Yes the score is expected to be within the alpha beta window at the root. How is that different from PV nodes in general?Henk wrote:In the root node you certainly don't have a fail high or fail low so you have exact values for all moves unless there is a check mate. But if you use LMR at the root the values are computed at different levels.lucasart wrote:Complety agree. I never understood why people always want to treat the root node differently. I use the same search function for all kinds of nodes, and view the root node like any other PV node.bob wrote:For the record, LMR at root works, and is effective. If it is good at ply=2, why not at ply=1? Ply=1 also has "lousy moves" near the end of the list.tpetzke wrote:I don't do LMR at the root. I just do a scout search for the moves beyond the first. If it fails high, I research with an open window and only if it still increases alpha, it is a new best move.
Thomas...
At the root they do aspiration search. I don't know it that gives profit.
Futility pruning and null move makes no sense in the root node.
Null move, Futility pruning, Razoring, and generally anything used to predict a fail low or a fail high without searching anything makes no sense at PV nodes, not just the root node.
This is the theory. In practice none of that is true in the presence of aspiration windows and search instability...