A hybrid discrete-polynomial dynamical system modeling board positions in the game of go

Thomas Wolf
Brock University

It is well known that the game of go poses a special challenge to computer technology, from modeling to programming, from Artificial Intelligence to hardware. Despite three decades of world wide effort and recent progress with Monte Carlo type go programs it is still possible for young children of Dan level playing strength to beat the strongest programs running on clusters with 100s of nodes.

The talk will describe the dynamic creation of a model that aims at describing board positions in go. The model exploits the partially local nature of go (more precisely of the capture rule in go) and leads in its simple version to a polynomial dynamical systems for over 300 unknowns which to formulate needs computer algebra support. Better models require the solution of hybrid discrete-polynomial systems.

The talk will further report on the numerical solution of these dynamical systems. A demo shows the real time operation of such a model and compares its move predictions to moves made in professional games.