D Sceviour wrote:I tried that and it is not quite correct. First, the new move trying to refute the PV may not be in the PV line. That is, the new move may be another one in the root move list.
True, sometimes the PV refutation cannot be repaired in the current PV node, and has to be repared closer to the root. Sometimes it cannot be repaired before it reaches the root, or even in the root itself. But that doesn't mean the current node isn't the best place to try to repair it. Of all the nodes between the root and itself it is the node with the shallowest side branches, and when you can get a move there that was close to the originalscore of the PV move that was now refuted, you are back to 'business as usual' in the still lower levels.
Second, PV nodes are exact scores and have already been searched full width. It is better to concentrate on fail high hash entries because these are the move lists that have not been searched full width.
They have all been searched to nearly the required depth, but not with a relevant window. They could all have failed low against the (now refuted) high score of the hash move. And that would mean you have no idea which of them were good and which truly stink. Nevertheless it seems very dangerous to restart at lower depth just for that. Because you might do it again and again. If the PV hash move scored +200 at d=7, and now that you are at d=8 scores only -400, many other moves might run into the same deep tactics that eventually refuted the PV. So you would re-search all alternative moves up to d=7 to find the second best there, which might be +190, and then at d=8, boom, it drops to -410. You could then launch a new IID search to eventually get the 3rd-best move at d=7, see that fail low at d=8 too, and so on, and so on...
It is a very tricky problem. There is a definite possibility that all moves look good at d=7 (say between 0 and +200), and most of them then are dramatically refuted at d=8 (because of the same weakness they failed to repair), except the very few that address the weakness. Then you would always plunge in at d=8 blindly, with very little chance of success. IID won't help, and is thus just a waste of time. That is the worst case, however, and there will also be cases where the d=8 refutation could only be masked at d=7 because of a unique series of delaying tactics, and for any other sequence of moves the axe would fall already at d=7.
I guess this is where aspiration can prove its worth.