On the representation of quadratic forms by quadratic forms (Q2442200)

Let \(A\) and \(B\) be symmetric positive definite integer matrices of dimension \(n\) and \(m\), respectively. The interest of authors in this paper is to find \(n\times m\) integer matrices \(X\) such that \[ X^{T}AX=B \] thereby generalizing the classical problem of representing a positive integer as a sum of squares. The usual approach is to fix \(m,n\) and \(A\) and then try to represent `large enough' \(B\) for dimension \(m\) as large as possible in terms of \(n\). Hsia, Kitaoka and Kneser [\textit{J. S. Hsia} et al., J. Reine Angew. Math. 301, 132--141 (1978; Zbl 0374.10013)] have shown the local-global principle to hold for \(n\geq 2m+3\) and \( \min_{x \in \mathbb Z^{m} - \{0\}} x^{T}Bx \geq c_{1}\) for some constant \(c_{1}\) depending only on \(A\) and \(n\). \textit{J. S. Ellenberg} and \textit{A. Venkatesh} [Invent. Math. 171, No. 2, 257--279 (2008; Zbl 1247.11048)] used ergodic theory to show that the condition on \(n\) can be greatly improved to \(n \geq m+5\) under the additional assumption that the discriminant of \(B\) is square-free. \textit{R. Schulze-Pillot} [Acta Arith. 138, No. 3, 289--299 (2009; Zbl 1261.11038)] had refined the latter condition. The methods just described do not yield any quantitative information about integer solutions to \(X^{T}AX=B\). Let \(N(A,B)\) denote the number of integer matrices \(X\) satisfying \(X^{T}AX=B\). \textit{C. L. Siegel} [Ann. Math. (2) 36, 527--606 (1935; Zbl 0012.19703)] gave an exact formula for a weighted version of \(N(A,B)\). The focus of this paper is on deriving not an exact but rather an asymptotic formula for \(N(A,B)\), yet one that is valid for all forms \(A\). The previous approaches to this problem have used modular forms, whereas the strategy in this paper is to consider equation \(X^{T}AX=B\) as a system of \(R=\frac{m(m+1)}{2}\) quadratic equations and then apply circle method.