Author: Ricardo Gibert
Date: 23:07:37 10/07/00
Go up one level in this thread
On October 08, 2000 at 02:00:25, Dave Gomboc wrote: >On October 08, 2000 at 01:37:20, Ricardo Gibert wrote: > >>On October 08, 2000 at 01:27:15, Dave Gomboc wrote: >> >>>On October 07, 2000 at 22:29:44, Ricardo Gibert wrote: >>> >>>>On October 07, 2000 at 19:32:18, Vincent Diepeveen wrote: >>>> >>>>>On October 07, 2000 at 15:16:17, Peter McKenzie wrote: >>>>> >>>>>>The position where the solution is most in doubt: >>>>>> >>>>>>[D]2k2K2/8/pp6/2p5/2P5/PP6/8/8 w - - >>>>>> >>>>>>The authors say that after 1.Ke8 Kc7 2.Ke7, black draws by 2...b5 with a >>>>>>stalemate motif after 3.Ke6 b4 4.a4 Kb6. >>>>> >>>>>nice motif for humans. peanut for the computer to see. >>>>>however, after 1.a4 i get a 0.00 score from diep initially, >>>>>now there are sure some bugs in this version, as i'm busy rewriting >>>>>its hashing to 64 bits (which asks for bugs of course), but 0.00 is >>>>>pretty hard. it is basically doubting between 0.50 and 0.00 on most >>>>>plies. where the stalemate position is 50 moves of shuffling around >>>>>with king and score +1.31. this version not showing +3.x scores weirdly. >>>>> >>>>>what is the win with a4 which i'm missing? and 20 plies of search too, >>>>>which is hard to believe in this position. with all 3 vs 2 egtb attached >>>>>and 20 plies of search with the white king already penetrated i either >>>>>expect to see +mateXX or see many pawns go or see draw score if it's >>>>>a draw. Now i get a slight draw score depending upon depth it is 0.00 or >>>>>+0.50 for white. That's not very convincing. >>>>> >>>>>>Of course white can vary, and they quote: 4.axb4 cxb4 5.Kd5 a5 6.Kc5 a4= >>>>>> >>>>>>Or: 2.a4 b5 which is supposed to be drawn too. >>>>> >>>>>>I haven't checked these lines thoroughly, but quickly playing some of them vs my >>>>>>program suggests they are probably correct. Certainly its possible there is a >>>>>>mistake though. >>>>>> >>>>>>The other controversial positon: >>>>>> >>>>>>[D]8/1k6/p4p2/2p2P2/p1P2P2/2P5/P1K5/8 w - - >>>>>> >>>>>>Kc1 is analysed using the 'theory of corresponding squares', something I don't >>>>>>really understand :-) I haven't analysed this one at all, I will just quote the >>>>>>main variation: >>>>> >>>>>i have wasted a full evening to go to a meeting where the writers >>>>>about the 'corresponding square' theory were there. >>>>> >>>>>it's all big nonsense. the problem is to figure out what the corresponding >>>>>squares are. it's like saying: "find best move M and play perfect >>>>>chess". Now the problem is to find move M. So is the problem to >>>>>find the corresponding squares. There is no algorithm for it at all. >>>>>After wasting hours of talk to the authors who themselves aren't strong >>>>>chessplayers at all, they couldn't give any algorithm for it, and it >>>>>all appeared to come down to how well you can define squares as being >>>>>the corresponding square! >>>>> >>>>>>1.Kc1! Kc7 2.Kd1! Kd7 3.Ke1 Kc7 4.Kf2 Kd8 5.Ke2 Ke8 6.Kd3 Kd7 7.Ke3 Kd6 8.Ke4 >>>>>>"(forcing the pawn to advance)" a3 9.Kd3 a5 10.Kc2! a4 "The posiiton on the >>>>>>Q-side is blocked; a quadratic system with non-ambiguous rear (711) now >>>>>>operates." Go figure! 11.Kc2! Ke7 12.Kd3 Kc6 13.Ke2 Kd6 14.Kf2 Kd7 15.Ke3 Ke7 >>>>>>16.Kf3 and wins >>>>> >>>>>that's 16 moves. I'm searching 40 plies. that's 20 moves. So i see >>>>>4 moves deeper as this. Also i have made afew moves and then also searched >>>>>40 plies. that's like 23 ply in the diagram position. So obviously this >>>>>trick isn't the problem here! >>>>> >>>>>>I didn't play thru. that variation, but clearly its at least 31ply and white >>>>>>hasn't even captured a pawn yet! Let me see, finished with white K on f3, so it >>>>>>needs another 3 moves at the very least to capture c5 so this problem looks like >>>>>>it is at the VERY least 34ply deep and probably more. >>>>> >>>>>I searched if i count the moves made first with it 44 plies or something >>>>>and don't see a win at all. >>>>> >>>>>>cheers, >>>>>>Peter >>>> >>>>If you are really interested in the theory of corresponding squares, you may >>>>want to have a look at the book, "The Final Countdown" by Willem Hajenius & >>>>Herman van Riemsdijk. >>>> >>>>As for translating it into a computer comprehensible algorithm, I don't think it >>>>is practical. >>> >>>Your opinion notwithstanding, Murray Campbell did it over ten years ago. :-) >>> >>>Dave >> >>Very interesting. Is there a download article on this? I'm curious if this a >>practical algorithm, why it is not being used in programs. Does it include >>positions where pawns can move forward? Or only positions where they are >>blocked? > >In fact, there's an entire Ph.D. thesis on it. :-) I imagine it isn't being >used in (at least most) programs because it's not perfect either, and the >brute-force method worked somewhat better overall than it did. However, it did >handle some positions much better than a brute forcer. The pawns didn't have to >be blocked, but of course that makes it easier (mind you, that makes it easier >for a brute force program too!) > >Dave Then I seem to have guessed right about its practicality.
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.