Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Exploiting databases in GK-v-WT

Author: guy haworth

Date: 14:24:16 09/18/99

Go up one level in this thread


OK - I think we are in agreement though perhaps my notation was not clear
enough.

KQQKQQ, because of its symmetry and 2Qs on each side, involves less positions
than K**K** by a factor of 8 - 1,546,346,340 to be precise.  Do NOT, as Robert
Hyatt said, hold your breath for this endgame.

KQQKQQ also involves no pawns which require the 'solver' to consider P=R, P=B
and P=N as well as P=Q for a 'full on' solution.  There ARE wins to be had only
through underpromotion.  That figure includes unreachable positions (both sides
in check, double check - not possible in KQQKQQ) but the Ks are apart.

Some more figures from Lewis Stiller.  The 1st player wins 82.81% of these
positions;  95.5% are wins in 1 or 2 moves (Q en prise or skewer).  99.3% of
these wins are in 6 moves (11 ply) or less.  I guess a chess program will be
able to full-width search 11 ply even in KQQKQQ.  The EG-db becomes useful only
by augmenting the search-tree-vision of the program.

An analysis of KQP(g7)KQP(xn) where 'x' = b OR d and 'n' =6-2 (for 'd') and 7-2
for 'b' would be useful but is harder.  How one builds the simplifications P=Q,
no captures by the pawns, and 'bK doesn't move too far from square X' into a
retrograde analysis program is not clear to me.  Maybe that's the next step
forward in EG-db generation;  generation with heuristics and simplifying
assumptions - Wirth/Nievergelt have proposed as much.

I have asked W/N for an electronic copy of their paper.

Meanwhile, the GK-v-WT BBS is beginning to think that ending 'G' is at least as
good as ending 'D';  this has the wP on h7 rather than on g5.  Looks worse
because the pawn is 2 further on but the battle is really only joined on the 7th
rank anyway.  The 'h' file is worse for White than the 'g' file.  So maybe it's
the KQP(b7)... EG-db the World Team needs rather than the one above.



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.