For a while now I've been playing with the idea of defining chess pieces in strictly mathematical terms. Originally this was meant as a more flexible/descriptive alternative for Betza Funny Notation, perhaps as a way of specifying pieces for chess-variant-playing AIs, but there is also the possibility of defining operations and functions over them, and maybe even proving some things algebraically.

My early attempts were to define pieces as sets of possible paths, where paths are sequences of steps, and steps are vectors; notation would use the Kleene star for unlimited equal steps. So a wazir, for example, would be {(0,1),(1,0),(0,-1),(-1,0)}, and a rook would be {(0,1)(0,1)*,(1,0)(1,0)*,(0,-1)(0,-1)*,(-1,0)(-1,0)*}. Since most pieces are symmetrical it would make sense to parameterize this as a kind of shorthand, but since some pieces aren't symmetrical it shouldn't be assumed (in the way that "a 1,2 leaper" usually implies the (1,-2), (2,1), etc. leaps). This works for the usual steppers, leapers, and riders, and even bent riders; it's basically how people usually think of chess moves, just in more formal terms.

But this starts to get awkward where pawns are involved (and a way of describing chess pieces that has trouble with pawns is seriously flawed!). First off, we need to separate passive moves from captures. That's not too bad; we can even do so by introducing a "capture step" distinct from a passive step, which must be to a square occupied by an opposing piece, and this lets us do fun things like define the Chu Shogi Lion and the Locust. Then there's the initial double move, and promotions. We can introduce a new kind of "step" that is a promotion to another piece type to deal with these: the initial double-step can be implemented by defining a "starting pawn" that has it, where every move ends with a "promotion" to a "basic pawn" piece that doesn't. The promotion step is also strange, because it always happens at the end, but if you define operations that derive pieces from other pieces (like a "double move" operation that derives a Hook Mover from a Rook) in terms of sequence concatenation, you can end up with a promotion in the middle, or even more than one. So we've unfortunately introduced a distinction between "well-formed" and "ill-formed" pieces, and a need for some sort of normalization.

Introducing hoppers also complicates things. We have to introduce another step, one that requires a hurdle like the capture step but doesn't capture it. But now we've introduced the possibility of moves that "collapse": after all, what is the difference between a (1,2) knight move, and a pair of a regular mao-move and a mao-move over an obligatory hurdle?  Or between the latter and a pair of a regular moa-move and a moa-move over an obligatory hurdle? Equivalence is a mess.

I never found a satisfying way of dealing with en passant, or castling. Defining a piece in terms of the path is travels doesn't lend itself to a move where two pieces are repositioned.

Finally I realized that this approach fundamentally allowed for pathological alternate definitions of pieces. For example, take the rook. One or more equal orthogonal steps, right? Now let's define an "inside-out rook": a piece that leaps orthogonally directly to the square right next to its final destination, then slides by orthogonal steps until it reaches its starting square, then finally leaps directly to its final destination. This piece behaves exactly like the basic rook. It's entirely equivalent: any square it can reach, a standard rook can also reach under the same conditions, and vice versa. Yet it is considered a different piece. While the example is obviously contrived, similarly weird things could be produced by operations that concatenate or interpolate paths.