The asymptotic formula in Waring's problem: higher order expansions (Q1668307)

For a fixed integer \(k\geq 2\), let \(R_s(n)\) be the number of ways of writing \(n\) as the sum of \(s\) strictly positive \(k\)-th powers. The classical Hardy--Littlewood asymptotic formula states that \[ R_s(n)\sim\frac{\Gamma(1+1/k)^s}{\Gamma(s/k)}\mathfrak{S}_s(n)n^{s/k-1} \] as \(n\to\infty\), provided \(s\) is sufficiently large in terms of \(k\). Here \(\mathfrak{S}_s(n)\) is the singular series. The aim of this paper is to investigate lower order terms in this asymptotic. In general it is shown that \[ R_s(n)=n^{s/k-1}\{\mathfrak{C}_{0,s}+\mathfrak{C}_{1,s}n^{-1/k}+ \ldots+\mathfrak{C}_{J,s}n^{-J/k}\}+o(n^{(s-J)/k-1}) \] as \(n\to\infty\), provided \(s\) is sufficiently large in terms of \(k\) and \(J\). The paper makes this last condition explicit. The constants \(\mathfrak{C}_{j,s}\) take the form \[ \mathfrak{C}_{j,s}=\begin{pmatrix} s\\ j\end{pmatrix} \frac{\Gamma(1+1/k)^{s-j}}{\Gamma((s-j)/k)}\mathfrak{S}_{s,j}(n), \] where \(\mathfrak{S}_{s,j}(n)\) is a type of singular series. The analysis and results take two forms, depending on the parity of \(k\). When \(k\) is even one merely has \[ \mathfrak{S}_{s,j}(n)=(-\tfrac12)^j\mathfrak{S}_{s-j}(n). \] In this case it is clear that the lower order terms in the asymptotic expansion are merely an ``edge effect'' caused by the sharp cut-off involved in requiring that the \(k\)-th powers are strictly positive. Indeed the authors' result implies that if \(R^*_s(n)\) is the number of \(s\)-tuples \((m_1,\ldots,m_s)\) in \(\mathbb{Z}^s\) for which \(n=\sum m_i^k\), then \[ R^*_s(n)=\frac{\Gamma(1+1/k)^s}{\Gamma(s/k)}\mathfrak{S}_s(n)n^{s/k-1}+ o(n^{(s-J)/k-1}) \] if \(s\) is sufficiently large in terms of \(k\) and \(J\). Conversely, such an estimate implies the authors' more complicated asymptotic formula for \(R_s(n)\). Thus for even \(k\) there is nothing mysterious. The function \(R^*_s(n)\) is far more natural than \(R_s(n)\) for even \(k\). The case of odd \(k\) is more interesting, and here the expression for \(\mathfrak{S}_{s,j}(n)\) involves the sums \[ \sum_{r=1}^q\left(\tfrac12-\tfrac{r}{q}\right)e(ra^k/q), \] which have no nice multiplicative property. The proofs assume that one has a suitable minor arc estimate and analyse the major arcs carefully to extract the main terms described above.