Transitions of Solutions and Their Efficiency

Abstract

We broaden the basis of non-cooperative game theory by considering miscoordination on a solution concept. For any solution concept, we extend the solution set of a strategic-form game to a transition set. This set contains profiles where various agents simultaneously follow different solutions, e.g.~different Nash equilibria. This models the fact that in practice, complicated agents are rarely perfectly coordinated on the same equilibrium. We define two efficiency measures, called the price of transition anarchy and stability, and bound them. We also refine the notion of transition to the notion of limited transition, where only a limited number of solutions is simultaneously played, and to stable transitions, which allow for only minor lack of coordination. We compare the above mentioned efficiency measures and bound the efficiency of transitions in important cases, including the important cases of constant-sum and potential games, which span the set of finite games with the same number of strategies for each agent. We also prove tight efficiency bounds for routing games and coordination games on graphs. Finally, we study algorithms to find the transition degree required to make a given profile a transition, or to render all the profiles transitions. We conclude that for the sake of efficiency, it is crucial to avoid uncoordinated transitions, besides certain cases, such as constant-sum games, identical utility games, some types of routing games, limited transitions in potential games, and stable transitions in coordination games.