Author: Uri Blass
Date: 06:50:06 04/17/04
Go up one level in this thread
On April 17, 2004 at 08:16:21, Joachim Rang wrote: >This position is an easy mate in 4: > >[D]6Q1/8/8/7k/8/8/3p1pp1/3Kbrrb w - - 0 1 > >Certain engines however have problems to see the mate and return instead a draw >score. So ie Crafty. > >I think it is related to nullmove, since the solution involves a zugzwang for >the black side and since it's so many material on board most engines will use >nullmove in this position. However I thought Nullmove will produce a reduction >rathern than a complete cutt-off and hence the solution should be found a few >plies later. Or did I understand the mechanism of Nullmove wrong? You understood null move wrong. The idea of null move is clearly to make programs so stupid so that they will never be able to see not only mate in 4 but also mate in 2 when zugzwang is involved. It is probably better for program not to use null move pruning in that way but programmers usually do not care about improvement of not more than 10 elo that chess programs can probably get from it. simply to say reduction is not the solution because if the reduction is a small reduction then the program may spend too much time on that reduction and if the reduction is a big reduction the program may get wrong result because of not searching deep enough and suspect zugzwangs inspite of the fact that they do not exist. I think that the best general solution may be to compare scores based on searching to the same reduced depth(in the middle game you can decide about D-5 when the remaining depth is D) and repeat the process to bigger depth in the rare cases when you suspect zugzwang based on doing it(your result in small depth may be misleading). Another practical option that is not theroetically sound is to have better conditions when not to use null move pruning(for example you can use the number of moves of the opponent that do not obviously lose material before decision if to do null move pruning). Uri
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.