Geometry-Aware MCTS Solves Combinatorial Geometry Problems More Efficiently.

Researchers have developed a novel Geometry-Aware Monte Carlo Tree Search (MCTS) framework designed to address challenging extremal problems in combinatorial geometry. These problems involve finding specific point configurations on a grid that adhere to strict global geometric rules, often leading to computational bottlenecks for traditional solvers due to combinatorial explosion. Standard AI methods like reinforcement learning also struggle with sparse rewards and token limits. The new MCTS framework overcomes these issues by strictly enforcing geometric constraints through incremental updates to the feasible action space. For instance, in problems involving collinear points, this mechanism reduces constraint checking complexity significantly. Furthermore, the framework leverages geometric symmetries through canonical pruning during node expansion and symmetric batch transitions, which accelerates the discovery of promising configurations. Extensive experiments demonstrated the effectiveness of this approach, establishing new best-known computational results for five out of six considered problems. Notably, it found larger configurations for the classic No-Three-in-Line problem and improved upper bounds for the Smallest Complete Set problem. This work positions Geometry-Aware MCTS as a highly adaptable tool for discovering novel configurations in complex geometric scenarios.