On a question of B. H. Neumann (Q4804047)

Let \(F_k\) be the free group of rank \(k\) freely generated by \(x_1,\dots,x_k\) and write \(d(G)\) for the minimal number of generators of the finitely generated group \(G\). It is said that a normal subgroup \(N\) of \(F_k\) is a \(G\)-defining subgroup of \(F_k\) if \(F_k/N\cong G\). Denote the set of all \(G\)-defining subgroups of \(F_k\) by \(\Sigma(G,k)\). Let \(k\geq d(G)\) (note that \(\Sigma(G,k)\) is not empty if and only if \(k\geq d(G)\)), then there is a natural action of \(\Aut(F_k)\) on \(\Sigma(G,k)\) defined by \(N\longmapsto N\sigma\) where \(N\in\Sigma(G,k)\) and \(\sigma\in\Aut(F_k)\). The orbits of this action are called \(T_k\)-systems of \(G\). \textit{B. H. Neumann} and \textit{H. Neumann} [Math. Nachr. 4, 106-125 (1951; Zbl 0042.02102)] introduced the notion of generating \(G\)-vectors which enabled them to define an equivalent action of \(\Aut(F_k)\) that is more manageable. A generating \(k\)-vector of a group \(G\) is an ordered \(k\)-tuple \({\mathbf v}=(g_1,\dots,g_k)\) with \(g_1,g_2,\dots,g_k\in G\) such that \(\langle g_1,g_2,\dots,g_k\rangle=G\). The set of all generating \(k\)-vectors of \(G\) is denoted by \({\mathcal N}_k(G)\). Consider an arbitrary \(\sigma\in\Aut(F_k)\). We have \(x_i\sigma^{-1}=w_i(x_1,x_2,\dots,x_k)\) where \(w_i\) are uniquely determined words in \(x_1,x_2\dots,x_k\) for each \(i\in\{1,\dots,k\}\). For each \({\mathbf v}=(g_1,g_2,\dots,g_k)\in{\mathcal N}_k(G)\), it is defined \[ {\mathbf v}\sigma=(w_1(g_1,g_2,\dots,g_k),w_2(g_1,g_2,\dots,g_k),\dots,w_k(g_1,g_2,\dots,g_k)). \] This defines an action of \(\Aut(F_k)\) on \({\mathcal N}_k(G)\). Also it is defined an action of \(\Aut(G)\) on \({\mathcal N}_k(G)\) by \((g_1,g_2,\dots,g_k)\theta=(g_1\theta,g_2\theta,\dots,g_k\theta)\), for all \(\theta\in\Aut(G)\). Let \(\widehat{{\mathcal N}}_k(G)\) denote the set of \(\Aut(G)\)-orbits of \({\mathcal N}_k(G)\) and let \(\widehat{{\mathbf u}}\) denote the orbit containing \({\mathbf u}\in{\mathcal N}_k(G)\). Now it can be defined an action of \(\Aut(F_k)\) on \(\widehat{{\mathcal N}}_k(G)\) by \(\widehat{{\mathbf u}}\sigma=\widehat{{\mathbf u}\sigma}\) for all \(\widehat{{\mathbf u}}\in\widehat{{\mathcal N}}_k(G)\) and all \(\sigma\in\Aut(F_k)\). In [loc. cit.], Neumann and Neumann proved that the action of \(\Aut(F_k)\) on \(\Sigma(G,n)\) is equivalent to its action on \(\widehat{{\mathcal N}}_k(G)\). The orbits of the action of \(\Aut(F_k)\) on \(\widehat{{\mathcal N}}_k(G)\) are also called \(T_k\)-systems of \(G\). We denote by \(\widetilde{{\mathbf u}}\) the orbit of the latter action containing \(\widehat{{\mathbf u}}\in\widehat{{\mathcal N}}_k(G)\) where \({\mathbf u}\in{\mathcal N}_k(G)\). Let \(w\) be a non-trivial word in the free group \(F_k\). It is said that \(w\) is `invariant' on \(T_k\)-systems of \(G\) if, for every two generating \(G\)-vectors \({\mathbf v}=(g_1,g_2,\dots,g_k)\) and \({\mathbf u}=(h_1,h_2,\dots,h_k)\) of \(G\) such that \(\widetilde{{\mathbf v}}=\widetilde{{\mathbf u}}\), we have \[ \{w(g_1\theta,g_2\theta,\dots,g_k\theta)^{\pm 1}\mid\theta\in\Aut(G)\}=\{w(h_1\theta,h_2\theta,\dots,h_k\theta)^{\pm 1}\mid\theta\in\Aut(G)\}. \] By a result of Higman, the commutator \([x_1,x_2]\) is invariant on \(T_2\)-systems of every group \(G\) with \(d(G)\leq 2\) (see [\textit{B. H. Neumann}, Arch. Math. 7, 87-90 (1956; Zbl 0075.23903)] and [\textit{I. Pak}, Ohio State Univ. Math. Res. Inst. Publ. 8, 301-347 (2001; Zbl 0986.68172)]). In [loc. cit.], \textit{B. H. Neumann} asks whether there exists a generalization of Higman's result for \(T_k\)-systems with \(k\geq 3\). In the paper under review, the authors give a negative answer to this question: (Theorem 1.1) For every non-trivial word \(w\in F_k\), where \(k\geq 3\), there exists a finite group \(G\), such that \(w\) is not invariant on \(T_k\)-systems in \(G\). The proof of Theorem 1.1 is based on a result by \textit{R. Gilman} [Can. J. Math. 29, 541-551 (1977; Zbl 0332.20010)], that for each \(k\geq 3\), the group \(\text{PSL}(2,p)\) (\(p\) is a prime \(\geq 5\)) has a unique \(T_k\)-system. In fact, the authors prove that the map \(\phi_w\colon{\mathcal N}_k(G)\to G\) where \(\phi_w(g_1,g_2,\dots,g_k)=w(g_1,g_2,\dots,g_k)\) for every generating \(G\)-vector \((g_1,g_2,\dots,g_k)\) of \(G\), takes unboundedly many values on this \(T_k\)-system, when \(p\to\infty\). The proof idea was motivated by a recent paper [\textit{M. W. Liebeck, A. Shalev}, Ann. Math. (2) 154, No. 2, 383-406 (2001; Zbl 1003.20014)]. The second result of the paper under review is Theorem 1.2.: The number of \(T_2\)-systems in \(\text{PSL}(2,p)\) (\(p\) prime) is unbounded as \(p\to\infty\). This result is due to \textit{M. Evans} [Ph. D. Thesis, University of Wales (1985)], who proved it by an explicit construction. A similar result for alternating groups \(A_n\) was obtained in [\textit{I. Pak}, loc. cit.] by using Higman's result and a simple combinatorial construction.