Castling bonus and Chess960

[Oliver Uwira]

Hello everybody,

my own experience when playing Chess960 was that castling too early can be very devastating, especially when the draw assigned the kings a square near a corner.

The reason is that your opponent could start an all out attack on your king while enjoying the safety of the corner given to its own king.

This is similar to castling to opposite wings in standard chess. But the trouble is now that when you start your counter attack, the enemy king can still hop away to the other corner when the air becomes too hot... and this may probably leave your forces shooting holes into the sky while the enemy attack is still firmly in place and you would have to spend time to regroup...

(I've experienced this at last year's FiNet Chess960 open in Mainz - this was the first time I played Chess960... and in the second round I was thrashed by a 1750 player in exactly the described scenario... )

Now I don't like the idea of a castling bonus any more because I don't want to distinguish in my evaluation whether Chess960 or standard chess is being played.

Has anybody contemplated this, too? I'm tending to leave it all to the king safety evaluation.

Viele Gr??e,
Oliver

[Tord Romstad]

Now I don't like the idea of a castling bonus any more because I don't want to distinguish in my evaluation whether Chess960 or standard chess is being played.

Has anybody contemplated this, too? I'm tending to leave it all to the king safety evaluation.

I haven't tried it, but I suspect leaving it to king safety eval alone is not good enough. The point of castling isn't just to bring the king to a safe square, but also to make it easier to activate the rook. Rooks are powerful pieces, but they also tend to be the most difficult to activate in the opening, especially rooks which are located outside a king on the back rank, with no open files except possibly on the other side of the king.

I agree that trying to find good rules for when to castle early is an interesting idea, but I think it will be rather difficult. In addition to king safety, I think you should consider the position of the opponent's king, the number of moves you need before you can castle in the opposite direction, and whether the rook on the short side of your king can be easily activated without castling. Tuning it all to work well will be a real challenge.

Tord

[Reinhard Scharnagl]

SMIRF has an evaluation without any castling bonus. Thus a castling is not played simply because of getting a bonus but instead to reach a better evaluated position. There are indeed situations in Chess960 (or also in 10x8 Capablanca Random Chess), where to castle would not be optimal. Maybe in traditional Chess a castling bonus would not do too much harm, nevertheless it is a basically silly approach to handle that problem.

Regards, Reinhard.

[Tord Romstad]

SMIRF has an evaluation without any castling bonus. Thus a castling is not played simply because of getting a bonus but instead to reach a better evaluated position.

Hardly anyone uses a castling bonus (in the literal sense of the word), I think, although some programs (including mine) consider the castling rights in the eval. What some people do is to give the king a small bonus for being close to the corner instead of on the central files. I don't do this directly, but I have other eval terms which tend to have a similar effect. My space and mobility eval makes the program push central pawns in the opening, and the king then tends to move towards a corner in order to get a better pawn shelter. Usually the program will prefer to castle rather than use normal king moves torwards the corner, because otherwise the rook outside the king would be difficult to activate.

There are indeed situations in Chess960 (or also in 10x8 Capablanca Random Chess), where to castle would not be optimal. Maybe in traditional Chess a castling bonus would not do too much harm, nevertheless it is a basically silly approach to handle that problem.

I don't have any experience with 10x8 chess, but I don't think normal chess is very different from FRC in this respect. Castling at the wrong moment is no less dangerous in chess, and there are many examples where it is a grave mistake (like several sharp Sicilian lines, and classical anti-computer Stonewall setups, just to give two very different examples).

Tord

[Oliver Uwira]

Hi Tord,

I've seen your Deep Sjeng - Glaurung analysis. This demonstrated EXACTLY what I had in mind. Glaurung shoved his pawn forward for a strong attack while having the emergency exit of castling queenside at hand in case white would have managed to create some threats on the exposed black king.

The way to go for the castling boni (without distiguishing #518 and the other 959 positions) might be to not give boni for castling but penalties for the king in an airy environment. An program whose evaluation is tuned for #518 appears to inevitably castle within the first 4 moves in positions with king on f1 and rook on g1...

Viele Gr??e,
Oliver

PS: I reached my train last Friday so I didn't have to sleep on a bench...