Improvements in Birch's theorem on forms in many variables (Q2408334)

Let \(F(x_1,\ldots, x_n)\in \mathbb{Z}[x_1,\ldots, x_n]\) be a homogeneous polynomial of degree \(d\) and consider the projective hypersurface \(X\) given by \[ F(x_1,\ldots, x_n)=0. \] Work of \textit{B. J. Birch} [Proc. R. Soc. Lond., Ser. A 265, 245--263 (1962; Zbl 0103.03102)] provides an asymptotic formula for the number of integral solutions of bounded height if \(n-\sigma>(d-1)2^d\), where \(\sigma\) is the dimension of the singular locus of \(F=0\) viewed as an affine variety. The leading constant is given as a product of local densities which leads to a local-global principle for rational points on \(X\), i.e., if \(X\) has non-singular local points in all completions of \(\mathbb{Q}\), then it has a \(\mathbb{Q}\)-rational point. The work under review improves on the condition on the number of variables \(n\) that is required for such a local-global principle. The main theorem states that, assuming \(d\geq 3\), if \[ n-\sigma \geq (d-\frac{1}{2}\sqrt{d})2^d, \] then the smooth Hasse principle holds for \(X\). Moreover there are sharper results for small degrees, namely the for \(3\leq d\leq 9\) the condition on \(n\) can be replaced by \[ n-\sigma > \frac{3}{4}d\, 2^d-2d. \] For general degree \(d\) and general forms \(F\) the paper under review gives for the first time improvements in \(n\) compared to work of Birch [loc. cit.]. For smooth degree \(3\) forms see work of \textit{D. R. Heath-Brown} [Proc. Lond. Math. Soc. (3) 47, 225--257 (1983; Zbl 0494.10012)], and generalizations to number fields by \textit{C. M. Skinner} [Duke Math. J. 75, No. 2, 409--466 (1994; Zbl 0848.14009)]. These results inspired both the work under review as well as earlier developments for degree \(4\) forms by \textit{T. D. Browning} and \textit{D. R. Heath-Brown} [J. Reine Angew. Math. 629, 37--88 (2009; Zbl 1169.11027)] and refinements by \textit{M. A. Hanselmann} [Rational points on quartic hypersurfaces. München: Univ. München, Fakultät für Mathematik, Informatik und Statistik (Diss.) (2012; Zbl 1296.14001)]. In fact, for \(d=4\) the work under review exactly recovers the earlier results by Browning and Heath-Brown. Let \(P\geq 1\) and \(\omega(\cdot)\) be a smooth compactly supported weight function on \(\mathbb{R}^n\). The proof proceeds via providing an asymptotic formula for the smoothened counting function \[ N_\omega(F;P)=\sum_{\overset{\mathbf{x}\in \mathbb{Z}^n}{F(\mathbf{x})=0}}\omega\left(\mathbf{x}/P\right), \] that is obtained using the Hardy-Littlewood circle method. Define the exponential sum \[ S(\alpha,P)=\sum_{\mathbf{x}\in \mathbb{Z}^n}\omega(\mathbf{x}/P)e(\alpha F(\mathbf{x})), \] such that \[ N_\omega(F;P)=\int_0^1 S(\alpha,P) d\alpha. \] There are a number of ways to estimate the exponential sum \(S(\alpha,P)\) on the minor arcs, i.e., on the set of \(\alpha\in [0,1]\) which are not close to rational points with small denominators. Birch applies a process of Weyl differencing. One step in this process would correspond to the inequality \[ |S(\alpha,P)|^2\leq \sum_{\mathbf{h}\in \mathbb{Z}^n}|\sum_{\mathbf{y}\in \mathbb{Z}^n}\omega((\mathbf{y}+\mathbf{h})/P)\omega(\mathbf{y}/P)e(\alpha (F(\mathbf{y}+\mathbf{h})-F(\mathbf{y})))| \] If the polynomial \(F\) is of degree \(d\), then the polynomial \(F(\mathbf{y}+\mathbf{h})-F(\mathbf{y})\) is of degree at most \(d-1\) in \(\mathbf{y}\). Applying this process \(d-1\) times hence leads to a family of linear exponential sums in \(n\) variables, which can be evaluated directly. In this setting each process of Weyl differencing roughly leads to one factor of \(2\) in the bound \(n>(d-1)2^d\). Another approach to bound the exponential sum \(S(\alpha,P)\) is via van der Corput differencing which starts with the identity \[ S(\alpha,P)= \frac{1}{H^n}\sum_{1\leq \mathbf{h}\leq H}\sum_{\mathbf{x}\in \mathbb{Z}^n}\omega(\mathbf{x}/P)e(\alpha F(\mathbf{x})), \] for a new parameter \(H\in \mathbb{N}\). For \(1\leq H\leq P\) an application of Cauchy's inequality then leads to the bound \[ |S(\alpha,P)|^2\ll \frac{P^n}{H^{2n}}\sum_{1\leq \mathbf{h},\mathbf{h}'\leq H}\sum_{\mathbf{x}\in \mathbb{Z}^n}\omega_{\mathbf{h},\mathbf{h}'}(\mathbf{x}/P)e(\alpha (F(\mathbf{x}+\mathbf{h})- F(\mathbf{x}+\mathbf{h'}))) \] and again after a variable substitution this can be interpreted as a family of sums of degree at most \(d-1\). On the other hand cubic exponential sums can be treated with a direct application of Poisson summation, as for example used in work of Browning and Heath-Brown [loc. cit.]. The paper under review combines all of the above techniques, to obtain two different estimates for \(S(\alpha,P)\). For some \(1\leq k\leq d\), the first uses \(d-k\) applications of van der Corput differencing together with Weyl differencing for the family of exponential sums with degree at most k polynomials. The second result combines \(d-3\) applications of van der Corput differencing with bounds for the resulting cubic exponential sums using Poisson summation.