Author: Majd Al-Ansari
Date: 23:02:23 01/09/06
Go up one level in this thread
Thank you for your excellent insight and great examples. I too remember that
Topalov vs. Moro game. You are right maybe I am wishing too much for Rybka in
some positions, but Rybka seems to play the "intuitively" correct moves so much
in the middle game that I think if the same type of knowledge based system is
used in the ending, it will bring Rybka to new heights. Although I did not
analyze totally the game where it plays Kh5 to see if it can save the draw, I
was thinking more of how a "human" would play it by intuition. I believe Rybka
has some incredible knowledge in the opening stage and the middle game ... I
mean really INCREDIBLE, and hopefully that same type of knowledge can be ported
to the endgame.
I also agree with you totally that Rybka can play some endings better than any
other engine. I have posted some examples of fantastic endgame play in my post
"THE GOOD". If you have more interesting games with some analysis please post
them, I would be very interested in seeing them. Thanks
On January 09, 2006 at 17:05:30, Dagh Nielsen wrote:
>Hi, thanks for a great post. I have some comments for two of the positions:
>
>[D]8/5p2/6k1/pp6/3P3p/7K/PP4P1/8 b - - 0 40
>
>>40...Kh5?? Even a weak player here can figure out that trying to protect the h
>>pawn with a d runner is futile. Black's best chance (if he has one) is to play
>>Kf5 and somehow hope that after snatching the d pawn he can hold the fort by
>>trying to get his King to get the opposition. 1.37/1911 41.d5 0.76/20 7 41...Kg5
>>1.02/20 14 42.d6 0.84/20 15 42...Kf6 0.90/19 9 43.Kxh4 1.64/20 13 43...Ke6
>>1.90/19 7 44.Kg5 1.67/20 13 44...Kxd6 (b4) 2.25/20 10 45.Kf6 2.30/20 7 45...Kd5
>>2.31/19 7 46.Kxf7 2.65/20 10 46...Ke5 2.54/18 6 47.Ke7 2.30/16 7 47...b4 2.80/19
>>6 48.b3 3.64/21 7 48...Kd4 (Kf4) 11.09/22 51 49.Kf6 (Kd6) 7.57/17 6 49...Kc3
>>9.22/16 9 50.g4 7.85/20 11 50...Kb2 9.28/16 11 51.g5 8.01/16 5 1–0
>
>I agree that all the red flags should have been up before entering this ending,
>but I am no so sure 40...Kh5 is such an unreasonable move. To me, it seems like
>everything loses both the g and h pawn for the d pawn, so, for the engine, it is
>only a question of delaying and pushing the rapidly deteriorating position as
>far away in the horizon as possible. My analysis is simply Kf5 (instead of Kh5)
>Kxh4 Ke5 (or f6 Kh5) Kg5 and black loses the g-pawn right away. I guess, in
>losing positions engines will do all kinds of counter-intuitive stuff!?
>
>[D]2r3k1/1R2K1p1/6B1/7P/8/8/8/8 w - - 0 113
>
>>Here white has a completely winning position. He is a bishop up and he still has
>>a pawn that can Queen. The easiest ways to win this position is to ofcourse
>>flush out the King and try to keep him away from h8. Then you use strength of
>>the extra bishop to take the black pawn. The rest is just moving your King in
>>front and Queening the pawn. Here is an example of how to win this ending.
>>5.88/13 3 [65.Bf7 Rd6 66.Kc5 Rf6 67.Bg6 Rf1 68.Ra8+ Ke7 69.Re8+ Kd7 70.Rg8 Ke6
>>71.Rxg7 Rh1 72.Rf7 Rh4 73.Rf1 Ke5 74.Bf7 Rh3 75.Kc6 Rd3 76.Kc7 Ra3 77.Kd8 Rh3
>>78.Ke7 Ke4 79.Kf6 Ke3 80.Kg7 Rg3+ 81.Bg6 Rh3 82.h6 Rh2 83.h7 Ke2 84.Rf8+-]
>
>I will not deny that the position may be winning, but I don't think it is so
>easy for engines to figure out the right plan :-)
>
>OK, flushing out the King from h8, that means Bf7 or Kh7. Let's take your Bf7.
>One winning plan is to move white king to h7. But what happens if black simply
>keeps his rook on the 8th rank, and when white puts king on g6, black checks and
>move back to the 8th?
>
>OK, I figure your simpler plan is then Bf7-a2 and Rf7+ and Rxg7. But still,
>there are lots of checks than can "push" such variants out beyond the horizon.
>
>I don't know, figuring out the two above ideas, I see that your variation maybe
>is the "logical" winning solution, it's just that maybe there are sidelines that
>are very difficult to calculate so that the initial position can't be solved so
>easily by search alone. And I don't think it is THAT easy to add knowledge to
>the engine so that it will choose Bf7 as needed move; I faintly recall the
>Topalov-Morozevich game from San Luis, just before the time control Topalov had
>a very big advantage and had to choose between essentially two different winning
>tries, but he opted for a "safe" one that gave Black a fortress somewhat similar
>to this example, so that there were no longer no immediate threats to keep the
>pressure, and Morozevich ended up drawing by a miracle perpetual. We sat there
>on Playchess observing the game armed with our engines and said "oh-oh, did
>Topalov just allow a fortress instead of this easy win"... I mean, Topalov
>entered the fortress to avoid some practically risky line, and this seems quite
>similar to Rybka allowing Kg8 because it cannot figure out the pawn win by
>search!?
>
>>113.h6? Rybka is forced to do a pawn move to stop the 50 move draw rule. It
>>obviously has no clue how to win the won ending. Once the pawn is gone this is a
>>draw.
>
>If the engines have lost track, it is in my experience very unlikely they will
>somehow be forced to find the right track once the 50 move draw looms... :-)
>
>I firmly agree that Rybka has some endgame issues, but let me include two of my
>earliest blitz games from playchess with Rybka where she impressed me a lot in
>the ending, vs. Deep Shredder 9 and Deep Fritz 8. They both had far superior
>hardware. The interesting part is that the evaluations differed a lot at the
>last crucial moments; Rybka saw the eventual "truth" far earlier :-)
>
>
>[Event "Rated game, 3m + 0s"]
>[Site "Machines A"]
>[Date "2005.12.05"]
>[Round "?"]
>[White "Nátas, Deep Shredder 9"]
>[Black "Flying Saucers, Rybka 1.0 Beta 32-b"]
>[Result "0-1"]
>[ECO "C45"]
>[WhiteElo "2601"]
>[BlackElo "2661"]
>[PlyCount "182"]
>[EventDate "2005.12.05"]
>[TimeControl "180"]
>
>1. e4 {0} e5 {0} 2. Nf3 {0} Nc6 {0} 3. d4 {0} exd4 {0} 4. Nxd4 {0} Qh4 {0} 5.
>Nc3 {0} Bb4 {0} 6. Be2 {0} Qxe4 {0} 7. Ndb5 {0} Bxc3+ {0} 8. bxc3 {0} Kd8 {0}
>9. O-O {0} Nf6 {0} 10. Be3 {0} Re8 {0} 11. Bf3 {0.62/12 6} Qc4 {0.04/10 5} 12.
>Rb1 {0.54/13 0} Ne5 {0.07/9 4} 13. Bd4 {0.99/12 7} d6 {0.18/10 10} 14. Be2 {
>0.53/13 30} Qxa2 {0.03/8 1} 15. Ra1 {0.64/13 6} Qe6 {0.38/10 0} 16. Nxa7 {
>0.70/13 5} Rb8 {0.28/10 0} 17. Nxc8 {0.77/12 4} Kxc8 {0.39/10 4} 18. c4 {
>0.94/12 6} c5 {0.00/10 3} 19. Bb2 {0.92/12 4} Kc7 {-0.36/10 3} 20. Re1 {
>0.75/12 0} Nxc4 {-0.56/9 1} 21. Bc3 {0.94/11 4} Ne4 {-0.34/11 0} 22. Qd3 {
>0.24/11 6} f5 {-0.47/10 8} 23. Bxg7 {0.63/11 4} d5 {-0.23/9 1} 24. Qh3 {
>1.31/11 7} Qg6 {-0.21/11 0} 25. Bh6 {1.35/10 1} Ned6 {0.27/8 2} 26. Bxc4 {
>1.22/12 3} dxc4 {0.33/10 0} 27. Bf4 {1.24/12 8} Ra8 {0.37/10 3} 28. Rad1 {
>1.11/12 9} Rxe1+ {0.52/11 10} 29. Rxe1 {1.36/17 0} Kc6 {0.32/11 2} 30. Bxd6 {
>2.12/11 1} Kxd6 {0.59/11 3} 31. Qe3 {1.87/12 1} Qf6 {0.13/10 2} 32. c3 {
>1.75/12 0} Ra3 {0.44/9 3} 33. Qf4+ {1.52/11 1} Kd7 {0.12/10 4} 34. Rd1+ {
>1.86/12 9} Kc6 {0.22/11 3} 35. Qxc4 {2.16/13 0} b5 {0.71/10 8} 36. Qg8 {
>2.05/14 0} Qe5 {0.59/9 4} 37. Qxh7 {1.88/11 0} Rxc3 {0.28/8 1} 38. Qd7+ {
>1.41/10 2} Kb6 {0.01/0 0} 39. Qd8+ {1.41/10 1} Kc6 {0.53/10 3} 40. Qa8+ {
>1.66/10 1} Kb6 {0.27/9 1} 41. h4 {1.38/10 0} b4 {0.30/9 1} 42. Qd8+ {1.93/9 1}
>Qc7 {0.33/10 1} 43. Qf6+ {1.91/10 1} Kb5 {0.49/11 1} 44. Qxf5 {1.81/12 0} b3 {
>0.43/10 1} 45. h5 {2.19/10 3} b2 {0.33/10 0} 46. Qb1 {2.00/11 1} Rb3 {0.81/11 3
>} 47. h6 {2.25/12 0} Qc8 {1.21/11 20} 48. h7 {2.82/10 1} Qh8 {1.33/9 2} 49. Qg6
>{2.82/10 2} Ra3 {0.86/8 1} 50. Qg8 {2.82/11 0} Ra8 {0.94/10 1} 51. Qb3+ {
>2.82/12 0} Kc6 {0.94/10 0} 52. Qf3+ {1.82/10 5} Kb5 {0.52/11 0} 53. Qb3+ {
>2.57/9 0} Kc6 {0.00/16 1} 54. Qd5+ {1.81/10 0} Kb6 {0.44/10 4} 55. Qe6+ {
>1.81/12 0} Kb5 {0.46/8 0} 56. Qd7+ {1.81/11 0} Kc4 {0.34/9 3} 57. Qe6+ {
>1.56/8 1} Kb5 {0.00/14 1} 58. Qd7+ {1.56/11 0} Kc4 {0.00/14 0} 59. Qg4+ {
>1.09/9 1} Kb5 {0.47/9 2} 60. Qe2+ {1.09/11 0} c4 {0.05/10 4} 61. Rd5+ {
>1.55/10 1} Kb4 {0.21/10 0} 62. Qe7+ {1.80/11 0} Kb3 {0.08/10 4} 63. Rb5+ {
>1.91/10 1} Ka2 {0.00/10 0} 64. Qe4 {0.54/10 3} c3 {0.00/11 0} 65. g3 {0.66/10 1
>} Ra7 {0.00/10 1} 66. Qd5+ {0.15/11 3} Kb1 {0.00/13 0} 67. Qd1+ {0.08/11 1} Ka2
>{0.00/14 0} 68. Qb3+ {0.08/10 0} Kb1 {0.00/13 1} 69. Rb6 {0.20/9 1} Rd7 {
>-0.97/9 1} 70. Rc6 {0.07/8 1} Qxh7 {-1.03/11 0} 71. Rxc3 {0.10/11 1} Qf5 {
>-1.17/11 0} 72. Re3 {0.55/7 2} Rc7 {-0.74/11 0} 73. Kg2 {0.55/7 0} Qc8 {
>-0.79/10 2} 74. Qd3+ {0.36/7 2} Rc2 {-0.88/11 0} 75. Re1+ {0.09/7 1} Ka2 {
>-0.97/10 0} 76. Qd5+ {0.08/8 0} Qc4 {-0.94/10 1} 77. Qa8+ {0.04/8 1} Kb3 {
>-1.16/10 2} 78. Qb7+ {0.02/9 0} Qb4 {-1.21/10 2} 79. Qf3+ {0.03/9 1} Rc3 {
>-1.39/7 1} 80. Qd5+ {-0.64/8 5} Qc4 {-2.66/10 0} 81. Qb7+ {-0.89/8 2} Ka2 {
>-3.71/10 0} 82. f4 {-1.27/7 1} Rc1 {-3.77/8 1} 83. Qa7+ {-3.91/7 4} Kb1 {
>-3.53/11 0} 84. Qa5 {-3.91/8 0} Rxe1 {-8.32/7 1} 85. Qxe1+ {-1.43/5 0} Kc2 {
>-8.32/10 0} 86. Qf2+ {-2.09/6 1} Kc3 {-8.38/11 0} 87. Qe1+ {-2.09/5 0} Kb3 {
>-8.38/11 1} 88. Kh3 {-2.16/6 1} Qc8+ {-8.38/12 0} 89. Kh4 {-7.68/8 0} Qh8+ {
>-8.38/12 1} 90. Kg4 {-8.46/9 0} Qg7+ {-8.38/12 1} 91. Kf5 {-8.46/9 0} Qh7+ {
>Nátas,Deep Shredder 9 rinde -9.23/11 3} 0-1
>
>
>
>[Event "Rated game, 3m + 0s"]
>[Site "Machines A"]
>[Date "2005.12.05"]
>[Round "?"]
>[White "Flying Saucers, Rybka 1.0 Beta 32-b"]
>[Black "Firstjuni, Deep Fritz 8"]
>[Result "1-0"]
>[ECO "E03"]
>[WhiteElo "2641"]
>[BlackElo "2705"]
>[PlyCount "91"]
>[EventDate "2005.12.05"]
>[TimeControl "180"]
>
>1. d4 {0} Nf6 {0} 2. c4 {0} e6 {0} 3. g3 {0} d5 {0} 4. Bg2 {0} dxc4 {0} 5. Qa4+
>{0} Nbd7 {0} 6. Qxc4 {0} c5 {0} 7. Nf3 {0} a6 {0} 8. O-O {0} b5 {0} 9. Qd3 {0}
>Bb7 {-0.07/11 11} 10. Be3 {0.10/11 0} Be7 {-0.14/10 4} 11. Nc3 {0.19/9 4} O-O {
>-0.30/10 4} 12. b3 {0.01/9 7} b4 {-0.56/10 5} 13. Na4 {0.00/12 0} Rc8 {
>-0.58/10 6} 14. Rac1 {0.08/9 5} Qa5 {-0.46/10 4} 15. Nd2 {0.15/10 3} Bxg2 {
>-0.63/11 1} 16. Kxg2 {0.14/10 3} Rfd8 {-0.76/10 0} 17. Rfd1 {0.13/10 3} cxd4 {
>-0.46/10 4} 18. Rxc8 {0.11/10 1} Qd5+ {-0.49/11 0} 19. Nf3 {0.00/12 4} Rxc8 {
>-0.44/12 0} 20. Bxd4 {0.00/11 3} Rd8 {-0.40/11 2} 21. Qc2 {0.45/10 5} Ne5 {
>-0.50/10 3} 22. Qc7 {0.19/11 4} Nxf3 {-0.51/12 7} 23. exf3 {0.20/14 0} Qd6 {
>-0.45/12 4} 24. Qxd6 {0.24/15 0} Rxd6 {-0.45/13 0} 25. Bxf6 {0.19/15 5} gxf6 {
>-0.39/16 0} 26. Rxd6 {0.21/16 3} Bxd6 {-0.46/16 0} 27. Kf1 {0.31/15 4} f5 {
>-0.47/15 3} 28. Ke2 {0.71/14 3} Kg7 {-0.20/13 2} 29. Kd3 {0.84/15 3} a5 {
>0.43/14 5} 30. Kc4 {1.12/14 3} Bc7 {0.68/13 0} 31. Kb5 {1.26/14 1} f4 {
>1.23/13 3} 32. Nc5 {1.47/14 4} h5 {1.75/13 0} 33. Nb7 {1.90/14 2} Be5 {
>1.96/14 0} 34. Kxa5 {2.26/15 3} Bd4 {2.25/14 4} 35. Kxb4 {2.32/17 0} Bxf2 {
>2.64/14 2} 36. gxf4 {2.43/16 0} Bg1 {2.97/14 2} 37. h3 {2.84/12 1} Kf6 {
>3.36/14 2} 38. Nc5 {3.18/16 1} Be3 {3.87/14 2} 39. Nd3 {3.20/16 2} Ke7 {
>4.02/15 3} 40. a4 {3.31/17 0} Kd6 {4.05/14 3} 41. a5 {3.44/16 2} Kc7 {4.33/15 0
>} 42. Kb5 {3.64/16 2} Kb8 {4.83/15 0} 43. Ka6 {4.94/15 4} Bd4 {7.35/14 3} 44.
>b4 {5.59/17 2} Bg7 {7.91/15 0} 45. b5 {8.72/16 17} h4 {14.61/16 0} 46. b6 {
>Firstjuni,Deep Fritz 8 gibt auf (Lag: Av=0.43s, max=1.5s) 10.12/14 2} 1-0
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.