Opening novelty, positional exchange sacrifice by lc0 and incredible position

[Javier Ros]

From one of my tests, lc0 played the opening novelty 10..h5 putting the rook in play in an original way and sacrificing it later for a bishop.

19. Bg5 {lc0 0.00/27 16} Rxg5 {Sf9 1.31/14 7.7s}

After 74..Rxg6 the next incredible position was reached, where Sf9 is cornered and lost.

[pgn][Event "Silver16sec"] [Site "Ros"] [Date "2018.10.12"] [Round "42"] [White "Stockfish9"] [Black "Lc011261"] [Result "0-1"] [ECO "B66"] [Annotator "R,J"] [PlyCount "153"] [EventDate "2018.??.??"] [SetUp "1"] [FEN "8/8/6r1/p7/P2p1p1p/KPb1pPkP/4R1P1/1R6 w - - 0 75"] [Termination "unterminated"] [PlyCount "0"] [WhiteType "human"] [BlackType "human"] * [/pgn]

Stockfish 9 with two logical cores of the i7 980x, which allows calculating about 2.9 million nodes per second against 3300 of lc0 11261(GTX 1060 3Gb), with a ratio of:
LeR = 3300 * 875/2900000 = 0.996
Both played from Silver Test Suite without tablebases at 16 seconds per move.


[pgn][Event "Silver16sec"] [Site "Ros"] [Date "2018.10.12"] [Round "42"] [White "Stockfish9"] [Black "Lc011261"] [Result "0-1"] [ECO "B66"] [Annotator "R,J"] [PlyCount "153"] [EventDate "2018.??.??"] 1. e4 {book} c5 {book} 2. Nf3 {book} d6 {book} 3. d4 {book} cxd4 {book} 4. Nxd4 {book} Nf6 {book} 5. Nc3 {book} Nc6 {book} 6. Bg5 {book} e6 {book} 7. Qd2 {book } a6 {book} 8. O-O-O {book} h6 {book} 9. Be3 {0.42/24 16} Be7 {-0.37/8 16} 10. f3 {0.54/24 16} h5 {-0.39/9 14} 11. h3 {0.62/24 16} h4 {-0.32/8 9.9s} 12. Kb1 { 0.65/26 16} Bd7 {-0.31/8 10} 13. a3 {0.58/24 16} Rb8 {-0.10/8 13} 14. Be2 {0. 70/25 16} Qc7 {-0.01/9 9.8s} 15. Nxc6 {0.50/24 16} bxc6 {0.32/8 7.7s} 16. Bc4 { 0.32/24 16} Qb7 {0.42/9 7.3s} 17. Qc1 {0.00/28 16} d5 {0.39/11 6.3s} 18. Bb3 { 0.45/26 16} Rh5 {0.36/10 3.5s} 19. Bg5 {0.00/27 16} Rxg5 {1.31/14 7.7s} 20. Qxg5 {0.47/30 16} Nxe4 {1.19/16 4.4s} 21. Qxg7 {0.43/32 16} Bf6 {1.12/15 2.7s} 22. Nxe4 {0.37/34 16} Bxg7 {1.07/15 1.6s} 23. Nd6+ {0.47/32 16} Ke7 {1.06/14 5. 2s} 24. Nxb7 {0.30/34 16} Rxb7 {1.08/13 1.6s} 25. Rhe1 {0.21/32 16} a5 {1.19/9 7.4s} 26. a4 {0.40/32 16} f5 {1.09/9 16} 27. Ka2 {0.25/31 16} Kd6 {1.06/8 8.5s} 28. c3 {0.25/36 16} Rb8 {0.97/8 8.5s} 29. Ka3 {0.30/35 16} Bf8 {1.00/7 14} 30. Bc2 {0.31/32 16} e5 {1.22/8 8.3s} 31. Ka2 {0.35/32 16} Be7 {1.34/8 7.6s} 32. Re2 {0.16/28 16} Be6 {1.36/8 7.5s} 33. Bb1 {0.00/33 16} Bf6 {1.52/7 8.1s} 34. Bd3 {0.00/35 16} Ke7 {1.38/7 11} 35. Rde1 {0.00/36 16} Kd6 {1.31/9 12} 36. Rd1 {-0.07/34 16} Ke7 {1.30/8 11} 37. Ka1 {0.00/39 16} Rg8 {1.30/7 7.6s} 38. Rde1 { 0.00/36 16} Kd6 {1.30/8 6.9s} 39. Rd1 {0.00/40 16} Bh8 {1.25/7 14} 40. Kb1 {0. 00/36 16} Kc7 {1.22/8 9.9s} 41. Rf1 {0.00/36 16} Rb8 {1.25/7 11} 42. Kc1 {0.00/ 37 16} Bg7 {1.30/8 12} 43. Bc2 {0.00/35 16} c5 {1.56/8 12} 44. Rg1 {-0.07/35 16 } Bh6+ {2.11/11 6.9s} 45. Kb1 {-0.34/30 16} Bf4 {2.24/11 3.9s} 46. Bd3 {-0.50/ 32 16} d4 {2.26/11 7.7s} 47. cxd4 {-0.08/32 16} cxd4 {2.18/12 2.4s} 48. Rc2+ { -0.15/35 16} Kd6 {2.13/10 6.1s} 49. Re1 {-0.34/34 16} Rg8 {2.13/10 12} 50. b3 { -0.12/30 16} Rb8 {2.09/9 5.9s} 51. Rb2 {-0.45/30 16} Be3 {2.12/9 6.9s} 52. Ka2 {0.00/38 16} Rc8 {2.07/8 11} 53. Rc2 {0.00/33 16} Rf8 {1.95/11 11} 54. Rf1 {-0. 15/30 16} Bd5 {2.53/9 7.6s} 55. Rb2 {-0.08/30 16} Ke6 {2.49/10 5.1s} 56. Bc4 { -0.50/30 16} Rb8 {2.33/11 7.0s} 57. Rbb1 {-0.77/28 16} Bd2 {4.61/8 7.9s} 58. Bb5 {-1.55/30 16} Bc3 {4.94/8 8.7s} 59. Rbc1 {-1.82/32 16} e4 {4.70/10 8.0s} 60. Rf2 {-1.97/30 16} e3 {6.22/8 8.1s} 61. Rfc2 {-2.05/37 16} Rc8 {6.64/8 9.0s} 62. Re2 {-2.05/41 16} Ke5 {7.61/7 8.5s} 63. Ba6 {-2.05/42 16} Rc5 {7.78/8 8.3s} 64. Rec2 {-2.52/34 16} Kf4 {8.02/7 8.3s} 65. Rd1 {-3.16/32 16} Kg3 {8.71/7 7.0s } 66. Bd3 {-3.87/32 16} Be6 {9.30/7 8.4s} 67. Re2 {-4.32/33 16} Rc8 {10.30/7 13 } 68. Rc2 {-4.74/30 16} f4 {11.44/7 14} 69. Rdc1 {-5.32/30 16} Rb8 {13.12/7 8. 6s} 70. Rb1 {-6.05/32 16} Rb4 {12.36/7 13} 71. Ka3 {-6.50/34 16} Bf7 {12.57/8 15} 72. Re2 {-6.72/34 16} Rb6 {12.61/8 7.3s} 73. Be4 {-6.93/34 16} Bg6 {16.57/ 7 8.9s} 74. Bxg6 {-9.02/31 16} Rxg6 {15.42/8 6.4s} 75. Rd1 {-10.05/26 16} Kh2 { 15.56/8 8.7s} 76. g4+ {-10.61/26 16} Kg3 {19.45/7 9.6s} 77. Rc2 {-12.25/26 16s, Black wins by adjudication} 0-1 [/pgn]

[jdart]

It can play really nice positional chess. But it still seems to have occasional tactical blindness. The modern alpha-beta engines are so selective, they are throwing out huge pieces of the search tree. But it is still kind of amazing to me that these NN engines can play as well as they do, since they are effectively visiting even less of the tree.

--Jon

[peter]

Hi!

From one of my tests, lc0 played the opening novelty 10..h5

That's not a novelty. It's just transposing to other less seldom played move orders and therefore not in all books to be found at once, but it's even in Brainfish- Cerebellum too.
Not to be played actively but to be answered with 11.h3 by Brainfish.
Answering 11.Kb1 instead leads to lines in Hiarcs online and in Noomen- book again.

[Javier Ros]

Hi!

From one of my tests, lc0 played the opening novelty 10..h5

That's not a novelty. It's just transposing to other less seldom played move orders and therefore not in all books to be found at once, but it's even in Brainfish- Cerebellum too.
Not to be played actively but to be answered with 11.h3 by Brainfish.
Answering 11.Kb1 instead leads to lines in Hiarcs online and in Noomen- book again.

You are right, although it doesn't appear in the bigbases, after 11.h3 it appears by transposition, the novelty is 11..h4.

[MikeGL]

From one of my tests, lc0 played the opening novelty 10..h5 putting the rook in play in an original way and sacrificing it later for a bishop.

19. Bg5 {lc0 0.00/27 16} Rxg5 {Sf9 1.31/14 7.7s}

After 74..Rxg6 the next incredible position was reached, where Sf9 is cornered and lost.

[pgn][Event "Silver16sec"] [Site "Ros"] [Date "2018.10.12"] [Round "42"] [White "Stockfish9"] [Black "Lc011261"] [Result "0-1"] [ECO "B66"] [Annotator "R,J"] [PlyCount "153"] [EventDate "2018.??.??"] [SetUp "1"] [FEN "8/8/6r1/p7/P2p1p1p/KPb1pPkP/4R1P1/1R6 w - - 0 75"] [Termination "unterminated"] [PlyCount "0"] [WhiteType "human"] [BlackType "human"] * [/pgn]

Stockfish 9 with two logical cores of the i7 980x, which allows calculating about 2.9 million nodes per second against 3300 of lc0 11261(GTX 1060 3Gb), with a ratio of:
LeR = 3300 * 875/2900000 = 0.996
Both played from Silver Test Suite without tablebases at 16 seconds per move.


[pgn][Event "Silver16sec"] [Site "Ros"] [Date "2018.10.12"] [Round "42"] [White "Stockfish9"] [Black "Lc011261"] [Result "0-1"] [ECO "B66"] [Annotator "R,J"] [PlyCount "153"] [EventDate "2018.??.??"] 1. e4 {book} c5 {book} 2. Nf3 {book} d6 {book} 3. d4 {book} cxd4 {book} 4. Nxd4 {book} Nf6 {book} 5. Nc3 {book} Nc6 {book} 6. Bg5 {book} e6 {book} 7. Qd2 {book } a6 {book} 8. O-O-O {book} h6 {book} 9. Be3 {0.42/24 16} Be7 {-0.37/8 16} 10. f3 {0.54/24 16} h5 {-0.39/9 14} 11. h3 {0.62/24 16} h4 {-0.32/8 9.9s} 12. Kb1 { 0.65/26 16} Bd7 {-0.31/8 10} 13. a3 {0.58/24 16} Rb8 {-0.10/8 13} 14. Be2 {0. 70/25 16} Qc7 {-0.01/9 9.8s} 15. Nxc6 {0.50/24 16} bxc6 {0.32/8 7.7s} 16. Bc4 { 0.32/24 16} Qb7 {0.42/9 7.3s} 17. Qc1 {0.00/28 16} d5 {0.39/11 6.3s} 18. Bb3 { 0.45/26 16} Rh5 {0.36/10 3.5s} 19. Bg5 {0.00/27 16} Rxg5 {1.31/14 7.7s} 20. Qxg5 {0.47/30 16} Nxe4 {1.19/16 4.4s} 21. Qxg7 {0.43/32 16} Bf6 {1.12/15 2.7s} 22. Nxe4 {0.37/34 16} Bxg7 {1.07/15 1.6s} 23. Nd6+ {0.47/32 16} Ke7 {1.06/14 5. 2s} 24. Nxb7 {0.30/34 16} Rxb7 {1.08/13 1.6s} 25. Rhe1 {0.21/32 16} a5 {1.19/9 7.4s} 26. a4 {0.40/32 16} f5 {1.09/9 16} 27. Ka2 {0.25/31 16} Kd6 {1.06/8 8.5s} 28. c3 {0.25/36 16} Rb8 {0.97/8 8.5s} 29. Ka3 {0.30/35 16} Bf8 {1.00/7 14} 30. Bc2 {0.31/32 16} e5 {1.22/8 8.3s} 31. Ka2 {0.35/32 16} Be7 {1.34/8 7.6s} 32. Re2 {0.16/28 16} Be6 {1.36/8 7.5s} 33. Bb1 {0.00/33 16} Bf6 {1.52/7 8.1s} 34. Bd3 {0.00/35 16} Ke7 {1.38/7 11} 35. Rde1 {0.00/36 16} Kd6 {1.31/9 12} 36. Rd1 {-0.07/34 16} Ke7 {1.30/8 11} 37. Ka1 {0.00/39 16} Rg8 {1.30/7 7.6s} 38. Rde1 { 0.00/36 16} Kd6 {1.30/8 6.9s} 39. Rd1 {0.00/40 16} Bh8 {1.25/7 14} 40. Kb1 {0. 00/36 16} Kc7 {1.22/8 9.9s} 41. Rf1 {0.00/36 16} Rb8 {1.25/7 11} 42. Kc1 {0.00/ 37 16} Bg7 {1.30/8 12} 43. Bc2 {0.00/35 16} c5 {1.56/8 12} 44. Rg1 {-0.07/35 16 } Bh6+ {2.11/11 6.9s} 45. Kb1 {-0.34/30 16} Bf4 {2.24/11 3.9s} 46. Bd3 {-0.50/ 32 16} d4 {2.26/11 7.7s} 47. cxd4 {-0.08/32 16} cxd4 {2.18/12 2.4s} 48. Rc2+ { -0.15/35 16} Kd6 {2.13/10 6.1s} 49. Re1 {-0.34/34 16} Rg8 {2.13/10 12} 50. b3 { -0.12/30 16} Rb8 {2.09/9 5.9s} 51. Rb2 {-0.45/30 16} Be3 {2.12/9 6.9s} 52. Ka2 {0.00/38 16} Rc8 {2.07/8 11} 53. Rc2 {0.00/33 16} Rf8 {1.95/11 11} 54. Rf1 {-0. 15/30 16} Bd5 {2.53/9 7.6s} 55. Rb2 {-0.08/30 16} Ke6 {2.49/10 5.1s} 56. Bc4 { -0.50/30 16} Rb8 {2.33/11 7.0s} 57. Rbb1 {-0.77/28 16} Bd2 {4.61/8 7.9s} 58. Bb5 {-1.55/30 16} Bc3 {4.94/8 8.7s} 59. Rbc1 {-1.82/32 16} e4 {4.70/10 8.0s} 60. Rf2 {-1.97/30 16} e3 {6.22/8 8.1s} 61. Rfc2 {-2.05/37 16} Rc8 {6.64/8 9.0s} 62. Re2 {-2.05/41 16} Ke5 {7.61/7 8.5s} 63. Ba6 {-2.05/42 16} Rc5 {7.78/8 8.3s} 64. Rec2 {-2.52/34 16} Kf4 {8.02/7 8.3s} 65. Rd1 {-3.16/32 16} Kg3 {8.71/7 7.0s } 66. Bd3 {-3.87/32 16} Be6 {9.30/7 8.4s} 67. Re2 {-4.32/33 16} Rc8 {10.30/7 13 } 68. Rc2 {-4.74/30 16} f4 {11.44/7 14} 69. Rdc1 {-5.32/30 16} Rb8 {13.12/7 8. 6s} 70. Rb1 {-6.05/32 16} Rb4 {12.36/7 13} 71. Ka3 {-6.50/34 16} Bf7 {12.57/8 15} 72. Re2 {-6.72/34 16} Rb6 {12.61/8 7.3s} 73. Be4 {-6.93/34 16} Bg6 {16.57/ 7 8.9s} 74. Bxg6 {-9.02/31 16} Rxg6 {15.42/8 6.4s} 75. Rd1 {-10.05/26 16} Kh2 { 15.56/8 8.7s} 76. g4+ {-10.61/26 16} Kg3 {19.45/7 9.6s} 77. Rc2 {-12.25/26 16s, Black wins by adjudication} 0-1 [/pgn]

Thanks for this instructive endgame. Funny how a B pair of LC0 defeats R-pair of SF9.
Amazingly done by ensuring few or no open files are made by Lc0 so as to make the rooks weaker than bishops.

[peter]

You are right, although it doesn't appear in the bigbases, after 11.h3 it appears by transposition, the novelty is 11..h4.

Also only if you don't look for transpositions in online- database, e.g. chessbase.
After 12 moves of the engine- game that transposes to this one after 11 moves. Difference comes from LC0 playing h6-h5 in two steps, which may be new, but one could simply call it loss of tempo too, even if LC0 is not to blame for that fully, having had 8...h6 in "book".
[pgn] [Event "NED-ch62"] [Site "Hilversum"] [Date "2007.06.23"] [Round "7"] [White "Nijboer, Friso"] [Black "Bosboom, Manuel"] [Result "1-0"] [ECO "B54"] [WhiteElo "2606"] [BlackElo "2423"] [PlyCount "82"] [EventDate "2007.06.16"] [EventType "tourn"] [EventRounds "11"] [EventCountry "NED"] [EventCategory "12"] [Source "ChessBase"] [SourceDate "2007.08.01"] 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 e6 5. Nc3 a6 6. Be3 Bd7 7. Qd2 Nc6 8. O-O-O Nf6 9. f3 h5 10. Kb1 Be7 11. h3 h4 12. f4 b5 13. Bd3 b4 14. Nce2 Qb8 15. Nf3 a5 16. Bf2 a4 17. Bxh4 b3 18. cxb3 Rxh4 19. Nxh4 Nb4 20. a3 Nxd3 21. Qxd3 Bc6 22. Ng3 d5 23. Qc3 Qb7 24. Nf3 dxe4 25. Nd4 Bd5 26. b4 Rc8 27. Qe3 Bd8 28. Rc1 Rc4 29. Nge2 Bb6 30. Rhd1 Qa7 31. Rc3 g6 32. Ka1 Kf8 33. Rxc4 Bxc4 34. Qc3 Bd5 35. Nc6 Qd7 36. b5 Ng8 37. Qb4+ Kg7 38. Nc3 e3 39. Nxa4 e2 40. Re1 Bc4 41. Qc3+ Nf6 1-0 [/pgn]

And we are in Noomen- book again and in Cerebellum too.

[Guenther]

...
After 12 moves of the engine- game that transposes to this one after 11 moves. Difference comes from LC0 playing h6-h5 in two steps, which may be new, but one could simply call it loss of tempo too, even if LC0 is not to blame for that fully, having had 8...h6 in "book".
...
And we are in Noomen- book again and in Cerebellum too.

There is no loss of tempo, you missed that a full move pair is inserted with Bg5 h6.

[peter]

There is no loss of tempo, you missed that a full move pair is inserted with Bg5 h6.

There is loss of tempo yet, but for both sides compared to Nijboer-Bosboom.
That's why the same position is reached one move earlier there with same side to move, that's what I meant.