Mate in 12, but impossible ...

[Werewolf]

I've known about this puzzle for 15-20 years. It used to be unsolvable, now a number of engines find it quickly.

[chessica]

WinCHEST Ver.3.19i+, 08-Mar-2011
Options = -M64 -Z12 -11 -el -U -R7 -K3 -X35 -0 -5 -rS
Input file: STDIN
Reading job:
% created by ChestUCI Ver.5.2
W: Kf6 Qc8 Bd8 Pg6 (4)
B: Kg8 Qe8 Na8 Pa5 Pf7 Ph7 (6)
FEN: n1QBq1k1/5p1p/5KP1/p7/8/8/8/8 w - -
analysing (first special-mate in 12 moves):
Solution (in 12 moves):
d8c7
Time (virt) = 0.239 sec

PV= d8c7 e8c8 g6f7 g8h8 c7e5 c8c5 e5b2 a8c7 b2a1 a5a4 a1b2 a4a3 b2a1 a3a2 a1b2 a2a1q b2a1 c7d5 f6e6 d5c3 a1c3 c5e5 c3e5
d8c7 e8c8 g6f7 ...+10
a8c7 g6f7 ...+5
h7g6 c8e8 ...+2
e8f8 g6f7 ...+1
g8f8 g6g7 ...+1
a5a4 c8e8
a8b6 c8e8
e8d8 c8d8
f7g6 c8e8
g8h8 c8e8
h7h5 c8e8
h7h6 c8e8

end of solution tree (no duals except at top)
Total Time (virt) = 0.239 sec

[chrisw]

Is Crystal a matefinder, because obviously matefinders will find mates. What's interesting is what prevents (allegedly) the SF-model from finding a M12 with 70 plies of nominal search. Give SF the first move and the solution is almost immediate.

Ah, the notorious null move zugzwang.

Both Cstal an Rebel within less of a second here.

Code: Select all

 25/39	00:00	 1.191k	1.360k	 0,00	g6xh7+ Kg8-h8 Qc8-c3 Qe8-e6+ Kf6-g5+ Kh8xh7 Qc3xa5 Kh7-g8 Kg5-f4 f7-f5 Qa5xa8 Qe6-e4+ Qa8xe4 f5xe4 Kf4xe4
 25/39+	00:00	 1.240k	1.359k	+0,10	Bd8-c7
 25/25+	00:00	 1.286k	1.368k	+5,06	Bd8-c7
 25/28	00:00	 1.351k	1.406k	+M13	Bd8-c7 Qe8xc8 g6xf7+ Kg8-h8 Bc7-e5 Qc8-c5 Be5-b2 Na8-c7 Bb2-a1 a5-a4 Ba1-b2 a4-a3 Bb2-a1 a3-a2 Ba1-b2 a2-a1Q Bb2xa1 Qc5-d6+ Kf6-f5+ Qd6-f6+ Kf5xf6 h7-h6 f7-f8Q+ Kh8-h7 Qf8-g7+
 26/26	00:00	 1.356k	1.407k	+M12	Bd8-c7 Qe8xc8 g6xf7+ Kg8-h8 Bc7-e5 Qc8-c5 Be5-b2 Na8-c7 Bb2-a1 a5-a4 Ba1-b2 a4-a3 Bb2-a1 a3-a2 Ba1-b2 a2-a1Q Bb2xa1 Nc7-d5+ Kf6-e6+ Nd5-c3 Ba1xc3+ Qc5xc3 f7-f8Q+

Is it nullmove? Doubtful.

I suspect it’s because SF allows pruning at the root, disable that and it will probably find it. If so, begs the question, does pruning at the root actually give Elo, while completely eliminating certain moves from being considered?

[Werewolf]

Stockfish Cluster gets this, but not normal SF.

[Joerg Oster]

This one has been on Agadmator's channel for ten days, surprising nobody posted it, I guess some might think it falls into the dreaded forbidden class: chess positions unsolvable by engines, but, then again, it's a very beautiful position AND a forced mate in 12. 70 ply deep doesn't find it though.

https://www.youtube.com/watch?v=ZNnIsmHjcjE

Impossible? ChestUCI finds #12 in about 1 min (in 'Normal' mode) and in 3 sec (in 'AutoTurbo' mode)!

ChestUCI Ver.5.2:
FEN: n1QBq1k1/5p1p/5KP1/p7/8/8/8/8 w - - (4+6)
Stellungs-Analyse: C0/R0/K3/P6/X22 W:4/25
Suche nach Matt in 12 ... (Hash=256MB)
12 00:53 46.678.802 969.224 +M12 1.Bc7
Suche abgeschlossen ... (Zeit=79.80s)
Matt in 12 gefunden ! (1 Lösung in 01:19)
12/12 01:19 74.607.332 934.870 +M12 1.Bc7 Qxc8 2.gxf7+ Kh8 3.Be5 Qc5 4.Bb2 Nc7 5.Ba1 a4 6.Bb2 a3 7.Ba1 a2 8.Bb2 a1Q 9.Bxa1 Nd5+ 10.Ke6+ Nc3 11.Bxc3+ Qe5+ 12.Bxe5+
Best move: Bc7, Value: +Mate in 12, Depth: 12/12, Time: 01:19,854, 74.607.332 Nodes, 934.870 N/sec

ChestUCI Ver.5.2:
FEN: n1QBq1k1/5p1p/5KP1/p7/8/8/8/8 w - - (4+6)
Stellungs-Analyse: C0/R0/K3/P6/X22 W:4/25
AutoTurbo-Suche nach Spezial-Matt [C1/R0/K0/P0/X0] in 12 ... (Hash=256MB)
AutoTurbo-Suche nach Spezial-Matt [C0/R1/K0/P0/X0] in 12 ... (Hash=256MB)
AutoTurbo-Suche nach Spezial-Matt [C0/R2/K0/P0/X0] in 12 ... (Hash=256MB)
AutoTurbo-Suche nach Spezial-Matt [C0/R-2/K0/P0/X0] in 12 ... (Hash=256MB)
AutoTurbo-Suche nach Spezial-Matt [C0/R3/K0/P0/X0] in 12 ... (Hash=256MB)
AutoTurbo-Suche nach Spezial-Matt [C0/R-3/K0/P0/X0] in 12 ... (Hash=256MB)
AutoTurbo-Suche nach Spezial-Matt [C0/R0/K1/P0/X0] in 12 ... (Hash=256MB)
AutoTurbo-Suche nach Spezial-Matt [C0/R0/K2/P0/X0] in 12 ... (Hash=256MB)
AutoTurbo-Suche nach Spezial-Matt [C0/R0/K3/P0/X0] in 12 ... (Hash=256MB)
12/12 00:02 575.109 857.092 +M12 1.Bc7
12/12 00:02 1.078.847 933.258 +M12 1.Bc7 Qxc8 2.gxf7+ Kh8 3.Be5 Qc5 4.Bb2 Nc7 5.Ba1 a4 6.Bb2 a3 7.Ba1 a2 8.Bb2 a1Q 9.Bxa1 Nd5+ 10.Ke6+ Nc3 11.Bxc3+ Qe5+ 12.Bxe5+
AutoTurbo-Suche nach Spezial-Matt [C0/R0/K3/P0/X35] in 12 ... (Hash=256MB)
12/12 00:04 428.867 785.470 +M12 1.Bc7
12/12 00:04 932.250 904.219 +M12 1.Bc7 Qxc8 2.gxf7+ Kh8 3.Be5 Qc5 4.Bb2 Nc7 5.Ba1 a4 6.Bb2 a3 7.Ba1 a2 8.Bb2 a1Q 9.Bxa1 Nd5+ 10.Ke6+ Nc3 11.Bxc3+ Qe5+ 12.Bxe5+
Matt in 12 gefunden ! (1 Lösung in 00:01)
12/12 00:04 932.250 904.219 +M12 1.Bc7 Qxc8 2.gxf7+ Kh8 3.Be5 Qc5 4.Bb2 Nc7 5.Ba1 a4 6.Bb2 a3 7.Ba1 a2 8.Bb2 a1Q 9.Bxa1 Nd5+ 10.Ke6+ Nc3 11.Bxc3+ Qe5+ 12.Bxe5+
Best move: Bc7, Value: +Mate in 12, Depth: 12/12, Time: 00:04,058, 932.250 Nodes, 904.219 N/sec

About 1 min in normal mode is still excellent.
Matefish's Proof-Number Search also finds it instantly, whereas the default AB search takes ages.

[ImNotStockfish]

surprising nobody posted it

Maybe because it's so old and has been posted so many times that it's boring at this point
Here is a post from 2008 viewtopic.php?t=22391

[Rebel]

Is Crystal a matefinder, because obviously matefinders will find mates. What's interesting is what prevents (allegedly) the SF-model from finding a M12 with 70 plies of nominal search. Give SF the first move and the solution is almost immediate.

Ah, the notorious null move zugzwang.

Both Cstal an Rebel within less of a second here.

Code: Select all

 25/39	00:00	 1.191k	1.360k	 0,00	g6xh7+ Kg8-h8 Qc8-c3 Qe8-e6+ Kf6-g5+ Kh8xh7 Qc3xa5 Kh7-g8 Kg5-f4 f7-f5 Qa5xa8 Qe6-e4+ Qa8xe4 f5xe4 Kf4xe4
 25/39+	00:00	 1.240k	1.359k	+0,10	Bd8-c7
 25/25+	00:00	 1.286k	1.368k	+5,06	Bd8-c7
 25/28	00:00	 1.351k	1.406k	+M13	Bd8-c7 Qe8xc8 g6xf7+ Kg8-h8 Bc7-e5 Qc8-c5 Be5-b2 Na8-c7 Bb2-a1 a5-a4 Ba1-b2 a4-a3 Bb2-a1 a3-a2 Ba1-b2 a2-a1Q Bb2xa1 Qc5-d6+ Kf6-f5+ Qd6-f6+ Kf5xf6 h7-h6 f7-f8Q+ Kh8-h7 Qf8-g7+
 26/26	00:00	 1.356k	1.407k	+M12	Bd8-c7 Qe8xc8 g6xf7+ Kg8-h8 Bc7-e5 Qc8-c5 Be5-b2 Na8-c7 Bb2-a1 a5-a4 Ba1-b2 a4-a3 Bb2-a1 a3-a2 Ba1-b2 a2-a1Q Bb2xa1 Nc7-d5+ Kf6-e6+ Nd5-c3 Ba1xc3+ Qc5xc3 f7-f8Q+

Is it nullmove? Doubtful.

I suspect it’s because SF allows pruning at the root, disable that and it will probably find it. If so, begs the question, does pruning at the root actually give Elo, while completely eliminating certain moves from being considered?

I have seen it so many times with engines in those special cases, also my own, give it the right move in advance and engines are immediately on track.

You could be right after all, SF in the early plies even prunes in root moves, but look at the forced main line, black is fully paralyzed and can only delay the upcoming mate by moving the a-pawn.

[Robert Flesher]

This one has been on Agadmator's channel for ten days, surprising nobody posted it, I guess some might think it falls into the dreaded forbidden class: chess positions unsolvable by engines, but, then again, it's a very beautiful position AND a forced mate in 12. 70 ply deep doesn't find it though.

https://www.youtube.com/watch?v=ZNnIsmHjcjE

Rebel 16.3 finds it in a few seconds on my machine.

New game Line 0.0
n1QBq1k1/5p1p/5KP1/p7/8/8/8/8 w - - 0 1

Analysis by Rebel-16.3:

1.Bc7 Qxc8 2.gxf7+ Kh8 3.Be5 Qc5 4.Bb2 Nc7 5.Ba1 a4 6.Bb2 a3 7.Ba1 a2 8.Bb2 a1Q 9.Bxa1 Nd5+ 10.Ke6+ Nc3 11.Bxc3+ Qxc3 12.f8Q#
+- (#12) Depth: 102/102 00:00:10 58621kN
White mates
(, 03.08.2024)

[Ajedrecista]

Hello:

The null move issue regarding this study was pointed long time ago, no later than 2003:

By Matous-1975. An incredible position!

This study comes back from time to time and was claimed to be solved by SF back in 2018, although 6-man Syzygy EGTB were used:

Mario Matous study from 1975 solved by Stockfish

(Thread found through a post from other thread).

It sounds familiar to me that Zappa Mexico II was able to solve the problem with null move disabled, in the golden times of Rybkaforum.

------------

On a totally unrelated note, while searching old references of this problem at the old CCC and rec.games.chess.computer newsgroup, I came across old references of 'neural networks', 'machine learning', 'pattern recognition' and so on... even before year 2000. It must be vox populi for sure, but I was surprised that those techniques were considered that early in the field of computer chess.

Regards from Spain.

Ajedrecista.

[Pat51]

This one has been on Agadmator's channel for ten days, surprising nobody posted it, I guess some might think it falls into the dreaded forbidden class: chess positions unsolvable by engines, but, then again, it's a very beautiful position AND a forced mate in 12. 70 ply deep doesn't find it though.

https://www.youtube.com/watch?v=ZNnIsmHjcjE

Impossible? ChestUCI finds #12 in about 1 min (in 'Normal' mode) and in 3 sec (in 'AutoTurbo' mode)!

ChestUCI Ver.5.2:
FEN: n1QBq1k1/5p1p/5KP1/p7/8/8/8/8 w - - (4+6)
Stellungs-Analyse: C0/R0/K3/P6/X22 W:4/25
Suche nach Matt in 12 ... (Hash=256MB)
12 00:53 46.678.802 969.224 +M12 1.Bc7
Suche abgeschlossen ... (Zeit=79.80s)
Matt in 12 gefunden ! (1 Lösung in 01:19)
12/12 01:19 74.607.332 934.870 +M12 1.Bc7 Qxc8 2.gxf7+ Kh8 3.Be5 Qc5 4.Bb2 Nc7 5.Ba1 a4 6.Bb2 a3 7.Ba1 a2 8.Bb2 a1Q 9.Bxa1 Nd5+ 10.Ke6+ Nc3 11.Bxc3+ Qe5+ 12.Bxe5+
Best move: Bc7, Value: +Mate in 12, Depth: 12/12, Time: 01:19,854, 74.607.332 Nodes, 934.870 N/sec

Hi.
When i personnaly enter the problem in arena with the ChestUCI engine , i got only the lines :

ChestUCI:
FEN: n1QBq1k1/5p1p/5KP1/p7/8/8/8/8 w - - (4+6)
Stellungs-Analyse: C0/R0/K3/P6/X22 W:4/25
Parameter in Datenbank gefunden: #12; 00:00 @ C0/R7/K3/P0/X35;
Suche nach Matt in 10 ... (Hash=64MB)
Suche abgeschlossen ... (Zeit=11.90s)
Kein Matt in 10 gefunden ! (00:11)

I don't get in particular the last line

12/12 01:19 74.607.332 934.870 +M12 1.Bc7 Qxc8 2.gxf7+ Kh8 3.Be5 Qc5 4.Bb2 Nc7 5.Ba1 a4 6.Bb2 a3 7.Ba1 a2 8.Bb2 a1Q 9.Bxa1 Nd5+ 10.Ke6+ Nc3 11.Bxc3+ Qe5+ 12.Bxe5+
Best move: Bc7, Value: +Mate in 12, Depth: 12/12, Time: 01:19,854, 74.607.332 Nodes, 934.870 N/sec

Can you tell me why please and how to get all the information ?