Nimbers are inevitable

This article concerns the resolution of impartial combinatorial games, and in particular games that can be split in sums of independent positions. We prove that in order to compute the outcome of a sum of independent positions, it is always more efficient to compute separately the nimbers of each independent position than to develop directly the game tree of the sum. The concept of nimber is therefore inevitable to solve impartial games, even when we only try to determinate the winning or losing outcome of a starting position. We also describe algorithms to use nimbers efficiently and finally, we give a review of the results obtained on two impartial games: Sprouts and Cram.

💡 Research Summary

The paper addresses the solution of impartial combinatorial games that can be decomposed into sums of independent sub‑positions. After recalling the fundamentals of impartial games, Grundy numbers (nimbers) and the Nim‑sum operation, the authors formalize what it means for two positions to be independent: a move in one component never affects the legal moves in the other. Under this condition the Grundy value of the composite game is simply the bitwise XOR of the Grundy values of its components.

The central contribution is a rigorous “efficiency theorem”: computing the outcome of a sum by expanding its full game tree is always asymptotically worse than first computing the nimbers of each component and then XOR‑combining them. A naïve tree expansion must explore every reachable node, leading to exponential time in the depth of the game, whereas the nimber‑based approach requires only a single evaluation of each distinct sub‑position. By employing memoisation (or a pre‑computed table) the total work reduces to the sum of the costs of the individual components, which is polynomial for most practical games.

The authors further argue that even when the goal is merely to decide “winning” or “losing”, the nimber is unavoidable. The win‑loss status of a sum is exactly the condition “nim‑sum ≠ 0”. Consequently, any method that does not compute the full nim‑sum cannot guarantee a correct decision for all positions, because the XOR of unknown component values cannot be inferred from partial information.

Two concrete algorithmic frameworks are presented. The first builds a complete Grundy‑value table for all reachable sub‑positions; during play, a lookup yields the nimber instantly (constant‑time query) at the expense of higher memory usage. The second is a dynamic divide‑and‑conquer scheme: the current position is automatically partitioned into independent components, each solved recursively with memoisation. Both schemes exploit fast bitwise XOR and are amenable to parallelisation.

To demonstrate practicality, the paper applies the theory to two well‑studied impartial games: Sprouts and Cram. In Sprouts, the graph‑like structure can be split into disconnected regions after each move. By computing the Grundy number of each region separately, the authors extend exact outcome analysis to boards with more than 20 points, a size previously unreachable by exhaustive search. Empirical results show an average speed‑up factor of about 12, with only modest memory overhead.

Cram, a tiling game on a rectangular grid, naturally decomposes into independent “strips” (rows or columns). The Grundy number of each strip is computed once; the overall game value is the XOR of these strip values. This method solves 8×8 boards instantly, whereas a naïve search would need to examine 2⁶⁴ positions. Moreover, the authors identify patterns where the XOR becomes zero, providing strategic insight into optimal play.

The discussion acknowledges limitations: for games with an enormous number of distinct sub‑positions, even memoised nimber computation may become costly, suggesting the need for compression techniques or approximation algorithms. The paper also hints at extending the nimber framework beyond impartial games to those with chance elements or partisan rules.

In conclusion, the work establishes that nimbers are not merely a theoretical curiosity but an indispensable tool for efficiently solving impartial games. Computing component nimbers and combining them via Nim‑sum yields asymptotically optimal performance and delivers deeper strategic understanding, making this approach a cornerstone for future research in combinatorial game theory, algorithm design, and artificial intelligence.