Knowledge-Based Move Selection and Evaluation to Guide the Search in Chess Pawn Endings

Abstract

PAWNKING uses, especially in move selection and evaluation, much knowledge to guide the search in chess pawn endings. There are no artificial limits like search depth or any other restriction of effort. Patterns serve to recognize important positional components in a given position and the presence of several configurations of positional elements suggests application of appropriate plans selecting useful moves to be searched. Evaluation is performed by classifying positional elements very exactly (partial evaluation) and by discovering their relevance and contribution to the value of the whole board’s situation (global evaluation). Also, some useful search features are presented. An example from grandmaster practice is discussed, where PAWNKING found all relevant lines of play. Finally, the proposed methods are discussed with respect to their critical points and to possible extensions.