Not all links are concordant to boundary links (Q5906466)

This paper gives the first examples of links (with vanishing Milnor \(\overline \mu\)-invariants when \(n = 1)\) which are not concordant to any boundary link. A link is a collection of disjointly imbedded \(n\)-spheres in \((n + 2)\)-space. It is boundary if the components bound disjoint submanifolds of \((n + 2)\)-space. The \(\overline \mu\)-invariants are a collection of numerical invariants defined by Milnor in 1957, which, in general, have a complicated indeterminacy, but whose vanishing is well- defined. They are concordance invariants and vanish for boundary links. Most previous knowledge of link concordance was confined to boundary links, which are well-understood. There have been some speculative concordance invariants, introduced at various times, which do vanish for boundary links, but all of these, except for the \(\overline \mu\)- invariants, either were found subsequently to always vanish or remain a complete mystery. The examples produced in this paper are all odd-dimensional homology boundary links (the components bound disjoint singular submanifolds). Subsequent to this work, more examples were constructed by others using different techniques, but the existence of even-dimensional examples and the question of whether every link (with vanishing \(\overline \mu\)- invariants when \(n = 1)\) is concordant to a homology boundary link are still open problems. A related open problem is whether every null- concordant boundary link is boundary null concordant. The method of the present paper is roughly as follows. Given a link \(L\) one can choose a component and take a \(p\)-fold branched cover of the sphere branched along this component. The result is a \(Z/p\)-homology sphere \(\Sigma\). Lifting the components of \(L\) into \(\Sigma\) gives a new link \(\widetilde L\) -- a covering link of \(L\). This process can be repeated. If \(L\) is a boundary link, then so is \(\widetilde L\). The authors now consider two invariants of \(\widetilde L\). The complexity is a measure of the failure of the meridians to generate the homology of the complement. If \(\Sigma\) is the standard sphere or if \(\widetilde L\) is a boundary link then they do generate the homology and the complexity is 1. The second invariant is given by the Blanchfield pairing of \(\widetilde L\). This is considered as an element of a Witt group of torsion Hermitian forms over a Laurent polynomial ring with \(p\)-adic integers as coefficients. Combining these two invariants in the right way produces a concordance invariant. Despite the complicated nature of the invariant, the examples which the authors produce, and which their invariant is able to distinguish from any boundary link, are remarkably simple and natural.