Author: Slater Wold
Date: 21:55:34 07/24/02
Go up one level in this thread
On July 24, 2002 at 23:55:58, John Merlino wrote: >On July 24, 2002 at 23:34:17, Slater Wold wrote: > >>[D]1N4br/3pkPp1/6PP/8/pp3P2/8/P4r1P/6K1 w - - 0 1' >> >>Every program I tested found the correct move in about 5 seconds. >> >>Every program I tested then failed low for 10-13 ply. >> >> >>The real test here is not "finding" the move. But finding that it's winning. >> >>How long does it take your program to find hxg7, and _know_ it's winning? > >Chessmaster 9000 (PIII-600) finds the move instantly, and DOES fail low at ply >10. After that, the eval stays steady through to ply 13. > >Time Depth Score Positions Moves >0:00 1/3 3.32 1973 1.hxg7 Rhxh2 2.f8=Q+ Ke6 3.Qxg8+ > Kd6 >0:00 1/4 4.70 9531 1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 > Bxf7 4.gxf7 Kxf7 >0:00 1/5 4.40 16406 1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 > Bxf7 4.gxf7 Kxf7 5.Kg3 Kxg7 6.Nxd7 >0:01 1/6 4.40 74263 1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 > Bxf7 4.gxf7 Kxf7 5.Kg3 Kxg7 6.Nxd7 >0:01 1/7 4.39 87914 1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 > Bxf7 4.gxf7 Kxf7 5.Nxd7 Kxg7 6.Kg3 > Kg6 >0:02 2/8 4.42 124497 1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 > Bxf7 4.gxf7 Kxf7 5.Nxd7 Kxg7 6.Kg3 > b3 >0:02 3/9 3.80 203060 1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 > Bxf7 4.gxf7 Kxf7 5.Nxd7 b3 6.Nf6 > Kxg7 7.Ne8+ Kg6 >0:05 4/10 2.01 422430 1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 > Bxf7 4.gxf7 Kxf7 5.Nxd7 b3 6.axb3 > axb3 7.Ne5+ Kxg7 8.Nc4 Kf6 >0:33 5/11 2.00 2574161 1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 > Bxf7 4.gxf7 Kxf7 5.Nxd7 b3 6.axb3 > axb3 7.Ne5+ Kxg7 8.Nc4 Kf6 9.Kg3 >1:05 6/12 2.07 5089526 1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 > Bxf7 4.gxf7 Kxf7 5.Nxd7 b3 6.axb3 > axb3 7.Ne5+ Kxg7 8.Nc4 Kf6 9.Kg3 > Ke6 >3:04 7/13 2.03 14522913 1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 > Bxf7 4.gxf7 Kxf7 5.Nxd7 b3 6.axb3 > axb3 7.Ne5+ Kxg7 8.Nc4 Kf6 9.Kg3 > Ke6 10.Nb2 > >HOWEVER, hxg7 still appears to lose anyway! After forcing CM9000's main line of >1.hxg7 Rg2+ 2.Kxg2 Rxh2+ 3.Kxh2 Bxf7 4.gxf7 Kxf7 (and you should agree that all >of White's moves are forced), we get this position: > >[D]1N6/3p1kP1/8/8/pp3P2/8/P6K/8 w - - 0 5 > >And Chessmaster says this: > >Time Depth Score Positions Moves >0:00 1/3 3.06 356 5.Kg3 b3 6.Nxd7 bxa2 >0:00 1/3 4.38 392 5.Nxd7 Kxg7 6.Kg3 Kg6 >0:00 1/4 4.41 945 5.Nxd7 Kxg7 6.Kg3 b3 >0:00 1/5 3.77 4520 5.Nxd7 b3 6.Nf6 Kxg7 7.Ne8+ Kg6 >0:00 2/6 2.00 11934 5.Nxd7 b3 6.axb3 axb3 7.Ne5+ Kxg7 > 8.Nc4 Kf6 >0:00 3/7 1.99 25482 5.Nxd7 b3 6.axb3 axb3 7.Ne5+ Kxg7 > 8.Nc4 Kf6 9.Kg3 >0:00 4/8 2.06 50786 5.Nxd7 b3 6.axb3 axb3 7.Ne5+ Kxg7 > 8.Nc4 Kf6 9.Kg3 Ke6 >0:00 5/9 2.01 99429 5.Nxd7 b3 6.axb3 axb3 7.Ne5+ Kxg7 > 8.Nc4 Kf6 9.Kg3 Ke6 10.Nb2 >0:02 6/10 -2.43 204435 5.Nxd7 b3 6.axb3 a3 7.b4 a2 8.b5 > a1=Q 9.b6 Qxg7 10.Kh3 Ke6 >0:04 7/11 -3.32 479598 5.Nxd7 b3 6.axb3 a3 7.Nf6 Kxg7 > 8.Ne8+ Kf8 9.Nd6 a2 10.b4 a1=Q > 11.b5 Qd4 >0:07 8/12 -3.73 740137 5.Nxd7 b3 6.axb3 a3 7.Nf6 Kxg7 > 8.Ne8+ Kf8 9.Nc7 a2 10.Ne6+ Ke7 > 11.Nd4 a1=Q 12.Nc6+ Kd6 13.Ne5 >0:13 9/13 -4.20 1263381 5.Nxd7 b3 6.axb3 a3 7.Nf6 Kxg7 > 8.Ne8+ Kf8 9.Nf6 a2 10.Nh7+ Kg7 > 11.Ng5 a1=Q 12.Ne6+ Kf6 13.Nc5 Qd4 >0:24 10/14 -4.44 2318389 5.Nxd7 b3 6.axb3 a3 7.Nf6 Kxg7 > 8.Nh5+ Kf8 9.f5 a2 10.Nf4 a1=Q > 11.Kh3 Qc3+ 12.Kg4 Kf7 13.Ne6 Qxb3 >0:47 11/15 -4.52 4484987 5.Nxd7 b3 6.axb3 a3 7.g8=Q+ Kxg8 > 8.Kh3 a2 9.Kg4 a1=Q 10.Ne5 Qd1+ > 11.Kg5 Qxb3 12.f5 Qe3+ 13.Kf6 Qh6+ > 14.Ng6 Qg7+ 15.Ke6 Qf7+ 16.Ke5 >1:09 12/16 -4.60 6779430 5.Nxd7 b3 6.axb3 a3 7.g8=Q+ Kxg8 > 8.Kh3 a2 9.Kg4 a1=Q 10.Ne5 Qd1+ > 11.Kg5 Qxb3 12.f5 Qd5 13.Ng6 Qd2+ > 14.Kf6 Qd6+ 15.Kg5 Kg7 > >Is CM9000 missing something? > >jm Yes, yes, there is. It's called about 80 ply. Ok, here's the situation: When black comes down with it's rooks, which are pretty much forced moves, *every* programs natural reaction is to take the rooks. Why? Because they are free rooks! However, if you DO take both rooks, you lose. Here is how the game goes, all moves are forced: 1.hxg7 Rg2+ 2.Kf1 Rf2+ 3.Ke1 Re2+ 4.Kd1 Rd2+ 5.Kc1 Rc2+ 6.Kb1 Rb2+ 7.Ka1 Rxa2+ 8.Kb1 Rb2+ 9.Kc1 Rc2+ 10.Kd1 Rd2+ 11.Ke1 Re2+ 12.Kf1 Rf2+ 13.Kg1 Rg2+ 14.Kxg2 Rxh2+ 15.Kxh2 Bxf7 16.gxf7 Kxf7 17.Nxd7 a3 18.Nc5 a2 19.Nb3 Kxg7 20.Kg3 Kf6 21.Kf3 Kf5 22.Na1 Ke6 23.Ke4 Kf6 24.f5 Kf7 25.Ke5 Kg7 26.Ke6 Kg8 27.Ke7 Kg7 28.f6+ Kg6 29.f7 Kf5 30.f8Q+ Ke5 31.Qb8+ Kd4 32.Qxb4+ Ke3 33.Qb2 Kd3 34.Qf2 Kc3 35.Kd6 Kc4 36.Kc6 Kc3 37.Kc5 Kd3 38.Kd5 Kc3 39.Qd4# 1-0 So perhaps the REAL problem here is not the original position, rather this position: [D]1N4br/3pkPP1/6P1/8/pp3P2/8/P5rP/6K1 w - - 0 2 However, in order to solve this, you have to attain 2 things: 1.) Taking the rook would be BAD! Most programs will see that, eventually. However they are unable to see it in the original position. 2.) That if you race down to the other side of the board, let the rook take the pawn, and then race back, you will be able to stop the advancing pawns on a and b, as well as win the pawn race on the other side of the board. #1 is obtainable. #2 is not. Most programs I have tested, if given enough time will eventually realize that taking the rook will result in a promotion of the a or b pawns. So they go with Kf1 however they give a stupid line that kind of goes no where. I thought it was an interesting position anyway. ;)
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.