Large gaps between sums of two squareful numbers (Q6569633)

A non-negative integer \(n\) is \textit{squareful} or \textit{powerful} if either \(n = 0\) or \(\alpha_i \ge 2\) for all \(i\) in its prime factorization \(n = p_1^{\alpha_1} \cdots p_s^{\alpha_s}\). In contrast, \(n\) is \textit{squarefree} if \(\alpha_i = 1\) for all \(i\) in its prime factorization. Let \(\mathcal{V}\) be the set of all numbers that are sums of two squareful numbers. Recently, \textit{V. Blomer} [J. Lond. Math. Soc., II. Ser. 71, No. 1, 69--84 (2005; Zbl 1166.11312)] and \textit{V. Blomer} and \textit{A. Granville} [Duke Math. J. 135, No. 2, 261--302 (2006; Zbl 1135.11020)] obtained good estimate on\N\[\NV(x) := | \mathcal{V} \cap [1, x]| = \frac{x (\log \log x)^{O(1)}}{(\log x)^{1 - 1 / \sqrt[3]{2}}}.\N\]\NIn the current beautiful paper under review, the authors study large gaps in \(\mathcal{V}\). Let \(M(x)\) be the length of the largest subinterval of \([1, x]\) without any elements of \(\mathcal{V}\). The authors prove the new result that, for \(x \ge 3\).\N\[\NM(x) \gg \frac{\log x}{(\log \log x)^2}.\N\]\NThe starting point is that every squareful number \(n\) can be written uniquely as \(n = x^2 a^3\) for some positive integer \(x\) and squarefree number \(a\). Hence, \(\mathcal{V}\) consists of numbers of the form \N\[\Nx^2 a^3 + y^2 b^3. \tag{1}\N\]\NThis prompts the authors to obtain an upper bound for quadratic congruences \(A x^2 + B y^2 \equiv u \pmod P\) for some squarefree number \(P\).\N\NLet \(M\) be a large positive integer and \(N = M^D\) where \(D\) is some large exponent. The idea is to construct a positive integer \(u \le \exp (C M (\log M)^2)\) for some constant \(C > 0\) such that the numbers \(u+1, u+2, \ldots, u+M\) do not lie in \(\mathcal{V}\). The authors accomplish this by means of probabilistic arguments to come up with many size \(K = \lfloor 22 \log M \rfloor\) disjoint subsets \(P_j\) of \(\mathcal{P} = \{ N \le p \le 2N: p \text{ is prime} \}\) such that each \(P_j\) intersects with \(R(c) = \{ c \le M^8 : (\frac{- c}{p}) = -1 \text{ for all } p \in \mathcal{P} \}\).\N\NBy Chinese remainder theorem, one can find \(u \le P =: \prod_{j, p \in P_j} p \le (2 M^D)^{MK}\) such that \(u + j \equiv 0 \pmod p\) for all \(j \le M\) and all \(p \in P_j\). Let \(U\) be the set of all numbers \(v \in [0, P^{10}]\) such that \(v \equiv u \pmod{P}\). The upshot is that, for most \(v \in U\), the numbers \(v + j\) cannot be of the form (1) for otherwise\N\[\Nv + j = x^2 a^3 + y^2 b^3 \Rightarrow - \frac{a^3}{b^3} \equiv \frac{y^2}{x^2} \pmod{p}\N\]\Nfor some \(p \in P_j\) which contradicts the requirement \((\frac{-c}{p}) = -1\) with \(c\) is the squarefree part of \(ab\). Here, there is a misprint in the paper with \(- \frac{a^3}{b^3} \equiv \frac{x^2}{y^2}\) instead of the correct expression \(- \frac{a^3}{b^3} \equiv \frac{y^2}{x^2}\) above.\N\NThere are technicalities such as situations with \(v + j\) divisible by some \(p^2\), or when the size of \(ab\) is in the intermediate range \(M^8 < ab \le M^{40}\) and in the large range \(ab > M^{40}\), or when \(ab\) is not coprime to \(P\). The authors use probabilistic arguments and clever reductions to deal with them.\N\NIt is worthwhile to note that one key element of the proof is the establishment of the following variant of Linnik's theorem on the least prime in arithmetic progression: There exists a real number \(A > 0\) such that for all \(x \ge q^A\) and all real primitive Dirichlet character \(\chi \pmod q\), the inequality\N\[\N\sum_{x < n \le 2 x} \Lambda(n) \chi(n) \le 0.02x\N\]\Nholds as \(x \rightarrow + \infty\) with \(\Lambda(n)\) being the von Mangoldt lambda function supported on prime powers. This is used to show that the size of \(R(c)\) is big and \(A\) enters into the definition of \(D\) above. Its proof hinges on the observation that the possible contribution from exceptional character is non-positive.