If a saddlepoint does not exist for a payoff matrix, a probabilistic strategy is optimal to maximize expected (average) rewards.