Author: Olaf Jenkner
Date: 10:59:50 08/14/00
Go up one level in this thread
On August 14, 2000 at 13:35:59, Heiner Marxen wrote: >On August 14, 2000 at 10:20:59, Uri Blass wrote: > >>On August 14, 2000 at 08:43:08, Enrique Irazoqui wrote: >> >>>This one is a MES position: >>> >>>[D]3B4/1r2p3/r2p1p2/bkp1P1p1/1p1P1PPp/p1P1K2P/PPB5/8 w - - 0 1 >>> >>>After 6.f5, all programs I tried evaluate it from +11 to +17. >>> >>>Enrique >>> >>>[Event "?"] >>>[Site "?"] >>>[Date "????.??.??"] >>>[Round "?"] >>>[White "MES.941"] >>>[Black "?"] >>>[Result "*"] >>>[SetUp "1"] >>>[FEN "3B4/1r2p3/r2p1p2/bkp1P1p1/1p1P1PPp/p1P1K2P/PPB5/8 w - - 0 1"] >>>[PlyCount "11"] >>> >>>1. Ba4+ Kxa4 2. b3+ Kb5 3. c4+ Kc6 4. d5+ Kd7 5. e6+ Kxd8 6. f5 * >> >>This example proves that programs do not know to search the right lines. >>It is possible to prove that white has at least a draw by search+hash tables if >>you search the right lines. >> >>The number of relevant positions that programs needs to look at in order to >>prove by a tree that white has at least a draw is >>4*(22*21)/2*19*17*2=78*231*34<1000000 so it is not a problem for the right >>program to prove the draw. >> >>4 is the number of squares for the black bishop. >>19 is an the number of squares of the black king(after choosing a square for the >>bishop). >> >>22*21/2 is the number of squares of the black rooks(after choosing the squares >>for the black bishop and the black king). >>17 is the number of squares for the white king. >>2 is the number of sides to move. >> >> >>The only problem is that programs need to be selective in searching white moves >>and to avoid searching captures of white and chess programs do not know to do it >>(the first move that they analyze is capturing the rook in case that it is >>possible). >> >>Uri > >That is not enough. >The program also has to find that white can force a "draw by repetition". >When you try to do this as part of the normal search, the minimum search >depth needed to recognize it is the length of the longest cycle that the >opponent can manage to set up. Here it is in the order of 4*(22*21)/2*19, >= 17556. That if far too much for a search depth. > >I.e. you also have to have a special detection algorithm. The normal search >is not sufficient to recognize such large loops. > >Of course, it can be done ;-) it is just not very easy. > >Heiner You don't have to detect a draw to find the right move. Each other move will lead to breakthrough of the black pawns. After that the evaluation will get much worse. I analysed the position with Crafty some month ago. The results were intersting. It takes me some time to find them. May be I post them a little bit later. OJe
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.