Win or draw

[Lyudmil Tsvetkov]

In most cases a reasonably good chess judgement will tell apart a win from loss for one side, and even a loss from draw and a draw from a win.

However, there are a range of positions in which neither humans, nor current top engines will be able to definitely assert what is the state of the game. Reason for this might be that the position is very deep, very peculiar, has lots of branching lines, etc. Your judgement tells you so and so, but, in actual case, the truth might lie in a single refutation line out of the zillion possible, so everybody is prone to wrong conclusions.

[d]r2qk2r/1p1b1pb1/3p2p1/p1n1pPP1/2PpP2p/3P3P/PP2N1B1/R1BQ1RK1 w kq - 0 16

Above position is one of those. I am to 85% certain how the game should end, but, with multiple reasonable ramifying lines, some very deep at that, it is indeed very difficult for me to claim I know for certain what the outcome is. So I would request your and your engines' opinion about above position.

Who wins above, or is it a draw?

[Eelco de Groot]

I could find nothing threatening for Black, sorry Lyudmil. Maybe if you gave a few of those deep lines, but Stockfish thinks it is a dead draw...

Multi PV, four best moves.

r2qk2r/1p1b1pb1/3p2p1/p1n1pPP1/2PpP2p/3P3P/PP2N1B1/R1BQ1RK1 w kq -

Engine: Rainbow Serpent 20160903_014 HT (512 MB)
Latest Serpent version is 016. 3 hyperthreads and four normal threads, one hyperthread left for other uses and more responsiveness when leaving the engine calculating for long. One hyperthread is not enough for most things of course.
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott

35 417:34 -0.25 1.f6 Bf8 2.b3 b5 3.Kh2 Qb8 4.Ba3 b4
5.Bb2 a4 6.bxa4 Rxa4 7.Rg1 Qa8 8.Qd2 Qc8
9.Rgb1 Kd8 10.Rg1 Qa6 11.Nc1 Ne6
12.Bf3 Nf4 13.Bg4 Bxg4 14.Rxg4 (112.182.098.635) 4477

35 417:34 -0.35 1.Kh2 f6 2.fxg6 fxg5 3.Rf7 Bf6 4.Qf1 Rf8
5.Rxf8+ Kxf8 6.Bxg5 Kg7 7.Bd2 Qf8
8.Ng1 Kxg6 9.Nf3 Qh8 10.Kh1 Kf7
11.b4 axb4 12.Bxb4 b6 13.Nh2 Qh6
14.Rb1 (112.182.098.635) 4477

35 417:34 -0.62 1.b3 f6 2.fxg6 Qe7 3.Rf5 Qe6 4.Qd2 fxg5
5.Qxg5 Rh6 6.Qg4 Rxg6 7.Bg5 Qf7
8.Raf1 Bxf5 9.exf5 Rxg5 10.Qxg5 Bf6
11.Qh6 O-O-O 12.Qd2 Rg8 13.Kh1 Qc7
14.Bd5 (112.182.098.635) 4477

34 417:34 -0.66 1.Kh1 gxf5 2.exf5 Qc8 3.f6 Bf8 4.Rf3 Bc6
5.Qf1 Rg8 6.Ng1 Qd7 7.b3 O-O-O 8.Rf5 Bxg2+
9.Kxg2 Re8 10.Kh2 Kb8 11.Rb1 Qe6
12.Bb2 d5 13.cxd5 Qd7 14.Rd1 (112.182.098.635) 4477

best move: f5-f6 time: 418:26.594 min n/s: 4.477.538 nodes: 112.414.615.629

[Milos]

It's a draw. Most lines end by perpetual. White can force it.

[Lyudmil Tsvetkov]

I could find nothing threatening for Black, sorry Lyudmil. Maybe if you gave a few of those deep lines, but Stockfish thinks it is a dead draw...

Multi PV, four best moves.

r2qk2r/1p1b1pb1/3p2p1/p1n1pPP1/2PpP2p/3P3P/PP2N1B1/R1BQ1RK1 w kq -

Engine: Rainbow Serpent 20160903_014 HT (512 MB)
Latest Serpent version is 016. 3 hyperthreads and four normal threads, one hyperthread left for other uses and more responsiveness when leaving the engine calculating for long. One hyperthread is not enough for most things of course.
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott

35 417:34 -0.25 1.f6 Bf8 2.b3 b5 3.Kh2 Qb8 4.Ba3 b4
5.Bb2 a4 6.bxa4 Rxa4 7.Rg1 Qa8 8.Qd2 Qc8
9.Rgb1 Kd8 10.Rg1 Qa6 11.Nc1 Ne6
12.Bf3 Nf4 13.Bg4 Bxg4 14.Rxg4 (112.182.098.635) 4477

35 417:34 -0.35 1.Kh2 f6 2.fxg6 fxg5 3.Rf7 Bf6 4.Qf1 Rf8
5.Rxf8+ Kxf8 6.Bxg5 Kg7 7.Bd2 Qf8
8.Ng1 Kxg6 9.Nf3 Qh8 10.Kh1 Kf7
11.b4 axb4 12.Bxb4 b6 13.Nh2 Qh6
14.Rb1 (112.182.098.635) 4477

35 417:34 -0.62 1.b3 f6 2.fxg6 Qe7 3.Rf5 Qe6 4.Qd2 fxg5
5.Qxg5 Rh6 6.Qg4 Rxg6 7.Bg5 Qf7
8.Raf1 Bxf5 9.exf5 Rxg5 10.Qxg5 Bf6
11.Qh6 O-O-O 12.Qd2 Rg8 13.Kh1 Qc7
14.Bd5 (112.182.098.635) 4477

34 417:34 -0.66 1.Kh1 gxf5 2.exf5 Qc8 3.f6 Bf8 4.Rf3 Bc6
5.Qf1 Rg8 6.Ng1 Qd7 7.b3 O-O-O 8.Rf5 Bxg2+
9.Kxg2 Re8 10.Kh2 Kb8 11.Rb1 Qe6
12.Bb2 d5 13.cxd5 Qd7 14.Rd1 (112.182.098.635) 4477

best move: f5-f6 time: 418:26.594 min n/s: 4.477.538 nodes: 112.414.615.629

thanks Eelco.

I wonder what SF static eval above would be, if even after search it still sees some black advantage.

My static eval of the diagrammed position would be some 100cps white edge, so probably winning, by there are some long tricky lines, so I could not be certain.

No deep lines from me, just general considerations: after f5-f6 the black bishop on f8 is trapped, which should incur a huge penalty, a whole piece out of the game. At some point, no matter what black plays and how long it takes, 30-50 moves (100 plies), white will have the upperhand with a piece more. And only way to revive the bishop is by playing d6-d5, sacrificing a central pawn.

[Lyudmil Tsvetkov]

It's a draw. Most lines end by perpetual. White can force it.

Quoth the engine: 'It's a draw!'
Giving forth an output raw
It should all end by perpetual
It's my understanding natural.

Man, I do not trust engines...

after f5-f6, the black f8 bishop is trapped, how can one side draw with a piece less?

[Milos]

I wonder what SF static eval above would be, if even after search it still sees some black advantage.

My static eval of the diagrammed position would be some 100cps white edge, so probably winning, by there are some long tricky lines, so I could not be certain.

No deep lines from me, just general considerations: after f5-f6 the black bishop on f8 is trapped, which should incur a huge penalty, a whole piece out of the game. At some point, no matter what black plays and how long it takes, 30-50 moves (100 plies), white will have the upperhand with a piece more. And only way to revive the bishop is by playing d6-d5, sacrificing a central pawn.

Sorry but you got it all wrong. Black has and will have advantage all the way till white doesn't equalize either by perpetual or by endgame (TB draw).
After given line by SF - 16.f6 Bf8 17.b3 b5 18.Kh2 Qb8 19.Ba3 b4
20.Bb2 a4 21.bxa4 Rxa4 22.Rg1 Qa8 23.Qd2 Qc8
24.Rgb1 Kd8 25.Rg1 Qa6 26.Nc1 Ne6
27.Bf3 Nf4 28.Bg4 Bxg4 29.Rxg4
the position is closed and there is just infinite shuffling unless white pushes "a" pawn which results in perpetual.
I don't analyse by letting engine analyse till eternity. I pick myself the lines and follow them very deep and then trace back and force the score from hash. This way you can make the engine "see" beyond infinite shuffling.
White's black bishop also gets eventually voluntarily "trapped" on a1 to prevent white from losing so both sides are equal in trapped pieces.

[tpoppins]

I don't analyse by letting engine analyse till eternity. I pick myself the lines and follow them very deep and then trace back and force the score from hash. This way you can make the engine "see" beyond infinite shuffling.

That's by far the best way to conduct computer analysis. However if you are out working 12 hours a day and the position doesn't seem too inspiring sometimes the best you can manage is something like this:

[d]r2qk2r/1p1b1pb1/3p2p1/p1n1pPP1/2PpP2p/3P3P/PP2N1B1/R1BQ1RK1 w kq - 0 16

Code: Select all

Analysis by CorChess beta 151016 x64 POPCNT (16 GB hash/11 threads):

1. =   (-0.20): 16.f6 Bf8 17.Kh2 b5 18.b3 Qb6 19.Ba3 b4 20.Bb2 a4 21.bxa4 Rh5 22.Qd2 Ne6 23.Bf3 Rxg5 24.Qe1 Nc5 25.Qxh4 Nxd3 26.Bxd4 exd4 27.Qxg5 Rxa4 28.Bg2 Qc5 29.Qd5 Bh6 30.Rad1 Qxd5 31.cxd5 Bb5 32.Nxd4 Bc4 33.Rf3 Ne5 34.Rf2 Rxa2 35.Rxa2 Bxa2 36.Ne2 Bg5 37.Kg3 Nc4 38.Nc1 Ne3 39.Nxa2 Nxd1 40.Nxb4 Kd7 41.Bf3 Nc3 42.Nd3 Bxf6 43.Kf4 Kc7 44.Ke3 
2. =/+ (-0.52): 16.Kh2 f6 17.fxg6 fxg5 18.Rf7 Bf6 19.Qf1 Rf8 20.Bxg5 Rxf7 21.gxf7+ Kxf7 22.Bxh4 Kg6 23.Bxf6 Qxf6 24.Qxf6+ Kxf6 25.Rd1 Rh8 26.Rd2 b5 27.cxb5 Bxb5 28.Nc1 a4 29.a3 Bd7 30.Rf2+ Ke7 31.Rd2 Ne6 32.Ne2 Rg8 33.Bf1 Rf8 34.Bg2 Rh8 35.Bf1 Rg8 36.Bg2 Rg6 37.Bf1 Rf6 38.Bg2 Rf8 39.Kg3 Kf6 40.Kh2 Rg8 41.Ng1 Bb5 42.Bf1 Rc8 43.Ne2 Rh8 44.Kg3 Rg8+ 45.Kh2 Ng5 46.Ng1 Ke7 47.h4 Rh8 48.Kg3 
Depth: 51/78   20:37:05  815462 MN, tb=3741944

Code: Select all

Analysis by Houdini 4 Pro x64 B (Tactical/16 GB hash/16 threads):

[...]

16.f6 Bf8 17.b3 d5 18.cxd5 Bb5 19.Rf3 Qb6 20.Bb2 Rh5 21.Rc1 Rxg5 22.Kh1 Rc8 23.Bf1 Rh5 24.Ng1 Rd8 25.Rc2 a4 26.b4 Nd7 27.a3 Nxf6 28.Bc1 Be7 29.Rff2 Kf8 30.Nf3 Kg7 31.Nd2 Nd7 32.Be2 Rhh8 33.Nc4 Bxc4 34.Rxc4 
  = (-0.21)  Depth: 30/95   03:36:12  331572MN, tb=1500
16.f6 
  = (-0.13 ++)  Depth: 31/95   04:06:45  378204MN, tb=1509
16.f6 Bf8 17.b3 d5 18.exd5 Qc8 19.Rf3 Bd6 20.Kh1 Kf8 21.Ba3 Kg8 22.Ng1 Rh5 23.Qc1 Bf5 24.Qd2 b6 25.Raf1 Qd7 26.Bc1 Re8 27.Rxf5 gxf5 28.Qd1 Rh8 29.Nf3 f4 30.Kg1 Qf5 31.Ba3 Kh7 32.Qe2 e4 33.dxe4 d3 34.Qe1 Rxe4 35.Qxh4+ Kg8 36.Qg4 
  = (-0.16)  Depth: 31/95   04:27:57  410520MN, tb=1519
16.f6 Bf8 
  = (-0.24 --)  Depth: 32/95   05:34:21  517268MN, tb=1598
16.f6 Bf8 17.b3 d5 18.exd5 Qc8 19.Rf3 Bd6 20.Ba3 Rh5 21.Qd2 Kf8 22.Kh1 Kg8 23.Ng1 b5 24.Bxc5 Bxc5 25.Rd1 Bb4 26.Qc1 a4 27.Rf2 axb3 28.axb3 Bd6 29.Bf3 Rh8 30.Kg2 Kh7 31.Re1 Ra3 32.Qd1 Kg8 33.Qc2 Bf5 34.cxb5 Qb8 35.Be4 Bd7 36.Nf3 
  = (-0.16)  Depth: 32/95   09:21:39  878764MN, tb=3262
16.f6 Bf8 17.b3 b5 18.Rb1 Rb8 19.Kh2 bxc4 20.dxc4 a4 21.b4 Qc7 22.Bd2 Ne6 23.Qc1 Rc8 24.Ng1 Qc6 25.Re1 Qxc4 26.Qxc4 Rxc4 27.Rec1 Rxc1 28.Rxc1 Bb5 29.Nf3 Kd7 30.Kg1 Bc6 31.Re1 Kc7 32.Bf1 Kb6 33.Kf2 Nd8 34.Bd3 Bd7 35.Rh1 Rh5 36.Ke2 Be6 37.a3 Bd7 
  = (-0.20)  Depth: 33/95   11:39:43  1101907MN, tb=3309
16.f6 
  = (-0.12 ++)  Depth: 34/95   19:37:13  1866621MN, tb=9634

Total time: 21:16:28

[Lyudmil Tsvetkov]

I wonder what SF static eval above would be, if even after search it still sees some black advantage.

My static eval of the diagrammed position would be some 100cps white edge, so probably winning, by there are some long tricky lines, so I could not be certain.

No deep lines from me, just general considerations: after f5-f6 the black bishop on f8 is trapped, which should incur a huge penalty, a whole piece out of the game. At some point, no matter what black plays and how long it takes, 30-50 moves (100 plies), white will have the upperhand with a piece more. And only way to revive the bishop is by playing d6-d5, sacrificing a central pawn.

Sorry but you got it all wrong. Black has and will have advantage all the way till white doesn't equalize either by perpetual or by endgame (TB draw).
After given line by SF - 16.f6 Bf8 17.b3 b5 18.Kh2 Qb8 19.Ba3 b4
20.Bb2 a4 21.bxa4 Rxa4 22.Rg1 Qa8 23.Qd2 Qc8
24.Rgb1 Kd8 25.Rg1 Qa6 26.Nc1 Ne6
27.Bf3 Nf4 28.Bg4 Bxg4 29.Rxg4
the position is closed and there is just infinite shuffling unless white pushes "a" pawn which results in perpetual.
I don't analyse by letting engine analyse till eternity. I pick myself the lines and follow them very deep and then trace back and force the score from hash. This way you can make the engine "see" beyond infinite shuffling.
White's black bishop also gets eventually voluntarily "trapped" on a1 to prevent white from losing so both sides are equal in trapped pieces.

As usual, I am rigth.

I requested computer help from the big guns, as I myself do not quite have the time to do a full analysis of the position.

It stems from one of my engine games, and I am pretty much certain in my eval assessment. I also ran a couple of short manual shootouts with SF and Komodo, and, in most cases, score invariably changes to white's favour, but only after quite a few moves. So I have good reason to believe white has (winning) advantage.

Not able to consider all lines now, but, taking only Milos' line, which also appears a bit modified in Mr. Poppins' output, after f6 Bf8 b3 b5 Kh2 Qb8 Ba3 b4 Bc1 (better than Bb2) a4 white plays Rb1(compromising white's pawn structure with ba4 is stupid of course):

[d]rq2kb1r/3b1p2/3p1Pp1/2n1p1P1/ppPpP2p/1P1P3P/P3N1BK/1RBQ1R2 b kq - 0 6

take a shot with your big guns at above position, and tell me whether the score does not change in white's favour after a while.

[tpoppins]

Code: Select all

rq2kb1r/3b1p2/3p1Pp1/2n1p1P1/ppPpP2p/1P1P3P/P3N1BK/1RBQ1R2 b kq - 0 21

Analysis by CorChess beta 151016 x64 POPCNT (16 GB hash/11 threads):

21...axb3 22.axb3 d5 23.cxd5 Bb5 24.Rf3 Bd6 25.Bb2 Rh5 26.Kh1 Qb7 27.Rc1 Rxg5 28.Rc2 Rc8 29.Bc1 Rh5 30.Bb2 Kf8 31.Ng1 Qa6 32.Bf1 Re8 33.Rf5 Rh7 34.Rff2 Nxb3 35.Nf3 Rh8 36.Kg1 Ba4 37.Rc4 Rh5 38.Nh2 Rc8 39.Rxc8+ Qxc8 40.Ng4 Nc5 41.Qd2 Qb7 42.Be2 Nb3 43.Qd1 Rh8 44.Rf1 Rh7 45.Rf2 Rh5 46.Bf1 Qa6 47.Be2 Nc5 48.Qd2 Bb5 49.Nxe5 Rxe5 
  -/+ (-1.06)  Depth: 46/81   03:06:55  158165MN, tb=1033077

Houdini Tactical analysis to follow once the 33rd iteration is complete. At the moment it likes the immediate 21...d5 (-0.74), which you should have expected given the analysis in my previous post.

The story so far:
[pgn]
[White "Lyudmil"]
[Black "Big Guns"]
[Result "*"]
[SetUp "1"]
[FEN "r2qk2r/1p1b1pb1/3p2p1/p1n1pPP1/2PpP2p/3P3P/PP2N1B1/R1BQ1RK1 w kq - 0 16"]
[PlyCount "18"]

16. f6 Bf8 17. b3 b5 18. Kh2 Qb8 19. Ba3 b4 20. Bc1 a4 21. Rb1 axb3 22.axb3 d5 23.cxd5 Bb5 24.Rf3 Bd6 *

[/pgn]

Black is not really being a piece less now, is he? Note the scope of the white knight and light-square bishop, too.

[tpoppins]

Code: Select all

Analysis by Houdini 4 Pro x64 B (Tactical/8 GB hash/16 threads):

21...d5 22.cxd5 Bb5 23.Rf3 Bd6 24.Bb2 Rh5 25.bxa4 Rxa4 26.Qd2 Rxa2 27.Nc1 Ra8 28.Rf1 Kf8 29.Bf3 Rh8 30.Ra1 Rxa1 31.Bxa1 Kg8 32.Rg1 b3 33.Bb2 Bd7 34.Rg2 Qb5 35.Qd1 Qa4 36.Qd2 Bb5 37.Qd1 
  -/+ (-0.76)  Depth: 33/83   03:58:52  397333MN, tb=1487