Author: Albert Silver
Date: 07:35:31 03/03/01
Go up one level in this thread
On March 03, 2001 at 08:50:26, Vincent Diepeveen wrote:
>>I believe the 60 ply analysis is posted on Ed's web site. I have not looked
>>at it in quite a while, but it was pretty comprehensively covered there the last
>>time I looked at it. I didn't make up the 60 ply number...
>
>My maximum search depth during this game is around 100 plies at
>both processors.
>
>In openings position i see already several 60 ply lines :)
>
>Of course majority is completely nonsense of those lines, but
>it *does* cover checking lines.
>
>It's obvious that you need to see certain tactics and checks to
>get a near to draw score. With little extensions i evaluate
>Qxc6 highest for sure. Only adding extensions DIEP is giving
>Qxe3 a low score, as there are a few crucial draws to see
>in the lines!
>
>>>I also did not see a forced draw in the line Qe3 Qd6 Re8 Qd7 Re7 Qc6 and in this
>>>line white gives the bishop.
>>>
>>>I do not know if the line leads to a draw and I know that I did not read an
>>>analysis of this line by the GM's at that time probably because the GM's did not
>>>believe that wasting tempos can be a good idea(I also did not believe in it)
>>>
>>>The only thing that can be proved is that programs cannot see that white can win
>>>material after Qe3 and I believe that this is the reason that the score is only
>>>0.xx.
>>
>>Roman was one of the first GM players that suggested that line. But after we
>>went over it for a _long_ while everyone became convinced that white could not
>>win that way either... but it is very hard to prove this and I don't think a
>>program has a chance in hell of following that analysis from the point where it
>>has to find Qe3 with a draw score...
>
>Perhaps Roman missed a shorter side line. But again,
>i can't rule out DIEP *did* checkout some 60+ lines to get the draw
>score.
I think that proving a forced draw is beyond the current software/hardware
combinations, and that programs that claim it now, right from Qe3, are missing
resources for both sides. In other words, it is simply a horizon effect. Let me
give you an example. Take the Qc6 line 45...Qe3 46.Qxd6 Re8 47.Qd7+ Re7 48.Qc6
which is already 6 plies deep. Now let's go further a few moves with 48...Qxe4
49.d6 Qd3 50.Kg1 Re8 51.Qc7+ which is now already 12 plies deep from and
including 45...Qe3:
[D]4r3/2Q2kp1/R2P1p1p/1p2pP2/1Pp5/2Pq3P/6P1/6K1 b - -
Here, how long does it take for DIEP (other programs have similar problems)
to see that 51...Kf8 is MUCH worse than 51...Kg8 (I can't find anything
conclusive for White with 51...Kg8). Remember this line isn't some crazy
sideline and that a program would be evaluating this from 12 plies already. If
it doesn't see that 51...Kf8 is a worse choice, it will also not properly
estimate the danger of 51.Qc7+ either.
Now feed it 51...Kf8 and see how long it takes to go for 52.Ra1 which is
White's most promising move IMO. You'll notice that after some time thinking
DIEP will probably want to play 52...Kg8 because it needs to make room for the
rook on f8. So feed it the further moves 52...Kg8 53.d7 Rf8 (this is already 17
plies deep including Qe3 BTW) and now again let DIEP (or any engine) analyze
this for a long time.
[D]5rk1/2QP2p1/5p1p/1p2pP2/1Pp5/2Pq3P/6P1/R5K1 w - -
I let Tiger 13 do this and it changed its mind regularly until it finally
changed its mind AGAIN at ply 16 (!) to 54.Rf1 with a +1.04 evaluation. It may
not be a forced win, but it certainly isn't a forced draw here. Yet the program
would already have to evaluate that Kf8 is bad (and why) from 12 plies and see
Kg8 to neutralize Qc7+. If it doesn't see this, then it saw a draw or possible
draw only because it didn't see and neutralize White's resources.
Albert
>
>Any depth limitation on the extensions and it already doesn't
>find it!
>
>I'll try tonight to limit the total search depth to 32 plies. then 40 plies.
>So only applying evaluation then (not a qsearch as i can see perpetual
>lines of 32 plies easily in qsearch too)
>
>>
>>
>>
>>>
>>>I do not believe that Deeper blue evaluated these positions as +2.xx and I
>>>believe that other programs that can see the 0.xx evaluation and the line Qe3
>>>Qd6 Re8 Qd7 Re7 Qc6 simply can see deeper than deeper blue.
>>>
>>>Uri
>>
>>The analysis for DB is readily available. I have the log files if you can't
>>find them anywhere. I don't remember what its analysis actually was, but we
>>do have it...
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.