Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: If you gave a program unlimited time, could it solve a 5 piece ending?

Author: Robert Hyatt

Date: 16:43:39 06/02/02

Go up one level in this thread


On June 02, 2002 at 11:43:49, Jesper Antonsson wrote:

>On June 02, 2002 at 10:50:52, Robert Hyatt wrote:
>
>>On June 02, 2002 at 04:47:21, Terry Ripple wrote:
>>
>>>Could Fritz,Century,Tiger,Junior,or Crafty solve any or all the 5 piece endgame
>>>problems if given no time limit and not able to use Nalimov's tablebases?
>>>
>>>If the answer is no, then why not if the program could take any amount of time
>>>that it needed to find the correct answer to the solution?
>>>
>>>Thanks in advance for any information!
>>>
>>>Best Regards,
>>>Terry
>>
>>
>>Solve = See the mate, or Solve = play moves good enough to win?
>>
>>In many endings, the computer can do fine.  But not all.  IE KNN vs KP
>>is very difficult.  Particularly if you want to actually find the mate
>>which can be well beyond 200 plies...
>
>That's true, but I percieved the question as more theoretical than practical, so
>it should be the case that using unlimited time, any correct program with no ply
>limit or stack limit that prohibits 230 or whatever recursive calls to search(),
>should be able to solve any five man ending. It may take billions of years,
>though, on reasonable hardware.

OK... In that case, the answer is "yes".  In the practical case, the answer
is "no" unless unlimited time means a year or more.  IE to compute KPPKP would
be quite a lot of work, starting from scratch.

>
>However, experiment with the though that we have a terabyte of RAM or something
>like that (some high end database servers do, I've heard). Then it should be
>possible to solve any five man tablebase through ordinary search, since the
>program will in essence construct the relevant subset of the tablebases, albeit
>distributed in it's hash tables. It shouldn't have to be more than an order of
>magnitude or two slower than ordinary tablebase construction. With specific
>coding, it might even be faster than tablebase construction since we only need
>to walk through a small subset of the positions in the relevant EGTBs.



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.