Yet another bug !

Discussion of chess software programming and technical issues.

Moderator: Ras

Henk
Posts: 7251
Joined: Mon May 27, 2013 10:31 am

Yet another bug !

Post by Henk »

Is there any engine who can find the right move here. Best move should be Re6! Position is from (famous) game Reshevsky - Petrosjan 1953.

[d] 3rq1k1/4rppp/2n3b1/pp2P3/2pP1QB1/P1P1R3/1B4PP/4R1K1 b - - 3 25
Joost Buijs
Posts: 1671
Joined: Thu Jul 16, 2009 10:47 am
Location: Almere, The Netherlands

Re: Yet another bug !

Post by Joost Buijs »

Henk,

I don't think it is a bug if your engine can't find it.
Nightmare thinks after searching for 1.5 minute that Re6 is the best move, but at higher depths it thinks Ra7 is the best move.

Code: Select all


FEN: 3rq1k1/4rppp/2n3b1/pp2P3/2pP1QB1/P1P1R3/1B4PP/4R1K1 b - - 3 25

Nightmare MP:
 1	00:00	 80	80	-0.47	Bg6-d3
 2	00:00	 237	237	-0.47	Bg6-d3
 3	00:00	 2k	1,811k	-0.73	b5-b4 a3xb4 a5xb4
 4	00:00	 4k	1,486k	-0.58	b5-b4
 5	00:00	 17k	1,580k	-0.79	Bg6-d3 Re3-g3 b5-b4 a3xb4 a5xb4
 6	00:00	 34k	2,120k	-0.74	Bg6-d3 Re3-h3 b5-b4 Bg4-f5 b4xc3 Bb2xc3
 7	00:00	 97k	3,343k	-0.59	b5-b4 Bg4-f3 b4xc3 Bb2xc3 Bg6-d3 d4-d5
 8	00:00	 182k	3,705k	-0.93	b5-b4 Bg4-f3 b4-b3 d4-d5 Nc6-a7 d5-d6 Re7-e6 Qf4xc4
 9	00:00	 435k	5,300k	-0.74	b5-b4 Bg4-f3 b4-b3 d4-d5 Nc6-a7 Qf4xc4 Qe8-b5 Qc4-h4
 10	00:00	 708k	5,900k	-0.75	b5-b4 a3xb4 a5xb4 Bg4-f3 b4-b3 d4-d5 Nc6-a5 Bb2-a3 Re7-b7
 11	00:00	 1,749k	7,138k	-1.03	b5-b4 Bg4-f3 b4-b3 d4-d5 Nc6-a7 Qf4xc4 a5-a4 e5-e6 Qe8-b5 Qc4-b4
 12	00:00	 3,262k	8,073k	-0.84	Re7-a7 e5-e6 f7-f6 Re3-g3 b5-b4 Qf4-e3 Ra7-e7 a3xb4 a5xb4 Bg4-e2 b4xc3 Bb2xc3
 13	00:00	 4,394k	8,582k	-0.69	Re7-a7
 14	00:00	 6,564k	9,458k	-0.72	Re7-a7 e5-e6 f7-f6 Re3-g3 b5-b4 Bg4-f5 Nc6-e7 Bf5xg6 Ne7xg6 Qf4-f1 b4xc3 Bb2xc3 Qe8-c6
 15	00:00	 9,634k	9,994k	-0.79	Re7-a7 e5-e6 f7-f6 Re3-h3 b5-b4 Qf4-f1 h7-h5 Bg4-f5 b4xc3 Bb2xc3 Nc6xd4 Qf1xc4 Nd4xf5 e6-e7+ Bg6-f7
 16	00:02	 24,812k	11,397k	-0.73	Re7-a7 e5-e6 f7-f6 Re3-h3 Nc6-e7 Qf4-f1 Ne7-d5 a3-a4 b5xa4 Bb2-a3 h7-h5 Bg4-f3 Bg6-d3 Qf1-f2 Nd5xc3 Rh3xh5
 17	00:08	 96,293k	11,906k	-0.92	Nc6-a7 Re1-f1 Na7-c6 Bg4-f3 f7-f6 e5xf6 Re7xe3 Bf3xc6 Qe8xc6 f6-f7+ Kg8-f8 Qf4xe3 Bg6xf7 Qe3-g5 Rd8-e8 Qg5-f5
 18	00:11	 134,867k	11,906k	-1.04	Nc6-a7 Qf4-g5 b5-b4 a3xb4 a5xb4 e5-e6 f7-f6 Qg5-c5 b4xc3 Bb2-a3 Na7-c8 Re3xc3 Bg6-d3 d4-d5 Nc8-d6 Qc5-b6 Re7-b7
 19	00:18	 222,490k	12,086k	-0.95	Nc6-a7 Bg4-f3 Na7-c8 Qf4-h4 Bg6-d3 Bb2-c1 Re7-e6 Bf3-e4 Bd3xe4 Qh4xe4 Nc8-e7 Re3-h3 Re6-g6 Re1-f1 Ne7-d5 Qe4-h4 h7-h6 Rh3-f3
 20	00:20	 249,762k	12,150k	-0.93	Nc6-a7 Bg4-f3 Na7-c8 Qf4-h4 Bg6-d3 Bb2-c1 Re7-e6 Bf3-e4 Bd3xe4 Qh4xe4 Nc8-e7 Re3-h3 Re6-g6 Re1-f1 Ne7-d5 Bc1-d2 Qe8-d7 Rh3-f3 h7-h6 Rf3xf7
 21	00:28	 347,474k	12,264k	-1.08	Nc6-a7 Bg4-f3 Na7-c8 Qf4-h4 Bg6-d3 Bb2-c1 b5-b4 c3xb4 a5xb4 a3xb4 Nc8-b6 b4-b5 Re7-d7 Bf3-c6 Qe8-e7 Qh4xe7 Rd7xe7 Bc1-a3 Re7-a7 Ba3-c5
 22	01:26	 1,028,802k	11,911k	-1.22	Re7-e6 Bg4xe6 Qe8xe6 Re3-f3 b5-b4 a3xb4 a5xb4 c3xb4 Nc6xb4 Re1-a1 Qe6-d5 Qf4-h4 Bg6-e4 Rf3-g3 Nb4-c6 Rg3-g4 Be4-f5 Rg4-g5 Bf5-d3 Qh4-h6 Bd3-g6
 23	01:36	 1,165,373k	12,033k	-1.22	Re7-e6 Bg4xe6 Qe8xe6 Re3-f3 b5-b4 a3xb4 a5xb4 c3xb4 Nc6xb4 Qf4-h4 Qe6-b6 e5-e6 f7xe6 Re1xe6 Qb6xe6 Qh4xd8+ Bg6-e8 Qd8-g5 Nb4-d3 d4-d5 Qe6-b6+
 24	03:41	 2,683,767k	12,131k	-0.49	Re7-a7 e5-e6 f7-f6 e6-e7 Nc6xe7 d4-d5 f6-f5 Bg4xf5 Bg6xf5 d5-d6 Bf5-d3 a3-a4 Qe8-c6 d6xe7 Rd8-e8 a4xb5 Qc6xb5 Bb2-a3 Qb5-f5 Re3-f3 Qf5xf4
 25	04:02	 2,957,710k	12,185k	-0.40	Re7-a7 e5-e6 f7-f6 e6-e7 Nc6xe7 d4-d5 f6-f5 Bg4xf5 Bg6xf5 d5-d6 Bf5-d3 a3-a4 Qe8-f7 Qf4xf7+ Kg8xf7 d6xe7 Rd8-e8 Re3-f3+ Kf7-g8 Rf3-f8+ Re8xf8
 26	05:07	 3,741,195k	12,154k	-0.44	Re7-a7 e5-e6 f7-f6 a3-a4 Nc6-e7 a4xb5 Ne7-d5 Qf4-f2 Nd5xe3 Qf2xe3 Qe8xb5 Bb2-a3 Ra7-a6 Bg4-f3 Qb5-f5 e6-e7 Rd8-e8 Qe3-e2 Qf5-b5 Ba3-c5 Bg6-f7
 27	06:55	 5,102,553k	12,287k	-0.40	Re7-a7 e5-e6 f7-f6 e6-e7 Nc6xe7 d4-d5 f6-f5 Bg4xf5 Bg6xf5 d5-d6 Bf5-d3 a3-a4 Qe8-f7 Qf4xf7+ Kg8xf7 d6xe7 Rd8-e8 Re3-f3+ Kf7-g8 Rf3-f8+ Re8xf8

Stockfish also thinks Ra7 is the best move in this position.

Petrosian was a good player, but not as good as the current crop of engines.
Henk
Posts: 7251
Joined: Mon May 27, 2013 10:31 am

Re: Yet another bug !

Post by Henk »

Not so sure about it. Why not for instance fxe6. Blocking both d4 and e5 pawn is the theme here. Be6: is probably a bad move too. White exchanging his only good bishop ? When knight gets to d5 here there is more than enough compensation.
PK
Posts: 913
Joined: Mon Jan 15, 2007 11:23 am
Location: Warsza

Re: Yet another bug !

Post by PK »

to play Re6 your engine needs positional compensation of about 125-150 centipawns, depending on your material values:

Code: Select all

rook - (bishop + bishop pair)
This is pretty speculative. After all, Petrosian was a speculative player (except that he speculated about opponent's threats, not about his own attacking potential, as it is usually the case).

Beside that he worked on notions which are not yet programmed even in the top engines (certainly not in Stockfish). Black's compensation is about white central pawns being unable to move (though not necessarily blocked), black queenside pawns being mobile (which becomes important only because of immobilisation of white's center) and black knight having several good white squares to head for (which becomes more important than any single outpost (and d5 is not even an outpost by computer definition, since white has a pawn on c3).
Joost Buijs
Posts: 1671
Joined: Thu Jul 16, 2009 10:47 am
Location: Almere, The Netherlands

Re: Yet another bug !

Post by Joost Buijs »

Well, it depends, this is not so easy to answer without thorough examination of all subsequent variations.

If white takes the rook you will probably get a blocked position with that knight on d5 but you are still an exchange behind which can wreck you in the endgame.

I don't consider it to be a bug when an engine plays something else than Re6.
Henk
Posts: 7251
Joined: Mon May 27, 2013 10:31 am

Re: Yet another bug !

Post by Henk »

A knight in the centre in a (closed/blocked) position where it can not be chased away should get enough bonus. Perhaps value of a rook is overestimated in opening and middle game in closed positions.
Last edited by Henk on Tue Oct 13, 2015 1:08 pm, edited 1 time in total.
PK
Posts: 913
Joined: Mon Jan 15, 2007 11:23 am
Location: Warsza

Re: Yet another bug !

Post by PK »

No engine that I know of will understand within its evaluation function that Nd5 cannot be chased away.

Even if it understood, what is a reasonable bonus for a central outpost not defended by a pawn? Certainly not enough to sack the exchange just for the outpost.

BTW, it might be interesting to run current engines through this collection of games (all of them having one thing in common: Petrosian's opponents were in no position to activate their pieces after the exchange sac):

http://www.chessgames.com/perl/chesscol ... id=1013349
Henk
Posts: 7251
Joined: Mon May 27, 2013 10:31 am

Re: Yet another bug !

Post by Henk »

Problem of chess engines is that they lack the chess knowledge that had not been implemented for it slowed down search too much. That's one of the reasons why their evaluation often can not be trusted.
Vinvin
Posts: 5316
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Yet another bug !

Post by Vinvin »

At depth 42/60 SF prefers Ra7 (+0.51) over Re6 (+0.85).

Code: Select all

FEN: 3rq1k1/4rppp/2n3b1/pp2P3/2pP1QB1/P1P1R3/1B4PP/4R1K1 b - - 3 25

Stockfish_15100312_x64_modern:
...
---------------------------------------------------------------------------
 42/60	46:05	 24.459.029k	8.844k	+0,85	25. ... Re6 26.Bxe6 fxe6 27.Rf1 Ne7 28.Ref3 Nd5 29.Qc1 Rb8 30.R1f2 Qd7 31.Qf1 Qc8 32.h3 Bh5 33.Rg3 Bg6 34.Qa1 Ra8 35.Qe1 Rb8 36.Rgf3 Qe8 37.Qc1 Qe7 38.Rg3 Qd7 39.Rff3 Qe7 40.Rf1 Bd3 41.Rf2 Rf8 42.Rxf8+ Kxf8 43.Qd2 Kg8 44.Rf3 Bg6 45.Rf2 h6 46.Rf3 Qh4
 42/60	46:05	 24.459.029k	8.844k	+0,51	25. ... Ra7 26.e6 f6 27.Bf3 Ne7 28.a4 bxa4 29.Ba3 Nd5 30.Bxd5 Rxd5 31.Qf3 Rf5 32.Qe2 Rc7 33.e7 Bh5 34.Qb2 Rb5 35.Qc2 Rd5 36.Bc5 Rg5 37.Qb2 Bf7 38.R3e2 h5 39.Qb6 Rc8 40.Qxa5 h4 41.Qa6 Rb8 42.h3 Rb3 43.Bb4 a3 44.Qd6 Rf5 45.Qc7 Bd5 46.Qa5 Rg5 47.Qd8 Bf7 48.Ra2 Rg3 49.Qc7
But SF sees no progress after 25. ... Re6 26.Bxe6 fxe6 :

Code: Select all

FEN: 3rq1k1/6pp/2n1p1b1/pp2P3/2pP1Q2/P1P1R3/1B4PP/4R1K1 w - - 0 27

Stockfish_15100312_x64_modern:
...
 31/36	00:01	 17.052k	9.375k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qg5 Qd7 30.h3 Rb8 31.R1f2 Bf5 32.Qc1 Be4 33.Rg3 Bg6 34.Rff3 Qe7 35.Kh2 Bd3 36.Rf2 Qd7 37.Qe1 Bf5 38.Kg1 Bd3 39.Rgf3 Qe7 40.Qd2 Be4 41.Rg3 Bg6 42.Qe1
 32/36	00:02	 24.093k	9.353k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qg5 Qd7 30.h3 Rb8 31.R1f2 Bf5 32.Qc1 Be4 33.Rg3 Bg6 34.Rff3 Qe7 35.Kh2 Bd3 36.Rf2 Qd7 37.Qe1 Bf5 38.Kg1 Bd3 39.Rgf3 Qe7 40.Qd2 Be4 41.Rg3 Bg6 42.Bc1 b4 43.axb4
 33/36	00:03	 35.863k	9.398k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qg5 Qd7 30.h3 Rb8 31.R1f2 Bf5 32.Qc1 Be4 33.Rg3 Bg6 34.Rff3 Qe7 35.Kh2 Bd3 36.Rf2 Qd7 37.Qe1 Bf5 38.Kg1 Bd3 39.Rgf3 Qe7 40.Qa1 Bg6 41.Rg3 Qh4 42.Rff3 Qe7 43.Qc1
 34/36	00:05	 53.552k	9.468k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qg5 Qd7 30.h3 Rb8 31.R1f2 Bf5 32.Qc1 Be4 33.Rg3 Bg6 34.Rff3 Qe7 35.Kh2 Bd3 36.Rf2 Qd7 37.Qe1 Bf5 38.Kg1 Bd3 39.Rgf3 Qe7 40.Qa1 Bg6 41.Rg3 Qh4 42.Rff3 Qe7 43.Qc1 Bd3 44.Kh2
 35/36	00:08	 79.933k	9.439k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qg5 Qd7 30.h3 Rb8 31.R1f2 Bf5 32.Qc1 Be4 33.Rg3 Bg6 34.Rff3 Qe7 35.Kh2 Bd3 36.Rf2 Qd7 37.Qe1 Bf5 38.Kg1 Bd3 39.Rgf3 Qe7 40.Qa1 Bg6 41.Rg3 Qh4 42.Rff3 Qe7 43.Qc1 Bd3 44.Kh2 Qd7
 36/37	00:10	 100.301k	9.435k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qg5 Qd7 30.h3 Rb8 31.R1f2 Bf5 32.Qc1 Be4 33.Rg3 Bg6 34.Rff3 Qe7 35.Kh2 Bd3 36.Rf2 Qd7 37.Qe1 Bf5 38.Kg1 Bd3 39.Rgf3 Qe7 40.Qa1 Bg6 41.Rg3 Qh4 42.Rff3 Qe7 43.Qc1 Bd3 44.Kh2 Qd7 45.Rf2
 37/46	00:36	 355.874k	9.738k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qg5 Qd7 30.h3 Rb8 31.R1f2 Bf5 32.Qc1 Be4 33.Rg3 Bf5 34.Qe1 Bd3 35.Qa1 Ra8 36.Qc1 Rf8 37.Rxf8+ Kxf8 38.Rf3+ Kg8 39.Qd2 Qe7 40.Qe1 Bf5 41.Rf2 h6 42.Qd2 Bd3 43.Rf3 Qh4 44.Kh2 Qe7 45.Qe1 Bf5 46.Kg1 Bg6
 38/53	01:00	 576.981k	9.587k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qg5 Qd7 30.h3 Rb8 31.R1f2 Bf5 32.Rg3 Be4 33.Qc1 Bf5 34.Qe1 Bd3 35.Qa1 Ra8 36.Qc1 Rf8 37.Rxf8+ Kxf8 38.Rf3+ Kg8 39.Qd2 Be4 40.Rf2 Bd3 41.Qd1 h6 42.Qc1 Qe7 43.Qe1 Bf5 44.Qd2 Bd3 45.Rf3 Qh4 46.Kh2 Qe7
 39/53	01:16	 730.349k	9.541k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qg5 Qd7 30.h3 Rb8 31.R1f2 Bf5 32.Rg3 Be4 33.Qc1 Bf5 34.Qe1 Bd3 35.Qa1 Ra8 36.Qc1 Rf8 37.Rxf8+ Kxf8 38.Rf3+ Kg8 39.Qd2 Be4 40.Rf2 Bd3 41.Qd1 h6 42.Qc1 Qe7 43.Qe1 Bf5 44.Qd2 Bd3 45.Rf3 Qh4 46.Kh2 Qe7 47.Qe1
 40/57	03:05	 1.784.761k	9.620k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qd2 Rb8 30.R1f2 Qc6 31.h3 Bh5 32.Rg3 Bg6 33.Kh2 Qd7 34.Qe2 Bd3 35.Qd1 Bg6 36.Qc1 Bd3 37.Qa1 Ra8 38.Kg1 Bg6 39.Qe1 Rc8 40.Rf1 Bd3 41.Rff3 Rf8 42.Rxf8+ Kxf8 43.Rf3+ Kg8 44.Qh4 Qc7 45.Rf2 Qd7 46.Qg4 Qe7 47.Qd1 Bg6 48.Qd2 h6 49.Qc1
 41/57	03:58	 2.298.375k	9.641k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qd2 Rb8 30.R1f2 Qc6 31.h3 Bh5 32.Rg3 Bg6 33.Kh2 Qd7 34.Qe2 Bd3 35.Qd1 Bg6 36.Qc1 Bd3 37.Qa1 Ra8 38.Kg1 Bg6 39.Qe1 Rc8 40.Rf1 Bd3 41.Rff3 Rf8 42.Rxf8+ Kxf8 43.Rf3+ Kg8 44.Qh4 Qc7 45.Rf2 Qd7 46.Qg4 Qe7 47.Qd1 Bg6 48.Qd2 h6 49.Qc1 Bd3 50.Qe1
 42/57	05:42	 3.316.607k	9.687k	+0,85	27.Rf1 Ne7 28.Ref3 Nd5 29.Qd2 Rb8 30.R1f2 Qc6 31.h3 Bh5 32.Rg3 Bg6 33.Kh2 Qd7 34.Qe2 Bd3 35.Qd1 Bg6 36.Qc1 Bd3 37.Qa1 Ra8 38.Kg1 Bg6 39.Qe1 Rc8 40.Rf1 Bd3 41.Rff3 Rf8 42.Rxf8+ Kxf8 43.Rf3+ Kg8 44.Qh4 Qc7 45.Rf2 Qd7 46.Qg4 Qe7 47.Qd1 Bg6 48.Qd2 h6 49.Qc1 Bd3 50.Qe1 Bf5
Henk
Posts: 7251
Joined: Mon May 27, 2013 10:31 am

Re: Yet another bug !

Post by Henk »

Position of Nd5 is underrated.


It's almost the same error as assigning a wrong material value to a piece. So searching deeper won't help here. It's just an evaluation error.