Ralph Stoesser wrote:Gerd Isenberg wrote:
But isn't NM fails high but verification fails low the definition of zugzwang?
Yes, these are zugzwang positions by definition, within the (reduced) search horizon.
Maybe a few position examples from Marco could clear up the mystery..?
As a chess player the definition of zugzwang is different for me and it is not dependent on the search of chess programs.
A zugzwang is a position when
not moving is better than moving.
The result of the search of chess program is clearly irrelevant for the definition.
The program may score not moving at reduced depth as better because it does not search deep enough but it does not mean that the position is zugzwang.
Even without that problem the program may score not moving as better than moving because of not seeing deep enough and it may be possible that in some position every move reduce the positional score if you search to depth 1 but it is not a zugzwang.
I think that the right solution is to detect positions as "probably zugzwang" only when the computer detects losing at least a full pawn(or maybe at least 1/2 pawn but something significant) by moving relative to not moving when you search to the same reduced depth and hopefully in this case zugzwang detection is going to detect real zugzwangs in most cases of positive detection.
Hopefully if the search is reduced enough(so Crafty is not more than 1% slower in nodes per second by correct zugzwang detection) Bob is going to find a small improvement in playing strength of less than 5 elo by correct zugzwang detection.
Uri