An aspiration window is a heuristic used in pair with alpha-beta pruning in order to reduce search time for combinatorial games by supplying a window (or range) around an estimated score guess.
Alpha-beta pruning achieves its performance by using cutoffs from its original range. Aspiration windows take advantage of this by supplying a smaller initial window, which increases the amount of cutoffs and therefore efficiency.[1]
However, due to search instability, the score may not always be in the window range. This may lead to a costly re-search that can penalize performance.[1] Despite this, popular engines such as Stockfish still use aspiration windows.[2]
The guess that aspiration windows use is usually supplied by the last iteration of iterative deepening.[3]