Locally contractible coset spaces (Q2353620)

\textit{J. Szenthe} [Acta Sci. Math. 36, 323--344 (1974; Zbl 0269.57019)] stated the following result: Let \(G\) be a locally compact group acting transitively on a locally compact space \(X\). If \(X\) is locally contractible and \(G\) is \(\sigma\)-compact and has a compact group \(G/G^\circ\) of connected components, then \(G\) is a Lie group and \(X\) is a manifold. The first author [Topology Appl. 159, No. 9, 2235--2247 (2012; Zbl 1247.22008)] discovered a serious gap in the proof given by \textit{J. Szenthe} [Acta Sci. Math. 36, 323-344 (1974; Zbl 0269.57019)]. The present authors now give a completely different proof of that result. Independently, \textit{K. H. Hofmann} and \textit{L. Kramer} [J. Reine Angew. Math. 702, 227--243 (2015; Zbl 1330.22023); erratum ibid. 702, 245--246 (2015)] have given another proof of Szenthe's claim. Both proofs have in common that they use many classical ingredients from the structure theory of locally compact groups, and that they first treat the case of a compact group and then use some reduction argument to treat the general case. For the compact case, both use some homotopy theoretic tools. In the article under review, the latter are somewhat more dominant. The main result states that the following properties of the coset space of a locally compact group \(G\) modulo a closed subgroup \(H\) are mutually equivalent: (1) \(G/H\) is locally contractible; (2) \(G/H\) is finite-dimensional and locally connected; (3) \(G/H\) is a manifold. The hitherto unproved implication is \((1) \Rightarrow (2)\). For \(G\) compact, the implication \((1) \Rightarrow (3)\) is obtained in three major steps, by showing first that \(G/H\) is an absolute neighborhood extensor (ANE) for compact spaces (essentially by proving that it has the homotopy extension property (HEP)), then that \(G/H\) is a metrizable absolute neighborhood retract (ANR), and finally, that \(\dim G/H\) is finite. For the last one of these three steps, it is shown that an infinite dimensional coset space \(G/H\) of a compact group satisfies the disjoint cells property (DCP), i.e., every map of a closed Euclidean cell into \(G/H\) can be approximated by pairs of maps having disjoint images. This implies that the identity map of \(G/H\) is inessential, a contradiction to properties of \(G/H\) that also occur in the proof by \textit{K. H. Hofmann} and \textit{L. Kramer} [J. Reine Angew. Math. 702, 227--243 (2013; Zbl 1330.22023); erratum ibid. 702, 245-246 (2015)].