Does your engine take the poisoned pawn?

[Patatov]

If you can only expect a draw by taking the pawn I don't undertsand why Fischer didn't chose some other easiest way to achieve this result

[Vinvin]

I analyzed this position some years ago and the capture on h2 is not losing because you get a lot of pawns for the bishop and an active king.

[d]5k2/pp4pp/3bpp2/1P6/8/P2KP3/5PPP/2B5 b - - 0 29

From Euwe and Timmans book on the Fischer Spassky Match, game one was close to a draw.

Funny note on the move Bxh2??!!! One doesn't see comments like this in many books!

"The capture is a curious bloomer, imcompatible with the reality of high level chess"

"It is almost certain that this move is based on a miscalculation.All other moves lead to a draw. It iis nevertheless strange that Fischer takes the poisoned pawn, considering that it is the only variation in the position requiring calculation - for which Fischer had ample time"

Prodeo selects this move, while Hiarcs does not.

[Jouni]

Yes this is familiar looking position! After 1.-Bxh2 I did fast 30s autoplay and both ended draw- so this is probably draw and bad test position.

Rybka:

2. g3 Ke7 3. a4 h5 4. Ke4 h4 5. Kf3 h3 6. Kg4 Bg1 7. Kxh3 Bxf2 8. Bd2 b6 9. Kg2
Bxe3 10. Bxe3 Kd6 11. Kf3 Kd5 12. Kg4 g6 13. Bh6 e5 14. Bg7 Ke6 15. Bh8 f5+ 16.
Kg5 Kd6 17. Bf6 Kd5 18. Bg7 Ke6 19. Bh6 Kf7 20. Kh4 Ke6 21. Bd2 f4 22. gxf4
exf4 23. Kg4 g5 24. Kxg5 f3 25. Be3 Kd5 26. Kf4 Kc4 27. Ke4 Kb4 28. Kd5 Kxa4 0,5-0,5

Fritz:

2. g3 Ke7 3. Bd2 Kd6 4. Bb4+ Ke5 5. Ke2 h5 6. Kf3 h4 7. Kg2 hxg3 8. fxg3 Bxg3
9. Kxg3 f5 10. Be7 g6 11. Kh4 f4 12. exf4+ Kxf4 13. Bd6+ Ke4 14. Bb8 a6 15. b6
e5 16. Kg4 Kd5 17. Bc7 Kc5 18. Bd8 a5 19. Kg3 g5 20. Kf3 Kb5 21. Ke3 g4 22. Ke4 0,5-0,5

Jouni

[pijl]

I think the thing that makes it interesting is that a whole bunch of super-strong engines that clearly understand poisoned pawns want to take it anyway (at least for some time period).

I disagree. It takes only a few ply to see that the bishop will be lost. CTD has no code to recognize trapped bishops. Nonetheless it avoids Bxh2 pretty quickly. When thinking in simple patterns, it sees that after 1..Bxh2 2.g3 black will lose the bishop. Plans to free the bishop with h5 do not work (takes a few more ply to see that).
So is it simply a matter of counting wood to avoid taking it. It probably requires understanding of the pawn formations and the chances that a 4 agains 2 majority on one side plus a 2-1 majority on the other side can accomplish that is required to entice programs to take the pawn. Especially when programs count a bishop as equal to 3 pawns.
Richard.

[Dr.Ex]

I think the thing that makes it interesting is that a whole bunch of super-strong engines that clearly understand poisoned pawns want to take it anyway (at least for some time period).

I disagree. It takes only a few ply to see that the bishop will be lost.

Maybe you don't have a super strong engine, yet.
My Fritz doesn't spot that at depth 20.

5k2/pp4pp/3bpp2/1P6/8/P2KP3/5PPP/2B5 b - - 0 1

Analysis by Fritz 11:

1...Lxh2 2.f4 Kf7 3.Ld2 Lg3 4.e4 e5 5.f5 Lf4
-/+ (-0.86) Tiefe: 9/15 00:00:00 35kN
1...Lxh2 2.g3 h5 3.Ke2 Ke7 4.e4 h4 5.gxh4 Le5 6.Le3
= (-0.21) Tiefe: 10/17 00:00:00 80kN
1...Lxh2 2.g3 h5 3.Ke2 Ke7 4.e4 h4 5.gxh4 Ld6 6.Kd3 Kd7
= (-0.11) Tiefe: 11/16 00:00:00 119kN
1...Lxh2 2.g3 h5 3.Ke2 Ke7 4.e4 h4 5.gxh4 Ld6 6.Kd3 Kd7 7.Le3
= (-0.11) Tiefe: 12/19 00:00:00 250kN
1...Lxh2 2.g3 h5 3.Ke2 Kf7 4.e4 g5 5.Le3 h4 6.Kf3 b6 7.Kg4 hxg3 8.fxg3
= (-0.17) Tiefe: 13/22 00:00:00 395kN
1...Lxh2 2.g3 h5 3.Ke4 h4 4.gxh4 Ld6 5.Lb2 Kf7 6.f4 Le7 7.Kd3 Ld6 8.Ke4
= (0.00) Tiefe: 14/23 00:00:02 971kN
1...Lxh2 2.g3 h5 3.Ke4 h4 4.gxh4 Ld6 5.Lb2 Kf7 6.f4 a6 7.bxa6 bxa6 8.a4 Le7 9.Kd3
= (0.01) Tiefe: 15/24 00:00:05 2446kN
1...Ke7 2.f4 a6 3.bxa6 bxa6 4.Lb2 g5 5.g3 g4 6.Kc4 f5 7.Kd3 Kd7 8.e4 Lc5 9.exf5 exf5
= (0.01) Tiefe: 15/24 00:00:06 3044kN
1...Ke7 2.f4 e5 3.Ke4 g6 4.f5 b6 5.g4 Lc5 6.a4 Lb4 7.Lb2 g5 8.Kd5 Kd7 9.e4
= (0.03) Tiefe: 16/25 00:00:21 5247kN
1...Lxh2 2.g3 h5 3.Ke4 h4 4.gxh4 Ld6 5.Lb2 Ke7 6.f4 Kd7 7.a4 Lb4 8.Ld4 b6 9.h5 f5+
= (0.01) Tiefe: 16/25 00:00:25 5632kN
1...Lxh2 2.g3 h5 3.Ke4 h4 4.gxh4 Ld6 5.Lb2 Ke7 6.f4 Kd7 7.a4 a6 8.Lc3 axb5 9.axb5 Lc5 10.h5 f5+
= (0.01) Tiefe: 17/25 00:00:35 7173kN
1...Lxh2 2.g3 h5 3.Ke4 h4 4.gxh4 Ld6 5.Lb2 Kf7 6.a4 f5+ 7.Kd3 e5 8.f3 Ke6 9.Lc3 Lc5 10.h5
= (0.01) Tiefe: 18/26 00:00:47 12089kN
1...Lxh2 2.g3 h5 3.Ke4 h4 4.gxh4 Ld6 5.Lb2 Kf7 6.h5 f5+ 7.Kd4 Le7 8.Kc4 Lf6 9.Ld4 b6 10.Lxf6 Kxf6 11.f4
= (0.06) Tiefe: 19/28 00:00:57 17702kN
1...Lxh2 2.g3 h5 3.Ke4 h4 4.gxh4 g6 5.Lb2 Kf7 6.Ld4 Ld6 7.a4 f5+ 8.Kd3 b6 9.a5 bxa5 10.Lxa7 Le7 11.b6 Lxh4
= (0.04) Tiefe: 20/30 00:01:39 40696kN

[d]5k2/pp4p1/4pp2/1P5p/8/P3P1P1/4KP1b/2B5 b - - 0 1

Analysis by Fritz 11:

3...h4 4.gxh4 Ld6 5.e4 Ke7 6.Le3 Lxa3 7.Lxa7 e5 8.Le3 Ke6
= (-0.01) Tiefe: 9/12 00:00:00 5kN
3...h4 4.gxh4 Ld6 5.e4 Ke7 6.Le3 Lxa3 7.Lxa7 e5 8.Le3 Ke6
= (-0.06) Tiefe: 10/14 00:00:00 11kN
3...h4 4.gxh4 Ld6 5.e4 Ke7 6.a4 Lc5 7.h5 e5 8.Le3 Lxe3 9.Kxe3 Ke6
= (-0.01) Tiefe: 11/16 00:00:00 17kN
3...h4 4.gxh4 Ld6 5.e4 Ke7 6.Kd3 Lc5 7.f4 a6 8.bxa6 bxa6 9.Lb2 Kd7
= (-0.01) Tiefe: 12/17 00:00:00 35kN
3...h4 4.gxh4 Ld6 5.a4 Ke7 6.Ld2 e5 7.Kd3 Ke6 8.Lc3 b6 9.h5 Le7 10.Kc4 f5
= (-0.04) Tiefe: 13/19 00:00:00 101kN
3...h4 4.gxh4 Ld6 5.a4 Ke7 6.e4 Lc5 7.h5 e5 8.f4 Ke6 9.Kd3 Ld4 10.f5+
= (-0.07) Tiefe: 14/20 00:00:00 191kN
3...h4 4.gxh4 Ld6 5.Kd3 Kf7 6.h5 e5 7.Kc4 Ke6 8.e4 b6 9.a4 Lc5 10.f3 f5 11.exf5+
= (0.00) Tiefe: 15/23 00:00:01 562kN
3...h4 4.gxh4 Ld6 5.Kd3 Kf7 6.h5 e5 7.Kc4 Ke6 8.e4 b6 9.a4 Lc5 10.f3 Ld4 11.Ld2
= (-0.01) Tiefe: 16/24 00:00:02 1056kN
3...h4 4.Kf3 hxg3 5.fxg3 Ke7 6.Kg2 Lxg3 7.Kxg3 Kd6 8.e4 Kc5 9.Le3+ Kxb5 10.Lxa7 b6 11.Lb8 f5 12.Kf3 fxe4+
+/- (0.78) Tiefe: 17/26 00:00:05 2473kN
3...Ke7 4.Kf1 Kd6 5.Kg2 Lxg3 6.Kxg3 e5 7.Ld2 Kd5 8.Kh4 g6 9.f3 Kc4 10.a4 Kd3 11.Lc1 b6
+/= (0.68) Tiefe: 17/27 00:00:10 5638kN
3...Ke7 4.Kf1 Kd6 5.Kg2 Lxg3 6.Kxg3 Kc5 7.f4 Kxb5 8.Kh4 g6 9.Lb2 f5 10.Kg5 Kc4 11.Ld4 Kd5 12.Kxg6
+/= (0.68) Tiefe: 18/31 00:00:14 7681kN
3...Ke7 4.Kf1 Kd6 5.Kg2 Lxg3 6.Kxg3 Kc5 7.f4 Kxb5 8.Kh4 g6 9.Lb2 f5 10.Kg5 Kc4 11.Ld4 b6 12.Kxg6 h4
+/= (0.66) Tiefe: 19/28 00:00:20 11248kN
3...Ke7 4.Kf1 Kd6 5.Kg2 Lxg3 6.Kxg3 Kc5 7.f4 Kxb5 8.Kh4 g6 9.Lb2 f5 10.Kg5 Kc4 11.Ld4 b6 12.Le5 Kd5 13.Kxg6
+/= (0.66) Tiefe: 20/32 00:00:32 16433kN

[pijl]

I think the thing that makes it interesting is that a whole bunch of super-strong engines that clearly understand poisoned pawns want to take it anyway (at least for some time period).

I disagree. It takes only a few ply to see that the bishop will be lost.

Maybe you don't have a super strong engine, yet.
My Fritz doesn't spot that at depth 20.

This is exactly what I meant. The point is not whether the bishop is lost. If white wants it, it can get it and that is not some deep line. Therefor having evaluation terms for trapped bishops does not help to avoid the capture on h2 in this position.

The point is to assess what kind of compensation black will have in its pawn structure. Simple programs like CTD do not see much compensation and see that black does not have pieces left with sufficient pawns on the board. And by just counting wood (plus perhaps some bonusses/penalties for last piece captured) it will decline the offered pawn.

Fritz does seem to think that there is compensation in the black pawn structure for the bishop and accepts (as white) a black pawn sac on h4 instead of collecting the bishop.

The Baron 2.22 (quite a bit stronger than CTD) does have trapped bishop evaluation and could play Bxh2 in a tournament game as it sticks with it for quite a long time.

So I just disagreed with Dann that it was interesting that super-strong engines would play Bxh2 despite their understanding of poisoned pawns as:
- poisoned pawns evaluation has noting to do with wanting to play Bxh2
- there are good explanations for why they go for it while others seem to refute it pretty quickly.

Richard.

[Jouni]

I also run tournament with 10s level after -Bxh2:

Code: Select all

                         1    2    3    4    
1   Spike 1.2 Turin      **** ½½   1½   ½1     4.0/6
2   Rybka 2.3.2a 32-bit  ½½   **** ½½   ½½     3.0/6  9.00
3   Fritz 10             0½   ½½   **** 1½     3.0/6  8.00
4   Fruit 05/11/03       ½0   ½½   0½   ****   2.0/6

So Spike succeeded to win with black! Game:

[Event "New"]
[Site "?"]
[Date "2007.12.18"]
[Round "1.1"]
[White "Fritz 10"]
[Black "Spike 1.2 Turin"]
[Result "0-1"]
[Annotator "0.16;1.38"]
[SetUp "1"]
[FEN "5k2/pp4pp/3bpp2/1P6/8/P2KP3/5PPP/2B5 b - - 0 1"]
[PlyCount "41"]
[EventDate "2007.12.18"]
[EventType "tourn"]
[Source "Player"]

{Intel(R) Pentium(R) 4 CPU 2.40GHz 2393 MHz W=15.8 ply; 730kN/s; 6 482 TBAs
B=15.8 ply; 796kN/s; 1 325 TBAs} 1... Bxh2 2. g3 {0.16/16 8} h5 {1.38/17 60} 3.
Ke2 {(Ke4) 1.81/16 4} Ke7 {(h4) 1.58/16 27} 4. Kf3 {2.02/15 4} Kd6 {
(Bg1) 1.50/15 6} 5. a4 {(Kg2) 1.82/15 5} Kc5 {1.15/16 8} 6. Bd2 {
(Ba3+) 1.68/15 4} Kc4 {(h4) 1.04/17 19} 7. Kg2 {1.68/15 3} Bxg3 {1.00/16 12} 8.
Kxg3 {1.24/14 2} g5 {1.00/15 7} 9. f4 {1.17/14 9} g4 {(e5) 1.19/14 17} 10. Ba5
{(f5) 0.95/15 17} b6 {0.96/14 5} 11. Bd2 {0.82/15 6} f5 {0.88/15 9} 12. Be1 {
(Kh4) 0.61/16 11} Kb3 {(Kd5) 0.80/16 16} 13. a5 {0.88/17 8} Kc4 {0.66/15 3} 14.
axb6 {0.84/16 6} axb6 {0.41/16 6} 15. e4 {(Bf2) 0.73/16 6} Kxb5 {0.56/16 6} 16.
Bf2 {0.41/16 16} Kc6 {(Ka5) -0.07/16 13} 17. Kh4 {0.06/16 11} b5 {0.00/15 7}
18. Be1 {(Kxh5) -2.62/15 18} Kc5 {-0.54/13 3} 19. Kg3 {(exf5) -2.92/14 11} fxe4
{(b4) -2.21/13 4} 20. Kf2 {-6.56/13 11} Kd4 {-2.46/14 4} 21. Kf1 {
(Bd2) -6.38/13 7} Ke3 {(Kd3) -3.82/13 4} 0-1

Jouni

[bob]

crafty certainly will never take it at any depth greater than 1 ply because it understands the concept of the bishop being trapped by g3...

[bob]

It might turn out to be a safe capture, since crafty doesn't like it until it can see the g3 trap being broken apart... But one thing is for certain, crafty won't make the capture if g3 can be played and kept trapping the bishop...

Looking at the PV, that is exactly what is going on with black playing h5 and h4...

[George Tsavdaris]

I analyzed this position some years ago and the capture on h2 is not losing because you get a lot of pawns for the bishop and an active king.

I analyzed it now and it is a draw.

Key move in the best line: 29...Bxh2 30. g3 h5 31. Ke2 h4 32. Kf3 Ke7 33. Kg2 hxg3 34. fxg3 Bxg3 35. Kxg3 Kd6 36. a4 Kd5 37. Ba3

It's the 37...a6! with draw.

There are many many more lines of course but all lead to draw....