It's just a terminological thing. There are "safe cutoffs" – like usual alpha-beta cutoffs. And "unsafe" ones – futility pruning, razoring, null-move pruning, forward pruning and so on. For example, most of the modern engines going to cut-off all "quiet" moves at the bottom part of the search tree in the case if we're already lost some material. If the opponent won a queen – it's usually worthless to move some pieces w/o checks, captures and promotions – everything is probably already lost, it's better to skip this subtrees to save time for the more perspective ones.Sven Schüle wrote: Can you explain what you mean by "a cutoff changes the best move at root", "a cutoff affects PV", and "a cutoff results in nothing"? Which kind of cutoffs are you thinking of - most probably not the usual "beta cutoffs"?
In my view, performing a cutoff *never* changes the best move at root or the PV. That is somehow inherent to a cutoff: you save searching more nodes based on the observation that searching these nodes would *not* have any effect on the PV, best move and evaluation at root.
Maybe we only have a different understanding of certain terms?
Sven
Of course non-safe cutoffs are able to change root eval/move – if you will cut some move that should be in PV for exhaustive search. Or the refutation for non-exhaustive search PV.