Knotting and linking in the Petersen family (Q902242)

This paper studies the relationship between knotting and linking in spatial embeddings of the Petersen family of graphs, that is, the family of graphs obtained from the complete graph on \(6\) vertices by \(\nabla Y\)- and \(Y\nabla\)-moves. H. Sachs, and J. Conway and C. Gordon proved that these graphs all contain at least one non-splittable link in every spatial embedding. An embedding \(f:G\rightarrow {\mathbb R}^3\) is \textit{knotted} if there is a nontrivial knot in \(f(G)\), and is \textit{complexly algebraically linked} if \(f(G)\) contains either a \(2\)-component link \(L\) whose linking number, \({\mathrm {lk}}(L)\), satisfies \(|{\mathrm {lk}}(L)|\geq 2\) or at least two \(2\)-component links \(L_1\) and \(L_2\) with \({\mathrm {lk}}(L_i)\neq 0\) for each \(i\). The main theorem of the paper establishes that for a graph \(G\) in the Petersen family, if \(f\) is a complexly algebraically linked embedding of \(G\) into \({\mathbb R}^3\), then \(f\) is knotted. A graph that has the property that every embedding that is complexly algebraically linked is also knotted is called \textit{\(K\)-linked}. A key technical ingredient in the proof of the main theorem is a formula that relates the linking numbers for pairs of disjoint cycles in an embedding of the graph \(K_{3,3,1}\) with the coefficients of the Conway polynomials for certain cycles within the embedding. In particular, letting \(\Gamma_{3,4}(K_{3,3,1})\) denote the set of pairs consisting of a disjoint \(3\)-cycle and a disjoint \(4\)-cycle in \(K_{3,3,1}\), \(\Gamma_k\) denote the set of \(k\)-cycles in \(K_{3,3,1}\) for \(3\leq k\leq 6\), \(\Gamma_H\) denote the set of all Hamiltonian cycles in \(K_{3,3,1}\), \(A\) denote the vertex of valency \(6\) in \(K_{3,3,1}\), and \(a_2\) denote the second coefficient of the Conway polynomial, the author proves that for every embedding \(f\) of \(K_{3,3,1}\), \[ \sum_{\lambda\in \Gamma_{3,4}(K_{3,3,1})} {\mathrm {lk}}(f(\lambda))^2=2\left (\sum _{\gamma\in \Gamma_H}a_2(f(\gamma))-2\sum_{\gamma\in \Gamma_6\atop A\notin \gamma} a_2(f(\gamma))-\sum _{\gamma\in \Gamma_5\atop A\in \gamma}a_2(f(\gamma)) \right) +1. \] The main theorem follows as a consequence of this formula, a similar formula for embeddings of \(K_6\) due to R. Nikkuni, and another lemma in this paper that proves that if \(G\) is a graph that is \(K\)-linked and \(G'\) is obtained from \(G\) by a \(\nabla Y\)-move, then \(G'\) is \(K\)-linked. The author finishes the paper by showing that the converse of the main theorem is false, and that some other types of complexity in linking for an embedding do not imply that the embedding is knotted. To demonstrate the latter, she gives examples of an unknotted embedding of \(K_6\) that contains a Hopf link and a nonsplittable link with linking number equal to \(0\), and an unknotted embedding of \(K_6\) that has a single nontrivial \(2\)-component link with unknotted components and linking number \(-1\) that is not the Hopf link.