PEASANT: An endgame program for kings and pawns

Abstract

Tournament chess programs have had considerable difficulty in playing the endgame. The game scores presented by Professor Mittman in Chapter 1 and by Newborn [73] provide ample evidence of this weakness. Levy [67] has frequently pointed out this shortcoming. It is not clear whether the poor play which is generally observed is due to a basic weakness of the Shannon—Turing minimax approach [81, 82, 95, 96] as Harris suggests (see Chapter 7) or whether it is because the evaluation functions that have been used were not intended for the endgame. To examine this question, a program was written which uses a conventional α—β minimax search procedure but employs an evaluation function and tree-pruning heuristics which are specifically tailored for the endgame.