On incompleteness of the deleted product obstruction for embeddability (Q1432034)

The deleted product of a space \(X\) is \(\widetilde X = X \times X \setminus (\Delta X)\), where \(\Delta X\) is the diagonal. If \(f:X \to E^m\) (Euclidean \(m\)-space) is an embedding, then \(f\) induces a \(\mathbb{Z}_2\)-equivariant map \(\widetilde f :\widetilde X \to S^{m-1}\). One may ask after the existence of an embedding of \(X\) into \(E^m\) if there exists a \(\mathbb{Z}_2\)-equivariant map \(\varphi: \widetilde X \to S^{m-1}\). Restricting to PL or DIFF this is in many cases true, but not in general, and depends on the dimension of \(X\), which we denote by \(n\). In the PL case the question is open for \(m=3, n=2,3\). The authors offer a counterexample for \(m=n=3\), which is weaker since it holds up to suspension: There exists a manifold \(M^3\) with boundary (a certain punctured homology 3-sphere) that is not embeddable into \(E^3\) but there is an equivariant map \(\varphi : \Sigma \widetilde M^3 \to \Sigma S^2\) (\(\Sigma\) denotes the suspension). In the DIFF case the following result is proved: For \(k > 1\) there exists a closed smooth \(4k\)-manifold \(M^{4k}\) which PL embeds into \(E^{4k+2}\) (hence there exists an equivariant map of \(\widetilde M^{4k}\) into \(S^{4k+1} \subset S^{6k-2}\)) but does not smoothly embed into \(E^{6k-1}\).