The decomposition of 3-dimensional Poincaré complexes (Q1583921)

A \(PD^n\)-complex is a connected finitely dominated CW-complex \(P\) with \(w: \pi = \pi_1 (P) \to \{\pm 1\}\) which exhibits the equivariant Poincaré duality of a closed \(n\)-manifold with orientation class. It is a homotopy analogue of closed manifolds. \textit{V. G. Turaev} [Math. USSR, Sb. 67, No. 1, 261-282 (1990); translation from Mat. Sb. 180, No. 6, 809-830 (1989; Zbl 0693.57013)] proved that \(PD^3\)-complexes are distinguished up to homotopy equivalence by their fundamental groups, orientation class and fundamental class. A challenging problem is to determine which \(PD^3\)-complex is homotopy equivalent to a 3-manifold. Wall showed that if \(\pi\) has 0, 1 or 2 ends, then \(P\) has a universal cover \(\widetilde{P}\) homotopy equivalent to \(S^3\), aspherical, or one of \(\mathbb{R}\mathbb{P}^3 \# \mathbb{R}\mathbb{P}^3\), \(S^1 \times \mathbb{R}\mathbb{P}^2\), \(S^1 \times S^2\), \(S^1 \widetilde{\times}S^2\) respectively, otherwise \(\pi\) has infinitely many ends. Analogous to results of 3-manifolds, Wall asked (1) if \(\pi\) is a proper free product; (2) if such a decomposition of \(\pi\) implies a corresponding connected sum of \(PD^3\)-complexes. Turaev answered the second question affirmatively. The paper under review is to study the first question. For an orientable \(PD^3\)-complex \(P\), if \(\pi\) has infinitely many ends then \(\pi\) is either a proper free product or virtually free. Thus every \(PD^3\)-complex has virtually torsion-free fundamental group. Note that there are examples of \(PD^3\)-complexes which are not homotopy equivalent to any 3-manifolds (but they carry finite fundamental groups). The method in this paper is from ideas and techniques from \textit{J. A. Hillman} [Math. Proc. Camb. Philos. Soc. 114, No. 2, 215-218 (1993; Zbl 0815.57015) ]. The author shows that the homology group \(H_*(C, H^1(\pi, Z\pi))\) with module \(H^1(\pi, Z\pi)\) coefficients can be computed independently from duality, where \(C\) is a cyclic subgroup of \(\pi\). By the accessibility of \(\pi\) and Mayer-Vietoris sequence for graphs, the author derives the main result, from comparing with independent calculations. Theorem 14 of the paper states that for an orientable \(PD^3\), \(\pi\) either has 1 end (a proper free product, spherical complex), or is virtually free of finite rank (a nontrivial connected sum by Turaev's result). The author also was able to recover the main result of Hillman in the above paper. For a pair of \(PD^3\)-complexes, Poincaré duality is replaced by Lefschetz duality of a manifold with boundary. Then the author extends Theorem 14 to pairs of \(PD^3\)-complexes. It would be nice to extend Turaev's result to pairs of \(PD^3\)-complexes.