.11)
where c and c' are constants, 11·11~~ is the norm of the operator acting
from B1 to B 2 •
Proof. Let b =1= 0. Denote z = a +
ib. Since Im z =1= 0 thefunction
<p: <p (x) = (x- z)- 1 belongs to w~ (R) for any k, so (jl E~ N (R).
Since 1 E jg N (R) the function 'ljJ:
'ljJ (x) = x (x - z)- 1 = z (x- z)-1 +1
belongs to ~ N (R). By virtue of the preceding lemma <p* (A) A =
= (z<p*) (A) +
1E, where 1E is the identical image of E to itself.
Thus,
<p* (A) (A - z) = 1E,
i.e., the operator (A - z) has the inverse:
(A-zt 1 = fP* (A) IR(A-z>·
Here
II (A-zt 1 ll~~~ll <p* (A) II~~~ -Jzn II fJllla?N<RJ·
1
Set 
11*
we obtain for any g ECN (R)

00 00
Jr-cro (t) g (t) dt = TbT

1c-J crt (t) g (t)
TbT dt.
-00 -00
Since
~~j
lgh-h)l<~llgllcN<R>~~E (1+1ti)N
(1+1ti)N 1+1btl =
=II g llcN(R) max { 1, I ~11N } '
we have
II cro llssN(R):::; II crtllssN(R) max { N' I b ~~+1 } '
hence we obtain (3.11). This completes the proof of the lemma. The
following lemma is proved in the same way.
Lemma 3.6. The operator [A - (a + ib)]-"-, k > 0 satisfies the
estimate
cl
II [A
- (
a + z"b)]-k IIB2< Ck +
B! ~ Tbj"k
li
I b lk+N '
where ck and ck are constants.
Proof. Obviously, for Im z =1= 0 and 0::;;; j < k the function
cpi (x) = x 3 (x - z)-k belongs to ~ N (R). Since 1 E !iJ N (R), it
follows that the function
cpk(x)=xk(x-ztk=1- (x-z)k-xk =
(x-z)k
k
= 1- ~ Ck (-1)i zixn-i (x-ztk
i=1
belongs to .9!! N (R). Therefore, using the corollary of Lemma 3.4 we

have
cp 0 * (A) (A - z)k g = g
for any g E E. Consequently, (A - z)-k = cpo* (A) IR(A-z>''' whence
II (A- zrk II~~:::; II cpk* (A) II~~:::; v12n II cro llssN(R):::;

:::; V12JT II -cro llssN<R> max { lb"lh,
1 1
1b IN+k
}
'
where cp 0 (x) = (x- i)-k. Thus the lemma is proved. Note that
the statement of this lemma is not directly derived from Lemma 3.5.
CH. I. FUNGTmN'S OF A REGULAR OPERATOR 165
The operator h: .98 N(R)-+ B 2 corresponds to each vector h E E

by means of the formula
hqJ = '11* (A) h.
The operator h is determined in $ N (R) and is bounded:
II h II= sup II <Jl* (A) h 112 :s:::::-1-11 h II (3.12)

QJESSN(R), QJ=FO II <Jl llsgN(R) -o::: V 2n 1'
Put the norm ll·llmid in E by the formula
Jlhllmid def V2nllhll.
We shall call the completion of E in this norm an intermediate Banach

space Bmld•
Lemma 3. 7. The following inequalities are valid:
II· ll1:>ll · llmtct:>ll · ll2·
Then
B1-< Bmtct-< B2.
Proof. From (3.12) it follows immediately that II· llmtd~l! · 1!1·
On the other hand,
II h llmtd = sup Vzn II cp* (A) h 112 >- Vzn II h 1!2 = II h liz,
QJESSN(R), QJ.fO II cp II llillsgN(R)
since 111 llssN<R> =II V 2n 6o llc_tv<R> =V 2n. This completes the proof
of the Lemma.
Lemma 3.8. For any t 0 ER the following equalities are valid:
(1) U (to)= (e-iAto)*;
(2) fP*(A)(e-iAto)*=[e-iAtoqJ(A)]*, \fqJE.!BN(R).
Proof. First of all, for any hE E we have

00
(e-iAto)* h = V12n J V2n 6t

-00
0 U (t) h dt = U (t0) h.
It remains to be proved that

qJ* (A) U (t 0 ) h = [e-iAtoqJ (A)]* h.
166 OPERATIONAL· METHODS
We have
00
q:>* (A) U (t 0) h=
1
V 2n JI ~ (t) U (t) U (t 0) hdt=
-oo
00
= V12n Jr -q:> (t) U (t 0 +t) hdt.

-oo
Here we note that if the function !FE ,9; N (R) is represented

as !F (x) = cp (x) e-itox, then F- 1y = (F- 1cp) * V 2n 6t 0 , so for any
gE CN (R)
00 00
) (F- 1y) (t) g (t) dt = ) (F- 1cp) (t) g (t + t0) dt.

-co -oo
For this reason

00
[e-iAtow(A)]*h= J{j)(t)U(t +t)hdt=q:>*(A)U(t )h,

0 0
-oo
this completes the proof.

r .
Lemma 3.9. Let '¢ (x) = ~ ake-ttk"' Then
k=i
V 2n II"'* (A) h llmld::::;; II 'i' 11$N(R) II h llm!d·

Proof. The following inequalities are obvious (for ljl =I= 0):
sup II <:p* (A) h llzl~ "UP II [(<:p¢)* (A)) h 1!2:?;:
q>Ef,BN(R) II <:p llfBN(R) -:::-- q>E.93N(R) II <:p·'ljl llfBN(R) -;?
q>40 q>*O
>- sup 1
II (<:p·'ljl)* (A) h liz
q>EfBN<R>
q>40
v-2n 11"' 11 (;{)
ViJN
(R) u<:pJI ViJN
(;{) (R)
this completes the proof.

CH. I. FUNCT:DONIS OF A REGULAR OPERATOR 167
Lemma 3.10. The junction u in the definition of the generator

satisfies the following inequality for any th t 2 ER
\1 U (t2) llm!d~(1 +I t1--t2l t II u(t1) llrn!d·
Proof. By virtue of Lemma 3.1 we have
u(t 0 + t) = U (t) u (t 0).
Consequently, by Lemma 3.8 we can establish

u (t0 + t) = (e-iAt)* u (t0).
Hence according to Lemma 3. 9 we obtain
llu (to+ t) II mid~ II Bt llcj\r(R) llu (to) II mid= (1 +It I)N llu (to) II mid•
The lemma is proved.
Lemma 3.11. For any lJlE.%'N(R), hEE the element Im1ct["¢; h]
of the space B;h~d defined by the formula
00
Im!ct['i'·; h](h*)= . /2n ) ~(t)h*(U(t)h)dt, h*EB;hict, lJl=FljJ

-00
belongs to Bm 1ct, where

1
II I mid["¢; hlllm!d~ V 2n II"¢ llssJV<R> II h llmid·
Denote by "¢mid (A) the continuation of the operator h-+ I mid["¢; h]

to the homomorphism Bmid-+ Bmid· Obviously
1
II 'i'm1d (A) II~ V 2n II"¢ llssN<R>·
Lemma 3.12. The mapping \jl-+ "¢mid (A) is a homomorphism of

the algebra .%' N (R) into the algebra of operators of space Bmld•
·~
Pl'oof. We must prove that 'i'mtd (A) IPm!d (A)= ("lJl<p)mid (A) for
any <p, "¢E.%' N (R). For any g* EB;hid we have
00
1
g*(IPmict(A)h)= ~~- ) -cp(t)g*(U(t)h)dt.
v 2n -00
Since "¢m 1ct (A) is the homomorphism,
00
g* ("lJlmid (A) IPm!d (A) h)= V 1n

2 Jr -(p (t) g* ("¢mid (A) U (t) h) dt.
-00
168 OPERATIONAL :VIETIWDS
By considering A as a generator with the producing pair of spaces

(Bt. Bm 1d), and by applying Lemma 3.8 we obtain
00 00
J cp (t) g* (\(lmid (A) U (t) h) dt = ~ ~ (t) g* ((\(1 (A) e-iAt)mid h) dt =

-oo -00
00 00
=
-00
Jcp(t) ( J (1Ji*8 )(t')g*(U(t')h)dt')dt.
-00
1
On the other hand,

00
g* ((\(1 · cp)m!d (A) h)= -/2n 1(~ * cp) (t) g* (U (t) h) dt,
-00
The proof of the lemma will be completed if we prove the follo-

wing formula: for any \il Ect (R), t EcN (R)q;,
00 00 00
.\ q; (t) dt ~ (¢ * 8t) (t') t (t') dt' = J(lP * cp) (t) t (t) dt.
-oo
(3.13)
-oo -00
Note that
00 00
J ('P * 8t) (t') f (t) dt' =

-oo
V12n J 1ji (1:) f (T--1- t) dT,
-oo
so (3.13) now becomes

00 00 00
V12n ~
-oo
cp(t)dt ~
-oo
'\i'(T)j(T--1-t)dt= J (lP*cp)(t)f(t)dt.
-oo
(3.14)
But (3.14) had been established in Sec. 1. Thus the lemma is

proved.
Lemma 3.13. There is a sequence {cpn} EC'Q (R) so that for any
hE Bmld, h* E B:'nid
lim h* [(cpnmld (A)-1) h] = 0.
n-+oo
Proof. Choose the sequence {cpn} in the following way:

cpt(x) = 1 for IxI< 1, cpt(x) = 0 for I xI> 2,
cpn (x) = cp 1 (xjn).
CH. I. FUNCTJlONIS OF A REGULAR OPERATOR 169
Such a sequence {<"Pn} is bounded in $ N (R). In fact, let qi'n be the·

Fourier transform of the function <"Pn. Then
00 00
~ (t) =
<"Pn v- 1 rI eitx fPn ( X) d X =
2:rt -00
"
v-1
2:rt
)
-00
eitx <"Pi ( -X
n
) d X=
00
= VnZ:n; )' eitxn<p 1 (x) dx = ncp 1 (nt).

-00
Next we have, for any function f E eN (R):

00 00
~ 'CPn (t) f (t) dt

-oo
= J ncp (nt) f (t) dt -
-00
1
= Jf;;;-ri (t)f (.!....)

-00
n
dt :::::::::11;;; II+ sup 1 I (*)I =
-...-:: -r1 CN(R) tER (1 +It I )N
-II m II + sup I 1 (t) I : : : : :

- -ri CN(R) tER (1+nlti)N-...-::
I I (t) I
~II <"Pi llct_,<Rl ~~f (1 +I t I )N =II <"Pi llct_,<Rl !If llcN(Rl·
Hence, it follows that II <"Pn 11.93N<Rl~ II <"Piii$N(Rl· Set 1Jln (x) =

= !Jln+(~)
X !
• Then 1Jln (x)-
X
+1 . -+0 in any Sobolev
!
space W~ (R).
For this reason 'i'n(x)-+ x~i in $N(R), so that the sequence of
1-)
operators 'i'nm!d (A) converges in norm to the operator ( - •
A+i mid
Next we have
<"Pnmld (A)= 1jl"mld (A) (A+ i) h (3.15)
for any hE E. Note that
( A~i )mid (A+ i)h=h (3.16)
for any hE E. Consequently, for any hE E the following limiting

relationship is valid:
lim CPnmld (A) h =h. (3.17)
n~oo
From (3.17) we obtain for any hE E, h* EB:'nid:

h* [(<p"mld (A) -1) h]-+ 0
for n-+ oo.

170 OPERATIONAL METH,ODS
Now let h be an arbitrary element of Brn 1d, h* EBinid· Since

the sequence of operators {Cflnrnld (A) -1} is uniformly bounded by
some constant M, for a given e > 0 the element hE E can be chosen
:so that II h-h llrnid <e and we obtain
lh* [(cpnrnld (A)-1)h] ~~
~I h* [(cpnrnld (A)-1)/iJ !+M II h* II e_,.M II h* II e
ior n-+ oo, and this means th?t
lim h* [(cp"rnld (A) -1) h] = 0,
n-+oo
since e is arbitrary. This completes the proof of the lemma.

Corollary. The set
M = U R (Cflrnid (A))
q>ECg"(R)
;is dense in Bmid·

Proof. Let us assume that the opposite is true, i.e., that M =I= Bmld·
Then from the Hahn-Banach theorem there follows the existence of
.a non-zero element h* E Bin1d such that h* (h) = 0 for any h EM.
Let h * (h 0 ) =I= 0. Consider a sequence {hn} = {Cflnmld (A) h 0 }, where
{Cfln} is a sequence of functions of Co
(R) featured in Lemma 3.13.
We have h* (hn) = 0 for any n. On the other hand, by Lemma 3.13
we have
lim h* (hn) = h* (h 0) =I= 0.
n-+oo
The resulting contradiction proves that M is dense in Brnld•

Lemma 3.14. If Im z =I= 0 then the operator [(A - z)- 1 lmld is
,invertible.
Proof. Let hE Brnid and let [(A - z)- 1 lmld h = 0. If {cpn} is a
:Sequence of the functions of C0 (R) considered in Lemma 3.13 then
0 =[(A -z) Cfln (A)Jmid [(A -zt 1lmid h = Cflnmld (A) h
for all integers n. From Lemma 3.13
0 =lim h* [ Cflnmld (A) h] = h* (h)
U-+00
for any h'* E Binid· Therefore h = 0, Q.E.D.

Lemma 3.15. There exists a closure A of operator A on Bmlrl·
Proof. Let Im z =1= 0. Then the operator, which is inverse to
{(A - z)- 1 lmtd• is closed. By virtue of Lemma 3.5 this operator is
CH. I. FU:NC'T]QNIS OF A REGULAR OPERATOR 171
an extension of operator A - z. This means that operator A =

= (A - z) + z has a closed extension, Q.E.D.
Operator A will be called a generator. Now we shall give a new
interpretation of the formula
00
Cf!mld (A) h = 1_ ) -cp (t) U (t) h dt, -cp = p- 1cp. (3.18)

11 2n •
-00
Consider space C N (Rn, B), which consists of functions continuous

on R" with values in a Banach space, having the finite norm
II j II _ sup II f (x) II
CN(Rn,B)- xERn (1+1xi)N
Example. For any h E E the function f:

I (t) = u (t) h
belongs to C N (R, Bm 1ct), where
1\ f \\cN(R, Bmtct>~ll h llmtd·
Consider next for any IE C N (Rn, B) the operator I 0 [I]: F-1 JO>~ (Rn)-+-
-+- B defined by the formulas
10 [fJ 6~ =I(£),
Io [/] cp = ) cp (x) I (x) dx, cp EC0 (Rn).
Rn
Problem. Prove that

II lo [/] II~ II I llcN(Rn, B)'
Note. For any cp EC0 (Rn) and any non-equal vectors s~o ... , Sk
from Rn we have
k 00 k
cr+ ~ a;{)~i = ~ lcr(x)j;(1+1xltdx+ ~latl(1+1sd)N.

i=l C"JV(Rn) -oo i=1
Extend the operator I 0 [I] to the homomorphism I [j] : Cty. (Rn) -+-
-+- B. We shall use the notation
I [fl cp = J" cp (x) f (x) dx,

Rn
which provides the necessary interpretation of equality (3.18). Ob-

viously, both definitions of the operator Cf!mtd (A) are in fact iden-
tical.
172 OPERATIONAL l\IETHJODS
Let f E C N (Rn, B) and let the function !F defined by the formula

!F (x) = Cf (x), where C is a closed operator on B, belong to
C N (Rn, B). Then for any cp E F-1§!'Jv (Rn)
i cp (x) Cf (x) dx = C ) cp (x) f (x) dx. (3.19}
If cp is an arbitrary element of C"Jv (Rn) then, by choosing the

sequence {cpk} c: F-1§!'Jv (Rn), which converges to cp and going to
the limit k ~ oo in the equality
JIJlk (x) Cf (x) dx = C ) cpk (x) f (x) dx,
Rn Rn
we achieve that formula (3.19) is also valid in this case.

Lemma 3.16. Let Im z =F 0. Then
(A-zt 1 = [(A -zt 1lmid·
Proof. We have to prove the following two formulas:
[(A - z)- 1 lm 1d (A - z) h = h, hE DA., (3.20)
(A - z) [(A - z)-1 1m!d = 1. (3.21)
Formula (3.20) is valid for h E E according to Lemma 3.5; if
hn ~ h, hn E E, h E DA, then in (3.20) the limit can be reached be-
cause A - z is a closed operator and [(A - z)- 1 1m!d is a homomor-
phism. This completes the proof of formula (3.20).
r
Now let h E E again. Denote by z (t) the Fourier transform of the
function rz (x) = (x- z)- 1 • We have
h= [(A-zt 1 ]mid (A-z) h=
00
= V 2n JI
1 ~
r2 (t) U (t) (A-z) hdt=
-00
00
=,,- \
1 r ~
rz(t)(A-z)U(t)hdt=
v 2n -oo
"'
-"'
- 1 I ~
=(A-z) V 2n J r2 (t)U(t)hdt=
-oo
= (A-z) [(A -zt 1lmid h.

Here we have made use of formula (3.19). Thus, for h E E
(A- z) [(A - z)- 1 lm 1d h = h.
CH. I. FUNGTmNlS OF A REGU:LAR OPERAT"OR 173
From the fact that A is closed and E is dense in Bmict• it follows

that the equality is valid for any hE Bmid• i.e., formula (3.21) is valid.
This completes the proof of the lemma.
We shall summarize the main results obtained in this section in
the form of the following theorem.
Theorem 3.1. (1) The generator A as an operator on Bmid has a
closure A.
(2) There exists a homomorphism M of the Banach algebra !}g N (R)
into the algebra Op (Bm 1ct) of homomorphisms Bmld-+ Bmld• besides
the operator (A - z)- 1 corresponding to the function
1
x-+--
x-z'
Imz:foO.
Notation. Let the homomorphism <D : Bmid-+ B mid correspond

to the function <p E !}g N (R). Denote <D = <p (A). The image of the
homomorphism M will be denoted by the symbol 12!.
The function <p E .CJJ N (R) will be called the symbol of the operator
<f' (A).
Sec. 4. The Extension of the Class of Possible Symbols

Let A be a generator of degree N with the defining pair (B 10 B 2 )
of Banach spaces, every space containing a dense linear manifold
D = D A. In the previous section the operators <p (A) were introduced,
where <p E .CJJ N (R). By definition we have assumed <p (A) = <pm 1ct (A),
where IPmid (A) is an operator defined onE. Now we shall put forward
a formula for <p (A) which is more evident. Namely,
for any hE Bmid• In fact, let {hn} be a sequence of elements of E

which converges to h. Then
I e-iAth- U (t) hn II= I e-iAt (h-hn) II~ (1 +It \)N II h-hn II·
Consequently, the functions U(t)hn converge to e-i.At h inC N (R, Bmtct)·
For this reason
00 00
1
, r-
v 2n
) ~
<p (t)
.-
e-tAth dt = lim
n .... oo
v-
1 ) ~
2:rc -oo
<p (t) U (t) hn dt =
-oo
= lim <pmid (A) hn = <p (A) h,

n-+oo
Q.E.D.
174 OPERATIONAL :VIETHODS
Theorem 4 .1. Let <p E !Jil N (R) and let

'\jJ (x) = xh<p (x),
where k is an i1J,teger also belonging to !Jil N (R). Then

1jl (A)= .J~<cp (fl) = cp (A) A~<= cp (A) A~<.
To obtain the proof we shall need the lemma that follows.
Lemma 4.1. For any h ED-x_ the following formula is valid:
e-iAtAh= .Ae-iAth.
Proof. Let {hn} be a sequence of elements of E which converges
to h, where Ah = lim Ahn.
Q.E.D.
Proof of Theorem 4.1. For any h E E the following equality is valid:
'¢mid (A) h = <pmld (A) A k h, i.e.
00
1Jlm1ct (A) h = -v-

1
2:n:
\ ~
~
-00
<p (t) U (t) Ahh dt. (4.1)
By virtue of (3.19f the above equality may be written as follows:
1Jlm1ct (A) h =
r
00
V12:n: A-h j cp~ (t) u (t) h dt =A-k <pm1d (A) h.

-oo
Here we have made use of the law of closedness of the degree of a

closed operator. Besides, the fact has been taken into account that
(A +i)-1 is the homomorphism Bmld-+ Bmld· We have proved that
the operator '\jJ (A) coincides with the operator A h<p (A) on the set E
which is dense in Bmld· Since <p (A) and '\jJ (A) are homomorphisms
and the operator A h is closed, operator A k<p (A) coincides every-
where with operator '\jJ (A).
Further, the equality (4.1) means that the operator <p (A) Ah
is a restriction of the homomorphism '\jJ (A) of the dense set E. For
this reason the following formula is valid:
1jl (fl) = <p (A) A~<.
It remains to prove the formula
1p (fl) = cp (A) A~<.
CH. I. FlJNGTfONS OF A REGULAR OPERAT'OR 17S:·
In other words, it is necessary to show that the operator cp (A) Alt

has a closed extension, since it coincides with the homomorphism
'ljJ (.if) on the dense set E. Let hn E DAit' hn ._ 0 and the sequenc&
cp (.A')kA hn converges. We have
00 00
cp(A)A"hn= V1211 J'iP(t)e-iAtAkhndt= / J'iP(t) Alte-iAthn dt=

211
-oo -oo
00
1
-- ,;·- A-It j' ~cp (t)e-iAtJ~n dt-- 1
,;·- A-" cp (A) h n-+ 0
v 211 -00
v 211
for n-+ oo due to the fact that the operator A is closed, Q.E.D.
Thus, we have shown that the formula
[fg] (A) =I (A) g (A)
is valid not only in the case when /, g E .* N (R) but also when f
is a polynomial,
g E S!l N (R) and fg E J& N (R).
We shall adhere to the following notation: for the sake of simpli-
city the closure A of the generator A in Bmid will be denoted by A.
We shall mention this specifically only in cases where some misun-
derstanding may arise.
Lemma 4.2. If cp E J& N (R) and if cp (x) does not turn zero at any
point of R, then the operator cp (A) is invertible (i.e. cp (A) h =1= 0
for h =I= 0).
Proof. Let {en} be such a sequence of functions in C0 (R) that
lim h* (en (A) h)= h* (h)
for any h E Bmid• h* E B* mid· The existence of such sequences has

been established in Lemma 3.13. If cp (A) h = 0 then
'IJln (A) cp (A) h = 0,
where 'IJln (x) =en (x)/cp (x). But 'IJln (A) cp (A) = en (A). Consequent-
ly for any h* E B:hict.
h* (h)= lim h* (en (A) h)= 0.
n->oo
Hence it follows that h = 0 and the lemma is proved.

Theorem 4.2. For any f E J& N (R) and a polynomial P there exists
a closure of the operator f (A) P (A) on Bmid·
176 OPERATIONAL METH.ODS
Proof. Let hn-+- 0 and let f (A) P (A) hn-+- q for n-+- oo. Show
that q = 0. Denote by <p a function in S which does not turn to zero
.anywhere. We have
lim <p (A) f (A) P (A) hn = <p (A) q.
n~co
But cpfP E .98 N (R). And therefore by virtue of Theorem 4.1 we have
<p (A) q = lim cp (A) f (A) P (A) hn = limP (A) <p (A) f (A) hn = 0.
n~oo n~oo
Hence it follows that q = 0, due to the invertibility of the operator

and tj:le theorem is proved.
·<p (A),
Theorem 4.3. Let f E .98 N (R) and let P be a polynomial. Then the
<{)perator P (A) f (A) is an extension of the operator f (A) P (A).
Proof. If h belongs to the domain of the operator P (A), then
co
f (A) P (A) h = V12n

-co
I T(t) e-iAtp (A) h dt,
where Tis the Fourier transform of the function f. Hence by using

Lemma 4.1 and the fact that the operator P (A) is closed we obtain
co
f (A) P (A) h = Vi~n J f(t) P (A) e-iAthdt =

-co
co
= V12n P (A)! J T(t) e-iAth dt = P (A) f (A) h.

-co
Therefore the operator P (A) f (A) is an extension of the operator

j (A) P (A).
Now let h be an arbitrary element of the domain of the operator
f (A) P (A). Then there exists such a sequence {hn} of elements of
the domain of the operator P (A) that
::-;-;-:---=--;-:-:-
lim hn = h, :}im f (A) P (A) hn =!(A) P (A) h.
n-+oo n-+oo
Since f (A) is a homomorphism and f (A) P,_(A) is a restriction of

the operator P (A) f (A) we obtain
lim f (A) hn = f (A) h, limP (A) f (A) hn = f (A) P (A) h.
n~co n~co
Hence due to the closed nature of the operator P (A) it follows that
P (A) I (A) h = t (A) P (A) h,
Q.E.D.
CH. I. FUNG'l'IIQNIS OF A HEGULAR OPEHATOR 177
Lemma 4.3. Let / 11 / 2 , / 3 be functions belonging to !J1J N (R) and let

P and Q be polynomials, where
/1 (x) P (x) = ! 2 (x) Q (x) + / (x).
3
Then for any h E D P(A)/i(A) nD Q(A)fz(A) the following equality

is valid:
P (A) / 1 (A) h = Q (A) 12 (A) h + /3 (A) h.
Proof. Let a function ep E S not turn zero at any point. Then
according to Lemma 4.2 the operator ep (A) is invertible; it suffices
to show that
ep (A) [P (A) / 1 (A) - Q (A) 12 (A)- 13 (A)l h = 0.
The functions epP and epQ belong to !J1J N (R), besides
ep (A) P (A) c: [Pep] (A), ep (A) Q (A) c: [Qep] (A).
Therefore
ep (A) [P (A) / 1 (A) - Q (A) / 2 (A) - ! 3 (A)] h =
= [(Pep) (A) ! 1 (A) - (Qep) (A) 12 (A) - ep (A) 13 (A)l h = 0,
since the operator [Pep] (A) 11 (A) - [Qepl (A) 12 (A) - ep (A) 13 (A)
has the symbol ep (PI1 - Ql 2 - / 3 ) = 0, Q.E.D.
Now we shall define the increasing functions of a generator.
Definition. Let g (x) = f (x) P (x) + 11 (x), where I E !J1J N (R),
/1 E !J1J N (R) and where P is a polynomial. Set
g (A) h dcr f (A) P (A) h + f 1 (A) h.
Note. Element h of Bm!d belongs to the domain of the operator
g (A) if and only if there exists such a representation of the function g
g (x) = I (x) P (x) +/ 1 (x),
where j, 11 E !J1J N (R) and P is a polynomial, that
hE Df(A)P(A)'
From Lemma 4.3 and Theorem 4.3 it follows that the definition
is correct in the sense that the operator g (A) does not depend on the
type of representation of the function g in the form IP 1 • In the +/
case when g E .91J N (R) the definition is obviously in line with the
definition of the homomorphism g (A) mentioned in the preceding
section. In the case when g is a polynomial the definition is also in
line with the definition of a polynomial of an operator. In fact, let
g (x) = I (x) P (x) + l1 (x),
12-01225
where f, i 1 E !B N (R) and where P and g are polynomials. First of

all, note that the degree of P is not lower than the degree of g
since the functions I and it are bounded. Therefore, the domain of
the operator P (A) is contained in the domain of the operator g (A).
It must be shown that the domain of the operator I (A) P (A) does
not exceed the domain of the operator g (A). Let h be an arbi~rary
element of the domain of the operator 1 (A) P (A). Then there exists
such a sequence {hn} of elements of the domain of the operator P (A)
that lim hn = h. Hence the sequence {I (A) P (A) hn} is fundamen-
n-+oo
tal. But hn E D g CA>· Consequently, according to Theorem 4.3 and
to Lemma 4.3 the following equality is valid:
g (A) hn = I (A) P (A) hn +
f1 (A) hn.
Hence the sequence {g (A) hn} converges. Since the operator g (A)
is closed, hE Dg !A>• Q.E.D.
Suppose now that the closed operator A is invertible.
Theorem 4.4. Let <p E !B N (R) and let
'¢ (x) = x-k<p (x),
where k is an integer, also belong to !B N (R). Then
11' (A) = A-h<p (A).
Proof. We have
<p (x) = xh'¢ (x).
Therefore
<p (A) =A k'¢ (A), '¢(A) =A -h<p (A).
Now define the operators which correspond to symbols with sin-
gularities at zero.
Definition. Let g (x) = f (x) x-k, where f E !B N (R) and k is an
integer. Set
g(A) ctcr A-hf (A).
Verify the correctness of the given definition. Let / 17 i2 E ,(jlj N (R),

x-k1 it (x) = x-k2 / 2 (x). Show that
A -ktft (A)= A -hzf2 (A).
Let k 2 > kt. We have / 2 (x) = x"z-ht ft (x). Consequently
f 2 (A) =A"2-k1 ft (A). By multiplying both members of this equality
by A -1<2 from the left we obtain the relationship in question.
1CH. I. FUNC'TrONIS OF A REGULAR OPERATOR 179
Lemma 4.4. Let f E .91! N (R), h ED A-k· Then

I (A) A -kh = A -kt (A) h.
Proof. We have
By multiplying both members of the obtained equality by A -k

we assert. the proof of the lemma.
Definition. Let g (x) = / 1 (x) P (x) 2 (x) x-k +/
fs (x), where +
fu / 2 , fs are functions belonging to .91! N (R), P is a polynomial, k is
an integer. Set
g (A) h = g1 (A) h + g 2 (A) h + fs (A) h,
where g1 (x) = / 1 (x) P (x), g 2 (x) = / 2 (x) x-k, h is an arbitrary
n
element of Dg!CA> Dg 2<A>·
Verify the correctness of this definition. Let
g (x) = g1 (x) +g 2 (x) +Is (x) = g4 (x) +g 5 (x) + /s (x),
where
g1 (x) = /1 (x) P (x), g 2 (x) = / 2 (x) x-k,
g 4 (x) = / 4 (x) Q (x), g6 (x) = /5 (x) x-l~
j = 1, 2, 3, 4, 5, 6, P and Q are polynomials, k and l

fi E .91! N (R),
are integers. First of all, note that the functions g 2 and g 5 may be
modified in a bounded neighborhood of zero with a view to obtain
functions belonging to .91! N (R). Since
gl- g4 = g5- gz + fs + /3
and since the function g1 - g 4 coincides with some function belong-
ing to .91! N (R) in a neighborhood of zero we obtain
gl (x) - g4 (x) = !1 (x),
where / 7 E .91! N (R). Therefore
g1 (A) = g4 (A) +
!1 (A).
Further, the following equality is valid:
g 2 (x) = g 5 (x) + /8 (x),
12*
where / 8 = / 3 + / 6 - /1 Egg N (R). Let

hE D A-"MA> n D A-z 15<Ar
Set
q = [g 2 (A)- g5 (A)- / 8 (A)] h =[A -"/ 2 (A)-A - 1/ 5 (A) - /8 (A)] h.
It is required to prove that q = 0. We shall prove this statement

in the following way: let cp E S be a function which does not turn
zero anywhere and let cpm (x) = xmcp (x), m = 1, 2, .... We shall
show that the operator cpm (A) is invertible and that cpm (A) q = 0
for a sufficiently large m. We have cpm (A) = A"'cp (A). For this
reason
(cpm (A) u = 0) '* (A"'cp (A) u = 0) '*
=?- (cp (A) u = 0) =?- (u = 0).
Here we have made use of the invertibility of operators A and cp (A).

The invertibility of operator cpm (A) is proved.
Let m ~max (k, l). Then the functions x-"cpm (x) and x- 1cpm (x)
belong to the space !}g N (R). By making use of Theorem 4.4 and
of Lemma 4.4 we obtain
cpm (A) q = [Am-kcp (A) / 2 (A) -Am-zcp (A) / 5 (A)-A"'cp (A) /s(A)] h=O.
Q.E.D.
Example. Let V x = i
-def v-
I x I for x < 0. Then the function
xN+l Vx is N +1 times continuously differentiable along a real
straight path. Let 1 = e1 (x) +
e 2 (x), where e1 E C't' (R), e1 (x) = 1
for sufficiently small values of x. Set / 1 (x) = e1 (x) xN+l Vx,
/ 2 (x) = e 2 (x)!V x. Then / 1 , / 2 E 3Y N (R1 ) and
Vx = gt (x) + g2 (x),
where g1 (x) = x-N-1/ 1 (x), g 2 (x) = xj 2 (x). Consequently, for any
h E D gt<A) n D g2{A) the following formula is valid:
VAh=A-N-ift (A)h+A/2 (A) h.
Problem. We have defined the infinitely differentiable functions

of a commutative operator which are increasing at a rate not exceeding
the degree of the argument. Show that they form an algebra with
11-structure, where the algebra A is an algebra of commutative un-
bounded operators, and the set M consists of generators and real
functions of them.
CH. 1. FUNGTTIONIS OF A REGULAR OPERATOR 181
Sec. 5. Homomorphism of Asymptotic Formulas. The Method of

Stationary Phase
Let A be a closure in Bmld of a generator of degree N and let k
be an integer. From Theorem 4.1 it follows: if <p, 'ljJ E 9J N (R) and
<p (x) - 'ljJ (x) = 0 (x-k) in a sense that the function xk [<p (x) - 'ljJ (x)]
belongs to 9J N (R), then for any g E Bmid the vector <p (A) g -
-- 'ljJ (A) g belongs to the domain of the operator A k. This is a ho-
momorphism.
To illustrate the point we shall consider the so-called method of
the stationary phase. Let <p E CO' (R), let .'F E Co (R) be a real
function, and let the equation § ' (s) = 0 have a unique solution
so on the support of the function <p, where .liT (s 0 ) =/= 0. Consider 11
the integral
00
I(:r)= J cp(s)eixEl'<s>d£. (5 .1)
The method of the stationary phase provides an asymptotic expan-

sion of the function I (x) in negative powers of x:
n . 1
I (x) = eixff(t;o) ~ aj (cp, §)X -J- 2 + Rn+i (x),
4=0
where Rntl (x) = 0 (x-n-~) . We reproduce here the derivation of
the formula of the method of the stationary phase to obtain the
estimate of the remainder Rn+l in the required form.
Now make the substitution of variables in the integral (5.1)
[ff (s) - JF (so)l = t 2 sgn L'F (s)l II
and denote (!) = I x I, a = sgn [x.F (s 0 )]. We obtain

11
I
00
I (x) = eixEl'(so) eiwat 2ljl (t) dt, (5.2)
where 'ljJ (t) = <p (s (t)) dsldt, 'ljJ E CO' (R). Transform the formula
(5.2) in the following way:
00 00
e- ixEl'(so) I (x) = ljJ (0) j eiwat2 dt + j eiwat2 [ ljJ (t) -ljl (0)] dt =
= v: eicrn/4ljl (0)- 2i~a r eiwcrf2ljl1 (t) dt,
where 'ljl 1 (t) Here we have used integration by

= :t ¢ (t) -;-¢ (O) •
parts. It is easy to see that outside the support of function 'ljJ (t)
the following equality is valid:

'i't (t) = 1Pt~O)
00
By transforming the integral J eiwat 2 '¢1 (t) dt in the same way as

-oo
00
the integral J eiwat 2 '¢ (t) dt we obtain the formula

-oo
e-ixfh~o>J (x) =-.V/ :n: eian/1•

w
('I' (0) + _i_
2wcr lPt (0)) +
00
+ [ z!cr ]2 J eiwat2¢2 (t) dt,

-oo
where \jl 2 (t) = :t 1Pt (t)-;1Jlt (O). Continuing the process we obtain
e-ixff(~o>J (x} = v: eian/4 ~

j=O
( r
z!a "'j (0) +
+ ( z!a r+i J 00
-oo
eiwat2¢n+l (t) dt, (5.3)
where
1h. (t) =_!:_ ')Jj(t)-')Jj (0) j=i, 2, ... , ¢o='l't·
'rJ+I dt t '
To prove the formula (5.3) we must verify that for any integer j
. 1PJ (t) -'Pi (0) 0
l lnl t = . (5.4)
t-+oo
By using the explicit formula for '¢1 (t), where t is large it is easy
to establish by induction that for t ~ supp \jJ we have '¢J (t) =
= t- 2 P (t- 1), where P is a polynomial in j. The formula (5.4) is
thereby derived.
Thus we obtain the following expansion:
00
Vx e-ixff(£o) ~ eixfilmcp md~ = Vn ein/4 sgn fil"<~o) X

-00
n
X~ (zxsgn~"(GoJj'i'J(O)+rn+!(x), (5.5)
j=O
r n+! (x) = Vx ( i sgn ;"(Go) r+ 1

00
J
-oo
eixt2 sgn ffH (6ul¢n+l (t) dt, (5 .6)
CH. I. FUNCT:UONIS OF A REGULAR OPERA'J.10R 183
where for negative x

V -x =def v-
1 x eirr./2 sgn fl" <so).
1
We cannot replace the variable x directly by the generator A in

formula (5.5) because the function !i
does not belong to JSJ N (R).
Let us correct the expansion (5.5) in the following way. Let rad x
be an infinitely differentiable function which coincides with Vx
outside a finite neighborhood of zero, and let Pi (q;) be an infinitely
differentiable function which coincides with x-' outside a finite
neighborhood of zero. Then
00
(rad x) e-ixfl(so) .\· eixfl(£)<p (~) d~ = Vn eirr./4 sgn fl"<~o) X

-00
n
; sgn fl" (so) ) i -
X~ ( 2 Pi(x)\jlj(O)+rn+1(x), (5.7)
i=O
where rn+l (x) = rn+l (x) for sufficiently large I X I·
Note that all terms in (5. 7) except rn+l (x) are obviously infinitely
differentiable. Hence, it follows that the function rn+l (x) is also
infinitely differentiable.
Now evaluate the remainder term rn+l (x) at infinity. For the
sake of certainty set sgn !F" (~ 0 ) = 1. Continuing the expansion
(5.5) we obtain for any integer m ;;;::: n + 1:
+Vii: ein/4
i=n+2
m
~ ( ~ r xn+l-i'l'i (O) +
00
+ Vx ( ~ )m+ 1 xn-rn ) eixt2'i'm+1 (t) dt. (5.8)

-00
The first term in the right-hand member of (5.8) is a constant and

m
this is why it belongs to !J3> N (R). Next all terms in h in the
i=n+2
right-hand side of (5.8) are square-integrable at infinity and all their
derivatives have the same property. Finally, consider the function
f (x) = -v- xn-m Jf etX. 12'iJm+1 (t) dt= v

X
00
def 11-
X xn-mg (x).
-oo
184 OPERATIONAL :\IETHiODS
The function g is bounded. Estimate its derivatives. We have

00 00
g' (x) = J it2eixt2¢m H (t) dt = - 2~ \ eixt• [Nm+t(t) ]' dt.
We have carried out integration here by parts and have made use
of the fact that lim t'\jJm+ 1 (t) = 0. For calculating g" (x) we note
f..,.oo
that the function [t'IJlm+t (t)l' and the function '\jJm+ 1 (t) can be repre-
sented in the form of the product of t- 2 by a polynomial in t- 1 in
such a way that
00
g
" ( ·)
X = (2x)2
1
Jreixt2 dt
d t d
dt flh
'Ym+l
(t) dt •
-oo
By calculating the higher derivatives in the same way we verify that
the function g (x) is infinitely differentiable and gU•J (x) = 0 ( 1x 1-k).
Hence, it follows that for sufficiently large m the function is integ··
rable at infinity for k ::( N 1. +
From the obtained estimates it follows that the function xn+~i=;;+ 1 (x)
can be represented in the form of the sum of a constant and a function
in w~+ 1 (R). Consequently, it belongs to ~ N (R). Obviously the
function rn+ 1 itself belongs also to ~ N (R). Thus for any g E Bmtri
rn+1 (A) g EDl +1 .
Now consider the left-hand member of formula (5.7). This function
belongs to~ N (R) because the right-hand member has this property.
Set
def
rad x-I(x)=K(x).
The following expansion is valid:
K (A)= Vn e'
. (Aczz: <sol+ n sgn ....qz, <sol ) X
<7 4
X ~
n
j=O
[ i sgn ~"(sol J'i'i (0) Pi (A)+ rn+t (A),
where the range of the operator 1;.+ 1 (A) is contained within the
domain of the operator An+l. Now set
g (x, ~) = (x!i)2 <p (S) eixff'(sJ, L (x) (x!i)2 I (x).
The function g (x, s) generates the function G (s) with values in

w~+ 1 (R) c ~ N (R). The function G is continuous. Indeed, tho
following estimate is valid:
ai+lg (x, s) - 0 ( 1 )
as axi - (x2+ 1)1/2 '
C:H. I. FUNC'TIIQNIS OF A REGULAR OPERATOR 185
which is uniform in 1;. Consequently

00
('
L (A) = J g (A, £) d£.
-00
Next, it is not difficult to verify that IE$ N (R). This means that
the latter equality may be rewritten in the form
00 00
(A+ it I (A)= ~ g (A,£) d£,

2 I (A)= (A+ i)2 ~ g (A, £) d£.
Show that K (A) = rad A ·I (A). LethE E. Then for any integer
k and any x E $ N (R) we have
00
x(A)(A+i)kh= -/2n ~ x(t)(A+i)~<U(t)hdt=(A+i)~<x(A)h.

-00
Hence, in particular, it follows that X (A) h belongs to the domain

of any degree of the operator A. Set
radx
radk (x) = - - ;
(x+i)k
for sufficiently large k we have radk E $ N (R). For any hE E, rad A X

X I (A) h = radk (A) (A + i)k I (A) h = radk (A) I (A) (A + i)~<h.
On the other hand,
(A + i)-k K (A) = rad~< (A) I (A);
for this reason
rad A ·I (A) h = (A + i)-k K (A) (A + i)kh = K (A) h.
Now let h be an arbitrary element of Bm 1n, h1 E E, h1 -+ h for·
j-+ oo. Then
K (A) h1 -+ K (A) h, I (A) h1 -+ I (A) h for j-+ oo,
consequently,
K (A) h =lim rad A· I (A) hi= rad A· I (A) h,
j ....... oo
Q.E.D. \Ve can now write the formula

00
K(A)=radA·(A+i) 2 ) g(A, £)d1;.

-00
186 OPERATIONAL METHIO[)S
Hence it follows that for any h E E

00
K(A)h=radA·(A+i) 2 ) g(A, s)hd£=

-oo
00 00
=radA Jg(A, £)(A+i)2hd£=radA J cp(s)eiAs;-<s>hd£.

-oo -oo
To extend the equality obtained for h E E
J cp (£} eiAff<s>h d£
00
K (A) h = rad A
-oo
for the case of an arbitrary h E Bmid we shall first show that the
function
s--+ cp (£) eiAff(s)h (5.9)
with values in Bmid is continuous along the real straight path. It is

sufficient to show that the mapping
t --+ eiAth, t ER
is continuous. Due to the boundedness of the operator eiAt this
corresponds to the following statement: for any e > 0 we may find
a positive number 6 so that for I 't I < 6 we have
II eiAc h- h II < e.
Choose such h 0 E E that the equality

II (eiA• -1) (h -ho) II< ~ (5.10)
is valid for any 't of a fixed bounded neighborhood of zero. Set q0 =

= (A + i) 2 h 0 • Then
II (eiA<- 1) h 0 II = II (A + i)- 2 (eiA<- 1) q0 II-+ 0
for 't--+ 0, because the function G with values in Wf+1 (R), defined
by the formula
[G ( 't) l (x ) = (x+
1 ( ix't 1]
i)2 e -
is continuous. Choose 6 > 0 so that for I 't I < 6 inequality (5.10)

is valid, and the inequality
1\ (eiA•-1) ho II<~
is also valid.
CH. I. FUNC'TJlONIS OF A REGULAR OPERATOR 187
Then for I 't' I < 6

II (eiAt_1) h II<;: II (eiAt_1) ho II+ II (eiAt -1) (h-ho) II< e.
Thus the continuity of the function (5.9) is proved. Consequently,
the integral
00
~ cp (£) eiA£PC£>h d£ .
understood as the limit of integral sums in the norm of Bm 1d, exists.
Further, if hn-+ h for n -+ oo, then II cp (£) eiAEP<s> (hn - h) liB . -----'.:>
mid
s
-+ 0 strives uniformly in to zero as n-+ oo since
IJ eiAEP<S> II~ (1 +I.¥(£) I)N.
Let h be an arbitrary element of Bm!d and {hm} be a sequence
of elements of E which converges to h. Then
lim I (A) hm =I (A) h,
00 00
r~~~ j cp (£) eiAS:(s)hm ds = .\ cp (£) (?iAEP(slh d£.

-00 -00
00
But ) cp (£) eiAs:'<s>hm ds--:- I (A) hm· For this reason

-00
00
I (A) h = ) cp (£) eiAY:<s>h d£:

-00
hence it follows that

00
K (A) h = rad A ) cp (£) eiAEPCslh d£.

-00
Denote
.(
a 1 cp,!f'-
6E') _ v-neit sgnEF"<sol [ i sgn EF"
2
(£0) ]i '". (O)
_ 'fJ ,
where the functions 1Jli are defined from the functions cp and .¥ des-
cribed above. The numbers ai (cp, .¥) depend only on values of the
functions cp, !F and their derivatives at the point so up to some order
dependent on j.
Thus we obtain the following statement.
Theorem 5.1. Let cp E Co

(R) and let.¥ be an infinitely differentiable
real function on R; it has a unique critical point so on the support of
the function cp, where !T" (£ 0 ) -=1= 0. Next let rad x be an infinitely
differentiable function which coincides outside a bounded neighborhood
188 OPERATIONAL :IIETHODS
of zero with the function

x"21 def { v-x for x :;>-0,
VIXT eht/2 sgn Y:"(so) for X< 0,
and, finally, let pi (x) be infinitely differentiable functions wh(ch coin-
cide outside a bounded neighborhood of zero with functions x- 1 •
Then for any integer n ;:;;:: 0 the following ·expansion is valid:
00
rad A ~ !Jl (£) eiA?i'<s>h d£ =

-00
n
= eiAS"(s) ~ ai (!Jl, .¥) Pi (A) h + qn+b (5.1'1)
i=O
where .qn+ 1 belongs to the domain of the operator A n+I.

Note. Let the operator A be invertible. Then in (5.11) we may
replace Pi (A) h by A-ih on condition that hE DA-n. To prove thiE·
statement it is sufficient to show that for any h ED [i
(Pi (A)- A-i] hE D An+l·
Denote [P.i (A)- A-i] h = q. Then Aiq = [Aipi (A)- 1] h. Tho

function x 1 Pi (x) belongs to $ N (R) and is identical to unity in tho
neigh):wrhood of infinity. Therefore for any integer m the function
xm (x1 pi (x) - 1) belongs to$ N (R) so that the range of the operator
Aipi (A) - 1 is contained in DAm' This means that for any m we
have qED Ai+m; in particular, qED An+1·
Sec. 6. The Spectrum of a Generator

Let A be a generator of degree N.
D~finition. The minimal open subset of a real axis with the following
property
(!Jl E CO' (R) and supp !Jl c p (A)) ==? (!Jl (A) = 0)
is called the resolvent set p (A) of the operator A. The set a (A) = R""' p (A)
is called the spectrum of the operator A.
Problem. Prove that a (A) =1= 0.
· Theorem 6.1. Let 'A E p (A). Then an operator (A - 'A)-1 exists
which is defined everywhere on Bmid and satisfies the estimate
II (A- 'A)-1 II~ f (d),
GIL I. FUNC'TrONS OF A REGULAR OPERATOR 18!)
lj)here d is the distance between the point 'A and the spectrum of the opera-
tor A and f is some non-increasing (numerical) function.
Proof. Let cp be an infinitely differentiable function and cp (x) =
= (x- 'A)-1 for I x - 'A I ;;:,:8, where 8 <d. Show that
cp (A) = (A - 'A)-1, II (A- 'A)-1 II~ t (d). (6.1)

From the equality
cp (x) (x - 'A) = 1 - X ('A),
where suppx c [/..,- 8, 'A+ 8] c p (A), it follows that
(A - 'A) cp (A) = 1,
cp (A) (A - /..,) h = h, Vh E DA.
The latter two equalities show that
cp (A) = (A- 'A)-1 .
Estimate the norm of the operator (A - 'A) - 1 = cp (A).
II (A- 'At 1 11 ~ -/2n: II cp 11$N<R>~ -/2n: II cp llwf+\R) ·

The norm of the function cp in W:l" +1 (R) depends on 8 and on the
method of continuing the function (x - 'A) - 1 into the interval
I x - 'A I < 8. Let <!> 8 be the set of all possible cp, where 8 is fixed.
Then
II(A-'At 1 II~V·12n: O~e<d
inf llcr\1 N+t
IV 2 (R)
=/(d).
q>Ee

Theorem 6.2. Let /.., E R. If (A - /..,)- 1 is a homomorphism, then
'A E p (A).
Proof. Let 0 < 6 < 1 . Show that cp (ii) = 0 for
II (A-A.)- 1 11
any function cp E Co (R) which satisfies the condition supp cp c
c- ('A -
6, 'A +
6). Let cp be the described function. Then
II cr (A) II= II (A- 'At~< (A-'A)~<cr (A) II~
~ V12n: II (A- 'At1 Ilk 111P~t llwf+1<R>'
where tpk (x) = (x - 'A)k cp (x).
It is easy to obtain the following estimate
l\1iJ~t\l N+1
w (R) =0(6/t).
2
Consequently,
II {j) (A) II = o [(8 II (A- t.)-1 ll)ltl-+ o for k-+ oo.
This means that II cp (A) II = 0 and the theorem is proved.

From Theorems 6.1 and 6.2 it follows that the spectrum of a gene-
rator coincides with the spectrum of its closure in Bmld (in the com-
mon sense), i.e., (A- E p (A))~ [(A- A-)- 1 is a homomorphism].
Theorem 6.3. For any function f E 1$1 N (R) with the support in
p (A) the the following equality is valid:
t (A)= o.
Proof. First of all, suppose that the support of the function f
is compact. Let {fn} be a sequence of infinitely differentiable func-
tions which converges to f in 1$1 N (R), g is a function belonging to
Co (R) with the support in p (A), which is equal to 1 on supp f.
Then gfn E Co (R), supp (gfn) c P (A) and gfn-+ f in 1$1 N (R).
Since g (A) fn (A) = 0, it follows that f (A) = 0.
Reject the assumption that the set supp f is compact. Let {cpn}
be a sequence of functions belonging to Co (R) such that for any
h* E B* and for any h E B the following relationship is valid:
lim h* (cpn (A) h)= h* (h).
n-+oo
Set fn (x) = f (x) (j)n (x). Then supp fn c p (A), so that fn (A) = 0.
For any h* E B* and any h E B we have
0 =lim h* Un (A) h)= lim h* (cpn (A) f (A) h)= h* (/(A) h).
n-+oo
Hence it follows that f (A) = 0.

It is clear that Theorem 6.3 remains valid if $ N (R) is replaced
by$ N (a (A)), whereby the estimate from above of the norm of thP
operator qJ (A) will be improved, and the kernel of the homomorphism
M will shrink. However, this mapping is not a monomorphism (i.e.,
not one-to-one) either, even when $ is finite-dimensional and N
is the exact degree of the operator A. The fact is that different J or··
danian blocks of a matrix A may generate groups of homomorphisms
with an arbitrary order of growth.
Example. Let :\1 , 1. 2 be different real numbers and let

CH. I. FUNC'TFONIS OF A REGULAR OPERATOR 191
Then
e-iAt = (e-~7-.11 e-~1>2t - u~-i1>2t) ,

0 0 e-i7-.2t
so that A is a generator of first degree but not a generator of zero
degree. For any function cp E !B 1 (R) we have
rp (A. 1) 0 0 )
cp (A)= ( 0 cp (A.z) cp' (A.z) •
0 0 (jl (A. 2)
The spectrum a = a (A) of the operator A consists of two points:
"- 1 and A2 • Let cp E COO (R), A2 ~ supp cp, cp (A.l) = 0, cp' (A- 1 ) =I= 0.
Then cp (A) = 0, but II cp llfE 1<a> > 0 (here the norm of the class of
.equivalency, which the function cp belongs to, is denoted by II cp 11$ 1<a>
as usual). Despite the fact that the homomorphism Ma: !B N (a (A))-+
-+ ~N is not a monomorphism, we obtain a sufficiently good esti-
mate for II cp (A) II from above.
Definition. An eigenelement of the operator A at the point A. is called
such a vector g ED-;.. that (A - A.) g = 0. The associated element of
degree k of the operator A at the point is called such a vector g E D-h
A
that (A- A.)h g = 0 (but (A - A.)h-t g =1= 0). The eigen and asso-
ciated elements of the operator A will be called the e.a. elements of the
operator A.
If the operator A has .only isolated points of the spectrum and
the number of linearly independent e.a. elements corresponding
to each point of the spectrum is finite, then the operator A is said
to possess a discrete spectrum.
Problem. Let h be an eigenvector of the operator A at the point A..
Prove that e-iAt h = e-i'Mh, where h E D:4.
Lemma 6.1. Let A be an isolated point of the spectrum of the generator
A of degree N. If g is an e.a. element of the operator A at point A.,
then (A- A.)N+i g = 0.
Proof. Let g be an e. a. element of the operator A at the point A..
Then for some k
(A - A.)h g = 0. (6.2)
Show that if k > N +
1, then for (6.2) it follows that (A - 'V)h-t g =
= 0. Suppose that k > N +
1. Let { Xn} be such ·a sequence of in-
finitely differentiable functions that Xn (A.) = 1, and that supp Xn
192 OPE.RATIONAL METHODS
is contained in the interval n

1- , 1-) .
n
(A- - A+ -
Therefore, the
support of a function Xn contains only one point of the spectrum of
the operator A for sufficiently large n. Besides, let
lim II l¥n llssN<R> = 0,
ll-->00
where '¢n (x) = (x - A) 11 - 1 Xn (x). It is easily seen that such a sE~

quence exists. We have
II Xn (A) (A- A) 11 - 1g II-+ o for n-+ oo.
On the other hand, the vector

h = (A - A)k-1 g
is an eigenelement of the operator A since
(A-A)h=(A-A) 11 g= o.
This means that
00
Xn (A) h = V12rt ) Xn (t) e-iAth dt =

-CO>
00
- -1- ) ~ (t)e-,f..thdt-x
xn . (A)h-h
- V2rt - n - '
-00
i.e.,
Xn (A) (A- A)k-1 g =(A- A)k-1 g.
From (6.3) it follows that (A- lv) 11 - 1 g = 0. By induction we obtain

that (A - lv)N+ 1 g = 0.
Theorem 6.4. Let the spectrum of the generator A consist of isolated
points Az, i = 1, 2, .... Then the set of all e.a. elements of the operator
A is complete, i.e., the linear span of the set is dense in Bmid•
Proof. Let xi be an arbitrary function of C'i) (R) with a support
containing only one point A; of _the spectrum of the operator A and
let x' be a function such that x' (x) = 1 in some neighborhood of a
point A;. Then (A - A;)N+ 1 Xi (A) = 0. Indeed, let {X~} at a fixed
i be the sequence considered in Lemma 6.1 (fork= N 2, }, •= A1), +
where x~ (x) = 1 in some neighborhood (dependent on n) of t lw
point A;. With sufficiently large n we have
x~ (A)= xi (A),
CH. I. FUNCTl'ONIS OF A REGULAR OP.ERA'l10R 193
and
\\(A-Ai)N+tx~(.A)II--+0 for n--+oo,
Q.E.D.
In Sec. 3 it was proved that the linear span of the set of all vectors
of the form q> (A) h, q> E Co (R), hE Bmid is dense in Bmid· This
means that for the proof of the theorem it suffices to verify that for
any h E Bmid, q> E Co (R) the vector q> (.A) h can be represented in
the form of a finite sum ~ gh where gi satisfies the equality
i
("4 - Ai)N+t gi = 0. Let {Qi} be the open covering of the sul!port
of the function cp, where Ai ~ Qi for i =I= j, Ai E Qi. And let {x'} be
a C"' -partition of unity which is subordinate to the covering* {Q;}.
Then cp (A) h = 21 cp (A) x; (A) h. Set gi = cp (A) xi (A) h. We
have
(A- Aj)N+t gi = <p (A) (A- Aj)N+tXi (A) h = 0.
From the proof of Theorem 6.4 it also follows that for every isolated
point Ai of the spectrum of the generator A there exists a nonzero
eigenvector of the operator A at this point (i.e., Ai is an eigenvalue
of the operator A). In fact it suffices to verify that there exists a
nonzero e.a. element of the operator A at the point Aj· Let (A) =1= 0 (such a
function exists in accordance with the definition of a generator).
def - ·
Then there exists h E Bmid, such that g = q> (A) h =I= 0. If x' is
the function considered in the proof of Theorem 6.4 then xi (A) X
:< q> {A) = q> (A) so that g = "t (A) q> (A) h. Hence it follows that
- N+1 -
(A - Ai) g = 0. If A is an eigenvalue of the operator A then
II f (.If) II ~ I f (A) I for any f E 1# N (R). In fact if g is a nonzero
eigenelement of the operator A at the point A, then f (A) g = f (A) g.
Thus, in the case when the spectrum of the operator consists of
isolated points we obtain the following estimate from below for
\If (A) II:
II t (A) II>- sup If (A) I· (G.4)
I.Ecr(A)
* This means that Xi E coo, xi= 0 outside Q;, 0 <Xi< 1 and 21 Xi =1 on

i
supp <p •. In treatises on mathematical analysis it is proved that such a set of
functions exists.
13-01225
194 OPERATIONAL METHJODS
Sec. 7. Regular Operators

In this section we extend the results of the preceding sections to
the case of n-pi:uameter groups.
Let E be a vector space and let { U (t) }, t E Rn be an n-parameter
group of homomorphisms E-+ E:
u (t + 't) = u (t) u (1:).
Next, let E have two norms II lit and II 11 2 , moreover, II lit~ c II liz,
c = const, and let
au (t) h = - iA .u (t) h
uti 1 '
where the derivative is understood as II lit and where Ai: E-+ E

are linear operators. We shall call A = (At, ... , An) a generating
set of operators of degree N = Nt, ... , Nn (with respect to thegroup
{ U (t)} and norms II lit, II liz), if
n N
II u (t) h Ib ~ c1II (1 + I t i I) i II h 111.
i
Theorem 7 .1. If A is the generating set of degree N with respect to the

group { U (t)} and norms II 111 , II 11 2 , then in E there exists such a norm
II llmid that A is the generating set of degrees N with respect to the same
group and norms JJ·IImid, JJ·IImid·
Proof. Define ll·llmili by the formula
II h llm!d = sup . II U (t) h II~ (7.1)
IER n IJ (1 +I tt I) i
i
Obviously the axioms of the norm have been carried out and II h llmid~
~ II h 11 2 • From (3.3) it follows that
ll hJJ m!d"""tERn
:::::::sup ct(1+ltllNIIhll1 =c Jlhll·
(1+lti)N 1 1
For this reason the derivative of any function with values in E with
the norm of the form II IJ 1 is simultaneously the derivative with the
norm of the form II llmid· Further,
II u (t) h llm1ct = su II u (t+'t'l h 112 ~
(t+lti)N tER~ (1+j't'I)N (1+Jti)N""""'
< {IIU(t+'t')hll2 (1+1t+'t'I)N }

"""t,~~in (1+lt+'t'I)N" (1+Jtj)N(1+l't'I)N •
CH. I. FUNCTl!ONIS OF A REGULAR OPERA'J.10R 195
Since the latter factor does not exceed 1 it follows that

II U (t) h llm!d~(1 +It I)N II h llmld•
Q.E.D.
Let Bmid be the completion of E in the norm 11·\lmid· The extension
of the operator U (t) to the homomorphism Bmid -+ Bmtd will be
denoted by e-iAt, where At = A 1 t1 + ... + A 11 t 11 • By A =
'= (A h . • . , A 11 ) we shall denote the set of closures of the operators A i
in Bmid· Note that the operators Ai exist (see Theorem 3.1).
Definition. Let cp E §J N (Rn) and let A be a generating set of degree
N. Define the homomorphism cp (A) : Bmid-+ Bmtd by means of the
formula
cp (A} h = (2n)-n/Z J7P (t) e-iAth dt,
Rn
where 7P = F-1c.p. '{he function cp will be called the symbol of the opera-
tor cp (.A).
Obviously the following estimate is valid:
II cp (A) II ~c 1\ c.p II~N(Rn), c = const,
where S& N (R11) = $ Nt• • • • , Nn (R11) .
Thus the mapping M : cp-+ cp (A) is a homomorphism of the
Banach space $ N (R11) into the Banach space Op (Bruid) of the
homomorphism Bmid -+ Bmid·
Theorem 7 .2. The homomorphism M is a homomorphism of algebras.
The proof is analogous to the proof of Theorem 3.1.
The following theorem is well known.
Theorem 7 .3. Let (A 1 , A 2) be a generating set in a separable Hilbert
space H which consists of self-adjoint operators. Then
11/(Al, A2) II~ II/ llc(R2)·
Definition .. The minimal open set in Rn with such a property that
(cp E Ca (R11) and supp cp c p (A))=>- (cp (A) = 0)
will be called a resolvent set p (A) of the generating set A. The set
a (A) = R11"'-.P (A) will be called a spectrum of the generating set.
Definition. The operator of the form T = A 1
->
+ iA 2, where (A 1 , A 2 )
is a generating set of degree N = {N, 0}, and
e-i(Attt+.A2t2) = 6 -iAitte-iA2t2 == e-Ditt2e-iA1t2
will be called the regular operator of degree N.

13*
Lemma 7 .1. There exists a closure T of a regular operator Tin Emid·

Proof. Let f (x) = lxl 2\ 1 , x E R2 and let T = A1 + iA 2 ,
where A = (A 1 , A 2) is the generating set. The operator f (.A) has
an inverse. In fact, let { !pn} be such a sequence of functions of Co (R 2 )
that
limh* (!pn (A) h)= h* (h)
n-+oo
for any h E Emid, h* = E~id· If f (A) h = 0, then

'i'n (A) f (A) h = o,
where 'i'n (x) = Cf!f (~~) = !pn (x) (I x 12 + 1).
But 'i'n (x) f (x) = !pn (x), for this reason ¢n (.A) f (A") = !pn (A).
Therefore for any h* E E~id
h* (!pn (A) h) = 0 = h* (h),
hence it follows that h = 0.
Consider now the operator f (A) T. This operator has a closed
·
extensiOn g (A-) , where g (x) = lxiZ+i
x1+ixz . Thus
(xn-+ 0 and f (A) Txn-+ z) ==?- (z = 0).
Let Xn-+ 0 and Txn-+ y. Then f (A) Txn-+ f (A) y = 0. Since the
operator f (A) is invertible it follows that y = 0, and the lemma is
proved.
Theorem 7 .4. The spectrum of a closure T of a regular operator
T = A1 + iA 2 in Emid coincides with the spectrum of a generating set.
The proof is analogous to the proof of Theorems 6.1 and 6.2.
Theorem 7 .5. Let T be a regular operator and let the spectrum of its
closure T in Emid consist of isolated points. Then the system of e.a.
elements of the operator T is complete in Emld·
The proof is analogous to the proof of Theorem 6.4.
Lemma 7.2. LetT be an operator in a normed space E, and {Ea.}
be a family of vector subspaces of E such that
(a) Ea. are finite-dimensional, dim Ea. ~ N +
1, where any system
{ga.} is such that when ga. E Ea., g =1= 0, it is linearly independent;
(b) Ea. are invariant with respect to T.
Then shrinkage T 0 of the operator T on E = U Ea. is a regular
operator of degree N (with respect to some pair of norms).
CH. I. FUNCT~ONIS OF A REGULAR OPBRA'l10R 197
Proof. Let T a be the shrinkage of the operator Ton Ba : T a = T P a

where Pa : E-+ Ba is the projector from E onto Ba. Without rest-
riction of generality one may say that the normal Jordanian form
of the matrix T a consists of one Jordanian block (otherwise one
may represent Ba as the direct sum of invariant subspaces of the
operator T a)· Thus, let T a be reduced to the form:
'Aa 1 0 . . . 0)
(
~ .'A~ .1. 0. ·. .. ... ~ .
0 0 . . . . . Aa
Consider (N 1) X (N + +
1) matrices A~ and A~ which have
r);
the following forms respectively in the same representation:
( R~A>:A.
0 . . . Re 'Aa 1
~
0
.
Tml.. .. ·
(ImO'Aa
Im 'Aa
) . ~.
0 Re 'Au
It is obvious that A~ is a generator of degree N and A;;. is a scalar
operator so that A;;. commutes with A~, and A~ is a generator of
zero degree. The generating set (A~, A;;.) generates the group
-iA~t'-iA~t"
e -iAat -e
_
.
In fact, let f.t be a .real number and A be a Jordanian block of the
size N +
1:
f.t 1 0 0
0 f.t 1 0 0
0 . . 0 f.t 1
0 . . . . . f.t
Then the vector-function u (t) = e-iAth, h E cN+i represents a
solution of the Cauchy problem
.!!!!:.. = -- iAu u (0) = h
dt ' '
or, in the extended form,
du1 • .
----at= - ~f.tUi- w2,
duN . .
( i t = - tf..tUN-WN+i•
duN+i .
--cu-= -~f..tUN+h
Uj(O)=hj, j=1, ... , N+1,

198 OPERATIONA L METHODS
where UJ> hi are the components of vectors u and h. By solving the

equations successively from the lower one upwards we obtain
N+1-j
UJ (t) = 2J
k=O
e-JJ.tPJ, k (t) hi+k•
where PJ.k (t) are polynomial s of degree k (it is not difficult to cal-
culate them precisely). Hence, it follows that
II e-iAa.th lln~Ca. (1 +It I)N II h lin,
where Ca. is a constant dependent on a. Besides, T a. = A~ iA~.
This decomposition corresponds to the decomposition of the opera-
+
tor T 0 :
T 0 =A'+ iA ",
where A' = 2J A~Pa., A"= 2J A~Pa.
The pair (A', A") induces the group
U (t'' t") h = 2J e-i(A~t'+A~t"> Pa.h.
Introduce in E the norm II lit:
def " '
II h lit= LJ Ca. II Pah lin•
It is clear that II lit~ II lin·
The following estimate is valid:
II U (t', t") h lin ~ (1 +
I t' I)N II h lit·
Thus (A', A") is a generating set of degree (N, 0) with respect to
norms II 111 , II lin. and the theorem is proved.
From this theorem the theorem given below follows directly.
Theorem 7 .6. Let T be a closed operator in the Banach space B with
a discrete spectrum having a complete system of e.a. elements, the operator
T being without any associated elements of the order exceed~ng N.
Then the restriction of the operator T to a linear manifold dense in B is
a regular operator.
Sec. 8. The Generalized Eigenfunctions and Associated Functions

We shall start with a physical analogy. In quantum mechanics
the state of a system of particles with n degrees of freedom can be
described by a function of the space L 2 (Rn) while operators on L 2 (Rn)
correspond to physical values. The logic of quantum mechanics
provides the possibility of the decomposition of an arbitrary element
of L 2 (Rn) in "eigenfunctions" of an operator corresponding to any
physical value. (The inverted commas are used here because the so-
called eigenfuncti ons of physics do not necessarily belong to L 2 (Rn).)
CH. I. FUNCTFONIS OF A REGULAR OPERA'l10R 199
For example, let n = 1 and let A be a multiplication operator

by the independent variable of L 2 (Rn). This operator corresponds
to one of the most important physical values, namely, the coordinate
of a particle. The equations
(A - A,) 1Jl = 0
for eigenfunctions of the operator A have no ordinary solutions. In
physics one assumes, however, that 61. is a generalized eigenfunction
of a multiplication operator by an independent variable, and so one
writes
x& (x - A,) = M (x - /.,). (8.1)
The equality (8.1) may be interpreted in the following way. Consider
the closure A - A of the operator A - A, as an operator which acts
from the space W2 8 (R) to the space L 2 (R), where s is sufficiently
large. Then
(A-A,)6A=0.
Note that the equality A&A = M>. is not valid, since &A~ L 2 (R).
The statement that A - A, =1= A - A, is justified because the mul-
tiplication operator by a scalar is not bounded as an operator acting
from Wi" (R) to L 2 (R).
Next (in physics) the following expansion is used:
· \jJ (x) = .\ \jJ (A,) 6 (x- /.,) dA, (8.2)
or, in a more general form, the following expansion is written:

\jJ (x) = ) c (A) l!J>- (x) dt.,, (8.3)
where {'\JA (x)} is the family of generalized eigenfunctions of the .
operator of some physical value. Since \jJA (x) is feaLured in (8.3)
as a function of the parameter A, it is quite natural to assume that
the operator (A - /.,) acts on the space of functions of the argument A,
i.e., to consider A as a multiplication operator by an independent
variable.
in the present section we shall obtain the generalization of (8.2)
for the case of the expansion of an arbitrary vector of Bmid in gene-
ralized associated eigenfunctions of a regular operator.
Let us proceed to the definitions and notation. Consider
Hom (.93' N (Rn), C). By definition, Hom (.93' N (Rn), C) is the space
ofboundedfunctionalsin !B N (Rn), i.e., the space S6''N (Rn) for whose
elements the following notation was agreed upon:
f[g]=) f(A,)g(A,)dA, fEHom[!BN(Rn), C], gES6'N· (8.4)
200 OPERATIONAL METH10•DS
As it was agreed above, this integral which has meaning only for
continuous f (A.) is formally written down for any functional f, f (A.)
being called a generalized function.
Let G (A.) be a continuous function of A. with values in B. Then the
integral
) G (A.) f (A.) dA., f (A.) E ~N (Rn)
Rn
determines the element of B so that the operator G

Gfdet IG(A.)f(A.)dA.
Rn
is an element of the space Hom (~ N (Rn), B). For any element

G E Hom (~ N (Rn), B) we agree to write formally
Gf = ) G (A.) f (A.) dA..

By analogy with (8.4) ~e shall call G (A.) a generalized function of A.
with values in B. In accordance with this definition, ordinary gene-
ralized functions are generalized functions with values in the complex
plane.
Let T be a regular operator of degree N and Bmid be the corres-
ponding intermediate Banach space, and let A. = At +
iA- 2 E C.
T<k> = (T- A.)k,
which acts from Bmid to Bmid· We shall consider it as an operator
acting from the space of generalized functions of A, with values in
Bmid to the space of continuous functions of 'A with values in Bmid
so that the values of the operator (T - 'A)k should belong to Bmid
for any fixed 'A.
Thus, we shall considet (T - 'A)k cter T<k> as an operator acting
from Hom (~ N (R 2), Bmtd) to C (R 2, Bmtct), where 'A = At + i'A 2 =
= ('AI> 'A 2). We shall make the meaning of this statement more pre-
cise. Denote by means of T~k>
Hom(~ N (R2), Bm!d)-+ C (R2 , Bm!d)
the operator defined by such continuous functions with values in
Bmid that the mapping
A.-+ (T- A.)k f ('A)

is a continuous function in R 2 with values in Bmid· The operator is
acting according to the formula
T~f (f..) = (T- A.)k f (A.).
CH. I. FUNCTOONIS OF A REGULAR OP.ERA'IiOR 201
We shall extend the operator T~k> up to the operator T<k> as fol-

lows. Let {/n} be a sequence of functions belonging to D T(k) such
0
that there exists a distrib11tion f with values in Bmld possessing
the property to make for any function '\j) E C';' (R 2)
lim ) fn (A-) \jJ (A-)"dA, =) f (A-) \jJ (A-) dA,.
n-+oo Rn R2
Let lim T'ok>t n =g. Then set r<k>t =g.
Verify the correctness of the given definition. Let In E DT(k)

0
and
for any '\j) E C';' (R 2)
lim \ fn (A-) ljJ (A-) dA, =0.
n-+OQ R2
Then allow for the existence of lim T~k>fn = g.
n-.oo
We must show that g = 0. For any '\j) E C';' (R 2):
lim ) (T- A_)k fn (A-) ljJ (A-) dA, = J
(A-) '\j) (A.) dA-.
-00~ ~
Let P be a smooth sufficiently rapidly decreasing function, which.
is strictly positive, and let T = A 1 +
iA 2 , where (A 11 A 2 ) = A
is the generating set, which determines the regular operator T. Set
k
q (x) =P (x) (x 1 + ix 2 - A-)k = ~ qi (x) J..k.
i=O
Then the operator [P (A)l - 1 q (A) is a closed extension of the operator

(T - A-) 11 , the operator [P (A)] - 1 being closed as an operator inverse-
to the homomorphism. For this reason
lim
n-+oo
J(T- A)k fn (A-) \jJ (A-) dA, =
R2
k
=[P(A)r lim 1 .2} Jqi(A)A_kfn(A.)lp(A.)dA-=0

n-+oo j=O R2
for any '¢ E C';' (R 2 ). Hence, it follows that g = 0.

Definition. The kernel of the operator T<k> will be called the system
of generalized e.a. functions of the order ~ k of the operator T.
Definition. The system ~ of generalized e.a. functions of the order
~k of the operator T will be called complete if for any vector h E Bmid
there exists a generalized function G E ~ and a function f E $ N (R2}
202 OPERATIONAL METHIO!DS
.such that
h= ~ f ('A) G ('A) d'A.
R2
Theorem 8.1. The system of the generalized e.a. functions of the

.order ::::;;;; N + 2 of a regular operator T of degree N is complete.
Lemma. Let <Jln (x), x E R 2 be the following 6-formed sequence
cpn (x) = n 2cp (nx), cp EC0 (R), \ <p (x) dx = 1.
"
R2
'Then
Proof. Denote 'l'n (x) =(xi+ Cix 2)N+2 <Jln (x). We have 'l'n (x) =
=n-N1j) 1 (nx). Consequently, lPn (p) =n-<N+Z):;j;i { ~), where lPn =
= F- 1'l'n· Let f be an arbitrary continuous function in R 2 which
:satisfies the condition sup If (p) I ~ 1. Then
PER2 (1 +I Pi DN '"""
112\Pn (p) f (p) dp I= n-N-11 J2lPi (p) f (np) dp I::::;;;;
~n-N-111'1' II sup lf(np)l ~

'""" 't'i .9$N,o(R2)PER2(1+1Pd)N"'
~ -N-1jj II (1+nltl)N -1jj'h II ~N,o(R2)"

"'n 1!'1 .98N,O(R2)~~f (1+1ti)N =1t 't'i
Hence, it follows that

H'l'n 11.98N, o(Rz):::;;;n-lll'l'tii.9$N, o(Rz) -r 0
for n -roo.
And the lemma is proved.
Proof of Theorem 8 .1. Consider in the space 119 N 0 (R2 ) the operator
T with the domain D = GO' (R2 ) acting accordi~g to the formula
Tf (x) = (x1 + ix 2 ) I (x).
It is not difficult to see that T is regular, its space Bmid coinciding
-with 119 N.o (R2). Now show that the identity operator
IE Op (119 N, 0 (R2)) =Hom(~ N, 0 (R2), $ N, 0 (R 2))
belongs to the kernel of the operator r<N +1). Let <p be a monotonously
non-increasing infinitely differentiable function defined in [0, oo)
CH. I. FUNCTIONS OF A REGULAR OPERATOR 203
and vanishing with all derivatives in [1, oo) and let ) £cp (s) ds= 2~ •
0
Set
CJJn (x) = n 2 cp (n I x 1).
Every A. E R 2 will have a corresponding mapping
n (A.) : X -+ CJJn (x - A.).
Obviously, <Dn is a continuous function with values in §J N.o (R 2).
Further, let {en} be a sequence of functions in C0 (R 2) such that
en (x) = 1 for I x I < n and 0 ~ en (x) ~ 1. Set ·
In (A.) = en (A.) <Dn (A.).
Let 'ljJ be an arbitrary function in C0 (R 2). Consider the sequence of
integrals
JIn (A.) \jJ (A.) dA. = J<Dn (A.) en (A.) \jJ (A.) dA..
R2 R2
For suffi.ciently large n

) [In (A.)- I (A.)]\jl (A.) dA. = ) <Dn (A.) \jJ (A.) dA.-\jJ def 'IJln -\jl.
R2 R2
Show that II 'IJln - 'ljJ II w 2N+1 (R2) -+ 0 for n-+ oo. This will mean in
particular that 'IJln-+ 'ljJ in §J N.o (R 2). We have
~'n (x) = JCJJn (x- A.) ljJ (A.) dA. = JCJJn (y) 'ljJ (x- y) dy. (8.5)
R2 R2
From (8.5) it is seen that 'IJln (x) uniformly converges to 'ljJ (x) along
with all its derivatives. Since the supports of functions 'IJln belong
to some compact set independent of n it follows that 'IJln-+ 'ljJ in any
Sobolev space.
Now consider the sequence of functions
Gn = f<t'+ 2)In; Gn (A.)= ('f- A.)N+ 2 en (A.) <Dn (A.)
or in the amplifi.ed form
Gn (A.) (x) = (x1 + ix 2 - A1 - iA. 2)N+2 en (A.) CJJn (x - A.).
It is evident that
II Gn (A.) llssN, o<R2)~ II Gn (0) II!SN, o<R2) =II hn llssN, o<R2>'
where hn (x) = (x1 +
ix 2 )N+ 2 CJJn (x). It is easy to verify that II hn II-+
-+ 0 for n-+ oo. Hence, it follows that II Gn (A.) llssN,o(R2)-+ 0
uniformly in A. for n -+ oo so that

j<N+2)/ =0.
Thus, the system of generalized e.a. functions of the order ~ N +2

of the operator T is complete. Actually for any IE fB N,o (R 2 )
I= 1I (A.) I (A.) d.A..
R2
Let T now be an arbitrary regular operator T = A 1 + iA 2, where

A = (A 11 A 2 ) is a generating set and let h be a vector in Bmid· Con-
sider a mapping
n: A.-+ CJJn, ~(A) h, where CJJn, ~ (x) = CJJn (x- A.).
nis obviously a continuous function with values in Bmid· Set
Kn (A.) = en (A.) n (A.).
Let'IJl, as before, serve as an arbitrary function belonging to Co (R2).
Consider a sequence of integrals
1Kn (A.) 'IJl (A.) dJ.. =) <Dn (A.) en (A.)

R2 R2
\jJ (A.) df... =gn.
Let K be the following distribution with values in Bmid

1K (A.) I (A.) d.A. =I (A) h.
R2
(8.6)
Show that
!~ gn = !~~ 1<Dn (A.)
Rn
\jJ (A.) dJ.. = 1K (A.)
Rn
\jJ (A.) df... = \jJ (A) h.
We have
1<Dn (A.) 'IJl (A.) df... 1CJJn, ~(A) (A.) df...] h = 'IJln (A) h,
R2
= [
R2
\jJ
where 'l'n (x) 1CJJn (x- A.) (A.) df... so that

= \jJ
R2
- -
'ljln-+ 1jJ in fB N, 0 (R2) and 'ljln (A) h-+ \jJ (A) h
in Bmtd·
Now consider a sequence of distributions gn = TbN+ 2> Kn:
gn (A.)= (T- A.)N+ 2 en (A.) <Dn (A.) = (T- A.)N+2 en (A.) CJJn, 1. (A) h.
CH. I. FUNC'TJ'ONtS OF A REGULAR OPERATOR 205
It is obvious that
II gn (A.) IIBmtct ~II (T- A.)N+t CJln. ~(A) II= II 'i'n. ~(A) h 11.
where
'i'n, ~ (x) = (xt + ix2- At- iA.2)N+2 CJln (x- A.).
For this reason
II gn (A.) IIBmld~ll 'i'n, ollssN, O(R2) II h IIBmld -+0 for n-+ oo,
which means that TN+t K = 0, so K belongs to the family of gene-
ralized e.a. functions.
In accordance with (8.6) the vector h may be represented in the
form
h =--= JK (A.) ·1 dA.
R2
and the theorem is proved.
Sec. 9. Self-Adjoint Operators as Transformers in the Schmidt Space':·

Let A and G be self-adjoint operators in the separable Hilbert
space H, and let T be a Schmidt operator, i.e., T E B 2 (H), where
B 2 (H) is a set of operators which have a bounded Schmidt norm.
One may compare operators A and G in the space B 2 (H) (let us
denote them by 1 and G) acting according to the rule
AT defAT, TED ... = {S EB 2 (H): AS EB 2 (H)},
A
GT det TG, T E Da ={S E B 2 (H): SG E B 2 (H)}.
Theorem 9.1. The operators A and Gare self-adjoint in the Hilbert

space B 2 (H) and commute in some dense set.
Proof. First of all show that D~ n D ..... is dense in B 2 (H). Let
A G
En (A) be an operator with the symbol e (x + n) e (n - x) (E~ (M
for A, = -n, 11. = 2n) and let {e~n>} be the complete orthonormal
system in En (A) H, where {ein+v} contains {e~n'} as a subsystem.
Let En (G) and {gln'} be similar projectors and orthonormal systems
for the operator G. Then the set {e;} of all vectors of the type ein.
forms a complete orthonormal system in H, and the same is valid
for vectors gt'. Indeed, any vector in H may be approximated in
* We shall cite some known results here which are necessary for the succeed-
ing sections.
norm by linear combinations of vectors of the form cp (A) h, where

cp E C';' (R) and a vector of the form cp (A) h belongs to En (A) H
for sufficiently large n.
We have for any T E B 2 (H):
II TEn (G)-T II12(H) =II T II12(H)- ~II Tg~n)W-+0
•
for n-+ oo. Similarly for n-+ oo
II En (A) T-T II~2(H)-+0,
because this is a sequence of remainders of the convergent series
2J
i,
j
I (Teit e1) 12 • Thus the set K of operators of the form En (A) X
X TEm (G), T E B 2 (H) is dense in B 2 (H). But the setD is contained

in D_,.
A
n D..- because the operators A En (A) and Em (G) G are bounded.
G
This completes the proof of the density D_,. n D..- in B 2 (H).
A G
Let {Ti}}': 1 be such a complete orthonormal system in B 2 (H) that
Tj ED- n D+-. j = 1, 2, .... It is obvious that A GT, =GAT,
A G
so that A and G commute in a set which is dense in B 2 (H). This
n
set is a set of linear combinations of the form
.
2J
j=i
a1T i·
Prove that A and Gare self-adjoint operators. Let T ED ..... , i.e.

A
ATE B 2 (H). Then for SED_,. we have (AS, T) 2 =(AS, T) 2

A
=
=(S, AT) 2 =(S,ATh, i.e., TED- and A*T=AT. Prove the
A*
inverse inclusion: D_,. c: D_,..
A* A
Let T ED_.. , i.e., there exists such an operator WE B 2 (H) that
A*
(AS, T) 2 = (S, W) 2 for any SED-. Choose in H such a complete
A
orthonormal system {e 1 } that ei E DA, j=1, 2, .... Set Tiiu =
= (u, e1) ei. Then Ttiek = {)ikei so that
00
II Til 11: = 2J 116tke, 11

k=1
2 = 1.
It is easy to see that operators Tii, Tkz with non-coincident pairs of

subscripts are orthogonal in B 2 (H). Thus {T 11 } is an orthonormal
system in B 2 (H). It is easy to verify that this system is complete in
B 2 (H). We have II AT 1, 11 2 = II Ae, II < oo so that Tii E D 1 . Next
we have the relationships
(ATu, T) 2 = (TIJ, W) 2 , i, j = 1, 2, ....
CH. I. FUNCTJ10N<S OF A REGULAR OPERATOR 207
By definition of the scalar product in B 2 (H) the latter equalities

may be rewritten as follows
(Aeit Tei) = (eit Wei);
Hence, it follows that
ATe, = Wei, i = 1, 2, ... ,
i.e., AT = WE B 2 (H). This completes the proof of the self-adjoint-
ness of the operator A.
For the proof of the self-adjointness of the operator Gconsider a
complete orthonormal system {ej} in H which consists of vectors
belonging to D 0 , and let
Tiiu = (u, e~) eit 't/u E H.
Determine the adjoint operator (TiiG)* of the product TiiG.
Since the range of the operator Tii is contained in D(j we have
(TijG)* = GTij = GT}i·
For this reason, if T E"D(j* then there exists such an operator
WE B 2 (H) that
(TijG, T) 2 = (T*, (TijG)*) 2 = (T*, GTii) 2 = (Tij, W) 2 •
Hence
(ej, TGei) = (ej, Wei), i, j = 1, 2, ... ,
i.e., the operator TG has a closure which is equal to WE B 2 (H).
This means that T ED-. Prove the inverse inclusion fl.._ c fl.._
G G G*
and the equalityG*S = GS for S ED-. This will complete the proof
G
of the statement that Gis a self-adjoint operator.

Let T E fl.._, i.e. TG E B 2 (H), where (TG)* E B 2 (H). Determine
G
the action of the operator (TG)* on the vector ei. We have, for any
u EH:
(TGu, ei) =
(u, (TG)* ei).
For this reason for u E D 0 we get:
(Gu, T*ei) = (u, (TG)* ei).
Hence, it follows that T*ei E D(j and (TG)* ei = GT*ei. Taking
this into account for any S ED- we obtain
G
(S, GT)z=(S, TG)z=((TG)*, S*)z=

=~ ((TG)* e;., S*ei.) = ~ (GT*e~~., S*e~~.);
II. II.
208 OPERATIONAL METH10DS
(GS, T) 2 =(SG, T)z=(T*, (SG)*)z=

= ~ (T*e/t, (SG)* e/t) = 2Jk (T*e!t, (GS*) e/t).
k
Hence it follows that TED ..... and G*T = GT. The statement is
G*
proved.
The operators A and Gform the groups {e-i.At} and {e-iat} which
act aooording to the formula
.... .,_.
e-iAtT = e-iAtT, e-WfT = Te-iGt. (9.1)
For example, let us prove the second equality. Consider the two
functions
u 1 (t) = [e-iGtT]* h,
u 2 (t) = (Te-iGt)* h, T*h ED(;.
We have u1 (0) = u2 (0) = T*h. Next
du
a/ = (- iGe-,GtT)* h = i l(e-t
+- .+- .+-
0
• +-
fT) G]* h = iG (e-tGtT)* h=iGu 1 (t);
lduz a eiGfT*h-
i t -- dt - z'GeiGfT*h-
- z·Gu 2 (t) ,
so the functions u1 and u 2 , satisfying the same equality ~~ = iGu
and one and the same initial condition, are identical. Since the set
of operators T satisfying the condition T*h E D(j for the set of vectors
h, dense in H, is dense in B 2 (H), it follows that the formula
e-tiitT = Te-iGt
is valid for any T E B 2 (H).
From (9.1) it follows that the set K of operators of the form En (A) >C
X TEm (G), which is dense in B 2 (H) (where En (A), Em (G) are
the above-considered projectors), is invariant with respect to the
groups {e-iAt}, {e-iGt}, where
~ +- +- --+-
e-iAte-iG't = e-iGTe-iAt.
Consider the two-parameter group {e-iAt-iG"}:
+- def
-+
e-iAt-iG't = --+- .,_.
e-iAte-iGT.
For any T E K we have
-
a .__,. .+-
e-tAt-tGTT = _ iAe-tAt-tGTT,
__,.. . __,.. ::--
at
a__,.. ... .___,.. ...
_ e-iAt-iGTT = _ iGeiAt-iG'tT.
8-r;
........
Thus (A, G) is a generating set of operators.
CH. I. FUNCTIIQNlS OF A REGULAR OPERATOR 209
Theorem 9.2. Let /E~ 0 (R 2 ), f=F- 1/. Then
112~ lzf(t, -r)e-iAtTe-iG"tdtd-rllz:'(;llfllc<R2>1iTIIz·
The proof follows from the formula
2~ ) f(t, -r) e-iAtTe-iG't dt d-r=

n2
= 2~ i
ll2
f{t, -r) e-iAt-icq dt d-r = f (A, G) T.
Problem. We have defined infinitely differentiable functions of

commutative operators, whose growth rate does not exceed the degree
of the argument. Show that they form an algebra with the ~A-struc
ture, where the algebra .Jt is an algebra of commutative unbounded
operators, and lhat the set Jl;f consists of generators and their real
functions.
H-01225
II. CALCULUS OF NONCOMMUTATIVE
OPERATORS
In this chapter we shall develop a 1-1-structure for the case when the
set jl![ consists of two noncommutative vector generators A 11 and
A 2 *. Without additional assumptions it is impossible to extend
this development to a case when M consists of a greater number of
noncwnmutative unbounded generators. For this reason the main
"working" formulas of the theory of 1-1-structures (the formula of
commutation and the K-formula) should be specifically proved with
additional assumptions even for the case of two noncommutative
operators because these formulas contain their commutaLor as well.
These difficulties, of course, do not exist for bounded operators.
In the latter case the given patterns of functions of ordered operators
coincide with 1-1-structures.
By relying on the logic of Chapter I we present an independent
construction of functions of ordered operators here and it is only in
Sec. 9 that we shall construct the growing symbols leading us to
1-1-structures.
Sec. 1. Preliminary Definitions

Let {B,J be a family of Banach spaces satisfying the following
conditions (-r varying over the totality of integers):
(a) there exists a linear manifold D, which is dense in B, for
any -r;
(b) (-r <-r') ==?-B-. -<B-., i.e., II·IIB-.;;;::: II·IIBr··
The family {B 17 } will be called a Banach scale and the norm in B 17
will be denoted by 11·11-.·
ThQ linear mapping A : D -+ D will be called a generator of degree
s with step k on the Banach scale B 17 if, for any 't, the operator A is
a generator of degree s with the defining pair of spaces (B.0 B-r;+k)·
* Moreover, we shall concern ourselves with a slight generalization of the
1 2
u-structure since we shall define only the operations 11: (x1 --+ A1 , x2 --+ A 2)
' 1 2
and ~t: (x1 --+A 2 , x 2 --+ A 1).
CH. II. CAIJCULUIS OF NONOOMMU'I\ATIVE OP.ERA"I10RS 211
Denote by {UA (t)} the group of homomorphisms generated by A,

and by UA.-r. (t) the closure of the operator UA (t): B-r.-+ B-r.+k·
Let {T -r.} be a family of homomorphisms B-r.-+ B-.:+ 1, such that
1'-.:D c: D and T 1 h = T-.:·h for hE D for any 't and 't'. The family
T = {T-.:} will be called a translator (or translating operator) with
step l of the scale {B-r.}· In cases when it will lead to no misunderstan-
ding, we shall write Th instead of T 'th.
Let A 17 • • • , AN be generators on the scale {B-.:} with steps corres-
pondingly k1, ... , k N and degrees Sl, ... , SN, and let 1'(1), • • • , J'(N-i)
be translators with steps l17 • • • , ZN-H respectively. Now, to the
operators AH ... , A N• r 0 >, •.. , r<N- 1) we shall add superscript-
ions in the following way:
1 2 3 2N- 2 2N- 1
A1, r 0 >, A2, ... , r<N-1), AN (1.1)
Here we assume that operator A 1 acts first, operator r<1) acts second
and operator AN acts as (2N - 1)th. The set (1.1) will be called
a vector-operator. Note that in general, some components of the
vector-operator may commute and be equal.
Let the vector-operator (1.1) correspond to the following homo-
morphism u't (t) depending on the parameter t E RN:
U(t 1, t2 , ••• , tN)=
= ij AN (tN) J'(N-1)ij AN-l (tN-t) · · · ij A2 (tz) J'(!)ij At (ft) (1.2)
(here and below we shall skip the subscript 't in the notation U -r (t)
whenever possible).
Theorem 1.1. For any hE Bt the function
t-+U(t)h, tERN, U(t)hEB t+kt+ ... +kN+lt+ ... +IN-1
is continuous.
Proof. At N = 1 the statement is obvious. The proof is made by
induction. To avoid the cumbersome details of the derivation, how-
ever, we shall only show the transition from N = 1 to N = 2,
because the transition from an arbitrary N to N + 1 does not differ
in principle from this easier case.
Let N = 2 and denote r< 1> by T. We have
U + (\, t 2 + Bz) h- U (t 17 t2 ) h =
(t1
= U A (tz+ 6z) TU At (tt + 61) h- U A (tz) TU At (tt) h=

2 2
= U A (tz + 6z) T [U At (tt + 61)- U At (tt)] h +

2
+ [U A (tz + 6z)- U A (tz)] TU At (lt) h.

2 2
It is obvious that each of the two latter terms tends to zero in the
norm of the space Bt+kt+ ... +ZN-t whenever 61 and 6 2 -+ 0.
4*
212 OPERATIONAL :\iETHODS
Expand the spaces~ N (R11) constructed above. Denote by C N (R11)

the Banach space of all continuous functions with the finite norm
II f II _sup If (x) I
CN(Rn)-, (1 +I x J)N
and by Cj. (R11) the space conjugated to C N (R11). Fix a linear sub-
space of functionals suitable for our purpose. We shall introduce
the following preliminary definition.
Definition. We shall say that a functional L E C1¥ (R11) is equal to
zero on an open manifold U c: R11 if for any function f E C N (R")
such that supp f c: U the relation is valid: L (f) = 0.
The smallest closed subspace F c: Rn such that L is equal to zero
on Rn - F will be called the support of the functional L and denoted
supp L.
Denote by :£ N a linear subspace of C1¥ (R11) consisting of all
functionals with compact supports. These functionals will be called
finite functionals.
Let 'ljln E C't' (Rn) and the following conditions be satisfied:
0 ~ 'ljln (x) ~ 1 for all x ERn and 'ljln (x) = 1 for x E {x : I x I ~ n }.
It is easily seen that for any L E :£ N there exists n 0 such that for
all f E C N (Rn) the following relation is valid:
L ((1 - 'ljln) f) = 0 for any n ~ n0•
Hence
L (f)= L ('1Jn0 /) + L (( 1-'1Jn 0) f) = L Nnof)
and
II'IJnof llcN(R 11 )~ II f llcN(R 11)"
Consequent constructions are based upon this characteristic property

of finite functionals. Note that the 6-function and functions from
CO' (R11) belong to :£ N· Therefore, as before, we shall denote by
Civ (R11) the strong closure :£ N in C1¥ (R11). Below we shall consider
Civ (R11) only in this sense and not as a strong closure of the subspace
spanned by 6-functions and functions from C't' (R11).
An important property of the functionals of Civ (R11) is given in
the lemma that follows.
Lemma 1.1. Let T E Civ (Rn), then for any e > 0 there exists an
integer A > 0 such that for all n > A and all f E C N (R11) the following
inequality is valid:
(1.3)
Proof. If T E :£ N then the statement is obvious. But if T E Civ (R 11)
and is not a finite functional, then there exists a sequence {T m}m>o c:
GH. II. CALCULUS OF NONOOMMUTATIVE OPElRAT'ORS 213
c X N strongly convergent to T, i.e. for any e > 0 there exists

M > 0 such that for all m > M the following inequality is valid:
II T-T m llcJ'V(Rn) <e.
Let m.0 > M, then
IT ((1- 'iln) /)- T mo ((1- 'iln) f) I< e II f llcN(Rn)"
Next, there exists an integer A > 0 such that supp T mo n
n supp (1 - '¢n) j = 0 whenever n >A. Hence
IT ((1-'IJn) /) l~e II f llcN(Rn)
for any n > A. Lemma 1.1 is proved.
Corollary. For any function f E C N (Rn) there exists a sequence
{cpn}n:>o c C'((' (Rn) such that
lim L (cp 11 ) = L (f) (1.4)
n-+oo
for any L E CF. (Rn). Iff =I= 0 then there exists T 1 E CF. (Rn) such
that T f (f) =I= 0.
Proof. Let L E c-;. (Rn). For any f E c N (Rn) we have
L (f) = L ('¢n/) +
L ((1 - '¢n) /). (1.5)
The function '¢nf is finite and continuous for any n > 0. Conse-
quently, for any n there exists cpn E C~(Rn) such that ll'¢n/- (jln llcN(Rn)~
~1hz.
From Lemma 1.1 it follows that for any e > 0 there exists M > 0
such that I L ((1 - '¢n) /) I < e II f lieN for n > M. By virtue of
(1.5) we obtain
IL (f)- L ( (jln) I~ II L llcJV(Rn) ·1/n + e II f llcN(Rn)" (1.6)
Relation (1.4) follows from this inequality.

If 7=I= 0 then there exists x 0 E Rn such that 7 (x 0) =I= 0. The
functional llxo defined by the formula .
llxo (/) = f (xo) for any f E C N (Rn)
belongs to X N and llxo (f) = f (x 0 ) =I= 0. Hence the proof is obtained.
We shall introduce the operation of convolution of functionals
from C"jy (Rn) with functions from C N (Rn). But first we define some
continuous linear mappings 1:h and "~'' of the space CN (Rn) into
Cx (Rn) (which will be convenient for future usage) in the following
way: for any f E C N (Rn) let
def
Thf (x) = f (x-h)
214 OPERATIONAL METH10·DS
and
~ def
f (x) = f (- x).
Definition. The function F defined by the formula
F (h) = T ((•hi)~),
where T E C"Fr (Rn) and f E C N (Rn), will be called the convolution of

the functional T E Cf,; (Rn) with the function f E C N (Rn) and denoted
by
, dcf
1 *f(h)=T((•hf)~). (1.7)
The function (• hf) v E C N ( Rn) for any h E Rn. Hence the con-
volution is defined on Rn everywhere.
From (1. 7) we hrJVe
IT* f (h) I::::;;; II T llcj\(11 f lieN (1 +I h I)N. (1.8)
Moreover, the convolution belongs to C N (Rn). In fact let

T E Cf,; (Rn), I E C N (Rn) and h 0 be a point in Rn. For any h satis-
fying the inequality I h - h 0 I < 6 the following estimate is valid:
II •hi llcN(Rn)::::;;; II I llcN(Rn) (1 +I hoI+ 6)N.
Consequently, the family of functions (• hf)lh-hol<o is uniformly
bounded and then from Lemma 1.1 it follows that for any e > 0
there exists M > 0 such that for all n > 1H the following inequality
is valid:
IT ((1- 'l'n) (•d))v I < e/2. (1.9)
The function h-+ 'l'n (•d) v considered as a function with values
inC N (Rn) is continuous due to the fact that '~n E C'Q (Rn), and there-
fore for 211 /'llc'}V > 0 there exists 6 > 0, such that
( 1.10)
From (1.5), (1.9) and (1.10) it follows that for any e > 0 there
exists 6 > 0, such that
* f (h) - T * t (h 0 ) I::::;;; e
IT
whenever I h - h 0 I < 6.
The continuity of the convolution is thereby proved.
Moreover, from (1.8) we see that the convolution of any functional
from Cf,; (Rn) with functions from C N (Rn) is a continuous linear
CH. II. CALCULUS OF NONOOMMUTATIVE OPER.~TIQRS 215
mapping from C N (Rn) into C N (Rn), which i~ invariant with respect

to a shift, i.e., for any h' ERn we obtain
-r:h' (T * f (h)) = T * (-r:h' f) (h).
The opposite assertion is also correct: for any continuous linear
mapping A of the space C N (R") into C N (Rn), which is invariant
with respect to a shift, i.e., forj any h E Rn and f E C N (Rn) the
following equality is fulfilled:
A-r:hf = T~tAf; (1.11)
there exists a functional T E C'N (Rn), such that, for any f E C N (Rn),
we obtain
Af = T *f.
Prove the assertion. Note that
(1.12)
Since A is continuous, the mapping f-+ Af (0) yields a continuous
linear functional from C';, (Rn) which will be denoted by T.
Thus
T (1) = Af (0).
But from (1.12) we obtain T * f (0) = Af (0). By replacing f by
-r:hf and taking into account (1.11) we further obtain
T * f (h) = Af (h) for any h E Rn.
The assertion is proved. Note that T may not belong to C"h (Rn).
This assertion, however, allows us to define the operation of the
convolution of functionals of the space Ch (Rn).
Let Tu T 2 E Ch (R"). Then it is obvious that the mapping
Af = T1 * (T2 *f)
is continuous and invariant with respect to a shift.
Thus, as proved above, there exists T E C'N (Rn) such that for
any f E C,y (R11)
T*f= T 1 *(T2*f). (1.13)
Now we may give the following definition.
Definition. The functional T defined by relation (1.13) will be called
a convolution of the functionals T 1 , T 2 E C"h (Rn) and denoted by
def
T=T 1 *T 2 •
By using (1.12) and (1.13) we obtain the relation
T*f(O)=T (/) =Tt((T 2 *f)'') (1.14)

from which it follows that T E C"N (Rn) due to the fact that f-+
-+ (T 2 *f)~ is a continuous linear mapping from C N (Rn) into
C N (Rn) and T 1 E Cjy (Rn).
Thus, we have proved the theorem that follows.
Theorem. The space C1v (Rn) is invariant relative to the operation of
convolution.
Some important properties of the operation of convolution are
given by the theorem that follows.
Theorem 1.2. The operation of the convolution of functionals of the
space C"N (Rn) is commutative and continuous in factors.
Proof. The continuity of the convolution follows directly from
(1.14). In fact,
IT (f) ~~II T1ilck II T2 * f lieN~ II T1 lick II T2llck II f lieN.

Hence
II T1 * TzllcN(Rn)~ll T1llcN(Rn) II T21!cJV(Rn) (1.15)
which means the continuity of the convolution in factors.

Now show the commutativity of the convolution, i.e., prove that
for any f E C N (Rn) ~he equality
T 1 * T 2 (!) = T 2 * T 1 (!) (1.16)
holds.
If we show that (1.16) is fulfilled for any f E C': (Rn), then from
Lemma 1.1 it follows that (1.16) is also correct for any f E C N (R").
For this purpose we shall need the lemma that follows.
Lemma 1.2. Let 'Pe E co;; (Rn) and have the following properties:
'Pe (x) ;;?: 0 for any x ERn and ~ 'Pe (x) dx = 1, 'Pe (x) = 0 for
I x I ;;?: e > 0; then for any cp E C': (Rn) the following relation is
valid:
lim T*('Pe*'P)=T*(jl·
e ~ 0
Proof. From the obvious equalities we obtain

T * 'Pe * cp - T * cp = T * ('Pe * cpcp).
By virtue of the continuity of the convolution we have
II T * ('Pe * (jl- 'P) llcN(Rn)~ II T llcf..r<Rn) II Cfle * (P -- 'PIIcN(n.nr (1.17)
But lim II 'Pe * cp - cp llcN(Rn) = 0. This completes the proof

etO
of the lemma.
CH. II. CAlJCULUS OF NONOO!VDIUT A TIVE OPERoATORS 217
We shall now proceed with the proof of the theorem.

Let {T7}n>o be a sequence of finite functionals convergent to Tlt
and {TT}m>o be a similar sequence convergent to T 2 • By considering
(1.17) we obtain
lim (T~ *ere)* (Tr;' *ere) (/) = Ti * T 2 (/).
EtO
But T1 * ere and TT * ere are finite continuous functions, and for
this reason we have
(T1 *ere)* (TT *ere) (f) = (TT *ere)* (T1 * <pE) (f).
Thus, the commutativity of the convolution is proved, and at the
same time Theorem 1.2 is also proved.
We shall establish another important property of the functionals
of space C~ (R11 ) . Let f E C N (R11) . With no loss of generality one may
set n = 2 here. Bring the function F (x 17 x 2 ) in correspondence with
any function f (x1 , x 2 ) according to the formula
i f (xi, £) d£.
X2
F (xi x 2) = (1.18)
0
Show that the mapping I 2 : f (x 1 , x 2 )-+ F (x1 , x 2 ) given by (1.18)

is a continuous linear mapping from C N (R11) into CN+I (R11) .
In fact,
X2
~ f (xi£) d£ ::::;; II f llcN(R2l I x2! (1 +I x i)N.

0
Consequently,
II F (xi, x2) llcN+i(R2)::::;; II f llcN(R2l· (1.19)
Since the linearity of the mapping I 2 is obvious the assertion is

proved. Define for any functional T E CFH1 (R 2 ) the functional T 0
given by the formula:
T 0 (f) = T (I 2 (f)) for any f E c N (R 2 ). (1.20)
Since I 2 and T are continuous mappings, it follows that T 0 E C'N.
Besides, T E CFHt (R11), hence T 0 E C'N (R11 ) . By considering (1.19)
it is not difficult to note that the following inequality is valid:
II ro llc~(Rn)::::;; II T llclV+i(R")'
Thus by means of relation (1.20) we have defined a continuous
linear mapping l 2 : C~+ 1 -+ Ctv which has the following property:
iff (xlt x 2 ) is a continuous function from C N (R 2) such that
218 OPERATIONAL l\IETH,ODS
1~ 2 (xu x 2 ) E C N (R2 ), then for any T E Cf, + 1 (Rn) we have

f 2 (T) (/~ 2 (x 11 x 2 )) = T (! (x 11 x 2) ..:_ f (x11 0)).
Formulate the obtained result in the general form.
Theorem 1.3. There exists a continuous linear mapping f;:Cf,+ 1(Rn)-+
--+ Civ (Rn) which has the following property: iff E C N (Rn) and
1~ 2E C N (Rn), then for any T E Civ+l (Rn) we have
fi (T) (/~.) = T (f (x 11 ••• , X;, ••. , Xn) - f (x 11 ••• , 0, ... , Xn)),
'
where fi is defined by the formula identical to (1.20).
Now pass on to the construction of spaces !EN (Rn). For this
purpose we shall define the Fourier transform of the functionals of
the spaces Civ (Rn).
Definition. The mapping !F which brings the function
L (t) def 1 , L (e-ixt) (1.21)
(2n)n/2
in correspondence with each L E Civ (Rn) will be called the Fourier

transform of the functionals of the space Cf, (Rn). It is obvious that
the Fourier transform is a linear mapping. We shall establish the
properties of the Fourier transform of the functionals of the space
Civ (Rn). By C+ (Rn) denote the space CN (Rn) for N = 0.
Lemma 1.3. The Fourier transform of the functionals of the space
c+ (Rn) has no kernel and maps C+ (Rn) into C (Rn).
Proof. Let T E C+ (Rn) and T =I= 0, and let T (e_ux) = 0 for any
t E Rn. Then there exists f E C (Rn) such that T (f) =F 0. Further,
according to Lemma 1.1 there exists a sequence {Qln }n>O c C0 (Rn)
such that lim T (QJn) = T (!). Next for any Qln E C 0 (Rn) there
n~oo
exists q)n E S such that IPn (x) = 1

(2n)n12
.~.1 e-itxq;n (t) dt. The con-
vergence of this integral is uniform.
Consequently,
T (Qln) = 1 T (e-itx) fPn (t) dt = 0.
12 \
(2n)n J
Hence lim T (QJn) = 0. The obtained contradiction illustrates
n~oo
the fact that the Fourier transform of the space C+ (Rn) has no kernel.
Now shpw the continuity of T. For any n > 0 the mapping t-+
-+ '¢ne-ttx ('¢n being defined by Lemma 1.1) is obviously continuous
as a mapping from Rn into C (Rn). Then
T (e-itx) = T Nne-itx) + T (( 1- 'li'n) e-itx).
CH. II. CALCULUS OF NONCOMMUTATIVE OPERaAT'ORJS 219
For any e > 0. there exists A > 0 such that if n >A, then
I T ((1 - 'ljln) e-' 1x) I < e/3 for any t ER11 • At the same time there
exists 6 > 0 such that I T ('\jln (e-itx - e-itox)) I < e/3 for any t
satisfying the inequality I t - t 0 I < 6 for some n > A.
Hence IT (e-itx- e-'tox) I < e whenever I t - t 0 I < 6. Lem-
ma 1.3 is thereby proved.
Now let N > 0. Then the mapping t ~ e-itx is continuous as a
mapping from R 11 into C N (R11). Moreover, for any polynomial
p (x 10 ••.• , Xn) with a degree no greater than N - 1 the mapping
t ~ pe-ttx from R 11 into C N (R11) is continuous and the function
Fp (t) = T (e-itxp (x))
is also continuous for any T E C'lV (R11 ). This assertion follows direct-
ly from the definition of the norm in C N (R 11). If the degree of the
polynomial p (x1 , . . • , Xn) is equal to N, however, then, generally
speaking, F P (t) will not be a continuous function.
If f!})f is a differential operator of degree no greater than N - 1,
then by virtue of the continuity of the mapping t ~ f!})fe-itx we have
T (f!})fe-it"') = :!iJfT (e-itx) = T ((- ixte-itx)
for any T E C'lV (R11 ). This means that the Fourier transform of any
functional T E C'N (Rn) is an N - 1 continuously differentiable
function limited with its derivatives, i.e.,
§ (Cf,) c C(N-1).
Besides, functionals from C'Fl (R 11) are characterized by the lemma
that follows.
Lemma 1.3*. The Fourier transform of the functionals of the space
CN (R") has no kernel and maps C'Fl (R11) into c<N> (Rn).
Proof. The triviality of the kernel in the Fourier transform is
proved in the same way as in Lemma 1.3.
We have already seen that !F (C'Fl (Rn)) c c<N-t)(Rn); there-
fore, it remains to be proved that for any polynomial p of degree N
the function F P (t) = T (e-' 1xp (x)) is continuous for any T E C'Fl (R11).
The mapping from . C N (Rn) into C (Rn) described by the relation
f~ f N is an isometric isomorphism of C N (Rn) and C (R 11).
(1+1 xi)
Hence for any L E C'Fl (Rn) from the formula
L (pe-ixt) = L ((1 +I X I)N pe-ixt ) =

(1+1 X i)N
N ( pe-ixt \ ~ ( p (x) e-ixt )
=(Lo(1+Jxi)) (1+1xi)N J=L (1+lxi)N
220 OPERATIONAL METH;ODS
it follows that [, E C+ (Rn). As in Lemma 1.3 it remains to be shown

~ ( p ( x) e-ixt )
that L N is a continuous function, and this will com-
(1+lxl)
plete the proof of Lemma 1.3*.
Denote by $ N (Rn) the Fourier transform of the space Cjy (Rn)
into c<N> (Rn). As has been proved, the Fourier transform realir.es
an isomorphism between Clv (Rn) and $ N (Rn). Therefore, we may
introduce the norm in $ N (Rn) as follows:
(1.22)
Since Cjy (Rn) is closed in C"N (Rn) it follows that $ N (Rn) becomes
a Banach space with respect to the norm (1.22).
Let ~r be a differential operator with degree no greater than N,
then for any function .

(2:rt) N
For this reason the convergence in $ N (Rn) leads to the convergence

in the norm c<N> (Rn). Thus $ N (Rn) is continuously embedded in
c<N> (Rn).
Let T 1 , T 2 E Cjy (Rn). By definition of the convolution of func-
tionals and by virtue of Theorem 1.2 there exists the Fourier trans-
form of the convolution
!F (Tz * T1) = !F (T 1* Tz) = 1 T1 (T 2 (e-i<x+Y> 1)) = (2nt 12 'f/f2 •
(2n)n12
Next from (1.15) we obtain
~ ~ 1
IITrTzllm
vuN
(Rn)=
(2:rt)
n;ziiTz*Tillc*(Rn)~
N
1
~ (2:rt) n/2 II Tzllc*N ·II T1llc*N ·
This provides the proof of the theorem that follows.

Theorem 1.4. The space $ N (Rn) is a Banach algebra with respect
to the ordinary operation of multiplication of functions and is conti-
nuously embedded in the Banach algebra c<N> (Rn).
CH. II. CALiCULUS OF NONOOMMUT'ATIVE OPERoA:T'ORS 221
Let the function I E C< 1 > (Rn). Introduce the notation for the
difference derivative of the function I with respect to x 1:
{jf , def
-,.- (x 1 , ••• , x;; Xi, xi+l• ... , Xn) =
ux;
j(Xt. ••• ,X;, ..• , X 11 )-f(Xj, ••• , Xj, ... , Xn)
Xi-Xi
With no loss of generality in what follows below, we set n = 1.

In complete analogy with Theorem 1.3 we may obtain the following
X
result: let I E C N (W), then the mapping I : I~ Jf (~, x - ~) d~

0
is continuous and maps linearly C N (R 2) into C N+I (R), and for
any T E C1V+ 1 (R) the functional L defined by the relation
L (f) = T (I (f)) (1.23)
belongs to C!v (R 2 ). Thus we have defined a continuous linear map-
ping from Ch 1 (R) into C1V (R 2 ) provided by the formula (1.23).
Put in (1.23) the function e-it1x 1-it2"'2 instead of f.
Then the function L E $ N (R2 ) and we have
L (t~, t2) = T (I (e-ittx1-it~x2)). (1.24)
But
Consequently,
T(I(e-i+ 1x 1 -it2 x 2))= -iT(t1)-T(t2).
t1-t2
By virtue of (1.24) the function T (t:)- T{t 2) belongs to $ N (R 2),

1-t2
and the mapping from .<J,9 N+l (R 1) into $ N (R 2) given by the relation
<p ~ ~~ (x; x') is continuous and linear due to the fact that the map-
ping from C!v+t (R) into C!v (R 2 ) defined in Theorem 1.3 is continuous
and the norm in $ N (Rn) is determined by formula (1.22). Thus we
have established that the theorem that follows is valid.
Theorem 1.5. Difference differentiation is a continuous linear map-
ping from $ N+I (Rn) into $ N (Rn+ 1).
Very often it is not easy to establish whether a function belongs
to the algebra .'!/l N (Rn). It is well known that for k > ~ N +
the space W~ (Rn) continuously embeds in c<N> (Rn). We have
established that !iJ N (R") is continuously embedded in c<N> (R").

The lemma that follows will be valid.
Lenuna 1.4. The space W~ (R") for k > ~ +N is continuously
embedded in !iJ N (R").
Proof. Let cp E S and f E CN (R"). Then the following inequality
is valid:
n
Jfi;(x)f(x)dx= Jfi;(x)(1+Jxi)N+z+e(l+Jxfl)~~n/ 2 +edx~
~ t llcN<R"> fi; llwf+n/2+e<R"> ··v J +~: D"+e~
11 11 (t
~c II t llcN<R"> II cp II N+~+e •
w2 2 <R"l
Next, S is dense in W~ (R"); consequently for any function cp E

E W~ (R") the above inequality holds and any function cp from
W~ (R") generates a continuous linear functional from Civ (Rn)
described by the formula
Trv(f) dct Jfi;(x)f(x)dx

for any f E C N (R").
Show that if cp E w~ (R") and cp =I= 0 then there exists t E cN (R")
J
such that (j;j dx =1= 0. Suppose the opposite. Then for any function
'ljJ E C0 (R") we have JfP'ljJ dx = 0. The space W~ (R") is continuous-
ly embedded in L 2 (R") and Co
(R") is dense in L 2 (R"). Hence
cp = 0. Considering the inequality obtained above we see that
W~ (R") is continuously embedded in Cjy (R"). Consequently
w~ (R") is continuously embedded in !iJ N (R"). The Lemma is
proved.
Denote by C N [R", B] the set of all continuous functions on R"
with values in the Banach space B for which the norm
dcf II f (x) lin
II f llcN[R", BJ= sup (i +I x I)N
is finite.
Define the operation of the pairing of the spaces Civ (R") and
C N [R", B] as follows.
Let L E Cjy (R") and f E C N [R", B]. Define the linear functional
(L, f) on B* by the formula
(L, f) (h*) = L (h*f). (1.25)
CH. II. CA:UCULUS OF NONOOMMUTATIVE OPER.A:TORS 223
Then
I(L, f) (h*) I~ II L llcN(Rn) ·II h* liB* ·II f llcN[Rn, BJ
and, consequently, (£, /) E B**. The mapping I: C"N (Rn) X
X cN mn,
B]-+ B** defined by the formula (1.25) will be called
the pairing of Cf.r (R") and c N B]. mn,
Lemma 1.5. The functional (L, f) defined by (1.25) for any f E
E cN mn,
B] and L E C"N (Rn) belongs to B (considered as a subspace
in B**).
Proof. It is known that a functional g E B* * belongs to B if and
only if g is continuous in the weak* topology of the space B*.
Let the sequence (hin)m>o c B* converge weakly* to zero. Show
that (£, f) (h~,) also converges to zero, which will complete the
proof of the lemma. From Lemma 1.1 it follows that for any e > 0
there exists Jl![ > 0 such that for n > M we obtain
IL (h* (1 -'\jln) /) I+ e II h*f II ~ e II h* liB• II f II CN[Rn, B]"
Any weakly* convergent sequence is bounded. Consequently for
any e > 0 there exists M > 0 such that for all n > M the estimate
is valid:
I (L, f) (hin)- (L, 'i'nf) (h;';,) I< e. (1.26)
The function 'i'nf is continuous and of compact support. Therefore
there exists a function fn with a finite number of non-equal values
such that II fn - f llcN[R", BJ <e. But since in the weak* topology
h':n-+ 0 it follows that (£, fn) (kin)-+ 0 for m-+ oo. By virtue of
(1.26) we thus obtain that for any e > 0 there exists M' > 0 such
that
I (L, J> (hi,,) I < 3e
for any m > M'. The lemma is thereby proved.
Thus the pairing of Cf;; (Rn) with c N mn,
B] described by (1.25)
is a continuous linear mapping from Cf.r (Rn) X Cn [Rn, B] into B.
In complete analogy with the preceding section we may construct
the algebra :Iff sr, ... , s ,l (RN), where su ... , sN are integers. For this
1
purpose denote by c"t· ... , SN (RN) the Banach space of functions
continuous on RN and having a finite norm
def I g (x) I
II g lie (RN) =sup s s
8 1' .. • , 8N xERn (1+1 Xti) i · ·. (1+1 XN I) N
The closure of continuous functionals finite in C,~, ... , sN (RN)

will be denoted by c;l' ... , "N (RN). Construct $81' .. . , SN (RN)
224 OPERATIONAL :11ETHODS
from C;1, .•.• sN (RN) in a manner similar to the one we have used in
constructing !8 N (Rn) from C"N (R"). The isometric isomorphism,
which is made use of, will be called the Fourier transform and deno-
ted by ,¥.
Sec. 2. The Functions of Two Noncommutative Self-Adjoint

Operators
Let A and B be self-adjoint operators in a separable Hilbert space
H and let T be an operator of the Schmidt class B 2 (H).
In Sec. 9 of Chapter I the pair (A, B) of the operators commutative
in the everywhere dense set defined on a Hilbert space B 2 (H) cor-
responded to the pair of operators (A, B). There exists a homomor-
phism
eft->-+-: C (R 2) -+- Op (B 2 (H)),
A,B
which translates the function f E C (R 2 ) into the operator f (A, B),

II f (A~ B) IIB2 (H) ::::;;; II f llc<R2J· The homomorphism cfh'-+
A,B
..- induces
the mapping
Q/ft 3 2 1 : C (R 2) __,_ B 2 (ll)
A, T,B
for a fixed T by means of the formula

def -> -<-
Q/ft 3 2 d= j (A, B) T.
A,T,B
We shall write Q/ft instead of Q//(3 2 1 and use the notation
A,T,B
2 ( 3 1)
Q/ftj = Tj A, B .
Obviously, the following estimate is valid:
lih(A, B) ~~~l\f\\c(R2)1\T\b,
where \1 II is the norm of an operator H. Thus oft is a homomorphism
2 ( :; 1)
of normed spaces. Note that the definition of the operator Tf A, B
agrees with the definitions of the preceding section: if f E .'J!l 0 (R 2)
then the following formula is valid:
Tt(A, B)h= 2~ .\ (F- 1f)(t 1 , t 2 )e-iAttTe-iBt2hdt 1 dt 2

. R2
or any h E H. Besides, the following equality is valid:
Tt (A) / (A) =ft(A)Tf2(B).

1 2
'CH. II. CADCULUS OF NONOOMMUTATIVE OPERoA'T'ORJS 225
Lemma 2.1. The following estimate is valid:
I Trp3 \A, B1)

I :J
1jJ
(A,3 B1) I 2~ \1 cp llc<R2) I T'IJ
2 (A,:J B1) I 2.
Proof. We have
\\rcr(l, .B)'~'(A, .8)/Jz=llcr(l, B)'IJ(A, B)TII2~

~II cp IIC(n~) II 'I' (A, B) T ll2 =II cp llc(R2) J T'IJ (A, B) //2.
From Lemma 2.1 it follows that the kernel of the homomorphism
.dl is an ideal in C (R 2).
Defmition. The point 'A E R2 will be called a point of the resolvent
set p ( A, T, B of the pair of operators A, B with respect to the operator
~ 2 1) 3 1
2
T if there exists such a neighborhood U oj this point that any function
'lj: E Co (R 2) with the support in U belongs to the kernel of the homo-
morphism cit.
The complement of the set p A, T, B ( 3 2 1) in R2 will be called the
( 3 2 1) 3 1 2
spectrum cr A, T, B of the pair of operators A, B with respect to T.
Theorem 2.1. (1) Any function in C (R 2 ) with the support in
p (A, T, B) belongs to the kernel of the ho~omorphism &It.
(2) If cp belongs to the kernel of the homomorphism al/t then any point
'A E R 2 for which cp ('A) =I= 0 belongs to p A, T, B . ( 3 2 1)

Proof. Let/ E C (R 2), supp fc p f, Note that p (.A, B). (A., T, B)
is the maximal open set in R 2 which satisfies the condition
[cpEC0 (R 2) and supp cpcp A, T, B ( 3 2 1)] =?lTcp
12 (3 1) J
A, B =0.
Suppose the support f is compact. Then there exists such a sequence
.3 2 1)
{In} of functions in Co that supp In c. p (A, T, B and
lim II In (x) - f (x) llc(R2) = 0. For this reason
2 (3 1' 2 (3 1)
Tf A, B) =lim Tfn A, B = 0.
n-+oo
Remove the assumption that supp f is compact. Consider the

linear manifold L of vectors of the form
q = 'ljJ (B) h, 'ljJ E co (R),
15-01225
226 OPERATIONAL ::IIETHODS
which is dense in H. Obviously it suffices to prove that the restriction

2 ( 3 1)
of the operator Tf A, B on L is equal to zero. Let q = 1jJ (B) h,
1jJ E COj (R) and let {fPn} be such a sequence of functions in CO' (R)
that fPn (A) converges point-by-point to the unit operator. We have
2 ( 3 1) 2 ( 3 1) ( 1)
Tf A, B q = Tf A, B 1jJ B h =
= limT<pn
n-+oo
(A) f (A, B) 1jJ (s) h=O,
since the support of the function fPn (x1) f (xH x 2 ) 1jJ (x 2) is compact
and lies in p (.A, T, B). Item (1) is proved.

2 ( 3 1 )
Now let T 0 the
function F = e1j<p belongs to C (R 2 ). But e" = <pF. Hence the func-
tion <p belongs to the kernel of the homomorphism cJ/1. Since the
kernel of the homomorphism cJI is an ideal inC (R 2 ), Te" (A_, B) =0.
Theorem 2.2. The point "A = ("AH "A 2) belongs to p (.A, T, B) if

and only if there exist functions <p 1 , <p 2 E C'f:' (R) such that <p 1 ("A1 ) =I= 0,
<p 2 ("A 2) =I= 0 and <p 1 (A) T <p 2 (B) = 0.
Proof. Let "A = ("AH "A 2) E R 2 and there exist functions <p 11 <p 2 E
E Co (R) such that <p 1 ("A 1) =I= 0, <p 2 ("A 2 ) =I= 0 and <p1 (A) T<p 2 (B) = 0.
( 3 2 1)
Prove that "A E p A, T, B . The inverse statement follows directly
from the definition. It is obvious that there exists such a neigh-
borhood U of the point "A in which the function <p 1 ("A1) <p 2 ("A 2) does
not vanish. Let 1jJ be an arbitrary function in CO' (R 2) with the support
in U. Denote X (x) = ( 1jJ) (x) ( ) • Then X E CO' (R 2) and 1jJ (x) =
Cfli x1 cpz xz
= X (x) <p1 (x1) <p 2 (x 2). Since the kernel of the homomorphism &!I
2 ( 3 1) = 0. The theorem is proved.
is an ideal in C (R 2), T'IJ A, B
3 2 1)
Corollary. (1) a (A, T, B c: a (A) X a (B);
3 2 1)
(2) the point ("AH "A 2) belongs to a (A, T, B if and only if the point
( 3 2 1) .
("A 2, "A1) belongs to a B, T*, A
CH. II. C.AlJCULUS OF NONOOMJ\WTATIVE OPER.A'T'ORlS 227
Let C (p) be a set of all complex functions continuous in R 2 with

supports in the open set p c R 2 • C (p) is an ideal inC (R2). Consider
the Banach factor-algebra C (R 2)/C (p). Denote by a the complement
to p in R 2 • Let C (a) be an algebra of continuous complex functions
in a with the ordinary norm
II f lba> =sup If (x) I·
xEa
If f 1 and f 2 are two functions in C (R 2) which correspond to the
same element of C (R2)/C (p) then the shrinkage of the function f 1
coincides with the shrinkage of the function f 2 into a. This means
that the mapping of the shrinkage
f ~-+ f Ia
induces the mapping :rt 0 : C (R 2)/C (p) ~ C (a), which is the map-
ping onto all C (a) because any function continuous in a may be
extended to the function which is continuous in R 2 •
Lemma 2.2. The mapping :rt 0 preserves the norm. The proof is left
to the reader.
From Lemma 2.2 it follows that :rt 0 is an isomorphism. By means
of this isomorphism let us identify the algebras C (R2)/C (p) and
C (a).
Theorem 2.3. Let :rt : f ~-+ fl (3 2 1) be the projection of C (R2)
a A, T,B
into C (a (.A, T, B) ). Then the following expansion is valid:

(2.1)
Here dl a : C ( a ( A, T, B
:J 2 1))
~ B 2 (H) is the homomorphism of
Banach spaces defined uniquely by (2.1).
The estimates are valid:
II c!/ta II~ II T !12, II c~!talP¢ l12 ~II cp lie (a <~. t, il)) II c~lta¢ ll2·
The proof is left to the reader.
Finally, we shall consider the following theorem which is impor-
tant for some applications.
Theorem 2.4. Let cp be a real-valued continuous function in
a C4, T, B) . Then II Teirp (!J) 11 2 = II T 11 2 •

This theorem follows directly from the unitariness of the operator
eiCfiA,B) (see the preceding chapter and Theorem 4.3).
15*
228 OPERATIONAL METHIODS
Sec. 3. The Functions of Noncommutative Operators

Let the space Cs,, .•. , sN (RN, B) be defined similarly to the defi-
nition of cs (R, B) in Chapter I. If I E c;!, .... SN (RN) and G E
ECs 1, .•• , sN (RN, B) then the integral
~ f (x) G (x) dx
RN
will be understood in the same sense as in the case N = 1.

Let there exist a vector-operator (1.1). Set for any cp E.93' 81 , ••• , sN(RN)
2( 1) 4( 2) 2N-2N_ 1 ( 1 3 2N-1) dcf
T T ... T cp Ai, A2, ••. ' AN h=
Jr ~ <t> u (t) hdt,

= 1N (3.1)
12
(2rr)
RN
where cp = F- 1cp and U (t) is a function of the operator (1.2).
Lemma 3.1. The following estimate is valid:
2(1) 2N-2 ( 1 2N-1) II ::;:::
11 T ... T (N- 1lcp Ai, ••. , AN B, .... Bt+ht+ ••• +kN+lt ... +lN-1""'
~ c II r< 1>IIB-r+ht-+ B-r+ht+lt X ...
• . . X 112Ni211B-r+kt+• . • +kN-1+lt+ .• . +!N-2-+ Br+llt+· . . +hN-1 +It+· .. +!N-1 X
xllcrl1.99 si' •.. , sN ,

where c depends only on -r and on the operators Au ••• , AN·
The proof follows immediately from the inequality
II JN f (x) G (x) dx liB~~~ f llcs1, .•. , sN<RN) II G lies!, ... , sN(RN, B).
Note. It is obvious that in the above constructions one may consi-
der the degrees si of the generators A i : B, -+ B,;+k.l as functions
of -r. We have not reflected this in the notation for the sake of sim-
plicity. In every formula it is clear at what point -r one has to consi-
der si. An alternative approach is to consider the Banach scales
{B-r} where -r varies only over a finite set of values. Then one may
choose the degrees si independent of -r, but the operator
2( 1) 2N-2(N- 1) ( 1 2N-1)
T ... T <p Att ... , An : B-r-+
-+ B-r+ht+ ... +kN+lt+ .•• +lN-1
will now be defined only for certain values of -r.
CR. II. CAIJCULUIS OF NONOOMMUTATIVE OPERATIORJS 229
Example. Let
B-r=W2-r(R), D=Co(R),
A 2cp (x) = xcp (x), A 1 cp (x) = - icp' (x),
Then
-'t' for 't'<O,
k1 =k 2 =0, l =m, s1 =0, s2 ('t') = { 0 for 't':;?-0.
It is not difficult to verify that
2 (3 1) ljl(x)=
Tcp A 2 , A1 V12n J . (x, v
r etpxcp p)
~
1+p2 m¢(p)dp,
-00
where ¢' = F¢. We shall denote
Tcp (.4 2 , 1J ljJ (x) =L u. -i L)'I' (x),
where L (p, x) =cp (p, x) Vi+ p 2m. By virtue of Lemma 3.1 we have
II L (;, - i fx ) llwr(R)->
2
wr-m(R)~c II cp IIS6l"r-m, o<R
2
2)'
where
( x 1 x 2) L (xt. x2)
m = , c=const.
'I' V1+x~m
Problem. Let cp E S6l"z, 0 (R2),
L (x, p) =cp (x, p) V1 + p2m.
Set
2 1) def
L ( .r, p U(x)=
= 2:h re~
-00
P·X L(x, p)dp r
-oo
e- i-P·'6 U(s)d£, UECo(R).
Let S be an infmitely differential real-valut>d finite function.

Define tho operator T =
def V 1 + (p + S')2m in the following way.
Consider the generator A det p+S':
AU (x) = - ihU' (x) + S' (x) U (x).
2~0 OPERATIONAL :.VIETH·ODS
Set T n m
= <vn (~) V 1 + ~zm, where {<vn} is a sequence of functions
in C0 (R) defined in Lemma 3.13 of Chapter I.
Let TU =lim Tn (A) U.
n->oo
(1) Prove that T is a translator with the step m in {w;' (R)).
(2) Prove the equality
e- ~ S(x) L G, ;) e~ S(x) U (x) =L (~, p+S') U (x),

where
2 1 ) def 2 ( 3 1 \
L ( x, p+S' =V1+(p+S')mcp x, p+S'), UEJV;'(R).
(3) Prove that this equality remains valid in the case of an arbi-
trary function S E coo, if it is to be considered for finite functions U.
Notation. If X = <p'¢ then we shall write
2( 1) 2N-2(N- 1) ( 1 2N-1)
T ... T X Ai, ... ' AN =
2( 1) 2N-2(N- 1) ( 1 2N-1) ( 1 2N-1)
=T ... T cpN_ 1 (AN-1) ... T< 1> (jl1 (At)·
The proof is easily obtained from (3.1).

Lemma 3.3. The linear span of the set of functions ljJ of the form
ljJ (x) = ¢ 1 (xt) ... ¢N (xN), ¢i E$ s.l (R) (3.2)
is dense in $ s1, ••• sN (RN).
Proof. It suffices to show that by means of linear combinations
of functions of the form (3.2) it is possible to approximate in the
space $ 81 , ••• , sN (RN) the Fourier transforms of functions in
CO' (RN) and also the exponentials
X-+ e-ia1x1- ... -iaNxN.
For the exponentials, this is obvious because

e-ia!Xl-···-iaNxN=e-ialX! .. • e-iaNxN.
Now let <p be the Fourier transform of a function in (R11) . Co
Then cp E W~ (RN) for any l. Any function in the Sobolev space may
CH. II. CAI.JCULUS OF NONOOMMUTATIVE OPERoAT'ORS 231
be approximated in the norm of this space with any degree of accu-

racy by means of linear combinations of products of the form (3.2),
where all "\jli belong to C'f/ (R) (prove it). To end the proof it remains
to make use of embedding Theorem 2.1 of Chapter I and 1.2 of
Chapter II.
Lemmas 3.1 and 3.2 signify the existence of the homomorphism
of the Banach spaces:
rd/: ${}s 1, ••. ,sN(RN)--+Hom (B,, Bt+kt+ ... +kN+l 1+ ... +lN-t),
which is defined by (3.1) and which translates the function cp into
the operator
}(!) 2N-2(N-l) ( 1 2N-1)
T ... T cp Ail ... ' AN ;
besides, it has the property
2(!) 2N-2(N-l) ( 1 ) ( 3 ) (2N-1)
T ... T {j)t . At X <:p2 .A2 X ... X cpN AN =
= cpN (AN) r<N- 1) cpN-t (AN-1) ... r< 1) cpt (A). (3.3)
It follows from Lemma 3.3 that this homomorphism is uniquely
defined by the property (3.3).
Thus the following theorem is valid.
Theorem 3.1. There exists a unique homomorphism
2(i) 2N-1(N-!) ( 1 2N-1)
ell: cp--+ T ... T cp At, .•. , AN
of the Banach space ${} 81 , ••• , sN (RN) into a Banach space of everywhere
defined linear bounded operators from B-r. into B-r.+k 1+... +kN+lt+ ... +lN-t
with the property (3.3).
Sec. 4. The Spectrum of a Vector-Operator

Let us make further agreements as to the notation. If a translator
T is identical, then we shall omit it in the notation of functions
of a vector-operator and number all other operators in succession.
For example, instead of lcp (l,A')
we shall write cp(l,}).
Let 0: Do--+ B,, where D 0 ::J D. Then the norm of the shrinkage
0 lv, which is considered as the operator from B-r.' into B,", will
be denoted by II 0 IJ,;,..,.,...
The following lemma is obtained directly from the definitions.
Lemma 4.1 . Let A i, i = 1, ... , N be generators of degree si
with step ki on the scale {B-r.} and let T(il, i = 1, ... , N- 1 be
translators. Let A i be the operator A i considered as a generator with
232 OPERATIONAL METfWDS
step ki +
f.ti· Then for any 'ljJ E S9 s 1, ... , sN (RN) and h ED the
following equality is valid:
2 2N- 2 ( 1 2N- 1 )
T(1)... J' (N-1)'\jl Ai, ... , AN h=
2( 1) 2N-2(N 1) ( 1 2N-1)
=T ... T. - 'ljJ A~, ... , Aiv h.
Lemma 4.2. Let
Then the following estimate is valid:

2( 1) 2N-2(N 1) (2m-1 2m+1) ( 1
11 T ... T - rp Am , Am+! X 'ljJ A1,
N-1
~cMII¢11.58si, .. ,sN
(Rn) lJ
a=i
IIT~~ht+ ... +kr"+lt+ ... +la-!11·
a=j=m
Proof. We shall prove this lemma for N = 2. Namely, prove

the following estimate
II T(cp'ljl) Ct. 1J ll,_.,+kt+k2+l~

~c//rcr LL, A.J ll<+ht-.<+kt+zii¢11SBs ,s 1 2 <R2)·
Let T': BHkt-+ BHkt+Zt be the closure of the shrinkage of the

2 ( 1 3 )
operator Tcp A1 , A 2 on D. Then
2
T(rpljl) A 11 A 2
(1 3) = T'ljl
2 (1A Az3) . 1,
Indeed, if h E D then
= 2~ j 1P (t1,
R2
tz) u A2 (tz) [rcr LL. 1J] u A1 (t1) hat=
= ( 2 ~) 2 ~ lP (t1, t2.) dt ~ cp ('tf> 'tz) UA2 ('tz + t 2) TU A1 X

R2 R2
CH. II. CALCULUS OF NONOO:VIMUTATIVE OPER.A'TIQR:S 233
X (T 1 +t 1) h dT = 2~ ) (cp*ljl) (t 1, t 2 ) U A 2 (t2) TU A 1 (t 1) h dt =
R2
2 ( 1 3 )
= T (cpljl) At. A 2 h.
To complete the proof it remains to make use of Lemma 3.1.
The following theorem is proved in exactly the same way as:
Lemma 4.2.
Theorem 4.1. For any integer l' the following estimate is valid~
llr (cpljl) Ct. lJ ILHk1+k2+l'~

~c II T(p LL, AJ ILki-+'t+ki+l' JJljJ ll~s 8
1' 2
(R2)·
Lemma 4.3. There exists a set of functions ljJ E!1l s, s' (R2) such
2 ( 1 3 )
that Tljl A, A' = 0 is an ideal in the algebra $ s, s' (R2 ).
Proof. Let hE B,. Then for any cp E !1l s, s' (R2)
II Tljl (1, A') (1, A') h IIHk+l+k' ~

X cp
~c II cp ll~s -II Tljl (A, A') IIHk-+'t+k+zll h JJ, = 0.

8
1' 2
(R2)
Consequently,
Tljl (1, A') X cp (.4, A') =0

for any cp E$ s, s' (R2).
1 2(1) 3 2N-2(N- 1)2N-1)
Let % = ( At. T , A 2 , . . . , T , AN be a vector-operator.
In this case we denote by X an ideal in the algebra $ si' ••. , sN (RN)
containing the totality of all functions cp such that
2 2N-2 ( 1 2N-1)
r<1) ... T (N- 1>(\jlcp) Ab ... , AN =0,
'1\jl E$ si' ... , sN (RN).
Lemma 4.3 means that in the case N = 2 the ideal X coincides

with the kernel of the homomorphism
2 ( 1 3 )
M : cp -+ Tcp A11 Az •
234 OPERATIONAL :WETHIODS
Definition. A point 'A E RN will be called a point of the resolvent

set of the vector-operator
( 1 2(1) 3 2N-3 2N-2(N- 1) 2N-1)
~q:; = At, T , A2 , ••• , Al\--t, T , Al\. ,
if there exists such a neighborhood U of this point that any function
cp E Co (RN) with the support in U belongs to the ideal X constructed
according to the homomorphism
2( 1) 2N-2(N i) ( 1 2N-1)
M: ljl-+T ... T - ljJ Ab ... , AN .
The complement to the resolvent set will be called the spectrum a of
the vector-operator !1:.
Lemma 4.4. Let the function cp E Coo (RN) belong to the ideal X
constructed according to the homomorphism
2( 1) 2N-2(N- 1) ( 1 2N-1)
ljl-+T ... T ljJ Ab ... , AN .
Then any point at which the function cp is non-zero belongs to the resol-
vent set of the vector-operator
( 1 }(1) 2N-1)
~q:;= At, T ' ... , AN .
Proof. Let cp ('A) -=1= 0 and let eA be a function in Co (RN) which
is equal to unity in the 6cneighborhood of the point 'A and va-
nishes outside the 6 2 -neighborhood of this point. For sufficiently
small 6 2 the function.¥= eA/cp belongs to Co(RN). But eA = cp ·.¥.
Hence the function cp is included in the ideal X; consequently,
eA EX. Let x E Co(R 2) and let the support of the function x lie
in the 6cneighborhood of the point 'A. Then x (x) = X (x) e1. (x),
i.e. X EX, Q.E.D.
In the case N = 2 Lemmas 4.3 and 4.4 result in the criterion that
follows.
Treorem 4.2. For a point 'A E R 2 to belong to the resolvent set of
the vector-operator (A, T, A' )
it is necessary and sufficient that there
exist such functions cp1 and cp 2 in Co (R) that cp 1 ('A1 ) -=/= 0, cp 2 (/c 2 ) -=/= 0,
and, at the same time, cp 2 (A') Tcp 1 (A) = 0.
Let A = (A 1 , . • • , Ak) be a generating set of degree s with the
defining pair of spaces (B 1 , B 2 ). Let T be a homomorphism T: B 2 -+
-+ B 3 and A'= (Ai, ... , AI,,) be a generating set of degree s'
with the defining pair of spaces (B 3 , B 4 ). We shall assume that there
exist embeddings Bt c B 2 and B 3 c B 4 •
Let { U (t)} be the k-parameter group generated by the set A,
and let {V (t')} be the k'-parameter group generated by the set A'.
CH. II. CALCULUS OF NONOOMMUTATIVE OPERj\.'T10RS 235
In analogy with the above procedure we shall define a Banach al-

gebra .tf5J s, s' (R" X R"') on the basis of the space Cs, s' (R" X R"')
of functions continuous in R" X R"' with a finite norm
II f II - sup If (x, y) I
Cs,s'<n"xn"'>- xERk (1+1xl)s(1+1YI)s' •
uEn"'
If cp E $ s, s' (R" X R"') and h E B 1 , then set by definition
3 (1 3)
Tcp A, A' h= (2:n:)-
~
2 J1 {p (t, T) e-iA'TTe-iAthdtdT,
nk+k'
where e-iAt: B 1 -+ B 2 is the closure of the operator U(t); e-iA't: B 3 -+
-+ B 4 is the closure of the operator V (t).
Definition. The point (A., f..l) E R" X R"' will be called a point
1 3
of the resolvent set of the pair of generating sets A, A' with respect
2
to the translator T if there exists such a neighborhood U of this point
that any function cp E Co (R"+h') with the support in U belongs to
the following ideal X of the algebra $ s, s' (R"+"') constructed accor-
ding to the homomorphism
2 ( 1 3 )
rdl : 'P-+ 1'~) A, A'
of the Banach algebra $ s, s' (R"+"') into the Banach space
Hom (B 1 , B 2 ):
I
X= ( cp cl!cp¢ = 0, 'V¢ E $ s, s' (R"+k)) •
The complement to the resolvent set will be called the spectrum a
1 3 2
of the pair A, A' with respect to the operator T.
1 3
Let (A., f..l) be a point of the resolvent set of the pair A, A' with
2
respect to T, and cp be a function which is infinitely differentiable
everywhere except the point (A., f..l). Next, let U be a neighborhood
of the point (A., f..l) mentioned in the_ definition, and let infinitely
differentiable in U functions (ji and cp belonging to $ s, s' (Rk+k')
coinc0e with the function cp outside_some closed set in U. Then
<P - cp E Co (Rk+"') and supp ((ji - {ji) c: U so that the function
belongs to the ideal X. For brevity we shall denote the element
{cp} = {rp} of the factor-algebra $s,s' (Rk+k')IX by {cp} (sometimes
even by cp) though cp does not necessarily belong to .~ s, s' (R"+"').
For the homomorphism ell!, the following expansion is valid:
QJ!! = :n:clft(J'
236 OPERATIONAL :\1ETH10:DS
where
k+k' :rt k+k' <Yfta
Sfl s, s' (R ) __,. Sfl s, s' (R )/X _ __,. Hom (B~> B4).
For this reason
0 ( 1 3 ) 2- ( 1 3 )
T(i> A, A' = Tcp A, A' .
We shall denote this operator also by Trp (A., A') and the function q>
will be called its symbol.
Lemma 4.5. Let the support of a function 'ljJ E Sfl 8 , s' (Rk X Rk')
1 3 2
lie in the resolvent set of the pair A, A' with respect to T. Then 'ljJ be-
longs to the ideal X constructed according to the homomorphism c;Jt.
Proof. First of all, we shall note that if the point (.A, f.L) belongs
1 3 2
to the resolvent set of the pair A, A' with respect to T, then any
function in Sfl s, s' (Rk X R"') with the support in a suffldently
small neighborhood of the point (.A, 1-t) belongs to X. This follows
from the fact that such a function may be precisely approximated
in Sfl s, s' (Rk, R"') by the function in (R" X R"') having the Co
support in a small neighborhood of the point (.A, f.L).
First of all, suppose that the function 'ljJ is finite. Then it can be
represented in the form of a sum of a finite number of functions
in X and, consequently, it belongs to X itself.
Now let 'ljJ be an arbitrary function satisfying the conditions of the
lemma. There exists a sequence {ln} c (R" X R"') such that Co
k+k'
lim (2n)- - 2 - J (F- 1 Zn) (t) g (t) dt = g (0)
n-oo Rk+k'
for any g E Cs, s' (R" X R"'). Set ¢n =¢ln. We have supp ¢n c
c supp 'ljl. For this reason T¢n (A., A') = 0. Next
lim (2n)- -
n-oo
h+k'
2-
Rk+h'
J (F- 1¢n) (t) g (t) dt =
k+h'
=lim (2n)- - 2- [ (F- 1¢) * (F- 1Zn)] (t) g (t) dt =
n--.oo
J (F- ¢) (t) g (t) dt.

1
Rh+k'
CH. II. CAIJCULUS OF NONOOMMUTATIVE OPER•A'T10R:S 237
Hence, it follows that for any hE B 1 , h* E B;
n~oo
This means that z\~ = 0. (A., ;l')

Replacing 'P by 'PI in this discourse we obtain
2 (1 3)
T¢ A, A' X f A, A' =0
(1 3)
li+k'.
for any f E &J s, s' (R ). This means that 'ljJ E X and the lemma
is proved.
Theorem 4.3. Let A = (A 1 , . • • , An) be a generating set of degree
s with the defining pair of spaces (B 1 , B 2), A' = (A 1, . A l) be 0 • ,
a generating set of degrees~ with the defining pair of spaces (B 3 , B 4 ),

let 1': B 2 -+ B 3 be a homomorphism. Let A- 0 E Rk, f.Lo E R 1 and let
there exist such a neighborhood U of the point (1, 0 , !-to) in Rk+l that
for some 6 > 0 for (A-, ~t) E U and e < e 0 the following estimate is
valid:
1)2 1 2 13 2 -~<h+l+s+s'+o)l'
1 T[(A-A-) + ~A'-1-1) +e2 ] IBr•Ba<c.
Then the point (A- 0 , f.!o) belongs to the resolvent set of the vector-operator
( 1 2 :J )
fC = A, T, A' (c not dependent on e)
Proof. Let 'ljJ E Co (Rk+t) and supp 'ljJ c Uo We have to prove
that 'ljJ (%) = 0. Without restriction of generality we set !-to = 0,
A. 0 = 0.
Let (P (x, y), x E R\ y E R 1 be such a function in Co (Rk+Z) that
00
2i fP(X-i, y-j)=1,
it=-oo
ik)' j = (j 1l ••• ' iz).
i = (ill . 0 • '
Then such a constant c1 will be found that for sufficiently large n

'P(x, Y)= 2J 'ljl(x, Y)fP(nx-i, ny-j) def 2] ~;~1?-l(x, y),
(i, j)EMn (i, j)EMn tJ
where the number of elements of the set M n does not exceed

c 1nk+l and (...!_,
n
j_)
n
EU for (i, j) EMn.
Denote
Re (x, y) = (V x2+ y2+ e2)k+l+st-s'+6.
238 OPERATIONAL i\'IETH.O:DS
We have
cp(nx-i, ny-j)R1;n(x- ~, y- ~)=
=cp(nx-i, ny-j)Rt(nx-i, ny-j)n-h-1-s-s'-b .
Hence, it follows that the estimate is valid:
'P(nx-i, ny-j)R 1;n (x-..!:_, y - i ) II
n
~
n f!as,s•(Rk+l)
'' ~II'PR1II,gas s•(RR+l)n-h-1-b.
From the identity

\f~j> (x, y) = [ 1PW> (x, y) Rt;n ( x- ~ , y- ~ ) JX
. 2 . 1 1 ~(-h-1-s-s'-b)
x[(x-,i-) +(y- ~) +-;2]
and from the estimate for the operator
2 [ ( 1 . ) 2 ( 3 . - ~ (h+l+s+ s'+b>
T A-...:_
n '
+ A' _ _L)
n
+_i_]
2
n2
it follows
II1P?j (.2') II~ c2 • n -h-l-b ,
where c2 is a constant. This means that
II'P (v't") !l~c2·n-h-l-b.c1nh+l=c1c2n-b-+O
for n-+ oo. The theorem is proved.

Theorem 4.4. Let A = (A 1 , . . . , Ah) be a generating set of degrees
with the defining pair of spaces (B1 , B 2 ), let A' = (A;, ... , A i) be
a generating set of degree s' with the defining pair of spaces (B 3 , BJ
and let T: B 2 -+ B 3 be a homomorphism. Suppose that for any suf-
ficiently small real-valued e > 0 there exists an estimate
II T[ (.4- A) 2 + (1,- ~) 2 + e2 rNJZIIB2_.Ba~cN,

where c is a constant and N is any integer. Then the point (A, ~t)
1 2 3 )
belongs to the resolvent set of the vector-operator .2' = ( A, T, A'
(c not dependent on e).
Proof. Suppose (without loss of generality) that A=O, ~=0.
V
Denote ,;8 (x, y) = x2+ y2+ ::,2,
CH. II. CALCULUS OF NONC'Ol\J:MUTATIVE OPERAT'OR!S 23!}
Let x(x, y)EC;;"(R"+ 1) bean arbitrary function with the support.

in the <'>-neighborhood of zero. Then
II X (x, y) [T0 (x, y)]N 1/fEs, s•(Rh+l) = 0 (<'>N).
Choose 6 < J...c . Then
II X (v"t') lln1_.IJ2 =II [T[ T0 (A, lr)] -N] X (A, 1) X

X [To (A, Ar) JN lln _.B.t ~ 1
~C·cNI!x(x, y)[T0 (x, y)JNI!ns,s•<Rh+ZJ-+0 for N-+oo.

2
Here the brackets [ ] mean that the numbering of operators inside-
the brackets does not extend to the operators outside the brackets,
i. e., the expression in the brackets is considered to be the operator
acting second (see Introduction). The theorem is proved.
Sec. 5. Theorem on Homomorphism

Theorem 5.1. Let lu be a function in .9& s, s' (Rk X Rk'). It is
1 3
equal to unity in the spectrum a of the pair of generating sets A, A'
2
with respect to the translator T. Then the following inequality is valid:
I Tcp (A, A') X 'P (A, Ar) lln _.n4 ~ cII Tcp (A, Ar) lln
1 2 _.B3 X
X II ¢luI Ins, s•<R"xRii') •

Proof. Let .9& s, s' (Rii+h' Ia) be a sub algebra of the algebra
.9& s, s' (Rh+"') which is the closure of the algebra of functions in
.9&s,s' (R"+Ii') with the support in R"+'i'"'a. For any function
f E .9& s, s' (Rh+k'"'a)
2 ( 1 3 )
Tj A, A' =0.
Since .9& s, s' (RIH"'"' a) is an ideal in .9& s, s' (Rh+k') it follows that
for f E .9& s, s' (Rh+li'"'a) we have
2 (1 3) 2
T(lp·cp) A, A' =T(('P-f)cp) A, A' .
(1 3)
Consequently,
II T(cp1j)) (l, 1) IIBj->BJ ~

~c II Tcp (A, Ar) IIB2_.B3II11'- f JIBs, s•(Rii+h') •
240 OPERATIONAL METHDDS
It remains to be noted that the function la - 1 belongs to the algebra

!iJ s,
s' (Rk+k'""'a).
Theorem 5.2. (The Main Inequality.) Let a be a spectrum of the

1 3 2
pair of generated sets A, A' with respect to the translator T. There
exists a unique homomorphism
2 ( 1 3 )
cJ!t : cp-+ Tcp A, A'
of the Banach space !iJ s, s' (a) into a Banach space of uniformly defined
linear bounded operators from B 1 into B 4 with the property (3.3) and
.the inequality
II Tcp (l, } ) 'P (1, A') 11Bi->B4 ~

~c II Trp (A, A') IIB2-.Ball 'P ii&~s, s' (a)
is true.
1 ,'J 2
Corollary. Let a be the spectrum of the pair A, A' with respect toT.
The following inequality is valid:
~
I •~"
T
~!;(A') TF;(A)
I{
lls,-n, '~"
2J cp; (y) \~; (x)
N
i=O
r
1; (y) F; (x)
}
Definition. Let the sequence {(])n} converge to the function (]) in

!il s, s' (R k X Rk') and let the functions 'Pn: x -+ [ (])n (x) ]-1 also
belong to !JJ 8 , s' (Rk X Rk'). At that, let the following condition be
fulfilled:
I fi.• U. 1- l II.,.•,,;;; II;. (Lt· lll

s~p lll'il·llcril&~s s•<Rkxn"'>
Then the function (]) will be called the spectral weight.
We shall say that the spectral weight (]) is subordinated to the
spectral weight !F if the sequence of norms is bounded:
II ~ 1193s 8 •(RkxRh'> •
CH. II. CALCULUS OF NONOOMMUTATIVE OPERATIORS 241
k k'
Introduce in !19 s, s' (R X R ) the norm
def
II (p ll<t> =II cpcD ll$s, s•<R"xR11 '> (5.1)
and the product
def
(P1 X cpz = cptcpz()); (5.2)
we have
II cpt X cr211<1> =II cpt(Plll211$s, s•<R"xR"'> ~
~II cpj(J) ll$s, s•<R"xR"'> II cp2(}) ll$s, s•<R"xR"'> =II (Pi ll<t>ll crzii·
Thus the norm (5.1) and the product (5.2) induce the structure
of the normed algebra into !19 s, s' (R" X R"').
The Banach algebra obtained by the completion of this normed
algebra will be denoted by !19 s, s'. (R" X R"').
Let !19 s, s', <1> (R"+"'"-a) be the closure in !19 s, s', <1> (R" X R"')
of the subalgebra consisting of finite functions with supports in
R" +1" ""'a. The factor-algebra !19 s, s', (R" X R"')/.5W s, s', <1> (R"+"'"- a)
will be denoted by !19 s, s'. <1> (a).
Theorem 5.3. The following inequality is ualid:
2 ( 1 3 ) II
T'ljJ A, A' ~IJ'IJlll$s,s',<t><a>·
Proof. We have
llr'IJl(l, A') ll=llr'IJl (l, A') (}); 1 (l, A')(Dn (l, A') 11~
~II T(J),:; (A, A') IIB2-+B3JJ1J:cDn JJ$s, s•<R"xR"'> X
1
X supiiTJillcpll
<Jl
II /cp u. 1~) II
.98s, s' (l~kxRk
' ~II'IJl<DII-98
) s, s
•<R"xR"'> +en,
where Bn-+- 0 for n-+- oo.

If the support of the function cp·E !19 s, s' (R" X R"') lies in R"+"' "-a,
then T'ljJ (A, A') =T [ 'ljJ (A, }) -(p (A, A')]. Hence
llr'ljJ (.4, A')~II('IJl-cp)cDJif-Bs,s'<R"xR"'> =

=II 'ljJ- cp 11.9.]8 , s•<R"xR"'> ·
By taking the greatest lower bound in cp in the latter ine-
quality we obtain the proof of the theorem.
16-01225
Sec. 6. Problems
(1) Let
A d 2x 4, d
+~
0 0 o
=~ dx xZ +1 ' • = ~ dx ' T=1,

B1 = B2 = B 3 = B4 = £ 2 (R).
The function
u (x, t) = ((x;;tt i] j (x + t)
0
~
!!!:._ _ !!!:._
at -
o
~
2x
ax
+~ 0
x2 +1 u. 0
vr[
For this reason
lleiAtflk2(R)= (x~2t~ti rJj(x+t)i 2 dX~

~~~;I (x~2t~t1 IV f if(x+t)i2dx~
-oo
~c (1 +It I ) 2 11 f IIL2(R)o
It is obvious that
II eiA't f li£2(R) =II f lk2(R)o
For any f ES we have
Je-iAt<D(t)
- f (x) dt=
00
<D (A) f (x) =-==-

1
V 2n -oo
00
=___!__ r (x-t) 2 1 + f(x-t)fb(t)dt=

V2n J
-oo
xZ-f-1
= xz ~ 1 <D { - i :x ) (xz + 1) f (x).

For this reason for any <D, cp ECO' (R) and f ES
<D (A) cp (A') j (x) =
= x 2 ~ 1 <D { - i :x) (x 2 +1) cp ( i d~) f (x)=
00
= Vn Jr ei px ( 1 -
1 d2 ) -1
dp2 cD (p)
(
1- d2 )
dpz
-
cp (- P) I (P) dp.
-oo
iCJI. II. C'A:UCULUS OF NONCIOMMUT'A TIVE. OPERA:TIORIS 243
If 'A =1= I! and the support of the function lies in the
small neighborhood of the point f.t, then the operator
CD (p) ( 1- :;2 ) <p(p)

1 2 '
(
is equal to zero. For this reason the spectrum of the pair A, A')
lies on the diagonal of the direct product R X R.
On the other hand, the spectrum of the pair {A_, covers the A')
whole space R2 , because for any 8 > 0, 'A E R and I! E R there exist
such functions and (p)=J=0
(the operator ( 1 - :;2 ) - 1 is non-local) .

Thus, the spectrum of the. pair of non-self-adjoint operators de-
pends on the order in which these operato.rs are considered: in gene-
ral, the spectrum of the pair (A., B) is not equal to the spectrum
of the pair (j, B). Besides, one may say that for self-adjoint opera-
tors these spectra coincide with one another (correct to the reflection
in relation to the bisectrix of the coordinate angle).
(2) Let. B be the completion of Co (R) in the norm
00
Jl f (x) /11 = J Iex

-00
212 f (x) 12 dx
and B'- the completion of C0 (R) in the norm

00
II f (x) II~·= ~ Ie-x 212f (x) l2 dx.
It is obvious that B c B'. Take the operator of embedding 1: B ~ B'

as T. Consider the "birth" and "death" operators
• d ......I ~X
A =~- • A , = ~. dx
d
-~x.
.
dx 1 '
The following theorem is valid.

Theorem 6.1. (a) A is a generator of degree 0 with the defining pair
(B, B) and A' is a generator of degree 0 with the defining pair (B', B');
16*
244 OPERATIONAL METH,Q:DS
(b) the spectrum of the pair (A., A') is the whole plane R 2 ;
(c) the function
1
(J>: (x, y)-+ c (1 + (x _ y)2)2 e-<x-y)2f2
for sufficiently large c is the spectral weight of the vector-operator

(A. A') so that
Proof. (a) The solution of the equation

.au .au+.
lat=laz_lXU
+-
x2
takes the form u (.r, t) = e 2 cp (x + t). For this reason
x2 (x+t)2
e-iAtj(x) =e- 2e- -2-f(x+t)
so that
II e-iAtj (x) liB= II e ~2 e-iAtj (x) IIL2(Rxl =
=II e (x~t)2 f (x + t) IIL2(Rxl =II e ~2 f (x) IJL2(Rxl =II f (x) liB

Co
for any f E (R). This means that II e- iAt II= 1.
(b) Similarly, for any cp E (R) Co
1
(x+t)2 2
J
00
II e-iA'tcp liB'= [ j e-x2ex2e- -2-1 cp (x+t) 12dx =

-00
1
oo x2 2
= (Joo e- 2 1cp(x)l 2 dx) =llcriiB'·
Let Cl>, f E (R). Co

In analogy with Example 1 we obtain
x2 x2
CI>(A)f(x)=e-2ci>(i ddx)e2j(x),
x2 x2
CI>(A')f(x)=e2 ci>(i d~ )e- 2 /(x).
CH. II. CAIJCULUIS OF NONOO:IIlVIUTATIVE QoPERoAT'ORIS 245
Let cD, /, <pECo(R). Then
x2 ( d ) ( d ) x2
<p (A') <J) (A) f (.:r) = e 2 <p i "dX e-x 2Q) i dx e 2 f (x) =
x2
= e2fp ( i :x ) e-x 2cD ( i d~ ) ft(x).
Let Fp~x: L 2 (Rp) ~ L 2 (Rx) be a Fourier transform and r;.lv :

L 2 (Rx)~L 2 {Rp) be the inverse Fourier transform. Denote Fx-.vfdx)
by !1 (p). Then
x2 d2
(p (A') (D (A) I (x) = e- 2 r;;~x<p
1 -
(p) edpZ <ll (p) !I (p).
-
d2
The expression <p (p) e dP 2 '1jJ (p) does not diminish to zero for any
finite fP and 'ljl. For this reason the spectrum of the vector-operator
( 1 2 )
A, A' is the whole plane.
(c) Let <ll (x, y) = <p (x- y). It is not difficult to verify that
in this case <ll {A, A') is a multiplication operator by some func-

tion, namely
CD (A, A') f (x) = ·2~ eX2j (x) r

-00
(p (t) e-(x-t)2dt,
where q; = F- <p. 1
Let
1
<Iln (x, g)= (1 + (x + y)) 2 [ e-<x-y) 12 + ! J= fPn (x- y).
2
Then
00
;n ex2 ~ q;n (t) e-(x-t)2 dt =

-00
1 2 .• - .!. ~2j2
=--wex I e-<x-t)2+its(1-j-~2) e d~dt=
2n ~ J
R2
2
1 + e'§ 1.2 jn_
2
It is not difficult to verify that

- .!. £2j4 - ~ ~
sup (1 t2)
sER
~o + ''
~ e ; 212 = 2 2 e4 -1- d1,
1-1-e jn
246 OPERATIONAL METHOiDS
where dn-+- 0 for n-+- oo. Therefore, the latter integral is bounded
as a function of n. Hence the statement (c) is obtained because
a multiplication operator by ex 2 is bounded as an operator acting
on B to B'.
Problem 1. Let
A= - £.ddx +"
£
x cos x-sin x
x sin x+x2
1 d2
A = dx2 ' T=1,
B 1 =B 2 = B3 = B,.=L 2 (R).
Show that the spectrum of the vector-operator (A_, A') lies in the
1
region A2 ~ 0, A1 - 1~ I A2 12 ~ A1 + 1.
Solution. Note that A = -i :x + i :x [In ( 1 + si: x ) J. Hen-
ce, by virtue of the problem of Sec. 3 we obtain
ln(t+sinx) d -ln(t+sinx)
ljJ (A) f=e x tjJ ( - i dx) e x f (x) =
= (1 + si~ X ) ljJ ( - i :X ) ( 1 + si: X rt j (x).
Let q>, tjJECo(R). Then

'P (A') ljJ (A) f =
(/1 + . i (·d))
sm ~-
= F;;-1x q> ( - p 2) -i-P X
Consider in greater detail the function
11 (p) = (1 + . (·*P
d))
!d
sm
X (p) =
i
00 00
= 1 l dy \ d£eiv<P-!il (1 + sin y ) Xm,

V2n J
-oo
~
-oo
Y
xECo (R).
CH. II. C'ALCULUtS OF NONOOMMUTATIVE OPERoAT'D-RIS 247
l\iote that
+1 -
Fx_. 6 (8(1-jxj))=_!__ I e-isxdx=-./~ sins
y2n: J
-1
V n: s ,
or
F;~p(1+ si;y)=V~ 8(1-jpj)+V2n8(p).
Hence
TJ (p) =~X (p) + + P+1
~
p-1
X(£) d£.
If supp X c [a, b], then supp 11 c [a- 1, b 1]. +
Return to operator cp (A') 'ljJ (A). From the above it is clear that
i:f supp 'ljJ c [a, b] and supp cp (-p 2) c R"'-.[a- 1, b 1] then +
qJ (A') 1p (A) = 0. Hence
a(A, A')c (A ::(;0, 2 (A 1 -1):::;;VIAzi::(;(AJ+1)l.

Problem 2. Let
A . a A' . a
= ~ ox 1 ' = ~ ox2 '
B 1 = B 2 = B 3 = B,. = Cs 1 , s2 (R X R), T = 1.
Show that in this instance
i.e., the algebra induced in Hom (Cs 1 , s2 (Rx~' Rx2), Cs 1 , s2 (Rx1 , Rx2))
by the homomorphism r.dl is a normed algebra.
Solution. There exists the inequality
jcp(x+t)j (1+jx 1 1)-"1(1+jx2 j)-"2 ::(;
11 CfJ llcs(R~h
::(; (1 +I t1 I )" 1 (1 +I tzl )"2
where
t =(til t 2 ), x = (x 17 x 2 ), s = (s 17 s 2 ), cp E C. (R 2 ).
Consequently, the set (A., A')

is a generating set of degree s.
=
Next, it (p F- 1cp EC0 (R2) then
I cr C4, 1, hllc. =
= s~p (1 +I x 1 j) -•q1 +I x 2 1) -•zl ~ ~ (t) X(x + t) dt I::(;
::(; ~ I~ (t) I (1 +I t1 I )" 1 (1 +I tzl )82 dt·il Xlies= 'II CfJ ll,:s ·II Xlies·
8
248 OPERATIONAL :IIETHO!DS
The same estimate is obtained if (j) (t) = 6 (t- t 0 ). For this reason
it follows from the definition of the space fB (R2 ) that for all s
cpE !B s (R2 ) the following estimate is valid:
llcp(J, A')llcs-.Cs~Jicpll$ 8 •
On the other hand, for cp EfB .• and for any e > 0 snch Xe EC
8,
I!Xe lies= 1 will be found that
Ij (j) (t) Xe (t) I>II cp lla?s- e.
dt
Then
llcp (1, I)xellcs:>J) cp(t)Xe(t)dtj:>Jicpl!_qs 8 -e
so that II cp (1, A') Jlc,-.c.,:>ll cp 11.98s· Hence, we obtAin
II cp (1, } ) llcs-.Cs =II cp 11.98 8•
It is not difficult to obtain the final inequalities now:
II (cp· P) (A, A') llcs-.C• =II cp·1p lla?s~jj cp 11.98 ·II~' 11&1
1 8 8 =
=II cp (1, A') llcs->CJ 1p (1, A') llc -.Cs·

8
Problem 3. Let .8 1 = .8 2 '= .8 3 = .8 4 = .8, T

1 and let A -A' be =
[(
a bounded operator. Show that the function +. for a suffl-
x1-x2 t
ciently large K is the spectral weight of the vector-operator
(1, A').
Solution. Let en E Co (R), en-+ 1 in !B 1 (R) for all l. Consider
the sequence of functions
'i'n (xt. X2) = ! [en (xt) en (x2) (xt- X2) + i] EfB s (R 2 ),
where s= (st. s 2 ), s 1 is the degree of A and s2 is the degree of A~.

We obtain
1 [(
'Pn =en (xt) en (x2) (x1-x2) +i=
=K [ ien (~t) e( (~() (xt-~2)
en (x1 en x2 Xf-X2 +i
i] EfB 8 (R2),
_1___ .. ~ K [ i(Xt-X2) ·] f[( .

'l'n
---,..
n-.oo (xt-X2 +i
) ~ = Xt-X2Tt
.1_. lll :A! s•
tJll
GH. II. CAIJCULUJS OF NONOO~I~IUTATIVE OPERAT'ORiS 249
Next
I 'Pn (A, A') II= {-II (A _:_A') en (A) en (A')+ i ~~~

~ c211 A-A' 11+1
"""' /( '
where
II en (A) II~ c, II en (A') II ~c.
.
Let ill= sup
I cpII ( 1.II } ) II . Choose
qJ cp 9.1,.
K';J::-M (c2 ll A -A' II+ 1).
Then
JVl.II'PnCL A')ll~1.
Thus the sequence 'Pn saLisfies all conditions of the definition of the
K) +.
:o:pectral weight, i. e., the function (x -xz is the spectral
L 1
( 1 2 )
weight of A, A' .
(3) One of the most important particular cases of Theorem 5.1
is the following estimate. Let A (A') be the generating sets of degree
s (s') and let B 2 c B 3 • Then
II f (A)·- f (A') 11Bt-->B4~
~c II cp (A)- cp (A') lln2~B311 ~ ~:~=~(r~) I ."Es, s' (Rh+k')
.
(4) Let {B-r:} be a Banach scale, T1 and T 2 be the regular operators
of degrees with the defming pair (B-r:, B,+-r: 1), where 'tis an integer.
Let the operator T/- T 2 be bounded as the operator acting from B-r:
to B,_,2 , where ,; 2 > 2,;1 • Let T1 = A 1 +
iA 2 , T 2 = B1 iB 2 , +
where (A 1 , A 2 ), (B 1 , B 2 ) are the generating sets. Let the functions
(D1 (x 1 , x 2 ) and <J) 2 (y 1 , y 2 ) be such that the function ID~ ~x) -ID 2 (y:
x1 lx 2 -Yt-lYz
belongs Lo ~ s, s' (a ( T 1 ) X a ( T 2 )). Then the operator cD 1 (A 1 , A 2 ) -
-- cD 2 (B 1 , B 2 ) is bounded as an operator from B, to B,+2, 1_, 2 •
(5) Let r be a smooth curve in R 2 • Introduce the coordinates r, t
in the neighborhood of r, where r is the length of the normal, t is
the parameter on the curve. Let
cp E Co (R2 ), cp (x, y) = (ji (r, t),
N k ,-
~ _r_
-(/JN (r, t) -- Ll k! !0£.
iJrh
(0 ' t ) = cpN (x, y ) .
h=O
250 OPERATIONAL METH10!DS
If the spectrum of the vector-operator T' lies in rand (A.' AI)

the sum of the degrees of the generators A and A' does not exceed N,
then
2 (1 3) 2 (1 3)
Tcp A, A' = TcpN A, A' .
(6) Let P (x, y) = ~ aiixixi be a polynomial in two variables
and A and A' be generators in some Banach space B. Set
~ auAti Ai.
' 1 2 )
P {A, A' cter
Problem. Let A and A' be operators in £2 (R) considered in (1):

A . d A' _ . ~ -1- . 2x
= ~ dx ' - ~ dx ' l 1 + x2
( 2 1) k
Show that A' -A = 0 if and only if k> 3.
Solution. It is easily seen that -A')=!=O for k=O, 1, 2. (A.

Let k ~ 3. Obviously A is a generator of degree sA = 0. Next,
e-iA'tj (t) = e-ln(1+x2)+Jn(1+(x+!)2)j (x + t) = 1 ~~~2t)2 f (x + t);
consequently A' is a generator of degree SA' = 2. For cp, ljJ E CQ'(R)
we have
1p(A')cp(A)f= 1 ~x 2 '1J(i :x)(1-+x2)cp(i :.-c)f(x)=
= 1 ~ x2 [ 'iJ (p) ( 1 - :;2 ) Cf! (p) (Ff) cp J.

For this reason if supp cp n supp ljJ =I= 0 then 'iJ (A') cp (A) = 0,
i.e., the spectrum a (A., A') lies on the diagonal in R2 .
In view of the preceding item the operator F is equal (A., A')
to zero if on the diagonal in R2 the function F has zero of the order
~sA +SA' + 1 = 3. Consequently, for k ~ 3
(g, (l-Jt) It f)= lim (g, en (1~) (A-A') em (l) 1)

m, n~oo
for f E Co (R) and g E £2 (R) ( {en} being the sequence of functions

1 2 )h
from Lemma 3.13 of Chapter 1). Hence ( A- A' = 0 fork:;:;:, 3.
(7) Let B 1 = B 2 = L 2 (R), B 3 = B 4 = W;1 (R), and T = i :.[
and let A and A' be multiplication operators by a smooth real-
GH. II. •CALCULUIS OF NONOOMMUTATIVE OPERATORS 251
valued function f (x). It is not difficult to verify that the spectrum

of the vector-operator
From the equality
(A., T, A') lies on the diagonal.
(i :x) e-if(x)t = e-it(x)t ( i a:)+ (_it) e-if(x)t (if' (x))

it is not difficult to obtain the formula
TF(A, A') q;=F(f(x), f(x))i d~;r) +

+F'(f(:r), f(x))i d~~) q;(x),
where F' is the derivative of F with respect to the first argument

and q> E Co
(R).
Sec. 7. Differentiation of the Functions of an Operator Depending on

a Parameter
Theorem 7 .1. Let A~"\ A~"' be generators with the defining pairs
. l y, wh ere II e - iA (n)t
of spaces (B 1 , B 1 ) and (B 2 , B 2) , respectwe 1 II : : : ; ;
~ c1 (1 +
I t 1) 81 and 11 e-iA~n)t II~ c2 (1 +
I t 1) 82 , c1 , s1 , c2 , s2
being constants independent of n. Let T 11 : B 1 -+ B 2 be homomorphisms
bounded in the norm by the same constant. Next, let the sequence {A i"'}
converge point-by-point in some set dense in B 1 to the generator A 1 and
let the sequence {A;"'} converge point-by-point in a set dense in B 2
to the generator A 2 • Let the sequence { T n} converge point-by-point to
the homomorphism T: B1 -+ B 2 • Then, for any function f E .%l s1 • 52 (R 2),
the sequence
2 (1Ai"', A~m
{Tnf 3 J\} converges point-by-point to the homo-
. 2 ( 1 3 )
morphwn T/ ,A 1 , A 2 : B1 -+ B 2 •
Proof. FirsL of all, suppose that the function f is of the form
f (x, y) = / 1 (x) / 2 (y), where / 1 E .%ls1 (R), / 2 E .%ls2 (R). Then the
following formulas are valid:
T') nf ( A\n),
1 3
A~n)
)
= fz(A~n)) T nfi (Aln)),
lJ Ct, AJ = /2 (Az) TfdAt).

For any hE .8 1 we have
lim ft (A\n)) h = ft (ilt) h.
n_,.oo
252 OPEHATIONAL METHO[)S
Since T n-+ T point-by-point and II T n ll~const it follows that

( (n)) n-+oo
Tnft At h ---7 Tft (At) h
in the norm of .82 • Since / 2 (A~11l)-+ / 2 (A 2 ) point-by-point in .8 2
and II/ 2 (A;n')ll~ ,?
v 2n
ll/ \\on
V<Jsz
2 (R) it follows that
f (A~nJ) T nf (A~nl) h ~~ f (Az) Tf (At) h

in the norm of .8 2 • Theorem 7.1 is proved for the particular case
in question.
Now let h be a fixed element of .8 1• Then
2 ( 1( ) 3( ) ) n-->oo 2 ( 1 3 )
Tnf A 1n, A 2n h --?Tj At, Az h
in the norm of B 2 for any f of the set P dense in .98 , 1 , 82 (R 2) of
functions of the form
f (x, Y) = "
~ fiJi (x) lPi (y).
j=l
Denote by An: .98 8 1' s

2
(R2 )-+ .8 2 the operator acting according to
the formula
A nj = T2 n j ( 1
A <nl
1 '
A32<nl ) h
and denote by A the operator

def 2 ( 1 3 )
Af=Tf A 11 A 2 h.
Then the sequence {An} converges point-by-point to A in a set P
dense everywhere, and is bounded in the norm
II An II~ c;~2 I\ T n II·

By virtue of the Banach-Steinhaus theorem An -+ A point-by-point
in fBs 1, s2 (R 2), Q.E.D.
Note. In this theorem the generators may be replaced by the
generating sets.
Let A be a generating set consisting of k commutative operators
with the defining pair of spaces (.8 1 , B 1), let A' be a similar genera-
ting set of k' operators with the defining pair of spaces (.8 2 , B 2 )
and let T be a homomorphism B 1 -+ B 2 • Let P (x, y) be a polyno-
. mial in variables x E R", y E Rh:
P (x, y ) = ,, i j
Li aux y ,
CH. II. CADCULUIS OF NONOOMMUTATIVE OPERoAT!ORIS 253
where
... , ... ,
Set
. ·) (1 3)
TP A, A' = ~ auA'iTAi.
Lemma 7 .1. Let f E .CJ3> s, s' (R" X R11 ') and f P E !J6l s, s' (R" X R"').
2 ( 1 3 )
If the operator T P A, A' is bounded as the operator from B1 to B 2 ,
then
ft P (A, 1:) = r' t (A, 1,) ,
where T' is the extension of the operator TP (A., A') to the homomor-
phism B 1 -+ B 2 •
Proof. Let {In} be a sequence of functions of the form
fn (x, y) = Un (x) Vn (y), X E R", Y E R"'.
It is evident that if the operators A and A' are bounded, then the
lemma is valid for functions fn:
2 (1 3) 2 (1A, -A'3) .
TfnP A, A' = T'f, (7 .1)
If fn-+ f and fnP-+ fP in !J6l 8 , s' (Rh+h'), then passing over to the
limit for n -+ oo in (7 .1) we obtain the required lemma. The proof
of the existence of such a sequence is left to the reader. The lemma
is proved for the case of bounded operators A and A'.
Now let A and A' be arbitrary generators. Set
A'- A'
n- 1 +A'2jn
Then An and A~ are bounded operators converging to the operators

A and A', respectively, on sets dense everywhere. There exist con-
stants c, c', N, N' such that
II e-iAnt ll~c (1 +It It, I e-iA~t ll~c' (1 +It It'.
Denote: T, = cp (A~) T(p (An)· Then T n -+ T point-by-point in B 1
and II Tn IIB 1... B2 ~ const. Consequently, by virtue of Theorem 7.1
for imy function f E .CJ3> 0 (R 2 ) and for any h E B1 , the following
formula is valid:
2 (1
lim Tnf An,
3) h=Tf2 (1A, A'3) P (1A, A',
3) P (1An, A~
A~
3) h.
n-+oo
On the other hand, the sequence of operators
T~ = P (A, A') (jl (A~) Tcp (An)= (jl (A~) T'cp (An)
254 OPERATIONAL l\LETHODS
acting on B 1 to B 2 is bounded in the norm and strongly converges

to T'. For this reason
2 (1 3)
lim T;,j .An, A;, h = T'f A, A' h.
2 (1 3)
n-+oo
Now consider the family {T (;)} of regular operators
T (;) = A 1 (£) iA 2 (s),+
where (A 1 (s), A 2 (s)) = A (£) is a generating set of degree s with
the defining pair of spaces (B, B). Here the domain D of the opera-
tor T (;) does not depend on s,
lle-iA(~)tll
~~ (1 +It I )8 < 00
and the spectrum of the operator T (s) is contained in the closed

set a c R 2 which is independent of S·
Lemma 7 .2. Let f E .911 s (a), let the family { T (s)} have the defining
pair of spaces (B', B'), where B c B', and let the spectrum of the pair
1 2
of generating sets A (s'), A (s ") with re~pect to the embedding of B into
B' be contained in the closed set 2: c a X a c R 4 for any ~', s"
from the domain of the function T (s). Next, let
lim II T (£)- T (£0 ) IIB-+B' = 0.
s-+so
Then, if the set of functions of the form
g (x) -g (y)
is dense * in .911 s (2:), then

lim II I (A(£))- I (A (so)) IIB-..w = 0.
s-+so
Proof. Let the functions In (x), xER2 be such that
<Dn (x, y) = fn ~x)- fn (y). C .911 8 (2:). (7.2)
Xt +tx2-Yt-!Y2
Let Ts, so be the extension of the operator T (£) - T (s 0 ) to the

homomorphism B-+ B'. Then, according to the above lemma, we
have
2 ( 1 3 )
/n(A(s))-/n(A(£o))=Ts,so<Dn A(£), A(so) .
* Note that the eondition is carried out if {T (£)} is a family of generators.
·In this case one may set a = R, 2; = R2 .
GH. II. C:AIJCULU.S OF NONCIOMMUTATIVE OPER.A'T.Q>R!S 255
Consequently,
II f n (A (s))- In (A (so)) IIB-+B' ~
~ c II T (s)- T (so) 1/B-+B, II <l>n llsss<~>-+ 0
for £-+So·
Let In-+ f in 1& s (cr X cr). Then
II t (A (s))- t (A (so)) IIB-+B' ~II t (A (s))- fn (A (s)) IIB-+B' +
+II In (A (s))- In (A (so)) IIB-+B' +II In (A (£o))-
- f (A (so)) IIB-+B' ~ c {II f-In llsss<'Jl +
+II T m- T (so) IIB-+B' II <Dn llsss<~>}·
Fix s > 0 and choose such an n that the inequality
will be fulfilled. For a given n there exists such a 8 > 0 that
II T (£)- T (so) IIB-+B' II <Dn llsss<~> < ~

for ls-£ol<l'l. Consequently, for ls-sol<l'l the following ine-
quality is carried out:
II t (A (s))- t (A (sc)) IJB-+B' < cs.

Problem. Formulate and prove the multidimensional analogue
of Lemma 7 .2.
s
Theorem 7 .2. Let the parameter vary over an interval of the real
axis, let the family of regular operators {T (s)} (excluding (B, B))
have a defining pair of spaces (B', B'), where B c B', and let for any
£' and
1
s"
2
of the interval the spectrum of the pair of generating sets
A (£'),A (s ") with respect to the embedding of B to B' be contained
in the closed set ~ c <J X a. Next, let there exist a derivative T' (£)
considered in the following sense:
limjj T(~+e)-T(£)_T'
e-+0 e
(s)l/ B-+B
,=0,
where II T' (s)/1 B-+B' < oo. Let f E.99 s (cr). Then, if
<l>(x, y)= f(~)-f(y) . E.99s (~), xER 2 , yER 2 ,

Xt + 1X2-Yt-1Y2
then in the topology of the space Hom (B, B') there exists the deriva-
tive :~ f (A (~)) so that the following formula is valid:
-:- f (A(~))= T' Cf.) dl ~~))- ~ (l (~)) .

~ T (~)-T (~)
Corollary. The following estimate is justified:
II:~ f (A (~)) //B~B' ~c II T' (s) IIB~B' 1\ x1: ~=~=~~~ iy2 11.93 8 <};>'
Jl e-iA(~liJJB~B
where c depends only on sup and on
tEH2 (1+Jtl)s
Jl e- iA (~) t IJB'~B'
~~ (1+jtJ)S •
Proof. From the existence and the boundedness of the operator

T' (~) it follows that II T (£ +
e) - T (£) IIB~B' < oo. By virtue
of Lemma 7.1 we have
f (A (~+e))- f (A(!;))
8
1 3
(7.3)
8
T (l;+e)-T (!;)
where T (£ +
e) - T (£) is the extension of the operator T (£ e)- +
- T (£) to the homomorphism B-+ B'. Passing in (7.3) to the
limit for e-+ 0 and using the multidimensional analogue of Lem-
ma 7.2 we obtain the proof of the theorem.
It is not difficult to formulate and prove the multidimensional
analogue of Theorem 7.2 for the functions of a number of generating
sets. The definition of such functions can be obtained almost word
for word by using a similar definition from Sec. 4.
Sec. 8. Formulas of Commutation

Let A and B be generators of degrees s and s', respectively, acting
in a Banach space B which contains a linear manifold E dense
everywhere and invariant with respect to A, B, eiAt and e-iBt.
In this section we shall assume that the following condition of
agreement between the operators A and B is carried out: for any
function cp E C0(R) the operators cp (A) B and Bcp (A) are defined
on E and are closed.
CH. II. CAIJCULUIS OF NONOOMMUTATIVE OPER.ATIORJS 257
Lemma 8.1. Let the operator A as well as the commutator [A, B)

be bounded. Then, for any f E fJ!! s+l (R) the following formula is valid:
[B, f (A)]= [B, A] f

2
(1~ -~ U) '
A-A
where the line above means the closure.
Proof. Let cp E CO' (R) be a real function which is contained bet-
ween 0 and 1 and is equal to 1 in the neighborhood of zero. Denote
<vn (x) = cp ( ~ ) • Set An = cpn (B) A cpn (B) and first prove the
formula
[B, f(An)J= f U~)-~ UJ 2

[B, Anl· (8.1)
An-An
\Ye have
fAn, B)= cpn (B) Acpn (B) B- Bcpn (B) Acpn (B)= cpn (B) [A, B] cpn (B).
This means that the right-hand member of (8.1) is equal to
[ {jln (B) [B, A] {jln (B)] f U~)- !3 UJ =

An-An
= [cpn (B) BAcp 11 (B)] f U~) -~ UJ-

An-An
- [ Cfln (B) ABcpn (B)] f Un)- /3 UJ =

An-An
=[AnB]f Un) -f3 (1J

+[BAn]f Un) -f3 (1J,
An-An An-An
Denote Bm = Bcpm (B). Then for a fixed n and sufficiently large m
A 1]/
[2[B' n UJ -f (lJ
1 3
= -[2A B ]f
nm
UJ -t (lJ +
1 3
~-~ ~-~
+[2BT]'
mn
UJ -t UJ _-1 nm
1 3
i3 t UJ -t UJ +
- 1 3
An-An An-An
2 1 1
( ') ( )
3 2 3 1
+ BmAn f A~ - ~ An = Bm ( - f (An) + f (An)) =
An-An
= [Bm, f(An)].
17·.. 01225
258 OPE-RATIONAL METHODS
Thus for a fixed n and sufficiently large m the formula is valid:
[[B, AnJ] f (A~)-~ (lJ = [Bm, f (An)J.

An-An
In particular, this means that the right-hand member of this for-
mula does not depend on m. If h = 'ljJ (B) q, where 'ljJ E Co (R), then
f (A-n) Bmh = f (An) Bh
for sufficiently large m. Consequently, Bmf (An) h does not depend
on m for a sufficiently large m and since IPm (B) -+ 1 point-by-point,
B = Bcpm (B) and B is a closed operator, it follows that
Bmf (An) h = Bj (An) h
for a sufficiently large m. This means that (8.1) is valid in a set
dense everywhere and, consequently, it is valid everywhere.
In (8.1) we shall pass over to the point-by-point limit for n-+ oo
with the additional assumption that the Fourier transform of the
function f has a compact support. For any h
(8.2)
uniformly in t in any interval a ~ t ~ b and

II e-iAnt \l~c (8.3)
for t E [a, b]. This follows from

N
e-iAnt- ~ (iAn)k -0 (e)
Ll kl - t '
k=O
where Ot (e) does not depend on n. Passing over to the limit in the
right-hand member of (8.1) for n-+ oo, by virtue of (8.2) and (8.3)
we obtain
As to the limit in the left-hand member of (8.1) we consider f (An)-+

-+ f (A) point-by-point. Hence f (An) B-+ f (A) B in E. Conse-
quently, the sequence {Bf (An) h} converges for any hE E. By
virtue of the closed nature of the operator B, we have, for h E E:
lim Bj (An) h = Bj (A) h.
n-+oo
ICH. II. CAIJCULUS OF NONOOMMUTATIVE OPER.ATIORIS 259
Thus, for any h E E the following formula is valid:
[B, f (A)] h= lB, A] f (1)1 - !3 (1) h.

A-A
The lemma is proved for the function f which has a Fourier trans-
form with compact support. Since the set of such functions is dense
in 1IJ s+l (R) the lemma is valid in the general case too.
The following theorem states that the assumption that the opera-
tor A is bounded may be rejected.
Theorem 8.1. Let the commutator [A, B) be bounded. Then for
any f E 1IJ s+I (R) the following formula is valid:
[B, f(A)J=[B, AJ 1 U~-~ (1).

A-A
Proof. Let {ern} be the sequence of functions considered in the
proof of Lemma 8.1. Set Bn = ern (A) Bern (A), An = Aern (A).
Consider the commutator
[Bn, AnJ =ern (A) Bern (A) Aerm (A) -Aerm (A) ern (A) Bern (A)
Let n be fixed. Then for a sufficiently large m we obtain
[Bn, Aml h = ern (A) [B, AJ ern (A) h, hE E.
Consequently, for a sufficiently large m the operator [Bn, Aml does
not depend on m and is bounded in norm by the number II [B, A] 11.
By applying Lemma 8.1, we get
[
B n• f (A m)l = :,[) CJln (A) [B ' A l ern (A)] f UJ - f UJ
1 3 •
Am-Am
For a sufficiently large m this formula may be rewritten in the
form
( 3) 2_
[Bn, f(Am)l=CJln A [B, Alcpn A
( 1) f (l) -1 (l)
1 3 • (8.4)
A-A
Let h E E. Then f (Am) Bnh = f (Am) CJln (A) Bcpn (A) h =
= f (A) ern (A) Bcpn (A) h = f (A) Bnh, if m is sufficiently large for
a fixed n. In the same way, Bnf (Am) h = Bnf (A) h for a sufficiently
large m. Consequently, the following formula is valid:
3 ) 2-
[Bn, f (A)]= CJln ( A [B, A]
f u) - f u)
1 3 CJln
(A 1)
.
A-A
17*
260 OPEHATIONAL METHiO,DS
In this formula let us pass over to the limit for n-+ oo. The right-
hand member converges point-by-point to
[s, Alt U} -: (l).

A-A
Concerning the left-hand member we may say that for h E E, q =
= 'ljJ (A) h, 'ljJ E Co (R) we have
[Bn, f (A)] q = CfJn (A) Bcpn (A) f (A) 'ljJ (A) h-
- f (A) CfJn (A) Bcpn (A) 'ljJ (A) h =
= CfJn (A) [B, f (A)] CfYn (A) 'ljJ (A) h.
If n is sufficiently large, then CfYn (A) 'ljJ (A) h = 'ljJ (A) h = q so
that
lim [Bn, f (A)] q =lim CfJn (A) [B, f (A)] q = [B, f (A)] q.
Since the linear core of the set of elements of the form 'ljJ (A) h,
whereljl E C'f:' (R), h E E, is dense everywhere, the theorem is proved.
Theorem 8.2. Let the commutator [A, B] be bounded. Then for any
I (x, y) E .%' •+1, •'+1 (R2 ) the following formula is valid:
1 2) ( 2 1) 3 _ 62j ( 2 4 1 5)
f ( A, B - / A, B =[A, B] 6x 6y A, A, B, B • (8.5)
Proof. It suffices to consider the case when f (x, y) = / 1 (x) f 2 (y)

since the set of linear combinations of such functions is dense in
.96's+!, s'+ 1 (R 2). Then (8.5) is obtained by passing over to the clo-
sure. We have
[ft (A), !2 (B)]= lit (A), B] fz (A~ -~2 (~).

B-B
Next,
[ft(A). B] = lA, B] It (1) 1 - /3

1 (l).
A-A
For this reason
fz (A) -fz (~)

1 5
B-B
Q.E.D.
CH. II. CALCULUrS OF NONOOMMUTATIVE OPERJAT~ORIS 261
Theorem 8.3. Let f (x) E nSf

s
s (R) and j (x, y) E n Sf
s, s'
s, s' (R2) and
let the commutator [A, B] be bounded. Then the following formula
is valid:
f([g (l, .Bh) =[/ (g (l, .B) h+

5__ 15g ( 3 7 9) 15g ( 3 4 6)
+[fA, B]& A, A, B bY A, B, B X
{52f ( ( 1 9) 2 ( 1 2) 8 ( 1 2) )
X l5x 2 g A, B , [g A, B ], [g A, B ] ]. (8.6)
Proof. If the operator A is bounded, then the proof is identical

with that of Introduction.
A
Let the operator A be unbounded. Set An AZ • Then An
1+-
n
is bounded and II e-iAn t II ~ c (1 +
I t l)S, where c and s are con-
stants independant of n (see Chapter I). Besides
IJ[An, BJII~II[A, BJIIII g;~ llsss,s'<R>'

X
where fn (x) = - -
2 •
1+::._
n
We have
X y
1+ x2 -1+ y2
~,2jn(x, y)= nx-y n =62/1 (.Vii' JJ·
Hence it follows that
1162/n (x, y) llsss, s' (R2)~1162j1 (x, y) llsss,s' (R2)·
Therefore, the commutators [An, B] are bounded uniformly in n.
If in (8.6) A is replaced by An~ then the obtained formula will be
valid. Passing to the point-by-point limit for n -+ oo in the latter
formula, we obtain (8.6) for the general case, Q.E.D.
Sec. 9. Growing Symbols

Let A 1 and A 2 be generators of degrees s1 and s 2 , respectively,
acting on a Banach space B. As usual we assume that A 1 and A 2
are defined in the same set E dense everywhere. The closures of the
operators A 1 and A 2 in B we shall again denote by A 1 and A 2 , res-
pectively.
262 OPERATIONAL METHiOIDS
Definition. Let there exist such non-negative integers k, l for the

function f (x, y) that the function
( f (x, y)
g x, Y) = (x+i)k (y+i)l
belongs to space !18 81 , 82 (R2). Then set
( 1 2def (1 2)
fAt, Az)=(Az+i) 1 g At, A 2 (At+i)kh (9.1)
for any vector hE E, so that g

domain of the operator A~.
Lt, .AJ (A 1 + i)kh belongs to the
Theorem 9.1. Let (in the notation of the definition)

k ' .~
. . . _ k ' l' .-::?
. . . _ l ' g ' (X, y ) = f.(x,
k'
y)
. !'
(x+~) (y+~)
and h be such a vector in E that
g Ct. .AJ (At +i)k hE D Al· 2

Then
(A 2 + i) 1 g (.,.L, .AJ (At+ i)k h = (A 2+ i( g' (At, .AJ (A + i)k• h.
2
(9.2)
Proof. We have
g(x, y)=g'(x, y) (x+i)k'-k(y+i)l'-l.
Let the sequence {<Jln} of functions of the form
mn
<Jln (x, Y) = ~ U..jn {x) ~Jn (y), U..in E!18 81 (R), ~in EBs2 {R)
j=i
converge to g(x, y) in !l881 , 82 (R2 ). Then the sequence of functionR

mn
'I' n (x, Y) = ~ Yin (x) flin (y),
5=1
~nW ~nM
where Yin (x) = (x+ i)k'-k and (y) = (y+i)l' _ 1
flin , converges to
g' (x, y) in !18 8 1, 82 (R 2). For any q EE we have
( 1 2 ) ~
<Jln At, Az q = k! ~in {Az) U..jn (At) q =
j=1
CH. II. GAIJCULUIS OF NONOOMMUTATIVE OPEIMT:ORIS 263
Passing over to the limit for n-+ oo we obtain
g LL. lJq = (A2 i)l'-l g' + Ct. lJ

(A1 i)k'-k q (9.3) +
(here we have used the closed nature of the operator (A 2 i) 1'- 1). +
Putting q = (A 1 +
i) "hand applying the operator (A 2 i) 1 to both +
members of (9.3) we obtain (9.2).
Theorem 9.1 shows that the element/ Lt. A.J
h does not depend
on the choice of powers k and lin (9.1). Note that if the right-hand
member of (9.2) exists, then the left-hand member also exists. This
follows directly from formula (9.3).
Theorem 9.2. Let f Ct. .AJ

be an operator defined everywhere
in E and bounded in B and let T be the closure ofl this operator in B.
Then for any function g E ~ 81 ,82 (R 2) and any h E E the following
formula is valid:
[fg]
1 A2)2 h = Tg2(1A1, A3) h.
(A1, 2
Proof. Let
f LL, lJ h= (A + i) Ct. lJ (A + i)" h,

2 1/ 0 1
where
fo(x, Y)=f(x, y)j(x+i)h(y+t) 1, fvE~8 1 ,8 2 (R 2 ).
Then for any h E E we have
2 ( 1 2 )
Tg A 1 , A 2 h=
= 2~ Jg(t,
R2
T) e-iA2-r (A 2 + i) 1 fo (1 1, AJ X
X (A 1 + i)" e-iA t h dt dT = (A 2 + i) 1 2~ Jg(t,

1 T) X
R2
xe-iA2-rfo (1 1 , lJ e-iA2t(A+i)hhdtdT=
= (A 2 + i) 1 [f0g] (1 1, AJ (A + i)" h = (fg) (1t. AJ h.
1
Corollaty. The following estimate is valid:
II (fg) (11, lJ IIB~c I f (11, lJ liB II g llss 8 1• 8 2

R2• (9.4)
where c is the norm of the homomorphism &It: .98 8 1' 82 (R 2) -+
1 2
-+Hom (B, B) constructed on the ordered pair of operators A 1 , A 2 •
264 OPERATIONAL METHiOJ:JS
Next, we shall need the symbols f (xh ... , Xn) satisfying the
condition
f (xb . • ., Xn) E $3' (Rn)
(xt+i)k •t· ... ,sn
for some integer k. If A 1 , • • • , An is a set of generators of degree~
s10 • • • , sn, respectively, acting on a Banach space B and defined
on a linear manifold E dense everywhere, then we set, for any h E
E E,
1 n ) def ( 1 n )
f ( At, ... , An h=g At, ... , An (At+i)~th,
where g (x1 , . . . , Xn) = f (xu ... , Xn)l(x + i)~t. As in Theo-
rem 9.1 this definition may be shown not to depend on the choice
of k.
It is not difficult to verify that in this case the estimate analogous
to (9.4) is also valid: if f 1, ••• ,
in B and g E 53' 81 , ... , sn (Rn), then
(A .AJ is an operator bounded
II (/g) Lt, ···•;J IIB~C II f CL ···' AJ \\B II g 11.58 sf' .•. ,sn
(Rn)'
where c is the norm of the homomorphism alii: 53'S!, ...• sn (Rn)-+-

1
-+- Hom (B, B) constructed on the ordered set of operators A 1 , .••
n
• .. , An.
Problem. We have defined infinitely differentiable functions of
two non-commutative generators with a growth rate not exceeding
the degree of the argument. Show that they form an algebra with
the ~-structure (with the same proviso as in the footnote on page 210).
Finally, define growing functions of several non-commutative
operators. In this case additional assumptions are necessary. Name-
ly, let {B-r} be a Banach scale, where B-r c B-r' for 't < 1:'. Let A 11 • • •
. . . , An be generators in the scale {B-r }, where the operators are also
translators in the same scale. Denote by c!f"" the space of functions
f (x1 , • • . , Xn) growing with all their derivatives at a rate not exceed-
ing 1 x lk (power k corresponding to the junction f).
For any function f (xu ••• , Xn) in rif"" there exist such integers
k1 , . . . , kn that
( ) f(xb ... , Xn)
g X1, • • ·' Xn = (Xt + l·)kt •••
( Xn + l·)kn
belongs to 53' z1, ... , ln (Rn) for any lh .•. , ln. Set
( 1 n ) def ( 2 ) k1 ( 2n ) hn ( 1 2n- 1 )
j At, ... ,An = At+1 ... An+1 gAt, ... ,An ,
(9.5)
CEI. II. C:AIJCULUS OF NONCIOMMUTATIVE OPER.AT:O·RIS 265
where the operators (A 1 + +

i)h1, ... , (An i)hn are considered a~
translators. The investigation of the correctness of the definition
is left to the reader.
Problem. Verify that in the algebra of generators and translator&
in the scale {B,;} formula (9.5) defines a fl.-Structure with an
operation f-t: (x1 -+ AH ... , Xn-+ An) translating f (xH ... , Xn)
into f (..t, ... , .4J .
Sec. 10. The Factor-Spectrum
Let {B,;} be a Banach scale and A and A' be generators in the
scale of degrees s and s' with steps k1 and k 2 , respectively. Let T
be a translator with step l. Let S 1 and S 2 be operators defined on D
and acting on B,; for any -r. We shall call the operators S 1 and S.,.
the equivalent operators if, for any 't and 't',
II sl- s2 ll"t..,."t' < oo.
Definition. We shall say that a point ')., E R 2 belongs to the factor-
resolvent set of the vector-operator (A, T, A')
if there exists such a neigh-
borhood U of the point A that .for any function f E f8 s, s' (R 2 ) with
support in U and for any 't and 't 1
II h Ct. A') IL"t' < oo,

2 ( 1 3 )
i.e., the operator Tf A, A' is equivalent to zero.
The complement in R 2 of the factor-resolvent set will be called the
( 1 2 3 )
factor-spectrum a of the vector-operator A, T, A' .
Lemma 10.1. Let f be a finite function in f8s, s' (R 2 ) and the opera-
2 ( 1 3 )
tor Tf A, A' be equivalent to zero. Then any point ')., for which
f ('A) =1= 0 belongs to the factor-resolvent set of the vector-operator
1 T,2·3)
(A, A' .
Proof. Let U be such a neighborhood of the point ')., that f ('A) =I= 0
for ')., E U and let cp be a function in $ s, s' (R 2 ) with support in U.
Then there exists such a function '¢ E f8 s, s' (R2 ) that cp = '¢/.
According to Theorem 4.1 we have for arbitrary ~t
II Tcp (l, A') 1Lt+h1+h2W~

~c II Tf (A, .J.t) ll"t-l-k1-+t+k1+11 II'¢ 11.58 8, 8, (R) < oo,
Q.E.D.
Theorem 10.1. For the point ').0 to belong to the factor-resolvent set
1 2 3 )
of the vector-operator ( A, T, A' it is necessary and sufficient that for
.sufficiently small e > 0 there exist functions <p, 'ljl E Co
(R) with
.support in the e-neighborhood of the points ').~ and ').~, respectively,
.such that the following condition is carried out:
<p ('A~) = 'ljl ('A~) = 1
and that the operator

'ljl (A') T<p (A)
is equivalent to zero.
Sec. 11. The Functions of Components of a Lie Nilpotent Algebra

and Their Representations
This section will be mainly devoted to the study of functions of
differential operators which form an ordered representation L 11 • • •
. . . , Ln, a of components of some Lie nilpotent algebra (see Sec. 9
of Introduction). We shall show that these operators are generators
and translators on a suitable scale and prove the relation (9.5)
of Introduction for the case when, instead of a polynomial, there
exists an arbitrary function in Cz.
We shall also prove formula (8.10)
which is necessary for the proof of the main theorem. Besides, we
shall obtain formulas for reducing a function of the ordered opera-
1 2 n n+1
tors Lll L 2 , • • • , Ln, a to a pseudodifferential operator.
First point out some properties of the ordered representation.
Let X be a set of components All ... , An of a Lie nilpotent
algebra (see Sec. 9 of Introduction). Since, by hypothesis, any com-
mutator of length N is equal to zero, it follows that a commutator
of length N - 1 commutes with all elements of the set X. By hypo-
thesis, all commutators of operators are included in the set of ope-
rators All ... , An-
Let X 1 be a subset in X of elements commuting with all elements
in X. Let X 2 be such a set of elements that their commutator with
any element of X belongs to X 1 . In general, let Xi c X be such
a subset of elements that their commutator with any element belongs
to the sum X k for k < i. Obviously the number of such sets Xi
does not exceed N.
Now consider the construction of an ordered representation of
the set X. The method of its construction is described in Sec. 9
n+1 ( 1 n )
of Introduction. By transferring in Aiq A 11 • • • , An the opera-
tor A i from position n + 1 to position i we obtain the expansion
CH. II. CAIJCULUS OF NONC:OMMUTATIVE OPERAT'ORIS 267
of q
n-t-2
Ct....,A.J in a Taylor series in powers of An at the point
An. Hence we obtain the terms of the form
( n n+2)"n+1
An -An A; iJh ( 1 n-1 n+2)
k'
.
"
iJxn
q At, · · · • An-t• An •
1 3 )" 2
Let An E X;(n)· Then ( An -An At (the right-hand commutator
being of the order k) belongs to such a class Xi that j ~ i (n) - k.
Similarly one may conclude that, in general, the coefficient of the
derivative iJxit • ~~ iJxi of the operator L; is a constant or a linear
h
homogeneous function of Xr such that i (r) <max i (j 8 ).
1:o::;s:%:k
Note that this remark is also relevant to the construction of the
representation G; defined in Sec. 9 of Introduction.
~ ( 2 1 )
Lemma 11.1. Let L = L x,- i8/8x be an ordered representation
of one of the components of the Lie nilpotent algebra. Then, for any
integer k or l there exist such integers s (k, l), m (k), n (l) that
II e-iLt llcm(h)_,.ch
l n(l)
~const (1 +It l)s(k, 1),
where c~ is the space of k times continuously differentiable functions

with the finite norm
h.
, 0 \vl
II <p lick= sup L} (1 +I xd) 1 - , <p (x).
l E n iJx'
X R IYJ=O
Proof. Denote by H7 the completion of the space S in the norm
n
l k
11 .
l n(l)
Let F x-+~ be a Fourier transform. Denote L = F x-'f.LF~l._,x. Since

the Fourier transform of the space H~ is II1, it suffices to obtain
268 OPERATIONAL 1\'IETH!ODS
the estimate
lle-iLtiiHz Hn(Z)~const(1 +I ti)s(k,l).
m(k)-+ k
The operator L has the following form (the coordinates are num-
bered respectively):
r1 r2
L=i ~ ai a~--i ~ Pi(St• · .. , Sr1) a~1 - • ·•

• 1 J • 1
J= J=ri+
rp+1
... - i ~ Pi(St> ... ,£rp)a:.- •.•
. 1 J
J=rp+
n
···-
.
~
J=rq+ 1
Pi (sh · · · , Sr a) a:. + Po (s),
J
where Pi are polynomials with real coefficients and ai are real num-
bers. If '¢ (s) E S, then the function '¢ (£, t) = e- 1L 1'¢ {s) is a solu-
tion of the following Cauchy problem:
fJijJ ~ fJijJ ~ fJijJ •

{ m+ ~ ai a£ 1 +. LJ Pi a£ 1 + lP '¢ = 0 0,
J=1 J=r 1+1
lJl <£. o> = (jl <£>.

Next, write the characteristic equation for this problem:
( d£i .
dt=ait ]=1, ... ,rtt
l
a: = Pi,
d£·
j = r1 + 1, ... , n,
t s(0) = s 0' tP (0) = (jl (s 0 ).
By solving the first n equations successively we obtain

sJ+ait, j=1, ... ,ri,
Si=Si(S0 , t)= { s~+Qi(S~ •... , s~p' t), j=rp+1, .. . ,rpH•
where Q1 are polynomials. Hence we have
s;-ait, j=1, ... , rit
{
+
so= so(£, t) = Si Ri (st, · · ·, srp• t), j = r P + 1, ... , r p+h
(11.1)
CH. II. CAIJCULUS OF NONOOMMUTATIVE OPER,AT:ORIS 269
where Ri are polynomials. Let 1ji (~ 0 , t) be a function tp (~, t) of

coordinates ~ 0 and t. From the latter equation of the characteristic
system we have an explicit formula for 1ji (~ 0 , t)
t
1ji (~o, t) = cp (~o) exp { - i) P 0 (~ (~ 0 , T)) dT}. (11.2)
0
n~
Note that D:o = 1. Consequently
II e-iLt fJli!I2 = ) 11il (~o, t) 12 d~o =II rpJI£2.

By putting in (11.2) ~ 0 = ~ 0 (~, t) and applying to both members
an operator of the form ~ 13 ( :£ r and using (11.1) in analogy with
an estimate in L 2 , we obtain the required estimate II e-i'Lt II 1 nU>
Hm(h)-Hh
for the case l ~ 0. Here, one may even set n (l) = l. Besides, for
a given m (k) one may always choose k = k (m). Now make use
of the fact that the spaces H~ and H=~ are conjugated in relation
to the scalar product in L 2 • In other words, the norm
II fJlll~ = sup I (cp, 1~)L 2 1, rp, tp E S
ll'~'llu=~=1
is equivalent to the norm in H~. The operator conjugated to the

operator e-iLt in L 2 has the form eiu since L is symmetrical in L 2 •
Consequently
II e-iLt rpllk"l = sup I (e-iLtcp, 1Jl)L21 = sup I (cp, e-iLt tp)L21·
II¢11HI =1 ll¢11HI =1
k -h
If l > 0, then, according to the relation proved above, there exist

such numbers k' and s that for II tp II 1 = 1
H_k
Jl e-iLt1jJ IIH1 ~const (1 +It I)'.

k'
For this reason

II e-iLt rpiJk"' = sup
II¢11H 1 =1
II cp llk",lll eiLt 1jJ IIHl
h'
~ const II cp llk",l (1 +It ns.
-k

We shall now pass over to the proof of (8.5) and (8.10) as well
as of their analogues given in the item "the rule of reduction of the
main problem" (see Sec. 9 of Introduction).
270 OPERATIONAL :VrETHIODS
As it has been mentioned, this proof is based on axioms (~-t 7 ) and

(~-t 8 ).
It consists of two parts.
(1) Let g (x) E Coo; then the operator
iA t
e kg A 1 , A 2 ,
( 1 2
••• ,
n)
A = e
t_i2 t
" g A 1,
( 1
••• , An
n )
(11.3)
1 2 n )
is equal to the operator'¢ ( A 11 A 2 , • • . , An, t , whose symbol is
expressed by the symbol g (x) in the following way: '¢ (x, t) =
= eiLkt g (x). Here Lk is an ordered representation of the opera-
tor A k· The operator exists by virtue of the established properties Lk.
Proof. Consider derivatives of the symbols eixn+2 1 g (x) ctcf h and
'¢ (x, t) with respect to t, i.e., the symbols ixn+ 2h and '¢i (x, t) =
= iLk'¢ (x, t). It is obvious that for the former the relation (~-t~)
of axiom (~-ts) is fulfilled. Bear in mind that the operation 11 'brings
n+2 k ~
into correspondence Xn+ 2 and Ak and xk and Ak· Since Lk is an
ordered representation of A k it follows that
~
[iLk'iJ (x, t)] 1 n =
(1 2
iA" ['IJ A1, A2, ... , An ].
n)
Xt=At•·, .,xn=An
Hence, it follows that the relation (~-t~) is also fulfilled for the
iA t ~ ~
symbol'¢ (x, t). At t = 0 the two operators'¢ and e " g are equal
to g. For this reason their difference satisfies the relations (~-t~)
and (~-t~). This means that the operators are equal, as required.
From the second part of the axiom (~-t 7 ) it easily follows that equa-
lity (8.5) can be extended to symbols of the class C'f.
(2) Consider the operators
n n+1+k
i ~1 Ah tk { 1 n ) ( 1 n )
e"= gAt. ... ,An and¢ A1o ... ,An, t 1 , ..• ,tn,
(11.4)
n k
i ~ Lkth
where '¢ (x, t) = e k= 1 g (x). From the above item it follows
that they are equal.
Let the pseudodifferential operator
f ( L, {)t1{) '
. {)
L {)t2 ' ... ' L {)tn
. {) ) def
=
j ( , {) )
L 7ft '
act on the symbols of both operators (11.4), where f E r!f 00
• By virtue
of the properties of the operator Lk we have
n k
~
f
.
(LTt) {)
'IJ(x, t)=e
i
k- 1 / ( Lt.
Lk tk 1 n
... , Ln g(x).
)
CH. II. CADCULUS OF NONOOMMUTATIVE OPERATORS 271
Obviously, for symbol (11.4) we have
n
. ~ t
' ~ xn+1+h h
= e h=i f
(xh · · ·, Xn)•
(Xn+b · · ·, Xn+Hh) g
Because of axiom (~t 7 } the corresponding operators are equal for all
t (t = 0 included). Namely, for t = 0 we obtain the equality
(1 n) (1
[f A1, ... , An ] [g A1o ... , An ] --: \jJ A1o ... , An ,
n) (1 ")
where 'ljJ (x) is expressed}by g (x) in the following form:
\jJ (x) = f (£1. ... , £J g (x)

as required.
The other equalities of Sec. 8 are proved in a similar way. There-
exists an obvious generalization for the case when the argument
n+1
is B.
We shall again use formula (8.10) for a complex-valued function
of ordered operators. Let fi C.L, ... ,.AJ EM, i = 1, ... , k.
Then j (J~, ... , JJ can be represented in the form of the funr,tion
'ljl(A 17 ••• , .AJ, the symbol 'ljl(x) being defined by the formula
J([J1 (Lb · .. , LJ], · .. , [Jh (£1. ... , L71 h) 1='\jl(x). (11.5)
Here 1 is a unity belonging to the space of symbols J'"". We shall
apply the above formula to describe the function of representations
I n
L 1 , • • • , L 11 as a complex function of the components of another
nilpotent algebra, namely, that of - i :x ,
x, in other words, we shall
1 n
write the function of L 17 • • • , Ln in the form of a pseudodifferential
operator. Thus we shall obtain an ordered representation of opera-
. a
t OI'S -lfiX' X.
It is obvious that
( l-. a
3 )
OXj f
(
-l-
. a1
Dx '
~) =•h. (-il
't'JDx ' ~)'
wl1ere lJli (y, x)--:- { - i a~i + Yi) f (y, x).
272 OPERATIONAL :11ETHODS
Consequently, .the operators ( - i 1x +y, ~) constitute an orde-
red representation of the set ( i 1x , ~). Then, from (11.5), it fol-

lows that
t(lb ... ,EJ=cvU. -ifx),

where
cp(x, y)=f i
([£1 (~, - i +Yh.
... , n[Ln (2x, -~ax-+Y
. J )]) 1
is the symbol of the pseudodifferential operator.
III. ASYMPTOTIC METHODS
Sec. 1. Canonical Transformations of Pseudoditlerential

Operators
We have already seen in Sees. 1 and 2 of Introduction that the
commutation formula of a pseudodifferential operator with an ex-
ponential plays an essential role in Heaviside's calculus. The com-
mutation formula of Hamiltonian with exponential featured in
Ch. II in the construction of the calculus of ordered operators is also
a key formula. In the theory of I-t-structures (rather in its application
to pseudodifferential operators) this formula was in effect accepted
as the definition of functions of operators
.1 a+ as
- Z ~ r)x Tx ·
It should be noted, however, that the commutation formula of
• • • • .!:_S(x)
the pseudodifferentlal operator With exponential e h of the form
i i
e- h S(x) p ( - ih _!_)
fJx
eh S(x) = p ( - ih _!_
fJx
+ fJSfJx )
iS( )
may be interpreted in two ways. By taking T = eT. x as a multi-
..i.scxl
plication operator by the function e h we obtain
(1) r-' P ( - ih _!_) T = P ( - ih _!_ + as )

fJx fJx fJx
or
(2) T* P ( - ih _!_) T
fJ.r;
= P ( - ih _!_ -l- .!§__)\ .
fJx ' fJx
. (2 1)
Up till now, for the case when T = eh ' cp x, p we gene-
...!:..sxp (2 ')
ralized formula (1) and thus obtained the commutation formula
of Hamiltonian with exponential. This section is devoted to the
18--01225
274 OPERATIONAL ·l\1ETHiODS
study of the operator
T*P (': -ih

' 1)
ox T, where
_2s(~,~)- (. 1 2)
T* = e h <p x, p .
A transformation of this type will be called the canonical trans-

formation of the operator P ( :, ; ) . It turns out that in many cases
it is more convenient to use the canonical transformation but not
transformation (1) of the differential (or pseudodifferential) opera-
tor P ( ;, - ih l ).
First, we consider the composition
T *T = e ~ [ 8 (;,~)-s(;,~)]
<p
(2x, pt)-(2 3)
<p x, p ,
where S is a real function.

From Theorem 1.1 (see below) it follows that
T*T = cp
(~
X
(2x, p,1 p3) 'p1) cp (~ ( x,2 p,1 p3) ' p3)
X
IJ ( ~ (2 p,1 p3) ; p;1 p3) I

X X,
,
where x (x, p, p') is a solution of the equation
~ {)S ~
x+6P (x; p, p') = x
and it is supposed that the Jacobian J does not vanish:
~ 0 {)S ~ 0
J (x; p; p )
'
== 1 +---;::-
OX
fJP ( x;
I
p, p ) =I= •
The operator T*T has the following symbol:
e
! [S(x, p)-S(x, p')] (
<p x,
) - (
p <p x, p .
')
It turns out that with an accuracy of up to 0 (h the operator T* T 00

)
depends only on the values of its symbol on the diagonal p = p'.

If the terms of the order 0 (h are omitted, then the condition on
00
)
the Jacobian J may be also weakened.

Let us formulate precisely and prove the above assertions.
CH. III. ASY'MPTO•TIC METHODS 275
Lemma 1.1. Let <1> E Coo (Rn X Rn X Rn) and let it grow no
faster than a polynomial. Then
( 3 x,2 p1) f= 1 r j_(P·X-p'·x")

<D p, (2nh)2n Jell. X
X f (x") dx" dp dp' ) e! (p' -P) x' <1> (p, x', p') dx' =
= F.;;1xF x'-+pF.;;'~x'<l> (p, x', p') Fx"-+p'f (x"),
is the so-called 1/h-Fourier transform.

The proof is obvious.
In the space L 2 (Rn) we set the operator
. (2 1)
T =cp x, p e( 2 1) ~Sx,p
, (1.1)
where cp E C~ (R2n) and SEC~ (R2n), S being a real function.

{)S
Define the vector 6P (x, p', p) ERn whose ith component is a
dilference derivative
{)S (
{)p; x; P1,' · · ·, Pi-!;
'
Pi,' P;; Pi+b · · ·, Pn ) ,
where p, p' ERn. Consider the equation

~
x + 6P (x;
{)S -
p', p) = y (1.2)
with respect to the vector X' ERn. Denote by J the Jacobian of

this equation
J (';;, p', p) = det ll()ii + 8~i g~ ('; ; p~, . · . , pi, Pi, · · ·, Pn) II·
Theorem 1.1. Under condition that J =1= 0 the identity holds
T*T = cp (~ (x,2 p,1 p3) ' p1) -cp (~ (x,2 p,1' p3) ' p3) X
X X
X IJ-i (:; (~, ~. ~)' ~. ~)I'

where "; (y, p', p) is the solution of the equation (1.2).
18*
276 OPERATIONAL "METHiOUS
Proof. By virtue of Lemma 1 .1

i[ (2 p1) -8 (2x, p3)] (2 1) (2 3) I-
T*TI-
- eh 8 X,
rp x, p -
rp x, p -
= 1
(2:n:h)2n j
r e ~ {P·X-P'·x")l( X
")d "d d
X p p X
1
Xje
r ~ [(p'-p)•x'+8{x', p')-8{x', p)] ( 1 1 ) -( 1
(j)x,prpx,p X.
) d 1
In the inner integral we pass over to a new variable y according

to the formula X 1 = x
(y, p 1 , p), where (y, p 1 , p) is the solution x
of equation (1.2). By virtue of the condition J =1= 0 the solution
of (1.2) ~ (y, p 1 , p) exists and belongs to C"" (R3 n). Taking into
account the identity
(P 1 - p)·Y = (P 1 - P)·;;+S (;;, P1 ) -S (x, p)
we obtain
T*TI= ( 2 :n:~) 2 n J i
eh(P·x-p'·x")l(x")dx"dpdp1 X
X ) e~ Y·(p'-p) I J-i (x (y, P1 , p), P1 , p) I X
x rp (x (y, P1 , p), P1 ) q; (; (y, p 1 , p), P) dy, ·
where rp E CO' (R 2 n).

The proof of Theorem 1.1 then follows from Lemma 1.1.
Example. Let S (x, p) = -(x- x 2 ) p; p, x E R, h = 1 and
rp (x, p) = rp (x). Then
Tl = rp (x) ei 8 (~. ~)I= rp (x) I (x 2}

(here
•o
b=- i fx because h=1).
In this case equation (1.2) has the form
~ (_;+';2) p 1 -( _;+;2) p ~
y= x+ p 1 -p
or y = x2 •
The Jacobian has the form J = This is why if rp (x) is equal2x.

to zero in the neighborhood of the point x = 0, then
T*TI= I cp (Vx) 12+lcp(- l/x) 12 e (x) I (x).
2Vx
CH. III. ASY1MPTOTIC METHODS 277
By the same method (i.e., by an analogous change of variables)

we shall derive a similar formula for the operator T 1 :
T f cter 1 . \ e~ [s(~.~)-!s) (~ 1) d~·f(x)

1 (2rth)n/ 2 J cp ' p
whose symbol is equal to the 1/h-Fourier transform of the symbol
of the operator T. The formula which is to be obtained will help
us to clarify the important role which the 1/h-Fourier transform
plays in canonical transformations of pseudodifferential operators.
Let J 1 (t p', p) be the Jacob ian of the equation
'6S - 6S --;;- -
Y=8jj (~; p', P)+B[ (£+p-p', ~; P) (1.3)
which is considered in regard to f (the vector 66~ is defined in the

same way as the vector ~~ ) . Thus we obtain
Jt(f, p', p)=
=detlla~i [66~ ~; p', P)+ g~ (~+p-p', ~ p)JII·

Theorem 1.2. Under condition that J 1 =I= 0 the identity holds
T!Tt = cp (-(2 1 3) 1)
~ x, p, p , p X
-(-(2 3) 3 3)]-1(-(2
X cp ~ x, p, p + p- p, p
1
J
1 3) 1 3)1 ,
~ x, p, p , p, p
1
1
(1.4)
where 1 (y, p', p) is a solution of equation (1.3).
Proof. By virtue of Lemma 1.1 we obtain
T!Td = (2rth) 2n J eh
1 r i < , ">
P·x-p ·x f (x") dx" dp dp' X
i ._ ,....,
X ) eh [S (\;, p')-S <Hp-p', p)] qJ {f + p- p'' p) cp ~' p') d~.
We make a change of variables f-+-

y in the inner integral accor-
ding to formula {f::3). By virtue of J 1 =I= 0 the solution f (y, p', p)
exists and belongs to Coo (R 3 n). By taking into account the identity
(p' - p) ·Y = S (f, p') - S (f + p - p', p)
we obtain the desired equality (1.4) as in Theorem 1.1, Q.E.D.
Now let us consider the general situation including the formulas
of Theorems 1.1 and 1.2. For simplicity let us restrict ourselves
278 OPERATIONAL METH!O·DS
to a two-dimensional case: x = (xlt x 2 ) and p = (PH p 2 ). Consider

the operator
v r
21) -X1S1
hi [ 8 ( (;1, X2,P 2] ( !: 2 1) df- j ( )
T f def _1_
2 - 2nh J e (j) 'oil .T2, p S! X • (1.5)
In analogy with Theorems 1.1 and 1.2 we obtain

i
T~Tz/=( 2 ~1 ) 4 ~ eh"<P·x-p'·x")f(x")dx"dp'dpX
X (ji (£t + P1- p~, z, p) cp (£~> z, p') dz d£ 1•

By a change of variables (£ 1 , z)-+ (ylt y 2 ) we wish to reduce the
exponent in the inner integral to the form
:! [(p~-Pt)·YI+(p~-pz)·Yzl= ~ (p'-p)·y.
Therefore, we set
Y1 = g: 1 (£1; z; P~, P1; Pz) + g: 1 (£1 +PI-p;, £1; z; p~, Pz), (1.6)
Yz = z + :~ {£1; z; p;; p;, Pz).
Then the following identity is valid:

Y · (p'- p) = (p~- Pz) · z+S (£~> z, p~, p;}-S (£t + P1- p~, z, P~> P2)·
Suppose that the Jacobian J 2 of system (1.6) is not equal to zero:
~~: o%zl)
lz (£~> z, p'' p) = det ( oyz oyz * 0. (1.7)
0~1 dz ·
Then there exist smooth solutions £1 (y, p', p) and z (y, p', p)
of system (1.6). In analogy with Theorems 1.1 and 1.2 after a change
of variables (£ 1 , z)-+ (YH y 2 ) we obtain
( (2 1 3) (2 1 3) 1)
TzT 2 =cp s1 .x,p,p ;z x,p,p ;p X
( (2 1 3) 3 1 (2 1 3) 3)
X q; Sl x, p, p +PI- PI; z x, p, p ' p X
xll; 1 (SI(~. ;, ;); z(~, ~. ;); ~;;)I. (1.8)
Thus, we have proved the theorem that follows.

CH. III. ASY'i.VIPTOTIC METHODS 279
Theorem 1.3. Let the operator T 2 be defined by formula (1.5) and

let condition (1. 7) be fulfilled. Then the identity (1.8) holds.
Note. In Theorems 1.1 and 1.2 it is sufficient to stipulate that the
J acobians J and J 1 should be non-zero only for (x, p) Esupp cp (x, p)
and (£, p) E supp cp (£, p), respectively. Besides, Theorems 1.1 and
1.2 can be generalized to cover the operator T*'ljl p - p T. Here
3 ('' 1)2
if the Jacobian J is non-zero for (p, p') E R2n such that p - p' E
E supp 'IJ, then the correlation is valid
3 (''p-p1) T=cp;;
T*'IJ
2 ( (2x,p,p1 3) ,p1) X
-(~(2 1 3) 3)1 (-(2x, p,1 p3) ' p,1 p3)1 'P (3p- p1) '
X cp p, p ' p J-i
X X, X
.r
where (y, p', p) is a solution of equation (1.1).
Under similar conditions Theorem 1.2 can be generalized:
:J ( 4 1) 2
TN p- p Tt = cp '£ x, p, p ' p
(~ ( 2 1 3) 1) X
X cp
1 + p-
3 p,1 p3)1 J~
£ x, p, p3)
-(~(2 1
(~(2
£ X,
1 3) 1 3)1 X
p, p , p, p
X¢(;_;),
where ~ (y, p', p) is a solution of equation (1.3) and the Jacobian
J 1 =I== 0 for (p', p) E R 211 such that p - p' E supp 'ljl.
An analogous assertion also takes place in the situation described
in Theorem 1.3.
Lenuna 1.2. Let the function 'ljl 0 (z) = 0 in a 6-neighborhood of
point z = 0, 'ljl 0 (z) = 1 outside a 26-neighborhood of point z = 0
and 'ljl0 E Coo (Rn). Next, let cp E Co (R 2 n) and S (x, p) E C00 (R 2 n),
S being a real function and let the Jacobian
JO=detll6··+o2S(x, P)JJ=i==O. (1.9)

1 ' ox; opi
Then, for any N and k>O lee have
II T*¢ (~- ~) T/lw~~w~~cN,

6 khk.
Proof. By virlue of (1.9) the equation

oS (x', p) + _ oS (x', p') + ,
ox' P- ox' p
has a unique solution in regard to p ERn, namely, p = p'. Thus,

for jp-p'j?:-6 the vector
a
ax' [ -S (x', p) +S (x', p') + (- P+ p') x'] ::;l= 0. (1.10)
By representing 6
3 (/• 1)2
T*'¢ p - p Tf in an integral
form it is not
difficult to verify, in analogy with Theorem 1.1, that the obtained
integral will differ from the integral in the proof of Theorem 1.1
by an extra factor '¢0 (p - p'). Since the derivative of the exponent
in the inner integral does not turn to zero on the support of
'¢0 (p - p') (see (1.10)), the inner integral may be integrated by parts
any number of times (by putting the exponential inside the differen-
tial). This immediately leads to the proof of the lemma.
Condition (1.9) can be shown to be essential here.
The meaning of Lemma 1.2 is in the fact that the operator T*T
_j_[S(x, p')-S(x, p)]
depends largely on the values of its symbol e" x
X cp (x, p') cp (x, p) on the diagonal p = p' (correct to 0 (h"")).
Thus, if in Theorem 1.1 the condition J ::;l= 0 is weakened by
replacing it by condition (1.9), then by virtue of Note of 1.1 and
Lemma 1.2 we obtain the theorem that follows.
Theorem 1.4. Let '¢0 be the function of Lemma 1.2 and condition (1.9)
be fulfilled. Then for a sufficiently small 6 the following equality is
valid:
T*T = cp (-(2x, p,1 p3) ' p1)1cp (-(2x, p,1 p3\J' p3) X
X X
X IJ- (~ U, ~. ;), ~.;)I (1-~,o (;_;))+A,

1 (1.2')
where II A llwN2 .... wN~cN,
2
khk for any N, k~O.
The latter lemma and theorem can also be formulated for the
case of the operators T 1 and T 2 • For the case of the operator T 1 we
note that the equation
~~ <£. p')- ~: <£+p-p', p)=O
has a unique solution p = p' if the Jacobian
J~ (£, p) cter det I (a~z:£i + a~z:Pi) (s, p)ll

does not become zero. For this reason, in analogy with the preceding
theorem, the theorem that follows is obtained under the same comli-
tions on S and cp.
CH. III. ASY'MPTOTIC METHODS 281
Theorem 1.5. Let J~ =I= 0. Then for sufficiently small6 the following
equality is valid:
TtTt={j) (% (~, ~. ~), ;) q; (% U. ~. ;) + ;_;,;)X
xiJ~~(I(~, ;, ;), ;, ;)1(1-lPd;_;))+A,
where IIAIIwN~wN~cN.khk for any N, k?;-0.
2 2
Theorem 1.3 can be reformulated in a similar way.
In the given formulas one may notice a certain duality in
regard to the operators T*T and T~T 1 • First of all we note that we
3 1
may set p = p correct to 0 (h). Then, instead of the J acobians J
and J 1 we may substitute the J acobians J 0 and J~. If we denote
<p 1 = V rp and <p 2 = <p then the obtained formulas will have
IJ0 I V IJ~ I
the form
T*T=/<Jltlz(J; (~, ;), ;) +O(h),
(*)
TtTt =I cp2/ 2 (~, (f
+0 (h), ;) , ;)
where 3; (x, p) is a solution of the equation
-x+ as -
ap (x, p)=x
and 1 (x, p) is a solution of the equation
as (t'
p
)+as~, p) __
aP ~, - X.
- a(;
Let us now dwell on the geometrical meaning of the observed
deality in the case n = 1. Consider a one-parameter family of
infinitely smooth curves in the phase plane (~, x):
~ = 8 (y, p), x = X (y, p), (1.11)
where the parameter p E R and y is a coordinate.
The one-parameter condition has the form
,Ty
ax ap
ax)
D =, det ( 03 03 -:f= 0.
ay ap
By a suitable change of coordinate y we may always obtain D = 1
in new variables *. Hence without loss of generality we can set
D=1. (1.12)
* This assertion is provrd in Chapter IV.
282 OPERATIONAL :HETIWDS
From this condition it follows that at degeneration points of

projection onto the axis x, i.e., at points where aslay = 0, the
derivative aslap is non-zero. Similarly, at degeneration points
of projection onto the axis £ we have aXIap =!= 0. At other points,
either axlap =!= 0 or aslap =!= 0.
This is why we can cover the family of curves (1.11) by the system
of intersecting neighborhoods, in each of which whether aXIay and
as I ap or a X I ap and as I ay are non-zero. The first will be called
non-singular patches and the second-singular patches. The set of
patches is called a canonical atlas of the family of curves.
Along with (1.11) we shall also consider a "shifted" family of curves
£= s (y' p) - p' X = X (y' p) 0
Consider a finite real function ~ (y, p) E coo with support in one

of patches of the canonical atlas.
We shall study two separate cases when the support ~ lies either
in a non-singular or a singular patch.
Case 1. Non-singular patches. Since axlay =!= 0 in this patch,
it is possible to make a change of variables x = X (y, p), hence
_ dcf
y = y (x, p) E coo. Denote £ (x, p) = S (y (x, p), p). We have
D=
-
D(X,B) I
D (y, p) y=y(x, p)-
_D(x,'t(x,p))X
D (x, p)
X D (X' p)
D (y, P)
I Y=Y
_,1
(x, P ) - i{ i{
0 0 fJX !!.._ =
fJy fJp
I 1 0
'
fJx fJp 0 1
consequently,
fJX i{
;:;-
uy
(y (x, p), p) :;-
up
(x, p) = 1
or
8% ( ) _ fJy (x, p)
fJp X, p - fJx • (1.13)
Let the patch under consideration contain the point (Yo (p), p)
and let x 0 (p) =X (Yo (p), p). Denote
X
S(x, p)= ~ [1'(x', p)-p]dx'+S 0 (p),

xo(P)
where
~~0 (p) =Yo (p)- Xo (p) + (S (Yo (p), p)- p) ~~0 (p). (1.14)
Show that
fJS
ap(x, p)+x=y(x, p). (1.15)
CH. III. ASYMPTOTIC METHIODS 283
From the definition of the function S it follows that :~ =

= 1'(x, p)- p. Then by virtue of (1.13)
ay (x, p)
ax
at (x '
ap p
)= 1 + azsax(x,app) = _!_
ax
(x + ~)
ap '
i e., :~ (x, p) +x-y (x, p) =I (p) is a function of p only. We have
as (x, p)
ap
I ap
X=XJ(P) a--c
= aso (p)-
as (xo (p), p) axo"(p) =
ap
aS 0 (p) (""' ( (P) P) -p ) axo
=---.::,Yo -ap-(p).
ap '
Therefore, from (1.14) we obtain
f (p) =a:;- (S (Yo (p), p)- p) ax~;p) + Xo (p)- Yo (p) = 0,

from which follows (1.15).
Denote by T' an opera tor in the form
i (2 1) v ( (2 1) 1) 1 1
T' = eh" s x, v cp Y x, P ' P '
P= "h
-~
a
ax· (1.16)
·v1 a~ (u (;, ~), ;) I
azs au ( ax ) - 1 Therefo-
From (1.15) it follows that 1+ axap=a;= Ty ·
re, it follows from (*) that
v
(T')*T'=Jcpl 2 x, p +O(h).
( 2 1)
In fact, a more precise theorem exists (in its formulation in the
text below we shall omit the signs' and to avoid any ambiguity). v
Theorem 1.6. Under the above conditions an equality is valid:
T*T=cp (~, ~) cp (~, ;) +h2A(h),

where II A (h) llwN-wN~CN
2 2
for any N';;::::-0, and lm cp = 0.
Before passing over to the proof of this theorem we shall consider
the case when the support ~ belongs to a singular patch.
Case 2. Singular patches. In this case fJB/fJy =1= 0 and, consequent-
ly, the equation
£+p = 8 (y, p)
284 OPERATIONAL l\lE'l'H:ODS
dcf
has a solution y (£, p) E coo. Set x (£, p) = X (y (£, p), p). Just
as in Case 1, we have
1=DI-
Y-Y(~. P)
= D(x(£, p),
D (£, p)
£+P) D('B(y, p)-p, p)
D (y, p)
I
y=y (~. p)
=
OX OX 1 -
- ~ iJp
iJ'B
au
iJ'B -
ap ---ay ar- ap
iJ'B ( ox ox )
.
1 1 0 1
Hence
iJy ox ox
~ <£, p) =~ <£. p)- ap <£, p). (1.17)
Define the function

s
S(£, p)= J x(£', p)d£'+S (p),
~0 (p)
0
where
as~~p) =X (Yo (p), p) [ 8£~~) + 1]- Yo (p) (1.18)
(we assume that the patch contains the point (y 0 (p), p); so (p) =
= 8 (y 0 (p), p) - p). Note that at the intersection of singular and
non-singular patches the function S (£, p) is the Legendre transform
of the function S (x, p) with respect to argument x, i.e.,
S(£, p)=sx(£, p)-S(x(£, p), p).
Show that
as as
-ap-(£, P)+ar(£, P)=Y (£, p). (1.19)
From the definition of the function S we have

as
a[(£, P) =X(£, p).
Hence and from (1.17) it follows that
au azs a2s _ a ( as as ) .
~=- iJpo£ + o£2 -~ ar-iii '
therefore
as as
-ap-+ar-y<£, p)=t·
CH. III. ASY!l\'IPTOTI'C METHODS 285
But at point s=so

(p) we have
as aS0 (p) as
7JjJ (S, p) i;=;o(P) = ---ap-- '8f (So (p), P) asoiJp(P) =
= as~;p) _X (Yo (p), p) as~;P) .

For this reason and by virtue of (1.18) we obtain
a8o oso (p)
f(p)= - 0 P+X(yo(P), P)ii]J+X(yo(P), p)-Yo(P)=O,
from which follows (1.19).
Equality (1.19) may be rewritten in the form
~ (8 (y, p)-p, p)- ~; (8 (y, p)-p, p)=y ( 1.19')
because
£+p==8(y(£, p), p) or
y' =y (8 (y', p)- p, p).
Denote
T' =
1
A
V -2nih J
r e~ [~;-8(6.~)1 ~ L (6, ~)' ~) Id£,
--./ (1.20)
v I ~~ ( s· ~L ~)
y (
where "A is a constant in modulus equal to 1 introduced for the sake

of convenience. Its meaning will be made clear below.
By virtue of (1.19) we have
as = [!.ILJ-1 = ( a2s- _ o2S) -1 = (Jot1
ay - iJs o6 2 iJs iJp 1 •
For this reason from (*) it follows that

~ (2 1)
(T;)*T;=Icpl 2 x, p +O(h).
In fact, there exists a more precise relation; namely, the theorems
that follow are valid (again we shall omit ' and~).
Theorem 1. 7. There exists an equality
T!T 1 =cp(~, ~)cp(~. ;)+h2A(h),

where A (h) is such an operator that II A (h) llw~- wr ~ c N for
any N;;:;::: 0, and Im cp=O.
The proof of this theorem will be cited somewhat later, whereas
now we shall formulate the result related to the canonical trans-
286 OPERATIONAL METHiOIDS
formation of the pseudodifferential operator f (~, ~) whose symbol

belongs to coo (R 2 ) and grows no faster than a certain degree of its
arguments. In the case of a non-singular patch there exists the
relation
For a singular patch there exists an identical relation
In fact, there exists a more precise statement. Namely, the theorem

that follows is valid. Its proof will be cited later.
Theorem 1.8. Let f E Coo (R3 ), where
I (p, x, p') =I (p', x, p) and I f<a> I:::;;; ck (1 + I p' I k + I P lk)
for all a and some k > 0, ck do not depend on x. Let T be an operator

defined by the formula (1.6). Then, if Im cp=O,
T*[f(;, ~. ~)]T=cpU, ;)f(sU, ;);
~[xU, ;)+xU,~)]; s(~.~))cp(~.~)+h2A 2 (h), (1.21)
where A2 (h) is an operator satisfying the following relation:

II A 2 (h) llwN2 ... wN::;;;;cN
2
for all N';;;?O. Let T 1 be an operator defined
by the formula (1.20). Then, if Im = 0, cp
n[f (;, ~. ;) JT1=cp U, ;) f(s U, ;);
~ [x (~, ~)+XU, ;) l s U, ~)) cp (~, ~) +h2 As (h),
where A 3 (h) is an operator satisfying the following relation:
II Aa (h) llwf... w~:::;;cN for all N';;;?O.
The proof of this theorem will be cited at the end of the section.
Note. Below the following construction of the family of curves

B (y, p), X (y, p) will be used. Let x (x 0 , p 0 , t) and p (x 0 , p 0 , t) be
a solution of the Hamiltonian system
• aH • aH
x=Tp, x(O)=x 0 , p=-ax, p(O)=p0 , H(x, p)ECoo.
CH. III. ASYIMPTOTI'C METHIODS 287
Then according to Liouville's theorem (see Ch. IV) we obtain

ax ax
a;; % 0 = 1 at any moment of time t.
ax 0 ap0
The equations
x = X (x 0 , p 0 , t),
p = p (xO' pO' t)
define the family of curves (p 0 being a parameter) satisfying the
above requirements at fixed t.
Proof of Theorem 1.6. According to Theorem 1.4 we have
T*T=cp(y (X:,;),;) cp(y (;:, ;), ;) X

X J,-d!.~.Z)I X
vI ~~ (y (;. ~), ~) ~~ (y (;, ;) . ;) I

X [ 1- \jl 11 ( ~-;) J+ 0 (h 00
), (1.22)
where
~
X=X
~(2 1 3) X, p, p ,
2 1 3) a; ( 2 1 3)
J-1 (x, p, p = Tx x, p, p '
//0 (h 00
) 1/wN-wN~ck, Nhk
2 2
for all k and N':;::.-0.

Recall that ';; (y, p, p') is a solution of the equation
~ + S (_;, p) -S (_;, p')
(1.23)
X p-p' =Y·
Now observe that from (1.15) we have
X (y, p)
as (X (y,
+ ap p), p) = y.
Therefore, it is quite natural to seek the solution of (1.23) in the
form
a; (y, p, p') = x (y, p) + (p' - p) xl (y, p) +
+0 (p'- p)2. (1.24)
Note that by virtue of Lemma 1.2 we may consider all functions

only in a small strip near the diagonal p = p'. Here and below we.
denote the function of the form (p' - p) 2 Cl> (y, p, p') by the symbol
0 (p' - p) 2 , where Cl> E C"".
By putting (1.24) in equation (1.23) and expanding all functions
in p' in the neighborhood of p we shall easily find the coefficient X 1 :
1 azs
X 1 (y, p) = --zlo (y, p) apz (X (y, p), p), (1.25)
where
1o_1 ( y, p ) -_ 1 + axazsap (X ( ) ) _ ay (x,
y, p ' p - ax
p) I
x=X(y 1 p) •
Or, in other words,

/ 0 (y, p) = a"x
uy
(y, p) =I= 0 (in a non-singular patch).
Prove that
1
Xt(y, p)=-z7ij)(y, p),
ax (1.26)
i.e., expansion (1.24) has the form
- 1 ~
x(y, p, p')=X(y, p)+-z(p'-p)Tp(y,p)+O(p~-p) 2 • (1.27)
In fact, from ( 1.15) we obtain
ay (x, p) azs
~(x, p).
ap up~
However,
ay
ap (x, (1.28)
For this reason from (1.25) we have
Xt(y, P)=-2lo(Y, P)
1 [ 1 oX
-Io(Y, p) fJp
J
1 ax
(y, p) =2Tp•
i.e., (1.26) is proved.
By differentiating (1.27) with respect to y we obtain an expansion
of the Jacobian J- 1 (y, p, p') = ~;:
uy
(y, p, p') in the form
1 _1 ( y, p, P') _ oX-au +:r1 ( P, -p) fJ2X

fJyfJp
+ O ( P, -p) 2_
-
1 f)/
=10 (y, p)+-z(p'-p) a: (y, p)+O(p'-p) 2 • (1.29)
Now we shall expand the Jacobians

fJX - fJX - ,
:;- (y (x, p), p) and -0 (y (x, p ), p')
uy Y
CH. III. ASY'MPTOTI'C METHODS 289
in degrees of (p' - p). In an analogous manner after simple trans-

formations and by taking into account (1.26) from (1.27) we obtain
(here =x x (y, p, p'))
ax ~
Ty(Y (x, p), p) = lo (y, P) +
ax
- ( y , p)
+ _i_( ,_) op
2 p p 1 0 (y, p)
olo(Y, p) +O( ,_ )2
up p p ·
For the second Jacobian we have
~~ (y(x, p'),. p')=lo(Y, p)+(p'-p)[aio~~, P) _
_ _i_ 1 o/0 (y, p)oX (y, p)]+O( ' - )2
2/o(y,p) oy up p P •
For these two expressions and from (1.29) we obtain
I J-i (y, p, P') I = 1 + 0 (p'- p)Z. (1.30)
vi ~~ (y (;, p), p) ~~ (y (;, p'), p') 1
Out of all functions in the right-hand side of (1.22) only the func-
tions cp (y (x, p), p) and cp (y (x, p'), p') remain to be expanded.
By virtue of ('1.28) their expansions in degrees of (p' - p) will have
the form
--: _ a<p (p' -p) ax ,
cp (y ( x, p), p) -cp (y, p)+- oy (y, p) 210 (y, p) ap (y, P)+O (p -p)
2
and
cp(y(x, p'), p')=cp(y, p')-
a<p (p'- p) ax
- iJy (y, p) 2Io (y, p) ap (y, p) + 0 (p'- p)2 ..
Note that in the second case we do not expand cp (y, p') in the
argument p'. Besides, here the remainders 0 (p' - p )2 have the
form
(p' _ p)2 (]) (y, p, p'), where <l> E CO'.
From these equalities and from (1.30) we obtain that the symbol
of the operator in the right-hand member of (1.22) is equal to the
expression
cp (x, p) cp (x, p') [1 - 1jl1 (p --p')] +
+ (p' - p)2<J) (x, p, p'), (1.31)
where <D E CO'. (In (1.30) we replaced the argument y in the symbol
' 2
by the operator x; this is why in (1.31) we raplace y by x.)
19-01225
290 OPERATIONAL !METHODS
where
8m (s, p, x, p', S') = (s-s')m ~:~;;)~) cp (x, p) cp (y, p') ECO',
m being any positive integer.
Evidently, we have the following estimate:
JJ ¢6 (~-~)cpU, ~)cpU, ~) llwN-.wN~ 2 2

(1.32)
i.e., this operator is 0 (h"") in our notation.
In the same way,
( 3 1)2 (2 1 3)
p-p x, p, p =h2 Ab where II A1llwN-.wN~cN. (1.33)
2 2
Therefore, the equality of Theorem 1.6 follows from (1.32) and

(1.33). Q.E.D.
Proof of Theorem 1.7. By virtue of Theorem 1.5 we have
T1T1=cp (y (f, ~); ~) cp (y (f+~-~' ;); ;) X

X 'Jit(;, ~.;)I X
VI~~ (y (r. ~); ~) ~~ L (r+~-;; ;) ; ;) I

x[t-¢ 6 (~-;)]+0(h""), (1.34)
where
~ a% ( z
s=s~ ( 2 1 3) ( z 1 3)
x, p, P , J~l x, p, P =a;: x,
1
p,
II 0 (h"") llwN2-> wN~cN,

2
khk for all N, k';;;:;-0.
Here the function f (y, p, p') is a solution of the equation
s{'[+p-p', p')-s{%, p'J S(f, p'J-S[, Pl_y (1. 35 )
p-p' p'-p - .
Taking into account the equality (1.19) it is natural to seek the
solution in the form
f(y, p, p') = B (y, p)- P+ (p'- p) Bt(y, p) +0 (p'- p) 2 •
CH. III. ASYIMPTOTI•C METHIODS 29f
Put this expansion in (1.35). As in Theorem 1.6 we obtain the

coefficient 8 1:
.....,
1::!.1 (y, p) = 1-
/1 (y, p) (a2S
2
a2S ) I
a£2 (£, p)- ap2 (£, p) ~=8(y, p)-p' (1.36)
where
I -1_ (a2S- a2S)
( 1) - a£2 a£ ap
I~=S<y. p)-p--3!...1
as ~=8(y, p)-p •
or
88
I1 (y, p) =ay (y, p).
From (1.19') wo obtain (by differentiating with respect to s
and p and summing up the obtained equalities)
a2s _ a2s -3!...+ ay
as;2 ap2 - as ap •
From the identity £+ p= 8 (y (£, p), p) we also obtain
2 = as ( ay
ay as
_ay )
ap
+ as .
ap
+
By virtue of (1.36) those two equalities produce
1 as
p) =-zap (y, p).
M
~:.t(y,
Thus, tho expansion of function 1 (y, p, p') in degrees of (p'- p))

has the form
f(y, p, p') = 8 (y, P)- P+ ~ (p'- p) ~~ (y, P) +
+O(p'-p) 2. (1.37)
By differentiating this equality with respect to y we obtain an
expansion of the Jacobian
J;l (y. p. p') = ~r (y. p, p') =It (Y. p) +

+ ~ (p'-p)fJ/1~~ P)+O(p'-p)2. (1.38)
Further, in a manner analogous to the proof of Theorem 1.6,

a>< a><
~
+
we expand the J acobians uy (y (£, p); p) and+
uy
~
(y (£ + p- p', p'); p')
as well as the amplitudes cp(y{%, p); p) and cp(y([+p-p', p'); p')
in degrees of (p'- p).
Finally, we obtain the symbol of the operator in the right-hand
member of (1.34) in the form
cp (x, p) cp (x, p') [1-'\J 6 (p-p')] + (p'- p)2<D (x, p, p'),
where <DE CO'.
19*
292 OPERATIONAL !METHODS
By repeating the discourse of Theorem 1.6 we can obtain that

the operator with the above symbol is equal to
(jJ G. ;) (jJ (:, ; ) +h2A2(h),

where II A 2 ilwN-wN~cN for all N';;:::-0, Q.E.D.
2 2
Before passing over to the proof of Theorem 1.8 we shall make
a comment analogous to Note 1.1.
Note. Let f EC"" (R 3 ) he the function defined in Theorem 1.8.
Then for the operator T defined by formula (1.16) we obtain the
result in a manner analogous to the proof of Theorem 1.6:
4 (53 1)2
T*f\p,x,p T=QJ
(2x,p3) f (3~(2 1 3) 1) (2 1)
p;x x,p,p; p QJ x,p +h A 2 2,
where X'· is a :solution of equation (1.23). Since X' (y, p', p) =

=X'(y, p, p') from (1.27) we have
~ 1
x(y, p', p)=z-[X(y, p')+X(y, p)J+O(p'-p)2.
Then in a manner analogous to the proof of Theorem 1.6 we

obtain
4 (5p, x,3 p1) T2= cp (2x, p3) X
T*f
x t (;; {[xU. ;) +X(~. ~) l ;) U. ;) +h A;, QJ 2
where II A; (h) liwN wN~cN for all N';;:::-0.

2- 2
The same equality is obtained from Theorem 1. 7 for the opera-
4 (5 3 1) 2
tor Tff p, x, p T1.
Proof of Theorem 1.8. Denote
ljJ (x, P) def fJl ~~ (x, p), p) •
vl7iii (y (x. p), p) I

By using the commutation formula with an exponent we obtain
5 ( 4 3 2) 1 _ i.s(~. ~) (6 7)
T*f p, x, p T = e h ljJ x, p X
.
X f ([P+ ~~ (x, ~)], ~. [P+ ~~ (x, ;·h ) ljJ (:, ~) e~ s (4 1)
x,p •
Next, let us transform the operators with numbers 3 and 5 by

using
CH. III. A:SYIMPTOTIIQ METHODS 293'
the K-formula and commutation formula, and putting the operator

p =- ih !
in the fust and seventh place, respectively. We shall
get
5 ( 4 3 2) 1
T*j p, x, p T=
i (2 3) ( 3 2 1)
x, P + hN+2R
i (2 1)
- - h s x, p '*'N
-e '"
p, x, p eh"s N· (1.39)
In this equality the function <DN has the form
<DN (p, x, p') = \jJ {x, p) N (x, p') + {- ih) \jJ {x, p) X .
X
at a'ljl
[ ap' 7ii' (x,
,
P)
a2t azs
+ 21 , ,
ap'2 axz (x, P) lf {x, P)
J+
+(~h)
. [ a'ljl at 1 azs azt
Tx (x, P) 7ij) +2 \jJ (x, P) axz (x, p) apz +
J
N+i
+~ h"F,. (p, x, p'), (1.40)
h=2
where we omit the arguments of the function f and its derivatives

(P + ~~ (x, p); x; p' + ~~ (x, p'));
FkECo(R3 ), k=2, ... ,N+1.

The remainder RN in (1.39) has the estimate
II hNRN!I Wz_,.W2
N N-:::;;;JN for any N?;::-0. (1.41)
Since P+ :~ (x, p) = 8 (y (x, p), p), from Theorem 1.6 and by

taking into account formula (1.40) for the term of the zero order
in h in equality (1.39) we obtain
r*t (8 (y U. ; ), ; ), ~. 8 (y U. ; ),~)) r=
=cpU,;) J (8 U, ;) , ~[xU,;) +X U, ;)] ,
(1.42)
where II A; II w2_,.w2
N N~CN.
The term of the first order in h in (1.39) can be transformed

in an analogous manner. By taking into account the symmetry of
the symbol f we reduce it to the form h p- p Q p, x, p , ( 1 3) ( 3 2 1)
where Q ECO' (R 3 ). In a manner similar to the proof of Theorem

1.5 we obtain the estimate
llh (;_ ;) Q(;, ;, ;) "wf--wf=

=hz\1 ( :x- !J Q (;, ~. ;) 1Lf_.wf~h2cN. (1.43)
The remaining terms in (1.39) (of orders h2, h3 , • • • , hN+i) can

also be transformed by virtue of Theorem 1.5 and the following esti-
mate for them can be obtained:
II ... IIWzN ->WzN~h2cN.
Consequently, from (1.39) and (1.42) and from estimates (1.41)
and (1.43) the desired equality for the operatou T follows.
The equality for the case of the operator T1 is established in an
analogous manner. The Theorem 1.8 is proved.
Sec. 2. The Homomorphism of Asymptotic Formulas

Let A 1 and A 2 be generators of degrees s1 and s 2 , respectively,
with the same defining pair of Banach spaces (B, B). The generators
are defined on the same linear manifold D. Denote again the closures
of the operators by A 1 and A 2 •
Theorem 2.1. Let cp E ~ 81 , 82 (R 2) and the function ')l:
'¢ (x1 , x 2 ) = x~cp (x11 x 2),
where k is an integer; k also belongs to ~ 8 ~, 82 (R 2). Then
'I' Lt. A.J = A~cp (Ab .4J.

=
Proof. Let e E C't' (R), e (0) =1= 0. Set cp = cp
- - k-
+ +
(j), ')l = ')l ')l,
where cp (x1 , x 2 ) = cp (x1 , x 2) e (x 2), 'ljJ (x1 ,x 2) = x2 cp(x1 , x 2 ). Let {cpn}
be a sequence of functions converging to cp in ~ 81 • 82 of the form
m(n)
cpn(Xi, X2)= ~ Un.(x1)vn.(x2),
j=i J J
where each function Un.} and Vn.} either belongs to (R) or is theCo
Fourier transform of the 6-function. Identically let {'Pn} be a sequen-
ce of functions converging to ')l in ~ 8 ~, 82 (R 2 ) of the form
l(n)
'Pn (xi> X2) = ~ Wn. (xi) Zn. (x2),
i=i } }
CH. III. ASYIMPTOTI•C METHODS 295
where the functions Wn.] and Zn.] are taken from the same class as
Un 1. and Vn .• Set
] k
(jl1n (xi, X2) = cpn (xl, X2) e (x2), '¢In (xl, X2) = (jl1n (xl, X2) X2,
'¢ 2n (x1, x 2) = '\jJ (x1, x 2) (1 - e (x 2)), cp 2n (x1, x 2) =
= '¢ 2 n (x 1 , X2 ) x2h.
Then the following relations of convergence exist in .96' 81 , 82 (R 2 ):
lim cp 1n = Cp, lim tP1n = 1ji,
n-+oo n-+oo
lim tP2n = ljl,

n-+oo
lim (jlzn =
n-+oo
cp,
Thus,
lim ((Pin+ (jlz,) = (jl, lim N1n + tPzn) = 'iJ, (2.1)
the corresponding sequences converging in the norm of operators.

Let h be an arbitrary element of the space B. From Theorem 4.1
of Chapter I it follows that
[ lP1n Ct, .t) + lP2n Ct, AJ Jh=

Since the operator Afis closed, from (2.2) and (2.3) it follows that
ljJ
Q.E.D.
Ct ..4J h=A~cp Ct ..4J h,
Theorem 2.1 shows in particular (see Note in Sec. 8) that, in the
essence of the fundamental problem, we are looking for the asymp-
totics "in some operators": in the operator A 2 in this case, i.e.,
if we know the behaviour of the symbol of the "remainder" for x 2 -+ oo
then we can arrive at the conclusion that the "remainder" belongs
to the domain of A~.
Strictly speaking, Theorem 2.1 is equivalent to the following
statement. Let cp and ljJ belong to the space .96' s 1, s2 (R 2 ) and cp (x1 , x 2) -
- 'ljJ (x1 , x 2 ) = 0 (x2h) in the sense that the function

(x1 , x 2 )-+ x~ [qJ (x1 , x 2) - '\jJ (x1 , x 2)1
belongs to .91J 81 , 82 (R 2 ). Th-en for any g E B the vector qJ (A A

1, 2) g --
- 'ljJ (A A_J
1, g belongs to the domain of A~.
As an example consider the method of a stationary phase in the
case when the phase linearly depends on a parameter. Let qJ E (R) Co
and !F be an infinitely differentiable in R real function. Let the
equation !F' (£) = p have a unique solution £ = 8 (p) on the S>lp-
port of the function qJ for any p E R, where !T" (8 (p)) =1= 0 for
any point p E R. Consider the integral
()()
I (x, p) =X (p) J qJ (£) eix[Y(~)-p,~] d£, (2.4)

-oo
where x E (R). Co
In the integral (2.4) proceed to a new variable of integration t
according to the formula
x [ ff (£} - p ·£ - F (8 (p)) + p8 (p)) =
= I xI t2 sgn [x !T" (8 (p))l.
Denote as in Sec. 5 of Chapter I
I x I = w, sgn x !T " (8 (p)) = a.
We obtain
()()
I (x, p) =X (p) eix[Y(S(p))-p(S(p))] ,} eiwatz1Jlo (t, p) dt, (2.5)

-00
where '\jl 0 (t, p) = qJ (£ (t, p)) 0 ~ ~; p) • The function '\jl 0 is infinitely

differentiable in all arguments; moreover, there are such positive
continuous functions c (p) and d (p) that ~Jo (t, p) = 0 for t ~
~ [-c (p), d (p)]. Construct a sequence of functions '\jlj (t, p) accord-
ing to the following recurrent formula:
'"· (t )=_!__ tlJi(t, p)-tpj(O, p)
'~'Hi ' P at t ·
It is obvious that for t ~ [ - c (p), d (p)]

1Jli(t, p)=t-2Pi(t-1, p),
where Pi (-c, p) is a polynomial in the variable -c with infmitely
differentiable coefficients depending on p. In Sec. 5 of Chapler I
CH. III. ASYIMPTOTIC METHIODS 297
it is shown that the following expansion exists:

Vx e-ix[SC<E<Pl-P<E<Plll I (x, p) =
=
~
- ;~ sgnSC"(E(p)) LJ
V :n;e' [ i
2wa
Jj 'Pi (O , P)X(P)+rnH(x,
-
p),
i=O
(2.6)
where
rn+i (x, p) = Vx [ ~~ r+l X (p) \

00
-00
ei 0l<JI 2 ¢n+1 (t, p) dt, (2.7)
the function Vx
being defined in Sec. 5 of Chapter I. In the expan-
sion (2.6) replace Vx
by rad x and ffi-i = x-i (-1)i sgn x by
Pi (x) (-1)i sgn x, where the function rad and Pi are also defined
in Sec. 5 of Chapter I. Here the remainder rn+l (x, p) is replaced
by the function rn+l (x, p) which coincides with rn+l (x, p) for
I x I> ex, where ex is an integer independent of p. Note that in (2.6)
sgn .¥" (8 (p)) det a 0 = const because the function .¥" (8 (p))
does not become zero anywhere. For x =1= 0 all terms of the expan-
sion (2.6) are obviously infinitely differentiable in x and p with
the exception of-rn+I (x, p). This means that the function -rn+I (x, p)
also possesses this property. Let e (x) be a function in (R) which Co
is identically equal to 1 for I x I < ex. Set
rn+l (x, p) e (x) = ~L (x, p), rn+l (x, p) [1 - e (x)] = v (x, p).
We have
rn+l (x, p) = ft (x, p) + v (x, p).
The function ft belongs to Co (R 2) c .%'51 , 82 (R 2). This is why the

function
(x, p) -+ xn+I~L (x, p)
also belongs to Co (R 2) c .%' 81 , s2 (R 2). Show that the function

(x, p)-+ xn+Iv (x, p)
also belongs to .%' 8 1 , 82 (R 2) (this also means that v E .%' 81 , 82 (R 2)).

For the sake of certainty set sgn .¥" (8 (p)) = 1. Just as in Sec. 5
of Chapter I for any integer m ?: n +1 we obtain:

- .2::_ i n+i
xn+ 1v(x, p)=Vne 4 ( 2 ) 'i'n+L(O, p)X(p)(1-e(x))+
+
Vn e-
in m
4 ~ ( +r
xn+!-i (1-e (x)) 'i'i (0, p) X(p)
i=n+2
+
+Vxxn-m[1-e(x)] ( ~ r+i X
00
X j eixt 2'\Jm+i (t, p) X (p) dt. (2.8)

-00
The first term in the right-hand member of (2.8) obviously belongs

m
to pg 81 , 82 (R 2). The same applies to each term of the sum . Lj . Consi-
J=n+2
der the latter term in (2.8). Set
J eixt2'i'm+dt, p)X(p)dt=
00
f(x, p) def Vxxn-m[1-e(x)l
= Vx xn-m [1- e (x)Jx (p) gm (x, p),

where
00
gm(X, p)= j eixt2'\JmH(t, p)dt.
Just as in Sec. 5 of Chapter I we establish that for x =I= 0 the

function gm is infinitely differentiable and
akl+k2 gm (x,p ) =0(ixJ-h1),
ax" 1 op"2
where 0 is an estimate, uniform in p for p E supp X· Consequently,
for m ?: n +
2 and for any non-negative integers k1 and k 2
jI
iJkl+k2 f (x, p)
h h dxdp<oo
I
OX I iJp 2
R2
so that f EW21+ 1 ' 82 + 1 (R2) c: pg 81 , 82 (R 2). Thus, we have shown that

the functions r n+i (x, p) and xn+ir n+i (x, p) belong to the space
pg .~, 82 (R 2). Denote
K (x, p) = rad xe-ix[g-(E(p))-pE(p)][ (x, p).
From the expansion
in: n
K(x, p)=Vne4ao ~ ( i~o )iPi(x)'\Ji(O, p)X(p)+rnH(x, p)
i=O
(2.9)
CH. III. ASYlMPTOTl'C METHJOD,S 299
it follows that K c: ~ s1, s2 (R2). By using Theorem 2.1 we obtain

the following result: for any hE B the vector
K Lt. JtJ h = Vn e i~<Jo ~

i=O
c~o r Pi (At) Xi (Az) h, (2.10)
where Xi (p) = 'tjli (0, p) x (p) belongs to the domain of the opera-
tor A1+ 1 •
Theorem 2.2. Let A 1 (A 2) be generators of degree s1 (s 2) with the
same defining pair (B, B) and with the same domain D A 1 = D A 2 = D.
Let the functions cp and 'tjJ E ~ s1, s2 (R 2) be correlated by the formula
'tjl (x1 , x 2) = P (x1 ) cp (x1 , x 2),
where P is a polynomial. Then
'I' CL. lJ = cp (Jtb AJ p (At)·

Proof. As in the proof of Theorem 2.1, we construct the sequen-
ces {{jltn}, {(jlzn}, Ntn} and {'1' 2 n} of functions of the form
l(n)
.2} uni (xt) Vni (x 2 ) possessing the following properties:
i=1
lim (cptn -t- (jlzn) = (jl, lim Ntn
n_,.oo
+ 'l'zn) = 'IJ,
lJltn(Xt, Xz)=P(xt)cptn(Xt. Xz), lJlzn(Xt, Xz)=P(xt)cpzn(Xt, Xz).
Let hE D. Then from Theorem 4.1 of Chapter I it follows that
['l'tn CL, AJ +'l'zn (At, AJ Jh=

= [ {jltn CL, AJ + {jlzn LL Az) JP (At) h.
Passing over to the limit for n -+ oo we obtain
'I' (Jtb .4J h = cp Lt. .4J p (At) h.

Q.E.D.
Theorem 2.2 brings out the following statement. Let cp and ljJ
belong to the space ~ s 1, s2 (R2) and cp (x 11 x 2) -ljl (x 11 x 2 ) = 0 (xjk)
in the sense that the function
(x 11 x 2) -+x~[cp(x 11 x2)-ljl(x 11 x 2)]
belongs to ~ s 1, s2 (R2). Then the operator
[cp Ct..4J -ljl (Jtb .4J JA7

is bounded.
We can now write an analogue of the expansion (2.10) of the

stationary phase method in a case when the operator A 1 acts first
and A 2 second. Specifically, the operator
[K LL, lJ- Vn i:ao ±(i~o )i Xi (A2) Pi (A1) JAf+

e 1 (2.11)
i=O
is bounded.
Apply the obtained results of the stationary phase method to the
pseudodifferential operators. The expansion (2.10) has the following
form:
K (- i d~ ' n h (x) = Vn e in~o ±(i~o

i=O
)j X
X Pi ( - i :x) Xi (x) h (x) + qn+i (x),
where qn+1 E w~+n+t (R) if hEW~ (R).

The expansion (2.11) for the pseudodifferential operators has the
form
K (- i d~ ~) h (x) =
,
=
i=O
r
Vn e i~.ao ~ c~o xi (x) Pi (- i :X) h (x) +
+rn+i ( - i d~, ~) h(x),

where the pseudodifferential operator r 11 +1 ( - i :x , ~) possesses
such a property that for any k the operator r n+i ( - i :x , ~) X
X (- i ! r+ 1is bounded in W~ (R). Consequently, for any hE

E w~-n-l (R) the function
r n+1 ( - i d~ , ~) h (x) =
=rn-i-1 ( - i :X, ~) (- i d: +ir+l ( - i d: +irn-l h(x)
belongs to W~ (R).
CH. III. ASY'MPTOTIG METHODS 301
Sec. 3. The Geometrical Interpretation of the Method of Stationary

Phase
Consider the integral
Jr
00
rad (J) 07:

I (p, cu) = 1 eiw(-px+o-(x))cp1 (x) dx, (3.1)
(2ni) 2 -oo
1
where p, cu are real variables, ( 2 = eirt/ 4 and q:>1 E (R); rad cu Co
is an infinitely differentiable function defined for sufficiently large
I cu I by the formula
1 in
ra d cu = cu
2 =def e-~,-<1-sgn w)l cu 121 •
Note that for sufficiently large I cuI the function i 1 / 2I (p, cu) is !-
Fourier transform of the function eiw.~(x)q:> 1 (x) depending on the
parameter cu.
Apply the method of stationary phase to the integral I (p, cu)
using only the first term of the asymptotic expansion. Supposing
that .FF" (x) =!= 0 on the support of the function q:> 1 (x) and that the
equation -p +
ff' (x) = 0 has a unique solution x = x (p) for
x E supp q:>1 , we obtain
00
-i!i-(sgnfil"(x(p))-1) sgn w rad w \ qz;
e q • eiW(-px+v(X))cpt(X)dX=
(2ni) 112 -
-00
cpt(x (p)) . eiw[ -px(p)+ff(x(p))J + g (p cu)

1-~" (x (p)) 11/2 ' '
where the function g (p, cu) possesses the following property: for
any X E C0 (R) and any s 17 s2
X (P) e-iw[ -px(p)+ff(x(p))]g (p, cu) E!B 81, 82 (R2),
ffi)( (p) e-iw[ -px(p)+fil(x(p))J g (p, cu) E !B 81, 82 (R2).
We shall write further ft(p, cu) ~ / 2 (p, cu) if ft- / 2 E !B 8 1, 82 (R 2)
and cu [ft(p, cu)- fz(p, cu)] E ~ 81, 82 (R2 ). Thus
e-i : [sgnff" (x(p))-1Jsgn wX (p) X
00
X rad w ~ eiw[ -px+ff(x)+px(p)-ff(x(p))J X

(2ni) 112
-00
X (x) dx c.:: -v (p) cp 1 (x (p)) (3.2)

Cjlt -"' Iff" (x (p)) 1112
In the space of points (x, p) which is usually called the phase

space in physics the equation
!F' (x) = p (3.3)
describes some line. Introduce the parameter a on the curve (3.3):
x = x (a), p = p (a).
The condition.¥" (x) =I= 0 means that equation p = p (a) with respect
to a has a unique solution a = a (p), where the function a (p)
is infinitely differentiable. In this case the curve (3.3) is said to be
diffeomorphically projected onto the axis p. Thus any function
on the curve (3.3) may be regarded as a function of x and p or a.
Allowing for inaccuracy in the notation we shall sometimes write
an equality of the type
cp (x) = cp (p) = cp (a),
as well as the expressions a = a (x) for the solution of the equation
x = x (a), and a = a (p) for the solution of the equation p = p (a)
(of course, this applies to cases where there could be no misunder-
standing). We have
x a
!F(x)=!F(x(a0 ))+ ~ ff'(x)dx=ff(x(a 0 ))+ Jpdx, (3.4)
x(ao) ao
a
-px(p)+ff(x(p))= -px (p)+ ~ pdx+ff (x(a0))=
ao
a
=- ~ xdp-p(a0 )x(a0 )+ff(x(a0 )),
ao
a a
where Jx dp and Jp dx are curvilinear integrals along the cur-
ao ao
a
ve (3.3). The integral Jp dx along the curve in the phase space
ao
will be called the action (as accepted in physics).
By analogy with (3.2), set
cpt(x) = I~ 11/2
rp (a) I
da a=a(x)
By virtue of (3.4) and of equalities
!F"(x)= ~~' !F" :~ = ::

CH. III. ASYIMPTOTl'C METHODS 303
we obtain
r e-i(J) I/ eiffi f
00
p dx ) dx,..._.
1 X (p) rad <p (a)
I:~T/2
(J) px ao
w+i V2ni }oo . \ a=a(x)

a
-1--z-o-
. tt sgn ffi ( iffi[- S x dp-p(ao)x(ao)] <p ( a ) )
,..._. e e ao I:: 11/2 a=a(p)

X
1
X X (p) w+i ' (3.5)
where
G = 21 ( 1 - sgn "'dx""
dp ) .
Sec. 4. The Canonical Operator on an Unclosed Curve

Let M 1 : x = x (a), p = p (a) be an unclosed infinitely smooth
curve without self-intersection with coordinates (x, p) in the phase
space and let cp (a) be an infinitely differentiable function with
a compact support. Define the real functions on M 1
a a
S(a)= ~ pdx, S(a)= Jxdp+x(a )p(a )= 0 0
= -S(a)+p(a)x(a).
Let an open arc U of the curve M 1 containing the support of the
function cp be diffeomorphically projected onto the axes x and p,
i.e., :: =1= 0, ~~ =1= 0 on U. Then, instead of a, we may choose
either x or p as a parameter on U:
a = a (x), a = a (p).
The relations are valid:
as as ~
P=ax, x="(fj)', S+S=px,
i.e., the transition from x and S (x) to p and S (p) is a Legendre

transformation.
Exchanging x and p in (3.5) and passing over to the complex-
adjoint values we obtain, for any infinitely differentiable function
p (w) which is equal to zero in a neighborhood of zero and is equal
to unity in a neighborhood of infinity:
e-iffiS(cx(x))X (x) p (ro) vr ~:ni X

co 1
X l eiffipxe-iWS(a)~~~- 2 cp(a)l dp"'

J da cx=a(p)
1
"'X (x) p (ro) (i sgn ro)cr j ddx

a
,- 2 <:p (a) I
cx=a(x)
under the condition that the support of the function X is contained

in the projection of U onto the axis x. But if the support of an infi-
nitely differentiable function X (x) does not intersect with the sup-
poft of the function <:p (a (x)) and x (x) = 1 in a neighborhood
of infinity, then the function
g(x, ro)=x(x)p(ro) , / w
2m. X
V -
co 1
X
Jr I
ei(J) px- iWs(cx) .:!.E._ ,- 2 <:p (a)
da
I CX=CX(p)
dp
belongs to the space W~ (R 2 ) for any s. The function rohg (x, ro),
where k is an arbitrary integer, has the same property. Consequently,
the equality is valid:
X (x) p (ro)
co
y -~:ni X
1
X r
J
eiffi(px-Sicx)) j.!:!!...j-
da
2 <:p (ex) I CX=CX(p)
dp =
,-2
1
=X(x)p(ro) (isgnro)creiffiS(cx)l ~: cp(a)la=a(x) +

+ eiffiS(a(x))g1 (x, ro) +g 2 (x, ro), (4.1)
where g1 and g 2 are infinitely differentiable functions g1 (x, ro) = 0
for x E supp <:p (ex (x)), the functions rog1 (x, ro) and rog 2 (x, ro)
belonging_ to the space !J35 81 , 82 (R 2) for any non-negative integers s1
and s 2 •
In the right-hand member of (4.1) the function
1
eiffiS(a) I:~ ,- 2
is infinitely differentiable in U, but outside the arc U it may no

dx
longer be smooth in points, where dP = 0. These points will be
CH. III. ASYIMPTOTI-C METHiOD.S 305
called focal points. Besides, the left-hand member of (4.1) is pro-

portional to the ffi-Fourier transform of the function
1
e-iwS(a) I!:E_~-T.(jl (a) I '

da a=a(p)
1
where the factor e-iWs(a) 1 ~~ r-z

is an infinitely differen-
tiable function in the neighborhood of a focal point. This conside-
ration suggests the idea of defining the following operator K which
is a canonical operator translating an infinitely differentiable finite
function in M 1 into a function of variables ffi and x.
Let the family { U i} iEI of open arcs cover M 1:
M1= u uj
iEI
(in M 1 the ends are not included), every arc Ui intersecting only
a finite number of arcs Ui. Next, let on every arc Ui
dx dp
aa-=FO or -aa=FO.
Choose on every arc Ui either x or p as a parameter (as a local coor-

dinate). If x is chosen as a local coodrinate on Uh then we call
Ui a non-singular patch of the curve M 1 ; otherwise, Ui will be called
a singular patch. The function a= ai (x) or a = ai (p) corresponds
to every patch Ui. The set of singular or non-singular patches on
Jl.[l will be called the canonical atlas of the curve M 1 •
Let the function fJJ (a) have a compact support supp fJJ E M 1 .
Consider the partition of unity corresponding to the canonical
atlas, i.e., infinitely differentiable functions {ei (a)} which satisfy
the condition
1= 2Jj ei (a) for a E supp qJ, supp ei cUi.
Then
fJJ (a)= 2j ei (a) (jJ (a)= 2j fJJi (a), supp fJJi cUi.
For the sake of certainty let Ui 1 be a non-singular patch which

contains a point a 0 • For the function 'ljJ (a) with the support in Ui 1
we set
1
[K· 'ljJ] (x) = eiwS(a) ~~ ,-2 'ljJ (a) I . (4.2)

11 da a=alxl
Let Ui 2 be a singular patch crossing with Ui 1 • Define a canonical

operator Ki 2 acting on the function with support in Ui 2 by means
2•}-0 1225
306 OPERATIONAL 'METHODS
of the formula
[K12'1Jll (x) =c-./ 2 .
w X
V - m
I_!_E._ ,- I
00 1
Jr
X eioopx-iooSca.) 2 'ljl (a) dp,
da a.=a.1• (p)
-00 2
where the constant c will be defined below. Require that for any
infinitely differentiable finite function '\jJ (a) with the support con-
tained in U; 1 n
U12 the following relationship should be valid:
e-iooS(al(x))X (x) p {ffi) [Ki!'IJl- K 12'1jl] (x, ffi)-"./ 0
regardless of the function x E Co

(R) with the support contained
in the projection of Ui! n
Uh onto the axis X and regardless of the
function p E Coo (R), which is equal to zero in a neighborhood of
zero and equal to 1 in a neighborhood of infinity.
By virtue of (4.1) we have
e-iooS(ai!(x))X (x) p (ffi) [K;2'1jl] (x, ffi)-"./
~ c2 (i sgn ffitui!nu;2e-iooS(a1(x))X (x) p {ffi) [Ki!'IJll (x, ffi),

where
aulnu2= 21 (1-sgn?)
X
IaEUjlnuh
.
Consequently, we may set
c = (i sgn ffi) -cru;lnu;2.
Identically for non-singular patch Uia which intersects with Uh
we set
[K; 3'1jl] (x) = (i sgn ffi) -cru;lnu;2+cru;2nuia x
1
X eiooS(a.) I~ ~-2 'ljl (a)/ ;

da a.=a;a<x>
here supp '\jJ E U a·
In general let U1 , U2 , • • • , Un be a finite sequence of patches
of a given canonical atlas, where singular and non-singular patches
of the sequence follow in succession and two adjacent patches of
the sequence intersect.
Then set
-cru. nu· +cru. nu· - •.. +cru; nu;
[K; 'ljl] (x) = (i sgn ffi) H 32 32 33 n-1 n X
n
1
X I:: ~- 2 \jJ (a) la=a.j

n
(x)
CH. III. ASYIMPTOTrC METHODS 307
for odd n and

-au. nu· +... -au. nu·
Kj 11 ljl (x) = (i sgn ro) 11 12 1 n-1 1n x
00 1
X-./ w_ leiw(px-Sca))ldp,-2"ljl(a)l dp.
V -2m J
-oo
da a=aj
n
(p)
For even n define the index and the canonical operator. The
number
will be called the index of the chain of patches Uiu . . . , Ui n . For

the given definition of the canonical operator Ki in a patch to be
uncontradictory it is necessary and suffiCient that the index
ind (j1 , • • . , jn) should depend only on h and jn or identically that
the index of the closed chain of patches should be equal to zero.
This statement holds for an unclosed curve M 1 • In fact, let U be
an arc of the curve M 1 which is diffeomorphically projected onto
the axis p and a' and a" be two points on the arc U which are non-
focal (i.e., ~~ =1= 0 at these points) . Set
'\' (a', a")= ~_[ sgn :: (a')- sgn dx (a") J.
If the points a', a" are on the arc which is diffeomorphically pro-
jected on the axis x, then we set
v(a',a")=O.
It is easy to see that both definitions yield the same result in cases
when they are applied simultaneously. Now let l be a path on the
curve M 1 with ends~',~", i.e., a continuous mapping
t -+ a (t)
of the segment [0, 1] in M 1 satisfying the conditions
a (0) = W, a (1) = ~".
Let the points ~' and W' be non-focal. We set

n-1
'\' [l] = 2J
k=1
'\'(a (tk), a (tk+i)), (4.3)
where W= a (0), ~" = a(1) and 0 = t 0 < t1 < ... <tn-t<

< tn =1, where the numbers t 1 , • • • , tn _1 are chosen in such a way
that on the right-hand side of (4.3) every segment of the path with
ends a (tk), a (tk+ 1) either has no focal points or entirely lies on the
20*
308 OPERATIONAL 'METHODS
arc diffeomorphically projected onto the axis p (this choice is always

possible). Call y [l] the index of the path l. This term is justified
because a (t) does not depend on the choice of intermediate points
a (t1), . • • , a (tn_1 ) on the path l. In order to verify this statement
it is sufficient to note that y [l] does not change if the point t' is
placed between th. and tn+l' In this approach a summand is to be
added to y [l]:
y (a (tn), a (t')) + y (a (t'), a (tn+ 1 )) - y (a (th.), a (tn+ 1 )).
The summand is always zero as follows directly from the definition.
Let two paths l: a = f (t), l': a = g (t) be homotopic on Ml,
i.e., there exists a continuous function .!F(t, s), 0:::;;; t:::;;; 1, 0:::;;;
:::;;; s :::;;; 1 with such values in M 1 that .!F(t, 0) = f (t) and .!F(t, 1) =
= g (t). If for any s E [0, 1] the points .¥(0, s) and .¥(1, s) are
not focal then the indices of the two paths l and l' coincide. The
proof of this statement is left to the reader. Now, it is not difficult
to verify that the index of the closed chain of patches on M 1 is
equal to zero. In fact for any chain of patches Uiu ... , Ui n the
number ind (j1 , • • • , jn) is equal to the index of some path l with
the initial point in ui1 and the final point at the intersection uin-1 n
n Ui n . Let jn = j 1 . Then we may assume that the initial and final
points of the path l coincide, say with a point a 0 , i.e., the path
l is closed. But any closed path of the unclosed curve may be conti-
nuously transformed into the path l': a = const (into the point).
It is obvious that y [l'] = 0. Consequently,
y [ l] = ind (j11 • • • , h) = 0.
The concept of the path index on M 1 becomes especially clear
if in every compact subset M 1 there exists only a finite number of
focal points and at every point the derivative ~; changes its sign
(if Lhese conditions are fulfilled, then the curve M 1 is said to be in
the general position with respect to the projection onto the axis x).
Then we can consider the index y (a) of the point a as an integer-
valued function on M 1 in the following way: y (a 0 ) = 0; by the
transition through a focal point in the direction of decreasing dx/dp
the index y (a) increases stepwise by one; by the transition in the
direction of increasing dx/dp the index y (a) decreases by one; at
focal points the function y (a) is not defined and at other points
it is continuous (hence locally constant). We shall stress that the
initial point a 0 should be non-focal. For any path l on M' which
begins at the point a 0 and ends at the point a we have
y [ l] = y (a). (4.4)
In the case when the curve M-1 is not in a general position we shall
consider (4.4) as the definition of the function y (a). Finally, define
CH. III. ASYIMPTOTI•C METHlODS 309
the canonical operator by formula

1
-2
[K<p] (x, w) = ~ (i sgn w)"~'i eiwS(o:) j :~ I X
iEit
X ei (a) <p (a) lo:=o:.(x)+
]
00
+ ~ (i sgn w)"~'r( ~ni ~ eiw px-iwSco:) X

~~ -00
1
X I ~Pa ~- 2 ei (a) <p (a) Io:=o:lp)

dp,
where I 1 is a set of numbers of non-singular patches and I 2 is a set

of numbers of singular patches and
Yi = Y (a), a E Ui
if Ui is a non-singular patch. If Ui is a singular patch, then y (a)
may accept different values at different points of the arc Ui. Let
Yi = min'\' (a).
o:EUJ
An operator f!Jt will be called a smooth canonical operator if it satis-

fies the following conditions:
(a) [6.if<p] (x, w) = [K<p] (x, w) if I w I is larger than some constant
independent of <p;
(b) [f!/t<p] (x, w) E Coo (R 2 ).
Let A 1 and A 2 be generators acting on a Banach space B. Since
[Q~<p] (x, w) is a smooth function of x and w we may consider ope-
rators [f!Jt<p]
notation:
CL, 1J and [f!Jt<p] (..,t, .4J . Introduce the following
1 2
Definition. We shall say that the pair of generators A 1 and A 2
( 2 1 )
A 1 and acting on B is in agreement with the curve M 1 in the
A2
phase plane (x, p) if for any non-singular patch U of the curve M 1
and for any infinitely differentiable function X (x) with the support
contained in the projection of the arc U onto the axis x the operator
X (J.J ei1z'l' Ct) (respectively, X (J.J eitq> (! 1) ) , where <p (x) =
== S (a (x)) is bonded in B.
Theorem. Let f!Jt and f!lt' be two canonical operators on Jlf1 corres-
ponding to different canonical atlases {U i LEI and { Uj LEI' to parti-
tions of unity {ei} and {ej} and to different techniques of "smoothing"
in w, but to the same initial point a 0 •
Then
1 2
(a) if the pair A 11 A 2 is in agreement with M\ then the operator
AJf!/f: Ct, A.J cp(a)-f!lt' CL A.J cp(a)]

is bounded in B;
2 1
(b) if the pair Ail A 2 is in agreement with M 1 , then the operator
[f!/f: (.41, .4J cp (a) -f!lt' Ct, AJ cp (a)] A2

is bounded in B.
Proof. Consider the sum {Vi} iEI" of the canonical atlases {U i} iEI
and {ViLer', i.e., the set I" is the sum (disjunctive) of the sets I
and I', Vi = u, for j E I and v, = Uj for j E I'. Continue the par-
titions of unity {ei} and {ej} in the following way:
e,(a)=O forjEJ',ej(a)=O forjEI.
Here I and I' are regarded as unintersecting subsets of the set r.
Then f!Jt and f!lt' may be regarded as two canonical operators cor-
responding to one and the same canonical atlas {Vi}, but to dif-
ferent partitions of unity {e,} and {ej}. For this reason and without
any loss of generality we may consider that from the start the cano-
nical atlases { Ui} and { Uj} have been identical.
For sufficiently large modulus w we have
[(&%- f!lt') cp] (x, co)=
t
= ~ (i sgn w)'~'JeiwS(a) I:~ ~-2 [ei (a)- ej(a)] X

iEr 1
X cp (a) la=at<x> + ~ (i sgn w)'~'FV -~:rri

iEh
f
-oo
eiw px-iwBca) X
X I ~Pa ,-
2
[ei (a)- ei(a)] cp (a) Ia=at<p) dp.
Without any loss of generality one may suppose that any two
different non-singular patches of the atlas {U i} do not intersect
and any two different singular patches of this atlas do not intersect
either. In fact, let Uio be a non-singular (singular, respectively)
CH. III. ASYIMPTOTJ'C :.viETHIOD.S 311
patch of a given atlas and let I 0 be a set of the numbers of all non-
singular (singular) patches intersecting with Uio· Then the arc U Ui
iElo
with x (or p) as a local coordinate may be chosen as a patch of some
canonical atlas in M 1 . In general, call two elements j' and j" of the
set ·I equivalent if the patches Ur and Ur are simultaneously
non-singular or singular and intersecting.
Let 1 be the quotient set of the set I according to a given equi-
valency relation. For every equivalence class ] E] we set
U-:= U Ui, e..,.= 2]ei>
3 3
iEI iEI
where we choose x (or p) as a local coordinate on U; if Ui is a non-
singular (singular) patch for j E ]. Then { u1}j E 1 is the canonical
atlas on M 1 and {e7 }je! is partition of unity corresponding to this
atlas. Here any two different non-singular (or singular) patches of
the atlas {u1 } do not intersect and the canonical operator corres-
ponding to the atlas {Ui} and to partition of unity {ei} obviously
coincides with the canonical operator corresponding to the atlas
{u1} and to partition of unity {e1}.
Thus, suppose that the atlas {Ui} coincides with the atlas {U1 }.
Then [ei (a)- ej (a)] cp (a) = 0, if the point a is not included
in any of the non-singular (singular) patches of the atlas, because
in this case ei (a) cp (a) = ej (a) cp (a) = cp (a). Let Ui be a singu-
lar patch of the atlas {Ui} and let U ku • • • , U km be non-singular
patches intersecting with Ui. Then the support of the function
m
[ei (a)- ej (a)l cp (a) is contained in U (U1 n Uk ), where at
1111= 1
any point a E Uk l.t n U1 the following equalities are valid:
ei (a) cp (a) + ek (a) cp (a) = 1,
l.t
ej (a) cp (a) + e~ 11 (a) cp (a) = 1.
Hence at the intersection U1 n Uk the following equality is
11
valid:
cp (a) [cp 1 (a) - ej (a) 11
+
ek (a)- ek (a)] = 0.
l.t
Here we note that since the functions ek and ek are equal to zero
. 11 11
in a neighborhood of each end of the arc it follows that the support
of the function [e1 (a) - ej (a)] cp (a) is contained in some open
arc which is diffeomorphically projected onto the axis x. By apply-
ing (4.1) we obtain that the difference [(K - K') cp] (x, w) is the
sum of a finite number of terms of the form
eiw5(a.i(x))gt (x, w) + g2 (x, w),
where g1 and g 2 are infinitely differentiable functions g1 (x, ro) = 0

outside some closed set which is contained in the projection of the
non-singular patch ljj onto the axis X and the functions rog1 (x, ro)
and rog 2 (x, ro) belong to !!& 81 , 82 (R 2 ) for any non-negative integers
( 1 2)
s1 ands 2 .TheoperatorsA 2g1 A 17 A 2 ,g1 At.A 2 A 2 ,A 2g 2 A 17 A 2
(2 1) (1 2')
and g 2 CL, .AJ
A 2 are bounded in B. The proof of the theorem
now follows directly from Theorem 9.2 of Chapter II.
Sec. 5. The Method of Stationary Phase

Let h E R, y E Rn, <D be an infinitely differentiable real function
and cp be a complex function in Co
(Rn+ 2) whose support is contai-
ned in (a, b) X Y X (-c, c), where' a < b, c > 0 are real numbers
and Y is a ball in Rn. Consider the integral
I (
y, h) = rJ eh<l>(p,
00 i
y, h)
cp(p, y, h)dp. (5.1)
-oo
Suppose that the equation

8<1>
ap-(P, y, h)=O, PE(a, b), yEY, !hl<c
has a unique solution p= p (y, h), where
82<1>
8p2 (p (y, h), y, h) =I= 0.
We shall now make in (5.1) a change of variables

<D (p, y, h)- <D (p (y, h), y, h)= t 2a, (5.2)
82<1>
where a= sgn 8P2 (p (y, h), y, h). We obtain
i i
r
00
I <y, h) = eh<'P(p(y, h), y, h)

j e~t2 '"
'I'
<t, y, h) d t, (5.3)
-00
where p (t, y, h) is the solution of the equation (5.2),
ljl (t, y, h)= cp (p (t, y, h), y, h) 8P (t~t h) •
The function ljl belongs to Co

(Rn+ 2 ) with support contained
in (d, e) X Y X ( -c, c), d, e being real numbers. By applying to (5.3)
the results of Sec. 5 of Chapter I we obtain the following expan-
CH. III. ASYMPTOTI-C METHODS 313
sion
(5.4)
i=O
where
.:rt N
1 4 cr sgn h ( iah) +1
(y, h )=
-t
rN~t e - 2- X
· Vlhl
(5.5)
-oo
'l't(t, y, h)='l'(t, y, h),

8 'i'i (t, y, h)-"ljJj (0, y, h)
'l'rrt (t, y, h)= Tt t • (5.6)
Note that for sufficiently large in modulus t every function

'l'i(t, y, h) has the form*Py,h (-}),where Py,h is a polynomial
whose coefficients are infinitely differentiable functions of y and h
supported in Y X (-c, c). Besides, the functions
Xi (y, h) det 'Pi (Ol y, h)
belong to C0 (Rn+l) and are supported in Y X ( -c, c).
Estimate the remainder. Set h = -(i)1- and consider the function
RN+t (y, ro) def rN+t (y,-&--).
This function is non-zero only for
I ro I> 1/c. Show that the function roN+lRN+t (y, ro) belongs to
!JB 81 , 82 (Rn X R) for any s1 , s 2 • In fact, for any integer m > N + 1
we have
i;
roN+1RN+i (y, ro) = Vn ( )N+1 XN+i (y, ~) +
i=N+2
+V -I-,
ro e-i ~.. cr sgn ro ( -ia )N+1 ro N -m X
2
00
X ) eirocrt 2 'l'm+t ( t, y, -&-) dt. (5.7)

-oo
The first term in the right-hand side of (5. 7) belongs to

g; s~o a2 (Rn X R)
since
AN+1 ( y, --&-) = XN+1 (y, 0) +--&-X (y, -&-)'

where x
(y, h) is an infinitely differentiable function. Hence the
function !
X( y, !)
belongs to any Sobolev space. Next all terms
m
in the·sum 2j in the right-hand side of (5.7) are square-integrable
i=N+2
and all their derivatives possess the same property. Finally, consider
the function
00
X j eiwat~¢m+1 ( t, y, --&-) dt def VIWi X

-oo
_,_a
.n Egn w
X e 4 wN-mg(y, w).
The function g (y, w) is non-zero only for y E Y and is bounded.
We shall estimate its derivatives. For any multi-index k =
= (ku ... , kn) and any number l we have
l
iPl+ 1g (y, w)
~ Ci. ( - 1 )j X
iJyk iJwl 2co
i=O
! r(0~ r-i (:y r ~·m+i

00
X J
-00
eiwat 2 ( 1 +t ( t, y, -&-) dt. (5.8)
From (5.8) it follows that the function g (y, w) is infinitely diffe-

rentiable and all its derivatives are bounded. Consequently, for
a sufficiently large m the function f (y, w) belongs to any Sobolev
space given beforehand. This proves that the remainder r N+l in the
expansion (5.4) satisfies the condition
wN+irN+i ( y, ! }E.91J s~, s2 (Rn X R).
In conclusion, we shall consider the case when the critical point

of the phase <D (p, y, h) is absent. Consider the integral (5.1) and
let the functions <D and <p satisfy all the above conditions excluding
the solvability of the equation a<D = 0.
op
CH. III. ASYIMPTOTIC METHJODS 315
Instead of this, we suppose that

iJ<D
ap (p, y, h) =I= 0
for p E (a, b), y EY and I h I <c. So we can make the following

change of variables in the integral (5.1):
«t> (p, y, h) = t, p = p (t, y, h).
We obtain
I(y,h)= re~t'P(t,y,h)dt,
-oo
where
ljJ (t, y, h)= rv (p (t, y, h), y, h) ap (t'at h) •
As before, the function 'ljJ belongs to CO' (Rn+ 2 ), where its support
is contained in (d, e) X Y X ( -c, c), d, e being integers. By integ-
rating by parts N times we obtain
oo it N
I (y, h)= (iht I eTa 'P (t,J' h) dt. (5.9)
J
-oo
at
Seth= !.
From (5.9) it follows that if in (5.1) ~~ =1=0 on the
support of the functions rp, then for any integer m the function
wml (y, 1/w) belongs to any Sobolev space which means that it
also belongs to any space !JJ 81 , 82 (Rn X R).
Sec. 6. The Canonical Operator on the Unclosed Curve Depending on

Parameters Defined Correct to 0 ( !)
1. Consider the family of unclosed curves M 1 (~, w) in the phase
plane (x, p). Let ~ E Rn, w belong to an extended numerical axis
and the equations of curves M 1 (~, w) are
x = x (a, ~. w), (6.1)
p = p (a, ~. w), (6.2)
where a is the (same) parameter on the curves of the family in ques-
tion and x (a, ~. w) and p (a, ~. w) are infinitely differentiable
functions (also for w = oo).
Introduce the following functions S (a, ~. w) and S (a, ~. w):
dS = p (a, ~. w) dx (a, ~. w),
S (0, ~. w) = 0,
S (a, ~. w) = -S (a, ~. w) + x (a, ~. w) p (a, ~. w). (6.3)
Let {U (~, ffi) heB, ooEQ be a family of the arcs of the curves M 1 (~, ffi)
corresponding to the interval (a, b) containing the parameter a,
and let Q be an open domain without zero on an extended numerical
axis and B be a domain in Rn. Suppose that every arc U (~, ffi)
is diffeomorphically projected onto the axis x and also onto the
axis p so that the equalities (6.1) and (6.2) have on (a, b) X B X Q
a unique solution with respect to a.
a =a (x, ~. ffi), a = a (p, ~. ffi).~
Let <p (a, ~. ffi) be an infinitely differentiable function with
a compact support which is contained in (a, b) X B X Q. Consider
the integral
1
I (x, ~' ffi) =}I _~ni f

-oo
eioop · x-iooS(a, ~. oo) I !~ lT X
x rp (a, 3, ffi) 1a=a(p,

=
~. oo)
dp.
For the calculation of this integral we can apply the method of

stationary phase. If X (x) is an arbitrary function in CO' (R) then
X (x) I (x, ~. ffi) =
1
. ax
=X (x) (i sgn ffi)cr etooS(a, B, oo) 7ki I ~-2 +
+ eiooS(a(x, B. oo), ~. oo)g1 (x, ~' ffi) + g2 (x, ~' ffi).
Here the functions g 1 (x, ~. ffi) and g 2 (x, ~. ffi) are non-zero only
for ffi E Q, the functions ffig1 (x, ~. ffi) and ffig 2 (x, ~. ffi) belonging
to the space .$ s~, 82 , sa (R X Rn X R) for any sb s 2 , sa;
G =1
-
2
( 1-sgn--
iJpfiJa)
iJxfiJa
Ia=a(x, ~. w)
.
Now let X (x) be an infinitely differentiable function equal to

zero on the combined projection of the arcs U (~, ffi) onto the axis x
and equal to 1 in a neighborhood of infinity. Then we can use the
results of the preceding section which deal with the absence of
critical points of the phase if we set
We obtain that for any N and any Su s 2 and sa

ffiNX (x) I (x, ~. ffi) E .$s1, s2, sa (R X Rn X R).
CH. III. ASY'MPTOTfC METHJOD.S 317
Putting unity in the form 1 = X (x) +

(1 - X (x)), where X (x) E
ECO' (R), we finally obtain the following relation:
I ,-2 x
1
I (x, p, w) = (i sgn w)G eiroS(a, 13, ro) ax

7ki
x cp (a, p, w) la=a(x, (3, ro) + eiroS(a(x, (3, ro), (3, ro>g 1 (x, p, w) +
+ g 2 (x, p, w), (6.4)
where the functions g1 and g 2 have the same properties as above.
2. The definition of the index. The formula (6.4) which is a gene-
ralization of (4.1) permits us to define the canonical operator in the
family of curves M 1 (p, w). Let C be a compact set in the space of
parameters a, p and w containing no points at which w = 0. Next,
let {U 1 }, j = 1, ... , N be a family of open connected sets in the
parameter space which covers C and let {e 1 }, j = 1, ... , N be
partition of unity corresponding to the covering {U i}
N
e1 (a, p, w) EC0; ~ ei (a, p, w) = 1
.1=1
for (a, p, w) E C.
Obviously, the covering {U1 } of the compact set C may be chosen
so that for any j one of the following conditions is fulfilled:
(a) ;~(a, p, w)=FO for (a, p, w)EUi,
(b) ;~ (a, p, w) =I= 0 for (a, P. w) E Ui·

Divide the set of numbers 1, 2, ... , N into two classes-I1 and
I 2 -in such a way that for j E I 1 the condition (a) will hold and
for j E I 2 the condition (b) will hold. The family {U 1 } with the
pair {I u I 2 } will be called a canonical atlas of a neighborhood
of the set C in the space of parameters. The set U1 for i-E I 1 will
be called a non-singular patch of this atlas and for j E I 2 -a sin-
gular patch. The function ai (x, p, w), which is the solution of the
following equation with respect to a
x = x(a, p, w),
(a, p, w) E Uh
corresponds to every non-singular patch Ui and the function
ai' (p, p, w), which is the solution of the following equation
p = p (a, p, w),
(a, p,
w) E Ui'
corresponds to a singular patch ui'.
318 OPERATIONAL :\iETHODS
The canonical atlas {U 1 } and {/1 , / 2} of a neighborhood of the

compact set C may always be chosen in the space of parameters
in such a way that for any hE / 1 and j 2 E / 2 the intersection Ui, n
n Ui. is connected. We shall suppose that this condition is fulfilled.
The number
will be called an index of intersection of the non-singular patch U 1,

and singular patch uj which have a non-empty intersection.
Set by definition 2
If both patches ui, and ui. are non-singular, or singular, then set
au.11 nu·12 = 0.
Let Ui, ... , Uim be a finite sequence of patches of the cano-
nical atla~ such that any two successive patches of this sequence
intersect. Such a finite sequence will be called a chain of patches.
The number
ind U11 ... , jm)=au.1! nu1·2 + ... +au.'m-1 nu·1m
will be called an index of this chain of patches.

By means of the index of the chain of patches we introduce the
index of an arbitrary path l in the space of parameters (a, ~. w)
which begins and ends at points belonging to some non-singular
patches. Let U 1 , •.. , Uim be such a chain of patches begun and
ended by non-siAgular patches that the path l is equivalent to the
sum of the segments
l = l1 + ... + lm
satisfying the condition z,L c Uj!J>· Then the number
y [l] = ind (j 1 , .•• , jm)
will be called an index of the path l. This defmition of the index

of the path l is correct, i.e., it is independent of the choice of the
chain .of patches Uiu ... , Ui m and of the partition of the path l
into segments l10 • • • , lm.
In order to verify this statement, first suppose that by the same
partition of the path l into segments we use another chain of patches
u11.. , ... , U.,1m for the definition of the index of the path.
CH. III. ASYI1\IIPTOTJ<C METHiODS 319
Then
ind (jH ... , jm) = ind (j;, ... , j:,). (6.5)
In fact, let M~ be the end of a segment lw Set
if :~ (M~) =/= 0, ;~ (M~·) =/= 0 and O"~ ctef 0 in the opposite case.
Regard the path l as a road which in every part l 11 has a highway
(a non-singular patch) or a canal (a singular patch) (or the former
and the latter). At the points there exist stations where one can
change from a car to a motor-ship (if the highway and the canal
intersect at this point). The passenger begins and ends his trip in
a car. At every station he has the right to change the type of trans-
port. Leaving the car he receives the sum a~ for the loss of the speed
and, leaving the motor-ship, he pays the sum a~. The sum does not
change r-egardless of the part of the road where the highway is along-
side the canal. It is clear that the sum total for the whole trip does
not depend on the choice of the scheme for changing transports
and this means that (6.5) is valid. Since one may go over singular
or non-singular patches several times it should be noted that the
index of the road does not change by partition of the segment l"
into parts. ·
From the definition of the index of the path it follows that y [ l)
does not change by continuous deformation of l if the ends of the
path remain unmoved. Since, in our case, the parameter a changes
in some part of the real line it follows that any closed path l may
be transformed into a point by continuous deformation in the space
of parameters a, ~. (J). This means that y [ l] = 0 for any closed
path l.
Hence it follows that the index of the path depends only
on its initial and final points.
Therefore, the index of the chain of patches beginning and ending
with non-singular patches depends only on the first and last patches
on the chain. .
This fact is identically proved for an arbitrary chain of patches.
3. The definition of the canonical operator. Now we can define
the canonical operator K on the family of curves {M1 (~, ffi)}.
Fix a point P 0 in the space of parameters and let this point belong
to some non-singular patch Ujo· For any function qJ (a, ~. ffi) sup-
ported in the set C by definition we set

[K<p] (x, ~' ro) = ~ (i sgn w)vieiooS(rx, 13. oo) x
iElt
I ox ,-2
1
ei(a., ~' ro)<p(a., ~, w)lrx=rxlx,J3,oo>+
v -~l"ti r
X oa
+~ (i sgn ro) Vj eioo(px-Scrx, 13, oo)) X

iEk -oo
1
xj ;~ ~- 2e1 (a, ~. ro)<p(a, ~. ro)lrx=rxJ(P.J3,oo)dp, (6.6)

where Yi is the index of the chain of patches starting with the patch
U 10 and ending with the patch U 1•
The canonical operator K defined by (6.6) depends on the choice
of the canonical atlas { U1 } and on the partition of unity {e1 } (as
well as on the choice of the initial point P 0 and on the arbitrary
choice of the definition of the action S associated with the possibi-
lity of the reparametrization of the family of curves).
Let Alt ... , An+z be generators acting on a Banach space B.
( k 2 k-1 k+1 n+2 1 }
[K<p] Ah A 2 , ••• , Ak-h Ak, ... , An+t• An+z .
Definition. We shall say that the ordered set of generators A
k 2 k-1 k+1 n+2 1
= (Ah Az, ... , Ak-h Ak, ... , A n+h An+2) acting on B is in
agreement with the family of curves M 1 (~, ro) in the phase plane
(x, p) if for any non-singular patch U in the space of parameters
a, ~. ro (for family of curves M 1 (~, w)) and for any infinitely dif-
ferentiable function x (a, ~. w) supported in U, the operator <1> (A)
corresponding to the symbol
<l>(x, ~' ro)=x(a(x, ~, ro), ~' ro)eiooS(rx(x,J3,oo),j3,oo)
is bounded in B.
Theorem 6.1. Let K and K' be two canonical operators on the
family {M1 (~, w)} corresponding to different canonical atlases
{Ui}iEI• IH I 2 and {Uj} 1 er•, I~, I; (of the neighborhood of compact
s;et C in the space of parameters) and to the partitions of unity
{ e1 } i E r and {ej}1 EI' respectively. Next, let the set of generators
( k 2 k-1 k+1 n+2 1 )
A= A1, Az, . · ·, Ak-h Ak, · · ·, An+u An+z
acting on a Banach spaceB be such that the operators A 11 • • • , Ak-t
commute with each other and with the operator An+z and the
CH. III. ASYIMPTOTIC METHODS 321
operators Ak, ... , An+l commute with each other and with the
operator A 1 ; and let this set be in agreement with the family
{M1 (~, w) }. Then for any infinitely differentiable function cp (a,~. w)
with support in C the operator
{[Kfp] (A) - [K'cp] (A)} An+2
is bounded in B.
The proof of this theorem is based on (6.4) and is analogous to
the proof of the theorem of Sec. 4.
Note. Below we shall also need the dependence on the translator.
This case is analogous to the preceding case. For the sake of simpli-
city let n = 1. In Theorem 6.1 one can replace the vector-operator A
by the vector-operator
(A1 A2 3, , 2,
3
T, At
4)
where A 2 commutes with A 3 and T: B-+ B is a translator, or by
the vector-operator
(
1 2 3 I,)
A 3 , T, A 2 , At ,
where A 2 commute with A 1 •
Sec. 7. V -Objects on the Curve

1. The definition of the spaces LA and L'£. Let M 1 be an infi-
nitely smooth curve on the plane R 2 and let the Cartesian coordi-
nates x and y be fixed on this plane. If an open arc U of the curve
JV/1 is diffeomorphically projected onto the axis x, then one can
choose x as a parameter on this arc. If such a choice is made, then U
will be called a non-singular patch of the curve M 1 . In the same
way, the arc of the curve 111' 1 on which y is chosen as a parameter
will be called a singular patch of the curve M 1 •
A set of all singular and non-singular patches of this curve will
be called a complete canonical atlas .Jf oo of the curve M 1 • Any
finite family {UiL=J, ····N of patches of the complete canonical
atlas dl oo covering M 1 will be called a finite canonical atlas of the
curve M 1 • We shall suppose that the finite atlases of the curve JY/1
exist (for example, M 1 is a compact curve).
Let .!1 H be a set of a non-singular patches of the atlas .:11 oo and
.J/: 0 be a set of singular patches of this atlas. Every patch U E .:1. H
is defined by the equation
Y = Yu (x)
and every patch U E .:!1 0 is defined by the equation
X= Xu(y).
21-01225
It is obvious that in the case when the two families of functions

{Yu}u E AH• {xu}uE Ao
are given they are equivalent to the curve M 1 •

For any of the two intersecting patches tJ E AH and U' E J! 0
there exists a linear operator V u, u' acting on infmitely differen-
tiable functions of y with support in the projection onto the axis y
of the intersection U n U' and translating them into infinitely
differentiable functions of x with support in the projection onto
the axis X of the intersection u n
U'. It is required that the follow-
ing two axioms be fulfilled:
V1 (the locality of the operators V u, U')· The value of the func-
tion V u. u'Cj) at the point x depends only on the germ of the func-
tion cp at the point Yu(x) (i.e. only on the restriction of the func-
tion cp by any small neighborhood of the point Yu(x)).
V 2 (the conditions of concordance)
[Vu 1 , u{cp] (xo) = [Vu 2 , u2cpl (x 0 )
for any function cp with support in the projection onto the axis y
of the intersection u; n lj~ and for any point Xo of the projection
onto the axis X of the intersection ljl n U 2·
Let .f!: = {Ui} be the finite atlas of the curve M\ / 1 be the set
of numbers of non-singular patches and / 2 be the set of numbers
of singular patches of this atlas. Consider the triple
(7.1)
where cpi (x) is a function in CO' (R) with support in the projection
uj onto the axis X and {jJj (y) is a function in c~ (R) with support
in the projection Ui onto the axis y. The set of all possible elements cp
of the form (7.1) becomes a vector space for a fixed .!6 if linear ope-
rations will be introduced by the formulas
A(j) = (.Jl:, {A,cpj (x)}JE11• {A,cpj (Y)}JEI2),
(.Jl:, {cpi (x)}, {cpi (y)}) + (.Jl, {~'i (x)}, Ni (y)}) =
= (.If:, { cpi (x) + lJldx)}, {{jJj (y) +'Pi (y)}).
The obtained vector space will be denoted by L .:~t· Denote by L~
the join of all possible sets Ldl.
2. The definition of V-objects. Let
cp=(.Jl:, {cpi(x)}JEii• {cpj(y)}JEh), Jl={Ui},
ljJ =(A', {'Pi (x)}JEii• Ni (y)};o;), .A'= {Uj}
CH. III. ASY':Y!PTO"TIC li'IETH.ODS 323
be two elements of L"2.. Consider the following four functions on M 1 :

c:pH (x, y) = 2J
jEI1
cpj (x), cp 0 (x, y) = 2J
jEI2
{jlj (y),
(x, Y)EUj (x, Y)EUj
"¢H (x, y) = 2J \jlj (x), "IJlo(x, Y)= 2J \jlj(y).

jEJ{ jE12
(x, v;Euj (x, Y)EUj
We shall say that the elements c:p and '\jJ are equivalent and we shall
write c:p = '\jJ if such elements satisfy the following conditions:
(a) there exists a finite set~ of the pairs (U, U'), where U E .Jl;H•
U' E .JI 0 such that for two different pairs ( U1 , u;), (U 2 , u;) in ~
the intersection ul n n n
u; u2 u~ will be empty and
,.
(jlu-lPn= z.;
(U, U')E~
CXu,U', cro-1Jlo=
~l
LJ
(U, U')EL\
Bu,u-,
where au, u' and Pu, u' are infinitely differentiable functions with
support in U n U';
(b) for any pair (U, U') E ~ the following equality is valid:
+
cxu, U' (x, Yu (x)) Vu, U' rBu, U' (xU' (y), y)] (x) = 0.
Denote by :t the factor-set Lr. according to the equivalence rela-
tion ==·
Introduce the structure of a vector space on :£. First of all, we
note that if cp, '\jJ E Ly, then there exist c:p', 'ljl' E L"2. such that c:p = c:p'
and '\jJ == 'ljl' and the atlases corresponding to the elements c:p' and 'ljl'
coincide. The elements cp' and '\jl' can be added. Define the sum of
the classes of the equivalency {c:p} and {'\jl} by the formula
{cp} + {~1} = {c:p' +1Jl'}. (7.2)
It is easy to see that the sum is well-defined by (7.2). The product
of the class {c:p} by a scalar will be defined in the ordinary way:
A {<p} def {"Ac:p }.
The introduced vector space:£ will be called the space of V-ob-
jects on the curve llf 1 •
Example. The functions on the curve as V-objects.
Consider a V-object on M 1 which corresponds to the following
family of the "transition operators" V u, u-: if U is a non-singular
patch of the atlas .:!too and U' is a singular patch intersecting with.
it, then
Vu, u-cp (y) = c:p (Yu (x)),
(7.3)
VU', ucp (x) = (jl (xu' (y)).
Let c:p be a V-object which corresponds to the family of transition

operators (7 .3) and constitutes the class of the equivalency contai-
21*
324 OPERATIONAL :\1ETHODS
ning the element

(.ff, {<pi (x)}, {<pi (y)})
of the set L"E.. Bring the object <p in correspondence with the function
(ji on M 1 defined by the formula
cp(xo, Yo)=~ <JJi(xo)+ ~ <JJi (Yo), (7.4)
iEK1 iEK2
where K 1 is the set of numbers of non-singular patches and K 2 is

the set of numbers of singular patches (of the atlas .II) containing
the point (x 0 , y 0 ) E M5. From (7 .3) it follows that the definition
of the function does not depend <m the choice of the representative
V -object <p in Ll:..
Conversely, let ~ be an infinitely differentiable function with
a compact support in M 1 and let .J! be a finite atlas in M 1 • Consider
some partition of unity {ei} corresponding to the atlas .:11:
N
eiECoo(M), suppeic:.Uh ~ ei=1;
j=1
then every function (iii = eicp is supported in Ui. If Ui is a non-

singular patch of the atlas .JI; then set
<JJi (x) = ~i (x, Yu.J (x)), (7.5)
and if Ui is a singular patch then
<JJi (y) = <Pi (xui (y), y). (7.6)
The formulas (7 .5), (7 .6) bring the function ~ in correspondence
with an element (.:fl, {<p 1 (x)}, {<JJi (y)}) of the set L"E. such that (7.4)
is valid. This establishes a one-to-one correspondence between the
space :£ of V-objects corresponding to transition operators (7.4)
and the space C"((' (M 1). Below we shall identify the V-object <p and
the correspemding function cp.
3. Linear operators on the space of V -objects. We shall now
describe the method whereby linear operators on the space :£ of
V-objects will be given. Since:£ is a factor-set it is natural to first
-set some operator A 0 on the set L"E. and then, as usual, to bring
an operator A on :£ in correspondence with A 0 according to the
formula
A {/} = {Ao/}. (7.7)
Of course, for (7. 7) actually to define the operator A on :£ it is neces-
sary that the operator A 0 should be in agreement with the relation
,of the equivalency ==
(f =g)-+ (A of= Aog).
CH. III. ASY'iVIPTCYI'IC METHODS 325
To set a linear operator on :£ in the usual way one needs to set

a linear operator A 0 on LJ:.. However LJ:. does not possess a structure
of the vector space. For this reason the operator A 0 acting on LJ:.
will be called a linear one if for every fmite atlas A of a curve M 1
the restriction of the operator A 0 on LA is a linear operator acting
on LA:· A structure of the vector space on :£ is introduced in such
a way that the above-mentioned linear operator A 0 on LJ:. corres-
ponds to the linear operator A on :£.
The above method of setting the linear operators on :£ is not yet
versatile enough. One may now consider the operators A 0 defined
on a part of LJ:.. It suffices to suppose that the operator A 0 is defined
on the join of a family {LA-} of linear subspaces such that A 0 LAc:
c: L dt' LA-c: L dt' where for any fPC::£ there exists a representative
in one of the subspaces LA-.
Example. A differential operator on the curve.
Consider again the space c:(M 1) as a space of V-objects on M 1 .
Fix an arbitrary finite atlas A on M 1 and the partition of unity
{ei} corresponding to this atlas. This partition will be called a weigh-
ted partition of unity.
In the space L dE consider a linear subspace LA- consisting of ele-
ments of the following form:
(A, {ei (x, Yu1 (x)) IJl (x, Yui (x))}, {eJ (xui (y), y) 'P (xui (y), y)}),
where 'P is an arbitrary function in C;;"(M1). It is obvious that for
any function 'P E c:
(M') there exists its representative in LA.
Let / 1 be the set of numbers of non-singular patches and I 2 be the
set of numbers of singular patches of the atlas .:fl. A differential
operator L acting on C;'(M1) will be defined by means of the sets
of operator {Li}iEr 1 and {LihEI2 with differentiable coefficients
according to the formulas
(.!1:, {l\li (x)}, {l\li (y)}) = L (A, {'Pi (x)}, {'Pi (y)}), (7.8)
'i'i (x) = Li'Pi (x),
(7.9)
'i'i (y) = Lj'Pi (y),
In order to define pseudodifferential operators on M\ we intro-

duce Sobolev spaces W~ (M1). Let again the finite atlas .ll and the
weighted partition of unity {ei} be fixed. For any function 'P E
E c: (M1) we set
II 'P llwk(Ml) =
2
tJ
1EI1 2
tJ II 'Pi (y) llwk(R)'
II 'Pi (x) llwk(R) + 1El2 2
(7 .1 0)
where rp1 (x) and rp1 (y) are the components of a representative of
the function cp in the subspace L'.-:f!.
Problem. Prove that in the case when the curve M 1 is compact
the norms of the form (7 .10) corresponding to the different finite
canonical atlases and to the different weighted partitions of unity
are equivalent.
Example. Pseudodifferential operators on a curve.
Define the space W~ (M 1) as the completion of the vector space
C;;'(M1) in norm (7 .10). The elements of spaces W~ (M 1) will be
called generalized functions in M 1 .
Let q> E C';(M1 ), (.:11:, { rp 1 (x) }, {cp 1 (y) }) be the representative of
(j) in L'.-:f!. Consider the two families of pseudodifferential operators
{Il1 (2'
··x-l-
. 1 )}
d {H 1 (2y,
· -l-
. 1 ) }jEI2
d
(7 .11)
dx iEI!' dy
whose symbols H 1 satisfy the condition

Il j (~, YJ)
+
(YJ i)m X mE!J65 s. n(R 2) (7.12)
for any x E C;;' (R). The operators of this family translate. the space S
in itself. Suppose that the operators II1 (;, -dL!ax) and
H1 (~, -idldy) do not extend the support of the function on which
they act (e.g. differential operators, the operator with the symbol
H (£, 'Y]) = e'll does not fit in). Bring the families of functions
{ ff>i (x) }iEI 1 and {cp i (y) }iu 2 in correspondence with the two
new families {'ljli (x) }i E 11 and {'IJli (y) }iu 2 according Lo the formulas
~J(X)=Hj (~, -i }x) cpj(X),

~i(Y)=lli (~, -i :Y) cp (y). 1
Then
(.:11:, {'Pi (x) }iEJI, {1PJ (y) }jEiz)
belongs to the space L.:!f· Thus some operator If: C'(;' (M 1)--+ C'(;' (M1 )
is defined. From the conditions (7 .12) it follows that H is bounded
as an operator acting on W~+m (M 1 ) to W~ (M 1 ), where IZ I :::;;;; s.
It is natural to consider the extension of the operator If to the homo-
morphism W~+m (M 1 ) --+ W~ (M 1 ) as a pseudodifferential operator
.corresponding to the family of symbols H 1 (~, 11).
C:H. III. ASYC\IPTOTIC METHODS 327
Sec. 8. The Canonical Operator on the Family of Unclosed Curves

In this section the class of canonical operators will be constructed
correct to 0 (1/wN) where N is any number.
1. r -objects for the canonical operator. Let the equations
x = x (a, B, w), p = p (a, B, w)
define the family {M 1 (B, w)} of infinitely smooth curves on the
phase plane x, p which satisfy the conditions mentioned in Sec. 6.
Let C be a compact set in the space of parameters a, B, w which
does not contain points where w = 0. In Sec. 6 we introduced the
concept of the canonical atlas of a neighborhood of the set C in the
space of parameters. The set of all singular and non-singular patches
of all canonical atlases corresponding to the totality of compact
sets C will be called a complete canonical atlas of the space of para-
meters.
Proceed now to the defmition of objects to which the canonical
operator 2ir 1 will be applied. These will be some V-objects in the space
of parameters. We allow some deviation from the general definitions
given in the preceding section. This corresponds to our wish to avoid
cumbersome details in the definitions of Sec. 7 in order to get them
across more clearly. Now describe the mentioned deviations. In the
defmitions of the previous section the complete canonical atlas
of the curve is to be replaced by the complete canonical atlas of the
space of parameters and the fmite canonical atlas of the curve by the
canonical aLias of a neighborhood of the compact set C in the space
of parameters. The elements of the space L df: will have function
components
(jlj (x, B, w), j E I1, (8.1)
cp j (p, B, w) , j E I 2,
where I 1 is the set of numbers of non-singular patches and I 2 is the
set of numbers of singular patches of the atlas .Jf. Here we shall
identify those functions (8.1) which differ only by 0 (w- 1- 1), i.e., who-
se difference is the product of w- 1- 1 by an infinitely differentiable
function; for a patch containing the point w = oo we shall identify
any two functions (8.1) if the paLch does not contain points where
w = oo. In other words, we make a corresponding factorization
in every space L df:· By virtue of the preliminary factorization the
conditions defining the relation of equivalency = are to be changed
in the obvious vvay.
Define the transition operators V u, U' which translate the func-
tions of x, B and w into the functions of p, B and w if U' is a non-
singular patch and U is singular, and the functions of p, B and w
into the functions of x, ~ and w if U' is singular and U is non-singu-

lar. Let U be a non-singular patch and U' be a singular patch, both
belonging to the atlas .JI ao· The patch U corresponds to the function
au (x, ~. w) which satisfies the equation x = x (au (x, ~. w), ~. w)
and the patch U' corresponds to the function au• (p, ~. w) which
satisfies the equation p (au• (p, ~. w), ~. w) = p. Let cp (a, ~. w)
be an infinitely differentiable function with support contained
in U' U,n .
(ji (p, ~. w) = cp (au (p, ~. w), ~. w).
Consider the integral
P. w) = .. /
I ( X'p, V (!)
-2m
. r
J
-oo
eiOO.J·x-iwS(CG, 13, oo) X
X I ~~ r 2
1
cp (a, ~. w) la=CGU'(P, 13, 00) dp,
where the action S (a, ~. w) is defined by (6.3). On calculating

the integral I (x, ~. w) by the method of stationary phase (for x =
= X (a, ~. w), (a, ~. w) E u U') we obtain n
l
I (x, ~' w) = (i sgn w)a ~ { ( ~a; ) j eiooS(au(x, 13, oo), 13, oo) X
i=O
1
X I.!!_ ,-
Or:J.
2
CG=CGu(X, l3, 00)
'X,j (x, ~. w)} +r + 1 1 (x, ~. w) X
X iooS(au(x, 13, oo), 13, oo)

e '
where
0'-
- 2
i.(t- •sgn (~)/(~))I
oa oa (CG, 13, oo)EUnu•'
and the function w1+1 rl+ 1 (x, ~. w) belongs to any of the spaces
!J3I s 1 , ••• , s n+r (Rn+ 2 ). Concerning the functions 'X,j we may say that
by virtue of the method of stationary phase the functions 'X,j (x, ~, w)
for a fixed function S depends only on values of the function cp
and its derivatives in a up to order l inclusive at the point (cxu (x,
~. w), ~. w).
CH. III. ASYMPTOTIC .METHIOD.S 329'
We define the operator V u, u' in such a way that the following

equality is valid:
I (x, ~, ffi) = eiwS<au(x, B, w), 13. w) { (i sgn ffi)cr X
1
xj.!:.._j- 2
iJr:x a=au(x, B. w)
lVu,u,qi](x, ~. ffi)+r!+i(x, ~. (!))}.
In other words, we set
[Vu, u•(ji] (x, ~. ffi) = ~

i=O
( ~~ r Xi (x, ~. ffi). (8.2)
2. The construction of the canonical element. Consider the spa-

ce X of V-objects in the space of parameters a, ~, ffi corresponding
to the family of transition operators V u, u' defined in the present
section. The vectors of the space :£will be called canonical elements.
We shall describe a standard method for constructing a canonical
element. Let cp (a, ~, ffi) be an infinitely differentiable function
supported in a compact set C, let .Jl = (Ui)i E I be the canonical
atlas of a neighborhood of the set C and {ei} be a partition of unity
called a weighted one corresponding to the atlas .Jl:
ei Ecoo, supp ei cUi,
L}ei(a, ~. ffi}=1, (a,~. ffi}EC.
The function cp, the atlas .Jl and the weighted partition of unity
{ei} are brought in correspondence with the canonical element by
means of the following procedure: let .Jl' = { Uj }i E I' be an arbitrary
canonical atlas of a neighborhood of the set C and let {ej} be the
partition of unity corresponding to the atlas .Jt'. We set
CVi (x, ~. ffi) = ej (a, ~, ffi} ea (a, ~. ffi} X
X cp (a, ~, (!)) la=au'.(x, B, w), iEIJ.
J
(8.3)
CVi(P, ~. ffi)=ej(a, ~. ffi)e 0 (a, ~. ffi)X
X cp (a, ~, W) la=au .(P, B. w), iEI2,
}
where
eH=L}ei, e0 =L}ei>
jEI1 iE/2
I 1 is the set of numbers of non-singular patches of the atlas .Jl, I 2 is
the set of numbers of singular patches of the atlas .Jl and (J~) I;
is the set of numbers of non-singular (singular) patches of the atlas
.Jt'. Thus for every canonical atlas .Jl' of a neighborhood C we define
the element
of the space L df'. It is left to the reader to verify that for any cano-
nical atlases .fl:' and .Jt" there exists an equivalency qJ df' qJ df" ==
so that all elements of the form qJdf' define one and ihe same class
{ qJ·"*'} E :£ .
Let k 1 , . . . , kn+ 2 be such non-negative integers that l lei- ... >
... + kn+!· Intoduce in the space of infmitely differentiable
functions of x, p and (!) the following relation of equivalency ~=
cp (x, p, ffi) ~ 1p (x, p, ffi) if and only if
(1)1-k!- ... -hn+![([l(X, B, ffi)-tjJ(x, p, ffi)]E.%'"1' ... ,hn+2(R"+2).
The quotient space corresponding to the relation of equivalency ~

will be denoted by ShJ, · · ·• ""+2• 1• Allowing for the common inac-
curacy in the notation we shall deal with the elements of the space
Sk1, · · ·• hn+2· l as with functions of x, p and ffi, keeping in mind
their actual representatives in C""(R"+2). In the same way we shall
often use notation and expressions as if the components cpi (x) and
qJj (p) of an element of Lr, were functions and not classes of equiva-
lent functions.
3. The definition of the canonical operator. We may now define
the canonical operator K 1 on the family of the curves M 1 (p, ffi)
with a fixed parametrization and a f1xed "initial" point P 0 in the
space of parameters, the operator acting on Lr. to S"!· · · ·· ""+2• 1.
Let .Jf be the canonical atlas of a neighborhood of a compact set C
in the space of parameters, / 1 be the set of numbers of non-singular
patches of the atlas .JI; and I 2 be the set o£ numbers of singular patches
of this atlas. Let P 0 belong to a non-singular patch Uio E .Jf; and y
be the index of the chain of patches of the atlas J6 which starts with
the patch Uio and finishes with the paLch Ui.
Definition. For any element
qJ = (.Jl, {qlj (x, p, ffi) }H Iu { qlj (p, p, ffi) }H rz)
of the space L Jl we set
[KzqJ] (x, p, ffi) = 2J (i sgn ffi}"i eiwS(a, (3, w) x
iEI1
1
XI!!!_
iJr:z
-'2~ a=au .(X, (3, W) CfJj(x)+ .~ (isgnffi)ViX
1 1El2
X v /"--
_w__
-2m
r eiw:(p~·)-S{a,
00
J
(3, W)) X
X I ~~ r 2
la=au. (p, (3,w) qlj (p) dp.
J
(8.4)
CH. III. ASYI.MPTOTIG METHJODS 331
Problem. Show that the definition is correct in a sense that if

cp' = (.I!, {cpj (x, ~. ro) }, {cpj (p, ~. ro) })
and
cp" = (Ji, {cpj (x, ~. ro) }, {cpj (p, ~. ro) })
are two representatives of one and the same element cp E L d!: then
Kzcp' = Kzcp".
Theorem 8.1. Let cp, '\jJ E L'£ and cp = 'tjl. Then
Kzcp = Kz'\jl.
Proof. First we note that wiLhouL loss of generality one may
eonsider that cp and '\jJ belong to one and the same space L df:· Let
cp = (.Jl, {cpj (x, ~. ro)hEIJ• {cpi (p, ~. ro)hu2),
'\jJ = (.ft, Ni (x, ~. ro)}jEip Ni (p, ~. ro)}iEI2).
Then
[K 1cp- K 11pl (x, ~' ro) = [Kz (cp- 1jJ)] (x, ~. ro),
where cp - '\jJ = 0. Hence it is necessary to show
(cp = 0) ==? (K 1cp = 0).
Let
According to the defmition of the relation of equivalency= consider

the following two functions
rpu (a, ~. ro) = ~ cpj (x (a, ~. ro), ~. ro),
jEll
cro (a, ~. ro) = ~ Cj)j (p (a, ~. ro), ~. ro).
iEI2
Then there exists a fmite set !1 of the pairs (U, U'), where U is
a non-singular patch and U' is a singular one, both belonging to the
atlas .JI oo such that the following conditions are fulfilled:
(a) for any two different elements (U1 , u;) and (U 2 , U~) of the
set /). the intersection ul n n n
u; u2 u~ is empty;
(b) CJln (a, ~' ro) = ~ !lU, U' (a, ~' ro),
(U, U')E~
fPs (a, ~' ro) = ~ Vu, U' (a, ~' w),

(U, U'JE~
where Jlu. u• and 'Vu, u· are infmitely differentiable functions with

support contained in U n U',
(c) for any pair (U, U') E !1
Jlu, u' (au (x, P, ro), p, ro) +
+ [Vu, u·'Vu, u' (au• (p, p, ro), p, ro)l (x, p, ro) = 0.
Let y ( U) be the index of the chain of patches starting with the
patch Uio and finishing with the patch U. Then
[KYcpJ (x, p, ro) = ~ { (i sgn ro)v(U) eiws(rx, ~. w) x
(U, U')Eil
-00
X I ;~ ,- 2 vu, u• (a, p, ro) lrx=au•(P, ~. w) dp} · (8.5)
According to the method of stationary phase every summand in

the right-hand member of (8.5) is of the following form:
1
(i sgn ro)v(U) eiwS(a, ~. w) I~ ~-2~ X

ar:t. rx=au(x, B. W)
X {J.tu, u· (au (x, p, ro), p, ro) +

+ [Vu, u•Yu, u• (aU' (p, p, ro), ~. ro)] (x, p, ro )} +
+ eiwS(au(x, ~. w), ~. w)r! (x, B, ro) + r2 (x, p, ro),
where the function r 1 (x, ~' ro) is supported in the domain of the
function au (x, ~. ro) and the functions r 1 and r 2 satisfy the condi-
tion
ro 1+1ri (x, p, ro) E $k 1, ... , kn+z (Rn+ 2),
where j = 1, 2 for any k1 , ••. , kn+ 2 • Hence the function

wl-k 1 _ ••• -kn+t [eiwS(au(x, ~. w), ~. w) r 1 (x, p, ro) +r 2 (x, p, ro)]
belongs to the space $ k1, .•• , k n+2 (Rn+ 2), i.e., it is equivalent
to zero.
From this theorem it follows that the operator K 1 induces some
operator on :£ to Skt, ···• hn+2. 1. This operator will be called
a canonical operator f!J[' 1•
CH. III. ASY1MPTO'Tl'C METHODS 333
Now let A 1 , • • . , An+ 2 be generators of degree k1 , • • • , kn+ 2 ,

respectively, acting on a Banach space B and defined on a dense
linear manifold D. The operator
(
8 2 s-1 s+1 n+2 1 ) Al-k 1 - ••• -hn+i
[K1cp] At. A 2, ... , As-t. As, ... , An+t. An+z n+2
is bounded in B if cp = 0. Let Op be the space of linear operators

in B defined on D and factorized according to the following relation
of equivalence ~= A 1 ~A" if and only if the operator (A 1 - A ") X
X A~f.~f- ... -kn+1 is bounded. If the elements cp 1 and cp" of Lr.
are representatives of one and the same class cp EX then
s-1 s+1 n+2 1 )

· · ·, As-b As, · · ·, An+b An+2 •
(8.6)
We shall denote the class of equivalent operators on Op which the
left and right members of (8.6) belong to by
(
8 2 s-1 s+2 n+2 1 )
[K1cp] At. Az, ... , As-t. As, ... , An+t. An+2 •
Sec. 9. The Canonical Operator on the Family

of Closed Curves
1. The definition of the canonical operator [{ l on LA· Consider
the family of closed curves M 1 (~, ffi) in the phase plane (x, p) defi-
ned by the equations
x = x (a, ~' ffi), p = p (a, ~' ffi),
where a is a point on the circle, ~ E nn, (I) belongs to the extended
numerical axis and the functions x (a, ~' ffi) and p (a, ~' ffi) are
supposed to be infinitely differentiable, for ffi = oo as well.
For the reader who is unfamiliar with the theory of manifolds
we shall explain some topological concepts related to the space
of parameters a, ~ and ffi considered in this section. The convergence
of a sequence of points on the circle will be understood as the con-
vergence of this sequence on the plane where the circle is lying.
The convergence to oo of the sequence of points of an extended
numerical axis is understood in the obvious way. We shall say
that the sequence {(am, ~m' ffim)} of points of the space of para-
meters converges to the point (a, ~' ffi) if
lim am= a, lim ~m = ~ and lim ffim = ffi.
m-+oo m-+oo
334 OPERATIONAL :\fETHODS
Next, the concepts of a closed set and of a closure of a set in the

space of parameters (a, ~' w) are defined in the same way as in Rn.
A compact set in the space of parameters (a, ~' w) is a closed set
contained in some set of the form I ~ I < const. An open set is the
complement of a closed set. If the parameter t is introduced on th~
circle in a neighborhood of the point a 0 then in the neighborhood
of the point (a 0 , ~ 0 , w0 ) of the space of parameters every function
of a, ~ and w may be considered as a function of t, ~ and w. This
enables us to introduce the concept of a differentiable function
defined on the space of parameters a, ~ and w. Below we shall assu-
me that the points where w = 0 are excluded from the space.
As before, a complete canonical atlas .r.' oo and a canonical atlas .f.~
of a neighborhood of the compact set are introduced.
Let (a 0 , ~ 0 , w 0 ) be a fixed point in the space of parameters and
C = (U1 , • • • , Um) be the chain of patches, where (a 0 , ~ 0 , w 0 ) E U1 •
Consider the function S 1 (a, ~' w) defined in the patch U1 by the
formulas
dSt(a, ~' w) = p (a, B, w) %~ (a, B, w) da,
S t( a 0 , Bo, m0) = 0.
In order to interpret the symbols :a and da from now on we shall
identify a moving point a on the circle with the polar angle of this
point.
Let (a1 , ~ 1 , w1 ) E U1 n
U 2 • In the patch U 2 defme a function
s2 (a,~' w) by the formulas
ox (a, ~, w) d a,
dS (a, ~, w) = p (a, B, (J)) arx
2
S 2 (a 11 ~ 1 , m1) = St(a 11 ~ 1 , m1 ).
Continuing this procedure we shall come to a function Sm (a,~' w)

in Um, which will be denoted by Sc (a,~' w). Defme a function
Sc (a, ~' w) by the formula
Sc (a, ~. w) = -Sc (a, B, w) +p (a, B, w) x (a, ~' w).
Define the index yc of the chain of patches C as before in Sec. 6.
Besides, introduce V-objects in the space of parameters in the same
way as in the previous section.
Bring every patch Ui of the atlas .Jl; in correspondence with the
chain Ci of patches of this atlas starting with the patch Uio contai-
ning the point (a 0 , ~ 0 , w 0 ) and fmishing with the patch Ui. Denote
YCj = Yh Scj = sj and Scj = sj. Define on Ldl the canonical
operator K 1 in the following way: let
cp =(.It', {cpi (x, B, w)}iEJt, {cpi (p, B, m)}iEI2);
CH. III. ASY':\1PTOTIC l\1ETHODS 335
then
[K 1cp] (x, B, co) dcf ~ (i sgn w)"~i eiillBFk• 13 • ill) X
iEl!
1
X I ~: ~-T la=au }
.(x, J3, ill) CJli (x, B, cu) +
""' (" sgn w)"hV/
+ LJ L - ul2ni f[
.J e
iill(px-Sla, fl, ill)) X
jEJ2 -00
1
X\ ;~ ~--zl=au.Cl', fl, ill) CJli (p, B, ul) dp.

}
2. The condition of quantization. The following lemma is valid.

Lemma 9.1. Let C = (U1 , • . • , Um), where Um = U 1 be the
closed chain of patches of the atlas .II oo· Then yc is even.
Proof. The index of a closed path in the space of parameters is
equal to the index of the circle
B= const, w = const,
[x(cz, B, cu)-x 0 ]2+fp(cz, B, cu)-p0 ] 2 =R 2
which passes on the phase plane integral number of times clockwise
and counterclockwise. Calculate the index of the circle, it is equal
to 2. The lemma is proved.
Suppose now, for simplicity, that B is a one-dimensional parame-
ter. From the lemma it follows that the function
[K 1cp] (x, B, cu)
does not depend on the choice of chains of patches cj at the points.
w .\ pdx-:n=O (mod2:rc). (9.1)
ll:fl(fl, ul)
Equation (9.1) correlates the variables B and w when the family

{M1 (B, w)} is given. Equation (9.1) is known as the "condition
of quantization".
Example. Let Lhe curve M 1 (B, cu) have the equation x 2 +p 2 = ~·
Then the condition of quantization is of the form
w;n;~ - ;n; = 0 (mod 2:rc). (9.2)
2 [Ew]+1
Let e be an arbitrary real number. Set B((J)) (J)
..
where [ecu] is the integral part of scu.
If ~ = ~ (co) is substituted in (9.2) then this condition is identically

satisfied. Note that ~ (co) is a bounded function with discontinuities
at isolated points. This situation is typical.
Proceed to the general case. Let the equation
~ p dx=a
MI(fl, ro)
with respect to unknown ~ have a unique solution

~ = [3 (a, co),
where p (a, co) is an infinitely differentiable function, and the
derivatives of j3 in co of all orders are bounded. Set
(J(e, co)= 2[ew]+1 ,
(l)
where [ecol is the integral part of the number eco and denote
~ (e, co)= if (a (e, co), co).
Then the substitution ~ = ~ (e, co) turns the condition of quanti-

zation (9.1) into an identity.
The operator Kquan defined by the formula
[Kquan cp] (x, e, co) = [K 1cp] (x, ~,(e, co), co)
will be called a quantized canonical operator. The quantized cano-

nical operator does not depend on the choice of chains cj.
3. The regularization of the canonical operator. The function

{Kquan cp] (x, e, co) is not a smooth function. For this reason it can
be used for the construction of a function of operators. In order
to avoid this difficulty we shall regularize the canonical operator.
Let r be an integer. We shall correlate the function Krcg cp defined
in R3 with the element cp E L-1:.:
00
[Kreg(jl] (x, e, co)= co-r ~ p (e- e') [Kquan(jl] (x, e', co) de', (9.3)
-oo
where p (e) E Co
(R). The operator Kreg will be called a regularized
canonical operator.
Lemma 9.2. Let r;;;;:;: s1 + s + 1.
3 Then
[Kreg(jl] E !il s1, s2, sa (R 3 ).
CfL III. ASY,:v.IPTOIJ'I1C METHlODS 337
Proof. Denote f (x, E, w) = [Kquan{jl] (x, E, w),

f (x, E, w) = IKregfP] (x, E, w) =
00
= :r ~ p(e-e')f(x, e', w)de'.

-00
The function f (x, E, w) is of the form

f (x, e, w) = h (x, ~' w) if3=B([ero], ro)> (9.4)
where h (x, ~' w) =
:::
l&fzfP] (x, ~' w), ~ (lc:wl, w)
= w - (2[ew]
~ w
+1, )
and the functions h and fl satisfy the following conditions:

(a) h is infiniLely differentiable;
(b) h (x, ~' w) = 0 for sufficiently large I~ I and h (x, ~' w) = 0
for sufficiently small I w I;
(c) for any function e (x) E Co
(R) which is equal to unity in
a sufficiently large neighborhood of zero, the function 11 - e (x)l X
X h (x, ~' w) belongs to any Sobolev space;
(d) for any function e (x) E Co
(R) the function
I(_!_)r
iJB
(_!_)r2 (-i) )rae
i.Jx
(x) h (x, B, w)
i.Jw
c
wr
I& V. +
-.-:::: w2 1
for r ~ r1 +r +12 and any ra (c = const depending only on
r 1 • r 2 , ra, r);
(e) _!he function tf
(£, w) is infinitely differentiable;
(f) ~(lew], w)-+ oo f~ e-+ oo uniformly in w;
(g) the deriva.0-ves of~ (£, w) in w of all orders are bounded in
any set, where ff (£, w) is bounded.
In (9.3) we shall make a change of variables = e' w. Then we s
obtain
00
g(x, E, w)= (.)+ 1 ~ p (e-!) h(x, ~([£], w), w)d£. (9.5)
On differentiating the right-hand member of (9.5) and by using

the properties (a)-(g) we obtain the proof of the lemma. Note that
g (x, e, w) = 0 for sufficiently large I e I·
Consider also the operator K;eg defined by the formula
00
[Kregcp] (x, e, w) =·X~~) ~ e-<e-e') 2 ro 2• [Kquancp] (x, e', w) de',

-00
(9.6)
where x (~::) E Co
(R).
Lemma 9.3. Let r ~ s1 + rss + rssa + rssa + 1 if s ~ 1 and
2
r ~ s1 + +
sa 1 if s = 0.
22--01225
338 OPEHATIOKAL METHODS
Then
[K;eg<p] EfB s1, s2 , sa (R3 ).
Proof. Make a change of variables of integration
s= s'co
Then we obtain
')((8)
00
) -(e-J..)wZs
[K;eg<p] (x, e, co)=--
wr+1
e w X
-00
x h (x, 13 ([£ co], co), co) ds, (9.7)

where the functions h, ~ are the same as in the proof of the previous
lemma. By differentiating the right-hand member of (9.7) we obtain
[K;eg<p] E W2si, s,;, s~ (R 3),
where si > si + {,
i = 1, 2, 3. Hence follows the proof of the
lemma.
Consider in the space !B s 1, s2 , sa (R3 ) the following relation of
equivalency:
l
<p (x, s, co) ~ 'ljJ (x, s, co)
if and only if
co 1+1 [cp (x, e, co)- 'ljJ (x, e, co)] E fBst,s 2,sa·
Theorem 9.1. Let <p, 'ljJ ELl:. and <p == 'ljJ. Then for r ~ s + s + 1 1 3
l
Kreg<j) ~ Kreg'ljJ;
for r ~max {s1 + s + 1,
3 s1 + rss + rss + rs- s
2 3 3}
l
K;eg<j) ~ K;eg1J'.
Proof. Following the proof of Theorem 8.1 it suffices to show
l l
that if <p =
0 then Kreg<j) ~ 0 and K;eg <p
the previous section we obtain
~ 0. Let <p = 0. As in
00
[Kreg<p] (x, s, co)= - 1-

wr+l+Z JI p (s-..l) w
h (x, ~ ([s], co), co) ds,
-00
1 )oo - (e-l..)Z wzs =

[K;eg<p] (x, s, co)=--
wr+l+Z
e w h (x, B([s], co), co) dt.,
~
where the functions h and 13 satisfy the above conditions (a)-(g),

Q.E.D.
CH. III. ASYtNIPTOTI'C ME'f'HiODS 339
Now let A 1 , A 2 , A 3 be a generator of degree s1 , s 2 , s 3 actin~

on a Banach space B and defined on a linear manifold D dense every-
where.
In the space Hom (B, B) introduce the relation of equivalency
~= A'~ A if and only if the operator (A' -A
II A~~ is bounded. 11
)
The factor-space Hom (B, B) I""" will be denoted by Op (B).

From Theorem 9.1 it follows that if the elements cp' and rp of II
LJ; are representatives of the same class cp E :tJ and r satisfies the
condition of Theorem 9.1, then
[Kregcp']
(A3 A2 A1') ~ [KregfP (3A A2 A1)
1, 2, 3
11
] 1, 2, 3 , (9.8)
(9.9)
The class of equivalent operators on Op (B) which the right and
3 2 1 )
left members of (9.8) belong to will be denoted by [QJ( regcpl ( A 1 , A 2 , A3 ;
and the class which the left and right members of (9.9) belong to
will be denoted by ·
( 3 2 I )
[Qf(~cgcp] At. A2, Aa .
In the same way, the regularized canonical operators correspon-
ding to other methods of ordering the triple A 1 , A 2 and A 3 such as
are defined, where cp is an arbitrary canonical element.
Sec. 10. An Example of Commutation of a Canonical Operator with

a Hamiltonian
[n this section we shall consider the simplest case of a curve
ML (~, w) which does not depend on parameters ~ and w. For the
convenience of notation we assume that w = 1/h. In order to illust-
rate general situation we shall first consider the following example.
Example. Find a "nearly" exact solution of the equation y + h; II
+ xy = 0 for h > 0 in the region I x I <a to an accuracy of 0 (h3),

i. e. find such a function y (x) which on substitution in the given
equation yields a residual term 0 (h3 ).
This problem is trivial because after the {-Fourier transform
we arrive at the equation ~2y - ih~~ = 0 which is a linear equation

of first order. Nevertheless we shall construct its solution by means
22*
of the canonical operator so as to illustFate the general situation.

For this reason we shall not use the simplest canonical atlases.
Seek y in the form y = K 1 c:p, where [{1 is a canonical operator
(mod 0 (h2 )) on the curve A= {x . ~2 , p =a} associated with
the Hamiltonian - h22 : :2 - x. {In the canonical operator we make
the substitution ro = ! .)
Consider the canonical atlas on A which
consists of non-singular patches U 1 and U 2 and a singular patch U 3 ,
where the patch ul corresponds to the region a> 1, u2 to the
region a< -1 and U 3 to I a I< 2. The situation in question
is essentially general.
The canonical element c:p which the operator [( acts on is defmed
by the three functions: c:p1 (a), c:p 2 (a), and c:p 3 (a) also depending
on h with support contained in the patches U 1 , U 2 , U 8 , respectively.
These functions are defined correct to modulo 0 (h 2). The olher triple
of functions 'lj)1 (a), 'lj) 2 (a) and 'ljJ 3 (a) defines the same canonical
element c:p if the following conditions are fulfilled (modulo 0 (h2 )):
fP1 (a) = '\j)1 (a) for a > 2,
([) 2 (a) = '\j) 2 (a) for a < -2,
fPa (a) = '\j)a (a) for I a I < 1,
fP1 (a) + V1, afPa (a) = 'lj)l (a) + V1, a'\j)a (a) for a E U1 n Ua,
fP2 (a) + V2, afPa (a) = 'lj)2 (a) + V a'\j)a (a) for
2, a E U2 n Ua,
where V1 , 3 (V 2 , 3 ) is a transition operator on the patch U 3 to the
patch U1 (U 2). In the case in question according to (8.2) it is of the
form:
def
Vi, 3 = V2, 3 = V =
= 1- ih [ 2~ a!2z - 2~2 a! + 2:a3 Jmod 0 (h2).
The canonical operator is of the form
-00
where
3 3
2
Sdx) = (2xi2 S 2 (X
) _- - (2x)3
Y1 (x) = V2x, y2(x) = - V2x.

CR. III . .AJSYIMPTOJ'IG METHODS 341
h2 d2
Dtmote H = - 2 dx 2 • We have
HK 1 = -ih K 1 P 1 , HK 2 = -ih K 2 P 2 , HKa = -ih K 3 Pa,
where
·{ 5 1 d
P 2 -P
-
d
1 -----~h
da
-
8a4
- - -d
a3 da + -1- -da2
2rx2
d2}
- ' Pa=-.
da
The triple of operators P 1 , P 2 and P 3 defines an operator P acting

in the space of canonical elements since the following conditions
are fulfilled
VP 3 = P 1 V mod 0 (h2 ), VP 3 = P 2 V mod 0 (h2).
Thus, there exists the following formula of commutation of a cano-
nical operator with a Hamiltonian
HK = -ihKP.
Therefore the equation Pep = 0 is to be solved in a neighborhood
of the region a 2 <. 2a, i.e., it is necessary to find such a canonical
element cp that the element '\jJ = Pep will be defined by the functions
'\jJ1 (a), '¢ 2 (a) and '¢ 3 (a), where '¢i (a) = 0 for I a I = V2a + e, e
being an arbitrary positive number. The element cp is defmed by
the three functions: cp1 (a), cp 2 (a) and cp 3 (a). The equation Pep = 0
+
for I a I < V 2a e will be satisfied if the functions CfJi satisfy the
following conditions:
(a) P 1cp 1 (a)= 0 for 2< a<V2a+e,
(b) P 1cp!(a)+VP3 cp 3 (a)=0 for 1 <a<2,
(c) Pacra(a)=O for I aj<1,
(d) Pzcp2(a)+VPacra(a)=0 for -2<~<-1,
(e) P 2cp 2 (a)=0 for -:V2a-e<a<-2.
Since P 1 , P 2 and P 3 define the operator P conditions (b) and (d)
may be rewritten in the form
(b') P 1 (cp1 (a) + Vcp 3 (a)) = 0 for 1 <a< 2,
(d') P 2 (cp 2 (a) +
V(p 3 (a)) = 0 for -2 <a< -1.
Start with equation (c)
cp; (a) = 0.
The solution o£ this equation (correct within an arbitrary constant
factor) is as follows
cp 8 (a) = 1 for I a I < 1.
Now denote
er 1 (a) + Vera (a) = X (a),
The equation (b') becomes
ax . { 5 1 a 1 az }
da - ~h 8a4 ---;xa da + 2a2 aaz X (a)= O,
x(1)=Ver 3 (1)=1- 254 ih.

Solve this equation by means of the perturbation theory
x = x<o> hx(l)' +
ax(O)
d<X=O, xe(a)=X0 (1)=1,
dx(1>-
da -
+i {-5____a31_ _!}__+_1_~}
Sa4 da 2a2 da2
')((Ol(a)
" '
ax(i> - ~ Xd> (1)- -~

da - Sa4 ' - 24 ·
Hence we obtain
5i 5ih
X(i> (a)=- 24a3 ' X (a)= 1 - 24a3 ·
Let e (a) = 1 in a neighborhood of 1 and e (a)= 0 in a neighborhood
of 2. We can set
Vera (a) = e (a) X (a), er 1 (a) = (1 - e (a)) X (a),
1 <a <2.
Hence we find
fP3 (a)= v-Ie (a) ( 1 - 2~:3 ) =
=[ 1 +ilz{ 2~ d~2--2~2d~ +245a3}Je(a) ( 1 -2~:3)-
= e (a)+ ih [ 2~ !
e" (a)- 2 2 e' (a) J, 1 <a< 2.
At a = 2 we obtain the following initial condition for er 1 (a):
5ih
erd 2) = x (2) = 1- 192 •
Next we solve a Cauchy problem:
drn'l0l
--;Ia--= 0, (jJ~Ol (2) = 1,
drp~l)
~= ~
. { 5
Sa 4 --
1
--;;:3 da
d + 2a21 d2
da2
} (Ol (
fP 1
)
a '
u>
(P1 (2) = - 5i
192 , 2 < a <V-2a + s.
CH. III. ASYr:\IPTOTIC METHJODS 343
We obtain
cp!<0) ( a ) = 1 ' cp!<1) ( a ) = - 5i.

24o:3 '
(j)[ (a)= 1 - 5ih

24 o: 3 for
2 <a< v-2a +B.
In the same way we find
l5ih
cp 2 (a)+Vcp 3 (a)=1- 24a 3 ,
cp 3 (a)=e(a)+il{ 2~ i'(a)- 2~ 2 ? (a)]

for -2<a<1,
e
where (a)= 1 in a neighborhood of 1 and e( a) = 0 in a neigh-
borhood of -2,
cp 2 (a)=(1-e(a)) (1- 2~i:3 ) for -2<a<1
and
(p 2 (a)= 1- 245ih
a 3 for - v-2a- B <a<- 2.
Now it is easy to write the solution and find out how it is related
to the exact solution which can be obtained for the case in question.
Now let M 1 be an arbitrary smooth curve in the phase plane
(.:r, p) and K be a canonical operator mod 0 (h 1+1) on the curve
M 1 = {x (a), p (a)}. The operator acts on canonical elements.
Let .Ji= {Ui}iEI be a canonical atlas on the curve 1~f. 1 • The cano-
nical element cp is given by the family of functions {cpi (a)}iE r.
where supp cpi E Ui. The functions cpi are defined by mod 0 (h'+ 1),
cpi E C~ (Ui)·
Let A' = {Uj} iE I ' be anotn er canonical atlas. The family of
funclions '\jlj (a) E C~ (Uj), j E I' defines the same element cp as
the family {cpi (a)}j E I if the following conditions are met:
(a) in a sufficiently small neighborhood of any non-local point
on M 1 the following equality is valid:
~ cpi(cG)+lJ Vcpj(ct)=h '\jlj(a)+~ V\pi(a),
jEI1 .iEI~ .iEii iEiiJ
where
I 1 is the set of numbers of non-singular patches of the atlas .11:,
I; is the set of numbers of non-singular patches of the atlas A',
I 2 is the set of numbers of singular patches of the atlas .f{,
I~ is the set of numbers of singular patches of the atlas .:11:',
V is a transition operator from singular patches to non-singular
ones (it is defined in a neighborhood of any non-focal point and
has the form

l
V= 1 + k=i
~ (ih)k Lk mod 0 (h 1+1),
where Lk is a differential operator of order 2k);

(b) in a sufficiently small neighborhood of any focal point on jJ;J1
~ cpi(a)=~ lJli(a).
iEI2 iEI2
The canonical operator K acts according to the formula

Kcp = ~ iviK;cpi (a),
iEI
where
for j E 11 and
hi -
(px-S(rx))
e {jlj (a.)
dp
for j E I 2 •
Let H be a Hamiltonian or a pseudodifferential operator of the
form
H=&'t (~,~'h),
~ d ~
where p = -ih dx (the operator h commutes with x and p, hence
it is not important in which order it acts). Let the curve }11] 1 be
associated with the Hamiltonian H, i.e.~
&'t(x (a), p (a), 0) = 0,
and the parameter a be chosen in such a way that the relations are
carried out on M 1
da = dx
dp
H p (x, p, 0) Hx (x, p, 0) ·
Then for any j E I we have

HKiCfJi (a)= -ihKiPiCfli (a),
CH. III . .ASYIMPTO'rl:C METHODS 345
where Pi is of the form

l
Pi= ~ (ih)k P7'> mod 0 (h 1+1).
k=O
Here P(ll)
)
are differential operators and
P)ol = :X + f (a),
where f (a) does not depend on j. Hence in a neighborhood of any
point on 1111 the operator Pi depends only on whether Ui is non-
singular or singular.
The family of operators {Pi} defines some operator P acting
on the space of canonical elements. In other words, if Ui, is non-
singular and Uh is singular, then for any function X E C';(Uil n Ui2)
Pi!VX = VPi2X·
In fact we have
HKrp = -ihK¢,
where the canonical element 'ljJ is defined by the family of functions
'IJli = P iCPi· It remains to be shown that the canonical element 'ljJ
is defined uniquely by the canonical element rp or, what is the same,
that the canonical operator is monomorphic:
(K'Ijl = 0) ~ (~1 = 0).
Let the canonical element ;tJ.l be non-zero; prove that K"¢ =I= 0. For
x
example, let 1jJ (a) =1= 0. Denote = x (a) and consider the function
K"ljl in a sufficiently small neighborhood of the point Choose x.
on }11!1 such a canonical atlas .JI; that any two patches of the atlas
whose projections onto the axis x contain the point have non- x
intersecting projections onto the axis p. There exists a canonical
element W defined by the family of functions {lJli} which correspond
to the atlas .:1/; such that the following conditions are fulfilled:
(a) lJli = 0 if the projection of the patch Uj onto the axis X does
not contain the point X,
(b) KW = K1p in a neighborhood of the point It suffices to x.
verify that KW =1= 0. Denote Kqt (x) = <!> (x). In a sufficiently
small neighborhood of a point
;~
we have x
CD (p) = ('ie- h S(a) '~i (a) mod 0 (h 00 ) ,
·(dp
da a=a/p)
where Ui is a patch containing the point a. Thus iii (p) =I= 0, Q.E.D
Consider the problem: find a function y (x) such that the equation
Hy = 0 is satisfied for I x I <a with an error 0 (h 1+ 2). We shall
seek y in the form
y = Kcp.
For a canonical elemenl cp we obtain the condition
supp Pep n Ma = 0,
where Ma is an image of the segment I x I <a by projection of Jl!J 1
onto the axis x. The canonical element cp satisfying this condition
can be constructed by means of the perturbation theory in the
same way as in the above example.
Sec. 11. Commutation of a Hamiltonian with a Canonical Operator

Let {p 1 (a), p 2 (a)} be a weighted partition of unity on M\ i.e.
p 1 , p 2 ECoo, let supp p1 not contain points which do not belong to
any non-singular patch of the complete canonical atlas dfoo of the
curve 1l1\ and let supp p 2 be contained in the union of singular patches
of the atlas ,J;' oo; p1 +p 2 = 1. The weighted partition of unity
{p 1 , p 2 } corresponds to the operator x
translating the function cp
(defined by mod 0 (h 1+ 1)) on M 1 into a canonical element on M 1
which is described in the canonical atlas .:fl: = {U i LEI by the
formulas
cpi (a) = ei (a) p1 (a) cp (a) + Vei (a) p 2 (a) cp (a),
if uj is a non-singular patch and
cpi (a) = V- 1ei (a) p1 (a) cp (a) +
ei (a) p 2 (a) cp (a),
if Ui is a singular patch. Here {ei} is the partition of unity cor-
responding to the atlas .JI. Denote
S = K'£.
Lemma 11.1. The operator S is invertible.
The proof is analogous to that of the inverlibility of the canonical
operator K.
Lemma 11.2. Let the curve M 1 be associated with the Hamiltonian H.
Then for any function cp of the domain of the operator S the function
HScp is of the form -ihS 'ljl, where'ljl is a function on M 1 •
Proof. We have HScp = H K 32 cp = -ihKPxcp. Lel in the atlas
.J{= {Ui }i E I a canonical element Pxcp be defined by the family
of functions {1jJ i (a)}. It suffices to show that for any j E I there
exists such a function Xi on M 1 that the element XXi is defined
CH. III. A.SY\MPTOTIC ME1'HODS 347
by the family of functions

'h = 'IJlkoki·
Let Ui be a non-singular patch. Then the function 'Xi is defined
by the equation
(Pt + Vpo) 'Xi = 'IJli· (11.1)
The operator V has the form
l
V= 1 + k=l
~ (ih)k L,..
For this reason (11.1) can be rewritten in the form
[ '1 + ±
k=l
l
(ih),. Lit 'Xi= 'IJli·
_j
The function 'Xi can now be determined by means of the perturba-

tion theory. In the case when Ui is a singular patch the functions
'Xi are found in an analogous manner. The lemma is proved.
Theorem 11.1. Under the assumptions of Lemma 11.2 there exists
the following commutative relation:
HS = -ihSPq,, (11.2)
l
P<IJ = !, + r (a) + ~ (- ih)i P;f}, (11.3)

j=1
where r (a) is a function and P!r., are differential operators on .1111 •

Proof. According Lo Lemmas 11.1 and 11.2 there exists the com-
mutation relation (11.2) with an operator P<IJ acting on functions
defined on M 1 correct to modulo 0 (h 1+ 1). It remains to be shown
that P,D is of the form (11.3). For this it suffices to consider the action
of this operator on a function with support in a single patch of the
canonical atlas. For example let cp (a) be a function with support
in a non-singular patch Ui. Then the canonical element J!cp may be
determi11ed by the set of functions: cph (a) = 0 for k =1= j and
<Pi (a) = p1 (a) cp (a) Vp 2cp (a). +
By applying the operator P to the element xcp we obtain the
Hlement P?£cp which may be determined by the following family
of functions {'IJlh (a)}:
'IJlk (a) = 0 for k =1= j and
l
l\Jj(a)= d(pd;o:) -f-r(a)cpj(a)-f- ~ (-ih) 11 P) 11lcpi(a),

~t=1
where r (a) is a function, P}~t> are differential operators. On cal-

culating now the function 'ljJ = Pci>fJJ = ~- 1 P~qJ by means of the
perturbation theory in a manner analogous to the proof of Lemma 11.2
we obtain formula (d. 1.3) for the case in question.
The case when qJ (a) is a function with support in a singular
patch can be treated in a similar manner, Q.E.D.
Theorem 11.1 enables us to reduce the equation H y = 0 to the
equation P<ll qJ = 0. The solution of the latter equation can easily
be obtained by means of the perturbation theory.
Sec. 12. The General Canonical Transformation

of the Pseudodifferential Operator
1
Let w = ih fx
1
( 2 1 )
and T = [KqJ] x, w , where K is a canonical
operator on a family of curves depending on the parameter w and
qJ is a real function. Consider the transformation of the pseudodif-
ferential operator L = f
. (2x, w,1 w3) by the formula
L-+ T*LT. (12.1)
This transformation contains the canonical transformations of
Sec. 1 as particular cases. In this section an asymptotic expansion
correct to 0 (h 2 ) of the transformation (12.1) will be obtained.
Lemma 12.1. Let M 1 : {x = x (a), p = p (a)} be a smooth curve
in the phase plane (x,p), K be a canonical operator (correct to 0 (w- 1))
on M\ f (x) be a smooth function. Then for any function fp 1 (a), fjJ 2 (a) E
E C'; (M 1) the following equality is valid:
(f (x) [KqJ 1 ] (x, w), [Kfp 2 ] (x, w))L 2(x) =
=(f(x(a)) qJt(a), fJJ2(a))L 2(rxJ+O (!) ·

Proof. Obviously it suffices to consider the case when the support
of the function f is sufficiently small. Then on the curve M 1 such
a canonical atlas A can be chosen that if the projections of two
patches U i and U k• j =I= k of the atlas .fl: onto the axis x intersect
with the support of the function f, then the projections of these
patches onto the axis p do not intersect. Let {U i }JEJo be a family
of patches of the atlas A whose projections onto the axis x inter-
sect wiLh supp f. Denote by Uj the set n -l [(supp f) n (n Ui)], where
n is a projection onto the axis x. Then correct to 0 (w-"") we obtain
KqJ 1 = 2J
jEJo
KqJu, KqJ 2 = 2J
5EJo
KqJ 2j, on support /,
'CH. III. ASY1MPTO'l'IC METHODS 349
where supp cplj c uj! supp cpzj c uj and cplj (a) = cpz (a) for
n E Uj and cp 2 i (a) = cp 2 (a) where a E Uj. If j =I= k then
(f (x (a)) fPti (a), cp 2 k (a)bca) = 0. Besides
(f (x) [Kcp 1 j] (x, w), [Kcp 2 k] (x, w)) = 0 (w- 1). (12.2)
In fact, denote
'¢1i (x) = [Kcp 1i] (x, w), '¢ 2 k (x) =
= [Kcp 2 k] (x, w), '¢ 1 i (x) f (x) = '¥ 1i (x)
(the functions '¢u, '¢zk depend on w as on a parameter). We have
'ilu (p) = e -iwSr,alP>>xli (p)' liJzk = e -iwS(ak(P))Xzk (p)'
·where the functions Xti and Xzk have non-intersecting supports.
Denote Si (p) = S (ai (p)). We have
Wu (p) = f ( ~ a~) e-iwB/P>xu (p) =
= e-iriS/P>f (_!__ ~ +S] (p)) Xii (p) =

w op '
= e-iWs/P>f (Sj (p)) Xu (p) + 0 (w-1)
so that the function 'I'li (p) is equal to 0 ( w - 1) on the support of
the function 1i'zk·
By virtue of (12.1) it suffices to prove the lemma in the case when
q1 1 (a) = cp 1 i (a) and cp 2 (a) = cpzi (a). Then, however, the proof of
the lemma directly follows from the definition of the canonical
operator.
Lemma 12.2. Let M~: {x = x (a), p = p (a)} be a smooth curve
on the phase plane (x, p), K 1 be a canonical operator on A (to an accu-
racy of 0 ( w - 2 )), f (x) be a smooth function and x
be the operator
translating smooth functions on JV/ 1 into canonical elements in the
usual way and corresponding to a real weighted partition of unity.
Then for any real function cp (a) E C'; (M1) the following equality is
valid: 0
(f (x) [K 1 3:cp] (x, w), f (x) [K1 xcp] (x, w))L 2cx> =
=(If (x (a))l 2 cp (a), cp (a))L2(a) +0 (w- 2 ).
The proof of the lemma is analogous to that of Lemma 12.1.
We shall note only a few additional features which distinguish the
proof of Lemma 12.2 from that of Lemma 12.1.
(1) Let the function cp have support in a non-singular patch U.
Then as is easily verified by using the definition of operators V u, u'
350 OPERATIONAL :.\1ETHODS
we obtain
1
[K 13!cp] (x, w) = ceiwS(a(x)) 1.3.:__ ~- 2 [ ( 1 + ~L) cp (a)] ,

da w a=a(x)
where L is a differential operator (of second order) with real coef-

ficients and I c I = 1. The statement of the lemma is valid in this
case because
I(1+! L) cp(a) j =I cp(a) J2+0(w-2).
2
(2) An analogous formula exists for the w-Fourier transform of

the function K 13!cp, if U is a singular patch.
(3) Let U be a singular patch. Denote 'IJl (x) = [K 1 3!cp] (x, w),
lf(x)=f(x)ljl(x), Sl(p)=S(a(p)), :: =Jt(p(a)).
By expanding the function
1
1¥ (p) = Cie-iwSiP)j ( ~ i)~ + s; (p)) IJi (p) ,- 2 X
x[(1+_!_L~)cr(a)j
W a=a(p)
J (12.3)
in powers of w we obtain that in equality (12.3) we can replace
the operator f ( ~) {)ap + s;
(p)) correct to 0 ( (!) - 2 ) by the following
operator:
-
j (S; (p)) + wi L2,
where L 2 is a differential operator with the real coefficients in the
case of a real function f (the latter can be assumed without loss
of generality). Consider the set of the operators depending on w -+- oo
which are of the form
(12.4)
where A 1 and A 2 are continuous operators acting on and trans- Co
lating real functions into real ones. The set of operators of the type
(12.4) form a circle x relative to the composition of operators
(A + ~ 1 A 2 +0(w-2 )) (Aa+ ~ A 4 +0(w-2 )) =
=AtAa+~A 5 +0 (w- 2).
(J)
Consequently,
1
1f (p (a))= c1e-ic.ls(a) j :: ,- 2 ( f (x (a))+ ~ L 2 ) cp (a)

which leads to the proof of the lemma in the case in question.
CH. III. ASY'MPTOTIC METHODS 351
Note. Lemma 12.2 remains valid if in the left-hand side of (*)

2 . 1
f (x) is replaced by the operator f ( x,--; a/ax), and in the right-hand
side of (*) f (x (a)) is replaced by f (x (a), p (a)). The proof is com-
pletely analogous.
Theorem 12.1. Let r(i) = {(x, p): X= X (a, w), p = p (a, w)}
be a one-parameter family of curves in the phase space R~ X R~,
where D (X, P +
w)ID (a, p) = 1.
Consider a canonical operator [Kcp] (x, w) on the family r w applied
to a real function cp (a, w) E C'Q.
Let T=[Kcp] ( x,
2 1
w), where
1
w= -ih :x.
1
Then in the space L 2 (R)
T*T =cpU, ~) cp (~, ~) + h 2R 2 (h),

where the operator R 2 is bounded in L 2 uniformly in h: II R 2 (h) II :s;;;
:s;;; c.
Proof. From the definition of the canonical operator it follows
that [Kcp] (x, w) has compact support Q. Let Q"' be a projection of
() onto the axis w and
Q1 = {(x, w', w): (x, w) E Q and (x, w') E Q}.
Consider in a neighborhood Q1 a finite partition of unity
{f11 (x, w', w)} c Co
in which the supports f 11 are sufficiently
small. Then
,-1 (2 1 3) [Kcp] (2X, 1)-(2
T*T = 2.J fl1 X, w, (l) [KcpJ x,
3)(l) (l) '
11
where the line means complex conjugation. Prove that
2 1 3) (2x, w1)-
f 11 (x, w, w [Kcp]
(2 3)
[Kcp] x, w =
=cp(~. !) cp(~. ~) /11 (~ (x (~. !) +

+X(~.~)),~. !)+O(h 2). (12.5)
Summing up these operators with respect to ft we obtain the

proof of the theorem.
Let n (ri) be a projection of the curve r"' onto the axis x (or p).
Take into account that the support / 11 is arbitrarily small. Then
if the projections n- 1 (supp / 11 (x, w, w')) for w, w' E Qw intersect
X
with the patches ul! ... , u l of the canonical atlas rw then we

may consider that

ei (n - 1 (p + ()) - ()) 1
' ffi), ffi) ej (;:t -1 (p, ()) "), ()) ") = 0
for i=l=j and ffi, ffi', ffi"EQ 00 , where {e; (a, ffi)}isthe partition of unity
corresponding to the canonical atlas. (If this condition is not ful-
filled, then, for a given t11 , we may choose another canonical atlas
and use the invariance property of the canonical operator when
changing the atlas.)
On the support t11 the canonical operator is of the form
l
[Kcp] (x, ffi) = Lj [K ((pej)] (x, ffi)
j=1
and
t11 (~, ~' !) [Kcp] (~, ~) [Kcp] (~, !) =

= ±
i, j=1
[K (cpe;)J{~, !) [K (cpei)J{~, ~) / (~, ~, !) · 11 (12.G)
Prove that
[K(cpei)l(~, !) [K(cpei)J{~, ~) t11 U, ~' !) =h2R 2 (h), (12.7)

for i =1= j,
where II R 2 (h) II :s;;c. Indeed we have
i ~
- hSi(p, W)
Fx-p [K (cpei)l (x, ffi) =e 'i'i (p, ffi), ( 12.8)
i ~
Fx-.p [ K (cpei ) ] (x, ffi ') _

- e-h 8/P, W')
'i'i ( p, ()) ') '
where
¢; (p, ffi) lJli (p + ()) - ffi 1 , ffi') = 0 for ffi, ffi 1 E Qw. (12.9)
Since the supports lj;;, 'i'i do not intersect we can assume that in
(12.8)
si (p, ())> = s (p, ffi), sj (p, ffi) = s (p, (!)),
where S (p, ffi) is a smooth real function. Then
(12.10)
CH. III. ASY:\IPTOTIC METHODS 353
where
T g =ctcr Ji e+[p~-s(v,~)]
' ·
(
g p, w, w dp,
1 3)
Expanding the operator in the right-hand side of the equality

in powers of h we obtain
Xi= fu ( ~ (p, w'), w', w) lJli (p, w') +
m-1
+~ (-ih)kV(k)lJli(p, w')+(-ih)mQ)m)(p, w', w),

n=1
where V< 1 l, • • • , v<m-t) are known differential operators, and

Q/m' E w~ (W) for all r.
By virtue of (12.9)
lJli(p, w) iJph
a" lJli(p+w-w', w')=O (i:i=j).
Consequently, according to the results of Sec. 1 we have

n 1 Txi = (-ih) 111 T$lQ<.ml(i
J
=i= j).
Therefore, by virtue of (12.10) to obtain the proof of (12.7) we
must estimate in !ll 0 (R 3 ) the norm of the symbol of the operator
T~iTQ'.nn, i.e., the norm of the function
J
11(x, w', w)=(-ih) 111 Ji Ji dpdp' { eh...!:... (p'- p)x X
Xe
~ [S(j>, Ul)- S(p', Ul')]ljl. (
, p, w
) Q(m) ( 1
P, w, w
1 ) }
1 ·
We have
llllii,98 0 (RxXRUl'XRwl::::;;c1h ~ ~ 111

dpdp 1 X
II ~ [S(p, Ul)-Scp', Ul')]ljl., ( p, w) Q(m) ( P ' w , w II

X e 1 ,98 0<nUl'xR(J}l::::;;;
1 1 )
:::;;;c 2 hm ) ~ dp dp 1 II e ~ [S(p, Ul)-Scp', Ul')J X

Xljli(p, w)Q)m)(p~, w~, w)llwt<Rffi•XRUl)::::;;c3hm-2.
23-01225
354 OPERATIONAL :\IETHODS
This completes the proof of (12.7). Then from (12.G) it follows
f,, (x,
2 w,1 CD.3) [K(p] (2x, w1) [Kcp] {x,
'2 cu3) =
= ±
j=1
[K (cpei)l (~, ~) [K (cpej)] (~, ~) f 11 (~, c~, ~) +
+ 0 (h 2 ).
According to the theorems of Sec. 1 this operator is transformed

in the following manner:
l
~ (cpej)
(2 3) (cpej) (2 1)
X, W X, W X
j=1
xf11 (;x(.~, ~)+~X(~,~),~. !)+O(h 2). (12.11)

Next we have
(qJei) (x, w) (cpei) (x, w') = 0 for i =I= j and w, w' E Qw,
because we have assumed, in particular, that
supp ei (x, w) nsupp ej (x, w') = 0
X X
for i =I= j; w, w' EQ"'.

The expression (12.11) then coincides with the right-hand member
of (12.5) which completes the proof of (12.5), Q.E.D.
Corollary. Let f E Coo(Rx X Rw' X Rr,) and
cp = cp 0 (x, w) + (-ih) cp1 (x, w) + h2 cp 2 (x, w, h),
where (jlc, cp 1 are real functions, cp 0 , cp 11 cp 2 E C';, sup II cp 2 lln·nitc
2
xR ) ~
O<h<1 "'
~ cr for sufficiently large r. - -
Then if the conditions of the theorem are fulfilled the following
equality is valid:
[[K(p] (x,2 1) f (3x, w,1 w3) [[Kef] (2x, w1) ] =

C•) ]
2 3\ (2 1) f (t(
= fjJo ( w) cro x,
X, 2 X
(2
(l) X,
+xU, c~)), ~. !)+h2R 2 (h),

where II R 2 (h) II ~c.
IV. GENERALIZED HAJIILTON-JACOBI
EQUATIONS
Notation. Let g (x, y), x ERn, y E Rm, be a Coo function. Then,

by gxy we denote the n X m matrix (a:; :Yi).
2
Let y (x) be a Coo function with values in Rm, defined on an open

subset of Rn. Then we shall use the notation oylox = Yx for the
Jacobi matrix: (~};i = ~~~· If m = n, then Dy/Dx denotes the
Jacobian deL (oy/ox).
Let u = (u 1 , . . . , un) and v = (v1 , . . . , vn) be two ordered
n-tuples and let I and K be two disjoint subsets of {1, 2, ... , n}.
Then {ur, VK} will denote the following subsequence {w;}; E r u K
of the finite sequence (u1 , v1 , u 2 , v2 , • • • , un, Vn):
u; for i E I
{
w; = V; for i E K.
For instance, if I = {3, 1} and K = {2}, then {u 1 , vK} =
= (u17 v 2 , u 3 ). We shall mostly deal with K = 7, where l =
= {1, ... , n}"'I, or K = 0· In the latter case we shall write
{ur, vK} = Ur·
One should bear in mind (see Introduction) that
dcf n
(u, v)= ~ u;v;.
i=1
Let a be a multi-index, say a = (a 1 , ••• , ak)· Then
def I I
ct ! = ct1 . . • . ct" ..
For two multi-indices a = (a1, ... , ak) and ~ = (~ 17 ••• , ~k),

we write a~~ if a;~~;, i = 1, ... , k. If x = (x17 ••• , xk}
and a is a multi-index, then we set
xa = x~i x~2 x~k.
23*
356 OPERATIONAL :\IETHODS
Now let 6 (x) and cp (x), x E R'\ be two smooth complex functions,
the matrix C = 86/8x being non-degenerate. Then we define the
formal derivative 8cp/86 as follows:
ar- tc-1 --;;;-·

acp - acp
Second-order formal derivatives are given by

n
a2cp =tc-1 a2cp c-1+ "5; (tc-~~) tc-j azsm c-t •
iJ£2 iJx2 LJ ax m iJx2
m=l
In particular,
iJ2cp iJ2cp
a~;i iJsj = iJsj a~i
for every smooth function cp.
Sec. 1. Hamilton-Jacobi Equations with Dissipation

The examples of the wave equation, the crystal oscillation equa-
tion and the Schri:idinger equation for an oscillator (see Sec. 8 of
Introduction) show that the problem of fmding asymptotic solutions
of these equations leads in "zero approximation" to a first-order
partial differential equation called the characteristic equation or
the Hamilton-Jacobi equation. The latter equation does not involve
the parameter in which the asymptotics for the initial problem
is to be found. Its solution S (x, t) is real. In case of the Schri:i-
dinger equation, for example, this function determines the phase
of oscillations and the solution can be expanded, after factoring
·Out exp { ~ S (x, t)}, into an asymptotic power series in the small
parameter h. For a number of examples (involving the "absorption"
phenomenon), however, the analogous function is complex, so
that exp [* J
S (x, t) tends rapidly to zero in the domain
def
S 2 = Im S (x, t) > 0
when h ~ 0. If the initial problem is to be solved modulo 0 (hN),
then it is clear that the value of S 1 def R e S (x, t) in the domain
S 2 (x, t) > 0 is inessential. Therefore, the asymptotic solutions of
this kind lead to the more complicated statement of the problem
for a first-order equation than those we dealt with in the above
examples.
Now state the problem.
CH.' IV. GENERALIZED HAIMILTON ·J A'COBI EQUATI1011.,S 357
Definition 1.1. Let f (x), x ERn, be a non-negative function and

g (x) be a smooth function. We shall write g = 0 1 (ha) if
I
alli
g
(x)\ e -f(x)
h =O
(
h
a--'1_1)
2
oxl
for a -I ;1 > 0, h-+ 0, 0-estimate being locally uniform in x.

Similarly we define o (ha).
Problem Let f ~ 0, a> 0. Then the inequalities
1.1.
Jgjexp (- ~)~cha and JgJ~c1jfY., where c1 =a-aeac, are equiva-
lent. Now let two real smooth functions f (p, P, x, t) and T(p, p, x, t)
satisfy the conditions
fp= -fp, fy;=fp, fpp= -fPP' (1.1)
T~o for p=O.

The problem is to find such two real smooth functions S 1 (x, t)
and S 2 (x, t) that
(1.2)
at
as 2 -t ~
1 ( as 1
as2
ox ' Tx' x,
(h)
= 082 '
t)
t E (0, T).
We shall call (1.2) a dissipative Hamilton-Jacobi system.
Note that if
where H = H (p, x, t) is a real function, then the system (1.2)

with the initial condition S 2 (x, t) = 0 reduces to the Hamilton-
Jacobi equation
rJSt
iJt
+H ( rJSt
OX ' X,
t) =O •
The Hamilton-Jacobi equation is characterized by the Hamilton

function H (p, x, t). We shall transform the dissipative Hamilton-
Jacobi system Lo such a form which will be also characterized by
a single function QJ8(p, x, t), this function being complex.
First we prove two lemmas, which will be used frequently in the
treatise.
Lemma 1.1. (The Garding inequality.) Let F (x) E C2 (Rn) be
a non-negative function, then
j ~~ (x) j2 ~cF (x) · iuf IF x 1h (x) j, c = const.

358 OPERATIONAL METHOHS
Proof. We have
... ,
:::;:;,F(xi, ... , Xn)+tFxk(xi, ... , Xn)+
t2
+2sup I Fxkxkl'
X
Since the non-negativity condition for a quadratic trinomial

consists in the non-positivity of its discriminant, we obtain
I ::k 1
2
:::;;;,2F S~p I Fxkxk 1,
which proves the lemma.
Corollary 1.1. Let F E C""(Rn) be a non-negative function and let
P N be a homogeneous polynomial of degree N in oF lox with coefficients
smoothly dependent on x. Then P N = OF (hN/ 2 ).
Proof. In view of locality of Orestimates we may assume without
loss of generality that F E C'; (Rn). Then, by Lemma 1.1 we have
I F x I : :; :;, const ·F112 •
Hence, for I l I < N,
all I N-1 l I
- PN (x):::;;;,const [F (x}] 2 •
axl
According to the result of Problem 1.1 this proves the required

assertion.
Lemma 1.2. (A representation of the remainder term in Taylor's
formula.) If f E C""(Rn), then
n
i, j, 1<=1
where giJk E C""(Rn).

The proof of this lemma follows from the identity
1 n 1
f (x) = f (0) + Jdfd~x) dt = f (0) + ~ xi ) a~~ix) dt. (1.3)
0 i=1 0
Note. There is a similar representation of the remainder term

in Taylor's formula containing any number of terms.
Now we shall turn directly to transforming the system (1.2).
CH. IV. GENERALIZED HA/MILTON-JACOBI EQUATIIONIS 359
Set
~
r(p, p, x. t)=f(p, p, x, t)-f(p, 0, x, t)-

~ 1 ~ ~
-(p, /p(p, o, x, t)>-z<P, tr;r;(P, o, x, t)p),
r(p, iJ, x, t)=f(p, p, x, t)-t(p, o, x, t)-

~ ~ 1 ~ ~ ~
-(p, /p(p, 0, x, t))--z(p, /pp(p, 0, x, t)p).
Lemma 1.2 and Corollary 1.1 imply that

as2 , x, t ) = 0 s 2 (ha12) ,
as 1 , --;;;;-
r (----a;-
'"'(
r --;;;;-, as2 x, t ) = Os2 (h a12 ).
as, --;;;;-,
Therefore (1.2) is equivalent to the following system:
a~ I + f ( aa~l • 0, x, t) + <: 2 ' fr; ( aJxl • 0, x, t)) +

+ ~ ( aJx2 , fr;r; ( aJx1 , 0, x, t) aa~2 ) = os2(h), (1.4)
a~2 + f ( aJxl , 0, t) + <aJx fp ( aJxl , 0, t)) +

X, 2 , X,
+ ~ ( aJx fr;r; { aJx 0, x, t) aa~2 ) = (h).

2 , 1 , Os2
Set
H (p, x, t) = f (p, 0, x, t), H (p, x, t) = f (p, 0, x, t),
JJ8 (p, x, t) = H (p, x, t) + iff (p, x, t).
Now we multiply the second equation of the system by i and add

it to the first one. Making use of (1.1) we arrive at
FQ7f][S,+iS2]det a(SttiS2) +JJ8(aJxt' x, t)+
+ i ( JJ8 P( a8~1 , t) , a:: ) -

x,
as2
-21 ( ox '
no
QIVPP (as, ) as2)
a;;-' x, t --;;;;- = os2 (h). (1.5)
It is easily seen that each pair of functions satisfying (1.5) satis-

fies (1.4) as well.
The relation (1.5) will be called the Hamilton-Jacobi equation
with dissipation associated with the Hamilton function cJJe(p, q, t).
Sec. 2. The Lagrangean Manifold with a Complex Germ

In this section we shall develop a geometric theory which will
be used for solving Hamilton-Jacobi equations with dissipation.
Consider the 2n-dimensional Euclidean space Rn X Rn with
coordinates p 1 , . . . , Pn 1 q11 • • • , qn. This space will be called
the phase space.
Definition 2.1. An n-dimensional submanifold An of the phase
space is called a Lagrangean manifold if it satisfies the conditions
det ( op oq ) ( op oq )
[p, q]jk = Daj ' oak - oak ' oaj = 0 · j, k= 1, ... , n,
(2.1)
where a = ( a 11 • • • , an) are local coordinates on An and q = q (a),
p = p (a) are local equations of An. The expressions [p, q]u are
called the Lagrange brackets for the vector functions p (a) and q (a).
(2.1) implies that the form pdq on An is locally the differential of
J
a function. Therefore p dq, where l is any path on An, does not
l
change under continuous deformation of l with fixed initial and
final points. If it is additionally assumed that
J
£p
dq = 0 for each
closed path on An, then there will be a function s on An satisfying
the equation ds = p dq. We shall call such a function an s-action
on An.
Let D (a) be a given non-negative Coo function on An, called the
dissipation. We shall denote by r the set of all its zeroes.
Definition 2.2. A pair rn = (w (a), z (a)) of smooth n-dimensional
complex vector functions on An is called a complex germ, if the following
conditions (the complex germ axioms) are satisfied:
g 1) z(a) = On(h 112 ), w(a)=On(h 112 )
g 2)
g 3) there exists a smooth function E (a) = 0 n (h) on A'\ called

a potential, such that
dE=(w, dq)--(z, dp)+<w, dz>;<z, dw)+(f, da),
where f = 0 n (h).
CH. IV. GENERALIZED HAIMILT'ON-JAOOBI EQUATJ10NIS 361
Note 2.1. We shall consider a Lagrangean manifold with a com-

plex germ (or simply germ) to be given if all the four objects: An,
rn, D and E are given.
Note 2.2. If there is an s-action on An, then the existence of a po-
tential on An is equivalent to that of a function W(a) on An, satis-
fying the condition
dW = (p + w, d (q + z)) +0 D (h).
This function will be called an enthalpy of the complex germ. The
potential and the enthalpy of the same complex germ are related
by the formula
<
W=E+s+ P+ 21 w, z) +On(h 3/2 )+const.
Let H (p, q, t) E C""(R2n+l) be a real function. Denote by p (a, t),

q (a, t) the solution of the Hamiltonian system
(2.2)
satisfying the initial condition p (a, 0) = p (a), q (a, 0) = q (a),

where (p (a), q (a)) E An. The equations
p = p (a, t), q = q (a, t)
determine for a fixed t a Lagrangean manifold which we shall denote
by A~.
Proh Iem 2 .1. Verify that (2 .1) is satisfied on A~.
Let p (p 0 , q0 , t), q (p 0 , q0 , t) be the solution of the system (2.2)
with the initial condition p (p 0 , q0 , 0) = p 0 , q (p 0 , q0 , 0) = q0 • The
mapping
(Po, qo) ~ (p (Po, qo, t), q (po, qo, t)) dcf gh (Po, qo)
will be called the canonical transformation of the phase space asso-
ciated with the Hamiltonian function H.
There is an (n + 1 )-dimensional submanifold (with a border)
M 2n+l c R 2n X R 1 naturally associated with the family {A?}.
As a rule we shall identify this family with Mn+I.
Let a be local coordinates in A~, then (a, t) are local coordinates
in Jllr+l. It is these ones that will be used constantly. We shall
denote by (a, t) the corresponding point of Mn+1 as well, and \Ve
hope there will be no confusion in using the same symbols.
Problem 2.2. Suppose that there is an s-action on the Lagrangean
manifold
A~: {p = p (a), q = q (a)}
and let A1 = g:kA~ be given by p = p (a, t), q = q (a, t). Then

the function s (a, t) defined by
t
s(a, t)=s 0 (a)+ .\ [(p(a, -r), Hp(p(a, -r), q(a, -r), -r))-
o
-H (p (a, -r), q (a, -r), -r)] d-r (2.3)
is an s-action on A~.
Note that s-action on a Lagrangean manifold is not unique but
defined only up to an additive constant. When dealing with the
family A?= gkA~ of Lagrangean manifods we always assume the
s-actions on these manifolds to be co-ordinated by (2.3).
Let H (p, q, t) and if (p, q, t) be real Coo functions and IJ.B =
= H + iii. We shall correlate with rJI8 the following t-dependent
transformation of R 2 n X C2n:
(Po• qo, Wo, Zo)--+ (p, q, w, z),
where (p, q) = g}I (p 0 , q0 ) and the functions w = w (p 0 • q0 , w 0 ,z0 , t)
and z = z (p 0 , q0 , u· 0 , z0 , t) satisfy the system
dw .~
dt = - lH q- JJ8 qqZ- QJ8 qpW,
(2.4)
~~ =iff P +JJ8 pqz+ JJ8 ppW
with the initial condition w lt=o = w 0 , z lt=o = z0 • Denote by
dg}I+iH the mapping (p 0 , q0 , w 0 , z 0 )--+ (w, z); allowing for a slight
inaccuracy, we shall write (w, z) = dgk+ ;li (w 0 , z0 ).
Definition 2.3. The transformation (g}r, dg}I+ tE) will be called the
complex canonical transformation associated with the complex Hamil-
tonian function JJ8 = H +
iii. This transformation will be said to be
dissipative if the condition if·~ 0 is satisfied.
In what follows all the canonical transformations under conside-
ration are assumed to be dissipative.
Note 2.3. Let H = 0. Then dg}I+iH = dg:k is the differential of
the mapping g}Io (2.3) being the variation system for (2.2).
Let
(p (a, t), q (a, t)) = g},- (p 0 (a), q0 (a)),
(w (a, t), z (a, t)) = dgii+tH (w 0 (a), z 0 (a)),
and let F (p, q, w, z, t) be a function. The integral
t
~ F(p(a, -r), q(a, -r), u•(a, -r), -r)d-r
0
CH. IV. GENERALIZED HAIMILT'ON-JAGOBI EQUATI'ONlS 363
will be called an integral along the &'8-trajectory starting from.

(a, t) t
a, J
(a, 0)
Fdt or JFdt in symbol.
0
We shall use also the notation dF/dt or F for the derivative of the
composite function
F (p (a, t), q (a, t), w (a, t), z (a, t), t)
with respect to t.
Theorem 2.1. Let (A~, r~) be a Lagrangean manifold with a complex
germ and let D 0 and E 0 be the dissipation and the potential of this
germ. Let us define a dissipation D (a, t) on A~ = gj1 A~ by
t
D(a, t)=D0 (a)- JHdt.
0
Then
rn (a, t) = dgii+ilir~ (a)
is a complex germ on A~ and
t ~ ~
E (a, t) = E 0 (a)- i J (E + (w, H p);-(z, Hq)) dt

0
is a potential on A~.
In order to prove this theorem we need the following three lemmas.
J~emma 2.1. Let u (y)., 0 ~ y ~ a, be an m-dimensional real vector
C1-function, and let u satisfy the following inequality:
I n' (y) I ~ c [f (y) + I u (y) 11,
where f is a continuous non-negative function and c = const. Then
where c1 = c1 (c, a) is smooth in c and a.

Proof. We have
y
I u (y) I:(; I u (O) I+ c JU (£)+I u (£) ll d£.

0
Hence
y
Ju(y)J~c0 (F(y)+ ~ Ju(S)Jds),

where
y
co= 1 + c, F (y) = 1u (O) I+ j f (s) d£.

0
Let I be the operator defined by

y
If (y) = Jt (s) d£.

0 .
Then we have
I u I :::;;; c 0 (F + I I u I):::;;; c 0 (F + c0 IF + c 0I 2 Iu 1).
It follows by induction that
lui::;;; (]1 c~r- 1 F) +C~INiul.

The identity
11
Inf(Y)= (n~i)! j (y-£)n- 1 f(s)ds
0
implies the inequality
II"f (y) I~ Y~ max If (s) I:::;; a~, max If (s) j.

11 • 0"'£"'11 n. 0"'£"'11
The function F being monotonic, we obtain the inequality

N
C"an-1 cNaN
lu (y) I~ ~
n=l
(n"-1)! F (y) + --N! o~~:a Iu (£)I.
-.....:~'""
Passing to the limit for N-+ oo finally gives

I U (y) I :::;;; c0eco" F (y),
Lemma 2.2. Let p (a, t), q (a, t), w (a, t), z (a, t) be a solution
of the system (2.2), (2.4) smoothly dependent on a parameter a E R2",
an d l et J = D(p+w,Da
q+z) Then
•
[)J
at= ((p, u:) + (lp, z),
where q> = cp (a, t) and 1jJ = 'P (a, t) are smooth vector functions.
CH. IV. GENERALIZED HAMILTON-JACOBI EQUATrONS 365
Proof. We have
P+~= -QJllq-0JllqpW-JJ8qqZ,}
(2.5)
q+~ = QJ.f] p+ QJC ppw+JJB pqZ.
Differentiation of these equations with respect to a shows that
n
+ 2j (z jA + w jB<j>), (2.6)
j=i
(q +;)a= JJ8 pq (qa + Za) + dJC pp (Pa + Wa) +

n
+ 2J (zjcU> + WjD(j)),
j=l
where A <h B<h C and D<h are smooth matrix functions of a
and t. ' '
By the rule of differentiation of a determinant
2n
fjJ- ~ J
at - LJ h•
h=!
where J h is the determinant obtained by differentiating the kth

row of J. It follows from (2.6) that
-Jd!Cqhph+(lllh, z)+(q;k, w), k~n
Jk= {
JQJ.Cphqh+Nh, z)+(q;h, w), k>n,
where lflk, 1JJk are smooth functions of a and t. Thus,
2n
fjJ ...._.,
ift= L (('lJ,, z)+(IPk• w)),
h=i

As a corollary of Lemma 2.2 we obtain the following classical
result:
Theorem (Liouville). Let g}I be the canonical transformation of the
phase space associated with a Hamiltonian function H and let Q 0 be a
measurable subset of the phase space. Set Q 1 = gkQ 0 • Then
J
Qt
dqi · · · dqn dpi · · · dpn = ~
Qo
dqi · · · dqn dpi · · · dpn.
To prove this theorem it suffices to put ii = 0, w = 0, z = 0,

a = (p (0), q (0)) in Lemma 2.2.
Lemma 2.3. Let

A~:
{p =Po (a), q = q0 (a)},
'\'~:{w=w 0 (a), z=z 0 (a)}
be a Lagrangean manifold with a complex germ, let p (a, t), q (a, t),
w (a, t), z (a, t) be the solution of the system (2.2), (2.4) satisfying
the initial condition p (a, 0) =Po (a), q (a, 0) = q0 (a), w (a, 0) =
= w 0 (a), z (a, 0) = z 0 (a) and let a 0 E A~. If
a(Po+wo))
(
rank a (q::zo) = n
arx ' a=ao
and w (a 0 , t) = z (a 0 , t) = 0 for 0 ~ t ~ T, then
a(p+w))
rank
(
a (::z)
arx a=ao
= n.
t=T
Proof. Let ~ E Rn. In a neighbourhood of the point a = a 0 ,
~ = 0 in R 2n we set
Po (a, ~) =Po (a), qo (a, ~) = qo (a), }
w 0 (a, ~) = w 0 (a) + A~, (2.7)
z0 (a, ~) = z 0 (a) + B~,
where A and B are n X n matrices. Obviously, we can choose A
and B so that the matrix
a (Po+wo, %+zo)
a (rx, fl)
is non-singular at (a 0 , 0).
Let p (a, ~. t), q (a, ~. t), w (a, ~. t), z (a, ~. t) be the solution
of (2.2), (2.4) satisfying the initial condition (2.7). By applying
Lemma 2.2 with a = (a, ~), we have
D (p+w, q+z) I = D (Po+wo, qo+zo) =I= O I
D (o:, fl) la=ao D (rx, fl) a=ao
B=O B=O
for 0 ~ t ~ T, which proves the lemma.
Proof of Theorem 2.1. I. We first verify the validity of g 1). By
Lemma 2.1,
lw(a, t)J+Iz(a, t)J~
t
~c(a, t) (iz0 (a)J+Jw0 (a)l+) (jHpi+JHql)dt),
0
where c E Coo.
CH. IV. GENERALIZED HAI:IIILT'ON-JAGOBI EQUATIONIS 367
Applying Lemma 1.1 now and the Schwartz inequality we have

t t
J(iHPI+IHq\)dt~c!(a, t) I VI HI dt~
0 0
~c!(a, t)vtUIHidtr 12 , C1EC"";

0
it follows that there exists such a smooth function c2 (a, t) that

t
Iw I + I z I~ c2 [ D6 + ( ~ I .H I dt) ] ~
12
~ V-2 c2 (D 0 - jf H- dt ) 1/2 = v-2 c2 D

1/2
.
() I
II. To verify the validity of g 3) we consider the following func-

tion:
1 1
fi(a, t)=Ea..-(w,
"1
qa.)--
1 2 (z, Wa.)+-
1 2 (w, Za.).
1
It follows from the systems (2.2), (2.4), the equation

; . -- i ,. ._, i ,..._,
E= - l l l - 2 (Hp, w)- 2 (Hq, z)
and the expressions of qa, Pa,

(va, ~a• Ea being obtained by differen-
tiating these equations with respect to a that
fi (a, t) = (u•, aplw) + (w, a~lz) + (z, a~lz),
where aY; 2, 3 are smooth matrix functions of t and a. Thus,
~ ii(a, -r)d-rl~c(a,
0
t) i 0
(lw(a, -r)l 2 +lz(a, -r)i2)d-r,
t • t
I of;~:·
0
1:) d-r ~c(a, t) I (iw(a, -r)l+iz(a,
0
-r)i)d-r,
where c (a, t) is a smooth function. Since D (a, t) is monotonic

in t and
z =On Wl2), w =On (h112),
we have
t
Jji(a, -r)d-r=Ov(h).
OPERATIONAL METHODS
It remains to note that

fi (a, 0) = Ov 0 (h),
for E 0 is a potential on A~ and it follows from

D0 (a)~ D (a, t)
that
t
fi(a, t)=fi(a, 0)+ ~ ii(a, T)d't=On(h).
0
The reader is offered to verify that E = 0 n (h) holds.

III. If D (a 0 , t) = 0, then D (a 0 , 1:) = 0 for 0 ~ 't ~ t, and
therefore z ( a 0 , 1:) = w ( a 0 , 1:) = 0 for 0 ::::;;; T <
t. Thus g 2) is
valid by Lemma 2.3. The theorem is proved.
In what follows we shall always suppose that if a family of Lag-
rangean manifolds Af with complex germs rt' is obtained from (A~, r~)
by a complex canonical transformation, then the dissipation D (a, t)
and the potential E (a, t) on At' are correlated with D 0 (a) and E 0 (a)
by the formulas of Theorem 2.1.
Example 2.1. Let JJ8 1 = H 1 = ~ (pJ + qi) and let (A~, y~) be
given by
P =Po (a), q = qo (a),
w = w0 (a), z = z0 (a).
Then (Af, rt') is given by

p 1 (a, t) = p 10 (a), PI. (a, t) = py0 (a) cost- qio (a) sin t,
q1 (a, t) = q10 (a), q1 (a, t) = q10 (a) cost+ p10 (a) sin t,
w1 (a, t) = w10 (a), wy(a, t) = wy0 (a) cos t-z10 (a) sin t,
z1 (a, t)=z 10 (a), z1 (a, t)=zy 0 (a)cost+u~10 (a)sint,
D (a, t) = D0 (a),
E (a, t) = E 0 (a),
. s(a, t)=s 0 (a)+! [p~0 -q~0 ]sin2t+ ~ (q 10 , py 0)(cos2t-1),
where D 0 (a), E 0 (a) and s 0 (a) are the dissipation, the potential
and the s-action on A~, respectively.
CH. IV. GENERALIZED HAI.iVliLTON-JAQOBI EQUATrONIS 369
In particular, we obtain for t = +n/2:

A:.:!. : {PI = P1o (a), Py = ±qy0 (a),
-~
qr=qio(a), qy=+Py0 (a)},

rn :rt :{Wr=Wr 0 (a), Wy=+Zy0 (a),
±:r
z1 =z 10 (a), z1 =+wy0 (a)}.
Note Lhat
(P (a, + ~ }, q (a, + ~ } ) = (Pio (a), q1 0 (a))-
-(P!o(a), qyo(a))=(Po(a), qo(a))J
which makes the notation (u, v)J reasonable.

Definition 2.4. Let I be a subset of {1, ... , n} and let An:
{p = p (a), q = q (a)}, rn = (w (a), z (a)) be a Lagrangean mani-
fold with a complex germ. The subset of An defined by
(2.8)
will be called the Ith zone of (An, rn), Q 1 in symbol. The zone Q 10
with I 0 = {1, 2, ... , n} will be called non-singular. By dealing
with the non-singular zone we omit the index I 0 : for example, we
write Q in place of Q 10 , E in place of E 10 and so on.
It is obvious that Q 1 is an open subset of An. We shall show that
the axioms g 2) and g 3) imply that the family of zones {Q 1 } with I
running through all subsets of {1, 2, ... , n} covers r.
Lemma 2.4. Let u (a) and v (a) be smooth en-valued functions
defined in a neighborhood of a point a 0 E Rn, and let the following
conditions be satisfied for a = a 0 :
(i) [u, v]u=O, i, j=1, ... , n;
(H) <ank ( ~: ) "u,

Do.
Then there exists an I c {1, ... , n} such that the matrix

iJ(u 1 , V]}
iJa
is non-degenerate at a = a 0 •
21-01225
Proof. Denote by A the matrix ~~ (a 0 ) and let rank A = k. If

k = n, then the statement of the lemma is valid. Let k < n. We
may assume without loss of generality that the first k rows of A
are linearly independent. Let C be the matrix composed of the
first k rows of A and the last n - k rows of the matrix B = ~~ (a 0 ).
We shall show that rank C = n, which means that the lemma is
true. Suppose that this were not the case, i.e., rank C = m < n.
Then rank ( ~ ) = m. Let ~ E en and a = M~, where M is a com-
plex non-degenerate matrix. Set
n
a -v M a 1, = au ( )
"8if Cto ' 8 , = i5(f
r!v ( )
~= L..J ii aa· ' J . Cto '
l i=i J
and let C' be the matrix composed of the first k rows of A' and the
last n- k rows of B'. It is obvious that rank ( ~: ) = n and rank
( ~; ) = m. Now me choose M so that the last n - m columns of the
matrix ( ~;) vanish and det (Aii)~, i=i does not vanish. The condi-
tion [u, vl,ii = 0 implies that
< au au )
aj3, ' aBn -
<aBnau ' k
au ) - '\' aur avr -0
aj3; - ..:....J aB; aBn - .
r=i
.
Smce det ( au
ul/JJ· ) 1,J-1
k •
.. _ =I= 0, 1t follows that the last column of ( A'
B'
)
I
vanishes, contradicting the condition rank ( ~: ) = n. The lemma

is proved.
Corollary. rE u QI.
IC{1,2, ... ,n}
In fact, set u = q +
z, v = p.+ wand let a E r. Then tl.te con-
dition (ii) of the lemma is identical with the axiom g 2) of the com-
plex germ, and the condition (i) follows from the axiom g 3).
In what follows we shall consider only a little neighborhood of
r in An, therefore we can always suppose that the set of all zones
covers r.
Definition 2.5. The function
1
~t(et)= 2 (z(a), (IE(a)z(a)),
CH. IV. GENERALIZED HAMILTON-JACOBI EQUATIONS 371
where
1/, = BC-1, B = a(~:w) ' C -- a (q+z)
- aa '
will be called the [t-action in the non-singular zone. The phase of
germ in the non-singular zone will be defined by
1
(J)(a)=- 2 (w(a), z(a))+~t(a)+s(a)+E(a).
Note that
a (p+w)
Wo= a(q+z) '
where a/a (q + z) is the operator of formal differentiation with

respect to q + z. Since an enthalpy W (see Note 2.2) satisfies the
condition
aw
P+w= D(q+z) +On(h),
we must have
D2W 1/2
1/, = 0 (q+z)Z +On (h ).
An immediate consequence is that the matrix 1/, is "almost sym-

metric":
t
1/,-1/,=0n(hi/2).
Now we introduce the s-action, the [t-action and the phase in the
zone Q1 • To this end, consider the canonical transformation
(g'%~, dg1/:) associated with the Hamiltonian function H 1 introdu-
ced in Example 2.1. Note that the image of the zone Q 1 under g~~
lies in the non-singular zone of g1/: An because
Definition 2.6. We define the s-action, the [t-action and the phase
in the zone Q 1 by the formulas
si(a) =S (a, ~) ,
[LJ{CG) = f1 ( CG, ~ ) ,
CDr (a)= Q) (a, ~ ),
24*
where s {a, },
~ f.t ( a, ~ } and <D (a, ~ } are the s-action, the
f.t-action and the phase in the non-singular zone of gfjZAn, respecti-
r
vely.
The result of Example 2.1 leads to the explicit formulas
sr(a)=s(a)-(Pz(a), qy(a)),
1
f.lr (a)= 2 ({zr, wy}, ~r {z1 , wy}),
1
<Dr= -2(w, z)r+f.tr+sr+E,
where
a{qr+zi> Py+wl}
Cr= aa
Note that any s-action s (a, t) on a family of Lagrangean mani-
folds Af = gi=,An satisfies the equation
ds 1 = (p 1 , dqr)- (qy, dpy) -H dt,
which is transformed into the Hamilton-Jacobi equation
as ( as as )
at+H axr' ~I' Xr, - a£- ' t =0
I
under the change of variables

(a, t)-+ (xr, ~I' t), }
x 1 =qr(a, t), ~I=Pz(a, t), (2.9)
S(xr. ~1 , t)=s 1 (a, t).
We shall see below that if Im <D 1 ~ 0, then <D 1 (a, t) satisfies an
equation which becomes a Hamilton-Jacobi equation with dissipa-
tion under the change (2.9).
Sec. 3. 'l' -Atlases and the Dissipativity Inequality

In this section we introduce a special sort of atlases of An which
makes it possible to formulate a condition for a complex germ,
called the dissipativity condition, essential in solving Hamilton-
Jacobi equations with dissipation.
Lemma 3.1. Let a 0 E Q 1 • Then there exists a real n-dimensional ve-
ctor function y (a) such that each component of it is one of that of the
CH. IV. GENERALIZED HAIMIL'I'ON-JAGOBI EQUATI'ONIS 373
(2n+1)-dimensional vector
(0, Rezi(a), Imzr(a), Rew7(a), Imw1 (a)),
and
The following fact of linear algebra will be needed to prove this

lemma.
Lemma 3.2. Let u 1 , ••• , um, v 11 ••• , Vm be elements of a real
vector space V. Then
u1 + iv1 , ••• , Um + iVm
are linearly independent in the complexification CV of V if and only if
u1 EB v11 ••• , Um EB Vm, ( -v1 ) EB U11 ••• , ( -vm) EB Um
are linearly independent in V EB V.
Proof. Let ail bi be real numbers. Then
(ai = bi = 0, all j) <=> (2J (ai + ibi) (ui + iv;) = 0) ~·
i
<:=> (~ (ajUj- bjVj) = ~ (ajVj
j j
+ bjVj) = 0) <:=>
<=> (2J ai (ui EB v;) + 2J bi ((-vi) EB ui) = 0 ).
j j
Proof of Lemma 3.1. Considering the canonical transformation

gf/I2 of Example 2.1 shows that Qr may be assumed without loss
of generality to be non-singular. Under this assumption we have
det C (a 0 ) =1= 0. Fix a = a 0 • Let rank (aq/aa) = k; we may, of
course, assume that it is the first k rows of i)q/i)a which are linearly
independent. Consider the matrix
_ ( a (q~aRe z) ·)
M- iJimz ·
-;ra
By Lemma 3.2 non-degeneracy of C implies that
iJ(q+Rez)
( aa ----a;;,-
iJlmz )
det iJ Im z a(q+ Re z) =I= O,
---aa aa
hence rank M = n. Choose lk+ 1 , ••• , ln so that the matrix M 1,
obtained from i)q/aa by replacing its last n - k rows by those of M,
is non-degenerate. Set
0 for 1~j~k,
"?i = { Re z 1i for j > k, li~n,
Imz 1i for j>k, li>n,
then fJ (q + y)JfJa is non-degenerate. In fact,
a (q+v> =
aa.
(1A 0)1 M b
where A is an (n - k) X k matrix. The lemma is proved.

Corollary. There exists such a neighbourhood uy j a 0 that the res-
triction n:~ of the mapping
a-+ {qi (a) + "?1 (a), Pi (a) + "?i (a)}
to ui is a diffeomorphism into R".
Definition 3.1. we say that Uy is a y-domain, n;~ is a y-diffeomorphism
and (uv, n:~) is a y-patch of A". Sometimes we shall use the term
"y-patch" as a synonym of "y-domain".
Any set of y-patches covering A" will be called a y-atlas of A".
If y (a) = Fv (a), where F is a matrix and
v = (Re z1 , Im z1 , Re w1, Im w1),
then the corresponding y-patch of the zone Q 1 is said to be of type (I, F).
A patch of type (/, 0) is said to be non-singular.
Lemma 3.3. Let (u-, n:~) and (u=, n:t:) be y-patchesof the zone Q 1 •
y y y y
Then cr = (n:~)- 1 on:~ has the following asymptotic expansion:
y y
cr (a)= a+ cr<t> (a)+ cr< 2 >(a)+ Ov (h 312 ),

where
cr< 1>(a)= A- 1 (y-y),
Proof. Assume for simplicity (which does not affect the genera-
lity) that Q 1 is the non-singular zone. Then cr (a) satisfies the equa-
CH. IV. GE?i:ERALIZED HAI.MIUI'ON-JAJGOBI EQUATJ10NIS 375
tion
q (a (a)) + y (a (a)) = q (a) + y (a). (3.1)
Set
a (a) = a + a< 1> (a) + a< 2> (a) + y (a).
Substitution of the expansion of q (a (a)) + y (a (a)) in powers of
a< 1> +
a< 2 > +
r into (3.1) leads to
(qo:+Ya)(a< 1>+a< 2>+r)={ ~ a;d~t_~> (a(1)+a< 2 >+r); (a< 1>+

• • z J
t, J
+ a< 2 > + r) i + F 3 ( a(i) + a< 2 >+ r) = y(a)- y(a),

where F3 is a cubic form smoothly depending on a. Since a< 1>
and a< 2 > satisfy the equations
(qa + Ya) a·t> (a)= y (a) -y (a),
(q +y) a< 2> (a)+_!_~ 82 (q+y) a11lajl>,
a CG 2 ~. 8a.; 8U..j
1, J
and
a<t> =On (h 112 )
one has
"l/2
A 1r=On(h')
with A 1lr = Ar, so det A 1 =1= 0 in a neighborhood of each a 0 Er.
This proves the lemma.
Corollary. Oma>-estimate is equivalent to 0 ma<a>>-estimate.
Proof. Set D(a)=D(a(a)). Since a(a)-a=On(h 112 ), the
expansion
D(a) = D(a)+(Da, (a(a)-a))+(a(a)-a, f(a)(a(a)-a)),
where f (a) is a smooth vector func.tion, implies that D=On (h).
The same argument with a replaced by a-t shows that D=015(h).
Definition 3.2. The function
<D1=1Dr+({Pr, -q!}+Yl 1 , 1')+-}(y, ~jy),
where
111 ={wr, -zy}-~Hzr, Wr},
~~= ~ (~r+ 1 ~r),

defined in a given y-patch of the zone Q 1 , will be called the. y-phase

of the complex germ in this y-patch.
Lemma 3.4. Let (u.,., ni) and (u-, n~) be two y-patches of the
1' 1' 1'
zone Q1 • Then
rr.V rr.'l' ( I -1 o Jt;;,+
WJ = 'VJ o Jt'l')
I 0 n (h3/2)
whenever (nit 1 on!. is defined.
1' 1'
Note 3.1. (n 1 )-1 on.!. is obviously defined in a neighbourhood of
1' 1'
u.,. n u;; n r.
Note 3.2. Lemma 3.4 shows that if the system
qi (cr (a)) = qi (a) +'VI (a},
p 1 (a (a)) = Py(a) Yi (a) +
has a solution cr (a), then modulo On (h312), the y-phase <D1 (a) is
obtained from the phase <D 1 (a) by a change of variables: <D1 (a),....,
"""'<1> 1 (cr (a)). Note, however, that Definition 3.2 defines <Di on
u.,. whether cr (a) exists or not.
Proof of Lemma 3.4. We may assume without loss of generality
that Q 1 is non-singular. Set cr = n;,;Ion;;. Expanding <D" (cr (a))
in powers of (cr (a) - a) and using Lemma 3;3 we get
def - -
R (a)= <D'~' (cr (a)) -(D" (a)= <D'~' (a)- CD'~' (a)+ ((D~, a< 1l (a)+
+ cr< 2> (a))+-} (<D~a(:; (a), cr< 1l (a))+ On (ha 12 ). (3.2)
To calculate the derivatives fJ<D"!fJa and fJ 2<D"!fJa 2 it is convenient
to express <D'~' in terms of an enthalpy of the complex germ:
<D'~'=W-<p+w- ~ ~ (z-y), z-y)+On(h 312 ).
By the definition of an enthalpy
ow tc o(q+z)
a;:-= ow =
tc
(P+ w) + 0 n (h ),
so
CD~= tc (p + w)- ts (z-y) + ~ t(z- 'V)a ~ (z-y)-
_t(z-y)a(P+w- ~ ~(z-y))+On(h)=
= tA (p+ w)- tc~ (z- y) + t (z- y)u ~ (z-y) +
+On (h)= tA [p+ w- ~ (z-'\')1 +On (h), (3.3)
CH. IV. GENERALIZED HA2V!ILTON-JACOBI EQUATrONIS 377
where A = a (q +
-y)/oa. Differentiating once again and using the
identity
1 - C- 1 (z - '?)a = 1 - C- 1 (q z)a +
C-l (q + +
'?)a = C- 1A
we obtain
az<D'V
8a.2
= ~ (
L.J
az (q+v)
8a.; 8a.j '
P) + t A (B- ~ (z- -y) a H-0 D
(h 1/2) =
i, j
(3.4)
Substitution of (3.3), (3.4) and the formulas of Lemma 3.3 for a< 1 >
and a< 2 > into (3.2) shows that R = 0 D (h3 12 ), which proves the
lemma.
Definition 3.3. Let uy be a -y-domain of the zone Q 1 of a complex
germ. We say that the germ is dissipative in uy, if there exist smooth
functions e (a) > 0 and c (a) ~ 0 satisfying the condition
Im CD}+ cDa 12 ?::--eD, a Euy,
which will be called the dissipativity inequality.
Note 3.3. The condition of dissipativity for uy is equivalent
to the following one: each a E uy has a neighbourhood in which
a dissipativity inequality holds. This statement is a consequence
of the following fact:
Lemma 3.5. Let A be a submanifold of Rm, let {uB} be an open
covering of A (u13 c A), and let {c13} be a family of real numbers. Set
f (a)= inf c13 •
u 13 3a
Then there exists such a C"" function <p on A that <p ~f.
Proof. Any submanifold of Rm is a locally compact space with
a countable base of open sets. Therefore, there exists a sequence
{Kn} of compact subsets of An such that
Kn C Kn+t; UKn =An
n
(the sign o means the interior of a set). Since Kn is compact, it can

be covered 'by a finite subfamily of {u 13 }, which implies that there
exists such a dn ~ 0 that f (a)~ dn for a E Kn. Let An= Kn "Kn-I·
An is compact and
An c Kn+h Ann Kn-2 = 0.

It follows that there exists such a smooth function <pn on A that

0 ~ <pn ~ dn and
dn if aEAn,
(j)n (a) { a
0 if a~KnTI or aEKn-2·
00
The function <p (a)= 2j <pn (a) is easily seen to satisfy the require-
n=1
ment of the lemma.
Note 3.4. We can choose e and c in the dissipativity inequality
SO that C (a) = 0 in a neighbourhood of the set r Uy. n
Definition 3.2. A complex germ '\'n on a Lagrangean manifold An
with a fixed -v-atlas is called dissipative if it is dissipative in each -v-do-
main.
Note 3.5. Since s1 (a) is real and the dissipativity condition
does not involve the real part of the phase the concepL of dissipa-
tivity of a germ can be used also even if there exists no s-action
on An.
Theorem 3.1. The property of a complex germ to be dissipative is
independent of the choice of a -v-atlas on An.
The result of this theorem may be interpreted to mean that if the
n
inequality (3.4) holds in a neighbourhood u c u~ u~ of a point
a 0 E r, where u~ is a -v-domain of Q 1 and u~~ is a -v' -domain of
QK, then there exist a non-negative function c1 and a positive func-
tion e1 such that
v'
Im cDK = c1D 3/2 >e 1D.
For I = K, this result follows immediately from Lemma 3.4 and
the corollary to Lemma 3.3. In the general case, the theorem will
be proved in Sec. 5.
From now on the complex germ will be assumed to be dissipative,
the word "dissipative" being omitted.
Sec. 4. Solution of the Hamilton-Jacobi Equation with Dissipation

In this section we shall express the solution of the Hamilton-
Jacobi equation with dissipation in terms of that of the Hamilto-
nian system (2.2) and the system (2.4).
Let
A?: {p = p (a, t), q = q (a, t)}, 0 ~ t ~ T,
be the family of Lagrangean manifolds with complex germs
r;1 : {w = w (a, t), z = z (a, t)},
CH. IV. GENERALIZ~m HAMIL'IION-JAGOBI EQUATrONIS 379
obtained from (l'l~, r~) by the complex canonical transformation

associated with a Hamiltonian function aJ8 = H +
iff. Let Mn+1 be
the manifold associated with the family {A?}. We define a y-atlas
on M"+ 1 as follows. Let U.., be an open subset of M"+l such that
its intersection with any plane t = const is a y-domain on Af
(whenever it is non-empty), the type of this y-domain being inde-
pendent of t. Define a diffeomorphism
TI~: (a, t) ~ (x, t)
of U v into R"+ 1 by the formulas
xr = qr (a, t) + Yr (a, t), x 1 = Pr (a, t) + y1 (a, t).
The pair (U..,, ll~) will be called a y-patch of type (I, F) of the
manifold Jl/"+ 1 •
Definition 4 .1. We say that a family {Af, r?} is dissipative if for
each y-patch (U..,, II~) there exist smooth functions e > 0 and c ~ 0
such that
e(a, t)D(a, t)~Imc:D)'(a, t)+c(a, t)[D(a, t)] 312 , (a, t)EU..,.
The last relation will be called a dissipativity inequality in the patch

(U..,, II~).
Theorem 4.1. If a complex germ r~ on a Lagrangean manifold An
is dissipative, then the family {Af, r?}, where
at A"
n
.1\ t =,H o, rtn = (d gH+iH
t ) ro,
n
is dissipative.
In particular, the property of a germ to be dissipative is invariant
under complex canonical transformations.
The proof of this theorem will be given in Sec. 5.
Theorem 4.2. Let Mn+l be the manifold associated with the family
{A?}, and suppose that there is a y-atlas of M"+l which consists of
a single y-patch (U..,, II..,). Then the function
s= (<D" + icD 312 ) o IIv\

where c:D v is the y-phase of the germ and c is the same as in the dissi-
pativity inequality, satisfies the relation (1.5).
The following lemma will be needed to prove this theorem.
Lemma 4.1. There is a smooth function g (a, t) such that
~ 2/3
t max I JJ (p (a, -r), q (a, -r), -r) j~g (a, t) [D (a, t)] .
o~·~t
Proof. Omitting the argument a, set

I (t) = -il (p (a, t), q (a, t), t).
Set
a=a(t)= max I(T), a=I(T 0}, T0 =T0 (t).
O<:;,;<:;t
By the G~rding inequality,
If' (T) I~ g1 V t (T).
Set Vf=h, then lh'l~g 1 /2. It follows that
h (T) >-Va- g; IT- To I,
which implies
h (T)
v-2 a
>- for IT- T0 I~ _ _
a
v- ,
g1
that is,
I (T) >- 4a for I T- To I~ v~
g1
.
Consider the following two cases:
(1) if Va;g 1 ~t/2, then
a Va aa;z
D(a , t)~-·-=-
/ 4 g1 4g1 '
so
(2) if Va!g 1 > t/2, then D(a, t)>! ta, so

t max I (T)~SD (a, t)~c (a, t) [D (a, t)] 213 •
O<S't<St
Proof of Theorem 4.2. I. Here we shall calculate <D". To do this
rewrite <D" in the form
1 1
<D"=z(z-y, ~(z-y))+(p+w, y)- 2 (w, z)+s+E.
We have
• 1 • • •
<D" =z (z- y, ~ (z-y)) + (z- y, ~ (z-y)) +
• • • 1 •
+<p+w, v>+<p+w, y)-z(w, z)-
1 • • • . .
- 2 (w, z)+s+E+(z-y, Ov(h)). (4.1)
CH. IV. GENERALIZED HA!MILTON-JACOBI EQUATrONlS 381
First we calculate
i = iJc-t- sc-~cc-~.
It follows from (2.6) that
• 1/2
.8= -QJ8qqC-OJ8qpB+ Ov(h ),
C = OJ8 pqC + QJ8 ppB + Ov (h
• 1/2
),
therefore
~ =-rJJ.8qq-$8qp~-~aJ8 pq-~r!/8 pp~ +0n(h 1h)
Now recall that
• ~ i ~ ~
E= -iH- 2 [(w, Hp)+(z, Hq)],

S= -H +<P, Hp)
by the definitions of a potential and an s-action on A7. Set . £= . .

= w- 'G (z - 'V)· Substituting the above expressions of 'G, s, E
and those of p,
~ and ~ (see (2.2), (2.4)) into (4.1) and, using the
fact that 'G is "almost symmetric", we obtain
• 1 .
<D" = 2 (z- j', - rfJIJqq (z- 'V)- 2QJ8qp~ (z- 'V)-
- 'GQ/8 ppG (z- 'V)) + (~ (z-j'), ill P+ QJtJ pqz+QJ8 ppw) +
+ (j', p +G)- (j', Q/tJ q + dJtJ qqW + rfJ8 qqZ) +
1 ~
+2 (z, iHq+ 0Jtqqz+ OJ8qpw)-
.-
- 2 (w, lHp + QJtJ pqz +OJ8ppw) -QJe + (p, Hp)-
1
i - - - • 3/2
-2[(w, Hp)+(z, Hq)]+(iHp+'V, On(h))+On(h )=
= -Q/8-i(lip, £>-<"'' OJeq>-i-<"'' QJ{!qqj'>-

1
- (Q/8 pqj', ~)- 2 (~, Qc;.e ppS) + (p, H p) +
+(y, P+~>+ZiHp+y, On(h))+On(h 312 ).
A simple calculation which uRes Taylor's expansions of QJ8 (p, q +
+ j', t), QJ8 P (p, q + j', t) and QJ8 PP (p, q + j', t) in a neighbourhood
of (p, q, t) and the relation £=On (h 112 ) yields
<Dv = _ QJ8 (p, q + j', t)- (0Je 11 (p, q + j', t), ~)-
1 •
--2(£, QJ!!pp(p, q+'V, tH>+<Hp(p, q, t)+'V, P+~>+
- • 3/2
+(iHp+'V, On(h))+On(h ).
Since
• 1/2 • ~ 1/2
Rez=On(h ), Imz=Hp+On(h )
by (2.4), it follows from the definition of y (a, t) that
• ~ 3/2
(y, Ov(h))=(Hp, Ov(h))+Ov(h ). (4.2)
Thus,
m"= -JI8(p, q+v. t>-<OJBp(p, q+v. t), s>-
1
- 2 <£. OJB pp (p, q + y, t) £> +
• ~ 3j2
+(Hp(P, q, t)+y, P+s)+(Hp, On(h))+Ov(h ). (4.3)
II. Given a smooth function cp on M"+l, we shall denote by cp
the composite function cp 0 TI; 1 • Set also
<Di = Re cD", <D~ = Im <D", s=1 Re £, £2 = Im £,
~ ~., ~ ~- :Jj2
f :---- CDx- p- £, / 1 = Re /, / 2 = Im /, r = cD ,
where c is the same as in the dissipativity inequality. We have:
rx = 015 (h) by the G~rding inequality,
<D~x=On (h 112 ) by (3.3).
Therefore,
"' ~., .~ ~.,
FJf{) [S (x, t)] = <Dt + Lrt + OJ8 (<D1x, x, t) +
~., ~., ~
+ i(OJ8 p (<D1x, x, t), <D2x + rx)-

1 ~., §:J/i ~., ~., 0 3/2
- 2 (<D2x, QIV pp (<D1x, x, t) <D2x) + 15 (h ).
Expand QJ8, o'J8p and OJBPP in this formula in powers of 1+~;
since
~ 1/2 ~., ~ ~
£=015(h ), <D2x=~2+f2
and 7=On (h) by (3.3), we obtain
FJ!{)[S(x, t)J=d5i+irt+&78(P, x, t) +<rf!8p(j, x, t),£~+]';>+
1 ~ ~~ ~ ~ ~ ~
+2(0J8pp(p, x, t)Sb Si)+i(OJ8p(p, x, t), £2+f2+rx)+

. ~ ~~ t~ ~ ~
+L<&'t?pp(P, x, t)st. £2>- 2 <£2. OJBpv(P, x, tH2>+
+ 015 (h 3/2 ) = <Dt +&'8 (p, x, t) +
~., ~
~ ~ ~ 1 ~ ~
+(OJ8p(P, x, t), £+!>+2(£, OJ8pp(P, x, t)SH-
~ ~ ~ 3/2
+i(rt+(OJ8p(p, x, t), rx))+Ofi(h ).
CH. IV. GENERALIZED flAJ\IIL'l'ON-JACOBI EQUATJ!ONIS 383
To calculate <15", we use (4.3), (4.2) and (3.3):

~ ~
~y • y y • ~ ~ -
<Dt =<D - (<Dx, v+HP (p, q, t)) =
= - 618 (p, x, t)- (JJ8 p (p, x, t), [)-
1~ ~ ~ ~~ ':"~
-z-(~, r:fJtpp(p, x, t)S)+(Hp(p, q, t)-v, P+~>-
~v • ~ ~ ~ ~ ~ 3/2
-(<Dx, v+Hp(p, q, t))+(Hp(P, q, t), OJ5(h))+OJ5(h )=
= - QJt (p, x, t)- (Q/8 p (p, x, t), '[)-
1 ~ ~ - ~~ ~
--y(S, JJ.Bpp(p, x, t)S)-(Hp(p, q, t), f)+
~ - ~ 3/2
+(Hp(p, q, t), Of5(h))+Oy;(h ).
Now we note that
~ + (JJ.8 P (p, x, t), r_,J = 71 + (Hp (p, q, t), rx) +
~ ~
-· - ~ 3/2 • ~ •
+<H]J(p, q, t), OD(h))+Oy;(h )=r-(rx, V) +
- - - 3/2
+<Hp(Jl, q, t), 0 15 (h))+Oy;(h )=
~
=~+(Hp(p, q, t), Oy;(h))+Oy;(h 312 ).

The last three formulas give
• ~ ~ ~ 3/2
FQ16'[S(x, t)]=ir+(Hp(p, q, t), OJ5(h))+OJ5(h ). (4.4)
For t > 0, the right-hand member of (4.4) is o15 (h). In fact, it
follows from Lemma 4.1 that
;= _ ~ C'IJ112fi(r, x, t)+~IJ3!2= 015 <h7/6),
I Hp (P, q, t) l~cdx, t) V -H (p, q, t)~c2 (x, t) D113 .
It should be noted that Oy;-estimate implies Os2-estimate because
the germ is dissipative. The theorem is proved.
Now we turn to the Cauchy problem for the Hamilton-Jacobi
equation with dissipation. Let a compact set K 0 c: Rn be given,
and let S 0 = S10 + iS 20 be a smooth function with the non-nega-
tive imaginary part S 20 , defined in a neighbourhood of K 0 • Let
(A~, r~) be the Lagrangean manifold with the complex germ defined
by the formulas
as10 (a) . as2o (a)
q = a, P = aa • z = 0 ' w = ~ aa '
(it is easy to verify that all the axioms of a complex germ are satis-
fied). The whole of this manifold lies in the non-singular zone and
has a )'-atlas consisting of a single non-singular patch. Let 6Jr 0 denote
the inverse image of K 0 under :n: 0 :
?% 0 =:rt~ 1 K 0 cA~,
and let K 1 be the image of

def
?h':t = gk61Co c A~
under the canonical projection of the phase space onto the plane
p = 0.
Our explication of the Cauchy problem consists in fmding a smooth
function S (x, t) which satisfies (1.5) in a neighbourhood of the set
K = {(x, t) I0 < t < T, x E Kt}
and the initial condition S (x, 0) = S 0 (x) in a neighbourhood
of K 0 • The pair (K 0 , S 0 ) will be called Cauchy data for the Hamil-
ton-Jacobi equation with dissipation.
Proposition 4 .1. Given a complex Hamiltonian function, suppose
that the germ (A~, rTI) generated by Cauchy data (K 0 , S 0 ) for Hamilton-
Jacobi equation with dissipation is such that the whole of 6Jr 1 lies in the
non-singular zone of A~1 for 0 :s;; t :s;; T.
Let {U h II.,.}, j = 1, ... , N, be a )'-atlas of a neighbourhood of
1
the set
6/C = {(a, t) I 0 :s;; t :s;; T, a E c%t} c Mn+1.
Set Vi = II.,. (Ui) and extend the family {Vi} to an open covering
J
of K by adding a set V N+1 which does not intersect the image of the
set of all zeroes of the dissipation under the canonical projection onto
the plane p = 0. Let {ei} be a partition of unity in K such that
supp ei c Vh let S be the solution of the Hamilton-Jacobi equation
with dissipation given by Theorem 4.2 with ( U .,, II.,) replaced by
(Uh II.,.) and let S(N+l) be any smooth function with the positive
1
imaginary part.
N+1
Then S = 2J eisU>satisfiestheCauchyproblemfortheequation(1.5)
i=1
with the Cauchy data (K 0 , S 0 ).
This proposition is a direct consequence of the following lemma.
Lemma 4.2. 4et S< 1 > (x, t), S< 2 > (x, t), ... , S<Nl (x, t) satisfy the
relations F QJt [S< 3 >} =on (h), where D is a smooth non-negative june-
CH. IV. GENERALIZED HAL\IILTON-JACODI EQUATI'ONS 385
tion, the relations

S<i)- S<") =On (h3;2), I IDa;;-=
as 0 n (/1;2)
L
being also satisfied, and let e1 , e 2 , • • • , eN be smooth non-negative

functions such that e1 + . . . + eN = 1. Then
N
F&[LJ ejS<iJ]=on(h).
j=i
Proof is accomplished by induction on N. Let N = 2. Set X =

= S< 2 J - S< 1 J, so x = 0 n (h 3 12 ). We have
:t (e1S<i' + e2S< 2') = Stl) + e2x +On (h

1 312 ),
i}f (e 18<1'
_!!___ + e S< 2') = S<~> + e X + 0 n (h 312 ) •
2 X 2 X
Set
SijJ = ReSdJ, X1 =ReX,
S = e1S< 1' + e2S< 2', S 1 =ReS.
The expansion of OJB(S 1 x, x, t), OJBp (S 1 x, x, t) and GMpp (S 1 x, x, t)

in powers of S 1 x - S\1~ gives, by virtue of the relation F BC[S< 1 '] =
=On (h),
F&[SJ=xte2+<cf!B"(sn x, t), Xx"'2>+0n(h).
On the other hand, the expansion of cf/8 (S\2~, x, t), QJB (S\~, x, t)
and cf/8pp(S\2~, x, t) in powers of S\~~-S\1~=x 1 x leads, by using
the relation F Be [S< 2l) =On (h), to
Xt+(OJBp (S\~, x, t), Xx)=On(h),
which implies that F CJf) [S] = 0 n (h).

Now suppose that the assertion of the lemma is true for N = N 0
def
andproveitforN = N 0 +
1. We may assume that e' = e1 + ...
... + e No =I= 0 (since the statement of the lemma is of local cha-
racter). We have:
eNo+1 + e' = 1,
No+1 No
'V e.S<jl=e S(No+i)+e' 'V 2s<jl
L.J 1 No+1 .LJ e' ·
j=i j=i
25-01225
386 OPERATIONAL ;>.IETHODS
No
Since 2] eije' = 1, it follows from the induction hypothesis that
j=i
No
F&\9[~ :~ s<v>]=Ov(h).
i=1
To complete the proof use the result for N = 2.
Using the canonical transformation associated with the Hamil-
tonian Hr (see Example 2.1), we obtain the following generaliza-
tion of Theorem 4.2.
Theorem 4.3. Let Mn+1 be the manifold associated with the family
{A?}, and suppose that there is a y-atlas of Mn+l which consists of
a single y-patch (Uv, IT~). Then the function
S = (<D1+ icD 312 ) o (II~t 1 ,
where <D" is the y-phase of the germ in the zone Q 1 and c is the same
as in the dissipativity inequality, satisfies the following Hamilton-
J acobi equation with dissipation:
as
Tt + JJ8 ( axas xi, xr, - as
1
ax 1
1
' t + )
1 '
+ L·(no ( as 1
aro PI oxr ' XI' Xr,
_ as 1 t)
OXj ' '
as 2) -
oxr
-L ·(no
aroqf (as1
oxr , xi, Xr, - -8as)
-
XI
, -as2)
8-XI -
Sec. 5. Preservation of the Dissipativity Inequality. Bypassing

Focuses Operation
In this section we shall prove Theorems 3.1 and 4.1.
Lemma 5.1. Theorem 4.1 is valid under the conditions of Theo-
rem 4.2.
Proof. Let the germ {AB, rB} be dissipative, that is
e 0 (a) D (a, 0) ~ Im <D'~' (a, 0) +c 0 (a) [D (a, 0)] 312 ,
CH. IV. GENERALIZED HAIMILTON-JACOBI EQUATl!ONIS 387
where e 0 > 0 and c0 are smooth functions. We wish to show that

there exist such smooth functions B (a, t) > 0 and c (a, t) that
B (ex, t) D (a, t) c::;;; Im " (a, t) = c (a, t) [D (a, t)] 3f2
for 0 c::;;; t c::;;; T.
Set as above
~1 = Re (w - ~ (z - j')),
~ 2 = Im (w - ~ (z - j')),
" = <Ill' + i<l>~.

By virtue of (3.3), we have
a<D¥
aa
=[ a(q+v)
aa
JE- +0n (h) '
2 (5.1)
so
E =A a<D~
-2 aa +0 D
(h) with A= t[ a (q+y)
aa ] -
1•
Taking the imaginary part of (4.3) we obtain

~= -fi(p, q+l'. t)-(Hp(p, q+'\', t) ~ 2 )
-(Hp(p, q+j', t), ~1)-(~1• Hpp(p, q+j', tH2>-
1 - 1 ~
-2(£u Hpp(p, q+'\', t) £1>+2<£2, Hpp(P, q+'\', t)£2)+
• ~ 3j2
+(Hp(p, q, t)+'\'s2)+(Hp, On(h))+On(h ). (5.2)
It follows from (5.1) and (5.2) that <l>~ satisfies the following dif·
ferential equation:
~=G+<tAw~cz. g+v>. (5.3)
where
G= -H(p+£ 1 +/, q+'\'. t)+On(h 312 ),
f=On(h),
g=Hp(P, q, t)-Hp(p, q+'\', t)-Hpp(P, q+'\'. tH1 -
1 H~
-2 pp(p,q+j',tH2= On(h 112 ).
Here we have used (4.2).

Denote by d/dl the vector field
a ( A(g+'\'),
&+ · aaa ) ·
25.
.388 OPERATIONAL METHODS
With this notation we can rewrite (5.3) in the form

d<D~ =G
dl .
:Set
G1 = _jj (p + £ + /, q + y,
1 t),
and let '¢ be the solution of the following Cauchy problem:
~i =Gt, 1jl(a, 0)=e 0 (a)D(a, 0).
Thus, '¢ is defined on each trajectory of the vector field d/dl starting
from the initial manifold A3. Note that dt/dl = 1, so'¢ is non-nega-
tive for t ~ 0. Furthermore, '¢ does not decrease along every trajec-
tory of the field d/dl. We shall see that'¢ may be regarded as a "new
dissipation", so the following estimates are valid:
'ljJ (l, t) ~ c1 (a, t) D (a, t), (5.4)
D (a, t) ~ c 2 (a, t) '¢ (a, t).
Here we encounter, however, some diffi.culty, because, in general,
l.jJ is not defined everywhere (the trajectories of the field d!dl need
not be continuable). Therefore, we localize the situation as follows:
by Note 3.3 it is suffi.cient to prove the dissipativity inequality in
a neighbourhood of an arbitrary point (a 0 , t 0 ). Moreover, only the
case D (a 0 , t 0 ) = 0 is to be considered, as otherwise the dissipati-
vity inequality holds obviously.
The trajectory of the field dldl, starting from a point (a 0 , t 0 )
with D (a 0 , t 0 ) = 0 coincides for 0 ~ t ~ t 0 with that of the field
d!dt. Furthermore, we note that each point of a sufficiently small
neighbourhood u of (a 0 , t 0 ) lies on a trajectory of the field d/dl
starting from A3. In fact, let a = ~ (t) be the trajectory of dldl
passing through the point (a1 , t 1 ), let a = (; (t) be the trajectory
of d!dl starting from (a 0 , 0), and let 6 0 be R positive number such
that ;;: (t) is defined for 0 ~ t ~ t 0 +
6 0 (note that (; (t) = a 0 for
D ~ t ~ t 0 ). Denote by V the closed domain
0 ~ t ~ t0 +6 0, Ia - ;;: (t) I ~ 6,
and let
C= max IFa (a, t) j, where F (a, t) = ~~.
v
Since dtjdl = 1, we have
aa. : : :;
at= F (a (t), t),
CH. IV. GENERALIZED HA.MIL'l'ON-JACOBI EQUATI'ON1S 389
so
I:t (~{t)-a(t)) I~ cl~(t)-a(t) I
for t with (~ (t), t) E V. It follows that
1 ~ (t)- a(t) 1~ 1 a 1 - a: (t 1) 1 eclt-t11.
This estimate implies that the trajectory a = ~ (t) does not meet
the boundary of V for 0 < t < t 0 +
l) if u is sufficiently small.
The same argument shows that all segments of trajectories of d/dt
(that is, the lines a = const) starting from and finishing at u Aa
lie in the range of definition of '¢.
Now we prove that (5.4) holds in U. We have
-fi (p, q, t) ~G1 + l<Hp (P+ Gi + j, q+y, t), Gt) I+
+I <fiq (P+ £1 + f, q+y; t), y) I +caD~
~Gt+c4jil(p+£t+f, q-f-y, t)l 112 (!£tl+ll'i)+
+ caD ~c 5 (G 1 +D), (5.5)
where c 3 , c4 , c5 are smooth functions of a and t. Since

g =O D (h112) , aD
iJa
=O (hi/2)
D '
y= M fi P +On (hi/ 2),

where M is a smooth matrix function, and
~ aD 1 ~ 112 112 ~
IHpiX I
aa ~IHI D ~c6(1Hl+D)~c7(Gt+D),
it follows that
As is seen from Lemma 2. 1,

D(a 17 t 1 )~c 9 (a 17 t 1) (D(a, 0)+ ) G 1 dt)~
L(a1, t1)
~c 10 (a, t)(e 0 (a 1)D(a 17 0)+ ) Gidt)=c 11 (a 17 tt)lJl(ab tt),

L(a1, l1)
where L (a1 , t 1 ) is the segment of a trajectory of d!dl starting from

AIT and finishing with (a1 , t 1 ). This proves the second estimate
in (5.4).
On the other hand,
I~? I=IG~-<!!. A(g+v))l:s;

:s;Gi +ct 2¢ 11 2 (D 112 -+:- I Hl112 ):s;c13 (G 1 +¢+D). (5.6)
The calculation similar to that in (5.5) shows that
Gt = -li (P+ 'Gt + f, q+ y, t):s;ctd -li +D). (5.7)
By Lemma 2.1 we obtain from (5.6) m~d (5. 7):
t
'IJ(a, t):s;c 15 (D(a, 0)- )" lidt) =c15 D(a, t),
0
and (5.4) is proved.
Now we are able to obtain the required inequalities for <DL
We have
d (¢-CD¥) = 0 (h3!2)
dl D '
so
l!J(a 1 , t 1)-<D¥(ah t 1):s;l¢(ah 0)-<l)~(ah 0)1+
+Mdat. t 1) j" D 312 dt:s;le 0 (a)D(at.O)-CD¥(ah0)1+
L(a, t)
+M2(a1, t 1) J ¢ 12dt:s;
3
L(a, t)
:s; I e0 ( a 1) D (at. 0)- <D~ ( ah 0) I+ M 3 ( a 1 , tt) X

X[¢ (at. t 1)] :s;c0 (a 1) [D (at. 0)] 3' 2 +
+Mt,(ah tt) [D (ah t 1)] 312 :s;c (~h t 1)[D (ah t 1)·] 312 •
Thus,
..!. D :s; lP :s; CD~+ cD3!2.
C2

Note 5.1. In the same way it can be proved that Theorem 4.1
is valid under the condition· of Theorem 4.3.
Corollary. Theorem 4.1 is valid if A? c Q1 for any t (with I being
independent of t).
Proof. Let L : {a = a 0 , 0 :s; t :s; T} be a segment of the tra-
jectory of d!dt starting with a 0 E AU. Then there exist such numbers
t 0 , t 1 , • • • , tN with 0 = t 0 < t 1 < t 2 < ... < tN = T that th_e
CH. IV. GENERALIZED HAIMIL'l'ON-J.AJCOBI EQUATJ10NIS 391
segment Li of L corresponding to [ti-I• til lies in Uh where (Ui, IT~)

I .
is a y-patch of Mn+I, IT.,, (a, t) 1 = (q (a, t) +
Yi (a, t), t). By
Note 5.1 we can choose smooth functions 8 1 > 0 and c1 in such a way
that the dissipativity inequality
8 1D ~ <D:f! +c
1D3f2
holds in a neighbourhood of L 1 in .llr+l.

Since the property of a germ to be dissipative does not depend
on the choice of a y-atlas (we have proved this in the case of a y-atlas
consisting of patches of the same zone), we conclude that the dissi-
pativity inequality
8 'D....--rnv2+,
::::::::::w1 2 c1 D3!2
with some 8' and c; holds in a neighbourhood of (a 0 , t 1). It follows
from Note 5.1 that there exist smooth functions 8 2 > 0 and c2
such that
8 2 D~cD1i + c2D 312
in a neighbourhood of L 2 in Mn+I, and so on.
Now let us return to Theorem 3.1. Suppose, for the sake of simpli-
city, that a 0 E r belongs to the intersection of a non-singular patch
of the non-singular zone with a non-singular patch of Qr, and let
the dissipativity inequality for the non-singular zone
8D rn
~ w 2 + cD3/2
with some 8 and c hold in a neighbourhood of a 0 • We have to show
that the dissipativity inequality for the zone Qr
(5.8)
with some 8' and c' holds in a neighbourhood of a 0 •
Let gkr be the canonical transformation considered in Exam-
ple 2.1, and let <D (a, t) be the phase on gkrAn. Then <Dr (a) =
= <D (a, n/2). If the matrix fJ (q (a, t) +z (a, t))loa is non-dege-
nerate for a = a 0 and 0 ~ t ~ ~ , then the validity of (5.1) fol-
lows from the corollary to Lemma 5.1.
If, however, for t E (0, n/2) the image of a 0 by the canonical
transformation g~ r is a zero of the Jacobian D (q + z)/Da (such
a point (a 0 , t) is said to be focal), then the above argument fails.
Note that even if the transformation gfi: (which is the rotation
through 90° of some axes in the phase space) were embedded in the
family of canonical transformations gk associated with another
Hamiltonian function H, then, nevertheless, the trajectory of the
Hamiltonian vector field d!dt starting with a 0 would be sure to have

focal points.
It turns out that this difficulty can be avoided by using some
other rotations of axes. This method will be called "bypassing
focuses operation".
Definition 5 .1. Let ao E r c An belong to the intersection of a non-
singular patch of the zone QK with a non-singular patch of the zone Q 1 •
A real Hamiltonian function H will be called a bypassing focuses
Hamiltonian associated with the point a 0 and the pair (K, I) if the
following conditions are satisfied:
(i) det c[{ (a, t) =I= 0 for 0 ~ t ~ ~ '
(ii) K {a, T) = <1> 1 (a) in a neighborhood
of a 0 , where CK (a, t), and K (a, t) are the matrix introduced in the
definition of fl-action and the phase in the zone Q 1 of gkAn, respecti-
vely.
Theorem 5.1. Let ao E r belong to the intersection of a non-singular
patch of the non-singular zone with a non-singular patch of the zone
Q 1 , and let satisfy a dissipativity inequality in a neighborhood of
a 0 • Then there exist an orthogonal matrix u = (uo)i , J'EI and a dia-
gonal matrix e = (eu)i, iEI with eu det ei = +1 such that the
function
1
H = 2 {(p1, u- 1euP]) + (fJ]_, u- euq1)}
1 (5. 9)
is a bypassing focuses Hamiltonian associated with a 0 and (, J), the

image of a 0 by gk belonging to a non-singular patch of the non-singu-
lar zone of gkAn fortE [o, T J.
Proof. To simplify the notation we shall renumber the coordinates
in the phase space so that I should be equal to {1, ... , k}. Set
A (a)= PrArl = pq,
then A (a) is symmetric. Denote by A <k> the left upper block o£ A.
Choose an orthogonal matrix u so that the matrix uA <k>u - 1 is dia-
gonal, and a diagonal matrix with eu = +1 so that the condition
euA <k>u - 1 ~ 0
is satisfied. Defme the n X n matrices ~ and e by
~
u=~o
!u 0)1 '
CH. IV. GENERALIZED HAIMIL'l'ON-JA.COBI EQUATJ10NIS 393
Let us consider the canonical transformation g generated by the

orthogonal operator 'U in R~. This transformation is given by
p' (a)= (tp (a), q' (a)= uq (a),
w'(a)=uw(a), z' (a)= uz (a),
D' (a)= D (a), E' (a)=E(a)
(we mark off functions on gAn by '). We have:
(J)'(a)=CD(a), Ci=(;,Cr, B!=(;,Br, er=Ueru-1,
in particular, the image of a 0 by g belongs to Q Rr. It is obvious n
that the function (5.9) is a bypassing focuses Hamiltonian associated
with a 0 and (Q, Rr) if and only if the function
H' = ~ ((p 1, sp1) + (q 1, sq1)) (5.10)
is a bypassing Hamiltonian associated with ga 0 and (Q, Rr)· There-
fore, we shall omit strokes, assuming from the outset that Ak is
diagonal and putting u = 1 in (5.9).
On solving the Hamilton-Jacobi system and the system (2.4)
correspoding to the Hamiltonian function (5.10) we obtain
p1 (a, t)=p1 (a)cost-sq1 (a)sint, Pr(a, t)=Pr(a),
q1 (a, t)=q1 (a)cost+sp 1 (a)sint, qr(a, t)=qr(a),
w1 (a, t)=~(a)cost-sz 1 (a)sint, wr(a, t)=wr(a),
z1 (a, t)=z1 (a)cost+sw 1 (a)sint, zr(a, t)=zr(a).
It follows that
p1 (a, n/2)= -sq 1 (a), w1 (a, n/2)= -sz1 (a),
q1 (a, n/2) = sp1 (a), z1 (a, n/2) = sw1 (a),
(w(a, n/2), z(a, n/2))=(-sz1 (a), sw1 (a))+(wr(a), zr(a))=
= -(z1 (a), w1 (a))+(wr(a), zr(a))=(w(a), z(a))r. (5.11)
Set P=p+w, Q=q+z. Then
iJQ- (a) iJP- (a) )
(
: cost+ e - ; - sin t
C(a, t)= a ua .
iJQr (a)
iJa
So
eiJP1 (a) )
iJa A iJ (P 1 (a), Or (a)) A
(
C(a,n/2)= =e 8 =eCr(a).
iJQr (a) a
iJa
In the same way we obtain

B (a, n/2) = eBI(a).
Therefore
lb (a, n/2) = elbr (a) e.
By noting that e = 1,
2 we obtain
1
!!(a, n/2)=-z(z(a, n/2), lb(a, n/2)z(a, n/2))=
={(e (:I<~)), elbi(a) ee (:I<~)))=

=~((:I(~)), lb 1 (a) (:I(~~)))=!!r(a). (5. '12)
We also have
n/2
s(a, n/2)=s(a)+z- 1
Jr {(Pz(a, -r), ePz(a, -r))-
o
-(qi(a, -r), eqi(a, -r))}d-r=
n/2
=s(a)+ ~) {(p1 (a)-q1 (a), e(p1 (a)-q1 (a)))cos2T-
o
-2(Pz(a), qi(a)sin2-r)}d-r=
=s(a)-(pi(a), q1 (a))=s 1 (a).
Since the potential is independent of t, the f1rst assertion of the
proposition is proved.
Now it is easily seen that
1- 0 )
C(a,t)= ( 01 . 1 1 (Ccost+eC 1 sint),
SIll t + COS t I
where 11 is the identity k X k matrix and 11 is the identity (n - k) X

X (n- k) matrix. Hence for any t E [ 0, ; J, the matrix C (a, t)
is non-degenerate if and only if C (a) cost + ~C 1 (a) sin t is non-
degenerate.
We are thus led to the criterion: H is a bypassing focused Hamil-
tonian if and only if the matrix eC 1 (a 0 ) C- 1 (a 0 ) has no negative
eigenvalue.
CH. IV. GENERALIZED HAt;\:IIL'DON-J.AJCOBI EQUATION'S 395
We have
8P-
ic,c-• ~ ,~; ( (5.13)
Denote by ~<k> the upper left k X k block of ~ (a 0):

8P-
~<k> = 8Q~ ( tXo).
I
By (5.13), it is sufficient for non-existence of negative eigenvalues

of eC1 C- 1 la.=a.o that 8~<k> have no negative eigenvalue because
the spectrum of the latter matrix contains that of ;C1 (a 0 }CJ.1 (a 0 ).
In fact, let x be the left eigenvector of ;c
1 (a 0 ) C-1 (a 0} belonging
to an eigenvalue ').,, then
x-8'2<h>
I c; -- J..x-I'
8P-
X18 aQ; (a0) Xr = J..x1•
Note that xy=/= 0, for otherwise the latter equation implies that
x1 = 0, so x = 0. Thus, x 1 is the left eigenvector of 8~<k> belon-
ging to the eigenvalue ')..
Similarly, for the matrix qa. (a, t) to be non-degenerate when
J,
t E [ 0, ~ it suffices for the matrix 8A <h) to have no negative eigen-
value. The last is ensured by the choice of 8. We shall show that
the same choice of 8 ensures that 8~<h> has no negative eigenvalue
either.
Let us accept the following agreement: for the rest of the proof
all differentials in question are taken at a 0 • Let E 0 be the subspace
of Rn consisting of all vectors annihilating d2D. Then the relations
z =On (h112) and w =On (h112 ) imply that dz lEo = 0 and dw lEo =
= 0. In fact, let f =On (h112) be a real function, then
d2j2 (u) = 2 [df (u)l 2
for each u. Since / 2 ::=:;;; cD and j2 - cD has a maximum at a 0 it
follows that
d2j2 - cd2D ::=:;;; 0,
hence
[df (u}l 2 ::=:;;; const d2D (u),
which implies immediately that
(u E E 0) * (df (u) = 0).
Now suppose that e~<k> has a negative eigenvalue:

e~<k> (u + iv) = -'A (u + iv), (5.14)
where 'A> 0. Rewrite (5.14) in the form of a system of two equa-
tions:
~lk)u- ~~k)v = - 'Aeu,}
ce(k)
e1 v
+ ce(k)
(2) 2
~
u = - "'ev,
where ~lk) and ~~k) are the real symmetric matrices defined by
~<k> = ~\k) +
i~\J'>. Multiplying the firSt equation by v and
subtracting it from the second one multiplied by u we obtain
(~~k>u; u) + (~~>v, v) = 0. (5.15)
Let us choose q as local coordinates at a 0 • Then d2<D 2 (u) = (~ 2 u, u).
By the dissipativity inequality,
d 2 <D 2 ~ cd2D ~ 0,
where c is a positive number. If u' E Rn and v' E Rn are defined by
uj = u, vy = v, u[ = 0, vi = 0,
then
(~~k>u, u) = d2<D 2 (u') ~ cd2D (u') ~ 0,
(~~k>v, v) = d2<D 2 (v') ~ cd2D (v') ~ 0.
Hence (5.15) implies that
(~\J'>u, u) = (~~k>v, v) = 0, (5.16)
u' EE0, v' EE0•
Since ~~k) ~ 0, it follows from (5.16) that ~~k)u = ~~k)v = 0;
hence u + iv is an eigenvector of the real matrix e~~k) belonging
to the eigenvalue -'A. At least one of the two real vectors u and v
must be non-zero. Let, for example, u =1= 0. Then
~<k>u = -e'Au, u' E E 0• (5.17)
Let us show that
~<k> u = A <k>u.
We have
C (a 0 ) u' = [1 + Zq (a )1 u'
0 = u' + dz (u') = u',
hence C-1(a 0) u' = u'. Next,
~ (a 0 ) u' = B (a 0 ) C- 1 (a 0 ) u' = B (a 0 ) u' =
= pq (a 0) u' + dw (u') = pq (a 0 ) u'.
CH. IV. GENERALIZED HA!MILTON-.TACOBI EQUATI'ONIS 397
Since u] = 0, we obtain
[pq (a: 0) u'l:r = A <11 lu
and
[~ (a: 0 ) u'l:r = ~< 111 u.
It follows from (5.17) and (5.18) that
A< 11 lu = -e'Au,
contrary to the condition eA <11 l ~ 0. The theorem is proved.

Corollary. Let a: 0 E r belong to the intersection of a non-singular
patch of QK with that of 0 1, the phase <DK satisfying at a: 0 the dissipa-
tivity inequality
eD ~ Im <DK + cD 312 •
Then there exist smooth functions e1 > 0 and c1 such that the dissi-
pativity inequality
e1D ~ Im <1> 1 + c D 312
1 (5.19)
holds in a neighborhood of a: 0 •
In fact, in the case with K = 0 this statement is a direct corol-
lary of Lemma 5.1 because a bypassing focuses Hamiltonian exists.
The general case can be reduced to that with K = 0 by applying
the canonical transformation gfi~, where H K is the same as in Exam-
ple 2.1.
Now let us turn to the singular patches case. Our treatment de-
pends on the following observation: the image of a: 0 E An by an
appropriate "small" canonical transformation is in a non-singular
patch of the zone Q 1 for every I. To state this result more precisely
we introduce the following definition.
Definition 5.2. A point a: 0 of a Lagrangean manifold
An = {q = q (a:), p = p (a:)}
is said to be non-singular with respect to the canonical projection onto

the coordinate plane p 1 = 0, q1 = 0 of the phase space if
D (q[, PJ.)
Da
I
a=ao
=f= 0,
i.e., if the mapping

a:-+ (qi (a:), PJ: (a:))
is a diffeomorphism of a neighborhood of a: 0 •
Proposition 5.1. Let A be a Lagrangean manifold. Then for any

a 0 E A and, sufficiently small positive t, the point
g}p0 Eg1:A, where H = 21 (p2 + q2),
is non-singular with respect to the canonical projection onto the coor-
dinate plane q1 = 0, PI = 0 of th, phase space.
Note 5.2. Proposition 5.1 remains valid if H is any Hamil-
tonian function with 8 82~ (p (a 0), q (a 0)) > 0.
ql PI
This result is essentially the same as the known lemma of M.
Morse which states that focal points of a submanifold of a Rieman-
nian space corresponding to a geodesical are isolated in R.
Proof of Proposition 5 .1. It suffices to check the result for I = 0.
In fact we have the following commutative diagram
n/2
gHI
A A'
gkl 1 gil
t n/2
gHI t
At __,.At
where HI is the Hamiltonian function introduced in Example 2.1.
Since gf/ 2 sends (p, q) E R2n into (p', q') with p~ = (pi, -qy),
I
q' = (qi, p 1), the point g1:a 0 E g1:A is non-singular with respect
to the canonical projection onto the plane Pr = 0, q1 = 0 if, and
only if, the point (g1:ogfi:) a 0 E At is non-singular with respect
to the canonical projection onto the plane p = 0.
So let I = 0 and let At be given by p = p (a, t), q = q (a, t).
We have to show that the matrix oq (a, t)/oa is non-degenerate for
a = a 0 and sufficiently small positive t. We have
8q(a, t)
aa =
8q(a)
----r;a cos
t+ ---ga-
8p(a) . t
sm = cos ac;:- +tan t ---ar;:-
t ( 8q(a) 8p(a))
•
Set
8q(a)
aa
I
a=ao
=C, 8p(a)
aa
I
a=ao
=B,
tan t= 'A.
Thus, what we have to show is reduced to

det (C + 'AB) =I= 0
CH. IV. GENERALIZED HAIMILTON-JACOBI EQUATJ10NIS 399
for small positive 'A. Let rank C = k. We may assume without loss
of generality that the first k rows of C are linearly independent.
Then the matrix
( 8qrf8a )
8p·J)8a a=ao
with I= {1, 2, ... , k} is non-degenerate (cf. the proof of Lem-
ma 2.4). Choose q1 , p1 as local coordinates in A at a 0 • Then we can
express B and C in the form
apr apr )
(
B= a~r a:1 ,
(for the rest of the proof the argument a 0 will be omitted everywhere).
so non-degeneracy of C +
'AB is equivalent to that of
1 + 'Aaprfaqr .!..E.!_)
(
api
Mt'A)= aq- .
_I_ 1
aqr
Since M ('A) is continuous at 0, it is sufficient to check that M (0)
is non-degenerate. Let s1 (q 1 , p 1) be a generating function of the
Lagrangean manifold A (an s-action):
asr as
q1 = - iJp-' Pr=-.
iJqr
I
Then
so to complete the proof, it suffices to apply the following result:

Lemma 5.2. Let A be an m X n matrix. Then the matrix
is non-degenerate.
Proof. Suppose that this is not the case and let u be a non-zero
vector such that .Jlu = 0. Write u in the form
with u1 ERn, u 2 E Rm. Then

u1 + Au 2 = 0, - 1Au1 + u 2 = 0,
which implies that
(u1 , u 2) = -(A 1Au1 , u 1 ) = - ( 1Au1 , tAu1 ).
Thus u1 = 0, contradicting the assumption u =fo 0.
Proof of Theorem 3.1. We can now give the promised proof of
Theorem 3.1. We have to show that if a 0 Er is in the intersection
of two patches (uy, n:) and (uy', :rt~·) of types (K, F) and (I, F'),
respectively, <Dk- satisfying a dissipativity inequality in a neigh-
borhood of a 0 , then there are 8 > 0 and a such that
8D ~ Im cp:f' + cDSf2
near a 0 •
Let
80 (a) D (a) ~ Im <Dk- (a) +c 0 (a) [D (a)J31 2
in a neighborhood of a 0 • Set H = ~ (p 2 + q2 ) and consider the

canonical transformation gi£, where t is so small that the image of
a 0 under gi£ is in the intersection of a non-singular patch of QK
with that of Q 1 (such at exists by Proposition 5.1), being yet in the
intersection of a patch of type (K, F) with that of (I, F'). Then for a
near a 0 we have
8 1 (a, t) D (a) ~ Im <Dk- (a, t) + c1 (a, t) [D (a)J31 2
by Lemma 5.1;
8 2 (a, t) D (a) ~ Im <DK(a, t) +c 2 (a, t) [D (a)J31 2
by Lemma 3.4;
8 3 (a, t) D (a) ~ Im cD 1 (a, t) +c 3 (a, t) [D (a)J31 2
by the corollary to Theorem 5.1;
8 4 (a, t) D (a) ~ Im <D:f' (a, t) + c4 (a, t) [D (a)J31 2
again by Lemma 3.4; and finally,

8 (a) D (a) ~ Im <D:f' (a) + c (a) [D (a)J31 2
by Lemma 5.1 with gi£ replaced by gt_H which coincides with gff.
Theorem 3.1 is proved.
Proof of Theorem 4.1 is similar to that of the corollary to Lem-
ma 5.1, we need only replace Proposition 5.1 used there by this
corollary, Lemma 3.4 by Theorem 3.1 and consider zones instead
of patches. Details may be left to the reader.
CH. IV. GENERALIZED HAIMILTON-JAOODI EQUATrONIS 401
Sec. 6. Solution of Transfer Equation with Dissipation

We have before by numerous examples illustrated the role that
J acobians and transfer equations play in asymptotic expansions
and geometric constructions. Here we shall solve the transfer equa-
tion in the general situation corresponding to a Lagrangean mani-
fold with a complex germ.
First of all we introduce the notation g = 0 1 (ha) for functions
dependent on the parameter h.
Definition 6.1. Let f (x) be a smooth non-negative function and let
io
g(x, h)=~ g 1 (x)hi 12 •
j=O
Then we write g =Of (hv) if gi = Of (hv-i 12 ) for j :s;,;; 2v.

I I is clear that
Ot (h''1) Ot (hv2) =Of (hvt+V2),
o, u~V!) + o, (hV2) = Ot (hmin{vt, V2})
and that any polynomialg in h 112 wilh coefficients smoothly depen-
dent on x is at least Ot (h0l
Definition 6.2. We write g = (CJ 1 (hv) if
g=h-" 12 0n (hv+ ~)
for a non-negative integer k. This notation makes sense even if v is
negative.
Let
s (x, t) = s1 (x, t) + is 2 (x, t)
be the solution of the Hamilton-Jacobi equation with dissipation
given in Theorem 4.2. Thus we consider the following object to be
given: a family {At', rf} of Lagrangean manifolds with complex
germ and the corresponding to this family diffeomorphism
Tiv : (a, t) -+ (x, t)
which allows us lo regard x, t as coordinates in the (n +
i)-dimen-
sional manifold {At}. As before, differentiation with respect to t
with fixed a will be denoted by :e
or by the point over the symbol
of a function, while that with fixed x will be denoted by ! or by
the subscript t.
26-01225
Now we introduce a class fff' of differential operators which will

be regarded as "perturbations".
Definition 6.3. We say that a differential operator
.Jl=
io
~
lii=O
aj(a, t, h) ( i)~ r
dependent on the parameter h belongs to the class JP if
ai=Ov(h ljl;-i )+<Hp(p(a,t),q(a, t), t), Gv(h l;l))+
lil-1)
+h 112 0v ( h -2- for ljl:;>1
and
a0 = Gv (h 112 ) + (fiP• Gv (h0)).
t
Set ) j (a, -r) d-r = If (a, t). The following shows why it is natural
0
to regard operators belonging to fff' as "perturbations".
Lemma 6.1. For any .Jl E fff' there exists an integer-valued junction
(r, s) of two non-negative integer variables such that the following
:rt
conditions are satisfied:
(i) .JI; (I .Jl )r Ov (hs/2)
hsf:!.
= (9
D
(h :rt (r, s) ) •
2 '
(ii) lim :rt (r, s) =
r~oo
+ oo.
Proof. Let .Jl E (ff'. Then thBre is an l :;;>0 such that
.:!i0v(hsf2)=h-l/20v (hs~l)
and the opera tor I .Jl can be represented as A+ B, where
s+l+l)
AOv (h 512 ) = h -l/ 20v ( h -2- ,
BOn (hs/2) = fh 112 0v (h

s;i) for s:;;>1
l 112
h 0v (h 0 ) for s = 0.
This partition of I A corresponds to the natural partition of ai:
ai = a}+a},
lii+l ) .
aj = Ov ( h-2-, + (Hp, Gv (hiJi/2)),
1/20 ( h Iii; 1 )
a':-}
1 - L D ,
CH. IV. GENERALIZED HAIMIL'I'ON-JACOBI EQUATl'ONIS 403
the estimate of A being obtained by noting that

t t
I j (Hp, x)dtl~c(a, t) [D(a, t)]v j IHvldt~
0 0
t 1
:::;;;c!(a, t) [D(a, t)] v[\j IHidt

~· ]1/2 ~ct(a, t) [D(a, t)]
v+z
0
holds for each vector-function x (a, t) = On (hv).

Consider the operator (A + BY. We have
r ~ ~ p(1) p(k) p(k+1) p(r) def
(A+ B) = LJ LJ
k=O PEG
s ... s A ... A = hk, p ck. p,
where G is the group of all permutations of {1, ... ; r}. It is easily

seen that
C On (h s/2)
= h
-l(r-k)+s-k
2 On
(
hmax
( (1+l)(r-h)+s-l:
2 ,
o}) •
k' p ,_....=:h_:.~..,/2~
It follows that
On (hs/2) = h P(1·-k)+s-h ( (1+l)(r-k)+s-k ) ( r-s )
C 2 On h 2 = <:Yn h2Cl+2)
k, p hs/2
f or k ~_ (1+l)r+s d
l +2 , r > s, an
0 s/2 l(r-k)+s-k ( r- s )
Ch, P
n(h ) =h-
hs/2
2 0 (h 0 )=<3
n n
h 2<1+ 2>
f or k > (1+l)
t+Z
r+s
, r .> s. At last for r~s, we have
C On (h"/2) - (9 (ho)
k, P hs/2 - n •
Thus, one may put

0 for r~s,
:n: (r, s) =
{
[ r-s
Z+Z
Jf or
.
r>s, Q.E.D.
Definition 6.4. Any function :n: (r, s) satisfying the conditions (i)
and (ii) of Lemma 6.1 will be called a ff'-function of .II:. The small-
est non-negative integer l with the property
.:!iOn (hs/2) = h-lf20n (h s~l)

will be called the type of .Jf.
26*
Problem 6. i. Let
Then
[ 2r;-s J+i for { 2r;-s} >{
n (r ' s) = {
2r-s-4 [ -5
2r-s
-J for { - 2r-s
5- } ~5
1
is a ~-function of .It; here [x] denotes the greatest integer not excee-
ding x, and {x} = x - [xl. In particular, n (r, 0) ~ 1.
Given a Hamiltonian function rJ/8 (p, q, t), a solution s (x, t) of
the Hamilton-Jacobi equation with dissipation corresponding to $-8,
a smooth function g (p, q, t) and an operator .!! E ~. the following
relation will be called the transfer equation with dissipation in the
N d . .
2 - or er appronmatwn:
(/!t+(d1'8p(Stx, x, t)+ic!J8pp(Six, x, t)S2x. (/Jx)+
+{-}tr[c!J8pp(Six, x, t)Sxx]+g(S1x, X, t)+.it}cp=
=('Js~(hN/2). (6.1)
We shall consider only those solutions of the transfer equation with

dissipation which have the form cp = h-m/ 20s2 (hm12 ), where m
is a non-negative integer.
The relation (6.1) with the right-hand side replaced by
0s2 (h 112 ) + ( <9s 2 (h 0 ), Hp (Six, x, t))
will be called the transfer equation with dissipation in the zero

approximation. Recall that
~ 1/3
Hp (Slx• x, t) = <9s 2 (h )
for t > 0.
For convenience we shall identify cp (a, t, h) by the diffeomor-
phism II~ with a function (dependent on the parameter h) on the
(n + i)-dimensional manifold {An. Since
cpt = ~- (q + y, cpx) = ~- ((qa + '\'a.)- 1 (Hp + y), era}

CR. IV. GENERALIZED HAMILTON-JAOOBI EQUATI'ONIS 405
and
Sx = l1 + w- t«J (z -y) + u 1 ,
Sxx = t«J + u2,
OJ8 P (Slx• x, t) = c!J81, (p, q, t) + JJ8 PP (p, q, t) X
X (w - 8 (z - y)) + rfJ8 pq (p, q, t) y + u 3 ,
rJJ8pp (Slx• x, t) = OJ8pp (p, q, t) + u4 ,
g (Slx• X, t) = g (p, q, t) + U5,

where
U1 =On (h), u 2 =On (h 112), u 3 =On (h),
u4 = On (h112), U 0 =On (h112 ),
then the transfer equation with dissipation can be rewritten in the
coordinates a, t as follows:
~P + (fPa., (qa + Ya.r 1[iH P - ~ + rJJ8 PP (w- ~ (z -y)) + JJ8 pqy]) +
+ {-}- tr (&8 pp~) + g (p, q, t) + pg} cp = 0n (hN 12 ), (6.2)
where $ E rff' is related to .Jl by the formula

.CJIJ = .11: + <
(qa + Yut 1 [Us+ iu4S2X + irJJ8 pp Im ut], a~ ) +
'1
+2tr (CJf! ppUz + u4Sxx) + u 5 •
Define the transfer operator pv by
pv = :t + (t. a~ ) ' (6.3)

where
f = (qa + Ycxt 1 [iH p - ~ + rJJ8 pp (w- 8 (z-y)) + OJ8 pqy].
We begin by solving the transfer equation with dissipation in
the zero approximation which has the form
pVQJ+[ -}tr(&'z9ppB)+g+$JqJ"-'0, (6.4)
where the sign,....., means equality modulo

1/2 ~
0n(h )+(0n(h0 ), Hp).
By analogy with the real Hamiltonian function case, we introduce
a new function '¢ to be found, related with QJ by
QJ (a, t, h)= V J 1(a, t) \jJ (a, t, h),
where J (a, t) = det C (a, t). In order to rewrite (6.4) in terms

of 'ljl, we shall use the following result.
Lemma 6.2. The Jacobian J satisfies the relation
(6.5)
Proof. By the rule of differentiation of a determinant we have
a (qi +z1) a (qi +z!) )

OCXj OCXn
(
+det · · · · · · • 0 0 • • •
(6.6)
a (qn +zn) a (qn +zn)
()a;! OCXn
In virtue of (2. 6)
a (q+z) §10
pqC + QlV PP.B +On (h
£10 1/2
arx QIV )=
= (QJ8 pq+rJ/8 pp~) C +On (h 112 ).
Thus,
. . n
o(qb+zb) = L.J
'\. . ., (6/8 +&'8 l)) a(q,+zk) +O (hf/2)
OCXj pq pp lk acxj D '
k=1
so the Z-th determinant in the right-hand side of (6.6) equals
(6/8 pq + OJ8 PPl))u J +On (h 1/2),
and it follows that (6.5) holds.
Corollary.
d 1
-d ,(- = -
t v J 2
1
J
v- tr (QJ8 ppl) + QJ8 pq) + x. (6.7)
where X = 0 n (h1 / 2).

It follows by noting the corollary that 'ljJ must satisfy the relation
(Pv+G(p, q, t)+$')¢---0 (6.8)
with
G= g - + tr JJ8 pq $' = VJ ( $ J:r + X) E ~ ·
CH. IV. GENERALIZED HAJMILTON-JAGOBI EQUATl-0:-IIS 407
Note that in spite of pv being equal to :t + (t, iJ~ ) with f ,. . .,. 0,

we may not ignore the Lerm (t, ~~) if we wish to find a solution
m
of (6.8) of the form 'ljJ = h- 2 0 D (hm 12 ) with m > 0.
The idea of our method of solving (6.8) is as follows. If we could
transform pv into :t
by an invertible operator L:
PvL=L Tt,
d
(lU))
then the problem would be trivial. In general it is impossible;

nevertheless, it turns out that there is an "almost invertible" operator
"almost satisfying" (6.9).
Namely, let
def N ·
L'fv = -v [B <~; t)]J ( a )i (6.10)
LJ J. iJa '
li 1=0
where
~= -c-t(z-v). (6.11)
Lemma 6.3. L'fv satisfies the relation

d
P VLVN = LVN-aT +e
~+ ;;
L.. gk ( iJa
i) ')k
lki=N+1
with eEfP and
gk=OD (hN~i) + (oD (h ~) Hp>.

Proof. We have
PvL'fv-L'fv ~ = ~
N
lk\=0
*Bk, (6.12)
where
B
h
= pV~k (-a
iJa
)k _ ~h (-iJ-)h
iJa
_:!:_.
dt
(6.13)
Let 1i denote the multi-index (oil> ... , oin), where

1 if k= j
{
0ih= Oifk=/=j.
Then
Bk= ~ ki~k- 11 (~ 1 +<1, ~:)) (a:)"+

i=1
n
+ ~k ~ IJ (
]=1
a: r+1J. (6.14)
Substitution of (6.14) into (6.12) giv8s
pV£'},-L't, :t = ~ ( ~
i=1
n N-1
\ki=O
~~ Bk-1j[PV(aJ+~;)] (a: r +
'\'
+ L...J 1::._ 1
k! J
(_a
iJa
)k+1J) •
\k\=N
Since ~=On (h 112 ) and 11 =On (h 112 ) +(a, fl p), where a is a
smooth function of ex; and t, it suffices only to check that
Pv(a1 +~ 1 )=0n(h)+(On(h 112 ), Hp) (6.15)
to complete the proof of the lemma.
We have
P
v •
(aJ+Bi)=IJ+BJ+ I, 8~
aaj < ) def
=Aj.
Let A denote the vector (At> ... , An), then
A = ~ + (1 + Ba) I. (6.16)
Using the formula
~ 1/2
C=0J8pqC+JJIJppB+On(h ),
we obtain
~ = c- 1 (<f/8 pqC + JJ8 ppB) C- 1 (z -y) -C- 1 (;-~)+On (h)=
= c- 1 [&'8 j!Q (z- y) + JJ8 ppW. (z- y)- iff p -
-JJIJ ppw-J78 pqZ +'\']+On (h)=
=- c- 1 [c%' PP (w -lt (z -y)) + iil]l + JJ.81JQ y- ~~+On (h)=
= - c- 1 (qa + Yu) I+ On (h). (6.17)
Further,
1 + Ba = 1- c-i (za- Ya) + 0 n (h 112 ) =
= c- 1 (q_a +'\'a)+ On (h 112 ). (6.18)
CH. IV. GENERALIZED HAIMIL'l'ON-JAGOBI EQUATFONIS 409
Equations (6.17) and (6.18) imply that

1/2 ~
A=On(h)+On(h )Hp,
Q.E.D.
The following two lemmas give the construction of an "almost-
inverse" of L'fv.
Lemma 6.4. Let p (ex, t) be a smooth vector function with
det (1 + Pa) =I= 0. Then there exist two sequences {8\ (p)} and {Qk (p)}
of homogeneous polynomials in p with vector coefficients smoothlyc
dependent on ex and t, ff'k and Qk being of degree k, such that
R-1 R-1
P+ ~
li 1=0
f/ (:a)j ~ ff'R.(p)=QR(P)
k=1
(6.19}
holds for every R > 1.

Proof. We shall construct {ff'~t} and {Qd by induction. Set
ff>l (p) = -(1 + Pa) -l p.
Then
1 .
+ ~ ..e.!_ a' J lg'i (p) =
p L.i il aai
I i 1=0
= p-(1 +Pat 1 p-(1 + Po.t 1 PaP+ Q2 (p) = Qz (p),
so (6.19) holds for R = 2.
Suppose we have constructed ff' 1 , . . . , ff>m and Q2 , • • • , Qm+r
so that (6.1 9) holds for R = 2, ... , m + 1, and let ff> m+t be any
homogeneous polynomial in p of degree m + 1 with vector coefficients
smoothly dependent on ex and t. Then
m+i m+i m+1 I.
P+ ~ .E.!._(_!!___)i ~
LJ jl aa L.i
@'
It
(p)= ~ ..£!._a J!gum+t(P)
LJ jl aai
+
lii=O k=1 lii=O
+ Qm+i (p) + ~ ~: (a~ )j ~[!>It (p) = Qm+i (p) +

li l=m+1 k=1
m+i I •1
~ 2.!_ a 3 .a?m+i (p) (6.20)'
+ I iLJ1=0 il aai '
where iJm+t (P) is a homogeneous polynomial of degree m 1 +

with smooth coefficient which does not depend on the choice of
ff> m+t· Since
ali!gum+t(P)
aai =
('
Pa
a); ff>m+t (p) + Qi.
ap
=
m+2-i (p),
where 7h,m+ 2 -i (p) is a homogeneous polynomial of degree m + 2 - j

with smooth coefficients, (6.20) implies that if ff> m+I satisfies the
-equation
m+i
~ pi (t a )i - -
L..J JT Pa ap ~m+dP) = -Qm+i (p), (6.21)
I j J=O
then there is a Qm+ 2 such that (6.19) is satisfied for R = m + 2.

Thus, the lemma will be proved if we show that (6.21) has a solution
of the form
cfJ'm+i (p) = Lj bjp1,
lil=m+i
where bi is a smooth vector function of a and t. Let
Om+i (p) = - Lj ajp 1•
I i l=m+1
'Then ff> m+I satisfies (6.21) if b1 , ••. , bn satisfies the equations
[ k,;;j
~ k!J~k)! CPa) 1k 1Jbi=ah j=1, ... , n. (6.22)
So it suffices to show that for any j the matrix

def ~ j! t Ik I
Aj= L..J k! (j-k)! (~a)
k~j
is non-degenerate. But note that

~ j!xlkl Iii*
~ k! (j-k)! = (1 +x) '
k~j
hence
det Ai = det [(1 + t~a) 1 i 1] = [det (1 + ~a)] 1 i 1:i= 0
.as required.
Lemma 6.5. Let {8\} be the sequence of Lemma 6.4, and let
N
~N = Lj ff>d~).
k=i
* This follows from the generalized binomial formula

'/
(x + yf. = ~ kl u'-k)! .;ckyi-Jt;
k~j
:here j and k are multi-indices of length n, and x =- (x1 , ••• , xn), !I = (y 1 ,

·• · ., Yn).
CH. IV. GENERALIZED HALVIILTON-JAGOBI EQUATION'S 411
Then the operator

N
R",= ");
I\ ~
_1 p)
i! PN
(-a )iar:x (6.23)
I i 1=0
satisfies the conditions
I' I'
LNRN= 1 +e 1 ,
A
(6.24)
I' I'
RNLN=1-e 2 ,
A
(6.25)
u•here
2N
I r 1=1
a: r '
2N
~2 = ~ br (a, t) (
1r 1=1
N+1
Gr (a, t) = OD ( h -2-) ,
Proof. First we check that Rf.v is a right "almost in verse" of Lf.v,
i.e., (6.24) holds. We have
a: r ~ i- ~~ (a~ )
N N
Lf.vRf.v = ~ :, ~k ( t =
1k 1=0 111=0
S N -lilA!
- "' )1 "' 1 ~ (-f)-)l+k-j
~ -..J ~
k
- LJ j!(k-j)!Z! or:x1 ar:x •
I k 1=0 ll 1=0 O~j~k
Introducing the index r = l + k- j instead of j, by a change of

order of summation we obtain
N N
L"N R"N-- LJ
.....::, "'
~ "'
LJ
pk
(l+k-r)! (r-l)! l! X
I k 1=0 111=0 l~r~l+k
fjll+h-r lp~
~__,....
( f) r
= 1+
2N
"' ( f) )r ,
ar:x 1 + -r
X ....,.k___:_::.:._ -:;;::;-)
u~ LJ
ar -
0 ~~
I r 1=1
where
f)ll+h-r lp~
"'
LJ
P"
(l+k-r)! (r-l)! l! f)r:xl+k-r
(6.26)
O~l~rh>r-l
lli>Nihi~N
Now introduce the index j = l +k - r instead of k, then (6.26)

becomes
N+lll-lrl ~i+r-l ali I~Jv
ar= ~ ~ j! (r-l)l ll aa)
O<Sl<Sr I i l=O
lli<SN
~r-l
N+llt-lrl . aliiRl
. (~J PN
= ~ ll (r-l)l ~ 71 arxi •
(6.27)
O<Sl~r I i 1=0
lli<SN
Let us show that
N . alii~! ( N+1)
" ~ N = (- ~)! + 0 h-2- • (6.28)
.LJ 1! arxi n
li 1=0
To do this, we observe that since ~ = 0 n (h 112 ) and the relation
1 +
~a = 1 - (qa +
Za)-1 (za - 'l'a) +On (h 112 ) =
= (qa Za)-l (qa +
'l'a) +On (h 112 ) +
implies that the matrix 1 + ~a
is non-degenerate on the zero-set
of the dissipation, it follows from Lemma 6.4 and the definition of
~ N that
Ij I N+ 1)
"
N
.LJ
~ a
7!
.
arx1
~N = _ ~ + 0 n ( h-2- .
I ii=O
Thus, to prove (6.28) it is sufficient to show that
"
N .
~ o ~N-
,I j 1 1 [
"
N .
!:_ alJI~N
. J=0
l ( N+ 1
h-z). (6.29)
.LJ 1! 8rx1 .LJ Jl orxJ n
Ji 1=0 u 1=0
Let f (a, t) be a smooth vector function. Consider the following
polynomial F 1 (~) with coefficients smoothly dependent on a and t:
F 1 (~) =
N
LJ)1 --:-
~i 8111
--. -
. 1
1 [ N
"
~i 01J'f
-. --.
. ]! = "
.LJ c,. (a, ,.
t) ~ .
l 1 8rx1 .LJ 1.1 fJrxJ
I i 1=0 1i 1=0 · h
If x runs over a small neighborhood of 0 in Rn, then

F!(x)=f 1 (a+x, t)+O(IxiN+t)-
- [/(a+ x, t) +0 (I x IN+i)]z = 0 (I x IN+t).
It follows that ck (a, t) = 0 for I k I~ N, so
N+1)
Ft(~) =On ( h-2- .
CH. IV. GENERALIZED HAIMILTON-JACOBI EQUATrONS 413
Putting f = ~N in this relation gives (6.29). Equation (6.28) implies

that
(6.30)
for l~r. Putting (6.30) into (6.27), we see that

- 'V flr-l l ( N~1)-
ar- ~ ll(r-l)l (-~)+On h -
l::;;r
= (~- Br +On ( h-N+1) =On ( h-N+1)

2- 2-
for l ~ r. To complete the proof of (6.24) it remains to note that

N+1
in the case I r I > N the estimate a 7 = 0 n (ft2) obviously holds.
Turning to the proof of (6.25) we note that only the following
two properties of ~ (a, t) have been used for constructing the right
"almost-inverse" of L'fv: (1) B =On (h 112 ) and (2) det (1 ~a)aErt =I= 0. +
The function BN (a, t) has these properties too, for
PN= -(1+Bat 1 ~+0n(h),
(1 + a:: )
= 1- (1 + Bat 1 Bu +On (h 112 ) = (1 + Bd- 1 +Ov (h 112 ),
hence there is a function i:f = Ov (h 112) such that the operator
N -
"" -
c N- ""' j!
..W ~ (_?___
u'a )j
15!=0
is a right "almost-inverse" of R'fv, that is,
'V
RNSN=1
'V
+e ~
3, (6.31)
where
2N
e3 = 2..J ~1 dr (a, t) ( a
iia )r ,
I rl=1
( N+1)
dr = On h---zs- .
The multiplication of (6.24) from the left by R'fv and from the
right by gives srv
(6.32)
414 /OPERATIONAL METHODS
It follows from (6.31) and (6.32) that

(6.33)
Set
(6.34)
Because of (6.33), e is
2 a differential operator of order 2N:
2N
e=
2 ~ bi ( a~ ) J•
iJ i=O
It is easily seen from (6.34) that bi= On (h N~t), which proves the
lemma.
Theorem 6.1. If .!l

0 (h '; )
n m ,.._. 0, then, for any function v0 (a, h)
of the form
On (hm/2)
vo(a, h)= hm/2 '
the function
t
- ~ G(p (a, 1:), q (a, 1:), 1:) d't
cp(a, t, h)= V J 1(a, t)
L'/nv0 (a, h)e 0
satisfies the transfer equation with dissipation in the zero appro-

ximation.
Note 6.1. If the condition .Jl Onh~~12 ) ,.._. 0 is not satisfied, then
the function cp (a, t, h) introduced in the theorem can be used for
solving the transfer equation by the theory of perturbations (see
Theorem 6.2 below).
Proof of Theorem 6.1. Set
t
- ~ G(p (a, 1:), q (a, 1:), 1:) a't
v(a, t, h)=v0 (a, h)e o
Then v satisfies the equation v+Gv=O, and
v = h- ~On (h ~).
CH. IV. GENERALIZED HAG.YIILTON-JAGOBI EQUAT:DONIS 415-
By Lemma 6.3 we have

(P" +G) L;;,v= {a, (!+G) +e+
+ .:S g" ( 8~ ( + [G, L;;,]} v =
lhl=m+i
= { e+ .:S
lh i=m+i
g" ( r
8~ + [G, L;;,]} v.
It is easy to see that
[ 8+ .:s gh ( 8~ ( ] v,......, 0.
lkl=m+i
Moreover,
m
[G, L~] = .:S G j; ( 8~ )i-
Ui=O
m
Bi ali-"IG
- .:S .:S kl(j-k)! 00)-k (a~)"==
li 1=0 h~j
=- .:S';, .:S Bi
k! (j-k)!
ali-"IG ( a )"
aai-k aa E<f?, (6.35)
li 1=0 h<i
hence [G, L;;,] v ,......, 0. So 'ljJ = L;;,v satisfies (6.8). The theorem is
proved.
Now let us tackle the transfer equation with dissipation in the
1
app.r;oximation of order N. As before, we set ') 'ljJ = f)D (hN/2), (6.36)
(note (6.7), (6.8)).
We shall look for 'ljJ in the form
ljJ = Lfu-v (6.37)
with m to be determined later. Putting (6.37) into (6.36}, we obtain
by Lemma 6.3
[ L;;, (!+G)+~'+ .:S gk ( 8~ )"]v=fJD(hN12 ), (6.38)
1kl=m+1
where
e' = e+ [G, L;;,] + ,95>' L;;, E 3', (6.39)
the operator e and the functions gh being those of Lemma 6.3.

If v= 0 :~~~ 2 ), then it follows, because of gk=On(hm 12 ), that we
may drop the term ~ gh ( 8~ ) h v in (6.38) provided that

lhl=m+1
m?i:;N +s.
Now, by virtue of Lemma 6.5, we obtain
AI v v A AI v A" A A'
e = (LmRm- e!) e =LmB -e 1 E,
where
e" = Rinel Eff' (6.40)
the operator e being as in Lemma
1 6.5, in particular,
A AI On(h 8 12 ) _rc. (hN/2) (6.41)
e1 e h 812 - v n ,
whenever m "?;; N + s + k - 1, where k is the type of .!!. Noting

that
L'/nfJ D (hN/2) = f) D (hN/2),
we arrive at
(! +G+e") V=f:in(hN;z). (6.42)
We can construct a function v satisfying (6.42) by the perturbation

theory, '8" being considered as a perturbation. We have the following
result.
Lemma 6.6. Let I be the operator of Lemma 6.1, let eE <fP, and set
t
JG dt - s G dt
6= - e0 o eo e 0
Then for any

On (hso/2)
Vo (a, h)= hso/2
the function
t
- ~ G dt /!, A '
v= e 0 2,; (!6) 1 vo
.i=O
satisfies the relation
n(N, s))
(:t +G +B) v=Bn ( h - 2- ,
CH. IV. GENEHAUZED HAIMILTON-JAGOBI EQUATl'ONS 417
where n is a E!'-function of 6.
Proof. We have
t
-I G dt
• A A N
~-= -e 0 8(18) v0 ,
and the lemma is proved.

Now it is easy to formulate the main result of this section.
Theorem 6.2. Let .11: in (6.1) be of type k. Then for any function
Vo (a, h) of the form Vo = h-sofZQD (h' 012 ), the following function
satisfies (6.2):
t
1 -) G(p(a, 1:), q(a, 1:), 1:) d1:
rp(a, t, h)= L'/ne 0 X

V .J (a, t)
N(h+2)+s 0
X ~ (Ib)iv 0 (a, h),

i·-·0
where
i t
, ~ G dt , - ~ G nt
8 ~"' - e 0 o c," o e 0
(with ~" given by (6.40)) and

m = N (k + 1) 2 + s 0 (k + 1) +max {k- 1, 0}. (6.43)
Proof. Essentially, the theorem has already been proved above.
There are only two points to be noted.
(1) By Lemma 6.6 the function
t
- ) G dt N(h+2)+so A •
v= e 0 2] (18)'v 0 (6.44)
j=O
'latisfles (6.42). In fact, if .Jl; is of type k, then it is easy to see that
the same is true of 6. It follows from the proof of Lemma 6.1 that
( r-s
if r~ s
n(r, s)=·) k+2
CO ifr<s
is a cfJ'-function of fJ, so
n (N (k + 2) + s0 , s 0 ) = N.
(2) It is easy to see that the function (6.44) has the form
v = h-s/20n (hs/2)
with s = k [N (k 2) s0 ] + +
s0 • +
For this s, (6.41) is satisfied whenever m ;;;:, N (k + 1) +
2
+ s0 (k 1) + +
k - 1, and the condition
'\:l ( a )k On (hs/2)
LJ gk 7fa hs/2
lk J=m+i
is satisfied whenever m;;;:: N (k + 1) + s

2 0 (k + 1).
Note 6.2. Let n (r, s) be a cfJ'-function of 6, and let r = r (s 0 , N)
satisfy the equation n (r, s 0 ) = N. Then Theorem 6.2 remains valid
N(k+2l+so r(so, N)
if we replace 2J
j=O
by 2J
i=O
and (6.43) by
m = max {r + s, r + s + k - 1 }.
Note 6.3. Let :z

= ( 8~ 1 , • • • , afn) be a set of commuting complex
vector fields on uv 'which are linearly independent at every point
(a· complex vector field on a real manifold M is an element of the
complexification of coo module of all vector fields on M). Theorem 6.2
remains valid if one replaces 8/arx by 8/al in the definitions of
J (a, t), Lin and 8, for instance:
m .
Lv
m
= '\:l <Bv)J
LJ j!
(_!_)i
az '
ri
li 1=0
where ~v= - [ o(qaiz) (z-y).
V. CANONICA L OPERATO R ON A LAGRANGE A.N
MANIFOLD WITH A COMPLEX GERJti
AND PROOF OF THE MAIN 'J!HEOREJ tf
Notation. In this chapter we shall use the spaces Hi, introduced

in Sec. 11 of Chapter II:
II I (x) Ilui,= (1 I +I x 12)kfz ( 1 +I a~ 12) 1/2 I (x) IILz.
We shall write H~ = Hk, HY = H 1• Thus H 1 is the Sobolev space
W!. The symbol Hf; ~ will denote the completion of the Schwartz
space S with respect to the norm
lll(x)ll;i1· h = ~ 1\Cir(:x)BI(x)IJ~z'
l,m jaj~l
I f31~m
where h is a (small) positive parameter.
Sec. 1. Quantum Bypassing Focuses Operation

In this section we shall give an asymptotic expansion for h-+ 0
of the Fourier transform of a function of the form
[ V 1 2._ <J>'I'(a)
eh cp (a)
J , Im cD'~';;;:::,O, x ERn.
J (a) - a=n);I(x)
We have to do this to define the canonical operator as we have

already done in the case Im <D'~' = 0, n = 1 (see Chapter III).
In the case under consideration, a method which is similar to the
saddle-point method is likely to be applied to obtain an asymptotic
.2.s
expansion of the integral, for, in general, e h decreases rapidly
ash-+ +O. However, a difficulty arises because it would be desirable
to obtain as the result an expression similar to the integrand with
the phase and the Jacobian replaced by those corresponding to the
zone Q1 , if tho Fourier transformation Fx[-+ P[ is considered (it is
27*
420 OPERATlONAL METHODS
such a geometric interpretation of the asymptotic expansion of the

integral (3.1) that has led in Chapter II to the definition of the cano-
nical operator and thus to the global construction of solutions of
(pseudo)differontial equations). It appears to be very difficult to do
this by the saddle-point method because a complex-valued function
is not uniquely expressible as the sum of an s-action, a f.t-action and
a potential. This difficulty will be avoided by a method similar to
that of Sec. 8 of Introduction. The method is based on a special
transformation which is connected with the bypassing focuses oper-
ation and will be called the quantum bypassing focuses operation.
We shall first introduce some terminology.
Definition 1.1. Let {<pi (a, h)} be a family of smooth functions of a
dependent on the standard small positive parameter h, with the index j
running over a certain set J. Given a smooth non-negative function
D (a), the series 2j <Jli (a, h) will be called D-asymptotic if
jEJ
(i) <Jli =(3D (h vi), supp <Jli lying in a compact set independent
of j and h;
(ii) for any natural number m, there is such a finite set J 0 c: J
that <Jli = C3 D (hm) for j ~ J o·
Definition 1.2. Let { <Jli (x, h)} be a family of smooth functions of
x E Rn dependent on a small positive parameter h, with the index j
running over a certain set J. The series 2j <Jli (x, h) will be called
iEJ
h-asymptotic if
(i) for any j, k and l there is such a number mik 1 that
II <Jlj II Cz(Rn)
k = o (hmikl)
(for the definition of C~ see Sec. 11 of Chapter II);
(ii) for any N, k and l, there is such a finite set J 0 c: J that
k n =O(hN) for j~J 0 •
li<Jlill Cz(R)
In the following we shall mean, by an asymptotic series, either

D-asymptotic series, or an h-asymptotic series.
Example 1.1. Let <p (a) E Co and set ~ 1 = - C[ 1 (w1 , zr), then
the series
dcf oo
Lr<p= ~ -J.1. 7kX
f3i ( a ) i <p
I ii=O
is D-asymptotic.
CH. V. CANONICAL OPERAT'OR ON A LAGRANGEAN MANIFOIJD 421
Example 1.2. Let A E tP, QJ (a) E CO', and let I be the same opera-
tor as in Sec. 6 of Chapter IV. Then the series
is D-asymptotic by Lemma 6.1 of Chapter IV.

We now define addition and multiplication of asymptotic series.
Definition 1.3. The sum QJ + 'ljJ of two asymptotic series, QJ = 2] QJJ
iEJ
and 'P = 2] 'ljJk, is the following asymptotic series:
· kEK
QJ-f-'ljJ= 2] Xz,
IE(JVK)
where the sign V stands for disjoint union and

QJz for l EJ
{
Xz = 'ljJ 1 for ZEK.
.
Let -'ljJ be the asymptotic series 2] (-'lj.lk). The difference QJ - ~
kEK
is the asymptotic series QJ (-'ljJ). +
The product QJ'IjJ is the following asymptotic series:
qJ'IjJ= 2] (jlj'lj.lk.
(j, k)E(JXK)
The reader is invited to check and see that this definition is correct,
i.e., to verify that both QJ + 'ljJ and QJ'IjJ are asymptotic.
Definition 1.4. Let QJ E 2] Qlj and 'ljJ = 2] 'ljJ 1 be two asymptotic
j j
series, with the index j running over one and the same set J. Then the
term-by-term sum QJ +
'ljJ is the following asymptotic series:
QJ+'P=lJ (QJJ+'Pi)·
iEJ
Definition 1.5. We write:
I
fp(a)=On(h 8 ), if e--h-cp(a) ~ch•;
D(a) I
io
QJ(a, h)=On(h 8 ), if QJ(a, h)= 2] QJJ(a)hi/2,
i=O
zl'here Qlj (a)= On ( ·- ~) for i~2s;

k
QJ(a, h)=On(h 8 ) if qJ=h-20n(h8 +k/Z),
for some integer k > 0. Here all the estimates are assumed to be locally
uniform in a.
Definition 1.6. We say that a D-asymptotic series 2j <Jli is weakly
jEJ
equivalent to zero if for any natural number N, there is such a finite
set that
2J Cf!i=c:>n(hN),
JEJ1
whenever J 1 c J is any finite set containing J 0 •
An h-asymptotic series 2J
Cf!i will be called weakly equivalent to
iEJ
zero if, for any natural numbers N and k, there is such a finite set
J 0 c J that
II iEJ1
2J Cf!i lie~ = 0 (hN),
whenever J 1 c J is any finite set containing J 0 •

We write cp ,..., 0 to denote that cp is weakly equivalent to zero.
Definition 1.7. We say that an asymptotic series 2J
iE.f
Cf!i is equivalent
to zero, cp ~ 0 in symbol, if for any multi-index k,
~
iEJ
(a: r Cf!j,..., 0 ·
Two asymptotic series, cp and 'ljJ, are said to be weakly equivalent,
cp ,.., 'ljJ in symbol, if cp _:__ 'ljJ ,.., 0, and equivalent, cp ~ 'ljJ in symbol,
if cp _:__'ljJ ~ 0.
The equivalence relation ~ is compatible with the arithmetical
operations defined above:
. .
((cpl ~ Cf!2) and ('IJJt ~ 'IJJ2)) '* ((cpl + 'IJJ1) ~ Cf!2 + 'IJJ2)
and (cpl'ljJl ~ Cf!2'1JJ2).
Moreover, the term-by-term sum of two asymptotic series, 2J Cf!i
iEJ
and 2J 'IJJi, is equivalent to their sum. Because of this, we shall, as
iEJ
a rule, write +
in place of +
and use the standard symbol 2J to
denote the summation operator for asymptotic series.
Definition 1.8. An asymptotic series cp = 2J Cf!i is called an asymp-
iEJ
totic solution of the equation Acp = 0 if the series 2J
jEJ
Acpi is asymptotic
and equivalent to zero.
CH. V. CANONICAL OPERAT'OR ON A LAGRANGEAN MANI:Il'OLD 423
Example 1.3. The series

00
J- 112 L ~ (I8)iv(a)
j=O
is a D-asymptotic solution of the transfer equation with dissipation

for a non-singular patch (see Theorem 6.2 of Chapter IV).
Definition 1.9. A D-asymptotic differential operator L is such an
operator series L = ~ Li that if qJ = ~ fPk is a D-asymptotic
jEJ k=K
series, then the series
LfP = ~ ~ LifP~t
jE.J kEK
is D-asymptotic, and LqJ ~ 0 whenever qJ = 0.

In the case where Li is differential for every j, L will be called a differ-
ential D-asymptotic operator.
Definition 1.10. We shall say that the series ~ fPi (a, h) is asymp-
iEJ
totic at a 0 if there is a function e (a) E Co equal to 1 near a 0 such that
the series
~e(a)cpj(a, h)
iE.J
is asymptotic.
Definition 1.11. Let both cp and 'ljJ be asymptotic at a 0 • We say that
IP and 1p are equivalent (respectively weakly equivalent) at a 0 , cp~ 'ljJ
(respectively cp ~ ¢) in symbol, if there is a function e (a) E C't' equal
to 1 near a 0 such that ecp ~ e¢ (respectively eqJ,...., e'ljl).
Definition 1.12. Let cp E ~ cpi be a D-asymptotic series. If cp is
jEJ
not equivalent to zero, then the order of ([J, ord qJ in symbol, is the smallest
half-integer k with the following property: there is such a finite subset
J' c J that
LJ cpi=F<9D(hk+1!2), cpi=<9D(hk+1!2) for j~J'.
jEJ'
If cp ~ 0, then we set ord rp = oo.

Definition 1.13. Set p (cp) = 2J
iEJo
cph where J 0 is the set of all j
such that ord qJ i ~ ord cp (it is easy to see that ord p (cp) = ord qJ). Let
k2
p (cp) = LJ hk12 ck (a).
k=kt
The smallest k with the property ord (hk 12 ck (a)) = ord cp is called
the type of.cp. The principal monomial of cp, m (cp] in symbol, is hk 12 ck (a),
where k is the type of cp.
If ord cp = oo, then we set m [cp) = 0.
Problem 1.1. Let A and I be the operators of Example 1.2. Then
00
~ (IA)J is a D-asymptotic differential operator preserving the

}=0
principal monomial.
Definition 1.14. A D-asymptotic operator A is called a quasi-identity
operator if
m [Acp] = m [cp] + hkOn (h"+ 1/ 2 )
for any D-asymptotic series cp, where k is the type of (p and s k == +
= ord cp.
Given a Lagrangean manifold with a complex germ, let a 0 be in
Q n QI n r, and suppose that the assumptions of Theorem 5.1
of Chapter IV are satisfied. Let a (x) be the solution of the equation
q (a) = x. For the integral
~ ell(a) - 1/2 }
I(xr, Sy, h)=Fxy-+6y { e [J(a)] Lcp(a) a=a(x)=
m inm
= (2nh)- 2 e- -~..- X
X Jr e - ~ (Xp 6y>{
e
~ <ll(a)(J (
a )J-lf2 L cp ( a )~J a=a(x) d :xy,
'
Rm
where L is the D-asymptotic differential operator of Example 1.1,

(J(a)j 112 =JJ(a)J 112 exp ( ~ ArgJ(a)),
Arg J (a) is continuous near a 0 , and cp is supported near a 0 , we
shall obtain an h-asymptotic expansion, i.e., an h-asymptotic series
which is equivalent to I (x1 , £1 , h) and does not involve integration
(that series will be defined via a D-asymptotic differential operator
applied to cp).
Let H be a bypassing focuses Hamiltonian such as that of Theo-
rem 5.1 of Chapter IV:
H = 21 { (p 1,u -1 eup
'
1) + (q1u -1
euq1 ) } , (1.1)
and let J (a, t) be the complex Jacobian corresponding to the bypas-
sing focuses operation associated with H. Further let Arg .T (a, t)
be the function determined by the following conditions:
(i) Arg J (a, t) is continuous in t,
CH. V. GANONI•CAL OPERATOR ON A LAGRANGEAN MANIF'OIJD 425
(ii) J (a, t) = I J (a, t) I ei Arg J(a,t>,

(iii) Arg J (a, 0) = Arg J (a).
Proposition 1.1. Under the above assumptions, for any «p (a) E Co
supported near a 0 , we have
t_ h),..... hi i (
<J>(a)- 2Arg J a, 2
n) - inu
-2-
I (XJ, <;,1 , ,.....e X
X IJ1 (a) l- 112 LrVrCP (a) la=al(xl' !:.y>' (1.2)
where a 1 (x 1 , £1 ) is the solution of the system

q1 (a)=xr,
{
Py (a)= Sf,
a is the number of negative eigenvalues of e, Jr = det C1 , and vr

is a D-asymptotic differential quasi-identity operator.
Note. Since J (a, n/2) = (-1)a J 1 (a), the function

e ~ [Arg J(et, :TI/2}+:nu] I J 1 (a ) 11/2 rs
. (a ) .
a b ranc h of v -1 J 0
The proof of the proposition will be by the technique which

we have used in Sec. 8 of Introduction to obtain an asymptotic
expansion for h-+ 0 of the integral I (h).
Consider the Cauchy problem
- ih ~
()t
+H ( - ih _!_
ax ' x) •h
'Y·
= 0'
{ (1.3)
ljl(x, 0)='Jl 0 (x),
where H is the bypassing focuses Hamiltonian. The solution of (1.2)

is given by
i:nm i:na
--4- +---:!
'Jl (x, t) = e(2nh sin t)m/ 2 X
\ ~ (sint)-l[H(y,x)cost-(xpu-leuvy>l ( )d ( . )
X
Je 'Po xr, Yy Yy 1 4
where m is the number of elements ofT. Note that if '¢ 0 is an asymp-

totic series, then the last formula defines an asymptotic series which
satisfies (1.3) (both the equation and the initial condition) term by
term. It is natural to regard such a series as a precise solution of the
Cauchy problem in contrast to asymptotic solutions introduced in
Definition 1.8.
Putting t = n/2 in (1.4) we obtain
(1.5)
if
1Jlo(x)=[e- i~a + ~ <t>(a)- ~ ArgJ(a)IJ(a)ri/2 Lcp(a)Ja=a(x)· (1.6)
So the problem of constructing an asymptotic expansion for
I (x1 , ~ 1 , h) is reduced to that of the solution of the Cauchy problem
(1.3). The standard way for obtaining an asymptotic expansion of
the solution of a differential equation is as follows:
(a) one constructs an asymptotic solution of this equation;
(b) by an appropriate estimate of the solution of the corresponding
non-homogeneous Cauchy problem one proves that the asymptotic
solution is equivalent to the precise one.
Therefore, we give some estimates of solutions of the Schrodinger
equation corresponding to the bypassing focuses Hamiltonian.
Lemma 1.1. 1°. Let 'ljl 0 (x), f (x, t), x E Rm, t E [ 0, ~ J be smooth
functions vanishing for Ix I> R, and let 'ljJ (x, t) be the solution of the
I
Cauchy problem
'h o'ljl (x, t) L .
- ~ at '
I + ~ j~8j [ ( -ih r
:X + x 2 ]1Jl(X, t)=f(x, f) (1. 7)
'ljJ (x, 0) = 'ljl 0 (x),

u here ei = + 1.
Then the following estimate holds for sufficiently large l:
II'\' (x, t) llc?(R~t)~ck, z (R) h -k-i- [~]X

X (II 'ljl 0 lick+! +h- 1 max II x llck+l(Rm)) .
O<t<.::!. X
--z
2°. Let ¢ 0 ES, f=O. Then
-~-k
li¢(x, t)ll k( m)~ck.z(t)h
Cz Rx
2 II1Jlollc!+k
l+k+m
, tE(O, ~].
where ch, 1 (t) is a continuous function.
CH. V. CANONICAL OPERATOR ON A LAGRANGEAN MANIF-OLD 427
Froof. (1) Let f = 0. Then

_ inm + ;n<J
e t, 2
'ljl(x, t)= X
(2rrh sin t)m/ 2
X Jexp{hs~nJ~ Ej ( x]~YJ cost-XiYi) ]} ¢ 0 (y)dy=

nm i=1
def 1 imn ina ) . - ..::::_

= (2rt)m/2 exp ( --4-+-2- (hsmt) 2 J[¢ol- (1.8)
Set
m 2 2
""' \X,
'-.V
'· y, t)- "V Ej. ( -xj-+
_ L.! Yj
2 - cost-,?;.j Yi ) .
j=1
Then we have
a<D
oyi =ei(Yicost-xi), (1.9)
· h es on 1y 1'f y = x(cos t.
a<D vams
so Ty
Let Q (x, t) denote the ball
{ yE x I< lxl+i}
Rml y __t . 2 t
cos. cos
Then o<D/oy=/=0 for y(:Rm"'Q(x, t). Therefore the integral
/1[¢ 0 ]= J exp{hs!nt<D(x,y,t)}¢ 0 (y)dy

nmxQ(x, t)
can be transformed by partial integration as follows:

. i i)g
l1 [lJJol = - i ) exp ( -h-.-<D(x,
Stll t
y, t))'ljlu(Y)-
8 n dcr+
iJQ{x, t)
+ h sin t exp ( h s!n t <D (x, y, t)) L¢ 0 (y) dy,
where
(1.10)
dcr is the area element on the sphere 8Q (x, t), and 8/on is the
exterior normal derivation. Repeating integration by parts M times
428 OPERATIONAL METHOThS
yields
/1 ['¢ol = - i ) exp(hs~nt<D(x,y,t))x
oQ(x, t)
M-1
X .L~ [(h sin t) 1 L 1¢ 0 (y)]

1=0
!! da +
+ (h sin t)M exp ( i t
h sin <D (x, y, t) ) L M '¢o (y) dy.
)
It follows from (1.9) and (1.10) by a simple calculation that

ILM'¢ol~constlycost-xrM sup
1i \:SM
I(_!_)i'¢
ay
(y)l 0
for x, y and t such that I y cost- x J~a > 0, const depending on a.

Hence
I It (lJlcH~.const r~
M-1
l=O
(h sin t/ (I X 1-i-1) -!-1+ 2
m
11 '¢u llci(Q(x, t)) +
+(hsint)M(Ixi+1)-M f sup [(:)j'¢ 0 (y),dy]. (1.11)

J Ij
llm
\:SM y
As to the integral
/2['¢ol=l['¢ol-/d'¢ol= ~ exp (~<D(x, y, t))'¢o(y)dy,
Q(x, t)
it is easy to obtain the following estimate:
I /z ['ljld I~ J I '¢o (y) I dy~const·J x !

Q(x, t)
111
• sup I tl'o (y)
yEQ(x, t)
I~
~const (I xI+ 1)m-z.
Combining the estimates for / 1 ['¢o] and / 2 ['¢ol gives
I I [¢ull~const (I X I+ 1)-M II l\lo II~~I+m" (1.12)
Now derivating (1.8), we see that

as the sum of terms of the form
(:X r ¢ (x, t) can be expressed
m
c (x, t) (h sin tf 2 - 1 xkl[ynljl 0],
where
; I j I~ z ~I j I, In I~ 2Z- I j I, I k I= 2Z-1 n I- U J,
and c (x, t) is a smooth function. Hence 2° follows from (1.12).
CH. V. CANONICAL OPERATOR ON A LAGRANGEAN MANIFIODD 421}
(2) Now let the assumptions of 1° be satisfied. First we estimate

'\jJ for large x. We have
'\jJ = tj) + tj),
where 1j) and '¢ are the solutions of (1. 7) with f replaced by 0 and tj)
replaced by 0, respectively.
If I xI ~ .. • 2R +
1, then'\jJ 0 vanishes on Q (x, t), so 1 2 ['1jJ 0 ] = 0 and.
I 11 ['ljJ 0 ] I ~ const (h sin t)M I x 1-M II '\jJ 0 llcM·
Therefore, we can estimate ~ for I x I ;;;:: 2R + 1 as follows:
M-~-lil
~const (h sin t) 2 I x 11 i !-M II ¢ 0 llcM· (1.13)
We further have
t
~(x, t)= ~) '\jJ(x, t, -c)dT,
0
where ¢ (x, t, -c) is the solution of the Cauchy problem

{ m
J -ih ~~ + ~ ~ [( -ih ~ ) 2 +x2J¢=0,

l ¢ (x, t, -c) i=f (x, -c).
Estimating ljJ (x, t, -c) by (1.13) and integrating with respect to ,;

on [0, t] we obtain for I x 1;;;::2R + 1:
J(;:rljl(x, t)~~CM,j(R)hM- ~ -lil-itM- ~ -til+i X

Xlxl 1 ii-M max 11/(:r, t)llcM(n~')'
O<t< n
--~
Combining the lasl estimate wilh (1.13) gives for I x j;;;:: 2R + 1
J(!r¢(x, t)J~cJI[.j(R)hM- ~ -utlxl'j'-1\fx

X (ll¢ollcM+mfxll/l\cM(n~))· (1.14)
430 OPERATJ!ONAL METH~
Now lot us give estimates on the whole of nn. The following

is well known:
t
II¢ IIL 2 (n~) ~II ¢o IIL2 + ! ) II f (x, T) IIL2 (n~) dT.

r
0
(:X
(:x r
Set ¢Ul= ¢. Then ¢0> satisfies (1.7) with ¢o replaced by
¢o and f replaced by a linear combination of the functions
(:x)if, ¢< 1> and Xsljl with !ZI=UI-2, IPI=UI-1, s--

=1, 2, ... , n. Using this fact we provo by induction that
II t~ IlL (nm)~const h-I i 1 (II ¢o II . [ m] +

2 x e'' I+ 2
+ h-t max II f II
<"'"'
. m] ), (1.Hi)
o::;;t..., 2 eIJI+[T (nm)
x
with canst depending on j and R. For j = 0, (1.16) follows from

(1.15) by noting that any function cp with support in a ball Qn c nn
of radius R i'atisfies the inequality
II cp IIL2(Hm)~ II cp lie. Vh 12 ,
where V R is the volume of ITn. Suppose (1.16) holds for I j I~N
and estimate ¢U•> with I k 1= N+ 1. vVe have
II (:X r ¢o III,z ~canst II ¢o lieN+!,
II ( i}~ r IIL2(R~) ~canst

f II f lleN+i(n~)·
If Ill = N -1, IpI = N, then
II ¢<1) IIL2(n~)~const ht-N (II ¢o lieN+ [ ~] +

+ h- 1 max II I II [ m 1 -~ )
t e N+ 2 (nm)
X
by the induction hypothesis, and
llx8'1jJ<Pli!Lmm)~consth-N(
2
11'1-'oll
X
[m]+h- 1 maxllfll
N+- t
[m]
N+- }
e 2 C 2 (R~
CH. V. CANONICAL OPERATOR ON A LAGRANGEAN MANIJ<'OLD 431
by (1.14) (with M = N + [; J+ 1) and the induction hypothe-

sis. These estimates imply that
1/ ljJU<l I!£2(Rm)
x
~con st h -N - 1 ( II 'Po II [ m] +
CN+ z +1
+h-lmaxll/lllii+2 [r:-']+1 )•
t C 2 (R~)
which proves (1.16).
It follows from (1.16) by the Sobolev :embedding theorem that
llJ/3)(x, t)J~consth-(lii+[;]+i)(ll'l'oll . [m] +

CIJI+2 2 +1
+h-1maxllflllil+2[~]+1 )·
t G 2 (R~')
This estimate together with (1.14) prove 1".
Corollary 1.1. Let lj1 0 (x, h) be an h-asymptotic series equivalent
to zero. Then Fx-....
I
~-'Po~ 0.
I
Proof. Note that

Fxr~:r lj10 (x, h)= 1jJ ( x1 , £1, ~ , h) ,
where ljl(x, t, h) is the solution of (1.3) with e=1, u=1, and
use 2° of Lemma 1.1.
Proof of Proposition 1.1. Parallel with the precise solution
of (1.2) consider an asymptotic solution of the following form:
i
hs (x, t)
'ljlr(x, t, h)= e v (x, t, h). (1.17)
We have
a a ..i..s ~
[ -ihTt+H ( -iha;, x)] '!' 1 =eh {Hv+S 1v-ihv 1},
where
fi = H (Sx, x)- ih [-} tr (H ppSxx) +
Set
+(Hp(Sx, x), :x) ]- r tr (HPP :;2 ) •
(At', rt')=(gkAn, (dgic)rn).

Let <D(a, t) be the phase on At', and let II: (a, t)-+(x, t) be the
diffeomorphism corresponding to a non-singular patch of the non-
singular zone. Put S =<Do n- 1 • We have already shown that

S (x, t) satisfies the Hamilton-Jacobi equation with dissipation
as def 3/2
at+H(Sx, x) =g(x, t)=0s2 (h ).
Substitution of (1.17) into (1.2) leads by noting the last equa-
tion for S to the following transfer equation to be satisfied by v:
Vt+(Hp(Sx, x), Vx)+[~ tr(HppSxx)+.Jl: Jv~O, (1.18)
where
i ih fj2 )
.Jl; = h g (x, t) - T tr ( H Jl P fJxZ E,fJ> •
By Example 1.3 this equation has (in the coordinates ex, t) a D-
asymptotic solution of the form
i 00
v=Jl(ex, t)r 112 e--zArg.J(a,t)Lh (IB)ivo(ex) (1.19)

j=O
with 6 E ff' determined by the Lagrangean manifold with the complex

germ and the bypassing focuses Hamiltonian. (Note thaL if p (ex, h)
is aD-asymptotic series equivalent to zero, then exp ( ~ <1> (ex)) p (ex, h)
is an h-asymptotic series equivalent to zero.) In order to satisfythe
initial condition '¢tit=O ~ '¢ 0 , where '¢ 0 is defined by (1.6), we put
in
--(J
v0 (ex)=e 2 cp(ex). (1.20)
Now we observe that
so L turns for t= ~ into Lr; moreover,
J (ex, ~) =(-1t lr (ex).

Recall further that
<t>( ex, ~)=(!)I( ex)
and
q (ex,~) ={qr(ex), u- 1eup1 (ex)}.
Therefore, we obtain the following expression of ;p 1 (.r, t, h) for
t=n/2 and x={qr(ex), u- 1eup1 (ex)}:
CH. \". CANONICAL OPERAT'OR ON A LAGRANGEAN MANIFOLD 433
where
00
v1 = .:2}
J=O
(/b)j J-..::..2
t-
Note here that u1 satisfies the requirement of the proposition (see

Problem 1.1).
To complete the proof, we shall show that '¢ 1 (x, t, h) ~
~ 1p (x, t, h). Let
'¢ = 2J
hEK
'¢h,
'¢1 = ~ '¢1h•
hEK1
and set
'\j)(R) = ~ '¢h'
kER
- •h(R)
XRR1- 1h(R1)
'r - -r1 '
where both R and R 1 are finite. Fix a natural N. Since '¢ lt=O ~
~ '¢11t=o and
[ -ih :t +H ( -ih :x, x) J('¢-'¢1)=/,

where f has the form
i
- (CG, t)
eh p<a. t, h) ka. t) =II-1 (x, t)
with some D-asymptotic p ~ 0, it follows that for any k and l,
there are fmite sets R c K and R 1 c K 1 such that for any finite
R' :::J R and R; :::J R 1 ,
N+(; ]+k+i)
ll'l·w, Ri (x, 0, h) llck+l (R~) = 0 ( h '
[ -ih :t +H( -ih :x, x)]xw,R;=fwR{(x, t, h)

with
( hN+(; ]+k+2)
II fw, Ri 1/ck+l (R~) = 0 .
Furthermore, XR'lt R'1 (x, 0, h) and fR, ' R'1 (x, t, h) vanish for x being
outside a ball independent of h and t. Hence
II Xw, Ri llc7 (Rril = 0 (hN)
by lemma 1.1, which means that '¢ ~ '¢1 •
28-01225
434 OPE:RATIONAL METHODrS
The proposition is proved.

Now we give a generalization of Proposition 1.1 to the case where
a 0 need not be in the intersection of non-singular patches of the zones
Q and Qr. To do this we use the fact that for the canonical trans-
formation g1 0 associated with the harmonic oscillator Hamiltonian
function H 0 = ~ (p 2 +q2), a neighborhood in g1 0 An of the point
g1 0a 0 satisfies the condition of Theorem 5.1 of Chapter IV provided
that 't is sufficiently small.
For the rest of this section we shall fix the following notation.
Let J (a, -r) be the complex Jacobian corresponding to the family
{g10 An, dg1 0 rn}. Given a continuous branch Arg J (a) of thephase
argument of J (a), we define Arg J (a, -r) as the continuous function
J (a, -r) = I J (a, -r) I exp (i Arg J (a, -r)),
Arg J (a, 0) = Arg J (a).
For every small 't > 0 choose a bypassing focuses Hamiltonian H,;
associated with g1 0a 0 and the pair (CD, I) and having the ,form (5.9)
of Chapter IV. Let J (a, 't, t) be the complex Jacobian corresponding
to the bypassing focuses operation associated with HT. We define
Arg J (a, -r, t) as the continuous in t function satisfying the con-
ditions
J (a, 't, t) = I J (a, 't, t) I exp (i Arg J (a, 't, t)),
Arg J (a, -r, 0) = Arg J (a, -r).
Set
Arg Jr (a, -r) = Arg J (a, 't, ~) +:rca (-r),
where a (-r) is the number of negative eigenvalues of the matrix
e (-r) which occurs in the Hamiltonian HT. In Sec. 3 (see Lemma 3.5)
we shall show that the limit
lim ArgJr(a, -r) def ArgJr(a) (1.22)
T-++0
exists and does not depend on the choice of H,.,.
Proposition 1.2. Let ao E r be in the intersection of two patches
(u11 , :n: 11) and (u7 ,, Jt~,). Then for any cp (a) E Co
supported near a 0 ,
we have
i i 1
F [ h<D1'(a)-2ArgJ(a)JJ( )J-2£1' ( )] - ,.....,
xr..,.S] e a (jJ a a=n/ (x) ,.....,
""'"'Lr T<l>r
i v' i
(a)-zArgJrCa)JJ
1
J-2£11' Y!V' ( )]
""'"' e r 1 V <1> I
'
(jJ a a=(nl )-l (x
V' ~I'
•-)'
'1:!1
(1.23)
CH. V. CANONICAL OPERAT'OR ON A LAGRANGE.AN MANIFIOLD 435
where
L"{ = ~ (~~.)i (_.!._)1

L..J j! aa '
!jJ=O
P~·= -Ci 1 [{zr, w:c}-1''1

and v"!.:. v' is a D-asymptotic differential quasi-identity operator.
<l>, I
To prove this proposition we need the following lemma.
Lemma 1.2. Let cx 0 E r [be in the intersection of two patches,
(uv, n~) and :(uv'• n~·), of the zone Q 1 • Then for any q>(tx)ECo SUp- 00
ported near cx 0 , we have

1 i
- - -<J>Y(cx) y
[Jr (ex)] 2 e h I LI![! (ex) lcx=(nl)-1 (x) ~
y
i
-
y'
<l>J (ex) --z1 y' Y y'
~eh [Jr(cx)] Lrvr' q>(tx)la=(n~,)-l(x)' (1.24)
where vi' v' is a D-asymptotic differential quasi-identity operator.

1
Here [J 1 (cx)]2 is a smooth branch of the root of the Jacobian in the
neighborhood of cx 0 •
Proof. As before, it suffices to check the result for the special case
where Q 1 is the non-singular zone. ·
First we prove that
exp { ~ (Dv (cr (ex))}~ exp { ~ wv' (rx)} tjJ (a),
where cr = nv on);} and 1jJ is the D-asymptotic at cx0 series defined by
~
1jJ = Li
00
1 [ hi
if rr-.Y
(w rr-.Y' (
(cr (ex))- w a)) Jj .
i=O
Recall that
cr (ex) = ex +On (hlf2)
by Lemma 3.3 of Chapter IV, and
<Py (cr (ex)) = wY' (ex) +On (h312)
by Lemma 3.4 of Chapter IV.
We have to show that
(a~) j exp { ~ CI)Y (cr (a))}~ (a~ r exp { ~ (J)Y• (tx)} tjJ
28*
for any multi-index j. For j = 0, this follows from the identity

exp { ~ <D" (a (ex))}= exp { ~ <D"' (ex)} X
X exp {* [(J)~'(a(ex))-<D"' (ex)]}=

! [<D" (a (ex))- <D"' (ex)] r+
N
= ~ exp { ~ <D"' (ex)}* {

j=O
1 { i 1' 1'' } N +1
+(N+ 1)! ;;[<D (a(ex))-<D (ex)] x
X exp {_~ [<D"' (ex)+ 0 (ex) (<D" {a (ex))- <D"' (ex))]} ,
where 0 < 0 (ex) < 1. To prove it for all j, note that
(:a ) exp { ~ <D"' (ex)} ;tV= exp { :~ <D"' (ex)} ( a~ + ~ acD;~(a))

i i ljl
and
(a~ )i exp { ± <D" (a (ex))}=
= exp { ~ <D" (a (ex))} (a~+! arT!~'~~ (a)) ) 5 1~

~ exp { ! <D"'.(ex)} 1jJ (a~+! acD~' ~~(a)) r 1,
which shows that it suffices to check the relation
( _!__
aa
+ ..!:_h acD~''aa (a) ) i ,h ~ •h (_!__ , ..!:_ acD" (cr (a))
'Y
'Y ' - aa T h aa
)i 1 ' Ij
I
>- 1 ·
The proof will be by induction on I j I· We have
( _!__
aas
+ ..!:_h acD~''
aas
(a)) tV=
;! (! r {r[<D"(a(ex))-<D"' (ex)t-1
co
= ~ X
r=O
X ( acD~' (cr (a)) acD~'' (a) ) +
aas aas
+ ..!:_h acD~''aa (a) ( <D" (a (ex)) -<D"' (ex) )r} ~
8
~ ~ _1 (_!:_)r+tacD'~''(a) (<D"(a(ex)-<D"'(exW=
. Li r! h aas
r=O
=tp (~+ _£_ acD~'' (a)) 1.

aa 8 h aa 8
CH. V. CANONI'CAL OPERAT'O'R ON A LAGRANGEAN MANIFOLD 437
So the relation in question holds for I j I= 1. Assume that it holds

for UI~N and let jki=N+1, k8 =/=0. Then
( _!___ +~ iJ(])'l'' (a) )k 'ljl =
oa h oa
=(_!___ ~iJ(])'l''(a))(_!____ .!..._iJClJ'l''(a))k-1s ~
oas +h oas [oa +h oa 'ljl
~ (__!!__
8as
+ _!:__h iJClJ'l'' (a)) [¢ (~+ _!:__ iJ(fl'l' (a (a)) )k-1s 1J~
oas . oa h oa
~ 'ljl [_!:__ iJ(])'l' (a (a)) (~ _!:__ ()(fl'l' (a (a))) k-1s 1 +
.-L
h Oas Oa I h Oa ,
+ ..!....__ ( ~+_.£. ~ClJ'l' (a (a)) ) k-1s 1 ] ='ljl ( ..!....__ + _!:__ iJ(fl'l' (a (a)) ')k 1
oas oa. h oa oa h oa '
which proves the desired result.
Next we have
[J (a (cx))]-1/2 = [J (ex)] -1/2 (1 +X (ex))
with x(cx)=On(h 112 ), and
v "" oo (a (a)-a]k alliB~ alHk-llcp
[L cp] (a (ex))~~ ~ ~ ll j! (k-l)l "'7kil aai+k-l ~
li!=O lkl=O l~k
~ ~ "'V ~ [a (a) -ajl+s 01l!Br-s 0lrlcp ~

Ll Ll Ll l! s! (r-s)! iJal oar
lrl=O s~r 111=0
00
1 olrlcp def
~ ~ rr[a(cx)-cx+~v(a(cx))t oar=L'cp.
lrJ=O
Let R'~'' be the D-asymptotic differential operator defined by
lii=O
00
R'~'' = ~ /1 ( ~
00
lkl=1
cfJ\ (~v') r( r' a~
where {cfJ\} is the sequence of polynomials of Lemma 6.4 in Chap-
ter IV. According to Lemma 6.5 in Chapter IV,
LV' R'~''1Jf ~ 1Jf
for any D-asymptotic series 1Jf supported in a small neighborhood
of cx 0 • Put
v'~'· v' = R'~'' ¢(1 +X) L'.
It remains to be shown that v'~'• v' is a quasi-identity operator. To do
this, we use the observation that L' - L 'l'' raises the order. In fact,
438 OPE~ATIONAL METHODS
we have
00
(L-Lv')rp(a)= ~ -h-{[a(a)-a-t-~.,(a(a))f-
lri=O
- [~v· (a)n alr~:Yx)

for any rp (a) E Coo supported near a 0 • It follows from Lemma 3.3
of Chapter IV that
a (a) - a + ~v (a (a)) - ~v· (a) = On (h),
which implies
[a (a)- a-t- ~v (a (a))r- [~v' (a)r =On (hrii),
so
ord [(L ' -L'V' )rp(a)L>z-+ordrp(a).
1
Now the required assertion follows from the identity

v'V· V' = R"''L'V' + R"'' (L' -L'V') + nv' ('¢ -1 +X'¢) L'
by noting that x and 'ljJ- 1 raise the order, Lv' and R"'' do not
lower the order, and Rv'Lv' is a quasi-identity operator. The lemma
is proved.
Proof of Proposition 1.2. We need some further notation. For every
h-asymptotic series '¢ 0 , let Rt'¢o (x, h) = ~' (x, t, h) be the solution
of the Cauchy problem
- ih ~~ + [ ~2 - h2 ( :X ) 2 J'ljJ = 0, (1.25)
'¢1t=o='¢0 (x, h), (1.26)
and let R~ 'ljJ 0 (x, h) = '¢1 (x, t, h) be the asymptotic solution of
8
(1.2fi), (1.26) with the special Cauchy. datum

i i
hiiJ'V(a) -1/2 --zArgJ(a) 'V
1jl0 (x, h)=e IJ (a)J e L rp(a, h)la=n:;/(x)
corresponding to the path (u.,, nv):
i i
h iD'V (a, t)- 2 Arg J (a, t)
'\jlt(x, t, h) = e X
00
xJJ(a, t)I- 112 L"'j~VI8)jrp(a, h)la=a'~'<x,t)'

where a'V (x, t) is the solution of the equation q (a, t) + y (a, t) = x,
f) E ~ and <!>"', J, L v are determined via the (n + 1)-dimensional
manifold {gk0 An} (with the complex germ dgk 0 rn). Quite similarly,
we define the operator R.rs which sends any h-asymptotic series 'ljl"
CH. V. CANONICAL OPERAT'OR ON A LAGRANGEAN i\IANIFOLD 439
of the form
_ [ hi <Drv' (a, 1:)- 2i Arg J 1 (a, 1:)
'ljJ't ( x, h) - e X
xlfr(a, T)I- 112 Li'!g' ncp(a, h)] ,

HoA a=ai (x 't)
V' ,
where a~· (x, 't) is the solution of the system

qr(a, 't) = y1 (a, T) = x 1 , P1 (a, 't) yj (a, 't) = x'i +
into the asymptotic solution of (1.25) with the initial condition
'ljJ lt='t ~ 'ljJ't. Let F be the h-1-Fourier transformation with respect
to the .fth group of variables. According to Proposition 2.1, for
any D-asymptotic series x" supported near g10a 0 E g1 0 An, we have
F'ljJ (x, 't', h) ~ Fas'i' (x, 't, h),
where
i i
"'x, 't, h)-
'"( -eh (a, 1:)-2 Arg J (a, 't) IJ ( a, 't) ~-1/2 L ('t ) X ( a ) Ia=a(x,'t)
and
i
"
F as'i' (;_{,,
dcf
h)- hi (a, 't)-2 Arg J I (a, 't)
't, -e X
X IJr (a, 't) l- 112 Lr ('t) Vr ('t) X(a) la=ai(x, t)'
L (T), L 1 ('t) and v 1 ('t) being determined via (g1 0 An, dg'k 0rn).
Finally, we let ri· v' (T) denote the operator corresponding to
(g1 0An, dg1 0 rn), which sends, for any cp supported near g'k0 a 0 , the
asymptotic series on the left of (1.24) into that on the right.
Now the proposition is proved by considering the diagram:
440 OPE:RATIONAL METHODS
where
i i
•h ( )- -,;:<t>'l'(a)- 2 Arg.J(a)JJ( )J-1/2£"1 ( )J
'i'O x - e ex rp ex a=nvl (x).
The desired result follows from the fact that F'\j) 0 ~ 'IJ'r. Justification
of the equivalences which occur in the diagram is as follows:
eq1 is by Lemma 1.1,
eq2 follows from eq 1 and Lemma 1.1,
eq4 follows from eq 3 and Corollary 1.1,
eq6 follows from eq 4, eq 5 and Lemma 1.1.
This completes the proof.
Note 1.1. Let 8/[)l = (8/8l1 , • • • , 8/8l 11 ) be a set of commutating
complex vector fields on u., which are linearly independent at every
point. Lemma 1.2 and Propositions 1.1, 1.2 remain valid if we set
D (qr+zr. p1 +w1 ) 8 (qi+zi, P]+w1}
JJ(ex) = Dl = det 81 , (1.27)
L"= ~ w;lj (.!__)j

I LJ j! i.Jl '
\ji=O
where
Sec. 2. Commutation Formulas for a Complex Exponential and a

Hamiltonian
Let x = (x1 , • • • , x 11 ) be the generating set of multiplication ope-
rators by independent variables operating in the Sobolev scale
{Hk (Rn)} with D = Co (see the definition of a Banach scale in
p
Sec. 1 of Chapter II), and let = (p1 , • • • , Pn) be the generating set
of operators in the same scale defined by Pi = -ih 8/oxh where h
is a positive parameter. Further let
def oo
QJ8(x, p)E.%loo(R 211) = n .%l 8 (H 2 "),
s=O
and let S (x) be an infinitely differentiable finite function with

Im S (x);;;;:: 0.
In this section we shall obtain commutation formulas for the
Hamiltonian rJJ8
. (x,2 p1) and the multiplication operator by
CH. V. CANONICAL OPERAT'OR ON A LAGRANGEAN MANIFOLD 4.41
*
exp { S} . To this end we first consider the following two special
cases: (1) Re S = 0 and (2) Im S = 0.
Lemma 2.1. Under the above assumptions suppose additionally
that ReS = 0. Then for any natural k,
_!:_ S
rJ/8 ( 32) .i_ 8 (~) =
x,p eh
( - ih) lrxlalrxle h
(x)
alai rJf& ( 2 1) +
a! axa apa x, p
O~lal~h-1
+ (- ih)h rh (S), (2.1}

where rh (S) is an operator such that
h+l
II rh (S) llnz .... nl ~ ch,zh --z (2.2)
for all l~O.
Proof. Using the commutation formula of Chapter II, it is easy

to prove by induction that (2.1) holds with
n
rk(S)= ~ ~
i=1 lrxl='k-1
rxj+t=· . . =<Xn=O
ao )(41
(apa
3
(2.3)
X {)pi rJJ8 x, Pt> · · ·' Ph Ph
According to Corollary 1.1 of Chapter IV,

2..s
aheh h+l
~const h--2-
axrxaxj cl(Rn)
hence, letting A denote the multiplication operator by

acting on H 1 to itself, we get
h+l
II A Jl~const h--z
Since
lilt (~,~17 ... ,ph ~i> ..• , ~Jllnz . . nz~

<;;.const II A llnz . . nzll f IJ _g& 1,
this implies (2.2), Q.E.D.
As to the case of a real S, it is not difficult to obtain the expansion

which generalizes (7. 7) of Chapter II. To state the result we introduce
non-linear differential operators P~~ 13 as follows: if !~ (x 0 ) = 0,
then P~~ 13 (S) lx=xo is defined by the identity
i i
e
-hS(x) ( - 'h_!_)a
l ax e
hS(x) ( -
X Xo
)131 x=xo --
(a(
= ~ (- ih) 1 a!~! P~~ 13 (S) lx=xu; (2.4)

1=0
in the general case we set

P~~ 13 (S) /x=xo = P~~ 13 (So) lx=xo•
where
as
S 0 (x)=S (x0) - (x-x 0 , ax-(x0 )}.
In particular,
I
Pb~) 0 (S) = 1'
P~.> 13 (S) = o if I a I= 1, I ~ 1= o or I a I= 2, I ~I = 1,
P~!I3(S)=(ct, ~) if lctl=l~l=1,
p(i)"
a,"
(S) = __!_ aas
al axa
if I ct I= 2.
lemma 2.2. Under the assumptions of Lemma 2.1 with: Re S =0
replaced by Im S = 0 we have for any natural k
QJ8 (!, ~) e ~ 8 (!) = e ~ 8 <x> (i{ (- ih) 1 U>1 (S) + hkrk (S)}, (2.5)
l=O
where
rk (S) E Hom (Hs+k, H"), all s E R,
and
ala I Qfe ( as ) 1 all31
U>z (S) = ~ ----;;-a- ax , x Pa, 13 (S) axil (2.13)
l<:;(a(<:;21 p
In particular, U> 0 (S) =JJ8 (x, Sx).

Proof. Obviously, (2.5) can be rewritten as
aJ8 ( ;, !~ +p) ~ 1
l=O
( - ih) 1 U>z (S) + hkrk (S), (2.5')
CH. V. CANONICAL OPERAT'OR ON A LAGRANGEAN MANIFOLD 443
but the left-hand member of (2.5') has the form f (c, A+ B), so
we use the following generalization of (6.8) of Introduction:
k-1 n n { 2 21
f(A+B)=f(A)+ ~ .2i .... ~
1=1 1 =1 11=11_ 1
Bi 1 • •• Biz X
1
( 1 1 3 2!-121+1 2!+1)}
X 6i 1 ••• 8izf A1 , ••• , Aj 1 , Aj 1, ••• , Aiz, A;z, ... , An +
2 2k
+ ~
1~ji~· . . ~jk~n
{
Bi 1 ••• Bik X
1 1 3 2h-1 2k+1
(
X 6i 1 ••• 6ikf A 17 ••• , Ai 1 , Aj 1 , ... , Aik' Aik +
(2.7)
where, as well as in the one-dimensional case, some operators may
be additionally included as parameters (cf. the notes following
. Theorem 6.4, 6.6 and 6.7 of Introduction, respectively). Thus we
obtain (2.5') with
Cl>z (S) = ~ · {a~ . .. ·a:. x

1~ji~· . • ~iz~n 11 1z
( 1 1 3 21-121+1 2!+1)}
[6j 1 ••• 6}jdf'8) Sx 1, •• • , Sxji' Sxj 1 , •• • , Sxi' Sxjl' ... ,Sxn
(here we have dropped x being the last to operate). In the expression
of cD 1, let us change the order of operating so that f)/f)x would be the
first to operate. Using the commutation formula of Chapter II, we
arrive at
CVz = ""
LJ
aa61e (~as
--;;a 7fX ' X
) Q<Z> ( a ) B,
a, B 7fX
l~la\~2! p
O"'illl~l
where Q~~ B (S) is an independent of 018 polynomial (with constant

coefficients) in the derivatives of S of orders 2, ... , l 1. +
What remains to be proved now is that Q~: B = P~~ B· It is easily
seen that for this it suffices to show that
Q~~ B (S) lx=O = p~~ 13 (S) lx=O
for all S with Sx(O) = 0. To do this, put dl8 (x, p) = pa. In spite of
pa ~ S&oo, both (2.7) with f = J!8, A= Sx, B = p and the commu-
tation formula of Chapter II remain clearly valid in the case, the
444 OPE:RATIONAL ;vrETHO:DS
remainder in (2. 7) vanishing if k is large enough. Hence
;a exp { ! S (~) } = exp { ~ S (x)} X

1 alo\
X ~ ( - ih) 1~ a. sa-vQ<Z> (S) - . (2.8)
.LJ .L.l(a-y)! x y,o axo
z v,o
Applying this operator indentity to xB and putting x = 0, we obtain
_..!__ s (x) a a _2._ s (x)
e h ( - ih fiX) eh xi3Jx=0 = ~ ( - ih) 1 a!~~ Q~~ 13 (S) ix=O·
l
Comparison of the last formula with (2.4) produces the desired result,
Q.E.D.
We now generalize (2.5) to the case where Im S need not vanish.
Theorem 2.1. Under the assumptions stated at the outset of the section
we have
( 3 2) i ( 1) i
cf/8 x, p eh 8 x =ehS(x) ~ (-ih)k<D}<N>(S)+RN(S),
N-1
(2.9)
k=O
where
<D~N) (S) = { __!__ ( i a lm s ) v X
y! ax
o,_:;1v1,_:;2 (N-k)-1
k,_:;Jaj,_:;2k
0,_:;113\,_:;k
alal+lv1Q16' ( aRes) (k) all31
X apa+y x I -OX- P a, IJu(S)-
axJ)
(2.10)
and RN (S) can be estimated as follows:

N-.£..z
IIRN(S)IIHN+Z...,.Hl~CN,lh (2.11) 2 ' l=O, 1, 2, ...
Proof. Set S 1 =ReS, S 2 =ImS. We begin by commuting the
Hamiltonian QJ8 (~, ~) with the multiplication operator by
~xp [ - ~ S 2 (x) J; according to (2 .1), we get
3 2)
cf/8 (x, p e h
_2_ 8 (~)
= X
(2.12)
where
i
. 2N hSi(X)
RN, 1 (S)=(-zh) r2 N(S 2 )oe •
CH. V. CANON~CAL OPERAT'OR ON A LAGRANGEAN MANIFOLD 445
!
( x, 0 1 ) X
(2.13)
where
rN, a (St) E Hom (Hs+N, H 8 )
for any sER.
Thus (2.12), after substitution of (2.13), gives
( 3 2)
,fJ8 x, p eh
_!_g(~)
=
· N-1
.2_g
=eh ~ (-ih)k ~
h=O o,;;1v l,;;h
I<<: I a 1,;;2h
O,;;l f31"'2(N-h)-1
X (iS2x) 13 Qh, a, 13. v (S) (:X r} +

+ +
RN, 1 (S) RN, 2 (S) RN, 3 (S), +
where Qh. a, 13,., (S) is a polynomial in the derivatives of S 1 and S 2
of orders 2, ... , k+1,
RN,2 (S) = ~
O,;;la\,;;2N-1
and RN, 3 (S) is expressible as the sum of operators of the form

i
ehS(x)hkf (x) (a:: )f3 (:X)"
with t E Co, I y I ~ N - 1, k I~ I ;? N. +
To compute Qh, a, 13, .,, we apply the method similar to that used
in the proof of Lemma 2.2. Put ,fJ8 (x, p) = p 6 • The expansion (2.14)
is, of course, valid in this case, the remainders RN, 1 , RN, 2 and RN, 3
vanishing provided that N is large enough. Thus,
2, i (1) i
~ (6-~~~)!
-h S X -h S(x)
pue =e [ ( - ih)k
k, a, [3, l'
X (i 0 ! 2 ) f3 Qk, a, 13, 'l' (S) ( :x )"].

Let us apply this operator equality to x 8 ; if a~1 (0) = 0, which we

may assume without loss of generality, then
~ 2.. S(x)
plleh xe
Ix=O =
= e-h
i
S(x) 'V
L.J ( - ih) 6!e!
h. as2 ) ~ Qh., o-[3,
( i Tx (3, e (S) I
x=O. (2.16)
h, ~
On the other hand, (2.8) holds for complex S too. Hence

~ _i_ S(x)
pile h xe
I x=O =
i
= h
e
S(x) ~ (- '1 )h.~ (. as2
n ("'u-y)I L uX
"
)ll-v p<h)v e (S) /x=O •
'
h, y
Comparing the last formula with (2.16) we conclude that

1 (h)
Qh.. a, B. v = WPa, v·
To complete the proof, we have to check (2.11) for
:3
RN (S) = 2J RN, j (S).
5=1
To estimate RN, 2 , consider (up to a numerical factor) the general
term in the right-hand member of (2.15):
1
!a!-hs2 i
hN+Ia! iJ e (S)
•4 -
dcf hSt-
axa e rNa
• 1.
We have
\\rN, a (S1) 1\Hl+N_,.FJl~C:mst,
II e ~ 81 tz_,.FJl ~ const h- 1
and
!al+l
~ const h- - 2-
FJl_,.FJl
(for the last estimate see the proof of Lemma 2.1). Hence
N-~l
\[A 1\HN+l_,.FJz~const h 2
which implies the required estimate for RN, 2 (S). Estimation of

R N, 1 and R N, 3 may be left to the reader.
CH. V. C.ANONl'GAL OPERATOR ON A LAGRANGEAN MANIFOLD 447
Let F~~ ... x 1 be the inverse of the h -1-Fourier transformation with

respect to the /th group of variables:
where m is the number of elements in/. Next, we must obtain a com-

mutation formula for J/8 (~, ~) and the operator
-1
F ;14-Xl ~ S(~I' xy)
o e
generalizing (2.9).
Let r 1 be the operator defined by
rirp (x) = rp (-xi, x:r),
X . (x1 , ••. , Xn)·
Set
P~. 1~ = ( -1 )1 ~I I rio P~. ~ o r 1 •
As before, we assume that S E Co, Im S ;;;:: 0 and JJ8 E ~ oo, and
use the notation S 1 =ReS, S 2 = Im S.
Theorem 2.2. Under the above assumptions we have
where
cDf. I (S) = ~ ~ ~ ~ { y! (Br~~y)! 6! X

h~\ a 1~2k O~J ~ J~k '\'~~I O~JI)I..;;:2N -2k-1
olaJ+J'l'!+i61Qf6> ( DSt DSt)

X a +6a-,.+6 - o£r ' xy, SI, OX- X
ox/ I op/ I opj I
h, I
xPa.~(S)
( . os 2 ) 6 I (. oS 2
-lar l-0-
)61 --,~---
oi~J-Ivl }
bi xy ox-/ o£~1-'\'
and RN (S) can be estimated as follows:
N- ~~
IIRN(S)IIr,1 1 ~cn,zh 2
Hz.l+N->H
Proof. We start with the note that the required expansion can
be rewritten in the form
~Cio
Q/1)
( "h
~
~ 2
8Gr ' xy, <or, -
f "h
~
~
ax-
)
o e
~ S(~I' xy)-
-
I
i N-1
= eh S<~r xy) ~ ( - ih)k C!>f, r(S) + RN (S), (2.17)
k=O
( 2 1) _1 _1 (· i~ 2 1 ~ )
QJt X, p oF~r... xr=F~I-.xioQJt ~h 8Gr' Xj, £r, -ih axy .
S mce HI1; 1h IS con t"muons Iy em bedd ed mto

o
0
° F xr-.£r H 1, it suffi ces
to check the estimate
_ N-2_l
II RN (S) IIHI, 1 _.HI, h~cc, Nh 2 (2.18)
l,l+N l,l
The proof will be given in three steps.

(1) By an obvious modification of the proof of Lemma 2.1 we
arrive at
(2.19)
with the remainder
~J}' (2.20)
CH. V. CANONICAL OPERATOR ON A LAGRANGEAN MANIFOLD 449
where
y=(pi, xj), z=(iha~I' -ihiJ:_),
I
QJB(y, z)=QJB((zi, Yy), (y 11 z1 )),
8. = { 1 for j E~
1 0 for jE I.
Expression (2.20) implies the following estimate:
k+Z
II rk (S2) \\HI, h_,.HI, h~ck, zh- 2 (2.21)
Z,Z Z, z
(2) By the obvious modification of the proof of Lemma 2.2, we

arrive at
Pf ( 'l ~
l ~ a~I , Xj,
2 ~'oi, 'h b ) o e~ S!(sp x'j)-
(-
o)IJ - l ax- -
I
~,: 8
' 0 '' xy> r~ ih) 1 <D,, ,(S,) +h';:>(S.)}• (2.22)
where
- I h I h
rk (S 1) EHom (Hz,' z+n, Hz,' z )
for every non-negative integer l, and
01rx 1Q10 ( a~ 3 a~ )
<Dz, I (S) = 1 :-1
~
{
rx rxr - a~I '
1
xy, SI, -,-
uX-
X
Z~l rx 1~2Z O<l B 1~1 ax I r ilpy I
3 t 2 ) BI ( 2 ) Bj}
X [P~.IB (S)] ~ a~I 0:_ .
I
Commuting SI and a:I in the last formula, we obtain

2Z k
(J)z, I (S) = ~ ~ ~ { y! l/~y)! X

I rx I=Z I B 1=0 'l'<BI
aS )
Xj, SI, --;;;;:: X
I
(2.23)
29-01225
450 OP EHATION AL l\IETHODS
(3) By a slight modification of the proof of Theorem 2.1 (in par-

ticular, we must replace (2.1) and (2.5) used lhere by (2.19) and
(2.20), respectively) we obtain (2.17) with the remainder having the
following estimate:
- N-2z
h ~Hr, h~cn, zh
II RN (S) IIHr,l, l+N 2
l, l
It remains to be noted that the bound of the identity operator

I : H I,z, t+N
1
-+ HI, h
c, t+N does no t excee d h-l .
Corollary 2.1. Under the conditions of Theorem 2.2 we have for

any fJJECO':
_c:Jn ( 2
Q/1/ X, p
1) p- 1
s 1 ~x/
~ S(s I' xy> (~"
fjl SI, Xy
)
~
(2.24)
Jft being an S 2-asymptotic differential operator.
The same formula is valid in the case where fJl is an S 2 -asymptotic
series.
Example 2.1. Let 'lj.l (a) E CO' be supported in a y-domain u"'

of the zone Q.f of a Lagrangean manifold with a complex germ, and
let X be the image of r n supp 'lj.l under the mapping a-+ q (a).
Then, for any x (x) E Co
supported outside of X, we have
CH. V. CANONICAL OPERAT'OR ON A LAGRANGEAN MANIF\OLD 451
In fact, setting S = CDt o (n~t 1 , cp = ljJ o (n~t\ we obtain by Corol-

lary 2.1:
1 ..!:...sc£I' x 1l ,.
')( (x) F~1 .... x 1 e h cp (sr, xy) ~
i 00 2k h 00
"'"'p-1 hS(£p x"j)""' ""' ""' ""' { (-ih)h X

,...._. £1-xle L..J ~ L..J L..J o!
h=O Jal=h \131=0 16/=0
alal+/61x( ast ) h,I
X fJxaH - 8£r ' xy Pa, (3 (S) X
. as 2
X (- ~ ar ) 6 al 13 '
(3- /3 cp (sr, x7)
. }
.
I oxi o£/
It is obvious that the support of each term of the h-asymptotic
series on the right is a subset of the support of the function
· as 1
X (- o£r (sr, xy) ) cp (sr, x 7).
Set
x(a)=x ( - ~:: (pr(a)+yr(a), qy(a)+Yy(a)), qy(a)+l'y(a)).
It suffices to show that x
(a) ljJ (a) vanishes in the vicinity of r.
We have
X(a) = X (q (a) f (a)), +
where f = 0 n (h1 12) (see (3.3) of Chapter IV). Let u be a neighborhood
of X such that u n
supp X = 0. Then the inverse image of u under
the mapping a-+ q (a) +
f (a) is a neighborhood of r supp '\j), n
on which x
vanishes.
Problem 2.1. Given c;/8 and S, let cp (sr, x 1) be an S 2-asymptotic
series with the principal monomial of the form h 8 cp 0 (£ 1 , x1), and
let Jk' be the S 2 -asymptotic operator defined by (2.24). Then the
principal monomial of ~cp equals
h 8 Q/8 ( - ~:; , x 7 , sr, ~:~ ) Cflo (sr. xy)

I
provided that this functions does not vanish identically.

Problem 2.2. The expansion (2.24) remains valid in the cbse
JIB (x, p) E Soo.
Hint: represent QJ8 in the form
QJC(x, p)=(x2+f)Z(p2+f)l&'6'o(x, p)
with JJ8 0 E fJ1l oo·
29*
452 OPEiRATWNAL METHODS
Sec. 3. C -Lagrangean Manifolds and the Index of a Complex Germ

Let M be a real manifold of dimension n, and let f be a smooth
mapping of M into C2 n (identified with R 4 n), the 2n-dimensional
complex phase space with the coordinates P = (P 1 , • • . , Pn), Q =
= (Ql, ... , Qn). If f is given by
P = P (a), Q = Q (a)
in local coordinates a = (a 1 , •.. , an) in M, then we set
Pa = B, Qa = C,
a(Qr, Pr) Cr, (3.1)
fJa
Definition 3.1. We say that a point a 0 EM is C-Lagrangean with

respect to f if the following conditions are satisfied:
(C1)
rank ( B (rxo) ) = n ·
C (a 0) '
in other words, M is locally embedded at ao into C2 n = R"'n,
and the tangent space to M at a 0 is real-like.
(C2) The Lagrange brackets for the functions P (a), Q (a) vanish
at a 0 :
[< .!!._
iJai '
_!.2_)-(.!!!.._ !!L)J
fJak iJak ' iJai a=ao
=0·
'
(C3) The imaginary part of the quadratic form

F (g) = (B (ao) g, c (ao) g), g E en, (3.2)
is non-negative.
The conditions (C1)-(C3) are, of course, independent of the choice
of local coordinates.
Note that (C3) is equivalent to the following:
(C3') For some I, the imaginary part of the quadratic form
Fr (g) = (Br (a 0 ) g, Cr (a 0 ) g) (3.3)
is non-negative. .
In fact, a straightforward calculation shows that Im F 1 (g) does
not depend on I.
A subset of M consisting of C-Lagrangean points will be called C-
Lagrangean.
Example 3.1. The subset r of a Lagrangean manifold with a com-
plex germ is C-Lagrangean with respect to the mapping
Q (a) = q (a) +
z (a), P (a) = p (a) w (a). (3.4) +
In fact, (C1) and (C2) follow immediately from Definitions 2.1.
2.2 of Chapter IV. Let us verify (C3') for an arbitrary point a E r.
CH. V. CANOmGAL OPERAT'OR ON A LAGRANGEAN MANIF·OLID 453
Let a be in Q 1 and set ~ 1 = BrC1 1 , ~ 1 = ~}1) + ~}2 \ where

both ~}1) and ~7"> are real. We have shown before that ~}1 > and
~}2 > are symmetric, ~}2 > being non-negative because of the dissi-
+
pa.tivity condition. Hence we have for any h = h< 1> ih' 2>, h< 1>,
h<2> ERn:
Im(~rh, h)= -(~h< 1 >, h< 2 >)+(~h< 2 >, h< 1>)+
+ (~<2>h(1>, h(i)) + (~(2>h(2), h<2>) =
= (~<2>h(i), h(i)) + (~<2>h<2>, h<2>);;::0.
Putting h = C rg here, we obtain the required result

Im (Erg, Crg) ;;;;::: 0.
Now we shall produce some elementary consequences of (C1)-(C3).
Lemma. 3.1. If a EM is C-Lagrangean, then the matrix Er (a) +
+ it Cr (a) is non-degenerate for t > 0.
Proof. Suppose that this is not the case and let g be a. non-zero
vector such that
(E r + it CI) g = 0
for some t > 0. Since Erg = -it Crg, it follows that F (g) =
= -it II Crg 11 2 • Then Crg = 0 by (C3), so B 1 g = 0, contradicting
(C1). The lemma. is proved.
Lemma. 3.2. Let an EM be C-Lagrangean with respect to the mapping
P = P (a), Q = Q (a), let H (p, q) be a real quadratic form, and let
P = P (a, t), Q = Q (a, t) (3.5)
be the solution of the Hamiltonian system
P = -Hq (P, Q), Q = Hp (P, Q), (3.6)
satisfying the initial condition P (a, 0) = P (a), Q (a, 0) = Q (a).
Then an is C-Lagrangean with respect to the mapping (3.5) for any
fixed t E R.
Proof.
(1) First we show that
rank(Pa (a.o,t))
= n.
Qa (a.o, t)
Let P (a, ~), Q (a, ~) be linear in ~ERn functions satisfying
the conditions
P(a, O)=P(a), Q(a, O)=Q(a),
det a(P, Q) j =f= 0.
a(a, ~) a=cxo
Further, let P (a, ~. t), Q (a, ~. t) be the solution of (3.6) with the
initial conditions
P (a, ~. 0) = P (a, ~), Q (a, ~. 0) = Q (a, ~),
and set
J( P. t) = d t a (P (ex, B, t), Q (ex, B. t))
a, p, e a(ex, B) •
In the same way as in (the proof of) Lemma 2.2. of Chapter IV, we
conclude that :t
J (a, ~. t) = 0. Thus, J (a 0 , 0, t) =1= 0 which implies
the required assertion.
(2) Next we check (C2) for (3.5). We have
Qa (a, t) = HppPa (a, t) + HpqQa (a, t),
Pa (a, t) = -HqpPa (a, t) - HqqQa (a, t). (3.7)
A straightforward calculation yields
d
dt{Q, P}i~t=O.
(3) Now we prove that

lm (Pa (a 0 , t) h, Qa (a 0 , t) h) ~ 0.
Equation (3.6) implies that
d
dt(Pa(a, t)h, Qa(a, t)h)=-(HpqPa(a, t)h, Qa(a, t)h)-
-(HqqQa(a, t)h, Qa(a, t)h)+
+ (Pa (a, t) h, H ppPa (a, t) h)+ (Pa (a, t) h,
HpqQa(a, t)h) =(P:x.(a, t)h, H 1,pPa (a, t)h)-
-(Qa(a, t)h, HqqQa(a, t)h).
Thus the derivative :t
(Pa (a, t) h, Qa (a, t) h) is real, as the sum
of two quadratic forms corresponding to Hermitian bilinear forms.
Hence Im (Pa (a, t) h, Qa (a, t) h) does not depend on t, which com-
pletes the proof.
Now let us consider the Hamiltonian function H = ; (p; + q;).
The trajectory of H starting from (Q (a), P (a)) is given by
QI (a, t) = QI (a) cos t + PI (a) sin t,
Q1 (a, t) = Q1 (a), (3.8)
PI (a, t) = P 7 (a) cos t - QI (a) sin t,
Pr (a, t) = Pr (a).
CH. V. CANONI·CAL OPERAT'OR ON A LAGRANGEAN MANIF·OLD 455
Let a 0 E M be C-Lagrangean with respect to the mapping Q =

= Q (a), P = P (a), and suppose that the conditions
J ( ) def DQ O
a -na=F'
def D (Qr, Py)
lr(a)= Da =J=O
are satisfied at a 0 • Let a certain value Arg J (a 0 ) of the phase argu-
ment of the Jacobian J (a 0) be chosen. Using the transformation (3.8)
we define Arg J 1 (a 0) correlated with Arg J (a 0 ) as follows. First
we define the family (Q'(a), P' (a)), 't ~ 0, of mappings M-+ C2n by
Q' (a) = Q (a) - iTP (a),
P' (a) = P (a) + iTQ (a). (3.9)
Next we set
DQ'
J(a T ) = -
' Da
and define Arg J (a 0 , 't) as a continuous function satisfying the
condition
Arg J (a 0 , 0) = Arg J (a 0 ).
The transformation (3.8) induces the following homotopy of the
mappings
a-+Q'(a)
and
a--+ (Q 1 (a)- iTP 1 (a), PI (a)+ iTQ1 (a)):
Q'(a, t)= (Q}(a), QJ(a) cost+P:f(a)sint),
n
O~t~z-·
For a fixed T, the mapping [ 0, ~ J-+ C defined by
t-+ DQ~~· t) la=ao def J ( ao, 't, t),

(3.10)
n
O~t~ 2
is a curve in C starting from J (a 0 , and finishing at J 1 (a 0 , 't) der

T)
0
=
def DQ}
Da (a 0). Let us show that J (a 0 , T, t) =I= for. 't > 0. In fact,
it follows from Lemma 3.2 that a 0 is C-Lagrangean with respect
456 OPE:RATIONAL METHOD:S
to the mapping
a-+ (Q (a, t), P (a, t)),
where Q (a, t) and P (a, t) are defined by (3.8). By Lemma 3.1, for
-r > 0 we have
de t ( oQ (a, t) -l't
. aP (a, t) ) -1-
.,. 0.
oa oa a=ao
It remains to be noted that
Q' (a, t) = Q (a, t) - i-rP (a, t).
Since the curve (3.10) does not pass through the ongm, it defines
a certain value Arg / 1 (a 0 , -r) of the argument of the Jacobian
J 1 (a 0 , -r). Finally, we defme Arg / 1 (a 0 ) by
Arg J r (a 0 ) = lim Arg lr (aa, -r). (3.11)
"t-++0
The limit in (3.11) must exist as is seen from the following lemma.
Lemma 3.3. If J 1 (a 0 , 0) =I= 0, then Arg / 1 (a 0 , -r) is uniformly
continuous on (0, -r 0 ) for any -r 0 > 0.
Proof. Let 0 < -r1 < -r 2 < -r 0 • Consider the following four paths
in C:
l1 : 't-+ J (ao, -r), 't1 ~ T ~ 't2,
:n:
l2 : t-+ J (a 0 , -r1 , t), 0 ~ t ~2'
l3 : t-+ J (a 0 , T 2, t),
Let f.,i be the increment of the phase argument corresponding to li>

j = 1, ... , 4. Since the increment of the phase argument corres-
ponding to a closed path is a homotopic invariant and the path
l 1 + l 3 - l4 - l 2 in C {0} is homotopic to a point, we have
/5,1+ f., a = /). 2 +
/5, 4 • It is clear that
Arg lr (a 0 , -r 2) - Arg lr (a 0 , T1) = /5,1 + /),a - /5,2 = /5,4.
Let m= min Jl 1 (a 0 , -r)J. If
o~'~'o
max I / 1 (a 0 , T 1) - /1 (a 0 , -r) I< m,
't1~'~'2
then a simple calculation yields

I sin /5,4 I ~ m - 1 I lr (a 0 , 't1) - lr (ao, 't2) 1.
Thus uniform continuity of Arg / 1 (a 0 , -r) follows from that of
J 1 (a 0 , -r), and the lemma is proved.
CH. V. GANONICAL OPERATOR ON A LAGRANGEAN MANIF'OLD 457
Definition 3.2. Arg J 1 (a 0 ) defined by (3.11) will be considered

concordant with Arg J (a 0 ).
Let ao E Q QI n n
r be a point of a Lagrangean manifold with
a complex germ. For this situation we have introduced in Sec. 1
another rule of correlating the argument Arg J 1 (a 0 ) with Arg J (a 0 ),
that rule depending on bypassing focuses operation. Let us show that
these two rules are equivalent. In the case where a 0 is in the inter-
section of non-singular patches of the zones Q and Q 1 , this is an
immediate consequence of the following result.
Lemma 3.4. Let a 0 EM be C-Lagrangean with respect to the mapping
a-+ (Q (a), P (a)), and let Q' (a), P' (a) be given by (3.9). Further
letT= {1, .. . , k}, and set
,, 8 def , , •
Q1 (a, t) = Q1 (a) cost+ ePy (a) sm t,
, def ,
Qi 8 (a, t)=QI(a),
, 8 def , , •
P 1' (a, t) = Py (a) cost- eQy (a) sm t,
P'· 8 (a, t) def P} (a)

def DQ'• 8 (a, t)
Je(a, 't, t)= Da (3.12)
for any symmetrical orthogonal k X k matrix e. Define Arg J 8 (a 0 , 't 0 , 0)

for 't > 0 as a continuous function independent of e (note that
Jf: (a 0 , 't, 0) does not depend on e). Further define Arg J 8 (a 0 , 't, t)
as a continuous function of t for a fixed 't. Then
Arg Je ( ao, 't, ; ) + :rtO'e,
where a e is the number of negative eigenvalues of e, does not depend on e.
Proof. An argument similar to that used in the proof of Lemma 3.3

shows that Arg J 8 (a 0 , 't, n/2) is continuous in 't for 't > 0. Consider
the difference
11e, e' ('t) = [ Arg Je ( ao, 'C, ; ) + 'JtO'e J-
- [ Arg Je' ( ao, 't, ~ ) + 'JtO'e' J.
Since
(-1)cr 8 Je (a 0, 't, ~)=(-1)cre'Je(ao, 't, ~),
we have
11e,e' ('t) = 0 (mod 2n),
458 OPERATIONAL METHOJ:J.S
which implies by continuity that 11 e, e' ('t) does not depend on ,; .

Therefore, it suffices to show that !1 8 , 8 • (1) = 0 for any e and e',
i.e., that Arg J e (a 0 , 1, n/2) + na e does not depend on e. We have
J e (a 0 , ,;, t) = v (t) det (C-r: cost+ eCJ sin t),
where
.0 1 1n-k
v(t)=det( 10k (cost+smt)- )>0 '
D (P~, Qj)
Cj = ~a (a0),
and 1 i is the j X j identity matrix. In particular,

~ a (Pj, Qj) )
Je(a 0 , 1, t)=v(t) det C1 det ( cost+e aa a 0 sint.
Note that P 1 = iQi, hence

a (P-I1' Ori) 1 o
( i 0k )
aQ1 1n-k '
so we obtain
le (a 0 , 1, t) =
k
= v (t) det C1 (cost+ sin tt-k IJ (cost+ i"As sin t) =
s=1
k
= det C1 IT (cost+ i"As sin t),
S=i
where "At. ... , "Ak are the eigenvalues of e.

For any z EC ""- {0}, define arg z, the reduced phase argument
of z, in such a way that -n<argz~n. Since
arg (cost+ i"A8 sin t) =I= 'Jt
for t E [ 0, ~ J, it follows that

k
~ arg (cost+ i"A8 sin t)
s=1
is continuous in t on [ 0, ~ J. Therefore,
k
ArgJ 8 (a 0 , 1, t)=ArgJe(a0 , 1, 0) + 2}arg(cost+i"Assint).
8=1
.
P uttmg t = 2:n: , we ge t
Argle(an, 1, ~)=argl 8 (a0 , 1, O)+:n:;-Jta8 , (3.13)
so ~e (1) equals :n:2k which is independent of e, Q.E.D.

Now we turn to the case where ao E Q QI n
r is not in the n
intersection of non-singular patches of the zones Q and Q1 •
Lemma 3.5. The limit in (1.22) exists, Arg 1 1 (a 0 ) defined by (1.22)
being concordant with Arg J (a 0 ).
Proof. Let a 0 satisfy the assumptions of Proposition 1.2 and
let Q"· 8 (a, t, s), p<,e (a, t, s) be defined in the same way as Q'· 8 (a, t)
and P'· 8 (a, t) in (3.12) but with Q (a), P (a) replaced by q (a, s) +
+ z (a, s) and p (a, s) +
w (a, s), respectively, where
(p(a, s), q(a, s))=gif0 (p(a), q(a)),
(w(a, s), z(a, s))=dgi£0 (w(a), z(a)),
1
llo (p, q) = 2 (p2 + q2).
Set
J 8 (a, T, t , s) = det oQ"· 8 aa
(a, t, s)
and choose for any s E (0, s0 ) a symmetric orthogonal matrix e (s)

in such a way that
J e<s> (ao, 0, t, s) ¥=- 0
:n:
for 0 ~ t ~ z·
Let a certain value of the phase argument of J (a 0 ) be chosen. We
first define by continuity Arg J e<s> (a 0 , 0, 0, s), next Arg J e<s> (a 0 ,
1:, 0, s), T ;;-;::: 0 for a fixed s E (0, s0 ), and then Arg J e<s> (a 0 , T, t, s)
for fixed s and 1:. The statement of the lemma means that
lim [ Arg l 81 s> ( a 0 , 0, ~ , s) + na (e)]=
s~+O
To prove this we observe that by Lemma 3.4,
1t1Je<s>+Argle<s>(a0, T, ~, s)=ArgJ1 (ac, T, ~, s) (3.14)

for ,;> 0. Since
lim J (a0 , T, ~ , s) = J1 (a0) ¥=- 0,
'C-+0
s ... o
we obtain by an argument similar to that of the proof of Lemma 3.3

a
that Arg J 1 ( 0 , 't, ~ , s)
is uniformly continuous on the open
square 0 < 't < 't 0 , 0 < s < s0 • This, together with (3.14), imply
the required relation.
Definition 3.3. Given a real-smooth mapping a-+ (P (a), Q (a))
from a manifold M into C2n, the zone Q 1 is the set of all the points
a EM such that a is C-Lagrangean and
def
det C1 (a)= J 1 (a) =I= 0.
It follows from Lemma 2.4 of Chapter IV that the family {Q 1 } with
I running through all subsets of {1, . . . , n} covers the set of all C-
Lagrangean points of M.
n
Let ao E QI QK· We shall give the rule of correlating the phase
argument of J K (a 0 ) with that of J 1 (a 0 ). Let a certain value of the
phase argument of J 1 (a 0) be chosen. We set
D (Qj (a), p:_ (a))
J I (a, 't) = Da I
(Q\ P' being introduced in (3.9)), and define Arg J 1 (a 0 , 't) as a

continuous function satisfying the condition
Arg J 1 (a 0 , 0) = Arg fr (a 0 ).
Consider the Hamiltonian function
H (p, q) = ~ (Pit+ qJ1- Pkt- qkt),
where
Ii = I"'K, I2 = K"'I.
The trajectory of this H starting from (Q (a), P (a)) is given by
Qlt (a, t) = Qit (a) cost+ Pit (a) sin t,
Pit(a, t)=Pit(a)cost-Qit(a)sin t,
Qx 1 (a, t) = Qxt (a) cost- Pxt (a) sin t,
Pxt (a, t) = Pxt (a) cost+ Qxt (a) sin t,
QitllKt (a, t) = QytllKt (a),
Pyti!Kt (a, t) = Pyt1lif1 (a).

·CII. \'. CANONICAL OPERAT'OR ON A LAGRANGEAN MANIFOLD 461
The transformation (3.15) induces the following homotopy of the

curves 't-+ J 1 (a 0 , 't) and 't-+ J x (a 0 , 't):
1 1 (a, 't, t) =
a (Q} 1 (a) cos t+P} 1 (a) sin t, Q}nx (a), P_k1 (a) cost+
+ Q_k 1 (a) sin t, Pin.K (a))
=det----------------~a-a--~~-----------
(3.16)
Lemma 3.1 and Lemma 3.2 imply that 1 1 (a 0 , 't, t) =I= 0 for 't > 0.
Therefore, for 't > 0, one can define Arg J 1 (a 0 , 't, t) as a function
continuous in t. Finally, let us define Arg J x (a 0) by
ArgJK(a0 )= lim ArgJ 1 (a0 , 't, ~). (3.17)

, .... ·1·0
The proof of existence of the last limit is quite similar to that of

Lemma 3.3.
Definition 3.4. Arg J x (a 0) defined by (3.17) will be said to be con-
cordant with Arg 1 1 (a 0 ).
Lemma 3.6. If Arg JK (a 0) is concordant with Arg 1 1 (a 0) and
Arg J r ( a 0 ) is concordant with Arg J L ( a 0), then Arg J K ( a 0 ) is con-
cordant with Arg ·h (a 0).
Proof. We may assume without loss of generality that L = 0. Set
a(Q'I (a), Q:. (a) cost+ p~ (a) sin t)
I I
J(a, .-, t) = d et aa 1
a(Q_k(a), Q::..(a)cost+P_k(a)sint)
J(a, T1 t) = det K aa
It suffices to show that
~ (-r) def Arg J ( ao, T, -f)- Arg J (ao, 't, 0) +
+Argl1 (a0, 't, ~)-ArgJ 1 (a0 , 't, O)-
-ArgT(a0, 't, ~ )+Arg.T(a0, T, 0)=0
for every 't > 0; here J 1 (a, 't, t) is defined by (3.16), and all the
phase arguments of J acobians are assumed to be continuous functions
in 't and t. This, in turn, is equivalent to the equality
!1 (1) = 0,
because ~ (1:) is continuous for T > 0 and

~ (T) = 0 (mod 2n).
Let m1 , m 2 , m 3 be the numbers of elements in 1"'-K, K"'l, l"'K·
respectively.~We have already shown (see (3.13)) that
ArgJ(a 0 , 1, ~ )-ArgJ(a0 , 1, 0)= ~ (n-m 1 -m3 ),
ArgJ -( a 0, 1, 2 n) -ArgJ(a
- 0,
n
1, 0)= 2 (n-m 2 -m 3 ). (3.18}
It is easy to verify that
JI(a, t, T)=(sint+cost)-n+mt+m2 x
X
[
f) (Qj (a),
fJa
P] (a))
cos t +
fJ (Q1: (a),
fJa
P~ (a)) .
Slll
J
t .
Putting T = 1 here, one obtains
-iP (a)) .n-mt-m2 -ma
J r ( a, 1 , t) -_ D (Q (a)Da • l X
X (cost+ i sin t)m 1 (i cost+ sin t)m2 ,

hence
ArgJ1 ( a 0 , 1, ~) -ArgJ1 (a0 , 1, 0)= ~ (m 1 -m 2 ). (3.19)
The required result ~ (1) = 0 now follows from (3.18) and (3.19),
We can now define the index of a complex germ. To make the
treatment more transparent, however, we shall first defme the index
for a class of objects which are closely related to Lagr.angean mani-
folds with complex germs. For example, this class contains all (real)
Lagrangean manifolds.
Definition 3.5. A C-Lagrangean manifold is a real manifold to-
gether with a real-smooth mapping I : M-+ C2n such that each point
of M is C-Lagrangean with respect to f.
We start with the index of a closed path in a C-Lagrangean mani-
fold. First consider a path l (not necessarily closed) lying in a coor-
dinate neighborhood u EM. Let ~ (l, -r), -r > 0, be the increment
of the phase argument of det (C (a) - i-rB (a)) corresponding to l.
It is ebvious that ~ (l, -r) is independent of the choice of local coor-
dinates; in fact, on making a change of coordinates in u,
det (C (a) - i-rB (a)) acquires a positive or a negative factor. For
the case where l is not in any coordinate neighborhood, we define
~ (l, -r) by additivity. It is clear that ~ (Z, -r) is independent of 1:
CH. V. CANO~liCAL OPERAT'OR 0)1 A LAGRANGEA:-1 MAJ\IFCJLD 463
provided that l is closed. In this case we set

1
lndl=2n~(l, n), -r>O.
One can rewrite this as follows:
D(Q-ir:P)
i ~ Da
Ind l = -2il ';}' d ln D (Q-ir:P) . (3.20)
1 1
1 Da
It is easily seen that if lis homotopic to a point, then Ind l = 0;

hence, some 1-dimensional (characteristic) cohomology class of M
is defined.
Now, let l be a path joining two·non-singular points (i.e., points
where DQ!Da. =1= 0). Set
~l = lim ~ (l, -r); (3.21)
't->+0
~l will be called the rotation of Jacobian along l. Existence of the

limit in (3.21) can be proved in the same way as in (3.11).
Now suppose that Ind l = 0 for every closed path l in M, in
other words, that the cohomology class introduced above is trivial.
Then the rotation of Jacobian along a path joining a non-singular
point a. 1 to a non-singular point a. 2 depends only on a. 1 and a. 2 (but
not on the choice of a path). Let a point a. 0 E M, called initial, be
fixed; to be definite, assume ry} to be non-singular. Let us choose
a certain value of the argument of det Oa (a. 0 ), which determines,
in particular, a local orientation of M at a. 0 • We can uniquely define
Arg ~~ for any non-singular point a.1 which belongs to the connected
component of Jlil containing a. 0 by the formula
Arg ~~ (a. 1)=Arg ~~ (a. 0 )+~(l[a.o, a. 1]), (3.22)
where l [a. 0 , a. 1 ] is a path joining a. 0 to a. 1 • In particular, (3.22) deter-
mines a local orientation of M at a. 1 . Thus, we have defined a local
orientation of M at every point of the non-singular zone of a connected
component of 111.
We shall show that if M is connected, then the local orientations
introduced above determine a global orientation of the zone Q. Let
a.\ a. 2 be two points of Q, let u1 3 a.l, u 2 3 a. 2 be two coordinate neigh-
borhoods in Q, and let x, y be coordinates in u1 , u 2 compatible with
local orientations at a. 1 and a. 2 , respectively. If a. E u1 n
u 2 , then
both x and y are compatible with the local orientation at a.. To see
this it suffices to consider the rotations of Jacobian along the following
two paths joining a. 0 to a.:
(1) +
l [a. 0 , a. 1 ] l [a.\ a.],
464 OPERATFONAL MElTHODS
where the whole of l [a 2 , a] lies in u 2 ;

(2) l [a 0 , a 2 ] + l [a 2, a],
where the whole of l [a 2 , a] lies in u 2 •
So ~~ > 0, the global orientation of Q being actually defined.
Thus, we have shown that triviality of the characteristic class
Ind implies orientability of the non-singular zone. But, in fact,
triviality of this class implies orientability of the whole of M. To
show this consider the mapping
a-+ (Q (a) - i-r;P (a), P (a) i-r;Q (a)) + (3.23)
for some 1: > 0. With respect to this mapping, M is again C-Lagran-
gean (this is an easy exercise for the reader), and all points of M
are now non-singular as seen from Lemma 3.1
We now define uniquely Arg J 1 (a) for every point a E Q1 provided
that the initial point a 0 E M is fixed and a certain value Arg ~~ (a 0 )
of the phase argument of~~ (a 0 ) is chosen, by the following procedure
(correct if Ind = 0):
(1) For 1:;?;: 0, define Arg J (a0 , 1:), where
J( 0 )-D (Q-ir:P) ( 0)
a' 1: - Da a '
by continuity.
(2) For 1: > 0, define Arg J (a, 1:) by
Arg J (a, 1:) = Arg (a 0 , 1:) +
~ (l [a 0 , a], 1:),
so Arg J (a, 1:) is continuous in a.

(3) For 1: > 0, define Arg J 1 (a, 1:), where
D (Q 1 -ir:P 1 , Pr+ir:Qr)
Jr(a, 1:) = D rx '
to be concordant with Arg J (a, 1:).

(4) Set
ArgJ1 (a)= lim ArgJ 1 (a, 1:).
't-++0
Problem 3.1. If a E Qr n QK, then Arg JI (a) is concordant with
Arg JK (a).
Returning to a Lagrangean manifold with a complex germ, we
recall that it can be regarded as a manifold with the C-Lagrangean
subset r (see Example 3.1). For such a manifold, one could define
Indr, the index on r, by considering paths lying in a small neigh-
borhood of r. However, it is more convenient to define Indr in
terms of the Cech cohomology theory, where the index of a closed
chain of patches plays the same role as before the index of a closed
path did. To clarify the connection between these two approaches,
we shall describe Ind as a Cech cohomology class of a C-Lagrangean
manifold M.
We start with the case where M = Q, the non-singular zone. Let
{u 1 } be a covering of M by open subsets u1 such that the increment of
Arg J corresponding to any path lying in u1 is less than l't/2; such
a covering will be said to be admissible. For every j, fix a point
a/ E u1 which will be called the central point of u1. Define the 1-co-
chain ~ Arg J with real coefficients as follows: for every pair of
intersecting sets (uh uk), we let (~ Arg J) 1k be the increment of
Arg J according to a path l [ai, ak] which joins ai to ak and consists
of two arcs, the first lying in uh the second in uk. Obviously,
(~ Arg J) 1k does not depend on the choice of paths l [ai, ak], since
I ~ Arg J lik < 1t because of the admissibility property of the cover-
ing {u1 }. It is easy to check that ~ Arg J is a cocycle.
Definition 3.6. We define Ind to be the cohomology class induced
I
by the 1-cocycle 2n ~ Arg J.
Let us show that Ind is independent of the choice of central points.
In fact, let [{Xi] be another family of central points, and let 61 be
the increment of Arg J corresponding to a path joining ai to [;i
and lying in u1. Then the coboundary of the 0-cochain [6 1] is the
difference of the 1-cocycles ~ Arg J corresponding to the families
{a/} and {a;}, so these two 1-cocycles are cohomologic.
Problem 3.2. The class Ind is independent of the choice of an ad-
missible covering {u1 }.
I~emma 3.7. The class Ind of a C-Lagrangean manifold M c: M
is trivial if, and only if, there exists a continuous branch of the phase
argument of the Jacobian J.
Proof. 1°. Sufficiency. Let Arg J be a continuous branch of the
phase argument of J. Then {Arg J (ai)} is a 0-cochain with the
coboundary ~ Arg J.
2°. Necessity. Let Ind be trivial. Then for some admissible covering
{u 1 } (with any fixed family of central points), there is such a family
{a 1 } of real numbers that (~ Arg J) 1k = ak - ab where (~ Arg J) 1k
is the d.-cochain corresponding to {u 1 }. Without loss of generality ·
we may assume that M is connected and that aio is one of the possible
values of Arg J (aio) for ~orne j 0 • Then a1 is obviously one of the
possible values of Arg J (a 3) for any j (to see this it suffices to consider
a chain of elements of the covering joining UJ 0 to u1). This value of
Arg J (a 1) gives rise to a continuous branch of Arg J in ub so Arg J
30-01225
466 OPERATIONAL 1\IETHOtD.S
is defined as a 0-cochain with coefficients in the sheaf of germs of

smooth functions. To prove necessity it remains to show that this
cochain is a cocycle. But this is an obvious consequence of the fact
that
I aj - all. I < JI: if Uj ull. =f=. 0. n
We shall now remove the assumption made above that the whole
of M is in the zone Q. Let a-+- (P (a), Q (a)) be, as formerly, the
mapping that determines on M the structure of a C-Lagrangean
manifold. Recall that M is in the non-singular zone with respect to
the mapping (3.9) with 't > 0. So we obtain the family {Ind~}
of cohomology classes of M. But, in fact, Ind~ is independent of ·r.
To see this, let 't'1 and 1: 2 be positive numbers, 't'1 < 't' 2 , and let {ui}
be a covering of M by such open subsets that for every j, the incre-
ment of Arg J (a, 1:) corresponding to any paLh lying in ui is less
than n/2 if 0 < 1:1 ~ 't' ~ 1: 2 • Further, let {6~} and {6IU be the
cocycles ,1 Arg J corresponding to the mapping (3.9) with 't' = 't'1
and 1: = 't' 2 , respectively. Then we have
J:'t2 J:'tl Ail. A}
Ujll. - Ujk = Ll·q1: 2 - Ll'tft2 1
where ,1~n2 is the increment of Arg J (ai, 1:) corresponding to the

change of 1: from 't'1 to 1: 2 • So the cocycles {6Jn and {6Ik} are coho-
mologic,
We may now define the index for a C-Lagrangean manifold by
def
Ind = Ind\ 't > 0.
Lemma 3.7 leads to the following criterion for triviality of the
class Ind:
Proposition 3.1. The class Ind of a C-Lagrangean manifold M is
trivial if, and only if, for each zone Q 1 c M, there exists a continuous
branch Arg J 1 of the phase argument of the Jacobian Jr, these branches
being correlated in the intersections of zones.
Now let M be a smooth real manifold with a given mapping
M-+- C2 n, and let r c M be C-Lagrangean with respect to this
mapping. ,.,
Consider the following "almost constant" pre-sheaf II on M:
_ { R if n n r =I= 0
[II.(u) = 0 if u n f=F 0
with the naturally defined restriction homomorphism. We will define
Indr as a Cech cohomology class with coefficients in II.
First consider the special case where the Jacobian J does not vanish
on r. Let {ui} be a covering of M by open subsets with the following
admissibility property: if Uj n
f '=J=. 0, then J does not vanish On Uj
and, for any path lying in uh the increment of Arg J corresponding

to this path is less than n/2. For any j, fix a central point a/ E uh
and for j' k so that Ujn nuk r =I= 0' define {L1 Arg J} jk as in the
case of a C-Lagrangean manifold. Thus a 1-cocycle with coefficients
in II is defined, which will be denoted by Llr Arg J. We now define
Indr to be the cohomology class induced by 2~ Llr Arg J.
Problem 3.3. Indr is independent of the choice of a covering {ui}
and a family of central points {a/}.
Lemma 3.8. Let J (a) =I= 0 on r. Then Indr = 0 if, and only if,
there exists a continuous on r branch of Arg J.
Proof. 1o. Sufficiency. Let Arg J be a continuous on r branch of the
phase argument of Arg J. Choose an admissible covering {ui} of M
so that
- =
IArg J (a) -ArgJ (a) I < 2
:rt
a a
whenever there is a j such that both and are in Uj 1\ r. Further,
choose a family of central points {aj} so that aj E r n
Uj if r n
Uj =I=
=I= 0. Thus, a 0-cochain {Arg J (ai)} is defined, whose boundary is

L1r Arg J.
2°. Necessity. Let Indr = 0. Then there are an admissible covering
{uj} of M and real numbers ai (dependent on the choice of central
points ai E ui) so that
(.11 Arg J)ik = ah - ai
whenever Uj n nuk r =I= 0. Moreover, for some fixed jo such that

Uj 0 n f =f= 0, we can let a; 0 be one of the possible values Qf
Arg J (ai~). As before, we choose the family of central points {a1 }
so that a 1 E r if Uj n r =I= 0. Assume that any two points a' and
a" of r can be joined by a chain (ui 1 , • • • , uim) of elements of the
covering {ui} in such a way that
a'Euil a"Eum, Ujs n Ujs+t n r=t=O, S=1, ... , m-1
(which does not affect the generality). Putting a' = ai?, a" = ai,
we see that ai is one of the possible values of Arg J (a 1) whenever
Uj n r =I= 0. Define the 0-cochain with coefficients in the sheaf of
germs of smooth functions on f by assigning to each j with Uj n
f =f= 0
the continuous on ui branch of Arg J satisfying the condition
Arg J (ai) = ai. The same argument as in the case of a
C-Lagrangean manifold shows that this cochain is a cocycle, and
the proof is complete.
30*
We now remove the assumption that r c: Q. Again using the

family of the mappings (3.9), we obtain the family of cohomology
classes Ind~, 1: > 0. It is easy to verify that Ind~ is independent of 1:.
So we define the index on r by
Indr = Ind~, 't > 0.
In particular, Indr is now defined for any Lagrangean manifold with
a complex germ.
Lemma 3.8 implies the following criterion for triviality of Indr.

Proposition 3.2. The class Indr of M is trivial if, and only if, for
each I c: {1, ... , n}, there exists a continuous on QI n
r branch
Arg J I of the phase argument; of J I, Arg J I being concordant with
Arg JK on QI QKn r.n
The construction of the canonical operator in the next section
will depend on the existence of correlated continuous branches of
the roots of J acobians corresponding to various zones (but not of
such branches of the phase arguments of J acobians). Since VJ =
..!:..Arg J
= IJ Ie2 is defined uniquely, even if Arg J is defined only
modulo 4n, the condition of triviality of Indr is, in general, too
restrictive. Therefore, it is natural to introduce lndr modulo 2.
a
Given a E R, set Ra =Ria, where (x'""' y) ~ (x- y = 0 (mod a)).
Obviously, Ra is a module over Z. In the definition of lndr, let us
change the pre-sheaf II by using R 2 instead of R. Suppose that J =I= 0
on r, which is essentially the general case as we have already seen.
Let indik be the element of R 2 generated by the number{~ Arg JLk·
Then {indik} is a cocycle. The cohomology class induced by this
cocycle will be denoted by lndr (mod 2). The criterion for triviality
{)f Indr (mod 2) coincides word for word with that for lndr (see
Proposition 3.2) if Arg J 1 is regarded as a function with values in
R 4 n, Arg JI (a 0 ) and Arg J K (a 0 ) being considered concordant if
they have concordal!l.t representatives in R.
For some applications of the canonical operator (but not for the
proof of the Main Theorem) it is not sufficient to consider the class
Indr (mod 2) only, and the notion of index needs to be generalized
as follows. Let a/al = (a/al1 , • • • , a/aln) be ann-tuple of commuting
linearly independent at every point complex vector fields on M.
Then on replacing J (a) by det ~i in the above construction of the
index, we define a cohomology class of M which will be denoted by
Indr, OfOl (mod 2). It is easy to see that this new class differs from
Indr (mod 2), in general. The triviality criterion for Indr, O/Ol (mod 2)
is quite similar to that for Indr (mod 2).
Sec. 4. Canonical Operator

In this section we assume all the phase arguments of J acobians
to be defined modulo 4rc.
1. Definition of local canonical operators. We need some notation
and terminology.
Let us denote by .:ll (An, rn) the set of all equivalency classes of
D-asymptotic series on the Lagrangean manifold An with the com-
plex germ rn . .:11: (An, rn) is obviously a left module over the ring of
D-asymptotic operators. A point a E An will be called essential
for cp E .:11: (An, rn) if cp has a representative non-equivalent to zero
at a 0 • It is clear that all the representatives of cp are non-equivalent
to zero at a 0 provided that a 0 is essential for cp. The set of all points
essential for cp will be called the support of this element, supp q>
in symbol. For any cp E ./1; (An, rn), the support of cp is a compact
subset of r.
Let u be a subset of An. We denote by .Jl (u) the set of all elements
of ./I; (An, rn) supported in u. We shall use the notation .:11: (Rn)
for the set of all equivalency classes of h-asymptotic series in Rn.
Let us say that a y-patch of a Lagrangean manifold with a complex
germ is admissible if a dissipativity inequality with e = 0:
c (a) cvi (a) ~ D (a)
holds in this patch. Let (u, rc~) be an admissible y-patch of (An, rn).
Fix an n-tuple :Z = ( 8~ 1 , • • • , a:n)
of commuting linearly inde-
pendent at every point vector fields (in general, complex) on u.
a(qi + zr, P-+w-)
Set J 1 = det az 1 1 and fix a continuous on u branch of
the phase argument of J 1 • Then we define the local canonical ope-
rator
by
(ettv1(j))( X )=F-1 { ia<Dx<a)-~ArgJr<al)
s--x-
1 1
e X
XI Jr(a) ~-1/2 Ljcp (a) la=(n~)-l(xl' s1) (4.1)

(for definition of Li see (1.27)).
X
Proposition 4.1. ettjcp ~ 0 provided that x does not belong to the
image of r under the mapping a-+ q (a).
This proposition is an immediate consequence of Example 2.1.
Proposition 4.2. etti is a monomorphism.
To prove this proposition we need the followin" lemma.
Lemma 4.1. There is a function k (p), p ? ~ , such that
j=O, .. . , r
def
for all h>O and i=O, i, ... , [2p-i]=r whenever D>O,
c:;;>O, ai EC satisfy the inequality
l.± aihj
3=0
12 1e- ~ ~chP (4.2)
tor all h > 0.

Proof. Putting in (4.2) h = Dy 2 , where y is a positive variable,
we get
.± aiyiD~~~cy2Pe1/Y2 DP.
'J=O
It follows that for any (r +

i)-tuple y = (y 0 , • • • , Yr) of positive
numbers, and (r + i)-tuple b = (b 0 , ••• , br) of real numbers, the
following inequality holds:
(4.3)
-+
Fix y such that Yi =I= y 1 for j =I= l and consider the non-homogeneous
linear system of e,quations
l=O, ... , r, (4.4)
where sis a given integer and b is to be found. Since the determinant

of this system is exactly the Vandermonde determinant
i i i
11 = Yo Y1 · · · Yr
Y~ Yi · · · y; -+
corresponding to the (r +
i)-tuple y, it is nonzero, so (4.4) has a
solution b = b (s). Substituting this solution into (4.3), we obtain
ja 8 J~ck' (p)Dp-s/ 2 ,
where
k' (p) = max f (y, b(s)).
O:Ss:Sr
To complete the proof use the result of Problem 1.i of Chapter IV.
CH. V. CANONICAL 0PERATOR ON A LAGRANGEAN MANIFOLD 471
Proof of Proposition 4.2. It is clear that the operator

i
e- 2 ArgJ I(a) IJ I (ex) ,-1/2 Lt
is a monomorphism of .:1/; (u) to itself and that F~~ _, xy is a mono-
morphism of .Jl; (Rn) to itself. So it is to be shown that the multi-
.!:. 'l'(a)
plication operator by eh I cannot send a D-asymptotic series
non-equivalent to zero and supported in u into an h-asymptotic
series equivalent to zero.
Let 'ljJ = ~ 'ljJi be aD-asymptotic series supported in u.
jEJ
1o. First we show that
(e :, t 1jJ '""0) ~ (1Jl'"" 0).

i '\'
-r
In fact, if e h 1jJ'"" 0, then, for any N, there is a finite set
J 0 c J such that
I ~ e~ J(a)1Jli(ex, h),~const·hN
jcK
for any fm i te K => J 0 • Since

lo
~ 1Jli(ex, h)=h-s/2 ~ hlf2az(ex),
jEK l=O
it follows from Lemma 4.1 thaL

~
az=On h
(
N+~)
2
for l < 2N + s. Thus,

""'
~ ~
1Jli= <:Yn(h N ),
jEK
which means that 1jJ ~ 0.
_i_ i
2°. Now let e" 1jJ;::::;; 0, i.e.,
i '\'
( D~) heh J(a) 1jJ (ex)"-' 0
for any multi-index k. We shall prove by induction that

( D~ )" 1jJ (ex)'"" 0 for any k. For k = 0, this follows from 1°. Sup-
pose that ( :a r~j ,..__, 0 for Ill~ N and let I k I = N + 1. Then
(aa- )h eh "I1jl=e" " (-aa)

f) i iJ h1Jl+e" "IL-N-1£1jl,
i
I
i
472 OPERATFONAL METHODS
where L is an Nth order differential operator with coeffiCients

smooth in a and polynomial in h. By the induction hypothesis,
L1jJ--.- 0, hence
i v
eT<1Jih-N-1£1jJ--.- O,
i v
e
h" <IJr ( a ) k 'P"'-' 0 '
aa
which implies by 1o that (a:) k 'ljJ --.- 0. The proposition is proved.
2. Transition operators. Consider two admissible y-patches (u, n~)

and (u', n~) of a Lagrangean manifold with a complex germ, together
with fixed continuous branches of the phase arguments of fr and J x
defined on u and u', respectively, these branches being concordant
on u n u'. Let an n-tuple :Z of commuting linearly independent
at every point complex vector fields on u U u' be also fixed. Then
the local canonical operators 6'1t1 : .fl; (u) -+- .A (Rn) and r7t"k_ : .II; (u')-+-
-+- .fl; (Rn) are uniquely determined. Define the local operator
I K
n
V]"k_ acting on .ll: (u u') to itself (the transition operator from
·
(u, n.,) to (u', n.,,)) by the formula
(4.5)
The existence of VJ.k is a consequence of the following result
which generalizes Proposition 1.2:
Proposition 4.3. Let ao Ern u nu'. Then for any cp (a)~ CO' sup-
ported near a 0 , we have
i <!Jv(a) 1
[
Fxr1->sr/i";1-+xxt eli 1 llr(a)r2x
X e- ~ ArgJ1 (a) L"1 cp (a )] I _ ~

a=<nv> 1 <x1 , s1>
~[e~ <!Jk(a)IJK(a)le- ~ ArgJK(a)X
v'yvv'
X Lx 1xcr (a)
J -< K>-1<
a- nv' xK,
t-)'
~x
where I 1 = I"'- K, K 1 = K "'-I, Arg J x is concordant with Arg J 1

and V is a D-asymptotic differential quasi-identity operator.
This proposition can be proved in the same way as Proposition 2.2
and Lemma 3.5, or by using the canonical transformation corres-
ponding to the Hamiltonian function H 1 (see Example 2.1 of Chap-
ter IV).
The definition of Vl.i' is correct in the sense that there is only

one operator satisfying (4.5). This is due to the fact that local cano-
nical operators are monomorphisms. Moreover, it is easily seen that
Vtk is independent of the choice of vector fields a/al and concordant
branches of Arg J 1 and Arg J K·
It follows from the definition of transition operators that
V vv'vv'v"
IK KJ (jl =
TT'V'Y"
V i.J (jl
for any three admissible 1'-patches (u, :rr~), (u', :rr~), (u", :rr~") and
any cp E .It (u u' n n
u "). In particular, Vi?; is invertible, (Vtk) -1
being equal to v:kl.
3. Definition of the canonical operator on (A", rn). Consider· a
Lagrangean manifold An with a complex germ rn. Let us fix the
following objects:
(1) An admissible (i.e., consisting of admissible patches) 1'-atlas
{wh :rr ~H)} of a neighborhood of r such that the covering {ffii} is
locally finite. This 1'-atlas will be called weighting.
(2) A "weighting" C""-smooth partition of unity {Pi}:
~Pi=1, supp PiCffij.
j
(3) An n-tuple :l
= ( :Z 1 , • • • , a~J of commuting complex
vector fields on An, linearly independent at every point. Note that.
such vector fields always exist: for instance, one can put
a a
Tz =a (p+iq) •
(4) A set of concordant continuous branches of the phase arguments

of Jr I o1 nr (/ runs over all subsets of {1, ... , n}). Thus, the
cohomology class Indr,a;at (mod 2) is assumed to be trivial. We
define the canonical operator
fifC: .fl (An, rn)-+ .fl (Rn)
by the formula
~ N/' '\' (j)
cr = u> Picr·
N/'
Q/1.- ~ Q/1.- 1 (4.6)
j
Note that the number of non-zero summands in the sum in the right-
hand member of (4.6) is obviously finite.
Proposition 4.4. Let cp:E .It (An, rn), let {uk,:rr~~~l} be an admissible
1'-atlas of a neighborhood of supp cp, and let {elt} be a C"" smooth par-
titian of unity subordinate to the covering {u,J. Then

'\;1 vh v Cil vh
t!/Ccp = ~ t!/Cih VI(j)Ih Piehcp.
J,h
Proof. By the definition of transition operators,

vhvvCilvh '\;1 N?VCil '\;1 '\;1 '"viOl ('£"
f!/C Ih I (j)Ih piehcp = L,J I (j) pi .fl ehcp = 4 <Yt-J(j)
vt- Picp = vt- cp.
J "' J
Note 4.1. So far we have assumed that there exists an s-action on

An (or at least in the vicinity of r). This assumptioncanbeexpressed
in terms of triviality of a certain cohomology class of An. Namely,
let {ui} be a locally finite covering of An by open subsets satisfying
the condition: Uj Uh n n
r =/= 0, then ~ p dq vanishes for any
l .
cycle l lying in Uj n Choose
uh. a central point (J/ E Uj for any j'
and set
8ih = nih p p dq (mod 2),
l [a:i, rxh]
where l [ex/, ah] is a path lying in ui U uh and joining aj to ah. The

family {Bid determines a cocycle (depending on h) with coefficients
in the same pre-sheaf as for the class Indr (mod 2). This cocycle
induces a cohomology class of An, which will be de noted by p dq. :h p r
If h is a continuous parameter, then the existence of s-action is equi-
valent to triviality of this cohomology class for all h > 0.
In the case, where s-action does not exist in the vicinity of r,
the condition
_!__h .t, p dq = 0
rt ~
r
determines a certain set M of (permissible) values of h. If M has 0
as a limit point, then we can define the canonical operator by allowing
h to take on permissible values only.
Note 4.2. The case where (An, rn) depends on parameters can be
considered in the same way as in Chapter III. If both the classes
~h ~ p dq an d Indr, a;az (mod 2) are non-trivial, then, to define
""Jr
the canonical operator and to satisfy the quantization condition, it
is required that
nih pp dq = Indr, a;az (mod 2), (4.8)
r
which generalizes (4.7) and determines the permissible values of h

and the other parameters. For example, if (An, rn) depends on a para-
meter ~, then the condition (4.8) may be satisfied by a substitution
of the form p = p (e, h), where e is a new parameter. To substitute
operators for the parameters e and h, one can use the regularized
canonical operator in the same way as in Chapter III.
4. Commutation of a canonical operator with a Hamiltonian.
Proposition 4.5. Any canonical operator is a monomorphism.
Proof. Let X (x) be an infinitely smooth function which equals 1
for I x I < 1 and vanishes for I x I > 2. Set
Xe (x) = X (e-1x),
e 8 (ex) = Xe (q (ex)),
f 8 (ex) = Xe (p (ex)).
Let 0, Xe&t'<p = 0 and Xe67Z' X
X (1 - e 28 ) <p = 0 (see Proposition 4.1).
We have
Xe&t'eze<p = 2J
Xe&t'iYh Phe<p =
j
2J
67Z'yY]) x~)pie28<p = 0,
j
where x~) is a D-asymptotic differential operator preserving the

principal monomial at ex 0 (see Problem 2.1). This equation implies
that
Set
~ '\' " '\'
m; I = F x-+£&t' I .
Since the operator ifj_ can be explicitly expressed by the formula
(6.~}¢)(£)=Fxi-+~I {e~ <Dj(a)e- ~ ArgJI(a)X
1
xjJI(ex)J-2LI<p(ex)l r_ 1 }·
tx=(ny) (x 1 , ~I)
Proposition 4.1 remains valid if Qit'j is replaced by ifj_ and the map-
ping ex-+ q (ex) is simultaneously replaced by the mapping ex-+ p (ex).
Therefore, ·
~ ; ' y{j) A(j)
x8 ~ Qit' I ud28 Xe pje28e2ePir:p = 0, (4.9)
j
where x~> is a D-asymptotic differential operator preserving the

principal monomial at a 0 •
Let e be so small that the whole of supp (f 28e28 ) is in an admis-
sible y-patch (u, n~). Then (4.9) can be rewritten in the form
"'V''Vo '\:1 yvov Ur(J)f ~ (j) 0
o/vioLJ IoiUJXe 2eXe e2ePjfP= .
j
The element
'ljl = }J V1gimx~>f2eX~>e2ePicp
j
is non-zero since the classes r:p and 'ljl have representatives with the
same principal monomial. But this contradicts the fact that 6%1
is a monomorphism, and the proof is complete.
Let JJ8 (p, x) E Soo (R 2 n). Since the operator JJ8 (;, ~) preserves
the property of h-asymptotic series to be equivalent to zero, it
induces an operator acting on .Jl: (Rn) to itself which will be also
denoted by JJ8 (~, ~) .
Theorem 4.1. Given an $£ E Soo (R 2n) and a canonical operator Q/C
on (An, rn), there is such an operator P Qff) acting on .Jl (An, rn) to
itself, that
&e (;, ~) Q/C = 6% P QJO·

The following lemma is an essential ingredient in our proof of
this theorem:
Lemma 4.2. Every D-asymptotic quasi-identity operator is an auto-
morphism of .Jl: (An, rn).
Proof. Let V be a D-asymptotic quasi-identity operator regarded
as an endomorphism of ./1; (An, rn). It is clear that V is a monomor-
phism. It remains to be shown that Vis any.
Let r:p E A (An, rn). We have to prove that the equation V~J = cp
(where cp is to be found) has a solution. Let q;
be a D-asymptotic
series representing r:p. We shall seek a D-asymptotic series 'P
00
representing 'ljl in the form ,P = }J 'lJli· Setting 'ljl 0 = m [q;J, we define

j=O
'Ph j = 1, 2, ... by the following recurrence formula:
~Ji=m [ ~-
j-1
h V¢h J
l.
h=O
00
Let us check that the series h 'IJli is D-asymptotic. To do this we

i=O
observe that since
¢H 1 =m[~- ;~1 V'IJlh-vm[~- ii{v'IJlhJJ

k=O h=O
and V is a quasi-identity, it follows that for any j, at least one of the
following three statements is valid:
(1) ord 'Pi+t > ord 'IJli;
(2) the type of '1Jli+ 1 is greater than that of 'IJli;
(3) 'Pi = 0 (in this case 'lj.lh = 0 for k ~ j).
Since the type of a monomial does not exceed its order it follows
that for any natural N, there is j 0 such that ord 'lj.lj ~ ~ if j ~ j 0 ,
which means that the series under consideration is D-asymptotic.
We now show that V,;j) ~ ~. Let j 0 (N) be an integer-valued func-
tion of such a natural variable, that ord 'IJli ~ N/2 for j ~ j 0 (N).
Then
ord ( V . ~
J=Jo(N)
'Pi)> ~
and
0'd ( V ~:~:''I>;-~)~ O'd ( -'l>lo (N)) > ~ '

so ord (V;p- ~) ~ N/2. Hence ord (V;p- ~) = oo, and the proof
is complete.
Note. It is easy to see that if V is a quasi-identity then the same
is true of V-1 • .
Proof of Theorem 4.1. We begin by commuting J/8 (;, ~) with

a local canonical operator. Let (u, n~) be an admissible y-patch and
let <p E .It (u). By the commutation formula for a Hamiltonian and
the composition of the multiplication operator by a complex exponen-
tial with the Fourier transformation, we have
Q/8 (~, ~) 6'/Ct<p =

where QJe is the D-asymptotic differential operator defined by (2.24)

("lifted" to An by the diffeomorphism :n:~)· It follows that
C18 (b, ~) &rY =6i({P~, (4.10)

where
I,y
PQ/e =Rr
"V- ~
J C18 1 v-,
1
Jr
and RJ is the in verse of LJ in the group of automorphisms of .11: (u).
Equation (4.10) implies that for any qJ E .11: (An, rn), there is such
a family {'ljli} of elements of .J/: (An, rn) that
Let us show that there exists an element 1j1 E .Jl; (An, rn) satisfying
the condition
'1:1 f!/1,-
AV' '\' (i) • h "'/'
LJ l(j) 't'j = f!/l,'ljl. (4.11)
j
To do this, it suffices to find, for any j, such an element t!Ji E .11: (ffii)
that
y(j) -
f!ltr(i) ¢i = f!lt"¢i· (4.12)
Rewrite (4.12) in the form
f!Jt}H] tiJi = 2J f!Jt}gj Vii1Wt? p
i
j1ilj· (4.13)
The last equation is equivalent to the following one:

.JI:jlj)j = ¢i (4.14)
with
But note that .11: i is a quasi-identity operator, so the equation

(4.14) has a solution by Lemma 4.2.
Thus we have shown that for any qJ E .11: (An, rn), there is a '\jJ E
E .11: (N', rn) satisfying the equation
C18 (b. ~) f!JtqJ = f!/t'ljl,
such a 'ljl being unique by virtue of the fact that f!Jt is a monomorphism.
It is obvious that the operator P Qfe taking qJ to 'ljl is linear.
5. Canonical operator on the family {Af, r;'}. Let {Af, r?},
0 ~ t ~ T, be the family of Lagrangean manifolds with complex
germs obtained from (An, rn) by the family of complex canonical

transformations associated with a Hamilton function J/8°, and let
Mn+1 be the (n + i)-dimensional manifold (with the complex germ
w (a, t), z (a, t), the dissipation D (a, t) and the potential E (a, t))
associated with {A~, r?}. The canonical operator 6/t Mn+t on Mn+l
is defined almost word by word as that on An. Here we shall mention
only a few additions and modifications required.
(1) One must replace the space .Jl (An, rn) and Jl (Rn) by
.:/!: (Mn+l) and .J/; (Rn X [0, Tl), respectively; the definition of these
new spaces is obvious. Note that there is a natural embedding of
.J/; (An, rn) into Jl (Mn+l).
(2) The complex vector fields 8/ol must be chosen so that they
would be expressed locally via a only.
(3) One must replace the diffeomorphisms n~ in the definition of
local canonical operators by II~ (see Sec. 4 of Chapter IV).
It is not difficult to verify that for the canonical operators on
Mn+l, the results similar to that of Propositions 4.1, 4.2, 4.4, 4.5
and Theorem 4.1 hold.
It is of special interest to consider the commutation of the canonical
operator 6/t Mn+1 with a pseudodifferential operator of the form
~ a z ) (1
H=-ihTt+OJt p, x, t, h , rfJ8Ecffoo (4.15)
in the case with

OJ(] (p, q, t, 0) = 0J8° (p, q, t). (4.16)
We shall agree to say that Mn+1 is subordinated to the operator fi
defmed by (4.15) if the condition (4.16) is satisfied.
Theorem 4.2. Let Mn+1 be subordinated to the operator fi defined
by (4.15). Then
H6'/C Mn+i = - ih6'JC Mn+i ( :t + G (p, q, t)) X,
where x is a quasi-identity operator and
G(p, q, t)=- ~ tr $8~q(p, q, t)+i&'8h(p, q, t, 0).
Proof. We have
~
H H~ ~ M:/"vU> ~ MJ'vU> H-
6'/CMn+1 = ..:;:.J 'Yb I(j) Pi= .4J 'Yb I(j) iPi>
J J
where iii is the operator that arises when commuting fi with the
local canonical operator 6/ttm. By virtue of (2.24) and the results
of Sec. 6 of Chapter IV, iii has the form

li·=
J
-ih (~+G+A·)
dt J '
where Ai is a D-asymptotic differential operator belonging to the

class ff'. Let P be the operator taking any <p E .11: (~+1) to the
solution 'ljJ of the equation
iJ~.Mn+i<j) = ~.Mn+1'i' (4.19)
(recall that this equation has a unique solution). Set
'I' = - ih ~~ - ihe<p- ihG<p, (4.20)
where e is an operator to be found. Substituting 'ljJ = P [ • +A iP j<j) l ·
I(i) Pi<r (4.21)
J
Let {Vi} be such a family of quasi-identity operators that
n . .
~M) (pj<p +Ajpj<p) = ~An+1Vj (pj<p+ Aipi<p).
Then (4.21) can be rewritten in the form
e<p= Li vi (Pi<r+AiPi<r)·
j
Since :2]
j
Pi = 0, one of the following two statements concerning the
element x = Lj ViPi<r is true: (1) the order of X is greater than that
j
of <p; (2) the order of x is equal to that of typ <p. The
same is true if xis replaced by I= :2] AiPi<p, where I is the operator
j
on .11: (Mn+l) into itself induced by the operator of integration with
respect to t:
t
f (a, t) -+ \ f (a, -r) d-r.
0
It follows that
t t
-Sadt Sadt
x=1 +e o IeO e
is a quasi-identity. But this x satisfies the condition
:e +G+e= ( :t +G) x,
so the theorem is proved.
CH. V. CANONICAL OPERATO-R ON A LAGRANGEAN MANIFOLD 481
The equation ( ft + G)
X!Jl = 0, where x and are the same as G
in Theorem 4.2, will be called a transfer equation.
For any (p E .J! (M11 +I), we shall write !Jl lt=o to denote the element
of A (A", r") induced by ~ !Jli lt=o• where ~ !Jli is a D-asympto-
iEJ iEJ
tic series rep-resenting !Jl· The definition of !Jl lt=o is obviously correct,
.e., it does not depend on the choice of a representative of (p. Simi-
iarly we defme 'ljJ lt=o E .Jl; (R11) for any 'ljJ E .Jf (R11 X [0, Tl).
lmther, we set
It is readily seen that rJt An is a canonical operator on (A", r"). We

shall say that rlt 111n+l is correlated with &fAn. Given a canonical
operator i>'/{ on (An, r"), Lhere is always a canonical operator rJt 111 n+I
correlated with rJt (to construct rJt Mn+l, it suffices to co-ordinate,
in 1ho natural sense, the branches of the phase arguments of J acobians,
the weighting aLiases, Lhe weighting partitions of unity and the
vector fields 8/0l).
Consider the Cauchy problem for the transfer equation:
{:t +G) X(p = 0, (p lt=O = !Jlo· (4.23)
The (unique) solution of this problem is

(p = x-lV-l!Jlo•
where V is the quasi-identity operator in .! (A11 , r 11) defined by
V'ljJ = (x- 1 ~J)It=O·
As a direct consequence of Theorem 4.2, we have the following
resuh:
Theorem 4.3. The Cauchy problem
. ,;¢ c-•, ( 1 2 )
-~hTt+ci!t p, x, t, h ¢=0,
has the asymptotic solution
\jJ = Vi'Mn+i(j),
zohere !Jl is the solution of (4.23).

In conclusion, we note that Theorem _4.3 can be generalized in an
obvious way to the case where the Hamiltonian function QJ8 may
depend on an m-dimensional parameter ffi.
Sec. 5. Proof of the Main Theorem

Notation. If
.. ,
th~ components of B commuting, then we write
1 n n+ 1 ) I 1 2 )
f ( At, ... , An, B =f \A, B .
If a = (a1 , • . . , a p = (p1 , . • . , Pn), then we write
11 ) ,
(a 1 '}..p 1 , ••• , an'Ap")=a'AP;

here 'A is "one-dimensional".
Here we shall prove the Main Theorem (see Introduction) using
the lemmas of Appendix to this section. Let us first deal with the
most elementary versiori of the Main Theorem which is stated on
page 132.
Let f (x, a) be the symbol of the operator to be quasi-inverted and
let Li (11, x, ~) be the symbol of the differential operator Li =
= Li ( i !x, ~. -i a:)
representing Ai· According to the conditions
of the theorem, the composite function f (L (11, x, ~),ex) is asympto-
tically p-quasi-homogeneous in x and ~. where p = (p1 , • . . , Pn+m)·
One can check that Li (0, x, 0) = xh so that f (x, a) is asymptotically
p'-quasi-homogeneous in x, where p' = (p1 , . . . , p11 ). Lemma 3 of
Appendix implies that the problem of quasi-invertion reduces to the
following: one has to find a function g N (x, a) satisfying (9.6) of
Introduction with 1 replaced by a p' -quasi-homogeneous in x ftmction
F (x, a), supported in the vicinity of the zero-set of f 0 • LeL s~- 1
be the unit "quasi-sphere" in R":
s~- 1 = {x ER" I Ao (x) = 1},
where A 0 (x) = ( ~x 21 Pi) 1 1 2 . Using a partition of unity we can
additionally suppose the support of F (x, a) restricted to s~-l X Mm
to be so small, that there exists such a T > 0 that 0 < T <
< .-' (q0 , ffi, 0, 0) and
H (p (q 0 , ffi, p 0 , 11, T), q (q0 , ffi, p 0 , 11, T), ffi, 'l'J) < 0
for (q 0 , ffi) E Sl.J.PP F lsn-1 x Mm and (q 0 , ffi, p 0, 'l'J) E Qe (see the absorp-
P
tion condition). By the rule of reduction (see p. 108), it suffices to
find, for any natural N, a function 'ljJ N (a, 11, x, t) vanishing if x
is in the neighbourhood of 0 (say, for I x I< c), belonging to Cz
CH. V. CANONI'GAL OPERAT"OR ON A LAGRANGEAN l\IANIF10LD 483
for any fixed t and satisfying the following conditions:

lJl.v(a, 1'], x, T)=Oz(lxi-N), (5.1)
(
-l
.Ar- 1 ( .1. ) 8\jJN ( 2
o .x --a! a, lax'
• 3 x, t
1 )
+
1
+[f ( L11 ... , Ln, a
n n+i) (' 2
]lJlN a, i Ox' ~ ~. t) =
1=JBN(~, i:x' ~' t), (5.2)
l·53lN(a, 1'], x, t)=Oz(ixi-N) uniformly in t,
\jl;v(a, 1'], x, O)=F(x, a)p(lJ) for lxl>c, (5.3)
where p (lJ) = 0 for 111 I > e and p (lJ) = 1 in a neighbourhood of
2
the origin. Note that F ( x,
1 2
a) p (' i :X ) 1 = F (x, a) 0 Z (I x j-oo) +
which can be shown by partial integration with respect to y in
lhe formula
F (.~, ~) p (i!) 1 =F(x, a)+
+(2ntn) ei<1l.v-x>[-1+P(1'J)]F(y, a)dl']dy.

R2n
Further, it is convenient to replace A 0 (x) by such a Coo function

A (x) that A > 0 and A (x) = A 0 (x) for large x.
Now let us transform the Hamiltonian f ( First of all, note L, d).
that the condition for asymptotic quasi-homogeneity of a function
cp <S), s = (s 1 , . • . , sn), may be written as follows: for all A.> 0,
k
cp(A,Pt~h ... , A,Pnsn)= ~ A.ricpi(s)+A.ro-tcpn+t(S, l),
j=O
where r 0 > r1 > ... > r 1, cpi is quasi-homogeneous of degree ri
for j = 0, ... , k, cpk+ 1 E e'f 00 (Rn X (0, oo )) being bounded as
A,- oo as well as all its derivatives with respect to £. Thus, any
asymptotically quasi-homogeneous function cp (s) can be expressed
in the form
cp (s) = A.ro C~o A.rrro cpi (A, -ps) + A,-tcpk+t (A, -rs, A.)}. (5.4)
Set fji = r 0 - rh j = 1, 2, ... , k, so we correlate a k-tuple

(e1 , • . . , ek) of positive numbers with any asymptotically quasi-
homogeneous function.
31*
484 OPERATJONAL METHODS
Using the representation (5.3), the composite function

f (L (lJ, x, ~). a) can be expressed in the form
t (L (lJ, x, B), a) =
= 'Ar [$6' 1 (a, 1'], ~. w, x) + 'A- $81 2 (a, 1'], B, cu, 'A)l, (5.5)
where w = 'A -rx, x = ('A -et, ... , 'A -eh), JJ8 1 and JJ8 2 belong to <i! 00
,
lm J/8 1 ~0, and JJ8 2 is bounded as 'A-+ oo, together with all its
derivatives in a, 11, ~. w.
Taking into account (5.5) and using the note following Lemma 2
of Appendix, we obtain the following expression of the Hamiltonian
f (l, ~) =
2 1
. 1 2
= Ar["Jo
( 2 1'
a , - l·A-1
clir. 1 a, l. Tx
11'
fJ , x A-r ,
fJa
1'
A -e
)
+
+ - ( 1' )
iA- 1 o'J8 3 a, i
( 2
:r , - iA-
2
1'
1
1
!._ , xA-
1 1'
1,
1 ')
A ], (5.6)
where e >0, JJ8 3 (a, 11, ~. x, 'A) E<ffoo is bounded as 'A-+ oo, together
with all its derivatives with respect to a, I], ~. x, and A -e =
= (A -Et ' . . . , A-Ek) .
To solve the problem (5.2) with the Hamiltonian function f ~) (£,

of the form (5.6) we use the theory of canonical operator on a Lagran-
gean manifold with a complex germ developed in this chapter.
Set
def
JJ8 0 (X 0 , X; P 0 , P; w, Xt. X 2) = =
=&8 1 (Xo, X, P 0 , w+A 1 -PP, A -e),
where
XoEMm, X=(Xt. ... , Xn)ERn, PoERm,
P=(Pb ... , Pn)ERn, w=(wb ... , wn)ERn,
Xj-
~ ( -E1
A ' ... ,A '
-Ek)
def (A1-r 1 A1-pn)
%2 ' ••• ' •
The function Q/8 0 depends smoothly on the parameters w, x 1 and x 2

for w E Rn and x 1 -+ 0, x 2 -+ 0. Moreover, J/8 0 tends to the Hamil-
1 n n+1)
tonian function JJ8 corresponding to the operator f ( A 1 , • • • , An, B
as x 1 -+ 0, x 2 -+ 0 (for the defmition of the Hamiltonian function,
see Sec. 9 of Introduction).
CH. V. CANONICAL OPERATOR ON A LAGRANGEAN MANIFO:UD 485
We will regard a/8 0 as a function on the phase space

{Xo, X; Po, P} = (Mm X R") X Rm+n.
Denote by Mm+n the space JV!m X R" and by T* JVr'+" the phase
space. It is clear that Mm+n is a Lagrangean manifold. The condition
of the asym ptolic p-quasi-homogeneity of the Hamiltonian implies
that Im rf/e 1 ~ 0, hence
Im & 0 ~ 0.
Set H 0 = Re 6JI? 0 and consider the system of bicharacteristics
iJX 0 ·- all 0 X X ]J p )
~- DPo ( 0• ' O• (J)+X2 'XI'
Xo lt=O =a,
aX" -_ rJllo
-;;--p
X X j'J
w+ X2 p 'X1 ) '
- 8t (J It ( O• ' 0•
X It lt=O = 111" (5.7)

rJPo _ Dllo (X X p p )
----;;t" -- - iJXo 0• ' 0• ()) ·t X2 ' XI '
Pult=O = 0,
DP11 Dllo X )
---;;t= -- DX" ( O• X, Po, ffi+xzP, XI'
ph lt=O = 0,
k=1, .. . , n.
Since the lim iLiug function QJ8 = lim &8 0 satisfies the absorption
A~oo
conditions whenever the parameters of the system (5. 7) lie in Q 8 ,

this system has a solution for x 1 = x 2 = 0. Therefore, it is easy
to construct by the perturbation theory an asymptotic solution of
the system (5.7), i.e., functions X 0 , X, P 0 , P E coo which salisfy
(5.7) modulo A-N with N being as large as desired.
Denote by gJJo+illo (ffi, x 1 , x 2 ) the shift corresponding to this
asymptotic solution.
It is easy to verify that all the properties of the complex germ are
preserved if one replaces the solution of the Hamilton system by the
asymptotic solution of the system (5.7).
It follows that the family
( ]}f"'+" .m+n)
t ~ {Nfm+1t O} o-t-T
=
1 ' 1tgHo+-iHo ' ' ~ ~ '
t
obtained from Mm+n by the family of transformations g Ho+iHo ( ffi,
x 1 , x 2 ) (here T is the same as above) is a family of Lagran-
gean manifolds with complex germs smoothly dependent on the
parameters w = (w 11 • • • , ffi t E (0, T) and x 1 , x 2 • As formerly
11 ) ,
486 OPERATIIQNAL METHODS
we shall identify the family {Mr+n} with the (m n i)-dimen- + +

sional manifold Mm+n+I which obviously smoothly depends on ffi
for sufficiently small x 1 and x 2 •
Let 6'JCA be the canonical operator on Mm+n+I corresponding
to the germ rf<+n, with h replaced by A -1 *.
As is seen from (5.2), (5.3) and (5.6) we have to consider the fol-
lowing Cauchy problem:
. 1 _ 1 a¢ f ~ .~
-~A Tt \a, 2
~ ax '
.
X' t
\}
+
12 2
3
+ [&8 1-
1'
iAJ/8 3 ] lJ'
·
(a, i :x , X,
1 )
t = 0, (5.8)
'!' ( ~. i !x , ~, 0) = p (i !) P ( ~, ~),
where
2 2 1' ~ 1 1' 1' )
No
Q!Oj=CV'Ot
No ( ,..
vv, .
~ax'
a " aa.' xA- P, A-e '
-;A-t..!!....
,2 2 1' 1 11' 1')

rfJ8 3 = JJ8 3 ( a i .!.__ - iA-t _!_ x A -I, A
' Ox ' aa. ' I •
We will show that the problem (5.8) has an approximate solu-

tion of the form
'!'N (~, i fx, ~. t) = (6/l:lcpN) (~, i!, ~' A 1 -p, A- t), (5.9)
8,
1 1
where QJN is a function belonging to o'l' (Mm+n+1 ), and ffi = [xA-P]. 00
First note that by virtue of the choice of the initial manifold

Mm+n we have
(cifAcpN)(a, 'I'J, ffi, A 1 -P, A- 8 , t)lt=o=QJN(a, 'I'J, ffi, t)lt=O·

Consequently, for the operator (5.9) to satisfy the initial condition
of the Cauchy problem (5.8), it is sufficient to require that
cpN (a, 'I'J, ffi, 0) = p ('I'J) P (ffi, a). (5.10)
* More exactly, we interpret XA as follows: given a coo function IJl on

00
Mm+n+l (identified with an element of A (Mm+n+l)), let }.J ¢i (a., x, t, h) Le

i=O
an h-asymptotic series representing CYCIP (we omit the parameters w, x1 and
io
x 2 ); then (X AIJl) (a., x, t) means 2} ¢i (a., x, t, A-1), where j 0 is large
j=O
enough.
C'H. V. CANONfCAL OPERA'l"OH ON A LAGRANGEAN MANIFOLD 487
Making the Fourier transformation F-;,1~v in (5 .8), we obtain

.1 ' a'¢ ( 2
-zhTt a, y,
2 . 1
-Lay, t
)+
3 ( 2
+[&Bt a, y, -ih a~, -ihp
2 1' 1 1'
:Y,
1 1' )
h8 -
1' ( 2
-ih$8 3 a, y, -ih a~, -ih7iii, h- 1
2 1' 1 1' 1 1' )
] X
2 ! )
a,~,-i:Y,t =0,
n
( (5.11)
x¢
where h stands for the operator A - 1 ( -i
Now consider $ 11", i. e., iY(A with A replaced by h -1 • By Theo-
:J .
rem 4.2,
[ -ih_!_+JJ!Jt
at
(~ ' y ' 2
- i h l -ihpl+hP- 1(1) he\)1 -
aa ' ay '
-ihJ/8 3 (~, ~' -ih a:, -ih :Y +w, h- ](vrkcpJ= 1)
= ( -ih6%~ ( :t +q) xcpN) (a, y, w, hr-t, h t), (5.12) 8,
where q is a smoth function on Mm+1!+1 and x is a quasi-identity

operator.
Now we substitute the operator
\jJ N (
3 2
a, y, - Lh ay , t
. 1 )=
= (vr~cp_J (~, ;, -ihP L, hP- 1 , he, t)
into (5.11). Noting Lemma 5 of Appendix and (5.12), we obtain
- "h a C.vck(j)N) {\a,2 y,2 - ·hPl_ hp-1 he t) +

L iJt l ay , , , I
:J (2 2 ~
+[di'6'1 a, y, -ih-a
0
,
1 )
-ihp-0-, h 8 -
a ay
- ihJJIJ 3 ( ~' ~, - ih a: , - ih :Y , h- hx 1
X
/ (
~ cpN
/< .k ) (a,2 y,2 1
- L"hp iJy , hp- 1 , JB
L , t ) --
= - ·h[C'V*( -dt
L w1 q xcp J (2a, y,
d +) 2 _;,,Pl
" • a , hp-1,h,e). y I
(5.13)
1'
Substituting h=[A- 1 ( - i :Y)] here and making the Fourier
transform Fy .... x, we get
- l
'AI-1 fJ (~.lCjlN)
fJt
b
(a,2 l. a;;, 1
ffi, 1
\1 1-p
, 1
1\-e
, t
)
+ [3wo
#0
1-
. yJO ]
lwo 3 X
-1-q) 'XfPN J (2a, :x,

2
X [ <?%A1 ( ~ i
1
ffi, A1 1 -fl, A-e,
1
t ) , (S.14)
where ffik =A -rkxk, k = 1, ... , n, and we have clt·opped the argu-

ments
2 2 1
( . fJ 1' 11' 1' )
a, l_ - iA- 1 _!_ xi\ -p A -e
fJx ' fJa ' ' '
and
(
2
a, i _
~ . 1\' _ 1
-ll -
b 1 I'
xA- 1 A
1')
fJx' iJa' ' ,
of the operators Q/8 1 and QJ8 3 , respectively.
By virtue of (5.14) and (5.6), to solve approximately the Cauchy
problem (5.2), (5.3) it suffices to find a function rpN satisfying the
transfer equation
( ~ -j- q) 'XfPN = 0 (mod (9D (A-N)),

which is possible by Theorem 4.3.
Thus, we have constructed the required approximate solution of
the Cauchy problem (5.2), (5.3) for the operator f (L, a (take into
l 2)
account the estimates given in Lemma 4 of Appendix).
In fact, it remains only to verify that for t = T, the function
(5.15)
I
decreases as x-+ oo faster than any power of ~ . N ole that the
bicharacteristics (X, P) corresponding to H 0 (see (5.7)), which
depend on the parameters x 1 and x 2 coincide in the limiting case
X 1 = x 2 = 0 with the bicharacteristics corresponding to H def Rea'78,
where QJ8 is the Hamiltonian function of the operator f (A, B).
According to the condition of the theorem the latter satisfy the
absorption condition. It follows that the absorption condition is
satisfied as well for the bicharacteristics corresponding to H 0 , pro-
vided that min A (x) is large enough. Namely, the following in-
xERn
equality holds for t T:
=
Im 618 0 (X 0 , X; P 0 , P; ffi, x1 , x 2) ~ -6 < 0.
CH. V. CANONICAL OPERAT'OR ON A LAGRANGEAN MANIFOLD 489
It follows from this inequality and the construction of the cano-

nical operator, that the function ~Acp N it=T decreases faster than
any negative power of A as x-+ oo. One can check directly that (9.8)
of Introduction with '\jJ defined by (5.15) gives a solution of the
problem
{[!(.4, .Bh[gN(_4, .Bh=F(A, .B)+RN(.4., .8),

RN (x, ex)= Oz ( 1x 1-N).
The Main Theorem in the general form stated in item 4 of Sec. 9
of Introduction can be proved in the same way. Here we consider
in detail only the reduction of the generalized absorption condition
(see item 4 of Sec. 9 of Introduction) to the ordinary one (see item 3
of Sec. 9 of Introduction) and sketch a modification of the above
proof for the case of an operator f (A, A, £)
dependent on parameter
(but independent of the standard r).
The generalized absorption condition differs essentially from
ordinary one only in the case where the set Q 0 is non-connected.
In such a case Qe is non-connected as well, the function
~ (p (q0 , co, p 0 ,
H q ( q0 , co, p 0 , 'l'j, \,' T ') , co,\,' 'l'j )
' T ') ,
'l'j, \,
being strictly positive or strictly negative by the absorption condi-
tion on every connecLivity component Q~) of Q e (which corresponds
to a connectivity component of Q 0 ). Therefore, using the same argu-
ment as that at the beginning of this section, we can choose a cove-
ring of M 1 and a partition of unity corresponding to this covering,
so that the quasi-inversion problem reduces to the equations
i= 1, ... , s,
where s is the number of connectivity components of Q 0 and p is
such that the imaginary part of the Hamiltonian function
jj (P h=•'• q h=•'• co, v, 'I'J)

is strictly negative on supp p (if it is positive we multiply the
equation by -1). Thus the general case is reduced to the case of
an ordinary absorption condition.
Now consider a Hamiltonian
a: ]; ex, £1l
1
1 ('1 2 ) 2
f ( [L x, i Jx , - i
which does not depend on the standard and is (p 0 , • • • , Pn• Pn+ 1 )-

quasi-homogeneous of degree r with respect to the arguments
(-i a: = x0, x1 , ••• , xn; £), the numbers Po and Pn+I being equal
490 OPERATJ'ONAL MEJTHODS
to 1. Using the same argument as above (cf. (5.6)), we can express

f (£, :, ~) as the sum
1 2 ) 1 2 (' 2 ~ 1' 1 1' 1 1' 1' \
f ( L,a,~ =N[&'8 1 a,iTx, -iA- 1 0~,A-rx,v,A-e)+
+ ( -~"AI'_i) £!() ( 2. ~ 'A!'-1 ~ A1'-L1 1' AI')]

JQ/C· 3 a,~Tx• -~ iJa' x,v, , (5.16)
where A = A (x) is the same as above and

A-e=(A-el, . . . , A-ek), ei>0,j=1, ... , n.
The function 0/8 8 (a, 1), p 0 , w, v, A) is bounded as A-+ oo uni-
formly in all the parameters, provided that the functional space
dcf
where the Hamiltonian acts is such that the operator v = £-1 A
is bounded (see below).
According to the reduction rule, the quasi-inversion problem for
1 n n+1 )
the operator f ( A1, .•. , An, B , £ reduces to the Cauchy problem
of the form (5.2) with the Hamiltonian f ~. Now we express (l, £) .
the Hamiltonian in the form (5.16) and the further course of solving
the Cauchy problem does not differ from the one above. In parti-
cular, one constructs the Hamiltonian function
51 1-p _ 8 def
Ql'to (X 0 , X; P 0 , P; w, v, A ,A )=
=rfJft(Xo, X, P 0 , w+A 1-rP, v, A- 8 ),
where w = ( w1 , ••• , wn) are the parameters corresponding to the
operators
xA-P = (xlA-Pt, ... , XnA-Pn).
The function QJ8 0 tends as A -+ oo to the Hamiltonian function QJ8
1 2 )
corresponding to the operator f ( L, a, £ (see Sec. 9 of Introduction).
Further, one constructs, as above, the asymptotic solution of
the Cauchy problem (5.2) of the form
2 2 )
~)N ( ~~ i
; 9
:X , 1
X, t, ~ = ~A<pN
) (' 1 )
(
2 . {)
a, ~ Tx, w, v, A
1 1 1 -B 1 !-p
,A , t; ,
(5.17)
where 6%' A corresponds to Q/8 0 and wk = x"A -P,,, v = £- 1 A.
CH. V. CANONIGAL OPERATOR ON A LAGRANGEAN MANIF'OLD 491
The element g N of the quasi-inverse sequence has the symbol

T ' 2
. r I 2 . a 1 \ 1
gN(x, a,£)=~ J tPN \a, ~fu' X' t, £J A 1-"cp dt, (5.18)
0
where
cp = cp (x; a; £) = P 0 (x1£-fl1, . . . , xn£-Pn; a)
with P 0 (x, a) being as above.
The support of cp (x; a; s) is obviously contained in the domain
A (x) ~ c£. It follows that I v I ~ c on supp cp.
Thus, we have constructed a right-inverse sequence. A left-inverse
one can be constructed quite similarly with the aid of the same
Hamiltonian function. In fact, it suffices to nole that the operator
( n+ 1 2 1 ) ( 1 n n+ 1 )
[gN AI, ... , An, B] [/ At, ... , An, B ]
is equal to
n+ 1 2 1)
rN ( A 1, ... , An, B
with
( n+1 2 1 )
rN(x,a)=f L~, ... ,L~,a gN,
where
Lj (lJ, x, Po) def Li (-l], X, -Po)·
In conclusion, we consider the following generalization of the
quasi-inversion problem: it is necessary to find an approximate
solution ~ E :£, u; E :£ of equations
lv= P, zvl = P,
where
A ( 1 n n+1) A
f=f AI,··., An, B E.!!, vE:£

and
A
F=F A 1 ,
( 1 ••• ,An, B
n n+ 1) E.Jl
is such that supp F is contained in a f1xed domain 11 c S~- 1 X Mm.
More precisely, one has to find two sequences of elements of :£,
A ( 1 n n+ 1 ) A ( n+ 1 2 1 )
VN=VN AI, ... , An, B and WN=WN AI> ... , An, B
such that
JvN = P-+ R.N, wNJ = P-+ R.N.
where both RN and RN are in :£ and have the symbols decreasing

w
faster than I y 1-N. The elemenLs ~ N and N will be called respec-
tively right and left Ll-quasi-inverse, and the corresponding sequen-
ces {vn}, {wn} will be called a right- and a left-quasi-inverse sequen-
ce, respectively.
For the Ll-quasi-inversion problem the result similar to the Main
Theorem holds. It is only required to modify the absorption condi-
tion slightly, namely, initial points a 0 , q0 must now belong to the
intersection of Ll with a small neighbourhood, of the zero-set of f 0 •
Note also that the formula for a Ll-quasi-inverse element, which
was given for a particular case in Jntroduction, leads to the following
statement concerning the supports of Ll-quasi-inverse elements
(both left and right):
suppvNc::rt8 n_ 1 x,11 m U g~1 (LlnQe),
p o:st""T
where :rt 8 g-Ix,11 m is the projection onto s~- 1 X Mm, and
ik (Ll n Q e) denotes the image of Ll n
Q e under the canonical
transformation associated with the Hamiltonian function H.
Problem. Under the assumptions stated on page 138 and assu-
ming in addition that the operators A 1 , • • • , An commute, deduce
(9.75) of Introduction from the formulas (5.17) and (5.18) for the
right quasi-inverse sequence.
Note 1. In the particular case, when the operator f is not depen-
dent on B, the relation (9.1) of Introduction will be equiyalent to
the following:
,.!_ ~ ')- n/2 -
f ( Lt, · .. , Ln gN=(2:rt) 8-j-RN,
where {j is the Dirac 8-function concentrated at unity of the group G
induced by the generators iA 1 , ••• , iAn, RN=FRN is a sufficien-
tly smooth function on G, gN=Fg;v, Li=FLiF- 1 , and F is a Fou-
rier transform.
Note 2. If A1 , .•• , An, B are generators with the defining pair
of spaces (Bb .81), then by virtue of Note on p. 101 R 11 {.,L, ...
... ,An, n1/)gED c~A~)' VgEBt. In other \VOrds, the opera-
tor Rn is bounded as an operator on space BY into space B""+\
where BY is the completion D (see p. 211) in the norm
II u II BY= /I ( 1 -1- 1: -f- · .. +.Ad viZ u /In , Y ER.

Problem. Let f (x~> ... , Xn, a) be asymptotically homogeneous
function in x of degree k, where the leading term of f (x, a) =1= 0
CH. Y. CANONICAL OPERATOR ON A LAGRANGEAN MANIFOLD 493
for /x/=1. Prove that if Ai, j=1, ... , n, Bare generators of

1 n n+1)
zero degree, then the quasi-inverse operator gN ( At, ... , A, B
1 n n+1)
for the operator f ( At, ... , An, B for any y ER, 8 > 0, satisfies
the estimate II gl\. (rt, ... , An, nJ/) )/BY BY-m-e< const.
1 _,. 1
Hint. By virtue of Theorem 3.1 of Chapter II, to prove the

al1ove inequality il will be sufficient to prove that the symbol
grdx, ex) satisfies the condition (1+lxj2)k/ 2 -egi\.(x, cx)E$ 0 (Rn+t).
This fact follows from the estimate II (1 +I x j2)k/ 2 -egN lise n+t =
O(R )
=JJg.v ( - i -:-)
uy
(1-f1) 1' 12 -e)) C(Rn+t)..,.C(Rn+t) . By virtue of Lemma 3
of the appendix to this section, the symbol gN (x, ex) is an asymp-
totically homogeneous function in x of degree -k. As is well known,
the operator g 1y ( --i i~J) (1-f1)" 12 -e: C(Rn+ 1 )-+- C(Rn+ 1 ) is boun-
ded in this case.
Appendix to Sec. 5
Definition 1. Let (y, a; h), y E Rn, a E Mm, be a family of
/unctions belonging to r!f"" (Rn X Mm). We shall say that this family
is bounded in (!"" (Rn X Mm) if there exists such a k that
I(:U )P ( 0~ t (y, a; h) j ~ Cp, q (1 + 1y / )"

for any multi-indices p and q.
Lemma 1. Let (y, a; h) be a bounded family in r!f"" (Rn X Mm)
and let
n
Lj (11, x, ~. h)= 2J xkPk, i (11, ~.h)+ Pi (11, ~. h),
k=1
where Pk, h Pi are polynomials in 11 and ~ with bounded coefficients

dependent on h.
Set
. h' h ' ( 2 .1 j} 1 )
Lj=[Li f],X-lharj, ~,h ].
Then we have
<D {£',~;h) =<D ( L' (~, x-ih ~, (1)

where {Rh} is a bounded in r!f"" family of functions of the ordered ope-
0 (2 1 j} 1
rators11. x-iharj, ~, aand[L' 11,x-ih"Fr], ~,h)].
Proof. By the K-formula, we get
11 2 ) ( , ( 2 .1 iJ 1 ) 3, n+,1 n+2 )
<D ( L, a; h =<D, L 1 'I'J, X-lha:ij, ~' h , L2, •• • , Ln, a; h -
n 3 1 1
- ih ~ [ :~!
h=1
('I'J, x- ih :TJ , ~. h) Pk, 1 (_ ~' ~' h) X
()2<IJ ( ' ( 6 ! 8 1 ) 2 5 7
X <lyi L 1 'I'J, .T-lharj, ~. h , L;,L;,L;, ...
n+5
... , L;., a; h
n+6 ) J·
In the first term in the right-hand member let us change the order
of operating of '11 and L;. Using the commutation formula, we obtain:
' ( 3 _1 f) 1 2 4 n+1 n+2 )

= <D ( £1 'I'J, X- ~h Ol] ' ~. h) ' L;, L~ • . . . , /,~, a: h -
- ih ~ [Ph, ~' p, h) X 2 (
h=1
iJL' ( 2 2 4 6 6 .1 iJ 1
X iJTJ~ 'I'Jt, • · ·, 'llh-1• 'I'Jh, '111t+t• ... , 'I'Jn, X- ~h a:iJ, ~'h) X
X~
()2$ ( ~ ~
£1, £1, L;, L;, L;, ... ,
3 5 7 n+4 n+5
L~, a; h
) J, (2)
uYtuYz
where L1 stands for the operator expression
'(2 2 4 6 6 ~ f) 1 )
L1 '111• · · ·' 'I'Jh-1• 'Ilk• 'I'Jh-i 1• • · ·' 'lln• X-lh8TJ, ~' h ·
We now transform the expression of the first term in the right-

hand member of (2), first applying the K-formula to L; and then
changing the order of operating '11 and L; in the first summand of
the sum obtained.
Repeating this argument n times gives (1) with
n n-1 )
R = - i ( 2J rj + j=i
j=1
2J rJ ,
where
n+5-jn+tl-j )]
... , L~, ct.; h ,
with £<11.> standing for the operator expression
'(2 2 4 6 6 ! iJ 1 )
L "llt• ••. , "1111.-1• "Ilk• "1111.-JI• ••• , 1'] 11 , X-Lh7h], ~, h .
The assertion of the lemma now follows immediately.

Making use of estimates given in Sec. 11 of Chapter II, we obtain
the following corollary to Lemma 1:
Corollary. For any real r there is such an s that for all k, Rh is
bounded uniformly in h as an operator acting from c~ into c~.
Lemma 2. Let f (y, a) be the same as in the Main Theorem. Then
( 1 2) I (
f L, a =f(L\i :x' x, - i 8:
2 1 1) 2)
, a+
where {f,} is a family bounded in rff oo.

Proof. Define <D (y, a; h) and L' (1'], x, ~.h) by
hrf (y, a)= Q) (hPy, a; h),
hPjLj(1'], x, ~)=Lj(1'], hex, he~, h), (3)
then Lj (1'], x, ~.h) satisfies the condition of Lemma 1.
We have
f (l, ~) =h-rCD (hp£, ~;h)=

=h-rCD ([L' (- i :x, he~, -ih a:, hh, ~;h). (4)
We have shown in Sec. 11 of Chapter II that the operator
1 ( 2 1 1 ~ 2 I
CD ( [L' i :x, h 8 x, -ih 8 aaa, h 1 ], a; h)
can be expressed as the pseudodifferen tial opera tor
' 2 1 '
/ 2
\a, . a I e 1. 'he a h) . h
\jJ lax , ~ x, - l Ja , . wrt
tp(a, rJ, x, ~' h)=CD ([u (~, x-ihe a~1 , ~' hh, ~;h) 1. (S)
It follows from Lemma 1 that
cp (a, YJ, x, ~' h)=
=CD (L' (~, x-ihe ~,~'h),~; h) 1+hefh(a, r], x, B), (6)

where the family {h} = {Rhe1} is bounded in rffoo by the corollary
to Lemma 1. It is obvious that
CD (u (~ . .:r- ih a~J , ~' h) , ~; h) 1 = <D (L' (rJ, .:r, B, h), a; h), (7)
8
so
1 2) 2
f ( L, a =h-rCD ,L' i :x, h 8 x, -ih 8 a~,
j (
') 2 \ ( 1
h , a; h)+
+he-rfh (~ ' iax'

l he~ -ihel) =
' aa
= f (L (i lx , ~' - a: )' ~) -t-

i
/2 2 1 1 \
I lOx
+ I~ e-rfh \a, . a' he aJ'
x' - l'h 8 aa
Lemma 3. Let f (x, a) Er!J oo (Rn X Mm) be the same as in the
Main Theorem, and let / 0 (x, a) be the principal part of f (x, a).
CH. Y. CANONICAL OPERATOR ON A LAGRANGEAN MANIFOLD 497
Further, let § (x, a) be a p' -quasi-homogeneous in x function of degree

0, whose restriction to s~- 1 X Mm vanishes in a neighbourhood of the
zero-set of fo (x, a) restricted to s~-1 X Mm. Then, for any N, there
exists such a g N E ,f/oo (Rn X Mm) that
1 2'
(
f L, a) gN=§+RN, (8)
where RN E,f/oo (R" X Mm) and RN = O.z (I x 1-N).
Proof. Define F (YJ, x, B, a; h) by
hrf (L (11, x, B), a)= F (YJ, hp' x, hB, a; h).
By Lemma 2,
hrf (£, ~) =F (i :x, hp'~, -ih :a,~; h)+

+ hEj, ta, i :X
i 2 2 1 1 )
1 h 8X, - ih£ {)~ . (9)
Furthermore, by virtue of p-quasi-homogeneity of f (L (YJ, x, B), a)

in x and B,
. a 2 a 2 1 )
+hoFh ( z&, r' 1 .
h x, -lh&, a , (10)
where F 0 (YJ, x, B, a) is the principal part of f (L (YJ, x, B), a),

0 < 6::::; e and {Fn} is a family bounded in ,f/ F 0 (0, x, 0, a) 00
,
being obviously equal to f 0 (x, a).

· m
B y T ay 1or ' s expanswn iJ an d - l· iJa,
· l'h - 8 iJx iJ f or any pos1· t'1ve v we
get:
2iJ . iJ1 )
0 ( i 1 2
+h ff'v --iJ
he x , -l-iJ ,
a h 8 x, a, h , +
2 1 2
v ( i iJ . iJ . iJ e12)
+hQv 'h&• - l o a ' l&, hx,a,h, (11)
where ff'v (u, v, x, a, h) is a polynomial in u, v with coefficients

bounded in ,ffoo (if regarded as families of functions of x, a), and
32-01225
Qv (u, v, 1], x, ~.ex) is a polynomial in u with coefficients bounded

in ,ffoo as h ~ 0. +
Let A be a Coo function, A: Rn ~ R\ such that A> 0 and A (x) =
I x I for large x. Set w = A -p' x, a =A -ex, p = iAe.;.;. Replace
in (11) h by A- 1 (x) acting first:
1 2 (1 2) ]=/
A-r[/ L, ex 0 (w, ex)+
1
1 _6 (2 . a 1 2 1 _1 )
+A ff>v p, -~ aa., a, ex, A +
1
+A-vQv p, - i 8~, i
(2 1
:x,2 1
a, A- 1}.
1 \
Thus (8) becomes
[ !J(w, ex)+A
1 -0
ff>v
( 2 .
p, -~ aa., a, ex, A-
b 1 2 1 1)
+
+A-vQv
= A-r!F (w, ex) +A-rRN.
u, -i !a, lx, ~. ~. A- J
i 1) gN=
We will find such a gN that
[ f 0 (w, ex)+A
1 -0
ff>v
( 2
p, -~
. b a,
aa, 1 2 1 -1) ]
ex, A gN=
= A-r!F (w, ex)+ Oz (I X rN-rje) (12)

and
1 (2 1 2 1 2 1
N-VQ'I/ \P· - i :a.' i :X' a, ex, A- )gN=Oz(JxrN). 1 [(13)
We first solve (12). By the perturbation theory, we write
1
gN= fo(w, a) X
Ni j 1 1 0 ( 2 ~ 1 3 1 ) .
X~ (-1) [ (1 1) A- ff>v p, -iar;, a, ex, A- 1 ]1 X
J=O fo w, a.
(14)
Note that the right-hand members of (14) makes sense (and
belongs to rffoo) because !F (x, ex) vanishes in the vicinity of the
2 ~ 1 3 1 )
zero-set of f 0 (w, ex), and ff>v ( p, - i aa., a, ex, A- 1 , is a differen-
tial (hence local) operator.
Substituting (14) into (12) gives the condition
(15)
To estimate the left-hand members of (15) note that for large x,

pNq:>(<J)=A 8 (25s, q:>] (cr), (16)
where 25 s is a first-order differential operator.
Using this fact, one sees by elementary calculation that
1 1 -II 2 . a2 1 3 1 -1 ) N 1+ 1
[ (1 2) A ffov ( p, - ~ aa.., cr, a, A ] x
to w, a.
X A -r § (ffi, a) ck
r +(N!+1)6
< oo (17)
e
for any k. Pick such N 1 that (N1 + 1) l3 ~eN, then (15) is satisfied.
Similarly, (13) is satisfied provided that v > N, and the lemma
is proved.
Definition 2. We say that an infinitely differentiable function f (x)
belongs to the class if cr
\1 flies=
def
sup
I(1 + x 2) T--;:-a
alai
f (x) I< oo
1 xER''. lal:o::;s ox
for any natural s.

n 2, ) 1/2
As before, let. Pi?d, j=1, ... , n, and A(x)= ( ~ x/Pi
}=1
Given a C""-smooth function ,ff (x, s) defined on R~ X Rr, set
<D (s, x) =A (x) /1 (£, [A (x)]-p x),
<D 1 = Re <D, cD 2 = Im <D.
Lemma 4. Let g (x, £) be a function belonging (in x and jointly) s
to the class C~ and vanishing for I £ I > so and for I x I < B, and
suppose that <D satisfies the conditions
(1)for I£' I~ So, I£" I~ 0 , the mapping x._ x s 6: 1 (£', £•,x) +
is a diffeomorphism of the outside of the ball I x I < l3 into Rn;
(2) <D 2 is non-negative.

Then the operator
def ( 2 1) ( 2 1)
T = ei p, x g p, x ,
where p = -i ~' sends every function f E H-s, s ;;?: 0, into a function

belonging to the class co.::2s-2m- _.!._n-e '1/e > 0.
2 •
Proof.
(1) Set
T0 = ei!(;, ~) g (~, ~),
r; = e-i~(~. ;) g (~, ~),
and consider the operator
A l,r=To
def 1 *
(1 + x 2 ) l<pr (x) To,
where {jlr Ec~, O~<pr (x)~1, {jlr (x) = 1 for I X I< r.
We chim that
Az,r=Qz,r x, p, P , ( 2 1 3) (18)
where
Qz, r (x, ~', ~") = g (~",-;: (x, ~', ~")) X
X g(£', ~ (x, ~', £")) (1 +~2 (x, ~,, ~")) 1 X
X {jlr (~' (x, £', ~"))I J-i (~', ~", J; (x, ~,, ~"))I
and J; (x, £', ~") is the solution of the equation
~+ c5<D1 (1':' 1':" ~)
X T"''"''X=X.
In fact, for any hE CO', we have
Az, rh (x) =
= ( 2 ~)Zn ~ exp [i ( - :c"~" + x'~"- x'~' + x~' + <D 1 (~', x')-

R4n
-<Dd~", x')]g (~', x') g (r, x') (1 +x' 2) 1 <pr (x') h (x") X
X dx" d~" dx' d~'.
On making the same change of variables in the last integral as in
the proof of Theorem 1.1 of Chapter III, we obtain (18).
(2) The symbol Q1, r can be estimated as follows:
II Qz, r 1158o, 1!, I!(RnXRnXRn) ~Cz. k•
where k > l m + +·
n and C 1, k is independent of r. Therefore,
we obtain, for any h E CO':
~ I T~h (x) 12 (1 + x 2 ) 1 (/)r (x) dx =I (h, A!, rh)L2 l ~

Rn
~II h llifk II Az, r ilHk-+H_k ~Cz, k II h llhk

(note that(T*u, v)~=(u, Tv)L 2). Passing to the limit r~oo in the
last inequality gives
II T~h llii 1 ~Cz, k II h lih1"

so
T~ E Hom (Hk, lit)
for k> l+m +n. Let u, v ECO', then
J(Tou, v)L2 1=1(u, T~v)L 2 I~JiuiiH_ 1 IIT~IIHk-+H 1 JiviiHk'
so
II Tou IIH_k ~II T~ IIHh-+H 1 II u IIH_ 1,
hence
T 0 E Hom (H-z. 11 -h)
for k >l m + + n.
(3) We have
T=[ro(1+~f]e-<ll2(~.~)e(;, ~) (1+~2rr. (19)

where e (~, x) is such an infinitely differentiable function that
e (~; x) = 1 for I ~ I ~ ~ 0 and e (~, x) = 0 for I ~ I ;::;;:: 2~ 0 • The
r
operator T 0 (1 + x2 belongs to Hom (H -l+2r• H -k) for k > l +
+ m + n. Let us estimate the symbol
e-<1>2<~· xle (~, x) (1 + x fr =def h (s,
2 x):
Let f1 and v be multi-indices of length n. Then
I(:~ r (:X r e-<llz(~, x)e (~. x) I~ci1V (1 + x2)\ !11/4

which implies that hE J!Jk, 0 for r > : +~ n. Therefore, it follows
from (19) that
TEHom (H- H_ 8, 28 _ 2 m-; n-e)
for every s;::;;:: 0 and e > 0.
(4) Let {} m be a polynomial, )hen
{P(p)Tf(x)={P (~) ei(~.!)g(~, ~) f(x)=

=ei(~. !)g1 (~, ~) f(x),
where g1 (1;, x) = {}(1;) g (1;, x) satisfies the conditions of the lemma
as well as g (1;, x). It follows that if f E H _8 , then
{} (p) Tf EH 7 ,
-2s-2m- 2n-e
hence {P(p) Tf belongs to Hl 7 for any l (see Sec. 11 of Chap-

2s-2m-2n-e
ter II). To complete the proof, we must note the embedding
l- [!!] -1
Hi c. Ck 2 , Vk ER.
Let i!/C be a canonical operator on an n-dimensional Lagrangean
manifold with a complex germ dependent on a parameter ffi E Rn
in such a way that 6'/C<p induces class (6'/C<p) (y, ffi) of equivalent
modulo hoo (in the natural sense) series composed of a function
belonging to Cz (R 2n). Then, for any H E d/ (R 2n), 00
H u. -ih at+(()) [(6'/C<p)(y, ffi)]=(i!/CRH<p)(y, ffi),

where RH is the transfer operator (see Theorem 4.1).
Lemma 5.
[Hu, -ih }J][(i!/C<p) (;, -ih }J]=(i!/CRH<p) (;, -ih !) .

Proof. The pair (A 1 , A 2), where A 1 = -ih :Y,
A 2 = y, is a system
of generators for a nilpotent Lie algebra. Its ordered representation
(see Sec. 9. of Introduction) is (£ 1, .LJ, where L 1 = -ih-
8
8
x2
+
+ Xv L 2 = x 2 • Then one has for any f E d/ 00
(R 2 n), g E Cz (R 2n)
(see Sec. 11 of Chapter II):
[tCL.1J][g(lb 1J]=x (ll! 1J,

where X (x1 , x 2 ) =f (L1 , L 2) g (x1 , x 2).
Using this formula and taking into account the condition of the
lemma, we get the identity
[H(; ' -ih-l)][(~cp)

au .
(~, -ihl_)]=
au
=[H(~, -ih!+ffi)((~cp)(y, ffi))J .~
00=-thfii/, y =
2
y
=
Sec. 6. Cauchy Problem for Systems with Complex Characteristics

In this section the commutation formula of the canonical operator
on a Lagrangean manifold with a~ complex germ and the pseudodif-
ferential operator with a matrix valued symbol are introduced.
We also consider a method for the construction of the formal asymp-
totic solution of the Cauchy problem for systems of the form
a'l' (z 1 )
-ih7it+aJ8 ,x, p, h '¥=0
'I'it=o='I'o(x, h) (6.1)
where JY = -ihfx, ·078 (x, p, h) is an (m X m)-matrix with elements
being smooth functions of x E Rn, p E Rn, h E [0, 1] and gro-
wing (together with their derivatives) no faster than a power of
(I X 12 + Ip 12).
Denote
(i :h r078 (x, p, h) lh=O = J78(k) (x, p).
We have
aJ8 (~, ~. h)=~ (-ki!h)k &'8(k) (~, ~) +hN+1Q.N (~, Ah),

k=O
(6.2)
where the symbol Q.N (x, p, h) of the operator Q.N ( ~. ~, is an h)

(m X m)-matrix with smooth elements uniformly bounded with
respect to h.
Suppose the matrix r:!/8< 0 > (x, p), x E Rn, p E Rn is diagonalized

by a smooth non-degenerate transformation. Denote by A.1 (x, p), ...
. . . , Az (x, p), l ~ m, the eigenvalues of the matrix J/8< 0 > (x, p).
Suppose the multiplicity o: i of the eigenvalue A.i does not depend
on (x, p). Introduce the following functions
jj<h (x, p) = Im A,i (x, p), H (x, p) = Re Aj (x, p).
The inequalities
jj (x,p) ~ 0, j = 1, ... , l
are assumed to be valid everywhere in the text below.

Let Ei (x, p) be a projection operator on the subspace correspon-
ding to the eigenvalue A.i (x, p) of the matrix rf/8° (x, p). We see that
the matrix elements of Ei (x, p), j = 1, ... , l are smooth func-
tions of the variables x E Rn, p E Rn and the following equations
are valid:
l
~
E' j ( X, p ) E k ( X, p ) = Uj]l_ ~ 'AJ-EJ·
E j (X, p ) , L.J = JJ8<o>. (6.3)
j=i
Let An= {p = p (a), q = q (a)} be a Lagrangean manifold with

a complex germ r" and JJ8i (q, p) = A.i (q, p) be a Hamiltonian
function. Consider the family {Aj, 1, rj, 1}, 0 ~ t ~ T of Lagran-
gean manifolds with a complex germ generated by the complex
canonical transformation with the Hamiltonian function J/8 i (q, p)
from the Lagrangean manifold An = {p = p (a), q = q (a)}. Deno-
te by Mj+ 1 an (n +
i)-dimensional manifold
Mj+ 1 ={p=p<il(a, t), q=qUl(a, t);
(p<il(a, t), q<il(a, t))=g1(p(a), q(a))}
(w~th a complex germ . w (a, t), z (a, t), a dissipation function
D<'> and a potential E< 1> corresponding to the family {Aj, 1, r},t}).
Assume the following inequality being valid:
~i~ inf J A,i (p<il, q<il)-A.i (p<il, q<il) J?;:6,
1of=t (p<h, q<h)EMj+i
where 6 > 0 is a positive number.

We shall define asymptotic series in smooth m-dimensional vector-
functions similarly to Sec. 1 of Ch. IV. .
Let llm(Mj+l). be the set of all equivalence classes D< 1>-asymp-
totic series on Mj+ 1 •
Let .Am(Rm X [0, Tl) be the set of all equivalence classes h-
asymptotic series on Rn X [0, Tl.
Define the canonical operator on MJ+l by the mapping

r!/{ n+1: oftm (Mj+ 1) -+.,#m (Rn X [0, T])
Mj
similarly to Sec. 4, Ch. IV.

Now introduce the classes of operators defined in .Am(MJ+1).
Definition 6.1. Operator oft defined in .Jtm(MJ+I) and depending
on h-+ +O is of the @' i class if, for any admissible y-patch
uY E M']+I and the correspondi11g functional space oftm( U]), the
operator .:!1 is equivalent to the D< 1>-asymptotic differential operator
s az (a, t, h) ( a: r'
ili>O
the coefficients a 1 (a, t, h) being such (m X m)-matrices that
(a)· _ICI
l!,k-VD'Jl
·(h(/IJ+1)/2)+('<7 jji (3 D(J)·(hi!J/2))+
Vp,q,
+ h 1/20D<j> (h<lll-1)/2), Ill> 0,

(ao)t, k = (3 D<h (h 112 ) + (V p, qijU), eJ D<h (h(O))), i, k=1, ... , m.
Lemma 6.1. Let .Jl; E @' i• then the operator 1 + I oft is a quasi-
identity.
Definition 6.2. An operator B defined in .:Am (M']+I) and depending
on the parameter h -+ +O is of the class :£ i if for any admissible
y-patch Uj E Mj+l and the corresponcf,ing functional space .:Am (MJ+l),
the operator B is equivalent to the D 0 >-asymptotic differential operator
~
lli~O
bz (a, t, h) ( a: r'
the coefficients b1 (a, t, h) being such (m X m)-matrices that
(b)· _ICI • (h<lll+1)/2)+h1;z 0
l 1, h - VD<J>
. (huzi-1)/2) IZI>O,
D(J) '
(b 0 )t,k=GDUl(h'12), i, k=1, •.. , m.

The~ following lemma is obvious.
Lemma 6.2.
{1) For any B from :£ i the operator 1 +B is a quasi-identical
D 0 >-asymptotic operator.
(2) B1, B2 E Xi==>- B1B2 E Xi.
(3) Let .:fl: E ff' i> then I .J£ E :£ i·
(4) Let BE Xi> then B E ff'i.
Definition 6.3. Operator C defined in .Jtm (MJ+I) and depending
on the parameter h -+ 0 is of the class uY' i if, for any admissible
506 OPERATIONAL METH10DS
y-patch Ut c MJ+ 1 and the corr.esponding functional space, the

operator C is equivalent to the D<'>-asymptotic differential operator
c= ~ Cz(a, t, h)(a~r.
Ili>O
the coefficients c 1 (a, t, h) being such (m X m)-matrices that
(c 1);,k=C9vcJ>(h 11112 ), lli~O, i, k=1, ... , m.
Lemma 6.3. Let C E Jf/'i, then
(1) if Jl E fJSh then .JlC E fJSi, CJl E fJSi;
(2) if BE Xh then CB E Xh BC E Xi;
(3) if C1 , C 2 E .ff'h then [C1 , C 2 l E Xi.j
The proof is immediate.
Note that we write .fl E fJSi (ut} (corresp. BE Xi (u1}, C E
E Jfl'i (ut)) if .fl (corresp. B, C) is a nd>-asymptotic differential ope-
rator with coefficients satisfying the conditions of Definition 6.1
(or 2, 3) in an admissible patch u1.
Let 6'/CAn be a canonical operator on (An,rn). Suppose that the
canonical operators 6'!CMJ+1, j = 1, ... , l (cf. Sec. 4) correspond
to the operator 6'JC An.
Similarly to Sec. 4, suppose that the initial value 'l' 0 (x, h) of
problem (6.1) has the form
'I' o = 6'fC A n(j)o, (6.4)
where IPo E .:lim (An, rn).
Let F (x, p) be an (m X m)-matrix with elements from (1 00 (Rn X Rn).
We have the following lemma.
Lemma 6.4. For any qJ E .t~,m (MJ+l) there is an equation
[F(~, th 6'fC n+i(j)=6'!C n+iRjqJ,

Mj Mj
where Ri is an operator defined in .flm (Mj+ 1):

Ri= ~ ( -ih)k R~kl
k>O
R~0 l=F(q(a, t), p<il(a, t))+~i
~iEXi
~ (-ih)k R)klEXi.
k>1
The operator R~k) restricted on any admissible mapping of M)+ 1
is a differential operator with the coefficients being smooth func-
tions of h.
Proof. Consider the local canonical operator &r}. For the sake
of simplicity we restrict ourselves to the case of a non-singular
patch u of a non-singular zone, the general case being treated along
the same lines.
Let cp E .!l (U) where U is a non-singular patch. We have (cf.
Sec. 4)
(O)
6ir cp = e
. ( h1 <1> ( i> (a, t)- 1 Arg J (a, I) )
t 2 I J (a, t) ~-1/2 X
X Locp (a, t)a=u-i<xl' (6.5)
where <!>< 1> (a, t) is a phase in the non-singular patch U. On applying
the commutation formula of Hamiltonian and complex exponential
we have
where
s0 l (x, t) = uPl (a, t). n- 1•
The operators Gh have the form*

2h h
Gk(S(i))= ~
I v I>O
{
~ ~ ~I
I a l=h I (3 1=0
( ai~xSJ
0
f' X
X fp+vp (x aRe s<il) <t><h> (sch )~} (6.6)

apa+v ' ax a(3 ax!3 '
where <!>~'~ are polynomials with respect to the derivatives

al v ls<il
I'V I= 2, ... , 2k
having matrix coefficients. ~We have <D~~ = 1. The estimates

a Ims<il
ax = 0 n (h 1/2), (6.7)
aRe s<il
iJx
p{j) (a(x, t,
)
t)= 0 n<h (h1/2 )
proceeding from the theorems in Ch. IV. Here the functions

a (x, t) are solutions of the equation q (a, t) = x.
* From this point on, two D-asymptotic operators are defined to be equal
in the sense of the equivalence of the asymptotic series (cf. page 480).
On applying (6.6) we obtain

G0 (S(i)) = F ( x, 8 R;xs<il ) +
+ ~ 1 (.
L.J -y., ~
iJ lm s<il (x,
aX
t)) 'V -...,
iJ'V (
F x,
aRe s<il )
ax . (6.8)
iJp
1'V 1?1
On applying (6.7) and the Garding inequality, we can put G0 (S)

in the form
G0 =F(q<il(a(x, t), t), p<il(a(x, t), t))+G~, (6.9)
G~ EXi (U).
Note that according to Sec. 6 of Ch. IV, there exists such an
operator R E ../f!'i (U) that
L 0R = 1 (6.10)
in the ring of homomorphisms over .Jim (U).
Put (6.5) into the form
F (~, ~) [6%<0lcp) = 6%<DlPcp,

where
P=R { ~ ( -ih)k (J)1!2Gk (S<il) (J)-1/2} Lo cter ~ (-ih)k p<k>,
k?O k?O
(6.11)
where
p<kl = R (J/12 Gk (S<il) (J) -1/2 Lo.
On applying Eq. (6.9) and the definition of the operators L and R

(cf. Sec. 6, Ch. IV) we put the operator P< 0 > in the form
p<Ol = F (q<il (a, t), p<il (a, t)) + P', (6.12)
P' EXi (U).
In the same way for any patch U1 we obtain
F ( x, p
2 1) 6% 'V
1
'V
cp=6% 1 P 1 cp,
'V
where the operator PY is defined in the patch U1 and satisfies the

conditions of the lemma.
Let {U1, IT~} be an atlas of Mj+ 1 which may be different from
the weighting one, and let cp E .Jl (Ut). Then (cf. Sec. 4)
vv<i)'V
<'V'
<!/b n+ 1 (jl = ('f/''\' " '
<!/b I L.J I(i)I Pi(jl•
Mj i
CR. V. CANONICAL OPERATOR ON A LAGRANGEAN MANIFOLD 509
Hence
[F (~, ~ h(6% Mj
n+1cp) = 6'1CjP1 ~ viWipicp.
i
Let VIv n vkv' =I= 0 and let cpEA m (VIv n Vx).

v' Now prove the
equation
V' V'V VV' V
PxVxJ(jl=VIKPicp. 3
(6.1)
Indeed,
F ( x, p
2
~) V V V
[Q/l'Icp]=6%IPicp=6%xVIKPicp.
V' VV' V
(6.14)
On the other hand,
F U, ~) [6%tcp] =F (~,
The last equation and Eq. (6.14) involve Eq. (6.13) by the mono-
morphism property of the canonical operator.
Now define the operator Rk. in the patch Uk. of the manifold Mj+ 1
by the formula
v 1 ( ~ pv' ( ~ vv'(i) )
Rx Ul: = f v'v'Ci)
V KI(i) Pi
) - 1
x f
V KI(i>Pi , (6.15)
where the sum has taken over all indices i such that supp Pi n
nUl=!= 0 (cf. Sec. 4, Ch. V). Since the operator ~ Pi is
i
n;mr
quasi-identical, the operator Rl: iuv can be put into the form
K
Rl:iuv =F(qU>, pei>) iuv +R1, (6.16)

K K
where
Rx'v EXi Iuv.
K
Let the patches Uk. and Vi intersect. Prove that the operators
Ri iuv and Rk [uv' coincide on the space Am (Uk nUl).
I K
On applying (6.13) we have, for any rEAm (Uk. nUi}
V'
( Rx Iuk_ )r = -[ (f~ VKI(i) V''l''(i) V'
Pi ) -1 Px ( ~
f V'V'(i) )
VKI(i) Pi Jr =
= [ (
V'V'(i) ) - 1 V' V'V ( ~
f~ v KI(i) Pi Px v KI f v II(i)
VV'
Pi ) Jr =
= rl. (V""'
IK f~ KI(i) Pi )
yv'v'(i) -1 pVI ( f~ yvv(i)
II(i) Pi
) Jff]';"
F = (R"I IUj ) fffl'
F •
510 OPERATIONAL :viETHODS
Thus we obtain the operators Rk considering the operator Ri

defined Qn Mj+ 1 on the patch Uk.. The following equation is valid:
F G, ~ hQJC n+1CJ)]
Mj
= Q7C Mjn+1Rjcp.
Now the lemma follows from (6.16).
Let Bk, k = 1, ... , l be the operators defined by the following
formula
' 2 1)
Ek ~x, p [QJC M~+d = 67CM~+1Bk'·
On applying Lemma' 6.4 '
Bk = ~ ( -ih) 1 B~>,
l:?O
where
Bk.Ol = Ek (q<i)' p<h) + ~jl (6.17)
~jEXi.
Lemma 6.5. The following equations are valid:
l
(i)1 ~ B;o'= 1,
. i=1
(ii) B~0 >Bk_0 > = 6rkBk0 >.
Proof. Without an-y loss of generality consider a non-singular
patch of a non-singular zone. On applying Eq. (6.11) and the defi-
nition of the operators BjO>
B~0> Bk0>= (~ V~tHlPi) - 1 R (J) 112 go, rgo, k (J) - 112 Lo (~ V8iiJ.)pi),
i i
where
go, r-- L.J 1 (..!!._E )
"" -y! 0 v r (i iJ Im S )"
iJx •
IV I>O p
Hence
Thus (ii) is proved. In the same way we can prove (i),
Theorem 6.1. (1) The following equation is true for any j =
=1, ... , l:
[ - ih :t +df8 U, ~~ h) J[Q/CMf+t(j)} =67C Mj+1Ricp,

where Ri = ~ (- ih)k R]k> and the restriction of the opera tor
k>O
an admissible patch of Mj+ 1 is D-asymptotic differential opera-
tor. Operators R]k> are differential operators on Mj+ 1 having
CR. V. CANONICAL OPERATOR ON A LAGRANGEAN MANIFOLD 511
smooth coefficients with respect to h, such that

Rjo> = 2} ('All. (q<\ p<h)- 'Aj (q<\ p<j>)) Eli. (q<\ p<j>) + ~;v, ~t Ezj.
hof=j
(2) s~ol R'·l) B'·Ol = B\Ol (~- _.!_ tr a2'Aj
1 1 1 1 dt i 2 aq ap
+
+ aEapj a (Q!6'<0)- Aj)
aq
+ QJ8<il + ~\2)) s~ol
1 1 ,
~3\2) E ~J·'
where d~. is a vector field of the form

J
d~i = <H~> (q<h, p), :q)- <H¥> (q<h, p<h), a~) + :t .

~ a
Proof. Consider the opera tor H = - ih Tt OJ8 + (2x, 1 )
p, h . On
applying Eq. (6.2) and Lemma 6.4 we obtain
fj [6'/C M~+ 1 cp 1= 6'/C M~+ lR i(P'
1 1
where
R i = &'8jo> (q<h, pd>) _ 'Ai (qd>, p<h) + ~t,
'f,?>EZi n <fJ\.
We have taken into account that in the region QK the function
S'k (xK, ~K' t) =CD" (a, t) (TI}'d- 1
satisfies the dissipative Hamilton-Jacobi equation with the Hamil-
tonian function 'Ai (q, p) (cf. Sec. 1, Ch. IV). On applying (6.17),
(6.3) and Lemma 6.4 we obtain the first statement of the theorem.
To prove the second, consider the product of the operators
[Ej {:, JhH [6'/C M~+I!p].

1
On applying the commutation formula, we obtain
[EjU, ~hh=Ej(:;, ~) ( -ih :t )+[EjaW<o>(:, ~h+

+ ~ (-;:L)It [:;h Ei ah=~OJ (~, ~h+ ~ ~ (-ih)h+l X
11.~1 Z>!h~O
ah E. akc2f6'< z> ( 2 1) .a ( 2 2) ( 2 1 )·
--;!-
apn
ah
X
x,p =[Ei x,p ][-ih_Tt+'Ai x,p ]+
,, (-ih)lt
+ LJ [ ah£j ak (Q!6'<0) -Aj) ( ; } )] +
k! a It axlt ' p
k~! p
+ ~ ~ (- ih)k+l [ a;;i aha: U, ~h. (6.18)

l~! k>O
The following equation is a consequence of Sec. 4, Ch. V:
[ -ih! +J.j(~, ~)] 6%Mj+1cp=(-ih)6%Mj+lrjcp, (6.19)

where cp is an element of .Jim (Mj+1), r 1 is an operator on MJ+l
a
r1= dt 1 - 2 r
1 t az'},i
oqap-
t- 0<1?<4>
i '
<1?<4>
0 i
E riP
i·
(6 20)
·
On applying the operator E1 {:, ~) to both sides of (6.19) and

taking into account (6.18) we obtain
(-ih)[EJ {:, ~h(6%M~Hricp)=[Ei {:, ~)](6%M~+lRJcp)-

J J
(6.21)
where
M·= ~ (-ih)"' o"'EJ o"'(QJ£'<0>-'},i) ( : 2)
1 .:_)
R.;;,.i
k! [ ap k .
ax"' ' p ]+
"'Q "'Q (-ih)k+l ()kEj ()kQJ£'<1> ( 2 })
+ Z:>1
L.J L.J
h;;,.o
k! [ opk ox"' .r, p ].
On applying Lemma 6.4, we obtain
[Ej U, ~ h[6%M(l+lRjcp] = 6%M(l+l (.BjRj) cp.

J J
(6.22)
Now calculate the product B 1R 1•
On applying (6.15) in a patch Uj( of the zone QK, we have
B1R1 luv = (BJ luv) (RJ luv) =
K K K
= ( f
" ' VVV(i)
KI(i)Pi
) -1 p'VpV ( " ' vvv(i)
K K
)
KI(i)Pi • f (6.23)
Without loss of generality consider the product p'JP'k corresponding

to the non-singular patch of a non-singular zone. The sign ' denotes
the quantities of the operator B 1.
For the sake of simplicity denote P';/ and P'k by P' and P. By
(6.11) .
P' P = R (J) 112 ( L] (-ih)"' GJ. (S) L]( -ih) 1G1 (S<h) )(J) - 112 L 0 =
h;:,.O l>O
= R {(J) 1/ 2 + (- ih) x
(Go (S) G0 (S<h)
X [Gi (S) Go (S<h) +Go (S<h) G1 (S) + g]) (J) - 112 } L 0,
gEfP 1 (u). (6.24)

On applying Lemma 6.4 (cf. (6.8)), we obtain

2
G = "V _1_ _!!_(cfJt<O>-'A·)(i aimS<h)v +
0 L..J y! ap~' 1 ax
v=o
3 ·2 ~o
+Ov(h' )(1+go)+(H~, On<h(h)),
G'=
ll
~'
L..i
_'l_(iaimS<h)v_!!_E·(x aReS<h),
y! ax a ":' 1 ' ax I
I vt=O P
+Ov (h 312 ) (1 +go), go, giJ E Xi (u).

From the equalities
(&8< 0 >- 'Ai) E.i = Ei (c£1?< 0 >- 'Ai) = 0
we obtain
G 0G~ = G~Go = (On<h (h), iJ~l) + Ov (h312 ) (1 + g"), g"EXi(u).
(6.25)
Since the operators R and £ 0 belong to the class :£ i• we obtain
l>y Lemma 6.3 and Eq. (6.25)
P' P = (- ih) R {(!) 112 G;G 0 (J) - 112 +
+ (1) 112 GaGdJ) - 112} La+ g',
g' E5' i (u).
Similarly, on applying Lemma 6.3 and (6.25), we obtain
P' PP' = ( -ih) R (1) 112 [G~G,G~ + iJ (J)- 112· L 0, gEff'.i (u).
Hence, again applying Lemma 6.3 and (6.11), we obtain
BiRiBi=(--ih) [.8j0 'RjuBJD'+gi], giErfPi. (6.26)
Now, taking into account the monomorphism property of the cano-
nical operator and applying (6.21), (6.22), (6.26) and Lemma 6.3,
we obtain
B ,o, 13(l'B'·o'
• J 11 . J
= [a(o'
.• J
(~-_!_
dtj 2
tJ· a2"Aj)
{)q ap
+
+ {!Ej {I(Q!f,<O)_"J..j)+.fJ\O>c;c-y,r,(!)+<l?(2)].8(0) <1?(2)
cJp aq J C!v 0 1 1 , 0.i
cr\
Ev·j,
with cp = Bj0 >'Y, 'I' E.;tm (Mj+I).
The following equation
E· {!Ej {I (Q!f<O>-"Aj)- aEj {I (QJC<O>-"Aj)
] ap aq .- {ip {iq
is an easy corollary of Eq. (6.3) and the equation

E·
]
aEj
ap
+ iiEj
{lp
E-= aEj.
] {lp
33-01225
Finally, on applying Lemma 6.5 and (6.27), we obtain

B<.o> R<.r> B<.o> = B<.o> [-a--_!_ tr ..!!!:__ +
1 1 1 1 ati 2 aq ap
+ a:; a(&e;~- Aj) + Q18<1> + ?b}al JBjo>' w,;a> EfP i•
Q.E.D.
By Theorem 6.1 a formal asymptotic solution of Problem 6.1 can
be constructed with the elements
cpi(ex, t, h)E.Am(M'J+I), j=1, ... , l,
satisfying the equations
RicPi = 0, j = 1, ... , l (6.28)
l
~ cpi (ex, 0, h)= cp 0 (ex, h), (6.29)
j=1
where the element cp 0 (ex, h) E.Jlm (An, rn) verifies (6.4). Indeed, let
the elements be found. Set
l
1¥ (x, t, h)= ~ ~ MTJ+rcpi.

j=1 J
Then, by Theorem 6.1, H"'¥ = 0, and
W(x, 0, h)=J~O (67i'Mj+lcpj)lt=0=6'f{An c~1 cpj(ex, 0, h))=

=6fl;'Ancpo='¥o(X, h).
Theorem 6.2. There exist elements of the set .Jlm (Mj+I), j = 1, ... , l,
satisfying (6.28) and (6.29).
Proof. On applying Lemma 6.4 we obtain
+ +
Ri = Bj0 >RiBjo> Bj0 >Ri (1- Bj0 >) (1- Bj 0 >) RiBj0> +
+ (1- Bjo>) Ri (1- B] >). 0
By Lemmas 6.2-6.4
(- ih) Bjo> (R)ll + g]I>) Bjo>, g)!> Eff' i
Bj0 >R iB\0 >= (6.30)
Bj >Ri (1- .8]0 >) = (- ih) [Bj 0 >R]I> (1- Bj0 >) +
0
+Bj0 >gj2 > (1-.8]0 >)], g<J> E ffvj. (6.31)

To find the vectors cpi E.Jlm (M'J+I) satisfying (6.28) and (6.28),
it is sufficient to calculate the "projections" Bj~>cpi, k = 1, ... , l.
Consider the following equations:
+
B]0 >[Rju gjl>] (Bj0 >cpi) = - B?> [RJI> + gj2>] (1- .Bj0 >) cpi, (6.32)
+
(1 - .Bj0 >) R j (.8j 0 >cpj) (1 - .Bjo>) R j (1 - .8]0 >) cpj = 0. (6.33)
To solve Eqs. (6.32) and (6.33) we find the expression of B)0 )rp 1
by (1- .8}0 )) fiJi= ~ (Bh_0 lcpi) with the help of (6.32) and substiiute
koj=j
this expression in Eq. (6.33). Then we shall be able to find the
values
.B~)cph k = 1, ... , l.
Now turn to Eq. (6.32). Consider the function
Fi = -.8}0 ) [R}l) + g} 2 )] (1-Bj0 >) fPJ· (6.34)
On applying Theorem 6.1 and Lemma 6.4 we obtain
.B(o)
3
(R!.ll
·J
+ a<.O) B\o) =
"' 3 3
.B(o) (
3
....!!:.._ _..!. tr
dti 2
02 '}..,i
oqop
+
+ oEj o (Qf{}< 0l-"tvj) +d18<1) + g\3)) .8\0)
(6.35)
oP oq J 3 '
gj3) = g}l) + ~}2)' g?) EfP i.

Consider the system of equations
d 1 ()2'}..,. dB(O>
(-
at i
- - tr - -1-
2 oq op
B<.o>
3
+ B\o> - at- +
3
3-
+ B<,Ol _oE_
p
3
·
{)
J
{) (Qf{}<O)- "tv·)
{)q
J .B\0)
3
+ B\O)Jjg(!) B\0)
3 3
+
+ .BJD>gy;> Bjo>) ei = Fj,ej lt=o = eoj, (6.36)
where Fi is defined by (6.34) and the element 00iEAm(An, rn)
will be chosen later. We shall prove that if ei is a solution
of Eq. (6.36), then B}0 >0i is a solution of the system
+
.8)0) (Rjv g]V) .B)D>ei = Fi.
For this purpose it is sufficient to prove the following equality
a
B<,O)- (B<.o>e ·) = .B\Ol oe.
_ J _ + B\0)
( _aB<.o>
_3 - e.
)
3 at i 3 1 1 1
ot i at i n
which follows immediately from Lemma 6.5 and the Leibnitz for-
mula.
The solutions of Eq. (6.36) and the transfer equation are essen-
tially the same (cf. Sec. 6, Ch. IV). By Lemma 6.4 we have
B),O> = Ek + {)],_, {)k E ffo" k = 1, ... , l.
Therefore, on applying the lemmas of the beginning of this section
we can put (6.36) in the form
d 1 o2"tvj dEj oEj {) (Qf{}<O) -"tvj)
( -
dtj 2 r -{)q-{)p
--t -+E
+ EJ·dtj ·-
J 8p 8q
+
+ EiJ18<1)Ei+ ~i) ei = Fi, ~i E ffoi. (6.37)
33*
Let <Di be the fundamental matrix of the following system of equa-

tions:
d 1 o2"Ai dEi
(
dti -Ttr aqap Ei-1-Ei dti +
DEi iJ(Qfe<O>-"A·) )
1 +E-JJ8< 1
-1-E--
] op Dq ] >E.] <D·=O
J
<Dj lt=O = E,
where E is a unit matrix. Let ei = <Divi. We obtain the following
system of equations for vi:
{
~:; -1-(<l>jl~i<l>i)vi=<l>jlFj, (6.38)
vi lbo =Soh
where <l>j 1 ~i<Di E ff>i by Eq. (6.37). We also have
vi= 2J (I (<l>j 1 ~j<Dj)l [8 0 j -1- I (cDj 1 Fj)],
k>O
similarly to Sec. 6, Ch. IV. Turn to the solution of Eq. (6.33). Let
B~O) Cf>j = B}01 ej. Then
BjDlq>j = B}olql> B}olq>j (a, 0, h)+ .B)olq2l (1 _ Bjo>) Cf>i, (6.39)
where
qu.B)Glcpi (a, 0, h)= <Di [ 2J (I (<l>j 1 ~/Di))k] B}0 )q>i (a, 0, h),
k>O
q2> (1- .B}O)) Cf>j = <J)j [2J (I (<Djl~j<Dj))k] I (cDjlB}Ol X
k>O
(Rjl> -1- gj2') (1-.BjG') Cf>i)·
X (6.40)
Substituting the value of B}01 cpi into (6.33), we obtain
(1-Bjol) Ri (1-Bjol) Cf>i+(1-.B}o') Ri.Bjo>quB;o>cpi (a, 0, h) -1-
+ (1- .8}01 ) RiB} 0 )C} 21 (1- BJD') Cf>i = 0. (6.41)
Applying Lemmas 6.1-6.4, Theorem 6.1 and (6.40), we can put
(6.41) into the form
Lj (/..,k-/..,i)(B)illcpi)+ 2J B';,_01 RiB)0 >CjllBj0>cpi(a, 0, h)-1-
k~J k=Fj
+ 2J kf:.i
2J B)il)~kih 112 C] 21 (Bii: cpj) =
m=fr]
1 0, (6.42)
~kiEXi, k=1, ... , l, k=f=.j.

By Lemma 6.2 the following equality is true:
hlf2C?' = <J)j [ LJ (I (<Djl~j<Dj))k] I (C}3) (1- _B}O>) Cf>j),
k>O
CU. V. CANONICAL OPERATOR ON A LAGRANGEAN MANIFOLD 517
where
qz>=h 112 CI:r/.8~0 \RJ!> +g<j>)EuYi·
Let .JI; be a matrix with elements of the form
(.:m11.> h,m= (A-h-'Aj).
1 8 h,0) C>hj
<P. (I 1tzc<2l) (6.43)
~ i
and Y and F be vectors with coordinates of the form
1 B 10 >R i·B'o>c(nB<o>
Y i=.B.ico> Cf!i, Fi=-(l,h-'Aj) i i J i Cf!i ( a, 0 ' h) . (6.44)
Then (6.42) may be put in the form
(E + .It) Y =F.
By Lemma 6.2 and (6.42), (6.43) the operator E + .Jt is a quasi-
identity, therefore, by Lemma 4.2 there exists the operator (E + .Ji) - 1
and the equality
Y = (E + .ft)-1F
is true.
Choose the initial conditions 8 0h j = 1, 2, ... , l for Eq. (6.36)
to satisfy Eq. (6.29). We have
l l l
~ Cf!j lt=O =
i=1
LJ LJ (.B~>cpj) lt=O =
i=1 h=1
l
= LJ (Bteoi + hoti
j=1
LJ B~o>Cf!i lt=o). (6.45)
Defme the vector Bj,0 > Cf!i lt=o, k =1= j by (6.33). First we shall put
(6.33) in the form
~ B~>Ri(Bjo>eoi)+
Mi
h .B~)Ri
h=tj
~ Bii't> (Cf!ilt=o)=O.
m:J=j
(6.46)
By Lemma 6.4 and Theorem 6.1 we have

B),0 >RiB':// = B~o> [(A."- A.i) <\m + Ghml B~>, (6.47)
where Gmh E :£. Note that the index j of operator class is omitted,
since the introduced operator classes are obviously independent
of j when t = 0.
Let Y and F be vectors with coordinates of the form
Y h= B "w> ( (Pi I1=0'
) 1
F h= Aj-Ah B 10
h >R j (B'o>e
j
)
Oj'
k=1, ... ,l, k=l=j,

and G be an operator matrix with elements of the form
1
(G)h,m= ('Ah-'Aj) Ghm; k, m=1, ... , l, k=foj, m=foj.
By (6.47) and Lemmas 6.3, 6.4, Eq. (6.46) can be put in the form
(1 +G) Y =F.
By Lemma 6.3 the operator 1 + G is a quasi-identity. Hence the
following equations are true:
B k(Ol ( Cf>i It=O ) = s<o>
k L.J p<i)
"" B<O)[I
km· m Vom, (6.48)
m-j=j
Note that by Lemma 6.5

Bl~0 )RiB~O) E:£.
Hence the operators Pnm of (6.48) are of the class :£ as well.
Turn to (6.45) again:
l l
~
j=i
( mo>eoi + ~ B'(Jl m:f=i
k=Fj
~ P~},B~/Som) = ~ B}0 )q>0,
5=1
(6.49)
where
PkmEX, k, m=1, ... , l; k=Fj, m=Fj.
To define the elements 8 0 m E .Jlm (An, rn), consider the following
system of equations
[Iooz + ""
L.J
p<il B(O) [I
lm m vom =
s<ol
l Cf>o· (6.50)
m#-J
The elements B;;;.)eok obviously satisfy Eq. (6.49) if the elements
8o 1 , • . . , 8 0 z E .Jlm (An, rn) satisfy (6.50).
System (6.50) has a unique solution, since the operators P)~1 B;~J
are of the class:£. This statement is proved similarly to the solution
of Eq. (6.46). Theorem 6.2 is proved.
In the proof of Theorem 6.2 there is a method of constructing the
solutions of equations (6.28), (6.29). It is easy to sec that the problem
of solving equations (6.28), (6.29) reduces to that of constructing
inverse operators for some quasi-identities. This construction is
given in the proof of Lemma 4.2.
The following theorem is an obvious corollary of Theorems 6.1
and 6.2. Applying the same notation as in Theorems 6.1 and 6.2
we can formulate it in the following form.
Theorem 6.3. Let cp1 , • . • , cp 1 be such elements of .Jlm (Mr 1) that
the following equations are valid:
(i) Ricpi = 0
l
(ii) (Lj Cf>ih=o = Cflo, Cflo E .Jlm (An, rn), j = 1, ... , t.
i=1
l
Then an h-asymptotic series '¥ = 2J fliC Mn+i<Jli
5=1 i
is a formal asymptotic solution of the problem
{
-ih ~; +d78 (~, ~'h) 1¥=0
1Jf lt=o = 6)f: A niJlo.
Now we shall reformulate the theorems proved in this section
for the operator case.
Let QJ8 (x, p, h) be the same matrix as in (6.1).
Construct an h-asymptotic operator 1Jf (:, ~' t, h)= if satisfying
the equation
{
~ih
..
a"''P't.
u
+[cf/8 (~, ~, hh[l¥(~, ~, t, hh=o,
(6.51)
1¥/t=O = Ev (x),
A a
where p = -ih o:x, E is a unit (m X m)-matrix v (x) EC~(Rn).
Let Bm (M'J+1) be the set of equivalence classes of D<h-asymptotic
series in (m X m)-smooth matrices, depending on points of 111]+1
and h.
Let Bm (Rn X [0, T]) be the set of equivalence classes of h-asymp-
totic series in (rn X rn)-matrices defmed in Rn X [0, Tl. Within
the previous notation the following theorem is valid.
Theorem 6.4. The solution of Problem (6.51) has the form
lp (~, ~, t, h)=~ (~M'!+1CjJJ) (~, ~'

)=1 J
t, h), where
cpiE.Bm(Mj+ 1 ), j=1, ... , l.
The proof of Theorem 6.4 is similar to Theorem 6.3.
Sec. 7. Quasi-Inverse of Operators with Matrix Symbol

In this section the statement of the main theorem is introduced
for the case when the symbol of the operator to be quasi-inversed
is an (s X s)-matrix, s is any.*
Definition 7 .1. Let X be an .1:.' -module with a ~A--structure, A 0, •••
... , An, B EX. An element F =

A 2) 1 ( 1
{
g \B Ao+ f A
1 , • • • , An, E
n n+
B
1)
E .1/; is quasi-invertible if there exist such two sequences xk E :£, xk. E :;e
that the products Fxk and xi,} have the form
A ( 1
n+1 n+2)
Fxk=i+Rk A 0 , ••• ,An, B , xh.F=1+Rh. Ao, ... ,An, B,
(n+2A 2 1)
* Operaton F with matrix symbol F is a matrix consisting of operators Fij
with symbols FiJ· Product H = F·Gis the matix with elements fiii = _2jFik Gki·
k
where the functions Rk (x 0 , x, a) and Ri, (x 0 , x, a) satisfy the estimates

Rk=Oz(jxj-h), Rk=Oz(jxj-k), x=(x 1 , ••• , xn).
For the case g (B) = 0 this definition obviously coincides with that
of Sec. 9 of Introduction. The case g (B) = 1 is most useful for
various problems. The concept of a quasi-invertible operator is
changed for a .::.\-quasi-invertible one when the parameter manifold
Mm+I is not compact. Let L.\ E Rm+I be a compact domain.
. ~ ( 2) 1 ( 1 n n+ 1 )
Definition7.1'.AnelementF=g B A 0 +f A 1 , • • • , A,,B
is .::.\-quasi-invertible if there exist such sequences xk, Xk E :£ that the
product Fxk and xkf may be put in the form
~ ( 1 n+1 n+2)
Fxk=P(B)+Rk Ao, .. . , An, B ,
~
xi,.F = P'(B) + R/, ( n+2
A 0 , ••• ,
2 1)
An, B ,
where P (a), P' (a) E Co equal 1 when a E L.\ and equal 0 outside
a small neighborhood L.\. The junctions Rk (x 0 , x, a) and Rk (x 0 , x, ex)
satisfy the estimates
Rh=O:c (jxj-k), Ri.=Oz(jxj-h), x=(.t· 11 ... , Xn)·
Let the matrix elements fu (x, a) belong to rffoo (Rn X Mm+I)
and the matrix f (x, a) is asympLotically p-quasi-homogeneous of
degree r 0 in x, i. e.,
M
f (x, a)= 2j j (x, a)+ cr (x, a),
i=O
where the matrices j (x, a) are positively quasi-homogeneous in x

of degree ri with a unique set of p1 , • • . , Pn, r; ;;;:: r;+ 1 6, 6 > 0. +
The matrix elements of cr (x, a) satisfy the conditions
I a'k' I (p (xi)
oxk crii (x, a) ~ck
n
2jp.
'+ 1
)ro-1-lk/
.
,=1
Similarly to Sec. 9 of Introduction, the matrix j< 0 > (x, a) is called

a leading term of the matrix f (x, a) and the matrix / 1 (x, a) =
M
= ~ f (x, a) is a essential part of the matrix f (x, a). The
i=O
number r 0 is defined as an asymptotics to the quasi-homogeneous
degree of the matrix f (x, a). Suppose that the function g (a) is
smooth and uniformly bounded together with derivatives. The
matrix
F (x 0 , x, a) = g (a 0 ) x 0 + f (x, a)
is obviously asymptotically quasi-homogeneous in the variables x 0 , x
and of the same degree as j (x, a).
Without any loss of generality assume that the variable x 0 has
the weight Po = r 0 •
1 n+1 n+2
Let L 0 , ••• , L,_, a be a left-ordered representation of the ope-
1 n+l n+2
rators A 0 , ••• , A 11 , B. Suppose that the matrix
F (Lo (y, 1']), ... , Ln (y, 1']), a)
(variables Yt. i = 1, ... , n and Yn+h• k = 1, ... , m + 1, correspon-

ding to xi, i= 1, ... , n a n d 0- i - ° ,
ah-1
k= 1, ... , m+ 1, are in-
troduced as on p. 128) is asymptotically p-quasi-homogeneous in y of

degree r with the weights p0 , • • • , Pn+m+ 1 , such that P; = 1 for some
i E {n + 1, ... , m n + +
1}. Let :rt (y, 1'], a) be the leading term
of the matrix F (L 0 (y, 1']), .•. , Ln (y, 1']), a). Let J 1 be a subset
of all such numbers of the set (0, ... , n) that p; = 1 when i E J 1 .
Let J 2 be a subset of all such numbers j belonging to the set
(0, ... , m) that Pi+n+I = 1 if j E J 2 • Introduce the following
notation:
Yn+i+l=Pi, jElz; Yn+i+t=O, j"~_Jz/\)E(O, ... , m),
Yi= -Pm+i+i+wt. jEJ1, Yi=wi, if.Jd\iE(O, .. . , n),
ai=qt. i=O, ... , m; 1'Ji=qm+i+b j=O, ... , 1'].
On substituting the new notation into the matrix :rt(y, 1'], a) we

obtain the matrix J!S(p, q, w).
Let OJ8 1 be an essential part of the matrix F (L 0 (y, 1']), ...
. . . , Ln (y, 1']) a) for the new variables (p, q, w). We shall
suppose that the matrix OJ8 1 (p, q, w) is diagnalizable
by a smooth transformation and its eigenvalues p.1 (p, q, w), ...
• . . , p. 1 (p, q, w) (l ~ s) are of the class Coo and of constant multi-
plicity.
The following inequality is assumed to be true:
Im fti ~ 0, i = 1, ... , l.
Let A-1 (p, q, w), ... , A- 1 (p, q, w) be eigenvalues of the matrix

OJIJ(p, q, w) and let Q~i) be a manifold in the space of the variables
p, q, w defined by
n
p=O, (go, ... , gm)EJI11n+i, .2J q~+i+i <e,
J=O
w0 ER\ jA-;(0, g, w)l <e.

Note. If Mm+t = Rm+t, instead of Q~i) the set Q~i) (L\) defined by
n
Po= 0, (go, · · ·' gm) E L\e, ~ g~n+i+1 < e,
J=O
n
~ u)~/P;=f, w0 ER1, IA-;(0, q, w)l<e
t=i
is used, where L\ e is e-neighborhood of domain L\ used for the
construction of a L\-quasHnverse.
In the sequel, the eigenvalues !-! 1 (p, q, w), ... , f,tz (p, q, w) of
the matrix JJ8 1 (p, q, w), (p, g, w) E Q~') are supposed to satisfy
a condition
inf min lfli(P, q, w)-~ti(P, g, w)j:;>o>O, (7.1)
(p, q, w)EQ~<) if-i, h
where 8 > 0 is a consLant.

Suppose the matrix rJJ8 (p, q, w) to be diagonalizable and the eigen-
value mulLiplicities A-1 , • • • , A- 1 to be independent of p, q, w, which
involves that the set Q~i) does not intersect for sufficiently small s,
i. e.,
Q~i)nQ~)=0, i, j=1, ... , l, i=/=j.
Introduce the notation
H<;> (p, q, w) = ReA; (p, q, w),
(p, q, w) = Im A-; (p, q, w).
7f
Definition 7 .2. The ith family of bicharacteristics of the operator F
is a solution of the system
• . &H . .
q<t>=---ap(p<t>, q<t>, w), }
• . f)J{<i) . .
P<t> _ _ _ _ (p<t> q<t> w) (7.2)
- &q ' ' '
(p<;> (0)' q<;> (0)' w) E Q~i).
1 n+1 n+2)
Definition 7.3. An operator F ( A 0 , • • • , An, B satisfies the
absorption condition if such constants T > 0, e > 0, -r' > 0, 0 <
< -r' < T exist that for any i E {1, ... , l} with Qc!l =I= 0 we have:
(1) on the segment 0 ~ 't' ~ T there exists a solution of the class Coo
of system (7 .2)
q<;> (q<';, (0), ro), p (q (0), ro);
(2) when 0 ~ -r ~ T the function Ji (p, q, ro) is non-positive
and when -r = -r' it is strictly negative.
1 n+1 n+2)
Theorem 7.1. Let an operator F ( A 0 , • • • , An, B satisfy the
absorption condition, then it is quasi-invertible.
The proof is essentially the same as of the Main Theorem of Sec. 5.
The symbol of a quasi-inverse operator being a matrix, the construc-
tion of Sec. 6 to solve the initial value problem is applied.
We introduce a version of the Main Theorem useful for the study
of the Cauchy problem.
1 n+1 n+2) "
Definition 7.4. An operator F ( A 0 , • • • , An, B satisfies the
global absorption conditions, if there exist constants T > 0, e > 0,
-r' > 0, 0 < -r' < T such that for any i E {1, ... , l} we have:
(1) there exists a smooth solution of system (7.2) with initial condi-
tions satisfying the equations
p (p, q, ro) is non-positive

and when -r = -r' it is strictly negative.
Note that when Mm+1 = R 11H 1 Eqs. (7.3) in Definition 7.4 are
changed for
p<'> (0) = 0, (qbi), · · ·, q~l) 1.-=0 EL1e;
n n
'\,l
.L.J qm+i+1 )21 t=O
( (i) < e, 2.] ro~/P; = 1, roo ERi.
]=0 i=1 I
The global absorption conditions being satisfied, we can con-

struct the symbol of the quasi-inverse operator solving the
problem
2
.A1 r-1 a'¥ ( 1) ( 1 n+ 1 n+ 2)
- ~ ih; a, -r, - i 0:,, x' +[F L0, ••• , Ln, a ] X
a2 1 ) •
x [lf ( a, -r, - i ox' , x' ]=0 1
(7 .4)
where A = A (x), Eisa unit (m X m)-matrix, v (11) E Coo (Rn+1) 1 =

= v (11) when 111 I < e, and x' = (x 0 , x).
Following the arguments of Sec. 5 and applying the K-formula
(
1 n+i n+2)
we may put the operator F L0, ••• , Ln, a in the form
(, 1 ... , n+
FLo,
1 n+2)
Ln, a
1
Ar 0 [JI8 2
( 1 1'
X1
1'
A-P, -iA-PDa'
J
- 1
2
. ox'
a 2 Ae
, a,
1') + (
-
1' ) §:Jo
'A-t
~ wv 3
(1 ~A-r
X
1'
, -
1'
.A-P
l
o:
fJrx ,
2
. a
-~
ax '
By this formula the Problem (7 .4) is reduced to the following one:
1 1 1' -p 1' 1 2
( . -p a . a 2
X X A ' -~A fJrx ' - ~ ox' ' a,
x [1¥ \a,I
1:, - i
2
a!' , X1
1 )
] • 0,
lfh=o=Ev ( - i 8~,).
Let f.ti
( 1 I 1' -p
,x A .1' -p a
, -~A
1
. a2 2 Ae
fJrx , -~""'1£', a,
1' )
be the ith eigen-
value of the matrix 0/8 2 • Consider a function
det
'Vi (X 0 , X; P 0 , P; ro, x 11 X 2) = f.li(ro-
-A1-PP, A 1-pP0 ; X, X 0 , A- 8),

where
Xo = (qo, ... ' qm) ERm+t' X = (qm+1, · · ·, qn+m+2) E Rn+i,
Po= (Po, . · ., Pm)ERm+t, p = (Pm+1' · · · ' Pn+m+2) ERn+!,
ffii =A -pixil i=1, .. . , n, ro 0 =A-r0 x 0 ,
def def
xt = (Aet, ... ' A en), X2=(A1-p1, ... , A1-pn).
The function 'Vi is a smooth function of the parameters ro, x 1 , x 2

for A sufficiently large. The eigenvalue 'Ai of the matrix 618 (p, q, ro)
is a limit of the function 'Vi as A-+ oo. Note that Im 'Vi ~ 0.
Let (An+m+l, rn+m+ 1) be a germ defined by equations

An+m+ 1 ={Po=P=0, !Xj<s, XoEMm+ 1}, rn+m+ 1 =0.
We shall construct m n + +
2-dimensional manifolds Jl.;JV+m+ 2
with a complex germ w(i>, z(i>, a dissipative function D(i> and a po-
tential E(i> by a canonical transformation corresponding to the
Hamiltonian function Y; (X 0 , X, P 0 , P, ro, x 1 , x 2). We shall follow
the ideas of Sec. 4 to the effect.
Let QJC M·m-w+2 be a canonical operator on Mf+n+ 2 • Similarly to
t
Sec. 5 and Sec. 6 we may prove that a formal asymptotic solution
of the Problem (7 .4) is defined by the equation
2 2 1 ) l 2 2 1
(
'¥ a, T, - i a~', x' = (~QJCM'!-+m+2Cfli) (a, T,-ia~', x'),
i=1 J
where cpj E Bm (Mj+m+ 2 ).
( 1 n n+1)
Lemma 7 .1. Let an operator f .A 1 , . . . , An, B satisfy the
global absorption conditions. Then it has quasi-inverse and the symbol
of its quasi-inverse has the form
x(x, a)= i J{[A(r-t> ±(~

I) i=1
M'!-+m+2Cfli)
J
(a, T,
-i L, ~')]i(x)}d-r,
where 1 (x) is a junction of x equal to unity.
There exist analogues of Theorem 7.1 and Lemma 7.1 for !'!-quasi-
inverse operators; the absorption conditions should be restated
following the previous remarks.
Example 7 .1. Consider an equation (cf. Example in Sec. 9
of lnlroduction)
-l- . au
at
+ . auay
l-.--l . ( V (t ) ·,It -ay2-
aZ ) U= f
' (7.5)
.where v (t) is such a smooth bounded function that

v (t) = 0, t E [-T, T],
v(t)>O, ti:[-T,T], T>O,
f (y, t) E L 2 , supp j (y, t) c !'! c R1 X [-T, T].
Let
we have
[A 0 -A 1 -iv(B 0 ) Vi +A~] u =f. (7.6)
Now construct a quasi-inverse operator for the operator
1 1 2) def --
F ( A 0 , A 1 , B =A 0 -A 1 -iv(Bo)V1+A~.
Take the ordered representation of the operators
. a
L o=Xo-L-, . a
L 1=x1-L-a-,
8cxo a;1
a.
Hence, the Hamiltonian of the operator A 0 -A 1 -iv(Bo)Vt+A;
has the form
( 1 2 2) = (x
F L 0, L 1 , a .a) (
0 -L8a;;' - x1-L
.a)
acx
1 -
- iv (a 0) V 1 + (x 1- i a! 1 )
2
;
thus,
F(L 0 (y, YJ), Li(y, YJ), a)=
=(Yo+ Y2)- (Yi +Y3)- iv (ao) V 1 + (Yi + y3)2.
Consider the case
P1 = P2 = P3 = Po = 1.
We have
n: (y, f), a) = (Yo Y2) - (yl+ Ya) - iv (ao) + I Y1 + Ya I,
or, with respect to the variables (p, q, ro),
JJ8(p, q, ro)= (roo +Po- P2) - (rol + P1- Pa) -
+PI - Pa J.
- iv (qo) I ffi1 (7.7)
To construct the bicharacteristics of the operator F consider the
Hamiltonian system
. .
qo = q2 = 1
ql = qg = -1
Pi = 0, i = 0, 1, 2, 3
having the initial values p lt=O = p<O>, q lt=O = q<O> satisfying
Eqs. (7 .3). The following functions are obviously solutions of this
system
Po = P1 = P2 = Pa = 0, }
qo = t +
qbo>, qz = t q~o>, + (7.8)
ql = -t +
q~o>, qg = -t + q;o>.
By Eq. (7.5) the imaginary part of if (p, q, w) of the Hamiltonian

function is defined by the formula
if (p, q, w) = -v (qo) I WI +PI + Ps I= -v (qo)· (7.9)
By Eqs. (7 .8), (7. 9) the operator F satisfies the global absorption
conditions. Hence the construction of the symbol of the quasi-
inverse operator of F is reduced to the problem
- i ~~ (a, ~. - i t ,~) +
+[F(L, Lh ~)][W (a,~. (7 .1 0)
w(a, 0, -ilx, ~)=r(-i!)P(a),

where p (11) E C';, p (11) = 0, when 111 I> e, p ('YJ) = 1, when
111 I < ~ , P (a) E C';, P (a) = 1, when a E .1, P (a) = 0, when
aE .1e.
We may put the function 'I' (a,~. - i :x' ~) in the form

2
nr (I . fJ2 1) _ 1 (1x,
iAS -; aax , a, -c ) ( 1 . 82 )
T a,~, -~ax' X -e 8 X, -~BX' a,~,
where A = I xi 1.
To satisfy the initial conditions of Problem (7.10) we may put
the funcLions S and 8 in the form
S (:~::, 11, a, 0) = 0,
8 (x, 11, a, 0) = p (1J)P (a).
On substituting the function 1J.Y into Eq. (7.10) and applying the
commutation formula, we obtain
1 1
,,a '• )
(x,-iax,a,-c , ,
iAS 31 aS . a 1 aS
e [A--~-+xo
. a, +A--- a, aao
1 ' as . a 1 ' as . a 1' as
-A--~--x 1 -A--~- 1 1 A --
a'YJo aao aa 1 fJa 1 81']1
1' as 1' as a 2
-iv(a0) ( 1+ ( x 1 +A-a--A-a- -i-.-
a1 'YJ1 aa1
)2)1;2
][8]=0.
(7.11)
Ordering the operators by the K-formula, we obtain

3 1' as . a 1' as 1' as . a
[A--~--+xo-1-A--A--~--
o't o't Octo a'Y]o oet.o
- ( x1 +A -as- A -
1 ' as
1' a )
- i - _ - -iv(a0) X
oa1 a'YJ1 aa1
x (1+(xt+A as -A as -i-a-)z)1/2]x
aa 1 OYJ 1 aa 1
X
2
[8 x,
(1 .~
-~ax, a,
)=
't' ]
as as as as as
=A { [ -+w 0 +--+---w 1 - - + - -
o't oet.o ' Vllo oa1 aT] 1
- iv (ao) Iwt + ::1 + !~1 I:J + ( ~ (- iA- 1 )~< R~<)' e} x-

"""o
1 . ~ ) (7 .12)
X ( x, -~ox, a, -r .
Here Rk are differential operators, which we apply to the symbol
of the operator 8 (x, - i :x,
a, -r). The operators Rn are of an order
no greater than k, with the coefficients being smooth functions of
ffi1o a, -r and the derivatives of the functions S; w0 = x 0 A -1, w1 =
= xi A -1.
Let S 1 =ReS, S 2 = Im S
~ ~ ~ ~ ~
-+wo+----ffi~---1----
o't oet.o a'Y]o oa1 aT] 1
. I
-~v(ao) w1+ aa 1 -
as as I
aTJ 1 =0s2 (A
-a;z). (7 .13)
Eq. (7.13) is a Hamilton-Jacobi equation having the dissipative
function JJI](p, q, w) defined by Eq. (7.7).
On applying to Eq. (7 .13) the methods of Ch. IV and equating
to zero the right-hand terms in the parentheses in Eq. (7 .12) we
obtain the dissipative transfer equation for the function e.
This equation is solved in Theorem 6.2 of Ch. IV.
Consider
't
D=- JImJJI](p(q<O>, w, -r'), ,q(q<O>, w, -r'), w)d•'lq=q(q<o>, 0 ,,>·

0
We proved in Ch. IV that the soluLion of the equation (7.13)
satisfies the dissipativity inequality:
Im S (x, a, -r) ~ yD,
where y = const.
By the last inequality and the absorption condition we have

D > () > 0, () = const, when 't' = 2T. Hence the function
\f(a, 't', '1'}, x)=eiAS<x.Tl,et,'t)e(x, '1'}, a, -r)
satisfies the estimate
I \f (a, T, 'Y], x) I~ const e-lilxtl,
Thus, by the reduction rule (cf. Introduction, p. 109), the operator
having the symbol
2T 2 i
x(x, a)=i~ {['I'(a, 't', -i:x' x)]1(x)}d-r (7.14)
0
is a quasi-inverse for the operator F.

Note that by Eq. (7 .13) the function S has the form
S = -ro 0't' + S 1 (x1, a, -r), (7.15)
where S 1 satisfies an equation
iJS1
~
+ aS1- ro1- iJS1- iv (a) j ro1 + aS1 I= Os (A -3/2).
~ ~1 ~1 2
We have indicated that the coefficients of the operators Rk in

Eq. (7.12) do not depend on Xo· Hence the function e does not depend
on x 0 • Besides, the coefficients of Rk do not depend on 'tJ. Hence
the function e (xl, 'Y], a, 't') has the form (cf. (7.8))
e (xl, 'tj, a, 't') = el (xl, a, 't') p ('tJo + 't', '111 + 't'), (7.16)
where 81 (x1 , a, -r) satisfies the same transport equation as the
function e (xl, 'Y], a, 't') and the initial condition
el (xl, a, 0) = p (a).
The following equation is a corollary of Eqs. (7.15), (7.16):
['l' (a, -r, - i lx , ~ h1 (x) = e-ixo't [eilst(~t. a, 't) 9 u a, ,; ) X

1 11
Xp ( 0, - i ox~ 1 ) der •
+-r ]1(x)=e- 1xo't1jl'(x 1 , a, -r).
Thus Eq. (7.14) may be put in the form

2T
x (x, a)= i Je-ixo"tljl' (x 11 a, -r) d-r. (7 .17)
0
Let the function f (y, t), the right-hand side of Eq. (7.5), satisfy
the condition f (y, t) = 0 when t < 0.
34-01225
Then prove that the following equation is true
hx (-i!, -i :Y, l, ~hf(y, t)}L 0 =0, (7.18)
where x ( - il , - i:Y , l, ~) is an operator with a symbol

defined by Eq. (7.14). On applying Eq. (7.17) we obtain
[x ( - i l , :Y, l, ~hf(y,
-i t)=
=ilhe-t=t~'(-i!, ;, ~hf(y, t)}dT=
rh~' (-i!' ~' ~hf(y,

0
=i 0
t-T)}dT.
At the point t=O we have
{[x(-il, -i}y, ~. ~ht<Y. t)L=o=
=i rh~' ( :y, ~' ~hf(y,

0
-i -T) }dT=0.
Thus the function u (t, y), O~t~T
u (t, y) = [x ( - i !1 , - i
1
:Y , t,
2 2)'
y ] f (y, t)
{
- i
at + i !.!!:...
!.!!:... ay = f (y ' t)
ult=o=O,
where f (y, t) = 0, when t < 0.
Note that to prove Eq. (7.18) we have used on]y the additive
(A1 A1 B2)
dependence of the operator F 0, 1, on A 0
1 1 2) ( 1 2)
F (A A B =Ao+F' A11 B .
0, 1,
Hence the Main Theorem permits to construct a solution of the

Cauchy problem in a more general situation.
Example 7.2. Consider an equation

1
- i f)au
t
+ G {\ ~, - i .!_) u- iv (t) ( 1 -
uy
~
LJ
fJfJ2. ') 112 u =
Yi
f(y, t).
(7.19)
Here G (y, £) is an m X m-matrix positively homogeneous of the first
order in£, f (y, t) E L 2 , supp f (y, t) c 11 c Rnx [0, T], f (y, t) = 0,
when t < 0,
v (t) is the same as in the previous example,
11 is a compact domain.
Let the maLrix elements of G(y,£) belong to thespace C00 (Rnx
X (Rn"- 0)) and be uniformly bounded with the derivatives when I£ I=1.
Suppose that the matrix G (y, £) is diagonalizable for all values
of its arguments by a smooth transformation and its eigenvalues
"h• .. . , y 1 (l~m) are of the space C""(Rn X (Rn"-0)) and satisfy
the conditions Im Yi ~ 0, i = 1, ... , l, inf I Yi (y, £)- Yi (y, £) I:;:;:>-
~ const when I £ I = 1.
Let the Hamilton system
dqci> _ aRe "? i ( (i) (i))

d-r: - fJp q ' p '
dpci> _ _ aRe "li ( (i) (i))
d-r: - f)q q ' p , 1 pci> (O) 1= 1
have the solution for 0 ~ 't < oo. Then we shall construct a for-
mal asymptotic solution of the problem
{ -LTt+G
• au ( 2
\y, -i~)u=f(y, t)
ll (y, 0) = 0
with respect to the powers of ( 1 -l when 0 ~ t ~ T. We +I :y n

shall use a 11-quasi-inverse of the operator
-i :t +G (
2
y, - i
1 )
:Y
(
-iv(t) 1- 2}
n
i=1
::~
'
) 1/2
.
We mentioned in Introduction that solution of partial differential

equations reduces by quasi-inversion to that of integral equations
(in the present case of the Volterra type) with smooth kernels.
34*
Take the operators
A 0 = - i ! , A 1 =-i 0: 1 , ••• , An=-i 0:n, B=(t, y),

1 1 1 2) ( 2 1 1 )
F ( Ao, At, ... , An, B =Ao+G B, At. ... , An -
- iv (Bo) · ( 1 + i~1 Ar) 112 •

The Hamiltonian of the operator F has the form:
(1
F L0, ••• ,
1 2) (Xo-L.iJ)
Ln, a= 0a + 0
2 1 1 ) ( 1 1 ) 1 /2
+G ( a, L 17 . ••• , Ln -iv(a0 ) 1+Lr+ ... +L; ,
. a
1.. ·=Xi-L- i=1, ... , n .
z aai '
Hence,
F(Lo(Y, YJ), ••• , Ln(Y, YJ), a)= (Yo-1-Yn+t)+
+ G (a, Y1 + Yn+2• · · ·, Yn + Y2n+t)-
- iv (ao) ( 1 + i~1 (Yi + Yn+i+t)2 r/2.
Take p1 = ... = P2n+I = 1 and change the variables y, YJ, a
for the new variables p, q, ffi. We have
The eigenvalues A-1 , • • . , Az ofthe matrix J/8 (p, q, ffi) are obviously
defined by the equations
Ai (p, q, ffi) = ffio+ Po- Pn+i +
1 1 1 2)
We shall now verify that the operator F ( A 0 , A 1 , • • . , An, B
satisfies the global absorption conditions. Consider the ith family
of the bicharacteristics of the operator F

q~i)= 1
• (il _ aRe A.i j=1, ... , 2n+2
qi - ap1
(7 .20)
j=1, ... , n
n
p(i) (0) = 0, 2J w1 = 1,
i=1
n
.2J (q~~i+1 )21't=O <
J=O
8.
The following equations are obviously true:

p~>=O as n<k~2n+1, (7.21)
q~i) = "t + q~i) (0).
The remaining part of Eqs. (7 .20) can be solved with the help
of the following Hamilton system
;p> _
j -
aRe '\'i (q<il
ap1 1 ' • • ••
. q<i) ffi
n ' 1
+ p<i) 1 • • • ••
ffin + p<nil),
• 'il =
Pj
_ aRe 'l'i ( (i)
aq1 q1 , • • ••
(i)
qn •
+ P1(i) ,
ffi1 · · ·, Wn +Pi),
n (7 .22)
]. -- 1 • • • ••
n• (p<il
1 • • • •• n I't=O-
p) - 0•
(q1t(") ' •• •' q,iCl)l 't=O ELle•
A
The functions q3il, when n~j~2n+ 1, are determined by the

formulas
't
q i(i)-
- i
q<il (0) + Jf aRe A.i (p<il
ap. 1 ' ••• '
p<il q<il
n ' 1 ' ••• '
q(i)
n '
w) d..-'
••
0 J
By our assumptions Eqs. (7 .22) and, therefore, Eqs. (7 .20), have

a solution. Hence the operator F A 0 , ••• , An, B satisfies the (1 1 2)
global absorption conditions by Eq. (7.21) and the choice of the
function v (q0). Hence by Lemma 7.1 the symbol of the ~-quasi-inverse
operator x A 0 , (1 ••• ,
1 2)
An, B has the form
.
2T l 2
x(x0, x, cx.)=i) L}[(<?%Mj{oi)s!J>J) (a., -r, -ia~'' })]1d-r.

0 5=1
534 OPERATIONAL METROUS
+
Here M]n+ 3 ( ro) is a family of 2n 3-dimensional Lagrangean
manifolds depending on the parameters
ffi1
Xi
=A ' · · ·' (t)n
Xn
=A , A= I X I•
The family Myn+a (ro) is associated with a family of Lagrangean
manifolds with a complex germ:
{ M~n+2
,, 't"
(ro) • r~n+2
), 't"
(ro)} = D''Aj {M2n+2
0 ,
r2n+2}
0 ,
O:;;:::'t:;;:::2T,
~ ~
where r~n+ 2 = 0, M~n+ 2 ={Po= . · · = Pn = 0, (go, · · ·, qn) E~e}.

Finally, 6/C Mzn+3!pj is the value on the matrix <pi of the canonical
operator defined on the family M]n+ 3 • The elements of <pi belong
to the space of equivalence classes of D-asymptotic series on M]n+a (ro).
The formula
2T
x (x 0 , x, a)= i) e-ixoT.x' (x, a, 't) d't
0
can be proved as in Example 7.1 by using the expression of the

eigenvalue
Aj(p, q, ro)=ffio+Po+Pn+t+
- A.j (Pi• · · ·, Pzn+t• qt • · · ·, qn, ffip • • ·, ron)·
Then, again applying the method of Example 7 .1, we obtain
hx(- i t ,- ! ,t ~h
i
under the assumption f (y, t) = 0 when t < 0.
f (y, t)} Lo = 0
Hence the function
u(y, t)=[x ( - i l, -i!, ;, ~hf(y, t)

{
-i ~~ +G (~, -i!) u=f(y, t)
u (y, 0) = 0
when 0 ~ t ~ T.
Example 7 .3. The Main Theorem affords the construction of
a kind of asymptotic solution of equations with partial derivatives
with growing coefficients. Consider an example of a formal asymp-
totic (with respect to powers of the operator ( ~2 + 1:

2) -
112
) solu-
tion of the problem

fj2u f)2u
{ - at2 + ay~ -a (y) y2u=f (t, y),
(7 .23)
U lt=O = Ut lt=O = 0.
Suppose that
f (t, y) E L 2 , supp f (t, y) c A c [0, T] X R,
where A is a compact domain, a (y) is a smooth real function satis-
fying the conditions 0 < c1 ~a (y) ~ c 2 for some constants c1
and c2 •
Consider the operators
1 . a
A0 = - t -
at '
2 2
Bo=t, Bi=Y·
The left-order representatives of these operators have the form
Lo= (xo-i u~
,:0 ), L =x -i-
1
8--i-
ac~i
1
0-
OXz '
Eq. (7 .23) may be put in the form

[A~ -A~ -a (B 1) A;] U= f.
We shall reduce Eq. (7 .24) to a system of equations. Consider
the equations
[A 0 ] u (B)= v,
1 ( 2 ) 2 ) 1/2
(7.25)
[ ( A~+a B 1 A; ]u=w.
There is an equation
1 ( 2 )
[A 0]w-[ ( A; +a B 1 A;2 ) 1/2 ]v=O. (7 .26)
The product of the operators
1 ( 2 ) 2 ) 1/2 ( 1 ( 2 ) 2 ) 1/2
[ ( A~+a B1 A: ][ A;+a B1 A; ]
can be transformed by the commutation formula into the form
1 2 2 1/2 1 2 ) 2 1/2
[(Ai+a(B 1 )A:) ][(A~+a(B 1 A:) ]=
1 2 2
=Ai +a (B 1) A; +R (At. A2 , B), (7 .27)
where the symbol R (x1 , x 2 , a) is an asymptotic sum

R (x 1 , x 2 , a)= ~ R~t (x 11 x 2 , a),
k;;;,O
where the functions R~t (x11 x 2 , a) are homogeneous of a positive

degree (-k +
1) with respect to x1 , x 2 •
The following equation is a corollary of Eq. (7 .27)
1 ( 2 ) 2 ) 1/2
[ ( A:+a B 1 A: ]w=[A:+a(B 1)A;]u+
1 2 2)
+[R (A 11 A 2 ,B ]u.
1
Substitute the function u = [ ( A i a B 1 A; +
] -1 w into
( 2 ) 2 ) 1/2
the last equation. On applying the commutation formula

1 2 2 1/2
[(A~+a(B 1 )A:) ]w=[A:+a(B 1)A;]u+
1 2 2)
+[R 1 (A 1 ,A 2 ,B ]w, (7.28)
where the symbol R 1 (x1 , x 2 , a) is an asymptotic sum
R 1 (xl! x 2 , a)= ~ Rk. (xt, Xz, a).
k~O
The functions Rk. (x1 , x 2 , a) are homogeneous of degree -k with

respect to x.
Thus, Eq. (7.24) is reduced to the system
1 2 2 ) 1/2
[Aoiw-[(A: +a (B 1 ) A; ]v=O, (7 .29)
1 2 2 1/2
[A 0] v-[( A; +a ( B 1 ) A:) ] w+
( 1 2 2)
+[R1 A 1 ,A 2 ,B ]w=f.
Similarly to the previous examples the construction of a formal
asymptotic solution of Eq. (7.29) is reduced in the case to the con-
struction of a £\-quasi-inverse operator of the operator
where the function v (cx 0 ) is the same as above. The Hamiltonian

of the operator F has the form
( 1 1 2) =
F L 0 , L 17 ex
-iv(cxo)( (xi-i a!i a!2 +~:r 12 •

Taking the variables (p, q, ro), we obtain
-i r
Q/8 (p, q, ro) =
,roo+ Po- P2 -((rot+ Pt- P3) 2 + )
( +a (qt) ( ro2- P4) 2) 112
= -((rot+ Pt- P3)2+ -
+a (q1) (ro 2 - P4)2) 112 roo+ Po- P2
- iv (qo) ((rot+ Pt- P3) 2+ ( ro2- P4) 2 ) 112 .
The eigenvalues A17 A2 of the matrix 018 (p, q, ro) have the form
At, 2 (p, q, ro) =roo+ Po- P2 +((rot+ Pt- P3) 2+
+a (qt) ( ro2- p~,)2) 112 - iv (q 0 ) ((rot+ P1- P3) 2 + (roz- P4) 2) 112 •
The bicharacteristics of the operator F obviously satisfy the
conditions
Pbi> = p~i> = p~i> = p~i> = 0' i = 1' 2.
Hence the equation is true
A; (p' q' ro) =roo ± ((rot+ p~i))z +a (qli))ro;) 1/2-
- iv (q~i>) (( ro 1+ p 1)2 + ro:) 112 , i = 1, 2.
The following equation is easy to verify
q~i) = .. + q~i) (0).
Hence the operator F satisfies the global absorption conditions
and therefore is t1-quasi-invertible. The initial value conditions
in this case are verified similarly to the previous one.
We shall consider an analogue of Theorem 9.1 of Introduction
in the case of operators with matrix symbols.
Take a one-parameter families of m-dimensional matrix symbols

f (x1 ,
• • • , Xn, a,£), g(a, £, x0) where£ is a parameter 0< £<oo.
Suppose that
lim f(xi> ... ,'xn, a, £)=60 (x 11 ••• , Xn, a)Er!f 00
•
6 __,. 00
Consider a family of operators
~ ( 2 1 ) ( 1 n n+ 1 )
F=g B, £,Ao +I A1, ... ,An, B,£ •
1
where g is an m X m-matrix with elements of rf/ 00 • Let L 0 ,
n+1 n+2
... , Ln , a be a left ordered representation of the operators
A 0 , • • • , An, B. Suppose that the function
F (Lo (y, t]), ... , Ln (y, 'Y]), a,£),
Yi<='>X;, i = 0, ... , n;
Yn+k«='>-i~, u~k-1
k=1, 2, .. . ,m+1
is asymptotically p-quasi-homogeneous in y, £ of degree r with
the weights
Po• • • ., Pn• Pn+l• • • ., Pn+m+l• Pn+m+2
such that Pi= 1 at least for one i, i E {n 1, ... , m + + n + 1}.
Suppose that the spectrum of the set of the operators
1 -p n
£-PoAo, ... ' £ nAn
is contained in a ball, I x I < d, d is a constant.
Take the new variables. Let J 1 be such a subset of (0, ... , n)
that Pi = 1 only for i E J 1 • Let J 2 be such a subset of (0, ... , m)
that Pn+J+l = 1 only if j E J 2 •
Take
Yn+i+i=Pi, jEJ2, Yn+i=O, F'E.J,~. 1\ jE(O, · · ., m),
+
Yi = - Pm+i+i ffi;, i EJi> Yi = ffi;, i EJ1, i E(0, ... , n),
ai=qi, i=O, ... , m, 'Y];=qm+i+il j=O, ... , n, £=v.
Let :rt (y, 'Y], a, £) be the leading part of the matrix
F (L 0 , (y, 'Y]), ••• , Ln (y, 'YJ), a, £) and let JJ8(p, q, ffi, v) be the
matrix :rt (y, 'Y], a, £) transformed by the introduced change of va-
riables.
Let <f/8 1 be the essentia] part of the matrix F (L 0 (y, 'Y]), •••
• . . , Ln (y, 'YJ), a, £) with respect to the variables p, q, ffi, v.
Just as at the beginning of this section, introduce the absorption
conditions. The only difference constitutes a subsidiary inequality
I v I < d due to the presence of an additional parameter v. The
conditions defining the manifolds Q~i>, Q~i) (~) and Eqs. (7.3)
remain valid.
~ ( 2 1 ) ( 1 n n+ 1 )
Theorem7.2.LetanoperatorF=g B,£,A 0 +f A 1 , ••• ,An,B,£
satisfy the absorption conditions. Then there exists such a sequence
of symbols xN (x 0 , x, r:x, £) E C:Z, depending on a parameter £ that
( 2 1 ) ( 1 n n+ 1 ) ( 1 2 n+ 1 n+ 2 )
[g B,£,Ao +f Ab···,An,B,£ ][xN Ao,A~>···,An,B,£ ]=
1 n )
=Po ( £-PoAo, ... , £-PnAn + RN ( Ao,
1 n+1
... , An, B ,
n+2 )
£,
where RN(x 0 ,x1 , • • • ,xn,r:x,£)=0z(lxi-N) for any N>O
uniformly with respect to £, r:x and x 0 ; P 0 (x1 , • • • , Xn) is a function
of the class Cowhich is equal to 1 in the domain 1 x 1 < d.
A version of Lemma 7.1 and the corresponding statements about
A-quasi-inverse sequence are also true for operators with symbols
depending on a parameter.
As an application of the theorem, consider an example of a for-
mal asymptotic solution of the Cauchy problem for finite difference
schemes. Take a net Qh in the space R 1 X Rn
(t, Y)EQh-¢::> (t=kho, Y1=j1h, · .. , Yn=jnh),
where k, j 1 , • • • , jn are integers.
Consider the following finite difference Cauchy problem
u~+1 . -U~ . n
Ji' ... ,]n ]1, ••• , J n - ~ c<v>
ho -LJ ji' ... ,inX
V=1
u~+1 . . -u~+1 . .
11, ... ' 3 v+ 1' ... ' J n 3t, ... ' J v-1' ... , 1 n 1<+1
/
X· 2h +i1, ... ,jn'
. -0
ljQJJ, ... ,Jn- (7 .30)
where C}~~ ... ,in are the values on the net of the (m X m)
matrices Cv (y) with smooth bounded elements and UTJ-3 t•···• 3·n•
fii' ... ,in are net functions defined on the net. Suppose
that the function jfi' ... ,in equals zero at all knots of the net outside
a compact domain L\ c [0, T] X Rn.
On applying the Kotelnikov theorem, construct for a net
function cpf1, .•• ,jn satisfying the condition
ho·hn . ~ . I <rt ... ,jn J2< oo, (7 .31)
k,1f>••••Jn
an integral function <!lh (t, y) satisfying the conditions

(1) <!lh(k1:, j1h, • .. , jnh)=<p~i' ... ,jn;
(2) ) 1<llh (t, y) 12 ay at< oo
Rn+l
Take
!ph(t, Y)= ( 2~ r+i J
I 6ih 1::;:;n, i=1, ..• , n
ei < y, sl +i <t, sol X
I soh l::;:;n
X ho·hn ~
h, ji' .. . ,jn
!pt ... ,in X
n
-i6oMo-i ~ ikh\;k
X e d£k· k=i d~o (7.32)
It is easy to verify that the given function !ph (t, y) satisfies all
the necessary conditions.
Let fh (t, y) be a function of continuous argument defined by
Eq. (7.32) from the net function J}1, ... ,in·
Suppose we have constructed a continuous function which is
a solution of the problem
n
u(t+ho, y, h, ho)-u(t, y, h, ho)
ho
=""LJ Cv(Y) X
'11=1
Xu (t+ho, Yt. ... , Yv+h, ... , Yn, h, ho)- u (t+ho, Yt. . .. , Yn• h, ho)-+-
h I
+fh(t+h0 , y) (7.33)
u(O, y, h, h 0 )=0.
It is clear that the restriction of the function u (t, y, h, h 0 ) on
the net Qh is a solution of Problem (7 .30). Thus, the construction
of a formal asymptotic solution of Problem (7 .30) is reduced to the
construction of a formal asymptotic solution of Problem (7.33).
We infer from the estimate
II h a:i !h (t, y) //L2 ::=;;n II fh (t, y) IIL2

that the joint spectrum of the operators - ih-0 -
oyt '
... , - l'h -{j-
0Yn
is contained in a ball Ix l::=;;n Vn.
Take
1 . {j 1 • {j 1 {j 1
Ao= -L-
ot ' At= -L-
oy1 ' •. ' ' An- =i OYn' t=-h
"' '
2 2 2
Bo=t, Bt=Yt. ... , Bn=Yn·
It is easy to verify that the left ordered representation of the
indicated operators is the set of operators
1 {j n+1 . 0 n+2
L 0 =x0 - i -0- , .•• , Ln=Xn-t-{j-; ex.
ao an
CH. V. CA~ONICAL OPERATOR ON A LAGRANGEAN MANIFOLD 541
Applying this notation we may put Eq. (7.33) in the from
(7 .34)
Similarly to the previous examples, a formal asymptotic solution

of Eq. (7.34) is constructed by a ~-quasi-inverse operator. Indeed,
take an operator
1 1 1 2 ) def .
F ( Ao, A1 , ... , An, B, £ =£[1-e-•6-lAoJ+
n ( 2) 1 n 1/2
+£~Cv B sin£- 1 Av-iv(B0 )·(1+~A~) ,
V=i V=1
where the function v (a 0 ) is the same as in the previous examples.
Taking the variables p, q, co, v, we obtain
QJ8 (p, q, co, v) =+ (1-e-iv(wo+Po-Pn+i)j +
++ 1=1
n
~Ci(q 17 ... ,qn)sinv(coJ+PJ-Pn+i+t)-
- iv (ao) ( ~ (coJ + PJ- Pnti+t?r 1 ~ •

}=1
Let A.17 ••• , Az (l~m) be the eigenvalues of the matrix JJ8 (p, q,
w, v). The following equations are obviously true
A; (p, q, co, v) = ~ [1-e-iv (wo+Po-Pn+t>] +
+ "ri (qt. ... 'qn, COt+ Pt- Pn+t• ... ' COn+ Pn- Pn+t. v)-
n 1~
- iv (qo) ( ~ (coi +Pi- Pn+i 1-1) 2 ) , (7 .35)
}=1
where "ri are eigenvalues of the matrix
n
~ ~CJ(qb ... , qn)sinv(coi+Pi-Pn+J+1)·
}=1
Take
H = Re '\';, if = Im '\';·
Suppose the following inequalities to be verified
i = 1, ... , l. (7.36)
We shall define the sufficient conditions so that the absorption
conditions were satisfied in the present example.
By definition, the ith family of bicharacteristics of the operator

is a solution of the Hamilton system
I·
( • (i)
qo =COS'V (coo+ Po- PnH),
aRe J...i (
(i) _
qJ - iJpi p
ci> ci>
' q ' co, 'V ,
)
j = 1 , ... ' 2n + 1'
J "(i) "('l)
1 Po =Pn+i=O, (7 .37)
• (i) aHci> (i) (i) (i) (i)
Pi= --aq;-(q1, · · ., qn, co1+Pi -Pn+i•
.l ... ' COn+ Pn - P2n+1' v),

(i) (i)
1~j~n.
By Eq. (7.37) the equation
q'Qi> = T cos 'V co 0 + q'Qi> (0), (7.38)

p)i>=o, n<j~2n+1
is evidently true. Hence for any T there exists a solution of Eq. (7 .37}
if for any T there exists a solution of the following Hamilton system
iJHCi>
"(i)
qi = -,--
upj
(i)
(qi , · · ·, qn ,
(i)
C01 + P1(i) , · · ·, COn+ Pn , v)
(i)
• (il aHci> (il 

Pi = ---aq;- (qi ' ... , qn , COj
(i)
+Pi , ... ' COn+ Pn , v) (7 .39)
(i) (i) (i) (i)
Pi (0) = •· · = Pn (0), (qi (0), · · ·, qn (0)) E ,1e,
n
2j cof =1, lvl<nVn.
i=i
Note that ifci> ~ 1 -cosv (roo+Po-Pn+!) . So if I cos 'V co 0 I~
v
~ /3, where l3 is a sufficiently small positive number, then 'jjci> ~
~ 1 -v 13 < 0. Thus, by Eqs. (7.35), (7.36), (7.38) the operator F
satisfies the global absorption conditions if Eq. (7 .37) has a smooth
solution for any T and i = 1, ... , l. Thus we have proved the
following proposition.
i i 2 )
(
Proposition. An operator F A 0 , • • • , An, B, £ is a ~-quasi
invertible if inequalities (7 .35) are verified and the Hamilton sys-
tem (7.39) has a smooth solution.
Suppose the conditions of the proposition are satisfied. Then
i 1 2 )
(
the operator F A 1 , . . • , An, B, £ is ,1-quasi-invertible. Let
x (x 0 , x1 , a) be a symbol of its ,1-quasi-inverse. Similarly to the
previous examples we can show that the following equation is true

2T
x (x 0 , x, a,£)= Jexp {1:£ (e-i(;-lxo-1)} x' (x, a, 't, £) d't.
0
Prove the equality

( 1 1 1 2 )
[x A 0, A 17 ••• , An, B, £ ]h[t=o=O. (7.40)
We have
1 1 1 2 )
[x ( Ao, A17 ... , An, B, £ ]h=
=
2T
[I 1 1 1 2 )
exp {1:£ (e-i(;-lAo-1)} x' ( A 17 ... , An, B, T, £ dT] f =
0
= r{[x' (At. ... , An, B,

2T
0
2 1 1 2 1 1
T, £)][exp{T£(e-it;-tAo_
-1)}] ih} d't.

Thus the problem is reduced to the equation
a
hoTt
J = { exp { - 't 1-~o } !h (t, y)} lt=O'
where 't>O, fh (t, y) = 0 when t = kh0 , k:::;;;O.
We have
1-e-i~oho
J=
r
;t/h
J e
it~o-'t'
ho
~
fh (~o. y) d~o. (7 .41)
-nih
where {,. (s 0 , y) is a Fourier transform of the function fh (t, y) with

respect to t
(7.42)
By Eq. (7 .32)
ih (y, h)= ( 2~ r lsihl ~n.
I
i=1, ... , h
ei ((;, y) hn X
n
- i ~ ihhsh
~ j'!- •e h=1 d£.
X LJ 1 1• ···• 3 n
ji, · · ·' jn
ii44 OPERATIONAL METHODS
Denote e-itto by z. We may put Eq. (7.41) in the form
J= ~ e h:X ( h ~ fk (y, h) zh ) ~z = 0.
I z 1=1 k~1
Thus Eq. (7 .40) is proved. Hence the restriction of the function

1 1 2 )
u(t, y, h, h0)=[x ( A 0 , •• • , An, B, ~ ] /h(t, y)
-on the net Qh is a formal asymptotic (in the differentiable sense)
solution of Eq. (7.30) in the stripe [0, T] X Rn.
Using this asymptotic expansion with respect to the smoothness,
the asymptotic power expansion in h of the solution of Problem (7 .33)
(hence, of Problem (7 .30) as well) can be obtained as for the crystal
equation in Sec. 8 of Introduction.
APPENDIX
Spectral Expansion ofT-products
Here we shall outline a rigorous approach to the Feynman ope-

rator calculus wlich will generalize the concept of T-products,
widely used in modern quantum physics.
Define a T -product of functions of two self-adjoint operators A
and B in the Hilbert space H.
Let <Dk (x, y) be complex-valued functions such that for any
arbitrarily small B, [<Dk (x, y)]" E ~ 0 (R3). After Feynman we shall
use non-integer numbers over operators:
( s1 sz)
<D A,B =<D A,B
(1 2)
if s 1 < Sz. Consider a sequence of bounded operators
as N---?- oo and T > 0. Set fl.t =-wT and let

def
<D(t, x, Y)=<J)k(x, y) for t=tk = k·M.
Then
(1)
if <D (t, x, y) =F 0.
+
Assume ln <D (t, x, y) = <D (t, x) <p (t, x, y).
For the symbol in the right-hand side of (1) we obtain
exp{2j Mln<D(tk, x(tk), y(tk))). Hence it is evident that the limi-
ting symbol must be a functional on trajectories x (t), y (t).
35-01225
546 OPETIATIONAL 1\IETIIODS
When determining the spectrum of two operators, by virtue of

Chapter 2 it suffices to consider the products of spectral families
Ell.,xo(B)Eil.,yo(A), (Efl.,zo(z) being the characteristic function of
the interval z0- ~ ~z<zo+ ~ centered on the point zO) for any
(xo, yo) E R 2 and sufficiently small ~.
We shall apply the same approach for the T-product. Divide the
interval (- oo, oo) into subintervals of length 6 centered on xY and
setting 1 = ~ E 6 , Y~ (B) and in the same way 1 = ~ E 6 , xJQ (A) we obtain
' J
Here the sum is taken with respect to every possible sequence

{j1, ... , jN, it, ... , iN}· By setting
def 0 def 0
x(tk) = XJk' y(tk) = Yik
we can say that the sum is taken with respect to the totality of
discrete paths. In the limit it is natural to consider the set of all the
trajectories in R 2 • Under broad assumptions regarding the operators A
and B the spectrum of a T-product happens to be on the trajecto-
ries whose projection onto x is a set of stepwise trajectories. But
the y-component is discontinuous everywhere and unbounded on any
finite interval. When A= i :q , B = q, this fact reflects the
well-known uncertainty principle in quantum mechanics.
The value of (2) on the family of n-step x-trajectories with
arbitrary values at the points of discontinuity is equal to the n-th
term of the perturbation series in cp*. Therefore the spectral expan-
sion (2) in stepwise trajectories is identical with the expansion (1)
in the perturbation theory series in <p. We shall proceed to the
proof of the above assertions.
Let A and B be two self-adjoint vector operators A= A<1l, ... , A<,.l,
.8 = s<1l, ... , s<rl in a Hilbert space H, A being a generator of
degree l in the scale generated by the degree of B and vice versa.
* This fact is closely connected with the Feynman diagrams concept

(see V.V. Belov and V.P. Maslov, Appendix in V.V. Belov, E.M. Vorobyov,
V.E. Shat:ilov, The Theory of Graphs. Moscow, Vysshaya shkola, 1976.)
APPENDIX 547
Let H~ (H;) be the scale of Hilbert spaces with the norm
/fill~ ctcr li!i(B<J>+i)sf!l, (iifli~=liii(A<J>+Wfll), sEZ.

j j
By assumption A is a generator from H~ into H~ of degree l for

any integer s :S; r +
1, s > 0. In the same way B is a genera~
tor from HM. into ll)u of degree Z. Let Eg (x) be a characteristic func~
tion of the set Q belonging to the spectrum o (A) of the vector
operator A and Mes Q is the Lebesgue measure.
1 2
Definition. The speGtra of A, B will be called incompatible if
the following two conditions are fulfilled:
(1) there exists an integer Manda constant c>O such that when
.Mes Q ~ 0, Eg (A) as an operator from HM into H approaches
zero like c (Mes !:2) 112 :
1/ Eg (A) IIIIM--.H :S; c I Mes Q 11/2; (3)
(2) for all [p[;;?:c 1 (Mes Q) 112 +e (e>O being an arbitrarily small
number and c1 being a constant) there exists a constant c2 such that
i[Eg(A)eiBPEg(A)il:S;c2 (Mes !:2) 112+8 , P=P1, . .. Pr (4)
for Mes Q ~ 0.
Examples. (1) Let A= x, B = i :x , H = L 2 (R). Show that the
spectra of these operators are incompatible. We have
JJEg(x) (i :x+ir 2
f(x)j 2 ax:s;;
:s;;~maxj(i :'C +ir 2 f(x)I 2 :S;~!if(x)W\ ~=Mes Q,
i.e. II Eg (x) ( - i :X + i r 2
1[ :s;; v~. Besides
-t...!!.....
Eg(x)e ax Eg(:c)=Eg(x)Eg(x-t)=O
for t > c 1 ~.
(2) Let A= x 2 , B = i 2
0: 2 • Then for t > ~ 1 1 2 +e
II Eg (x2) eitBEg (x2) f (x) 112 :S;
:S; TJdx I Jei(x;t~l 2 f (£) d£ 2

1 :S; f ~2 /1 f (x) 1/ 2 :S; c~ 312 -e II f IF\
Q Q
i.e. condition (4) is fulfilled for e < 1/6.

35*
Condition (3) is fulfilled by virtue of the previous evaluation.

For simplicity, below we shall set r=1. Now assume that F(t, x)
is a function such that there exists a limit f (t, x) of the expression
N
II F (t,o
k=1
x)M, N =-t--+ oo
11t
belonging to the space .9Qiz(R) for all t >0 and !Jl (t, x, y) E .9Qi 1 (R 2)
is continuous in t with the function ~ e-ipy !Jl (t, x, y) dy being
ctmtinuous on the line p = 0*. To simplify the proof we shall
suppose in addition that F (t, x) :=/= 0 and let <D (t, x) = ln F (t, x).
Thus we assume that
t
S<l>{'t, x)d-r
e0 E.90iz.
Consider the T-product.
lim V N = lim IIN exp

{ [
<D tk, A
( tk)
+ cp ( th, tk th+M/2)
A, B
J!J.t }
N-.oo N->-oo k=i
(5)
with A and B satisfying the spectra incompatibility conditions.
It is known that under the above conditions lim V N det u (t) exists
N->oo
and satisfies the equation
a~t(t) = (& + ~) u (t), (6)
~ ( 1 2)
!Jl=CJl t, A, B ,
~
<D=<D(t, A).
The proof of this assertion is left to the reader.
Let Et.., xo (x) be a characteristic function of the interval
[xo+~, xo-~).
First we shall point out that under the above conditions
II Et.., xo~Et.., xo Jl ~ ca11 1 12+ 8
•
Here Et.., xO = Et.., xO (A). Hence

- - 1+e
II Et.., xo!JlEt.., xo IIH;\f_,.H~ IlEt.., xoqJEa, xo II II Ea, xo IIHM->H~c~~ , (7)
* Instead of the above conditions we can use the following:

lcpl<:(lxi+IYI)-M, fPELh and f(x, t)EC.
APPENDIX 549
c3 and c4 being independent of x. In fact,
II JJ E~. xo (A) eiBptE~. xo (A) X

1Ptl:::;c1~1/2+e
X eiAp2~
rp (t, Pt. P2) dpt dp2 ~ I
~ ~ II) e-iPtYrp(t, A, y)dy"dp1 ~
1Ptl:::;c1~1/2+e
~c 1 ~i/Z+e max I) e-ip yrp (t,

1 x, Y) dy ~~
x, 1Ptl:::;c1~1/2+e
~const ~i/Z+e,
(h:, ) ) E~.xo(A)eiBptE~.xo(A)X
1Ptl>ct~1/2+e
iAp ~
X e 2 rp (t, p 1 , p 2 ) dpt dp2h) ~
~ max
IP 1 i>c 1 ~ 1 /2+e, p 2E(-oo, oo)
I(h:, E~. xo (A) X
X eiBptE ~. xO (A) eiAp2h) In (ii llc+(!Ptl>ct~1/2+e) ~

~c/Jh:l] lihl/~ 112 +e; h:, hEH.
This completes the proof of the assertion. Consider
Prove that lim V N, xo (~) = E~. xo (A) 'IJl~. xo (t), with 'IJl~. xo (t) sa tis~
N-'>oo
fying the equation
(9)
~ ~
<D = <D (t, A), rp = rp t, A, B .
( 1 2)
E.\,xo(A)exp{[cD(t, A)+cp(t, A, B)]L'lt}E.\,xo=

=E.\,xo(A)exp{E.\,xo(A) [cD(t, A)+cp(t,A, B)] X
X E.\, xo (l ) M} = E.\, xo (l) { 1 + [E.\, xo (l) X

xcD(t, l)+cp(t, A, s)]E.\,xO (l)M}+
But
I {Ea,xo (l) LeD (t, 1) +cp (t, A, 8)]E.\,x0 (.4) rll~

~~C~IIE.\,xo<D 11 i!·llcp 11 (t, A, B)ll~
~ ~ c~ (max I cD I)~~ II cp II;~~= (max
x
IcD l +II cp ll.r:ez) n.
xER l
Therefore the term relative to the sign ~ in (10) is bounded in

the norm, hence
VN(fl)= fi E.\,xo (i) {1+fltE.\,xo(i)

k=i
[cD(tln lk)+
+cp ( th, A,tk t11 +M/2)
B
JE.\, xo ( A
tk+i) }
-1-0 (L'lt) =
tN) N {
=E.\,xO ( A IJ 1+ME.\,x0 ( th)
A [ (
CD lh, Ath) +
h=i
+cp ( th, A,
th tk+.\f/2)
B
l_,E.\,xo ( tk+i)}
A +O(M)=
=E.\, xo (A) ll exp { [hE.\, xo (A) [cD (tin A)+
h=i
due to the boundedness of the norm of the operator in the autono-

mous brackets. Hence (9) easily follows.
APPENDIX 551
Now sum expression (2) with respect to all E 6 , 11 q (B). We obtain

'
(11)
Now consider one term V N (6) of the sum corresponding to the

interval {.x:U X (0, T). Namely
t
s<1>(<, A)d<
Set 'P = e0 E6, x~ (A) 'Po·
We obtain
Hence
t t
~lll, xr = 'P- I [exp { I <D
0 t'
(1:, A) d,;} E{), xr~E{), xr] X
t
X 'P6, X~ (t') dt' det 'P-I Ko (t, t') 'Po, xr (t') dt' =
0
t
= 'P- JK
0
6 (t, t') 1p (t') dt' +
t t'
+ JK6 (t, t') dt' JK 6 (t', t") 1p6, x~ (t") dt". (12)
0 0
Evaluate the operator K 6 (t, t') from space H; into H;. We have
t
II (i +A) 8
K 6 (t, t') (i+At• II~ I exp { ) <:D(-r, A) d-r} E6, x~ I X
t'
t
xll(i+A) 8 ~(i+Ar·u~lle:\p{) <P(-r,x)d-rllfB x
t' l
X /1 <p (t, x, Y) 11.98 1 •

Hence it follows that lJl6. xf EH; and its norm is independent
of x~ and 8. Therefore
ll'i'6, X~ -ljl ll~clll E(), x~~E6, xr II X

X II E6, xr lfH,M-.H II E6, xr (A) (i + At• II' X
t
xljexp{j <P(-r,A)d-r
t'
}IJ, . ll(i+A) 'i'o/I.M+
HM-.HM
8
+ Czll E{), xr~E6, xf 112 1/ E6, X~ (A) (i + At• llll'i'6, xr ~~~

Hence by virtue of (7) we have
ll'i'6, xr -'i' II ~cs81+e (1 +I xn r•.
We shall evaluate one term of the sum (11) which corresponds
to a straight trajectory in the x-plane, i.e. to the interval {xn X
x (0, T). Now take the sum of the terms corresponding to the
intervals
{xnx(O,T), x~=(x~+j8), j=i, -1, ....
We have
t
I ~ 'i'6, xr+i6- exp { J<:D (1:, A) d-r:} ~ E6, xf+i6'i'o II=

j t' j
t
I
= ~ 'i'll, xf+il';- exp {
j
j CP (1:, A) d-r:} ljl
t'
0 II~
As 8-+ 0 we obtain that the sum of all the terms correspon-

ding to straight trajectories in (11) is equal to the first term of
the perturbation series in operator ~ for equation (5).
APPENDIX 553
Consider now the step {xn x (0, t') U {x~,} x (t', T) i :=1= n with
arbitrary values at the points of discontinuity. We put a tube-
around it x~- ~ ~x~x~+ ~ when O~t~t'-6' and -oo~x~oo·
when t' -6' < t< t+6' and x~,- ~ ~x~x~ ~ when t' 6' ~ + +
~t~T. We must now take the product in (11), which corresponds
to {x~} X (0, t'- 6'), sum with respect to all E~u,Xho for t E(t- 6',
+
t 6'), take again the product corresponding to the interval {x~} x
X (t' +6', T) and proceed to the limit as N-+ oo. Let Vin (6, 6')-
be the resulting limit. This means that we must solve equation (9)
for x0 =xL ~=6 up to the point t'-6', set the resulting solu-
tion at the moment t = t'- 6' as the initial condition for equa-
tion (6). The resulting solution must be put again as the initial
condition for (9) at the moment t' 6', for x0 = x?,, ~ = 6. We- +
have an integral equation of type (12) for t E (t'- 6', t' 6') +
1'+6'
'iJi (t' + 6') = ljl 1(t' + 6')- J Koo (t' + 6', t") ljl (t") dt" + 1
1'-o'
1'+6' I
+ J dt J Koo (t' + 6', t)" Koo (t, t") 'iJdt") dt";
t'-6' t'-6'
ljlt (t) = exp { J ('r, A) d-r:} lJl

1'-6'
6, x~ (t' -6');
ljli(t'-6')='iJ 0 x~(t'-6').
' l
+
At the moment t' 6' we act on 'i'i with the operator E 6 , x~ (A),
but ljl 1 = E~u, x.,~(A) '¢ 1 • Hence
t'+fJ' 1'+6'
EfJ, x~'i'dt' + 6') = J [exp { J cD (1:, A) d-r:}
1'-6' t
X
t'+fJ' I
+ E6, xR J dt J Koo (t' + 6', t) Koo (t, t") 'i'dt") dt".

1'-6' 1'-6'
Therefore
t'+6' T
Vin(6,6')=) dt[exp{~Cl>('t,A)d.-}x
t' -6' t
t
xE
6 xO
exp {IJ Cl>(.-, A) d.-} {1+
' i 0
T
+ 02 (61+e J x~ 1-s)} \jl 0 J+ exp { J cD (1:, A) d.-} X

t'+li'
t'+6'
xE6,x?,{1+0t(6 1 +eJx?,J-s)} ) dtx
t' -6'
t
X ) Koo (t' + 6', t) Koo (t, t") lPi (t") dt". (14)
t' -6'
Here 0 1 is independent of i, 0 2 is independent of n and K (t, t')

is independent of i and n. In the second term only the function
lJli (t') is dependent of i. By virtue of (13) we have
t
~ <P(-c, A)d-c
lPi = e0 E o{1+02(6 1 +ejx?l-s)}ljlo+
6, Xi
1'+6'
+ J Koo (t' + 6', t) lPi (t) dt.
t'-6'
Summing both sides of the equation with respect to we obtain

the equation for G = 2j \jli:
t
~ <l'>(-c, A)d-c t'+6'
G=eo (1+03 (6e))'i'o+ J Koo(t'+6', t)G(t)dt
t'-6'
where 0 3 is independent of n. Sum with respect to i and n both

sides of (14) for i 7'= n. For this we must sum with respect to all
n and i and subtract the sum in n = i from the expression obtai-
ned. By virtue of the above the latter sum is equal to 0 (6e) 6'.
APPENDIX 555
Hence
t'+o' T
= j [exp { l Q) (1:, A) dT} {1 + 0 (<'\e)} ~X
t' -0' t
t
X exp { j Q) (1:, A) dT} {1 + 0 (<'le)} 1Jlo] dt +
0
T t'+o' t
+exp{ ~
t+o'
CV(T,A)d,;}{1+0(<'1e)} J dt J
t'-0' t'-0'
KooX
X (t' +<'I', t) Koo (t, t") G (t") dt".
Divide the interval (0, T) into subintervals of the length 2<'1'. Set-
ting t' = tk_ =2M', sum with respect to k. It is easily seen that
approaching the limit as <'I' -+ 0 and <'I -+ 0 gives an expression
of the form
T T t
j [exp { j (J) (1:, A) dT} ~ exp { .\ cD (1:, A) dT} 1jl Jdt 0
0 t 0
which is to be proved.
Similarly for the sum corresponding to the steps with n discon-
tinuities we obtain the n-th term of the perturbation theory series
in cp.
Since the sum of the n terms of the perturbation series differs
from the solution of equation (5) by 0 ( :, ) , we obtain that the
sum (H) with respect to stepwise trajectories with no more than
n discontinuities differs from the sum (11) with respect to all
trajectories by 0 (
n.
+) .
This signifies that the projection of the
T -product spectrum onto the space of the trajectory x (t) consists
of stepwise trajectories. The spectral expansion (defined in the
Introduction) of the T -product in stepwise trajectories is identical
with the expansion in the perturbation series in ~.
Further note that since Ec,, xo (A) Ec,, yo (B) Ec,, xo (A) is less than
unity for sufficiently small ~ by virtue of (4), the terms of the
sum (2) containing the straight interval {x0} X (t 1 , t 2 ) decrease as
e-ar (a being a constant), where r is the number of the terms of
the form E..u, Ytq, for r of different values of time and finite ~. This
means that the spectrum consists of trajectories which are every-
where discontinuous and unbounded on any time interval not equal
to zero.
It is easily seen that for sufficiently small 6 the value
Ell x'lq exp {<P (tit, A) M} is close to the value exp {<P (tit, x~) ~t} Ell , xtq,
I
hence the sum (11) can he interpreted as an integral of the func-

T
tional exp { ) <P (t, x (t)) dt} relative to complex spectral measure*
0
concentrated on stepwise trajectories; the variation of the measure
on trajectories is bounded. It is established as above if we put
<11=0.
* The exact results involving this measure when A= i :X , B = x under

less limiting conditions obtained jointly by the author and A. G. Chebotarev
are given in the author's book Complex Markov Chains and the Feynman
Continuum Integral. M., Nauka, 1976.
Index
Absorption condition 132 Equation

Algebra bicharacteristics 126, 268
Banach 154 characteristics 126
commutative 26 A.-characteristics 57
of continuous function 149, 220 diff eren tial-diff erence 22
of functions of the generator Hamilton-Jacobi 65, 74, 121
group 159 Hamilton-Jacobi of the oscilla-
Lie nilpotent 104 tor 76
noncommutative 22, 25, 40 Hamilton-Jacobi with the dis-
normed 247 sipation 356
of operators 26, 40 of lattice oscillation 58
Associated element of operators 191 nonlinear, for a crystal 91
Asymptotic expansion 181 Schrodinger 76
Asymptotic quasiclassical 76 transfer 57, 481
D-asymptotic 420, 423 transport, with dissipation 137
h-asymptotic 420 wave 60
Asymptotic f:eries 421 Essential part of the function 128
Autonomous brackets 28 Essential point 469
extraction 29 Extension of the operator 176
Axioms of ~t-structures 41, 42
Factor-resolved set of operators 265
Banach scale 211 Finite functional 212
Bicharacteristic of the operator 129 Focus in a crystal 88
Bypassing focuses Hamiltonian 392 Formula
of the change of indices 44
of commutation 32, 35, 257, 259,
Canonical atlas of the curve 305 260, 273
Canonical projection 397 of commutation of Hamiltonian
Canonical transformation of the ope- and exponential 31
rator 274 K-formula 45
Cauchy data 384 Fourier transform 150, 218
Cauchy problem 384 Function
Cherenkov effect 83 asymptotically p-quasi-homoge-
Cohomology classes 465, 466, 469 neous 128
Commutator of operators 32 <'3-Dirac 149
Condition of the quantization 335 generalized 200
Convolution of the functional 214, 215 Hamiltonian 127, 144
p-quasi-homogeneous 127
smooth 64
Dissipativity inequality 377, 379 subordinate 128
Eigenelement of operators 191 Generating set of operators 191

Element, quasi-inverse 101 Generator 158, 171, 173
Enthalpy of the complex germ 361 of degree s with step k 211
558 INDEX
Index of the chain of patches 307 Phase in the nonsingular zone 71

Inequality of dissipation 139 Point of the resolvent set 225
Infinite polynomials 25
Isomorphic construction 14
Quasi-inverse sequences 101
Jacobian 66, 75
Regularizer of problem 61, 68
Relation of equivalency 27
Kernel of the operator 201 Representation of operator 104
Representation of nilpotent algebra
106
Resolvent sets 50, 225, 234, 188
Lagrangean manifold 360 Resolvent of vector-operator 234
Lagrangean manifold with complex Rotation of Jacobian 463
germ 360
Leading term of the function 128
Schmidt class 224
Solution, asymptotic 69, 140, 422, 481
Main problem 132, 481 Solution, general 19, 23
Space
Banach 154
Newton expansion 47 infinitely differentiable func-
tion 13
intermediate Banach 165
Operational Heaviside method 24 Hilbert 224
Operator 25, 40 Lk L~ 321
"birth" and "death" 243 phase 127
canonical 305, 319, 327, 333, 481 Sobolev 154
of difference differentiation 32 Spectral expansion of the operator 37
of differentiation 13, 25 Spectral weight 240
differential, D-asymptotic 423 Spectrum
differential, dependent on the discrete 191
parameter 402 factor 265
differential, S 2 -asymptotic 450 of generating set 195
generative 104 of matrix 35, 39
local canonical 469 of operators 40, 225, 234, 188
of multiplication 23, 25 of resolvent of vector-operator 234
pseudodifferential 52 Stationary phase method 75, 80, 181,
quantized canonical 336 301
quasi-identity 424 Stationary point 81
quasi-inverse 101 ~~-structures 40, 48, 51, 55, 147, 180
regular 195 Support of functional 212
regularized 336 Support of element 569
Schmidt 205 Symbol of mapping 17
Schrodinger 131 Symbol of operator 14, 173, 195
smooth 64 System
smooth canonical 309 of bicharacteristics 65, 74, 82, 84
of sum 17 dissipative Hamiltonian-Jacobi
translating (translator) 211 357
of translation 20 of generalized e.a. function 201
wave 60 of Hamiltonian 65, 77
Parseval 's equality 162 Tail of oscillations 81

Patch (singular and nonsingular) 305 Theorem of commutation 259, 260,
y-patch 374 261, 263
IXDEX 559
Theorem (Liouville) 365 V-object 322, 327

Theorem of product 44
Theorem of sum 42, 43 Weighting '\'-atlas 473
Trajectories of the Hamiltonian sy-
stem 77 Zone 369, 460

Maslov Operational Methods Mir 1976

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Maslov Operational Methods Mir 1976

Caricato da

Copyright:

Formati disponibili

operational

Translated from the Russian by

V. GOLO, N. KULMAN, G. VOROPAEVA

This translation has been read and approved by the author

© English translation, Mir Publishers, 1976

Sec. 11. The Functions of Components of a Lie Nilpotent Algebra and

Operational methods are such methods, which make it possible

detaching the non-smooth or rapidly oscillating part, serves as

In Sec. 6 the system of In Sec. 6 the system of axioms

condition for the completeness of eigen and associated elements.

Besides, the indices of closed paths on the Lagrangean manifold

* In the article "Operational Calculus Relating to Quantum Electrody-

Sec. 1. Solution of Ordinary Differential Equations

Clearly, the operator

is defined for every polynomial Pn (x), where a 1 are complex num-

There is a one-to-one correspondence between the operator Pn(D) E

Pn E K 1 [xJ. (To prove that this is a one-to-one correspondence, it

where a; is the operation (operator) of multiplication by the con-

To solve equation (1.3) we need the following lemma.

We prove this by induction. \Ve compute Deaxt (x):

where F (x) = vn-1eaxt (x). By the induction assumption

Equation (1.4) follows immediately from (1.5), and the lemma is

Thus equation (1.6) is equivalent to

We note that 1/D maps an element IE C"" into an entire class of

We return to equation (1.3). We have

Successively applying ('1.8), we obtain the formula

We have obtained a formula which gives a solution of equation

Here W (x) is obviously a linear combination of the derivatives

Conversely suppose yEP (D) _ 1 f. Then y (x) = P (D) u (x),

Proof. Let P, (x), Q, (x) (i = 1, 2) be polynomials and R, (x) =

R 1 (D) f (x) +R 2 (D) f (x).

and this proves the lemma.

where ai are complex roots of multiplicity ki of the polynomial

Hence, any solution y (x) of equation (1.3) can be expressed in

It is not difficult to verify*, that the general solution of (1.3) has

Pn(x)[Pn,s(x)+ ~ (Pn\xJ:k~O ~~]=1

* The statement is immediate since a solution of equation (1.3) may be

where Pn_1 (x) is a polynomial of degree equal to, or less than,

Sec. 2. Difference Equations

I (x+h) = ~ ~~ f<kl (x).

On the other hand, a formal expansion of ehD in a power series

Thus, the expansion in the Taylor series of f (x + h) is compared

where CC: is a space of infinitely differentiable finite functions on R

where A i are constants.

Consider that the series - ~ ekhD is a formal power series of the

be stated in .the form

Sec. 3. Solution of Systems of Differential Equations

where E is a unit matrix, 11 (x) = det B (x). Equation (3.2) for

We find by the argument stated above, that if Y (x) satisfies the

We see that in this case 11. (D) = (D + 1) 2 • Hence, the equation

It has thus been shown that the operational Heaviside method

Sec. 4. Algebra of Convergent Power Series of

The main difficulty is that the operators of multiplication and diffe-

of commutative variables but now two monomials, for instance,

'tl:f(x,y,z)-+f ,A,B,C = ~ au~tCkB3 Ai.

an operation of multiplication, which differs from multiplication

Example. denotes (A+B) Cor AC+BC AC+BC.

j (A,1 C3) B"'"'

j (A,1 B,2 C3) (2 1 c3)

j (A,1 A,1 C2) = g (A,1 C2) ,

Consider a substitution of an operational expression acting second

where H is an element of cff' equal to <p Hi is (c, D, ... , E),

where C~n is the corresponding binomial coefficient.

Then we shall consider

=dA.f-B) (A.f-B, 1)-A (A.f-B, 1h=