Hiarcs vs 400 years old chess puzzle

This forum is for general discussions and questions, including Collectors Corner and anything to do with Computer chess.

Moderators: Harvey Williamson, Steve B, Watchman

Forum rules
This textbox is used to restore diagrams posted with the fen tag before the upgrade.
herO
Member
Posts: 342
Joined: Wed Nov 12, 2014 8:49 am

Post by herO »

Peter Grayson wrote:
herO wrote:Chess engine are not able to solve these positions. These positions are draw.

8/6pp/1k2p3/p1p1P1P1/P1R2r1P/3P4/1P2P1K1/8 b - - 0 1

4k3/5p2/1p1p1P1p/1P1P1P1P/3P4/8/4K3/8 w - - 0 1

6kr/5b1p/2p3pP/rpPp1pP1/pP1PpP2/P3P3/1K6/8 w - - 0 1

k7/8/6p1/1p1p1pP1/1P1P1P2/1P1K4/3B4/8 w - - 0 1

4k3/2p5/8/1p1pPp1p/1P1P1P1P/8/2P2P2/4K3 b - - 0 1

Here Stockfish and Komodo need to calculate at least to depth 50, then they finally "see" that this position is a draw.
8/p7/kpP5/qrp1b3/rpP2b2/pP4b1/P3K3/8 w - - 0 1

No engine is able to find the right solution here.
3B4/1r2p3/r2p1p2/bkp1P1p1/1p1P1PPp/p1P4P/PPB1K3/8 w - - 0 1

This position is extremely difficult for chess engines! Even the newest version of Komodo cannot see the winning moves for white! If you do the right moves by yourself, Komodo will show you that this position is always draw. It is not true. After few moves, Komodo FINALLY change evaluation from 0.00 to mate in 4 :D Even Stockfish or Houdini cannot solve this position. Look at the position:
4r3/p1p1pPp1/P1P1P1P1/5K2/3p2P1/7p/3P1ppr/3R1nkq w - - 0 1

You can get to this position with perfect moves.
8/p1P5/P5K1/3p4/8/3p1ppp/3P1ppr/3R1nkq w - - 0 1

In this position, human players know that white is winning, but Komodo, Stockfish and Dumbini still "think" that this position is a draw. They always recommend a move c8=Q. I am glad that Hiarcs 14 WCSC is able to find the right solution here c8=N after 1 minute of calculation (depth 53). But tactical engines still recommend nonsence move c8=Q even at depth 130. If you play 1.c8=N d4 2.Nb6 axb6 3.a7 b5 4.a8=N by yourself, then those tactical engines find a "miracle"! :D They change drastically their evaluation from 0.00 to mate in 4 :D
You are being very selective in your choice of positions. In the original position Stockfish fails to solve because it does not consider a Knight under promotion for move 4.a8=N.

Try the following position with Stockfish and Komodo and then try with HIARCS. The reason HIARCS fails is for the same reason as Stockfish and Komodo missed your position. It ignores an underpromotion.

3n4/8/n6p/2pB1K1k/1b6/6P1/2p5/4n3 w - - 0 1

Analysis by Stockfish 8:
1.Bc4 c1N 2.Bb5 Nc6 3.Bxc6 Nc7 4.Ba4 Nc2 5.Bxc2 Ne2 6.Bd1 Na6 7.Bxe2#

Analysis by Komodo 11.2.2 32-bit:
1.Bc4 c1N 2.Bb5 Nc6 3.Bxc6 Nc7 4.Ba4 Nb3 5.Bxb3 Nd5 6.Bd1+ Nf3 7.Bxf3#

Analysis by Deep HIARCS 14 WCSC:

1.Bc4 Nd3 2.Bxd3 c1N 3.Be4 Nb8 4.Bf3#

The Earliest HIARCS engine that I have has the same problem

Analysis by Hiarcs8 Bareev:

1.Bc4 c1Q 2.Bxa6
-+ (-19.50) Depth: 2/6 00:00:00
1.Bc4 Nf3 2.Be2
-+ (-14.98) Depth: 2/10 00:00:00
1.Bc4 c1R
-+ (-14.05) Depth: 3/10 00:00:00
1.Bc4 Nd3 2.Bxd3 c1N 3.Bxa6
-+ (-8.18) Depth: 3/10 00:00:00
1.Bc4 Nd3 2.Bxd3 c1N 3.Bxa6
-+ (-8.18) Depth: 4/10 00:00:00
1.Bc4 Nd3 2.Bxd3 c1N 3.Bxa6
-+ (-8.18) Depth: 4/10 00:00:00
1.Bc4 Nd3 2.Bxd3 c1N 3.Bxa6
-+ (-8.18) Depth: 6/10 00:00:00 1kN
1.Bc4 Nd3 2.Bxd3 c1N 3.Bxa6 Nc6
-+ (-8.27) Depth: 6/11 00:00:00 2kN
1.Bc4 Nd3 2.Bxd3 c1N 3.Be4 Nd3 4.Bf3#
+- (#4) Depth: 7/13 00:00:00 3kN
1.Bc4 Nd3 2.Bxd3 c1N 3.Be4 Nd3 4.Bf3#
+- (#4) Depth: 7/13 00:00:00 3kN
1.Bc4 Nd3 2.Bxd3 c1N 3.Be4
+- (#4) Depth: 7/13 00:00:00 4kN

So for HIARCS this is a hereditary issue going back to at least HIARCS 8.

I would also comment that with PERFECT PLAY your last position
8/p1P5/P5K1/3p4/8/3p1ppp/3P1ppr/3R1nkq w - - 0 1
could never be achieved! The position may be legal that is probably what you meant however, I would like you to show the sequence of moves to get there, particularly the sequence with "perfect play".

In blockade positions, most engines tend to score based on material difference. Shredder is one engine that is capable of scoring some blockade positions correctly but perhaps the more valid question is "would engines play the positions correctly despite their scoring?" That is a far better measure and more telling than just the score of a fixed position.

PeterG
Hello PeterG,
I would like to show you the whole sequence of perfect moves to get into this position and even to the checkmate. If you will analyse these moves with Stockfish 8 or Komodo, then you will see that Stockfish and Komodo absolutely have no understanding of this position. Their evalution show that this position is still a draw.

Starting position:
[fen]4r3/p1p1pPp1/P1P1P1P1/5K2/3p2P1/7p/3P1ppr/3R1nkq w - - 0 1[/fen]

Solution:
1.fxe8=N d3 2.Nf6 gxf6 3.g5 fxg5 4.g7 g4 5.g8=N g3 6.Nf6 exf6 7.Kg6 f5 8.e7 f4 9.e8=N f3 10.Nd6 cxd6 11.c7 d5 12.c8=N d4 13.Nb6 axb6 14.a7 b5 15.a8=N b4 16.Nc7 b3 17.Ne6 b2 18.Ng5 b1=Q 19.Nxf3# *
User avatar
Peter Grayson
Member
Posts: 207
Joined: Sat Aug 04, 2007 3:23 pm
Location: South Wales, UK

Post by Peter Grayson »

herO wrote:
Hello PeterG,
I would like to show you the whole sequence of perfect moves to get into this position and even to the checkmate. If you will analyse these moves with Stockfish 8 or Komodo, then you will see that Stockfish and Komodo absolutely have no understanding of this position. Their evalution show that this position is still a draw.

Starting position:
[fen]4r3/p1p1pPp1/P1P1P1P1/5K2/3p2P1/7p/3P1ppr/3R1nkq w - - 0 1[/fen]

Solution:
1.fxe8=N d3 2.Nf6 gxf6 3.g5 fxg5 4.g7 g4 5.g8=N g3 6.Nf6 exf6 7.Kg6 f5 8.e7 f4 9.e8=N f3 10.Nd6 cxd6 11.c7 d5 12.c8=N d4 13.Nb6 axb6 14.a7 b5 15.a8=N b4 16.Nc7 b3 17.Ne6 b2 18.Ng5 b1=Q 19.Nxf3# *
Thank you. I understand the solution but cannot understand how the given start position would be achieved with perfect play. The position falls into the category of "Contrived", that is the position was not taken from an actual game. It may be of academic interest but has no value for analysis of game positions. Therefore there is little merit in an engine that can solve the position. More important for engines and of more interest to solve are positions that can arise in actual play. If however you can show the position was taken from an actual game or are able to demonstrate how the position could be achieved then I will accede to your comments.

In practice, there is no evidence to support the argument an engine and thus the end user benefits from incorporating code to solve these types of positions. There are however specialist engines available focused more towards analysis than game play. The general advice for analysis though is not to rely on one engine for the solution when the use of two or three with sufficiently different characteristics may give different insights into the finer considerations of positions. Stockfish or Komodo, Houdini in Tactical Mode with perhaps HIARCS or Shredder would give a good blend of engines with different styles.

PeterG
Post Reply