Vous êtes sur la page 1sur 7

Remark on the divergence of Fourier series

K. S. Kazaryan

Abstract. The problem of unconditional convergence almost everywhere of


series ∞ ∞
P
n=1 cn ϕn (x) where {ϕn (x)}n=1 is an orthonormal system and {cn } ∈
l , in relation to the completeness of system {ϕn (x)}∞
2
n=1 on subsets of com-
plete measure, is studied. Are given examples of orthonormal systems that
are uniformly bounded and complete on subsets with measures close to the
complete
P∞ measure, but, despite this, for any sequence {cn } ∈ l2 the series
n=1 cn ϕn (x) converges unconditionally almost everywhere. It is shown that
there exist subsystems of trigonometric and Walsh systems that are not com-
plete on any set of positive measure and are not systems of unconditional
convergence. A new criterion based on the metric properties of the system is
proposed, which makes it possible to clarify the incompleteness everywhere of
a given system.

In 1927, Kolmogorov [1] formulated the following assertion:


there exists a function f (x) ∈ L2 [0, 2π] such that the terms of its trigonometric
Fourier series can be rearranged in such a way that the resultant series diverges
almost everywhere on [0, 2π].
It was nearly 40 years later before Zahorski [2] gave a proof of this assertion.
Afterwards Ul’yanov proved a similar assertion for series in Haar and Walsh sys-
tems. Ul’yanov [4, 5], and also Olevskii [6, 7], proved the following
Theorem (Ul’yanov, Olevskii). Any complete orthonormal system can be rear-
ranged in such a way that the Fourier series of some function from L2 with respect
to the resultant system diverges almost everywhere.
Ul’yanov proved a more general fact in terms of bases in L2 , but in this paper
we will consider only orthonormal systems, and therefore we will not deal with
generalizations along these lines. The Ul’yanov-Olevskii theorem naturally leads to
the following question: to what extent the completeness condition in this theorem
can be weakened for particular orthonormal systems?
One direction of research that can lead to the solution of the problems is the
study of subsystems of a given complete orthonormal system that are complete on
subsets of the domain.
The problem of describing subsystems of classical orthonormal systems that are
complete on subsets of the domain has been dealt by the American mathematicians
Price and Zink [8],[9]. For the Haar system, they offered a complete solution of the
problem in a joint paper [9]. For trigonometric and Walsh systems, however, the
problem has yet to be resolved. Let us now give some definitions that we will use
in what follows.
1
2 K. S. KAZARYAN

Definition 1. We will say that a system of functions {fn (x)}∞n=1 is approx-


imately complete on measurable set E of positive measure if for any ε > 0 there
exists a measurable set Eε ⊂ E, |Eε > |E| − ε such that the system {fn (x)}∞
n=1 is
complete in L2 (Eε ).
Definition 2. We will say that a system of functions {fn (x)}∞ n=1 defined on
a segment [a, b] is nowhere complete in L2 ([a, b]) if for any subset F ⊂ [a, b] of
positive measure the system {fn (x)}∞ 2
n=1 is not complete in L (F ).

Definition 3. An orthonormal system {ϕn (x)}∞ n=1 is called


P∞ a 2convergence

system if for any sequence
P of coefficients {cn } n=1 such that n=1 cn converges
implies that the series n=1 cn ϕn (x) converges almost everywhere.
Definition 4. An orthonormal system {ϕn (x)}∞ n=1 is called an unconditional
convergence system if any permutation of the system is a convergence system.
In [10] Price proved that the systems {sin 2n πx}∞ n ∞
n=1 , {cos 2 πx}n=0 defined on
[0, 1] and the Rademacher system are nowhere complete in measure. He employed
the following criterion of incompleteness for the proof:
S for any ε > 0 there exist nonintersecting intervals {∆n } such that
if
∆n = [0, 1], |∆n | < ε and for any ∆n the graphs of all functions of
(*) the system {fi (x)}∞ i=1 , beginning from some number, are symmetri-
cal on this intervals relative to its center or to a perpendicular drawn
through the center, then {fi (x)}∞ 2
i=1 is nowhere complete in L [0, 1].
Obviously, Price [10] was aware that subsystems of the Haar system will be un-
conditional convergence systems if and only if they are nowhere complete in L2 [0, 1].
This follows directly from the Ul’yanov-Nikishin theorem [12] and from the crite-
rion given by Price and Zink [9] that describes subsystems of the Haar system which
are complete on sets of positive measure. Probably for this reason, and also in view
of the fact that the orthonormalized systems {sin 2n πx}∞ n ∞
n=1 , {cos 2 πx}n=0 and
Rademacher system are nowhere complete Price conjectured that an orthonormal
system will be an unconditional convergence system if an only if it is a nowhere
complete in L2 [0, 1].
From the following theorem it implies that even uniformly bounded orthonor-
mal unconditional convergence systems can be complete in subspaces L2 (F ), where
the measure of the set F can be close to the complete measure.
Theorem 1. For any a, 0 < a < 1 on [0, 1] there exists a uniformly bounded
orthonormal system {ϕn (x)}∞ 2
n=1 that is complete in L [0, a] and is an unconditional
convergence system.
Proof. A system {ϕn (x)}∞ n=1 that satisfies the conditions of Theorem 1 can
be written out simply. Let
 q
 21n a2 sin πa nx,
 for x ∈ [0, a]
ϕn (x) = 1 − 2−2n

p rn [a + (1 − a)x], for x ∈ (a, 1), n = 1, 2, ..
(1 − a)
where {rn (x)}∞ ∞
n=1 is the Rademacher system. That {ϕn (x)}n=1 is an orthonormal

system can be verified directly. Completeness of {ϕn (x)}n=1 follows immediately
∞ 2
P∞known result that the system {sin nx}n=1 is complete in L [0, π]. That
from the well
the series n=1 cn ϕn (x) is unconditionally convergent almost everywhere on [0, 1]
REMARK ON THE DIVERGENCE OF FOURIER SERIES 3

for {cn }∞ 2 1 ∞ 1
n=1 ∈ l is implied, on the one hand, by the fact that { 2n cn }n=1 ∈ l ,and,

on the other hand, by the fact that the system {rn (x)}n=1 satisfies to the conditions
of Theorem 1. 
Now we will prove a theorem which yields that the trigonometric and the Walsh
systems contain subsystems that are nowhere complete and are not unconditional
convergence systems. Thus, the situation for these systems differs substantially
from the case of the Haar system.
Definition 5. We will say that system {fi (x)}∞ i=1 of functions defined on
[0, 1] possesses (Γ) property if for any natural number m and N there exists a
natural number n > m such that all functions fj (2n x) 1 ≤ j ≤ N belong to the
system {fi (x)}∞i=1 where it is assumed that functions fi (x)(i = 1, 2, ...) are continued
periodically onto the real axis.
Theorem 2. Assume that {fi (x)}∞ i=1 is a complete orthonormal system with
(Γ) property. Then there exists a subsystem {fnj (x)}∞ ∞
j=1 of the system {fi (x)}i=1
∞ ∞
such that {fnj (x)}j=1 is nowhere complete on [0, 1] and {fnj (x)}j=1 is an uncon-
ditional convergence system.
Proof. By the Ul’yanov-Olevskii theorem there exists a series in the system
{fi (x)}∞i=1 , that is not an unconditionally convergent series almost everywhere.
Afterwards, using the (Γ)-property of the system {fi (x)}∞ i=1 , we construct a sub-
system of the given system such that it satisfies criterion (*) of incompleteness, and
we can write a series that converges in L2 [0, 1], whose partial sums (in the sense of
convergence almost everywhere) behave “badly”, like the series taken originally.
By Ul’yanov’s theorem (see [5], p. 132), we obtain that there exists a sub-
system {fnj (x)}∞ j=1 , that is nowhere complete on [0, 1] and coefficients {cj }∞ j=1 ,
P∞ 2 P∞
j=1 jc < ∞ such that the series j=1 cj fn j (x) diverges unboundedly after some
permutation.
Assume that {ai }∞ i=1 is a sequence such that

X
(1) a2i = 1
i=1

and for some permutation {ϑn }∞


n=1 of natural numbers the series
X∞
aϑn fϑn (x)
n=1

diverges unboundedly almost everywhere. Hence for any natural l we can choose a
number kl and set El such that
s
X
aϑn fϑn (x) > 4l , x ∈ El ,

(2) sup
1≤s<k l n=1

1
(3) El ⊂ [0, 1]; |El | > 1 −
.
2l
We choose natural numbers 0 = p0 < p1 < p2 < ... in such a way that functions
fj (2pl x)(1 ≤ j ≤ max1≤i<kl ϑi ) do not belong to the set of functions
{fj (2pl x), 1 ≤ j ≤ max ϑi ; s = 1, 2, ..., l}
1≤i<ks
4 K. S. KAZARYAN

Hence, in view of the fact that the system {fn (x)}∞


n=1 possesses (Γ) property the
system
(4) {fj (2ps x); 1 ≤ j ≤ max ϑi }∞
s=1
1≤i<ks

is a subsystem of {fn (x)}∞


n=1 and satisfies incompleteness condition (*). Conse-
quently, the system (4) is nowhere complete on [0, 1]. We will show that we can
write a series with respect to some permutation of this system which converges in
L2 [0, 1] and diverges unboundedly almost everywhere on [0, 1]. Consider the series
∞ kXl −1
X 1
(5) a f (2pl−1 x).
l ϑn ϑn
n=1
2
l=1

It is obvious from conditions (2) and (3) that the series (5) diverges unboundedly
on the set E = lim sup Ei , i.e. almost everywhere on [0, 1], since from (3) we have
|E| = 1. By (1) we have
∞ X kl  2 X∞
X 1 1
a
l ϑn
≤ l
< 1.
n=1
2 4
l=1 l=1

Remark 1. In view of Ul’yanov’s theorem ([15],p.132) that complete orthonor-
mal systems are not unconditional convergence systems in the sense of summability
for extensive classes of summation methods, Theorem 2 can be similarly proved for
these summation methods as well.
Theorem 2 directly implies the following corollaries.
Corollary 1. There exists a nowhere complete in measure subsystem of the
trigonometric system that is not an unconditional convergence system.
Corollary 2. There exists a nowhere complete in measure subsystem of the
Walsh system {Wn (x)}∞
n=1 that is not an unconditional convergence system.

We should note that the subsystems whose existence is claimed in Corollaries


1 and 2 can be written out precisely. For example, for the trigonometric system
the subsystem
{cos 2i nx, sin 2i nx; n = 1, 2, ..., 2i }∞
i=1
satisfies the conditions of Corollary 1.
Thus, adding the condition of uniform boundedness we can not save Price’s con-
jecture: the fact that a uniformly bounded orthonormal system is nowhere complete
does not imply that it is an unconditional convergence system, and, conversely, if
a uniformly bounded orthonormal system is an unconditional convergence system,
it is not necessary that the system is nowhere complete in measure.
As for weakening of the completeness condition in the Ul’yanov-Olevskii the-
orem, it seems at first glance that approximately completely uniformly bounded
orthonormal systems are not unconditional convergence systems.
Criterion (*) employed by Price is useful in constructing nowhere complete
systems, but it cannot be employed in general situations, since the criterion takes
account of only the structural features of functions. Here we will offer another
criterion for determining of systems nowhere complete in measure. This criterion
will be based on the metric properties of given systems.
REMARK ON THE DIVERGENCE OF FOURIER SERIES 5

It is well known that convergence in the metric


|f (t)|
Z
ρ(f ) = dt
[0,1] 1 + |f (t)|
is equivalent to convergence in measure in the space S of finite measurable functions.
Assume that (R, ρ1 ) is a complete metric space which is strictly contained in S
(there exists a function f ∈ S, that does not belong to R) and all functions that
coincide almost everywhere with some function from R are equivalent to it.
We will say that metric the ρ1 is stronger that metric ρ if convergence of any
sequence of functions in ρ1 implies that it converges in the metric ρ.
We have the following assertion.
Proposition 1. Let {gn (x)}∞ n=1 ⊂ R be a system of functions defined on the
segment [0, 1] and let E ⊂ [0, 1] be a measurable subset. Suppose that for any series
in the system {gn (x)}∞n=1 the convergence of some subsequence of its partial sums
in measure on E implies convergence of some subsequence of partial sums of the
series on E in the stronger metric ρ1 , then the system of functions {gn (x)}∞n=1 is
not complete in measure on E.
Proof. Proposition 1 can be proved by Talalyan’s [11] theorem (see also pa-
per [13]), which claims that removing a finite number of functions from a system
{fn (x)}∞
n=1 complete in measure on E we obtain a subsystem complete in measure
on E. Assume that {gn (x)}∞n=1 is complete in measure on E. Then for any function
f ∈ S we can define a sequence of linear combinations in the system {gn (x)}∞
n=1
nk+1
X
φk (x) = ai gi (x) (n1 < n2 < ...)
i=nk

such that  n 
X 1
ρ f− φk < n (n = 1, 2, ...).
2
k=1
P∞
This implies P convergence of the series k=1 φk in the metric ρ,i.e., convergence

of the series i=1 ai gi (x) with respect to the subsequence {nk }∞ k=1 . However, in
accordance withP∞ the conditions of Proposition 1, some subsequence of partial sums
of the series i=1 ai gi (x) should converge in the stronger metric ρ1 . Choosing
function f χE ∈ S in such a way that f χE does not belong to complete metric
space (R, ρ1 ), we arrive at a contradiction. 
By Proposition 1 we obtain the following criterion:
if for any measurable set E ⊂ [0, 1] the convergence in measure on E
of any series in terms of the system {gn (x)}∞n=1 implies convergence
of this series on E in a stronger metric, then the system {gn (x)}∞
n=1
is nowhere complete in measure on [0, 1].
In order to apply this criterion to multiplicative systems, which include the
trigonometric and Walsh systems, we recall some definitions and notations. We will
use the notation from a paper by Gaposhkin [14] because we are going to apply a
theorem from that paper. A system of complex-valued functions X = {χk (x)}∞ k=1 ,
defined on [0, 1] will be called a multiplicative orthogonal system if it is orthonormal
and, together with any two functions χk1 and χk2 , functions χ−1 k1 , χk1 χk2 also belong
to the system X. It is assumed that χ0 (x) ≡ 1. In the group of nonnegative
6 K. S. KAZARYAN

integers we define the operations of addition and subtraction in accordance with


the following rules:
ϑ = n u m, when χn χm = χϑ θ = n−̇m, if χn χ̄m = χθ .
For a trigonometric system the operations u and −̇ coincide with the operations
of addition and subtraction. Assume that N = {nk }∞ k=1 , nk ≥ 0, nk 6= nj for k 6= j
is some system of numbers. The system {χnk (x)}∞ k=1 for specified sequence N will
be denoted by X(N).
S∞
We will say that N ∈ D2−̇ , if N = i=1 ∆i , ∆l ∆j = ∅ if i 6= j, where ∆i
T
contains not more than s elements, and the number of solutions of the equation
nk1 −̇nk2 = ϑ (nk1 ∈ ∆i1 , nk2 ∈ ∆i2 , i1 > i2 )
for each ϑ ≥ 0 does not exceed A, A = const; N ∈ G−
2 , if

lim (nk1 −̇nk2 ) = ∞.


k1 6=k2 ,k1 −k2 →∞

From Theorem 4 of [14], it follows, in particular,Pthat if N ∈ D2−̇ G−̇


T
2 and the

absolute value of some partial sums of the series i=1 ci χki (x) in a multiplicative
orthogonal system are finite on some measurable set E of positive measure, then
P ∞ 2
i=1 |ci | < ∞. Hence from the above criterion of incompleteness we directly
obtain the following theorem.
Theorem 3. Multiplicative orthogonal system X(N), where N ∈ D2−̇ G−̇
T
2 , is
nowhere complete in measure on [0, 1].
Taking account of the remark to Theorem 4 in [14], we also obtain the following
assertion.
Theorem 4. If N ∈ D2−̇ then the system {eink 2πx }∞
k=1 is nowhere complete in
measure on [0, 1].
nk+1
Hence any lacunary trigonometric system {eink x }∞
k=1 where nk > λ > 1, is
nowhere complete in measure on [0, 2π].

References
[1] A. Kolmogoroff and D. Menschoff, Sur la convergence des series de fonctions orthogonales,
Math. Zeitschr., vol. 26 (1927), pp. 432-441.
[2] Z. Zahorski, Une serie de Fourier permutee d’une fonction de classe L2 divergent presque
partout, Compt. Rend. Acad. Sci. (Paris), vol. 251 (1966), pp. 501-503.
[3] P.L. Ul’yanov, Divergent series of class Lp , DAN SSSR, vol. 137, no. 4 (1961), pp. 786-789.
[4] P.L. Ul’yanov, Divergent series in the Haar system and in bases, DAN SSSR, vol. 138, no. 3
(1961), pp. 556-559.
[5] P.L. Ul’yanov, Divergent Fourier series, UMN, vol. 16, no. 3(99) (1961), pp. 61-142.
[6] A.M. Olevskii, Divergent series from L2 in complete systems, DAN SSSR, vol. 138, no. 3
(1961), pp. 545-548.
[7] A.M. Olevskii, Divergent Fourier series, Izv. AN SSSR. Seriya Matem., no. 27 (1965), pp.
343-366.
[8] J.J. Price, Sparse subsets of orthonormal systems, Proc. Amer. Math. Soc., vol. 35 (1961),
pp. 556-559.
[9] J.J. Price and R. E. Zink, On sets of completeness for families of Haar functions, Trans.
Amer. Math. Soc., vol. 119 (1965), pp. 262-269.
[10] J.J. Price, Topics in orthonormal fuctions, Amer. Math. Month., vol. 82 (1975), pp. 594-609.
[11] A.A Talalyan, Representation of measurable fuctions by series, UMN, vol. 15. no. 5 (1960),
pp. 77-141.
REMARK ON THE DIVERGENCE OF FOURIER SERIES 7

[12] E.M. Nikishin and P. L. Ul’yanov, Absolute and unconditional convergence, UMN, vol. 22,
no. 3 (1967), pp. 240-242.
[13] C. Goffman and D Waterman, Basic sequences in the space of measurable fuctions, Proc.
Amer. Math. Soc., vol. 11 (1960), pp. 211-213.
[14] V.F. Gaposhkin, Lacunary series in multiplicative systems of functions, Sib. Mat. Zhurn.,
vol. 12, no. 2 (1971), pp. 65-83.

Vous aimerez peut-être aussi