Estimates for complete multiple exponential sums (Q4486474)

Let \(f\) be a polynomial in \({\mathbb Z}[{\mathbf x}]\) in the \(n\) variables \({\mathbf x}=(x_1,\ldots,x_n)\) with integer coefficients and total degree \(d>1\). For a positive integer \(q\), define the complete exponential sum \[ S(f;q)=\sum_{{\mathbf x} \bmod q} e_q(f({\mathbf x})), \] where the sum is taken over a complete set of residues for \({\mathbf x}\) modulo \(q\) and \(e_q(t)=e^{2\pi i t/q}.\) As a consequence of his proof of the Weil conjectures, \textit{P. Deligne} [Publ. Math., Inst. Hautes Étud. Sci. 43, 273-307 (1994; Zbl 0287.14001)] proved that if \(p\) is prime, then \[ |S(f;p)|\leq (d-1)^n p^{n/2}, \] provided that the homogeneous part of \(f\) of highest degree is non-singular modulo \(p\). The emphasis here is on estimates when the modulus is composite. The results of \textit{J. H. Loxton} and \textit{R. C. Vaughan} [Can. Math. Bull. 28, 440-454 (1985; Zbl 0575.10033)] and \textit{W. K. A. Loh} [Can. Math. Bull. 41, 187-195 (1998; Zbl 0944.11025)] give very precise results when \(f\) is a polynomial of degree 1. In the multidimensional case, \textit{V. N. Chubarikov} [Mat. Zametki 20, 61-68 (1976; Zbl 0353.10028)] proved that if the content of \(f\) is prime to \(q\), then \[ |S(f;q)|\leq e^{7d'n}3^{n\nu(q)}\tau(q)^{n-1} q^{n-1/d'}, \] where \(d'\) is the maximum degree of \(f\) to any variable, \(\nu(q)\) is the number of distinct prime divisors of \(q\), and \(\tau(q)\) is the number of divisors of \(q\). (See also \textit{G. I. Arkhipov, A. A. Karatsuba} and \textit{V. N. Chubarikov} [Theory of Multiple Exponential Sums, Moscow (1987; Zbl 0638.10037)].) On the other hand, \textit{J. H. Loxton} and \textit{R. A. Smith} [J. Aust. Math. Soc., Ser. A 33, 125-134 (1982; Zbl 0493.10041)] obtained the much smaller bound \(|S(f,q)|\ll q^{n/2}\) when the projective variety defined by the equations grad \(f =0\) is non-singular. In this paper, the author attacks this problem by the use of the Newton polyhedron of a polynomial in several variables. By multiplicativity, it suffices to consider bounds for \(|S(f,p^\alpha)|\), where \(p\) is a prime. The results here imply estimates of the form \[ S(f;p^\alpha) \ll p^{n\alpha(1-1/(2e))+\text{dim}(\text{grad } f)\alpha/(2e)}, \] where \(e\) is the maximum order of a singularity of the variety \(\text{grad}(f)\), \(\alpha\) is sufficiently large, and the implied constant is independent of \(\alpha\). In the case where the variety \(\text{grad}(f)\) is non-singular, the results yield \[ S(f;p^\alpha) \ll p^{n([(\alpha+1)/2]+\delta)}, \] where \(\delta\) is the \(p\)-adic order of a certain discriminant and \(\alpha\) is sufficiently large. This is a more precise version of a result obtained by Loxton and Smith (loc. cit.). In the general case, the estimates here yield estimates that are of the same quality as the estimates of Arkhipov, Karatsuba, and Chubarikov (loc. cit.) when \(\alpha\) is large. The author also makes a detailed analysis of the case where the polynomials have degree 2, where there are classical evaluations. In a future paper, he will continue this analysis with explicit and precise estimates for polynomials of degree 3.