Author: Bruce Moreland
Date: 09:34:41 08/31/01
Go up one level in this thread
On August 30, 2001 at 10:38:40, Ed Schröder wrote: >On August 30, 2001 at 07:45:34, Kim Roper Jensen wrote: > >>Hi >> >>If my memory(heavily fragmented) serves me well there was some games, where the >>DB played against DT, and in one of them DB "announced" mate in 18, a very >>complicated position where the mate consisted of quiet moves and some >>counterchecks. >> >>Are there anyone that have this position ?? Or is my memory corrupted ??? :) >> >>With regards > > >[d]3r1r1k/1b4pp/ppn1p3/4Pp1R/Pn5P/3P4/4QP2/1qB1NKR1 w - - id MATE (16) > >I think you mean this one. > >Ed > >------------------------------------- > >00:00:04 12.00 0.00 1.Rxh7+ Kxh7 2.Qh5 Kg8 3.Rxg7 Kxg7 > 4.Bh6 Kh8 5.Bg5 Kg7 6.Qh6 Kg8 7.Qg6 > Kh8 8.Qh6 Kg8 9.Qg6 Kh8 10.Qh6 (3) > >00:00:09 13.00 17.26 1.Rxh7+ Kxh7 2.Qh5 Kg8 3.Rxg7 Kxg7 > 4.Bh6 Kh8 5.Bg5 Kg8 6.Qg6 Kh8 7.Qh5 > Kg8 8.Qg6 Kh8 (9) > >00:00:10 14.00 17.26 1.Rxh7+ Kxh7 2.Qh5 Kg8 3.Rxg7 Kxg7 > 4.Bh6 Kh8 5.Bg5 Kg8 6.Qg6 Kh8 7.Bf6 > Rxf6 8.exf6 Qxe1 9.Kxe1 Nc2 10.Kf1 > Ne3 11.fxe3 Rd7 12.Qe8 Kh7 (10) > >00:00:12 15.00 Mate in 16 moves 1.Rxh7+ Kxh7 2.Qh5 Kg8 3.Rxg7 Kxg7 > 4.Bh6 Kh8 5.Bg5 Kg8 6.Qg6 Kh8 7.Bf6 > Rxf6 8.exf6 Qxe1 9.Kxe1 Nc2 10.Kf1 > Ne3 11.fxe3 Rd7 12.Qe8 Kh7 (12) Mainly these comments aren't for Ed, who knows this stuff. This problem is solved with an extension known as "single response to check". Extending moves that check is something that people have done for many years, but if that is the only extension you do here, you don't find this, because you have <check> <response> <check> <response> etc., and while the <check> moves are free, each <response> eats up a ply of search depth. The solution in this case is to also extend the responses. You can't do this in normal cases because your search will blow up, but if there is only one response to check, you can find a certain category of mate very quickly. If you look at the position, you'll see that this is one of them. The checks mostly have one legal reply. This is also why DT found it quickly. A single legal move is going to trigger a singular extension. This kind of thing has been known for years. In "Inside Chess" circa 1995, there were some ads purporting to show how wonderful some commercial chess software was because it could find mates like this in a trivial amount of time. If you want to try to compare software based upon performance in tests like this, there is something you should consider. It's very cute to find mates like this, but it may not be the smartest thing to do in terms of performance. If you find a sequence like this, there is often some garbage time at the end, where the losing side throws away pieces in order to stave off mate. Why is it necessary to continue extending once you've determined that you've put the other side into such a state? It's possible to resolve the mate in a very short time, if you do the single response thing all the way out. It's also possible to find this very quickly with a big plus score, without necessarily resolving the mate. This is a design decision, not a matter of one program being better than another. bruce
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.