Weyl roots and equivalences of integral quadratic forms (Q2154301)

The paper under review is written in memoriam to Professor Daniel Simson, and is a continuation of some of his works. An \textit{integral quadratic form} \(q:\mathbb Z^n\to \mathbb Z\) is a map defined by a homogeneous polynomial of the second degree with integer coefficients, that is, \(q(x)=\sum_{i=1}^nq_ix_i^2+\sum_{i<j}q_{ij}x_ix_j\) with \(q_{i},q_{ij}\in\mathbb Z\). For any \(u\in\mathbb Z^n\) with \(q(u)\ne0\), the \textit{reflection} \(\sigma_u^q:\mathbb R^n\to\mathbb R\) at \(u\) is given by \(\sigma_u^q(w)=w-\frac{2q(u,w)}{q(u)}u\), where \(q(-,-): \mathbb Z^n\times\mathbb Z^n\to\frac12\mathbb Z\) now denotes the related symmetric bilinear form. The \textit{Weyl group} \(\mathbb W_q\) (defined if \(q_i\ne0\) for any \(i\)) is the subgroup of the general linear group \(\mathrm{Gl}_n(\mathbb R)\) generated by the simple reflections \(\sigma_i:=\sigma_{e_i}^q\) related to the vectors in the canonical basis of \(\mathbb R^n\). This paper is mainly devoted to \textit{Roiter}'s integral quadratic forms (RIQF's), which are those satisfying \(q_i>0\) and the following divisibility condition: \(q_{ij}/q_i,q_{ij}/q_j\in\mathbb Z\), which guarantees that the Weyl group is a subgroup of \(\mathrm{Gl}_n(\mathbb Z)\). Equivalently, the \textit{Coxeter matrix} defined as \(\Phi_q:=-\check G_q^{-1}\check G_q^{\mathrm{tr}}\) for \(\check G_q\) the upper triangular Gram matrix, has integral coefficients. Part of the motivation to study RIQF's is just their close relationship to root systems. For instance, if \(q\) is a connected, irreducible and positive RIQF, then the set of \textit{Weyl roots} \(\mathcal R_q:=\{\sigma(e_k):\sigma\in \mathbb W_q,k\le n\}\subset \mathbb Z^n\) is a reduced root system in the Euclidean space \((\mathbb R^n, q(-,-))\). One of the Roiter's purposes in [\textit{A. V. Rojter}, Proc. Steklov Inst. Math. 148, 207--217 (1978; Zbl 0443.10020); translation from Tr. Mat. Inst. Steklova 148, 201--210 (1978)] was to characterize Weyl roots independently on the explicit action of \(\mathbb W_q\). This work tackles this issue, providing a concrete description of \(\mathcal R_q\) in terms of \textit{roots} of \(q\), that is, integer solutions of the Diophantine equation \(q(x)=d\) for some \(d\in\mathbb Z\). Namely, the only case in which the inclusion \(\mathcal R_q\subset\bigcup_{i\le n}q^{-1}(q_i)\) is a proper containment occurs when \(\ell(q):=\sum_i|q_i-1|\) is equal to 1, that is, all \(q_i=1\) except for one index \(s\) with \(q_s=2\), in which case \(\mathcal R_q =q^{-1}(1)\cup \{\Phi_q^k(e_s):k\in\mathbb Z\}\). Some possible applications of this characterization are suggested. One of the main tools has been to employ various different equivalence relations slightly stronger than \(\sim\): two integral quadratic forms \(q\), \(q'\) are said \textit{weakly equivalent} (\(q\sim q'\)) if there is \(T\in\mathrm{Gl}_n(\mathbb Z)\) such that \(q'=q\circ T\). A careful comparison of other equivalence relations, here denoted by \(\sim_{\ell}\), \(\sim_{p\ell}\), \(\sim_{G}\) and \(\approx\), is developed. During the study a gap in the literature is detected in order to give a correct definition of the Dynkin type of a positive RIQF as well as the Euclidean type of a principal RIQF. (Recall that a \textit{nonnegative} integral form \(q\), that is, \(q(v)\ge 0\) for all \(v\in \mathbb Z^n\), is called \textit{positive} if the kernel \(\{v\in \mathbb Z^n:q(v)=0\}\) is zero, and \textit{principal} if the rank of the kernel is 1.) Any integral form \(q\) has associated a bigraph \(\Delta_q\): an undirected multigraph with \(n\) vertices, where two vertices labelled \(i\) and \(j\) are joined by \(|q_{ij}|\) solid edges if \(q_{ij}<0\) and by \(|q_{ij}|\) dotted edges if \(q_{ij}>0\), and besides each vertex \(i\) has \(|q_{i}-1|\) solid loops if \(q_i\le0\) and dotted ones if \(q_i>0\). It turns out that Dynkin and Euclidean bigraphs are just the bigraphs related to irreducible, connected and classic RIQF's, positive and principal respectively. But it is not true that Euclidean bigraphs are pairwise not \(\sim_{\ell}\)-weak equivalent: the only exception are the bigraphs named \(\tilde{\mathcal F}_{41}\) and \(\tilde{\mathcal C}_{4}\). Also through the paper many known results are extended within the possibilities: to integral quadratic forms not necessarily in the sense of Roiter, not necessarily positive, or not necessarily unit. Anyway, non-negative Roiter's integral quadratic forms are still far from understood.