Quiescence search

The search performed by a chess engine, a program that analyzes positions, must stop somewhere. Evaluating at that exact point can be misleading when an immediate capture sequence remains. A piece may appear to have been won because the program stopped after the first capture even though the opponent can recapture next. Quiescence search continues the calculation until it reaches a more settled position.

Settled does not mean that the position contains no plans or threats. It means that the evaluation is not dominated by a nearby forced sequence that would sharply change material or king safety. An implementation usually explores captures; if the king is in check, it must consider legal evasions. Some engines also include selected promotions or checks.

When the king is not in check, the engine may keep the current static evaluation as a reference without making another capture. This reference is called stand pat. It then continues only tactical moves that seem capable of improving that value. Tools such as static exchange evaluation help reject captures that are likely to lose material.

The technique reduces horizon-effect errors, where a tactical consequence lies just beyond the search limit. It is not, however, a complete exploration of every legal move. If too many moves are admitted, the search can expand dramatically; if too few are admitted, an important tactic may be missed. Filters and careful move ordering are therefore essential.