Counting in hyperbolic spikes: The Diophantine analysis of multihomogeneous diagonal equations (Q1707513)

Let \(d,k,n\in \mathbb{N}\) with \(n\geq 2\) and let \(a_j\in \mathbb{Z}\setminus\{0\}\) for \(0\leq j\leq n\). Consider the hypersurface \(X\subset \mathbb{P}^n \times \ldots \times \mathbb{P}^n \) (the product of \(k\) copies of \(\mathbb{P}^n\)) which is defined by \[ \sum_{j=0}^n a_j (x_{1,j}x_{2,j}\ldots x_{k,j})^d=0. \] Every rational point \(x\in X(\mathbb{Q})\) has a representation by a tuple of vectors \(\mathbf{x}_i=(x_{i,0},\ldots, x_{i,n})\in \mathbb{Z}^{n+1}\) with \(\gcd(x_{i,0},\ldots, x_{i,n})=1\). For such a representation one defines the height function \[ H(x):= (|\mathbf{x}_1| |\mathbf{x}_2| \ldots |\mathbf{x}_k|)^{n+1-d}, \] where \(|\mathbf{x}_i|\) is the maximums norm. The choice of the height function \(H(x)\) is motivated by Manin's conjecture on the number of rational points of bounded anticanonical height on Fano varieties. Note that in general the variety \(X\) is not smooth. One goal of this article is to find an asymptotic formula for the counting function \[ N(B):=\sharp\{x\in X(\mathbb{Q}): H(x)\leq B,\, x_{i,j}\neq 0 \,\,\forall 1\leq i\leq k, 0\leq j\leq n\}. \] Here it is necessary to remove some Zariski-closed subset as points where some coordinates are equal to zero would contribute more than the expected main term in Manin's prediction. One of the key techniques in the proof is the Hardy-Littlewood circle method. For this one needs good upper bounds on mean values of the exponential sum \[ S(\alpha)= \sum_{1\leq x\leq P} e^{2 \pi i \alpha x^d}. \] Let \(n_0(d)\) be the smallest even natural number such that for every \(\epsilon >0\) one has \[ \int_0^1 |S(\alpha)|^{n_0(d)} d\alpha \ll P^{n_0(d)-d+\epsilon}. \] For \(n\geq n_0(d)\) the authors prove the following theorem. There are real numbers \(\delta >0\), \(C\geq 0\) and a monic polynomial \(Q\in \mathbb{R}[X]\) of degree \(k-1\) such that one has \[ N(B)=CB Q (\log B) +O(B^{1-\delta}). \] The constant \(C\) is positive if and only if the diagonal hypersurface given by \(a_0y_0^d+\ldots + a_ny_n^d=0\) has non-trivial local solutions at all places. A similar statement and asymptotic is obtained for the number of integer solutions with no zero coordinates to the system of equations \[ \sum_{j=0}^n a_{l,j}(x_{1,j}x_{2,j}\ldots x_{k,j})^l=0,\quad 1\leq l\leq d, \] where \(a_{l,j}\), \(1\leq l\leq d\), \(0\leq j\leq n\) are non-zero integers. In this case the authors establish the expected asymptotic for the counting function as soon as \(n\geq n_1(d)\), where \(n_1(d)\) is the smallest even natural number such that \[ \int_{[0,1]^d} \left|\sum_{1\leq x\leq P}e^{2\pi i (\alpha_1 x +\ldots +\alpha_d x^d)}\right|^{n_1(d)} d\alpha \ll P^{n_1(d)-\frac{1}{2}d(d+1)+\epsilon}, \] for all \(\epsilon >0\). By the proof of Vinogradov's mean value conjecture though work of \textit{T. D. Wooley} [Adv. Math. 294, 532--561 (2016; Zbl 1365.11097); Ann. Math. (2) 175, No. 3, 1575--1627 (2012; Zbl 1267.11105)] and \textit{J. Bourgain} et al. [Ann. Math. (2) 184, No. 2, 633--682 (2016; Zbl 1408.11083)] we know that \(n_1(d)=d(d+1)\). A main new innovation in the article under review is a vast generalisation of Dirichlet's hyperbola method. With the circle method one can count integral solutions to the systems of Diophantine equations above with the restriction that all the variables lie in given intervals. However, the height function \(H(x)\) cuts out hyperbola-shaped regions and it is not clear how to combine these with the circle method in a direct way (there is some approach to this problem in a special case by \textit{C. V. Spencer} [J. Number Theory 129, No. 6, 1505--1521 (2009; Zbl 1171.11054)]) The idea of the authors is that once one understands the number of solutions in boxes one is able to deduce results for the region cut out by the height function \(H\) by covering the area carefully with boxes. More generally, let \(h: \mathbb{N}^k\rightarrow [0,\infty)\) be some arithmetic function. Assume that there are constants \(\alpha,c,\delta >0\) with \(\delta <\min\{1,\alpha\}\), \(0<\nu\leq 1\) and \(D\geq 0\), such that the following asymptotics hold. (I) \[ \sum_{x_i\leq X_i}h(\mathbf{x})= c(X_1\ldots X_k)^\alpha +O((X_1\ldots X_k)^{\alpha} (\min_{1\leq j\leq k} X_j)^{-\delta}) \] for all \(X_j\geq 1\), \(1\leq j\leq k\). (II) For every non-empty subset \(\mathcal{I}\subset \{1,\ldots, k\}\) with at most \(k-1\) elements there exists an arithmetical function \(c_{\mathcal{I}} : \mathbb{N}^{|\mathcal{I}|}\rightarrow [0,\infty)\) such that for any \((u_i)_{i\in \mathcal{I}}\in \mathbb{N}^{|\mathcal{I}|}\) one has \[ \sum_{u_i\leq V_i , i\notin\mathcal{I}} h(\mathbf{u)}= c_{\mathcal{I}}((u_i)_{i\in \mathcal{I}}) \prod_{j\notin \mathcal{I}}V_j^{\alpha} + O ( (\prod_{j\notin \mathcal{I}}V_j^{\alpha}) \max_{i\in \mathcal{I}}|u_i|^D (\min V_j )^{-\delta}), \] uniformly in \(V_j\geq 1\) and \(|u_i|\leq \prod_{j\notin \mathcal{I}}V_j^\nu\), \(i\in \mathcal{I}\). Under these assumptions there exist a positive number \(\eta\) and a polynomial \(P_h\in \mathbb{R}[x]\) of degree \(k-1\) such that \[ \sum_{u_1\ldots u_k\leq N} h(\mathbf{u})= N^{\alpha} P_h(\log N) +O(N^{\alpha -\eta}). \] The generalised hyperbola method that is developed in this article is a very flexible and useful tool. It has for example inspired work of \textit{D. Schindler} [J. Reine Angew. Math. 714, 209--250 (2016; Zbl 1343.11061)] and \textit{T. Mignot} [Acta Arith. 172, No. 1, 1--97 (2016; Zbl 1346.11030)] where modifications of this argument have been used.