Multicut Problems in Almost-Planar Graphs: The Dependency of Complexity on the Demand Pattern
Abstract
Given a graph $G$, a set $T$ of terminal vertices, and a demand graph $H$ on $T$, the \textsc{Multicut} problem asks for a set of edges of minimum weight that separates the pairs of terminals specified by the edges of $H$. The \textsc{Multicut} problem can be solved in polynomial time if the number of terminals and the genus of the graph is bounded (Colin de Verdi\`ere [Algorithmica, 2017]). Focke et al.~[SoCG 2024] characterized which special cases of Multicut are fixed-parameter tractable parameterized by the number of terminals on planar graphs. Moreover, they precisely determined how the parameter genus influences the complexity and presented partial results of this form for graphs that can be made planar by the deletion of $\pi$ edges. We complete the picture on how this parameter $\pi$ influences the complexity of different special cases and precisely determine the influence of the crossing number. Formally, let $\mathcal{H}$ be any class of graphs (satisfying a mild closure property) and let Multicut$(\mathcal{H})$ be the special case when the demand graph $H$ is in $\mathcal{H}$. Our fist main results is showing that if $\mathcal{H}$ has the combinatorial property of having bounded distance to extended bicliques, then Multicut$(\mathcal{H})$ on unweighted graphs is FPT parameterized by the number $t$ of terminals and $\pi$. For the case when $\mathcal{H}$ does not have this combinatorial property, Focke et al.~[SoCG 2024] showed that $O(\sqrt{t})$ is essentially the best possible exponent of the running time; together with our result, this gives a complete understanding of how the parameter $\pi$ influences complexity on unweighted graphs. Our second main result is giving an algorithm whose existence shows that that the parameter crossing number behaves analogously if we consider weighted graphs.