Collared and non-collared manifold boundaries (Q2154806)

Let \(M\) be a \textit{manifold-with-boundary}, that is, a topological manifold such that every point has a neighbourhood homeomorphic to some open subset of \(\mathbb R^{n-1}\times\mathbb R_{\geq 0}\), for a fixed value of \(n\in\mathbb N\); the set of points in \(M\) that have a neighbourhood homeomorphic to all of \(\mathbb R^{n-1}\times\mathbb R_{\geq 0}\) is known as the \textit{boundary} of \(M\) and denoted \(\partial(M)\). We say that \(\partial(M)\) is \textit{collared} in \(M\) if there exists an embedding \(h:\partial(M)\times[0,1]\longrightarrow M\) such that \(h^{-1}[\partial(M)]=\partial(M)\times\{0\}\) and the image of \(\partial(M)\times[0,1)\) is open in \(M\); if we can furthermore ensure that the embedding \(h\) is closed, then we say that \(\partial(M)\) is \textit{strongly collared} in \(M\). A classical result of \textit{M. Brown} [Ann. Math. (2) 75, 331--341 (1962; Zbl 0201.56202)] establishes that, if \(M\) is a metrizable manifold, then \(\partial(M)\) is collared in \(M\); this result was later strengthened by \textit{R. Connelly} [Proc. Am. Math. Soc. 27, 180--182 (1971; Zbl 0208.50704)], who showed that in fact \(\partial(M)\) is strongly collared in \(M\) whenever \(M\) is a metrizable manifold. (It seems that, historically, Connelly did not think of his result as a strengthening of Brown's, but rather simply a different proof; he used, however, stronger definitions and so in practice he improved the earlier result.) In the (extremely pleasant to read) paper under review, the author tackles the highly interesting question of which manifolds, when one does not insist that all manifolds be metrizable, have a collared boundary. The author claims no originality to most of the results, them being consequences of carefully following Connelly's proof; these results, however, are not systematically presented elsewhere in the literature and this paper does an outstanding job of carefully fleshing them out. After a brief introduction, the second section provides a detailed proof of the main family of results (stated in the paper as Theorem 1.2): amongst others, for a manifold (metrizable or otherwise) \(M\), if \(\partial(M)\) is Lindelöf then \(\partial(M)\) is collared in \(M\), and if \(\partial(M)\) is compact then \(\partial(M)\) is strongly collared in \(M\). In the reverse direction, notably, for any manifold \(M\) (metrizable or otherwise) such that \(M\setminus\partial(M)\) is metrizable, \(\partial(M)\) being collared in \(M\) is equivalent to it being strongly collared, and also equivalent to all of \(M\) being metrizable. The last section of the paper develops several exotic examples (the name of the section itself is, quite appropriately, ``gallery of horrors''). The first one is a non-metrizable manifold whose boundary is strongly collared; other examples include a normal surface whose boundary is non-collared (even more, the boundary is not a \(G_\delta\) set), a surface whose interior and boundary are both metrizable even though the boundary is not collared (\textit{a fortiori}, this implies that the surface itself is not metrizable), and a surface with metrizable interior whose boundary is, although connected, non-collared; these examples show that, if \(\partial(M)\) is not Lindelöf, then one cannot assume that \(\partial(M)\) is collared in \(M\) -- even if \(M\setminus\partial(M)\) is metrizable and even if \(\partial(M)\) is either Lindelöf or connected. It is also shown that, in the non-metrizable setting, strongly collared is not equivalent to collared: two of the examples (3.8 and 3.9) exhibit surfaces with a collared but not strongly collared boundary (satisfying various other combinations of properties that show the importance of various assumptions from the main theorem). The last three examples are highly interesting and show some fairly pathological combinations of properties, highlighting the richness that one can find when studying nonmetrizable manifolds.