Multiple Fourier Series and Fourier Integrals

I.
Multiple Fourier Series and

Fourier Integrals
Sh. A. Alimov, R. R. Ashurov, A. K. Pulatov
Translated from the Russian
by J. Peetre
Contents
Introduetion . . . . . . . . . . . . . . . . . . .
1. How Multiple Trigonometrie Series Arise . . . .
2. What Do We Mean by the Sum of the Series (I)?
3. Various Partial Sums of a Multiple Series
4. Forms of Convergenee . .
5. Eigenfunction Expansions .
6. Summation Methods . . .
7. Multiple Fourier Integrals .
8. The Kernel of the Fourier Integral Expansion
9. The Diriehlet Kernel . . . . . . . . . . .
10. Classes of Differentiable Functions . . . .
11. A Few Words About Further Developments
Chapter 1. Loealization and Uniform Convergenee
1. The Loealization Principle . . . . . . . .
1.1. On the Loealization Problem . . . . .
1.2. Loealization of Reetangular Partial Sums
1.3. Loealization of Spherieal Partial Integrals
1.4. Equieonvergenee of the Fourier Series and the Fourier Integral
1.5. Riesz Means Below the Critical Index. . . . . . . . .
1.6. Loealization Under Summation over Domains Whieh Are
Level Sets of an Elliptic Polynomial
2. Uniform Convergenee. . . . .
2.1. The One-Dimensional Case
2.2. Reetangular Sums . . . .
V. P. Khavin et al. (eds.), Commutative Harmonic Analysis IV
Springer-Verlag Berlin Heidelberg 1992
3
3
6
7
9
10
13
15
17
20
23
27
27
27
27
28
31
32
34
36
37
37
38
2.3. Uniform Convergence of Spherical Sums . . . .

2.4. Summation over Domains Bounded by the Level Surfaces
of an Elliptic Polynomial . . .
Chapter 2. L p - Theory . . . . . . . .
1. Convergence of Fourier Series in L p
1.1. Rectangular Convergence . . .
1.2. Circular Convergence . . . . .
1.3. Summation of Spherical lliesz Means
2. Convergence of Fourier Integral Expansions in L p
2.1. The Case N = 1 . . . . . . . . . . . . .
2.2. Bases and the Problem of Spherical Multipliers
2.3. Bases of lliesz Means . . . . .
3. Multipliers . . . . . . . . . . .
3.1. The Theorem of Marcinkiewicz
3.2. The Nonperiodic Case . . . .
3.3. Are Multipliers in RN Multipliers on T N ?
3.4. Pseudo-Differential Operators . . .
3.5. Fourier Integral Operators . . . .
Chapter 3. Convergence Almost Everywhere
1. Rectangular Convergence . .
1.1. Quadratic Convergence . . . . . .
1.2. Rectangular Convergence . . . . .
2. Convergence a. e. of Spherical Sums and Their Means
2.1. Convergence of Spherical Sums . . . .
2.2. Convergence a. e. of Sperical Riesz Means
Chapter 4. Fourier Coefficients . .
1. The Cantor-Lebesgue Theorem
1.1. The One-Dimensional Case
1.2. Spherical Partial Sums
2. The Denjoy-Luzin Theorem
2.1. The Case N = 1 . .
2.2. The Spherical Mean .
2.3. Rectangular Sums . .
3. Absolute Convergence of Fourier Series
3.1. Some One-Dimensional Results
3.2. Absolute Convergence of Multiple Series
3.3. On the Convergence of the Series of Powers of Ilnl
Remarks and Bibliographical Notes
Bibliography . . . . . . . . . . . . . . . . . . . .
40
41
42
42
42
43
45
49
49
50
52
54
54
56
58
60
65
67
67
67
69
72
72
73
78
78
78
78
80
80
81
82
83
83
84
87
87
91
I. Multiple Fourier Series and Fourier Integrals
Introd uction
1. How Multiple Trigonometrie Series Arise. The basie objects of study
in this article are functions of several variables whieh are periodic in each of
these variables. We may assume (this assumption is not very restrietive) that
the eorresponding periods are the same and equal to 211". Thus, we may take
the fundamental set, where our functions are defined, to be the N-dimensional
eube
1l'N = {x E RN: -11" < Xj ~ 1I",j = 1, ... ,N}.
This eube ean be identified in a natural way with an N -dimensional torus
(eiXl, . , e iXN ), where (Xl, . .. , XN) ERN),
whieh in what follows we shall not distinguish from 1l'N.
The fundamental harmonies with respect to whieh we take our expansion
have in appearanee exactly the same form as in the one-dimensional ease:
einx where, for N > 1, nx stands for the inner produet
(i. e. the subset of CN of the form
nx
= nlXl + ... + nNXN ,
nE ZN, x E RN.
Here, as usual, ZN is the set of all vectors with integer eomponents: n

(nl, ... ,nN) is in ZN ifnj = 0,1, ....
A multiple trigonometrie series
=
(1)
where the eoefficients Cn are arbitrary eomplex numbers, looks exactly as

an ordinary "one-dimensional" series. This exterior resemblanee leads one in
a eompletely natural way to the idea that the multidimensional version of
Fourier analysis is subject to the general principles whieh, in partieular, were
diseussed in the introduetory article of the series "Commutative Harmonie
Analysis" (Khavin (1988)). We run into a similar situation when we eonsider
the vibrations of astring, whieh is subjeet to the same principles as the
vibrations of eomplieated spatial objeets. This analogy, whieh may appear
superficial, has in fact a deep interior meaning.
Multidimensional harmonie analysis largely owes its appearanee to the
study of various vibrating systems. As is weIl-lmown, the early 19th eentury
witnessed new progress in the mathematieal deseription of physieal proeesses
taking place in real space, without assumptions of symmetry, whieh often
reduee the situation to functions of one variable. In the first place let us
here mention the classieal work of Fourier in the theory of the distribution
of heat (1807, the detailed study appearing in 1822), as weIl as the work of
Laplace and Diriehlet, where not only the eelebrated "method of Fourier"
was set forth but, essentially, also a eonsiderable portion of the problems of
harmonie analysis were formulated whieh eame to determine the development
of this subjeet for a long period of years.
Trigonometrical series on the realline ~l may be found already in the 18th

century in the work ofEuler, D. Bernoulli and Clairaut - mainly in connection
with the vibrating string. Now the era of "multidimensional" problems starts
- it concerns the distribution of heat and the vibrations of complicated threedimensional bodies. In the above mentioned work of Fourier the first attempts
were made to solve the problem of heat distribution in spatial bodies, in
particular, for the cube ']['3 by reducing it to the problem of expanding an
arbitrary function in three variables 1(Xl, X2, X3) in a tripie sine series, i. e. to
the problem of representing a given function in the form of an infinite series
L: L: L: b
00
1(x) =
00
00
n1 n2na
sin nl Xl sin n2x2 sin n3X3
n1=1 n2=1 na=l
The method 01 Fourier, applied with success in mathematical physics under

the course of two centuries, includes aseparation 01 variables. Perhaps, nothing describes it better than the renowned saying "di.vide and rule". (divide et
impera). One of the first problems which were effectively solved by Fourier's
method was the problem of transversal vibrations of a plane membrane n,
depending only On the expansion, and not on the curvature. The vertical distortions u = u( Xl, X2, t) of a point of the membrane with coordinates (Xl, X2)
satisfy the equation
{Pu
8t 2
= c2.u,
xE
n,t > 0,
(2)
where . = a~\ + a~2 2 is the Laplace operator. If the boundary 8n of the

"'1
"'2
membrane is stiffly fastened, then we have u = 0 along the boundary curve
for t ~ o. Usually one assumes that the position and the speed of each point
of the membrane are given for t = 0:
(3)
The method of separation of variables consists of first, neglecting the conditions (3), seeking a solution of equation (2) which is periodic in t and just
satisfies the boundary conditions (simple harmonics):
which leads to the following boundary problem for the eigenvalues and eigenfunctions of the Laplace operator:
.V
+ AV =0,
vlao=O.
xE
v2
n, A= 2'
c
When the eigenfunetions VII(X) (and thus the "speetrum of the problem",
i. e. the set of ,\'s for whieh the above harmonie oseillations with frequeney v =
c ...;>. do exist) are found, we obtain by virtue of the superposition principle,
a eonsequence of the linearity of our problem, the solution of equation (2)
subject to zero boundary conditions as the sum of the series
u(X, t)
= L VII (x)(all eos vt + bll sin vt).

11
Now we ean turn to the initial eonditions (3) and try to satisfy them by
adjusting the eoefficients all and bll , that is, we have to ehoose them in a such
a way that the following identities are fulfilled:
11
11
It is clear that the eigenfunctions vlI(x) depend on the form of the membrane {l, while it follows from the general speetral theory of operators that
the functions VII form a eomplete orthonormal system in L 2 ({l), which allows
us to find the coefficients in the expansion by the formulae
In the ease of a rectangular membrane

{l =
{x
E ]R2 : 0
< Xl < 7r, 0 <
X2
< 7r}
the variables separate also in the eigenvalue problem and we obtain (following
Fourier)
= cJn~ + n~,
nj
= 1,2, ... ,j = 1,2.
The expansion of f then takes the form
f(x)
L
00
00
~ fnln2 sinnlxI sinn2X2,
nl=l n2=1
where
f nln2
42 {21f (21f f(xI, X2) sin nixi sin n2 X2 dx l dx 2.
7r
10 10
The function g(x) admits an analogous expansion. It is not hard to see

that these expansions are special cases of double trigonometrie series.
2. What Do We Mean by the Sum of the Series (1)1 The peeularities of

the analysis on the tori ']['N in higher dimensions (N > 1), distinguishing it
from the one-dimensional situation, beeome visible at the very first attempt
to find the sum of the series (1). Of course, by the sum of (1) one has to
understand the limit of partial sums, sums of a finite number of terms of the
series.
However the speeifie character of the multidimensional ease shows itself in
the great variety of "natural" definitions of partial sums.
The situation is similar to the one which arises when we try to define a
multiple generalized integral over the entire space ]RN. If we are permitted to
seleet the expanding sequenee of eonnected domains exhausting RN arbitrarily, one ean show that the eonvergenee of the multiple integral is equivalent
to its absolute eonvergenee.
Exactly in the same way, if we take arbitrary expanding sequenees of
bounded sets Oh exhausting RN and define the partial sums of the series (1)
by the formula
Bh =
Cne inz ,
nEOh nzN
then only those series eonverge whieh are absolutely eonvergent. However,
the dass of absolutely eonvergent series, despite their importanee, is by far
too narrow. As each absolutely eonvergent trigonometrie series is uniformly
eonvergent, this dass is automatieally eontained in the space C(,][,) of all
continuous 21r-periodic functions but does not exhaust the latter.
Therefore, in order to be able to extend further the dass of multiple series,
to include series whose sums should have a eompletely determined meaning,
one has to restriet in a reasonable way the sets Oh over whieh one extends the
indices of the terms of the series entering in the partials sums. Resorting onee
more to the analogy with multiple integrals we may say that the nonexistence
of unconditionally convergent integrals nevertheless produces integrals whieh
are convergent in the sense of "principal values" (i. e. in the sense of limits
of integrals taken over expanding regular sets such as balls, eubes, ellipsoids
or parallelotopes, with restrietions on their halfaxes or side lengths).
In order to be able to better diseuss the various methods of defining partial
sums, to be set forth below, let us first consider the one-dimensional case.
As is well-known, the dassical (one dimensional) trigonometrie series
00
~ + L(aneosnx + bnsinnx)
n=l
can be written in the complex form (1) by putting

Cn
= ~(an -
ibn),
C-
n = Cn
The "natural" partial sum

k
Bk = ~ + L(aneosnx + bnsinnx)
n=l
then takes the form
Sk =
L eneinz ,
(4)
Inl:$k
i. e. it leads indeed to taking the sum in principal value sense. This defini-
tion generalizes to the multidimensional case in three obvious manners, each

leading to the following three forms of partial sums, which mainly will be
encountered within this article.
3. Various Partial Sums of a Multiple Series. A) The rectangular partial
sum Sm(x) of the series (1) is defined by a vector m E ZN with nonnegative
coordinates and takes the form
Sm(X)
L'" L
In ll:$mlln21:$m2
InNI:$mN
eneinz .
(5)
We say that the series (1) 1s rectangularly convergent (or convergent in the
sense 0/ Pringsheim) if the limit of the partial sums (5) exists for min mj -+
00.
Despite the apparent "naturality" of this definition, it is not deprived of

paradoxes. One such paradox is that there exists an uncountable set of partial
sums which can not be forced to convergence in any way. As an example we
may take the double series
00
00
L L
(6)
with coefficients en 1 n2 = n~(Dn20 - Dn2 2), where Djk is the Kronecker symbol.
The rectangular partial sums of this series at the point (0,0) take the form
It is clear that Sm(O)
for
m2 ~
2 so that the senes (6) converges
rectangularly to zero at the point (0,0), while Sml,l(O) =

ml -+
00.
ml
k=-ml
k2 -+ 00 as
This example shows also that rectangular convergence does not imply that
the coefficients tend to zero (or even their boundedness).
B) The quadratic partial sum (or cubical if N ~ 3) Sk(X) is defined by a
positive integer k and takes the form
Sk(X) =
L ... L
In ll9
eneinz .
(7)
InNI9
Iflimk-+oo Sk(X) exists we say that the series (1) is quadratically (or cubically)
convergent.
It is dear that the sum (7) is gotten as a special ease of the reet angular sum
(5) for m = (k, k, ... , k). Therefore rectangular eonvergenee implies quadratie
eonvergenee. The eonverse is not true, as is easily seen at the hand of the
double series
00
with eoefficients Cn 1n2 = n~6n20 - n~6nlO, where again 6 jk is the Kronecker

symbol. Obviously, this series is quadratically eonvergent at the point (0,0),
but not reetangularly eonvergent.
C) The circular partial sum (or spherical if N ;::: 3) 8R (x) takes the form
8R(X) =
(8)
Cne inx
Inl~R
The series (1) is circularly converyent (or spherically
if N ;::: 3) if
limR.... oo 8R(X) exists.

Rectangular eonvergenee does not imply spherieal eonvergenee. As an example we may take the above series (6). Indeed, this series is rectangularly
eonvergent at the point (0,0), but for integers l > 2 we have
81(0) =
Cn1n2
= 21 2 -
00.
n~+n~~/2
Also the converse is not true, i. e. circular convergence does not imply
quadratic convergence (and a fortiori not rectangular convergence). As an
example we may use a double series (1) with
all remaining coefficients being zero. It is dear that 8R (0) = for all R > 0.
However, the sequence of quadratic partial sums SOk = -k tends to -00 as
k-oo.
All three partial sums A)-C), dearly, coincide with (4) if N = 1. Let us
remark that any of these sums may be written in the form
Sw(X)
= L Cneinx ,
(9)
nEw
where w is some finite subset of the lattice ZN. Now, even if in the case of
spherical and cubical sums the subsets w form a family n of endosing sets,
this is not so in the case of rectangular partial sums. In fact, for an acceptable
definition it is not necessary to require that the sets wendose each other in
order to have among them a subsequence of expanding sets exhausting ZN.
These considerations lead us to the following general definitions.
D) Let 0 = {w} a family of finite subsets of ZN having the following

properties:
1) 'V w', w" E 0 3 w E 0 : w' U w" C Wj
2) UWEO W = ZN.
We define the partial O-sum 01 the series (1) by formula (9). We say that
the series (1) O-converges at the point x to I(x) if'Vc > 0 3we E 0 such
that
'VW E 0 (we C w) => 18w (x) - l(x)1 < c.
It is dear that if we take for 0 the family of rectangles (parallelotopes),
squares (cubes) or disks (balls), then we obtain the reet angular, quadratic
and circular partial sums, respectively.
If, on the other hand, the family 0 consist of all finite subsets of ZN, then
only the absolutely convergent sums are O-summable.
E) Let Q be any bounded subset of aN containing the origin and let
o consist of all sets of the form {n E ZN : n E >.Q}, where >. > O. The
corresponding partial sums have the form
8>.(x,Q) =
<:neinx .
(10)
nE>.QnZ N
In this case (1) is O-convergent if lim 8>. (x, Q) exists. In particular, if Q

>'-+00
is the unit ball or the unit cube, then 8>. (x, Q) is, respectively, the cubic or
the spherical partial sumo
Note that this convergence does not coincide, for no choice of Q whatsoever, with rectangular convergence.
4. Forms of Convergence. Everything that has been said in Sect. 3 refers,
of course, also to any multiple series and not neeessarily to the trigonometrie
series (1). On the other hand, as the series (1), which is the main object of
study for us, is a function series, one can introduce various forms of convergence. These forms differ from each other not only in the way one takes the
partial sums but also in the way they behave on ']['N. The most important
forms of convergence are uniform convergenee, eonvergenee at a fixed point,
a. e. convergence, convergence in the metric of Lp(']['N).
Let us assume that the series (1) converges in a determined sense to a
function 1(x):
(11)
I(x) =
<:neinx .
nEZ N
So far we have not put any restrietions whatsoever on the coefficients <:n,
which may have been arbitrary complex numbers. One of the first problems
which arises in the study of orthogonal series (to which also trigonometrie
series belong) is the problem of determing the coefficients. As in the onedimensional case, the solution to this problem depends on in which sense
the series converges to I. If the type of convergence allows one to integrate
10
the series term by term (for example, in the case of uniform convergence or
convergence in L p ), then upon multiplying both members of (11) with e- inx
and integrating over T N we obtain Cn = In, where the numbers
(12)
are referred to as the Fourier coefficients of I. It is clear that the Fourier
eoefficients exist only if I E L l (T N ), i. e. if I is a function summable over
T N (with respect to the N-dimensional Lebesgue measure). The Fourier coefficients In of a function I E Ll(TN ) tend to zero as Inl ---+ 00 (RiemannLebesgue theorem).
With each function I E L l (T N ) we ean thus associate amultiple trigonometrie series
(13)
I(x) '"
In einx ,
nEZ N
termed the Fourier series of I. The sign '" means that the series has been obtained in a purely formal way without any statements about its eonvergenee.
The following important problem arises now: must the Fourier series eonverge
in some sense, and if this is the ease, does it eonverge to the function I?
If the equality (11) is satisfied in such a way that term by term integration is not possible (i. e. in the ease of pointwise or a. e. eonvergenee), then
the eoefficients are in general not uniquely determined by this equality. The
uniqueness problem whieh arises in this way ean in somewhat different wording be stated as follows: does it follow from the eonvergence of the series (1)
to zero that an eoefficients are zero? In the one dimensional ease Riemann's
theorem gives an exhaustive answer to this quest ion in the ease of pointwise
eonvergenee, while the eounter-example of Menshov gives a negative answer
for a. e. eonvergenee (cf. Bari (1961)). Clearly, Menshov's example remains in
force also for N > 1. What eoneerns Riemann's theorem, then so far only an
analogue is known for N = 2: Cooke's theorem (1971) to the effeet that if a
double trigonometrie series eonverges cireularly to zero at each point then all
eoefficients must vanish. If N ~ 3 the problem of uniqueness remains open.
From now on (with the exeeption of Seets. 1 and 2 in Chapter 4), we
will restriet further diseussion to Fourier series only and we will not eonsider
general trigonometrie series.
5. Eigenfunction Expansions. In the beginning of the nineteenth eentury a
new eonnection between the theory of ~ultiple Fourier series and the theory of
partial differential equations was unvealed, namely in the subdomain of that
field which is known as speetral theory. In order to make the presentation as
transparent as possible we will begin our diseussion with the classical Laplace
opemtor
11
The operator 6. is considered in the Hilbert space L2(T N ) as an unbounded

operator with domain of definition c oo (T N).1
AB for any two functions u and v in the domain of definition holds2 ,3
(6.u,v) = (u,6.v),
(6.u, u) = -(Vu, Vu) $ 0,
the operator -6. is symmetrie and nonnegative. Consequently, by Friedrichs's

theorem (cf. Alimov, l1'in, and Nikishin (1976/77)) it has a nonnegative selfadjoint extension, which we denote by A. It is not hard to see that this
selfadjoint extension is unique and coincides with the closure of -6..
The operator A has in L2 (T N ) a complete orthonormal system of eigenfunctions
{(21r)-N/2 einx}, nE ZN,
corresponding to the eigenvalues {lnI 2 }, nE ZN. Like every selfadjoint operator, the extension A has in view ofvon Neumann's theorem a decomposition
ofunity {E~} with the aid ofwhich it can be written in the form (cf. Alimov,
l1'in, and Nikishin (1976/77))
It is easy to check that the operators
E~f(x) =
E~
have the form
fn einx ,
InI2<~
where In are the Fourier coeflicients of the function I E L2(l'N) defined

with the help of the identity (12). As we have seen, the family {E~J} called
the spectral expansion of f, coincides with the spherical partial sums of the
Fourier series (13).
Let us now consider instead of the Laplace operator an arbitrary differential operator with constant coefficients
A(D) =
aa Da ,
(14)
lal$m
where a = (al, ... ,aN) is a multi-index, i. e. an element of ZN with nonnegative coordinates, lai = a1 +.. +aN, Da = Dr 1 D~2 ... DC;:, Dj =
J
tk.
1 COO(TN) is the dass of infinitely differentiable functions on TN which are 211'-periodic

in each argument.
2 (u, v) is the inner product in L2(T N ).
Vu = (~, ... , (J8u ) is the gradient of u.

B:J:l
:J:,.
12
We associate with each differential operator (14) its symbol, an algebraie

polynomial in N variables:
A({) =
aa{a,
lai Sm
{r
where {= ({I, ... ,{N), {a = 1 . . . fft.

A differential operator is ealled elliptic if its principal symbol
Ao({)
aa{a
(15)
lal=m
is positive definite, i. e. if for any { E ]RN, { #- 0, we have A o({) > O.

If the eoefficients of the operator (14) are real, then A(D) is symmetrie,
i. e.
(Au, v) = (u,Av), u,V E coo(']['N).
If A(D) is also elliptie then by the well-lrnown Grding's inequality (cf.

Hrmander (1983-85), Vol. 111) it is bounded from below
(Au,u) ~ c(u,u),
cE ]RI.
Therefore, by Friedrichs's theorem just mentioned, A(D) has a selfadjoint

extension A and
A=
00
>'dE>.,
where {E>.} is the eorresponding resolution of identity. As in the ease of the

Laplace operator, the eigenfunctions of A are the functions (21l')-N/2 ein:.;,
while the eigenvalues equal A(n).
The spectral expansion of a funetion f E L 2(']['N) takes the form
E>.(A)f(x) =
fn einz .
(16)
A(n)<>.
Unfortunately, the eorresponding partial sum expansions do not refer to

the ones eonsidered in A)-C) and E) in Seet. 3, because the domains
are in general not similar for different values of >.. This is explained by the
possible presenee of lower order terms in the polynomial (14). The situation
improves considerably ifwe assume that the polynomial (14) ia homogeneoua,
i. e. coincides with its principal symbol (15). In this case, putting
13
we get the expansion E>..(Ao)f which coincides with the partial sums already
known to us in E) in Sect. 3:
It might appear that the study of the spectral expansions E>..(A) h~ a

very special character, as it refers only to operators defined on T N However,
analogous reasonings may be carried out also for the operator A(D) considered on an arbitrary domain n c ]RN. In this case one has in general infinitely
many selfadjoint extensions and the corresponding resolutions of unity E>.. do
not have any longer the form (16). However, it is remarkable that the spectral behavior of all the different selfadjoint extensions of the one operator
A(D) is basically the same and mimics the one of (16). The mathematical
formulations of this not completely dearly formulated statement comprises
the union of a great number of theorems known as equiconvergence theorems.
The uncontested importance of these theorems is seen by the fact that they
allow one to reduce the study of the immense variety of spectral expansions
to the study of one object: the multiple Fourier series (or integral). A characteristic feature of equiconvergence theorems is that their validity usually
involves a dass of functions: for each function f in some dass W we have
From this one can condude that if f E W then the convergence of E>..f
holds if and only if E>..f converges.
However, as often is the case, precisely this feature turns out to be a defect.
Namely, in many situations equiconvergence is established for precisely those
dasses one has the convergence of E>..f and, therefore, of E>..f. In those dasses
where E>..f does not converge one does not have as a rule equiconvergence
and therefore one cannot say anything about the behavior of E>..f. In other
words; equiconvergence theorems apply effectively when it is required to prove
convergence but they are somewhat harder to apply in divergence proofs.
Another important fact is that the "lower order" coefficients aa, lai< m,
in (14) do not infiuence the convergence of the spectral expansion E>..f of
the operator A(D), provided the function under view is sufficiently smooth.
In other words, one has equiconvergence not only for expansions of one and
the same operator but also for expansions of different operators Al and A 2 ,
again provided only their principal parts coincide. Using this fact one can,
in particular, reduce the study of the partial sum (16) to the study of the
simpler expansion E>..(Ao)f.
6. SUDlIllation Methods. As we have already noted in Sect. 4, one of the
main problems of harmonic analysis is the reconstruction of functions from
their expansion. The definition of the sum of a Fourier series as the limit of
partial sums does not always solve this problem. The point is that for not
14
very smooth functions (and exactly such functions have the most interesting and from the practical point of view important expansions) successive
strongly oscillating terms of the partial sums have a very big influence, which
does not correspond to their unsignificant rle in the characterization of the
function to be reconstructed. As a result the sequence of partial sums does
not approach the function but oscillates around it. However, these oscillations
have, as a rule, a regular behavior, so taking, for example, arithmetic means
of the partial sums we may expect that these means better approximate the
function under consideration. Mathematically this is equivalent to taking, in
the calculation of the partial sum, each term of the series with a weight which
decreases in size as we increase the index.
This heuristic device works in all cases where the partial sums are parametrized by a natural index, for example, for quadratic sums. Let, say,
be the partial sum of a double series. Then, after some simple transformations,
it is not hard to see that
where o:~ln2 = 1- max(lnll, In 21).

In particular, if nl = l - 1 or n2 = l - 1 then o:~~ n2 = t. The means 00/
usually behave better than Bk as l--+ 00.
We may proceed in an analogous manner when the partial sums are defined
by an continuous parameter, for example, in the case of the spectral resolution
E>.. In this case it is natural to define integral means as follows:
00>.1
I>'
"X 10 Eddt.
Integrating by parts, the last formula takes the form
Each time that the partial sums E>.I converge the mean 0) will also converge to the same value. The converse is not true. For example, if E>.I oscillates asymptotically as cos Athen 00).. tends to 0 as sin A. This leads to the
hope that one might be able to sum the series with the aid of the means 00)..
in cases when the partial sums do not have a limit.
It is natural to expect that the chances for succcess increase if we integrate
E>. several times. After an s-fold integration of Ed over the interval [0, Al
15
and a subsequent normalization we arrive after some transformations at the

quantities
EU
= 1>'
(1- ~)
dEd,
(17)
which are known as the Riesz means of order s. We say that the speetral
expansion (16) is summable to f(x) by the Riesz method of order s if
!im EU(x)
>'-+00
= f(x).
Let us note that the integral in (17) makes sense for any real s ~ 0 and
even for eomplex s with Re s ~ 0, thanks to which one ean use interpolation
theorems in the study of Riesz means.
If A is an elliptie operator then E>.f is given by (16). Inserting this into
(17) gives
EU(x) =
A(n)<>.
(1- A~n)r
fn einx .
(18)
In the ease of the Laplace operator the Riesz means take an especially
simple form:
o-U(x)
InI 2 <>.
(1- 1~2r
fn einx .
(19)
These Riesz-Bochner means for spherieal partial sums were first studied in.
detail by Boehner (1936).
The Riesz means may be viewed as a regularization of the partial sums, and
in many eases their asymptotic behavior gets better when Re s is inereased.
The regular summation methods in the ease of partial sums defined by
several parameters, for example reet angular partial sums, are somewhat more
diffieult to deal with. In this ease one ean likewise define means (cf. Zygmund
(1968), Vol. II, Chapter XVII).
7. Multiple Fourier Integrals. Exactly as in the classical (one dimensional)
ease trigonometrie series are intimately connected with Fourier integrals,
likewise multiple Fourier series admit a continuous analogue - the multiple
Fourier integrals.
If fELl (lRN ) then its Fourier transform is defined by the formula
(20)
The function can be reeovered in a unique way from its Fourier transform
by the formula
(21)
16
However, in contrast to (20) this formula does not make sense for all 1 E
L1(lR N ).
The formal requirement of the eonvergenee of the improper integral (21),
equivalent to the eondition j E L1(lRN ), restricts eonsiderably the dass of
functions 1 whieh ean be expanded in a multiple Fourier integral. Therefore,
there arises the question of the special eonvergenee of the improper integral
(21), analogous to the eonvergenee problem for multiple trigonometrie series.
As in Seet. 3D), let us eonsider a family of bounded sets W C lRN enjoying
the following properties:
1) \f WI, W2 E n 3 wEn: WI U w2 C Wj
2) UWEOw = lRN .
We define the eorresponding partial integral by the formula
(22)
We say that the integral (21) converges to I(x) with respect to the lamily
n if
It is dear that this indudes rectangular, quadmtic (cubic) and circular

(spherical) convergence 01 the integml (21).
If Ais a selfadjoint elliptie operator in L 2 (lR N ) generated by the differential
expression (14) and if E).. is the eorresponding speetral resolution, then the
spectral expansion of any element 1 E L 2 (lR N ) has the form
(23)
This formula needs some explanations, beeause we have defined j only for
1 E L1(lR N ). If 1 E L 2 (lR N ) then the integral (20) need not exist in Lebesgue's
sense so there arises the question of a workable definition of j. The key to
this problem is provided by Plancherel's theorem aceording to whieh for each
function 1 E LI (lRN ) n L 2 (lR N ) we have Parseval's lormula
This means that the Fourier transform is an isometry and so, being defined
on the dense subset LI (lRN ) n L 2 (lR N ) of the Hilbert space L 2 (lR N ), it admits
a unique extension to the whole of L 2 (lR N ) by passing to the dosure. This extension assigns to each function 1 E L 2 (lR N ) its Fourier-Plancherel tmnslorm,
whieh eoincides with the usual Fourier transform when 1 is summable. 4
4 If I is locally summable then j can be defined in the sense of the theory of distributions.
If I E LI or I E L2 then this definition coincides with the previous one.
1. Multiple Fourier Series and Fourier Integrals
17
The process of closure can be performed in several concrete ways, i. e. with

the aid of different sequences fn E L 1 (lR N ) n L 2 (lR N ). The simplest is to put
fn(x)
= Xn(x)f(x),
where Xn(x) is the characteristic function of expanding sets W n C lRN which

exhaust lRN . Then, irrespective of the choice of W n , the sequence (cf. (22))
converges in the metric of L 2 (lR N ) to a nmction j(~) which we take as the

Fourier transform of f. Usually one takes for W n concentric balls or cubes.
For the spectral expansion (23) we can, as in the case of series, define Riesz
means of order 8 ~ 0:
In Sect. 5, speaking of equiconvergence theorems, we noted that the spectral expansions E>.f, corresponding to different selfadjoint expansions of one
and the same operator A(D), behave on the whole in a unique way and so
make it sufficient to study one of these expansions in order to be able to
make conclusions about behavior of all the others. Thus, the most convenient object for a detailed investigation are the expansions of the form (23)
or their Riesz means (24). In the following section we shall try to elaborate
this assertion.
8. The Kernel of the Fourier Integral Expansion. The partial integral
Iw(x, f) defined in (22) may be transformed writing instead of j the integral to the right in (20) and then changing the orderof integration. This
yields the formula
Iw(x,f) =
f K(x JRN
y,w)f(y)dy,
(25)
with
(26)
Thus, every partial Fourier integral is an integral operator whose kernel
depends on the difference. It is aremarkable fact that the kernel (26) is the
Fourier transform of the characteristic nmction X(x) of the set w.
Let us assume that the set w has the form
18
where J.t
> 0 and Q is a suitable bounded neighborhood of the origin.
Then
i. e.
K(x, J.tQ) = J.tN K(J.tx, Q).

We see that one can get complete information about the kernel K(x, J.tQ)
by studying the integral
(27)
as a function of x. This function is continuous and further analyticj it is clear
that it is subject to the estimate
IK(x,Q)I::::; IK(O,Q)I = (21r)-NIQI

At infinity K(x, Q) tends to zero oscillating. The way it tends to zero depends
on the geometry of the set Q and determines, in turn, the behavior of the
partial sums of the Fourier integrals.
If A(D) is a homogenous elliptic polynomial of order m, i. e. A(te) =
t m A(e), then the integral (23) belongs to the previous dass with the domain
Q ofthe form
(28)
Thus, the behavior of the spectral expansion corresponding to the operator
A(D) is dosely connected with the geometry of the levellines of the polynomial A(e). This remarkable observation has stimulated the study of Fourier
transforms of characteristic functions of various sets. One of the most powerful methods in the study of oscillatory integrals is the method 0/ stationary
phase. The elaboration of this method has led to the discovery of the following rule: the "more convex" the set Q is, the faster the function K(x, Q)
decreases as lxi - t 00. The most convex set is the ball, and therefore the
function
(29)
has optimal behavior.
The integral in the right hand side of (29) (Titchmarsh (1958)) is computed
by passing to a system of spherical coordinates and turns out to be
(30)
where JI/(t) is the Bessel function of first kind and order v. As
1.1,
1/
(t)1 < const
0
- 0 ' t >,
(31)
19
it follows from (30) that
IK(x)1
const Ixl-----r-.
N+!
(32)
This is the best estimate for functions of the type (27), which is uniform
in all directions.
Of coUrse, in special directions, depending on the geometry of Q, one may
improve the estimate (32). Let, for example, Q be the cube
Then clearly
K(x,Q)
= rr- N
II smx
k=l
If x
-+ 00
then along the diagonal
Xl
IK(x)1 ~
X2
Xk
= ... =
XN
const
IxI N
'
which, apparently, is better than (32), but if x

then the estimate
const
IK(x)1 ~ Txf'
-+ 00
along a coordinate axis

(33)
is sharp, i. e. the kernel oscillates rather slowly. Of course, one can also construct more exotic examples of sets Q for which the kernel K(x, Q) oscillates
even slower than (33), but for a large class of sets of type (28) the uniform
estimates are included between (32) and (33).
Useful objects in this study are the elliptic operators of the special type
N
Am(D) = LDr, m = 2m',
(34)
k=l
which were first considered in detail for this very purpose by Peetre (1964).
The sets Qm corresponding to these operators have the form
(35)
If m = 2 the set Qm is the N-dimensional ball and therefore the spectral
expansion (23) coincides with the spherical partial integrals. If m increases
the boundary flattens, and as m -+ 00 the set approaches a cube and the
spectral expansion (23) goes over into the cubical partial integrals. These
examples convince us that the order of decrease of the Fourier transform of
the characteristic function of a set Q depends on the convexity of its boundary
20
BQ. As a numerical measure for the convexity of BQ we may take the number
ofthe non-zero principal curvatures at each point of BQ (cf. Ashurov (1983a)).
Indeed, if the boundary BQ of the domain (28) has r (0 ~ r ~ N - 1)
non-zero principal curvatures then the Fourier transform (27) satisfies the
estimate
(36)
Here we assurne that Q is convex.
Let us recall the simplest definition of principal curuatures of an (N - 1)dimensional hypersurface M at the point x E M. Let TxM be the tangent
plane and let YN = f(Yl, . .. ,YN-l) be the equation of M in a neighborhood of
x, (Yt. Y2, ... , YN-t) E TxM. Then the eigenvalues of the matrix
{~::~~}
are called the principal curvatures of M. Note that the principal curvatures
are invariants of M, i. e. they are independent of the coordinate system
considered.
One can obtain a more accurate measure of K(x, Q) if one takes into
account the order with which the principal curvatures vanish, which however
constitutes a much harder issue.
9. The Dirichlet Kernel. The discussion in the previous section suggests to
study in an analogous way the behavior of partial sums of multiple Fourier
series. Let us turn anew to Example D) in Sect. 3:
Sw(x,f) = L:fn einx .

nEw
If we replace the coefficients f n in this sum by their values given by formula

(12), we obtain
Sw(x, f)
= [
JTN Dw(x -
y)f(y)dy,
(37)
where
(38)
nEw
Thus, each set w E ZN defines an integral operator with a kernel which

depends only on the difference. The behavior of kerneis of the form (38) can
be studied by comparing them with the functions (26) (the proofs of most
of the equiconvergence theorems are based on this fact). For instance, in the
case of spherical partial sums it is, taking into account (30), natural to expect
that
'"' inx (2 )-N/2 .~J~ (JLlxl)
(39)
L.J e '" 'Ir
JL
lf
Inl~1l
lxi
The sign '" means that the ratio these two quantities in some sense behaves
in the same way for large values of JL. In order to give this assertion an exact
meaning we must estimate their difference, which is a far from simple issue.
21
Let us note that the comparison ofthe kernels (38) and (26) has an interest
of its own, independent of the problem of convergence of partial sums of
Fourier series. For example, the comparison (39) is extremely important in
number theory even for x = 0:
(40)
To the left stands the number of points of ZN contained inside a sphere of
radius JL. This can also be interpreted as LkQ,2 l/k(N), where l/k(N) is the
number of solutions of the Diophantine equation
In the case N = 2 estimating the difference of the quantities in (40) constitutes the so-called circle problem in number theory.
When comparing the kernels (26) and (38) we should bear in mind that
the corresponding integral operators act on different function spaces: on the
space LI(l~N) of summable functions on ]RN and on the space L1(']['N) of
functions on ]RN which are 2rr-periodic in each argument and summable on
'JI'N. It is dear that these dasses are different and that their intersection
consists of the zero function only.
Let us assume that fELl ('JI'N) vanishes near the boundary of the cube
'JI'N. Then the function
g
(x) = {f(X),
0,
xE 'JI'N,
X ~ 'JI'N
is in LI (]RN) and preserves all properties of f in the interior of'JI'N. Conversely,

if gELl (]RN) has its support in the interior of 'JI'N, then if we shift its graph
along the coordinate axes with steps which are multiples of 2rr, we get a
periodic function f E L1('JI'N) which coincides with g on 'JI'N, i. e.
f(x) =
g(x + 2rrn).
(41)
nEZN
Clearly, the Fourier coefficients f n of f coincide up to a factor with the

Fourier transform g of g taken at the point n:
In this case we have
f(x) =
nEZ N
fn einx = (2rr)-N/2
nEZ N
g(n)e inx ,
22
so comparing this with (41) we obtain
g(x + 27m) = (271')-N/2
g(n)e inx .
(42)
The equality (42) is called Poisson's summation formula. The preceding

reasoning has a heuristic character and cannot serve for the foundation of
(42). However, if the series to the left and to the right are uniformly convergent for x E T N , it is not hard to show the validity of the formula. Then it
is not necessary to assume that the function gELl (T N ) vanishes off TN.
For instance, if gis such that it and its Fourier transform are subject to the
estimates
Ig(x)1 ~ const(1 + Ixl)-N-e,

Ig(~)1 ~ const (1 + IW- N -
e,
> 0,
(43)
then the series converge absolutely and uniformly and (42) is in force.
The Poisson summation formula is very effective in the solution of various
problems in harmonie analysis, for instance, in the summation of multiple
Fourier series. As an illustration consider
K(x) =
<p(x + 271'n),
(44)
nEZ N
where <p satisfies estimates of the type (43). Then by Poisson's summation
formula we get
K(x - y) = (271')-N/2
cp(n)ein(x- y ).
nEZ N
Upon multiplying both sides of this equality by fELl (T N ) and integrating over T N we obtain:
* f(x) =
(271')-N/2
cp(n)fneinx,
(45)
nEZN
with the convolution K
*f
defined in the well-known manner:
* f(x) =
1
yN
K(x - y)f(y)dy.
If <p(x) is the characteristic function of a set w, then formula (45) gives

an expression for the corresponding partial sums of the Fourier series of f.
Unfortunately, the Fourier transform of characteristie functions of bounded
sets does not oscillate sufficiently fast in order to allow us to apply the Poisson
23
summation formula. It is a different thing if we "smoothen" the function t.p,

i. e. if we pass from the partial sums to their means.
Consider for example the function
(46)
Then
t.pl'(x) = (21r)-N/2 [
JI~I<I'
(1- I~~)S eix~df..

J.L
The integral is computed by passing to a spherical coordinate system and

equals
t.p/l-
(x)
= (21r)-N/22 f(s + 1)
S
J.L
N J ~+s(lxlJ.L) .
(lxlJ.L)~+s
(47)
Using an estimate for the Bessel function it follows that
so if s > N;1 then t.pl' satisfies an estimate of type (43). Therefore (45) is
applicable to t.pl" for which it takes the form
(48)
with
(49)
The formulae (48) and (49) give a representation of the Riesz means of
spherical partial sums for s > N ;1. The order s = N ;1 was termed the
critical order by Bochner, because many properties of the Riesz means for
higher orders faH for s = N ;1. The behavior gets even worse if s < N ;1. In
this case for uniform convergence, say, one has to invoke classes of smooth
functions, about which we will say a few words in the following section.
10. Classes of Differentiable Functions. The classes of differentiable functions in several variables which have been studied most are the Sobolev classes
W~(G) (Sobolev (1950)), where 1 ::; p < 00, l = 1,2, ... and G C JRN.
A function f E Lp(G) belongs to W~(G) if all partial derivatives DOlf (in
the sense of distributions) of order lai = l are in Lp(G), i. e. if the norm
Ilfllw~ = IlfllL +
p
L IIDOl fllL
IOlI=1
(50)
24
is finite.
If N = 1 membership of 1 in W~ means that 1 has 1 - 1 continuous
derivatives and that 1(1-1) is absolutely continuous with 1(1) belonging to L p
Simple examples reveal that for higher dimensions 1 E W~, in general, does
not even imply that 1 is continuous. However, if p and 1 are sufficiently large
then each function in W~ is continuous, i. e. the space W~ (G) is continuously
imbedded in C(G). Namely, if pl > N then the identity operator
I: W;
-t
is continuous (and even compact). This statement is the content of one of the
famous Sobolev imbedding theorems(Sobolev (1950)).
The classes W~(lRN) can also be described in terms of the Fourier transform. Namely, 1 E W;(RN) if and only if there is a (uniquely determined)
function 9 E Lp(RN ) such that
(51)
An equivalent norm in W~(RN) is provided by the quantity
(52)
That the norms (50) and (52) are equivalent can easily be checked for
= 2. Indeed, if p = 2 then (50) is equivalent to the norm

II/II~ = 1I/11L +
lal=l
11 Da IIIL
As (Dai)(~) = ~a j(~) it follows from Parseval's formula that
11/111 ~
(1 + 1~,le.12) 1/2 i<eJ

L2
while the norm in the right hand side is equivalent to the norm (51)-(52), as
clearly
Cl (1 + 1~12)1 :::; 1 +
I~a 12:::; c2(1 + 1~12)1.
I:
lal=l
If p =f:. 2 the proof of the equivalence is considerably harder.

In order to simplify the notation it is expedient to introduce the direct
and inverse Fourier transforms:
(53)
25
(54)
Then the definition of the norm (51)-(52) ean also be written as
Let us remark that the exponent 1 in (50) has to be a positive integer,

beeause it is not eompletely clear which sense has to be given to the norm
for nonintegrall. The situation improves if we look at the equivalent norm
(55). In this definition we ean take for 1 any positive number. The class of
functions f E Lp(IRN ) whieh for a given fixed number 1 > 0 make the norm
(55) finite is termed the Liouville class 5 L~(IRN). Thus
(56)
The classes L~ eoincide for 1 integer with the Sobolev classes W~ and eonstitute a natural eontinuation of the latter to noninteger values of l. Besides,
they are natural precisely from the point of view of harmonie analysis. If we
however want to introduee fractional derivatives, having in mind the exact
deseription of the traces of functions in Sobolev classes to subspaces of lower
dimension (and exactly these quest ions are the most appropriate in the theory of boundary problems for partial differential equations, for whose solution
also the classes W~ were first introdueed), then the most natural classes are
the Sobolev-Slobodetskiz classes.
Another approach to the problem of function spaces with a fractional
smoothness exponent has been developed in approximation theory. Here the
Nikol'skiz classes H~('II'N) are the most natural. They eonsist of those functions which admit an approximation in the Lp('JI'N)-metric by trigonometrie
polynomials of degree n 6 with remainder O(n- l ). Finally, a very general approach to this problem is provided by the Besov classes B~8 with an auxiliary
parameter (): if p = () these are the Sobolev-Slobodeckiz classes, if () = 00 the
Nikol'skil classes and if p = () = 2 the Liouville classes L~ (Nikol'skil (1977)).
In order not to eomplieate the presentation we will in the sequel limit
ourselves to the Liouville classes L~, justifying this ehoice by the fact that all
the above classes approximately eoincide with L~ (Nikol'skil (1977)):
where A~ denotes any of the classes of Sobolev-Slobodeckil, Nikol'skil or

Besov (the latter with arbitrary ()).
5
6
Translator's Note. Also known as Bessel potential dass.

That is, sums of the form L:1kl$n akeik"'.
26
If f E L~ then dearly f belongs to a Liouville dass for any smaller l, but,

even more important, then the summation exponent p can be improved. Let
us illustrate this at the hand of the function
It is not hard to see that for all a with lai = l the derivative Da f has the
singularity Ixlt:-l. Therefore Da f E L p , i. e. f E W~, if Ixlt:-l E L p , i. e. if
l- N/p < e.
Consequently, the function under consideration belongs to a whole family
of spaces W~ for which the quantity l- N / p has one and the same value. This
quantity characterizes the integrability of the top derivatives in (50), and l has
to do with the increase of the singularity which arises when differentiating,
while N / p is the decrease coming from the contribution (the larger the higher
is the dimension) introduced in the integral from the Jacobian.
These elementary heuristic considerations, arising by considering the simplest functions of a rather special type, have nevertheless a very general
character. It turns out that every Liouville dass L~ is imbedded in another
such dass L~ll with lt < l, provided only that the quantity l - N/p is the
same (cf. Nikol'skir (1977)). In other words, we have the imbedding L~ ~ L~l
provided
l- -
= 11 -
i. e.
11 = 1- N
-,
P1
(~-~),
p
P1
l > lt,
< P1.
(57)
The periodic analogue L~('fN) of L~{lRN) is obtained if we replace the

direct Fourier transform in (56) by the Fourier coefficient and the inverse
transform by the series:
IlflIL~
(1 + InI 2)1/2 fn einz
(58)
nEZ N
The local properties of functions in L~('fN) and L~{lRN) are the same:
if TJ E CIf('fN) then the sets {JTJ : f E L~('fN)} and {JTJ : f E L~{lRN)}
coincide.
Besides the spaces L~ we require also the dassical Hlder spaces Cl, which
for all positive integers I consist of all functions f having continuous partial
derivatives of all orders a with lai ~ I. For fractionall > 0 the dasses Cl are
usually defined as follows: if I = m + K., where m = 0, 1, ... , 0 < K. < 1, then
f E Cl if and only if
IDaf(x) - Da f(y)1 ::; const 'Ix -
yl~
27
for every 0: with 10:1 = l. As usual, CI(']['N) denotes the dass of 21l"-periodic
functions satisfying the above conditions.
The function f(x) = IxI E , 0 < c < 1, is a typical representative of a
function of dass Cl. We remarked already that this function belongs to every
Liouville dass L~ with l - N / p < c. This fact leads to the idea that there
must be a connection between the Liouville and Hlder dasses. Indeed, such a
connection is provided by the following imbedding theorem (Nikol'skit (1977)):
L~ -+ CE
N
where e = l - - > o.
p
(59)
The meaning of this theorem is that the existence of derivatives of high

order which are summable to a large power guarantees for a function its
smoothness in the dassical sense.
11. A Few Words About Further Developments. First we say a few words
about convergence of multiple Fourier series and integrals in the most important metrics - the uniform metric and the L p metric. Then we shall discuss
the problem of a. e. convergence, which is one of the most attractive problems
in the metric theory of functions. The problem of L p convergence is dosely
connected with the theory of multipliers, the development of which has been
one of the decisive motives for the introductions of pseudo-differential operators, and then Fourier integral operators.
Many important results, which are technically hard to formulate, fall outside the framework of this exposition. Adesire to adhere to the style dictated
by the intentions of this Encydopaedia, attempting to provide a general first
introduction to the subject, has forced us to put less emphasis on quest ions
of priority, rather concentrating on the highlights of the, in our opinion, most
important developments in the theory of multiple Fourier series.
To areader who wishes to penetrate deeper into the theory of multiple
series we recommend the books Stein and Weiss (1971), Chapter VIII and
Zygmund (1968), Vol. 11, Chapter XVIII, and further the surveys Alimov,
Il'in, and Nikishin (1976/77) and Zhizhiashvili (1973).
Chapter 1
Localization and Uniform Convergence
l. The Localization Principle
1.1. On the Loca1ization Problem. The dassical Riemann localization theorem states that the convergence or divergence of a "one-dimensional" Fourier
series at a given point depends only on the behavior of the function fELl
28
in an arbitrary small neighborhood of this point. Let us indicate the plan of

the proof of this theorem.
As is well-known, the Dirichlet kernel
Dn(x) =
Le
ikx
Ikl$n
in the one-dimensional case has the form

( ) _ sin(n + ~)x
Dn x 2'
sm 2x .
Let us fix a number 6,0< 6 < 71", and set Qs = {x E Tl,lxl > 6}. It
follows then from the summability of f that the function f (x + y) . (sin ~)-l
is summable in the domain Qs. Therefore we have
[ f(x + y)Dn(y)dy - t 0,
ins
- t 00,
uniformly in x E [-71",71"]. Therefore the partial sum of the Fourier series of f

is of the form
11
Sn(x, f) = - s f(x + y)Dn(y)dy + 0(1),

71" -s
from which the localization theorem follows.
Another formulation of the localization principle is: if two functions coineide in the neighborhood of a point, then the partial sums of their Fourier
series have the same behavior there. Taking the difference of these functions
we obtain an even more compact formulation: if f vanishes in a neighborhood
of the point then Sn(x, f) - t 0 for n - t 00.
In the multidimensional case the localization prineiple does not hold in the
class LI (T n ), neither for rectangular nor for spherical partial sums, which is
connected with the unboundedness of the Dirichlet kernel of a neighborhood
of the origin. In order to maintain localization one has to restriet the class of
functions under consideration, by raising their smoothness. Then the classes
are different depending on how the partial sums are formed.
1.2. Loca1ization of Rectangular Partial Sums. It is easy to establish the
non-existence of localization, say, for double Fourier series in the class of
continuous functions C(T2 ). It suffices to take (Tonelli (1928)) the function
where l1(xI) vanishes for lXII< 6 and 11 E C(T) is such that the sum of
partial sums Sml (Xl. 11) of its Fourier series at the point Xl = 0 is different
29
from zero for infinitely many indices ml (of course, then Sml (0, fd -+ 0,
ml -+ 00, by the classical Riemann localization principle). The second function !2 E C(,][,) is chosen such that for X2 = the sequence of its partial sums
Sm2(0,!2) is unbounded (the existence of such a function was first proved
by Du Bois-Reymond, cf. Bari (1961. Then the function f(XI,X2) E C(']['2)
thereby defined vanishes in a 8-neighborhood of the origin (0,0) (and even
in the strip {x E ']['2 : lXII< 8}), but the sequence of its rectangular partial
sums
Smlm2(0,0,f) = Sml(O,fd Sm2(0,!2)
at the point (0,0) is unbounded, which can be seen by choosing for each index
such that Sml (0, fd '# the corresponding index m2 sufficiently large.
This example extends in an obvious manner to an arbitrary number of
dimensions N > 2. In general, we remark that every example of a function
of N - 1 variables with a divergent Fourier series yields at the same time an
example of a function in N variables with an analogous property.
The first attempt to find an acceptable analogue of the localization principle for rectangular sums was made by Tonelli (1928). It was based on the
following observation: the rectangular partial sum Sm(O, f) is an integral operator
Sm(x, f) = - 1
Dm(x - y)f(y)dy,
(1.1)
ml
71'n
1TN
where
rr
N
Dm(x) =
Dmj(xj),
(1.2)
j=l
Dmj (Xj) being the "usual" Dirichlet kernel. The singularity of Dm (X) extends
over the subspaces Xj = 0, the kernel D being unbounded in a 8-neighborhood
of these subspaces. Therefore, if f(x} = in such c5-neighborhoods, then
Sm(x, f). -+ as min mj -+ 00. In particular, Tonelli's theorem guarantees
that the rectangular partial sums of the double sum at the point a = (al, a2)
of a function f E L l (']['2) vanishing in a "cross-shaped" 8-neighborhood
converge to zero.
Of course, this is not the principle of localization in its usual sense, as for
the convergence of the partial sums of the Fourier series of the function f one
has to impose a condition on f at points which are far away from the point
under consideration. A naturallocalization principle requires that one passes
from the class LI (']['N) to smoother functions.
It has tumed out that the Nikol'ski'l classes H!(']['N) (which are slightly
larger than the Sobolev and Liouville classes) are most suitable for this purpose. In these classes V. A. Il'in (1970) has found the exact conditions for the
localization of rectangular partial sums:
pi > N - 1,
> 0, P ~ 1.
(1.3)
30
Namely, if (1.3) is fulfilled, then for any function f E HtCH'N), vanishing in

some domain n C 'lrN , the rectangular partial sums Sm (x, I) of the multiple
Fourier series of f converge uniformly to zero on each compact set K c n.
If pI ::; N - 1 then there exists a function f E H~('lrN), vanishing in a
neighborhood of the origin, such that
. lim
mlnm;-+oo
ISm(O, 1)1> O.
The proof in the N-dimensional case reduces in fact, thanks to a clever

device, to the proof of the convergence on the (N -1)-dimensional torus. To
this end, each function vanishing, say, at the origin is written in the form
N
f(x)
= L /j(x),
j=1
where /j(x) = 0 in the strip {x E 'lrN : IXjl < 8}. If we replace f by /j in

(1.1), then the singularity of the integrand in the xj-variable disappears. In
fact, there occurs a smoothing in the Xj direction and the problem is reduced
to computing the partial sums of an (N - 1)-dimensional Fourier series of a
function in H~ ('lrN-1), pI > N -1 (which however not necessarily must vanish
near the origin). The last inequality (more about this in the next section) is
a necessary condition for uniform convergence in 'lrN - \ which then implies
the statement about localization. Let us however point out that the rigorous
proof of each of these steps involves considerable difficulties and technical
complications, which have to be overcome.
The connection between localization on 'lrN and convergence on 'lrN -1 of
rectangular partial sums becomes especially apparent in the construction of
so-called counterexamples. For example, let 9 E H~('lrN-l) be a function such
that its rectangular partial sums are unbounded at the point x = 0, where
we do not assume that 9 vanishes in a neighborhood of the origin. Take
h E C OO ('lr 1 ) such that h(t) = 0 for Itl < 8 and Sn(O, h) =F 0 for infinitely
many indices n. Then putting
f(x)
= g(XI, X2, ... ,xN-d h(XN),
we get a function f E H~('lrN) (with the same p and l) for which localization
fails.
The above argument illustrates anew the fact that each example of divergence in 'lrN - 1 gives an example of non-Iocalization in 'lrN Let us point
out that in the construction of counterexamples it is essential that we use
the method of rectangular summation. In Goffman and Liu (1972) there is
constructed a function in W;('lr N), p < N - 1, for which the localization
principle fails for square summation of the Fourier series.
An interesting generalization of the localization principle is given in BIoshanskir (1975), where the following is proved.
31
If / E L p (']['2), P > 1, vanishes on some domain n c ']['2 then the rectangular sum of / converges to 0 almost everywhere on n.
It is curious that this statement fails for p = 1, and does not extend to
Lp(']['N) with N > 2. Since the generalized localization principle is closer to
the problem of almost everywhere convergence, it will be discussed in greater
detail in Chapter 3.
1.3. Localization of Spherical Partial Integrals. As a rule, spherical partial

sums do not behave better than rectangular ones and, in particular, for them
also the localization principle fails in LI (']['N). Preservation ofthis class would
require passing to classes of smoother functions, exactly as this was done for
rectangular partial sums. However, it is much more effective to approach the
regularization of spherical sums by way of Riesz means.
Before studying the partial sums of a Fourier series, let us consider the
Riesz means of a Fourier integral:
The operator
R~
is an integral operator with kernel
For s = 0 this kernel is called the spectral function 0/ the Laplace operator
for the entire space lR N . The Riesz means (1.5) of the spectral function can
be computed explicitly by introducing polar coordinates, and take the form
(cf. (47)):
fJB(x, p,) = (27r)-N 28 r(s + 1)p,~-8 J ~+8~p,lxl) .
IxI T +8
(1.6)
In the introduction we remarked that the exponent s = N 21 was termed

by Bochner (1936) the critical exponent. This appellation is, in particular,
justified by the fact that the localization principle holds for the means (1.4)
for s ~ N 21 but not for s < N 21. In order to prove this statement we shall
turn to the identity (1.6). Using the estimate IJv(t)1 ~ const .1/0 for Bessel
functions, we obtain
(1.7)
Assume that the function / E L 1 (lR N ) vanishes in some domain n c lRN .
For an arbitrary compact set K c n, let 8 = dist(K, an). Then in the
integral
32
for x E K the variable y will not appear if it is closer than 6 to x, that is,
we have Ix - Yl > 6. In this case we obtain from the above formula and from
(1.6)
IR~f(x)1
N-1
:::; const J.L-2-- 8 I1fIIL 1 ,
xE K.
(1.8)
Ifs > N;l, it follows immediately from (1.8) that R~f--+ 0 for J.L --+ 00
uniformlyon K, that is, the localization principle holds true. The proof is
more complicated in the critical case s = N;l, because then the estimate
(1.8) takes the form
llfIIL
IR~f(x)1 :::; const
1,
xE
K.
(1.9)
The family of operators R~ : L1(1~N) --+ C(K) is uniformly bounded in

the Banach space B of functions f E L 1 (IR N ) which vanish on il, while it is
convergent on the dense subset CQ"'(IRN ). Therefore R~f --+ 0 uniformlyon
any compact set Keil for any f E B, that is, the localization principle
holds true also for s = N ;1 .
Ifhowever s < N;l the localization principle breaks down. In fact, assurne
that for each function f E L 1 (IR N ), which vanishes in some 6-neighborhood
of the point x = 0, we have
R~f(O) --+ 0,
J.L
--+ 00,
S<
N-1
-2-'
Then in view of the Banach-Steinhaus theorem for such functions we must

have
(1.10)
Set f(x) = (}8(X, J.L) if 6
obtain from (1.10)
< lxi<
26 and f(x)
0 otherwise. Then we
However, it follows from (1.6) and the asymptotics of Bessel functions that
the left hand side of the last integral grows like J.LN -1-28, while the right hand
N-1
side behaves like J.L-2- -8, which gives a contradiction to this inequality and,
thus, also to (1.10), so that the localization principle cannot be true.
1.4. Equiconvergence of the Fourier Series and the Fourier Integral. Turning to the study of Riesz means of order S of spherical partial sums of multiple
Fourier series,
(1.11)
let us fix our attention to the fact that the operator f

integral operator
(T~(X, f) = f D~(x jTN
with kernel
D~(x
33
I--t
(T~(x, f)
is an
y)f(y)dy,
- y) (the Dirichlet kernei) given by
D~(x-y)=(27r)-N
L
Inl<#'
(1_ln~2)8ein(x_y).
(1.12)
J.L
If one compares the identities (1.12) and (1.5), one sees that the Fourier
coefficient of the function D~(x) equals the value of the Fourier transform of
(J8(X, J.L) at the point { = n. Therefore, for s > N;1 we can apply Poisson's
summation formula, yielding
D~(x) =
(J8(X + 27rn, J.L).
(1.13)
nEZ N
Equality (1.13) constitutes one of the main tools with the aid of which one
studies spherical partial sums of multiple Fourier series. Let us split off the
term with n = 0:
(1.14)
where
p8(X, J.L) =
L (JB(X + 27rn, J.L).
(1.15)
n~O
Estimating each of the terms in the sum (1.15) with the aid of formula
(1.7), we obtain
If s
]RN:
> N ;1 the series to the right is uniformly convergent for x
{x
IXjl :::; 27r - }. Therefore we can rewrite the equality (1.14) as follows:
Assume that fELl ('lI'N) vanishes near the boundary of ']['N. Its continuation by zero off ']['N will likewise be denoted by f. Then we obtain from
(1.16)
(T~f(x)
N-l
= R~f(x) + 0(1) J.L-2-- BllfIlLll s>
N;I.
(1.17)
This equality shows that for s > N;1 we have uniform equiconvergence
of the Riesz means of the Fourier series and the Fourier integral expansion.
34
This implies, in particular, the validity of the localization principle in the

dass LI (T N ) of the means (1.11) with s > N 21.
Let us remark that relation (1.16) has been obtained only under the hypothesis s > N 21. Therefore it is not possibly to condude anything from it
for Riesz means of critical index. In particular, it does not allow us to prove
the localization ofthe means (1.11) for s = N 21. However, the reason for this
is not the weak equiconvergence of Poisson's formula but rather the essentially different behavior of spherical means at the critical index for Fourier
series and Fourier integral expansion.
We have seen above that the localization of the means (1.4) holds for
s = N 21. It is therefore rather surprising that for the means (1.11) the
localization fails for s = N 21 (Bochner (1936)). This means that in the
critical case the second term to the right in (1.14) begins to dominate and
the equiconvergence disappears. This example is yet another illustration of
the general truth that it is seldom possible to establish equidivergence with
the aid of the equiconvergence theorem.
1.5. Wesz Means Below the Critical Index. If s :::; N 21 , then for the
localization of the Riesz means (1.11) it is necessary to require appropriate smoothness of the function at hand. The most complete results hold for
Hilbert spaces of smooth functions (i. e. functions admitting square summable
derivatives). We give the corresponding formulation for the Liouville dasses
L~(TN).
If
1+ s ~
21,
1 ~ 0, s ~ 0,
(1.18)
then the localization principle for the Riesz means (1.11) of order s holds in
the dasses L~ ('D.'N).
If s = 0 (Le. for the partial sums themselves) this result was first established by V. A. Il'in (1957). In fact, the localization principle holds true in
the slightly larger Nikol'ski'i' dass H~ (cf. Il'in and Alimov (1971)).
As L~(TN) C L~(TN) for p > q, the condition (1.18) entails localization
also in the classes L~(TN) for p ~ 2. The extension of these results to the
case 1 :::; p < 2 is an extremely difficult problem. A partial solution of this
problem can be found in the paper Bastis (1983), where it is shown that in
the assumption (1.18) localization is in force also for p > 2 - (N), where
(2) = t7' (3) = ~, (N) = N~1 for N ~ 4. One can show (as is done in
Bastis (1983)) that the inequality 1+ s ~ N 21 guarantees the localization in
L~(TN) also for 1 < p :::; 2 - (N), but this succeeds not for all s ~ 0 but
only for s dose to N 21 .
There arises the quest ion how exact the condition (1.18) is and, in particular, whether it is possible to weaken it on the expense of increasing p. The
following theorem 0/ fl'in (cf. Il'in (1968)) on the non-existence o/localization
gives an answer to this question.
35
Theorem (V. A. Il'in). Assume that s ~ 0, l ~ 0 and l + s < N;l. Then

lor each point xO E TN there exists a finite 7 function 1 E G'(TN ) satisfying
the lollowing conditions:

1) I(x) = 0 in a neighborhood 01 the point XOj
2) liml'-+oo lu~(xO, /)1 = +00.
This theorem means that for l + s < N;l the localization in L~(TN)
cannot be saved even if we pass to p = 00, i. e. the sufficient condition (1.18)
is in fact independent of p (at any rate, for p ~ 2).
The proof of Il'in's theorem is based on an estimate of the Ll-norm of the
Riesz means of the spectral function (1.12):
f
N-l
JE ID~(x)ldx ~ a ./1-2-- 8 ,
(1.19)
where E is an arbitrary subset of T N of sufficiently large measure, lying off

a certain 8-neighborhood of the origin. Let us indicate how the divergence
of the multiple Fourier series expansion of the function 1 E L~(TN) follows
from this estimate in the case s = 0 (if s > 0 the argument differs in some
technical details). Let us denote by Lp(E) the set of functions in Lp(TN )
which vanish on E. By definition (cf. (58)) for each 9 E Lp(E) the function
I(x)
L (1 + InI )-!
2
nEZ
gn einx
(1.20)
belongs to L~(TN). The spherical partial sums of the Fourier series of 1 and
9 are connected by the formula
Here 1 ~ EI'l is an integral operator whose kernel KI'(x-y) is connected

with the Dirichlet kernel D I' (x - y) by the formula
(1.21)
This means that
(1.22)
The convergence of EI'I(O) for all 1 E L~(TN) of the above type would
imply, in view of the Banach~Steinhaus theorem, the boundedness of the
7 7mnslator's Note. That is, a function with compact support. This terminology is, regretfully, rarely used in the Western literat ure.
36
kernel KJ.L(x) in the Lq(E) metric, where ~ = 1- ~. But then it follows from
(1.21) that the norm ofthe kernel DJ.L(x) in Lq(E) is majorized by ,i, which
for l < N 21 and s = 0 contradicts Il'in's estimate (1.19).
Consequently, there exists a function f E L~(TN) of the form (1.20),
where gE Lp(E), with a divergent Fourier series at the origin for spherical
summation. With the aid of this function f it is also not hard to construct a
function satisfying all the conditions of Il'in's theorem. 8
In the last step one uses the property of pseudo-Iocalization of the operator
9 I--t f defined in (1.20). This property means that the operator at hand
does not increase the singular support, Le. if 9 E COO(il), il c T N then
fE COO(il) (cf. Hrmander (1983-85), Taylor (1981), Chapter 2, Sect. 2).
1.6. Localization Under Summation over Domains Which Are Level Sets
of an Elliptic Polynomial. Let A(e) be an arbitrary homogeneous elliptic
polynomial (cf. (15)) and set
(1.23)
For orientation let us remark that the behavior of the means (1.23) depends
in an essential way on the geometry of the surface
(1.24)
more exactly, on the number r of non-vanishing principal curvatures of this

surface. In view of this criterion we may make a dassification of homogeneous
elliptic operators, referring to one dass Ar all operators such that each point
of the surface (1.24) at least r of the N - 1 principal curvatures are different
from zero. It is dear that
where Ao is the dass of all homogenous elliptic operators. A typical representative ofthe dass A N - 1 is the Laplace operator. As an example of an elliptic
operator in Ar, r < N - 1, we may take
Mr(D) =
(~D1)m+! +
3=1
note that
(.t D1)m (t D1)

3=r+2
3=1
Mr~ Ar+!'
8 In view of the imbedding theorem (59) we have L~(l'N)

provided p p(e) is sufficiently small.
-->
Ci-erEN) for all e
>
0,
37
In the dass A o the exact sufficient condition for the localization of Riesz
means of order 8 of multiple Fourier series and integrals in the dass L~ has
the form (Alimov (1974)):
l+82: N;I,
l2:0,
82:0,
1~p~2
(1.25)
(if l = 0 this result is due to Hrmander (1969)). If p > 2 then an unimprovable condition is (1.18), whose union with (1.25) can be written
l+8>max{N-l
p , N-l}
2
,
1~p~2.
For operators in the narrower dasses Ar condition (1.25) can be weakened

to (provided the set QA is convex):
l
+8 >
-
N-l
- r(lp - 1)
p
2'
1 ~ p ~ 2.
(1.26)
The conditions (1.26) are sufficient for the localization of the expansion
in multiple Fourier integrals (Ashurov (1983a)). Concerning the expansion in
multiple Fourier series, one has so far only been able to prove the sufficiency
of these conditions for l = 0 (Ashurov (1985)).
In which sense is the condition (1.26) exact? For each fixed operator in Ar
(with r < N - 1) one can weaken them, but within the entire dass Ar they
are exact (Ashurov (1983a)).
Namely, assume that
l
+8 <
N-l
- r(lp - 1)
p
2'
1 ~ p ~ 2.
Then there exist an elliptic operator L E Ar such that the localization principle does not hold true for the Riesz means of the expansion corresponding
to it. This result, as weH as the sufficiency of (1.26), has been proved in the
general form for Fourier integrals and, if l = 0, for Fourier series. For r = 0
we may take for L the operator (34) in Pulatov (1981).
2. Uniform Convergence
2.1. The One-Dimensional Case. The problem of convergence is dosely
connected with the localization principle for the expansions. If the function
under examination belongs to a dass for which the localization principle holds
true, then the convergence or divergence at a given point is determined by
the smoothness of f in just a neighborhood of that point.
In the one-dimensional case the localization principle holds in LI (']['1), so
that for any fELl (']['1) the convergence of its Fourier series at the point
Xo depends only on the smoothness of f near xo. The required smoothness
condition comes in integral form and is known as the Dini condition:
1
0
If(xO
dt
+ t) - 2f(xO) + f(xO - t)l- < 00.

t
(1.27)
Sh. A. Alirnov, R. R. Ashurov, A. K. Pulatov
38
For the uniform convergence of the Fourier series in an arbitrary interval

I it suffices to require that the integral (1.27) converges uniformly for xO
in that interval. On the other hand, the well-known Dini-Lipschitz test for
uniform convergence on ']['1 does not assume that (1.27) is fulfilled. Let us
recall this classical test, for which purpose we introduce the not ion of modulus
of continuity:
(1.28)
w(c5, J) = sup If(x) - f(y)l
Ix-yl<o
In the right hand side of (1.28) the least upper bound is taken over all
points x, y E ']['1 such that Ix - Yl < 8. If
( 1)-1 '
w(c5, J) = 0(1) log 6
8 -+ 0,
(1.29)
then the Fourier series of f converges to f uniformlyon ']['1.

As is well-known (cf. Bari (1961)), the continuity only of fis not sufficient
for the uniform convergence of the Fourier series of fand, more generally,
not even for the convergence at a fixed point.
Indeed, if we consider Sn(x, J) as an operator from C(']['l) into itself, then
its norm equals
sup
sup
If
"'"09 xEyl }Tl
f(x + t)Dn(t)dtl =
}Tl
IDn(t)ldt,
Le. the Lebesgue constant, which grows as log n as n -+ 00.

Moreover, the divergence set may have the cardinality of the continuum
(although it always is a set of zero measure, as follows from Carleson's
theorem (1966)). Therefore it follows that auxiliary conditions of the DiniLipschitz type cannot be dispensed with.
The Dini-Lipschitz condition is exact in the sense that 0(1) cannot be
replaced by 0(1). Moreover, the condition f(xO + t) - f(xO) = 0(1) (log t )-1
does not gurantee the convergence of the Fourier series of f at the point xO,
that is, condition (1.29) is only a test for uniform convergence.
2.2. Rectangular Sums. A generalization of the theorem of Dini-Lipschitz
to the case of rectangular partial sums forN > 1 was given by L. V. Zhizhiashvili (1971), (1973). He show that if
( l)-N '
w(8, J) = 0(1) log 6
8 -+ 0,
(1.30)
then the Fourier series converges uniformly to f on']['N, provided we use

reet angular summation. The modulus of continuity in (1.30) is again defined
by (1.28) but now with x E ']['N, Y E ']['N.
39
Condition (1.30) is sharp, and also in the case of quadratic summability

it cannot be improved: there exists a function f such that
w(6, f)
1)-N '
= 0(1) (log t
6 -+ 0,
but whose Fourier series diverges at the point x = 0 under quadratic summation.
The proof of the necessity of condition (1.30) is based on the fact that
the Dirichlet kernel Dm(x) for reet angular partial sums is a product of onedimensional Dirichlet kernels (cf. (1.2)).
As
we have
Sm(x, f) - f(x)
= 'Ir- N
jTN
[f(x + y) - f(y)]Dm(y)dy.
(1.31)
Next, let us note that one has the following estimate for the rate of convergence in the one-dimensional case, which can be obtained upon a careful
analysis of the proof of the Dini-Lipschitz theorem: if
1)-~
w(8, f) = 0(1) log 8

then uniformlyon
'
A ~ 1, 6 -+ 0
']['1
Sn(X, f) - f(x)
= 0(1) (log n)->'+1 ,
n -+ 00.
Furthermore, one has to represent the difference f(x+y) - f(x) in (1.31) as

a sum of increments with respect to each coordinate Xk. Integrating the term
with index k with respect to Xj, we note that for j < k we integrate effectively
only the Dirichlet kernel Dm; (Yj) (the integral of which equals 'Ir) and that
for j = k we can use the aforementioned sharpening of the Dini-Lipschitz
theorem and for j > k the estimate
Hence we obtain
N
Sm(x, f) - f(x)
= 0(1) ~)logmk)-N+1
k=l
rr
N
j=k+1
logmj = 0(1).
40
Here we mayassume, without loss of generality, that ml
m2
...
mN
2.3. Uniform Convergence of Spherical Sums. In contrast to rectangular

partial sums, for the uniform convergence of spherical partial sums one has to
require quite a lot of smoothness (compared to the one-dimensional case). The
necessary condition for uniform convergence with respect to rectangles (1.30)
hardly differs from the condition for continuity and is furt her not detected
by the power smoothness scale, i.e. the usual Hlder classes are too crude
for the description of the situation at hand. This is connected with the slow
(logarithmic) growth of the LI(yN)-norm of the Dirichlet kernel (or, more
exactly, the Lebesgue constant) for the rectangular sums. In fact,
rr /,
N
Lm = /,
TN
IDm(x)ldx =
k=l
rr
N
IDmk (xk)ldxk =
Tl
k=l
logmk . (1 + 0(1)).
A completely different situation is encountered in spherical summation.

In the LI (yN)-norm the corresponding Dirichlet kernel grows like apower,
so that for convergence one requires high smoothness (in classical or in the
generalized sense). If the function under examination does not display high
smoothness then we must resort to averages of the partial sums, usually the
Riesz averages. For the expansion in a Fourier integral this follows at onee
from formula (1.6):
L>.= [
JRN I(}B(x,>')ldx=constlog>'(l+o(l)),
s=N:;I,>.-+OO.
And onee again, as in the localization problem, one is faced with the eritical
exponent s = N;l.
Boehner (1936) showed that if s > N;l then the Riesz mean of order
s (cf. (19) converge for any continuous function to the same function and
this uniformlyon yN. But if s = N;l this is not true anymore. In this
sense the Riesz means at the critical exponent behave as the partial sums of
the one-dimensional Fourier series: for their uniform convergenee eontinuity
alone does not suffiee and it is neeessary to impose "logarithmic" smoothness.
Namely, ifthe modulus of eontinuity of 1 E C(yN) satisfies the Dini-Lipschitz
eondition
w(8,1) = 0(1) log
8 -+ 0,
then the Riesz means of 1 of critical exponent converge to 1 uniformlyon

yN (cf. Golubov (1975)).
As in the one-dimensional case the convergence test 01 Dini holds true.
In order to formulate the eorresponding result let us make the following
definition: the function 1 E C(yN) satisfies the Dini condition at the point
x E yN, if for some 8 > 0 the integral
1
6
dr
11r(x) - 1(x)lo
r
41
is convergent, where fr(x) denotes the average value of f over a sphere of

radius r centered at the point x.
If f satisfies a Dini condition at the point x, then its Riesz means of critical
order at this point converge to f(x) (Bochner (1936)). If the Dini condition
is fulfilled uniformlyon 'll'N, then the convergence is uniform on 'll'N.
We have already remarked that the sufficient conditions for the uniform
convegence of the spherical partial sums, as well as of their Riesz means of
order lower than the critical, require increased smoothness of the function
under view. We formulate these conditions in the Liouville c1asses L~('ll'N),
p;::: 1, I > 0:
1+ s ;::: N;l, I p > N, s;::: o.
(1.32)
If (1.32) is fulfilled then the Fourier series of any function f E L~('ll'N) is

summable to it by Riesz means of order s, uniformlyon 'll'N (cf. Il'in and
Alimov (1971)). The first of the inequalities (1.32) is specially important.
That it is not possible to weaken it in the Hlder dass Cl follows from
V. A. Il'in's theorem on the non-existence of localization (cf. Sect. 1.5).
The second inequality lp > N is also essential, at least to make sure
that in the hypothesis of the opposite inequality lp ~ N there exists an
unbounded function in L~ whose Fourier series trivially is divergent and
cannot be uniformly summable on 'll'N (for lp > None has the imbedding
L~(1l'N) ....... C(1l'N)). However, this is not the whole story: the inequality
lp> N guarantees that all functions in L~('ll'N) are Hlder continuous with
That one cannot ignore the Hlder condition can be
exponent a = I conduded from the following theorem.
'i.
Theorem (Alimov (1978)). Assume that s ;::: 0, p;::: 1 and let I be an integer
such that
N-1
l + 8 = -2-' lp = N.
Then one can find for each point x O E 'll'N a function f E W;('ll'N)nC(1l'N)
such that
If I + s exceeds the critical exponent no additional smoothness is required
and the uniform convergence holds true for any f E W;(1l'N) n C(1l'N), lp =
N.
2.4. Summation over Domains Bounded by the Level Surfaces of an Elliptic

Polynomial. In Sect. 1.16 we remarked that the conditions for localization in
the c1ass L~ of partial sums depend in an essential way on the properties of
the principal symbol Ao(~) of an elliptic operator A, more precisely on the
degree of convexity of the set
42
Therefore it seems, at the first glance, quite unexpected that the conditions
for uniform convergence in L~, 1 > 0, do not depend on the symbol of the
operator, and take for all elliptic operators exactly the same form as for
the Laplace operator (cf. Alimov (1973)). As the level surface 8QA for the
symbol of Laplace's operator is a sphere, these conditions coincide with the
conditions (1.32) for uniform convergence of spherical partial sums.
One can explain this difference between uniform convergence and localization by noting that the most significant of the conditions (1.32) is the "limit"
case 1 + 8 = N;l, lp = N + c, c > O. The remaining cases can be reduced
to this case by applying the imbedding theorems, along with the following
property of Riesz means: if the means of order 8 converge, then the means of
order 8' > 8 are likewise convergent with the same limit. Now, the limit case
appears only if p > J:!l > 2, and therefore it is possible to compare (1.32)
with the conditions for localization (cf. (1.26))
1+ 8
~ max { N ;
1, N; 1 _ r
(~ _ ~) }
for p ~ 2, that is, for those p such that the conditions for localization do not
depend on the symbol of the operator.
The sharpness of condition (1.32) for all elliptic operators follows from the
fact that it is sharp for the Laplace operator.
Chapter 2
Lp-Theory
1. Convergence of Fourier Series in L p
1.1. Rectangular Convergence. By the classical theorem of M. Riesz the
one-dimensional Fourier series of a function f E Lp(T1 ) converges to this
function in the metric of Lp(T1 ) provided 1 < P < 00. Otherwise put, the
system {einx}~=_oo constitutes a basis in Lp(T1 ) for 1 1, as was proved by Sokol-Sokolowski (1947).
In fact, the following more general statement is true: if the operators Pn :
Lp(T1 ) - t Lp(T1 ) and Qn : Lp(T1 ) - t Lp(T1 ) are uniformly bounded, then
the operators
Rnmf(x, y) = PnQmf(x, y),
where Pn acts on the variable x E Tl and Qm on the variable Y E Tl, are
likewise uniformly bounded. As a system of functions is a basis if and only if
43
the operators of partial sums are uniformly bounded, this implies the theorem
of Sokol-Sokolowski.
In the study of bases it is often useful to make the following observation.
Let {v n }~=l be an orthonormal system in L 2 (Tl) and denote by Sn the
operator of partial sum:
n
Sn(J, x)
= L(J,Vk)Vk(X).
k=l
Usually one takes Vk E Loo(T I ). Then Sn extends in a natural way to

Lp(TI), P ~ 1. Let us assume that the operators Sn are uniformly bounded
in Lp. As S~ = Sn : Lq ~ Lq and
they are bounded in L q , + ~ = 1.

From this statement it follows that if the system {Vn}~=l is a basis in
Lp(T I ) then it is also a basis in Lq (Tl) and, therefore, in view of the Riesz
interpolation theorem, a basis in Lr(T I ), where r is any number between p
and q, !p + !q = 1.
We have already mentioned (cf. Chapter 1, Sect. 1.2) that there exists a
continuous function whose Fourier series is divergent at some point (whieh is
a consequence of the unboundedness of the Lebesgue eonstant). This means
that the trigonometrie system is not a basis in C(T1 ), and not in Loo(TI) and,
by duality, not in LI (Tl). As every function in one variable can be viewed
as a funetion of N variables, the multiple series do not give rise to a basis in
Loo(TN ) and LI (T N ), that is, the striet inequality 1 < P < 00 is neeessary
in the Sokol-Sokolowski theorem.
1.2. Circular Convergence. The fact that the property of being a basis of
the multiple trigonometrie system depends on the way how the partial sums
are formed, forees us to give a new definition of basis. Let il = {w} be one
of the families of sets, eonsidered in Sect. 3 of the introduetion, defining a
method of summation of multiple series. We say that the multiple trigonometrie system is a basis for il in Lp(TN ) if for every funetion f E Lp(TN )
its Fourier series converges to it with respect to [} in the Lp(TN ) metrie.
Let us consider from this point of view the question of basis in L2 (T N ). It
follows immediately from Parseval's formula that every eomplete orthonormal
system in a Hilbert spaee H is an unconditional basis in H (i. e. it remains
a basis after an arbitrary permutation of its elements). Consequently, the
system {einx}nEZN is an uneonditional basis in L 2 (T N ), and therefore the
multiple series of any function f E L 2 (T N ) eonverges to f in the L 2 (T N )metric with respect to any method of summation: with respect to rectangles,
balls, more generally, any system of sets exhausting ZN.
44
For p i- 2 the state of affairs is completely different. Indeed, if Pi- 2 then

the trigonometrie system is not a basis for spherieal summation. This follows
from Fefferman's theorem on spherical multipliers (cf. the following Sect. 2).
Namely, if the spherical partial sums
Sp.(f, x) =
fn einx
Inl~p.
for every function f E Lp(TN ) converge to f in the Lp(TN)-metric then, by

the Banach-Steinhaus theorem, we must have
(2.1)
From this it is not hard to deduce that the same inequality (with the same
constant M) holds in any cube
Letting h --t 0 we obtain it for the spherical means for the expansion in a
Fourier integral, whieh contradiets Fefferman's theorem (1970).
Let us present this reasoning in greater detail (cf. Mityagin and Nikishin
(1973a)).
The orthonormal trigonometrie system in Ti: is {(27r)-N/2hN/2eihnx}nEZN.
Introduce the corresponding Dirichlet kernel
D~(x)
= (27r)-Nh N
eihnx .
(2.2)
Inl~p.
Let fE COO(RN ) and let h > 0 be a number so small that the support of
f contains Ti:. Then f can be expanded in a multiple Fourier series in Ti:,
and the spherical partial sum equals
r D;(x - y)f(y)dy.
S~f(x) = jTN
(2.3)
Put A(x) = f(x/h). Then it follows from (2.2) and (2.3) that S~f(x) =
Sp.A(hx), which, in view of (2.1), gives
Consequently, for any
COO(lRN ) and for h sufficiently small

(2.4)
45
Let now h tend to O. It follows from (2.2) and (2.3) in an obvious manner
that
S:/h!(X) ~ u;!(x) == (211")-N/2 [
j(~)eiZed{.
Jlel~JI.
Therefore, we obtain from (2.4) the inequality
(2.5)
which for p :cf 2 is not true, by Fefferman (1970).
1.3. Summation of Spherical Riesz Means. The argument in the previous
section shows that for the approximation of functions in the Lp-metric by
multiple series it is, for p :cf 2, necessary to consider means of the partial
sums.
Bochner's result (1936) on uniform convergence (cf. Sect. 2 of Chapter 1)
shows that the Riesz means of order s > N ;1 for any continuous funetion
eonverge to it in the Loo('ll'N)-metric. In this case it follows from the duality
principle that for s > N;1, we have EU ~ ! (A ~ 00) in the L 1(']['N)-metric
for any ! E L 1(']['N) , and therefore
N-1
s>-2-
(2.6)
It follows from the orthogonality of the trigonometrie system that for each
fE L2(']['N) holds the estimate
(2.7)
It is intuitively clear that fr intermediate p, i. e. 1 < P < 2, an analogous estimate must hold with an s that deereases from N;1 to O. That
this is the case follows from Stein's interpolation theorem (Stein and Weiss
(1971)). Ordinary interpolation (8. la M. Riesz) allows one to interpolate the
inequalities (2.6) and (2.7) for a fixed operator, i. e. it does not provide the
possibility to change (in our case to lower) the order s when passing from L 1
to L 2 . However, if the dependence of the operators under consideration on
the parameter s is analytic (for this it is necessary to invoke Riesz means of
complex order s) then one ean carry out the interpolation in s.
Let us state Stein's interpolation theorem in a form suitable for our purposes.
We say that a function rp( r), r E IR 1 , has admissible growth if there exist
constants a < 11" and b > 0 such that
Irp(z)1
~ exp(bexpalrl).
(2.8)
Let A z be a family of operators defined for simple junctions (i. e. functions

which are finite linear combinations of characteristic functions of measurable
46
subsets of T N ). We term the family A z admissible if for any two simple

functions I and 9 the function
<p(z) = [
}TN
l(x)Azg(x)dx
is analytic in the strip 0 ~ Re z ~ 1 and has admissible growth in Im z,

uniformly in Rez (this means that we have an estimate in Imz which is
analogous to (2.8), with constants a and bindependent ofRez).
Theorem. Let A z be an admissible lamily 01 linear operators such that
IIAiT/IILpO(TN) ~ Mo(r) 1I/IILpo(TN) ,
1 ~ Po ~
IIAl+iT/IILpl(TN) ~ MI(r) 1I/IILp1 (TN) ,
00,
1 ~ PI ~
00,
lor all simple functions land with M j (r) independent 01 rand 01 admissible
growth. Then there exists lor each t, 0 ~ t ~ 1, a constant Mt such that lor
every simple function I holds
1
1- t
Pt
Po
PI
-=--+-.
Some of the most useful objects (but by far not the only one) to which
this theorem can be applied are the Riesz means, which analytically depend
on s. In this case, the admissible growth in practise does not cause any
great difficulty, as practically all functions encountered in the applications
have exponential growth (considerably weaker than (2.8)). The restriction
of the domain of definition of A z to simple functions does not diminish the
possibility of interpolation, as the simple functions constitute a dense subset
in Lp(TN ).
Let us turn to the inequalities (2.6) and (2.7). Let us fix an arbitrary
e > 0 and set s(z) = (N;I + e) z. Then the operators E~(t) satisfy all the
assumptions of Stein's interpolation theorem with PI = 1 and Po = 2, and
therefore for 0 < t < 1
where
N-l
s(t) = ( 2 - +e ) t,
whence eliminating t
1- t
-=-+P
2
l'
47
Thus, if 1 ::; p ::; 2 the Riesz means of order S of the Fourier series of any
nmction 1 E Lp(']I'N) converge for s > (N - 1)(~ - ~) to 1 in the Lp(']I'N)metric (cf. Stein (1958)). By duality, an analogous statement is true for p > 2
with s > (N - 1)(~ - ~), or for any p with
S>(N-1)1~-~I, p~1.
(2.9)
How sharp is condition (2.9)1 The divergence result, whieh we are now
going to state, joins up with condition (2.9) the doser p comes to 1 or 00
(Babenko (1973a)): if p does not belong to the interval IN = L~~l' J~11
(i. e. if I~ - ~ I > 2kr) there exists a function 1 E Lp(']I'N) whose Riesz mean
EUfor
0<
- s -<
NI~p ~I-~
2
2
(2.10)
do not converge to 1 in the Lp(']I'N)-metric. In other words, under the hypothesis (2.10) the multiple trigonometrie system is not a basis for Riesz
summability in Lp(']I'N). Let us show how this result can be obtained from
Il'in's estimate (1.19). To this end let us introduce the extremely important
zeta junction
(T(X)=
Inl-Tein:!:.
(2.11)
O~nEZn
As the series in the right hand side may diverge in the Lp-metrie and also
almost everywhere, it is necessary to make precise in what sense we take the
equality (2.11). We will interpret this equality so that to the right stands the
Fourier series of (T' In other words, (T(X) is a function, orthogonal to unity,
whose Fourier coefficients with respect to the multiple trigonometrie system
equallnl- T :
(2.12)
The first quest ion whieh arises from such adefinition is the quest ion of
existence of (T(X), If 'T > N/2 then this follows from the convegence of the
series En~o Inl- 2T along with the Riesz-Fischer theorem. The proof of the
existence of (T(X) in the case 0 < 'T ::; N /2 was first obtained by V. A. Il'in
(cf. (1958)). In this paper a detailed description of the singularities of (T(X)
for all 'T > 0 were given and, in partieular, it was proved that
(2.13)
with T(X) E C(']I'N) infinitely differentiable in the interior of ']I'N. The integral operator with kernel (T(X - y) coincides with the fractional power of
the Laplace operator (_~)-T/2 on functions orthogonal to unity (i. e. functions orthogonal to the kernel of Laplace operator on ']I'N). Therefore (T is
sometimes called the kernel 01 fractional order.
48
U Dp. (X) is the Dirichlet kernel for spherical summation:

Dp.(x) =
Lein:!:,
Inl~p.
then the spherical Riesz means of the series (2.11) can be written as
(2.14)
Comparing them with the means of the Dirichlet kernel
from which (2.14) differs only by the factor T- t in the integral, it is natural
to conjecture (this is so in the case of interest to us) that
(2.15)
Let 1 ~ p < ;~1' U s satisfies the condition
then one can always choose T such that

N -
N-l
P < T < -2- -
s.
In view of (2.13) the left hand side of this inequality guarantees that
(T belongs to Lp(TN), while the inequality to the right, together with the
estimate (1.19) and relation (2.15), shows that the norm of R~(T in Lp(TN )
cannot be bounded. The limiting case s = N(* requires somewhat
more tedious considerations.
The "necessary" condition (2.10) so far known and the sufficient one (2.9),
for convergence in Lp(TN), remain in force also for partial sums defined by
elliptic polynomials and their Riesz means (Hrmander (1969)).
A comparison of (2.9) and (2.10) reveals that they deviate from each
other. The Carleson-Sjlin theorem and also the theorem of Fefferman and
V. Z. Meshkov, to be set forth in the following section, give reasons to believe that the "necessary" conditions (2.10) are sharp and that this deviation
comes from the circumstance that the sufficient conditions are not complete.
!) - !
49
2. Convergence of Fourier Integral Expansions in Lp

2.1. The Case N = 1. The convergence of one-dimensional Fourier expansions in the metric Lp(JR1 ) is a consequence of the boundedness in Lp of the
Hilbert trans/orm
00
dt
H/(x) = /(x - t)(2.16)
11
7r
-00
(the integral is taken in the principal value sense). It is not hard to see that
F(HJ)(x) = HJ(~) = (isign~) . j(~),
(2.17)
and, consequently, cp(~) = sign~ is a multiplier from L p into L p Let us recall

the definition of a multiplier in general. A function cp(~), ~ E JRN, is called
a multiplier from Lp(JRN) into Lq(JRN) if the operator T'P' defined by the
equality
(2.18)
is of strang type (p, q), i. e. one has the inequality
(2.19)
The class 0/ multipliers from Lp(JRN) into Lp(JRN) is of great interest and
will be denoted by the symbol Mp(JRN). It is clear that if cp E Mp and 1jJ E Mp
then <p + 1/J E M p and <p .1/J E M p . If <p == 1 then Tcp is the identity operator
so that 1 E M p Moreover, if 1jJ(~) = cp(~ + h) then
T,p/(x) = e-ixhT<p[eiXh /(x)J,

so that cp(~) E M p entails that cp(~ + h) E M p
In the case N = 1 the operator (2.16) is bounded from Lp(JRl) into Lp(JRl)
so that (2.17) implies that sign~ E Mp(JRl). Then h(~) = ~(1 + sign~), i. e.
h(~) = { 1, ~ > 0,
0, ~ < 0,
is likewise a multiplier in Lp (i. e. from Lp into Lp). From this follows that
the characteristic function of any halfaxis, and thus also any segment on the
axis JRl is a multiplier in Lp , when 1 < P < 00.
Obviously the norm of the operator
(2.20)
from L p into L p does not depend on J.L. Therefore the boundedness (and so
the continuity) in L p of the partial sums of Fourier integrals holds if and only
50
if the characteristic function of the interval {~ : I~I ~ I} is a multiplier in L p ,

i. e. 1 < P < 00.
For p = 1 the situation deteriorates, since the operators (2.20) are not
bounded from LI (JRI) into LI (JR I ). Recall that this follows by duality from
the unboundedness of the operator (2.20) from C(JR I ) into C(JR I ), i. e. the
fact that there exists a continuous function with an unboundedly divergent
Fourier integral.
Thus, the convergence of the Fourier integral expansion of a function f E
Lp(JRI ) to itself in the Lp-metric is guaranteed only for 1 1, let us turn our attention to the following intuitively c1ear fact: if
ip(~), ~ E JRN, does not depend on the variable ~k then the operator TI(J'
defined by formula (2.18), is the identity in this variable, i. e. acts efIectively
only in the remaining variables. From this follows a more precise statement:
if ip E Mp(JR N- 1 ) then the same object ip, considered as a function in N
variables, belongs to Mp(JRN).
In particular, it follows from the properties of the Hilbert transform that
the characteristic function of a halfplane in JR2 is a multiplier in L p (JR2), 1 <
p < 00. Taking the product of such multipliers we see that the characteristic
function of any rectangle {a ~ Xl ~ b, c ~ X2 ~ d} is a multiplier. Exactly
in the same way we get that the characteristic function of any polyhedron in
JRN is a multiplier in Lp(JRN), 1< p < 00.
Thus, the partial Fourier integrals of a function in Lp(JRN), taken with
respect of the dilations of a polyhedron, converge in the metric of Lp(JRN),
1< P < 00.
Is the same property true for spherical domains, i. e. is the characteristic
function of the unit ball B = {~ E JRN : I~I ~ I} a multiplier in Lp(JRN),
1 < P < oo? If we do not put any restrictions to p then the answer to this
quest ion is negative. Set
N
(2.21)
The operator (2.21) is an integral operator whose kernel is given by formula
(29), so that
(2.22)
If we take for f the characteristic function of a ball of sufficiently small
radius (so that inside this ball the Bessel function does not oscillate), then
the function T f obtained cannot, in view of the slow decrease at infinity,
belong to L p (JRN) for I~ - ~ I 2: 2kr (this follows readily from the asymptotics
of the Bessel function).
51
The quest ion if the operator (2.21) is bounded from Lp(JRN) into Lp(JRN)
for I~ - I <
is known as the multiplier problem for the ball, which for a
long time was open. Of course, in the trivial case p = 2 the answer to this
question is positive. However, it was a surprise that only this case exhausts
the set of p's for whieh (2.21) is bounded in Lp(JRN), N ~ 2.
2k
Theorem (Fefferman (1970)). The chamcteristic junction of the unit ball

is a Fourier multiplier in Lp(JR N ) if and only if p = 2.
At this juncture we interrupt our presentation and turn to a problem, at
the first glance very remote from our subject, namely the Kakeya problem
which can be formulated as follows. Suppose that in the plane there lies a
needle covered by paint. It is required, without lifting it from the plane, to
turn the needle 3600 moving it in such a manner that the part which gets
colored by the needle has the least area. The determination of this minimal
area is precisely the Kakeya problem.
In the first (erroneous) solutions of this problem there appeared as minimal
domains colored domains lying in the interior of a curve of asteroid type.
However the correct answer , due to Besieovich, contained different ones: by
complicating the movement of the needle one can make the colored area
arbitrarily small. At the basis of Besieovieh's construction are "needleshaped"
domains, constructed as follows. The interval [0,1] on the abscissa axis is
divided into 2k equal parts, each taken as the basis of a triangle. The heights
of the triangles (and, thus, likewise their areas) are the same, but the vertiees
are situated in inverse order: the more to the right the basis of the triangle
is, the more to the left one takes its vertex. When the heights for k ~ 00
grow as log k, it is dear that the sum of the areas of the triangles grows as
log k. Therefore it is natural to expect that the set E equal to the union of
the triangles likewise grows as log k. However, it is remarkable that one can
position the vertices in such a way that the measure of E stays bounded (and,
in partieular, does not exceed 17). This means that the triangles intersect
many times and have a large common portion.
Precisely this circumstance was at the basis of the construction of Fefferman's counter-example. If one prolongs the sides of each triangle below, under
the abscissa axis, the corresponding prolongations of the triangles (whieh follows in a straight forward manner from the position of their vertiees) do not
intersect pairwise. Moreover, one can find a function whose support is in the
interior of this prolongation. A computation reveals that the spherical partial Fourier integrals of this function accumulate on the set E and that their
norm in Lp(E), when k gets large, grows faster than the Lp(JR 2 )-norm of the
function itself.
The fact that Kakeya's problem, whieh is so remote from harmonie analysis, proved to be the decisive step in the final solution of the multiplier
problem for the ball (however not in the expected way) leads to speculations
and to far reaching considerations. Let us limit ourselves just to state some
consequences of the properties of the above construction. An analysis reveals
52
that there exist no multipliers in L p (lR 2 ), p # 2, which are characteristic functions of bounded convex sets bounded by, say, the level lines of an elliptic
polynomial. Moreover, if the boundary of a convex set is a piecewise smooth
curve and its characteristic function is a multiplier in Lp (lR 2 ), p # 2, then it
must be a polygone.
From the negative solution of the multiplier problem for the ball in ]R2
follows also its negative solution for N > 2 (which is obtained by "stretching" the ball along the corresponding coordinate axies). Furthermore, as was
proved in the last section, it follows from here that the multiple trigonometrie
system is not a basis in Lp{'Jl'N), p # 2, for spherical summation. Finally, in
the following chapter, as well as from this result, one can get a function in
Lp{T N ), p # 2, with a Fourier series which is divergent on a set of positive
measure for spherical summation.
2.3. Bases of Wesz Means. The result on divergence in Lp{]RN) forces one
to return once more to the Riesz means of spherical partial integrals. By
standard arguments one can reduce the summability problem in Lp{lR N ) to
the following: for which 8 is the function
(2.23)
a multiplier in Lp{]RN)?
It is not hard to see that the corresponding operator
(2.24)
is an integral operator with the kernel in (4 7) taken with J.l
R S f{x) = const
r Iyl-fJRN
s J l!+s{lyl)f{x
2
= 1. Therefore,
+ y)dy.
(2.25)
Taking for f the characteristic function of a ball of sufficiently small radius

it follows immediately from the asymptotics of the Bessel function and from
(2.25) that RS f{x) decreases at infinity as Ixl-s-~. Consequently, RS f
does not belong to Lp{]RN) provided
N
1~p~1Y!
2
+8
Otherwise put, if 1 ~ p ~ N~l and
(1 1) 1
O<8<N - - p
2
-2'
53
then there masts a function in Lp(lRN ) with divergent Riesz means RB f(x)
in Lp(lRN ). By duality this statement is true for
p
~ IN =
[ ; : l'
;~ 1]'
~ s ~ N I~ ~I-~
(2.26)
Hence, it is too much to hope that the summability problem in Lp(JRN)

can be solved for Riesz means of small order (independent of N) for p ~ IN'
In the same way as in the case of multiple series (cf. Sect. 1.3) one can obtain
results on convergence by interpolating between the L l and the L2-estimates,
and then the sufficient conditions for convergence of R~f take the form
S>(N-1)1~-~I, p~1.
(2.27)
Thus, the function (2.23) is a multiplier in Lp(JRN ) if (2.27) is fulfilled and

is not a multiplier if (2.26) holds. As one sees, between the conditions (2.26)
and (2.27) there is a "margin" for p ~ IN, in size equal to
~ -I~ ~I > o.
In the two-dimensional case one has been able to eliminate this margin
thanks to the following remarkable result.
Theorem (Carleson-Sjlin (1972)). The junction (2.23) is a multiplier in
> O.
L p (JR 2 ) for pE 12 = [~, 4J and any s
The proof of this theorem is hard. Let us just remark that it suffices to
+ = 1)
prove it for p = 4. The result for p = ~ then follows by duality
and the remaining cases follow by interpolation.
If p ~ 12 , i. e. if I~ - I ~ 4, it follows by interpolation from the CarlesonSjlin theorem and condition (2.27), that the function (2.23) is a Fourier
multiplier in L p (JR2) provided
(! i
s>
21~ - ~I-~,
~h
Thus, the theorems of Fefferman and Carleson-Sjlin give an exhaustive

solution to the question of the convergence in Lp (lR2 ) of spherical Riesz means
of Fourier integral expansions.
It is natural to hope that the Carleson-Sjlin theorem extends also to the
case N > 2, i. e. that the function (2.23) for every s > 0 is a multiplier in
Lp(JRN), pEIN. However, at the present moment this has not been proved.
Assuming that the N -dimensional analogue of the Carleson-Sjlin theorem
is true, it would follow from Stein's interpolation theorem that the function
(2.23) is a multiplier in Lp(JRN) provided p ~ IN and
s>N
I~ ~I- ~.
(2.28)
54
For certain p rI, IN this result has in fact been verified. Namely, Fefferman
(1970) has proved that condition (2.28) is sufficient for the function (2.23)
to be a multiplier in Lp(JRN) for I~ - ~ I > ~t/. Refining this proof, V. Z.
Meshkov (1978) extended the validity of condition (2.28) to p's satisfying the
inequality
This result serves as a supplementary argument to the belief that condition

(2.28), apparently, guarantees the convergence of the means R~f in L p for
allprl,IN .
3. Multipliers
3.1. The Theorem of Marcinkiewicz. Let us recall that in Sect. 2.1 we
termed a function <p(~) a multiplier from Lp(JRN) into Lq(JRN) if the Fourier
transform of any function in Lp{JRN) becomes the Fourier transform of some
function in Lq(JRN) after multiplication by <p(~). One can likewise introduce
an analogous notion in the periodic case.
A sequence of complex numbers >.(n), n E ZN, is called a multiplier from
Lp(yN) into Lq(yN) if the operator
T>.f(x) =
>.(n)fn einx
(2.29)
nEZN
is of strong type (p, q), i. e. if

(2.30)
The norm of the operator T>. is called the multiplier norm of >..
In the most important case p = q multipliers from Lp(yN) into Lp(yN)
will simply be referred to as multipliers on Lp(yN) and the set of all such
multipliers will be denoted Mp(yN). The following example illustrates the
importance of multipliers.
Let us consider the Poisson equation
in periodic boundary conditions. A solution which is orthogonal to unity can

be obtained by the method of separation of variables:
()
U X
~
L..
O#nEZ N
Inl 2
f einx
n
55
A major problem in the theory of elliptic partial differential equations (to

which Poisson's equation belongs) is the estimating of the second derivatives
{Pu
ox ox =
J
L...J
O#nEZ N
njnk f inz
Inl2
ne
.k
},
= 1, ... , .
(2.31)
If >.(n) = njnk/lnl2 is a multiplier on Lp(TN ) (that this is in fact so

follows from Marcinkiewicz's theorem infra) then from (2.31) we get at once
the estimate
which yields, in view of the definition (50) of the norm in the Sobolev dasses
W~,
lIuIIW;(TN) ::; const IIfIlLp(TN) , 1 .(n) E M2(T N), then in view of (2.30), for each f E L 2('ll'N) we
must have
1>.(nWlfnI 2 ::; const
Ifnl 2
Clearly this holds if and only if >'(n) is a bounded sequence.

The case p =F 2 is considerably harder. Let us remark that if>' E Mp(T N )
then >: E Mp(T N ) so that by duality >. E Mq(T N ), where ~ + ~ = 1.
Upon applying the dassical Riesz interpolation theorem we find that >. E
Mp(T N ) for any r beween p and q, and in particular for r = 2. Therefore,
by what has been said, 1>.(n)1 ::; const, i. e. each multiplier in Lp(T N ) is
bounded, but in contrast to the case p = 2 the converse is not true.
A sufficient condition for >. E Mp(T N ) was found by Marcinkiewicz. For
the formulation of this result we introduce the notation
where ej is the vector with components e{ = Djk. If a is a multi-index, we

set
o. >.(n) = r 1 ';t >.(n).
For arbitrary multi-indices k and a we write
56
Set further I = Ef=l ei

means that D:j ~ j for all j
= (1,1, ... ,1).

= 1, ... , N.
As usual, the inequality
D:
Theorem (Marcinkiewicz (1939)). Assume that the bounded sequence A(n)

satisfies for each D:, 0:::; D: :::; I, and each k ~ 0 the inequality
Ict A(n)1 :::; const,
(2.32)
nE(k,ct)
where the constant does not depend on k. Then A(n) is a multiplier on

Lp(T N ), 1< p < 00.
3.2. The Nonperiodie ease. Marcinkiewicz's theorem has turned out to be

rather fruitful also in the nonperiodic case, i. e. in the study of the Fourier
transform. Using this theorem one has been able to obtain sufficient conditions, of considerable interest for the applications, for the membership of a
bounded function rp(~) in the class Mp(]RN), 1< p < 00 (cf. Lizorkin (1963),
Mikhlin (1956), Hrmander (1960)). Let us state the most useful variant for
our purposes of these conditions due to P. I. Lizorkin.
Theorem (Lizorkin (1963)). Assume that the junction rp(~) for each multiindex D:, 0 :::; D: :::; I, has a continuous derivative Dctrp(~) at points ~ E JRN
with ~i -I 0 if D:j = 1 satisfying the condition
(2.33)
where the constant does not depend on~. Then rp(~) is a multiplier in Lp(]RN),
1< p < 00.
Note that (2.33) for D: = 0 implies the boundedness of rp in ]RN. This is not
a coincidence, as precisely as in the periodic case it is not hard to prove that
Mp(]RN) = Mq(]RN), where ; + ~ = 1, and by the M. Riesz interpolation
theorem, Mp(]RN) C Mr(]RN) for any r between p and q. In particular, if
rp E Mp(]RN) then rp E M2 (]RN) so that
[ Irp(~)12Ij(~W~:::; constl Ij(~)12~,
JRN
RN
which again implies that rp(~) is bounded.

Note however that if rp(~) is a multiplier from Lp(]RN) into Lq(]RN), where
p< q, then rp(~) need not necessarily be bounded. Indeed, let us show that
for 1 < P < q < 00 the function
(2.34)
is a multiplier from Lp(lRN) into Lq(JRN). To this end we consider the function
(the integral is taken in the principal value sense)

(2.35)
57
Setting () = x/lxi and making a change of variable, we get
In view of the spherical symmetry of factor in front of the exponential, the

integral must be independent of () so that
(2.36)
The convolution of the kernel (2.36) with a function
as
Kr
* f(x) = f Kr(x JRN
f E Lp(JR.N), defined
y)f(y)dy,
is estimated with the aid of Sobolev's lemma (Sobolev (1950)), which is the
N-dimensional analogue of a well-known theorem by Hardy-Littlewood:
It remains to remark that, in view of the properties of the Fourier transform

and inequality (2.25), this convolution equals
The last equality together with the estimate (2.37) implies that the function (2.34) is a multiplier from Lp(JR.N) into Lq(JR. N ).
Now we interrupt our discussion on the theory of multipliers in JR.N and
turn to the periodic case. The obvious analogue of the multiplier (2.34) is the
sequence
(2.38)
and then to the kernel (2.35) corresponds the zeta function (cf. (2.11))
(r(x) =
Inl- r einx .
O#nEZ"
In view of (2.13) the zeta function (r(x) has exactly the same properties
as the kernel Kr(x) and therefore the sequence (2.38) is a multiplier from
Lp(JR.N) into Lq(JR.N).
But let us turn anew to multipliers in JR.N. The generalization of Marcinkiewicz's theorem formulated above allows us, as we have already recorded, to
conclude in a very simple manner that a given function is a multiplier in
Lp(JR. N ). Let us illustrate this at the hand of the following example:
(2.39)
58
If we write b(t)
that
= [t(1 + t 2)-1/2t then 'Ij;(~) = b(IW. It is not hard to see
Ib(k)(t)l~const.rk,
t>O,
k=I,2, ... ,
and this shows at once that the function (2.39) satisfies condition (2.33) and
therefore must be a multiplier in Lp(JRN ), 1< p < 00.
Another important example is provided by the functions
'lj;k(~)
= sign~k,
= 1,2, ... , N.
Each of these functions trivially satisfies inequality (2.33) and is a multiplier on Lp(JRN), 1 < P < 00 (note that this is not the case for p = 1). In
particular, this implies the boundedness of the Hilbert transform in L p (cf.
Sect. 2 of this chapter).
Likewise the functions
are multipliers on Lp(JRN).

Let us have a look at yet another important function:
'Pr(e)
= (1 + le/ 2)-r/2,
= 1,2, ... ,N,
= N (~ - ~) > o.
(2.40)
It follows from the properties of the multipliers (2.34) and (2.39) that 'Pr
is a multiplier from Lp(JR N ) into Lq(JRN).
Let us observe that T'f'.,.' in view of the definition of the Liouville classes
(cf. Sect. 10 of the introduction), maps Lp(JRN) into L~(JRN). This leads to
the idea that it might be possible to obtain imbedding theorems for Liouville
classes from the multiplier properties of the function (2.40). Let us prove, on
the basis of these properties, the following imbedding theorem (cf. Lizorkin
(1963):
(2.41)
(here l > m > 0, 1 0 we can write
The multiplier (2.40) maps Lp(JRN) into Lq(JRN), while the multiplier
(1 + 1~12)-m/2 takes Lq(JRN) into the Liouville class L~(JRN). Since their
composition then maps Lp(JRN) into L~(JRN), the required imbedding (2.41)
follows.
3.3. Are Multipliers in JRN Multipliers on 1'N? In the preceding section we
had several occasions to turn the attention to the analogy between multipliers
59
in ]RN and in ']['N. That this analogy is not a coincidence is established in the
following theorem due to de Leeuw.
Theorem (de Leeuw (1965)). Let A({) be afu,nction which is continu01J.S at
points 01 the integer lattice ZN and assume that it is a multiplier in Lp(]RN),
1 ~ p ~ 00. Then the sequence A(n) is a multiplier on Lp(']['N).
Let us remark that the assumption ofthe continuity of A({) at points ofZN
is necessary, because if we change a function in Mp(]RN) in a set of measure
zero its multiplier properties are not affected. In order to illustrate the power
of de Leeuw's theorem let us consider the function (2.39). As 'Ij; E Mp(]RN),
1 < p < 00, then by this theorem
is a multiplier in Lp(']['N). Taking into account that the sequence (2.38) is a

multiplier from Lp(']['N) into Lq(']['N) we conclude that
AT(n) = (1 + InI 2)-T/2,
= N
(t -~)
is a multiplier from Lp(']['N) into Lq(']['N). From this statement, similarly to

how was done in Sect. 2.3, it is not hard to derive an imbedding theorem for
Liouville classes:
l-!:!..=m-!:!..
p
q'
where 1 > m > 0, 1 'j(n) = I~I' j = 1,2, ... ,N,
are multipliers in Lp(']['N) and therefore also all products Aj(n) Ak(n). From
this follows, in particular, the estimate of the second derivatives of the solutions of the Poisson equation in terms of the right hand member (cf. (2.31)).
De Leeuw's theorem allows to translate multipliers in Lp(]RN) into multipliers in Lp(']['N). There arises the question whether the converse is true.
Namely, if the function ZN is continuous at the points of ZN and if A(n) is
a multiplier on Lp(']['N), does it follow that A({) E Mp(]RN)? The following
simple example gives a negative answer to this question:
A({)
={
Icl < 1
." - ~'
1 if
Ifl{1 > "2.
The operator T>., defined in (2.29), has in this case the form
T>./(x)
r I(y)dy
= 10 = (211")-N jTN
60
and is continuous in Lp{'JI'N) for all p ~ 1. However >'(~) is not a multiplier

in Lp(RN ), p", 2, which follows from Fefferman's theorem on spherical multipliers (cf. Sect. 2.2, Chapter 2). This depends on the fact that the class
of multipliers in Lp{R N ) is dilation (homothety) invariant: if cP E Mp(RN )
and CPe{e) = cp(c . ~), c > 0, then CPe E Mp(RN ). Moreover, the norms of the
corresponding operators do not depend on c. In fact, let
and write Ecp(~) = cp(c~). Then, as is readily seen,
TE",f(x) = E-1T",Ef(x),
so that
I/TE",fl/Lp{RN)
= I/E-1T",Efl/Lp{RN) = c~ I/T",Efl/Lp{RN)
N
$ cl> I/T",I/ . I/EflILp(RN) = I/Tcpl/ . I/fl/Lp(RN) ,

i. e. I/TE",I/ $ I/Tcpl/.
In an analogous way one establishes the opposite inequality.
It is clear that sequences >.(n) E Mp('JI'N) in general do not have such
properties. Therefore, for the converse of de Leeuw's theorem we have to
impose additional requirements, making this circumstance more precise.
Theorem (cf. Stein and Weiss (1971), Chapter VII). Let >'(~) be a contim.wus function on RN. Let us assume that for each e > 0 the sequence
>'e{n) = >.(en) is a multiplier on Lp('JI'N), 1 $ p $ 00, with the norm depending uniformlyon c. Then >'(~) is a multiplier on Lp(R N ).
3.4. Pseudo-Differential Operators. The theory of multipliers is closely
connected with the theory of partial differential operators. Let
P{D)
L:
caDa
(2.42)
lal:$m
be a differential polynomial with constant coefficients. Recall that Da =
h
D
1 8
D 1Oll ... DaN
N' W ere j = r 8z.
J
The imaginary unit has been included in D j in order to simplify the action
of the Fourier integral expansion. With such a normalization we have
The action of the operator (2.42) takes the following form
61
Let us consider the differential equation
P(D)f(x)
= g(x), x E ]RN.
(2.44)
If the functions f and 9 drop off sufficiently fast at infinity then, passing
to Fourier transforms and taking account of (2.43), we obtain
P(~)j(~) = g(~),
whence
f(x) = (27r)-N/2 (
JRN
_l_g(~)eix~~.
(2.45)
P(~)
Prom formula (2.45) one can also find the partial derivatives of f:
(2.46)
If r,o(~)
estimate
J(~) is a multiplier on Lp(]RN) then we get from (2.46) the

(2.47)
Estimates of this type playa great rle in the study of equation (2.44).
We will not enter into questions connected with the strict foundation of the
above reasoning, in particular, the problem of division with the algebraic
polynomial P(~). It is dear that the best estimates of the type (2.47) can be
obtained for polynomials having in some sense a small number of real zeros,
i. e. polynomials for which the set
N(P)
= U E ]RN : P(~) = O}
is not too big. For example, for a homogeneous elliptic polynomial p(e), i. e.
a polynomial satisfying the condition
N(P) consists of the origin only. For such a polynomial,

in L p for all a, lai::; m, so that by (2.46)
JU) is a multiplier
(2.48)
This estimate is best possible, as the exponent m in the left hand side can
not be replaced by m + e. Moreover, if (2.48) is fulfilled for all solutions of
(2.44) (i. e. solutions corresponding to an arbitrary 9 E Lp(]RN)), then the
operator P must be elliptic.
The ease with which the Fourier transforms converts differential operations
into algebraic operations leads to the wish to work out a similar method for
62
solutions of equations with variable coefficients. Let us try to apply the same
reasoning to the operator
P(x,D) =
:E ca(x)Da.
lal:~m
Proceeding as above, we obtain
P(x, D)f(x)
= (211')-N/2
:E
ca (x)
lal$m
= (211')-N/2 f
JRN
1e
a
](e)eiXedf, =
RN
p(x,e)}(e)eixedf,.
The factor in front of the exponential in the last integral is no longer the
Fourier transform of the function to the left. However, in this case one can
introduce the operator
and show that it is an "almost inverse" of the operator P(x, e).

Viewing the formula
P(x, D)f(x) = (211')-N/2
JRN
P(x, e)](e)eiXedf,
(2.49)
as the definition of the operator P(x, e) leads to the idea of extending the
definition to functions P(x, e) of a more general type. This route leads to the
notion of pseudo-difJerential operator.
Let us define the class of symbols sm as the set of functions p(x, e) E
COO(R N x RN) such that for each compact set KeRN and all multi-indices
a, one has the inequality
(2.50)
with a constant which does not depend on x E K and E RN.

Formula (2.49) assigns to each symbol p(x, e) E sm a pseudo-differential
operator p(x, D). Their merit is that the action of such operators actually
leads to algebraic operators on their symbols. For example, if P(x, D) is an
elliptic pseudo-differential operator, i. e.
for sufficiently large lei, then the inverse operator likewise is a pseudo-differential operator and its symbol is "principally" equivalent to p(;,e)'
63
e,
Let us point out that condition (2.50) with respect to reminds of condition (2.33) in the nonperiodic analogue of Marcinkiewicz's theorem. Of
course, this is not coincidental, because the symbol class S(lRN ) has multiplier properties. Indeed, if P(x, e) E SO(JRN), then for any function f E
Lp(JRN), 1 < P < 00, with compact support one has P(x, D)f E Lp(K) for
any compact set K (cf. Taylor (1981), Chapter XI, Sect. 2). Let us remark
that the local character of this result is connected with the conventional form
of condition (2.50). If this condition holds uniformly in xE JRN (and not only
on compact sets) then the corresponding operator P(x, D) is continuous on
Lp(JRN), 1 < P < 00. One has analogous theorems for operators whose symbols are subject to less restrictive conditions than (2.50) (cf. Taylor (1981),
Chapter XI).
Thus, we have established a connection between zero order pseudo-differential operators and multipliers in Lp(JR N ). A special case of zero order
pseudo-differential operators are singular integral operators.
The Hilbert transform (cf. (2.16)) is a one-dimensional singular operator
and that it is a pseudo-differential operator follows from (2.17). In order to
pass to the multivariate situation let us rewrite the operator (2.17) as
H f(x) = lim.!.
J1tl?e
e-+07r
SignlI t fex - t)dt.

t
The order of the corresponding kernel equals the dimension of the space, i. e.
is equal to 1. Therefore it is essential that the function sign t satisfies the
"cancellation condition" J~A sign t dt = o.
Now let ilex) be a function in N variables which is homogenous of degree
0, i. e. il(cx) = ilex), c > O. This is the same as to say that il is constant
on rays issuing from the origin. In particular, il is completely determined by
its restrietion to the unit sphere SN -1. Assume further that il satisfies the
cancellation condition
il(x)da = 0
JSN-l
and a condition of "Dini type": if

sup
Ix-x'I:::;c5,x,x'ES n -
then
lil(x) - il(x') I = w(6),
r w(6)d
6
Jo
< 00.
Let us define the operator Te with the aid of the formula
TeI(x) =
il(y)
-INI fex -
Iyl?,e Y
y)dy,
The following statements are valid (cf. Stein (1970)):
64
1) the limit
Tf(x) = lim Te/(x)

e-+O
exists in Lp , 1 < P < 00;

2) the operator T is bounded in Lp(JRN), 1 < p < 00;
3) there exists a bounded measurable function (a multiplier) m(y) such
that (f])(y) = m(y)j(y) for all f E L 2 (JR N ), where m(y) is of degree 0 and
has the form
m(y)
hN-l [~i sign (xy) + log I:yl] il(x)do-(x), lyl = 1.
The operator T is called a singular integral operator.

Next let us consider the limit Te/, c --t 0, in the sense of a. e. convergence.
As in other cases when a. e. convergence is studied, it is expedient to invoke
the corresponding maximal operator:
T* f(x) = sup ITe/(x) I

e
An important property of the operator T* is that it is of strong type (p, p)

for 1 0 such that (cf. Stein (1970))
(2.51)
We remark that in the proof of estimates for the maximal operator a major
role is played by the maximal function
Mf(x) = sup IB( 1 )1

r>O
X,
J1B(x,r)1
If(y)'dy,
(where B(x,r) is a ball ofradius r and center x), for which the following
estimate is fulfilled:
(the maximal function and its properties are studied in detail, e. g., in Stein's
monograph (1970)).
Usingthe estimate (2.51) the existence oflime-+o Te/(x) a. e. can be proved
rather easily.
If fE CQ"'(JRN) then Te/(x) converges uniformlyon JRN as c --t O. Indeed
Te/(x) =
J1Y1 ?1
il, (,~ f(x - y)dy +

Y
il, (,~ !f(x - y) - f(x)Jdy
e~lyl9 y
(by the cancellation condition). The first integral is a continuous function

on JRN while the second converges uniformly in x as c --t 0 in view of the
differentiability of f.
65
Denoting by Af(x) the oscillation of the function Tel as e -+ 0:
Af(x)
= lim Tel(x) e-+O
lim Tel(x) ,
e-+O
it is dear that Af(x) :::; 2T* fex). Let us write f = !I + 12, where !I E
elf(RN ) and 1I12IlLp(RN) :::; 6. It follows from (2.51) that IIAfllLp :::; cp6 so
that Af(x) = 0 a. e. and, consequently, lime-+o Tel(x) exists a. e. provided
1< p < 00. That the limit is precisely Tf(x) follows from the strang convergence Te -+ T in Lp(RN ).
It will be expedient to make the following remarks concerning a. e. convergence. The preceding scheme is very general and is often encountered. The
proof of a. e. convergence breaks up into two parts (this refers also to the
results set forth in the following chapter of this part). One of them is very
deep and indudes the essence of the result. This part is expressed in terms of
inequalities for the maximal operators. The second part is as a rule somewhat
simpler: it amounts to proving the a. e. convergence on adense subset of the
space under consideration.
Singular integral operators have numerous applications in the theory of
multiple Fourier series and integrals, and in mathematical physics (for exampIe, in the study of boundary values of harmonie functions and in the study
of generalized Cauchy-Riemann equations; for details see the monographs
Garcia-Cuerva and Rubio de Francia (1972), de Guzman (1981), Hrmander
(1983-85), Taylor (1981), Stein and Weiss (1971)). Moreover, it was precisely the study of singular integral operators which paved the way for the
introduction of the notion of pseudo-differential operator (cf., for instance,
Hrmander (1983-85)).
3.5. Fourier Integral Operators. One is led to operators of a more general
type if one looks in a different way at the definition (2.49). Writing out in
the integral the value of fee) we obtain
P(x, D)f(x)
= (21T)-N
JRN JRN
P(x, e)ei(x-y)e f(y)dya,e.
Ifwe replace the symbol p(x,e) by the amplitude

a result
Af(x)
= { {
JRN JRN
a(x,y;~),
we obtain as
a(x, y; ~)ei(X-y)e f(y)dya,e.
At first sight the dass of such operators looks much bigger than the dass
of pseudo-differential operators hut it turns out that it is not so. Indeed (cf.
Taylor (1981), Chapter II, Sect. 3), for any amplitude a(x, y; e) one can find
a symbol P(x,~) such that
A = P(x,D).
66
Operators which are definitely more general than pseud<rdifferential operators can be obtained if one replaces (x - y)~ in the exponential by aphase
function 'IjJ(x, y, ~):
Af(x) =
JRN JRN
a(x, Yj ~)ei1/J(x,y,~) f(y)dyrJe.
(2.52)
The phase function is assumed to be real, smooth and homogeneous of

degree one in~. The operators (2.52), which were called Fourier integral operators by Hrmander, were introduced by him in (1968) in the study of
eigenfunction expansions of elliptic operators with variable coefficients. Operators of a more general form than (2.52) were constructed by V. P. Maslov
(1965) in connection with the development of asymptotic methods in the
study of problems of mathematical physics.
Since the paper of Hrmander (1968) one has conjectured that the operator
of partial sums of the eigenfunction expansion of an elliptic operator p(x, D)
equals asymptotically
but one has not been able to obtain satisfactory estimates for the remainder. Hrmander proved that the principal part of the operator E).. can be
represented by the Fourier integral operator
with aphase function 'IjJ defined with the aid of the symbol of the operator
P(x,D).
The study of Fourier integral operators of the type (2.52) constitutes a
problem which is considerably harder than the corresponding problem in the
case of pseud<rdifferential operators. One of the main difficulties is connected
with the fact that the phase function 'IjJ is not uniquely determined by A. The
function 'IjJ(x,y,~) can be defined on XxYx]RN where X, Y and]RN need not
necessarily coincidej in particular, they might all have different dimension.
Furthermore, the amplitude can in different representations of A be defined
on different sets, which complicates the problem of defining the "principal
symbol" of A.
Closing this chapter devoted to the Lp-theory, we remark that the problem
of the continuity of Fourier integral operators in L p is still far from being
solved.
67
Chapter 3
Convergence Almost Everywhere
1. Rectangular Convergence
1.1. Quadratic Convergence. In 1915 in his dissertation "Integrals and
trigonometrie series" N. N. Luzin made the eonjeeture that the Fourier series
of any function in L2(T l ) eonverges a. e. In other words, if
(3.1)
then the series
(3.2)
eonverges for almost every x E Tl.

Over aperiod of many years the Luzin conjecture attracted the attention
of the specialists inducing a large number of investigations. This long story
was closed only in 1966 when Carleson (1966) published aremarkable result.
He proved that the Fourier series of any function in L2 (T l ) eonverges a. e.,
that is, Luzin's problem has a positive solution.
It is easy to extend Carleson's theorem to N-fold Fourier series with
quadratic or cubic summation (cf. Tevzadze (1970)). Let us indieate what
this is about if N = 2. Let f E L 2 (T2) be expanded in a double Fourier series
whose quadratic partial sums equal
Sk(Xl,X2)
L L
fnln2ei(nlxl+n2x2).
In21$k Inll9
Let us divide this sum into "one dimensional" partial sums with eoefficients
such that the series of their squares is eonvergent. To this end we divide the
"square"
into "triangles" with a eommon vertex at the origin and set
Then clearly
Sk(X1! X2)
A nl (x2)einlXl
In ll9
As
nl=-OO
j1r A!l (x2)dx2 =

-'Ir
B n2 (xt}ein2x2.
In 219
nl=-oo In21$lnll
Ifnln21 2
<
00,
(3.3)
68
then by Levi's theorem the series
will converge for almost every X2 E Tl. Therefore, the first of the sums in
the right hand side of (3.3) is for almost every X2 E Tl the partial sum of
a Fourier series of a function in L2 (T 1 ) and, by Carleson's theorem, it must
then converge for almost all Xl E Tl. In an analogous way one shows that
the second sum in (3.3) for almost every Xl E Tl converges in X2 a. e. on Tl,
i. e.
a. e. on T 2 . (That the series converges to f follows from the fact that Bk
converges to f in L 2 and therefore in measure, so that a subsequences Bk(j)
must converge to f a. e. on Tl.)
If N > 2 the proof is similar. Then the "cube"
Qk = {n E ZN:
In;1 ~ k,j =
1,2, ... ,N}
has to be divided into ''pyramides'' with a common vertex at the origin.

Indeed, if fE L2(TN ) has the multiple Fourier expansion
f(x) '"
fn einx ,
nEZN
then the cubic partial sum can be written as

N
Sk(X) =
L L
1=1
Here we have introduced for

(Xl, X2,"
Anl(xl)einIXI.
(3.4)
Inzl!Sk
(Xl, X2, .. , XN) E
,XI-l, XI+1,' ,XN) E jRN-1
JRN the notation xl =
putting
note, the inequality for the indices must in some cases be taken to be strict. In
other respects the proof does not difIer from the proof in the two dimensional
case.
Let us point out that the coefficients An! (xl) in the inner sum (3.4) are not
allowed to depend on k, because otherwise they will not be partial sums of a
Fourier series and their behavior will not be govemed by Carleson's theorem.
It is precisely this circumstance that dictates the way of writing the cubical
sum in the form (3.4).
69
Carleson's result has been improved by Hunt (1968), who showed that
the Fourier series of any function f E Lp(']['l), P > 1, eonverges a. e. on
']['1. On the other hand, as early as in 1922 A. N. Kolmogorov eonstructed a
remarkable example of a function whose Fourier series diverges a. e. and even
at every point. Thus, in the classes Lp(']['l) the problem of a. e. eonvergence
is completely solved.
Hunt 's theorem has also been extended to multiple Fourier series with
quadratic summation. There is even a somewhat stronger result due to Sjlin
(1971). He showed that if
iN
If(x)1 (log+ If(x)I)N log+ log+ If(x)ldx <
00,
(3.5)
then the quadratic partial sums of the Fourier series of f converge to f a. e. on

Here log+ t = 10g(max{l, t}). We remark that this theorem is new also
for N = 1. As log t grows slower than any power of t as t - 00, every function
in Lp(']['N), p > 1, satisfies condition (3.5). Therefore it follows from Sjlin's
theorem that the problem of a. e. convergence of multiple Fourier series with
quadratic summability in the classes Lp(']['N) is completely solved (for p = 1
a counterexample is provided by the function constructed by Kolmogorov).
']['N.
1.2. Rectangular Convergence. The problem of quadratic a. e. convergence

of functions in L p does not depend on the number of dimensions and the result
has exactly the same formulation for N > 1 as in the one dimensional case.
However, rectangular convergence leads to completely new phenomena.
In (1970) Fefferman gave an example of a continuous periodic function
f (Xl, X2) whose Fourier series when summed rectangularly diverges unboundedly at each interior point of the square ']['2. In this example some of the
paradoxieal properties of rectangular partial sums, casually mentioned in the
introduction, become manifest.
Let us illustrate this at the hand of the following remark. Let
'"'
~
'"'
f nln2 ei(n1xl +n2 X2)
~
!nl!:Sml !n2!:Sm2
be the partial sums of the Fourier series of the function f E L 2(']['2) and let
ml (k), m2 (k) be arbitrary decreasing sequences of integers. Then for k - 00
a. e. on ']['2. The proof of this statement is done in complete analogy to the

reasonings in Subsect. 1.1 of this section. Consequently, each sequence of
rectangular partial sums, taken with respect to any fixed exhausting system
ofrectangles, converges a. e. As C(']['2) C L 2(']['2) this remark applies also to
the Fourier series of the function constructed by Fefferman. Nevertheless, the
reet angular partial sums of its Fourier series diverge at each point.
70
There is of course no contradiction here. What then is the point? It is a

question on the sets of measure zero where the sequence Sml(k)m2(k)(Xb X2)
diverges. If there were no such sets of measure zero, i. e. if the convergence occured at all points, then we would indeed get a contradiction, because a numerical multiple series with rectangular summation cannot both
converge and diverge. In Fefferman's example one can find at each point
(XbX2) E 'll'2 a suitable subsequence (ml(k),m2(k)) such that the sequence
Sml(k)m2(k)(Xl,X2) diverges precisely at the point (XbX2) (but converges at
all other points). In other words, to each subsequence (ml(k),m2(k)) there
corresponds a divergence set of measure zero, and these sets taken together
cover the whole square 'll'2.
Thus, for all p ~ 1 the problem of a. e. convergence of rectangular sums
of Fourier series of functions in Lp('ll'N) has a negative answer for N > 1. In
particular, Luzin's conjecture concerning the a. e. convergence of the series
(3.2) (in the hypothesis of (3.1)) is not true for N-fold series under rectangular
summation.
The problem of a. e. convergence is often (especially in the theory of orthogonal series) formulated in terms of Weyl multipliers. A monotone sequence >.(n), n E ZN, is called a Weyl multiplier for a. e. convergence of
Fourier series if the condition
IInl 2 >.(n) < 00
(3.6)
nEZ N
entails the a. e. convergence ofthe Fourier series of 1 E L 2 ('ll'N). Then Luzin's

problem can be formulated as follows: is it true that the sequence >.C n) ::= 1 is
a Weyl multiplier for a. e. convergence? For quadratic summation the answer
is, as we have remarked, positive, while for rectangular convergence it is
negative.
Which are the exact Weyl multipliers for rectangular summation? So far
the answer to this question is known for N = 2.
Theorem 1 (Sjlin (1971)).
>.(n)
11
= 10g2(min{lnlI2 + 2,
In212
+ 2}),
then (3.6) entails the a. e. convergence on 'll'2 01 the rectangular partial sums
01 the Fourier series oll.
Theorem 2 (E. M. Nikishin (1972a)). Let there be given a sequence >.(n)
such that the assumption (3.6) always implies the a. e. convergence on 'll'2 .
01 the rectangular partial sums 01 the Fourier series oll. Then there exist a
constant c > 0 such that
To the problem of a. e. convergence pertains the question of generalized

localization, as studied by 1. L. Bloshanskii. RecaH that the validity of the
71
principle of localization means that the convergence of the Fourier series of

the function ! depends only on the behavior of ! in a small neighborhood of
the point. More exactly, if ! = 0 in an open set 0 C 11'N then the Fourier
series of ! converges to 0 at every point of O. For generalized localization one
requires that the convergence to 0 occurs a. e. on O.
Up to this moment, all what we have said might give the impression that
the main problem of the theory of multiple series is the extension of "onedimensional" results to the N-dimensional case and that the results are classified according to the following test: do they extend to N-fold series or not.
When passing from N = 2 to N ~ 3 such a dichotomy does not take place.
Generalized localization of rectangular partial sums belongs indeed to
those properties which discriminate between two- and three-dimensional Fourier series. In Bloshanskil' (1975) it is proved that generalized localization
holds in the classes L p (11'N), p > 1, for N = 2 but not for N ~ 3. In Chapter 1 we have noted an heuristic idea concerning to the behavior of rectangular
partial sums: if under the hypothesis of certain conditions convergence does
not hold in 11'N-l then under the same assumptions localization does not take
place in 11'N. In some sense even more is true: if convergence takes place in
11'N -1 then one has also localization in 11'N. These intuitive conclusions can
be confirmed. Rectangular partial sums of Fourier series of functions in the
classes L p (11'N), p > 1, converge if N = 1 (the result of Carleson and Hunt)
and diverge for N = 2 (Fefferman's example). Let us indicate how one can
obtain the results of Bloshanskil' (1975) from this.
Ifthe function !(Xl,X2) E L p (11'2) vanishes for lXII< 8, IX21 < 8, then it
can be written as the sum oftwo functions, one ofwhich vanishes for lXII< 6,
the other for IX21 < 8. For a function ofthis type the rectangular partial sum
can be written as follws:
Sml m2(X17 X2)=7r- 2
J6~lsl~7r
D m1 (s)j7r D m2 (t)!(Xl+S,X2+ t )dtds.

-7r
In view of Hunt's theorem (Sect. 1.1), the inner integral converges for a. e.
points (Xl + s, X2) E 11'2 and m2 ~ 00 to !(Xl + s, X2)j here we must take
into account that in view of Fubini's theorem the function !(Xl,X2) belongs
for a. e. Xl E 11'1 to L p (11'l) in X2. In the outer integral the kernel D m1 (s) is
uniformly bounded in the domain of integration so that it is natural to expect
that the outer integral too converges for m2 ~ 00. After some not very hard
technical calculations the above reasoning can be given asolid foundation.
The absence of generalized localization in L p (11'3) (and also in C(11'3)) can
be established with the aid of Fefferman's example of a continuous function
!o (Xl, X2) whose Fourier series summed rectangularly diverges at each point
(Xl, X2) E 11'2. As in Subsect. 1.2 of Sect. 1, Chapter 1, let us set
!(x) = !O(X17 X2) . h(X3),
X = (Xl. X2, X3),
where h E C(11'l) is chosen such that h(t) = 0 for Itl :::; 8, while the partial
sums Sk(t,h) ofitsFourier series at each point t E ']['1 are different from zero
72
for an infinite number of indiees. The function ! vanishes in the strip

but its rectangular partial sums
Smlm2m3(x,f) = Smlm2(Xl,X2,!O)' Sm3(X3,h),
IX31 < 8
X E Tl,
diverge everywhere on this strip (and even everywhere on T 3 ). This is not

hard to realize if we take for each m3 the indices ml, m2 large enough. Let
us also remark that the problem of generalized localization of quadratie or
cubic partial sums, which also has been studied in detail by I. L. Bloshanskil
(1976), is much harder.
We have spoken about the classes Lp(TN ) for p > 1. If p = 1, generalized
localization for rectangular partial sums does not hold for any N ~ 2, which
readily follows from Kolmogorov's example of a function in L l (Tl) with an
unboundedly divergent Fourier series.
2. Convergence a. e. of Spherical Sums and Their Means

2.1. Convergence of Spherical Sums. Recall that the spherical sums of a
function
take the form
E,..!(x) = (27r)-N
!neinx.
(3.7)
Inl~1'
One of the first questions which arise in the study of a. e. convergence of

the sums (3.7) is the question of the validity of the Luzin conjecture: is it true
that the spherical sums (3.7) of the Fourier series of an arbitrary function
! E L2(TN ) converge a. e. on TN? In other words, does Carleson's theorem
extend to N-fold Fourier series when the latter is summed spherically? The
answer to this question is open so far.
What is known is only that Hunt's theorem (Subsect. 1.1 of Sect. 1) does
not extend to N-fold (N > 1) series summed by circles. This follows from
general results by B.S. Mityagin and E.M. Nikishin (cf. (1973a), (1973b))
on the divergence of spectral expansions and lower estimates of partial sums.
Indeed, for each p E [1,2J there exists a function ! E Lp('lIoN ) such that on a
set of postive measure
lim IEI'!(x) I = 00.
1'-+00
E. M. Nikishin has remarked that this result (slightly less general than
in (1973a), (1973b)) is an immediate consequence of Fefferman's theorem to
the effect that the multiple trigonometrie system is not a basis in Lp(TN ),
1 ~ P < 2, under spherieal summation. To explain this we resort to Stein's
theorem on sequences of operators.
Let us consider a sequence of translation invariant linear operators T k :
Lp(TN ) -+ Lp(TN ), k = 1,2, ... , and let us introduce the maximal opemtor
T.!(x) = sup In!(x)l.

k
73
Theorem (Stein (1970)). 1110r each I E Lp(yN), where 1 $ P $ 2, the

function T* I (x) is finite a. e. on yN, then T* is 01 weak type (p, p).
Let us assume that for some p, 1 $ P $ 2, the multiple Fourier series of
any function I E Lp(yN) eonverges a. e. on yN under spherical summation.
Then for a. e. x (3.7) is bounded in J.L and so the maximal operator
E*/(x) = sup IE,J(x) I
(3.8)
I'
is finite a. e. on yN for each function I E Lp(yN). Therefore by Stein's

theorem (infra) E* is ofweak type (p,p). As L 2 (yN) C Lp(yN) for 1 $ P $ 2
it follows that E*!(x) is finite a. e. on yN also for any I E L 2 (yN) and
therefore, again by Stein's theorem, E* must be of weak type (2,2).
Now we ean apply the Marcinkiewiez interpolation theorem (cf. Stein
(1970), Seet. 2, Chapter 5) aceording to which E* is of strong type (Pl,PI)
for every PI in the interval P < PI < 2, i. e.
IIE*/IIL
p1
(TN)
$ eonst . II/IIL p1 (TN)
It follows from this inequality and from (3.8) that the operators EI' are
uniformly bounded in Lp(yN), eontradicting Fefferman's theorem, so that
the statement to the effect that the partial sums (3.7) eonverge a. e. for every
I E Lp(yN), 1 $ P < 2 is not true.
2.2. Convergence a. e. oC Spherical Riesz Means. The results on the divergenee of the spherieal sums (3.7) set forth in the last seetion force us to turn
onee more to Riesz means of these sums, whieh have the form
(3.9)
In the study of questions of a. e. eonvergenee it is eonvenient to introduee

the maximal operator
(3.10)
E!/(x) = sup IE~/(x)l.
I'
As was proved in Sect. 3.4 of Chapter 2 it follows from the boundedness of

the maximal operator in Lp(yN) that the means (3.9) converge a. e. for every
E Lp(yN).
Let us start with the simplest ease P = 2. The estimate for the maximal
operator in this ease follows from the following lemma, which is known as
Kaczmarz's lemma (cf. Kaczmarz and Steinhaus (1951)).
Lemma. Assume that Re s > 0, Re 0: > 0 and let F be an arbitmry measurable subset o/yN. Thenlor every function I E L 2 (yN) we have the inequality
74
Prool. The Riesz means (3.9) of different orders are related by the following
formula
\BEBI _
A
,x -
r(s + 1)
t\(\
r( + l)r(s _ + 1) 10 A
)B--1 Eld
t t t,
(3.12)
valid for Res > Re+! > O. This follows readily using the definition (11).
If we apply the Cauchy-Bunyakovski'l inequality to (3.12), we obtain
From the inequalities
~ 1,x IEf 11 dt ~ ~ 1,x IEf11 dt + ~ 1,x IEf 1 2
~ 21E~ 11 2 +
21
00
IEf 1 -
Efl 2 dt
Ef 11 2 ~t
follows the estimate
To finish the proof of (3.11) we have to prove that for Reet >
-! the following equality holds:
As
we get using Fubini's theorem
1,x
o
d(Ed,J)
11
0
[(1- u) - (1- u)al2~,

u
-!, Re >
75
concluding the proof of (3.13), and hence also of the lemma.

This proof of Kaczmarz's lemma is due to Peetre (1964).
What is the significance of Kaczmarz's lemma? It means that in the question of a. e. convergence in the class L 2(l'N) all Riesz means of positive order
are equivalent. In other words, if the Riesz means of some order Cl! > 0 converge a. e. on l'N, then the Riesz means of any other order s > 0 converge
also. Moreover, the same proof remains in force for a. e. convergence on an
arbitrary measurable subset F C l'N. Kaczmarz's lemma holds true not only
for multiple Fourier series expansions, but also for arbitrary orthogonal expansions (cf. Alimov, Il'in, and Nikishin (1976/77), Kaczmarz and Steinhaus
(1951)).
Thus, it is question of proving the a. e. convergence of Riesz means for
some (for example, a very high) order s > o. For s > N 21 we can use the
Poisson summation formula (1.13) or, more exactly, its consequence (1.17),
which reduces the problem of the convergence of spherical means of multiple
series to the analogous problem for spherical means of Fourier integrals (1.4):
From this formula and the well-known estimate for the Bessel function
we obtain after passing to spherical coordinates and integrating by parts

IR~f(x)1 ::; const
Jo(XJ IHd(x)l (tJ.L)-2

N-l
-8.
lIIH
dt
min{l, (tJ.L)-r+ 8} T. (3.14)
Here we have put

Hd(x) =
TT
1N
v Nt
Jf1yl 5:t f(x + y) dy,
(3.15)
where VN is the volume of the unit ball. Introducing the maximal operator
H .. f(x)
= sup IHd(x)l,
t>O
(3.16)
we obtain from (3.14) the required estimate

R!f(x) ::; const . H .. f(x).
(3.17)
The maximal operator H.. plays a major rle in analysis and has been
much studied (cf. Stein (1970), Stein and Weiss (1971)). In particular, for
any p > 1 the operator H .. is of strong type (p,p):
(3.18)
76
It follows from (3.17) and (3.18) that

N-1
8>-2-
This estimate implies that the lliesz means R~! with 8 > N;l of any function
E Lp(RN ), P > 1, eonverge a. e..
In view of (1.17) the same estimate is true for E!! with 8 > N;l:
N-1
8>-2-'
(3.19)
while from Kaczmarz's lemma we obtain for p = 2 an estimate whieh is valid

for an 8 > 0:
(3.20)
Consequently, for every ! E L 2('lr N ) the lliesz means EU of any positive
order converge a. e. on 'lrN For multiple series this result is due to Mitehell
(1951). As we have already told, the quest ion of a. e. eonvergence of EU,
! E L2('lrN ), remains open for 8 = o.
Let us now pass to the case 1 < P < 2. It follows from (3.19) that for such
values of p the lliesz means eonverge above the critical order. This is a good
result for p near 1, but its gets eruder when p approaches 2, as for p = 2
a. e. eonvergence holds true for all 8 > O. There arises the natural desire
to interpolate between (3.19) and (3.20), taking p in (3.19) as dose to 1 as
possible.
Let us turn attention to the diflieulties which arise then. The interpolation
theorem of Stein (cf. Sect. 1.3 of Chapter 2), which we would like to apply,
pertains to an analytic family of linear operators, but the maximal operator
E! is nonlinear . This difliculty is resolved in the following standard way.
Denote by M the dass of positive measurable functions on 'lrN taking finitely
many different values. If J.L E M then by the definition (3.8)
(3.21)
It is dear that one ean pick a sequenee J.Ll
such that
J.L2 ~ of elements in M
.lim IE~.(:t:)!(x)1
= E!!(x).
.-,
3-+ 00
This allows us to invert (3.21) as follows:

(3.22)
77
Fix now J-L E M and consider the linear operator E;(x)' which depends
analyticallyon the parameter 8. To this operator it is possible to apply Stein's
interpolation theorem, and proceeding as in Sect. 1.3 of Chapter 2, taking
acount of (3.21), we obtain from (3.19) and (3.20) the estimate
for an 8 > (N -1)(~ - ~), 1 < p ~ 2, with a constant independent of f,J-L.

From this estimate and (3.22) follows the desired estimate:
The estimate (3.23) says that the means EU of any function f E Lp(T N),
1 < p ~ 2, converge a. e. for the values of 8 indicated. This result (as weIl
as the preceding argument) is due to Stein (1958). Let us note that the
convergence theorem is valid also for p = 1. In this case, the maximal operator
H. is not of strong type (1,1), so the estimate (3.18) is not true, but it is of
weak type (1,1), which in view of (3.17) is sufficient for the a. e. convergence
of the means R~f (and therefore, in view of (1.17), also of the means E~f)
for 8 > N ;1 for any integrable function f.
To sum up, we may conclude (cf. Stein (1958)) that a sufficient condition
for the a. e. convergence of the means E~f of a function f E Lp(TN) is that
we have
(3.24)
How sharp is this condition? To elucidate this quest ion we begin with the
case ofRiesz means of order s = N;1. Condition (3.24) shows that for the a. e.
convergence of Riesz means one requires an order above the critical index only
if p = 1. That this requirement is essential follows from the following result
of Stein's (cf. Stein and Weiss (1971)): there exists a function f E L 1(T N ),
n 2: 2, such that a. e. on T N
N-1
8=-2-
Thus, for p = 1 condition (3.24) is sharp. For p > 1 one has a result due
to K. 1. Babenko (1973b): if
o -< 8 -< N (~p - ~)

- ~
2
2'
2N
l<p<-- -N+1'
(3.25)
then there exists a function f E Lp(T N ) such that E~f is divergent on a set
of positive measure. We turn the reader's attention to the fact that between
the conditions (3.24) and (3.25) there is a gap, which as in the case of L p
convergence is, of course, not accidental (cf. Subsect. 2.1 of this section).
78
Chapter 4
Fourier Coefficients
l. The Cantor-Lebesgue Theorem
1.1. The One-Dimensional Case. As is well-known, the convergence of a
series implies that its terms must converge to zero. In partieular, the convergence of a trigonometrie series
00
L(an cosnx + bn sinnx)

n=l
at a point x E ']['1 forces the expression an cos nx + bn sin nx to tend to zero

at this point as n - 00. But does this imply that an - 0, bn - O? If it
is a question of just one point the answer is, of course, negative, but if the
set of such points is sufficiently large the answer may be positive. In terms
of measure theory such an answer is provided by the well-known CantorLebesgue theorem: if
an
cos nx + bn sin nx - 0,
n-
00
on a set of positive measure, then an - 0, bn - O. This theorem, whieh is

rather easy to prove, plays a major rle in the theory of trigonometrie series
and, notably, in the problem of the unique representability of functions by
convergent trigonometrie series.
Before looking for an analogue of the Cantor-Lebesgue theorem for multiple series (whieh likewise is of great importance in the uniqueness problem)
we restate it in the case when the trigonometrie series is written in complex
form:
In this case it follows from the condition
L Cne
Inl=k
inx -
0,
k-
00
on a set E C ']['1 of positive measure that Cn - O. In this formulation a

possible generalization of the Cantor-Lebesgue theorem to N-dimensional
series with N > 1 becomes quite transparent.
1.2. Spherical Partial Sums. In 1971 Cooke (1971) proved that if N = 2
and
(4.1)
Ak(x) =
Cne inx - 0, k - 00
2
Inl =k
a. e. on T2 then
ICnI 2 -0,
79
(4.2)
k-oo.
Inl 2 =k
Cooke's proof is distinguished by its shortness and ingenuity, of which the

reader can easily convince himself. The main step consists of an estimate of
the norm of Ak in L 4 (T 2) by the norm of the same sum Ak in L 2(T2).
It is clear that
IA k (x)1 2
L: L:
cnCmei(n-m)x
= L: e ilx
Inl 2 =k Iml 2 =k
L:
CnCm,
n-m=l
InI 2 =lmI 2 =k
where the outer sum extends over all vectors l connecting points in the lattice
of integers, situated on a sphere of radius Vk. Let us now apply Parseval's
formula:
f., IA.(x)I'dx ~ 4~2 C~.1c,,12)' +4~2 ~ n~~1
c"c.. 2
(4.3)
InI 2 =lmI 2 =k
Until now we have not used the fact that N = 2. In particular, an identity
analogous to (4.3) holds also for N > 2. However, it is only in the case N = 2
that the inner sum to the right in (4.3) contains not more than two terms.
Therefore, applying the inequality la + bl 2 ~ 2(lal + Ib1)2, we obtain
L:
CnCm 2
~ 2 L: L:
ICnCml 2 = 2 (
Inl =k Iml =k
n-m=l
InI 2 =lmI 2 =k
L:
ICnI 2) 2
Inl =k
2
From this inequality and (4.3) we deduce the desired estimate

(4.4)
This estimate shows that the norms of the "spherical" sums (4.1) in the
spaces Lp (T 2 ) for 2 ~ p ~ 4 have the same order of growth.
Cooke's theorem follows in an obvious way from (4.4). Indeed, set
Then, applying the Cauchy-Bunyakovskil inequality, we obtain
IIAkllL(T2) ~ IEk(cW/2 IIAkIlL(E~) + c2IT2\Ek(c)1

~
IEk (c)1 1/ 21I A kIlL("f2) + 47r2c2.
80
It follows from this estimate and (4.4) that
IIAkIlL('f2) S 47l'2c: 2(27l' - v3I Ek(c:)1 1/ 2)-1v3

for IEk(c:)1
< l47l'2. If klim

Ak(X) = 0 a. e. on ']['2 then IEk(c:)I- 0 for k ...... oo
00
and any c: > O. Therefore
so that, as c: was arbitrary,

lim IIAkili 4 ('f2)
k ...... oo
= O.
(4.5)
It is clear that the desired result (4.2) follows from (4.5) (and the CauchyBunyakovskil' inequality).
Cooke's theorem was the last missing link in the solution of the problem
of the unique representation of functions by multiple trigonometrie series
(cf. Sect. 4 of the introduction). Before this it was proved in Shapiro (1967)
that from the convergence to zero of the circular partial sums of a double
trigonometrie series
at each point of xE
']['2
it follows, in the additional assumption of (4.2), that
Cn = 0 for all n E ZN. Cooke's theorem shows that the validity of condition
(4.2) is guaranteed by the condition of convergence at each point.

Moreover, as has been shown by Zygmund, for the relation (4.2) to hold
true it is sufficient to require that (4.1) holds on a set of positive measure
(Zygmund (1972)).
In order to solve the uniqueness problem by analogous methods in the case
N > 2 one has to extend the theorems of Shapiro and Cooke to N-fold series.
So far only the following result by Connes (1976) is available: if the spherical
partial sums of the series
(4.6)
converge in some nonempty open set w C ']['N, N ~ 2, then the relation (4.2)
is fulfilled. However, in the case N > 2 this result does not suffice to establish
the uniqueness.
2. The Denjoy-Luzin Theorem

2.1. The Case N = 1. If
(4.7)
81
then the corresponding trigonometrie series converges absolutely and uniformly, so in partieular
L la
00
cos nx + bn sin nxl < 00.
(4.8)
n=l
Is the converse true, i. e. when does (4.7) follow from (4.8)? An answer to
this question is provided by the theorem 0/ Denjoy-L'I.I.zin: if (4.8) holds on a
set of positive measure, then (4.7) is fulfilled. In other words, the series (4.8)
either converges everywhere or diverges a. e.
2.2. The Spherical Mean. For the series (4.6) the natural analogue of (4.8)
looks as follows:
L L
00
eneinx < 00.
(4.9)
k=O Inl2=k
Concerning condition (4.7), the affair is somewhat more complicated. One

might think that the most immediate counterpart is
(4.10)
but in the case studied here this is not so. In V. S. Panferov (1975) it is
proved that there exists a double trigonometrie series (4.6), N = 2, for which
condition (4.9) is fulfilled a. e. on ']['2, but
Llenl =
00.
In the same paper (Panferov (1975)) it is shown that for a square double
series the natural analogue of (4.7) is the condition
(4.11)
Namely, if (4.9) holds on a set of positive measure on ']['2 then (4.11) is

fulfilled. Also the converse is true: if (4.11) is fulfilled, then the series (4.9)
converges a. e. on ']['2. Consequently, for N = 2 the series (4.9) either converges
a. e. or diverges a. e..
The second half of V. S. Panferov's result is valid for all N ~ 2 and follows
immediately from Levi's theorem to the effect that the convergence of aseries
obtained by integrating term by term implies the convergence a. e. of the
function series itself. Indeed, appplying the Cauchy-Bunyakovski'l inequality
82
to the integral, we obtain
f: iN
k=O T
eneinz dx ~
Inl2=k
f: I']['NI (iN
I/2
k=O
eneinz 2 dX) 1/2
Inl2=k
~ I'lI'NIt,CE 1c..I')
/
1 ',
and from the convergence ofthe last series (4.9) follows for a. e. xE ']['N.
The coefficient condition (4.11) can be interpreted in terms ofthe smoothness ofthe function which is expanded into the Fourier series (4.6). We state
one such result due to S. P. Konovalov (1979). He showed that if 1 > 1 then
for each function fE C'(']['N) the series (4.9) converges a. e. on ']['N. For 1 = 1
this result is not true, as in Konovalov (1979) there is constructed a function
fE C 1(']['N), N ~ 2, such that the series (4.9) diverges on a set of positive
measure.
Combining the result of S. P. Konovalov with the theorem of V. S. Panferov
we find that the Fourier coefficients of any function f E C'(']['N) with 1 > 1
satisfy condition (4.11), while for 1 = 1 and N = 2 this is not the case.
In Konovalov (1979) the non trivial part is the construction of a function
fE Cl(']['N), N ~ 2, whose Fourier series (4.6) is not absolutely convergent.
Concerning the positive result on convergence in C'(']['N) for 1 > 1, the latter
can be obtained as an easy consequence of condition (4.11). Indeed, if f E
C'(']['N), 1 > 1, is given by the Fourier expansion (4.6), then applying the
Cauchy-Bunyakovskil inequality we get
t, C~, 1c..I') '/' ~ t,(1 + W'I'. C~, 1c..I'(1 + Inl'),) '/'

~
f:(1
+ k)-l] 1/2 .
k=O
[L len1
2 (1
+ In I2 )1] 1/2
nEZ N
~ const IlfIIL~(TN) ,
where L~(']['N) is the Liouville class defined in Sect. 10 of the introduction

(cf. (58. As C'(']['N) C L~-e(']['N), e > 0, this implies the required relation
(4.11).
2.3. Rectangular Sums. In this section we give another analogue of the
Denjoy-Luzin theorem, contained in the paper Revez and Szasz (1957), pertaining to rectangular absolute convergence. Let Zlj. stand for the set of
elements of the lattice of integers ZN with nonnegative coordinates, and set
for any m E Zlj.
Am={nEZN:lnjl=mj,
j=1, ... ,N}.
In Revez and Szasz (1957) it is shown that if N
L L
Cne inx
83
= 2 and
< 00
mEZ;:' nEAm
on the set of positive measure on T2 then
We note that the converse is trivially true.
3. Absolute Convergence of Fourier Series

3.1. Some One-Dimensional Results. In this section we consider ''real''
absolute convergence of Fourier series:
f(x) '"
fn einx ,
(4.12)
nEZ N
i. e. the convergence of the series of its moduli:
Ifnl <00.
(4.13)
nEZ N
In the one-dimensional case there is a classical result due to S. N. Bernshte'ln (cf. Zygmund (1968), Vol. 1, Chapter VI, (3.1)) on the absolute convergence ofFourier series offunctions in the Hlder classes CI(T N ): if l > 1/2
then the Fourier series is absolutely convergent, while there is a function in
C1/2(T N ) whose Fourier series diverges absolutely.
There is a generalization of this due to Szasz (cf. Zygmund (1968), Vol. 1,
Chapter VI, (3.10)): if CI(T 1 ) then
(4.14)
for any > 1;21. If = 1;2/1 l > 0, it may happen that the series (4.14)
is divergent, which is seen at the hand of an example due to Hardy and
Littlewood (cf. Zygmund (1968), Vol. 1, Chapter V, (4.2)):
L n-!-leinlnneinx.
00
f(x) =
n=l
(4.15)
Sh.A. Alimov, R.R. Ashurov, A.K. Pulatov
84
The function (4.15) belongs to CI(TN ) for any l, 0 < l

series (4.14) is divergent if = 1';21'
< 1, but for it the
3.2. Absolute Convergence of Multiple Series. In the multivariate case one

of the conditions for the convergence of (4.13) is easy to obtain. Comparison
with a multiple integral shows that the series
> N /2. Taking account of this, we can estimate the series

(4.13) as follows using the Cauchy-Bunyakovski'l inequality:
is convergent if l
(for the definition of the norm, see (58) in the introduction). This means that
the Fourier series (4.12) of any function f E L~(TN) for l > N/2 is absolutely
convergent.
From the imbedding CI(TN ) - t L~-e(TN), c > 0, follows now the analogue
of Bernshte'ln's theorem: the Fourier series of a function f E C1(T N ) with
l > N /2 is absolutely convergent.
How sharp is this result for the classes L~(TN)? In order to answer this
question let us consider the numerical series
nEZ
(1 + InI 2 )-.If [log(l + InI 2 )t o ,

N
which by Cauchy's convergence test converges if a > 1 and diverges if a

By the Riesz-Fischer theorem the function
g(x) =
(1 + InI 2 )-11 [log(l + Inl2)t1einx
l.
(4.16)
nEZ N
is in L2 (T N ) so that the function
f(x) =
L (1 + InI )-11 gn einx ,

2
(4.17)
nEZ N
where gn are the Fourier coefficients of g, belongs to the Liouville class

L:/ 2 (T N ) (cf. Sect. 10 ofthe introduction). However, it is manifest that the
Fourier coefficients of f do not satisfy (4.13). Thus, the condition l > N/2
for the absolute convergence of Fourier series in L~ (T N ) is sharp.
85
The question of absolute convergence in the classes L~ is more difficult. If

1 N/p and, as exampIes show, this condition is sharp, because the opposite condition 1 ~ N /p
entails that there exists such a function J(x) which is unbounded. If the
Fourier coefficients of this function would satisfy (4.13) then this would imply that the series (4.12) converges uniformly to J, which again ~ontradicts
the unboundedness of J.
The case p > 2 is different. Then we have the imbedding
where the smoothness exponent 1 does not change. Therefore, for p > 2
the sufficient condition for absolute convergence of the Fourier series takes
the same form as for p = 2, i. e. 1 > N /2. The fact that it is not possible
to improve this condition was first established by V. A. Il'in (1958), using
special properties of Rademacher functions.
Recall that the Rademacher functions
pk(t)
= signsin(2 k +17l"t),
= 0,1, ... , t E [0,1]
form an orthonormal system (not complete!) on [0,1]. We require here its

N -dimensional analogue: we set
rk(t)
= { P2k(t),
k2
P-2k-l (t), k < 0,
and consider the functions

nEZN ,
tE
Q,
defined in the cube Q = {t E RN, ~ tj ~ 1, j = 1,2, ... ,N} and likewise

forming an orthonormal system. One of the most remarkable properties of
this system is the fact that if
F(t)
nEZ
then for any p, 1 < p <
00,
enrn(t),
the following inequality holds:

(4.18)
86
(cf. Stein (1971), Appendix C). Let 9 be the function defined by (4.16). Set
F(x, t)
9nTn(t)einx,
xE ']['N,
tE Q,
(4.19)
nEZ N
where 9n are the Fourier coefficients of 9. It follows from the orthogonality

of the Rademacher system that
IIF(x, t)IIL(Q)
L N 19n1
= (21r)-NII911L(TN)'
InlEZ
Taking account ofthis relation we get from the rightmost inequality (4.18)
Consequently,
lTN
dx { IF(x, t)IPdt < 00.
lQ
Now we can refer to Fubini's theorem, according to which for a. e. tE Q

holds
{ lF(x, t)IPdx < 00,
lTN
t. But
then, in view of the definition of Liouville classes (cf. introduction, Sect. 10),
viewed as a function of x
i. e. (4.19) (considered as a function of x) belongs to Lp(']['N) for a. e.
!(x, t)
Tn (t)(1 + InI 2 )-lf 9n einx
(4.20)
nEZ N
belongs for a. e. x to L:/ 2(']['N). If t E Q has dyadic-irrational coordinates,

then ITk(t)1 = 1 and for such t the series (4.20) and (4.17) are simultaneously absolutely convergent or absolutely divergent. As (4.17) isabsolutely
divergent, it follows that (4.20) must be absolutely divergent for a. e. t. Consequently, for each p, 1 < P < 00, there exists a function in L:/2(']['N) with
absolutely divergent Fourier series.
Let us remark that the above proof shows more: that for each function
! E L~(']['N) with Fourier series (4.12) "almost all" functions of the form
l(x) '"
NEZ N
fneinx ,
obtained by applying an arbitrary permutation of signs, belong to L~(']['N),

1 < P < 00. Hence, taking ! E L~/2(']['N) with absolutely divergent Fourier
series we obtain the requested counterexample.
87
Thus, in the classes L~('JrN) the sharp conditions for absolute convergence
of Fourier series take the form:
l > N . max {~ ~}
p' 2 '
If l = N.max{~,
Fourier series.
n then there exist /
1 < P < 00.
E L~(TN) with an absolutely divergent
3.3. On the Convergence of the Series of Powers of I/ni. For multiple

Fourier series one furt her studies the question which smoothness of the function / will guarantee the convergence of the series
(4.21)
In Szasz and Minakshisundaram (1947) it is proved that for every function

/ E CI(TN ) the series (4.21) converges for > 2N/(N + 2l). In that paper
it is assumed that 0 < l < 1, which excludes (for N ~ 2) the case = l.
However, Bochner (1947) has remarked that the result remains in force for
any l > O. In particular, if l > N /2 this implies the convergence of the
series (4.13), and this statement, as we observed in the previous section, is
an analogue of Bernshte'ln's theorem.
That the condition > 2N/(N +2l) is sharp follows easily by the example
of aseries considered by Wainger (1965)
/(x)
Inl- N/ 2- I (log Inl)-Weilnl(log Inl)d einx
(4.22)
Inl>1
(compare with the Hardy-Littlewood series (4.15)). In Wainger (1965) it is
proved that for any d > 0 and l > ~ N the function /(x) defined by (4.22)
belongs to CI(TN ). The corresponding series (4.21) for = 2N/(N +2l) has
the form
l/nI J.f2 1 =
Inl-N(loglnl)-:-til
nEZ N
Inl>1
and, clearly, is divergent for every l >
o.
Remarks and Bibliographical Notes

Multiple Fourier series are treated in the books Zhizhiashvili (1969), Yanushauskas
(1986), Stein and Weiss (1971), Chapter VIII and the surveys articles Alimov, Il'in, and
Nikishin (1976/77), Zhizhiashvili (1973), Ash (1976). In the bibliographical survey Golubov
(1982) most results in the theory of multiple Fourier series and Fourier integrals, reviewed
88
in Referativnyi Zhurnal "Matematika" between the years 1953 and 1980, are set forth (and
in some eases likewise their generalization to eigenfunctions and to spectral expansions is
mentioned) .
The theory of general orthogonal series is developed in the book Kaczmarz and Steinhaus (1951).
Fourier series expansious in eigenfunctions of elliptie operators and also the eorresponding spectral expansions are treated in Titehmarsh (1958) and in the expository papers
Alimov, Il'in, and Nikishin (1976/77), Il'in (1958), (1968). Moreover, the papers Il'in and
Alimov (1971), Hrmander (1968), (1969) contain a short survey of results and methods,
along with a bibliography.
We remark that many results on multiple Fourier series and Fourier integrals, mentioned
in the present survey, have been proved in a more general situation (for example, for Fourier
series expansions for Laplace operator or general elliptie operators). The results of the
following papers belong to this eategory: Alimov (1973), (1974), (1976), (1978), Alimov,
Il'in, and Nikishin (1976/77), Ashurov (1983b), Babenko (1973b), (1978), Il'in (1957),
(1958), (1968), Il'in and Alimov (1971), (1972), Mityagin and Nikishin (1973), Nikishin
(1972b), Pulatov (1978), Hrmander (1968), (1969), Peetre (1964), Titchmarsh (1958).
In the present survey we have not diseussed more general questions of the theory of multiple Fourier series (for example, Fourier series on groups). These questions are diseussed
in detail in the introductory article by V. P. Khavin (1987) in Vol. 15 "Commutative Harmonie Analysis I" of this Eneyclopaedia (cf. further other parts of the series "Commutative
Harmonie Analysis"). There one finds also a bibliography and a detailed aceount of the
historical development of harmonie analysis, in particular, of the theory of Fourier series.
The majority of the results in the one-dimensional theory of trigonometrie series ean be
found in Bari (1961), Zygmund (1968). Moreover, we mention the survey article Kislyakov
(1988) in this series.
Introduetion. Sects. 3 and 4. Coneerning results pertaining to partial sums and forms of
eonvergenee, not eonsidered in the present work, we refer to the papers Golubov (1982), Ash
(1976). Note that the hyperbolie partial sums (cf., for example, Golubov (1982), Belinskii
(1979 Ll nll ..... lnNI:5R ene in", do not fall within the framework ofDefinition D) in Seet. 3,
because the set {n E ZN : Inll ... lnNI ~ R} is not bounded. Such partial sums playa
great role in approximation theory.
Sect. 6. In this survey we eonsider only Riesz means. Other summation methods (cf.
Abel, Riemann and Poisson means) are eonsidered e. g. in Alimov, Il'in, and Nikishin
(1976/77), Trigub (1980), Yanushauskas (1986), Stein and Weiss (1971), Chapter VII,
Zygmund (1968), Vol. 11. In partieular, Abel means of multiple Fourier series are used
in the study of the boundary behavior of analytie functions of several eomplex variables
(cf., for example, Zygmund (1968), Chapter XVII, Stein and Weiss (1971), Yanushauskas
(1986.
Sect. 8. The asymptotie behavior of the Fourier transform of eharacteristic functions of
eonvex sets is studied in Herz (1962), Randol (1969).
Seetion 9. In the study of the eonvergenee of multiple Fourier series further the asymptotic behavior for large R > 0 of the Lebesgue eonstant
(i. e. the norm of the operation of taking the partial sum SRw(X, J) in C(']['N plays an
important role. The following two-sided estimate (Yudin (1979 holds true: cIlnN R ~
LR(W) ::; c2R(N-I)/2. Henee, within the whole dass of convex sets the Lebesgue eonstant
of the ball has the largest growth (cf. (1.19, while the Lebesgue eonstant of the eirc1e
has the smallest growth. For N = 2 A. N. Podkorytov (1984) has eonstrueted for any
given numbers p > 2 and 0 < q < ~ convex sets W = w(P) and 0' = O'(q) such that
cIlnP R ~ LR(W) ~ c2lnP Rand CIRq ::; LR(O') ::; c2Rq.
89
Let us remark that the left hand side of (40) equals N(!-'), the number of eigenvalues
not exeeeding 1". In the study of the problem of the cirde in number theory one uses in
the multidimensional ease effeetively the method of trigonometrie sums (cf., for example,
Hua (1959); note that in this paper, as weIl as in Babenko (1978), likewise estimates of the
type (39) are obtained).
Chapter 1. Seet. 1.2. In Goffman and Liu (1972) it is further proved that the loealization
principle holds in the dass Wf-1('ll'N) under square summation.
Seet. 1.4. The results on equieonvergenee of multiple Fourier series and Fourier integrals
are due to Bochner (1936). Stein (1958) sharpened Boehner's result on loealization for
8 = N21 in terms of the Orliez dass L log+ L (J E L log+ L if and only if the funetion
I/(x)llog(max{1, I/(x)I}) is summable).
Seet. 1.5. B. M. Levitan (1963) gave neeessary eonditions for the loealization in the
dasses W~. For odd N these eonditions eoincide with (1.18) and are of final eharacter. An
analogue of V. A. Il'in's theorem (on the failure of loealization for I in Cl) in the Nikol'skii
dasses H~ was obtained in Pulatov (1978).
Seet. 1.6. That the eondition (1.25) is sharp for multiple Fourier integrals within the
dass of all elliptie polynomials was established in Pulatov (1978). In A. I. Bastis (1982)
there is eonstrueted a selfadjoint extension of the Laplace operator such that for the eigenfunetion expansion eondition (1.25) eannot be improved for p 1.
Seet. 2.2. If I E H~(1rN), pi > N, then by the imbedding theorem I E CE (']['N),
e = 1- ~ (cf. (59, so that for such values of p and I the eondition of uniform eonvergenee
(1.3) ean not be fulfilled.
Seet. 2.3. Stein (1958) proved that uniform summability of Riesz means of the eritieal
order holds likewise for funetions in the Orliez dass L log+ L. In the dasses of I times
eontinuously differentiable funetions a eondition for uniform eonvergenee was obtained by
B. M. Levitan (1963). For odd I this result is sharp.
Seet. 2.4. If the funetion I E Lp(G), 1 ~ p ~ 2, is eontinuous in a subdomain D c G,
then the Riesz means of order 8 >
1 of the speetral expansion of the funetion, eorresponding to an arbitrary elliptie operator, eonverge to I(x) uniformlyon every eompact set
K C D. This result is due to Hrmander (1969).
Chapter 2. Seet. 1.1. L. V. Zhizhiashvili (1970), (1971), (1973) obtained a definitive
eondition for eonvergenee in L1(']['N) for funetions in L1(']['N) in terms of an integral
modulus of eontinuity.
Seet. 1.2. The eonvergenee in L q of speetral expansions of funetions in L p is studied in
Mityagin and Nikishin (1973a), (1973b).
Seet. 1.3. In K. I. Babenko (1973a), (1973b) the order of growth in ,\ of the norm of the
operator E~ : L p ~ L q is studied in the assumption of (2.10). Condition (2.9) for p = 1
was sharpened in Orliez dasses by Stein (1958).
Seet. 3. The theory of multipliers and its applieations is set forth in the books Nikol'skii
(1977), Hrmander (1983-85), Stein (1970), Stein and Weiss (1971), Zygmund (1968).
Pseudo differential operators and Fourier integral operators playamajor rle in the theory
ofboundary problems ofmathematieal physies (cf., e. g., Maslov (1965), Hrmander (198385), Taylor (1981.
Seet. 3.3. If the eharacteristie funetion of a set E C lR N is a multiplier on Lp(lR N ),
then the same property is enjoyed by any set E' obtained by rotating E. In the periodie
ease this property may fai! (cf. Belinski'i' (1979.
Seet. 3.5. If the eUiptie partial differential operator A(x, D) has a eomplete set of eigenfunetions {Un(X)} in L2(fl), then the partial sum of the Fourier series with respeet to this
system ean be written as an integral operator:
N;
E)./(x)
in
9(x, y, '\)/(y) dy,
E L2(fl).
90
The kernel 9(x, Y, >') is called the spectral function of the operator A(x, D) and takes the
form
9(x, Y, >') =
Un(X)Un(Y),
An<A
where >'n is the eigenvalue corresponding to Uno Let N(>') be the number of eigenvalues
not exceeding >.. Then
N(>')
In
9(x, x, >') dx.
The asymptotic estimate for N(>') goes back to the work of Weyl, Courant and Carleman.
For a large dass of boundary problems one knows that (cf. Hrmander (1983-85
N(>') '" (211')-N
1(1
n
A(Z,e)<A
de,) dx.
This estimate is a generalization of formula (40).

So far two approaches are known to the study ofthe asymptotics ofthe spectral function
9(x, Y, >'), where x and Y lie on a compact subset of O. To the first group belongs the work
of Carleman, B. M. Levitan, Grding, Hrmander and others, which authors in order to
obtain estimates for the remainder term of the spectral function employ the method of
Carleman, which amounts to considering a suitable function of the operator A followed by
a subsequent application of an appropriate Tauberian theorem. For elliptic second order
operators V. A. Il'in (cf. Il'in (1958), (1968), Alimov, Il'in and Nikishin (1976/77), Part 11)
has developed an alternative approach, which is based only on the mean value theorem.
Let us here mention the most general result, due to Hrmander:
where m is the order of A and 'IfJ(x, Y, >') aphase function (cf. (2.53.
Chapter 3. Sect. 1.2. In Bakhbukh (1974), Oskolkov (1974) conditions are given for
the a. e. convergence of a multiple Fourier series of a continuous function in terms of
the modulus of continuity. These conditions are almost definitive (Bakhbukh and Nikishin
(1974. Analogous results in terms of integral moduli of continuity in the multidimensional
situation are given in Zhizhiashvili (1971), (1973).
Sect. 2.1. When / E L2(TN) is "logarithmically smooth" in terms of an integral modulus of continuity, it has been shown by V. I. Golubov (1975) that EA/(x) --> /(x) a. e. The
a. e. convergence of the Riesz means EU(x) of critical order 8 = N;;l for / E L(log+ L?
was proven by Stein (1958), (1961). In a recent publication Sunouchi (1985) has shown
that this result extends also to functions / E L log+ L log+log+ L.
Sect. 2.2. Divergence on a set of positive measure is also studied in Mityagin and
Nikishin (1973a). Let us remark that the condition (3.24) for a. e. convergence of multiple
Fourier series and integrals, in contrast to localization, does not depend on the geometry
of the surface 8QA (Ashurov (1983b.
Chapter 4. Sect. 2.2. The Denjoy-Luzin theorem for N = 2 is also proved in Golubov
(1975).
Sect. 3.2. In several papers necessary conditions for the absolute convergence of multiple
Fourier integrals are studied. Let us mention R. M. Trigub (1980).
91
Bibliography*
Alimov, Sh. A. (1973): Uniform convergence and summability of spectral expansions of
functions in L~. Differ. Uravn. 9, No. 4, 669-681. English translation: Differ. Equations
9,509-517 (1975), Zbl. 259.35064.
Alimov, Sh. A. (1974): On the localization of spectral decompositions. Differ. Uravn.
10, No. 4, 744-746. English translation: Differ. Equations 10, 576-578 (1975), Zbl.
284.35059.
Alimov, Sh. A. (1976): On spectral decompositions of functions in H~. Mat. Sb., Nov. Sero
101, No. 1,3-21. English translation: Math. USSR, Sb. 30, 1-16 (1978), Zbl. 344.46069.
Alimov, Sh. A. (1978): On expanding continuous functions from Sobolev classes in eigenfunctions of Laplace operator. Sib. Math. Zh. 19, No. 1, 721-734. English translation:
Sib. Math. J. 19, 507-517 (1979), Zbl. 397.47023.
Alimov, Sh. A., Il'in, V. A., and Nikishin, E. M. (1976/77): Questions of convergence of
multiple trigonometrie series and spectral decompositions. Usp. Mat. Nauk 31, No. 6,
28-83; 32, No. 1, 107-130. English translation: Russ. Math. Surv. 31, No. 6, 29-86
(1976); 32, No. 1, 115-139 (1977), Zbl. 345.42002; Zbl. 353.42005.
Ash, J. M. (1976): Multiple trigonometrie series. In: Stud. harmon. Anal. Proc Conf.
Chicago 1974, MAA Stud. Math. 13,76-96. Zbl. 358.42011.
Ashurov, R. R. (1983a): On conditions for localization of spectral expansions corresponding
to elliptic operators with constant coefficients. Mat. Zametki 33, No. 6,847-856. English
translation: Math. Notes 33, 434-439 (1983), Zbl. 588.35065.
Ashurov, R. R. (1983b): On the a. e. summability of Fourier series in L p with respect to
eigenfunctions. Mat. Zametki 34, No. 6,837-843. English translation: Math. Notes 34,
913-916 (1983), Zbl. 536.42029.
Ashurov, R. R. (1985): On localization conditions for multiple trigonometrie Fourier series.
Dokl. Akad. Nauk SSSR 282, No. 4, 777-780. English translation: Sov. Math., Dokl. 31,
496-499 (1985), Zbl. 613.42009.
Babenko, K. I. (1973a): On the mean convergence of multiple Fourier series and the asymptotics of the Diriehlet kernel for spherical means. Preprint No. 52. Moscow: IPM Akad.
Nauk SSSR. [Russian]
Babenko, K. I. (1973b): On the summability and convergence of eigenfunction expansions
of a differential operator. Mat. Sb., Nov. Sero 91, No. 2, 147-201. English translation:
Math. USSR, Sb. 20, 157-211 (1974), Zbl. 283.35060.
Babenko, K. I. (1978): On the asymptoties ofthe Diriehlet-kernel of spherical means of multiple Fourier series. Dokl. Akad. Nauk SSSR 243, No. 5, 1097-1100. English translation:
Sov. Math., Dokl. 19,1457-1461 (1978), Zbl. 419.10046.
Bakhbukh, M. (1974): On sufficient conditions for rectangular convergence of double series.
Mat. Zametki 15, No. 6, 835-838. English translation: Math. Notes 15, 501-503 (1974),
Zbl. 314.42011.
Bakhbukh, M. and Nikishin, E. M. (1974): On the convergence of double Fourier series
of continuous functions. Sib. Math. Zh. 14, No. 6, 1189-1199. English translation: Sib.
Math. J. 14, 832-839 (1974), Zbl. 279.42021.
Bari, N. K. (1961): Trigonometrie series. Moscow: Fizmatgiz. English translation: Oxford:
Pergamon Press 1964 (2 vols.)
Bastis, A. 1. (1982): On the divergence of certain spectral expansions in L~. Mat. Zametki
32, No. 3, 309-314. English translation: Math. Notes 32, 634-637 (1983), Zbl. 504.35069.
* For the convenience of the reader, references to reviews in Zentralblatt fr Mathematik

(Zbl.), compiled using the MATH database, and Jahrbuch ber die Fortschritte der Mathematik (Jbuch) have, as far as possible, been included in this bibliography.
92
Bastis, A. I. (1983): Some questions of the summability of expansions corresponding to

elliptic operators. Abstract of candidate dissertation. Moscow: MGU. [Russian)
Bastis, A. Y. (1983): On the asymptoties of the Riesz-Bochner kernel. Annal. Math. 9,
247-258, Zbl. 543.42012.
Belinskir, E. S. (1979): Some properties of hyperbolic partial sums. In: Theory of functions
and mappings. Collect. Sei. Works, Kiev 1979, pp. 28-36, [Russian) Zbl. 493.42021.
Bloshanskir, I. L. (1975): On the uniform convergence of multiple trigonometrie series and
Fourier integrals, Mat. Zametki 18, No. 2, 153-168. English translation: Math. Notes
18,675-684 (1976), Zbl. 317.42018.
Bloshanskir, I. L. (1976): On the uniform convergence of multiple trigonometrie series and
Fourier integrals under square summation. Izv. Akad. Nauk SSSR, Sero Mat. 40, 685705. English translation: Math. USSR, Izv. 10, 652-671 (1977), Zbl. 348.42005.
Bochner, S. (1936): Summation of multiple Fourier series by spherieal means. Trans. Am.
Math. Soc. 40, 175-207, Zbl. 15,157.
Bochner, S. (1947): Review of "On absolute convergence of multiple Fourier series" by
Szasz and Minakshisundaram. Math. Rev. 8,376
Carleson, L. (1966): On convergence and growth of partial sums of Fourier series. Acta
Math. 116, No. 1-2, 135-137, Zbl. 144,64.
Carleson, L. and Sjlin, P. (1972): Oscillatory integrals and multiplier problem for the disco
Stud. Math. 44, No. 3, 287-299, Zbl. 215,183.
Connes, B. (1976): Sur les coeffieients des series trigonometriques convergentes spheriquement. C. R. Acad. Sei., Paris, Sero A 289, No. 4, 159-161, Zbl. 335.42005
Cooke, R. (1971): A Cantor-Lebesgue theorem in two dimensions. Proc. Am. Math. Soc.
90, No. 4, 547-550, Zbl. 222.42014.
Fefferman, C. (1970): Inequalities for strongly singular integrals. Acta Math. 124, 9-36,
Zbl. 188,426
Fefferman, C. (1970): The multiplier problem for the ball. Ann. Math., H. Sero 94,330-336,
Zbl. 234.42009.
Fefferman, C. (1971): On the divergence of multiple Fourier series. Bull. Am. Math. Soc.
77, No. 2, 191-195, Zbl. 212,94.
Gareia-Cuerva, J. and Rubio de Franeia, J. L. (1985): Weigthed norm inequalities and
related topics. Amsterdam: North-Holland (604 p.), Zbl. 578.46046.
Goffman, C. and Liu, Fon-Che (1972): On the localization property of square partial sums
for multiple Fourier series. Stud. Math. 44, No. 1, 61-69, Zbl. 241.42016.
Golubov, B. I. (1975): On the convergence of spherieal Riesz means of multiple Fourier
series. Mat. Sb., Nov. Sero 96, 189-211. English translation: Math. USSR, Sb. 25, (1976),
Zbl. 305.42022.
Golubov, B. I. (1977): On the summability of Fourier integrals of spherieal Riesz means.
Mat. Sb., Nov. Sero 104, No. 4, 577-596. English translation: Math. USSR, Sb. 99,
501-518 (1977), Zbl. 364.42019.
Golubov, B. I. (1982): Multiple Fourier series and Fourier integrals. In: Itogi Nauki Tekh.
Sero Mat. Anal. 19, 3-54. English translation: J. Sov. Math. 24, 639-673 (1984), Zbl.
511.42021.
de Guzman, M. (1981): Real variable methods in Fourier analysis. Amsterdam New York
Oxford: North-Holland, Zbl. 449.4200l.
Herz, C. S. (1962): Fourier transforms related to convex sets. Ann. Math., H. Sero 75, No. 2,
81-92, Zbl. 111,348.
Hrmander, L. (1960): Estimates for translation invariant operators in LP spaces. Acta
Math. 104,93-140, Zbl. 93,114.
Hrmander, L. (1968): The spectral function of an elliptic operator. Acta Math. 121,
No. 3-4, 193-218, Zbl. 164,132,
Hrmander, L. (1969): On the Riesz means of spectral functions and eigenfunction expansions for elliptic differential operators. In: Some recent advances in the basic seiences,
93
Vol. 2 (Proe. Annual Sei. Conf., Belfer Grad. School Sei., Yeshiva Univ., New York,
1965-66), pp. 155-202. New York: Belfer Graduate School, Yeshiva Univ.
Hrmander, L. (1983-85): The analysis oflinear partial differential operators I-IV. (Grundlehren 256, 257, 274, 275.) Berlin Heidelberg: Springer-Verlag, Zbl. 521.35001, Zbl.
521.35002, Zbl. 601.35001, Zbl. 612.35001.
Hua Loo-Keng (1959): Abschtzungen von Exponentialsummen und ihre Anwendung in
der Zahlentheorie. Leipzig: Teubner (169 p.), Zbl. 83,36.
Hunt, R. A. (1968): On convergence of Fourier series, in: Proc. Conf. on Orthogonal Expansions and their Continuous Analogues, Edwardsville, pp. 235-255. Carbondale: Univ.
Press, Zbl. 159,357.
Il'in, V. A. (1957): On uniform eigenfunction expansions under summation taken under increasing order ofthe eigenvalues. Dokl. Akad. Nauk SSSR 114, No. 4, 698-701, [Russian)
Zbl. 79,291.
Il'in, V. A. (1958): On convergence of eigenfunction expansions of the Laplace operator.
Usp. Mat. Nauk 13, No. 1, 87-180, [Russian] Zbl. 79,292.
Il'in, V. A. (1968): The problems of localization and convergence of Fourier series with
respect to the fundamental systems of functions of the Laplace operator. Usp. Mat.
Nauk 23, No. 2, 61-120. English translation: Russ. Math. Surv. 23, No. 2, 59-116 (1968),
Zbl. 189,357.
Il'in, V. A. (1970): Localization conditions for rectangular partial sums of the multiple
trigonometrie Fourier series in the classes of S. M. Nikol'skii. Mat. Zametki 8, No. 5,
595-606. English translation: Math. Notes 8, 803-809 (1971), Zbl. 212,95.
Il'in, V. A. and Alimov, Sh. A. (1971): Conditions for the convergence ofspectral expansions
corresponding to selfadjoint extensions of elliptic operators, I. Differ. Uravn. 7, No. 4,
670-710. English translation: Differ. Equations 7, 516-543 (1973), Zbl. 224.35014.
Il'in, V. A. and Alimov, Sh. A. (1972): On the divergence on a set of positive measure
of Riesz means of kerneis of fractional order. Differ. Uravn. 8, No. 8, 372-373. English
translation: Differ. Equations 8, 283-284 (1974), Zbl. 244.35033.
Kaczmarz, S. and Steinhaus, H. (1951): Theorie der Orthogonalreihen. New York: Chelsea,
Zbl. 45,336.
Khavin, V. P. (1987): Methods and structure of commutative harmonie analysis. In: Itogi
Nauki Tekhn., Sero Sovrem. Probl. Mat. 15,6-133. Moscow: VINITI. English translation:
Encyclopaedia Math. Sei. Vol. 15, pp. 1-111. Berlin Heidelberg: Springer-Verlag (1991)
Kislyakov, S. V. (1987): Classical themes of Fourier analysis. In: Itogi Nauki Tekhn., Sero
Sovrem. Probl. Mat. 15, 135-195. Moscow: VINIT1. English translation: Encyclopaedia Math. Sei. Vol. 15, pp. 113-165. Berlin Heidelberg: Springer-Verlag (1991), Zbl.
655.42006.
Konovalov, S. P. (1979): On absolute convergence of multiple Fourier series. Mat. Zametki
25, No. 2, 211-216. English translation: Math. Notes 25,109-112 (1979), Zbl. 398.42014.
de Leeuw, K. (1965): On LV multipliers. Ann. Math., 11. Sero 81,364-379, Zbl. 171,118.
Levitan, B. M. (1955): On the summation of multiple Fourier series and Fourier integrals.
Dokl. Akad. Nauk SSSR 102, No. 6, 1073-1076, [Russian] Zbl. 68,279.
Lizorkin, P. 1. (1963): Generalization of Liouville differentiation and the function space
L~(En). Mat. Sb., Nov. Sero 60, No. 3, 325-353, [Russian] Zbl. 134,317.
Mareinkiewicz, J. (1939): Sur les multiplicateurs des series de Fourier. Stud. Math. 8,
78-91, Zbl. 20,354.
Maslov, V. P. (1965): Perturbation theory and asymptotic methods. Moscow: MGU. French
translation: Paris: Dunod, Gauthiers-Villars 1972, Zbl. 247.47010.
Meshkov, V. Z. (1978): On spherical multipliers. Mat. Zametki 23, No. 1, 105-112 English
translation: Math. Notes 23, 58-62 (1978), Zbl. 387.42005.
Mikhlin, S. G. (1956): On multipliers of Fourier integrals. Dokl. Akad. Nauk SSSR 109,
701-703, [Russian) Zbl.. 73,84.
MitchelI, J. (1951): On the spherical summability of multiple orthogonal series. Trans. Am.
Math. Soc. 71, 136-151, Zbl. 44,286.
94
Mityagin, B. S. and Nikishin, E. M. (1973a): On a. e. divergenee of Fourier series. Dokl.

Akad. Nauk SSSR 210, No. 1,23-25. English translation: Sov. Math., Dokl. 14,677--680
(1973), Zbl. 287.42009.
Mityagin, B. S. and Nikishin, E. M. (1973b): On divergenee of spectral expansions in the
mean and a. e. Dokl. Akad. Nauk SSSR 212, No. 3, 551-552. English translation: Sov.
Math., Dokl. 14, 1417-1419 (1973), Zbl. 298.35045.
Nikishin, E. M. (1972a): Weyl multipliers for multiple Fourier series. Mat. Sb., Nov. Sero
89, No. 2, 340--348. English translation: Math. USSR, Sb. 18, 351-360 (1973), Zbl.
258.42029.
Nikishin, E. M. (1972b): A resonanee theorem and series in eigenfunctions of the Laplace
operator. Izv. Akad. Nauk SSSR, Sero Mat. 96, No. 4, 795-813. English translation:
Math. USSR, Izv. 6,788-806 (1973), Zbl. 258.42018.
Nikol'skii, S. M. (1977): Approximation of functions of several variables and imbedding
theorems. 2nd ed. Moseow: Nauka (456 pp.). English translation: New York: SpringerVerlag 1975 (Grundlehren 205), Zbl. 185,379.
Oskolkov K. I. (1974): An estimate for the rate of approximation of a eontinuous function
and its eonjugate function by Fourier sums on a set of positive measure. Izv. Akad.
Nauk SSSR, Sero Mat. 98, No. 6, 1393-1407. English translation: Math. USSR, Izv. 8,
1372-1386 (1976), Zbl. 307.42002.
Panferov, V. S. (1975): Analogues of the theorems of Luzin-Denjoy and Cantor-Lebesgue
for double trigonometrie series. Mat. Zametki 18, No. 5, 659--674. English translation:
Math. Notes 18,983-992 (1976), Zbl. 369.42016.
Peetre, J. (1964): Remarks on eigenfunction expansions for elliptie operators with eonstant
eoefficients. Math. Seand. 15, 83-92, Zbl. 131,98.
Podkorytov, A. N. (1984): Intermediate rates of growth of the Lebesgue eonstants in the
two dimensional case. Zap. Nauchn. Semin. Leningr. Otd. Mat. Inst. Steklova 199, 148155, [Russian] Zbl. 601.42021.
Pulatov, A. K. (1978): On the loealization of speetral expansions eonnected with homogeneous elliptie operators. Dokl. Akad. Nauk SSSR 248, No. 4, 80O--S03. English translation: Sov. Math., Dokl. 19, 107-110 (1978), Zbl. 393.35048.
Pulatov, A. K. (1981): On eonditions for the loealization of expansions in N-fold Fourier
integrals eorresponding to a homogeneous elliptie operator. In: Differential equations
and questions of bifureation theory. Tashkent, 110--123, [Russian] Zbl. 533.47044.
Randol, B. (1969): On the asymptotieal behaviour ofthe Fourier transforms ofthe indieator
functions of a eonvex set. Trans. Am. Math. Soe. 199,279-285, Zbl. 183,269.
Reves, G. and Szasz, O. (1942): Some theorems on double trigonometrie series. Duke Math.
J. 9, 693-705, Zbl. 60,191.
Shapiro, V. L. (1957): Uniqueness of multiple trigonometrie series. Ann. Math., 11. Sero 66,
No. 2, 467-480, Zbl. 89,278.
Sjlin, P. (1971): Convergenee almost everywhere of eertain singular integrals. Ark. Mat.
9, No. 1,65-90, Zbl. 212,417.
Sobolev, S. L. (1950): Applieations of funetional analysis in mathematieal physies. Leningrad: LGU (256 pp.). English translation: Providenee, Rhode Island. Am. Math. Soe.
1963 (239 pp.) Zbl. 123,90.
Sokol-Sokolowski, K. (1947): On trigonometrie series eonjugate to Fourier series in two
variables. Fundam. Math. 94, 166-182, Zbl. 33,113.
Stein, E. M. (1958): Loealization and summability of multiple Fourier series. Aeta M~th.
100, No. 1-2, 93-147, Zbl. 85,284.
Stein, E. M. (1961): On eertain exponential sums arising in multiple Fourier series. Ann.
Math., 11. Sero 79, No. 2, 87-109, Zbl. 99,55.
Stein, E. M. (1970): Singular integrals and differentiability properties of funetions. Prineeton: Prineeton Univ. Press (290 pp.), Zbl. 207,135.
Stein, E. M. and Weiss, G. (1971): Introduetion to Fourier analysis on Euelidean spaces.
Prineeton: Prineeton Univ. Press, Zbl. 232.42007.
95
Sunouehi, Gen-Ichir (1985): On the summability almost everywhere of multiple Fourier

series at the eritieal index. Kodai Math. J. 8, No. 1, 1-4, Zbl. 575.42017.
Szasz, O. and Minakshisundaram, S. (1947): On absolute eonvergenee of multiple Fourier
series. Trans. Am. Math. Soe. 61, No. 1,36-53, Zbl. 54,30.
Taylor, M. E. (1981): Pseudodifferential operators. Prineeton: Prineeton Univ. Press. (499
pp.), Zbl. 453.47026.
Tevzadze, N. R. (1970): On the eonvergence of the double Fourier series of a square
summable funetion. Soobshch. Akad. Nauk Gruz. SSR 58, No. 2, 277-279, [Russian]
Zbl. 203,69.
Titehmarsh, E. C. (1958): Eigenfunetion expansions associated with seeond order differential equations, II. Oxford: Clarendon Press (404 pp.), Zbl. 97,276.
Tonelli, L. (1928): Serie trigonometriehe. Bologna: Nicole Zanichelli (527 pp.), Jbueh 54,298.
Trigub, R. M. (1980): Absolute eonvergenee of Fourier integrals, summability of Fourier
series and approximtion by polynomials of functions on the torus. Izv. Akad. Nauk
SSSR, Sero Mat. 44, No. 6, 1378-1409. English translation: Math. USSR, Izv. 17,567593 (1981), Zbl. 459.42019.
Wainger, S. (1965): Special trigonometrie series in k dimensions. Mem. Am. Math. Soe. 59
(98 pp.), Zbl. 136,366.
Welland, G. V. and Ash, J. M. (1972): Convergenee, uniqueness and summability of multiple Fourier series. Trans. Am. Math. Soe. 163, 401-436, Zbl. 233.42014.
Yanushauskas, A. I. (1986): Multiple Fourier series. Novosibirsk: Nauka, Sibirskoe Otd.
(272 pp.). [Russian] Zbl. 604.42012.
Yudin, V. A. (1979): A lower bound for the Lebesgue eonstant. Mat. Zametki 25, No. 1,
119--122. English translation: Math. Notes 25, 63--{j5 (1979), Zbl. 398.42016.
Zhizhiashvili, L. V. (1969): Conjugate functions and trigonometrie series. Tbilisi: Izd-vo
Tbilisk. Gos. Un-ta. [Russian]
Zhizhiashvili, L. V. (1970): On the divergenee of Fourier series. Dokl. Akad. Nauk SSSR
194, No. 4, 758-759. English translation: Sov. Math., Dokl. 11, 1281-1283 (1971), Zbl.
217,140.
Zhizhiashvili, L. V. (1971): On eonvergenee and divergenee of Fourier series. Dokl. Akad.
Nauk SSSR 199, No. 6, 1234-1236. English translation: Sov. Math., Dokl. 12,1261-1263
(1971), Zbl. 259.42033.
Zhizhiashvili, L. V. (1973): On some questions in the theory of simple and multiple trigonometrie series. Usp. Mat. Nauk 28, No. 2, 65-119. English translation: Russ. Math. Surv.
28, No. 2, 65-127 (1973), Zbl. 259.42027.
Zygmund, A. (1959): Trigonometrie series I, II (2nd edition). Cambridge: Cambridge Univ.
Press, Zbl. 85,56.
Zygmund, A. (1972): A Cantor-Lebesgue theorem for double trigonometrie series. Stud.
Math. 43, No. 2, 173-178, Zbl. 214,325.
(1979) Harmonie analysis in Euelidean spaces. Proe. Symp. Pure Math. 35, part I, II.
Providenee, Rhode Island: Am. Math. Soe., Zbl. 407.00005, Zbl. 407.00006.

Multiple Fourier Series and Fourier Integrals

Uploaded by

Copyright:

Available Formats

Multiple Fourier Series and Fourier Integrals

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Multiple Fourier Series and Fourier Integrals

Uploaded by

Copyright:

Available Formats

I.

Multiple Fourier Series and

Sh. A. Alimov, R. R. Ashurov, A. K. Pulatov

2.3. Uniform Convergence of Spherical Sums . . . .

I. Multiple Fourier Series and Fourier Integrals

(i. e. the subset of CN of the form

= nlXl + ... + nNXN ,

Here, as usual, ZN is the set of all vectors with integer eomponents: n

where the eoefficients Cn are arbitrary eomplex numbers, looks exactly as

Sh. A. Alimov, R. R. Ashurov, A. K. Pulatov

Trigonometrical series on the realline ~l may be found already in the 18th

sin nl Xl sin n2x2 sin n3X3

n1=1 n2=1 na=l

The method 01 Fourier, applied with success in mathematical physics under

where . = a~\ + a~2 2 is the Laplace operator. If the boundary 8n of the

I. Multiple Fourier Series and Fourier Integrals

= L VII (x)(all eos vt + bll sin vt).

In the ease of a rectangular membrane

< Xl < 7r, 0 <

= 1,2, ... ,j = 1,2.

The expansion of f then takes the form

~ fnln2 sinnlxI sinn2X2,

42 {21f (21f f(xI, X2) sin nixi sin n2 X2 dx l dx 2.

The function g(x) admits an analogous expansion. It is not hard to see

Sh. A. Alimov, R. R. Ashurov, A. K. Pulatov

2. What Do We Mean by the Sum of the Series (1)1 The peeularities of

can be written in the complex form (1) by putting

The "natural" partial sum

I. Multiple Fourier Series and Fourier Integrals

then takes the form

tion generalizes to the multidimensional case in three obvious manners, each

Despite the apparent "naturality" of this definition, it is not deprived of

It is clear that Sm(O)

2 so that the senes (6) converges

rectangularly to zero at the point (0,0), while Sml,l(O) =

Sh. A. Alimov, R. R. Ashurov, A. K. Pulatov

with eoefficients Cn 1n2 = n~6n20 - n~6nlO, where again 6 jk is the Kronecker

The series (1) is circularly converyent (or spherically

limR.... oo 8R(X) exists.

I. Multiple Fourier Series and Fourier Integrals

D) Let 0 = {w} a family of finite subsets of ZN having the following

In this case (1) is O-convergent if lim 8>. (x, Q) exists. In particular, if Q

Sh. A. Alimov, R. R. Ashurov, A. K. Pulatov

I. Multiple Fourier Series and Fourier Integrals

The operator 6. is considered in the Hilbert space L2(T N ) as an unbounded

the operator -6. is symmetrie and nonnegative. Consequently, by Friedrichs's

It is easy to check that the operators

have the form

where In are the Fourier coeflicients of the function I E L2(l'N) defined

1 COO(TN) is the dass of infinitely differentiable functions on TN which are 211'-periodic

Vu = (~, ... , (J8u ) is the gradient of u.

Sh. A. Alimov, R. R. Ashurov, A. K. Pulatov

We associate with each differential operator (14) its symbol, an algebraie

where {= ({I, ... ,{N), {a = 1 . . . fft.

is positive definite, i. e. if for any { E ]RN, { #- 0, we have A o({) > O.

If A(D) is also elliptie then by the well-lrnown Grding's inequality (cf.

Therefore, by Friedrichs's theorem just mentioned, A(D) has a selfadjoint

where {E>.} is the eorresponding resolution of identity. As in the ease of the

Unfortunately, the eorresponding partial sum expansions do not refer to

I. Multiple Fourier Series and Fourier Integrals