Endgame Explorations 11: Castling
Every chessplayer knows that castling is the only legal way to move two pieces at once, and the only legal occasion for the king to leap over a square. In practical play this unique move usually becomes the basis for a stereotyped defensive formation around the king, but occasionally it has a specific tactical significance, as in the following memorable game (Feuer--O'Kelly, Liege 1951):
1 e4 e5 2 Nf3 Nc6 3 Bb5 a6 4 Ba4 d6 5 Bxc6+ bxc6 6 d4 f6 7 Nc3 Rb8 8 Be3! Rxb2(?)
9 dxe5 fxe5 10 Nxe5! fxe5? 11 Qxd8+ Kxd8 12 0-0-0+!
The king leaps to attack the Rb2, while at the same time the Rd1 gives check, so White picks up the rook and emerges the Exchange ahead, winning.
Endgame composers, predisposed towards the unusual and offbeat in chess, naturally like to feature castling, as in the following setting by Selesniev of the Feuer-O'Kelly theme:
The solution is simple, especially given the theme:
1 d7 Kc7 2 d8/Q+! Kxd8 3 0-0-0+
And White wins. But note the tries 1 Rd1? Rh2! (equivalent is 1 0-0-0? Rh2!) 2 d7 (else Rh8(a8)) 2 ... Rh1(a1)+ 3 Ke2(c2) R:d1 4 K:d1 Kc7. Also insufficient is 1 Rc1 Rb5, as well as 1 d7 Kc7 2 0-0-0? (still too early!) 2 ... Rb8.
The alert reader will ask: how do we know that 3 0-0-0+ was legal: given the initial position of Selesniev's study, couldn't the White king or rook have moved in the past, invalidating the castling? Well, yes, they could have, and indeed it is not possible to construct a position where castling can be proved legal without further information (for all we know, White could have opened with 1 Nc3, 2 Rb1, 3 Ra1, 4 Nb1, ...). To avail themselves of the use of castling, composers follow the convention that in composed endgames (and in problems too), castling is legal unless it can be proven illegal by retrograde analysis. (For instance, if Black has only Ke8 and Rh8, one of them must have made the previous move, so Black cannot castle. By the way, for the other move whose legality depends on the history of the position, namely the en passant capture, the opposite convention holds: on the first move, a pawn can be captured en passant only if it can be proven to have just moved two squares. But en passant captures and retrograde analysis do not concern us here.)
For an example of Kingside castling we turn to a study by G. Nadareishvili (7th Prize, Magyar Sakkelet, 1980):
Black's extra tripled pawns do not constitute a decisive advantage, but White is about to lose a piece (Rh1 and Bc2 are both attacked), and so must look for stalemate. The try 1 0-0+!? Kxc2 2 Rf8! Rxf8?, stalemate, fails to 2 ... Rh7! (3 Rf7 Bxf7!), so White first lures the Bd5 off the a2-g8 diagonal:
1 Be4! Bxe4 2 0-0+! Kd2 3 Rf8! Rh7 4 Rf7!
Now this works, because the threatened 5 Rxh7 Bxh7 would bring about a book draw-- the White king can be neither ejected from the corner nor mated in it). Black has one more try:
4 ... Rh6 5 Rf6! Bg6!
Exploiting the b1-h7 diagonal, but White laughs last:
Threatening a new stalemate by 7 Rxg6! Rxg6. There remains only 6 ... Be4+ 7 Kg1, and White draws by repetition of position or stalemate.
The castling in this Nadareishvili endgame was incidental to the main theme of an unusual draw by repetition. In our last example (for which I won first prize in the 1987 Israel "Ring" tourney) castling is again the main attraction:
1 g7 g2
Or 1 ... Rb8 2 Nf8 g2 and now not 3 0-0-0!? Rd8! 4 Re1 Re8! with repetition, since after 5 Rg1!? Kxg1 6 g8/Q Re3 White can hope for no more than a draw, but 3 Kd2! Rb2+ 4 Kc3! (directed against ... Rb3+ and ... Rg3) and wins after 4 ... Rf2 5 g8/Q Rf1 6 Ra2.
We know already that 2 0-0-0? fails to 2 ... Rb8 3 Nf8 Rd8! Now White anticipates 2 ... g1/Q+ 3 Qxg1+ Kxg1 4 0-0-0+! winning; but Black has ...
2 ... Rc2!!
Not 2 ... Rb8?! (trying to get the rook out of harm's way with tempo) because 3 Qxb8 is check, which is why 2 g8/R was not sufficient. But with 2 ... Rc2!! Black stops White from castling, and White can do nothing to halt the g-pawn (3 Ra2 Rxa2 4 Qxa2 Kh1, or 3 Qb8+ Kh1, and the Nh7 gets in the way of 5 Qh8+ while the c6 pawn prevents a diagonal pin with Qa8(d5); 3 Kd1 Rf2 is also useless). And yet...
3 Nf6! g1/Q+
4 Qxg1+ Kxg1 5 Ng4!
Black's king and rook are caught in a unique domination. White threatens 6 Ne3!, nabbing the rook after 6 ... Rb2 7 0-0-0+, 6 ... Rc3 7 Kd2+, or 6 ... Rf2(h2) 7 0-0-0+ Kh2(f2) 8 Ng4+, and Black has no good defense. 5 ... Rb2 or 5 ... Rc3 are again met with 6 0-0-0+ or 6 Kd2+; 5 ... Kg2 6 Ne3+ or 5 ... Rc4 6 Ke2+ Kg2 7 Ne3+ forks king and rook. Finally, 5 ... Rg2 (or 5 ... Kh1 6 Ne3 Rh2) runs into the thematic 6 0-0-0, checkmate!
Noam Elkies is now Professor of Mathematics at Harvard University and is the author of "Chess Art in the Computer Age," published in ACJ 2 (1993). This article originally appeared in Chess Horizons.
This was the last published installment of Endgame Explorations.
This page last modified on
28 April 2018.
Copyright (c) 1995-2018 Christopher F. Chabris. All rights