Position from Norway Chess 2015 SF struggles with

[OneTrickPony]

[pgn]
1. e4 c5
2. Nf3 d6
3. d4 cxd4
4. Nxd4 Nf6
5. Nc3 a6
6. h3 e6
7. g4 h6
8. Bg2 Nc6
9. Be3 Be7
10. f4 Nd7
11. O-O Nxd4
12. Qxd4 O-O
13. Qd2 Rb8
14. Ne2 b5
15. Rad1 Qc7
16. f5 Nf6
17. Ng3 Bb7
18. Kh1 Rbd8
[/pgn]

So today Anand played a combination SF misses even at quite significant depth (currently I am running a compile from June 17th at D30 in multipv 4 and it still gives 0.00 for 19.Bxh6; apparently it sees it at D30/D31 in 1pv mode but it's still quite late and would be missed in blitz).
It seems Houdini 4 struggles with it as well. I don't have the newest Komodo to test but I am interested to see how it fares.

The combination doesn't seem that complicated so it's very suprising to me the top engine misses it. Props for Anand for playing it though!

[Lyudmil Tsvetkov]

[pgn]
1. e4 c5
2. Nf3 d6
3. d4 cxd4
4. Nxd4 Nf6
5. Nc3 a6
6. h3 e6
7. g4 h6
8. Bg2 Nc6
9. Be3 Be7
10. f4 Nd7
11. O-O Nxd4
12. Qxd4 O-O
13. Qd2 Rb8
14. Ne2 b5
15. Rad1 Qc7
16. f5 Nf6
17. Ng3 Bb7
18. Kh1 Rbd8
[/pgn]

So today Anand played a combination SF misses even at quite significant depth (currently I am running a compile from June 17th at D30 in multipv 4 and it still gives 0.00 for 19.Bxh6; apparently it sees it at D30/D31 in 1pv mode but it's still quite late and would be missed in blitz).
It seems Houdini 4 struggles with it as well. I don't have the newest Komodo to test but I am interested to see how it fares.

The combination doesn't seem that complicated so it's very suprising to me the top engine misses it. Props for Anand for playing it though!

Not true, man.

My SF 6 on humble 4 cores picks it immediately, with a score of 200cps after gh6, and 130cps after d5.

This is the only SF choice really, and all other white moves are bad.

The combination is very obvious, extremely easy, and shallow, so I do not believe any engine will miss it. Engines miss nothing shallow.

[MikeGL]

[D]3r1rk1/1bq1bpp1/p2ppn1p/1p3P2/4P1P1/4B1NP/PPPQ2B1/3R1R1K w - - 5 19


Below sub-variation is nice, due to all heavy pieces ramming into blacks K
[pgn]
[Event "Norway 2015"]
[Site "Norway"]
[Date "2015.06.23"]
[Round "?"]
[White "V Anand"]
[Black "Maxime Vachier-Lagrave"]
[Result "1-0"]
[BlackElo ""]
[Time "05:35:41"]
[WhiteElo ""]
[TimeControl "600"]
[SetUp "1"]
[FEN "3r1rk1/1bq1bpp1/p2ppn1p/1p3P2/4P1P1/4B1NP/PPPQ2B1/3R1R1K w - - 5 19"]
[Termination "unterminated"]
[PlyCount "12"]
[WhiteType "human"]
[BlackType "human"]

19. Bxh6 gxh6 20. Qxh6 d5 21. g5 Qxg3 22. Rd3 {sub-variation follows, (didnt happen in actual game)} Qe5 23. gxf6 Bxf6 24. Rf4
*
[/pgn]

Critter1.6(32bit) on old dual core, considers Bxh6 as best in 15s, but 0.00 only all other moves are -0.30 something.
So basically considers Bxh6 as perpetual check only, and doesn't find the win, but as Lyudmil mentioned, there are no other better moves.

[OneTrickPony]

I am probably doing something wrong then but:

[whereagles]

Just out of curiosity, I ran it in SF6 @3 PVs and...

Code: Select all

1. Bxh6 gxh6 2. Qxh6 d5 3. Qg5+ Kh7 4. Qh4+ Kg8 5. Qg5+ 
= (0.00)    Depth=29/47    TB=25    0:04:38    811 MN

1. c3 d5 2. e5 Ne4 3. Nxe4 dxe4 4. Bd4 Bg5 5. Qf2 e3 6. Bxe3 Rxd1 7. Rxd1 Bxg2+ 8. Kxg2 Qxe5 9. Bxg5 hxg5 10. Qe1 Qxe1 11. Rxe1 exf5 12. gxf5 Rd8 13. Kf3 Rd3+ 14. Kg4 Rd2 15. Re8+ Kh7 16. Ra8 Rg2+ 17. Kf3 Rxb2 18. Rxa6 Rc2 19. Rb6 Rxc3+ 20. Kg4 
- (0.17)    Depth=29/47    TB=25    0:04:38    811 MN

1. fxe6 fxe6 
= (#0)    Depth=28/47    TB=25    0:04:38    811 MN

Check last line.... =(#0) mate in 0 moves??? What does this mean???

[Lyudmil Tsvetkov]

I did not use multi PV, and my SF 6 finds it instantly; it is true though that it also strangely considers the aboveshown draw variation with queen check on g5 instead of pushing pawn to g5 that wins. Maybe you just need a bit more time or a stronger hardware.

Vachier-Lagrave went wrong a couple of times in a relatively very shallow line - he first captured on h6 instead of the much better d5, where black might even achieve the odd draw, if white is not precise in converting, and then played Nh5 instead of Ne4, which should hold much longer, though also obviously losing.

I would think that Anand was won throughout and simply the line Lagrave played is dubious at the least - the black setup is very passive, you can not allow such a gorgeous white pawn structure, h3,g4,f5 long chain, plus a twice defended storming pawn on f5, this is simply irresistible. Of course, SF does not have proper evaluation for this, and it sees it, mainly because the lines are shallow, with checks and attacking moves all the time, no difficult quiet moves to make.

[MikeGL]

[D]3r1rk1/1b2bp2/p3pn1Q/1p1p1PP1/4P3/6qP/PPP3B1/3R1R1K w - - 0 22

I ran engines again on this position in a slower PC to investigate what it
was busy calculating on, and it seems it keeps on missing the difficult
zwischenzug 22.Rd3 because it considers 22.gxf6 Bd6! 23.Qg7+ Qxg7
which is a -+ for black. Therefor it is difficult to consider 21.g5 on
earlier move including the even more earlier 19.Bxh6


See below the line after 21... Qxg3 22.gxf6 seems normal and the only
way to recover the lost piece, missing an "in-between" 22.Rd3 which wins
by stopping a battery at b8-h2 diagonal of B+Q (Bd6+Qg3 > h2#)


[pgn]
[Event "Norway 2015"]
[Site "Norway"]
[Date "2015.06.23"]
[Round "?"]
[White "Vishy Anand"]
[Black "Maxime Vachier-Lagrave"]
[Result "1-0"]
[BlackElo ""]
[Time "05:35:41"]
[WhiteElo ""]
[TimeControl "600"]
[SetUp "1"]
[FEN "3r1rk1/1bq1bpp1/p2ppn1p/1p3P2/4P1P1/4B1NP/PPPQ2B1/3R1R1K w - - 5 19"]
[Termination "unterminated"]
[WhiteType "human"]
[BlackType "human"]

19. Bxh6 gxh6 20. Qxh6 d5 21. g5 Qxg3 22. gxf6 {[22.Rd3! +-]} Bd6 23. Qg7+ Qxg7 24. fxg7
Rfe8 {- + for black due to a clear 1 B up} {[possible reason why this 19.Bxh6! sac is a bit difficult for engines and mere mortals]}
[/pgn]

But of course at a bit deeper search the 22.Rd3 intermediary will be found.

[zullil]

Just out of curiosity, I ran it in SF6 @3 PVs and...

Code: Select all

1. Bxh6 gxh6 2. Qxh6 d5 3. Qg5+ Kh7 4. Qh4+ Kg8 5. Qg5+ 
= (0.00)    Depth=29/47    TB=25    0:04:38    811 MN

1. c3 d5 2. e5 Ne4 3. Nxe4 dxe4 4. Bd4 Bg5 5. Qf2 e3 6. Bxe3 Rxd1 7. Rxd1 Bxg2+ 8. Kxg2 Qxe5 9. Bxg5 hxg5 10. Qe1 Qxe1 11. Rxe1 exf5 12. gxf5 Rd8 13. Kf3 Rd3+ 14. Kg4 Rd2 15. Re8+ Kh7 16. Ra8 Rg2+ 17. Kf3 Rxb2 18. Rxa6 Rc2 19. Rb6 Rxc3+ 20. Kg4 
- (0.17)    Depth=29/47    TB=25    0:04:38    811 MN

1. fxe6 fxe6 
= (#0)    Depth=28/47    TB=25    0:04:38    811 MN

Check last line.... =(#0) mate in 0 moves??? What does this mean???

It probably means you have a buggy GUI.

[Vinvin]

2 others positions where SF seems weak compare to an average GM : one opposite B ending and one rook ending

SF gives a wrong eval (around +1.40) to this drawish positions in the Hammer - Vachier-Lagrave game :

[d]4R1r1/3P1pk1/5b1p/3K4/3p1PPp/8/8/5B2 w - - 13 43

[pgn]
[Event "3rd Norway Chess 2015"]
[Site "Stavanger NOR"]
[Date "2015.06.17"]
[Round "2"]
[White "Hammer,J"]
[Black "Vachier Lagrave,M"]
[Result "1/2-1/2"]
[WhiteElo "2677"]
[BlackElo "2723"]
[EventDate "2015.06.17"]
[ECO "A29"]

1. c4 Nf6 2. Nc3 e5 3. Nf3 Nc6 4. g3 g6 5. Bg2 Bg7 6. O-O O-O 7. e4 d6 8.
d3 h6 9. Be3 Kh7 10. h3 Nd7 11. Rc1 Nd4 12. Bxd4 exd4 13. Ne2 c5 14. h4 b5
15. cxb5 Rb8 16. a4 a6 17. b4 axb5 18. bxc5 dxc5 19. axb5 Rxb5 20. Nf4 Ne5
21. Bh3 Bb7 22. h5 Nxf3+ 23. Qxf3 Qd6 24. Rb1 Rb4 25. Nd5 Bxd5 26. exd5 c4
27. Rxb4 Qxb4 28. dxc4 Qxc4 29. d6 Qc5 30. d7 Qxh5 31. Qxh5 gxh5 32. Bf5+
Kg8 33. Re1 Bf6 34. Re8 h4 35. g4 Kg7 36. f4 Bd8 37. Kg2 Bf6 38. Bd3 Bd8
39. Kf3 Bf6 40. Bf1 Bd8 41. Ke4 Bf6 42. Kd5 Rg8 43. Kc6 Bd8 44. Rxd8 Rxd8
45. Kc7 Rf8 46. d8=Q Rxd8 47. Kxd8 h5 48. gxh5 Kh6 49. Ke7 Kxh5 50. Be2+
Kh6 51. Bf1 Kh5 52. f5 Kg4 53. f6 h3 54. Bxh3+ Kxh3 55. Kxf7 d3 56. Kg7 d2
57. f7 d1=Q 58. f8=Q Qg4+ 59. Kf7 Qf5+ 60. Ke7 Qxf8+ 61. Kxf8 1/2-1/2
[/pgn]

In the ending Carlsen - Nakamura, SF gives >1 for this drawish position :

[d]8/4Rpk1/6p1/8/5PP1/1r2PK2/8/8 b - - 0 52

[pgn]
[Event "3rd Norway Chess 2015"]
[Site "Stavanger NOR"]
[Date "2015.06.22"]
[Round "6"]
[White "Carlsen,M"]
[Black "Nakamura,Hi"]
[Result "1/2-1/2"]
[WhiteElo "2876"]
[BlackElo "2802"]
[EventDate "2015.06.17"]
[ECO "D56"]

1. d4 d5 2. c4 e6 3. Nf3 Nf6 4. Nc3 Be7 5. Bg5 h6 6. Bh4 O-O 7. e3 Ne4 8.
Bxe7 Qxe7 9. Rc1 c6 10. Qc2 Nxc3 11. Qxc3 dxc4 12. Bxc4 b6 13. O-O Nd7 14.
Rfd1 Bb7 15. h3 c5 16. d5 exd5 17. Bxd5 Bxd5 18. Rxd5 Nf6 19. Rd3 Rad8 20.
Rcd1 Rxd3 21. Qxd3 Rc8 22. a4 c4 23. Qc2 Qe4 24. Nd4 c3 25. Qb3 Qd5 26. Qc2
Qe4 27. Qxe4 Nxe4 28. b3 Nd2 29. Rc1 a5 30. Rc2 Nb1 31. Rc1 Nd2 32. Rc2 Nb1
33. Nb5 Kf8 34. f3 Rd8 35. Nd4 Rc8 36. Rc1 Na3 37. Nc2 Nxc2 38. Rxc2 Rc5
39. Kf2 b5 40. axb5 Rxb5 41. Rxc3 g6 42. f4 h5 43. Rc8+ Kg7 44. Ra8 Rxb3
45. Rxa5 Rb2+ 46. Kf3 Rb3 47. Ra7 Rc3 48. Re7 Rb3 49. Kg3 Rb2 50. Kf3 Rb3
51. g4 hxg4+ 52. hxg4 Rb1 53. Rd7 Rf1+ 54. Ke4 Rg1 55. g5 Ra1 56. Ke5 Ra3
57. e4 Ra5+ 58. Kd6 Ra6+ 59. Ke5 Ra5+ 60. Rd5 Ra4 61. Rc5 Ra1 62. Rc2 Rf1
63. Rc7 Re1 64. Ra7 Rg1 65. Ra4 Rf1 66. Ra6 Re1 67. Kd4 Rd1+ 68. Ke3 Re1+
69. Kf3 Rf1+ 70. Kg3 Re1 71. e5 Re3+ 72. Kf2 Rb3 73. Rd6 Ra3 74. Rd8 Rc3
75. Ke2 Ra3 76. Rd3 Ra1 77. Ke3 Re1+ 78. Kd4 Rf1 79. Ke4 Ra1 80. Rd7 Ra4+
81. Rd4 Ra5 82. Rc4 Kf8 83. Rc8+ Kg7 84. Rc7 Kf8 85. Kd4 Ra4+ 86. Rc4 Ra5
87. Rc8+ Kg7 88. e6 fxe6 89. Rc5 Ra7 90. Ke5 Rf7 91. Rc4 Rf5+ 92. Kxe6 Kg8
93. Rc8+ Kg7 94. Rc7+ Kg8 95. Rc8+ Kg7 1/2-1/2
[/pgn]

[Vinvin]

[pgn]
1. e4 c5
2. Nf3 d6
3. d4 cxd4
4. Nxd4 Nf6
5. Nc3 a6
6. h3 e6
7. g4 h6
8. Bg2 Nc6
9. Be3 Be7
10. f4 Nd7
11. O-O Nxd4
12. Qxd4 O-O
13. Qd2 Rb8
14. Ne2 b5
15. Rad1 Qc7
16. f5 Nf6
17. Ng3 Bb7
18. Kh1 Rbd8
[/pgn]

So today Anand played a combination SF misses even at quite significant depth (currently I am running a compile from June 17th at D30 in multipv 4 and it still gives 0.00 for 19.Bxh6; apparently it sees it at D30/D31 in 1pv mode but it's still quite late and would be missed in blitz).
It seems Houdini 4 struggles with it as well. I don't have the newest Komodo to test but I am interested to see how it fares.

The combination doesn't seem that complicated so it's very suprising to me the top engine misses it. Props for Anand for playing it though!

1min9sec to get a >+1 score on a 6 cores@4.4GHz

[d]3r1rk1/1bq1bpp1/p2ppn1p/1p3P2/4P1P1/4B1NP/PPPQ2B1/3R1R1K w - - 5 19

Code: Select all

Stockfish_15060522_x64_modern:
...
 28/31	00:05	 51.778k	9.169k	 0,00	19.Bxh6 gxh6 20.Qxh6 d5 21.Qg5+ Kh8 22.Qh6+ Kg8
 29/31	00:07	 70.034k	9.403k	 0,00	19.Bxh6 gxh6 20.Qxh6 d5 21.Qg5+ Kh8 22.Qh6+ Kg8
 30/31	00:08	 83.425k	9.452k	 0,00	19.Bxh6 gxh6 20.Qxh6 d5 21.Qg5+ Kh8 22.Qh6+ Kg8
 31/31	00:12	 115.751k	9.580k	 0,00	19.Bxh6 gxh6 20.Qxh6 d5 21.Qg5+ Kh8 22.Qh6+ Kg8
 32/35+	00:22	 212.308k	9.576k	+0,06	19.Bxh6
 32/37+	00:24	 234.836k	9.578k	+0,12	19.Bxh6
 32/39+	00:48	 459.435k	9.511k	+0,87	19.Bxh6
 32/49	01:09	 659.819k	9.436k	+1,18	19.Bxh6 d5 20.Qg5 Nh5 21.Qxh5 Qxg3 22.Bf4 Qh4 23.Qxh4 Bxh4 24.exd5 Bxd5 25.Bxd5 Rxd5 26.Rxd5 exd5 27.Kg2 Rc8 28.c3 b4 29.Rc1 bxc3 30.Rxc3 Rxc3 31.bxc3 Bf6 32.Bd2 Kf8 33.Kf3 Ke7 34.Ke3 Kd6 35.Kd3 g6 36.fxg6 fxg6 37.c4 Be5 38.h4 dxc4+ 39.Kxc4 Kc6
 33/49+	01:17	 726.663k	9.437k	+1,24	19.Bxh6
 33/49-	01:17	 735.358k	9.433k	+1,18	19.Bxh6 d5
 33/49+	01:18	 744.628k	9.426k	+1,24	19.Bxh6
 33/49	01:19	 748.210k	9.421k	+1,24	19.Bxh6 d5 20.Qg5 Nh5 21.Qxh5 Qxg3 22.Bf4 Qh4 23.Qxh4 Bxh4 24.exd5 Bxd5 25.Bxd5 Rxd5 26.Rxd5 exd5 27.Kg2 Rc8 28.c3 b4 29.Rc1 bxc3 30.Rxc3 Rxc3 31.bxc3 Bf6 32.Bd2 Kf8 33.Kg3 Be5+ 34.Kf3 Ke7 35.h4 f6 36.Ke3 Kd6