Predicting Non-Square 2D Dice Probabilities

TL;DR

This paper develops a model to predict the probabilities of a 2D rectangular die landing on different sides, validated through experiments and simulations, with implications for physics engine testing.

Contribution

It introduces a novel ratio-based predictive model for 2D die probabilities and validates it experimentally and computationally.

Findings

Good agreement between theory, experiments, and simulations.

Model approximates probabilities better on less bouncy surfaces.

Predicted ratios tend to overestimate actual probabilities.

Abstract

The prediction of the final state probabilities of a general cuboid randomly thrown onto a surface is a problem that naturally arises in the minds of men and women familiar with regular cubic dice and the basic concepts of probability. Indeed, it was considered by Newton in 1664 [1]. In this paper we make progress on the 2D problem (which can be realised in 3D by considering a long cuboid, or alternatively a rectangular cross-sectioned dreidel). For the two-dimensional case we suggest a model that predicts this based on the side length ratio. We test this theory both experimentally and computationally, and find good agreement between our theory, experimental and computational results. Our theory is known, from its derivation, to be an approximation for particularly bouncy or grippy surfaces where the die rolls through many revolutions before settling. On real surfaces we would expect (and we observe) that the true probability ratio for a 2D die is a somewhat closer to unity than predicted by our theory. This problem may also have wider relevance in the testing of physics engines.