Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How many programs can see the draw in the 2nd game of DB vs Kasparov?

Author: Uri Blass

Date: 15:17:37 02/28/01

Go up one level in this thread


On February 28, 2001 at 17:04:19, Robert Hyatt wrote:

>On February 28, 2001 at 14:13:56, Uri Blass wrote:
>
>>On February 28, 2001 at 13:41:31, Robert Hyatt wrote:
>>
>>>On February 28, 2001 at 12:19:42, Uri Blass wrote:
>>>
>>>>On February 28, 2001 at 11:46:37, Enrique Irazoqui wrote:
>>>>
>>>>>On February 28, 2001 at 10:54:13, Uri Blass wrote:
>>>>>
>>>>>>On February 28, 2001 at 09:57:06, Enrique Irazoqui wrote:
>>>>>>
>>>>>>>On February 28, 2001 at 07:21:44, Jeremiah Penery wrote:
>>>>>>>
>>>>>>>>On February 27, 2001 at 13:35:33, Enrique Irazoqui wrote:
>>>>>>>>
>>>>>>>>>On February 27, 2001 at 10:49:29, Uri Blass wrote:
>>>>>>>>>
>>>>>>>>>>On February 27, 2001 at 10:07:31, Robert Hyatt wrote:
>>>>>>>>>>
>>>>>>>>>>>On February 27, 2001 at 08:24:14, Jorge Pichard wrote:
>>>>>>>>>>>
>>>>>>>>>>>>I believe that the drawish move was Qe3! instead of the Qxc6?  Can somebody
>>>>>>>>>>>>post the FEN string that produce the graphical position for me. Plus I wonder
>>>>>>>>>>>>what program save the draw in the shortest time possible?
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>None can see this.  It is a 60+ ply repetition.  Way beyond anything we can
>>>>>>>>>>>see today.
>>>>>>>>>>
>>>>>>>>>>We need to test all the programs in order to say that none can see it.
>>>>>>>>>>
>>>>>>>>>>Some programs like dark thought are not available so we cannot know that none
>>>>>>>>>>can see it.
>>>>>>>>>>
>>>>>>>>>>The fact that it is 60 ply repetition is not a proof that none can see it
>>>>>>>>>>because programs only to need the right extensions to see the relevant 60 plies
>>>>>>>>>>forward.
>>>>>>>>>>
>>>>>>>>>>If you are wrong then testing only one program is enough to prove it.
>>>>>>>>>>
>>>>>>>>>>Uri
>>>>>>>>>
>>>>>>>>>Deep Fritz doesn't see the draw yet after one hour, but it's getting there. Now
>>>>>>>>>I am trying with Deep Shredder and I might leave one of them computing the
>>>>>>>>>position overnight.
>>>>>>>>>
>>>>>>>>>What has been the evaluation that came closest to 0.00?
>>>>>>>>>
>>>>>>>>>Enrique
>>>>>>>>>
>>>>>>>>>Comp Deep Blue - Kasparov,G
>>>>>>>>>1r6/5kp1/RqQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/5K2 b - - 0 1
>>>>>>>>>
>>>>>>>>>Analysis by DEEP FRITZ   :
>>>>>>>>>
>>>>>>>><SNIP>
>>>>>>>>>45...Qe3 46.Qxd6 Re8 47.h4 h5 48.Bf3 Qc1+ 49.Kf2 Qd2+ 50.Be2 Qf4+ 51.Kg1
>>>>>>>>>  ±  (1.00)   Depth: 19/50   00:25:38  1907813kN
>>>>>>>>>45...Qe3!
>>>>>>>>>  ²  (0.69)   Depth: 20/52   00:56:36  4195172kN
>>>>>>>>>
>>>>>>>>>(Irazoqui, Cadaqués 27.02.2001)
>>>>>>>>
>>>>>>>>I have let various versions of Crafty search this position for very long periods
>>>>>>>>of time in the past.  I've had it return what looks like a draw score at around
>>>>>>>>ply 21 or so, and then at the next ply the score will go back up and the line
>>>>>>>>will change a lot!  I never could figure out why this was.
>>>>>>>>
>>>>>>>>BTW, there is a good line posted at the bottom of http://www.rebel.nl/db2.htm
>>>>>>>>for this position.  It looks like Fritz is following it fairly well so far, but
>>>>>>>>Crafty also looked to be following it when it found the drawscore, then it
>>>>>>>>changed its mind away from the line, and again the score started to rise. :(
>>>>>>>
>>>>>>>Same thing with Deep Fritz. What I find surprising is that Deep Blue expected
>>>>>>>46... Qxc6, when Deep Fritz, Deep Shredder beta and Junior 7 beta, for instance,
>>>>>>>pick Qe3 very quickly.
>>>>>>>
>>>>>>>Enrique
>>>>>>>
>>>>>>>
>>>>>>>Comp Deep Blue - Kasparov,G
>>>>>>>1r6/5kp1/RqQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/5K2 b - - 0 1
>>>>>>>
>>>>>>>Analysis by DEEP FRITZ   :
>>>>>>>
>>>>>>>45...Qxc6
>>>>>>>  ±  (0.91)   Depth: 1/6   00:00:00
>>>>>>>45...Qxc6 46.dxc6 Be7 47.Bd5+ Ke8 48.c7 Rc8 49.Ra7
>>>>>>>  ±  (1.16)   Depth: 6/15   00:00:00  10kN
>>>>>>>45...Qxc6 46.dxc6 Be7 47.Bd5+ Ke8 48.c7 Rc8 49.Ra7
>>>>>>>  ±  (1.16)   Depth: 6/15   00:00:00  10kN
>>>>>>>45...Qxc6 46.dxc6 Be7 47.Bd5+ Ke8 48.c7 Rc8 49.Ra7
>>>>>>>  ±  (1.16)   Depth: 6/15   00:00:00  10kN
>>>>>>>45...Qxc6 46.dxc6 Be7 47.Bd5+ Ke8 48.c7 Rc8 49.Ra7
>>>>>>>  ±  (1.16)   Depth: 6/15   00:00:00  10kN
>>>>>>>45...Qxc6 46.dxc6 Be7 47.Bd5+ Ke8 48.c7 Rc8 49.Ra7
>>>>>>>  ±  (1.16)   Depth: 6/15   00:00:00  10kN
>>>>>>>45...Qxc6 46.dxc6 Be7 47.Bd5+ Ke8 48.c7 Rc8 49.Ra7
>>>>>>>  ±  (1.16)   Depth: 6/15   00:00:00  10kN
>>>>>>>45...Qxc6 46.dxc6 Bc7 47.Ra7 Rc8 48.Rb7 Ke7 49.Rxb5
>>>>>>>  ±  (1.37)   Depth: 7/16   00:00:00  31kN
>>>>>>>45...Qxc6 46.dxc6 Bc7 47.Ra7 Rc8 48.Rb7 Ke7 49.Ke2 Kd6 50.Rxb5
>>>>>>>  +-  (1.44)   Depth: 8/17   00:00:00  65kN
>>>>>>>45...Qxc6 46.dxc6 Bc7 47.Ra7 Rc8 48.Rb7 Ke7 49.Ke2 Kd6 50.Rxb5
>>>>>>>  +-  (1.44)   Depth: 9/19   00:00:00  115kN
>>>>>>>45...Qxc6 46.dxc6 Kf8 47.Ra7 Rc8 48.Rb7 Ra8 49.Rxb5 Ra3 50.Rd5 Ke7
>>>>>>>  +-  (1.53)   Depth: 10/22   00:00:00  305kN
>>>>>>>45...Qxc6 46.dxc6 Kf8 47.Ra7 Rc8 48.Rb7 Rc7 49.Rxb5 Ke7
>>>>>>>  +-  (1.56)   Depth: 11/21   00:00:00  1078kN
>>>>>>>45...Qxc6 46.dxc6 Kf8 47.Ra7 Rc8 48.Rb7 Rc7 49.Rxb5 Ke7
>>>>>>>  +-  (1.56)   Depth: 12/22   00:00:01  1597kN
>>>>>>>45...Qxc6 46.dxc6 Rc8 47.Ra5 Ke7 48.Rxb5 h5 49.Bd5 Rb8 50.Rxb8 Bxb8
>>>>>>>  +-  (1.78)   Depth: 13/26   00:00:02  3199kN
>>>>>>>45...Qxc6 46.dxc6 Rc8 47.Ra5 h5 48.Rxb5 h4 49.Rb7+ Rc7 50.Bd5+ Ke7 51.Bxc4
>>>>>>>  +-  (1.84)   Depth: 14/26   00:00:06  6898kN
>>>>>>>45...Qe3!
>>>>>>>  +-  (1.81)   Depth: 14/39   00:00:21  25200kN
>>>>>>>45...Qe3! 46.Qd7+ Kg8 47.Qxd6 Rf8 48.Qe6+ Kh7 49.Qe7 Rg8 50.Bf3 Qc1+ 51.Kf2
>>>>>>>  ±  (1.22)   Depth: 14/39   00:00:29  34959kN
>>>>>>>45...Qe3 46.Qxd6 Re8 47.h4 Re7 48.Bf3 Qc1+ 49.Kf2 Qd2+ 50.Kg3 Qe1+ 51.Kh3
>>>>>>>  ±  (1.16)   Depth: 15/39   00:00:58  71272kN
>>>>>>>45...Qe3 46.Qxd6 Re8 47.h4 h5 48.Bf3 Qc1+ 49.Kf2 Qd2+ 50.Be2 Qf4+ 51.Kg1
>>>>>>>  ±  (1.28)   Depth: 16/43   00:01:48  133552kN
>>>>>>>45...Qe3 46.Qxd6 Re8 47.h4 h5 48.Bf3 Qc1+ 49.Kf2 Qd2+ 50.Be2 Qf4+ 51.Bf3
>>>>>>>  ±  (1.28)   Depth: 17/45   00:04:03  301257kN
>>>>>>>45...Qe3 46.Qxd6 Re8 47.h4 h5 48.Bf3 Qc1+ 49.Kf2 Qd2+ 50.Be2 Qf4+ 51.Bf3
>>>>>>>  ±  (1.28)   Depth: 18/46   00:09:44  723368kN
>>>>>>>45...Qe3 46.Qxd6 Re8 47.h4 h5 48.Bf3 Qc1+ 49.Kf2 Qd2+ 50.Be2 Qf4+ 51.Kg1
>>>>>>>  ±  (1.06)   Depth: 19/50   00:22:41  1678502kN
>>>>>>>45...Qe3 46.Qxd6 Re8 47.h4 h5 48.Bf3 Qc1+ 49.Kf2 Qd2+ 50.Be2 Qf4+ 51.Kg1
>>>>>>>  ±  (1.00)   Depth: 20/49   01:04:45  4818191kN
>>>>>>>45...Qe3!
>>>>>>>  ²  (0.69)   Depth: 21/54   02:32:36  11295715kN
>>>>>>>45...Qe3! 46.Qxd6 Re8 47.h4 h5 48.Bf3 Qc1+ 49.Kf2 Qd2+ 50.Be2 Qf4+ 51.Kg1
>>>>>>>  ²  (0.31)   Depth: 21/54   04:29:02  19961605kN
>>>>>>>45...Qe3 46.Qxd6 Re8 47.Qd7+ Re7 48.Qc6 Qxe4 49.d6 Qxf5+ 50.Kg1 Rd7 51.Qxb5
>>>>>>>  ²  (0.31)   Depth: 22/57   09:15:31  41587594kN
>>>>>>>45...Qe3 46.Qxd6 Re8 47.Qd7+ Re7 48.Qc6 Qxe4 49.d6 Qxf5+ 50.Kg1 Rd7 51.Qxb5
>>>>>>>  ²  (0.47)   Depth: 23/56   17:29:25  78776519kN, tb=3
>>>>>>>
>>>>>>>(Irazoqui, Cadaqués 28.02.2001)
>>>>>>
>>>>>>Compare to the analysis of yace after few hours of playing against itself and
>>>>>>going back.
>>>>>>
>>>>>>1...Qe3 2.Qxd6 Re8 3.Qd7+ Re7 4.Qc6 Qxe4 5.d6 Qd3+ 6.Kg1 Re8 7.Kh2 Qxf5 8.Ra2
>>>>>>Rd8 9.Qxb5 Rxd6 10.Qxc4+ Kg6 11.Ra7 Kh7
>>>>>>
>>>>>>It seems that the lines are similiar and I suspect that humans did not analyze
>>>>>>this line to prove if it is a draw or not a draw.
>>>>>>
>>>>>>I did not believe that this traingle Qc6-d6-d7-c6 can be a good idea for white
>>>>>>but maybe humans are wrong and it is not a draw.
>>>>>>
>>>>>>It is clear that deeper blue could not find this idea otherwise the programmer
>>>>>>could tell the public that it is not a draw so practically kasparov could draw
>>>>>>by playing Qe3 but it does not answer the question if Qe3 is objectively a draw.
>>>>>>
>>>>>>Uri
>>>>>
>>>>>OK, fine, but obviously Qe3 is the only chance for black and Deep Fritz, Deep
>>>>>Shredder and Junior 7 beta see it very quickly. If Deep Blue was so extremely
>>>>>fast, why did it expect Qxc6 and didn't see Qe3?
>>>>>
>>>>>Enrique
>>>>
>>>>I believe that the answer is simply one of the following or both of them:
>>>>
>>>>1)Deeper blue is worse than Deep Fritz and could not see the perpetual because
>>>>it did not do the right extensions(the main line of Deep Fritz suggests that
>>>>Deep Fritz can see the perpetual after many hours so it has a main line that
>>>>avoids it)
>>>
>>>This is totally flawed reasoning.  For a program to see a perpetual in its
>>>search, it must meet two conditions.  The conditions are summed up in the
>>>following possibilities:
>>>
>>>1.  It finds a perpetual it can force, regardless of what the opponent does,
>>>and it believes that it is losing unless it takes the perpetual.
>>>
>>>2.  If finds a perpetual the opponent can force, regardless of what it does
>>>itself, and it believes that the opponent is losing and will choose to take
>>>the perpetual.
>>
>>I believe that Deep Fritz found that the line Qe3 Qxd6 Re8 h4 is not good
>>because black can force a perpetual check after h4
>>
>>I believe that this is the reason that it changed it's ,mind at depth 22/57 to
>>the line Qe3 Qxd6 Re8 Qd7+
>>
>
>I take a different view.  I believe that Fritz simply doesn't like the white
>position, and it thinks that black can do no better than repeat, and white
>_wants_ to repeat.  In reality, white is better and the perpetual is black's
>only hope.  If a program doesn't think white will try very hard to avoid the
>draw, then it might find a draw much easier.  But it isn't going to see the
>draw that is 60+ plies deep, ever.
>
>And white can push the draw off that far with best resistance.
>
>
>
>
>>I am not sure if the depth 22/57 mean that the longest line is 57 plies and it
>>is possible that there is a display bug(assuming that you need 60 plies to see
>>the repetition and 57 is not enough).
>>
>>Another possibility is that hash tables helped Fritz to see it in less plies.
>>It is possible that it found a draw in 57 plies in line A and line B that is 61
>>plies leaded to one of the positions from line A so it did not need to search
>>all the line.
>>
>>>
>>>If a program believes white is losing in this position, then it will think white
>>>is trying to repeat and black is not.
>>>
>>>If a program believes black is losing in this position, then it will think black
>>>is trying to repeat and white is not.
>>>
>>>Those two cases are vastly different.  If DB believed (1) and Fritz believes (2)
>>>then comparing their results of looking for a draw is pointless.
>>>
>>>In any case, the forced (absolutely forced by black with white struggling to
>>>avoid the repetition) requires > 60 plies to see.  I don't care _how_ long you
>>>give fritz, it is _not_ going to be able to find the forced draw.  It is simply
>>>too deep.
>>
>>Deep Fritz display depth 22/57 plies after 9:15:31 so depth>60 does not seem
>>impossible.
>>
>>Uri
>
>I max out on that position at depth=64.  But that happens on a very few
>branches.  To find the forced repetition, it is going to happen on _many_
>branches, which pushes the search time limit required into at least months
>and probably years...

Can you show the 64 plies?

45... Qe3! 46.Qxd6 Re8! 47.h4? h5! 48.Bf3 Qc1+ 49.Kf2 Qd2+ 50.Be2 Qf4+ 51.Kg1
Qe3+ 52.Kh2 Qf4+ 53.Kh3 Qxf5+ 54.Kh2 Qf4+ 55.Kg1 Qe3+ 56.Kf1 Qc1+ 57.Kf2 Qf4+

Are these the first 25 plies?

If they are then I am interested in the last 39 plies.
If they are not what is the wrong move except 47.h4 when after 47.Qd7+ programs
cannot prove a repetition?

Uri



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.