A Survey of Uniqueness Questions in Multiple Trigonometric Series

The issue is uniqueness of representation by multiple trigonometric series. Two basic uniqueness questions, one about series which converge to zero and the other about series which converge to an integrable function, are asked for each of four modes of convergence: unrestricted rectangular convergence, spherical convergence, square convergence, and restricted rectangular convergence. Thus there are eight basic questions in each dimension. In all dimensions, both uniqueness theorems are valid for unrestricted rectangular convergence, as is the …rst uniqueness theorem for spherical convergence. The second uniqueness theorem holds for circular convergence in dimension 2. All the other questions are still open. The positive work will be surveyed along with related work involving extensions of the Cantor-Lebesgue theorem. 1. Introduction In 1974 Marshall Ash gave a talk entitled “Multiple trigonometric series” in which were listed only 4 substantial uniqueness theorems for convergent multiple trigonometric series [A1]. They were 1 if a double trigonometric series is spherically convergent to zero everywhere, then all its coe¢ cients are zero, 2 if a double trigonometric series is spherically convergent to an everywhere …nite, Lebesgue integrable function, then it is the Fourier series of that function, and 3 if double trigonometric series is unrestrictedly rectangularly convergent to zero everywhere, then all its coe¢ cients are zero, and 4 if a double trigonometric series is unrestrictedly rectangularly convergent to an everywhere …nite, Lebesgue integrable function, then it is the Fourier series of that function. Not much happened for about 20 years, but in the last few years all of these results have been extended to d dimensions. We will outline how this was done and point out several of the obstacles to further progress. To begin with, we will sketch the very well known one dimensional uniqueness proof given by Cantor in 1870. Date : January 28, 2011. 1991 Mathematics Subject Classi…cation. Primary 42B99; Secondary 42B08, 42A63. Key words and phrases. Uniqueness, square convergence, spherical convergence, restricted rectangular convergence, unrestricted rectangular convergence, Cantor-Lebesgue theorem . The research of both authors were partially supported by a grant from Faculty and Development Program of the College of Liberal Arts and Sciences, DePaul University. 1 2 J. MARSHALL ASH AND GANG WANG Theorem 1.1 (Cantor). Suppose that the one dimensional trigonometric series P cne inx converges to zero in the sense that at each x; sn(x) = Pn = n c e i x tends to 0 as n tends to 1: Then all cn must be 0: The …rst step of the proof is to observe that at each x; c ne inx + cne = sn(x) sn 1(x) also tends to 0: The Cantor-Lebesgue theorem asserts that this does not happen because of cancellation between the two terms. Explicitly, Theorem 1.2 (Cantor-Lebesgue). If c ne +cne ! 0 for every x 2 R, then, q jc nj + jcnj ! 0. In other words, our …nal goal of getting all cn identically 0 has been partially achieved: at least the cn tend to 0 as jnj tends to 1: Next one forms the Riemann function F (x) = c0 x 2 2 P0 cn n2 e ; where the prime means that the sum is over all nonzero integers. This function has two important properties: 1) it is continuous, and 2) it has a generalized second derivative which is everywhere 0: From these two facts it quickly follows …rst that F must be linear and consequently that all cn are 0: We will see below that all known higher dimensional proofs except one more or less follow this approach of starting with a Cantor-Lebesgue type theorem. Even for the exceptional case of unrestricted rectangular convergent, one of the known proofs [AFR] follows the one dimensional approach. It will also be pointed out that most of the open questions seem totally inaccessible because of a breakdown at this very …rst step. In most settings the best possible Cantor-Lebesgue extension is known and in some settings the corresponding extension yields much, much less than the one dimensional Cantor-Lebesgue conclusion. Before leaving the one dimensional setting we will state one further theorem that also serves as a model for higher dimensional generalizations. Theorem 1.3 (de la Vallée-Poussin). If a one dimensional trigonometric series T is convergent to an everywhere …nite, Lebesgue integrable function f , then it is the Fourier series S[f ] of that function. Let us consider the relation between Cantor’s theorem and de la Vallée-Poussin’s theorem. By de…nition, S[f ] = P f̂(n)e; where f̂(n) = 1 2 R f(x)e dx: Since the Fourier series of the everywhere zero function has all coe¢ cients equal to zero, it is immediate that Cantor’s theorem is a special case of de la Vallée-Poussin’s. It is worth a moment’s thought to distinguish the questions. After all, Cantor’s theorem is called a uniqueness theorem because if T and S are two trigonometric series converging everywhere to the same value, then applying Cantor’s theorem to the series T S shows that all the coe¢ cients of T and S must agree. One might therefore try to deduce de la Vallée-Poussin’s theorem from Cantor’s theorem as follows. Let T converge to f 2 L: Then T S[f ] converges to 0 everywhere, so by Cantor’s theorem, T and S[f ] are identical. The fallacy in this argument is that S[f ] may very well fail to converge at all. In fact, Kolmogorov has given an example of an L function with an everywhere divergent Fourier series. The proof of de la Vallée-Poussin’s theorem requires all the machinery of the original theorem together with additional notions such as majorants and minorants from the theory of Lebesgue integration. UNIQUENESS OF MULTIPLE TRIGONOMETRIC SERIES 3 Although we could often save space by only discussing analogues of de la ValléePoussin’s theorem, it is worth distinguishing the cases because the proofs can be of quite di¤erent levels of di¢ culty, even in the one dimensional case. We will consider multiple trigonometric series from now on. We begin by introducing some de…nitions and notation used for the rest of the paper. As a general notation, one dimensional and multiple dimensional indices are both denoted by m;n; . We do not distinguish between them for the sake of notational simplicity. The dimensions of the indices will always be clear from the context. Let T be a multiple trigonometric series, so that for x 2 T we have T = X m2Zd cme : Fix x and set am = am(x) = cme: There are many ways to add up the terms of T = P am: We will consider 5 summation methods here. All methods will be symmetric, i. e., if am is included in a partial sum and if m0 di¤ers from m only by some coordinate signs so that mj = jmj j for every j; then am0 will also be included in that partial sum. De…nition 1.1. Let jmj = qP m 2 j and for each real r 0; de…ne T r = X