Vinogradov's mean value theorem via efficient congruencing (Q431649): Difference between revisions

The estimation of mean values of exponential sums plays a significant rule in analytic number theory, for example in connection with Waring's problem and zero-free regions for the Riemann zeta function. In particular, mean values of exponential sums of the form \[ f_k(\boldsymbol{\alpha},X)=\sum_{1\leq x\leq X} e\left(\alpha_1x+\dots+\alpha_kx^k\right), \] where \(\boldsymbol{\alpha}=(\alpha_1,\dots,\alpha_k)\) is a \(k\)-tuple of real numbers, are of great importance. Vinogradov [\textit{I. M. Vinogradov}, Dokl. Akd. Nauk SSSR 8, 195--198 (1935; Zbl 0013.05303)] developed a new method to estimate mean values of the form \[ J_{s,k}(X)=\int_{[0,1)^k} \left|f_k(\boldsymbol{\alpha};X)\right| \text{d} \boldsymbol{\alpha}. \] There has been a long history of research on this mean value. Using a new method, which he refers to as efficient congruencing, the author obtains estimates for \(J_{s,k}(X)\) that for the first time approach those conjectured to be the best possible. In particular, he proves that the conjectured bound \[ J_{s,k}(X)\ll X^{\varepsilon}\left(X^s+X^{2s-\frac{1}{2}k(k+1)}\right) \] holds whenever \(s\) and \(k\) are natural numbers satisfying \(k\geq 2\) and \(s\geq k(k+1)\). If \(s\) is large enough compared to \(k\), it can be shown that an asymptotic formula of the shape \[ J_{s,k}(X) \sim C(s,k)X^{2s-\frac{1}{2}k(k+1)} \] holds, where \(C(s,k)\) is the product of a singular series and a singular integral. Let \(V(k)\) denote the least natural number \(s\) for which this asymptotic formula holds. The author also proves that, when \(k\geq 3\), one has \(V(k)\leq k^2+k+1\), which is only a factor 2 away from the best possible bound for \(V(k)\). As an important consequence of these estimates, the author makes significant progress on Waring's problem. Let \(R_{s,k}(n)\) denote the number of representations of the natural number \(n\) as the sum of \(s\) \(k\)-th powers of positive integers. A heuristic application of the Hardy-Littlewood circle method suggests an asymptotic formula for \(R_{s,k}(n)\) if \(k\geq 3\) and \(s\geq k+1\). Let \(\tilde{G}(k)\) denote the least integer \(t\) with the property that, for all \(s\geq t\) and all sufficiently large natural numbers \(n\), this asymptotic formula holds. Hardy and Littlewood [\textit{G. H. Hardy} and \textit{J. E. Littlewood}, Math. Z. 12, 161--188 (1922; JFM 48.0146.01)] established the bound \(\tilde{G}(k)\leq (k-2)2^{k-2}+5\). Since then, a lot of research has been done to improve this bound, carried out in particular by the author. In the present paper, the bound \(\tilde{G}(k)\leq 2k^2+2k-3\) is established for \(k\geq 2\). As further applications, the author considers bounds for \(f_k(\boldsymbol{\alpha},X)\) with single \(\alpha\) satisfying certain Diophantine properties, and small values of \(\min_{1\leq n\leq N} \|\alpha_1n+\alpha_2n^2+\dots+\alpha_kn^k\|\). It is outside the scope of this review to give a comprehensive account of the process of efficient congruencing. The main ideas are described briefly below. Vinogradov's classical approach starts off with choosing a prime \(p\) in a range of the form \([X^{\theta},2X^{\theta}]\), where \(\theta\) must satisfy the inequality \(\theta\leq 1/k\). Then one divides the exponential sum in question according to residue classes of \(x\) modulo \(p\). Now one applies Hölder's inequality to bound \(J_{s+k,k}(X)\) and evaluates explicitly the resulting integral, which leads to the problem of counting solutions to a system of certain congruences modulo powers \(p^j\) with \(1\leq j\leq k\), which can be expressed as counting solutions to a single congruence modulo \(p^k\). If \(\theta=1/k\), this congruence is forced to obey a diagonal condition, forcing it into an equation. The number of its solutions can be re-interpreted as another mean value which can be bounded in terms of \(J_{s,k}(X/p)\). Thus, one has bounded \(J_{s+k,k}(X)\) in terms of \(J_{s,k}(X/p)\), from which a bound for \(J_{s,k}(X)\) in terms of \(X\), \(s\) and \(k\) can be derived. The author refined this approach [\textit{T. Wooley}, Mathematika 39, 379--399 (1992; Zbl 0769.11055)], which led him to a big improvement over the classical bound for \(J_{s,k}(X)\). He assumes that \(\theta<1/k\) and re-interpretes the above-mentioned congruences as equations in a different way. The number of their solutions is expressed as a mean value of a new exponential sum. An application of the Cauchy-Schwarz inequality bounds this mean value in terms of \(J_{s,k}(X/p)\) and a new mean value that counts integral solutions of a system of Diophantine equations. This application of the Cauchy-Schwarz inequality introduces a differencing step. After re-interpreting the last-mentioned counting problem as a mean value, the whole procedure can be applied again, based on a new prime \(\tilde{p}\). Repeating this leads to an iteration which the author terms \textit{repeated efficient differencing process}. In the classical approach as well as the author's refined treatment described above, the underlying idea is to introduce a \textit{differencing} step to create congruences with \(2k\) variables that are forced to satisfy a diagonal condition, thus producing a new equation. The paper under review takes another viewpoint, putting emphasis on the creation of efficient \textit{congruence} conditions. The starting point is an initial condition modulo \(p\) amongst \(2s\) variables, from which a stronger condition modulo \(p^k\) is derived. The next step is to extract, in a similar fashion, a fresh condition modulo \(p^{k^2}\) from the aforementioned condition modulo \(p^k\). This process can be repeated, creating stronger and stronger congruence conditions. The author terms this iteration process \textit{efficient congruencing}. The details, however, involve many subtleties whose description goes beyond this review. This is a remarkable piece of work which has opened a new line of approaching mean values of exponential sums. It is astonishing that essentially best possible bounds have now been reached.