7 men Rook endgames H4 vs SF in recent TCEC stage 1b.

[Nay Lin Tun]

I am interested to know whether it was a draw or a loss for H4(for those who can access lomosov 7men TB). It would be great if someone can share how to post pgn, diagram as well.
Thanks

[Vinvin]

Here's the game
[pgn]
[Event "TCEC Season 8 - Stage 1b"]
[Site "http://tcec.chessdom.com"]
[Date "2015.09.06"]
[Round "8"]
[White "Houdini 4"]
[Black "Stockfish 300815"]
[Result "0-1"]
[BlackElo "3216"]
[ECO "E14"]
[Opening "Queen's Indian"]
[Variation "4.e3"]
[WhiteElo "3190"]
[TimeControl "5400+30"]
[Termination "adjudication"]
[PlyCount "201"]
[WhiteType "human"]
[BlackType "human"]

1. d4 Nf6 2. c4 e6 3. Nf3 b6 4. e3 d5 5. cxd5 exd5 6. Ne5 Bb7 7. Bb5+ c6 8.
Bd3 Bd6 9. O-O O-O 10. b3 c5 11. Bb2 Na6 12. Nd2 Qe7 13. Re1 Nb4 14. Bb1
Rae8 15. a3 Na6 16. f4 Nc7 17. dxc5 bxc5 18. Bc2 g6 19. Ndf3 Rb8 20. Bd3
Ne6 21. Qc2 Ba8 22. Rab1 Rb6 23. Ba1 Rd8 24. a4 Ne4 25. Ng4 h5 26. Qb2 d4
27. exd4 cxd4 28. Nh6+ Kh7 29. Bxe4 Bxe4 30. Rxe4 Kxh6 31. f5 gxf5 32. Qd2+
Kh7 33. Rh4 Kg8 34. Rxh5 Bf4 35. Qe1 Qf6 36. Kh1 Ng7 37. Rh4 Re6 38. Qd1
Be3 39. Qd3 Qe7 40. Rd1 Re4 41. Rh3 Ne6 42. Rf1 Nf4 43. Rg3+ Ng6 44. Bb2
Qe6 45. Bc1 Bxc1 46. Rxc1 Re3 47. Qc4 Qxc4 48. bxc4 Rc3 49. Rd1 Rxc4 50. h4
Kf8 51. h5 Nf4 52. Rg5 Rd5 53. Nh4 d3 54. Rxf5 Rd6 55. Kh2 Ne2 56. Nf3 Rxa4
57. Rd2 Kg7 58. g3 Nd4 59. Nxd4 Raxd4 60. Kg2 R6d5 61. h6+ Kg6 62. h7 Rd8
63. Rf3 Kg7 64. h8=B+ Rxh8 65. Rfxd3 Rxd3 66. Rxd3 a5 67. Ra3 Rh5 68. Kf3
Rb5 69. Ra4 Kh6 70. Rh4+ Kg5 71. Rg4+ Kf6 72. Ra4 Kg7 73. Rg4+ Kf8 74. Ra4
Ke7 75. Ke2 Ke6 76. Kd3 f6 77. Re4+ Kd6 78. Ra4 Ke5 79. g4 Ke6 80. Rc4 Kf7
81. Rc7+ Kg6 82. Ra7 Rb3+ 83. Kc2 Ra3 84. Kb2 Ra4 85. Kc3 Kg5 86. Rg7+ Kh4
87. Rg6 Rf4 88. Kb3 a4+ 89. Kb2 Rf3 90. Rg7 Rf2+ 91. Ka3 Rf4 92. Rg6 Kg3
93. Kb2 Kf3 94. Ka3 Ke4 95. Rg8 Ke5 96. Rg6 Kd5 97. Rh6 Ke4 98. Rg6 Ke5 99.
Rg7 Ke6 100. Ka2 Rf3 101. Rb7 0-1
[/pgn]

And the first position with 7 pieces on the board (after 66.Rxd3) :
[d]7r/p4pk1/8/8/8/3R2P1/6K1/8 b - - 0 66

[crossroads]

According to Lomonosov it's a draw.

[AdminX]

[Nay Lin Tun]

According to Lomonosov it's a draw.

That was what i though. In a rook endgame with a or h file extra pawn and kings and pawns on the other side, the upper hand player is likely to win if the rook is behind the passed pawn and the passed pawn is already advanced to 5th rank. ( I dont know in detail theory about distance between extra pawns vs kings on the other side). In general it is draw if, 1. extra passed pawn on 4th rank is blocked by defending opponent's rook 2. own rook is in front of extra passed pawn's promotion and opponent's rook is controlling from behind.

[hgm]

According to Lomonosov it's a draw.

I guess this sort of falsifies the belief that strong engines would know how to handle such positions by their normal search and evaluation.

If I counted right there are only 208 possible Pawn constellations reachable from this position, when I allow the Pawns to be on 8th rank (where they would represent a Queen). I we would declare a position with such an 8th-rank Pawn for the side to move, and not in check, as won, (or at least give it a +9 evaluation), it would require only solving of 208 4-men (KRKR) P-slices to identify positions where you have a certain draw. It should be possible to generate one 4-men P-slice in about 100msec on a single core (at least that is what my experience with KBNK tells me), so that would mean en engine could generate the required EGT in 21 sec.(And from that moment on play instantly for the rest of the game from 1-ply searches through the EGT.) That should be quite affordable under TCEC time controls.

So having on-the-fly EGT generation does seem a worthwile even for top engines, and would prevent such mishaps as bungling a drawn Rook ending.

[BubbaTough]

According to Lomonosov it's a draw.

I guess this sort of falsifies the belief that strong engines would know how to handle such positions by their normal search and evaluation.

If I counted right there are only 208 possible Pawn constellations reachable from this position, when I allow the Pawns to be on 8th rank (where they would represent a Queen). I we would declare a position with such an 8th-rank Pawn for the side to move, and not in check, as won, (or at least give it a +9 evaluation), it would require only solving of 208 4-men (KRKR) P-slices to identify positions where you have a certain draw. It should be possible to generate one 4-men P-slice in about 100msec on a single core (at least that is what my experience with KBNK tells me), so that would mean en engine could generate the required EGT in 21 sec.(And from that moment on play instantly for the rest of the game from 1-ply searches through the EGT.) That should be quite affordable under TCEC time controls.

So having on-the-fly EGT generation does seem a worthwile even for top engines, and would prevent such mishaps as bungling a drawn Rook ending.

Very cool idea!

-Sam

[BBauer]

Following the game on a slow notebook with 6-piece syzygys the draw was immediatly seen. Later on Houdini moved his king to the queen side. Then the game was lost.
Stockfish used syzygys, Houdini used nalimov. I don't know whether this made a difference.
So, while egtb generation may help, 6-piece syzygys are enough to handle this end game.
Kind regards
Bernhard

[hgm]

Unfortunately the number of 208 was wrong. (I added something that should have been multiplied). The a-Pawn can have 8 'locations', if you count 8th-rank and being absent. The other Pawns can similarly have 8 or 7 locations if they have not captured, which would make 56 combinations. Then there are 5+6=11 situations where one would have captured the other (and switched file). So that is 67 constellations for the f- and g-Pawn, times 8 for the a-Pawn, is 536 constellations. (Some of these with multiple Queens might be deleted as irrelevant.)

So that would bring the estimate to 54 sec. Still very doable.

[Vinvin]

http://i59.tinypic.com/wu2xw9.jpg

Can you point when Stockfish got a winning endgame for the first time ?