Author: Christophe Theron
Date: 09:37:39 02/08/04
Go up one level in this thread
On February 08, 2004 at 04:47:35, Uri Blass wrote:
>On February 07, 2004 at 22:10:45, Christophe Theron wrote:
>
>>On February 07, 2004 at 18:44:14, Torstein Hall wrote:
>>
>>>On February 06, 2004 at 20:04:48, Christophe Theron wrote:
>>>
>>>>On February 06, 2004 at 19:33:22, Bob Durrett wrote:
>>>>
>>>>>On February 06, 2004 at 19:18:54, Christophe Theron wrote:
>>>>>
>>>>>>On February 06, 2004 at 13:48:48, Jaime Benito de Valle Ruiz wrote:
>>>>>>
>>>>>>>This endgame study can be game following the sequence
>>>>>>>
>>>>>>>1.Bd1 g1Q 2.Bxa4 Qc1 3.Bxd7 h5 4.Be8 h4 5.Ba4 Line
>>>>>>>
>>>>>>>[d] 8/3p4/p6p/k2N3B/p7/K6p/PP4pP/8 w - - 0 1
>>>>>>>
>>>>>>>Many engines (Fritz 8, Hiarcs 8, Ruffian 1.01) cannot win this endgame, probably
>>>>>>>due to the well known null-move problem for extreme and rare positions such as
>>>>>>>this.
>>>>>>>
>>>>>>>They just don't find 5. Ba4
>>>>>>>
>>>>>>>Try your engine after 4...,h4
>>>>>>>
>>>>>>>[d]8/3p4/p6p/k2N3B/p7/K6p/PP4pP/8 w - - 0 1
>>>>>>>
>>>>>>>All the above mentioned engines play 5.Bd7 and get a draw instead of winning.
>>>>>>>
>>>>>>>Regards,
>>>>>>>
>>>>>>> Jaime
>>>>>>
>>>>>>
>>>>>>
>>>>>>Chess Tiger X: 1.15 second on PIII-M 933MHz (Dell X200), 6Mb HT:
>>>>>>
>>>>>>N14 0.33s Bd1 g1=Q Bxa4 Qc1 Bxd7 h5 Be8 h4 Bd7 Qxb2+ Kxb2 0.00
>>>>>>N15 0.66s Bd1 g1=Q Bxa4 Qc1 Bxd7 h5 Be8 h4 Bd7 Qxb2+ Kxb2 +0.90
>>>>>>N15 1.15s Bd1 g1=Q Bxa4 Qc1 Bxd7 h5 Be8 h4 Ba4 Qxb2+ Kxb2 Kxa4 Kc... +3.26
>>>>>>
>>>>>>
>>>>>>
>>>>>> Christophe
>>>>>
>>>>>The key question to be answered is:
>>>>>
>>>>>"What is there about Chess Tiger X that is different from the other engines
>>>>>mentioned which accounts for the fact that they could not find the move but
>>>>>TigerX could? Is the "null-move problem" not applicable to TigerX or is there
>>>>>some other reason?
>>>>>
>>>>>Bob D.
>>>>
>>>>
>>>>
>>>>I have worked a lot on zugzwang detection in Tiger 15, and it has been improved
>>>>further in the new version (CT X). I think that's the answer. It's simply an
>>>>area I've been working on.
>>>>
>>>>
>>>>
>>>> Christophe
>>>
>>>Can I asume you belive there is quite a bit of extra strength to pick up for a
>>>chess engine fixing this problem? I belived this occured pretty rarely in most
>>>games.
>>>
>>>Torstein
>>
>>
>>
>>Many pawns endgames are decided by very deep zugzwangs. These endgames are
>>generally not very interesting but human players excel at understanding them. So
>>I believe a substantial improvement is possible here for a chess engine.
>
>I believe that chess engines usually do not use null move pruning in pawn
>endgames.
And so they compute less deep. And so they miss some deep zugzwangs.
The trick is too see the zugzwang in a pawn endgame before entering this
endgame. Before the exchange of the last non-pawn (piece) happens.
Many programs will happily enter a lost pawn endgame because they cannot see
deep enough and do not understand that they are rushing into a losing zugzwang.
Human players usually can see this, and they decide correctly to go for the pawn
endgame now, or to keep playing the endgame and keep the last piece.
So you need to see really deep, and just disabling pruning in pawn endgames is
not going to help reaching this goal.
Christophe
>The problems of zugzwangs was in positions that the side to move had pieces that
>are not pawns.
>
>I believe that in part of the cases both programs do not see the zugzwang and
>the same for the programmer who watch the game and this can cause
>underestimating the demage that null move pruning cause.
>
>Uri
>
>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.