Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: BS-2830 #second part

Author: Vincent Diepeveen

Date: 16:11:51 12/24/00

Go up one level in this thread


On December 23, 2000 at 18:47:09, Joshua Lee wrote:

>I have narrowed it down to positions 5,8,11,22,and 23. These aren't solved by
>gandalf and others. Any clue as to which programs solve these?

Position 11:exf6 i do not see how it wins personally. the move all engines
play looks way better to me, note they plan exf6 a move later.

Position 22 is solved by diep after 9 minutes or so single cpu.
dual faster, sometimes it needs a ply more too.

Position 23 Rxf3 is a big piece sacraficing fun activity. extremely
deep line to find, but not impossible as it's all checks and captures.
If a prog finds it it must be real lucky with hashtable or extend
all checks and captures and have some more extensions. No big deal
to put it in a testset IMHO.

In fact i find very little positions in BS2830 real attractive. A prog
scoring better on it you can't call even tactical better as other
engines, as it is not representing normal positions. BS2830 represents
positions which are very anti-nature and anti-nullmove, though from
human side seen all pos are real cool tricks.

Best example of what is not very good testmaterial for the
real computerchessprogrammer being position 15:Kh2-g3
Short-Timman (wasn't it played in Linares tournament if i remember well?)

If you do not nullmove then this is a real short combination.
[D]2r2rk1/1bpR1p2/1pq1pQp1/p3P2p/P1PR3P/5N2/2P2PPK/8 w - -

if you use recursive nullmove then to find this position you need to see:

Kg3
nullmove
Kf4
nullmove
Kg5
nullmove
kh6
random black move
then mate in qsearch.

So it's 8 ply + 3 nullmoves.

With R=3 that's 8 + 3*3 = 17 ply
With R=2 that's 8 + 3*2 = 14 ply
With adaptive nullmove it's 16 ply (but i say that from head),
of course in all example cases i assume checks getting done in
qsearch, something nearly all commercial programs do (correct
me if i'm wrong, which commercial program is NOT doing checks
in qsearch?)

If you have mate thread extensions you might arguably
find it after kh6 already so as mate is threatening
then it's 13 ply.

Any prog claiming to use intensively recursive nullmove with R=2 or 3
and finding this trick sooner as 13 ply is tuned to find
this position.

There are no exceptions possible to this like in contradiction
to a lot of tricks transpositiontable is not going to find this
trick at a smaller depth for you: no way a transposition gets to
a position with king at h6 and black queen on root position, for
a pretty simple reason: if it would be possible to do so then
you could have nullmoved with black which gives the above line!

Vincent




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.