Author: Richard A. Fowell (fowell@netcom.com)
Date: 12:47:36 12/05/98
Go up one level in this thread
I suspect we can make the question more clear-cut ... why don't we start with an easier problem: what programs find the key move when they get to the critical position: 2r5/3bk1b1/3pp3/p6P/5P2/1PN4R/1KP1B3/8 b - - bm Rxc3; (eval should show a pretty significant advantage). This is 8 ply into your sequence - a program that can't find the key move here in reasonable time is hardly likely to find it from the original position - 8 ply is a factor of 6500 or so in nodes for a null-move program. If this position is a Nolot-class position for computers, your original position would be ever much more so. Richard A. Fowell. I gave the position above to a few programs: - MacChess 5.0b6 doesn't see it in 12 minutes - HIARCS 7 Beta didn't see it in 35 minutes (full search of 11 ply ) By the way, after ... Rxc3, Rxc3, it isn't so clear to me that White can't set up a blockade to keep the Black King out. By placing the bishop on the d3-h7 diagonal, and the h-pawn at h6, White creates two "walls" to block the Black king from getting to any of his pieces - the c-file (with the Rook) and the white diagonal d3-h7, with the interstices blocked by pawn control of g7, g5, e5, and the d4 square is blocked by the need of maintaining the pin on the rook. What's Black's plan for penetrating that blockade? To the extent it involves shielding Black's king with his White-square bishop to cross the c-file or the 3-rank, can White foil this by covering the potential crossing points by controlling the White long diagonal with his Bishop? Perhaps Black's best plan is to trade off the white-square bishops - if this can be managed, simple Zugzwang will be White's downfall - all pawn moves lose pawns, and after the pawn moves are gone, White's rook is lost. White's plight is further enhanced by the fact that his only option to free a square to maintain King protection of his rook and ease out of the pin is b4, to which axb4 means that Kb3 is too late to stop axc3 or Bxc3. A pretty off-hand sequence is: 1. Rxc3 2. Rxc3 Bd4 3. h6 Kf6 4. Bd3 Be8 5. Be4 Bg6 6. h7 Kf7 7. Bd3 d5 8. Be2 Bxh7 9. Bg4 Kf6 10. Be2 Bg6 11. Bg4 Be4 12. Bh5 Kf5 13. Bf7 Bg2 14. Bg8 Bf1 15. Bf7 Be2 16. Bg8 Bh5 17. Bh7+ Kxf4 On December 05, 1998 at 09:19:08, Soren Riis wrote: > >>On December 02, 1998 at 11:35:12, Robert Ericsson wrote: >> >>>This a missing post that seems to have disappeared since yesterday... >>> >>>/Robert Ericsson >>>--------------------------------------- >>>I had the following position in a correspondence chess game some three >>>years ago. >>> >>>5b1r/3bkp2/3ppp2/p6P/4PP2/1PNB4/1KP5/7R b - - >>> >>>White's last move was 1. Kb1-b2 (1. f5 is better) and after 1. - f5! >>>2. exf5 Bg7 3. fxe6 fxe6 4. Be2 Rc8 5. Rh3 Rxc3! 6. Rxc3 Bd4! >>>black won easily within 10-15 more moves. >>> >>>I have tested this position on both Hiarcs 6 and Rebel 9 and they seem to >>>understand nothing about this position. They don't find 1. - f5 because >>>they obviously don't see the deadly pin long the diagonal. And even after >>>6. - Bd4 it will take a long time for them to realize that white is actually >>>lost. >>> >>>My questions are: when (at which PLY) do chess engines realize that: >>>a) 1. Kb2-b2 is bad due to the reply 1. - f5! >>>b) white is lost after 6. - Bd4 >>> >>>Robert Ericsson >>>---------------------------------------- >> >>Whoaw, this is one of the most computer hostile positions I have seen for >>a long time!! Are there any programs (including weaker experimental programs >>which might have been designed to make this kind of positional sacrifises) which >>can solve the above problem? >> >>Soren Riis > >I now realize most chess programes eventually will find 1-f5! but of course not >because they see the exchange sac 5-Rxc3!!!!! I would be DEEPLY IMPRESSED >by any computer who can find the 5-Rxc3 idea. Such a move is (to quote Kasparov >on his comments on moves like Be4 in game two against deepblue) a sign of real >inteligence. For us humans it is a very easy move, but for computers it >seems to be very difficult because it involves reasoning on a kind of >metalevel. Even a rather weak human chess player will probably reason >like this: "White can never get out of the pin and the black king can walk >freely around on the dark squares and eventyally after having picked up all >white pawns black can bring his king to d2 (by help of his bishop) and then >eventually pick up the rook on c3". > >Does any of you have any ideas how to impliment procedures which can find >a move like 5-Rxc3? Of course such a procedure might lower the programs >general playing strength but I think it is important to try to write >programs which are good at detecting "unusual" situations where an "unusual" >(for a computer!) approach is required. > >Soren Riis
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.