Tight immersions of highly connected manifolds (Q1087157)

The only known examples of compact (k-1)-connected 2k-manifolds (k\(\geq 2)\) admitting smooth tight immersions into Euclidean space are connected sums of copies of \(S^ k\times S^ k\) and the projective planes over \({\mathbb{R}}\), \({\mathbb{C}}\), \({\mathbb{H}}\) and the Cayley numbers. In this paper certain topological obstructions are given which restrict the class of such manifolds. In particular infinitely many simply connected 4- manifolds cannot be immersed tightly (among them the Kummer surface). The results are the following: Let \(M^{2k}\to E^{2k+l}\) be a substantial tight immersion of a (k-1)-connected manifold. If the k-th Stiefel-Whitney class \(w_ k(M)\) vanishes then \(l=1\) or \(l=2\), otherwise \(k=1,2,4,8\) and \(l=k\), \(k+1\) or \(k+2\) (in the cases \(k=4\) or \(k=8\) only \(l=k+1\) or \(k+2)\). Moreover if \(w_ k(M)=0\) and \(k>2\) M must have the integral cohomology ring of a connected sum of copies of \(S^ k\times S^ k.\) For 4-manifolds \((k=2)\) the following is shown: codimension \(l\geq 3\) implies that M splits into \(M\cong {\mathbb{C}}P^ 2\#N\), \(l=2\) implies that M splits into \(M\cong (S^ 2\times S^ 2)\#N\) and the Betti number \(\beta_ 2(M)\) is even. In particular there is no tight immersion \({\mathbb{C}}P^ 2\#(-{\mathbb{C}}P^ 2)\to E^ 6\). The results by Donaldson about the non-splitting of manifolds as a connected sum lead to the corollary that many simply connected 4-manifolds have no tight immersion into any Euclidean space (including the K3-surfaces). It remains an open problem if there are tight immersions \(M^ 4\to E^ 6\) of simply connected 4-manifolds with \(w_ 2\neq 0.\) The proof uses top sets and top cycles. The key observation is that a point \(p\in M\) which is a nondegenerate minimum for some height function is contained in convex top cycles of dimension k in a \((k+1)\)-dimensional affine subspace meeting transversally at p, and that these top cycles are parametrized over \(S^{l-2}\). Consequently for \(l\geq 3\) there are infinitely many of them, and hence at least one k-homology class has selfintersection \(\pm 1\). Then the machinery of quadratic forms, characteristic classes, Wu formula, signature theorem etc. is applied to get the results. \{Reviewer's remark: The assumption of smoothness is essential. There are tight polyhedral embeddings of connected sums of copies of \({\mathbb{C}}P^ 2\) into \(E^ 8\), and even for the Kummer surface it is open if there is a tight polyhedral embedding into Euclidean space.\}