Author: Miguel A. Ballicora
Date: 08:09:29 03/03/01
Go up one level in this thread
On March 02, 2001 at 17:51:12, Dieter Buerssner wrote: >On March 02, 2001 at 14:39:10, Sune Larsson wrote: > >> 1k3b1q/pP2p1p1/P1K1P1Pp/7P/2B5/8/8/8 w - - 0 1 >> >> Horwitz,B 1885 >> >> This one might prove as mission impossible. Black is totally paralysed >> from start and white is winning - but how?? Maybe more amusing for >> humans than for silicons. One could characterize black's position >> as slightly passive ;) but how could white make progress. Looks hard... >> >> Test: Win it! Draw scores like 0.00 are not trusted... > >Nullmove seems not to be a good idea for this one - so I disabled nullmove in >Yace. While Yace does some Zugzwang safeguards for nullmove, it seems, that the >depth needed to find the solution with nullmoves turned on would be too high. > >I also get strange fail highs and fail lows. Perhaps one reason is, that I >cannot detect stalemate inside quiescence search. Exactly! My program seemed to find the solution... but for the wrong reason! There are some positions where the quiescent search ends in a stalemate. Those positions, before entering the quies search hasd a positive score (after taking the queen), and got stuck in the hashtable giving the ilustion to the search that there is win. So, Gaviota found the solution because black gave up the queen thinking that the position in the hashtable was worse. 8 first plies where correct, then black played Qxf7 at the wrong moment. When I played out the position, Gaviota could not win even though it found the first part of the winning line! It kept repeating positions because at that time it saw the stalemate. When I added detection of stalemate to the quies search Gaviota was not fooled by this problem, but as a consequence found no solution. Does anybody detect stalemate in the quiescent search? I did a quick implementation but it is very expensive... Very interesting position! Miguel
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.