Dense products in fundamental groupoids (Q2280130)

The paper proves two main theorems for metric path-connected spaces. The first states that the well-definedness of transfinite products in the fundamental groupoid of a space is definitely a stronger property than the well-definedness of transfinite products in the fundamental group, and the second states that the existence of a generalized universal covering space in the sense as it was proposed by [\textit{H. Fischer} and \textit{A. Zastrow}, Fundam. Math. 197, 167--196 (2007; Zbl 1137.55006)] is equivalent to the well-definedness of transfinite products in the fundamental groupoid. Towards the end of the paper Theorem 1.2 is expanded to a version also covering non-universal generalized covering spaces, as will be described in the penultimate paragraph of this review. For understanding and appreciating these results, we should continue with some explanations of the terms used above: Observe that in a fundamental group, and within its representing paths, product-constructions are possible, that one would not associate with the operator ``\(\prod^\infty_{i=1}\)'', as it is used in analysis. E.g., one could use the standard middle-third Cantor Set to subdivide the unit-interval into the (countably many) complementary subintervals, and if one maps the Cantor set to the base-point and associates one path returning to the basepoint to each of the complementary domains (accordingly shorter paths to the shorter intervals), then one defines a kind of infinite product of paths, but by the order of factors it is quite different from those products that one would have associated with the symbol ``\(\prod^\infty_{i=1}\)''. Since the author had already in [\textit{J. Brazas}, Proc. Am. Math. Soc. 148, No. 6, 2655--2670 (2020; Zbl 1459.57025)] considered products of the latter type and called them there ``infinite'', he has probably chosen the word ``\textit{transfinite}'' in the paper under review, to make clear that all types of infinite products (in particular with all possible orders of factors) are to be considered. There are quite some different constructions of \textit{generalized covering space} theories in the literature. The one, to which the above-quoted theorem refers, can be briefly described as follows: As in classical covering space theory, one can use the set of all relative homotopy classes of paths starting at the base point but ending in an arbitrary point as point-set of the universal covering space, and the end-point projection as candidate for the covering projection. In the classical theory it is used that for a semilocally simply connected space one obtains evenly covered neighbourhoods, and when they are used for topologizing the covering space, the above described construction gives the classical properties that are expected from a classical covering projection; in the above quoted paper [Fischer and Zastrow, loc. cit.] it was pointed out, that there is a wider class of spaces, where even in the absence of neighbourhoods as provided by semilocally simply connected spaces, the above described construction with an according adaptation of how to define the topology on the covering space gives projections that satisfy the unique path-lifting property and that this suffices (although there are no evenly covered neighbourhoods) to retain most of the applications of classical covering-space theory. Neither the above-mentioned paper [Fischer and Zastrow, loc. cit.], nor [\textit{H. Fischer} et al., Topology Appl. 158, No. 3, 397--408 (2011; Zbl 1219.54028)] which compares in its final diagram a lot of related conditions to the existence of generalized covering spaces, were able to find an equivalent condition to the existence of generalized universal covering spaces. In order to finally understand the issue of \textit{well-definedness of transfinite products in fundamental groups and groupoids,} let us look at the following examples: First take countably many circles whose radii are given by a null-sequence, and embed them so into the plane, that they have just one common tangent point. Take the topology as being obtained by the embedding into the plane. If one imagines the space so that all circles lie on the same side of the tangent line through this common tangent point, one usually calls this space ``the Hawaiian Earring''. Also in the paper under review this phrase is used, but the author has meanwhile (cf. his latest post in [\textit{J. Brazas}, ``Homotopy groups of shrinking wedges of non-simply connected CW-complexes'', Preprint, \url{arXiv:2204.03751}]) switched the name to ``Earring Space''. If one puts half of the circles on one side of the just mentioned common tangent line, and the other half on the other, the resulting space will be homeomorphic, but can also be regarded as a one-point union of two Earring Spaces. Let us take the latter image, and then in three-space put two cones which are disjoint apart from the common tangent point over the two Earring Spaces. The resulting space is then called ``the double cone over the Earring Spaces'', or ``Griffiths' Space'', according to the paper [\textit{H. B. Griffiths}, Q. J. Math., Oxf. II. Ser. 5, 175--190 (1954; Zbl 0056.16301)] where it was first constructed. At least since [\textit{J. W. Morgan} and \textit{I. Morrison}, Proc. Lond. Math. Soc. (3) 53, 562--576 (1986; Zbl 0609.57002)] it is clear, that although being a one-point-union of two contractible spaces, it has a non-trivial fundamental group, namely the path which remains only on the basis of the cones, and passes through the biggest ring on one side, then through the biggest ring on the other side, the second biggest ring on the first side, the second biggest ring on other side, and so on, is known to be non-nullhomotopic. Since the rings are getting small, such a path can be continuously parametrized on the unit interval so as to pass over all infinitely many rings, where the \(i^{\text{th}}\) passage takes place on the interval \([\frac{i-1}{i},\frac{i}{i+1}]\). One can think of the just-described path as being an infinite product of ring-passages. On the other hand, each ring-passage takes place at the base of some cone, i.e. is a contractible path, and so we have here the situation, that an infinite concatenation of contractible paths is non-contractible. Conversely, if we would have taken all ring-passages on the same side of our double-cone, then an analogous concatenation of contractible paths would have been contractible. And so the conclusion is that the infinite concatenation of the contractible homotopy class (as elements of a fundamental group) is not well-defined in this case, because, depending on how we choose the representatives of our homotopy class, the result can be either the trivial or a non-trivial homotopy class. The main tool that the paper under review uses to achieve the above described results, is the test-space philosophy, which was introduced with the coauthorship of the author in [\textit{J. Brazas} and \textit{H. Fischer}, J. Topol. Anal. 12, No. 1, 37--85 (2020; Zbl 1454.55015)], and is hence not very known so far. In addition, the according terminology of ``closure pairs \((T,g)\)'' and discussion ``whether subgroups are \((T,g)\)-closed'' is not too easy to get used to, but seems to be an adequate tool in this context. Let us therefore try to explain it: The typical test-space will consist of (or at least contain) a subspace, which looks a bit more general like a classical graph with infinitely many loops, but with a topology that allows also to describe phenomena like accumulation of vertices or nullsequences of loops. In addition, some of these loops might be spanned by two-disks, where not necessarily, if a sequence of loops should form a null-sequence, the corresponding spanning disks will also form a null-sequence. In so far the Earring Space and Griffiths' Space as discussed above, are possible test-spaces. Of course, the test-spaces are also varying in their geometry; infinitely many loops might not only be arranged like for the Earring Space, but also so, that they might rather resemble a string of pearls. The test-space \(\mathbb{T}\) should not be confused with the space \(X\), whose fundamental group is actually to be investigated. When for a subgroup of \(\pi_1(X)\) it is to be investigated, whether it is ``\((T,g)\)-closed'', then the typical choice for \(T\) is some algebraically infinitely generated subgroup of \(\pi_1(\mathbb{T})\), and the typical choice for \(g\) is some element in \(\pi_1(\mathbb{T})\), which is only in \(\pi_1(\mathbb{T})\) described via an infinite concatenation of paths whose homotopy classes are contained in \(T\). So \(g \notin T\), but there are situations, where for any continuous map \(\mathbb{T} \to X\) that maps the elements from \(T\) into \(H\), the infinite description that \(g\) has via elements from \(T\) suffices to guarantee that also the image of \(g\) will be mapped by the corresponding induced map into \(H\). This is the situation where \(H\) will be called ``\((T,g)\)-closed''. Test-space characterizations (but with different \(T\) and different \(g\)) of the well-definedness of transfinite products in the fundamental group and the fundamental groupoid were from [Brazas and Fischer, loc. cit.] known before. The author now finds a subgroup of the fundamental group in one of these test-spaces which satisfies one, but not the other characterization, and proves his \textit{first main theorem} by attaching so many cells to this space, till this subgroup becomes the trivial group. The key-observation for proving the \textit{second main theorem} was made by the reviewer in an old unpublished preprint (as indirectly acknowledged in the introduction of [Brazas and Fischer, loc. cit.]). Namely, if the above described construction-mechanism from [Fischer and Zastrow, loc. cit.] fails to give the unique-path-lifting property, it must be possible to map a system of paths and some connecting homotopies (that together may be interpreted as a test-space) in a certain way into the base space. In [Brazas and Fischer, loc. cit., Theorem 4.13] this observation gets formalized as a criterion for the unique path-lifting property. The author of the paper under review can that way compare test-space conditions for the existence of generalized universal covering spaces with those for the well-definedness of transfinite products, and so prove his second main theorem. A necessary and sufficient condition via test-spaces for the existence of generalized universal covering spaces was already contained in [Brazas and Fischer, loc. cit., Theorem 6.1]. The paper under review now continues the work in particular for non-universal (so-called: intermediate) covering spaces. Already in [Fischer and Zastrow, loc. cit.] it was pointed out that, although the existence of generalized universal covering space allows by identification of leaves to come up with a natural candidate of any intermediate covering space, this identification-process will often spoil the unique-path lifting property, so that in these cases the intermediate covering spaces will not exist. The above-mentioned criterion from [Brazas and Fischer, loc. cit., Theorem 4.13] works in both directions and for all subgroups. However, the involved test-space is very complicated, its possibilities to map accordingly difficult to understand, and the author wants to find a criterion via a test-space that is more simple to be applied, because the loops are aligned like in a string of pearls. In his culminating Theorem 6.10, when translated into the language of test-spaces, from the method as it is proven the author comes up with such a necessary and sufficient test-space criterion for the existence of intermediate coverings with normal characteristic subgroups. He interprets this criterion also as ``well-definedness of transfinite products in the fundamental groupoid relative to the corresponding subgroup''. He is by a corresponding example able to point out that the condition ``normal'' cannot be omitted from this theorem. With K. Eda in his first papers on one-point unions [\textit{K. Eda}, Proc. Am. Math. Soc. 109, No. 1, 237--241 (1990; Zbl 0697.55016); Q. J. Math., Oxf. II. Ser. 42, No. 168, 443--456 (1991; Zbl 0754.55004)] taking up the work of [Morgan and Morrison, loc. cit.], a line of research has started, that in particular tries to shed light on the behaviour of classical (not just defined via limiting processes) algebraic invariants of spaces that are locally complicated and where these invariants therefore cannot be directly computed with the classical methods. The questions, how infinite products that are naturally around for many of these spaces, can be handled, and to what extent the methods from covering space theory can be applied, do naturally show up in this context. In so far the paper should be understood as a paper that makes a step forward in this line of research and contributes new insights by pointing out relations between the various conditions under discussion that had not been observed so far.