We evaluate the best-response algorithm in the context of pure-integer convex quadratic games. We provide a sufficient condition that if certain interaction matrices (the product of the inverse of the positive definite matrix defining the convex quadratic terms and the matrix that connects one player's problem to another's) have all their singular values less than 1, then finite termination of the best-response algorithm is guaranteed regardless of the initial point. Termination is triggered through cycling among a finite number of strategies for each player. Our findings indicate that if cycling happens, a relaxed version of the Nash equilibrium can be calculated by identifying a Nash equilibrium of a smaller finite game. Conversely, we prove that if every singular value of the interaction matrices is greater than 1, the algorithm will diverge from a large family of initial points. In addition, we provide an infinite family of examples in which some of the singular values of the interaction matrices are greater than 1, cycling occurs, but any mixed-strategy with support in the strategies where cycling occurs has arbitrarily better deviations. Then, we perform computational tests of our algorithm and compare it with standard algorithms to solve such problems. We notice that our algorithm finds a Nash equilibrium correctly in every instance. Moreover, compared to a state-of-the art algorithm, our method shows similar performance in two-player games and significantly higher speed when involving three or more players.
翻译:暂无翻译