The Hasse principle for systems of diagonal cubic forms (Q261431)

The paper under review is concerned with the local-global principle for systems of diagonal cubic equations of the form \[ X: \quad \sum_{j=1}^s c_{ij}x_j^3=0,\quad 1\leq i\leq r, \] where \((c_{ij})_{1\leq i\leq r,1\leq j\leq s}\) is an integral matrix of size \(r\times s\). The authors establish the smooth Hasse principle for this system of equations as soon as the number of variables \(s\) is larger than \(6r\) and the matrix \((c_{ij})\) contains no singular \(r\times r\) submatrix, i.e. as soon as the system defines a smooth complete intersection in projective \((s-1)\)-space. As a corollary of the main theorem, any such system, which has non-trivial \(p\)-adic solutions for \(p\leq 9^{r+1}\), has infinitely many primitive integral solutions due to a local consideration of \textit{O. D. Atkinson} et al. [Mathematika 39, No. 1, 1--9 (1992; Zbl 0774.11016)]. The proof of the main theorem builds on the Hardy-Littlewood circle method and produces at the same time a lower bound for \(N(P)\), the number of integer solutions on \(X\) with all coordinates bounded by \(P\), of size \(N(P)\gg P^{s-3r}\). The current piece of work is a sharpening of previous work of \textit{J. Brüdern} and \textit{R. J. Cook} [Acta Arith. 62, No. 2, 125--149 (1992; Zbl 0774.11015)] who establish the Hasse principle for \(r\) simultaneous diagonal cubic forms in general position in at least \(s>7r\) variables. The proof of the main theorem proceeds via a socalled complification process which consists of subsequent applications of Schwarz's inequality to mean values that appear in an application of the circle method and that lead to larger systems of diagonal cubic equations in every step. This strategy is inspired by work of \textit{W. T. Gowers} [Geom. Funct. Anal. 11, No. 3, 465--588 (2001); Erratum 11, No. 4, 869 (2001; Zbl 1028.11005)]. It is then combined with good upper bounds that are available for sixth moments of smooth cubic Weyl sums. As the main theorem only aims for lower bounds on \(N(P)\) there is no loss in assuming that some variables are smooth numbers. More precisely, let \[ \mathcal{A}(P,R)=\{n\in [-P,P]\cap \mathbb{Z}: p\text{ prime and } p\mid n \Rightarrow p\leq R\}, \] and define the exponential sum \[ g(\alpha)=\sum_{x\in \mathcal{A}(P,R)}e(\alpha x^3). \] Then it follows from \textit{T. D. Wooley} [Mathematika 47, No. 1--2, 53--61 (2000; Zbl 1026.11075)] that for \(\eta\) sufficiently small and \(2\leq R\leq P^\eta\) one has the bound \[ \int_0^1 \left| g(\alpha)\right|^6\, d\alpha \ll P^{3+\xi}, \] with some explicit \(\xi< 1/4\). On the side of the complification process, the authors introduce so-called congenial and bicongenial matrices which are close to diagonal block matrices. Studying mean values corresponding to these matrices in combination with the complification process they arrive at the following mean value estimate. Let \[ \gamma_j=\sum_{i=1}^r c_{ij}\alpha_i, \quad 1\leq j\leq s, \] and define for \(s\geq 3\) the mean value \[ K(P)=\int_{[0,1]^r}|g(\gamma_1)g(\gamma_2)g(\gamma_3)f(\gamma_4)\ldots f(\gamma_s)|^2 \,d\mathbf{\alpha}, \] with \(f(\gamma)=\sum_{|x|\leq P}e(\alpha x^3)\). Then, if \(r\geq 2\), \(s=3r\) and if the matrix \((c_{ij})\) contains no singular \(r\times r\) submatrix, one has the upper bound \[ K(P)\ll P^{3r+\xi+\varepsilon}, \] which is only by a factor of \(\xi+\varepsilon\) away from square root cancellation.