On the local density formula and the Gross-Keating invariant with an appendix `The local density of a binary quadratic form' by T. Ikeda and H. Katsurada (Q2205602)

Let \(F\) be a finite extension of the \(p\)-adic field \(\mathbb{Q}_p\), let \(\mathfrak{o}\) be the ring of integers of \(F\), and let \(\mathfrak{p}\) be the maximal ideal of \(\mathfrak{o}\). Let \(L\) be a free \(\mathfrak{o}\)-module of finite rank and let \(Q\) be a nondegenerate quadratic form on \(L\). We say that \((L,Q)\) is a quadratic space over \(\mathfrak{o}\). Given two quadratic spaces \((L,Q)\), \((L',Q')\) over \(\mathfrak{o}\) Siegel defined a quantity \(\alpha(L,L')\) known as the local density of \(L\) in \(L'\). Local densities appear in the formulas of Siegel and Weil for representations of quadratic forms over \(\mathbb{Z}\). This paper considers the problem of finding invariants of the quadratic spaces \((L,Q)\), \((L',Q')\) which are sufficient to determine the local density \(\alpha(L,L')\). The focus here is on the cases where \((L,Q)=(L',Q')\). When \(p>2\) the answer is well known: \(\alpha(L,L)\) is determined by the \(F\)-valuations of the coefficients of a diagonalization of \(Q\). For the general situation, when it may not be possible to diagonalize \(Q\) over \(\mathfrak{o}\), \textit{T. Ikeda} and \textit{H. Katsurada} [Am. J. Math. 140, No. 6, 1521--1565 (2018; Zbl 1417.11026)] defined a sequence of integers which they call the Gross-Keating invariant of \((L,Q)\), and denote by GK\((L)\). If \(p>2\) then GK\((L)\) is given by the valuations of the coefficients of the diagonalization of \(Q\); the interesting cases occur only when \(p=2\). GK\((L)\) admits several refinements, including the ``extended Gross-Keating invariant'' and the ``truncated extended Gross-Keating invariant''. In this paper, it is shown that when \(F\) is a finite unramified extension of \(\mathbb{Q}_2\) then the local density \(\alpha(L,L)\) is determined by (extended, truncated) Gross-Keating invariants of certain quadratic spaces associated to \((L,Q)\). In an appendix Ikeda and Katsurada compute \(\alpha(L,L)\) in the case where \(F\) is an arbitrary finite extension of \(\mathbb{Q}_2\) and the \(\mathfrak{o}\)-module \(L\) has rank 2. This computation shows that the main theorem is false without the assumption that \(F/\mathbb{Q}_2\) is unramified.