Analysis of Networks in Chess

In this paper we examine the value of the information contained in networks constructed from positions in chess games with respect to game analysis and predictive power. The interrelations of pieces in a chess game represent complex and evolving networks that describe both tactical and strategic elements of the game. Even when a chess game is represented as a network of interrelations among the pieces, abstracting away all of the specifics and details of the game, such as the value of pieces or the notion of the board ”center,” the network still contains valuable information about the game played, when the critical moments are, and what the likely outcome will be. By representing a game as a series of networks, each modeling a different aspect of the connections among the pieces and the board, even richer information is available. By additionally considering the evolution of the graphs as a time series through the course of the game, still more information can be gathered. With only minimal encoding of game rules and logic, this series of evolving graphs is able to capture with surprising fidelity the evaluation of the game. In this paper we examine the information that can be extracted from the evolution of these networks and show that their predictive power surpasses that of an evaluation function that directly employs the rules and strategies of chess.