Lectures on

PARTIAL DIFFERENTIAL EQUATIONS

By
G. B. FOLLAND

Lectures delivered at the

INDIAN INSTITUTE OF SCIENCE, BANGALORE

under the

T.I.F.R.-I.I.Sc. PROGRAMME IN APPLICATIONS OF

MATHEMATICS

Notes by
K. T. JOSEPH and S. THANGAVELU

t,
.

~
.

Springer-Verlag Berlin Heidelberg GmbH

1983
 Author

G. B. FOLLAND
University of Washington
Seattle, Washington 98175
U.S.A.

© Springer-Verlag Berlin Heidelberg, 1983

Originally published by Tata Institute of Fundamental Research in 1983.

ISBN 978-3-540-12280-7 ISBN 978-3-662-00729-7 (eBook)

DOI 10.1007/978-3-662-00729-7

No part of this book may be reproduced in any form

by print, microfilm or any other means without
written permission from the Springer-Verlag Berlin
Heidelberg GmbH

Printed by N. S. Ray at The Book Centre Limited,

Sion East, Bombay 400 022 and published by H. Goetze.
Springer-Verlag, Heidelberg, West Germany
 CONTENTS

PREFACE

CHAPTER 1. PRE LIMINARIES

§ 1. General theorems about convolutions 1

§ 2. The Fourier transform 6

§ 3. Some results from the theory of distributions 13

CHAPTER 2. PARTIAL DIFFERENTIAL OPERATORS WITH CONSTANT

COEFFICIENTS

§ 1. Local solvability and fundamental solution 16

§ 2. Regularity properties of differential operators •• 22

§ 3. Basic operators in mathematical physics 25

§ 4. Iaplace operator 29

§ 5. The heat operator 47

§ 6. The wave operator 51 .

CHAPTER 3. L2 SOBOLEV SPACES

2
§ 1. General theory of L Sobolev spaces 62

§ 2. Hypoelliptic operators with constant

coefficients 77

CHAPTER 4. BASIC THEORY OF PSEUDO DIFFERENTIAL OPERATORS

§ 1. Representation of pseudo differential operators .• 82

§ 2. Distribution kernels and the pseudo local

property 86

§ 3. Asymptotic expansions of symbols 90

§ 4. Properly supported operators ... 93

CHAPTER 4. BASIC THEORY OF PSEUDO DIFFERENTIAL OPERATORS (CONTD.)

§ 5. ,¥DO 's defined by multiple symbols 96

§ 6. Products and adjoints of '¥DO's 102

§ 7. A continuity theorem for '¥DO on Sobolev

spaces 106

§ 8. Elliptic pseudo differential operators ••• 108

§ 9. Wavefront sets III

§ 10. Some further applications of pseudo differential

operators 117

CHAPTER 5. rPAND LIPSCHITZ ESTIMATES 126

REFERENCES 160
 PRE F ACE

This book consists of the notes for a course I gave at the

T.I.F.R. Centre in Bangalore from September 20 to November 20, 1981.

The purpose of the course was to introduce the students in the

Programme in Applications of Mathematics to the applications of Fourier

analysis - by which I mean the study of convolution operators as well

as the Fourier transform itself - to partial differential equations.

Faced with the problem of covering a reasonably broad spectrum of

material in such a short time, I had to be selective in the choice of

topics. I could not develop anyone subject in a really thorough manner;

rather, my aim was to present the essential features of some techniques

that are well worth knowing and to derive a few interesting results which

are illustrative of these techniques. This does not mean that I have

dealt only with general machinery; indeed, the emphasis in Chapter 2

is on very concrete calculations with distributions and Fourier transforms

- because the methods of performing such calculations are also well

worth knowing.

If these notes suffer from the defect of incompleteness, they

possess the corresponding virtue of brevity. They may therefore be of

value to the reader who wishes to be introduced to some useful ideas

without having to plough through a systematic treatise. More detailed

accounts of the subjects discussed here can be found in the books of

Folland [1], Stein[2], Taylor [33, and Tr~ves [4].

No specific knowledge of partial differential equations or

Fourier Analysis is presupposed in these notes, although some prior

acquaintance with the former is desirable. The main prerequisite is a

 ii

familiarity with the subjects usually gathered under the rubric "real

analysis" : measure and integration, and the elements of point set

topology and functional analysis. In addition, the reader is expected

to be acquainted with the basic facts about distributions as presented,

for example, in Rudin [7].

I wish to express my gratitude to Professor K.G. Ramanathan

for inviting me to Bangalore, and to Professor S. Raghavan and the

staff of the T.I.F.R. Centre for making my visit there a most enjoyable

one. I also wish to thank Mr S. Thangavelu and Mr K.T. Joseph for their

painstaking job of writing up the notes.

 CHAPTER 1

PRE LI MINARIES

IN THIS CHAPTER, we will study some basic results about

convolutions and the Fourier transform.

§1. GENERAL THEOREM> ABOUT CONVOLUTIONS

We will begin with a theorem about integral operators.

THEOREM 1.1 Let K be a measurable function on R n x lRn such

that, for some c > 0,

. n
fIK(X,y) Idy ~ c~ fIK(X,y) !dx ~ c, for every x, Y .!Q R •

If 1 ~ P ~. 00 and f E: LP (R n ), then the function Tf, defined by

Tf(x) f K(x, y) f (y) dy for almost every x in Rn,

belongs to LP (Rn ) and further,

If P = 00 , the hypothesis fIK(X,y) Idx ~ c is superfluous

and the conclusion of the theorem is obvious. If P < 00, let q

denote the conjugate exponent. Then, by Holder's inequality,

ITf(x) I ~ {fIK(X,y) Idy}l/q {fIK(X,y) Ilf(y) IP dy}l/p

~ c l / q {fIK(X,Y) I If(y) IP dy}l/p

From this we have,

fITf(x) IP dx ~ c p/ q ffIK(X,y) I If(y) IP dy dx

~ cl+p/qflf(Y) IPdx = c l +p/ q Ilf Il~

 2

Therefore IITf IIp.s c Ilf lip'

Next, we define the convolution of two locally integrable

functions.

DEFINITION 1.2 Let f and g be two locally integrable functions.

The convolution of f and g, denoted by f * g, is defined by

(f*g) (x) J f(x-y)g(y)dy = J f(y)g(x-y)dy = (~*f) (x)

provided that the integrals in question exist.

(The two integrals are equal by the change of variable

y .... x - y.)

The basic theorem on convolutions is the following theorem,

called Young's inequality.

THEOREM 1. 3 (Young's Inequality) Let f e: L1 (En) and g e: LP (En) ,

for 1 < P ~ 00. Then f * g e: LP (En) and

Take K(x,y) = f(x - y) in Theorem 1.1 . Then K(x,y)

satisfies all the conditions of Theorem 1.1 and the conclusion

follows immediately.

The next theorem underlies one of the most important uses of

convolution. Before coming to the theorem, let us prove the following

LEl+1A 1.4 For a function f defined on It and x in En,

x
we define a function f (y) = f(y-x). If fe:LP,l~p<oo,

then linillfx-fll =0.

x .... 0 p
 3

If 9 is a compactly supported continuous function, then

x
9 is uniformly continuous, and so 9 converges to 9 uniformly

as x tends to o.
x
Further, for Ixl .$. 1, 9 and 9 are supported in a common

compact set. Therefore, lim IlgX - 9 II O. Given f 10 LP we

x .... 0 p
can find a function 9 which is continuous and compactly supported

such that II f-g II < 10/3 for 10 ~ O. But then IlgX - fX II < 10/3
p p
also holds. Therefore

IIf X _ flip .$.llf x - gX lip + IlgX-g lip + Ilg-f lip

.$. 2 10/3 + Ilg X - 9 lip.

Since lim IlgX-g II = 0, we can choose x close to 0 so that

x .... 0 p

IlgX- gil
p
< 10/3. Then IlrX-f 11< 10 and this proves the lemma since

10 is arbitrary.

REMARK 1.5 The above lemma is false when p = 00. Indeed, IIfx .... f
00
in L II means precisely that f is uniformly continuous.

Let us now make two important observations about convolutions

which we shall use without comment later on.

i) Supp(f * g) c: Supp f + Supp g, where

A+B = {x+y : x 10 A, Y 10 B}.

ii) If f is of class ck and aOf(lol .$. k) and 9 satisfy

appropriate conditions so that differentiation under the integral sign

is justified, then f * 9 is of class Ck and

 4

THEOREM 1.6 Let ge:Ll(:R n ) and Jg(x)dx=a. Let

9 (x)
e:
= e: -n g(x/e:) e: > O. Then, we have the following:

i) If converges to af

in LP ~ e: tends to O.

ii) If f is bounded and continuous, then converges

to af uniformly on compact sets as I e: tends to O.

By the change of variable x + e:x, we see that

Jg (x)dx = a for all e: > O. Now

e:

(hg ) (x) - af (x) Jf(x-y)g (y)dy - Jf(x)g (y)dy

e: e: e:

J[f(x-y)-f(x)] 9 (y)dy
e:

J[f(x-e:y) - f(x)] g(y)dy

J[fe:y(x) - f(x)] g(y)dy.

If f e: LP and p < 00, we apply Minkowski 's inequality for integrals

to obtain

The function y+ Iffe: y - f II p is bounded by 2 II f II

p
and tends to

a as e: tends to a for each y, by Lemma 1.4. Therefore, we can

apply Lebesgue Dominated Convergence theorem to get the desired

result.

On the other hand, suppose f is bounded and continuous.

Let K be any compact subset of Rn. Given o> 0, choose a compact

set G lRn such that

 5

n I I9 (y) Idy < 6

:R 'G

Then

Sup I (f*g ) (x) - af(x) I ~ 26 11f 1100+ Sup If(x-e:y)-f(x) I Ilgldy

x e: K e: (x,y)e: K x G G

Since f is uniformly continuous on the compact set K the second

term tends to 0 as e: tends to o. Since 6 is arbitrary, we

see that

Sup [(f*9e:) (x) - af(x) I + 0 as e: + o.

x e: K

Hence the theorem is proved.

CORO LIARY 1. 7 The space is dense in

1 ~ P < 00.
2
-II (1-1 x I) for Ix I < 1
let 4> (x) e

o for Ixl ~ 1

Then 4> e: COO (:Rn )

o
and f 4> (x) dx = lla > O. If f e: LP and has

compact support, then a(f*4> ) e: Coo(Rn ) and by Theorem 1.6,

e: 0

a(f*4> ) converges to
e:
f in LP as tends to o. Since

functions with compact support are dense in LP , this completes the

proof.

PROPOSITION 1. 8 Suppose K c:::. En is compact and Q .=;; K be an

00
open subset of lRn. Then there exists a C function
o
4> (x) = 1 for x e: K and Supp 4> C Q •
 6

Let v={X€n:C1(X,K)~~O} where o=d(K,~n'm.

Choose a $0 € C: such that Supp $0C: B(O, ~o) and

f$ o (x)dx ~ 1. Define

$ (x) ,. f.
V 0
(x-y)dy = ($0 * Y ) (x).
'V

Then $(x) is a function with the required properties.

§ 2. THE FOUIlIU TRANSFORM

In this section, we will give a rapid introduction to the

theory of the Fourier transform.

For a function f € Ll (:Rn ), the Fourier transform of the

,..
function f, denoted by f, is defined by

,..
REMl.RK 1. 9 Our definition of f differs from some others in the

placement of the factor of 21T.

BASIC PROPERTIES OF THE FOURIER TRANSFORM

The proof of this is trivial.

Indeed,

II e- 21Ti x·~f(Y) g(x-y)dy dx

If e-21Ti(x-Y)·~g(x_y) e-21TiY·~f(y) dydx

fe-21Ti(x-Y)·~g(X_Y) dx ff(y) e-21TiY·~dy

= "f(~)g(~).
"
 7

Let us now consider the Fourier transform in the Schwartz

class S.. S(:Rn ).

PROPOSITION 1.12 For f € S, we have the following :

0". "
i) a""f = 9 where g(x) = (-2nix)S fIx).

i) Differentiation under the integral sign proves this.

ii) For this,we use integration by parts.

(asf)'" (~) = f e -21Tix·~ (aSf) (x)dx

= (-1) lsi I as[e-21Tix·~]f(x)dx

(-1) lsi (-21Ti~)S J e-2nix.~ f(x)dx

(2ni~) S r(~).

COROLIARY 1.13 If f € S, then "f € S also.

~ For multi-indices a and S, using Proposition 1.12, we

have

~a(asf) (~) = ~a«-21Tix)Sf(X»"(~)

(21Ti)-lal[aa«-2nix)Sf(X»"'(~)]

(-1) lsi (21Ti) ISI-lal (aa(XSf(X»>'''(~)

Since f €
a S
S, a (x f (x» € L
1
and hence
a 0
(d (x""f (x»)
"00
€ L. Thus

~a(aSf) is bounded. Since a and S are arbitrary, this proves

that f € S.

COROLIARY 1.14 (RIEMMN-IEBESGUE IEMMI.) "f

is continuous and vanishes at 00.

 8

By Corollary 1.13,this is true for f € S. Since S is

1
dense in Ll and Ilf 1100 ~ II fill' the same is true for all f € L •

Let us now compute the Fourier transform of the

Gaussian.
2
THEOREM 1.15 Let f(x) = e-nalxl , Re a > O. Then,
f(~) = a- n/ 2 e-a-lnl~12.

=I e -anlxl2d x
A
-2nix.~
PROOF f (~) e
n
e -2lfix'~' -anx~J dx. .
00

I
A
i.e. f(~) '" 11 J J e
j"'l -00 J

Thus it suffices to consider the case n = 1. Further, we can take

-1/2
a=l by making the change of variable x + a x.
2
-nx , x
Thus we are assuming f (x) = e €. R • Observe that

f'(x) + 2nx f(x) = O. Taking the Fourier transform, we obtain

+ i f'(~) = o.

Hence
A

f '(~)/f(~) -2n~

2
which, on integration, gives f(~) = c e -n~ , c being a constant.

The constant c is given by

00 2
C =
A

f(O) I e- lTX dx = 1.
-00

A
2
Therefore f~(~) e -n~, which completes the proof.

We now derive the Fourier inversion formula for the Schwartz

class S.
 9

THEORElo1l.I6 (Fourier Inversion Theorem) For f E S, (f)Y f.

PROOF First, observe that for f, 9 ELI, I f 9= I f g.

In ' fact,

I f(x)g(X)dx If e
-2rriy.x
f(y)g(x)dy dx

I [I e-2rriy.x g(x)dx] f(y)dy

J g(y) f(y)dy.

Given E > 0 and x in Rn, take the function ¢ defined by

Now

~(y) I e-2rriy.~ e2rrix.~ -rrE21~12d~

I e
-2rri(y-x)·~ -rrE21~12
e d~

-2
E-n e -rrE 1x-y 12 •

I f we take g(x) e-rrlxl2 and define 9 (x) E- n g(X!E), then

E

A
¢(y) 9 (x-y).
E

Therefore

A
fey) ¢(y)dy

f (y) 9 (x-y) dy
E

(f * 9 ) (x)
E
 10

But as E tends to 0, (f * g) converges to f, by Theorem 1.6

E

and clearly

Therefore (f) v f.

COROLLARY 1.17 The Fourier transform is an isomorphism of S

onto S

Next, we prove the Plancherel Theorem.

THEOREM 1.18 The Fourier transform uniquely extends to a unitary

map of L2(]Rn) onto itself.

PROOF For f E S, define f(x) f(-x). Then it is easily

checked that f = '"f, so that

J f(x) f(-x)dx

(f * f) (0)

Therefore, the Fourier transform extends continuously to an isometry

It is a unitary transformation, since its image S is dense

Let us observe how the Fourier transform interacts with

translations, rotations and dilations.

 11

(1.19) The Fourier transform and translation If

f(y-x), then

f e
-2niy·E.;
f(y-x)dy

f e- 2ni (z+x)·E.;f(z)dz (by putting y-x z)

(1.20) The Fourier transform and rotations (orthogonal transformations):

Let T Rn + Rn be an orthogonal transformation. Then

(f Q T) " (E.;) f e -2nix.E.; (f 0 T) (x)dx

. -1
f e- 2n1T y·E.;f(y)dy (by putting y Tx)

f e-2niy.TE.;f(y)dy

f(TE.;) '"
(f 0 T) (E.;).

'"
Thus, (f u T) f 0 T i.e. commutes with rotations.

(1.21) The Fourier transform and dilation r

-n - f (x/r) •
Let fr (x)

Then

f e~2nix.E.; r- n f(x/r)dx

-2niry.E.;
f e f(y)dy = '"f(rE.;>.

The last equation suggests, roughly: the more spread out f is, the

more '"f will be concentrated at the origin and vice versa. This notion

can be put in a precise form as follows.

(1.22) HEISENBERG INEQJALITY (n = 1): For f E: S(R), we have

'" 2
Ilxf(x) 11211~ f(~) 1122: (1/4n) Ilf 112 •
 12

PROOF Observe that

d
dx
(xf (x» = xdf
- (x)
dx
+ f (x) .

Thus

--
II f II ~ f f (x) f(x) dx

d
f f(x)[
dx
df
(xf(x) ) - x dx (x)] dx

=-f xf(x)
df
dx
(x) dx - f x df
dx (x)f (x) dx

= -2 Re f df
x f (x) dx (x) dx.

df
.s 21Ixf(x) 112 Il dx 112 (by Cauchy-Schwarz)

df
i.e. II f II ~ .s 21Ixf(x) 112 Il dx 112

But

Therefore,

or

Ilxf(x) 11211~f(~) 112 ~ (1/41T) Ilf II~ .

A GENERALISATION TO n VARIABLES

We replace x by x. , d by
_a_
• Also for any a ., b. E lR ,
J dx aXj J J
_0_
we can replace x. and
ax.
by x. - a. and ~ - b. respectively.
J J J Xj J
J
The same proof then yields:
 13

(1.23)

let us now take Ii f 112 = 1 and Let

f be small outside a small neighbourhood of A.

In this case, II(x. - a.) f(x) 112 will be small. Consequently,

J J
the other factor on the left in (1.23) has to be large. That is, if
A

the mass of f is concentrated near one point, the mass of f cannot

be concentrated near any point.

REMARK 1. 24 I f we take

a. = f x. I f (x) I 2 dx, b. = f ~. I f (~) I

A 2
d~,
J J J J

then inequality (1.2~) is the mathematical formulation of the

position-momentum uncertainty relation in Quantum Mechanics.

§ 3. SO ~E RESU TIrS FROM THE THEORY OF DISTRIBUTIONS.

In this section,let us recall briefly some results from the

theory of distributions. (For a more detailed tteatment, see, for

example [7] or [8]).

In the sequel, V'(Q) will denote the space of distributions

on the open set which is the dual space of When

Q = R n , we will simply wr i te V, instead of V, (lR n ). In the same

way, S' = S'(lRn ) will denote the space of tempered distributions and
E' = E' (lR n ) will stand for the space of distributions with compact support.

00
The value of a distribution u € V, at a function ¢ € C
o
will be denoted by < u,¢ >. If u is a locally integrable function, then

u defines a distr ibution by < u,¢ > = f u (x) ¢ (x) dx. It will sometimes
 14

be convenient to write f u(x) ~(x)dx for < u,~ >, when u is

an arbitrary distribution.

The convergence in v I is the weak convergence defined

by the following:

Un' u E V', un ~ u in V, means < un' ~ > ~ < u,~ > for
00
every ~ in C
o

Let us now recall briefly certain operations on distributions.

00
We can multiply a distribution u E V, by ,a C function

~ to get another distribution ~u which is defined by

< ~u,1jJ > = < u,¢1jJ >

A COO function ~ is said to be tempered if, for every

multiindex a , a
d ~ grows at most polynomially at 00 We can multiply

an u E S' by a tempered function to get another tempered distribution.

The definition is same as in the previous case.

If u E V, and f E C
00

o
, we define the convolution u * f
by (u * f) (x) < u, fx > where f
x
(y) =f (x - y).

The function u * f is COO and when u EE' , u * f is in

COO
o
The convolution * : V, x COO ~ COO
0
can be extended to a map from

VI X E I to V I • Namely,if u E V I , vEE' , and ~

00
E Co ,

< u * v, ~>=<u, V * ~ > where v is defined by <v,1jJ>= fv(x)1jJ(-x)dx.

The associative law

u * (v * w) (u * w) * w holds for u,v,w E V,

 15

provided that at most one of them does not have compact support. For

U E S' and f E S, U * f can also be defined in the same way and

00
u * f is a tempered C function.

Since the Fourier transform is an isomorphism of S onto

S I and S f 9 = Sf g, the Fourier transform extends by duality to an

isomorphism of S' onto S'

A
- 2TT i ( . ) . t;
For u E E ' C S ' , we have u (t;) < u, e >

which is an entire analytic function.

 16

CHAPTER 2

PARTIAL DIFFERENTIAL OPERATORS WITH CONSTANT COEFFICIENTS

§l. LOCAL SOLVABILITY AND FUNDA~NTAL SOWTION

For the sake of convenience in taking the Fourier transforms,

from now onwards, we will use the differential monomials

By a partial differential operator with constant coefficients,

we mean a differential operator L of the form

Further, we assume that la a I + o. In this case, we say that

lat= k
a
the operator L is of order k. I f we write P(~) = t
lal ~ k
a
a
~ ,

then we have L = P(D) .

For u sS, taking the Fourier transform, we see that

(P(D)u) (~) =

Let us now consider the following problem:

co
given f in C , we want to find a distribution u such that

P(D) u = f.

We say that the differential operator L is locally solvable at

x E Rn, if there is a solution of the above problem in some

o
co
neighbourhood of the point x for any f in C
o
 17

REMARK 2.1 We can assume that f has compact support. To see

ex>
this, we can take any ~ in 00 such that ~ = 1 in some

neighbourhood of the point x If we solve the problem P(D)u f~

o
near x, then u is a solution of our original problem since we
o
have f¢ = f in a neighbourhood of x
o

In the following theorem we give an affirmative answer to the

question of local solvability of L = P(D). The simple proof exhibited

here is due to L. Nirenberg.

THEOREM 2.2 Let L = be a differential operator with

ex> ex>
constant coefficients. If U E C , there is a C function u
o
n
satisfying Lu = f on 1l. •

PROOF Taking the Fourier tranqform of the equation III = P(D)u = f,

we see that P(~) ~(~) = ,f(~). It is natural to try to define u by

the formula

In general, P will have many zeros:so there will be a problem

A

in applying the inverse Fourier transform to f/p. But things are not
ex> A
n
so bad. Since fEe , f is an entire function of ~ E ([ and P
o
is obviously entire. Hence we can deform the contour of integration to

avoid the zeros of P(~).

, To make this precise, let us choose a unit vector n so that

Y a na + O. By a rotation of coordinates,we can assume that

laT= k a
n (0,0, ••. ,0,1). Multiplying by a constant,we can also assume that
 18

aCt = 1 where Ct o = (O,O, ..• ,O,k). Then we have P(~) = ~kn +

o
(lower order terms in ~ ). Denote with
n
n-l
JR

Consider P(~) = P(~', ~n) as a polynomial in the last

n-l
variable ~n in ~ for each ~ 1 in JR • Let 1...1 (~ ') , ••• , Ak (~ ')

be the roots of P(~', ~n) = ° arranged so that if i ~ j,

1m A. (~ ')
J

Since the roots of a polynomial depends continuously on the

coefficients we see that 1m A. (~') are continuous in ~'. To proceed

J
further, we need the help of two lemmas.

n-l
IEMW\ 2.3 There is a measurable function <jl : JR -+- [-k-l, k+l]
n-l
such that for all ~' in JR , min {1<jl(V) - 1m A. (~ ') I } ~ l.
1 ~ j ~ k J

PROOF Left as an exercise to the reader : (cf. G.B. Folland [1]).

(The idea is that at least one of the k+l intervals [-k-l, -k+l],

[-k+l, -k+3], .•. , [k-l, k+l] must contain none of the k points

1m A.(~'), j = 1,2, .•• ,k).

J

LEMMA 2.4 Let P (~) ~kn + (lower order terms) and let N(P) be

the set {~€ ~n : P(~) O}. Let d(~),=d(~,N(P». Then, we have

Ip(~) I ~ (d(~)/2)k

Take ~ in JRn such that P(~) '" O. Let 1l = (0,0, .•. 0,1)

and define g (z) P(~ + zll) for z in ~. This g is a polynomial

in one complex variable z. Let 1.. 1 , 1.. 2 ' •.. ' Ak be the zeros of the
 19

polynomial g. Then

so that
k

\9Jtl
9 (0) 1 II
j=l
11 - AZ,IJ .

Since ~ + Ajn E N(P), IAjl ~ d(~) so that when Izl ~ d(~),

19Jtl 1 < 2k. Also

9 (0)

Ig(k) (0) I =

But g(O) = P(~) and

9 (k) (0) k!

Therefore,

or

Hence the lemma is proved.

Returning to the proof of the theorem, consider the function

u (x) = fn-l f
:R 1m ~
n

By Lemmas 2.3 and 2.4, we have

 20

00 "
Since f € C , f (E) is rapidly decreasing as IRe ~I tends to 00
o
as along as 11m ~I stays bounded, so the integral converges

absolutely and uniformly together with all derivatives defining

00
a C function u.

Finally, by Cauchy's theorem,

P (D) u (x)

r e2TIix.~ f(~) d~ f (x).

l{-n
This completes the proof of Theorem 2.2.

Let us now consider the local solvability of L P(D) in

the case when f is a distribution.

As before, we remark that it suffices to take f € E' .

Further, it is enough to consider the case where f 6. Indeed,

if K satisfies P(D) K 6, then we have for any f € E' ,

P(D) (K * f) = P(D)K * f 6 * f f.

DEFINITION 2.5 A distribution K satisfying P(D) K = 6 is

called a fundamental solution or elementary solution of the differential

operator L = P(D).

A remarkable theorem due to MAIGRANGE AND EHRENPREIS states

that every differential operator P(D) with constant coefficients

has a fundamental solution. In fact, we can prove this result by a

simple extension of the preceding argument.

 21

THEOREM 2.6 Every partial differential operator P(D) with

constant coefficients has a fundamental solution.

PROOF Proceeding as in the previous theorem, we try to define

K(x)
~-l f
lR 1m t: =q,(t:')
n

Here, however, the integral may diverge at infinity. So, we consider

the polynomial
n
P (t:) (1 + 4n 2 I
j=l

where N is a large positive integer. Let

~-l
lR

where q, is chosen appropriately for the polynomial PN Then on

the region of integration, we have

so the integral will converge when N > n/2. We claim that

00
To see this, take q, € Co and observe that the transpose

< KN, PN(-D)(jl >

2nix.t: (PN(-D)q,) (x)

~
lR
~-l
lR
f
1m t: =q, (t: ')
e
PN (t:)
dt: n dt: ' dx
n
 22

cp(O) < Q,CP >.

Thus,

N
P (D) (1 - t:.) KN

N
so if we put K = (1 - t:.) KN we have P(D)K Q. Thus K is a

fundamental solution of P(D).

§2. REGULARITY PROPERTIES OF DIFFERENTIAL OPERATORS

DEFINITION 2.7 The singular support of a distribution f E V'is

defined to be the complement of the largest open set on which f is

00
a C function. The singular support of f will be denoted by

sing supp f.

a (X)D a
00
DEFINITION 2.8 Let L = a
where a
a
E C be a

differential operator. L is said to be hy~oelliptic, if and only if,

for any u E V, , sing supp u ~ sing supp Lu. In other words, L

is hypoelliptic if and only if for any open set nell and any
00 00
u E V '(m , "Lu E C tn) implies u E C (n)".

RE Mll.RK 2. 9 The operator L is said to be elliptic at x E mn

if
a (x)~
a +0 for every ~ in
n
m ,,{O}.
lat= k a
L is said to be elliptic i f L is elliptic at every x E mn • Elliptic

operators are hypoelliptic, as we shall prove later. This accounts for

 23

the name hypoelliptic.

00
We know that an ordinary differential operator with C

coefficients is hypoelliptic as long as the top order coefficient

is non-zero. But this is not the case with partial differential

operators as seen from the following

EXAMPIE 2.10 Take the operator L = L

dXdY
in The general

solution of the equation Lu = 0 is given by u(x,y) f (x) + g (y)

where f and g are arbitrary Cl functions. This shows that L is

not hypoelliptic.

We observe that if L is hypoelliptic, then every fundamental

solution of L is a COO function in ll.n '- {O}.

In the case of partial differential operators with constant

coefficients, this is also sufficient for hypoellipticity. Indeed,

we have the following

THEORE t:. 2. 11 Let L be a partial differential operator with

constant coefficients. Then the following are equivalent

a) L is hypoelliptic.

b) Every fundamental solution of L is COO in ll.n ,{O}.

00 n
c) At least one fundamental solution of L is C in ll. 'dO}.

That (a) implies (b) follows from the above observation

and that (b) implies (c) is completely trivial. The only nontrivial

part we need to prove is that (c) implies (a). To prove this implication,

we need
 24

eo
IEl>'l<IA 2.12 Suppose f € V I is such that f is C in

Then sing supp(f * g)c= supp g.

eo
Suppose x ~ supp g. We will show that f * 9 is C

in a neighbourhood of x.

Since x ~ supp g, there exists an €>0 such that

eo
B(x, €) n supp 9 ~. Choose <j>€C (B(O, €/2))
o
such that <j> 1

on B(O, €/4). Now, f * 9 = (<j>f) * 9 + (1 - <j»f * g. Since

eo
(1 - <j»f is a ceo function, (1 - <j»f * 9 is C . Also

Supp (<j>f * g) c Supp <j>f + Sup~ 9

~{y : dIy, supp g) ~ €/2}

which does not intersect B(x, €/2). Therefore, in B(x, €/2),f*g (l-¢) hg
eo
which is C

PROOF OF THEOREM 2.11 ~t K be a fundamental solution of L

eo
Suppose u € V, and Ul € C (m
where n CEn is open.

For x € n, pick € > 0 small enough so that B(x, €) c n

eo
Choose <j> € C (B(x,€)) so that
0
<j> =1 on B(x, €/2). Then

L(<j>u) = <j>Lu + v where v =0 on B(x, €/2) and also outside B(x,€) •

We write

K * L(¢u) =K * <j>Lu + K * v

is a ceo function so that K * Ul is a ceo function. Also

o
K* v is a ceo function on the ball B(x, €/2) by Lemma 2.12.
 25

00
Therefore K * L(<j>u) is a C function on B(x, E/2) •
00
But K * L(<j>u) IK * <j>u 15 * <j>u = <j>u. Thus, <j>u is a C function
00
on B(x, E/2). Since <j> 1 on B(x, E/2) , u is a C function

on B(x, E/2). Since x is arbitrary, this completes the proof.

§3. BASIC OPERATORS IN MATHEMATICAL PHYSICS

In this section, we introduce the three basic operators in

Mathematical Physics. In the following sections,we shall compute

fundamental solutions for these operators and show how they can be

applied to solve boundary value problems and to yield other information.

n
(i) LAPLACE OPERATOR t:, = ~
L in If u (x)
2
j=l clx.
J
represents electromagnetic potential (or gravitational potential) and

p denotes the charge (resp. mass) density, then they are connected

by the equation t:,u - 4np. If the region contains no charge,

i.e., if P = 0, t:,u = O. This equation is called the homogeneous Laplace

equation, and its solutions are called harmonic functions.

n+l
(ii) HEAT OPERATOR : L = ~
at - t:, in lR • If u(x, t)

represents the temperature of a homogeneous body at the position x

and time t, then u satisfies the heat equation t:,u = 0 in

n+l
(iii) WAVE OPERATOR : 0 = in lR If u (x, t)

represents the amplitude of an electromagnetic wave in vacuum at

position x and time t, then u satisfies the wave equation []u 0

 26

n+l
in R .

The equation CJ u =0 can also be used to describe other

types of wave phenomena although in most cases, it is only an

approximation valid for small amplitudes. More generally, the

equation (J u = f describes waves subject to a driving force f.

The Laplace operator 6 is an ingredient in all the above

examples. The reason for this is that the basic laws of Physics are

invariant under translation and rotation of coordinates which severely

restricts the differential operators which can occur in them. Indeed,

we have

THEOREM 2.13 Let L be a differential operator which is invariant

under rotations and translations. Then there exists a polynomial Q

in one variable with constant coefficients such that L = Q(6).

PROOF Let L = P(D) = I a Da . Since P(D) is invariant

lal ~ k a
under translations, a are constants. Let T be any rotation.
a
We have
A A
P(T~) ~(T~) (P(D)~) (T~)

A
(P(D)~ 0 T) (~)

A
(P(D) (~ 0 T» (~)

= P(~) ¢(T~) , for every ~ •

Thus P(T~) P(~) so that P is rotation-invariant.

 27

k
Write P(E,;) = I P. (E,;)
J
where P. (E,;)
J
is the part of P
j=O
which is homogeneous of degree j. We claim that each p. is
J
rotation-invariant. Indeed, if t is any real number,
k
j
P(tE,;) I
j=O
t P. (E,;).
J
For a rotation T, since P(E,;) = P(TE,;) , we

have
k
P(tE,;) I
j=O
t j P. (TE,;)
J

so that
k
I
j=O
t j (P. (TE,;) - P . (0 )
J J
O.

This being true for all t, it follows that P.oT = P . . But the only
J ]
rotation-invariant functions which are homogeneous of degree j are

of the form P . (E,;) c.!E,;!j where c. are constants.

] ] ]

Indeed, since Pj is rotation-invariant, P. (E,;) depends only

J
on !E,;! and thus, for ! E,;! + 0,

Since is not a polynomial when j is odd, c. o in that

J
case. Therefore,

Taking O(x)

RE~RK 2.14 This theorem applies to scalar differential operators.

If one considers operators on vector or tensor functions, there are

first order operators which are translation-and rotation-invariant, namely,

 28

the familiar operators grad, curl, div of 3-dimensional vector analysis

and their n-dimensional generalisations.

DEFINITION 2.15 A function F(x) is said to be radial,if there is

a function f of one variable such that F(x) f(r), r = Ixl.

1 n
When F is radial and F e: L (lR ), we have
co
n-l
F(x) dx f f f(r) r da(x) dr
o Ixl=l

n-l
where da(x) is the surface measure on S ,the unit sphere in

Thus
00

f F(x) dx f f(r) r
n-l
dr
lR n o
n-l
where w is the area of S .
n

Let us now calculate We have

Now

f
00
-nr2 n-l
e-nlxl2dx w
n f e r dr
0

w 00
n -s (s/n)n/2- 1
2n f e ds, s
0
w 00
n -s
f e s n/2-1 d s
2n n/ 2 0

wn f(n/2)/(2n n/2 )

Therefore

w 2n n/ 2/f (n/2) •
n
 29

§4. LAPLACE OPERATOR

First, let us find a fundamental solution of 6 , i.e. we

want to find a distribution K such that 6 K = 0- •

Since 6 commutes with rotations and 8 has the same property,

we observe that if u is a fundamental solution of 6 and T is a

rotation, then u ~ T is also a fundamental solution. We therefore

expect to find a fundamental solution K which is radial.

Let us tarry a little to compute the Laplacian of a radial

function F. Let F(x) f(r), r = Ixl. Then,

n
_8_
6F(x) I
j"'l
ax .
[f '(r) x./r)
J
J
2
n x.
I (f" (r) -L +
2
o.u
r
-~ x 2. )
3
j=l r r J

f"(r) + ((n-l)/r)f'(r)

Now set K(x) f(r). If K is to be a fundamental solution of 6

we must have

f"(r) + ((n-l)/r)f'(r) o on (0, 00) •

From this equation

f"(r)/f'(r) - (n-l)/r.

Integrating, we get

l-n
f '(r) c r
o

with a constant c One more integration yields

o
 30

2-n
f(r) cl r + c2 when n f 2

= c l log r + c 2 when n = 2.

Since constants are solutions of the homogeneous Laplace equation,

we may assume that c 2 = 0. Thus if we set F(x) = Ixl2-n (n f 2)

or F (x) = log I x I (n=2) , we expect to find that b. F = c 0 for some

c f 0, and then K = C-1F will be our fundamental solution.

In fact,we have

THEOREM 2.16 If F(x) = Ixl 2- n on R n (n f 2), then

b.F = (2 - n) WnO, where wn is the area of the unit sphere in Rn.

PROOF For E > 0, we define Then

F is a COO function on lR n , and an application of Lebesgue's

E

Dominated Convergence Theorem reveals that FE converges to F as

E tends to 0, in the sense of distributions. Therefore b. F

E

converges to b. F , in the same sense. Let us now compute b. F •

e:
n
b.F (x)
E I
j=l

In {(2-n)(lxl
2 2 -n/2
+E)
2 2 (-n/2)-1
+(2-n)(-n)(lxl+E) x
j=l

1 n
Thus we see that b.F e: L (lR ). A simple computation shows that
E
-n
b.FE(X) = E b.Fl (X/E); so, by Theor em 1. 6, b.F E tends to <f b.F 1)0 as

E tends to 0. Therefore b.F = (f b.Fl)O, and we need only to compute

f b.F l •
 31

n(2-n) J (1 + JxJ2) (-n/2)-1 dx

co
n(2-n)Wn I (1 + r2) (-n/2)-1 r n - 1 dr.
o

putting r2 + 1 s, we see that

J s (-n/2) -1 (s-l) (n-l) /2

co
1
J 6 Fl "2° (2-n)w
n
ds
1

f
co
1 S-2(1 _ !) (n/2)-1 ds
= "2 n (2-n)w
n 1 s
1
1
= "2 n (2-n)w
n J (l-a) (n/2)-1
da, a = 51
0

"21 n (2-n)wn (2/n) (2-n)w

n

which completes the proof.

EXERCISE Show that i f F(x) = log JxJ on ll? , then 6F 2no •

{ Ixl'-n + 2)
(2-n)W n (n
COROUARY 2.17 Let K(x)=
1
2n log JxJ (n = 2)

Then K is a fundamental solution of the Laplacian.

COROLLARY 2.18 6 is hypoelliptic.

Follows from Theorem 2.11.

It is also instructive to compute the fundamental solution

of 6 by the Fourier transform method.

If K is a fundamental solution of 6, we have 6(K * f)

6K * f
 32

f (~) ,

so that, at least formally,

We observe that when n > 2, the function -1/(4n21~12) is locally

integrable and so defines a tempered distribution. We want to

show that its inverse Fourier trans{orm is our fundamental solution

K. For that purpose, we will prove a more general theorem.

THEOREM 2.19 For 0 < a < n, let F be the locally in,tegrable

a
function F (~) = I~I-o. Then
a

F" (x) (f (21 (n-o.) ) /f (20.»

1 n o.-n/2 Ix lo.-n .
a

Before proving this theorem, let us pause a minute to

observe that it implies

f«n/2)-D n 2-n/2 Ixl 2- n = (f(n/2)/(n/2-l»n 2-n/2 Ixl 2- n

2 2-n
= (-4n / «2-n)w n » Ix I ,so that

as desired.

PROOF OF THEOREM 2.19 The idea of the proof is to express F

a
as a weighted average of Gaussian functions, whose Fourier transforms

we can compute.

To begin with,
00 00

f e- rt t o. - l dt = f e
-s
s
0.-1
r
-:-0.
ds r
-a
r(a), (r > 0, a> 0).
o o
 33

In other words, for any r > 0 and a > 0,

00

r
-a 1
r(a)
f e- rt t a - l dt •
o
Taking r = TI I ~12 and replacing a by a/2, then

TI
a/2
r(a/2)
7e-TII~12t t(a/2)-1 dt
o
which is the promised formula for F as a weighted average of
a
Gaussians. Formally, we can write
00
a/2
fn
e2TIix·~I~I-~d~ = TI J f e2TIix·~e-TII~12tta/2-ldt d~
r (a/2) n 0
lR lR

a/2 00
TI (e-TII~12tr (x)t(a/2)-1 dt
r (a/2) f
0

TI
a/2
r (a/ 2)
7 e- TI (l x I 2 )/t t -n/2 t a/2-l dt
0
a/2 00 2
TI -TIIXI s (n-a)/2-l
r(a/2) J e s ds
0
a/2
TI r«n-a) /2) I x Ia-n
r(a/2) (n-a)/2
TI

In this computation, the change of order of the integration is

unfortunately not justified, because the double integral is not absolutely

convergent. This is not surprising, since F is not an Ll ftmction;

a
instead, we must use the definition of the Fourier transform for

distributions.

For every ¢ € S,
 34

Now multiply both sides by t (a/2) -1 dt and integrate from 0 to (x).

J .inr
CXI e-n/ t Ix I2 t a- n )/2-1 (j> (x)dx dt .
o R

The change of order in the integral is permitted now and integrating,

we obtain,

f(a/2) f( (n-a)/2)
a/2 (n-a) /2
n n

i.e. < Ie-I-a, ~ >

<, 'i'
=< f«n-a)/2)
f (a/2) n
a-n/2 I la-n
x ,
"'>
'i' •

This shows that

F" (x) f«n-a)/2) na -n/2 Ixla-n

a f(a/2)

as desired.

This analysis does not suffice to.explain _r4n21~12)-1

as (in some sense) the Fourier transform of (2n)-1 log Ixl in,the

case n = 2. .One way to proceed is to define

and to study the behaviour of G and GV as tends to 2.

a a
This analysis can be carried out just as easily in n dimensions

where the problem is to study G and G" as a tends to n.

a a
Let R
a
= GV
a
for o < a < n. (R is called the Riesz
a
~tential of order a) • By the preceding theorem, we have
 35

R (x)
(f «n-a) /2) Ix Ia-n
a 2a 1T n / 2 f(a/2)

Moreover, if f E S,

(_~ya/2 (f * R ) = f ,
a

in the sense that (21T1~I)a (f * R) (~) '"f (~) .

a

As a tends to n, the f - function in R blows up. However,

a
since (2nl~1 lao = 0, we see that (_~)a/2 1 = 0; so we can replace
R by R - c, c being a constant, still having
a a

If we choose c = ca appropriately, we can arrange that R - c

a a
will have a limit as a tends to n. In fact, let us take

f «n-a)/2)
c and define R' R c. Then
a a a. a

R '(x) f «n-a.) /2) ( Ix Ia-n - 1)

a 2a 1T n / 2 f (a/2)

2f(n-a) /2) +1) ( Ix Ia-n - 1)

2a 1T n / 2 f(a/2) n - a

Letting a tend to n, we get in the limiting case

l-n
R '(x) - 2 log Ix I.
n
1T n / 2 f(n/2)
1
When n = 2, -R'(x)
2
= -2n log Ixl and ~(f * (-Ri» = f, so we

recover our fundamental solution for the case n = 2.

 36

It remains to relate the function Gn{~) = {2nl~I)-n to the

Fourier transform of the tempered distribution F' One way to

n
make G into a distribution is the following.
n

Define a functional F on S by

¢ {[hP {OJ d~ +
< F,<p > f f
I~I ~ 1 (2n I~ I) n I~I > 1

Note that the first integral converges, since, in view of the mean

value theorem, I<p{~) - <p(0) I ~ cl~1 . It is easy to see that F is

indeed a tempered distribution. Further, we observe that when <p(0) 0,

< F,<p > = f <P{~)G {~)d~, Le. F agrees with G on lRn,{O}.

n n

Just as R' was obtained from R by subtracting off an infinite

n n
constant, F is obtained from G by subtracting off an infinite
n
multiple of the 6 function. This suggests that F is essentially

the Fourier transform of R' In fact, one has the following

n

EXERCISE Show that F - (R ') is a multiple of 6

n

We now examine a few of the basic properties of harmonic

functions, i.e., functions satisfying 6u = O. We shall need the

following resultS from advanced calculus.

THEOREM 2.20 (DIVERGENCE THEOREM) Let ~ be a bounded open set in

lR n with smooth boundary a~. let V be the unit outward normal

vector on a~. Let F: lR n -+ lR n be a Cl function on ~, the closure

of ~. Then we have
 37

J F.v do f div F dx f ~L ~ dx .
art rt rt j=l aX j

CONSE~UENCES OF THEOREM 2.20

When we take F grad u, F.v = grad u.v = au

av ' the normal

derivative of u, and div F = div grad u = ~u. Therefore

au do
dV

When we take F =u grad v - v grad u, div F u ~v - v ~u.

Therefore

f (u 3v - v aU)dO f (u ~ v - v ~ u)dx.
art av av rt

This formula is known as Green's formula.

For harmonic functions, we have the following mean value

theorem.

THEOREl'i 2.23 let u be harmonic in B(x, r). Then, for any

o
p < r, we have

u (x )
o
1
f u(x)do(x).
ix-xo i=p

That is, u(x ) is the mean value of u on any sphere centred at x

o 0

PROOF without loss of generality, assume that x

o
o. Since
00
~u =0 and ~ is hypoelliptic, u is a C function. Now

formally,
 38

u (0) < <5 , u >

I

f <5 (x) u (x) dx

Ixl < 0

f t:,K(x) u(x) dx
Ixl < 0

f (u t:, K- K t:, u) dx
Ixl < 0

(u 3K _ K au ) do , by (2.22).
J dV dV
Ixl p

Of course, we are cheating here by applying Green's formula to the

non-smooth function K. Nonetheless the result is correct, and we

leave it as an exercise for the reader to justify it rigorously (either

C/O
approximate K by C functions as in the proof of Theorem 2.16 or

apply Green's formula to the region € < Ixl < p and let € tend to 0).

On the circle Ixl = p K is a constant and so

au au
I K av do const. f av
do const. f t:,u do O.
Ixl = p Ixl = p Ixb p

Further on Ixl p , l l
dr
, the radial derivative. Therefore
dV

on p •

Thus we have

u (0)
1
f u(x) do as desired.
Ixl = p

As a corollary to the mean value theorem, we can derive the maximum

principle for harmonic functions.

 39

THEOREM 2.24 Suppose n is a connected open set in Rn. Let

u be a real-valued function harmonic in n If A sup u(x),

n
then, either u(x) < A for all x € n or u(x) =A on n

Suppose that u(x ) A for some x € n. From the mean

o o
value theorem

u(x )
o
1
f u(x) da(x)
Ix-x o I =p

where p is small enough so that {x: Ix-xol .s p}cn. By our

assumption, the integral does not exceed A. If u(x l ) < A for some

point of B(x , p) , then u(x) < A in some neighbourhood of xl by

0

continuity. Taking r = IXl - x I < p, we have

0

u(x )
o \-1
r
f
Ix-x I=r
u(x) da(x) < A ,
o

a contradiction. Therefore, if we set n l = {x € n: u(x) A}, then

nl is an open subset of nand nl f ~. Further

n 2 = {x €n: u(x) < A} is also an open subset of n and n l U Q2 = n.

The connectedness of Q and the non-emptiness of nl force n2 to

be empty. Thus u (x) - A on n

COROLlARY 2.25 If n is a bounded oeen set in Rn , u € Cm)

and 6u = 0 in n, then

sup I u (x) I sup lu(x)l.

n an
Since the function u is continuous on the compact set n,

Iu (x) I attains its supremum on n at some point x

o
By multiplying

u by a constant, we may assume that u(x) = lu(x

o 0
)1.
 40

If x E an, the corollary is proved. Otherwise x E n

o 0

and by the previous theorem, on the connected component of

containing x
o
Re u(x) = u(x ) and hence
o
Im u(x) = o. Since

u is continuous on n, u(x) = u(x ) for all x E an Thus

o
sup ju(x) I = sup ju(x) I.
n art

CORO ILARY 2. 26 If u and v are in C(rt), n is bounded,

/::,.u = /::,.v o on rt and u =v on art, then u =v everywhere.

PROOF Apply the previous corollary to u - v.

The following boundary value problem for Laplace's equation,

known as the Dirichlet Problem, is of fundamental importance:

given a function f on art, find a function u such that

/::,.u o on rt and u = f on art .

When n is bounded and f E C(an) the uniqueness of the

solution, if it exists at all, is assured by Corollary 2.26. The problem

of proving the existence of solutions of the Dirichlet problem is

highly nontrivial. We shall solve a special case, namely, when rt is

a half-space and then indicate how similar ideas can be applied for

other regions.

First of all a word about notation: we will replace F n by

lRn +1 with coordinates (Xl' x 2 "" ,x n ' t). Our Laplacian in lRn +1
n
will be denoted by where at = .l..
at
and I
j=l
Now our Dirichlet Problem is the following:

given a function f on lRn, find a function u such that

 41

(2.27) «()2 + t,) u (X,t) t > 0

t

u(x,O) f(x), X E: JR n .

Since the half space is unbounded, the uniqueness argument given

above does not apply. Indeed, without some assumptions on the

behaviour of u at 00, uniqueness does not hold. For example, if

u(x,t) is a solution, so is u(x,t) + t. However, we have

THEOREM 2.28 Let u be a continuous function on JRn x [0,00)

satisfying

i)

ii) u(x, 0) = 0 for x E: lRn, and

iii) u vanishes at 00

PROOF Applying the maximum principle for u on B(O,R) x (O,T),

we see that the maximum of lui on B(O,R) x (O,T) tends to zero as

R, T tend to 00. Hence u = O.

REWl.RK 2.29 Hypothesis (iii) in Theorem 2.28 can be replaced by

(iii) , u is bounded on JRn x [0,00). But this requires a deeper

argument (See Folland [lJ).

To solve the Dirichlet Problem, we apply the Fourier transform

in the variable x. We denote by u(~, t) the Fourier transform

u(~, t) = f e-27Tix'~ u(x,t) dx.

 42

If we take the Fourier transfor~ of (2.27), we obtain

° on R n x (0, eo)

"-
u (t.;, 0) f(t.;) on

The general solution of the ordinary differ~ntial equation (ODE)

(d~ - 4n 2 1t.;1 2 )u(t.;, t) = ° is given by u(t.;,t) = a(t.;)e- 2ni t.;lt+ b(t.;)e 2n1 t.;lt
"-
and the initial condition is a(t.;) + b(t.;) = f(t.;).

This formula for -u will define a tempered distribution in t.;,

provided that as It.; I tends to eo , la(t.;) I grows at most polynomially,

and ib(t.;) i decreases faster than exponentially. We remove the

non-uniqueness by requiring that u should satisfy good estimates in

terms of f which are uniform in t - for example, that u should be

bounded if f is bounded. This clearly forces b(t.;) = 0, since

e 2nlt.;lt blows up as t tends to eo . Thus we take

u(~,t) = f(~) e- 2n1 t.;lt. Taking the inverse Fourier transform in the

variable t.;

u(x,t) [e -2nt( 1·1)] V(x) is the

Poisson Kernel for R n x [O,eo).

We now compute Pt explicitly. When n = 1, this is easy:

eo
fOe2nixt.; e 2n t.;t dt.; + 2n ixt.; -2nt:t
Pt(x) f e e dt.;
-eo
°
1 -1 t 2 2 -1
[(t + ix) + (t - ix)-l] (x + t )
2n n

To compute Pt(X) for the case n > 1, as in the proof of Theorem 2.19,

the idea is to express e -2nlt.;lt as a weighted average of Gaussians.

 43

Here's how:
-s -S2/4s
IEM<IP. 2.30 When S > 0, e -s I ..::.e_....::e
IX>
_ _ _ ds •
o /(1IS)

PROOF First, observe that

This can be proved by using the residue theorem or by applying the

1
Fourier inversion theorem to P1 (x) =~ (1 + x 2 )-1 on R • We

also have

Therefore,

Putting T = 2110 and changing the order of integration,

IX> IX>

~ of f -00

co 2
-s (411S)-1/2 -S /4s
i.e. e -S 2 f e e ds
0
2
IX> -s -S /4s
e e
f ds
0 /(1Is)

which is the required expression.

Returning to the computation of Pt(x), we have

 44

Taking 2TII~lt in the lemma, we get

P t (x)

00 -s 2 2 2
Se (TIt )-n/2 e-(s l xl )/t ds
o /itS s
2 2
S e-scl+(lxl
00
TI-(n+l)/2 t- n /t ))s(n-l)/2 ds
o
TI-(n+l)/2 t- n f(n+l)/2
(1 + 1:12 ) (n+l)/2

Thus we have

(2.31) Pt(x)

1
,
00
In particular, we see that P t E: L n L so that f
* Pt is well

defined if, for example, f E: r?, 1 ~ P ~ 00. That (t. + ( 2 )p

t t = 0

follows by taking the Fourier transform, and hence

(t. + d t2 ) (f * P
t
) =f * (t. + ( 2 )p
t t
=0 for any f.

Further note that

"-
PI (0) = 1.

Therefore, by Theorem 1.6, when f E: LP , 1 ~ p < 00, f * P

t
tends to

f in the LP norm and when f is bounded and continuous f * Pt

tends to f uniformly on compact sets as t tends to O.

If we take f continuous and bounded, then for u(x,t)

we have lim u(x,t) = u(x,O) = f(x). Thus the function

t + 0
 45

u(x,t) = (f * P t ) (x) is a solution of the Dirichlet Problem for

the half space.

REM\RK 2.32 The Poisson Kernel Pt is closely related to the

n+l
fundamental solution Kn+l of the Laplacian in lR Indeed

Kn+l (x,t)

and hence

Pt(x)

EXERCISE Check the above equation using the Fourier transform.

(Start with (a~ + 6)K n + l = o(x)o(t) and take the Fourier transform

in the variable x to obtain

o(t).
Solve this equation to obtain

e - 2TT I~ II t I
4TTI~1

Then
1 -2TTI~lt
2 e , t > 0, so that at Kn+l

Our formula (2.31) for Pt makes sense even when t < 0 and

we have P_t(x) =- Pt (x), so that lim f * Pt +f. We further

t + o±
observe that

n+l
where and mean convolution on lR n and
* (x) * (x,t) lR

respectively.
 46

Therefore,

f *(x) Pt = f(x) o'<t) *(x,t) 2 Kn+l(x,t)

Le. u(x,t) 2 f (x) 0 '(t) * (x,t) Kn+l (x,t).

From this, we have

(0 2 + t,) u (x,t) 2f(x)0'(t) * o(x)o(t) 2f (x) 0 '(t).

t

EXERCISE Show directly that if

i) (t, + a~) u = 0 on lRn+l" R n x {O}

ii) u (x, -t) -u(x,t)

iii) lim u(x,t) ~f(x), then u is a distribution solution

t -+- 0+
of 2 f(x)o' (t).

[TO avoid technicalities, assume f is sufficiently smooth so that

lim au exists; e.g. f € C2 is sufficient].

t-+-O + at

We now indicate, without giving any proof, how these ideas

can be used to solve the Dirichlet Problem in a bounded open set n in

Rn. We assume that the boundary an of n is of class C2 •

We know that the solution of the Dirichlet Problem in the case

n-1
when n = lR x (0, 00) is given by u = f * 2 ~
ax (the convolution
n
is in lRn - 1 ) • We note that oK
ax
is the inward normal derivative of K.
n
2
For bounded with C boundary let us consider

u(x) 2 f fey) a~ K(x-y)do(y) •

an y
 47

Here a{y) is the surface measure on the boundary, V LS

the unit outward normal on em, and a (x, y) -- 1: v. ~

~ a .
Vy J Yj
The minus sign in K{x-y) compensates the switch from

inward to outward normal.

Since ~K{x-y) = o{x-y) we see that ~u = 0 in ~ •

What about the behaviour of u on a~? It turns out that if

u is defined as above, then

u € C{~) and ula~= f + Tf

where T is a compact integral operator on L2{a~) or

C{3~). Hence if we can find ~ in C{a~) such that ~ + T~ f,

and we set

u (x) 2 af ~(y) _3_

3v
y
K{x-y) da{y)

then u satisfies = 0 in and further ul = ~ + T~ = f.

~u ~
an
Hence the Dirichlet problem is reduced to solving the equation

f, and for this purpose, the classical Fredholm-Riesz

theory is available. The upshot is that a solution to the

Dirichlet problem always exists (provided, as we have assumed,

that an is of class
2
C ). See Folland [1] for a detailed

treatment.

§5. THE HEAT OPERATOR

The Heat operator is given by dt - ~. We want to find a

 48

distribution K such that (at - ~)K ~ 6{x)6{t). Taking the

Fourier transform in both variables we have

(2.33)

EXERCISE Show that R is locally integrable near the origin,

and hence defines a tempered distribution.

R is not globally integrable, however; so computing

its inverse Fourier transform directly is a bit tricky. Instead,

if we take the Fourier transform in the variable x only,

We can solve this ODE to get

2 2
a{~) e- 4n I~I t, t > 0
K{~, t) ~ 1 b{~) e- 4n
2 2
I~I t, t < 0

with a (~) - b{~) 1.

As in the previous section, there is some latitude in the

choice of a and b, but the most natural choice is the one which

makes K tempered in t as well as ~ , namely a ~ 1, b ~ o. So,

e-4n21~12t, t > 0
K{~,t)
{

o otherwise.
 49

Taking the inverse Fourier transform,

2
(4n t )-n/2 e-(Ixl /4t), t > 0
K(x,t) ={
0, t < O.

This is a fundamental solution of the heat operator.

REMl.RK 2.34 If we take the Fourier transform of K(~,t)

in the t variable, we obtain

K(~,T) £~(~,t) e-2nitT dt

= fooe-t(4n21~12 + 2niT) dt
o
(4n21~12 + 2niT)-1,

thus recovering formula (2.33).

EXERCISE Show that this really works, i.e., the iterated

Fourier transform of K first in x, then in t, is the

distribution Fourier transform of K in all variables.

OBSERVATIONS From the formula for K, we get

(1). K(x,t) vanishes to infinite order as t tends to 0 when

00 n+l
x f 0 and hence is C on ffi ,{(O,O)}. Therefore by

Theorem 2.11, at - ~ is hypoelliptic.

(2) • K(x,t) = t- n/ 2 K(Tt,l). Thus i f we set K(x,l) = <p (x) ,

2
then K(x,€ ) = €-n <p (x/€) = <p (x); so, as t tends to 0,
€
K(x,t) tends to 6, by Theorem 1. 6.
 50

From this, we infer

(3). If f E LP and if we set u(x,t) = f *(x) K(x,t), then

o for t > 0

u(x,O) = f(x).

i. e., as t tends to 0, II u (. , t) - f II
p
converges to O. Thus

we have solved the initial value problem for the homogeneous heat -

equation, when f E LP . Actually, since K(x,t) decreases so

rapidly as Ixl tends to 00, this works for much wider classes

of f 's. The convolution f * K(. , t) makes sense, for example,

(x)
Ix I 2 - E
if I f (x) I < C e If f is also continuous, it is not hard

to see that f *(x) K(. ,t) converges to f uniformly on compact

sets as t tends to O.

It is now a simple matter to solve the inhomogeneous

initial value problem :

(at - /).) u = F on lR n x (0,00),

u(x,O)= f(x) on

If we take F*(x,t) K and

u2 ;(f - u l (.,0» *(x) K,

then we see that

n
F on lR x (0,00), (at - /).) u 2 o
 51

on IR n x (0, <»), and fIx). Thus

solves the problem.

As another application of the fundamental solution K,

we can derive the weierstrass approximation theorem.

THEOREM 2.35 (WEIERSTRASS) If f is continuous with compact

support on lR n, then, for any compact nC IR n , there exists

a sequence (Pm) of polynomials such that Pm converges to f

uniformly on n.
PROOF Let u(x,t) = (f *(X)K(.,t» (x). Then u(x,t) converges

to fIx) uniformly as t tends to O. Moreover, for any t > 0,

n 2
- E (x.-y.) /4t
n/2 j=l ] ]
u(x,t) =J fly) (4nt)- e dy

is an entire holomorphic function of x E ~n. So u(. ,t) can be

uniformly approximated on any compact set by partial sums of its

Taylor series.

§6. THE WAVE OPERATOR

If we take the Fourier transform of the equation

(a~ - ll)K = o(x) o(t)

in both variables x and t, we have, formally

(2.36)

This function R is not locally integrable, so it is not clear how

 52

to interpret it as a distribution. Again, it is better to take

the Fourier transform in the x variable obtaining

o(t).
Solving this ODE,

and the coefficients a+, b+ must be determined so that

K(~,O+) = K(~,O-), d t K(~,O+) - ' \ K(~,O-) 1.

This givts two equations in four unknowns. In contrast to

the situation with the Dirichlet problem and the heat operator,

there is no way to narrow down the choices further by imposing

growth restrictions on K as It I tends to 00. Rather, it is a

characteristic feature of the wave operator that one can adapt the

choice of fundamental solution to the problem at hand. The two

which we shall use, called K+ and K, are the ones supported in

the half-spaces t ~ 0 and t ~ O. K+ and K are thus determined

by the requirements a = b o and o respectively,

from which one easily sees that

~K sin 211 t
(~,t) = -H(-t) 2111~

where H is the Heaviside function, i.e., the characteristic

 53

function of [0,00). Let us compute the Fourier transforms of

in t to see how to make sense out of (2.36).

are not integrable in t, but it is easy to

approximate them in the distribution topology by integrable

functions whose Fourier transforms we can calculate. Indeed,

if we set

-KE( r t) = -2nEt H(t) sin 2ni(it > 0

+ s' e 2n IE;' E ,

then is an integrable function and converges to K+

in SI as E tends to O. Therefore lim ~E where
E ->- 0
+

2 -1 2 2 -1
(4n) ( IE; I - (l - i E» .

EXERCISE Prove that ~ lim t<E in S I where

E ... 0

Thus we have two distinct ways of making the function

[4n 2(lrs I2 - T' 2 )J-l °

1nto a tempere d dO1S t r1°b u to10n. T
h edlOff erence

~
+
- R- is of course a distribution supported on the cone
 54

We now propose to use the fundamental solutions K+

and K to solve the Initial Value Problem or Cauchy Problem,

for the operator

f on IR n+l

u(x,O)

at u(x,O) = u l (x),
where uO,ul,f are given functions.

For the moment, we assume that uO,u l £ Sand

00
f £ Coo(ffit'S(IR:» (That is, t + ft. ,t) is a C function

with values in S(IR n ».

Taking the Fourier transform in the variable x in (2.37),

When f 0, the general solution of the ODE is

A (f,;) sin 21T1f,;lt + B(f,;) .cos 21T1f,;lt,

nl (f,;)
A (f,;) = 21T1f,;1 and B(f,;)

when 0, the solution is

u(f,;,t) = ~t f(f,;,s) sin(~;I~1 (t-s)) ds

 55

(This may be derived by variation of parameters; in any case

it is easy to check that this u is in fact the solution).

Therefore, the solution for the general case is given by

Ul (~)
u(~, t) GO(~) cos 2TII~lt + 2TII~1 sin 2TII~lt +

+ ~ f{~,s) sin{~~I~1 (t-s» ds.

For t > 0, we can rewrite this as

t
u(~,t) = Ul (~) K+(~,t) + Go(~)atK+(~,t) + ~ f(~,S)K+(~,t-S)ds.

Since K+{~,t-s) = 0 for t < s,

00

Take the inverse Fourier transform

Likewise, for t < 0

So, for arbitra~y t, our solution u can be expressed as

(2.38)
 56

So far, we have avoided the question of computing K+

and K explicitly. Indeed, since K+ and K are not Ll

functions, it is not an easy matter to calculate their inverse

Fourier transform.

However, for the case n = 1, we can find K+ and K

by solving the wave equation directly. We have

If we make the change of variables f,; = x + t, 11 = x - t, then

the wave operator becomes
a2
-4 a f,;d11 • The general solution
a2 u
of
a f,;a 11 = ° is given by

U(f,;,l1) = f(f,;) + gIn)

where f and g are arbitrary functions. Therefore

u(x,t) = fIx + t) + g(x - t).

To solve 0, u(x,O)

must have

fIx) + g(x)

u l (x) = f' (x) - g (x) • I

From these equations, we have

1
f I (x)
°
-(u'(x) +ul(x»
2
1
2 °
g' (x) -lU' (x) - ul(x».
 57

Thus u(x,t) is given by

x+t
u(x,t) ~(UO(x + t) + uO(x - t)) + ~ J u l (s) ds.
x-t

Comparing with the previous formula (2.38), we find that

EXERCISE Compute K+ directly from these formulas.

It turns out that, for n = 2,

1
H(:!:t - Ixl),

and, for n = 3,
+1
41ft 0 (.:!:t - Ix I ) .

For n = 1,2,K+,K_ are functions; for n = 3,K+,K_ are not

functions but they are measures. For n > 3,K+,K_ are neither

functions nor measures; they are more singular distributions.

The exact formula for K+ is rather messy and we shall not

write it out; it may be read off from Theorem 5.13 and 5.14 of

Folland [1] in view of our formula (2.38). The most important

qualitative feature of K+, however, is that it is always

supported in the light cone {(x,t): .:!: t > Ixl} and we shall now

prove this as a consequence of the following result.

THEOREM 2.39 Suppose u function on {(x,t): t ~ O}

 58

such that (a~ - A) u 0 for t > 0 and u = a

t
u = 0 on

the set BO={(X,O) : Ix - xol~ to}· Then u vanishes on

Q = {(x,t) : o ~ t ~ to' Ix' - Xo I ~ t - t}.

0

PROOF Assume u is real valued; otherwise, we can consider

the real and imaginary parts separately. Let

and E(t) = ~ f Igrad t ul 2 dx

B x,
t

i.e. E (t)

Then
n
dE 1
f 2[a u a2u + L au aax.at]
2u
dx -
dt 2 B at dt 2 1 aX j
t J

n
1
f [ (au)
at
2 + L (au ) 2] do(x).
23B 1 dXj
t

(The second term comes from the change in the region Bt . If this

is not clear, write the derivative as a limit of difference quotients

and work it out).

Since
2 2
au au +Iau au au
div (at grad x u) ,
I ax. (l'X.at ax.2 at
J J
J

applying the divergence theorem and using «(12 - A)u 0, we obtain

t

~ f [au au - 1 Igrad t u1 2 ] do
dt aB at a~ 2 x,
t
 59

where V is the unit normal to Bt in IRn. Now

Iat
OU au/ < ![/au/2 + /dU/2]
av 2 at av

1 2
= 2 /gradx,t u/ •

dE
Thus we see that the integrand is non-positive, and hence dt ~ o.
Also E(O) =0 since u = at u =0 on BO' so E(t) S O. But

E(t) ~ 0 by definition, so E(t) O. This implies that grad tU

x,
= 0 on ~ and since u =0 on BO' we conclude

that u =0
2
CORO LlARY 2.40 Suppose u £ C on IR n x [0,00), (a~ - fl)u =0

then supp u c:. ~ {(x,t): d(x,~O) ~ t}.

(The set ~ is the union of the forward light cones with

vertices in ~O).

PROOF Suppose (xo,t O) ~ ~. Then for some £ > 0, the set

BO = {x: d(x,x O) S t~ + £} is disjoint from ~o.

Therefore, by Theorem 2.39, u =0 on the cone

In particular, u =0 on a neighbourhood of (xo,t o )'

i.e. (xo,t O) is not in the support of u.

 60

CORO LIARY 2.41

Supp K_C{(x,t):-t ~ Ixl}.

00
PROOF Pick a ¢ E CO(B(O,l)) such that J¢ = 1. Put

-n -1
¢ (x) = E ¢(E x). Let u (x,t) =¢ *() K , t > O.
E E E x +

¢
00
Then 0, at u(x,O) (x) and u is C .
E E

By the previous corollary

suPpuEC{(x,t): Ixl ~t+d.

Now uE converges to K+ in S', as E tends to O. Therefore,

supp K+ C {(x,t): Ixl ~ t}.

The result for K follows then since K_(x,t) K+(X,-t).

REW>.RKS 2.42

(i) One could also deduce the above result from our formulas

for K+ by using the Paley-Wiener theorem.

(ii) Actually for n = 3,5,7, .•. , it turns out that

supp K+ = {(x,t): Ixl ~ t}. This is known as the Huygens

principle. See Folland [1].

(iii) The distributions K+ are smooth functions of t (except

at t 0) with values in E'(lR n ). Therefore, we now see that

our formula (2.38) for the solution of the Cauchy problem makes

sense even when and f E C (lR t ' V '( m~ ) ), and it

 61

is easily checked that u thus defined still solves th~ Cauchy

problem in the sense of distributions. Corollary 2.40 remains

valid in this more general setting, as can seen by an approximat-

ion argument as in the proof of Corollary 2.41.

n +1
EXERCISE If (a 2 -ll.)u=f on IR , u (x,O) = U o (x) ,. and
t

u l (x), figure out how supp u is related to supp f,

supp U o and supp ul.

 62

CHAPTER 3

L2 SOBOLEV SPACES

THERE ARE MruNY ways of measuring smoothness properties

of functions in terms of various norms. Often it is convenient

to use L2 norms, since L2 interacts nicely with the Fourier

transforms. In this chapter, we set up a precise theory of L2

differentiability and use it to prove Hormander 's theorem on

the hypoellipticity of constant coefficient differential

operators.

§l. GENERA L THEORY OF SOBO lEV SPACES

DEFINITION 3.1 For a non-negative integer k, the Sobolev

space Hk is defined to be the space of all tempered distri-

butions all of whose derivatives of order less than or equal

2
to k are in L •

Thus

From the definition, we note that f € Hk if and only if

~af(~) € L2 for 0 'S.. lal < k.

PROPOSITION 3.2 f € Hk if and only if (1 + 1~12)k/2 f € L2.

PROOF First assume that (1 + 1~12)k/2 f € L2. Since, for I~I ? 1,

 63

and for I~I < 1,

we have

and hence

~ a f
A
£ L2 for Ia I ~ k which implies f £ Hk .
n
Conversely, assume that f £ Hk . Since 1~lk and j~ll~jlk
are homogeneous of degree k and nonvanishing for ~ ~ 0,

we have

so that

The characterisation of Hk given in the above proposition

immediately suggests a generalisation to non-integral values

of k which turns out to be very useful.

DEFINITION 3.3 For s £ m, we define the operator AS S+ Sby

 64

s D. s/2
In other words, A = (I
• Clearly maps S
-~)
4n
continuously onto itself. We can therefore extend AS

continuously from S' onto itself.

The Sobolev space of order s is defined by

We equip H with the norm s is

s
a positive integer~ the proof of Proposition 3.2 shows that

II II(s) is equivalent to the norm

PROPERTIES OF H
s

(i) H is a Hilbert space with the scalar product defined

s
by (u,v) (s) = (Asu,Asv). Here the scalar product on the right

is that in L2. The Fourier transform is a unitary isomorphism

between H and the space of functions which are square

s
integrable with respect to the measure (1 + 1~12)sd~.

(ii) For every S E: lR, S is dense in H.

s

(iii) If s > t, Hs~ Ht with continuous imbedding. In fact,

for u E: Hs' lIu II (t) ~ Ilu II (s)· In particular, Hs Co L2 for

s > o.
 65

(iv) Da is a bounded operator from H into H5- Ia I

5

5 E: m.

(v) If f E: H_ s ' then f as a linear functional on S

extends continuously to Hs and II f 11(-5) is the norm of

f in (H ) *
5
. So we can identify (H ) *
5
with H
-5
For

f E: H , 9 E: H the pairing is given by

-5 5
-5 5 A A
< f,g > = <A f,A g> = If g.

(If 5 = 0, this identification of with its dual

is the complex conjugate of the usual one).

(vi) The norm 11·11 (5) is translation invariant. Indeed,

~ 2TIixo'~ A
if g(x) f(x - x O)' then g(~) = e f(~) and hence

II f II (5) II 9 II (5) •
PROPOSITION 3.4 For 5 > n/2, we have o E: H-5

2 -5/2 2
(1 + 1~)
1 E: L , whenever 5 > n/2.

This is a consequence of the following observation :

2
1(1 + I~I )- 5 d~ ~ 1 +
00
I r
-25
r
n-l
dr < 00 if and only
1
if 5 > n/2.

As an immediate consequence of this proposition, we have

CORO LIARY 3. 5 For

 66

n
THEOREM 3.6 (SOBOIEV IMlEOOING THEOREM) For s > "2 + k,

H c; Ck . Further, we have
s

PROOF Since S is dense in Hs' it suffices to prove (3.7)

for f e: S. Let Ox denote the Dirac measure at x. For

n
ICt I ~ k, since s > "2 + ICt I ' oCto x e: H-s

Since

and II oCto x II is independent of x,

(-s)

L sup
ICt\~k lR n

L sup
s. ICtI::;k IRn
IloCt o
x
11(_ jlfll( )=c kllfll()
s s s s

Now, given u e: H , choose a sequence (u . ) in S such that

s J
II u - u j 1\ (s) converges to 0, as j tends to 00 . The above

inequality with f=u.-U. shows that is a Cauchy

1 J
sequence in the uniform norm for ICt\ ~ k; so its limit oCtu is

continuous.

00
CORO LIARY 3. 8 If u e: H for all s e: lR, then u e: C i . e. ,
s

This argument can be extended to show that i f s > 2n + k,

then elements of H and their derivatives of order less than or

s
 67

equal to k are not merely continuous but actually Holder

continuous.

PROPOSITION 3.9 II O<a<l s = "2n + a, then

PROOF We have

Let R be a positive number, to be fixed later. When 1;1 ~ R

we use the estimate Ie -2nix.; -e -2n i Y';1 ~ 2nl;1 Ix-yl (by the

mean value theorem) and when I; I > R, we use I e -2nix.; -e -2TIiY'~1 ~ 2 .

Then we have

R CD
2 f (1+r2) -s r n +l dr + f r -2s+n-l dr]
~ c[lx-yl
o R

~ c'[lx- yI 2 R- 2s +n +2 + R- 2s +n ] as ~ < s <~ + 1

-1
When we take R = Ix-yl we get our result.

EXERCISE Show that the above argument does not work when a = 1.
Instead, we get lIox - 0y II (-s) ~ c Ix - Yilloglx - y 111/2 when x

is near y. What happens when a > l?

 68

CORO LIARY 3. 10 Let o < a < 1

-and Aa = {bounded functions
g : sup Ig (x) - g (y) I < co. If s = 2n + a + k and f € Hs' then
x,y Ix _ yla
DSf € Aa for lsi ~ k.

RE~RK 3.11 We shall obtain an analogue of this result for

UP norms in Chapter 5.

The following lemma will be used in several arguments

hereafter.

LEMMA 3.12 For all ~,n € mn and s € m, we have

PROOF I~I < Inl + I~ - nl gives

th
If s > 0, then raise both sides to the s power. If s < 0,
th
interchange ~ and n and raise to the -s power.

PROPOSITION 3.13 If ~ € S, then the operator f ~ ~f is bounded

for all s.

PROOF The operator f ~ ~f is bounded on H if and only if the

s
operator of g ~ As ~ A-s g is bounded on L2, as one sees by
 69

setting g = l\ s f. But

(1+isi 2 )s/2 (¢l\-Sg)~(S)

(l+isI 2 )S/2 [~(S) * (l\-Sg) ..... (S)]

(l+lsI2)S/2 J ~(S-1l)(1+11l12)-S/2 9(1l)dll

J gIll) K(S,ll)dll

where

By lemma 3.12,

,.,
Therefore, since ¢ is rapidly decreasing at 00,

J IK(S,ll)I d s ~ c for every II E: lR n ,

J IK(s' ll) I dll ~ c for every S E: Rn •

Thus from Theorem 1.1, the operator with kernel K is bounded on

L2. Hence our proposition is proved.

n
The spaces H are defined on IR globally by means of
s
the Fourier transform. Frequently, it is more appropriate to consider

the following versions of these spaces.

DEFINITION 3.14 I f stCIR n is oEen and s E: IR, we define

H1oc (m = {f E: V, (m : V st'@st, 3 gst' E: H such that

s s

gst' = f on st' }.
 70

PROPOSITION 3.15
Q)

(j) e: Co (m·
00
If f e: Hloc(m and (j) e: cO(n), then there exists
s
g e: H
s
such that f =g on supp~. Therefore ~f = ~g e: Hs
by proposition 3.13.

Conversely, if (j)f e: H for all

s
choose with (j) =1 on n'. Then ~f e: Hs and f = ~f
on n'.

COROLIARY 3.16 If L = L a (X)D a with a

ex
£ C
00
(n), then L
lal ~ k a
into H~~~(n) for all s £ IR.

It is a consequence of the Arzela-Ascoli theorem that if

(u j ) is a sequence of Ck functions such that lujl and laaujl

(Iexl ~ k) are bounded on compact sets uniformly in j, there

exists a subsequence ( v~) of (u) such that (aav .) converges

J j J
uniformly on compact sets for lal ~ k-l. In particular, if the

u. 's are supported in a common compact set, then (aav .) converges

J J
uniformly.

There is an analogue of this result for Hs spaces.

00
IEMW. 3.17 Suppose is a sequence of C functions supported

in a fixed compact set n such that s~p II uk II (s) < 00. Then there

exists a subsequence which converges in the Ht norm for all t < s.

 71

Co n
00
PROOF Pick a <I> £ such that <I> = 1 on so that uk = <l>u k

and hence
....
uk = ...~ * ....uk. Then

u
(1+1E,;1 2 )s/2I k (E,;)I = (l+1E,;12)s/21 J $ (E,;-n) Gk(n)dnl

~ J I $(E,;-n) lluk(n) 12 I sl/2(1+lnI2)s/2(l+IE,;_nI2) I s l/2dn

~ 21sl/211 <1>11 (lsi) Ii u k II (s) ~ cl independent of k.

Likewise, we have

independently of k. Therefore, by the Arzela-Ascoli theorem, there

exists a subsequence (v k ) of (uk) which converges uniformly on

compact sets. For t < s,

Ilvj-v k II~t) = J (l+1E,;12)tIOj-OkI2dE,;

J (1+1E,;12)tIO.-O 12dE,; + J (1+1E,;12)tIO.-O 12dE,;

IE,;I~ R J k lE,;I > R J k

< (1 +R2)max (t,O) sup I.......

v.(E,;)-vk(E,;) 12 J dE,; +
lE,;I ~ R J lE,;I ~ R

~ c(l+R 2 ) n+ I t I sup lv.

A(E,;)-vk(E,;)
o,. 12 +(l+R)
2 t-s II v,-v 112( )
k •
IE,;I~ R J J s

Given £ > D, choose R large enough so that the second term is less

than £/2 for all j and k. This is possible since Ilvj-v k II (s)~ c
 72

and t - s < 0. Then for j and k large enough the first term

is less than E/2 , since (~k) converges uniformly on compact

sets. Thus we see that (v k) is a Cauchy sequence in Ht , and

since Ht is complete we are done.

REWoRK 3.18 Lemma 3.17 is false, if we do ~ot assume that all

the u 's have support in a fixed compact set. For example, for
k
00
u € Co and xk E mn with IXkl tending to 00, define

uk(x) = u(x - x k )· Then the invariance of H

s
norms shows that

II Uk II (s) =llull(s)· But no subsequence of (uk) converges in

any Ht . For, if a subsequence (v k ) of (uk) converges, it must

converge to 0, since Uk converges to ° in S'. But then

lim Ilvk II (t) = ° which is not the case.

THEOREM 3.19 (RELUCH THEOREM) Let HO (m be the closure of

s
00
Co (m in H If Sl is bounded and t < s, the inclusion
s
HO{m <.+ Ht is compact, Le. , bounded sets in HO(m are relatively
s s
comEact in Ht ·

PROOF be a bounded sequence in °

Hs(Sl). To each k,
1
find a such that II Uk - v k II (s) ~ "k. Then we have

Ilvk II (s) ~ Ilu k II (s) + 1 ~ c (in·dependent of k).

Therefore, by Lemma 3.17, a subsequence (w k ) of (v k ) exists

such that (w k ) converges in Ht . If (Uk) is the subsequence of

 73

(Uk) corresponding to the sequence (w k ), we have

1
< i + .... 0 as i ,j .... (X).

Hence (u~) converges in Ht .

In the proof of the next theorem, we will use the

technique of complex interpolation, which is based on the

following result from elementary complex analysis called 'Three

lines lemma'.

LEMMA 3.20 Suppose F(z) is analytic in 0 < Re z < 1,

continuous and bounded on 0 < Re z ~ 1. If iF(it) I ~ Co and

I-s s
IF(l+it) I ~ c l ' then iF(S+it) i ~ Co c l ' for 0 < s < 1.

PROOF If € > 0, the function

g (z)
€

satisfies the hypotheses with Co and cl replaced by 1, and

also i g€ (z) I converges to 0 as i 1m z i .... (X) for o< Re z < 1.

From the maximum modulus principle, it follows that Ig (z) I ~ 1

for o ~ Re z ~ 1 and letting € tend to 0, we obtain the desired

result.

THEORE M 3.21 Suppose that -(X) < So < sl < (X) and T is a bounded

linear oEerator on H such that TIH is bounded on Hs . Then

So sl 1
TIH is bounded on H for all. s with So S s sl.
s s ~
 74

PROOF Our hypothesis means that

51 -5
and A TAl

2
ar.e bounded operators on L • For 0 ~ Re z ~ 1 we define
5 -5
and T = A z T A z.
Z

Then what we wish to prove is that T is bounded on for

Z

o S Z S 1. Observe that when w = x + iy, AW = AX Aiy and

Thus Aiy is unitary on H for all s.

5

For ~, * £ S ,we define

5 -5
Fez) =J (T~)*=<AZTA z~,*>.
Z

Then
5 -5
IF (z) I 1<11. Z T A z~, * >1
-5 5
I<TA Z <1>, II. Z * > I
-5 5

~ II Til. Z ~11(s ) 1111. z* II (-5 )

-5
o 5
0

~ c II A Z <1>11(5 0 ) 1111. Z II (-sO)

~ c II ~ II (5 0 - 51) Re Z II * II (5 1- sO) Re Z

~ c II <I> II (0) II * II (5 -5 )
1 0

F(z) is clearly an analytic function of z for

 75

o < Re z < 1. Further, by our hypothesis on T, when Re z 0,

we have

IF(z) I ~ Co 114> II (0) III!! II (0)

and when Re z = 1, we have

IF(z) I ~ c l 114> II (0) III!! II (0)'

Therefore, by the Three lines lemma

l-z z
IF(z) I ~ Co c l 114> II (0) III!! II (0) for 0 < z < 1.

Finally, by the self duality of HO = L2, this gives

which completes the proof.

RE~RK 3.22 The same proof also yields the following more general

result

Suppose -00 < So < sl < 00, -00 < to < tl < 00. If TO is a

bounded linear operator from H to Ht whose restriction to

So 0
H is bounded from H to Ht ' then the restr iction of T to
sl sl 1
H is bounded from H to H for o < 8 < 1 where
s8 s6 t8
s = (1-6)sO + 8s 1 and t6 (1-6) to + 8t l ·
e
As a consequence of this result, we obtain an easy proof

that Hloc is invariant under smooth coordinate changes.

s

THEOREM 3.23 suppose Q and Q' are open subsets of mn and

 76

00
~: ~ +~' is a C diffeomorphism. Then the mapping f + fo ~

HloCW' ) continuously onto H10C(Q) .•

s s

The statement of the theorem is equivalent to the assertion

Co W ) , the map Tf = (<pf)

00 ,
that for any <P E 0 ~ is boul)ded on H
s
for s E IR. I f s = 0,1,2, •.. , this follows easily from the chain
rule and the fact that H = if: Daf E L2 for lal ~ s}. By Theorem
s
3.21, it is true for all s > O. But the adjoint of T is another

map of the same form:

T*g
-1
= (\(Ig) 0 '¥ where '¥ = ~ and \(I = <p IJI <> ~, J being

the Jacobian determinant of ~

-1
. Hence T* is bounded on H
s
for

all s ~ 0 and by duali ty of H and H , this yields the bound-

s -s
edness of T on H for s < O.
s

Finally, we ask to what extent the H spaces include all

s
distributions. Globally they do not, since, if f E Hs' then f is
,..
tempered and f is a function. But locally they do, as we see from

the following result.

PROPOSITION 3.24 Every distribution with compact support lies in

some H : Le. E' c:

s U H
s
.
s E IR
00
PROOF If f E E' , then it is a continuous linear functional on C

Therefore, there exist a constant c > 0, a compact set K, and a

nonnegative integer k such that

00
1< f,<p > 1 < for all <p E C ,
 77

L
00
i.e. 1< f,¢ >1 ~ c sUP. IDa¢1 for all ¢ € C .
lal ~ k IR n

By the Sobolev imbedding theorem,

Therefore, I< f, ¢ > I ~ e" II ¢ II for all ¢ € S. Since S

(k~+€l
is dense in this shows that f is a continuous linear

functional on Hence

COROLIARY 3.25 If f € V, (m and rG' has compact closure in rG,

then there exists s in lR such that f € Hloc m') .

s

§2. HYPOELLIPTIC OPERATORS WITH CONSTANT COEFFICIENTS

We now apply the machinery of Sobolev spaces to derive a

criterion for the hypoellipticity of constant coefficient differential

operators. First, we have a few preliminaries.

DEFINITION 3.26 Let P be a polynomial in n variables. For a

multi-index a, p(a) will be defined by pea) (~) = (:~)ap(~).

PROPOSITION 3.27 (LEIBNIZ RULE) When f € Coo, 9 € V, and P(D)

is a constant-coefficient partial differential operator of order k,

we have

P(D) (fg) I (P (al (D) g) (Oaf) .

k
a'.

The proof of this proposition is left as an exercise to the reader.

 78

DEFINITION 3.28 We say that a polynomial P satisfies condition

(H) if there exists a 0 > 0 such that

~I=
I P (~) I 0 (IC-I- olal )
<, ,
as I~ I -+- 00, V a.

THEOREM 3.29 (HORMP.NDER) If P satisfies condition (H), then

P(D) is hypoelliptic. Mbre precisely, if f is in V, (n) and

loc loc
P(D) f E H
s
(n), then
-- f E Hs +ko (m, where 0 is as in condition
(H) and k is the degree of P.

PROOF We first observe that the second assertion implies the

first since
00
C W) =
s
n E lR
HlocW)
s
by Corollary 3.8.

Suppose therefore that P(D) f e: Hloc(S'a) f e: V, (£1) • Given

s '
00
~ e: C0 (m, we have to prove that <I> f E Hs+ko. Let n' be an open

set such that n' ~ nand supp <l>c: n ' By Corollary 3.25, there

exists t in lR such that f E H!OC w '). By decreasing t, we can

assume that t = s + k - 1- mo for some positive integer m. Set

<l>m <I> and then choose <l>m-l' <l>m-2 ' ... , <1>0' <1>-1 in Coo(n ')
0
such that

<1>.
J
1 on supp <l>j+l .
Then <l>jP(D) f E HsC: Ht-k+l+jo for o~ j < m and

<1>-1 f E Ht • Now

P(D) (<I> f) <I> P(D)f + J. ~ pta) (D) (<I> f)Da<l>

o oaf 0 a. -1 0

since 1 on the support of <I> • This

o
means that
 79

By condition (H),

J (1 + < 00

This implies that

Next,

since ¢o = 1 on the support of ¢l' so P(D) (¢If) E Ht - k+l +o .

By the same argument as above,

Continuing inductively, we obtain P(D) (~ f) E H which

~j t-k+l+jo '
implies that

For j m, we have

If

choose a with lal = k such that a

a
f o. Then a '. a
a
f 0,

and we are done.

 80

REMA.RK 3.30 The condition (H) is equivalent to the following

apparently weaker condition

(H ') I~I ~ 00 for a + o.

Condition (H ') is in turn equivalent to

(H") : 11m ~I ~ 00, as I~I ~ 00 in the set {~e:«:n : P(~) OJ.

The converse of Hormander 's theorem is also true, i.e.,

hypoellipticity implies condition (H).

The proofs of these assertions can be found in Hormander [6J.

The logical order of the proofs is

(H) ~ hypoellipUcity .=;> (H") => (H ') => (H) •

The implication (H ') ~ (H) requires the use of some results from

(semi) algebraic geometry.

DEFINITION 3.31 P(~) = I a ~a is called elliptic if

lal ~ k a
~ aa~a +0 for every ~ f O.
lal =k

EXERCISES

1. Prove that P is elliptic if and only if Ip(~) I > c 1~lk

o
for

large I~I

2. Prove that P is elliptic if and only if P satisfies condition

(H) with 0 = 1.
3. Prove that no P satisfies condition (H) with o> 1.

(Hint: If lal = k, pea) is a constant).

 81

4. Let P be elliptic and real valued. Define Q on :R n+l by

Q(~, T) = 21TiT + P(~), so that Q(D x ' Dt ) at + P (0 ).

x
Show that

Q satisfies condition (H) with <5 = 11k where k is the degree

of P and that 11k is the best possible value of <5

k
(Hint: Consider the regions 1~lk ~ ITI and ITI ~ I~I separately).
 82

CHAPTER 4

BASIC THEORY OF PSEUDO DIFFERENTIAL OPERATORS

§l. REPRESENTATION OF PSEUDO DIFFERENTIAL OPERATORS

Let L = L be a partial differential operator

.lal ~ k
(X)
with C coefficients on ~. Using the Fourier transform, we can

write

(Lu) (x)

where p(x,~) = L a (x)~a. This representation suggests that

lal ~ k a
p(x,~) can be replaced by more general functions. So, we make the

following definition.

DEFINITION 4.1 For an open set ~ C:R n and a real number m we

define Sm(~), the class of symbols of order m on ~ , by

such that sup ID~D~P(X'~) I ~ c(l+I~I)m-lal} .

x E ~' <.,

We note that Sm(~) C sm' (~) whenever m < m' , and we set

W) n
-(X)

S
m E :R

EXAMPLES
k
(m. (m.
(X)
(il a E C Then PES
a
 83
N
(ii) Let p(x,l:) = I a.(x)f.(I:)
J J
where a. e: C
J
00
m) and
j=l
f. e: Coo(lRn ) is homogeneous of degree m. for large I:
J J
that is,
m.
f. (rl:) r J f . (I:) for II: I .? c, r .? 1.
J J

m
In this case, p e: S (~), where m =

(iii) Let p(x,l:)

(iv) ~t p(x,l:) = ~(I:) sin log 11:1 with near the

origin and ~ = 1 when II: I .? 1. Then p e: sO (lRn ) .

RE~RK 4.2 We observe that when p e: Sm(~) , 0 8x D~P

s
e: sm-Ial (~).
m m2
Further, if p e: S l(~) and q e: S (~), then p+q e: Sm(~), where
ml +m 2
m = max{m l ,m 2 } and pq e: S (~).

RE~RK 4.3 Our symbol classes Sm(~) are special cases of Hormander's

classes S
m
p,\).r(m.
Namely, for ° ~ 0 .s p .s 1, and m e: lR,

S
m .r (m = {p e: C
o on....
(~x lR ) : Y a , 8,
\J ~ , «:: ~ ,
p,\)

In this terminology,

DEFINITION 4.4 For p e: Sm(~), we define the operator p(x,D) on

00
the domain C (~) by
o

(Sometimes, we shall denote p(x,D) by pl. Operators of the form

 84

p(x,D) with p £ Sm(D} are called pseudo differential operators of

order m on D.

The set of all pseudo differential operators of order m

m
on D will be denoted by 0/ (D). For brevity, we will sometimes write

"o/DO" instead of "pseudo differential operators".

The next theorem states that p(x,D) is a continuous linear

map of For the proof,

we need a result which depends on

m
¢l£C (m.
00
IEMM\ 4.5 Let P £ S (m and Then, for each positive
o
integer N, there exists cN > a such that for all ~, n in lRn,

PROOF For any ~ and n in Rn

=11 D~ e2TIix·n p(x,~) ¢l(x)dxl •

Integrating by parts, we have

Ina f e2TIix·~p(X'~) ¢lex) dxl

If e2TIix·nDa(p(X'~) ¢lex»~ dxl

x

~ C (1 +1~I)m for all a.

a

Therefore
 85

Since (1 + Inl)N < c L Inal, the required result follows.

lal ~ N
co
CORO LlARY 4. 6 and ~ € Co(Q), then the function

g(~) = J e2TIix.~ p(x,~)~(x) dx

is rapidly decreasing as ~ tends to co.

Set ~ = n in the lemma.

THEOREM 4.7 If P € Sm(Q), then p(x,D) is a continuous linear

map of CCO(Q) into Cco(Q) which can be extended as a linear map from
o
f '(Q) into V '(Q).

co
PROOF For u € C (Q),
o

p (x, D) u ()
x = f e 2TIix·~p(x,~)
<,
u (~)d~
.... <, <,

The integral converges absolutely and uniformly on compact sets, as

do the integrals

for all a

since p € Sm and G£ S .
co
This proves that p(x,D)u £ C (Q), and the continuity of
co co
p(x,D) from C (Q) to C (Q) is an easy exercise.
o

To prove the rest of the theorem, we will make use of Corollary

4.6.
co
For u € f ' (Q), we define p(x,D)U as a functional on C (Q)
o
as follows:

< p(x,D)u,<j> >

 86

where

g(;) 1 e 21Tix'; p(x,;)~(x)dx, for ~ €

co
C (n).
o

,.,
By Corollary 4.6, g(;) is rapidly decreasing, while u is of
v
polynomial growth; so the last ~ntegral is absolutely convergent
00
and the functional on C thus defined is easily seen to be
o
00
continuous. M:>reover, if U € C , the double integral is absolutely
o
convergent, and by interchanging the order of integration, we see

that this definition of p(x,D)u coincides with the original one.

Hence we have extended p(x,D) to a map from E 'en) to V'(n).

REWl.RK 4.8 It follows easily from the above argument that p(x,D)

is sequentially continuous from E 'en) to V '(n), i.e., if uk

converges to u in E'(n), then p(x,D) uk converges to p(x,D)u

in V'(n). Actually, p(x,D) is continuous from E 'en) to V'(n)

this follows from the fact (which we shall prove later) that the

transpose of a pseudo differential operator is again a pseudo diffe~ential

t 00 00
oper ator. Thus p(x,D) : C (n) + C (n) is continuous, and so, by
o
duality, p(x,D) = (p(x,D) t) t : E '(m + V, (m is continuous.

§2. DISTRIBUTION KERNElS AND THE PSEUDO IDCAL PROPERTY

DEFINITION 4.9 Let T be an operator from C

00

o
(m to C
00
(m. If

there exists a distribution K on n x n such that

< Tu, v > = < K, v ® u >

00
for u,v € C (n),
o

we say that K is the distribution kernel of the operator T.

In this definition, v ® L1 is defined by (v ® u) (x, y) = v (x) u (y) .

Formally, this definition says that
 87

Tu(x) = f K(x,y) u(y)dy.

K is uniquely determined since linear combinations of functions of

00
the form v® u are dense in C (n x n).
o

If P € Sm(n), it is easy to compute the distribution

kernel of p(x,D). In fact,

If p(x,~) e
2TIix·~
(v ® u) 2
A
(x,~)d~ dx

where (v ~u); (x,~) means the Fourier transform in the second variable.

It follows immediately from the definition of K that

< K,w >

or < K,w > fIf e 2TIi(x-y)·~ p(x,~)w(x,y)dy d~ dx.

From this it is easy to see that K(x,y) = p~(x, x-y) where p~(x,.)

is the inverse Fourier transform of the tempered distr~ution p(x,.).

In particular, this shows that p is uniquely determined by the

operator p(x,D). If P € ~m(n), we shall sometimes denote the unique

p(x,D) by Op.

The following theorem gives precise results on the kernel K

of p(x,D).

THEOREM 4.10 The distribution kernel K of p(x,D) with

m
p € s W) is in C
00
on m x m' t, where t, {(x,x): x € n} is the

diagonal. lobre precisely, if lal > m+n+j for a positive integer j,

then (x_y)(l K(X,y) € cj(n x n).

 88

co a
PROOF For w £ C (n x n), let us compute < (x-y) K, w> •
o

< (x-y) a K, w > < K, (x-y) a w >

Ife2nix·~p(X,~) (X+D~)a w;(x,~)d~ dx

If w2(x,~) (X-D~) a
A { 21Tix'~
e
.
p(x,~)}d~ dx.

Using the Leibniz.formula, it is easily seen that

Therefore

< (x-y) a K,w > fJw;(x,~) e21Tix'~(_ D~)a p(x,~)d~ dx

ffIw(x,y) e21Ti(x-y).~(_ D~)a p(x,~)dy d~ dx.

From the above expression, we infer that

(x_y)a K(x,y) = f e21Ti(x-y).~(_ D~)a p(x,~)d~

Since I (- D~)a p(x.~) I ~ c(l +1~I)m-lal , the integral converges

absolutely and uniformly on compact sets whenever m -10.1 < - n.

Also we can differentiate with respect to x and y j times under

the integral sign provided m - 10.1+ j < - n or 10.1, > m + n + j. Thus,

we see that (x_y)a K £ Cj ( n x n).

CORO LIARY 4. 11 (PSEUDO IDCA L PROPERTY OF '1'DO). If P £ ~ (m, then,

for all u £ E'(n), sing supp Pu is contained in sing supp u.

Let u £ E' and let V be an arbitrary neighbourhood of

co
sing supp u. Take a ~ £ Co(V) such that ~ = 1 on sing supp u. Then
co
U = ~u + (1 - ~)u = u1 + u 2;u 2 is a Co function and supp u1 = V.
'co
Therefore, Pu = PU 1 + pU 2 and pU 2 is a C function. Moreover, when
 89

x ~ V,
o

J K(x,y) u l (y)dy
V

o
00
is also a C function in a neighbourhood of x since
o
for y near x • This implies that sing supp Pu C V. Since V is
o
any neighbourhood of sing supp u, the corollary is proved.

COROLLARY 4.12 If P € S-OO(Q), then the distribution kernel K

00
of p(x,D) is in C (QxQ).

PROOF Follows from the theorem if we take lal = o.

-00
CORO IIARY 4. 13 If P € S (Q), then p(x,D) maps E '(Q) continuously
00
into C (Q).

PROOF If u € E'(~), in view of Corollary 4.12, it is easily seen

that p(x,D)u is a smooth function defined by

p(x,D)u(x) = I K(x,y) u(y) dy < u, K(x,.) >,

whence the result follows.

DEFINITION 4.14 A smoothing operator is a linear operator T which

00
maps E '(Q) continuously into C (Q).

00
If K € C (Q x Q), then the operator T defined by

(Tf) (x) < K(x,.), f > = I K(x,y)f(y)dy

is a smoothing operator. Conversely, every smoothing operator T can

be given in the above form with K(x,y) = (To) (x).

y
-00
As we have already remarked if P € S ,then the corresponding

~DO is smoothing. However, not every smoothing operator is a ~DO.

For example, we have

 90

00
PROPOSITION 4.15 Suppose p(x,.) is a C function of x with

values in E I Then p(x,D) (defined in the same way as in the case

of P € Sm(n» is a smoothing operator.

PROOF The distribution kernel K of p(x,D) is given by

K(x,y) = J e2ni(x-y).~ p(x,~)d~

< p (x,. ) , e 2ni(x-y).(.) >

00
and hence K(x,y) € C. Thus K defines a smoothing operator.

REMl.RK 4.16 Sometimes, it is convenient to enlarge the class of

pseudo-differential operators of order m by including operators of

the form P + S where P € ~(n) and S is smoothing. However, the

general philosophy is the following:

1. On the level of operators, smoothing operators are negligible.

2. On the level of symbols, what counts is the asymptotic behaviour

at 00, so that symbols of order - 00 are negligible.

§3. ASYMPTOTIC EXPANSIONS OF SYffiOIS

DEFINITION 4.17 mO > ml > m2 >

Suppose
m. € R, m. tends to ...
] 00 ]
m· m
-00, and Pj € S ] (m, p € S om). We say that P'" p. if I
m j=O ]
p- I p. € S k(m for all k.
j < k ]
m'
PROPOSITION 4.18 Suppose m. tends to -00 and p. € S ] (n).
J ooJ
Then there exists a p in sffio(m such that P'" 2
j=O
Pj This p
-00
is unigue modulo S (n).

PROOF Let mn ) be an increasing sequence of compact subsets of Q

00
whose union is n Fix <P € C with <P = 1 for I~I ~ 1 and <P = 0 for

t~ I ~
1
2
.
 91

There exists a sequence (t.), t. ~ 0 and t. tending to

J J J
00 so rapidly that we have

ID
x
SD~(cjl(Vt.)P.
'0 J J
(x,E;,» I for

and lal + lsi + i ~ j.

Granted this, we define

co

L cjl(Vt.) p.(x,E;,).
j=O J J

Note that for each E;"the sum is finite. Using (4.19), it is

x and
m 00

straightforward to show that p e: S oem and p '\, L p . • lobreover,

m 00 j=O J
suppose that q e: 5 0 (m and q '\, L Pj Then
j=O
mk
p - q = (p - p.) - (q - I p.) e: S em for all k. l.
j < k J j < k J

-00
Hence p - q e: 5

We now briefly indicate the steps involved in proving the

claim:

i) First observe that ID~~~tj) I ~ clE;,I-lvl uniformly in j.

m. - lal
ii) Hence we have IDS
x
D~(cjl(Vt.)P.
J J
(x,E;,» I ~ cj(l +IE;,I)
J for

x e: Qi , lal + lsi + i ~ j.

iii) Finally, pick t. so large that

J

Details are left to the reader.

The following theorem provides a useful criterion for the

00

asymptotic relation p '\, I p. to hold.

j=O J
 92

m'
THEORE M 4. 20 Suppose p j ,..c. S ] ' m
j
tends to - 00 and

P E: coo(n x Rn) satisfies the following conditions:

i) for all a and S and all n' c1;: n , there exist c > 0,

].l E: lR such that

ii) there exists a seguence (].lk) , ].lk tending to -00 so that

].l
Ip (x,E) - L p.(x,~)1
] .s Cn ' (1 +I~I) k for x E: n'
j < k
00
mo
Then p E: S In) and P"" L Pj
j=O

To prove this theorem, we need the following

LEMM\ 4.21 let nl and n2 be two compact subsets of lR n such

that nl is contained in the interior of n2 Then there exist

2
constants cl > 0 and c2 > 0 such that for all f E: C !n2)'

Sup la.fl 2 < c l (Sup If12) + c 2 (Sup I f I) (Sup la~fl).

nl ] - n2 n2 n2 ]

It suffices to assume that n 1 and f is real valued.

with this reduction, the proof becomes an exercise in elementary calculus.

The idea is roughly as follows: We wish to show that if II f II 00 and

II i" II 00 are both small, then so is II f 'II 00 . If, instead, II fll II 00 is

small and If'(x ) I is large, then If'l will be large in some sizeable
o
interval [a,b] containing x . But then If(b) - f(a) I and hence
o'
II f 1100 is large. We leave it to the reader to work out the quantitative

details of this argument.

 93

PROOF OF THE THEOREM By Proposition 4.18, there exists q

00

in sIDe such that L p.; so, it will suffice to show that

J
j=O
First,
,
Ip(x,~)-q(x,~)I= I(p(x,~) - L Pj(x,O)-(q(x,O-. L p.(x,~))I·
j<k ]<k]

For

00

Also q '\, I p ..
J
These show that, for any N-,
j=O

Ip(x,~) 1- q(x,O I .s cNst,(l+I~I)-N , x € st' .

We want to prove such an estimate for D~ D~(P-q) also. To this end,

we will apply Lemma 4.21 to the function (x,n) + (p-q) (x, ~+n) considering

~ as a parameter, and taking st l = st' x {O} and st 2 a small

•1,
neighbourhood of st l If lal + lsi we use the estimate just

established for p ~ q, together with the hypothesis (i) on the second

order derivatives of p; the lemma implies that D~ D~(p-q) is rapidly

decreasinq for lal + ISI= 1. Combining this with the hypothesis (i)

on the third order derivatives of p, we see that is rapidly

decreasing for lal+ ISI= 2. Proceeding by induction on lal + lSi we

get the required result.

§4. PROPERDi SUPPORTED OPERATORS

Since pseudo differential operators map COO to COO rather

o
then c~, it is generally not possible to compose two of them. The

problem may be remedied by considering a more restricted class of operators,

the so called "properly supported" ones.

 94

DEFINITION 4.22 A subset K of nx n is said to be proper if,

-1
for any comoact set n'c: n 'IT 2 m ') t"'I K

are compact.

Here and are projections of nx n onto the first

and second factors.

For example, the diagonal

~ = {(x,x): x € n} is proper. Most of the

proper subsets we will be considering are

A proper set
neighbourhoods of subsets of the diagonal.

00 00
DEFINITION 4.23 An operator T : C (n) ~ C (n) is said to be
o
properly supported, if its distribution kernel K has proper support.

EXERCISE Let T = E a a (x) Da be a differential operator on n

Compute the distribution kernel K of T and show that supp K is

a subset of the diagonal. Thus T is properly supported.

00
If T is properly supported then it maps C into itself,
0

since supp Tu c: 'ITl ('IT 2 (supp u)

-1
n supp K) , as is easily seen from the

formula Tu(x) = J K (ox,y) u(y)dy. M:>re generally, for any n'cc n ,

there exists such that the values of Tu on n' depend

only on the values of u on n", namely,

00
From this, it follows that T can be extended to a map from C (n)

(m n', n"
00
to itself. In fact, if u € C and are as above, we define

Tu on n' by

where This definition is independent

 95

of the choice of ~ and gives the same result on the intersection

of two n's; so Tu is well defined on all of n.

If T is a properly supported pseudo differential operator,

so that T extends to a map from E'(n) to V'(n), the same

arguments show that T maps E '(n) into itself and extends further

to a map of V, (m to itself.
00
Suppose T and S are properly supported operators on C (n)
00
with distribution kernels K and L which are C off the diagonal.
00
Then TS is an operator on C (n) with distribution kernel M formally

given by M(x,y) = fK(x,z)L(z,y)dz. In fact, if x I y, since K(x,.)

and L(.,y) are smooth except at x and y, the productK(x,.)L(.,y)

is a well defined element of E' , and the formula M(x, y) = < K(x, • ) L(. , y) ,1 >
00
displays M as a C function off the diagonal.

PROPOSITION 4.24 ~ M is proper. Thus, TS is properly supported.

Clearly,

-1 -1
Supp MC{(x,y): lT 2 (lT l (x) nsupp K) (llT l (lT 2 (y) nsupp L) +.IJ}

Suppose AC n is compact and set B = lT2 (lT -1

l . (A) ("\supp K), then B

is compact and

-1
lTl (A) (l supp MCA x {y: B n lTl (lT -1
2 (y) () supp L) + .IJ}

-1
A x {y: lTl (B) n lT2-1 (y) n supp L + .IJ}

-1
A x lT2 (lT l (B) () supp L)

-1
which is compact. Likewise IT 2 (A) n supp M is also compact.
 96

EXERCISE Suppose AC m x m is proper. Show that there exists

00
a properly supported ~ E C (n x n) such that ~ = 1 on A.

(Hint: Let {~.} C CooW x m be a partition of unity on n x nand

] 0

let L ~ .) .
A nsupp ~. ~ ]
]

§5. '¥DO's DEFINED BY MULTIPIE SYM30IS

We have arranged our definition of pseudo differential operators

ro agree with the usual convention for differential operators, according

to which differentiations are performed first, followed by multiplication

by the coefficients. However, in some situations (for example, computing

adjoints), it is convenient to have a more flexible setup which allows

multiplication operators both before and after differentiations. We,

therefore, introduce the following apparently more general class of

operators (which, however, turns out to coincide with the class of '¥DO,

modulo smoothing operators).

DEFINITION 4.25 For n open in Rn and m real, we define the

class of multiple symbols of order m on n,

Smmxm = {a Ecoom x Rnxm:Va,S,';/n'cc.n,3c=caSn'

such that sup IDS D~a(X'~'Y) I ~ c(l+I~I)m-lal}

x,y E n' x,y

When a E sm m x m we define the operator A a (x,D,y) by

Au(x) = 11 e2ni(x-Y)·~a(x,~,y)u(y)dy d~

Here the integral must be interpreted as an iterated integral with

integration performed first in y and then in ~ as it is not absolutely

convergent as a double integral.

 97

We observe that if a(x,~,y) = a(x,~) is independent of y,

then A = a(x,D); thus this class of operators include the ~DO IS.

We also observe that different a's may give rise to the same

(m
00
operator. For example, if a(x,~,y) = ~(x)~(y) with ~,~ E C
o
and supp ~ ()supp ~ = ~, then a E sO(n x n) and a(x,D,y) = O.
m .
DEFINITION 4.26 Given a E S (n x n) we define

La = {(X,y)E n x n : (x,~,y) E supp a for some

PROPOSITION 4.27 Let a E Sm(n x n) and let K be the distribution

kernel of A = a(x,D,y). Then

i) Supp KC La' and

ii) if the support of K is proper, then there exists

a' E SmW x m such that a '(x,~,y) = a(x,~,y) when (x,y) is near

the diagonal in n x n , La' is proper and a(x,D,y) = a'(x,D,y).

PROOF The kernel K is given by

< K, w > flI e

21Ti(x-y)·~
a(x,~,y)w(x,y)dy d~ dx.

From this, we see that < K, w > = 0 whenever supp w ~La ~.

Therefore, supp K C La This proves (i).

00
To prove ·(ii), choose a properly supported ~ECWXm

with ~ = 1 on 6. U supp K, 6. being the diagonal. Set

a(x,~,y) ¢(x,y)a(x,~,y). Then L

a
,c: supp ¢ and hence La' is
I
proper. Also a'(x,~,y) = a(x,(,y) when (x,y) is near the diagonal.

Now,
 98

< a(x,D,y)u,v > = < K, v ®u >

< (j>K, v ® u >

<K,~(v®U»

IIf e
21Ti(x-Y)'E;;
a(x,E;;,y)(j>(x,y) x

x v(x) u(y) dy dE;; dx

III e2ni(x-Y)'~al(x,E;;,y)u(y)dy dE;; dx

< a '(x,D,y)u,v :> •

Since v is arbitrary, we have a(x,D,y) = a'(x,D,y).

THEORE M 4.28 Suppose a e Sm(Q x Q) and A = a(x,D,y) is properly

supported. Let p(x,E;;) = e- 2nix .E;; A(e 2ni (.).E;;) (x). Then
m
peS (Q) and p(x,D) = A. Further,

p(x,E;;) '\, L ....!..

Ct I.
aCt
E;;
oCt
Y
a(x,E;;,y) Iy=x
CL

00
PROOF For u in Co (Q) , we have

u (x) I e21Tix·E;;~(E;;)dl: •

Therefore, by linearity and continuity of A,

AU(x) f A(e 2ni (.).E;;) (l¢);(E;;)dE;;

f e 2nix .E;; p(x,E;;) ~(E;;)dE;; = p(x,D)u(x).

By the proposition above, we can modify a so that La is proper

without affecting A or the behaviour of a along the diagonal x = y

(which is what enters into the asymptotic expansion of pl. Set

 99

b(x,~,y) = a(x,~.,x+y). Then as a function of y, b has compact

support. Indeed, for fixed x and ~, if y E: {y ': b(x,~,y ') + o}

then x+y E: n2(n~1(x) n La) which is compact. Now,

p(x,n) __ e-2nix.n fJe2ni(X-Y).~

b(x,~,y-x)e
2niy.n dy d~

where "b 3 is the Fourier transform of b in the third variable.

ex)
Since b(x,~,.) is in C ,this calculation is justified and
o
"-
b3(x,n,~) is rapidly decreasing in the variable ~.

More precisely, since a E: Sm(Q x Q), we have

Since

[
l+lnl]S ~ (l+I~-nl) lsi, taking n+~ instead of n
l+1~1

we have (1 + I~+nl)s ~ (1 + 1~I)s (1 <+ Inl) lsi. If we use this in

we get

I f we take N so that - N + m -I a I < - n we have

 100

On the other hand, taking the Taylor expansion of

,.
b3(x,~+n,~) in the middle argument about the point n , (4.29) gives

~ cl~lk sup laab3(X,n+t~,~) I

10.1= k n
o< t < 1

~ cNI~lk sup (l+ln+t~l)m-k(l+I~I)-N.

o< t < 1

When I~I < ~ Inl ' taking N = k we get

,. ~ m-k
Ib3(x,~+n,~) - L d~ b3(X,n,~) 0. 1 [ ~ ck(l+lnl) •
10.1 < k

1
When I~I ~ 2 Inl ' we have

(Actually, the exponent of (1+1~1) can be taken as m - N when

m- k > 0 and k - N when m-k < 0). Also we have

J d~ b3(X,n,~)~ad~ = o~ f e2TIiy.~ d~b3(x,n,~)d~ly=0

= 00. aa b(x,n,Y) I 0
y n y=

= 00. ao.a (x,n,Y) I

y n y=x

So finally

-
t
10. < k
0ao. 1 I
d a(x,n,Y) -,
y n 0.. y=x
I
 101

Now

J 1rr.l Ck(l+lnl)m-k d~
I~I < 2

In the second integral, we take N > n + m+ k so that

by the usual integration in polar coordinates. Therefore, since N > k

we obtain

Ip(x,n) - \'
L 1 a .a
- , 0 d a(x,n,y)
1 1
~ Ck(l+lnl)
~k
10.1 < k a· y n x=y

with ~k = m - k + n.

Combining this with (4.30), we see that all the conditions of

Theorem 4.20 are satisfied, so we are done.

COROLlARY 4.31 If P E ~m , there exists a Q E ~m such that Q

is properly supported and P - Q is smoothing.

00
If P p(x,D) choose ~ E C (n x n) which is properly

supported and ~ 1 near the diagonal. Set a(x,~,y) = ~(x,y)p(x,~).

Then aEsmWxm and by construction La is proper. So a(x,D,y) =

If and denote the distribution kernels of P and

Q then it is easily seen that This shows that

00
vanishes near the diagonal: so, by Theorem 4.10, K - K is C .
Q P
Therefore, Q-P is smoothing.

REM\RK 4.32 The idea of multiple symbols can clearly be generalised.

For example, if a E COO W x R n x n x Rn) satisfies estimates of the form

 102

m -lsi m -Ial
lov Da oS
x,y ~ n
a(x,~,y,n) I~ C(l+lnl) 2 (l+j~l) 1 ,

one can define an operator A = a(x,D,y,D) by

which agrees with our previous definition~, if a is independent of

the last one or two variables. As one would expect, under reasonable
m1 +m 2
restrictions on supp a, it turns out that A € ~ (n). The reader

may wish to amuse himself by working this out and computing the symbol

of A.

§6. PRODUCTS AND ADJOINTS OF ~OO'S

We are now in a position to compute products and adjoints of

~DO 's. First we clarify our terminology. let T and S be linear

operators from Crr;> W) into C

co
(m. We say that S = Tt if
0

T*
C'O
< Tu, v > = < u, Sv > for u,v € Co(m and we say that S if
co
< Tu, v > = < u, Sv > for u,v € C W).
0

RE~RK 4.33 If T is properly supported, then Tt and T * are

also properly supported. Indeed, if K is the distribution kernel

of T, then the distribution kernel of Tt is and that of T*

is K* where Kt (x,y) = K(y,x) and K* (x,y) = K(y,x).

We recall that if P € ~(n), we denote the corresponding

THEOREM 4.34 If P € ~(m is properly supported, then

\' J.::.!llal a a
~ af a~ Ox 0p(x, -~),
 103

<Xl
For u,v E C (n), we have
o

< Pu, v >

I (J e
2nix.E;, A
p(x,E;,)v(x)dx)u(E;,)dE;,

where
g(E;,) J e 2nix.E;, p(x,E;,) v(x)dx.

By Corollary 4.6, 9 is a rapidly decreasing function. Thus

< Pu,v > J ug,. =<

t
u, P v >

so that

9'" (y)
t Jfe 2ni (X- Y).E;, p(x,E;,)v(x)dx dE;,
(4.35) P v(y)

IJe 2ni (y-x).E;, p(x,-E;,)v(x)dx dE;,

a(x,D,y)V(Y)

where a(x,E;"y) = p(y, -E;,). Therefore, by Theorem 4.28,

and

The assertions about P* follows along similar lines.

m m
THEOREM 4.36 If P E '1' l(m and Q£ '1"2(n) are properly
m +m
suppor ted, then OF E'¥ 1 2 (m and
 104

t t
PROOF Since P = (P) , we have

(Pu) (x) =fI e2ni(x-Y)'~a t(y,-~)u(y)dy d~ by (4.35).

P

In other words,

(Pu)
A
(~)
S e -2TIiy.~ a t(y,-~)u(y)dy.
P

Therefore,

(QP)u(x) SS e
2TIi(x-y).~
aQ(x,~)a t(y,-~)u(y)dy d~
P

a(x,D,y)u(x)

where a(x,~,y) = aQ(x,~)a t(Y'-~)' Clearly

ml+m 2 P
which implies O? E: 'if (m . M:>reover

'\, L ( L
S,A V+C=A

(-1) C I I (x O - x O) A with
But L V!C! xo (1,1, ••. ,1)
V+C=A
if A= 0

= {: if A +0
 105

m
CORO LIARY 4. 37 (PRODUCT RULE FOR ~DO 's) Q = q(x,D) € ~ (n)

C m), then for any positive integer

<Xl
and f € N,

q(x,D) (fu) I where

lal < N

CORO LLARY 4. 38 The correspondences p + p(x,D) and P + Op are

m
* homomorphisms modulo lower order terms i.e., if p € 5 l(n) and
m2 m +m -1
q € S (n), then (pq) (x,D) - p(x,D)q(x,D) € ~ 1 2 (n) and
* _ ml-l ml m2
p(x,D) - p(x,D) € ~ and also, if P € ~ (n) and Q € ~ (n),
m +m -1 m -1
then Op Q - Op 0 Q € 5 1 2 m), Op* - Op c.<"' S 1 (n) ••

m
COROLLARY 4.39 I f P € ~ 1 m) and
ml +m 2-1
[ p ,Q] = p Q - OJ € ~ m)
where {f,g} stands for the Poisson bracket defined by

Following the philosophy that smoothing operators are negligible,

it is a trivial matter to extend these results to non-properly

supported operators.

Suppose P € ~m(n) is not properly supported. By Corollary 4.31,

we can write P Pl + 5 where Pl is properly supported and S is

pt = pt + st t
smoothing. Then where Pl is given by Theorem 4.34
1
st *
and is again smoothing (c. f. Remark 4.33). Likewise for P If

Q is a properly supported ~DO, the products PQ and QP are well

defined as operators from C

00.

0
m) to
<Xl
C m) . Again, we have

PQ = Pl Q + 5Q and QP = QP l + 03; Pl Q and QP l are described by

Theorem 4.36, while SQ and (l) are smoothing.

 106

§7. A CONTINUITY THEOREM FOR 'fDO ON SOBOIEV SPACES

We now state and prove a continuity theorem for pseudo

differential operators acting on Sobolev spaces. We are indebted

to Dr P.N. Srikanth for simplifying our original argument.

THEORE IV: 4.40 Suppose P = p(x,D) € ~(n). Then

i) P : H -+ Hloc (m continuously for all s € lR.

s s-m

ii) If P is properly supported, P : Hloc (m -+ Hloc (m

s s-m
continuously for all s € lR.

PROOF To establish (i) , we must show that the map u -+ <jlPu is

C (m.
00
bounded from H -+ H for any <jl € Replacing p(x,E) by
s s-m o
<jl(x)p(x,~) we must show:

if P € Sm(lRn) and p(x,~) =0 for x outside a comp'act

set, then P p(x,D) is bounded from H and H for every s in

s s-m
lR.
Suppose then that u € H
s
Then Pu € E' , and from the

definition of P on distributions, we see that

A -2ni n. (. )>
(Pu) (n) =< Pu, e

=If
=f

and hence, if V € S,

< Pu, v >

g(n) = (1 +lnI 2) (m-s)/2 ~(n)

 107

We wish to estimate K(n,~). For any multi-index a , since

ClO
pt· ,~) E Co , we have

so that for any positive integer N,

and hence

IK(n,~) I .s CN(l + 1~I)m-s(l + Inl)s-m(l + I~-nl)-N

.s CN(l + I~-nl) Im-sl-N

If we take N > n + Im-sl, we see that

Therefore by Theorem 1.1 and the Schwarz inequality,

I < pu, v> I .s C Ilf 112 Ilg 112 = C Ilu II (s) Ilv II (m-s)

From this, it follows that Ilpu II (s-m) .s C Ilu II (s) for u E H

s
which establishes (4.41) and hence (i).

Now suppose P is properly supported and u E Hloc (nl. If

s
~ E cClO(n) there exists n' ~ n such that the values of Pu on
o
supp ~ depend only the values of u on n'. Thus if we pick
ClO
~' E C (n) with ~'= 1 on n', we have ¢Pu = ~P(~'u). But
o
 108

~'u £ H and by (4.41), ~P is bounded from H to H This

s s s-m
establishes (ii) and completes the proof.

§8. ELLIPTIC PSEUDO DIFFERENTIAL OPERATORS

DEFINITION 4.42 P £ ~(Q) is said to be elliptic of order m if

DEFINITION 4. 43 If P £ ~(Q), a left (resp. right) parametrix

of P is a 'fOO Q such that CJ? - I (resp. P Q - I) is smoothing.

THEOREM 4.44 If P £ fID is elliptic of order m and properly

supported, then there exists a properly supported Q £ 'f- m which is

a two-sided parametrix for P.

co
PROOF Let P = p(x,D). We will obtain Q = q(x,D) with q'\, 2
j=O
qj
co
where the q. 's are defined recursively. Let r;(x,f;) bea C
J
function such that r;(x,f;) = 1 for large f; and r;(l(,f;) 0 in a

neighbourhood of the zeros of ~~x£~)

p. Define qo(x,f;) Then
p(x,f;)
s-m Let Q = q (x,D). We have
qo £
o 0

q p mod s-l = 1 mod s-l 1 + rl with

o

Let Ql ql (x,D). Then

1 + r2 with

etc. Having determined so that

1 + r j +l , r j +l £
S-(j+l)
 109

set

00

Let 0 be a properly supported operator wi th symbol q'\, L q.

j=O J
-co
Then crOP = 1 mod S , i.e. , QP- I is smoothing. Thus 0 is a

left parametrix. In the same way, we can construct a right

parametrix Q'. Then if S = ()? -I and S' = PQ' - I, we ~ave

QPQ' = SQ' + Q' = QS' + Q .

So Q - Q' = SQ' - QS' is smoothing. Hence Q is a two sided

parametrix.

EXERCISE What happens if P is not properly supported? (cf. the

remarks at the end. of § 6 ).

The left parametrices are used to prove regularfty theorems

and the right parametrices are used to prove existence theorems.

Indeed, we have:

CORO UARY 4.45 (E LLIPTIC REGU rARITY, THEOREM) If P is elliptic

€ H IO
s m(m.
of order m, u € V'(Q), Pu € Hloc(Q) implies U +C
s
particular, P is hypoelliptic.

Let Q be a left parametrix and set S = (J> - 1. Then

u = QPu - SUo Since Pu € HIOC(Q) and Q is properly supported,

s
co
Q€ ~-m we have QPu € H!~~(Q) by Theorem 4.40. Also Su € C since
loc
s +m(m.
S is smoothing. Hence u € H

THEOREM 4.46 Every elliptic differential operator is locally solvable.

In fact, if P is elliptic on Q, f € V'(Q) and x € Q , there

o
 110

exists U E V, (Q) such that Pu = f in a neighbourhood of x • (Of

a
course, if f E Coo, then u E"C oo ~ xo' by the previous corollary~.

PROOF By cutting f off away from xo ' we can assume that

fEE' and hence f E H for some s. Let Q be a properly supported

s
parametrix for P and let S = PQ - I. Then S is also properly

supported. If ~l c.c ~ is a neighbourhood of xa , then there exists

~2C:~ ~ such that the values of Su on ~l depend only on the values

of u pick such that on

Now observe the following:

i) Tu = Su on ~l

ii) T

From (ii) and the Arzela-Ascoli theorem, it follows that if (uk) is a

bounded sequence in Hs' (TU k ) has a convergent subsequence in

and hence in H. Therefore, T is compact on H So the equation

s s
(T + I)u = f can be solved if f is orthogonal (with respect to the

pairing of H and H_S) to the space N = {g: (T*+I)g This space

s
00
N is a finite dimensional space of C functions so we can always

make this happen by modifying f outside a small neighbourhood of x

a
Indeed, piCk a basis gl' g2'···' gv for N and pick a neighbourhood U

of x so small that g , .•. , g are linea r ly independent as functionals

a 1 V
on C
00

a
m" U). (Such a U exists; otherwise, by a limiting argument using

the local compactness of N, we could find a nontrivial linear comoination

of supported at {x }, which is absurd) . Then we can make

a
f 1 N by adding to f a function in C
00

a en" U). But then

 111

POl (I + S)u = (I + T)u on nl and (I + T)u = f

in a neighbourhood of x So Qu solves the problem.

o

§9. WAVEFRONT SETS

We now introduce the notion of wavefront sets, which provides

a precise way of describing the singularities of distributions: it

specifies not only the points at which a distribution is not smooth

but the directions in which it is not smooth.

All pseudo differential operators encountered in this section

will be presumed to be properly supported, and by '''¥DO'' we shall

always mean "properly supported 'l'OO" .

DEFINITIONS 4.47

(i) Let n be an open set in Rn. Then we define

TOn n x (Rn " {a}). (In coordinate-invariant terms, T< n is the

cotangent bundle of n with the zero section removed).

(ii) A set SeTon is called conic i f (x,l:) (: S => (x,rl:) (: S , Vr > a

(iii) Suppose P = p(x,D) (: ~(nl. Then is said to

be non-characteristic for P if Ip(x,l:) I ~ cll:l m for II: I large and

(x,l:) in some conic neighbourhood of (x
o
,1: 0 ).

( iv) The characteristic variety of P, denoted by char P, is defined

by char P = {(x,l:) (: TO n: (x,l:) is characteristic for pl·

(vl Let u (: V' (n). The wavefront set of u, denoted by WF (u) ,

is defined by

WF(u) n{char P: P (: 'l' (n), Pu (: cooWl}.

 112

o
The restriction P E~ (n) in the definition of WF(u) is merely a

convenient normalisation. We could allow ~DO of arbitrary order

without changing anything, for if P E ~(n), and Q E ~-m(n) is

elliptic, then QP E~O(m, char (CJ?) = char (P) (by Theorem 4.36),
00
and QPu E COO if and only if Pu E C (by Corollaries 4.11 and 4.45).

a
EXERCISE When p = 1. a
a
(x) [, show that the characteristic
lal '.S m
variety of the differential operator P =p(x,D) is

char P {(x,[,): ~ a (x) [,

a O} .
a
lal= m
The motivation for WF(u) is as follows. If u E E' , to

say that u is not smooth in the direction [,0 should mean that

u is not rapidly decreasing on the ray through [,

o
. On the other
00
hand, if Pu E C then (Pu) should be rapidly decreasing everywhere.
o
If these conditions both hold,then [,0 must be characteristic for P.

Localising these ideas, we arrive at WF(u).

00
Thus (x ,[, ) E WF(u) means roughly that u fails to be C
o 0

at x in the direction r For another interpretation of this

o So
statement, see Theorem 4.56 below. For the present, we show that

WF(u) is related to sing supp u as it should be. We denote the

projection of TUn onto n by 11

THEOREM 4.48 For u E V'(n), sing supp u = 1I(WF(u».

PROOF Suppose x ~ sing supp u. Then there exists

o
¢ = 1 near Xo and But multiplication by ¢ is a ~DO
 113

-1 -1
of order O. Call it P~. Then char P~ = n (~ (0)). Since

~ = 1 near x o ' n-l(~-l(O)) is disjoint from n-l(x o ). Therefore,

-1
n (x)(1 WF(u) =~, i. e. , x 3 ~ n(WF(u)).
o u

Conversely, suppose x ~ n(WF(u)). Then for each E; with

o

IE; I with Pu E C'" (~) and (x ,E;)~ char P.

o
Each char P is a closed conic set; so, by the compactness of the

unit sphere, there exists a finite number of ~DO 's, say,

Pl,···P N E~O (m with P.u E C'" (m and

)

N N
P.* P.
-1
((\ char P ).) () n
j=l
(x)
0
~. Set P L
j=l
) )

Then P is elliptic near x and Pu E C"'(m. Therefore by Corollary

o
4.45, u is C'" near x 0' i. e. , x ~ sing supp u.
0

DEFINITION 4.49 l.€t P = p(x,D) E ~mW) and U be an open conic

set in TO~ We say that P has order - '" on U , if, for all

closed conic sets K ~ U with n(K) compact for every positive integer

N and multi-indices a and S , there exist constants CaSKN

such that

The essential support of P is defined to be the smallest closed conic

set outside of which P has order - '" •

EXERCISE If P is a differential operator whose coefficients do not

all vanish at any point of ~, show that the essential support of

P is T ~
 114

PROPOSITION 4.50 Ess. sup'p PQ c:: Ess. supp P nEss. supp Q.

This follows immediately from the expansion

LEMMA 4.51 Let (x,~) € ~~ and U be any conic neighbourhood

o 0
of (x
o
,~
0
). Then there exists a P € '1'0 m) such that

00
Choose p (~) € C with the following properties:
o

ii) p is homogeneous of degree 0 for large I~I

o
00
Then take ~ € Co(n(U» with ~ 1 near xo and put

The following theorem and its corollary are refinements of

Corollary 4.11 (pseudo local property of '1'00) and Corollary 4.45

(elliptic regularity theorem).

THEOREM 4.52 If P € ~m) and u € V'm), then WF(Pu)c:

WF(u) rl Ess. supp P.

PROOF If (xo'~o) ¢ Ess. supp P, by ~mma 4.51, we can find a

o
Q € 'I' such that (x ,~ ) ¢ char Q and Ess. supp P nESS. supp Q = ~.
o 0
-00 00
Then QP € 'I' by Proposition 4.50, so that QPu € C. But this

implies that (x,~)

o 0
¢ WF(Pu).

Suppose now that (xo'~o) ¢ WF(u). Then there exists A € '1'0

with Au € COO and (x,~)

o 0
¢ char A.
 115

Claim There exist operators B, C C'

<- '1'0 with
-co
CA mod 'I'

co co
Granted this, BPu CAu + (C function) e: C which implies

that (x,~) ~ WF(Pu).

o 0

To prove the claim, let A be elliptic with (JA (JA on

0
0
a conic neighbourhood U of (x ,~ ). Then Ess. supp(A - A)n u =~ •
o 0 0

By Lemma 4.51, choose B with (xo'~o) ~ char B and

Ess. supp B C. u. Let E

0
be a parametri~ for A
0
and set C = BPE 0
-co
Then CA = BPE 0 A = BPE 0 (A - A ) + BPE A
0 o 0
Since E A
o 0
=I mod,!,
-co
and BPE (A - A ) also belongs to 'I' (by Proposi tion 4.50 again) ,
o 0
-co
CA = BP mod '!' This completes the proof.

COROllARY 4.53 If P is elliptic, then WF(u) = WF(Pu).

PROOF By the theorem, WF(Pu) c: WF(u). If E is a parametrix for

00
P, then WF(u) WF(EPu + C function)

WF(EPu)~ WF(Pu), by the theorem again.

Hence WF (u) WF(Pu) .

THEOREM 4.56 Let u e:V' (D). Then (x ,~ ) ~ WF(u) i f and only i f

o 0
co n ...
there exists <p e: Co (:JR ) such that ¢ (x o ) = 1 and (¢u) is rapidly

decreasing on a conic neighbourhood of ~o.

o
PROOF (Sufficiency) If such a ¢ exists, choose p e: S ,

p(x,~) = p(~) with p(~) =1 near ~o and p(~) o outside the region

where (¢u) is rapidly decreasing. Then p(¢u)

... e: S and hence

p(D) (¢u) e: S where p(D) = p(x,D). The operator Pu = ¢p(D) (¢u) is a

 116

pseudo differential operator of order 0 with symbol ~(x)2 p(~)

modulo S
-1
, so (x ,~ )
o 0
t char P. This implies that (x ,~ )
o 0
t WF(u).

(Necessity) Suppose (x ,~ )
o 0
t WF(u) Then there exists

a neighbourhood U of x
o
in n such that for
ex>
all x € U. Choose ~ € C (U) such that ~(x) = 1. Then
o 0

(x,~o) t WF(~U) for all x € mn. Let

1: {~: (x,~) € WF(~U) for some x} .

This L does not contain ~o and is a closed conic set. There exists

p(~) €
o
S , P 1 on a conic neighbourhood of r- and p o on a
"0

neighbourhood of L , say LO Since p(~) =0 on 1:0

Ess. supp p(D) n (mn x L)C

o
and hence Ess. supp p(D) ('IWF(~U) = fl.
ex>
But this gives Ess. supp(p(D)~u) = fl. So p(D)~u € C We now claim

that p(D) ~u € s.
,.
Accepting this, we see that p(~U)A € S and in particular, (~u)

is rapidly decreasing near ~ . To prove the claim, observe that

o

0(1 + I~I) lal-ISI n

€ L2 if lal-ISI < - 2

So if we put K(x)

application of Leibniz's rule and the Sobolev imbedding theorem shows

Isi-lal > ~n + 2. Hence K and all its

derivatives are rapidly decreasing at ex> and since ~u € E' , the same

is true of p(D)~u ~u * K.
 117

§lO. SOME FURTHER APPL[CATIONS OF PSEUDO DIFFERENTIAL OPERATORS

We conclude our discussion of pseudo differential operators

by giving brief and informal descriptions of some further applications.

th
We recall that an m order ordinary differential equation
(m) (1) (m-l)
u = F(x,u,u , ... u ) can be reduced to the first order

system

d
dx

u
m F(X,ul,···,U)
m)

simply by introducing the derivatives of u of order < m as new

variables, and this reduction is frequently a useful technical device.

th .
To do something similar for m order partial differential equations,

however, is more problematical. If one simply introduces all partial

derivatives of u of order < m as new variables and writes down the

first order differential relations they satisfy, one usually obtains

more equations than there are unknowns, because of the equality of the

mixed partials. Moreover, such a reduction usually does not preserve the

character of the original equation. For example, take the Laplace

equation in two variables:

a 2u
+
a2 u f.
2 2
ax ay

Putting ul u, u 2 au ~
u3 we have a 3 x 3 system given
dx ay
by
 118

f.

The original equation is elliptic. But consider the matrix of the

top order symbols of the 3 x 3 system obtained above. The matrix is

given by

E;. a a
21Ti n a a
a E;. n

which is not invertible. This means that the first order system is not

ell;i.ptic.

There is, however, a method, due to A.P. CALDERON, of reducing

an mth order linear partial differential equation to an m x m system

of first order pseudo differential equations which preserves the

characteristic variety of the equation in a sense which we shall make

precise below. In this method, one of the variables is singled out to

playa special role, so we shall suppose that we are working on R n +l

with coordinates (xl' x 2 ,···,xn ,t).

Let L be a partial differential operator of order m on

such that the coefficient of am is nowhere vanishing. Dividing

t
throughout by this coefficient, we can assume that L is of the

following form:
 119

m-l
L \' A . (x,t,D ) a tj
L m-) x
j=O

Here A . is a differential operator of order ~ m-j in the x

m-)
variable with coefficients depending on t.

We want to reduce the equation Lv = f to a first order

system. To this end we proceed as follows:

using the familiar operators AS with symbol (1 + 11:.1) s/2

(acting in the x variables) we put

m-l
ul A v

m-2
u2 A '\V
m-3 2
u3 A at v

For j < m, we observe that at Uj Therefore,

m-l
at um f +
j=O
t A
m-)
., (x,t,D) a j v
x t

m-l
f + L A . (x,t,D ) Aj - m+l u j +l
m-) x
j=O

m
j-m
f + l.
j=l
A
m-)+
. 1 (x,t,D ) A
x
u.
)

.j-m
Set B. (x,t,D ) = A ' +l(x,t,D) H This B. is a '1'00 of
) x m-) x J
order 1 in the variable x. Then the equation Lv = f is equivalent

to
 120

at u = Ku + f
t t
where (O,O, ••• ,f) • (Here ( •.• ) denotes

the transpose of the row vector ( .••. » and K is the matrix given

by

o I\. o••• 0
o o 1\. ••• 0

o 0 O••• I\.

Bl B2 B3 • .Bm

This K is a matrix of first order pseudo differential operators in x,

with coefficients depending on t.

Let us now make a little digression to fix on some notations

which will be used in the further development. For apE Sm , a

function Pm(x,~) homogeneous of degree m in ~ is said to be the

principal symbol of p(x,D) if p - Pm agrees with an element of sm-l

for large I~I. we remark that not all ~DO's have a principal

symbol; but most of them that arise in practice do. Moreover, the

principal symbol, if it exists, is clearly unique.

EXAMPIES

(i) If P is a polynomial,

p(x,~)

then
 121

Returning to our discussion, let ak(x,t,~) be the principal

symbol of Ak(x,t,D x ) (Here we are considering Ak as an operator

of order k, so if it happens to be of lower order, its principal

symbol is zero). Thenthe principal symbol of B. is

J
b. (x,t,~) a m_ j +l (x.,t,~) 1~ Ij-m and the principal symbol of K is
J
the matrix Kl ,

D I~I 0 0 0

o 0 I~I 0 0

o o o 1~ I . . . 0

o 0 0 0 ••• I~I

The principal symbol of L, on the other hand, is

m-l .
L (x,t,~,'r) = \' a . (x, t , r-) (21T iT) J
m J m-J <,
J=O

The characteristic variety of L, i.e., the set of ze~os of L ,can

m
easily be read off from the matrix K as follows:

PROPOSITION 4.57 The eigenvalues of Kl (x,t,~) are precisely

21Ti'1"'" 21Ti'm where ' 1 ' '2 r"" 'm are the roots of the polynomial

PROOF The characteristic polynomial pO,) of Kl is the determinant of

 122

A -I~I 0... a

a A -I~I ... a

a a a ... -I~I

Expanding in minors along the last row, we get

m-l
(_l)m+ j (-b.) Aj -1 (- I~ I)m- j + A.m- l (A -
ptA) L ]
b )
m
j=l
m-l
Am _ m-l j-l
A b - L b. 1~lm-j A
m ]
j=l
m m-l
j-l
Am _
L a = Am _ L a j
m-j+l A m-j A
j=l j=O

This proves the proposition.

It is also easy to incorporate boundary conditions into this

scheme. Suppose we want to solve Ehe equation Lv = f with the

boundary conditions Bjvlt=o = gj , 1,2, •.• ,V where

m.
B. n.
L b.k (x,t,D )d kt b~ is of order m.-k and m. < m-l.
] k=O ] x ] ] ]

If we set

B~] Am-m ]·-1 b.]k-l (x,O,D)

x
A
k-m
,

Ij>j Am-m·-l
] gj

then with u.
]
= Am-j j-l
dt v as above and u~(x)
]
u. (x,O), the
]

boundary conditions will become

 123

m·+l
1. J B~ u~ <P j , j 1,2, •.• V •
k=l J

This is a system of zeroth order pseudo differential equations.

The above method of reduction is useful, for example, in the

following two important problems.

(1) CAUCHY PROBIEM FOR HYPERBOLIC ECUATIONS

IY = f, a~vlt=o = gj' j = 0,1, ••• , m-l

Here the equation is said to be hyperbolic when the eigenvalues of the

matrix Kl (i.e., the principal symbol of K occurring in the linear

system corresponding to Lv = f) are purely imaginary. Equivalently,

the roots of L (x,t,~,.) are real.

m

("2) ELLIPTIC BOUNDARY VAlliE PROB IEM>

Lu = f on 51 , B. u = gj on an
J

where L is elliptic of order 2m and j = 1,2, ••. m. Here one works

locally near a point x e: an and makes a change of coordinates so

0

that becomes a hyperplane near x One then can apply Calderon's

o
reduction technique, taking t as the variable normal to an.

(See Michael Taylor [3]).

We shall now sketch an example of a somewhat different technique

for applying ~DO 's to elliptic boundary value problems.

00
Let 51 be a bounded open set of with C boundary an.

Consider the problem

t:,u =0 in a u + au g on an, a e: C
00
(am.
X
 124

Here X is a real vector field on the boundary which is nowhere

tangent to the boundary and aXu grad u.X •

If V is the unit outward normal to aQ, by normalising

we can assume that X= V + L where L is tangent to aQ. We want

to use ~DO to reduce this to the Dirichlet problem. Pretend for the

moment that ClQ = R n-l x {OJ and Q = {x: x < O} so that a

_a_
n V axn
n-l 2
If l!.u =0 L au being the
. 1
]= ax.2]
Laplacian on the boundary. So formally aV u = +/(-l!. u)
- b
Indeed, if

we compare this with our discussion of the Poisson kernel in Chapter 2

(taking account of the fact that Q is now the lower rather than the

upper half space), we see that the equation au=!(-l!.)u is correct

V b
if interpreted in terms of the Fourier transform in the variables

It turns out that something similar works for our original domain Q.

Namely, if u is smooth on Q and l!.u = 0 on Q then avuiaQ = p(ui aQ )

where P is a pseudo differential operator of order 1 on aQ which

,
equals !(-l!.b) modulo terms of order < 0 where l!.b is the Laplace-Beltrami

operator on aQ. In particular, P is elliptic and has real principal

symbol.

The boundary conditions become Pv + atV + av = g, v = ui aQ .

This is a first order pseudo differential equation for v. It is

elliptic (because a has imaginary symbol). So it can be solved modulo

L

smoothing operators. Since 3Q is compact, smoothing operators on aQ

 125

are compact~so our boundary equation can be solved provided g

is orthogonal to some finite dimensional space of smooth functions.

Having done this, we have reduced our original problem to

the familiar Dirichlet problem nu =0 in n

which has a unique solution.
 126

CHAPTER 5

LP AND LIPSCHITZ ESTI~TES

OUR AIM IN this chapter is to study how to measure the

smoothing properties of pseudo differential operators of nonpositive

order in terms of various important function spaces. Most of the

interesting results are obtained by considering operators of order

-A with 0 ~ A ~ n. Indeed, if P E ~-A with A > nand K(x,y)

is the distribution kernel of P then, by Theorem 4.10, we know that

K E Cj m x m when j < A - n. So these operators can be studied by

elementary methods. What .

1S more, when P ~~ w-
I
A, oap ~~ w-
I
A+ lal .
So, by a proper choice of a, we can make 0 ~ A - lal~ n and then

study oap rather than P itself. Actually, we shall restrict

attention to operators of order -A where 0 ~ A < n. The

transitional case A = n requires a separate treatment to obtain sharp

results; however, for many purpose, it suffices to make the trite

observation that an operator of order -n can be regarded as an

operator of order -n + E. Further, we restrict our attention to

~OO 's whose symbols have asymptotic expansions

00

where is homogeneous of degree - A - j. (These p .'s no longer

J
belong to our symbol classes, being singular at ~ = 0, but we can

still consider the operators p. (x,D).) By the preceding remarks,

J
it will suffice to consider the operators corresponding to the individual
 127

terms in the series EP j whose degrees of homogeneity are between

o and
00
- n. Thus we are looking at p(x,D) where pIx,s) is C

on and homogeneous of degree - A in S where

oS A < n.

Since A < n, for each x, p(x,.) is locally integrable at

the origin and hence defines a tempered distribution. Denoting the

inverse Fourier transform of this distribution. by p~(x,.) we then

have

p(x,D) u (x)

(p;(x,.) * u) (x)

J p~(x, x-y) u(y) dy •

Thus we see that the distribution kernel of p(x,D) is given by

K(x,y) = p~(x, x-y).

Let us digress a little to make some remarks on homogeneous

distr ibutions.

DEFINITION 5.1 A distribution f £ S' is said to be homogeneous

of degree I.l, if < f. <l>r > = rl.l < f.<I> > for all ~ £
'I'
S where ~
'l'r
-n
is the function defined by <l>r (x) = r <I> (x/r) .

EXERCISES

l. Show that the above definition agrees with the usual definition

of homogeneity when f is a locally integrable function.

2. Show that i f f £ S' is homogeneous of degree I.l , then Da. f

is homogeneous of degree I.l - lal·

 128

3. Show that Dao is homogeneous of degree - n - JaJ (where 0

is the Dirac measure at 0).

Let us now prove a proposition concerning the Fourier

transform of homogeneous distributions. It turns out that the Fourier

transform of such a distribution is also a homogeneous one. Precisely,

we have the following

PROPOSITION 5.2 If f is a tempered distribution,homogeneous of

A

degree ~,then f is homogeneous of degree - ~ - n. If f is also

00 ~
C away from the origin, then the same is true of f.

If f is homogeneous of degree ~, then for ~ € S

,.
< f, ~r > <f, (~r) > = <f, r- n ~l/r > = r-n-~< f,~ >

r-~-n < i, ~ >

This proves the first assertion. To prove the second assertion, choose
00
~ € Co with ~ 1 in a neighbourhood of the origin and write
00
f Since is C everywhere. On the
00
other hand, (l-~)f is C and homogeneous of degree for large I;

and hence lies in S~(Rn). Let p(x,l;) = (l-~) (l;)f(I;). The

distribution kernel of p(x,D) is given by

K(x,y) = f e- 2ni {y-x).t (l-~) (I;)f(t) dl;

{(I - ~)f)"(y-x) .

00 00
Since K is C away from the diagonal, we see that {(l-~)f)'" is C

away from the origin. ijence f is COO on R n " {OJ.

 129

Returning to our operators p(x,D) with p(x,~) homogeneous

of degree -A in ~, 0 ~ A < n, by the proposition above, we have

p(x,D)U = J K(x, x-y) u(y) dy

00
where K = p" is C on lRn x :Rn " {O} and homogeneous of degree
2
A- n in the second variable.

Finally, we restrict attention to constant coefficient case,

i.e., K(x, x-y) = K(x-y). The essential ideas are already present

in this case and the results we shall obtain can be generalised to

the variable coefficient case in a rather routine fashion. Let us give

a name to the objects we are finally going to study.

DEFINITION 5.3 A tempered distribution K which is homogeneous of

00
degree A - nand C away from the origin is called a kernel of type

A. If K is a kernel of type A , the operator Tf = K * f is called

an operator of type A.

We now classify the kernels of type A ~ o.

PROPOSITION 5.4 Suppose A > 0 and f e: COO (:Rn .... {O}) is homogeneous

of degree A - n in the sense of functions. Then f is locally

integrable and defines a distribution F which is a kernel of type A.

Conversely, if F is a kernel of type A and f is the

function with which F agrees on En" {O}, then < F,<p >= Jf<P for every

<P in S.

Since A > 0, f is locally integrable and of (at most)

polynomial growth at 00, so f defines an F in S'. It is easy to

check that F is homogeneous in the sense defined above, i.e.

 130

A-n
< F, <jlr >=r <F, <jl >, and so F is a kernel of type A.

For the converse, define G by < G,<jl > =< F,<jl > -f f<jl, <jl E: S .
Then G is a distribution supported at 0 and hence we have

Therefore,

On the other hand,

-n-A
r . <G,<jl >

and this is not O(r- n ) as r tends to 00 unless < G,<jl > o. Hence

G= 0 and this completes the proof.

Suppose that f E: Coo(R n ,{O}) is homogeneous of degree -no

Then f is not locally integrable near 0 and so does not define a

distribution in a trivial way. However, let us define

J.l f J f(x)do(x).
JxJ=l

If J.l = 0 there is a canonical distribution associated with f which

f
is called principal value of f, PV(f), defined by

< PV(f),<jl > lim f

f(x) <jl(x) dx.
E: + 0 Ixl> E:

To see that this limit exists, we observe that

1
-1
J f(x) dx = J.l f J r dr o.
E: < Ixl < 1 E:

Hence

< PV(f),<jl > lim

E: + 0
f
E:<lxl<l
f(x) (<jl (x)-<j> (0) )dx + f
Jx~l
f(x)<jl(x)dx,
 131

where the last integrals are absolutely convergent, since

I$(x) - $(0) I ~ clxl, so that

J If(x) I I$(x) - $(0) Idx ~ c f Ixl-n+l dx < 00 •

Ixl < 1 Ixl < 1

Further, the estimate on $(x) - $(0) depends only on the first

derivatives of $ via the mean value theorem, so it is easily

verified that the functional PV(f) is continuous on S

Finally, we observe that

< PV(f), $r > = lim J f(x) r- n $(x/r) dx

e: -+ 0 Ix I > e:
lim I f(y) r- n $(y) dy
e: -+ 0 Iyl > (e:/r)

r
-n < PV(f), $ > •

Thus PV(f) is homogeneous of degree -n and so is a kernel of type O.

The following theorem gives a sort of converse to the above

results.

THEOREM 5.5 Suppose F is a kernel of type 0 and f is the

n
function with which F agrees on lR ,,{O}. Then Il f =0 and F PV(f) + co,

for some constant c.

Define the functional G on S by

< G,$ > = J f(x) (<P(x) - <P(O))dx + J f(x)<P(x)dx.

Ixl~l Ixl>l

By the same argument as above, G is a tempered distribution and G =F

-n
on lRn " {O}. Therefore, G - F =Ec oa o . Now, < F, <Pr > = r < F,<p >
a
so that
 132

as r tends to 00. On the other hand, for r > 1,

-n -n
< G'~r > - r < G,~ > J r f(x) (~(x) -~(O) )dx +
Ixl ~ l/r

+ J r
-n
f(x)~(x)dx - r
-n
I 'f(x) (~(x)-~(O))dx-r-n I f(x)~(x)dx
Ixl > l/r Ixl ~ 1 Ixl >1

r-n~(O) f f(x)dx r-n~(O) log r ].If' for every ~

l/r ~ Ixl ~ 1

This is not O(r- n ) as r tends to 00 unless ].I = O. Finally,

f
F - PVXf) is a kernel of type 0 which is supported at the origin and

hence is a multiple of 0

To study the boundedness of operators of type A on LP spaces,

we need some concepts from measure theory.

DEFINITION 5.6 Let f bea measurable function defined on Rn.

Then the distribution function of f is the function of: (0,00) + [0,00]

given ,by 0f(t) = IEt l , where Et = {x: If(x) I > t}.

From the distribution function of f, we can get a large amount

of information regarding f. For example, we have

00

I If(x) IP dx = - I t P d 0f(t)
o

(To see this, observe that the Riemann sums for the Stieltjes integral

on the right are approximating sums for the Lebesgue integral on the

left) •

If t P 0f(t) converges to 0 as t tends to 0 and t tends to

00
00, we can integrate by parts to obtain Ilf(x) I P dx = p J
t p - l 0f(t) dt.
o
 133

Using the concept of distribution functions, we will now define

weak LP spaces.

DEFINITION 5.7 For 1 ~ p < 00 we define weak LP

as If: 0f(t) ~ (c/t)p for some constant c}. For f £ weak LP ,

the smallest such constant c will be denoted by [f] Thus, if

P
f £ weak LP , we have

0f{t) < ([f] /t)p .

- P

00 00
For P =00 , we set weak L = L.

PROPOSITION 5.8 (CHEBYSHEV'S INEQJALITY)

PROOF For f £ LP , if Et = {x: If(x)1 > t} ,

i.e.

From this, it follows that f £ weak LP and [f]p ~ Ilf lip·

REWl.RK 5.9 It is not true that LP = weak LP for 1 ~ P < 00 .

For example, f(x) = Ixl-n/p belongs to weak LP but not LP .

RE~RK 5.10 The function satisfies [cf] = Icl[f]

f ~ [f]
P P P
But it fails to satisfy triangle inequality and hence is not a norm.

However, since

{If+gl > t}C {If I > t/2} U{I gl > t/2} ,

we have
 134

which gives, when f and 9 are in weak LP

so that

[f + gJ < 2([fJ P + [gJP)l/p < 2([fJ + [gJ ).

P- P P - . P P

The functional [.J thus defines a topology on weak LP which will

P
turn it into a (non-locally convex) topological vector space.

DEFINITION 5.11 A linear operator T defined on a space of

functions is said to be of weak type (p,q), 1 ~ P, q ~ w, if T is

a bounded linear operator from LP into weak Lq , i.e. for every

f £ LP , [TfJ q ~ c Ilf lip for some constant c independent of f.

We will now state the Marcinkiewicz interpolation theorem and

use it to prove generalisations of Young's inequality (Theorem 1.3).

THEORE M 5.12 (J. Mll.RCINKIEWICZ) Suppose T is of weak types

(po' qo) and (PI' ql) with 1 ~ Pi ~ qi ~ w, Po < Pl ' qo +ql '
i. e., ['rf J < c.
qi - ~
II f II P for
-
i = 0,1. Then, i f
i

P
T is bounded from L e to

For the proof of this theorem, see A. Zygmund [5J or

E.M. Stein [2J.

THEOREM 5.13 (GENERAL FORM OF YOUNG'S INEQJALITY) If

(l/p) + (l/q) - 1 = l/r, 1 ~ p, q, r < w ,~, for f £ LP , 9 €: Lq

and we have
 135

(In fact, c < 1 for all p,q, although our proof does not
pq-
yield this estimate).

PROOF Fixing f € LP , consider the convolution operator 9 ~ Tg = f * g.

1
We know that for 9 € L , Tg € LP and II Tg II p ~ II f II p II gill Further

if p' is the conjugate of p, then, by Holder's inequality

p'
9 £ L Thus we see that the operator T

is of weak types (l,p) and (p', 00). Therefore, by the Marcinkiewicz

Ps qs
interpolation theorem, T maps L boundedly into L for every

o< S < 1 with

(l/PS) =1 - S + (S/p') 1 - (S/p) and l/qS (l/p) - S/p.

Given r with (l/p) + (l/q) - 1 = (l/r) ,set qs = r. Then

(l/PS) =1 + (l/r) - (l/p) = (l/q). Hence we get the required result.

THEOREM 5.14 (WEAK TYPE YOUNG'S INEQJA LITY) Let 1 ~ P < q < 00

(l/p) + (l/q) > 1 and (l/p) + (l/q) - 1 = (l/r). Suppose f € LP and

9 £ weak Lq Then, we 'have

a) f * 9 exists a.e. and is in weak Lr , and

[f * g] r ~ c pq II flip [g]q

b) If P > 1, then f * 9 € Lr and II f * gil r ~ c~ Ilf lip [g]q

PROOF a) It suffices to assume that Ilf lip [g]p 1 and to show

that [f * g]r -< c pq Given 0. > 0, let

p '/(p '-q) - l/(p'-q)

M = (0./2) (p '/ (p '-q)

where p' is as usual the conjugate of p. Define

 136

gl (x) = g(x), if Ig(x) I > M

= a otherwise

and g2(x) g(x) - gl (x). Then

and we shall estimate the quantities on the right separately. By

Holder's inequality,

Since (l/q) - (lip') = l/r > 0, we see that p' - q > 0, and hence

ex>

Ilg II P: = p'
2 p
J tP'-l
a
0 (t) dt
g2

l;.\ t P '-I
p' J (Og (t) - 0 (~) dt
a 9
M
~ p' f tP'-l t-qdt = (p'/(p'-q»MP'-q= (a/2)P' .
a

Thus (f * g2) (x) exists at all points and II f * g2 1Iex>s. a/2.

Consequently of (aj2) = O. Next consider
* g2
ex> M ex>

II gl III f0 (t) dt = f 0 (M) dt + fM 09 (t) dt

a gl a 9
M ex>

~ f M- q dt + f t -q dt •
a M
ex>

The integral f t- q dt converges since q > 1, and we obtain

M

~q ~q l~
Ilglll l s M + M I(q-l) = (q/(q-l»M •

Also, by Chebyshev inequality,

 137

II f * gIll
of * 9
(a/2) < (
- a/2
p )P
1
p(l-q)
~ --
P p(l-q) ( ' / ( ' »
P ' p'-q
~ (2 laP) (q/(q-l» (a/2) p -q p p-q

-r
c a
pq

Hence (a) is proved.

b) The operator f .... f* 9 is of weak type (l,q) by (a).

Also, if we choose p > p with (l/p)+(l/q) > 1 and put

(l/r) (lip) + (l/q) - 1 then by (a), f .... f * 9 is of weak type

p
(p, r). Therefore, by Marcinkiewicz, f .... f * 9 is bounded from L
qa
into L with

l/Pa - 1 - a+ (a/~), (l/qa)- «l-S)/q) + (Sir).

Putting qa - r we get Pa - p. Hence T maps LP continuously

to Lr and II f * 9 II r ~ c ~ II f II p [g] q

CORO LlARY 5. 15 If T is an operator of type >.., 0 < >.. < n, then

T is bounded from LP into Lq , whenever 1 < P < q < 00 and

l/q - lip - >..In • Also, T is of weak type (1, n/(n->"».

PROOF If K is a kernel of type >.. , we have

IK(x) I Ixl>..-n K(x/lxl) I < clxl,,-n

which implies that K € weak Ln/(n->..) Therefore, by Theorem 5.14 (b),

if f € LP and Tf K* f then Tf € Lq where

(l/q) (lip) + «n->")/n)-l - (l/p)-(>"/n) and

~

Also we see that T is of weak type (1, n/(n->..» by Theorem 5.14 (a).
 138

The limiting case of this result with A 0 is also true, but

this is a much deeper theorem: •

THEOREM 5.16 (CAIDERON-ZYG MJND) Operators of t~ o are bounded on

LP , 1 < P < 00

Let T be an operator of type 0 with kernel K i.e.,

*
00 00
Tf = K f. To begin with, we can regard T as a map from C to C
o
00
and we shall show that f € C , 1 < P < 00 ,
o
so that T extends uniquely to a bounded operator on LP

We have K = PV(k) + co so that Tf ~ PV(k) * f + cf. Since

the identity operator is continuous, we shall assume that c = 0 and

also we shall identify K with k. The proof now proceeds in several

steps.

A A A .Itt.
Step 1. T is bounded on L2. Indeed, we have (Tf) = k f. k is

smooth away from 0 and homogeneous of degree 0, hence is bounded on ~n.

Therefore,

1
Ixl
00
Step 2. Fix a radial function 4> € C with 4>(x) 1 for ~"2
o
and <I> (x) = 0 for Ixl ~ 1. For € > 0, we define k (x) = k (x) (1-4> (x/€) )
€
and T f = k * f. Then we claim that T is bounded on L2 uniformly
€ € €
in € To see this, we observe that
,. ,.
(T f)" f k "f (k - k <I> (x/€» "
€ €

A A A A A
f k - f (k. * (4) (x/ €» )

which gives
 139

Step 3. T is of weak type (1,1) uniformly in E.

E

The proof of this is more involved and will be given later.

Step 4. By Steps 2 and 3, using Marcinkiewicz, we get that T is

E

bounded on LP for 1 < P < 2 uniformly in E.

Step 5. T is bounded on LP , for 2 < P < 00 , uniformly in E

E
00
Indeed, for f, gEe ,
0

J (T Ef) (x) g(x) dx = f f(x) (T g) (x) dx

E

k (-x) (1 - 4> (X/E) ) • Since k

E

satisfies the same conditions as k, we see that T is bounded on

E E
Lq for 1 < q < 2 uniformly in E. So if (l/p) + (l/q) = 1,

1 < q < 2,
,....
IJ(T f)gl
E IITt:*
II TEf II
P
sup 00
Ilg Ilq
~ suPoo II f II P = cll f II
P
.
gEe g E C Ilg q
0 0

00
Step 6. If fEe , T f converges to Tf in the LP norm,
0 E
l~p~oo as E tends to o. Since 4> is radial and Ilk = 0,

J 4>(x/E)k(x) da(x) = c I k(x) da(x) 0:

Ixl=r Ixl=r

so, for E ~ 1, we have

(T f) (x)
E
f (f(x-y)-f(x» k(y) (l-<P(y/E» dy +
Iyl ~ 1

+
Iy
f >1 f(x-y) k(y) dy
 140

and hence

(T f) (x) - (Tf) (x) J (f(x-y)-f(x)k(y)*(y/€)dy.

€
Iy I ~ €
Now supp (T f - Tf)
€
c:: {x: d (x, supp f) ~ t} C A, a fixed compact

set. Since, on compact sets, the uniform norm dominates all LP

norms, it suffices to show that (T f - Ti) converges to 0 uniformly

€

on A. But

f clyllyl-n dy
Iy I~ €
€J l-n n-l
~ c' IIgrad f 1100 r r dr c" € -+- 0, as € -+- O.
o
00
Step 7. If f € LP and n > 0, choose g € Co with

Then < liT € (f-g) II P + liT €9 - T.rug lip + IIT.r(g-f)

liT € f - T.ruf II p- u lip

Since n is arbitrary and liT 9 - T.rg

€ u
II P converges to 0 as E,

c tend to 0 by Step 6, we see that (T f) is cauchy in the

€
LP norm.

Setting

Tf lim T f, IITf lip ~ c II flip

€-+-o E

Thus the theorem is proved modulo Step 3.

Let us now proceed to the proof of Step 3. First, we need a

lemma which will be used in the proof.

1
IE lof.1A 5. 1 7 Suppose F € L , F ~ 0 and a > O. Then there exists

a sequence (Qk) of closed cubes with sides parallel to the coordinate

 141

axes and disjoint interiors such that

00

b) If n =U ~ , then Inl ~ (l/a) II Fill'

1

c} F(x} ~ a for a.e. x ~ Q

let r = (IIF Ill/a}l/n and for j = 1,2, ••. let ~ be the

collection of closed cubes of side length r/2 j and vertices in

(r/2 j ) ~ n. Our sequence will be constructed in the following way.

Put those cubes Q € Q1 in the sequence which satisfy

Then

2-
IQI
f F < _1_ IIF II
I QI 1
Q

so that the first condition is satisfied.

Put those cubes Q € ~ into the sequence which are not

contained in one of the previously accepted cubes and satisfy

a< I~I ~F
Inductively, put those Q€ ~ which are not contained in one

of the previously accepted cubes and satisfy a < I~I f F. If Q € ~

Q
is in the sequence and Q' is the cube in ~-l containing Q then

1 f _1 f
TOT Q F ~ I QI Q' F
F •
 142

Since Q c:. C', Q' cannot be in the sequence and hence

I~'I ~, F ~ a which then yields I~I ~ F ~ 2 na.

Thus the condition (a) is satisfied. Also,

co
L IQkl ~ a
1
L JF ~ (l/a) IIF III so (b) follows.
1 k ~

Finally, by the Lebesgue differentiation theorem,

lim I~I f F = F(x) for a.e. x.

xe:Qe:Q.,j+co Q
J

So, if x ¢ Q, I~I f F ~ a for all those Q's and hence

Q
F(x) ~a a.e. on Rn,Q

Coming back to the proof of Step 3, given f e: Ll and a > 0,

let (~) be the sequence of cubes as in the lemma with F = If I·

We write
co
f g + I bk with
k=l

f (x) -
b k (x) = {

0, otherwise

I~ I J fly) dy, for

and g (x) - { k Qk

f(x), for x ¢ Q.
Now we make the following observations:
 143

b) Ig(x) I ~ 2n a for x £ nand

Ig(x) I ~ a for a.e. x ~ n

Therefore, Ig(x) I ~ 2na a.e. and

00

We put L bk = b
1
so that T f
e:
=T
e:
g + T band
e:

I {I Te: f I > a} I ~ I { IT e: g I > a/2 } I + I { IT£

b I > a/2}
.
I

We shall show that both terms on the right are dominated by Ilf Ill/a

uniformly in £ •

To estimate the first termon the right,we use Chebyshev

inequality, Step 2 and the estimate (5.19) obtaining

To estimate the second term, let Yk be the centre of the cube Qk

and ~ be the cube centred at Yk but with length side 2/n times

n.
00

that of Qk' We put U ~ = Then

1
00 00

Inl ~ I I'Ok I = (2/n)n I I~I ~ (2/n)n Ilf Ill/a = c( Ilf II/a).

1 1
 144

So I {ITe:bl >. a/2} I ~ 1S11 + I {ITe:bl > a/2} " nl

~ c( Ilf Ill/a) + 1{ITe:bl > a/2} " nl,

and it suffices to estimate Since f b = 0,

~
T b(x)
e:
=I k (x-y) b(y) dy
e:

k (x-y) b(y) dy
e:

<Xl

Lf (k (x-y) - k (X-Yk)) b(y) dy

e: e:
1 Qk

Therefore, we have

~ (2/a) j IT bl dx
R ,IT e:
<Xl

~ (2/a)
L J- f
1 ll.,n Q
k
I (k (x-y)-k (x-Yk)) Ilb(y) Idy dx
e: e:
<Xl

(2/a) L f
1 Qk

We now claim that

t _ I (k (x-y) - ke:(X- Yk )) I dx ~ c independent of k and

R " Qk e:
for y e: Qk. Accepting this.claim, by the estimate (5.18), we have

<Xl

(2/a)c L f Ib(y) Idy ~ 2n +2 c( Ilf Ill/a).

1 Qk

Thus 1{ITe:bl > a}1 ~ cot Ilf Ill/a) which completes the proof of

Step 3.
n -
Returning to the claim, we observe that if x e: ll. "~
 145

So, if we set z = x-Y k ,w Y-Y k we

must show that

J I (kE: (z-w) - k E:(z» Idz ~ c

Izl > 21wl

independent of wand E:.

First consider the case E: = 1. Now

kl(z) is a COO function which is homogeneous of degree -n for large

z. Therefore, Igrad kl (z) I ~ c" Izl- n- l By the mean value theorem,

Ik 1 (Z-w) - kl (z) I ~ c'lwl sup Iz-twl- n - l

o < t < 1
-n-l
~ c" Iw II zl for I z I > 21wl·

So

f 1 (k l (z-w) - kl(z»ldz
Izi > 21wl

-n-l
~ c" f Iwll z I dz
Izi > 21wl

r
00

e"· Iwl
-2
~ r dr c.
21wl

Now for general E:

-n
k (z) (l-<p (z/E:» k (z/E:) E: by the homogeneity of k.
E:

-1 -1
Therefore, if we set z' = E: z and w' = E: w, we see that

J I (k (z-w)-k (z» Idz f I (k 1 (z '-w ') - k 1 (z ') ) Id z '

Iz[ > 21wl E: E: Iz 'I > 21 w' I
which is bounded by a constant, by the result for E: = 1. Hence the

claim above is proved.

 146

To complete the picture, we should observe that operators

1
of type 0 are not bounded on L (and hence, by duality, not bounded
ex>
on L). Indeed, if T is an operator of type 0 with kernel k,

k f.
A
(Tf)'" = Since k is homogeneous of degree 0, it has a
1
discontinuity at 0 (unless k = co). Thus if f E L, (Tf)
A
is not

continuous at 0 whenever "f(O) +0 and this implies that Tf is not

in Ll

This reflects the fact that if

A
If = f(O) +0, then Tf will

not be integrable near ex> , because k is itself not integrable at

ex>. However, there are also problems with the local integrability of

Tf caused by the singularity of k at the origin. In fact, let

ex>
~ E C be a radial function such that ~ 1 near 0, and set
o
Sf =f * (~k). Then the argument used to prove Theorem 5.16 shows that
A A A
5 is bounded on LP for 1 < P < ex>. In this case (~k) = ~ * k
ex>
is in C but still 5 is not bounded on

This follows from the following general fact.

PROPOSITION 5.20 If k E S' and the operator f + k * f is bounded

on is necessarily a finite Borel measure.

Choose C
ex>
with
-n
I~ = 1 and put ~ (x) = E ~(X/E).
o E

Then I~E III is independent of E , so II ~e: * k III ~ c. Therefore

there exists a sequence Ek tending to 0 such that

* k converges ~
Ek
to a finite Borel measure ~ in the weak* topology of measures and

hence * k converges to ~ in S' also. On the other hand, since

~E
k
(4) ) is an approximate identity, ~
e: k *
k converges to It in S· .
Ek
 147

Hence ~ = k and the proposition is proved.

It is easy to see that kernels of type 0 are not measures even

when truncated away from the origin, as their total variation in any

neighbourhood of 0 is infinite. One can also see directly that they

do not define bounded functionals on C

o

EXERCISE let k (x) = l/x on R and let f be a continuous,

-1
compactly supported function such that f(x) = (log x) for

o < x < 21 and f(x) = 0 for x < O. Show that

lim (f * PV(k»(x) co.

x -+- 0-

Generalise this to kernels of type 0 on ll,n, n > 1.

This example also shows that operators of type 0 do not map

continuous functions into continuous functions. However, they do

preserve Lipschitz or Holder continuity, as we shall now see.

DEFINITION 5.21 For o < a < 1, we define sup

If (x+y) -f (x) I
x,y Iyla

lI.a = {f: II filII. d~fll fllco + Ifla < co}.

a
lI.a is called Lipschitz class of order a.

RE mRK 5.22 The definition makes perfectly good sense for a = 1

(When a > 1 it is an easy exercise to show that if then

f is constant). However, we shall not use this definition for a = 1,

because the theorems we wish to prove are false in this case.

We are going to prove, essentially, that operators of type 0

are bounded on lI.a However, if k is a kernel of type 0 and f € lI.a

 148

the integral defining k * f will usually diverge because f need

not decay at 00 • Consequently, we shall work instead with

Aa n rP (1 < P < 00 ), concerning Wlich we have the following useful

result.

PROPOSITION 5.23 If f E: LP , 1 ~ P < 00 and

f E: Aa and Ilflloo ~ c( Ilfll P + Ifla)· Conseguently, LP () A is a

a --
Banach space with norm II f II p + I f Ia·

PROOF let

Ax = (~f~l) I)l/a for x E: Rn

a
Then If(y) I ~ If(x) 1/2 for all y such that Ix-yl -< A So
x

J If(y) IP dy ~ If(x) IP 2- P c' An

Ix-yl ~ Ax x

c" I f (x) Ip+ (n/a) If I- n / a

a
or

Since this is true for all x, setting e= n/pa we have

THEORE M 5.24 Operators of type 0 are bounded on

(0 < a < 1, 1 < P < 00).

PROOF Let T: f + K * f be an operator of type o. Since we know

that IITfl1 p ~ c p Ilfll p , by Proposition 5.23, it will suffice to show

 149

As in the proof of Theorem

5.16, we may assume that K = PV(k) and identify K with k.

n
Given y E: R .... {OJ and

g(x) = I k(z) f(x-z) dz

Izl ~ 31yl

h (x) I k (z) f (x-z) dz

Izl > 31yl

so that Tf =g + h. Since \1
k
= 0, we have

Ig(x) 1 = 1 I k(z) (f(x-z) - f(x» dzl

Izl ~ 31yl

f clzl-nlfl Izladz S cllfl Iyla

Izl~31YI a a

Since this is true for all x,

Next,

h(x+y) lim I
k(z) (f(x+y-z)- f(x»dz
n-+ oo 31yl <Izl<n

lim I k(z+y) (f(x-z)-f(x»dz

n -+ 00 31 y 1 < 1z+y 1< n

Therefore,

h(x+y)-h(x) lim f (k(z+y)-k(z» (f(x-z)-f(x»dz + E:l + E:2

n -+ 00 31yl < \zl<T)

where E:l and E:2 are errors coming from the difference between the

regions of integration.

E:l is the error coming from the difference between the regions

Izl < T) and Iz + YI < T).

 150

The symmetric difference between

these two regions is contained in the

annulus n- Iyl < Izl < n + Iyl

If z is in this region and

n » 1y 1 then 1z 1::::; 1z + y 1 :::: n so

that

-1
= 0 (n ).... 0 as n"" 00.

The term £2 comes from the symmetric difference of the regions

Izl > 31YI and Iz + YI> 31YI which is contained in the annulus

21YI < Izl < 41yl·

In this region Iz + yl:::: Izl ::::IYI·

Therefore,

1£21 ~ c"'
21YI <
fzl < 41yl
Iyl-n+alfl dz
a

= ci1f1a lyl-n+a«4Iyl)n - (2Iyl)n)

= C21f1a IYIQ •

Finally coming to the main term, we have

Ik(z+y) - k(z) I ~ Iyl sup Igrad k(z+ty) 1

o < t < 1

1y 1
-n-l
~ sup Iz + ty I
o< t < 1
 151

Hence, since a < 1 so that -n-l+ a < -n,

I f (k(z+y) - k(z» (f(x-z) - f(x» dzl

31YI < Izi < n

~ c f Iyllzl-n-llfl Izla dz
o 31YI < Izl < n a

~ c J Iyllzl-n-l+alfl dz
03IYI<lzl a

Therefore,

ITf(x+y)-Tf(x) I I I and consequently

~ c f '"
Iyl<l '"'

For kernels of positive type, we have the following result.

THEOREM 5.25 Suppose . 0 < A < n, 1 < P < rVA < q < Q where

Q = CD .i.f A~ 1, Q = n/(A-l), if A > 1. Let l/r (l/p) - (Ain)

and a = A - (n/q). (Thus r < CD and 0 < a < 1). Then operators

of type A are bounded from

let Tf = k * f be an operator of type A By Corollary

5.15, T is bounded frbm LP to Lr , so by Proposition 5.23, it

is enough to show that ITfia ~ c Ilf Ilq •

We have (Tf) (x) = f f (x-z) k (z) dz

(Tf) (x+y) = f f(x+y-z) k(z) dz

= f f(x-z) k(z+y) dz,

so that
 152

(Tf) (x+y) - Tf(x) =J f(x-z) (k(z+y) - k(z» dz

f f(x-z) (k(z+y) - k(z» dz +

Izl ~ 21YI

+ J f(x-z) (k(z+y) - k(z» dz •

Izi > 21YI

If q' is the conjugate exponent of q,

I J f(x-z) (k(z+y) - k(z» dzl

Izl ~ 21YI

~ Ilf II {( J Ik(w)l q ' dw)l/q' +( J Ik(z)lq'dz)l/q'}

q Iwl ~ 31yl Izi ~ 21yl

~ 2 II f II ( f I k (z) I q' d z) 1/q ,

q Izl ~ 31YI

~ clilf II ( J Izl O,-n)q' dz)l/q'

q Izi ~ 31YI

~ ci Ilf Ilqlyl-n+A+(n/ q ') = ci Ilf IlqlYla ,

by the definition of a.

For the second integral, we estimate k('z+y) - k(z) by the mean value

theorem:

Ik(z+y) - k(z) I ~ IYI sup Igrad k(z+ty) I

o < t < 1

Thus
I J f(x-z) (k(z+y) - k(z» dzl
Iz I > 21y I

~cllfllq{ f (lyllzIA-n-llq'dz}l/q'
I z I > 21 y I
~ c 2 Ilf IIqlyllyIA-n-l+(n/q ') since (A-n-l)q' < - n

= c 2'IIf Ilq Iyla •

 153

Hence

ITf (x+y) - Tf (x) I ~ c Ilf Ilq which gives

Iyla

REMl.RK 5.26 As in the preceding theorem, the reason for taking the

domain of T to be LP n Lq instead of just Lq is that the integral

defining Tf will usually diverge when f is merely in Lq •

Nonetheless, the point of these theorems is that operators of type 0 are,

in essence, bounded from Lq to A (for appropriate q and a) • To

a
make this precise without losing simplicity, one can observe that

operators of type 0 map AaO E' into Aa while operators of type

A > 0 map Lq n E' into Aa

We now introduce spaces of functions whose derivatives upto a

certain order are in LP or Aa

DEFINITION 5.27 Suppose 1 ~ P ~ 00 and k is a positive integer.

We define

We equip L~ with the norm Ilfllk = L IIDSfl1 (Thus

,p 1131 ~ k P
in the notation of Chapter 3).

DEFINITION 5.28 Suppose k is a positive integer and k < a < k + 1.

We define

We equip Aa with the norm

 154

REMARKS 5. 29

(i) f £ A if and only i f Oaf is bounded and continuous for 0 ~ I a I ~ k

ex
and Oaf A for 1131 = k. Indeed, if 1131 < k, for Iyl ~ 1,
£ ex- k

loaf(x+y)-Oaf(X) I
< c
Iyl

(by the mean value theorem) and for IYI > 1,

This shows that Oaf £ Aex- k for 0 < 1131 ~ k, i.e., f £ Aex •

(ii) If k < ex < k + 1 then is a Banach space with the

norm

This follows from the corresponding fact that LP nil.ex is a Banach

space (Proposition 5.23).

THEOREM 5.30 Suppose 0 ~ A < n, 1 < P < n/A, llr (lip) - (A!n)

and k 0,1,2, ••• Then we have:

a) Operators of type A are bounded from L~ into

b) 1! A is an integer, A = 0,1,2, ••• and k < ex < k+l,

then operators of type A are bounded from r{ (\ 11.0. to

a) This is an easy consequence of Corollary 5.15 and

Theorem 5.16, since convolution commutes with differentiation. In the

same way, (b) follows from Theorem 5.24 when A = O.

We now proceed by induction on A. Therefore, assume that

Suppose Then and

 155

is a kernel of type A - l. So, if lis = (lip) - (A-I) In , by our

induction hypothesis d j (f * k) E:
s
Lk+A- l n 1\aH-l . Since

n 1\a+A-l .
r
s < r < 00 , LS n LC: Lr , so
00
a.J (f * k) E: Lk+A- l Also

f * k E: L
r
n C1 , hence in 1\
a+A
provided it is bounded. But

dj{f * k) E: 1\a+A-l implies that dj{f * k) is bounded. By the mean

value theorem, we then have

I (f * k) (x+y) - (f * k){x) I
< c
IYI

This together with f * k E: Lr implies that f is bounded (by Proposition

r
5.21) . Hence f * k E: Lk+An 1\a+A •

The above theorem can be generalised. For example, if

o~ A < n one can show that operators of type A map 1\ (l E I

a
into

Also generalisations of the LP Sobolev spaces LP can be

k

given for non-integral values of k. In fact, a theorem due to CALDERON says

that for 1 < P < 00, f E: L~ if and only if 1\k f E: LP {Here 1\ = (I _ ~)1/2).

Therefore, we can define LP for any real s by

s

with the norm II f II s,p = II1\ s f II p

Then part (b) of the above theorem is still true for 0 ~ A < n, A

not necessarily an integer in this case.

Refer to E.M. Stein [2].

We will now prove the Sobolev imbedding theorem for LP

Sobolev spaces L~ with positive integral k. This theorem can also be

 156

generalised to s in :R.

THEOREM 5.31 (SOBOIEV IM3EDDING THEOREM) Suppose 1 < P < ~ and

k a positive integer. If k < nip , then LP Co Lr for

k
p r
llr (lip) - (kin) (Hence also Lk+jC: Lj for any j). If k > nip,

a = k - nip is not an integer, then

Let N be the fundamental solution of t:,. given by

for n +2
N(x)

Then k. (x) wn x.] Ixl -n (true for all n) is a kernel of

]

type 1.

Now, if fE:~nE"
n n
f = f * o= f * t:,.N = f * L O. k.
] ]
L (o.f *
k.
]
) .
j=l j=l ]

Suppose k = 1. If 1 < nip , OJ f E: If=> o.f

]
* k. E: Lr ]
where

llr (lip) - (lin), by Theorem 5.25 and hence f E: Lr. If 1 > nip,

·.\ .f
0]
* k
j E: Al-(n/p) by Theorem 5.25 which implies that f E: A1- (nip) •
Thus the theorem is true for k = 1.

For k > 1, we proceed by induction. Now

f E: ~ fl E I ==> f and Cl.f

]
are in LP
k-l
nE I

Therefore, if P < nl (k-l) and l/q (lip) - (k-l)/n, we have f,

d/ q
E: L , i.e. , f E: Lq1 while if p > n/(k-l), we have f,

a.f E: Ak-l- (nip) , i.e. , f E: Ak-(n/p) In the second case, we are

]

done, and in the first case, we apply the result for k = 1 to see that
 157

f is in the required space. Finally, it is easy to check th.3t i f we

keep track of the norm inequalities that are implicit in the above

arguments, we obtain

c II f Ilk ,p or

as appropr iate, for f e: ~ {\ E'. Since ~ n E' is clear ly dense

in ~, the desired result follows immediately.

We can summarise these theorems in an elegant way using the

following picture.

For - n < a < 0 we define Xa Ln/l a I and when a > 0,

a not an integer, we define X

a = Aa .
L2 A A
II
L4/3
I I ,
L4
II - . ,1/2 II ,3/2 • II

, , I , I Jo
-n 0 1 2 3 a
(The small circles represent missing spaces)

In this terminology,we have

THEOREM 5.32 Operators of type into

o ~ A< n.

THEOREM 5.33 (SOBOLEV IMBEDDING THEOREM) If

o ~ lsi ~ k, then f e: Xa +k

We now indicate how to fill the gaps in this picture at

a 0,1,2, ...

For a = 1, we d~fine

A1 = {f: f is continuous, bounded and

sup
If (x+y) + f(x-y) - 2f (x) I < oo} .
x,y /y/
 158

For k = 2,3,4, .•. we define

The sudden jump from first differences in the definition of Aa for

a < 1 to the second differences at a = 1 is less mysterious than

it seems at first, because it can be shown that if 0 < a < 2 then

f E Aa if and only if f is bounded, continuous and satisfies

If(x+y) + fIx-v) - 2f(x) I

sup
x,y Iyla

To fill the gap at a = 0 we use the space BID ("bounded mean

oscillation") first introduced by F. JOHN and L. NIRENBERG in 1961,

which is defined as follows.

1 n
For f E Lloc (lR ), we denote by mEf the mean value of f

over a measurable set E C.lR n , that is,

let Q. denote the collection of all cubes in lRn with sides

parallel to the axes.

DEFINITION 5.34 BID {f E 2-·loc (lRn ) : sup mQ(lf - mQfl) < oo}.
Q E Q.

Clear ly Loo c... BMO, for, f E Loo -=;> ImQf I .s II f Il, for every Q E Q.
and consequently

mQ( If - mQfl) .s 211f 11 00 •

It can be shown that BMO Co ~ for every

loc

If we define X for a = 1,2, ••• and X = BMO then

a o
Theorem 5.32 remains valid for all a E (- n, (0), except that the

Sobolev imbedding theorem for a =0 must be slightly modified as follows:

 159

If OSf is in the closure of BMO n EI in BMO for 1 S1 ~ k

then f E: Ak •

WARNING. BMO is not an interpolation space between LP and A

a
i.e. it is not true that i f T is a linear operator which is bounded

on LP for some p <<lO and on A for some a > 0, then T is

a
bounded on BMO.

For proofs of the foregoing assertions, see E.M. Stein [2J

and also the following papers:

(i) E.M. Stein and A. Zygmund: Boundedness of translation invariant

operators on Holder spaces and LP spaces, Ann. Math. 85 (1967),

337-349, and

(ii) C. Fefferman and E.M. Stein: HP spaces of several variables,

Acta Math. 129 (1972), 137-193.

As we indicated at the beginning of this chapter, the arguments

we have developed can be extended in a fairly straightforward way to

give estimates for ~OO with variable coefficients. We conclude by

summarising the results in

THEOREM 5.35 Let P = p(x,O) be a properly supported ~OO of

00

order - A on fl , where A > 0 and p. p '\, I with p.

J
(x,~)
j=O J
homogeneous of degree A- j for large 1~ I· Then P maps rf:m, loc}
p
Lk+A (fl, loc) for 1 < P < 00 , and in the terminology of Theorem 5.32

P maps X (fl, loc) into X ,(fl, loc) for - n < a < 00 •

a ---- a+A ---
 160

REFERENCES

1. G.B. FOLLAND:

Princeton University press, Princeton, N.J., 1975.

2. E. M. STEIN: S..i.ngulM -in:te.gJta.£.4 and d-i66vr.enti.a..bUUy pl!.opvr.:UeA

06 6unction6, Princeton University press, Princeton, N.J.,

1970.

3. M. TAYIDR: (i)p~eudo d-i66vr.ent-<.ai opvr.a.toJr.~,

Lecture Notes in Math. #416, Springer-Verlag, New York, 1974.

(ii)P~eudo d-i66vr.ent-<.ai opvr.atoJr.~,

Princeton University press, Princeton, N.J., 1981.

4. F. TREVES: BM-iC. !...i.ne.aJt paJr.:Ua.!. d-i66vr.e.nt-<.ai e.quat-iOn6,

Academic press, New York, 1975.

5. A. ZYGMUND: TJr.-igonome.tk-ic. Svr.-ieA, Cambridge University Press,

Cambridge, U.K., ~959.

6. L. HORW\NDER: L..i.ne.aJt paJr.:Ua.!. d-i6 6vr.ent-<.ai opvr.atOJr.~,

Springer-Verlag, New York, 1963.

7. W. RUDIN: Functiona!. Ana!.y~~, McGraw-Hill, New York, 1973.

8. F. TREvEs: Topologic.al VectoJt Spac.e..6, V~tJr.~buUOn4 and KMneiA,

Academic press, New York, 1967.
 TATA INSTITUTE OF FUNDAMENTAL RESEARCH

LECTURES ON ~THE~TICS AND PHYSICS

General Editors: S. RAGHAVAN and B.V. SREEKANTAN

MATHEMATICS

1. On the Riemann Zeta-function. By K. Chandrasekharan*

2. On Analytic Number Theory. By H. Rademacher *

3. On Siegel's Modular Functions. By H. Maass *

*
4. On Complex Analytic Manifolds. By L. Schwartz

5. On the Algebraic Theory of Fields. By K.G. Ramanathan*

6. On Sheaf Theory. By C.H. Dowker *

7. On Quadratic Forms. By C. I.. Siegel *

8. On Semi-group Theory and its application to Cauchy's problem

in Partial Differential Equations. By K. Yosida*

9. On Modular Correspondences. By M. Eichler *

10. On Elliptic Partial Differential Equations. By J.L. Lions *

11. On Mixed Problems in Partial Differential Equations and

Representations of Semi-groups. By L. Schwartz*

12. On Measure Theory and Probability. By H.R. pitt *

13. On the Theory of Functions of Several Complex Variables.

By B. Malgrange*

14. On lie groups and Representations of Locally Compact Groups.

By F. Bruhat*

15. On Mean Periodic Functions. By J.P. Kahane *

16. On Approximation by Polynomi~ls. By J.C. Burkill *

17. On Meromorphic Functions. By W.K. Hayman *

18. On the Theory of Algebraic Functions of One Variable.

By M. Deur ing *

* Not available.
19. On Potential Theory. By M. Brelot *

20. On Fibre Bundles and Differential Geometry. By J.L. Koszul *

21. On Topics in the Theory of Infinite Groups. By B.H. Neumann *

22. On Topics in Mean Periodic Functions and the Two-Radius

Theorem. By J. Delsarte*

23. On Advanced Analytic Number Theory. By C. L Siegel *

24. On Stochastic Processes. By K. Ito *

25. On Exterior Differential Systems. By M. Kuranishi *

26. On Some Fixed Point Theorems of Functional Analysis.

By F.F. Bonsall*

27. On Some Aspects of p-adic Analysis. By F. Bruhat *

28. On Riemann Matrices. By C.L. Siegel *

29. On Modular Functions of One Complex Variable. By H. Maass

30. On unique Factorization Domains. By P. Samuel *

31. On The Fourteenth Problem of Hilbert. By M. Nagata

32. On Groups of Transformations. By J.L. Koszul

33. On Geodesics in Riemannian Geometry. By M. Berger *

34. On Topics in Analysis. By Raghavan Narasimhan *

35. On Cauchy Problem. By Sigeru Mizohata

36. On Old and New Results on Algebraic Curves. By P. Samuel *

37. On Minimal Models and Birational Transformations of Two

Dimensional Schemes. By I.R. Shafarevich

38. On Stratification of Complex Analytic Sets. By M.-H. Schwartz

39. On Levi Convexity of Complex Manifolds and Cohomology

Vanishing Theorems. By E. Vesentini

40. On an Introduction to Grothendieck's Theory of the Fundamental

Group. By J.P. Murre

41. On Topics in Algebraic K-Theory. By Hyman Bass *

42. On the Singularities of the Three-Body Problem.

By C.L. Siegel
43. On Polyhedral Topology. By J.R. Stallings *

44. On Introduction to Algebraic Topology. By G. de Rham

45. On Quadratic Jordan Algebras. By N. Jacobson *

46. On the Theorem of Browder and Novikov and Siebenmann 's

Thesis. By M.A. Kervaire

47. On Galois Cohomology of Classical Groups. By M. Kneser *

48. On Discrete Subgroups of Lie Groups. By G.D. Mostow

49. On the Finite Element Method. By Ph. Ciarlet

50. On Disintegration of Measures. By L. Schwartz

51. On Introduction to Moduli Problems and Orbit Spaces.

By P.E. Newstead

52. On Numerical Methods fair Time Dependent Equations -

Application to Fluid Flow Problems. By P. Lascaux

53. On Optimisation - Theory and Algorithms. By Jean Cea

54. On Deformations of Singularities. By M. Artin

55. On Sieve Methods. By Hans E. Richert

56. On Irregularities of Distribution. By Wolfgang M. Schmidt

57. On Expansion Techniques in Algebraic Geometry.

By S.S. Abhyankar

58. On Torus Embeddings and Applications. By T. Oda

59. On Forms of Higher Degree. By Jun-ichi Igusa

60. On Curves on Rational and Unirational Surfaces.

By M. Miyanishi

61. On Wave Propagation. By G.B. Whitham

62. On Equations Defining Space Curves. By L. Szpiro

63. On Topics in Finite Element Solution of Elliptic Problems.

By Bertrand Mercier

64. On Diffusion Problems and Partial Differential Equations.

By S.R.S. Varadhan

65. On Numerical Methods for Non-linear Variational Problems.

By R. Glowinski
66. On Representations of Complex Semi-simple Lie Groups.
By Thomas J. Enright

67. On Nonlinear Waves and Shocks. By Cathleen S. Morawetz

68. On Topics in Stochastic Differential Equations.

By Daniel W. Stroock

69. On fuduli of Curves. By D. Gieseker

70. On Partial Differential Equations. By G.B. Folland.