Author: Robert Hyatt
Date: 19:49:09 06/08/01
Go up one level in this thread
On June 08, 2001 at 17:33:01, Bruce Moreland wrote: >On June 08, 2001 at 16:37:57, Daniel Clausen wrote: > >>Hi >> >>On June 08, 2001 at 16:23:39, Robert Hyatt wrote: >> >>>On June 08, 2001 at 16:21:01, Robert Hyatt wrote: >>> >>>>On June 08, 2001 at 16:08:52, John Merlino wrote: >>>> >>>>>On June 08, 2001 at 15:19:09, Robert Hyatt wrote: >>>>> >>>>>>Here is a cute position that occurred between a commercial program and Crafty >>>>>>last week: >>>>>> >>>>>>[D] 8/p4pp1/3rk2p/4p3/2P5/2K4P/P2R1PP1/8 w - - 0 1 >>>>>> >>>>>>Crafty was black and moved the rook to d6, offering a trade. The opponent >>>>>>took it and was happy to do so. Unfortunately, white is lost. White saw >>>>>>the passed pawn and apparently was quite happy. Crafty's static evaluation >>>>>>for this position is -1.0 roughly. >>>>>> >>>>>>For those that "don't do endgames" black's king-side majority is the problem >>>>>>here. White's passer gets blockaded, white has to desert it to stop black's >>>>>>kingside passer he makes after a few pawn moves, and then black eats white's >>>>>>a pawn and promotes. >>>>>> >>>>>>Instructional, at least. These are the kinds of positions you want to >>>>>>see your program get right. I saw a very similar one against a GM today, >>>>>>playing Crafty. He calculated for a long time after crafty offered to trade >>>>>>the last piece on the board. He traded, and 10 moves later realized he was >>>>>>dead lost. :) >>>>> >>>>>How did this game end after the rooks were traded? >>>>> >>>>>jm >>>> >>>> >>>>Black wins easily. >>> >>> >>>I should have added, the point here is that white ends up in a king and pawn >>>ending with a passed pawn to none for black. Black has a kingside majority that >>>many programs ignore. Here it is decisive. >> >>I'm not a good chess player but trying to summarize. Please correct me if I'm >>wrong. Black can easily transform the pawn majority on the kingside into a >>passer. After that White and Black both have a passer but Black wins because his >>(or her, or its :) passer is farer (is that an English word?) away than white's. >>Basically a program (or a human) has to recognize that in this situation the >>pawn majority for black on the kingside is worth a passer. >> >>Does this sound about right? >> >>Regards, >> >>Sargon > >Black sits in front of the c-pawn and waits. If white does nothing, black >advances the k-side pawns, creates a passer, eats the c-pawn, eats the a-pawn, >and queens. > >If white goes after the black a-pawn, black can probably take the c-pawn, >advance the k-side pawns, create a passer, and win. If white actually takes the >a-pawn, the black king can come to the c-file and seal the white king in long >enough that black can queen and mate. > >The key is that the pawn can be blockaded, white is not able to use the pawn to >divert black from something critical, and black's pawns can be used to divert >the white king away from the pawn *and* something critical (the a-pawn). > >White is trying to win on a very narrow front and can't, while black can make >the area of action wider, at his leisure. > >My program would mess this position up, too. I'll give Bob some credit for >having heuristics that work here. However, I'd be a little concerned about >evaluating this as +1 statically, since white does have a passer, and it may be >somewhat rare that he has no effective targets that can be overcome by >diversionary sacrifice of the pawn. > >bruce I did a poor job explaining that -1. The -1 comes not from the position I gave, but from the position after the logical moves of Rxd6 Kxd6 Kb4 Kc6, not from the original position. In the original position, the static evaluation is just slightly negative (-.41 actually). The opponent, whom I won't name, thinks this is over +1 for _white_ when he plays Rxd6+. That is way off. Crafty thought it was just slightly in its favor there, until white let it blockaed the pawn (or reach a position where it could). IE after the above moves, the static eval is -.80... Sorry. I gave the starting position, but was really talking about the resulting position after the obvious moves following Kxd6. This is really reasonably accurate, approximately like the outside passed pawn code, although even that is wrong on occasion of course. But I got tired of getting tricked into trading in obviously lost positions. I watched "deep shredder" do this several times recently on ICC (there are a few of them running there). It (and others) have several times simplified into a trivially lost position where keeping on the rook(s) makes progress unlikely or difficult. A couple of GM players exploit this weakness. Roman ragged me mercilessly until I fixed it after "CptnBlueBear" found the weakness in Crafty. :)
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.