Author: Dieter Buerssner
Date: 12:02:35 10/01/01
Go up one level in this thread
On September 30, 2001 at 19:18:53, Miguel A. Ballicora wrote:
>Here is a much tougher test from a real game
>(A. Sokolov - Korchnoi Tilburg 1987)
>Easy endgame? not quite!
>
>A. Sokolov miscalculated and made a mistake, played Rxc5? which ends up losing
>easily. Rb8 puts up a much tougher resistance with good chances of drawing.
>
>[D]8/2p3pp/8/1RrkP3/8/3K4/6PP/8 w - - am Rxc5;
>
>Not even Crafty gets this right (15 min K6-II 400mhz) Probably
>a problem evaluating the rook endgame because it sees a clear
>disadvantage in the pawn endgame.
Is it really so easy? Unfortunately not for me :-( I tried a few lines fast
in Analyze mode. I ended in drawn position not only once.
And - from a engine point of view. It is not too difficult, to give
the farer away passed pawn (or candidate, while the canditate is
probably much more difficult) a very high score. I am
however not convinced, that this only helps. I did quite some
experiments with mixed results. I think often, it will be a matter of
one tempo, and this seems to be very difficult to evaluate by me.
Many artifacts can arise, by scoring too high. Assume a won pawn endgame
with a passer at Q-side (the winning passed pawn). If you give this a
very high value, it may happen, that you try to defend the pawn,
instead of walking to the K-side and win there. In positions in the
search tree, you will see, that you lost one valuable pawn, but have
a much better K-position. So, it is not so easy to guarantee progression.
Especially, when I take into account, that my engine does not reach
very high search depths in pawn endgames with several unblocked pawns.
Back to the position you posted:
I tried at 40 moves in 2 hours against Crafty.
First, I used an old batch file by an oversight, and gave Crafty also only
half of the hash of Yace.
[Event "Computer chess game"]
[Site "?"]
[Date "2001.10.01"]
[Round "1"]
[White "Yace 0.99.56"]
[Black "wcrafty-17.10"]
[Result "1/2-1/2"]
[TimeControl "40/7200"]
[FEN "8/2p3pp/8/1RrkP3/8/3K4/6PP/8 w - - 0 1"]
[SetUp "1"]
{--------------
. . . . . . . .
. . p . . . p p
. . . . . . . .
. R r k P . . .
. . . . . . . .
. . . K . . . .
. . . . . . P P
. . . . . . . .
white to play
--------------}
1. Rxc5+ Kxc5 2. Ke4 h5 3. h4 c6 4. g3 g6 5. g4 hxg4 6. Kf4 Kd5 7. Kxg4
Kxe5 8. Kg5
{adj., TB position} 1/2-1/2
I repeated with Crafty 17.14 and enabled the PGN of Yace with scores:
[Event "Computer chess game 40 7200 +0 0 0"]
[Site "?"]
[Date "2001.10.01"]
[Round "1"]
[White "Yace 0.99.56 MLbook"]
[Black "crafty"]
[Result "1/2-1/2"]
[SetUp "1"]
[FEN "8/2p3pp/8/1RrkP3/8/3K4/6PP/8 w - -"]
[WhiteElo "0"]
[BlackElo "0"]
[Timecontrol "40 7200 +0 0 0"]
[Time "Mon Oct 01 20:05:13 2001"]
[LogFile "logs\yace0562.log"]
1. Rxc5+ {-0.30/15 217s} Kxc5 {181s} 2. Ke4 {-0.24/17 215s} Kc6 {181s} 3. Kd4
{-0.29/15 161s} Kd7 {182s} 4. Kd5 {-0.42/16 190s} g5 {189s} 5. e6+
{0.00/16 118s} Ke7 {182s} 6. Kc6 {0.00/16 107s} Kxe6 {182s} 7. Kxc7
{0.00/19 139s} Kf5 {181s} 8. Kd6 {0.00/62 80s} 1/2-1/2 {adj.}
I think, 4...c6+ would have won. (Crafty score for g5 was around 1 for black).
This all was on a rather slow AMD K6-2 475, but at least it indicates,
that it is not trivally won for a computer.
With switched colors I get:
[Event "Computer chess game 40 7200 +0 0 0"]
[Site "?"]
[Date "2001.10.01"]
[Round "1"]
[White "crafty"]
[Black "Yace 0.99.56"]
[Result "0-1"]
[SetUp "1"]
[FEN "8/2p3pp/8/1RrkP3/8/3K4/6PP/8 w - -"]
[WhiteElo "0"]
[BlackElo "0"]
[Timecontrol "40 7200 +0 0 0"]
[Time "Mon Oct 01 16:18:16 2001"]
[LogFile "logs\yace0558.log"]
1. Rxc5+ {0s} Kxc5 {0.23/18 324s} 2. Ke4 {181s} Kc6 {0.23/16 317s} 3. h3
{183s} Kd7 {1.18/16 185s} 4. Kd5 {183s} c6+ {1.30/15 241s} 5. Ke4 {209s} Ke6
{1.60/16 191s} 6. Kf4 {341s} h6 {1.70/15 129s} 7. g3 {538s} h5 {2.18/16 120s}
8. Ke4 {207s} c5 {6.85/16 172s} 9. Kf4 {176s} c4 {11.62/13 201s} 10. Ke4
{970s} c3 {11.74/13 143s} 11. Kd4 {740s} c2 {13.14/11 172s} 0-1 {adj.}
Regards,
Dieter
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.