Lorentzian Coxeter systems and Boyd-Maxwell ball packings (Q2256250)

In this paper it is shown that the set of limit directions of roots of a Lorentzian Coxeter group is equal to the complement of the interiors of the balls defined by the space-like weights. To describe this result in more detail, let \((W,S)\) be a Coxeter system with \(S\) finite. Let \((m_{st})\) be the Coxeter matrix. The group \(W\) has a faithful reflection representation -- the so-called geometric representation -- on the real vector space \(V\) with basis \(\Delta := \{ \alpha_s \}_{s \in S}\). The \(\alpha_s\) are called the simple roots of \((W,S)\) and their image \(\Phi := W \Delta\) under the action of \(W\) on \(V\) is the set of roots. The geometric representation is in fact orthogonal with respect to any symmetric bilinear form \(B\) with Gram matrix \((B_{st})\) with respect to the basis \(\Delta\) satisfying \(B_{st} = -\cos(\pi/m_{st})\) if \(m_{st}<\infty\) and \(B_{st} = -c_{st}\) if \(m_{st} = \infty\) for arbitrary \(c_{st} \in \mathbb{R}\) with \(c_{st} = c_{ts} \geq 1\). We fix such a bilinear form from now on and assume that \(B\) has signature \((n-1,1)\) so that \((V,B)\) is a Lorentzian space. All of the following notions depend on the choice of \(B\). Since \(B\) is non-degenerate we can find a basis \(\Delta^* := \{ \omega_s \}_{s \in S}\) of \(V\) with \(B(\alpha_s,\omega_t) = \delta_{st}\). The \(\omega_s\) are called the fundamental weights and their image \(\Omega := W \Delta^*\) under the action of \(W\) on \(V\) form the set of weights. For a non-zero vector \(\mathbf{x} \in V\) let \(\widehat{\mathbf{x}}\) be the line in \(V\) passing through \(x\) and the origin. This defines an element of the projective space \(\mathbb{P}V\), called the direction of \(\mathbf{x}\). For a set \(X\) of non-zero vectors of \(V\) let \(\widehat{X} := \{ \widehat{\mathbf{x}} \mid \mathbf{x} \in X \}\). Let \(E(\Phi)\) be the set of accumulation points of \(\widehat{\Phi}\) in \(\mathbb{P}V\). This is the set of limit directions of roots. Similarly, we define the set \(E(\Omega)\) of limit directions of weights. In Theorem 3.3 of the paper it is shown that \(E(\Phi) = E(\Omega)\), and this equality is the basis for the main result. Ball packings now come into play as follows. For any space-like vector \(\mathbf{x} \in V\), i.e., \(B(\mathbf{x},\mathbf{x})>0\), we define \(\mathsf{Ball}(\mathbf{x}) := \widehat{H_\mathbf{x}^-} \cap \widehat{Q} \subseteq \mathbb{P}V\), where \(H_\mathbf{x}^- := \{ \mathbf{y} \in V \mid B(\mathbf{x}, \mathbf{y}) \leq 0 \}\) and \(Q := \{ \mathbf{y} \in V \mid B(\mathbf{y},\mathbf{y}) = 0 \}\). After a stereographic projection \(\mathsf{Ball}(\mathbf{x})\) becomes a ``ball'' in the Euclidean space \(\mathbb{R}^{n-2}\), where \(n\) is the dimension of \(V\). Here, ``ball'' is understood in a wider sense to be a closed ball in \(\mathbb{R}^{n-2}\) in the actual sense, a complement of an open ball in the actual sense, or a closed half-space (this amounts to ``balls'' with positive, negative, or zero curvature). Let \(\Omega_r \subseteq \Omega\) be the set of space-like weights. The Boyd--Maxwell ball cluster is the collection of the balls \(\mathsf{Ball}(\omega)\) for \(\omega \in \Omega_r\). The main result (Theorem 1.1) of the paper is now that the so-called residual set of the Boyd-Maxwell ball cluster, which is the complement of the interiors of the balls in the cluster, is equal to \(E(\Phi)\). This gives an explanation for patterns observed in \(E(\Phi)\). The Boyd-Maxwell ball cluster is not necessarily a ball packing, meaning that all balls are closed and have disjoint interiors. In fact, a result by \textit{G. Maxwell} [J. Algebra 79, 78--97 (1982; Zbl 0556.51003)] shows that this is the case if and only if \((W,S)\) is of level 2, meaning that any subgraph on \(n-2\) vertices of the Coxeter graph is a Coxeter graph of finite or affine type while this is not the case for subgraphs on \(n-1\) vertices. The connected Coxeter graphs of level 2 have been determined by Maxwell [loc. cit.], and the paper finishes with a discussion of an algorithmic verification of his result.