Author: Robert Hyatt
Date: 13:25:21 03/19/98
Go up one level in this thread
On March 19, 1998 at 16:06:11, Christophe Theron wrote: >On March 19, 1998 at 11:45:39, Ernst A. Heinz wrote: > >>On March 19, 1998 at 11:11:25, Christophe Theron wrote: >> >>>On March 19, 1998 at 04:25:22, Ernst A. Heinz wrote: >>> >>>> >>>>"DarkThought" is an aggressive null move searcher and finds 1. Rd1+! >>>>in iteration #2 after just 464 nodes. >>>> >>>>It returns the mate score in iteration #8 after 4380 nodes. >>>> >>>>=Ernst= >>> >>>Well done! >>> >>>Note that what is interesting is not when the key move is found (it can >>>be found by pure luck), but when the program announces mate. >> >>No luck involved here because "DarkThought" scores the position as >= >>-1.5 >>from iteration #2 onward. >> >>> Being an "agressive" null mover, DarkThought sure has a good zugzwang >>>detection scheme. I suppose you don't just turn null move off when there >>>is only one piece left? It would be very ineffective in most endgames... >>>Especially rook endgames. >> >>As reported in my ICCA Journal 20(3) article "DarkThought" performs >>recursive null moves until one side has no more pieces, i.e., it still >>does null moves in minor piece endgames with one piece per side. It >>does not use a special Zugzwang detection scheme but instead relies on >>its extensions and high search depths to overcome Zugzwang glitches. >> >>=Ernst= > >So it should never find the solution of the given position. I am curious >to know how you explain this miracle... > >The problem with zugzwang is that it doesn't add more plies to find the >solution in this case. It is that if you fail to detect it and keep on >using null move, you simply NEVER find the solution... > > > Christophe My approach has always been obvious: with a queen or more, try null moves everywhere. When the side on move has no pieces left, don't try null move at all. Between these two bounds, I only try null move within N plies of the horizon, so that every iteration exposes one more ply to a search with no null moves... Loses some efficiency, but lets me try null moves with only a knight left on the board. If you do that normally, you get killed in the right kind of endgames, because a pawn (or two pawns) can zug a knight in a heartbeat...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.