Counting rational points on biquadratic hypersurfaces (Q2417753)

In this article the authors verify Manin's conjecture for the number of rational points of bounded anticanonical height (over the field of rational numbers) for smooth biquadratic hypersurfaces in products of two projective spaces \(\mathbb{P}^{n-1}\times \mathbb{P}^{n-1}\) for \(n>35\). Let \(n\in \mathbb{N}\) and write \(\mathbf{x}=(x_1,\ldots, x_n)\) and \(\mathbf{y}=(y_1,\ldots, y_n)\). Consider a non-zero bihomogeneous polynomial \(F(\mathbf{x},\mathbf{y})\) of bidegree \((d_1,d_2)\) with integer coefficients. Then \(F(\mathbf{x},\mathbf{y})=0\) defines a hypersurface \(V\subset \mathbb{P}^{n-1}\times \mathbb{P}^{n-1}\). If \(V\) is in addition a smooth Fano variety of dimension at least \(3\), then \(\operatorname{Pic} V \cong \mathbb{Z}\times \mathbb{Z}\) and the Manin-Peyre conjectures predict that \[N_U(B):=\sharp\{ (\mathbf{x},\mathbf{y})\in U(\mathbb{Q}): H(\mathbf{x},\mathbf{y})\leq B\} \sim c B\log B,\] where \(U\) is a Zariski-open subset of \(V\) and \(H\) an anticanonical height function and where the leading constant \(c\) has received an interpretation by \textit{E. Peyre} [Duke Math. J. 79, No. 1, 101--218 (1995; Zbl 0901.14025)]. In particular it is conjectured that \(c>0\) if the set of rational points on \(V\) is non-empty. The authors of the article under review focus on the case of biquadratic hypersurfaces, i.e., the situation where \(d_1=d_2=2\). Let \(F_{\mathbf{x}}\) be the quadratic form that arises from the polynomial \(F(\mathbf{x},\mathbf{y})\) when we consider the coefficients \(\mathbf{x}\) as fixed and similarly for \(F_{\mathbf{y}}\). Then \[Z:=\{(\mathbf{x},\mathbf{y})\in V: \det (F_\mathbf{x})\det (F_{\mathbf{y}})=0\} \] is a Zariski-closed subvariety of \(V\). Let \(n>35\) and \(V\) be a smooth hypersurface. The main result of the article under review is an asymptotic formula \[N_U(B)\sim c B\log B,\] for \(U=V\setminus Z\) where \(c\) is the constant predicted by Peyre. There are general results for complete intersections in products of two projective spaces which establish the Manin conjecture if the dimensions of each of the projective spaces are sufficiently large, see [the reviewer, J. Reine Angew. Math. 714, 209--250 (2016; Zbl 1343.11061)], and more general work for hypersurfaces in some toric varieties by \textit{T. Mignot} [Acta Arith. 172, No. 1, 1--97 (2016; Zbl 1346.11030)] However for the case of smooth biquadratic hypersurfaces in products of two projective spaces they require \(n>193\). In the case of diagonal hypersurfaces in multi-projective space there are stronger results available by \textit{V. Blomer} and \textit{J. Brüdern} [J. Reine Angew. Math. 737, 255--300 (2018; Zbl 1408.11099)]. The proof proceeds via an application of the Hardy-Littlewood circle method and the delta method as developed in [\textit{W. Duke} et al., Invent. Math. 112, No. 1, 1--8 (1993; Zbl 0765.11038)] and [\textit{D. R. Heath-Brown}, J. Reine Angew. Math. 481, 149--206 (1996; Zbl 0857.11049)]. The basic strategy is to count integer solutions to the equation \(F(\mathbf{x},\mathbf{y})=0\) where the variables \(\mathbf{x}\) and \(\mathbf{y}\) lie in boxes of possibly very different side length \(X\), and respectively \(Y\). The counting result for the case that \(X\) and \(Y\) are of roughly the same size is importet from general work for hypersurfaces in products of projective spaces. The new ingredients are needed for situations where for example \(X\) is very small and \(Y\) relatively large. In this case the strategy is to fix a vector \(\mathbf{x}\) and then count the number of integral solutions of bounded size on the quadric given by \(F_{\mathbf{x}}(\mathbf{y})=0\). This is where the main technical part of the paper comes in -- the authors provide a version of the delta-method for a smooth quadric in at least 5 variables with an explicit dependence on the coefficients of the quadratic form. Once one has established the different counting results for integer points in boxes on the hypersurface \(F(\mathbf{x},\mathbf{y})=0\), one needs to use these results to obtain an asymptotic for the function \(N_U(B)\). The main step here (apart from Moebius inversions) is to handle the anticanonical height function. In most applications so far this was done using the machinery of the generalized hyperbola method as developed in work of \textit{V. Blomer} and \textit{J. Brüdern} [J. Reine Angew. Math. 737, 255--300 (2018; Zbl 1408.11099)]. In the article under review this argument is realized in a direct way with the goal of producing the main term (but with the same idea to cover the hyperbola shaped region with boxes). One extra technical difficulty that arises in the article under review is that the delta method produces an asymptotic formula for a weighted counting function. The authors follow an argument from [\textit{T. D. Browning} and \textit{D. R. Heath-Brown}, Discrete Anal. 2018, Paper No. 15, 29 p. (2018; Zbl 1444.11142)] to deduce asymptotics for a counting function where the smooth weights are replaced with indicator functions.