On link maps and their homotopy classification (Q1113490): Difference between revisions

Given integers m,p,q\(\geq 0\), a map \(f=f_ 1\coprod f_ 2: S^ p\coprod S^ q\to S^ m\) is a link map if the two spheres \(S^ p\) and \(S^ q\) have disjoint images. Two such link maps are called link homotopic if they are homotopic through link maps. In this paper the author continues his analysis of the resulting set (and often group) \(LM^ m_{p,q}\) of link homotopy classes. The central tool consists of two exact sequences, valid under the dimension assumptions \((*)\quad q\leq 2(m-1)/3\) and \((**)\quad p\leq 3(m-q-2),\) and closely related to James's homotopy theoretical EHP-sequence. They involve the invariants \(\alpha (f)\in \pi^ S_{p+q+1-m}\) (a generalization of the classical linking number) and, even more importantly, \({\tilde \beta}\)(f) (which decides whether f can be deformed until it embeds \(S^ q)\). The free and odd torsion parts of \(LM^ m_{p,q}\) are described completely while an analysis of the 2- primary torsion unavoidably leads to very subtle homotopy theoretical questions. Nevertheless, ker(\(\alpha\) \(\oplus {\tilde \beta})\) and \(LM^ m_{p,q}\) itself are calculated for many interesting examples, and so are the natural suspension homomorphisms \(\Sigma_ 1\) and \(\Sigma_ 2\); e.g. if \(m=p+2=2q-1=7(8)\) and \(q>8\), then \(\Sigma_ 1\) has infinitely generated kernel and cokernel both of which are detected by \({\tilde \beta}\). Also, various recent results of Kirk, Massey, Rolfsen and others follow as rather special cases. It may be worth pointing out that the methods developed in the present paper allow a staightforward generalization to link maps with any finite number r of components. In the corresponding dimension range the 2- component sub-link maps turn out to contain most of the crucial information so that also the explicit computations of this paper are important in the case of general r.