Global and local game tree search

Abstract

Computer programs based on minimax search have achieved great success, solving a number of classic games including Gomoku and Nine Men's Morris, and reaching a performance that approaches or surpasses the best human players in other well-known games such as checkers, Othello and chess. All these high-performance game-playing programs use global search methods, which evaluate complete game positions. Local search is an alternative approach that works well in the case where a game state can be partitioned into independent subgames. The method of decomposition search [M. Müller, in: IJCAI-99, 1999, p. 578] can solve such games by a combination of local combinatorial game search with evaluation techniques from combinatorial game theory. We compare local and global search in endgame problems in the game of Go, which has been traditionally regarded as beyond the range of exact search-based solution methods. An endgame solver based on decomposition search can solve problems with solution lengths exceeding 60 moves.