Mate in 11 Zugzwang Test Position

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
MikeGL
Posts: 958
Joined: Thu Sep 01, 2011 12:49 pm

Re: Mate in 11 Zugzwang Test Position

Post by MikeGL » Sat Oct 14, 2017 5:03 am

32bit Critter1.6 on standard (default) setting gets M11 in less than 10s
Tried 6 times, clearing hash on each run
1 thread on a 2Ghz archaic 2009 laptop

Code: Select all

Depth  Time
23     00:06
23     00:03
24     00:04
23     00:06
23     00:04
24     00:09
Could be faster for 64bit versions
11.Nh3# mate is a minimalist mating position so prettier in my opinion than 11.Qxg2#. But ok, any type of mate is still mate.

[d]8/8/8/8/5P2/6pN/3KBkpp/7r b - - 18 11

MikeGL
Posts: 958
Joined: Thu Sep 01, 2011 12:49 pm

Re: Mate in 11 Zugzwang Test Position

Post by MikeGL » Sat Oct 14, 2017 5:25 am

or 11.f8(R)# mate

[d]5R2/8/8/8/8/6pB/3KNkpp/7r b - - 0 11

11.Qe1# mate is also possible
Looks like there are 5 types of "mate in 11" position on this puzzle.

peter
Posts: 2132
Joined: Sat Feb 16, 2008 6:38 am
Full name: Peter Martan

Re: Mate in 11 Zugzwang Test Position

Post by peter » Sat Oct 14, 2017 5:44 am

Martin Hertz wrote:[d]8/8/8/1B6/6p1/8/4KPpp/3N2kr w - - 0 1
Stockfish 8 struggles heavily with this zugzwang position.
I've stored that as a study from Johann (Janis) Behting (Carl's brother?), and I think Vincent Lejeune has or had it in one of his collections of test positions once, at least his name appears in my .pgn too, together with the one of Tim Foden, the chess programmer, without any knowledge of mine, why that neither
:oops:

BTW fritz15 needs about 3 minutes single threaded for finding, correct number of moves to mate takes about 5
Peter.

User avatar
Eelco de Groot
Posts: 4265
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Mate in 11 Zugzwang Test Position

Post by Eelco de Groot » Sat Oct 14, 2017 6:18 am

24 seconds for Ancalagon (timestamp 20170705 15:38, versionnumber_014). But no result yet from a less changed Stockfish at depth 84, about 24 minutes

Searching for Johann Behting and the FEN string does indeed return it as a study by him, at least a reference as such: https://www.chessproblem.net/viewtopic.php?t=399 Robert Allgeuer put it in his Zugzwang testsuite here http://www.open-aurec.com/wbforum/viewtopic.php?t=258 with the reference 8/8/8/1B6/6p1/8/4KPpp/3N2kr w - - bm Kd3 Ke3; id "ZPTS.08"; CCC post by Tim Foden, id "MES.831"


8/8/8/1B6/6p1/8/4KPpp/3N2kr w - -

Engine: Ancalagon HT (4 + 3 threads, 256 MB)
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott

39/04 0:01 0.00 1.Ne3 g3 2.fxg3 (32.332.990) 17496

40/04 0:01 0.00 1.Ne3 g3 2.fxg3 (32.607.064) 17483

41/04 0:01 0.00 1.Ne3 g3 2.fxg3 (33.016.363) 17459

42/04 0:03 0.00 1.Ne3 g3 2.fxg3 (55.146.877) 17545

43/04 0:03 0.00 1.Ne3 g3 2.fxg3 (59.862.283) 17539

44/04 0:04 0.00 1.Ne3 g3 2.fxg3 (87.713.822) 17634

45/04 0:08 0.00 1.Ne3 g3 2.fxg3 (154.323.478) 17504

46/04 0:09 0.00 1.Ne3 g3 2.fxg3 (160.114.488) 17523

47/04 0:10 0.00 1.Ne3 g3 2.fxg3 (183.247.504) 17513

48/04 0:10 0.00 1.Ne3 g3 2.fxg3 (184.135.344) 17511

49/04 0:19 0.00 1.Ne3 g3 2.fxg3 (341.325.356) 17306


50/22 0:24 +M11++ 1.Kd3 (417.422.056) 17125

50/22 0:34 +M11 1.Kd3 g3 2.f4 Kf1 3.Kd2+ Kg1 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (588.691.789) 17256


best move: Ke2-d3 time: 0:34.187 min n/s: 17.256.098 nodes: 588.691.789
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

peter
Posts: 2132
Joined: Sat Feb 16, 2008 6:38 am
Full name: Peter Martan

Re: Mate in 11 Zugzwang Test Position

Post by peter » Sat Oct 14, 2017 6:43 am

Hi Michael!
MikeB wrote:SF-McBrain 3.0 to be released tomorrow. With 18 threads, this could have been a lucky run.
2.7 didn't find it in about half an hour and depth 50 on a single thread, so I stopped.
With 24 threads I did so after 10 minutes in depth 49 too.

Do you want to give 3.0 a try on a single core too maybe
:?:
Otherwise I'll do so by myself tomorrow, looking forward to 3.0 impatiently
8-)
Peter.

peter
Posts: 2132
Joined: Sat Feb 16, 2008 6:38 am
Full name: Peter Martan

Re: Mate in 11 Zugzwang Test Position

Post by peter » Sat Oct 14, 2017 7:07 am

Hi Eelco!
Thanks for the infos, latest SugaR struggles even with switched off nullmove, stopped after 5 minutes in depth 53.
This Position is good for testing save and reload of hash too.

Here is Daniel Jose's SF after restart of GUI and reload of 2G of already "knowing" hash:

8/8/8/1B6/6p1/8/4KPpp/3N2kr w - - 0 1

Analysis by Stockfish 190817 64 POPCNT:

1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Ld7 Kf1
+- (#11) Tiefe: 7/8 00:00:00 3kN
...
1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Ld7 Kf1 5.Lh3 Tg1 6.Lg4 Th1 7.Le2+ Kg1 8.Sc3 Kf2 9.Se4+ Kg1 10.Sg5 Kf2 11.Sh3#
+- (#11) Tiefe: 101/22 00:00:05 178MN

And here after "reload" of 4G, that were "knowing" already before too, but were "empty" again obviously after giving "SaveHashtoFile" and "LoadHashfromFile" command (with 4G shown in Win Explorer correctly yet too):

8/8/8/1B6/6p1/8/4KPpp/3N2kr w - - 0 1

Analysis by Stockfish 190817 64 POPCNT:

1.f3 g3 2.f4
= (0.00) Tiefe: 7/4 00:00:00 13kN
...
1.f3 g3 2.f4
= (0.00) Tiefe: 83/4 00:01:30 4582MN

Question remains, if hash bigger than 2G get deleted already by "Save" or later on by "Load".
Anyhow to me it seems like a problem of communication between engine and GUI, don't you think so too
:?:
Peter.

rcmaddox
Posts: 179
Joined: Mon Oct 14, 2013 6:59 pm
Location: Winder, GA
Full name: Robert C. Maddox

Re: Mate in 11 Zugzwang Test Position

Post by rcmaddox » Sat Oct 14, 2017 7:23 am

Martin Hertz wrote: Fritz 3 and Wchess cannot be tested, because they are buggy at this position, showing a wrong mate in 5.
Shredder 13 likewise reports an incorrect mate in 5.

User avatar
Eelco de Groot
Posts: 4265
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Mate in 11 Zugzwang Test Position

Post by Eelco de Groot » Sat Oct 14, 2017 8:03 am

Hi Peter,

Stockfish struggles a lot with it as well but if switching off nullmove does not work there may be some extra factor in search that is not helping, like LMP (Late Move Pruning). Somebody should tell Marco Z. that probably implementing this was maybe a little premature:

https://github.com/official-stockfish/S ... /pull/1183

He should have read other Marco's comments :lol: (This is all I have seen of SugaR, well, Stockfish has bugs too)

I have no idea about the hash saving functions. At least it works with 2 Gigabyte if I understand you correctly?

I do think the OpenChess version with learning from Jeremy Bernstein, I hope I have that correct was very interesting. I saw Zerbinati working on an update, I got some GitHub notification of his branch, but that was a while ago, maybe gave up.


I don't know why Ancalagon can do this, really. Of course the ProbCut verification search", I just invent that name now, is the main thing but how exactly I don't really know. Also I have never measured how much Elo that costs, could be a lot.

A second try; still about 20 seconds but a very different depth. The solution time at least is 'somewhat' reproducible :)

8/8/8/1B6/6p1/8/4KPpp/3N2kr w - -

Engine: Ancalagon HT (256 MB)
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott

37/04 0:00 0.00 1.Ne3 g3 2.fxg3 (10.845.716) 17635

38/26 0:20 +M11 1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (401.176.913) 19168

39/26 0:21 +M11 1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (413.216.686) 19177

40/26 0:21 +M11 1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (415.663.972) 19183

41/26 0:22 +M11 1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (431.353.991) 19202

42/26 0:22 +M11 1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (432.440.045) 19201

43/26 0:22 +M11 1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (434.301.820) 19196

44/26 0:30 +M11 1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (576.681.792) 19143

45/26 0:31 +M11 1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (610.789.730) 19100

46/26 0:32 +M11 1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (612.735.223) 19097

47/26 0:32 +M11 1.Kd3 Kf1 2.Kd2+ Kg1 3.f4 g3 4.Bd7 Kf1
5.Bh3 Kg1 6.Bg4 Kf1 7.Be2+ Kg1 8.Nc3 Kf2
9.Ne4+ Kg1 10.Ng5 Kf2 11.Nh3+ (613.604.505) 19096


best move: Ke2-d3 time: 0:32.203 min n/s: 19.096.368 nodes: 613.604.505
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

peter
Posts: 2132
Joined: Sat Feb 16, 2008 6:38 am
Full name: Peter Martan

Re: Mate in 11 Zugzwang Test Position

Post by peter » Sat Oct 14, 2017 8:23 am

Eelco de Groot wrote: Stockfish struggles a lot with it as well but if switching off nullmove does not work there may be some extra factor in search that is not helping, like LMP (Late Move Pruning). Somebody should tell Marco Z. that probably implementing this was maybe a little premature:

https://github.com/official-stockfish/S ... /pull/1183

He should have read other Marco's comments :lol:

I have no idea about the hash saving functions. At least it works with 2 Gigabyte if I understand you correctly?

I do think the OpenChess version with learning from Jeremy Bernstein, I hope I have that correct was very interesting. I saw Zerbinati working on an update, I got some GitHub notification of his branch, but that was a while ago, maybe gave up.


I don't know why Ancalagon can do this, really. Of course the ProbCut verification search", I just invent that name now, is the main thing but how exactly I don't really know. Also I have never measured how much Elo that costs, could be a lot.
Guess it's a combination of nullmove and LMRs mainly maybe, but "switching off nullmove" by UCI- option means very different things as you well know anyhow. At some branches it's mainly raising of verification only.

MateFinder 041017 needs 1 sec. for #11 by default (study mode not checked) single threaded.

McBrain 2.7 doesn't find so soon with nullmove off neither, but within a few seconds with "brute force", even single threaded.

SugaR get's #11 almost at once with disabled nullmove and activated Corr.Chess Analysis Mode, latter one alone not helping up to depth 64 (2 minutes at 24 threads).

Jeremy Bernstein's SF PA_GTB was updated by Marco Zerbinat for some more version till now up to 1.0 Sf_PA_MZ at beginning of October
Peter.

User avatar
Eelco de Groot
Posts: 4265
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Mate in 11 Zugzwang Test Position

Post by Eelco de Groot » Sat Oct 14, 2017 8:41 am

peter wrote:
Eelco de Groot wrote: Stockfish struggles a lot with it as well but if switching off nullmove does not work there may be some extra factor in search that is not helping, like LMP (Late Move Pruning). Somebody should tell Marco Z. that probably implementing this was maybe a little premature:

https://github.com/official-stockfish/S ... /pull/1183

He should have read other Marco's comments :lol:

I have no idea about the hash saving functions. At least it works with 2 Gigabyte if I understand you correctly?

I do think the OpenChess version with learning from Jeremy Bernstein, I hope I have that correct was very interesting. I saw Zerbinati working on an update, I got some GitHub notification of his branch, but that was a while ago, maybe gave up.


I don't know why Ancalagon can do this, really. Of course the ProbCut verification search", I just invent that name now, is the main thing but how exactly I don't really know. Also I have never measured how much Elo that costs, could be a lot.
Guess it's a combination of nullmove and LMRs mainly maybe, but "switching off nullmove" by UCI- option means very different things as you well know anyhow. At some branches it's mainly raising of verification only.

MateFinder 041017 needs 1 sec. for #11 by default (study mode not checked) single threaded.
Yes it seems Matefinder wins hands downs here!
SugaR get's #11 almost at once with disabled nullmove and activated Corr.Chess Analysis Mode, latter one alone not helping up to depth 64 (2 minutes at 24 threads).
Maybe Analysis Mode still has mate distance pruning? Removing that is really introducing inconsistencies.
Jeremy Bernstein's SF PA_GTB was updated by Marco Zerbinat for some more version till now up to 1.0 Sf_PA_MZ at beginning of October
That would be great I think, if somebody could get it working again! I don't think I could do it.
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

Post Reply