Complexity Results for Rainbow Matchings

Abstract: A rainbow matching in an edge-colored graph is a matching whose edges have distinct colors. We address the complexity issue of the following problem, \mrbm: Given an edge-colored graph $G$, how large is the largest rainbow matching in $G$? We present several sharp contrasts in the complexity of this problem. We show, among others, that * can be approximated by a polynomial algorithm with approximation ratio $2/3-\eps$. * is APX-complete, even when restricted to properly edge-colored linear forests without a $5$-vertex path, and is solvable in %time $O(m^{3/2})$ on edge-colored $m$-edge polynomial time for edge-colored forests without a $4$-vertex path. * is APX-complete, even when restricted to properly edge-colored trees without an $8$-vertex path, and is solvable in %time $O(n^{7/2})$ on edge-colored $n$-vertex polynomial time for edge-colored trees without a $7$-vertex path. * is APX-complete, even when restricted to properly edge-colored paths. These results provide a dichotomy theorem for the complexity of the problem on forests and trees in terms of forbidding paths. The latter is somewhat surprising, since, to the best of our knowledge, no (unweighted) graph problem prior to our result is known to be NP-hard for simple paths. We also address the parameterized complexity of the problem.