Author: Ricardo Gibert
Date: 16:55:48 12/10/02
Go up one level in this thread
On December 10, 2002 at 18:53:21, Mike S. wrote: >Maybe it should be called "The Mig Fortress" better, because it'S Mig's >comprehensive analysis of game no. 6 of the Bahrain match which ends with this >position: > >[D]8/5k1p/5q2/8/1R6/6P1/5P2/6K1 b - - 0 45 > >White's rook can go to f4 or h4, and "Black has no way through the fortress." > >http://www.chessbase.com/images2/2002/bahrain/games/mig6.htm (see after the last >move on the bottom of the page) > >I think the fortress idea is something especially difficult for chess programs, >because it's very "general", IOW not related to a special pattern or material >balance. There are a number of abilities that are closely associated with chess mastery, but nevertheless have very little or no effect on playing strength overall. Recognizing such fortresses is one of them. Don't get me wrong. I *do* think the topic is an interesting one. However, such positions are rare and detection would be too costly. Perhaps limiting detection to the root position would be okay, but that would not add to playing strength. It would only help the program to offer/accept draws in a more timely fashion. > >I wonder if an idea from the incomplete tbs. problem could be used to detect >fortresses: With the pawn just before the promotion, but the tablebases for the >resulting material missing, the idea was like: "If it's # in 14, but one move >later *not* # in 13 or less, then there must be something wrong and I have to >promote the pawn (even if the eval is less than mate then, at first)." > >The situation is somewhat similar in a fortress position, when one side has a >more or less huge material advantage (which would normally be capable of >winning), but can't make any significant progress for a long series of moves. > >Testing such a detection would probably require to have the engine actually play >some f moves, so the engine could recognise that there is no progress... >Although, with usual search depths of ~12 or more plies (+ extensions), maybe >that is sufficient for the idea too. > >Are there engines known to have a fortress detection? > >Regards, >M.Scheidl
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.