A search-tree pruning technique applied to minimax-style adversarial search that keeps track of two bounds (alpha and beta) to eliminate branches that cannot influence the root decision, thereby reducing the number of nodes evaluated without changing the optimal minimax outcome when applied correctly.

Principle

Maintain and propagate lower (alpha) and upper (beta) bounds for achievable values along the tree; prune any subtree whose bound cannot improve the current best option at ancestor nodes.

Demonstration

In a two-player game tree, when exploring a node for the maximizing player with current best value alpha, if a child subtree of a minimizing opponent yields a value ≤ alpha, that subtree can be skipped because the maximizer will avoid paths leading there; this can cut large portions of the tree in practice.

Misapplication

Pruning based on incorrect or improperly initialized bounds, using alpha–beta on non-zero-sum or non-minimax problems without adaptation, or assuming it always yields linear-time speedups irrespective of move ordering.

Consequence

When used correctly, search time and memory are reduced—often dramatically with good move ordering—while preserving the exact minimax decision; however, worst-case complexity remains exponential if pruning opportunities are limited.

Reversal

Expanding every node without pruning (full minimax) inverts the technique, guaranteeing evaluation of all leaves and producing the same decision but with maximal computation and no bound-based shortcuts.

Boundary

Applies to deterministic, perfect-information, turn-based minimax problems; excludes stochastic nodes, partially observable games, and situations requiring non-minimax criteria unless adapted (e.g., expectiminimax or heuristic evaluations).

Semantic Tension

Often conflated with move-ordering heuristics and other pruning methods (e.g., null-move pruning); alpha–beta is specifically the bound-propagation pruning rule within minimax, not a heuristic ordering strategy though ordering strongly affects its effectiveness.

Synthesis

Alpha–beta pruning is the disciplined use of propagated alpha and beta bounds to safely omit parts of a minimax search tree: an exact pruning rule that trades redundant exploration for identical optimal decisions, whose practical savings depend on problem structure and move ordering.