The complexity of computing a (quasi-)perfect equilibrium for an n-player extensive form game of perfect recall (1408.1233v3)

Abstract: We study the complexity of computing or approximating refinements of Nash equilibrium for a given finite n-player extensive form game of perfect recall (EFGPR), where n >= 3. Our results apply to a number of well-studied refinements, including sequential (SE), extensive-form perfect (PE), and quasi-perfect equilibrium (QPE). These refine Nash and subgame-perfect equilibrium. Of these, the most refined notions are PE and QPE. By classic results, all these equilibria exist in any EFGPR. We show that, for all these notions of equilibrium, approximating an equilibrium for a given EFGPR, to within a given desired precision, is FIXP_a-complete. We also consider the complexity of corresponding "almost" equilibrium notions, and show that they are PPAD-complete. In particular, we define "delta-almost epsilon-(quasi-)perfect" equilibrium, and show computing one is PPAD-complete. We show these notions refine "delta-almost subgame-perfect equilibrium" for EFGPRs, which is PPAD-complete. Thus, approximating one such (delta-almost) equilibrium for n-player EFGPRs, n >= 3, is P-time equivalent to approximating a (delta-almost) NE for a normal form game (NFG) with 3 or more players. NFGs are trivially encodable as EFGPRs without blowup in size. Thus our results extend the celebrated complexity results for NFGs to refinements of equilibrium in the more general setting of EFGPRs. For 2-player EFGPRs, analogous complexity results follow from the algorithms of Koller, Megiddo, and von Stengel (1996), von Stengel, van den Elzen, and Talman (2002), and Miltersen and Soerensen (2010). For n-player EFGPRs, an analogous result for Nash and subgame-perfect equilibrium was given by Daskalakis, Fabrikant, and Papadimitriou (2006). However, no analogous results were known for the more refined notions of equilibrium for EFGPRs with 3 or more players.