Crystal 8

[acase]

Here is an amusing mate in 5 that seems to give Stockfish trouble and humans of course, but ALL Crystal versions solve and The Huntsman 1 solves rather easily. It's by "Marín i Llovet, Valentí" from 1905:

[d]qb6/3N2p1/r2p4/pP1k2p1/1pp3R1/6BB/2P1P3/2N1K3 w - - 0 1

Here is a link to it--->https://www.yacpdb.org/#65449

Code: Select all

Matefish 3.0 64 by the Stockfish developers and J. Oster
position fen qb6/3N2p1/r2p4/pP1k2p1/1pp3R1/6BB/2P1P3/2N1K3 w - - 0 1

d

 +---+---+---+---+---+---+---+---+
 | q | b |   |   |   |   |   |   | 8
 +---+---+---+---+---+---+---+---+
 |   |   |   | N |   |   | p |   | 7
 +---+---+---+---+---+---+---+---+
 | r |   |   | p |   |   |   |   | 6
 +---+---+---+---+---+---+---+---+
 | p | P |   | k |   |   | p |   | 5
 +---+---+---+---+---+---+---+---+
 |   | p | p |   |   |   | R |   | 4
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   | B | B | 3
 +---+---+---+---+---+---+---+---+
 |   |   | P |   | P |   |   |   | 2
 +---+---+---+---+---+---+---+---+
 |   |   | N |   | K |   |   |   | 1
 +---+---+---+---+---+---+---+---+
   a   b   c   d   e   f   g   h

Fen: qb6/3N2p1/r2p4/pP1k2p1/1pp3R1/6BB/2P1P3/2N1K3 w - - 0 1
PositionKey: 236FC10937D981EE
MaterialKey: 8092EDDCAA3A09F5
Checkers:
Legal moves: 32
setoption name PNS Hash value 1024
setoption name ProofNumberSearch value true
go mate 5
info string Starting Proof-Number Search ...
info string Success! Mate in 5 found!
info time 725 multipv 1 depth 9 seldepth 9 nodes 389209 nps 536840 tbhits 0 score mate 5 pv g3h4 a8a7 e2e4 d5d4 c1e2 d4e3 h4f2 e3f3 e2g1
bestmove g3h4 ponder a8a7

Sure, but it's not Stockfish, it's Matefish 3.

[Werewolf]

Have You noticed that engines are not intended for mate solving . Huntsman is:
info depth 35 seldepth 10 multipv 1 score mate 5 nodes 1700221 nps 7265901 tbhits 0 time 234 pv g3h4 g5h4 c1b3 c4b3 c2c4 b4c3 g4a4 b8a7 e2e4 means 234 ms

Well....
Huntsman may be a great mate-finder, but it's not a great tactician c/p with SF for more general positions.
So now it's like we need 3 engines:
One for normal chess - SF
One for tactics - I think Black Diamond is great, but also Crystal
and one for mates like Chest / Huntsman.

[acase]

Have You noticed that engines are not intended for mate solving . Huntsman is:
info depth 35 seldepth 10 multipv 1 score mate 5 nodes 1700221 nps 7265901 tbhits 0 time 234 pv g3h4 g5h4 c1b3 c4b3 c2c4 b4c3 g4a4 b8a7 e2e4 means 234 ms

True, but The Huntsman 1 couldn't solve that mate in 20 I posted earlier in this thread. Perhaps it could though if we let it think for a few days.

[Chessqueen]

Here is an amusing mate in 5 that seems to give Stockfish trouble and humans of course, but ALL Crystal versions solve and The Huntsman 1 solves rather easily. It's by "Marín i Llovet, Valentí" from 1905:

[d]qb6/3N2p1/r2p4/pP1k2p1/1pp3R1/6BB/2P1P3/2N1K3 w - - 0 1

Here is a link to it--->https://www.yacpdb.org/#65449

I'm becoming seriously concerned about SF and its clones. In the position above - a mere mate in 5 - the moderns are really struggling.

All on up to 64 cores / threads unless otherwise stated:
SF Dev 2nd April 2024 - FAILED. Not within 5 mins.
SF Cluster 480 Threads - FAILED. Not within 5 mins.
SF Cluster 2400 Threads - FAILED. Not within 5 mins.
Berserk 13 - FAILED. Not within 5 mins.
Fritz 19 - FAILED. Not within 5 mins.
Patricia 2 - FAILED. Not within 5 mins.
CS Tal 2 NNUE - took over a minute to get the mate.
Crystal 8 - Passed (within a few seconds).
Black Diamond 13 - Passed (within a few seconds)
Rybka 4.1 - Passed instantly
Deep Junior Yokohama - (Passed within a few seconds)
HIARCS Paderborn on 8 threads - Passed (within a few seconds).
HIARCS 13.88 (Oct 2022) on iPhone - Passed (within a few seconds).
HIARCS 15.3.1 (16 threads) - Took about 15 seconds but got there in the end.
Chessmaster GM (King 3.50) - Passed instantly.

Monarch which is NOT even rated 2000 can solve it in less than 1 seconds

489 Monarch 1.7 1987 +15 −15 48.4% +15.9 19.3% 1511
490‑491 Goldfish 1.13.0 64-bit 1984 +18 −18 49.4% +7.7 26.9% 92

FEN: qb6/3N2p1/r2p4/pP1k2p1/1pp3R1/6BB/2P1P3/2N1K3 w - - 0 1

Monarch(v1.7):
3/14 00:00 2k 49k +6.09 b5xa6 Kd5-e6 Nd7xb8 Qa8xb8 Rg4xg5+ Ke6-f7 Rg5xa5
4/14 00:00 3k 99k +6.09 b5xa6 Kd5-e6 Nd7xb8 Qa8xb8 Rg4xg5+ Ke6-f7 Rg5xa5
5/16 00:00 18k 577k +6.09 b5xa6 Kd5-e6 Nd7xb8 Qa8xb8 Rg4xg5+ Ke6-f7 Rg5xa5
6/18 00:00 69k 1,099k +6.09 b5xa6 Kd5-e6 Rg4xg5+ Ke6-f7 Nd7xb8 Qa8xb8 Rg5xa5
7/24 00:00 262k 1,669k +5.99 b5xa6 Kd5-e6 Rg4xg5+ Ke6-f7 Nd7xb8 Qa8xb8 Rg5xa5 Kf7-e8
7/29+ 00:00 962k 1,983k +6.00 Bg3-h4
7/29 00:00 1,191k 2,061k +M5 Bg3-h4 g5xh4 Nc1-b3 c4xb3 c2-c4+ b4xc3/ep Rg4-a4 c3-c2 e2-e4+

[Steve Maughan]

Monarch which is NOT even rated 2000 can solve it in less than 1 seconds

489 Monarch 1.7 1987 +15 −15 48.4% +15.9 19.3% 1511
490‑491 Goldfish 1.13.0 64-bit 1984 +18 −18 49.4% +7.7 26.9% 92

FEN: qb6/3N2p1/r2p4/pP1k2p1/1pp3R1/6BB/2P1P3/2N1K3 w - - 0 1

Monarch(v1.7):
3/14 00:00 2k 49k +6.09 b5xa6 Kd5-e6 Nd7xb8 Qa8xb8 Rg4xg5+ Ke6-f7 Rg5xa5
4/14 00:00 3k 99k +6.09 b5xa6 Kd5-e6 Nd7xb8 Qa8xb8 Rg4xg5+ Ke6-f7 Rg5xa5
5/16 00:00 18k 577k +6.09 b5xa6 Kd5-e6 Nd7xb8 Qa8xb8 Rg4xg5+ Ke6-f7 Rg5xa5
6/18 00:00 69k 1,099k +6.09 b5xa6 Kd5-e6 Rg4xg5+ Ke6-f7 Nd7xb8 Qa8xb8 Rg5xa5
7/24 00:00 262k 1,669k +5.99 b5xa6 Kd5-e6 Rg4xg5+ Ke6-f7 Nd7xb8 Qa8xb8 Rg5xa5 Kf7-e8
7/29+ 00:00 962k 1,983k +6.00 Bg3-h4
7/29 00:00 1,191k 2,061k +M5 Bg3-h4 g5xh4 Nc1-b3 c4xb3 c2-c4+ b4xc3/ep Rg4-a4 c3-c2 e2-e4+

WOW — I'm delighted people still use and enjoy Monarch!

Maverick also finds the mate in 1 second:

Engine: Maverick 1.5 x64 (512 MB)
by Steve Maughan

Code: Select all

13/33  0:00   +6.15    1.bxa6 Ke6 2.Rd4+ Kf7 3.Nxb8 Qxb8 
                       4.Rxd6 Qa8 5.Be6+ Kf8 6.Bxc4 a4 
                       7.Bd5 Qc8 8.Rc6 (2.012.917) 

13/33  0:01   +6.16++  1.Bh4 (3.548.148) 3540 

13/33  0:01   +M5      1.Bh4 gxh4 2.Nb3 cxb3 3.c4+ bxc3 
                       4.Ra4 g6 5.e4+ (3.867.617) 3540

[Werewolf]

Have You noticed that engines are not intended for mate solving . Huntsman is:
info depth 35 seldepth 10 multipv 1 score mate 5 nodes 1700221 nps 7265901 tbhits 0 time 234 pv g3h4 g5h4 c1b3 c4b3 c2c4 b4c3 g4a4 b8a7 e2e4 means 234 ms

True, but The Huntsman 1 couldn't solve that mate in 20 I posted earlier in this thread. Perhaps it could though if we let it think for a few days.

That's a big ask of any Alpha-Beta though. It's 39 plies deep with lots of branches at the end.
Do you know of any engine that can solve it? Lc0 gets it, but only in the sense that it sees Kf2 has good chances.

I think Crystal / SF would need months, not days.

[hgm]

Fairy-Max needs 81 sec (i7 3.2GHz, single core, of course):

Code: Select all

 15	    #5 	115.6M	1:23.77	g3h4 g5h4 c1b3 c4b3 c2c4 b4c3 g4a4 g7g5 e2e4
 14	    #5 	112.8M	1:22.44	g3h4 g5h4 c1b3 c4b3 c2c4 b4c3 g4a4 g7g5 e2e4
 13	    #5 	110.3M	1:21.15	g3h4 g5h4 c1b3 c4b3 c2c4 b4c3 g4a4 g7g5 e2e4
 13	+4.53 	69.8M  	0:51.24	b5a6 d5e6 g4g5 e6f7 g5f5 f7e7 a6a7 a8a7 g3h4 e7d7 f5a5 d7c7 a5a7 b8a7 h3f5 c4c3 c1d3
 12	+4.60 	25.4M  	0:18.51	b5a6 d5e6 g4g5 e6f7 g5f5 f7e7 a6a7 a8a7 g3h4 e7d7 f5a5 d7c7 a5a7 b8a7 h4f2 a7f2
 11	+4.97 	10.7M  	0:07.80	b5a6 d5e6 g4g5 e6f7 g5f5 f7e7 a6a7 a8a7 g3h4 e7d7 f5a5 d7c7 a5a7 b8a7 h3f5
 10	+4.90 	2.25M  	0:01.66	b5a6 d5e6 g4g5 e6f7 g5f5 f7e7 d7b8 a8b8 f5d5 b8b6 g3d6 e7f6 d6e5
  9	+4.67 	985608	0:00.74	b5a6 d5e6 g4g5 e6f7 g5f5 f7e7 d7b8 a8b8 f5a5 c4c3 a6a7
  8	+4.68 	306416	0:00.24	b5a6 d5e6 g4g5 e6f7 g5f5 f7e7 d7b8 a8b8
  7	+4.49 	120575	0:00.10	b5a6 d5e6 g4g5 e6f7 d7b8 a8b8 g5f5 f7e7 f5a5
  6	+4.51 	78427  	0:00.07	b5a6 d5e6 g4g5 e6f7 d7b8 a8b8 g5a5
  6	+3.96 	24497  	0:00.03	h3g2 d5e6 g2a8 a6a8 d7b8 a8b8 g4g5
  5	+3.96 	9652  	0:00.01	h3g2 d5e6 g2a8 a6a8 d7b8 a8b8
  4	+3.96 	3570  	0:00.00	h3g2 d5e6 g2a8 a6a8 d7b8
  3	+3.96 	1060  	0:00.00	h3g2 d5e6 g2a8 a6a8
  2	+3.96 	354    	0:00.00	h3g2 d5e6 g2a8
  2	+3.28 	264    	0:00.00	g4g5 d5e4 b5a6
  2	+2.74 	184    	0:00.00	b5a6 c4c3 d7b8 a8b8 g4g5
  1	+2.36 	11    	0:00.00	b5a6
  0	#

[acase]

Have You noticed that engines are not intended for mate solving . Huntsman is:
info depth 35 seldepth 10 multipv 1 score mate 5 nodes 1700221 nps 7265901 tbhits 0 time 234 pv g3h4 g5h4 c1b3 c4b3 c2c4 b4c3 g4a4 b8a7 e2e4 means 234 ms

True, but The Huntsman 1 couldn't solve that mate in 20 I posted earlier in this thread. Perhaps it could though if we let it think for a few days.

That's a big ask of any Alpha-Beta though. It's 39 plies deep with lots of branches at the end.
Do you know of any engine that can solve it? Lc0 gets it, but only in the sense that it sees Kf2 has good chances.

I think Crystal / SF would need months, not days.

I couldn't find an engine that solved it, however Crystallite 1.1a did find a mate in 21 where it demonstrated that it saw the idea but only after about a 4 hour search on my old 20 core machine, but still it didn't find mate in 20 which is the stipulation of that direct mate problem.

If you want another nasty one that is shorter (25 plies) try out this white to move and mate in 13. The thing that makes this one so difficult is that there are several 15 move mates, but only one 13 move mate:

[d]3K1nbr/1p3pp1/6kp/2p2p1R/1p3P2/3Pb3/p1Q1B3/R2N2B1 w - - 0 1

I haven't had any engines solve this one yet, but I don't have every engine so I won't say that there isn't something out there that can solve it, I'll post the pgn later.

[acase]

Have You noticed that engines are not intended for mate solving . Huntsman is:
info depth 35 seldepth 10 multipv 1 score mate 5 nodes 1700221 nps 7265901 tbhits 0 time 234 pv g3h4 g5h4 c1b3 c4b3 c2c4 b4c3 g4a4 b8a7 e2e4 means 234 ms

True, but The Huntsman 1 couldn't solve that mate in 20 I posted earlier in this thread. Perhaps it could though if we let it think for a few days.

That's a big ask of any Alpha-Beta though. It's 39 plies deep with lots of branches at the end.
Do you know of any engine that can solve it? Lc0 gets it, but only in the sense that it sees Kf2 has good chances.

I think Crystal / SF would need months, not days.

I couldn't find an engine that solved it, however Crystallite 1.1a did find a mate in 21 where it demonstrated that it saw the idea but only after about a 4 hour search on my old 20 core machine, but still it didn't find mate in 20 which is the stipulation of that direct mate problem.

If you want another nasty one that is shorter (25 plies) try out this white to move and mate in 13. The thing that makes this one so difficult is that there are several 15 move mates, but only one 13 move mate:

[d]3K1nbr/1p3pp1/6kp/2p2p1R/1p3P2/3Pb3/p1Q1B3/R2N2B1 w - - 0 1

I haven't had any engines solve this one yet, but I don't have every engine so I won't say that there isn't something out there that can solve it, I'll post the pgn later.

Here is the pgn to the mate in 13 by Peter Krug:

[pgn][Event "Mate in 13"]
[Site "Columbus, Ohio"]
[Date "2024.02.22"]
[Round "?"]
[White "Peter Krug"]
[Black "?"]
[Result "1-0"]
[Time "21:56:16"]
[WhiteElo "0"]
[TimeControl "0"]
[SetUp "1"]
[FEN "3K1nbr/1p3pp1/6kp/2p2p1R/1p3P2/3Pb3/p1Q1B3/R2N2B1 w - - 0 1"]
[Termination "normal"]
[PlyCount "25"]
[WhiteType "program"]
[BlackType "human"]

1. d4 Ne6+ 2. Ke7 Nxd4 3. Nxe3 Nxc2 4. Nxf5 Nxa1 5. Bd4 cxd4 6. Bd3 Kxh5 7.
Be2+ Kg6 8. Nh4+ Kh7 9. Bh5 Nb3 10. f5 g6 11. Kf6 gxh5 12. Ng6 fxg6 13.
fxg6# 1-0[/pgn]

[Spliffjiffer]

Have You noticed that engines are not intended for mate solving . Huntsman is:
info depth 35 seldepth 10 multipv 1 score mate 5 nodes 1700221 nps 7265901 tbhits 0 time 234 pv g3h4 g5h4 c1b3 c4b3 c2c4 b4c3 g4a4 b8a7 e2e4 means 234 ms

True, but The Huntsman 1 couldn't solve that mate in 20 I posted earlier in this thread. Perhaps it could though if we let it think for a few days.

That's a big ask of any Alpha-Beta though. It's 39 plies deep with lots of branches at the end.
Do you know of any engine that can solve it? Lc0 gets it, but only in the sense that it sees Kf2 has good chances.

I think Crystal / SF would need months, not days.

I couldn't find an engine that solved it, however Crystallite 1.1a did find a mate in 21 where it demonstrated that it saw the idea but only after about a 4 hour search on my old 20 core machine, but still it didn't find mate in 20 which is the stipulation of that direct mate problem.

If you want another nasty one that is shorter (25 plies) try out this white to move and mate in 13. The thing that makes this one so difficult is that there are several 15 move mates, but only one 13 move mate:

[d]3K1nbr/1p3pp1/6kp/2p2p1R/1p3P2/3Pb3/p1Q1B3/R2N2B1 w - - 0 1

I haven't had any engines solve this one yet, but I don't have every engine so I won't say that there isn't something out there that can solve it, I'll post the pgn later.

Here is the pgn to the mate in 13 by Peter Krug:

[pgn][Event "Mate in 13"]
[Site "Columbus, Ohio"]
[Date "2024.02.22"]
[Round "?"]
[White "Peter Krug"]
[Black "?"]
[Result "1-0"]
[Time "21:56:16"]
[WhiteElo "0"]
[TimeControl "0"]
[SetUp "1"]
[FEN "3K1nbr/1p3pp1/6kp/2p2p1R/1p3P2/3Pb3/p1Q1B3/R2N2B1 w - - 0 1"]
[Termination "normal"]
[PlyCount "25"]
[WhiteType "program"]
[BlackType "human"]

1. d4 Ne6+ 2. Ke7 Nxd4 3. Nxe3 Nxc2 4. Nxf5 Nxa1 5. Bd4 cxd4 6. Bd3 Kxh5 7.
Be2+ Kg6 8. Nh4+ Kh7 9. Bh5 Nb3 10. f5 g6 11. Kf6 gxh5 12. Ng6 fxg6 13.
fxg6# 1-0[/pgn]

Chest can proof that its a #13 or less in around 10 min on my computer...given that it uses "optimal" settings....by sying so, here is my run:

Code: Select all

FEN: 3K1nbr/1p3pp1/6kp/2p2p1R/1p1P1P2/4b3/p1Q1B3/R2N2B1 b - - 0 1

ChestUCI:
FEN: 3K1nbr/1p3pp1/6kp/2p2p1R/1p3P2/3Pb3/p1Q1B3/R2N2B1 w - - (9+13)
Stellungs-Analyse: C0/R0/K3/P9/X18 W:8/35
Suche nach Spezial-Matt [C0/R-6/K3/P9/X21] in 126 ... (Hash=2044MB)
 13	10:18	 295.177k	1.635k	+M13	1.d3-d4
Suche abgeschlossen ... (Zeit=618.99s)
Matt in 13 gefunden ! (1 Lösung in 10:18)
 13/13	10:18	 1.046.936k	1.691k	+M13	1.d3-d4 Sf8-e6+ 2.Kd8-e7 Se6xd4 3.Sd1xe3 Sd4xc2 4.Se3xf5 Sc2xa1 5.Lg1-d4 c5xd4 6.Le2-d3 Kg6xh5 7.Ld3-e2+ Kh5-g6 8.Sf5-h4+ Kg6-h7 9.Le2-h5 g7-g6 10.Ke7-f6 g6xh5 11.f4-f5 d4-d3 12.Sh4-g6 h5-h4 13.Sg6-f8+

with these parameters given it thinks like this about the best responses for black after 1. d4!!? :

Code: Select all

FEN: 3K2br/1p3pp1/4n1kp/2p2p1R/1p1P1P2/4b3/p1Q1B3/R2N2B1 w - - 1 2

ChestUCI:
FEN: 3K1nbr/1p3pp1/6kp/2p2p1R/1p1P1P2/4b3/p1Q1B3/R2N2B1 b - - (9+13)
Stellungs-Analyse: S:9/19
Automatik-Suche nach 1. Spezial-Matt [C0/R-6/K3/P9/X21] in -12 ... (Hash=2044MB)
 12/12	10:17	 1.042.481k	1.689k	-M12	
Matt in -12 gefunden ! (19 Lösungen in 10:17)
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Kg6-f6 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Th8-h7 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Le3xd4 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Le3xg1 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Le3-f2 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Le3-c1 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Le3-d2 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Le3xf4 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Lg8-h7 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Sf8-h7 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... Sf8-d7 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... b7-b5 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... b7-b6 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... b4-b3 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... c5xd4 2.Dc2xf5+
 1/12	10:17	 1.042.481k	1.689k	-M1	1. ... c5-c4 2.Dc2xf5+
 4/12	10:17	 1.042.481k	1.689k	-M4	1. ... f7-f6 2.Dc2xf5+ Kg6-f7 3.Le2-c4+ Sf8-e6+ 4.Df5xe6+ Kf7-f8 5.De6-e8+
 9/12	10:17	 1.042.481k	1.689k	-M9	1. ... Kg6-h7 2.Dc2xf5+ Sf8-g6 3.Le2-d3 c5-c4 4.Sd1xe3 c4xd3 5.Th5xh6+ g7xh6 6.Df5-f6 Sg6-h4 7.Lg1-h2 h6-h5 8.Ta1-g1 Sh4-f5 9.Se3xf5 h5-h4 10.Tg1-g7+
 12/12	10:17	 1.042.481k	1.689k	-M12	1. ... Sf8-e6+ 2.Kd8-e7 Se6xd4 3.Sd1xe3 Sd4xc2 4.Se3xf5 Sc2xa1 5.Lg1-d4 c5xd4 6.Le2-d3 Kg6xh5 7.Ld3-e2+ Kh5-g6 8.Sf5-h4+ Kg6-h7 9.Le2-h5 g7-g6 10.Ke7-f6 g6xh5 11.f4-f5 d4-d3 12.Sh4-g6 h5-h4 13.Sg6-f8+