Which is the weaker engine that can find the draw in game #2

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

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
pichy
Posts: 2564
Joined: Thu Mar 09, 2006 2:04 am

Which is the weaker engine that can find the draw in game #2

Post by pichy » Sat Jun 30, 2012 10:08 am

It was discovered within 24 hours of the end of game 2 between GM Kasparov vs Deep Blue that Kasparov, who had resigned on move 45 had done so prematurely. Positionally, the game was actually a draw, which was possibly first noticed by players in Germany, who then informed the Kasparov team in New York.

Apparently, Kasparov had considered playing 45. …Qe3 in hopes of a draw by perpetual check after 46. Qxd6, but he resigned since he correctly calculated that the checks which follow were finite. However, the move that was discovered in the post-mortem was 46… Re8, which does lead to a perpetual check after several more moves.

Kasparov’s resignation was, in short, a blunder. That Kasparov could blunder away the draw is at least understandable on human terms, but a greater question remains unanswered. How could Deep Blue, with all its prodigious calculating abilities, deliberately play into a drawn position? The only answer is that Deep Blue had also blundered, but was lucky enough to get away with it! That the machine had blundered is indisputable, since even the PC software of the time could detect the draw. One of the machine’s programmers, Murray Campbell, even referred to it as a blunder. One possible explanation is that the well-known horizon effect was responsible. This essentially refers to a situation where the machine stopped calculating right before the crucial bit of information was found.

User avatar
marcelk
Posts: 348
Joined: Fri Feb 26, 2010 11:21 pm
Contact:

Re: Which is the weaker engine that can find the draw in gam

Post by marcelk » Sat Jun 30, 2012 12:28 pm

45. ... Qe3 and 46. ... Re8 should be easy finds nowadays even if the draw is still far away. Recognizing upcoming repetition in leaf nodes could be of good help in this kind of positions with exposed king vs. queen.

I think the real question is what program still plays 44. Kf1? Modern evaluation functions should shun away from it.

[D] R7/1r3kp1/1qQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/6K1 w - - 11 44

PS: The draws that PC programs of the time gave were not based on full exploration of the critical tree but more likely based on misevaluating the position itself. The end position was also too deep for them.

pichy
Posts: 2564
Joined: Thu Mar 09, 2006 2:04 am

Re: Which is the weaker engine that can find the draw in gam

Post by pichy » Sat Jun 30, 2012 12:50 pm

marcelk wrote:45. ... Qe3 and 46. ... Re8 should be easy finds nowadays even if the draw is still far away. Recognizing upcoming repetition in leaf nodes could be of good help in this kind of positions with exposed king vs. queen.

I think the real question is what program still plays 44. Kf1? Modern evaluation functions should shun away from it.

[D] R7/1r3kp1/1qQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/6K1 w - - 11 44

PS: The draws that PC programs of the time gave were not based on full exploration of the critical tree but more likely based on misevaluating the position itself. The end position was also too deep for them.
Looking at the 44.Kf1? move played by Deep Blue, I believe that GM Kasparov was too scare of playing versus DB and eventually lost the match, but any program rated 2750 today could and can beat DP in a match.

User avatar
JuLieN
Posts: 2948
Joined: Mon May 05, 2008 10:16 am
Location: Nantes (France)
Contact:

Re: Which is the weaker engine that can find the draw in gam

Post by JuLieN » Sat Jun 30, 2012 1:33 pm

marcelk wrote:45. ... Qe3 and 46. ... Re8 should be easy finds nowadays even if the draw is still far away. Recognizing upcoming repetition in leaf nodes could be of good help in this kind of positions with exposed king vs. queen.

I think the real question is what program still plays 44. Kf1? Modern evaluation functions should shun away from it.

[D] R7/1r3kp1/1qQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/6K1 w - - 11 44

PS: The draws that PC programs of the time gave were not based on full exploration of the critical tree but more likely based on misevaluating the position itself. The end position was also too deep for them.
Prédateur first plays Kh2 during the first 13 plies, and eventually choses Kf1 starting with the 14th ply (after 33 seconds on my Core i5).
"The only good bug is a dead bug." (Don Dailey)
Image [Blog: http://tinyurl.com/predateur ] [Facebook: http://tinyurl.com/fbpredateur ] [MacEngines: http://tinyurl.com/macengines ]

rbarreira
Posts: 900
Joined: Tue Apr 27, 2010 1:48 pm

Re: Which is the weaker engine that can find the draw in gam

Post by rbarreira » Sat Jun 30, 2012 2:31 pm

Here is what Feng-Hsiung Hsu (one of Deep Blue's authors) says about this game in his book, "Behind Deep Blue":

Image

Will Singleton
Posts: 128
Joined: Thu Mar 09, 2006 4:14 pm
Location: Los Angeles, CA

Re: Which is the weaker engine that can find the draw in gam

Post by Will Singleton » Sun Jul 01, 2012 10:43 pm

<from a wiki article>

However, it was discovered in 2007 by Internet analysis[7] that after 45...Qe3 46.Qxd6 Re8 47.Qd7+ Re7 48.Qc6 Qxe4 49.d6 Qd3+ 50.Kg1 Re8 51.Ra1 there is no perpetual check and the attack continues. The critical move being for White to force Black's rook on e7, which is later attacked by 49.d6, but also for White to retract his queen to c6, which can later move to g2, if necessary, to block the perpetual check. Various other lines have been analyzed, such as 48...Qc1+ or 48...Qf4+, but White (Deep Blue) invariably ends up with a superior position. Given this analysis, all of Deep Blue's moves make sense; it is not apparent that the computer made any mistake.

User avatar
marcelk
Posts: 348
Joined: Fri Feb 26, 2010 11:21 pm
Contact:

Re: Which is the weaker engine that can find the draw in gam

Post by marcelk » Sun Jul 01, 2012 10:53 pm

Will Singleton wrote:<from a wiki article>

However, it was discovered in 2007 by Internet analysis[7] that after 45...Qe3 46.Qxd6 Re8 47.Qd7+ Re7 48.Qc6 Qxe4 49.d6 Qd3+ 50.Kg1 Re8 51.Ra1 there is no perpetual check and the attack continues. The critical move being for White to force Black's rook on e7, which is later attacked by 49.d6, but also for White to retract his queen to c6, which can later move to g2, if necessary, to block the perpetual check. Various other lines have been analyzed, such as 48...Qc1+ or 48...Qf4+, but White (Deep Blue) invariably ends up with a superior position. Given this analysis, all of Deep Blue's moves make sense; it is not apparent that the computer made any mistake.
Then if you go to the [7]-reference, a youtube video by kingscrusher, you can see the comments are ahead of Wikipedia:
kingscrusher in reply to jdthom6 3 years ago wrote:Unless anyone has a flaw in the line of "2TheDeath" below, it seems&#65279; black has enough to draw after all.

pichy
Posts: 2564
Joined: Thu Mar 09, 2006 2:04 am

Re: What is Houdini and Critter analysis ?.................

Post by pichy » Mon Jul 02, 2012 3:56 pm

marcelk wrote:
Will Singleton wrote:<from a wiki article>

However, it was discovered in 2007 by Internet analysis[7] that after 45...Qe3 46.Qxd6 Re8 47.Qd7+ Re7 48.Qc6 Qxe4 49.d6 Qd3+ 50.Kg1 Re8 51.Ra1 there is no perpetual check and the attack continues. The critical move being for White to force Black's rook on e7, which is later attacked by 49.d6, but also for White to retract his queen to c6, which can later move to g2, if necessary, to block the perpetual check. Various other lines have been analyzed, such as 48...Qc1+ or 48...Qf4+, but White (Deep Blue) invariably ends up with a superior position. Given this analysis, all of Deep Blue's moves make sense; it is not apparent that the computer made any mistake.
Then if you go to the [7]-reference, a youtube video by kingscrusher, you can see the comments are ahead of Wikipedia:
kingscrusher in reply to jdthom6 3 years ago wrote:Unless anyone has a flaw in the line of "2TheDeath" below, it seems&#65279; black has enough to draw after all.
What is Houdini and Critter analysis on this position with a very fast system?
[D] R7/1r3kp1/1qQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/6K1 w - - 11 44

jmartus
Posts: 240
Joined: Sun May 16, 2010 12:50 am

Re: What is Houdini and Critter analysis ?.................

Post by jmartus » Mon Jul 02, 2012 4:57 pm

not a super fast computer but decent amd X4 635 6 GB Ram


FEN: R7/1r3kp1/1qQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/6K1 w - - 0 1

Houdini_15_x64:
27/82 04:21 1,805,516,339 6,913,000 +1.63 Kg1-f1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 Qd7-f7+ Kg8-h8 Kf1-e2 Rb8-d8 Qf7-b7 Rd8-b8 Qb7xb6 Rb8xb6 Ra7-a8 Kh8-g8 Be4-f3 g7-g6 f5xg6 f6-f5 g2-g4 f5xg4 Bf3xg4 Rb6xg6 Ke2-f3 Rg6-b6 Ra8-a7 Bf8-d6 Kf3-e4 Bd6-b8 Ra7-d7 Kg8-f8 Bg4-e6 Bb8-d6 Rd7-a7 Kf8-e8 Ra7-h7 Bd6-f8 h3-h4 Rb6-d6 Ke4xe5
27/82 04:50 2,005,677,311 6,905,000 +2.29 Kg1-h1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 d5-d6 Kg8-h7 Be4-d5 Qb6-f2 Ra7-a1 Qf2-b2 Ra1-f1 Qb2xc3 Qd7-e6 Bf8xd6 Qe6xd6 Rb8-c8 Qd6-d7 Rc8-b8 Qd7-c7 Rb8-e8 Bd5-c6 Qc3-d3 Rf1-f3 Qd3-d8 Qc7xd8 Re8xd8 Bc6xb5 Rd8-b8 Bb5xc4 Rb8xb4 Bc4-f7 Rb4-b7 Bf7-g6+ Kh7-g8 Rf3-c3 Kg8-f8 Kh1-g1 Kf8-e7 Kg1-f2 Rb7-b2+ Kf2-f3
28/82+ 05:12 2,161,432,995 6,920,000 +2.41 Kg1-h1
28/82 05:49 2,428,836,767 6,945,000 +2.46 Kg1-h1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 d5-d6 Kg8-h7 Be4-d5 Qb6-f2 Ra7-a1 Qf2-b2 Ra1-f1 Qb2xc3 Qd7-e6 Bf8xd6 Qe6xd6 Rb8-c8 Qd6-d7 Rc8-b8 Qd7-c7 Rb8-e8 Bd5-c6 Qc3-d3 Rf1-f3 Qd3-d8 Qc7xd8 Re8xd8 Bc6xb5 Rd8-b8 Bb5xc4 Rb8xb4 Bc4-f7 Rb4-b7 Bf7-h5 Kh7-g8 Rf3-c3 Kg8-f8 Bh5-f3 Rb7-b4 Kh1-g1 Kf8-e7 Rc3-c7+ Ke7-d6 Rc7xg7 e5-e4 Bf3-e2 Kd6-e5 g2-g4
29/82 07:25 3,120,114,258 7,010,000 +2.54 Kg1-h1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 d5-d6 Kg8-h7 Be4-d5 Qb6-e3 Qd7-f7 Qe3-c1+ Kh1-h2 Qc1-f4+ g2-g3 Qf4-f2+ Bd5-g2 e5-e4 Qf7-c7 Qf2-b6 d6-d7 Qb6xc7 Ra7xc7 Rb8-d8 Bg2xe4 Bf8-d6 Rc7-b7 Kh7-g8 Be4-c6 Bd6-e5 Bc6xb5 Be5xc3 Rb7-b6 Kg8-h7 Rb6-e6 Bc3xb4 Re6-e8 Bb4-a5 Bb5xc4 h6-h5 Re8-e7 Ba5-b4 Re7-f7 Bb4-a5 Kh2-g2 Kh7-h6 Kg2-f3 Ba5-c7 g3-g4
30/88+ 09:38 4,101,932,682 7,096,000 +2.67 Kg1-h1
30/88 10:06 4,313,383,353 7,111,000 +2.67 Kg1-h1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 d5-d6 Kg8-h7 Be4-d5 Qb6-e3 Qd7-f7 Qe3-c1+ Kh1-h2 Qc1-f4+ g2-g3 Qf4-f2+ Bd5-g2 e5-e4 Qf7-c7 Qf2-b6 d6-d7 Qb6xc7 Ra7xc7 Rb8-d8 Bg2xe4 Bf8-d6 Rc7-b7 Bd6-e5 Be4-c6 Be5xc3 Bc6xb5 Bc3xb4 Bb5xc4 Bb4-a5 Bc4-f7 h6-h5 Kh2-g2 Kh7-h6 Kg2-f3 Rd8-a8 h3-h4 Ba5-d8 Kf3-f2 Bd8-a5 Kf2-e3 Ba5-d8 Bf7-g6 Bd8-e7 Ke3-e4 Be7-d6 Rb7-a7
30/88 10:14 4,372,214,652 7,118,000 +2.67 Kg1-h1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 d5-d6 Kg8-h7 Be4-d5 Qb6-e3 Qd7-f7 Qe3-c1+ Kh1-h2 Qc1-f4+ g2-g3 Qf4-f2+ Bd5-g2 e5-e4 Qf7-c7 Qf2-b6 d6-d7 Qb6xc7 Ra7xc7 Rb8-d8 Bg2xe4 Bf8-d6 Rc7-b7 Bd6-e5 Be4-c6 Be5xc3 Bc6xb5 Bc3xb4 Bb5xc4 Bb4-a5 Bc4-f7 h6-h5 Kh2-g2 Kh7-h6 Kg2-f3 Rd8-a8 h3-h4 Ba5-d8 Kf3-f2 Bd8-a5 Kf2-e3 Ba5-d8 Bf7-g6 Bd8-e7 Ke3-e4 Be7-d6 Rb7-a7

pichy
Posts: 2564
Joined: Thu Mar 09, 2006 2:04 am

Re: What is Houdini and Critter analysis ?.................

Post by pichy » Mon Jul 02, 2012 5:25 pm

jmartus wrote:not a super fast computer but decent amd X4 635 6 GB Ram


FEN: R7/1r3kp1/1qQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/6K1 w - - 0 1

Houdini_15_x64:
27/82 04:21 1,805,516,339 6,913,000 +1.63 Kg1-f1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 Qd7-f7+ Kg8-h8 Kf1-e2 Rb8-d8 Qf7-b7 Rd8-b8 Qb7xb6 Rb8xb6 Ra7-a8 Kh8-g8 Be4-f3 g7-g6 f5xg6 f6-f5 g2-g4 f5xg4 Bf3xg4 Rb6xg6 Ke2-f3 Rg6-b6 Ra8-a7 Bf8-d6 Kf3-e4 Bd6-b8 Ra7-d7 Kg8-f8 Bg4-e6 Bb8-d6 Rd7-a7 Kf8-e8 Ra7-h7 Bd6-f8 h3-h4 Rb6-d6 Ke4xe5
27/82 04:50 2,005,677,311 6,905,000 +2.29 Kg1-h1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 d5-d6 Kg8-h7 Be4-d5 Qb6-f2 Ra7-a1 Qf2-b2 Ra1-f1 Qb2xc3 Qd7-e6 Bf8xd6 Qe6xd6 Rb8-c8 Qd6-d7 Rc8-b8 Qd7-c7 Rb8-e8 Bd5-c6 Qc3-d3 Rf1-f3 Qd3-d8 Qc7xd8 Re8xd8 Bc6xb5 Rd8-b8 Bb5xc4 Rb8xb4 Bc4-f7 Rb4-b7 Bf7-g6+ Kh7-g8 Rf3-c3 Kg8-f8 Kh1-g1 Kf8-e7 Kg1-f2 Rb7-b2+ Kf2-f3
28/82+ 05:12 2,161,432,995 6,920,000 +2.41 Kg1-h1
28/82 05:49 2,428,836,767 6,945,000 +2.46 Kg1-h1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 d5-d6 Kg8-h7 Be4-d5 Qb6-f2 Ra7-a1 Qf2-b2 Ra1-f1 Qb2xc3 Qd7-e6 Bf8xd6 Qe6xd6 Rb8-c8 Qd6-d7 Rc8-b8 Qd7-c7 Rb8-e8 Bd5-c6 Qc3-d3 Rf1-f3 Qd3-d8 Qc7xd8 Re8xd8 Bc6xb5 Rd8-b8 Bb5xc4 Rb8xb4 Bc4-f7 Rb4-b7 Bf7-h5 Kh7-g8 Rf3-c3 Kg8-f8 Bh5-f3 Rb7-b4 Kh1-g1 Kf8-e7 Rc3-c7+ Ke7-d6 Rc7xg7 e5-e4 Bf3-e2 Kd6-e5 g2-g4
29/82 07:25 3,120,114,258 7,010,000 +2.54 Kg1-h1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 d5-d6 Kg8-h7 Be4-d5 Qb6-e3 Qd7-f7 Qe3-c1+ Kh1-h2 Qc1-f4+ g2-g3 Qf4-f2+ Bd5-g2 e5-e4 Qf7-c7 Qf2-b6 d6-d7 Qb6xc7 Ra7xc7 Rb8-d8 Bg2xe4 Bf8-d6 Rc7-b7 Kh7-g8 Be4-c6 Bd6-e5 Bc6xb5 Be5xc3 Rb7-b6 Kg8-h7 Rb6-e6 Bc3xb4 Re6-e8 Bb4-a5 Bb5xc4 h6-h5 Re8-e7 Ba5-b4 Re7-f7 Bb4-a5 Kh2-g2 Kh7-h6 Kg2-f3 Ba5-c7 g3-g4
30/88+ 09:38 4,101,932,682 7,096,000 +2.67 Kg1-h1
30/88 10:06 4,313,383,353 7,111,000 +2.67 Kg1-h1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 d5-d6 Kg8-h7 Be4-d5 Qb6-e3 Qd7-f7 Qe3-c1+ Kh1-h2 Qc1-f4+ g2-g3 Qf4-f2+ Bd5-g2 e5-e4 Qf7-c7 Qf2-b6 d6-d7 Qb6xc7 Ra7xc7 Rb8-d8 Bg2xe4 Bf8-d6 Rc7-b7 Bd6-e5 Be4-c6 Be5xc3 Bc6xb5 Bc3xb4 Bb5xc4 Bb4-a5 Bc4-f7 h6-h5 Kh2-g2 Kh7-h6 Kg2-f3 Rd8-a8 h3-h4 Ba5-d8 Kf3-f2 Bd8-a5 Kf2-e3 Ba5-d8 Bf7-g6 Bd8-e7 Ke3-e4 Be7-d6 Rb7-a7
30/88 10:14 4,372,214,652 7,118,000 +2.67 Kg1-h1 Rb7-b8 Qc6-d7+ Kf7-g8 Ra8-a7 Bd6-f8 d5-d6 Kg8-h7 Be4-d5 Qb6-e3 Qd7-f7 Qe3-c1+ Kh1-h2 Qc1-f4+ g2-g3 Qf4-f2+ Bd5-g2 e5-e4 Qf7-c7 Qf2-b6 d6-d7 Qb6xc7 Ra7xc7 Rb8-d8 Bg2xe4 Bf8-d6 Rc7-b7 Bd6-e5 Be4-c6 Be5xc3 Bc6xb5 Bc3xb4 Bb5xc4 Bb4-a5 Bc4-f7 h6-h5 Kh2-g2 Kh7-h6 Kg2-f3 Rd8-a8 h3-h4 Ba5-d8 Kf3-f2 Bd8-a5 Kf2-e3 Ba5-d8 Bf7-g6 Bd8-e7 Ke3-e4 Be7-d6 Rb7-a7
I wonder if you let it analyze up to 80/80 close to 80 ply since Deep Blue was able to calculate 2,000,000 positions per second, if it will go back and select f1 as the best move, and what about critter 1.6a what does it say?

Post Reply