Systems of cubic forms in many variables (Q2009901)

The famous theorem of \textit{B. J. Birch} [Proc. R. Soc. Lond., Ser. A 265, 245--263 (1962; Zbl 0103.03102)] gives an asymptotic formula for the number of common zeros of a system of forms \(f_1,\ldots,f_R\) of degree \(d\) in an expanding box, under the assumptions that the forms define a smooth complete intersection in projective space, and that the number of variables satisfies \(n\ge R+R(R+1)(d-1)2^{d-1}\). In the degree 3 case this bound is \(n\ge 9R+8R^2\), and the present paper makes remarkable progress, by showing that one can take \(n\ge 25R\). This linear growth rate improves the Birch result for \(R\ge 3\) (and recovers it for \(R=2\)). The asymptotic formula is of the usual Hardy-Littlewood type, and establishes the Hasse principle for the varieties involved. Indeed, although it is not mentioned in the paper, one gets weak approximation as well, as a by-product. The paper builds on ideas from author's work [Invent. Math. 213, No. 1, 205--235 (2018; Zbl 1442.11057)], which laid down a general framework to handle forms of degree \(d\), and showed that \(n\ge 9R\) is sufficient for systems of quadratics. The approach uses the circle method, and requires one to estimate the number, \(N(B)\) say, of solutions to certain auxiliary inequalities. In the case of cubic forms \(f_i\), let \(f\) be a real linear combination of the \(f_i\), normalized to have height 1, and let \(B_j(\mathbf{x},\mathbf{y})\) (for \(1\le j\le n\)) be the associated bilinear forms. Then \(N(B)\) is the number of pairs \(\mathbf{x},\mathbf{y}\in\mathbb{Z}^n\) for which each of \(||\mathbf{x}||_{\infty}\), \(||\mathbf{y}||_{\infty}\), and \(\max_j|B_j(\mathbf{x},\mathbf{y})|\) is at most \(B\). It is then shown, for suitable \(n\) and \(R\), that there is a constant \(\delta>0\) such that \[ N(B)\ll B^{2n-24R-\delta}, \] uniformly over all linear combinations. According to the earlier work, this is sufficient to derive the asymptotic formula. It should be observed that Birch's approach uses equalities \(B_j(\mathbf{x},\mathbf{y})=0\) for a rational linear combination \(f\), rather than inequalities for a real linear combination. Thus, in addition to the new approach given in the author's earlier paper, one needs a number of further ideas to handle these real inequalities.