Bounding Game Temperature Using Confusion Intervals

TL;DR

This paper introduces the first general upper bounds on the temperature of combinatorial games using confusion intervals, providing a new method to estimate volatility across game classes.

Contribution

It establishes a universal upper bound on game temperature based on confusion interval lengths and introduces a method to compute these bounds for various game subclasses.

Findings

Bounded temperature for certain game classes with fixed confusion interval limits

A method to estimate confusion interval bounds using passing move analysis

Application of bounds to subclasses of Domineering and Snort

Abstract

For combinatorial games, temperature is a measure of the volatility, that is, by how much the advantage can change. Typically, the temperature has been measured for individual positions within specific games. In this paper, we give the first general upper bounds on the temperature for any class of games. For a position $G$, the closure of the set of numbers $\{g\}$ such that $G-g$ is a first player win, is called the confusion interval of $G$. Let $\ell(G)$ be the length of this interval. Our first main result is: For a class of games $\mathscr{S}$, if there are constants $J$ and $K$ such that $\ell(G^L),\ell(G^R)\leq J$ and $\ell(G)\leq K$ for for every $G\in \mathscr{S}$, then the temperature of every game is bounded by $K/2+J$. We give an example to show that this bound is tight. Our second main result is a method to find a bound for the confusion intervals. In $G^L-G$ when Left gets to go first, the number of passing moves required by Right to win gives an upper bound on $\ell(G)$. This is the first general upper bound on temperature. As examples of the bound and the method, we give upper bounds on the temperature of subclasses of Domineering and Snort.