Conventions in Retroanalysis

A Plea for Consistency and Completeness
by Alexander George, 2002

Retrograde analysis, or retroanalysis for short, consists in the application of logic to determine past play on the basis of information about a given board position and the rules of chess. Composers have shown much ingenuity in thinking up questions that  can be answered only on the basis of information about past play (e.g., “Which Bishop is promoted?,” “What was the Rook’s path?,” “Where is the Black King standing?”) and in arranging the pieces so as to force their answers. Even the general problemist, who is less keen on composing or solving retroanalytic chess problems, must take an interest in them: for a longstanding constraint on chess problems of all kinds is that their initial position be a legal one, that is, one obtainable by a sequence of legal moves from the initial game position. It is striking that despite this, and despite more than a century of intensive retroanalytic work, there is still unclarity and disagreement concerning basic principles. It is the aim of this essay to indicate some sources of discontent, to assess various responses, and, finally, to advance and defend a view that, while not new, is deserving of a more precise articulation and a broader support than it has hitherto received.

Retroanalysis in chess is enriched by the existence of moves whose legality depends on features of the history of the game position. There are only two such moves in chess: en passant (e.p.) capture and castling. They are in a certain respect dual to one another: e.p. capture is legal only if a certain move has been made, while castling is legal only if certain moves have not been made.

Because e.p. capture requires the existence of a particular previous move (by the captured pawn), it is natural to adopt as a convention that it shall be permitted when and only when it can be established that the required move has taken place. And indeed this is the convention that has been universally adopted.

It is tempting to extend this convention to castling and declare that it shall be permitted when and only when it can be established that the designated moves (by the king or the relevant rook) have not taken place. This would have the effect, however, of ruling out all castling, since it is impossible to prove that such moves have never taken place. The search is thus encouraged for a convention that makes reference to the existence of certain moves, rather than to their non-existence. And this leads, naturally enough, to the common convention for castling (CCC), which declares it legal unless it can be proved not to be, that is, unless it can be established that one of the designated moves has occurred. We might put this by saying that, while the convention regarding e.p. capture places the burden of proof on those who claim it is legal, the convention for castling places it on those who hold it to be illegal.

The CCC is widely assumed in the construction of retroanalytical chess problems. It was even officially adopted in 1958 at the International Congress of Problemists at Piran, Yugoslavia. What has come to be known as the Piran Codex asserts that: “CASTLING is always regarded as legal whenever its illegality cannot be proved.”

So stated, however, the convention leads to strange, even paradoxical, consequences: ...


Read the full article (PDF)