Plane curves in boxes and equal sums of two powers (Q2571393)

From the text: Let \(F \in\mathbb Z[x_1, x_2, x_3]\) be an absolutely irreducible form of degree \(d\), producing a plane curve in \(\mathbb P^2\). The central aim of this paper is to analyze the density of rational points on such curves, which are contained in boxes with unequal sides. We shall see below how such considerations may be used to obtain new paucity results for equal sums of two powers. Suppose that \(\mathbf P = (P_1, P_2, P_3)\) for fixed real numbers \(1\leq P_1 \leq P_2 \leq P_3\), say. Then we define \[ N(F; P) = \#\{\mathbf x \in\mathbb Z^3 : F(\mathbf x) = 0, |x_i| \leq P_i, (1\leq i \leq 3), \mathbf x \,\text{primitive}\}, \] where \(\mathbf x = (x_1, x_2, x_3)\) is said to be primitive if \(\text{h.c.f.}(x_1, x_2, x_3) = 1\). Our starting point is the recent work of the second author [Ann. Math. (2) 155, No. 2, 553--598 (2002; Zbl 1039.11044)], who has shown that \(N(F; \mathbf P) = O(P^{2/d+\varepsilon})\) for any \(\varepsilon > 0\), whenever \(P_i = P\) for \(1\leq i\leq 3\). The implied constant in this bound depends at most upon the choice of \(\varepsilon\) and \(d\), a convention that we shall follow throughout this paper. This is essentially best possible for curves of genus zero, and it is natural to ask what can be said about the quantity \(N(F; P)\) when the \(P_i\) are of genuinely different sizes. With this in mind we define \[ T = \max\{P^{e_1}_1 P^{e_2}_2 Pe^{e_3}_3\}, \tag{1} \] where the maximum is taken over all integer triples \((e_1, e_2, e_3)\) for which the corresponding monomial \(x^{e_1}_1 x^{e_2}_2 x^{e_3}_3\) occurs in \(F(x)\) with non-zero coefficient. Then for any \(\varepsilon > 0\), the second author's principal result for curves [see Theorem 3, loc. cit.] states that \[ N(F; \mathbf P) \ll \left(\frac{P_1P_2P_3}{T^{1/d}}\right)^{1/d+\varepsilon}. \tag{2} \] This clearly reduces to the previous bound whenever \(P_i = P\) for \(1\leq i \leq 3\). Turning to the case of unequal \(P_i\), it is easy to construct examples for which (2) is not best possible. Our aim is to establish a sharper bound for the interim case in which \(P_1 \geq 1\) and the region \(|x_i| \leq P_i\) is sufficiently lopsided. Theorem 1. Suppose that \(T \geq (P_1P_2)^d\), and let \(\varepsilon > 0\) be given. Then we have \[ N(F; \mathbf P) \ll P^\varepsilon_3 \exp\left(\frac{\alpha\beta + (\alpha + \beta-\alpha\beta)(\tau-\alpha-\beta)}{d(\tau-\alpha)(\tau -\beta)}\log P_3\right). \] Here \(\alpha =\frac{\log P_1}{\log P_3}\), \(\beta=\frac{\log P_2}{\log P_3}\), \(\tau =\frac{\log T}{d \log P_3}\), for given \(1\leq P_1\leq P_2 \leq P_3\) and \(T\) as in (1). Theorem 1 is applied to the diagonal equation \[ w^k + x^k = y^k + z^k. \] For any \(X\geq 1\), we denote by \(N_k(X)\) the number of positive integer solutions in the region \(\max\{w, x, y, z\} \leq X\). There are \(2X^2 +O(X)\) trivial solutions in which \(y, z\) are a permutation of \(w, x\), and so we write \(N^{(0)}_k (X)\) for the number of non-trivial solutions. Bounds were given by Hooley and the second author. The aim of this paper is to improve upon these previous bounds whenever \(k = 5\) or \(6\). Theorem 2. For any \(k\geq 4\) and any \(\varepsilon > 0\), we have \[ N^{(0)}_k (X) \ll X^{3/2+1/(2k-2)+\varepsilon}. \]