Least zero of a cubic form (Q662402)

Let \(C(x_1,\dots,x_n)\in\mathbb{Z}[x_1,\dots,x_n]\) be a cubic form, and let \(\Lambda_n(C)\) be the minimum of \(\|\mathbf{x}\|_{\infty}\), taken over all non-zero integer solutions to \(C(\mathbf{x})=0\). It follows from work of the reviewer [Invent. Math. 170, No. 1, 199--230 (2007; Zbl 1135.11031)] that \(\Lambda_n(C)\) is finite as soon as \(n\geq 14\). The goal of the present paper is to bound \(\Lambda_n(C)\) in terms of \(\|C\|\), defined as the maximum modulus of the coefficients of \(C\). This problem was considered by \textit{P. D. T. A. Elliott} [Some problems in analytic number theory. Ph. D. thesis, Cambridge (1966)], who showed that \[ \Lambda_n(C)\ll_n \|C\|^{e(n)}\qquad (n\geq 17) \] with an explicit exponent \(e(n)\). For example one has \(e(17)=2500417\). (It seems to the reviewer that if one has an admissible exponent \(e(17)\), then one may take \(e(n)=e(17)\) for all \(n\geq 17\), simply by considering a non-trivial zero of \(C(x_1,\dots,x_{17},0,\dots,0)\). However Elliott's exponents \(e(n)\) increase with \(n\).) The present paper shows that \[ \Lambda_n(C)\ll \|C\|^{360000}\qquad (n\geq 17). \] Moreover one gets considerably better exponents when \(C\) has at most isolated ordinary singularities. For example, in this case one may take \(e(17)=1071\) and \(e(20)=261\). (With a little effort the above remark can be applied here too, so one would have \(e(n)=261\) for all nonsingular forms with \(n\geq 20\), for example.) The proof adapts the usual treatment of cubic forms by the circle method to produce results in which the dependence on \(\|C\|\) is made explicit. It turns out that it is the treatment of the major arcs and the singular series which are most important, and in this regard the paper provides a general result of interest, concerning the number \(N(p)\) of solutions of the congruence \(C(\mathbf{x})\equiv 0\pmod p\). Specifically, if \(C\) is non-degenerate and \(n\geq 10\), then it is shown that \[ N(p)\geq p^{n-1}+O_n(p^{n-2}), \] uniformly in \(C\). (The first author has subsequently shown that \(N(p)=p^{n-1}+O_n(p^{n-2})\), see [Can. Math. Bull. 56, No. 3, 500--502 (2013; Zbl 1368.11026)].) The Lang-Weil estimate would give a weaker error term \(O_n(p^{n-3/2})\).