Embedded spheres in \(S^2 \times S^1\# \cdots \# S^2 \times S^1\) (Q818360)

Let \(M\) be a connected sum of copies of \(S^2\times S^1\). This paper gives criteria for an element of \(\pi_2(M)\) to be representable by an embedded \(2\)-sphere in \(M\). Algebraically, such embedded spheres correspond to splittings of the free group \(\pi_1(M)\) as a free product. Elements of \(\pi_2(M)\) correspond to elements of the second homology of the universal cover \(\widetilde{M}\) of \(M\). Since an embedded \(2\)-sphere must separate, a necessary condition to represent a class in \(H_2(\widetilde{M})\) by an embedding is that every proper map of the line into \(\widetilde{M}\) have algebraic intersection number \(-1\), \(0\), or \(1\) with the class. Using various results of \(3\)-dimensional topology, the author proves that this condition is also sufficient. The condition for two embeddable classes \(A\) and \(B\) to be disjointly embeddable is that there do not exist proper maps \(c\) and \(c'\) from the real line into \(\widetilde{M}\) having intersection numbers \(c\cdot A=1=c\cdot B\) but \(c'\cdot A=1=-c'\cdot B\). In particular, this tells whether a class in \(\widetilde{M}\) is representable by a sphere disjoint from all its translates, hence whether the corresponding element of \(\pi_2(M)\) is representable by an embedding. The criteria lead to finite algorithms for checking representability. The reason is that \(\widetilde{M}\) is an increasing union of punctured \(3\)-cells, so any class in \(H_2(\widetilde{M})\) can be represented by a map with image in some punctured cell \(\Sigma\) in \(\widetilde{M}\) (which may be chosen large enough to contain any translates that meet the image). It is only necessary to check the intersection conditions for a set of proper arcs connecting each pair of boundary spheres of~\(\Sigma\). A key ingredient in realizing a class satisfying the criteria by an embedding is a theorem of \textit{P. Scott} and \textit{G. A. Swarup} [Geom. Topol. 4, 179--218 (2000; Zbl 0983.20024)], which gives conditions under which almost invariant sets in \(\widetilde{M}\) give rise to splittings of \(\pi_1(M)\). To the reviewer, it appears likely that much of this nice paper could be extended to more general classes of \(3\)-manifolds.