Shocking misevaluation by SF

[Lyudmil Tsvetkov]

tested with the latest asmfish (05.08.2017), no TB's :

on my pc asmfish didnt refused to take on f3 with 40/2h time-control

[d]8/4k3/R4pp1/6np/7P/p4NP1/5PK1/r7 b - - 0 58

I'm sure that the Stockfish team would gladly accept code that raises Elo by improving evaluation of rook and pawn endgame positions.

The latest asmFish (with 6-man tables) does switch from Nxf3 to Ne6, though as you point out, it takes some time:

Code: Select all

-3.54 58... Ne6 59. Nd2 Nc5 60. Ra5 a2 61. Ra7+ Kd6 62. Ra5 Ke6 63. Ra3 Kd7 64. Ra5 Kc6 65. Kf3 Ne6 66. Ke2 Kb6 67. Nc4+ Kc7 68. Nd2 Nd4+ 69. Kd3 Nc6 70. Ra4 Nb4+ 71. Rxb4 Rd1 72. Ra4 a1=Q 73. Rxa1 Rxa1 74. Ne4 Ra6 75. Ke2 Kd8 76. Nc3 Ra5 77. Kd3 Ra3 78. Kd2 Kd7 79. Nb5 Ra5 80. Nd4 Kd6 81. Kd3 Ra4 82. Nf3 Ra2 83. Ke3 Kd5 84. Ne1 Ra3+ 85. Kf4 Ra4+ 86. Ke3 Ke5 87. Nd3+ Kf5 (depth 43, 0:03:45)

asmFish sees no progress for Black with 58... Ne6. But perhaps the Fish is blind. Do you believe Ne6 is winning?

Code: Select all

-2.32 58... Ne6 59. Nd2 a2 60. Ra7+ Kd8 61. Nf1 Nc5 62. Ne3 Nd3 63. Nd5 Ne1+ 64. Kf1 f5 65. Ra8+ Kd7 66. Nc3 Nf3+ 67. Ke2 Re1+ 68. Kxf3 a1=Q 69. Rxa1 Rxa1 70. Nd5 Ra3+ 71. Kg2 Ke6 72. Nf4+ Kf7 73. Nd5 Ra6 74. Nf4 Ra4 75. Kf3 Ra3+ 76. Kg2 Ra2 77. Kf3 Kf6 78. Nd5+ Ke5 79. Nf4 Ra3+ 80. Kg2 Ra6 81. Kg1 Rc6 82. Kg2 Ke4 83. Ne2 Rd6 84. Nf4 Ke5 85. Kf3 Rb6 86. Nh3 Kf6 87. Kg2 Ra6 88. Ng5 Ra1 89. Nh3 Ra3 90. Kh2 Ra5 91. Kg2 Ra7 92. Kh2 Ke5 93. Kg2 Ra4 94. Kf3 Ra1 95. Nf4 Kf6 96. Nd5+ Ke6 97. Nf4+ Kf7 98. Kg2 Re1 99. Kf3 Re4 100. Kg2 Rd4 101. Kf3 Kf6 102. Kg2 Rd2 103. Nh3 Rc2 104. Kf3 Rc3+ 105. Kg2 Rc7 106. Nf4 Rc6 107. Nh3 Rd6 108. Kh2 Rd8 (depth 78, 6:10:40)

I did not analyse it, but here on a SF-drunken head, Ne4 seems a lot stronger, with 2 concrete threats:

- Nc3, followed by a2
- and Ra2, targeting the f2 pawn

[zullil]

tested with the latest asmfish (05.08.2017), no TB's :

on my pc asmfish didnt refused to take on f3 with 40/2h time-control

[d]8/4k3/R4pp1/6np/7P/p4NP1/5PK1/r7 b - - 0 58

I'm sure that the Stockfish team would gladly accept code that raises Elo by improving evaluation of rook and pawn endgame positions.

The latest asmFish (with 6-man tables) does switch from Nxf3 to Ne6, though as you point out, it takes some time:

Code: Select all

-3.54 58... Ne6 59. Nd2 Nc5 60. Ra5 a2 61. Ra7+ Kd6 62. Ra5 Ke6 63. Ra3 Kd7 64. Ra5 Kc6 65. Kf3 Ne6 66. Ke2 Kb6 67. Nc4+ Kc7 68. Nd2 Nd4+ 69. Kd3 Nc6 70. Ra4 Nb4+ 71. Rxb4 Rd1 72. Ra4 a1=Q 73. Rxa1 Rxa1 74. Ne4 Ra6 75. Ke2 Kd8 76. Nc3 Ra5 77. Kd3 Ra3 78. Kd2 Kd7 79. Nb5 Ra5 80. Nd4 Kd6 81. Kd3 Ra4 82. Nf3 Ra2 83. Ke3 Kd5 84. Ne1 Ra3+ 85. Kf4 Ra4+ 86. Ke3 Ke5 87. Nd3+ Kf5 (depth 43, 0:03:45)

asmFish sees no progress for Black with 58... Ne6. But perhaps the Fish is blind. Do you believe Ne6 is winning?

Code: Select all

-2.32 58... Ne6 59. Nd2 a2 60. Ra7+ Kd8 61. Nf1 Nc5 62. Ne3 Nd3 63. Nd5 Ne1+ 64. Kf1 f5 65. Ra8+ Kd7 66. Nc3 Nf3+ 67. Ke2 Re1+ 68. Kxf3 a1=Q 69. Rxa1 Rxa1 70. Nd5 Ra3+ 71. Kg2 Ke6 72. Nf4+ Kf7 73. Nd5 Ra6 74. Nf4 Ra4 75. Kf3 Ra3+ 76. Kg2 Ra2 77. Kf3 Kf6 78. Nd5+ Ke5 79. Nf4 Ra3+ 80. Kg2 Ra6 81. Kg1 Rc6 82. Kg2 Ke4 83. Ne2 Rd6 84. Nf4 Ke5 85. Kf3 Rb6 86. Nh3 Kf6 87. Kg2 Ra6 88. Ng5 Ra1 89. Nh3 Ra3 90. Kh2 Ra5 91. Kg2 Ra7 92. Kh2 Ke5 93. Kg2 Ra4 94. Kf3 Ra1 95. Nf4 Kf6 96. Nd5+ Ke6 97. Nf4+ Kf7 98. Kg2 Re1 99. Kf3 Re4 100. Kg2 Rd4 101. Kf3 Kf6 102. Kg2 Rd2 103. Nh3 Rc2 104. Kf3 Rc3+ 105. Kg2 Rc7 106. Nf4 Rc6 107. Nh3 Rd6 108. Kh2 Rd8 (depth 78, 6:10:40)

I did not analyse it, but here on a SF-drunken head, Ne4 seems a lot stronger, with 2 concrete threats:

- Nc3, followed by a2
- and Ra2, targeting the f2 pawn

If Ne4 then Nd4 and 0.00---at least according to a short asmFish search.

[Lyudmil Tsvetkov]

tested with the latest asmfish (05.08.2017), no TB's :

on my pc asmfish didnt refused to take on f3 with 40/2h time-control

[d]8/4k3/R4pp1/6np/7P/p4NP1/5PK1/r7 b - - 0 58

I'm sure that the Stockfish team would gladly accept code that raises Elo by improving evaluation of rook and pawn endgame positions.

The latest asmFish (with 6-man tables) does switch from Nxf3 to Ne6, though as you point out, it takes some time:

Code: Select all

-3.54 58... Ne6 59. Nd2 Nc5 60. Ra5 a2 61. Ra7+ Kd6 62. Ra5 Ke6 63. Ra3 Kd7 64. Ra5 Kc6 65. Kf3 Ne6 66. Ke2 Kb6 67. Nc4+ Kc7 68. Nd2 Nd4+ 69. Kd3 Nc6 70. Ra4 Nb4+ 71. Rxb4 Rd1 72. Ra4 a1=Q 73. Rxa1 Rxa1 74. Ne4 Ra6 75. Ke2 Kd8 76. Nc3 Ra5 77. Kd3 Ra3 78. Kd2 Kd7 79. Nb5 Ra5 80. Nd4 Kd6 81. Kd3 Ra4 82. Nf3 Ra2 83. Ke3 Kd5 84. Ne1 Ra3+ 85. Kf4 Ra4+ 86. Ke3 Ke5 87. Nd3+ Kf5 (depth 43, 0:03:45)

asmFish sees no progress for Black with 58... Ne6. But perhaps the Fish is blind. Do you believe Ne6 is winning?

Code: Select all

-2.32 58... Ne6 59. Nd2 a2 60. Ra7+ Kd8 61. Nf1 Nc5 62. Ne3 Nd3 63. Nd5 Ne1+ 64. Kf1 f5 65. Ra8+ Kd7 66. Nc3 Nf3+ 67. Ke2 Re1+ 68. Kxf3 a1=Q 69. Rxa1 Rxa1 70. Nd5 Ra3+ 71. Kg2 Ke6 72. Nf4+ Kf7 73. Nd5 Ra6 74. Nf4 Ra4 75. Kf3 Ra3+ 76. Kg2 Ra2 77. Kf3 Kf6 78. Nd5+ Ke5 79. Nf4 Ra3+ 80. Kg2 Ra6 81. Kg1 Rc6 82. Kg2 Ke4 83. Ne2 Rd6 84. Nf4 Ke5 85. Kf3 Rb6 86. Nh3 Kf6 87. Kg2 Ra6 88. Ng5 Ra1 89. Nh3 Ra3 90. Kh2 Ra5 91. Kg2 Ra7 92. Kh2 Ke5 93. Kg2 Ra4 94. Kf3 Ra1 95. Nf4 Kf6 96. Nd5+ Ke6 97. Nf4+ Kf7 98. Kg2 Re1 99. Kf3 Re4 100. Kg2 Rd4 101. Kf3 Kf6 102. Kg2 Rd2 103. Nh3 Rc2 104. Kf3 Rc3+ 105. Kg2 Rc7 106. Nf4 Rc6 107. Nh3 Rd6 108. Kh2 Rd8 (depth 78, 6:10:40)

I did not analyse it, but here on a SF-drunken head, Ne4 seems a lot stronger, with 2 concrete threats:

- Nc3, followed by a2
- and Ra2, targeting the f2 pawn

If Ne4 then Nd4 and 0.00---at least according to a short asmFish search.

that 0.0 is bogus.

after Ne4 Nd4, even f5 does not give you a 0.0 score:

[d]8/4k3/R5p1/5p1p/3Nn2P/p5P1/5PK1/r7 w - - 0 3

but it might be the case that Ne4 is drawn, Ne6 too, and Nf3 too, i.e., with perfect play the position is draw, no matter what continuation black chooses.

[zullil]

after Ne4 Nd4, even f5 does not give you a 0.0 score:

[d]8/4k3/R5p1/5p1p/3Nn2P/p5P1/5PK1/r7 w - - 0 3

but it might be the case that Ne4 is drawn, Ne6 too, and Nf3 too, i.e., with perfect play the position is draw, no matter what continuation black chooses.

Code: Select all

0.00 3. Rxg6 Ra2 4. Kf3 Nf6 5. Nxf5+ Ke6 6. Ne3 Rd2 7. Rg7 Nd7 8. Rg6+ Kf7 9. Ra6 a2 10. Ra7 Kf8 11. Nc4 Rc2 12. Ne3 Rd2 (depth 61, 0:11:00)

[Spliffjiffer]

hi Louis

i think that 1...ne6 is winning, yes

...although, there was a very serious mistake in the line i gave !.. though i think that the endgame with r+n vs r i gave is the critical problem to solve...lets see ...

in the line i gave previously 7...nb7 is a stupid blunder because white can easily simplify to a position that is practicaly the same as after 1...nxf3 with a dead draw:

instead of 11. rc8+, white can play simply rxa5 and we have the same draw i was blaming SF for going into the 1...nxf3 line
although atm i see no possibility to avoid the ending i was talking about :

1...ne6 !?, 2. nd2 nc5, 3. ra7+ ke6, 4. ra8 a2, 5. ra3 kd5, 6. ra5 kc6, 7. kf3 nd3 !! ( instead of nb7 ??)...
[d]8/8/2k2pp1/R6p/7P/3n1KP1/p2N1P2/r7 w - - 6 65

8.nb3! (ne4 looses to g5 followed by g4)
8...kb6, 9. ke2 rb1, 10. rxa2 rxb3, 11. rd2 ne5, 12. rd6+ kc5, 13. rxf6 rb2, 14. kf1 kd5
[d]8/8/5Rp1/3kn2p/7P/6P1/1r3P2/5K2 w - - 3 72

this is tricky to win but im sure its winning although SF might not be a big help for you guys
a typical maneuvre i gave already in my post before...if there are questions, ill try to help

[Lyudmil Tsvetkov]

after Ne4 Nd4, even f5 does not give you a 0.0 score:

[d]8/4k3/R5p1/5p1p/3Nn2P/p5P1/5PK1/r7 w - - 0 3

but it might be the case that Ne4 is drawn, Ne6 too, and Nf3 too, i.e., with perfect play the position is draw, no matter what continuation black chooses.

Code: Select all

0.00 3. Rxg6 Ra2 4. Kf3 Nf6 5. Nxf5+ Ke6 6. Ne3 Rd2 7. Rg7 Nd7 8. Rg6+ Kf7 9. Ra6 a2 10. Ra7 Kf8 11. Nc4 Rc2 12. Ne3 Rd2 (depth 61, 0:11:00)

why not 4...Rf2+ with check and taking a full pawn?

the very best white could achieve from here is, indeed drawn, but KRN vs KR endgame, with the stronger side having a full piece more.

if that is 0.0 for you, I am fine.

exchanging knights on f3 give +300cps score for SF instead of 0.0, but is still a draw.

same +200-300cps in the Ne6 line, presumably might also be a draw, so I don't know which line is better.

I would always choose to have a knight more instead of advanced passed pawn and initiative.

anyway, I did not analyse this position at all and am not very interested in what its objective outcome is, so you might feel free to totally neglect my input.

[Spliffjiffer]

...2 clear lines that should help you (and SF also) to see that its winning

[d]8/8/5Rp1/3kn2p/7P/6P1/1r3P2/5K2 w - - 0 1

1. Kg1 Rc2 2. Kf1 Ng4 3. Rf8 Ke6 4. Kg1 Ne5 5. Re8+ Kf6 6. Ra8 Nd3 7. Rd8
Rd2 8. Re8 Kf5 9. Rf8+ Ke6 10. Rf3 (10. Kf1 Ra2 11. Rf3 Ra1+ 12. Ke2 Ne5
13. Rf4 Ke7 14. Re4 Kf6 15. Re3 Kf5) 10. .. Rd1+ 11. Kh2 Ne5 12. Re3 Rd2
13. Kg1 Kd5 14. Ra3 Nd3 15. Ra8 Ke6 16. Re8+ Kf6 17. Rf8+ Kg7 18. Rf3 Rd1+
19. Kh2 Ne5 20. Ra3 Rd2 21. Kg2 Nd3 22. Ra7+ Kf6 23. Ra6+ Kf5 24. Ra5+ Ke4
25. Ra4+ Kd5 26. Ra5+ Kd4 27. Ra4+ Kc3 28. Ra3+ Kc4 -+

or:

1. Kg1 Rc2 2. Rb6 Ke4 3. Rb4+ Rc4 4. Rb6 Ng4 5. Rb3 Rc1+ 6. Kg2 Rc2 7. Rf3
Ke5 8. Rf8 Ke6 9. Rf3 Rd2 10. Kg1 Ne5 11. Rf4 Nd3 12. Rf3 Rd1+ 13. Kh2 Ne5
14. Re3 Rd2 15. Kg2 Kf5 (15.Kg1 Kd5) -+

[Eelco de Groot]

...2 clear lines that should help you (and SF also) to see that its winning

[d]8/8/5Rp1/3kn2p/7P/6P1/1r3P2/5K2 w - - 0 1

[pgn]
[Event "?"]
[Site "?"]
[Date "2017.08.08"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]
[SetUp "1"]
[FEN "8/8/5Rp1/3kn2p/7P/6P1/1r3P2/5K2 w - -"]

1. Kg1 Rc2 2. Kf1 Ng4 3. Rf8 Ke6 4. Kg1 Ne5 5. Re8+ Kf6
6. Ra8 Nd3 7. Rd8 Rd2 8. Re8 Kf5 9. Rf8+ Ke6 10. Rf3
(10. Kf1 Ra2 11. Rf3 Ra1+ 12. Ke2 Ne5 13. Rf4 Ke7 14. Re4
Kf6 15. Re3 Kf5) 10... Rd1+ 11. Kh2 Ne5 12. Re3 Rd2 13. Kg1
Kd5 14. Ra3 Nd3 15. Ra8 Ke6 16. Re8+ Kf6 17. Rf8+ Kg7
18. Rf3 Rd1+ 19. Kh2 Ne5 20. Ra3 Rd2 21. Kg2 Nd3 22. Ra7+
Kf6 23. Ra6+ Kf5 24. Ra5+ Ke4 25. Ra4+ Kd5 26. Ra5+ Kd4
27. Ra4+ Kc3 28. Ra3+ Kc4 $17 *
[/pgn]

or:

[pgn]
[Event "?"]
[Site "?"]
[Date "2017.08.08"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]
[SetUp "1"]
[FEN "8/8/5Rp1/3kn2p/7P/6P1/1r3P2/5K2 w - -"]

1. Kg1 Rc2 2. Rb6 Ke4 3. Rb4+ Rc4 4. Rb6 Ng4 5. Rb3 Rc1+
6. Kg2 Rc2 7. Rf3 Ke5 8. Rf8 Ke6 9. Rf3 Rd2 10. Kg1 Ne5
11. Rf4 Nd3 12. Rf3 Rd1+ 13. Kh2 Ne5 14. Re3 Rd2 15. Kg2
(15. Kg1 Kd5) 15... Kf5 $17 *
[/pgn]

[zullil]

...2 clear lines that should help you (and SF also) to see that its winning

[d]8/8/5Rp1/3kn2p/7P/6P1/1r3P2/5K2 w - - 0 1

1. Kg1 Rc2 2. Kf1 Ng4 3. Rf8 Ke6 4. Kg1 Ne5 5. Re8+ Kf6 6. Ra8 Nd3 7. Rd8
Rd2 8. Re8 Kf5 9. Rf8+ Ke6 10. Rf3 (10. Kf1 Ra2 11. Rf3 Ra1+ 12. Ke2 Ne5
13. Rf4 Ke7 14. Re4 Kf6 15. Re3 Kf5) 10. .. Rd1+ 11. Kh2 Ne5 12. Re3 Rd2
13. Kg1 Kd5 14. Ra3 Nd3 15. Ra8 Ke6 16. Re8+ Kf6 17. Rf8+ Kg7 18. Rf3 Rd1+
19. Kh2 Ne5 20. Ra3 Rd2 21. Kg2 Nd3 22. Ra7+ Kf6 23. Ra6+ Kf5 24. Ra5+ Ke4
25. Ra4+ Kd5 26. Ra5+ Kd4 27. Ra4+ Kc3 28. Ra3+ Kc4 -+

or:

1. Kg1 Rc2 2. Rb6 Ke4 3. Rb4+ Rc4 4. Rb6 Ng4 5. Rb3 Rc1+ 6. Kg2 Rc2 7. Rf3
Ke5 8. Rf8 Ke6 9. Rf3 Rd2 10. Kg1 Ne5 11. Rf4 Nd3 12. Rf3 Rd1+ 13. Kh2 Ne5
14. Re3 Rd2 15. Kg2 Kf5 (15.Kg1 Kd5) -+

Thanks. It is nice to see lines posted that have been carefully considered! The latest asmFish is still searching the variation that starts with 4. Rf4 in your first line, but the current evaluation for that branch suggests that Black wins. So, unless someone else finds a mistake, it appears that the original position is indeed winning for Black.

This seems like a fine example of a very deep endgame position that engines struggle with.

[Spliffjiffer]

thanks Louis ;.)

to be honest im still not completely sure myself wether this rn vs r endgame is winning or not, there are still lines where i struggle by myself to win this unfortunately....so if it should be a draw i apologize for claiming a win here previously...at least there are good winning-chances to win this because the defense, if there is any, is not trivial
i hope your monster-machine can confirm a win...that would be very satisfying for me