Late move reductions

Late move reductions, abbreviated LMR, exploit move order. A chess engine uses heuristics, meaning practical rules and accumulated search statistics, to examine apparently promising choices first.

Basic sequence

1. The engine searches better-ordered moves first and establishes a provisional reference.
2. A quiet move, normally a move that neither captures nor gives check, much later in the list is initially tested at reduced depth.
3. If that test returns an unexpectedly strong result and crosses the score bound used by the search, the branch is searched again at greater depth.

Reduction does not mean deletion. The deeper verification matters because a genuinely strong move may have been ordered poorly. The reduction size can depend on remaining depth, the move's place in the list, whether it gives check or captures, whether it belongs to the principal variation, the line currently considered best, and the recorded history of similar moves.

Modern implementations tune many of these conditions. Excessively aggressive reduction saves time but can miss a strong quiet move or worsen errors related to the horizon effect.