Finite type link-homotopy invariants (Q1604963)

This paper addresses the question of constructing link-homotopy invariants which are of finite type. For example the linking numbers are such finite-type invariants, but are there any others? In [J. Knot Theory Ramifications 10, No. 7, 1025-1039 (2001; Zbl 1002.57029), see above] \textit{B. Mellor} and \textit{D. Thurston} showed that these are the only finite-type homotopy invariants if the number of components is less than 6, but, for 9 components or more, showed that there are finite-type homotopy invariants which are not just polynomials in the linking numbers. The approach of the present paper is to use the Milnor \(\mu\)-numbers (higher order linking numbers) to construct such invariants. The \(\mu\)-numbers are not, in general, well-defined link invariants, but certain polynomial functions of them are. However, for string links they are well-defined and, if they have non-repeating indices, are even homotopy invariants. Furthermore, a fundamental result of Habegger-Lin defines an explicit group action on the set of string links whose orbits correspond exactly to links, under closure. Thus one needs to find polynomials in the non-repeating \(\mu\)-numbers which are invariant under this group action. This reduces the problem to linear algebra and the author is able to show that such polynomials, using only the linking numbers and triple \(\mu\)-numbers, do exist and can be constructed explicitly, for 6 or more components. This fits nicely with the Mellor-Thurston result. A particular consequence is the existence of finite-type invariants which can distinguish a link from its inverse for 6 or more components. It is a major problem whether finite-type invariants can distinguish a knot from its inverse.