Deep Blue position (1993)

[JBNielsen]

White must play Qxb5 before black closes the position with Nd6.

This is easy for humans, but can any program find this?

[D]4n3/4k3/p4n2/Pp1p1p1p/QPpPpPpP/2P1P1P1/1R3BK1/7R w - - 1 1

I gave some positions to the Deep Blue Team just before Deep Blue should meet Bent Larsen and other danish top players in 1993.
This was one of the positions.

[Werewolf]

Presumably DB failed to find the solution?

[JBNielsen]

Presumably DB failed to find the solution?

I did not really get an answer. Not a clear one, at least. So presumably DB failed.

But perhaps some programs could detect, that a move produces long lines without any captures, pawn advancing and no improvement of the score. These moves could be discarded in favour of other moves that had captures or pawn advances - although the score is less (but still a winning score).

[Darrel Briley]

White must play Qxb5 before black closes the position with Nd6.

This is easy for humans, but can any program find this?

[D]4n3/4k3/p4n2/Pp1p1p1p/QPpPpPpP/2P1P1P1/1R3BK1/7R w - - 1 1

I gave some positions to the Deep Blue Team just before Deep Blue should meet Bent Larsen and other danish top players in 1993.
This was one of the positions.

I tried several programs (Houdini, Critter, Stockfish, Komodo etc.), and none found Qxb5. So, I fed it to Deep Shredder 12, and it was found immediately.




4n3/4k3/p4n2/Pp1p1p1p/QPpPpPpP/2P1P1P1/1R3BK1/7R w - -

Engine: Deep Shredder 12 x64 (4096 MB)
by Stefan Meyer-Kahlen

22 0:00 +9.43 1.Qxb5 axb5 2.Ra2 Nc7 3.a6 Nd7 4.a7 Nb6
5.Ra5 Kd7 6.Rc1 Ke6 7.Bg1 Kd6 8.Kf2 Ke7
9.Rb1 Kf7 10.Ke2 Nba8 11.Bf2 Nb6
12.Kd2 Ke7 13.Rc1 Kd6 14.Rca1

best move: Qa4xb5

DB

*Edit* I should add I gave the other engines very little time before I gave up on them, as I suspected that Shredder would find it right away due to knowledge in the engine

[Uri Blass]

Presumably DB failed to find the solution?

I did not really get an answer. Not a clear one, at least. So presumably DB failed.

But perhaps some programs could detect, that a move produces long lines without any captures, pawn advancing and no improvement of the score. These moves could be discarded in favour of other moves that had captures or pawn advances - although the score is less (but still a winning score).

I expected to see some position from deep blue's games but
this position has nothing to do with deep blue.

I hope that the moderators change the title because probably deep blue never analyzed this position and the year 1993 is clearly not relevant
for the subject.

Maybe some human composed this position in 1993 but it is clearly not relevant because it is not from a chess game from 1993.

[zullil]

Critter-1.6 finds it, but not quickly:


40/74 18:12 13853M 12675K +16.53 Qc2 Ng8 Rbb1 Kf6 Qe2 Ne7 Rhf1 Nc6 Bg1 Kg6 Ra1 Ne7 Rfe1 Kf7 Rec1 Kg6 Qc2 Nc6 Rf1 Kf6 Rfe1 Kg6 Red1 Nd8 Rf1 Nc6 Rac1 Kf6 Qe2 Kf7 Ra1 Kg6 Ra2 Nd8 Rc1 Nc6 Re1 Kf6 Raa1 Kg6 Rec1 Ne7
40/76+ 38:52 30142M 12925K +17.56 Qxb5 axb5 a6 Nd7 a7 Nc7 Ra2 Nb6 Rha1 Nba8 Ra6 Nxa6 Rxa6 Kd7 Rh6 Kc7 Rxh5 Nb6 Rh8 Kb7 Rb8+ Kxa7 Rxb6 Kxb6 h5 Kc7 h6 Kb8 h7 Kc8 h8=Q+ Kb7 Qh7+ Kc6 Qe7 Kb6 Qc5+ Ka6 Qxd5 Kb6 Qxf5 Kc7 Qe5+ Kd8 Qxe4 Kc7 f5

[Houdini]

Here's a Houdini run with FiftyMoveDistance parameter set to 15, it takes 13 seconds to prefer Qxb5!

Code: Select all

4n3/4k3/p4n2/Pp1p1p1p/QPpPpPpP/2P1P1P1/1R3BK1/7R w - -

Engine: Houdini 3 (2048 MB)
by Robert Houdart

16/21  0:00   +17.12   1.Qd1 Kf7 2.Qc2 Kg6 3.Rf1 Nd6 4.Rfb1 Nfe8 
                       5.Rh1 Kg7 6.Qe2 Kf7 7.Qe1 Kg6 8.Qf1 Nf7 (1.089.201) 4615 

17/30  0:01   +16.40-- 1.Qd1 Kf7 (16.059.372) 8434 

17/31  0:03   +14.98-- 1.Qd1 Kf7 (35.327.524) 8874 

17/48  0:07   +13.15   1.Qd1 Kf7 2.Qc2 Kg6 3.Rf1 Nd6 4.Rbb1 Ng8 
                       5.Rh1 Ne7 6.Rh3 gxh3+ 7.Kxh3 Ne8 
                       8.Be1 Nf6 9.Kg2 Ng4 10.Bf2 Ng8 
                       11.Bg1 Ne7 12.Ra1 Kf6 13.Rd1 Kg6 (67.952.430) 8999 

18/48  0:10   +13.12   1.Qd1 Kf7 2.Qc2 Kg6 3.Bg1 Nd7 4.Rb1 Nf8 
                       5.Rh3 gxh3+ 6.Kxh3 Nf6 7.Qa2 Ng4 
                       8.Qd2 Ne6 9.Kg2 Nd8 10.Re1 Kf6 
                       11.Qe2 Ke6 12.Rd1 Kd6 13.Qd2 Nc6 
                       14.Re1 (97.876.833) 9111 

18/48  0:13   +13.70++ 1.Qxb5 (118.372.982) 8937 

18/48  0:13   +13.68   1.Qxb5 axb5 2.a6 Nd7 3.a7 Nc7 4.Ra2 Nb6 
                       5.Rha1 Nca8 6.Ra5 Nc7 7.a8Q Nbxa8 
                       8.Rxa8 Nxa8 9.Rxa8 Kf6 10.Rb8 Kf7 
                       11.Rb6 Ke7 12.Rxb5 Ke6 13.Rb6+ Kf7 
                       14.Ra6 (118.545.325) 8929 

[melajara]

Nice, but how did this FiftyMoveDistance parameter be set to 15?

If it's from you, Deus ex machina, it's like cheating

Now, if this parameter is only accessed internally by the program and dynamically calibrated according to the type of position on the board
and used to tune the search (kind of meta parameter for the eval function), then congratulations!

[Houdini]

Nice, but how did this FiftyMoveDistance parameter be set to 15?

If it's from you, Deus ex machina, it's like cheating

Now, if this parameter is only accessed internally by the program and dynamically calibrated according to the type of position on the board
and used to tune the search (kind of meta parameter for the eval function), then congratulations!

A FiftyMoveDistance parameter of around 15 is consistently useful in positions with fortress or blockade potential.
I could easily program Houdini to activate this parameter automatically when it has a material advantage of over 8 pawns in positions with all the pawns blocked.
But would that really be useful? The position that is discussed here would never occur in a real game!

[JBNielsen]

Presumably DB failed to find the solution?

I did not really get an answer. Not a clear one, at least. So presumably DB failed.

But perhaps some programs could detect, that a move produces long lines without any captures, pawn advancing and no improvement of the score. These moves could be discarded in favour of other moves that had captures or pawn advances - although the score is less (but still a winning score).

I expected to see some position from deep blue's games but
this position has nothing to do with deep blue.

I hope that the moderators change the title because probably deep blue never analyzed this position and the year 1993 is clearly not relevant
for the subject.

Maybe some human composed this position in 1993 but it is clearly not relevant because it is not from a chess game from 1993.

[moderation]
Hmm, I don't think that was such a big deal, but ok: let's be exact.

I hope that's ok with you, Jens Bæk ?
[/moderation]

This is not a big deal, but you are doing things wrong now.

I wrote this in my first posting: "I gave some positions to the Deep Blue Team just before Deep Blue should meet Bent Larsen and other danish top players in 1993.
This was one of the positions."

My job was to prepare the danish players for the match, and I had collected 50 pages they could read (You can read that in Hsu's book about DB). The given position (which I constructed) and some other positions were given to the DB team. The aim was to get some last-minute insight into DBs capabilities.
I am quite sure Murray tried the positions, but when I asked him about DB solved them he just shaked his head or raised his shoulders (I don't remember it exactly as it is almost 20 years ago). I did not press them very much; they were busy preparing for the event.

I was together with the DB team for a week, and after the event teammember Joe Hoane visited me for a weekend.
More info in my 4 videoes http://www.youtube.com/user/JBN777?feature=mhum. Especially the video with a rare game contains more information.

The new title is wrong - the positions were not submitted to DB; they were handed to them. And it was not to Deep Thought; it was to Deep Blue as this name was used for the first time in Copenhagen 1993 (Deep Thought ran my Nielsen-ratingtest in 1988 via Larry Kaufman).

I hope you now understand that the position, DB and the year 1993 is relevant here.

Best, jens baek nielsen