The epimorphism problem for Hausdorff topological groups (Q1329400)

Let \(\mathcal G\) be the category of Hausdorff topological groups and continuous homomorphisms. The problem in question, circulated broadly in the 1960's by K. H. Hofmann, is this: Given \(G\) in \(\mathcal G\) and a proper closed subgroup \(H\) of \(G\), must there exist \(K\) in \(\mathcal G\) and distinct morphisms \(f\) and \(g\) from \(G\) into \(K\) such that \(f| H =g| H\)? In categorical language: Do all epimorphisms in \(\mathcal G\) have dense range? Since affirmative solutions to Hofmann's question have been obtained in several important special cases -- by A. G. Kurosh, A. K. Livshits and E. G. Schul'geifer (1960) when all groups are discrete, by D. Poguntke (1970) when all are compact, by E. Nummela (1978) when all are locally compact, by W. F. LaMartin (1979) and P. Nicholas (1986) when all are \(k_ \omega\)-groups -- it is surprising that the answer is ``No''. The author's example, verified here in detail, takes for \(G\) the automorphism group of a compact connected manifold \(M\) without boundary in the topology of uniform convergence and for \(H\) the stability subgroup at some point of \(M\). Given \(G_ i\) in \(\mathcal G\) (\(i = 0,1\)) with a common closed subgroup \(A\), the so-called free product \(P = G_ 0 *_ A G_ 1\) in \(\mathcal G\) with \(A\) amalgamated is characterized as follows: (a) \(P\) is an object of \(\mathcal G\); (b) there are canonical morphisms \(g_ i : G_ i \to P\) in \(\mathcal G\) which agree on \(A\), and (c) for every pair of morphisms \(h_ i : G_ i \to Q\) in \(\mathcal G\) which agree on \(A\) there is a unique morphism \(f : P \to Q\) such that \(h_ i = fg_ i\). The author notes, thus settling another question closely related to Hofmann's, that his group \(G\) and subgroup \(H\) have this property: the free product \(G *_ H G\) in \(\mathcal G\) with \(H\) amalgamated coincides with \(G\) itself, hence is not equal (considered as a group without topology) to the free product \(G *_ H G\) in the category of groups without topology. It remains unsolved, however, whether in \(\mathcal G\) the canonical morphisms \(g_ i : G_ i \to P = G_ 0 *_ A G_ 1\) must always be homeomorphic embeddings.