Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Only one move wins test suite

Author: Vincent Diepeveen

Date: 18:58:19 04/19/01

Go up one level in this thread


On April 19, 2001 at 04:52:09, David Dahlem wrote:

>I would like to put together a tough test suite of positions where there is only
>one move that will win. Any positions with the solution or suggestions on where
>i might find such positions would be greatly appreciated.
>
>Dave

Ah a cool try. the problem is that to make such a testset that
you can't already use too well known material as that's already
inside testsets :)

But i'll give you 2 tactical positions for free

Diepeveen - H.H.Hage  correspondence game
2rr2k1/pp3pp1/4b3/2qNp1b1/4P3/1B1Q4/PPP5/1K3R1R w - - Qf3!!

Forgot what year i played this. Some progs have real problems with this one.
If they play it usually compu's play it even for wrong reason. Takes
real HUGE search depths to get score for diep +2.0 for white or so here.

Mainline and how game went: Qf3!! Rc6 Qh5 Bh6 Qxe5 Bxd5
Bxd5! Qxc2+ Ka1 Rc7 a3! b5 Rxh6! gxh6 Qf6 1-0

Reason why black plays the same moves as the compu's think it plays
is easy. H.H.Hage clearly played this game with a computer... ...i
assumed chessmachine schroeder or something as that could reproduce
all his moves this game! Note that in those days i figured out
myself easily that Qf6 won. Only later programs started showing me
that Rg1 also wins for white. When i played the game the only program
finding that Rg1 won was The King! All other programs didn't find it.
Diep in those days was only an interface, not a chessplaying program
yet :)

Another cool position but probably easier for some programs and
nearly impossible for others is the next position:

Analysis of Diepeveen + De Haan
r1b1k2r/p2n1ppp/1p2p3/3p2B1/P2P4/1Nn1P3/5PPP/R3KB1R w KQkq - f3!!
f3 wins a piece. It's not just 0.01 better or so. the piece at c3
is completely hung. It's not a real fair position as it's an anti-nullmove
position actually. Progs not nullmoving or extending threats will
find this quick and others will need huge depths because of nullmove.

Try this one at tiger or rebel... ...they might also suffer
from forward pruning here...

Another cool one which is very simple for some programs as their
evaluation is optimistic about doing it, others dislike it and
will never play it, as the real tactical trick is so deep that
all progs miss it. So this one is no good
in computer testsets but it's a cool combi simply:

4qrk1/3nppb1/R1Np2p1/3P2P1/1Pr5/4B3/5Q1P/5R1K w - - 4 30 Ra8!!
Benjamin - Xu Jun 2001
Alexandre Come brought this move into my attention. Took
me some long analysis to realize how bigtime Ra8 wins here...

For computers to consider they solved it they should get +4.0 or
more for Ra8.
All other scores are in the positional range as white is already
cool passer up.

I doubt any program will show soon Ra8 with over +4 as the trick is
real deep and it's a matening trick again, so programs which do not
do mating extensions will perhaps never get Ra8, not even at 20 ply
if their positional evaluation says it's not ok to play.









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.