Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Null move generalization

Author: Ed Schröder

Date: 02:27:18 04/18/02

Go up one level in this thread


On April 17, 2002 at 17:45:03, Peter McKenzie wrote:

>On April 17, 2002 at 16:53:30, Gian-Carlo Pascutto wrote:
>
>>On April 17, 2002 at 16:40:01, Dann Corbit wrote:
>>
>>>I don't see how it can completely remove the tactic from ever being seen unless
>>>the implementation of null move is broken.
>>
>>If it has a zugzwang, you will _never_ see it if you don't check for it
>>(verification search, double nullmove), or use another trick like force
>>nullmoves away from the root (that is what Crafty does).
>>
>>You can always play the 'nullmove' no matter what depth, so you'll
>>never realize the nullmove is no good.
>>
>>i.e. you make an illegal move that is not possible in the real game, and
>>increasing search depth won't change this.
>
>Yes, you are correct.  Some programs can never solve the following:
>
>[D]8/8/2p5/pkp3R1/7B/P7/2p3K1/8 w - -

Rebel uses null-move since its latest version, no problem, just 8 seconds to see
the draw. Maybe it is a matter of eval too, could be...

Ed

===================

00:00:02.8	0,01	12	853807	Rxc5+ Kxc5 Bg5 Kd4 Kf3 Kd3 Bf4 c5
00:00:03.3	0,00	13	1771119	Rxc5+ Kxc5 Bg5 Kd4 Kf3 Kd3 Bf4 c5
00:00:05.6	-1,41	14	4456961	Rxc5+ Kxc5 Bg5 Kc4 Kf2 Kb3 Bc1 Ka2
00:00:08.7	0,00	14	7068951	a4+ Kb6 Bf2 c1Q Rxc5 Qxc5 Kh1
00:00:11.1	0,00	15	11124240	a4+ Kb6 Bf2 c1Q Rxc5 Qxc5 Kh1
00:00:18.6	0,00	16	20139015	a4+ Kb6 Bf2 c1Q Rxc5 Qxc5 Kh1
00:00:30.6	0,00	17	34891048	a4+ Kb6 Bf2 c1Q Rxc5 Qxc5 Kh1
00:00:53.2	0,00	18	64932256	a4+ Kb6 Bf2 c1Q Rxc5 Qxc5 Kh1

Revel XP (alpha), A1400, 60 Mb


>http://homepages.caverock.net.nz/~peter/eg_test/pet013.htm
>
>>
>>--
>>GCP



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.