Zwischenzugzwang wrote on 11/19/11 at 09:21:17:
A little oddity:
Chessbase shows (in "Kavalek in Huffington: Joys of Chess: From Krabbé to Hesse") the following problem:
(Leonid Yarosh - White mates in 4 moves, Shakhmaty v SSSR, 1983 [GM Lubomir Kavalek/Huffington Post])
The point is, that, after 1.a7!, whichever piece Black promotes his pawn to, White mates within the given number of moves by promoting to the same piece.
Of course, every chess engine (I tested Fritz 11, Hiarcs 13, Houdini 1.5, Stockfish 2.0.1) solves the problem in a fracture of a second. Oddly enough, only Rybka (all versions I have) has a problem with the variation 1...axb1B: It doesn't find the solution 2.axb8B. Does Rybka have a blind spot and doesn't know the underpromotion to a bishop?
Best regards,
Zwischenzugzwang
It's well-known that Rybka does not iterate bishop underpromotions in its search. It will allow them as legal moves if played against it, but it will not search them in its tree.
Despite a fair bit of flak from the community, apparently Rajlich's ultra-pragmatic design choice reportedly gained 2 or 3 Elo points from the reduction.
From a consumer viewpoint, the gain is negligible, and there's something not very aesthetic about an engine being entirely unable to solve a problem
even if it searched until the end of time just because it didn't fully support the rules of chess. Nevertheless, it is true that bishop underpromotions are sufficiently rare that one could say they pretty much only exist in contrived positions.
At any rate, it is what it is.