Author: Dieter Buerssner
Date: 03:51:12 10/21/00
Go up one level in this thread
On October 21, 2000 at 05:28:14, Uri Blass wrote:
>I composed this position to see if chessmaster can see perpetual check.
>
>[D]4k3/6RQ/6PP/8/r7/8/r7/7K b - - 0 1
>
>The target is to see 0.00 score
>
>I found that chessmaster does not see the draw even after some minutes but after
>Ra1+ Kg2 it can see the draw very quickly in clearly smaller depth.
>
>The same after Ra1+ Kh2
>
>I tried also Fritz5.32 and it does not see the draw at least after a short time.
>
>If I give Fritz5.32 the moves Ra1+ Kg2 it can see the draw but after Ra1+ Kh2
>it has problems to see the draw.
>
>Fritz5.32 expects Kg2 when it plays Ra1+ and wants to play Kg2 at depth 13 with
>evaluation 0.00.
>
>How many squares can the white king visit before the repetition?(it is clear
>that the white king cannot visit all the 21 squares (b1-b3... h1-h3))
I think one big problem with such positions is the hash table.
I tried this position with Yace and 10M hash, and at depth 12, after
23 seconds, a draw score was shown. I actually was surprised, and
tried with larger hash. In one try, it needed 10 times as long.
1927643 12.21 -7.40 11. : Ra2a1+ Kh1h2 Ra4a2+ Kh2g3 Ra2a3+ Kg3f4 Ra3a4+
Kf4e5 Ra4a5+ Ke5d4 Ra5a4+ Kd4e3 Ra1a3+ Ke3e2
Ra3a2+ Ke2d3 Ra2a3+ Kd3d2 Ra3a2+ Kd2e1 Ra2a1+
Ke1f2 Ra1a2+ Kf2f3 Ra2a3+ Kf3g2 Ra4f4 [-660]
3901167 23.35 0.00 12++: Ra2a1+ Kh1h2 Ra1a2+ Kh2g1 Ra2a1+ Kg1f2 Ra1a2+
Kf2f1 Ra2a1+ Kf1g2 Ra1a2+ Kg2f3 Ra2a3+ Kf3g2
Ra3a2+ [-660]
6266901 32.93 0.00 12t : Ra2a1+ Kh1h2 Ra1a2+ Kh2g1 Ra2a1+ Kg1f2 Ra1a2+
Kf2f1 Ra2a1+ Kf1g2 Ra1a2+ Kg2f3 Ra2a3+ Kf3g2
Ra3a2+ [-660]
9543141 48.06 0.00 12. : Ra2a1+ Kh1h2 Ra1a2+ Kh2g1 Ra2a1+ Kg1f2 Ra1a2+
Kf2f1 Ra2a1+ Kf1g2 Ra1a2+ Kg2f3 Ra2a3+ Kf3g2
Ra3a2+ [-660]
I try to be careful, when getting the move from the hash. I always
check if the move yields in a repetition, and only accept the score, when
it doesn't yield in a repetition, or when the score is a draw score.
(When the score is no draw score and the hash move yields in a
repetition, and draw-score >= beta, this is obviously a fail high move)
But this only helps a little bit.
Say you have analyzed a position after one move with depth 12.
At the same search depth, you get to the same position after more
moves. Now, the depth in the hash table seems sufficient. I think,
there is no efficient to detetect, that because of the different path,
the position was reached, you can see, that now a repetition may be forced.
Of course the path could be hashed, but this would mean more or less to use
the hash table only for move ordering.
If I analysed correctly, with depth 12 in this sort of repeated check,
Yace will search 30 ply, which may indicate, that the king visited 14
different squares.
-- Dieter
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.