Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: A Tough Endgame For Computers!

Author: Terry McCracken

Date: 20:59:14 04/07/02

Go up one level in this thread


On April 07, 2002 at 17:51:40, martin fierz wrote:

>On April 07, 2002 at 17:22:03, Dieter Buerssner wrote:
>
>>On April 07, 2002 at 15:48:22, Terry McCracken wrote:
>>
>>>Here's an Endgame that should make _all_ chess engines really work!
>>
>>[3n2k1/p2P2p1/1p6/7Q/5q1P/8/PP6/1K6 w - -]
>>
>>
>>>Fritz 7 sees the draw but doesn't understand the loss after 1.Qe8+..Qf8??
>>>
>>>Played by Mieses-NN 1903. Mieses won after Qf8??, I'll give the main line.
>>>
>>>
>>>1.Qe8+ Qf8 2.h5 Nf7 3.Kc2 Nd8 4.Kc3 Nf7 5.a4 Nd8 6.Kc4 Nf7 7.Kd5 Nd8 8.b4 Nf7
>>>9.Kc6 Nd8+ 10.Kc7 Line
>>
>>Which would be the winning line after 2...b5. My engine unfortunately fails to
>>see it ...
>>
>>You give Qf8 a "??". Should it be really obvious, that tis is a bad move?
>>
>>Regards,
>>Dieter
>
>hi dieter,
>
>yes, it is obvious that this is a bad move. the point is something which
>computers are worst at, an "eternal bind". after 1. ...Qf8 2.h5, no black piece
>can move any more: if the king moves, black loses the queen, the queen cannot
>move, because it's illegal, and so the only thing to do is to shuffle the knight
>back and forth, because if you move it away, you lose to d8Q. so all you have to
>see is that after 2.h5 black can only move the knight, and white can move the
>king right down to c7. ...g6 at any time is met by h6-h7-h8 winning. for a
>human, this is really easy - we recognize the fact that black is reduced to
>shuffling and can ask ourselves: where would i like my king to be - i don't have
>to calculate anything! a computer would have to see through the whole line until
>the knight is lost. which is quite deep...
>this type of bind is causing me headaches in my checkers program, and it is a
>very general class of positions where chess programs are weak at too.
>
>aloha
>  martin

You described it best Martin, with "Eternal Bind", and your quite correct this
position is fairly easy for experienced players, not computers and a true
"Nightmare" for chess programmers!

It would be interesting how Deeper Blue would have tackled this position! It may
have been able to search far enough.


As for your checker programme maybe http://games.cs.ualberta.ca/~chinook/book/
could help?

I have a friend near Ottawa, Martin Devenport who knows Jonathan Schaeffer, and
have been friends since 1971. Both are chessmasters as well. However I don't
know Jonathan Schaeffer.

Maybe when I catch Martin online if I remember...I'll ask him about Jonathan and
Chinook, maybe he can ask him about specific programming techniques, that could
help you.

I can't promise anything, but I'll ask anyways. If I remember?:o)

Regards,
 Terry



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.