Common hyperbolic bases for chains of alternating or quadratic lattices (Q2423415)

Let \(R\) be a complete discrete valuation ring with uniformizer \(p\) and field of fractions \(F\). Let \(b\) be a nondegenerate alternating or symmetric bilinear form on a finite-dimensional \(F\)-vector space \(V\) such that, in the latter case, \(b\) is associated to a quadratic form \(Q\) on \(V\) via \(b(x,y)=Q(x+y)-Q(x)-Q(y)\). In the alternating case one puts \(Q\equiv 0\). An \(R\)-lattice \(\Lambda\) of maximal rank on \(V\) is called \(p^r\)-maximal if it is maximal among all lattices satisfying \(Q(\Lambda), b(\Lambda,\Lambda)\subseteq p^rR\), where, in the case \(r=0\), one calls such a lattice maximal for short. If \((V,Q)\) is anisotropic, there is a unique such \(p^r\)-maximal lattice given by \(\{ x\in V\,|\,Q(x)\in p^rR\}\). Consider the orthogonal decomposition of \(V\) into its anisotropic and hyperbolic part, \(V=V_{an}\perp V_h\). A lattice \(\Lambda\) is called almost \(p\)-elementary totally even if \(\Lambda = \Lambda_{an}\perp\Lambda_h\) with \(\Lambda_{*}=V_{*}\cap \Lambda\), \(\Lambda_{an}\) is the unique \(R\)-maximal lattice on \(V_{an}\), and \(\Lambda_h\) is totally even, i.e., \(Q(x)\subseteq b(x,\Lambda_h)\subseteq R\) for all \(x\in\Lambda_h\), and in addition \(p\)-elementary, i.e., \(pR\subseteq b(x,\Lambda_h)\subseteq R\) for all primitive \(x\in\Lambda_h\). The main theorem of the present paper states the following. Let \(n\) be the Witt index of \((V,Q)\) and consider two maximal admissible lattice chains \[\Lambda_\ell^{\max}=\Lambda_\ell^{(0)}\supsetneq \Lambda_\ell^{(1)}\ldots \supsetneq \Lambda_\ell^{(n)}=\Lambda_\ell^{\min}, \ell=1,2,\] where \(\Lambda_\ell^{(0)}\) is maximal and each \(\Lambda_\ell^{(j)}\) is almost \(p\)-elementary totally even. Then there exist isotropic vectors \(e_1,\ldots,e_n,f_1,\ldots,f_n\in \Lambda_1^{(0)}\) spanning a hyperbolic sublattice, i.e., \(b(e_i,e_j)=b(f_i,f_j)=0\) and \(b(e_i,f_j)=\delta_{ij}\) for all \(1\leq i,j\leq n\), such that \[\Lambda_1^{(0)}=\bigoplus_{i=1}^n(Re_i\oplus Rf_i)\perp K\quad\hbox{and}\quad \quad\Lambda_\ell^{(j)}=\bigoplus_{i=1}^n(Rp^{r_i^{(\ell j)}}e_i \oplus Rp^{s_i^{(\ell j)}}f_i)\perp K,\] where \(K\) is the unique maximal lattice on the anisotropic orthogonal complement of the space generated by the \(e_i,f_j\), and where \(r_i^{(\ell j)},s_i^{(\ell j)}\) are certain integers (\(1\leq i\leq n\), \(0\leq j\leq n\), \(\ell=1,2\)). As the author points out, this result on simultaneous `hyperbolic bases' is useful in the study of Bruhat-Tits buildings and proofs have been given before, e.g., by \textit{P. Abramenko} and \textit{G. Nebe} [Math. Ann. 322, No. 3, 537--562 (2002; Zbl 1113.20028)] and \textit{W. Frisch} [The cohomology of \(S\)-arithmetic spin groups and related Bruhat-Tits buildings. Göttingen (PhD Thesis) (2002)]. The proof given in the present paper has the advantage of being rather elementary using only the theory of quadratic and bilinear forms. The author also mentions that his proof may readily be extended to the case of hermitian forms.