The distribution of spacings between quadratic residues. II (Q5932016)

In the first paper of this series [Duke Math. J. 100, 211-242 (1999; Zbl 0985.11038)] the author and \textit{Z. Rudnick} considered the spacing between the quadratic residues mod~\(q\) for squarefree~\(q\) in the limiting case as the number of prime factors of \(q\) tends to \(\infty\). The conclusion was to the general effect that the limiting distribution is Poissonian. In the paper under review this result is extended to the case of arbitrary \(q\). The author observes that if the exponents of the primes dividing \(q\) were bounded then the methods of the earlier paper would suffice. The fact that they are not introduces extra difficulties, in that the effect of truncating \(q\) has to be taken into account by a process involving the estimation of the divisor sums that arise. As a result the quantitative aspect of the result suffers somewhat. The result depends on an estimate for an autocorrelation function \[ R_r({\mathcal C},q) = N_q^{-1}\smash{\sum_{{\mathbf h} \in s{\mathcal C}}}N({\mathbf h},q), \] where \({\mathbf h}\in {\mathbb Z}^{r-1}\), \(s\)~is the mean spacing \(\smash{2^{\omega(q)}}/\sigma_{-1}(q)\) between squares mod~\(q\), \(N_q\) is the number of these squares, \(N({\mathbf h},q)\) is the number of solutions in squares~\(s_j\) of the equations \(s_{i+1}-s_i \equiv h_i, \bmod q\) when \(1 \leq i \leq r-1\), and \(\mathcal C\) is a convex set in \({\mathbb R}^{k-1}\) with the property that \(x_i \neq x_j\) whenever \(i \neq j\) and \((x_1-x_2,\ldots ,x_{k-1}-x_k)\) is in~\(\mathcal C\) (this condition excluding considerations of the correlation of a point with itself). The principal result of the paper is then a formula \(R_r({\mathcal C},q) = \text{ vol}({\mathcal C}) +O\bigl( \text{ exp}(-K\sqrt{\omega(q)}) \bigr)\). The error term is substantially weaker than that previously obtained in the case where \(q\)~is squarefree.