Saturday, 14 February 2009

Computers aren't accurate

Or more correctly, there exists some computers who are inaccurate (nb this isn't a reference to measurement)
The study on the right was published way back in 1929 on the magazine "64". It is White to play and draw, and White does this by trapping the Black King on h8. The drawing line is pretty straight forward ie 1.g5+ Kh7 2.Bf7 c2 (otherwise Bb3 wins!) 3.Kh5! c1(Q) 4.g6+ Kh8 5.Kg4=
With king trapped on h8 Black cannot mate with a lone queen, therefore draw.
However, in the book "Modern Chess Analysis" by Robin Smith, there was a claim that Hiarcs had found a mate for Black, simply by creating a zugzwang with the White King on g5/h5. In his review of the book, Tim Harding (in Chessmail) said he couldn't repeat Hiarcs claim, and I haven't been able to either. Indeed I can't see why the White king needs to stay anywhere near g5 or h5 as hanging on to the f pawn doesn't look necessary and both Qxf7 or Qxg6 simply lose. If I had this position in a game I'd head for a1, just to make a point.
However if you plug this position into your computer it would probably find the right line (my copy of Fritz 9 does very quickly), but then still claim Black is winning (as he has a Queen after all). And for those who have delved a little into logic, any reasoner who believes something (in this case, Black is winning) which turns out to be false (as it is in this case), can be termed "inaccurate"


Mark Weeks said...

Re 'hanging on to the f pawn doesn't look necessary', White needs to keep the f-Pawn. If you lose the f-Pawn and head for a1, as you suggest, Black creates zugzwang with Qc2, forcing the Bishop to move. Then Black wins the Bishop with a fork.

Without the f-Pawn, Black can force the King into the corner, winning as I just described. The problem then becomes: can Black win the f-Pawn by force? It looks possible. - Mark

Shaun Press said...

Ahh, there is more subtlety to this study than I imagined. Maybe computers know more than I give them credit for!