Scale-multiplicative semigroups and geometry: automorphism groups of trees (Q329579)

The authors' aim is to describe the geometrical structure of locally compact totally disconnected groups using intersections of semigroups for the dynamical structure. The intention is to create more (computable) invariants carrying as little information as possible, preferably only the containment relations between semigroups. The article has extensive results and follows the ideas of \textit{J. Tits} [in: Essays Topol. Relat. Top., Mém. dédiés à Georges de Rham, 188--211 (1970; Zbl 0214.51301)] on automorphisms acting on homogeneous trees. They rely on double-transitivity of the action on the boundaries of semihomgeneous trees. The authors investigate a totally disconnected locally compact group, denoted \(G\), that as a topological space has no non-trivial connected subsets. The set of open compact subgroups is denoted \(B(G)\); for the existence see [\textit{D. van Dantzig}, Compos. Math. 3, 408--426 (1936; Zbl 0015.10202)]. The elements of \(B(G)\) form a basis of neighbourhoods of the identity. The group of continuous automorphisms of \(G\) is denoted \(\mathrm{Aut}(G)\); the elements are denoted \(\alpha\). The metric in \(\mathrm{Aut}(G)\) is taken to be the displacement function, i.e., the minimum distance between \(U\) and \(\alpha(U)\) in \(B(G)\) and is defined in such a way that \(G\) acts as isometries. The \textit{displacement index} is the cardinality of \(\alpha(V)\cap V\). It is finite and independent of the choice of \(V\). The \textit{scale} \(s(\alpha)\) is the minimum value taken by the indices; it measures the `distortion' due to \(\alpha\). Since the indices are integers the minimum is attained at some \(V\in B(G)\). The scale can also be considered as a continuous function on \(G\) by using conjugation, i.e., by replacing \(V\) by \(xVx^{-1}\). Such a \(V\) is called \textit{minimising} for \(x\). A subgroup is called \textit{tidy} as opposed to \textit{strange} (see [\textit{K. H. Hofmann} and \textit{A. Mukherjea}, Math. Ann. 256, 535--548 (1981; Zbl 0471.60015)]), if and only if it is minimising. \textit{G. A. Willis} [Math. Ann. 300, No. 2, 341--363 (1994; Zbl 0811.22004)] constructed tidy groups \(U\) as intersections of \(\alpha^nV\alpha^{-n}\) and then unions, which have to be closed, of the intersections. Definitions using a similar construction were developed by \textit{G. A. Willis}, e.g., [J. Algebra 237, No. 1, 142--164 (2001; Zbl 0982.22001)]. Simple examples of tidy subgroups are the subgroups \(\mathbb{Z}^n_p\) of the group \(G\) underlying the \(n\)-dimensional \(p\)-adic field \(\mathbb{Q}_p^n\). For each dimension the exponents are all taken to be powers of the same prime \(p\). Here \(\mathbb{Z}\) is considered as a subring of \(\mathbb{Q}\), the ring of rational numbers. These subgroups are tidy for all \(x \in G\). A scale cannot generally be multiplicative, i.e, \(s(xy) = s(x)s(y)\); \(x,y \in G\), unless \(G\) is \textit{uniscalar}, i.e., \(s\) is identically \(1\), though this rarely occurs. Maximal multiplicative subroups do not always exist in \(B(G)\), e.g., the group \(\mathbb{Z}_p^n\) is not contained in a maximal subgroup of \(\mathbb{Q}_p^n\). As it has been proved using Zorn's lemma that all \(s\)-multiplicative semigroups are contained in a maximal one. All maximal \(s\)-multiplicative semigroups are closed but they are not necessarily compact. \textit{G. A. Willis} [Bull. Lond. Math. Soc. 33, No. 2, 168--174 (2001; Zbl 1020.22002)] proved that all non-unital \(G\) have a subgroup of the semi-direct product form \(H \rtimes_{\alpha}\mathbb Z\); here \(H\subset G\), and \(\alpha\in\mathrm{Aut}(H)\). Suppose there is a \(V \subset \alpha(V)\subset H\) and \(H\) is the union, for \(n \in \mathbb{Z}\), of the \(\alpha^n(V)\). This \(V\) will be minimising for both \(\alpha\) and \(\alpha^{ -1}\). When \(V\) is tidy for \(x\), the authors construct a \(V_{++}\in B(G)\) such that the maximal \(s\)-multiplicative semigroups of \(V_{++} \rtimes_{\alpha_{x}}\langle x\rangle\) extend to \(G\). Here \(\alpha_{x}\) denotes the inner automorphism \(y\mapsto xyx^{-1}\) of \(G\) and \(s_{V}(x) = [xVx^{-1}: xVx^{-1} \cap V]\). Multiplicity over \(V\) means that one uses an \(s_V\) instead of \(s\). Semigroups are \(s_V\)-multiplicative over \(V\) if and only if \(V\) is a normal subroup of \(H\). They also produce semigroups that are multiplicative over \(V\) but not maximal. Denote the elements of \(H \rtimes_{\alpha} \mathbb{Z}\) as \((h,n),\;h \in H,\;n \in \mathbb{Z}\). There are only two maximal \(s\)-multiplicative semigroups. These correspond to either \(n\geq 0\) or \(n \leq 0\). Proposition 3.2 describes the structure of a \(G\) which has only two maximal \(s\)-multiplicative subgroups; they are inverses and their union is \(G\). Their intersections are \(\{x\in G: s(x) = 1 = s(x^{-1})\}\). If the number of maximal \(s\)-multiplicative semigroups is odd, then at least one of them must be equal to its own inverse and so uniscalar. If \(G\) is not uniscalar there are at least two different \(s\)-multiplicative semigroups for each \(G\), one which would be \(G\) itself. Example 3.3 (5) illustrates cases where the number of maximal \(s\)-multiplicative semigroups can be greater than two. The authors take \(H\) to be the group of diagonal matrices in \(\mathbb{Q}_p^n\) whose exponents are of the same prime \(p\). The group acts by matrix multiplication on \(H \rtimes_{\alpha} \mathbb{Z}\). This leads to a representation \(\rho:H\rtimes_{\alpha} \mathbb{Z}\to T_{q+1}\) where \(T_{q+1}\) is homogeneous of valency two. Then \(\rho(H)\) is in the set of elliptic automorphisms, and there is a hyperbolic element in the semidirect product such that it fixes an end \(\tau\); see [\textit{U. Baumgartner} and \textit{G. A. Willis}, Isr. J. Math. 142, 221--248 (2004; Zbl 1056.22001)]. The authors characterise the maximal \(s\)-multiplicative semigroups into two classes depending on whether \(\tau\) is attracting or repelling. The authors decide that two maximal \(s\)-multiplicative semigroups are adjacent if and only if the intersection of two semigroups is maximal in both semigroups. They can thus construct graphs for each \(H\) of Example 3.3. Subgroups of \(G\) are obtained by imposing restrictions on the vectors in \(H\). When there are no restrictions, there are \(2^n\) maximal \(s\)-multiplicative semigroups in \(G\) and the graph will be an \(n\)-dimensional hypercube. When there is a linear dependence of the vectors, with integral parameters, there will be \(2n\) intersections of maximal \(s\)-multiplicative semigroups and the graph will be a \(2n\)-gon. A tree is called \textit{homogeneous} if the valency is the same for all vertices; this holds also for all vertices in the same orbit. In the semihomogeneous case there are two valencies; vertices at even distances having the same valency. Usually the authors use pairs of valencies \((q-1,q+1)\). There is no confusion with the definitions when dealing with double transitivity as then at all vertices the edges can be subdivided into more edges. Although trees without leaves are semihomogeneous the homogeneity is degenerate. The authors now assume that each vertex has valency at least three. Bruhat-Tits trees (for semisimple algebraic groups over local fields) [\textit{F. Bruhat} and \textit{J. Tits}, Publ. Math., Inst. Hautes Étud. Sci. 41, 5--251 (1972; Zbl 0254.14017)] are semihomogeneous with pairs \((q,q^3)\); here \(q\) denotes the order of the residue field. Tits [loc. cit.] introduced combinatorial geometries and group actions for homogeneous locally finite trees \(T\). Such a tree may be recovered from its automorphism group by identifying maximal subgroups in \(B(G)\) with the vertices and edges of the tree and defining the incidence relation in terms of how they intersect. Tits showed that every automorphism of a tree either fixes a vertex, inverts an edge or induces a nontrivial translation along an infinite geodesic. The first two are called \textit{elliptic} and there is at least one fixed point in the tree. The latter type is called \textit{hyperbolic}, in which case the geodesic of translation is unique and called the \textit{axis}. Define the \textit{length} of an automorphism to be the number of fixed points or the minimum graph-distance by which it moves points along the hyperbolic axis. The authors adopt the graph convention that the edge between vertices \(v\) and \(w\) comprises the ordered pairs of oriented (geometrical) edges \((v, w)\) and \((w, v)\). The set of ends of an infinite tree \(T\) make up its \textit{boundary} \(\partial T\). Semigroups are found which identify with ends of the tree and with open sets of ends. The authors pay considerable attention to double transitivity of automorphisms acting on the boundary. An automorphism group \(G\) is called \textit{doubly-transitive} or 2-transitive if for pairs of elements \((x,y), (x' ,y')\) in \(\partial T \times \partial T\), \(x \neq x' , y \neq y'\) there exists a \(g \in G\) such that \(g.x = x', g.y = y'\). Proposition 4.4 concerns a tree that has no leaves and for which there is an edge \(\{e,\overline{e}\}\) such that each other component of the subtree \(T \setminus\{e,\overline{e}\}\) has at least two ends; \(G\) is a group of automorphisms of \(T\) acting doubly-transitively on the boundary \(\partial T\). The hypotheses of Proposition 4.4 usually apply to the remainder of the article. The authors demonstrate various properties. For instance they show in 4.4(2) that a doubly-transitive action is also \textit{spherically transitive}, i.e., assuming at all vertices of valency greater than 2, the action is transitive on the sphere with centre \(\omega\) and radius \(r \geq 1\); by Proposition 4.4.1 of \textit{F. M. Choucroun} [Mém. Soc. Math. Fr., Nouv. Sér. 58, 1--166 (1994; Zbl 0840.43019)], the converse holds if \(G\) is closed in \(Aut(T)\). Hyperbolic elements satisfy multiplicative scale if and only if the lengths are additive. They show also that the group \(G\) has at most two orbits, say \(O\) and \(E\). Then for every vertex in \(T\) the distance from \(E \cup O\) is bounded. The scale function of an element \(g\) is then a power of (something like) the product of the valencies of the two orbits; the power is a fraction of the length of \(g\) and the above-mentioned bound; viz. their formula (12). The concern is the relationship between multiplicity of scale and additivity of the lengths of hyperbolic elements. The \textit{minimal set} of an automorphism \(g\), denoted \(\min(g)\), is defined to be the sets of fixed points if \(g\) is elliptic and the axis if \(g\) is hyperbolic. If two automorphisms are contained in a semigroup on which the translation length is an additive function, then the intersection of their minimal sets is non-trvial. Proposition 4.6 describes the circumstances in which translation lengths add. Theorem 4.11 is a major result in the article and relates the maximal \(s\)-multiplicative semigroups to the structure of the tree. The authors use the hypotheses of Proposition 4.4 to characterise all the maximal \(s\)-multiplicative semigroups for a subgroup of \(\mathrm{Aut}(T)\), \(T\) a semi-homogeneous tree, that acts 2-transitively on \(\partial T\). Let \(\mathcal{S}\) be an \(s\)-multiplicative semigroup in \(G\) and for a set \(\mathcal{F} \subset \mathcal{S}\) define \(\min(\mathcal{F})\) to be the intersection of the \(\min(g)\) for all \(g \in \mathcal{F}\). There can only be three types of semigroups: \(\min\mathcal{S}\) is empty, which is if and only if \(\mathcal{F}\) is infinite; or \(\mathcal{S}\) contains an edge inversion; or \(\min\mathcal{S}\) is not empty and, since \(\mathcal{S}\) does not invert any edge, contains a vertex. Maximality follows by showing that none of these are contained in others. Double-transitively on \(\partial T\) is required in the characterisation of the maximal \(s\)-multiplicative semigroups in \(G\) to derive the formula (12) for the scale and to ensure that distinct sets of edges determine distinct semigroups determined by a vertex and the set of its edges. The authors' combinatorial geometry is limited, for instance to cover groups \(G\) in Hecke pairs \((G,C)\), where \(C\) a subgroup of \(G\) which has a finite number of cosets in \(G\), cf. [\textit{U. Baumgartner} et al., Glasg. Math. J. 48, No. 2, 193--201 (2006; Zbl 1151.22300)]; some of these groups have unbounded representations. Another limitation is that there are not enough invariants for maximal \(s\)-multiplicative semigroups. The authors give as an example the inability to find invariants to deal with semigroups. Since \(\mathcal{S}\to \mathcal{S}\cap \mathrm{PGL}(2,\mathbb{Q}_p)\) is a bijection between the maximal \(s\)-multiplicative semigroups in \(\mathrm{Aut}(T_{q+1})\) and those of \(\mathrm{PGL}(2,\mathbb{Q}_p)\), what is needed is an invariant that distinguishes between these two groups. Nevertheless, these two groups are distinguished by a \textit{local} invariant, called the \textit{structure lattice}, and the authors suggest that this might help; cf., [\textit{P.-E. Caprace} et al., C. R., Math., Acad. Sci. Paris 351, No. 17--18, 657--661 (2013; Zbl 1279.22009)].