Exponential sum estimates in finite commutative rings and applications (Q2474575): Difference between revisions

From the text: The present paper has a twofold purpose. First it aims at unifying several recent results on exponential sum estimates, obtained by combinatorial methods (centered around sum-product set results). In particular, new estimates on exponential sums have been established in the following contexts: (i) prime fields \(\mathbb F_p\) [[B-G-K] \textit{J. Bourgain, A. Glibichuk} and \textit{S. Konyagin}, J. Lond. Math. Soc. (2) 73, 380--398 (2006; Zbl 1093.11057); (ii) general finite fields [\textit{J. Bourgain} and \textit{M. Chang}, C. R., Math., Acad. Sci. Paris 342, 643--646 (2006; Zbl 1127.11081)]; (iii) powers of prime fields \(\mathbb F_p \times\cdots\times \mathbb F_p\) [[B] \textit{J. Bourgain}, J. Am. Math. Soc. 18, 477--499 (2005; Zbl 1072.11063)]; (iv) residue classes \(\mathbb Z_q\) [[B-C3] \textit{J. Bourgain} and \textit{M. Chang}, Geom. Funct. Anal. 16, 327--366 (2006; Zbl 1183.11047)] assuming \(q\) with few prime factors and in [[B2] \textit{J. Bourgain}, J. Anal. Math. 97, 317--355 (2005; Zbl 1183.11045). in the general case). Most of the key ideas do in fact appear in [B-G-K] and [B]. The general scheme in these works is as follows. First, there is a `sum-product' theorem where the input is combinatorial (in particular, the Plünnecke-Ruzsa sumset theory) and algebraic, depending on the particular algebraic structure under consideration. Next, in order to derive the exponential sum bound for `large' multiplicative subgroups, we rely essentially (besides rather standard Fourier analysis) on the `Balog-Szemerédi-Gowers Theorem', which is of a graph theoretical nature. It allows us to pass from set-theoretical statements (such as a sum-product theorem) to results of a more `statistical nature' that may be phrased as inequalities on additive and multiplicative convolution of discrete measures. With the sum-product theorem at hand, this second step is quite similar to the argument given in the first paper [B-G-K] in the prime field case. Some adjustments need to be made in the other cases, mainly due to the presence of zero-divisors, but only requiring minor changes. From the algebraic point of view, the reasoning involved in [B] and [B-C3] is quite similar; and trying to present these in a unified setting was tempting. We have achieved this here in the context of a finite commutative ring \(R\) with unit, such that \(q = |R|\) has only few divisors. The main result stated in Theorems 1 and 2 below claims roughly that if \(H < R^*\) is a multiplicative subgroup, \(H > q^\varepsilon\), then there is a nontrivial exponential sum bound of the form \[ \max_\chi |\sum_{x\in H}\chi(x)<| H|^{1-\delta} \tag{0.1} \] for some \(\delta > 0\) (\(\chi\) running over the nontrivial additive characters of \(R\)) unless \(H\) is `largely contained' in a proper subring \(R_1\), \(1\in R_1\) or \(H-1\) in a proper ideal \(I\subset R\) (in which cases, one can obviously not expect (0.1)). This result easily applies in certain situations that would be rather lengthy to analyze on a case by case basis, using the methods from [B], [B-C], [B-C3]. More specifically, there are the following two applications (our second purpose). (1) The Egami problem on the minimum norm of representatives of residue classes in quotients \(O_K/I\), where \(I\) is an integral ideal in the integers of an algebraic number field \(K\) [cf. [B-C2] \textit{J. Bourgain} and \textit{M. Chang}, Duke Math J. 138, No. 2, 263--280 (2007; Zbl 1139.11035), \textit{S. Konyagin} and \textit{I. Shparlinski}, Character Sums with Exponential Functions and their Applications, Cambridge University Press, Cambridge (1999; Zbl 0933.11001)]). Theorem 3 below refines the results obtained in [B-C2] because we are now able to treat situations not covered by [B-C] or [B-C3], which are the analytical input in [B-C2]. (2) The general Heilbronn problem, as described by \textit{R. Odoni} [Math. Proc. Comb. Phil. Soc. 98, 389--396 (1985; Zbl 0579.10016)]. The problem consists in estimating nontrivially exponential sums of the form \[ \sum_{x=1}^pe_{p^m}\left(f(x^{p^{m-1}})\right), \] where \(m\geq 2\) is fixed, \(f(x)\in\mathbb Z[X]\) is a polynomial of degree \(d\) (also assumed fixed) and \(p\) a large prime. If \(m = 2\), \(f(x) = x\), such estimates were obtained by \textit{R. Heath-Brown} [in: Analytic Number Theory, Vol 2, Birkhäuser, 1996, Prog. Math. 139, 951--963 (1996; Zbl 0857.11041)] and \textit{R. Heath-Brown} and \textit{S. Konyagin} [Q. J. Math 51, 221--235 (2000; Zbl 0983.11052)] using Stepanov's method. In [B-C3], the general case \(m\geq 2\), \(f(x) = x\) is treated, using the sum-product result in \(\mathbb Z_q\). An alternative approach, starting from the \(m = 2\) result, is due to \textit{Yu. V. Malykhin} [Math. Notes 80, No. 5, 748--752 (2006); translation from Mat. Zametki 80, No. 5, 793--796 (2006; Zbl 1183.11048)]. We are not aware of any results in this context when \(f(x) \neq x\).