More Chess Games Than Atoms in the Universe: The Mathematics of Infinite Complexity

Numbers That Make Atoms Look Trivial

The observable universe contains approximately 10^80 atoms — a 1 followed by 80 zeros. This is an incomprehensibly large number by everyday standards. Yet the number of possible distinct chess games is estimated at 10^120 — a 1 followed by 120 zeros, a number roughly one trillion trillion trillion trillion times larger than the atom count.

This estimate — known as the Shannon number, after mathematician and information theorist Claude Shannon who calculated it in 1950 — is based on an average game lasting about 40 moves per player, with an average of about 30 possible moves available at each turn. The combinatorial math compounds rapidly: 30 possible moves for white on move 1, 30 for black in response, 30 for white on move 2, and so on, giving approximately 30^80 possible games before considering the exact branching factor at each position.

The actual number is larger still because the branching factor varies — some positions offer many more than 30 legal moves — and because the definition of "distinct games" allows for transpositions (reaching the same position by different move orders), which multiply the game tree further.

Why Computers Still Find Chess Hard

The sheer size of the game tree explains why brute-force search — simply calculating every possible outcome — is physically impossible even with the most powerful computers. If every computer ever built had been calculating chess moves since the Big Bang, they would not have reached a meaningful fraction of the game tree. Chess is not solved in the mathematical sense, and it may never be, because "solving" it would require evaluating positions at a depth and breadth that no physical system can achieve.

Modern chess engines like Stockfish and Leela Chess Zero dominate human players not by exhaustively searching the game tree but by combining sophisticated position evaluation functions with selective search strategies (looking further down promising lines and abandoning unpromising ones quickly) and, in the case of neural network-based engines, pattern recognition learned from millions of analyzed games.

The best chess engines play at a level human players cannot match, but they still operate by approximation and heuristic rather than by complete knowledge of the game. The game is simply too large for complete knowledge to be achievable.

The Difference Between Chess and Go

Chess's 10^120 possible games is enormous, but it is dwarfed by Go, the East Asian board game played on a 19x19 grid. The number of possible Go games is estimated at approximately 10^360 — a number so much larger than the chess game tree that direct comparison becomes meaningless. This is why Go was considered a harder problem for artificial intelligence, and why it took significantly longer for computer programs to match the best human players.

AlphaGo's defeat of world champion Lee Sedol in 2016 was a landmark in AI history precisely because Go's game tree is so vast that earlier search-and-evaluation approaches could not handle it — the breakthrough required training a neural network on millions of Go games and using it to guide search through pattern recognition rather than explicit calculation.

Shannon's Insight and Its Implications

Claude Shannon's 1950 paper "Programming a Computer for Playing Chess" is not only a foundational document of computer chess but of artificial intelligence more broadly. Shannon understood that the chess game tree was too large for exhaustive search and proposed the framework of position evaluation and selective search that still underlies modern chess engines. His insight — that intelligence might be approximated by the right combination of pattern recognition and targeted search, even in domains of vast complexity — prefigured decades of subsequent AI development.

The fact that chess, a game played on an 8x8 board with 32 pieces following simple rules, generates a game space larger than the observable universe's atomic content is a useful reminder that complexity is not always where we expect to find it.