Subconvexity for additive equations: pairs of undenary cubic forms (Q475828)

Let \(l,m,n\) be nonnegative integers with \(m\geq n\), and fix nonzero integers \(a_i,b_i,c_j,d_k\), where \(1\leq i\leq l, 1\leq j\leq m\) and \(1\leq k\leq n\). Let \(N(B)\) denote the number of integer solutions to the system \(a_1 x_1^3+\cdots+a_lx_l^3 +c_1 y_1^3+\cdots+c_m y_m^3=0, b_1 x_1^3+\cdots+b_lx_l^3 +d_1 z_1^3+\cdots+d_n z_n^3=0\), with \(x_i,y_j,z_k\in[-B,B]\). Define \(s=l+m+n\), the total number of variables, and \(q_0^{\ast}=\text{min}\{s-l,s-m,s-n\}\), a measure of the minimal number of variables across equations. The main result proved is Theorem 1.1: Let \(s\geq 1\) and \(q_0^{\ast}\geq 7\), and suppose the system above admits nonsingular \(p\)-adic solutions for every prime \(p\). Then, with the possible exception of the case \((l,m,n)=(5,5,2)\), one has \(N(B)\gg B^{s-6}\). The same conclusion holds for the exceptional case if Hooley's Riemann hypothesis (HRH, discussed in Section 6) is assumed. For \(s\geq 13\) the result follows from the authors' previous work, in particular, [Clay Math. Proc. 7, 57--76 (2007; Zbl 1230.11042), Theorem 2], so only \(s=11\) and \(s=12\) need to be considered. The method of proof is an unconventional application of the Hardy-Littlewood circle method. The convexity barrier imposed by this method is \(s\geq 12\) (i.e., twice the total degree of the system), so the cases where \(s=11\) are of special interest. For \(s\geq 12\) the asymptotic lower bound of Theorem 1.1 can be refined using the product of local densities, yielding Theorem 1.2: If \(s\geq 12\) and \(q_0^{\ast}\geq 8\), then \(N(B)\geq(\mathcal{C}+o(1))B^{s-6}\). (Here \(\mathcal{C} =v_{\infty}\prod_p v_p\), where \(v_{\infty}\) is the area of the manifold defined by the system in the box \([-1,1]^s\) and, for each prime \(p\), \(v_p=\lim_{h\rightarrow\infty}p^{h(2-s)}M(p^h)\), where \(M(q)\) is the number of solutions of the system with \(\mathbf{x}\in(\mathbb Z/q\mathbb Z)^s\).) To prove Theorem 1.1 it is sufficient to consider four possible shapes for \((l,m,n)\), called types A, B, C and D. The method developed for type A serves as a model for the other types, with appropriate adaptations, and for type D also assuming HRH. The authors describe the approach: ``Our strategy for obtaining a lower bound for \(N(B)\) involves constraining the first block of \(l\) variables. Restricting in such a manner that the corresponding diagonal forms \(a_1 x_1^3+\cdots+a_lx_l^3\) and \(b_1 x_1^3+\cdots+b_lx_l^3\) behave essentially as expected as far as multiplicity of representations is concerned, we obtain a modified counting function \(N_0(B)\) whose behaviour is mollified by this arithmetic smoothing. In Lemma 3.3 we show that the major arc contribution [\(N(B;\mathfrak{N})\)] within \(N_0(B)\) is close to the contribution within \(N(B)\). By means of a pruning operation discussed in Lemma 3.1, this contribution is seen via Lemma 2.1 to have order of growth \(B^{s-6}\). Meanwhile the minor arc contribution within \(N_0(B)\) exploits the available smoothing by means of Bessel's Inequality, and is described in mixed form in Lemma 3.4, and pure minor arc form in Lemma 3.5. In this way, we aim to show that \(N(B)\geq N_0(B)\geq N(B;\mathfrak{N})+o(B^{s-6})\gg B^{s-6}\), and thereby establish Theorem 1.1.'' Theorem 1.2 was already been proved by the authors for \(s\geq 14\) [Can. J. Math. 63, No. 1, 38--54 (2011; Zbl 1238.11046), Theorem 1.1]. For \(s=12,13\) there are three possible shapes for \((l,m,n)\), called types E, F and G, and the results for these types are proved by modifying the arguments used to prove Theorem 1.1. In conclusion the authors mention that their methods may be extended to other systems with a similar structure or, with additional complications, to systems of three or more equations.