Author: Ricardo Gibert
Date: 03:47:04 08/05/00
Go up one level in this thread
On August 05, 2000 at 05:32:16, Colin Frayn wrote: >This position came up in a game I just played between ColChess (Development >version) and Sjeng v7.1. > >[D]8/R7/8/8/7P/6P1/3k2K1/2q5 b - - > >ColChess, as Black, has to move, having just promoted a pawn to a queen, and >then white played a rook move (Ra5-a7). > >This looks like it should be an easy win for black, but in the game ColChess >just moved its pieces around aimlessly and drew by repetition. This annoyed me >quite a bit so I tried this position at longer time controls and sure enough it >still couldn't see anything. Then I tried it with Crafty and still nothing. > >Is this one of those positions where it's drawn and humans (if they're better >than me) can see it easily but a computer can't? If so then I'd better fix >ColChess' evaluation function before release so that it doesn't exchange off >(from a stronger position it sac'd its rook to promote the pawn) In that case, I would guess it erred, since instead of Ra5-a7, the move Ra2+ followed by Rf2 draws easily (at least). White can set up its Rook on f2, K on h2 and the h-pawn on h3 when he would have an impregnable fortress. The white Rook would then shuttle between f2 and any unguarded square on the f-file, then back again to draw. You may have to resign yourself to the very distinct possibility that this is not really fixable. "Fixing" such things can do more harm than good. Computers are not good at recognizing fortresses. On the other hand, there may be another reason why your program should not have played into this line that _is_ fixable. Why not post the position before the sac and see how other programs handle it? As for the position you actually posted, my feeling is it should be drawn, but do not ask me for proof. If you look up in a endgame book, you may find that this type of position is drawn even without the h-pawn, since the remaining g-pawn is probably close enough to the edge of the board for a draw. > >I can (kind of) see why it might be drawn, but I'd appreciate input either way >(maybe from those of you with 5-man TBs.) > >Cheers, >Colin > >P.S. New version of ColChess is nearing completion. Still unsure whether it is >actually stronger than the previous version but it's slightly faster and has >some funky new features and a lot of code improvements and bugfixes.
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.