Test Position bm Rxc3

[MikeGL]

This position arised after SF10 played 27.Kf2 and the winning response for black(Lc0) is [bm 27...Rxc3!]
[d]1kr1b3/pr6/3pq3/n2p1p2/3P1Pp1/P1N1P1Pp/RP1NQK1P/4R3 b - - 3 27
I tried to analyze where SF10 went wrong after Rxc3, but looks like it made excellent defense but is really lost after exchange sacrifice of black.
Positional dominance of black as white rooks are tamed by minor pieces. N stronger than R on this setup, and bad bishop nicely activated.
Similar to one of those brilliancy games of top GMs. This positional exchange sacrifice is deep.

Complete game:
[pgn] [Event "?"] [Site "?"] [Date "????.??.??"] [Round "?"] [White "Stockfish 10 181129"] [Black "Lc0 0.21.1 N:42100"] [Result "0-1"] [TimeControl "5400+5"] [SetUp "1"] [FEN "rn1qkbnr/ppp2ppp/8/3p1b2/3P4/5N2/PP2PPPP/RNBQKB1R w KQkq - 0 1"] [Termination "normal"] [PlyCount "141"] 1. Qb3 Nc6 2. Bg5 f6 3. Bd2 Qd7 4. Nc3 O-O-O 5. Na4 Bd6 6. e3 Nge7 7. a3 h5 8. Bb4 Kb8 9. Bxd6 cxd6 10. Nc3 h4 11. Nd2 Qe6 12. Ba6 Na5 13. Qb5 Rd7 14. Qxa5 bxa6 15. Qxa6 h3 16. g3 Rb7 17. Ra2 g5 18. O-O Rc8 19. Re1 Bg6 20. Qf1 Nc6 21. Raa1 Ka8 22. Ra2 Kb8 23. f3 Na5 24. Qf2 f5 25. f4 g4 26. Qe2 Be8 27. Kf2 Rxc3 28. bxc3 Bb5 29. Qd1 Bc4 30. Rc2 Bd3 31. Ra2 Nc4 32. Qa1 Kc7 33. Kg1 a5 34. Qd1 Nb2 35. Qc1 Na4 36. Qd1 Qe8 37. Nf1 Qb5 38. Rf2 Nb2 39. Qa1 a4 40. Rc1 Nc4 41. Rd1 Bxf1 42. Rdxf1 Qe8 43. Qd1 Rb3 44. Qc2 Qe4 45. Ra1 Kc6 46. Re2 Kb6 47. Qc1 Kb5 48. Rf2 Qd3 49. Qc2 Qxc3 50. Qxc3 Rxc3 51. Rfa2 Nxe3 52. Rb2+ Rb3 53. Rd2 Nc4 54. Rda2 Nb2 55. Rc1 Nd3 56. Rd1 Rc3 57. Re2 Kc4 58. Re7 Rxa3 59. Rc7+ Kxd4 60. Rc2 Rc3 61. Ra2 a3 62. Rda1 Ke4 63. Rxa3 d4 64. Ra6 Kd5 65. R6a2 Nc5 66. Kf2 Rb3 67. Ke2 Ke4 68. Kf2 d3 69. Ra4+ Nxa4 70. Rxa4+ Kd5 71. Ke3 0-1 [/pgn]

[AdminX]

Leela (JH.T6.532) does not care for 35. Qc1

Code: Select all

Stockfish 10 181129 - Lc0 0.21.1 N:42100 0-1
8/1rk5/3pq3/p2p1p2/3P1Pp1/P1PbP1Pp/Rn1N3P/3QR1K1 w - - 0 1

Analysis by Lc0 v0.22.0-Dev-JHT:

1. =/+ (-0.53): 35.Qa1 Na4 36.e4 dxe4 37.c4 Kb8 38.d5 Qc8 39.Qd4 Ka8 40.Kf2 Nc5 41.Ke3 Qb8 42.Rc1 Qa7 43.Qh8+ Rb8 44.Qd4 a4 45.Rd1 Qb6 46.Rc1 Rb7 47.Rc3 Ka7 48.Rc1 Ka6 49.Rc3 Rb8 50.Rc1 Qa7 51.Rc3 Qe7 52.Rc1 
2. -/+ (-0.74): 35.Qc1 Na4 36.Qd1 Qe8 37.Qc1 Qc6 38.Qa1 Kb8 39.Rc1 Ka7 40.Kf2 Rb6 41.Kg1 Ka6 42.Kf2 Rb8 43.Kg1 Qe8 
Black is slightly better

[MikeGL]

Leela (JH.T6.532) does not care for 35. Qc1

Code: Select all

Stockfish 10 181129 - Lc0 0.21.1 N:42100 0-1
8/1rk5/3pq3/p2p1p2/3P1Pp1/P1PbP1Pp/Rn1N3P/3QR1K1 w - - 0 1

Analysis by Lc0 v0.22.0-Dev-JHT:

1. =/+ (-0.53): 35.Qa1 Na4 36.e4 dxe4 37.c4 Kb8 38.d5 Qc8 39.Qd4 Ka8 40.Kf2 Nc5 41.Ke3 Qb8 42.Rc1 Qa7 43.Qh8+ Rb8 44.Qd4 a4 45.Rd1 Qb6 46.Rc1 Rb7 47.Rc3 Ka7 48.Rc1 Ka6 49.Rc3 Rb8 50.Rc1 Qa7 51.Rc3 Qe7 52.Rc1 
2. -/+ (-0.74): 35.Qc1 Na4 36.Qd1 Qe8 37.Qc1 Qc6 38.Qa1 Kb8 39.Rc1 Ka7 40.Kf2 Rb6 41.Kg1 Ka6 42.Kf2 Rb8 43.Kg1 Qe8 
Black is slightly better

Thanks for the line. I checked and analyzed it using engine and it seems black will still have a strong winning position after 37.c4 then not Kb8 but 37...Kb6! and win for black looks inevitable.
Funny to see until at the end of your first line, both white rooks were still unable to challenge/occupy the only open b-file because of the bishop controlling b1 square.

[chrisw]

30 Ra1 looks better, after Rc2 it seems black just gets a huge grip on the position

[chrisw]

Yup, 30 Ra1 and white is okay, but no computer will see it. Funny position.

[jp]

This position arised after SF10 played 27.Kf2 and the winning response for black(Lc0) is [bm 27...Rxc3!]

What roughly were the nodes per move for SF10 & Lc0 (in the game)?

[MikeGL]

This position arised after SF10 played 27.Kf2 and the winning response for black(Lc0) is [bm 27...Rxc3!]

What roughly were the nodes per move for SF10 & Lc0 (in the game)?

Sorry, forgot to mention the credit line.
Credit goes to Stefan Pohl, that was not my Lc0 testing,

I was just analyzing these types of games. Since I feel these are almost perfect games, just so I get a positional feel on my real OTB games against strong opposition.

Here's the source of the game and his hardware setup:
Stefan Pohl
Hardware details of Stefan for testing these engines
https://www.sp-cc.de/lc0-testing.htm

//////////////////////// copied from the above link //////////////////////
Hardware: i7-8750H (Hexacore) Notebook with RTX 2060 GPU, Windows 10 64bit, 16GB RAM

CPU-Speed: Stockfish with 97% CPU-Speed (to switch off the Intel Turbo Boost): 7.5 MN/s in starting-position, running on 11 threads.

GPU (used by LC Zero): Nvidia RTX 2060 (6GB). LC Zero calculates around 11500 n/s in the starting position (I used the MSI-Afterburner-tool to reduce the speed of the RTX-Card as far as possible) (measured with "go infinte") with Netsize 20x256, which means a Leela-Ratio (what is Leela Ratio? look here) of 1.3.The Leela-Ratio-value of AlphaZero (used a 20x256 net, too) in the match vs. Stockfish 8 was 1.0 - so 1.3 is a high value, but acceptable.

Hash: 1 GB for AB-engines and 1.000.000 size of NNCache for Leela
GUI: LittleBlitzerGUI, draw at 160 moves, resign at a -700 cp evaluation.
Tablebases: none
Openings: 4 SuperGM-moves. Download my 4-moves opening-sets and books here
Large Memory Pages: Off
Ponder: Off
Thinking time: 150'' (2.5 minutes) +1500ms (average game-duration: 8-9 minutes)
/////////////////////


Source PGN of this game:
https://www.sp-cc.de/files/nn_engine_games.zip
4,054th game entry from that PGN

[MikeGL]

30 Ra1 looks better, after Rc2 it seems black just gets a huge grip on the position

I tried to run analysis with MultiPV and just ends up with B vs R endgame, but a pawn up and slightly better for black.
Better space and possible K activity compared to K of white, but might require deeper analysis.

[pgn] [Event "?"] [Site "?"] [Date "2019.05.07"] [Round "?"] [White "SF"] [Black "SF"] [Result "*"] [BlackElo "2400"] [WhiteElo "2400"] [TimeControl "5400+5"] [SetUp "1"] [FEN "1k6/pr6/3pq3/n2p1p2/2bP1Pp1/P1P1P1Pp/R2N1K1P/3QR3 w - - 0 30"] [Termination "unterminated"] [PlyCount "14"] 30. Ra1 Bd3 31. e4 dxe4 32. Qa4 Nb3 33. Rad1 a5 34. Nxb3 Qxb3 35. Qxb3 Rxb3 36. Ke3 Rxc3 * [/pgn]

[chrisw]

30 Ra1 looks better, after Rc2 it seems black just gets a huge grip on the position

I tried to run analysis with MultiPV and just ends up with B vs R endgame, but a pawn up and slightly better for black.
Better space and possible K activity compared to K of white, but might require deeper analysis.

[pgn] [Event "?"] [Site "?"] [Date "2019.05.07"] [Round "?"] [White "SF"] [Black "SF"] [Result "*"] [BlackElo "2400"] [WhiteElo "2400"] [TimeControl "5400+5"] [SetUp "1"] [FEN "1k6/pr6/3pq3/n2p1p2/2bP1Pp1/P1P1P1Pp/R2N1K1P/3QR3 w - - 0 30"] [Termination "unterminated"] [PlyCount "14"] 30. Ra1 Bd3 31. e4 dxe4 32. Qa4 Nb3 33. Rad1 a5 34. Nxb3 Qxb3 35. Qxb3 Rxb3 36. Ke3 Rxc3 * [/pgn]

e4 is absolute key move, cutting blacks ability to pivot on that square.

30. Ra1 Bd3 31. e4 dxe4 32. Qa4 Nb3 33. Ra2 is better than the above Rad1 (which threatens nothing and allows black to improve his position)

after 33. Ra2 Nc1 is just drawing, Nc5 also a draw. NxN is draw, some positions are like your line, but a vital tempo ahead (various defences, including Ke3 and counter exchange sac, the tempo is important here)
a6 would be an attempt to keep pieces on the board, but it isn't enough and the position holds, I think.

Key was e4, it gave the king a blockade square on e3, prevented the mating threat lines with Qe4 and gained just enough time. Therefore draw, not decisive black win by strangulation.

[peter]

Hi Mike!
Thanks for the position!

Leela (JH.T6.532) does not care for 35. Qc1

Code: Select all

Stockfish 10 181129 - Lc0 0.21.1 N:42100 0-1
8/1rk5/3pq3/p2p1p2/3P1Pp1/P1PbP1Pp/Rn1N3P/3QR1K1 w - - 0 1

Analysis by Lc0 v0.22.0-Dev-JHT:

1. =/+ (-0.53): 35.Qa1 Na4 36.e4 dxe4 37.c4 Kb8 38.d5 Qc8 39.Qd4 Ka8 40.Kf2 Nc5 41.Ke3 Qb8 42.Rc1 Qa7 43.Qh8+ Rb8 44.Qd4 a4 45.Rd1 Qb6 46.Rc1 Rb7 47.Rc3 Ka7 48.Rc1 Ka6 49.Rc3 Rb8 50.Rc1 Qa7 51.Rc3 Qe7 52.Rc1 
2. -/+ (-0.74): 35.Qc1 Na4 36.Qd1 Qe8 37.Qc1 Qc6 38.Qa1 Kb8 39.Rc1 Ka7 40.Kf2 Rb6 41.Kg1 Ka6 42.Kf2 Rb8 43.Kg1 Qe8 
Black is slightly better

Thanks for the line. I checked and analyzed it using engine and it seems black will still have a strong winning position after 37.c4 then not Kb8 but 37...Kb6! and win for black looks inevitable.

How about 36.Nf1 instead of e4?
After deviation 31.Qa1 Na4 32.Nf1 Fish doesn't see any progress for Black anymore.
(
1. Qb3 Nc6 2. Bg5 f6 3. Bd2 Qd7 4. Nc3 O-O-O 5. Na4 Bd6 6. e3 Nge7 7. a3 h5 8.
Bb4 Kb8 9. Bxd6 cxd6 10. Nc3 h4 11. Nd2 Qe6 12. Ba6 Na5 13. Qb5 Rd7 14. Qxa5
bxa6 15. Qxa6 h3 16. g3 Rb7 17. Ra2 g5 18. O-O Rc8 19. Re1 Bg6 20. Qf1 Nc6 21.
Raa1 Ka8 22. Ra2 Kb8 23. f3 Na5 24. Qf2 f5 25. f4 g4 26. Qe2 Be8

27. Kf2 Rxc3 28. bxc3 Bb5 29. Qd1 Bc4 30. Rc2 Bd3 31. Ra2 Nc4 32. Qa1 Kc7 33. Kg1 a5 34. Qd1
Nb2

35. Qa1 Na4 36. Nf1
)

[d]8/1rk5/3pq3/p2p1p2/n2P1Pp1/P1PbP1Pp/R6P/Q3RNK1 b - - 0 1
Analysis by CF OPT 180419 x64 POP NUMA:

36...De4 37.Tf2 Sb6 38.a4 Sc4 39.Sd2 De8 40.e4 Lxe4 41.Tef1 Tb6 42.Tc1 Dc6 43.Te1 Kd7 44.Tc1 Ke7 45.Te1
= (0.00) Tiefe: 54/34 00:06:21 8123MN

After some Forward- Backward with 24 threads of 12x3GHz CPU, 32G hash