Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Who can explain F6 behaviour-EGTB problem?

Author: Sanjiv Karnataki

Date: 14:53:40 01/15/00

Go up one level in this thread


On January 15, 2000 at 12:07:26, Mike S. wrote:

>Hi Henk,
>this could be a "null move" problem. I have tried this with Fritz6 light (from
>the demo); it doesn't find 2.axb6. The program might think that after
>1.c7 Kxc7 2.axb6+ Kxb8 3.b7, if black makes the null move (=doesn't move) and
>white could move once again, the best (and only) move would be 4.bxa8Q+ Kxa8
>with a draw. So Fritz may think, if 3.b7 is a draw even if white moves twice, it
>must be draw anyway if black can move normally in between. The zugzwang
>situation after 3.b7 is obviously not very well handled by the program.
>But I don't really exactly understand the null move logic (I'm afraid I don't
>even understand alpha-beta...), so I hope one of the programmers here can tell
>if it's a null move failure.
>If the move 2.axb6 is made, F6 light immediately shows # in 18 (# in 9 after
>2...Kxb8) when 4 piece-tablebases are accessed.
>
>Regards,
>M.Scheidl
>
>P.S.
>What is the source of this position? I think I've seen it before.

I don't think the problem is with null moves. I think the problem is egtb. with
EGTB crafty after 2:00 minutes comes up with 1. c7 nc8 with adv. for white
. turn off egtb and it immediately comes up with 1. c7, kxc7 2.axb6 etc.

That is because with 8 pieces on the board, most programs are 3-4 pieces away
from known egtb values. So with egtb, both sides know that the correct
continuation is a mate by white in x no of moves. since both sides are using
negamaxing to find the best moves for both sides, they tend to push the
certainty of death beyond their search horizon by keeping more pieces on board.
so with egtb instead of evaluating to mate, it evals to +8.00 a huge adv. but
less than mate.

Thats my theory and I will admit I am pretty new at this, so be kind if I am
wrong.

Thanks



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.