Local Matching Dynamics in Social Networks

We study stable marriage and roommates problems in graphs with locality constraints. Each player is a node in a social network and has an incentive to match with other players. The value of a match is specified by an edge weight. Players explore possible matches only based on their current neighborhood. We study convergence of natural better-response dynamics that converge to locally stable matchings - matchings that allow no incentive to deviate with respect to their imposed information structure in the social network. For every starting state we construct in polynomial time a sequence of polynomially many better-response moves to a locally stable matching. However, for a large class of oblivious dynamics including random and concurrent better-response the convergence time turns out to be exponential. In contrast, convergence time becomes polynomial if we allow the players to have a small amount of random memory, even for many-to-many matchings and more general notions of neighborhood.