On the density of integer points on generalised Markoff-Hurwitz and Dwork hypersurfaces (Q269892)

This paper is concerned with counting solutions of a Diophantine equation of ``nearly additive'' shape \[ f_1(x_1)+\ldots+f_n(x_n)=ax_1^{k_1}\ldots x_n^{k_n}, \] where the \(f_j\) are integer polynomials of degree at most \(d\), the coefficient \(a\) is a non-zero integer, and the exponents \(k_j\) are non-zero integers (which could be negative). This is what the paper calls the generalized Markoff-Hurwitz equation. The solutions to be counted are in non-zero integers \(x_j\) each of which lies in an interval \((u_j,u_j+h]\) of length \(h\). The results are of interest primarily when the parameters \(u_j\) are large. The first result involves a positive integer \(r\) and requires the \(k_j\) all to be odd and positive, and that \[ n\geq (d+1)(d+2)2^r\max(2r\,,\,3r-9/2)+3. \] There is then a constant \(C(r)\) such that the number of solutions \(x_1,\ldots,x_n\) is of order at most \(h^{n-4r/9}\), uniformly for \[ \max_{1\leq j\leq n}|u_j|\leq\exp(C(r)h^{4/9}). \] The implied order constant is allowed to depend on the polynomials \(f_j\) as well as \(n,a\) and the exponents \(k_j\). This result does not cover the case in which \(n\leq d^2\), for example, as for the so-called ``Dwork hypersurface'' \[ x_1^n+\ldots+x_n^n=ax_1\ldots x_n. \] A second result, handling this equation in particular, requires that the \(k_j\) are all odd and positive, that each \(f_j\) has degree at least \(r\), and that \[ n\geq 2r^3+2. \] There is then an absolute constant \(C\) such that the number of solutions is of order at most \(h^{n-r/3}\), uniformly for \[ \max_{1\leq j\leq n}|u_j|\leq\exp(Ch^{1/3}). \] A third result concerns the situation in which \(f_j(x)=x^d\) and \(u_j=u\) for every \(j\), in which case the number of solutions is of order \(h^{d(d+1)/2+o(1)}\), uniformly for all \(u\). (This is only of interest when at least one of the \(k_j\) is negative, since one trivially gets a much sharper bound otherwise.) The basic idea is to replace the equation by a congruence to a large modulus \(q\), and to express the number of solutions in terms of a product of one-dimensional mixed character sums \[ S(\alpha,\chi;q)=\sum_{u<x\leq x+h}e_q(\alpha f(x))\chi(x)^k. \] One can then estimate these sums non-trivially even when \(h\) is small compared to \(q\), by one of two methods. Either one may take \(q\) to be a product of distinct small primes and use the \(q\)-analogue of van der Corput's method, leading to the first result above; or one may take \(q\) to be a large power of a small prime and use Postnikov's method, (see [\textit{A. G. Postnikov}, J. Indian Math. Soc., New Ser. 20, 217--226 (1956; Zbl 0072.27304)] for example) leading to the second result. It seems to the reviewer that the sums \(S(\alpha,\chi;q)\) only depend on \(u\) modulo \(q\), so that it might be possible to assume for the proof that \(0<u\leq q\). In this case one could hope to remove the size constraint on \(u\) entirely.