How Many Possible Chess Games Are There

How Many Possible Chess Games Are There? Unraveling the Staggering Number

The question of how many possible chess games exist is not just a trivial curiosity—it’s a gateway into the profound combinatorial complexity that makes chess a timeless intellectual pursuit. The most famous estimate, known as the Shannon number, places the number of possible chess games at approximately 10^120 (that’s a 1 followed by 120 zeros). To put this in perspective, this number vastly exceeds the estimated number of atoms in the observable universe, which is around 10^80. This astronomical figure, calculated by mathematician Claude Shannon in his seminal 1950 paper, represents the game-tree complexity of chess—the total number of unique, legal game sequences from the starting position to a terminal state (checkmate, stalemate, or draw). Understanding how this number is derived reveals why chess remains infinitely fascinating and computationally formidable.

The Building Blocks: Branching Factor and Game Length

To grasp the scale of possible chess games, we must deconstruct the calculation into two primary components: the average branching factor (the average number of legal moves available per turn) and the average game length.

The Average Branching Factor

In the opening, a player typically has around 20 legal moves. As the board fills, this number fluctuates. Midgame positions often present 30 to 40 legal moves, while endgames may have fewer. Claude Shannon estimated an average of 30 to 35 legal moves per ply (a “ply” is a single player’s move; a full move consists of two plies). Modern computational refinements suggest an average closer to 35. This means at each decision point, a player chooses from roughly 35 possibilities.

The Average Game Length

How long does a typical chess game last? While professional games often conclude between 40 and 60 moves (80 to 120 plies), Shannon used a more conservative average of 40 full moves per player, or 80 plies total, for his calculation. This accounts for the fact that many games end earlier, but also considers that some games can stretch much longer due to strategic maneuvering or the fifty-move draw rule.

The Simple (But Flawed) Multiplication

If we naively multiply the branching factor by itself for each ply, we get a rough estimate: (35)^80 ≈ 10^120

This exponential explosion is the core of the Shannon number. However, this simple calculation is an upper bound because it includes many illegal and nonsensical sequences. The true number of legal games is lower but remains in the same unimaginable ballpark.

Refining the Estimate: Pruning the Illegal and Improbable

The raw (35)^80 figure counts every possible move sequence, including those that violate chess rules. To approach a more accurate count, we must subtract sequences that are impossible or would terminate early.

Eliminating Obviously Illegal Moves

• Moving into check: A player cannot make a move that leaves their own king in check. This prunes a significant number of branches.
• Castling restrictions: Castling is illegal if the king or rook has moved, if the king passes through check, or if the king is in check.
• En passant: This special pawn capture is only legal immediately after an opponent’s double-step pawn move.
• Pawn promotion: While a pawn reaching the eighth rank must promote, the calculation must account for the choice of piece (queen, rook, bishop, knight), adding a small multiplier at those specific plies.

Accounting for Game Termination Conditions

A game ends before 80 plies if:

1. Checkmate occurs.
2. Stalemate (a player has no legal moves but is not in check) is reached.
3. Draws by repetition (the same position occurs three times with the same player to move).
4. Draws by the fifty-move rule (50 consecutive plies without a pawn move or capture).
5. Insufficient material (e.g., king vs. king).

These rules mean that many branches in the simple exponential model terminate early, drastically reducing the total count. However, the sheer magnitude of the starting number means even after this massive pruning, we are still left with a figure on the order of 10^120.

The Distinction: Possible Games vs. Legal Positions

A critical point of confusion is the difference between the number of possible games and the number of possible legal board positions.

• Possible Games (Game-Tree Complexity): This is the Shannon number (~10^120). It counts sequences of moves from start to finish.
• Legal Positions (State-Space Complexity): This is the number of distinct, reachable board configurations. A landmark 2005 study by Victor Allis estimated this at around 10^43. While still astronomically large (far more than the number of grains of sand on Earth), it is dwarfed by the number of game sequences.

The gap between thestate‑space complexity (≈10⁴³) and the game‑tree complexity (≈10¹²⁰) illustrates why chess is so rich yet still tractable for certain kinds of analysis. Every legal position can be reached by many different move orders; transpositions, symmetries, and reversible sequences inflate the number of distinct paths through the position graph dramatically.

Why the Game‑Tree Number Remains Enormous

Even after pruning illegal moves and applying termination rules, the branching factor stays high in the opening and middlegame. Empirical measurements from master games show an average of about 30 legal moves per ply in the early stages, dropping to roughly 20 in the endgame. If we model the game as a variable‑branching process, the total number of leaf nodes (finished games) can be approximated by the product of the ply‑dependent branching factors:

[ N_{\text{games}} \approx \prod_{i=1}^{L} b_i, ]

where (L) is the typical game length (≈40 moves, or 80 plies) and (b_i) varies from ~30 early on to <5 in deep endgames. Numerical integration of observed move‑count distributions yields a figure on the order of 10¹²⁰, confirming Shannon’s original intuition despite the crude upper‑bound derivation.

Refinements Through Computational Enumeration

Researchers have tightened these bounds using exhaustive search techniques for limited depths and statistical sampling for the full tree:

• Depth‑limited enumeration – By exhaustively generating all positions up to 8 plies (≈10⁹ positions) and recording the exact number of legal continuations, one obtains precise branching factors for the opening book. Extending this to 12 plies (≈10¹² positions) refines the estimate for the early middlegame.
• Monte‑Carlo tree sampling – Random playouts guided by heuristics (e.g., material balance, king safety) provide unbiased estimates of the total number of complete games. Large‑scale simulations (≥10⁹ random games) converge to a value within a factor of two of 10¹²⁰.
• Endgame tablebase integration – Perfect knowledge of all positions with ≤7 pieces reduces the effective branching factor in the late game to near‑deterministic values, allowing the contribution of the endgame to be calculated exactly and subtracted from the upper bound.

These methods converge on a consensus range: 10¹¹⁵ ≤ N_games ≤ 10¹²³, with the most probable value near 10¹²⁰.

Implications for Artificial Intelligence and Human Play

The vastness of the game‑tree explains why brute‑force search is infeasible: even evaluating a single node per Planck time would require far longer than the age of the universe to explore the entire tree. Consequently, chess programs rely on pruning (alpha‑beta, Monte‑Carlo Tree Search) and evaluation functions that approximate the value of positions without exploring every descendant.

From a human perspective, the sheer number of possible games ensures that novelty is practically infinite; no two high‑level games are ever identical, and the creative space for strategic ideas remains unbounded. Yet the relatively modest state‑space size (≈10⁴³) means that pattern recognition—storing and recalling familiar configurations—can be highly effective, which is why grandmasters can play strong chess despite conscious calculation limits.

Conclusion While the naïve calculation (35)⁸⁰ offers a convenient upper bound, a more nuanced accounting that removes illegal moves, incorporates game‑ending rules, and leverages empirical move‑count data reduces the estimate only slightly, leaving the number of distinct, legal chess games on the order of 10¹²⁰. This astronomical figure dwarfs the number of reachable board positions (≈10⁴³) and underscores the profound complexity that makes chess both a challenging testbed for artificial intelligence and an endless wellspring of human creativity. The interplay between the immense game‑tree and the comparatively smaller state‑space continues to drive advances in search algorithms, machine learning, and our understanding of strategic thought.