Author: Christophe Theron
Date: 00:23:51 12/24/01
Go up one level in this thread
On December 24, 2001 at 00:23:18, Sune Larsson wrote:
>
> [D]8/k1p5/P7/8/2K4p/1p4rP/6P1/1R6 w - - 0 47
>
>
> This is a possible position from the first game Gambit Tiger - IM Berg.
> While playing out different endings vs Tiger, I stumbled upon this little
> intermezzo. Previous moves were: 45.-Nxc4 46.Kxc4 h4 /diagram.
>
> Tiger here ponders on and plays 47.Rxb3, with an eval of -0.78 - just to
> find itself in the misery of -11.36, a single move later...
>
> I played out this ending on a PIII 800, 192 Mb hash, 1 hour per player.
>
> Would have made a great testposition, but after the better 47.Kb5 black
> just have to find 47.-Rg8! 48.Kc6 Rb8! which also wins.
>
> For a human it's easy to see that a move like 47.Rxb3 loses quickly for
> white, but Tiger - (and maybe other programs) - has clear problems here.
The current version of Tiger (14.8) avoids Rxb3 after 19 seconds on my K6-2
450MHz.
However it avoids it in favour of another move without seeing that Rxb3 is
losing so badly.
I suspect other programs avoid it also for the wrong reasons. You need to
compute at least 20 plies after Rxb3 Rxb3 to see that it is losing, and it's
very difficult to go that deep in the start position.
Naturally, once Rxb3 Rxb3 has been played, most programs see that black is
winning in a few seconds. Because now there is less material and the "branching
factor" is much lower.
Christophe
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.