I think the engine's program is written economically in a way that can be described as "following the principles".. In this special case (promoting to bishop) the engine is told to consider promoting to B (instead of to queen) only if the opponent is entering into a stalemate right after a direct queen promotion. And if not a stalemate right after the promotion then a new-promoted queen is considered to be better than a new-promoted bishop, which in most of the cases (maybe 99.9 %) is true. A problem arises in this particular case when (in 0.01% maybe) a special position like this one is arranged. Anyway, I still can't figure how the strength of the engine is affected by this arrangement of the engine's "thinking process".. Maybe these + ELO points are gained because of the fact that this particular engine is "thinking" faster in some positions than the other engines which are not modified by this rule.. Who knows? Only the engine programmers maybe.. 

As I recall there are one or two technical endings wbere poromotion to bishop figures in. But here Rybka no doubt relies on tablebases (so that she does indeed "recognize" this possibility). I suspect the cost of refusing to calculate bishop promotion other than in technical endings is very slight.

Zwischenzugzwang wrote on 11/19/11 at 09:21:17:


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.

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

Markovich wrote on 11/03/11 at 16:48:54:

Nope, white wins no matter what. The commentators on the game as I said are believing their chess engine -- which was the point of this thread.

And look at this one! Aronian - Volokitin:


After the game Aronian makes the remark: Bxh6! with excellent play for white.
Just try to give this to Houdini!.. 
Never trust your engine!.. Trust your mind!.. 

The same thing, but little exaggerated: