Shirov's Bh3, Houdini solves it

[Lyudmil Tsvetkov]

May be it was a lucky run:

Code: Select all

FEN: 8/8/4kpp1/3p1b2/p6P/2B5/6P1/6K1 b - - 0 1
St-mod-16-12-02:
Found 510 tablebases
 50/56	02:12	 173.020k	1.306k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Kg3-g2 Kd2-e3 Be4-c2 Ke3-d2 Bc2-b1 Bf6-a1 Kg2-h3 Ba1-f6 Kh3-g4 Kd2-c1 Kg4-f4 Kc1-d2 Bb1-f5 Bf6-d4 Kf4-g4 Bd4-f6 Kg4-f3 Bf6-a1
 51/56	02:17	 182.220k	1.323k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Be4-b1 Kd2-e3 Bb1-f5 Ke3-d2 Kg3-f3 Bf6-a1 Bf5-e4 Ba1-c3 Kf3-g2 Bc3-f6 Kg2-g3 Kd2-c1 Be4-b1 Kc1-d2 Kg3-g2 Bf6-a1
 52/57+	04:39	 436.623k	1.563k	+2,33	Bf5-h3
 52/57+	05:20	 538.195k	1.680k	+2,41	Bf5-h3
 52/57	10:53	 1.378.699k	2.109k	+3,53	Bf5-h3

but stockfish found it in 4:39 min on a amd laptop with 4 threads 1024 mb hash and 6-piece syzygy.

@Ludmil: As you can see it is a mate in 4 or 5 moves

Kind regards
Bernhard

i think its just lucky, but its difficult to know for sure; hash size, TB probe depth + limit, SSD/HDD are all variables that seem so flexible (minus SSD being superior to HDD ofc) the engine authors themselves only really seem to be able to guess at optimal settings.

re Bh3, it is decisively best & thus perfectly valid as a test position, humans & machines alike agree on this, end of story there.

yeah, it is easy to say end of story, but the problem is things are way more complicated.

Bh3 is indeed the best move, and very elegant apart from that, but the problem is how to make engines choose it, when they see very large advantage in simpler lines too? why sac when other simpler lines give big advantage too? as a programmer, what would you do to resolve that?

any draws resulting from above position are simply fabulous fairy-tale fortresses of the kind: KBP vs bare K and wrong bishop, though far beyond the engine horizon, KBPP vs KB, with pawns on a and g files, also far beyond the engine horizon, or KBPP, with d and g, or a and d pawns, also beyond the engine horizon; when you add couple more pawns, tablebases do not help either. so what you do to resolve that?

[Lyudmil Tsvetkov]

May be it was a lucky run:

Code: Select all

FEN: 8/8/4kpp1/3p1b2/p6P/2B5/6P1/6K1 b - - 0 1
St-mod-16-12-02:
Found 510 tablebases
 50/56	02:12	 173.020k	1.306k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Kg3-g2 Kd2-e3 Be4-c2 Ke3-d2 Bc2-b1 Bf6-a1 Kg2-h3 Ba1-f6 Kh3-g4 Kd2-c1 Kg4-f4 Kc1-d2 Bb1-f5 Bf6-d4 Kf4-g4 Bd4-f6 Kg4-f3 Bf6-a1
 51/56	02:17	 182.220k	1.323k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Be4-b1 Kd2-e3 Bb1-f5 Ke3-d2 Kg3-f3 Bf6-a1 Bf5-e4 Ba1-c3 Kf3-g2 Bc3-f6 Kg2-g3 Kd2-c1 Be4-b1 Kc1-d2 Kg3-g2 Bf6-a1
 52/57+	04:39	 436.623k	1.563k	+2,33	Bf5-h3
 52/57+	05:20	 538.195k	1.680k	+2,41	Bf5-h3
 52/57	10:53	 1.378.699k	2.109k	+3,53	Bf5-h3

but stockfish found it in 4:39 min on a amd laptop with 4 threads 1024 mb hash and 6-piece syzygy.

@Ludmil: As you can see it is a mate in 4 or 5 moves

Kind regards
Bernhard

i think its just lucky, but its difficult to know for sure; hash size, TB probe depth + limit, SSD/HDD are all variables that seem so flexible (minus SSD being superior to HDD ofc) the engine authors themselves only really seem to be able to guess at optimal settings.

re Bh3, it is decisively best & thus perfectly valid as a test position, humans & machines alike agree on this, end of story there.

yeah, it is easy to say end of story, but the problem is things are way more complicated.

Bh3 is indeed the best move, and very elegant apart from that, but the problem is how to make engines choose it, when they see very large advantage in simpler lines too? why sac when other simpler lines give big advantage too? as a programmer, what would you do to resolve that?

any draws resulting from above position are simply fabulous fairy-tale fortresses of the kind: KBP vs bare K and wrong bishop, though far beyond the engine horizon, KBPP vs KB, with pawns on a and g files, also far beyond the engine horizon, or KBPP, with d and g, or a and d pawns, also beyond the engine horizon; when you add couple more pawns, tablebases do not help either. so what you do to resolve that?

[Lyudmil Tsvetkov]

besides, some of the positions with a and g pawns and other configurations are won, while others drawn, so how you specify a common rule?

[MikeGL]

May be it was a lucky run:

Code: Select all

FEN: 8/8/4kpp1/3p1b2/p6P/2B5/6P1/6K1 b - - 0 1
St-mod-16-12-02:
Found 510 tablebases
 50/56	02:12	 173.020k	1.306k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Kg3-g2 Kd2-e3 Be4-c2 Ke3-d2 Bc2-b1 Bf6-a1 Kg2-h3 Ba1-f6 Kh3-g4 Kd2-c1 Kg4-f4 Kc1-d2 Bb1-f5 Bf6-d4 Kf4-g4 Bd4-f6 Kg4-f3 Bf6-a1
 51/56	02:17	 182.220k	1.323k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Be4-b1 Kd2-e3 Bb1-f5 Ke3-d2 Kg3-f3 Bf6-a1 Bf5-e4 Ba1-c3 Kf3-g2 Bc3-f6 Kg2-g3 Kd2-c1 Be4-b1 Kc1-d2 Kg3-g2 Bf6-a1
 52/57+	04:39	 436.623k	1.563k	+2,33	Bf5-h3
 52/57+	05:20	 538.195k	1.680k	+2,41	Bf5-h3
 52/57	10:53	 1.378.699k	2.109k	+3,53	Bf5-h3

but stockfish found it in 4:39 min on a amd laptop with 4 threads 1024 mb hash and 6-piece syzygy.

@Ludmil: As you can see it is a mate in 4 or 5 moves

Kind regards
Bernhard

i think its just lucky, but its difficult to know for sure; hash size, TB probe depth + limit, SSD/HDD are all variables that seem so flexible (minus SSD being superior to HDD ofc) the engine authors themselves only really seem to be able to guess at optimal settings.

re Bh3, it is decisively best & thus perfectly valid as a test position, humans & machines alike agree on this, end of story there.

yeah, it is easy to say end of story, but the problem is things are way more complicated.

Bh3 is indeed the best move, and very elegant apart from that, but the problem is how to make engines choose it, when they see very large advantage in simpler lines too? why sac when other simpler lines give big advantage too? as a programmer, what would you do to resolve that?

any draws resulting from above position are simply fabulous fairy-tale fortresses of the kind: KBP vs bare K and wrong bishop, though far beyond the engine horizon, KBPP vs KB, with pawns on a and g files, also far beyond the engine horizon, or KBPP, with d and g, or a and d pawns, also beyond the engine horizon; when you add couple more pawns, tablebases do not help either. so what you do to resolve that?

I remember this position was also analysed by some GM's and was once the whole article of Mig Greengard of TWIC, they claimed that Bh3 is required for the win. All other alternatives will result to a draw. But we need concrete proof with current powerful CPU and engines if Bh3 is really the only way to win.

[MikeGL]

May be it was a lucky run:

Code: Select all

FEN: 8/8/4kpp1/3p1b2/p6P/2B5/6P1/6K1 b - - 0 1
St-mod-16-12-02:
Found 510 tablebases
 50/56	02:12	 173.020k	1.306k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Kg3-g2 Kd2-e3 Be4-c2 Ke3-d2 Bc2-b1 Bf6-a1 Kg2-h3 Ba1-f6 Kh3-g4 Kd2-c1 Kg4-f4 Kc1-d2 Bb1-f5 Bf6-d4 Kf4-g4 Bd4-f6 Kg4-f3 Bf6-a1
51/56	02:17	 182.220k	1.323k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Be4-b1 Kd2-e3 Bb1-f5 Ke3-d2 Kg3-f3 Bf6-a1 Bf5-e4 Ba1-c3 Kf3-g2 Bc3-f6 Kg2-g3 Kd2-c1 Be4-b1 Kc1-d2 Kg3-g2 Bf6-a1
 52/57+	04:39	 436.623k	1.563k	+2,33	Bf5-h3
 52/57+	05:20	 538.195k	1.680k	+2,41	Bf5-h3
 52/57	10:53	 1.378.699k	2.109k	+3,53	Bf5-h3

but stockfish found it in 4:39 min on a amd laptop with 4 threads 1024 mb hash and 6-piece syzygy.

@Ludmil: As you can see it is a mate in 4 or 5 moves

Kind regards
Bernhard

i think its just lucky, but its difficult to know for sure; hash size, TB probe depth + limit, SSD/HDD are all variables that seem so flexible (minus SSD being superior to HDD ofc) the engine authors themselves only really seem to be able to guess at optimal settings.

re Bh3, it is decisively best & thus perfectly valid as a test position, humans & machines alike agree on this, end of story there.

yeah, it is easy to say end of story, but the problem is things are way more complicated.

Bh3 is indeed the best move, and very elegant apart from that, but the problem is how to make engines choose it, when they see very large advantage in simpler lines too? why sac when other simpler lines give big advantage too? as a programmer, what would you do to resolve that?

any draws resulting from above position are simply fabulous fairy-tale fortresses of the kind: KBP vs bare K and wrong bishop, though far beyond the engine horizon, KBPP vs KB, with pawns on a and g files, also far beyond the engine horizon, or KBPP, with d and g, or a and d pawns, also beyond the engine horizon; when you add couple more pawns, tablebases do not help either. so what you do to resolve that?

I remember this position was also analysed by some GM's and was once the whole article of Mig Greengard of TWIC, they claimed that Bh3 is required for the win. All other alternatives will result to a draw. But we need concrete proof with current powerful CPU and engines if Bh3 is really the only way to win.

Copied from the above line:

51/56 02:17 182.220k 1.323k +2,26 a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Be4-b1 Kd2-e3 Bb1-f5 Ke3-d2 Kg3-f3 Bf6-a1 Bf5-e4 Ba1-c3 Kf3-g2 Bc3-f6 Kg2-g3 Kd2-c1 Be4-b1 Kc1-d2 Kg3-g2 Bf6-a1


[pgn]
[Event "Shirov Immortal Endgame"]
[Site "Analysis"]
[Date "2016.12.05"]
[Round "?"]
[White "-"]
[Black "Shirov,A"]
[Result "*"]
[Time "16:36:56"]
[WhiteElo "2400"]
[TimeControl "60+0"]
[SetUp "1"]
[FEN "8/8/4kpp1/3p1b2/p6P/2B5/6P1/6K1 b - - 0 1"]
[Termination "unterminated"]
[PlyCount "54"]
[WhiteType "human"]
[BlackType "human"]

1. ... a3 2. Kf2 Bc2 3. Ke2 Be4 4. g3 Kf5 5. Bd4 Kg4 6. Bxf6 Kxg3 7. Kd2
Kf4 8. Bd4 a2 9. Bf6 Kg4 10. Ke3 Kf5 11. Bc3 Bh1 12. Kd2 Bf3 13. Kc2 Kf4
14. Kd3 Bg2 15. Kd2 Kg3 16. Bf6 Kg4 17. Kc1 Be4 18. Kd2 Kh3 19. Ke2 Kg3 20.
Kd2 Bb1 21. Ke3 Bf5 22. Kd2 Kf3 23. Ba1 Be4 24. Bc3 Kg2 25. Bf6 Kg3 26. Kc1
Bb1 27. Kd2 Kg2 28. Ba1 =
[/pgn]

This is a dead drawn, thanks to opposite B. The dark-squared B of white can even be sac'd on d-pawn (assuming g and h pawns were exchanged before giving it up on d-pawn of black.
So I think, the previous analysis of GMs were correct, the position will be a draw if Bh3 is not played.

But strangely the above line of SF gives +2,26 for black when that line is in fact a draw.

[Lyudmil Tsvetkov]

May be it was a lucky run:

Code: Select all

FEN: 8/8/4kpp1/3p1b2/p6P/2B5/6P1/6K1 b - - 0 1
St-mod-16-12-02:
Found 510 tablebases
 50/56	02:12	 173.020k	1.306k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Kg3-g2 Kd2-e3 Be4-c2 Ke3-d2 Bc2-b1 Bf6-a1 Kg2-h3 Ba1-f6 Kh3-g4 Kd2-c1 Kg4-f4 Kc1-d2 Bb1-f5 Bf6-d4 Kf4-g4 Bd4-f6 Kg4-f3 Bf6-a1
51/56	02:17	 182.220k	1.323k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Be4-b1 Kd2-e3 Bb1-f5 Ke3-d2 Kg3-f3 Bf6-a1 Bf5-e4 Ba1-c3 Kf3-g2 Bc3-f6 Kg2-g3 Kd2-c1 Be4-b1 Kc1-d2 Kg3-g2 Bf6-a1
 52/57+	04:39	 436.623k	1.563k	+2,33	Bf5-h3
 52/57+	05:20	 538.195k	1.680k	+2,41	Bf5-h3
 52/57	10:53	 1.378.699k	2.109k	+3,53	Bf5-h3

but stockfish found it in 4:39 min on a amd laptop with 4 threads 1024 mb hash and 6-piece syzygy.

@Ludmil: As you can see it is a mate in 4 or 5 moves

Kind regards
Bernhard

i think its just lucky, but its difficult to know for sure; hash size, TB probe depth + limit, SSD/HDD are all variables that seem so flexible (minus SSD being superior to HDD ofc) the engine authors themselves only really seem to be able to guess at optimal settings.

re Bh3, it is decisively best & thus perfectly valid as a test position, humans & machines alike agree on this, end of story there.

yeah, it is easy to say end of story, but the problem is things are way more complicated.

Bh3 is indeed the best move, and very elegant apart from that, but the problem is how to make engines choose it, when they see very large advantage in simpler lines too? why sac when other simpler lines give big advantage too? as a programmer, what would you do to resolve that?

any draws resulting from above position are simply fabulous fairy-tale fortresses of the kind: KBP vs bare K and wrong bishop, though far beyond the engine horizon, KBPP vs KB, with pawns on a and g files, also far beyond the engine horizon, or KBPP, with d and g, or a and d pawns, also beyond the engine horizon; when you add couple more pawns, tablebases do not help either. so what you do to resolve that?

I remember this position was also analysed by some GM's and was once the whole article of Mig Greengard of TWIC, they claimed that Bh3 is required for the win. All other alternatives will result to a draw. But we need concrete proof with current powerful CPU and engines if Bh3 is really the only way to win.

Copied from the above line:

51/56 02:17 182.220k 1.323k +2,26 a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Be4-b1 Kd2-e3 Bb1-f5 Ke3-d2 Kg3-f3 Bf6-a1 Bf5-e4 Ba1-c3 Kf3-g2 Bc3-f6 Kg2-g3 Kd2-c1 Be4-b1 Kc1-d2 Kg3-g2 Bf6-a1


[pgn]
[Event "Shirov Immortal Endgame"]
[Site "Analysis"]
[Date "2016.12.05"]
[Round "?"]
[White "-"]
[Black "Shirov,A"]
[Result "*"]
[Time "16:36:56"]
[WhiteElo "2400"]
[TimeControl "60+0"]
[SetUp "1"]
[FEN "8/8/4kpp1/3p1b2/p6P/2B5/6P1/6K1 b - - 0 1"]
[Termination "unterminated"]
[PlyCount "54"]
[WhiteType "human"]
[BlackType "human"]

1. ... a3 2. Kf2 Bc2 3. Ke2 Be4 4. g3 Kf5 5. Bd4 Kg4 6. Bxf6 Kxg3 7. Kd2
Kf4 8. Bd4 a2 9. Bf6 Kg4 10. Ke3 Kf5 11. Bc3 Bh1 12. Kd2 Bf3 13. Kc2 Kf4
14. Kd3 Bg2 15. Kd2 Kg3 16. Bf6 Kg4 17. Kc1 Be4 18. Kd2 Kh3 19. Ke2 Kg3 20.
Kd2 Bb1 21. Ke3 Bf5 22. Kd2 Kf3 23. Ba1 Be4 24. Bc3 Kg2 25. Bf6 Kg3 26. Kc1
Bb1 27. Kd2 Kg2 28. Ba1 =
[/pgn]

This is a dead drawn, thanks to opposite B. The dark-squared B of white can even be sac'd on d-pawn (assuming g and h pawns were exchanged before giving it up on d-pawn of black.
So I think, the previous analysis of GMs were correct, the position will be a draw if Bh3 is not played.

But strangely the above line of SF gives +2,26 for black when that line is in fact a draw.

Be4 is stronger.

the problem with eval is that, for me personally, it is very difficult to find an eval pattern to code with such fortresses, unless you want to spend 500+ lines of code just on precisely coding this single opposite bishop ending.

if this position still migth be a draw (which I am still not certain), there will be thousands similar positions, very similar indeed, but where a very small detail, worth 10cps or so, will already swing the balance to a win.

btw. guess how many hours on massive cores SF will need to find the best move below:

[d]8/8/4kpp1/3p1b2/p6P/2B4N/6P1/6K1 b - - 0 1

[AdminX]

May be it was a lucky run:

Code: Select all

FEN: 8/8/4kpp1/3p1b2/p6P/2B5/6P1/6K1 b - - 0 1
St-mod-16-12-02:
Found 510 tablebases
 50/56	02:12	 173.020k	1.306k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Kg3-g2 Kd2-e3 Be4-c2 Ke3-d2 Bc2-b1 Bf6-a1 Kg2-h3 Ba1-f6 Kh3-g4 Kd2-c1 Kg4-f4 Kc1-d2 Bb1-f5 Bf6-d4 Kf4-g4 Bd4-f6 Kg4-f3 Bf6-a1
51/56	02:17	 182.220k	1.323k	+2,26	a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Be4-b1 Kd2-e3 Bb1-f5 Ke3-d2 Kg3-f3 Bf6-a1 Bf5-e4 Ba1-c3 Kf3-g2 Bc3-f6 Kg2-g3 Kd2-c1 Be4-b1 Kc1-d2 Kg3-g2 Bf6-a1
 52/57+	04:39	 436.623k	1.563k	+2,33	Bf5-h3
 52/57+	05:20	 538.195k	1.680k	+2,41	Bf5-h3
 52/57	10:53	 1.378.699k	2.109k	+3,53	Bf5-h3

but stockfish found it in 4:39 min on a amd laptop with 4 threads 1024 mb hash and 6-piece syzygy.

@Ludmil: As you can see it is a mate in 4 or 5 moves

Kind regards
Bernhard

i think its just lucky, but its difficult to know for sure; hash size, TB probe depth + limit, SSD/HDD are all variables that seem so flexible (minus SSD being superior to HDD ofc) the engine authors themselves only really seem to be able to guess at optimal settings.

re Bh3, it is decisively best & thus perfectly valid as a test position, humans & machines alike agree on this, end of story there.

yeah, it is easy to say end of story, but the problem is things are way more complicated.

Bh3 is indeed the best move, and very elegant apart from that, but the problem is how to make engines choose it, when they see very large advantage in simpler lines too? why sac when other simpler lines give big advantage too? as a programmer, what would you do to resolve that?

any draws resulting from above position are simply fabulous fairy-tale fortresses of the kind: KBP vs bare K and wrong bishop, though far beyond the engine horizon, KBPP vs KB, with pawns on a and g files, also far beyond the engine horizon, or KBPP, with d and g, or a and d pawns, also beyond the engine horizon; when you add couple more pawns, tablebases do not help either. so what you do to resolve that?

I remember this position was also analysed by some GM's and was once the whole article of Mig Greengard of TWIC, they claimed that Bh3 is required for the win. All other alternatives will result to a draw. But we need concrete proof with current powerful CPU and engines if Bh3 is really the only way to win.

Copied from the above line:

51/56 02:17 182.220k 1.323k +2,26 a4-a3 Kg1-f2 Bf5-c2 Kf2-e2 Bc2-e4 g2-g3 Ke6-f5 Bc3-d4 Kf5-g4 Bd4xf6 Kg4xg3 Ke2-d2 Kg3-f4 Bf6-d4 a3-a2 Bd4-f6 Kf4-g4 Kd2-e3 Kg4-f5 Bf6-c3 Be4-h1 Ke3-d2 Bh1-f3 Kd2-c2 Kf5-f4 Kc2-d3 Bf3-g2 Kd3-d2 Kf4-g3 Bc3-f6 Kg3-g4 Kd2-c1 Bg2-e4 Kc1-d2 Kg4-h3 Kd2-e2 Kh3-g3 Ke2-d2 Be4-b1 Kd2-e3 Bb1-f5 Ke3-d2 Kg3-f3 Bf6-a1 Bf5-e4 Ba1-c3 Kf3-g2 Bc3-f6 Kg2-g3 Kd2-c1 Be4-b1 Kc1-d2 Kg3-g2 Bf6-a1


[pgn]
[Event "Shirov Immortal Endgame"]
[Site "Analysis"]
[Date "2016.12.05"]
[Round "?"]
[White "-"]
[Black "Shirov,A"]
[Result "*"]
[Time "16:36:56"]
[WhiteElo "2400"]
[TimeControl "60+0"]
[SetUp "1"]
[FEN "8/8/4kpp1/3p1b2/p6P/2B5/6P1/6K1 b - - 0 1"]
[Termination "unterminated"]
[PlyCount "54"]
[WhiteType "human"]
[BlackType "human"]

1. ... a3 2. Kf2 Bc2 3. Ke2 Be4 4. g3 Kf5 5. Bd4 Kg4 6. Bxf6 Kxg3 7. Kd2
Kf4 8. Bd4 a2 9. Bf6 Kg4 10. Ke3 Kf5 11. Bc3 Bh1 12. Kd2 Bf3 13. Kc2 Kf4
14. Kd3 Bg2 15. Kd2 Kg3 16. Bf6 Kg4 17. Kc1 Be4 18. Kd2 Kh3 19. Ke2 Kg3 20.
Kd2 Bb1 21. Ke3 Bf5 22. Kd2 Kf3 23. Ba1 Be4 24. Bc3 Kg2 25. Bf6 Kg3 26. Kc1
Bb1 27. Kd2 Kg2 28. Ba1 =
[/pgn]

This is a dead drawn, thanks to opposite B. The dark-squared B of white can even be sac'd on d-pawn (assuming g and h pawns were exchanged before giving it up on d-pawn of black.
So I think, the previous analysis of GMs were correct, the position will be a draw if Bh3 is not played.

But strangely the above line of SF gives +2,26 for black when that line is in fact a draw.

Code: Select all

[Event "Linares 15th"]
[Site "Linares"]
[Date "1998.03.04"]
[Round "10"]
[White "Topalov, Veselin"]
[Black "Shirov, Alexei"]
[Result "0-1"]
[ECO "D85"]
[WhiteElo "2740"]
[BlackElo "2710"]
[Annotator "Ftacnik,L"]
[PlyCount "106"]
[EventDate "1998.02.22"]
[EventType "tourn"]
[EventRounds "12"]
[EventCountry "ESP"]
[EventCategory "21"]
[SourceTitle "CBM 064"]
[Source "ChessBase"]
[SourceDate "1998.06.02"]

1. d4 Nf6 2. c4 g6 3. Nc3 d5 4. cxd5 Nxd5 5. e4 Nxc3 6. bxc3 Bg7 7. Bb5+ c6 8.
Ba4 O-O 9. Ne2 Nd7 10. O-O e5 11. f3 (11. Be3 Re8 12. Qc2 Qe7 13. f3 Nb6 14.
Bb3 Be6 15. Rad1 (15. Bf2 Bc4 16. a4 Qe6 17. Rfb1 Rab8 18. dxe5 Bxe5 19. a5 Nc8
20. Bxc4 Qxc4 21. Ra4 Qe6 22. f4 Bg7 23. e5 {1-0 Neverov,V-Smikovski,I/
Chigorin mem 1995/ (41)}) 15... Rad8 16. c4 exd4 17. Nxd4 Bc8 18. Ne2 Nd7 19.
c5 Bf8 20. Rc1 Ne5 21. Rfd1 Rxd1+ 22. Rxd1 Nd7 {Ulibin,M-Novik,M/Moscow
Alekhine op 1992/0-1 (51)}) (11. Ba3 Re8 12. Bb3 Nb6 $6 (12... exd4 $1) 13. f4
exd4 14. f5 gxf5 15. Ng3 $40 {Shirov,A-Smejkal,J/GER-chT/1992/}) 11... Qe7 12.
Be3 Rd8 $1 13. Qc2 $6 (13. Qc1 Nb6 14. Bb3) 13... Nb6 14. Bb3 Be6 15. Rad1 (15.
f4 exf4 16. Bxe6 fxe6 17. Bxf4 Nc4 $11) 15... Nc4 $1 16. Bc1 b5 $1 17. f4 $1 {
Topalov has to exploit his central pawn majority in an attempt to compensate
for the knight on c4.} exd4 18. Nxd4 Bg4 19. Rde1 (19. Rd3 $2 Qxe4 $19) 19...
Qc5 20. Kh1 a5 (20... Bxd4 21. cxd4 Qxd4 22. f5 $40) 21. h3 $2 {Very serious
mistake, since the bishop will be much better on d7 square.} (21. a4 $1 bxa4
22. Ba2 $132) 21... Bd7 22. a4 bxa4 23. Ba2 Be8 $1 24. e5 (24. Qxa4 Nb6 25. Qb3
a4 26. Qa3 $15) 24... Nb6 25. f5 Nd5 26. Bd2 $6 (26. Rf3 $5) (26. f6 $5) 26...
Nb4 $1 27. Qxa4 (27. cxb4 Qxc2 28. Nxc2 Rxd2 $17) 27... Nxa2 28. Qxa2 Bxe5 29.
fxg6 (29. Re4 Bg7 30. Bg5 Rd5 $17) 29... hxg6 30. Bg5 Rd5 $1 {Two bishops and
a spare pawn favour black, but he has to survive the dangers of white attack.}
31. Re3 (31. Ne2 a4 $17) 31... Qd6 32. Qe2 Bd7 (32... c5 33. Bf6 Bxf6 34. Rxe8+
Rxe8 35. Qxe8+ Kg7 36. Nb5 Qd7 $17) 33. c4 Bxd4 34. cxd5 Bxe3 35. Qxe3 Re8 $1 (
35... Qxd5 36. Bf6) 36. Qc3 (36. Qc1 Qxd5 37. Bf6 Bxh3 $1) 36... Qxd5 37. Bh6 (
37. Bf6 Re2 $19) 37... Re5 {White still preserves drawing chances due to the
opposite colour bishops.} (37... Qe5 38. Qc4) 38. Rf3 (38. Re1 $2 Bxh3 $1) (38.
Rb1 Bxh3 39. Qxh3 Rh5 $19) 38... Qc5 39. Qa1 (39. Qxc5 Rxc5 40. Ra3 Be6 41. Bd2
Rc2 42. Bxa5 Bd5 43. Rg3 $17) 39... Bf5 40. Re3 f6 (40... Rxe3 $4 41. Qg7#) 41.
Rxe5 Qxe5 42. Qa2+ (42. Qxe5 fxe5 43. Kg1 Kf7 $19) 42... Qd5 $1 43. Qxd5+ (43.
Qa3 c5 44. Qxa5 Bxh3 45. Qd2 Bxg2+ (45... Qc6 $17) 46. Qxg2 Qh5+ 47. Qh2 Qxh2+
48. Kxh2 g5 49. Kg3 Kf7 50. Kf3 $17) 43... cxd5 44. Bd2 a4 45. Bc3 Kf7 46. h4
Ke6 47. Kg1 Bh3 $3 {[%csl Yf5] Fascinating idea of Shirov, since black has to
push his two pawns a4,d5 as quickly as possible. The bishop would obstruct the
way of the king on the natural square e4, on h3 it wins a precious tempo.} (
47... a3 48. Kf2 a2 49. Ke3 Bg4 50. g3 Kf5 51. Bd4 $13) (47... Be4 {Shirov -+}
48. Kf2 Kf5 49. g3 $13) 48. gxh3 (48. Kf2 Kf5 49. Kf3 Bxg2+ 50. Kxg2 Ke4 51.
Bxf6 d4 52. Be7 Kd3 53. Kf2 Kc2 54. Bb4 d3 $19) 48... Kf5 49. Kf2 Ke4 50. Bxf6
(50. Ke2 a3 51. Kd2 d4 52. Ba1 f5 $19) 50... d4 51. Be7 (51. Ke2 $2 a3 $19)
51... Kd3 $1 52. Bc5 Kc4 53. Be7 (53. Bxd4 Kxd4 54. Ke2 Kc3 55. Kd1 Kb2 $19)
53... Kb3 (53... Kb3 54. Bc5 d3 55. Ke3 Kc2 56. Bb4 a3 $19) 0-1

[Lyudmil Tsvetkov]

after Bh3 Kf2 Kf5 Kf3:

[d]8/8/5pp1/3p1k2/p6P/2B2K1b/6P1/8 b - - 0 3

SF again can not find Bg2. guess why?

Supposedly, because Bg2 is a bad capture, and engine usually order bad captures at the end of the queue. but here this is the (probably) only winning move.

same with Bh3 in the original position, it is a quiet move, quiet moves are ordered lower than captures, and also a quiet sac, a notion that most engines probably lack altogether.

[MikeGL]

Be4 is stronger.

I analysed using some engines but I cannot find the winning line after Be4.
White can hold the position after Be4 instead of Bh3.

after Bh3 Kf2 Kf5 Kf3:

[d]8/8/5pp1/3p1k2/p6P/2B2K1b/6P1/8 b - - 0 3

SF again can not find Bg2. guess why?

Supposedly, because Bg2 is a bad capture, and engine usually order bad captures at the end of the queue. but here this is the (probably) only winning move.

same with Bh3 in the original position, it is a quiet move, quiet moves are ordered lower than captures, and also a quiet sac, a notion that most engines probably lack altogether.

True SF can't find Bxg2 on your above position, but critter1.6 32bit on my slow 32bit laptop finds it in 1 sec or less, with hashtable cleared on each run.

[peter]

Dead draw, Lyudmil.

BTW, don't you think that to be a fittle funny too:

...do not look at what engines suggest...
...I ran a 10 games shootout with Komodo 10.1...
...games were at 1 min. bullet...