In Prolog, I would like to implement a trivial logical induction predicate i as (1):
i(0).
i(N) :- i(N-1).
or (2):
i(0).
i(N+1) :- i(N).
But this doesn't work; with (1), query i(3). causes a stack overflow; with (2), i(3). returns false.
Instead, I have to do (3):
i(0).
i(N) :- M is N-1, i(M).
Naively, the difference between (3) and (1) seems like a trivial syntactic shift, and declaring intermediate variables feels like a hassle in bigger programs. What's the reason behind the Prolog compiler rejecting (1) and (2), does a work-around exist, and why do (1) and (2) behave differently?
Surprisingly, it isn't a trivial syntactic shift. N-1 in Python is always arithmetic; you can replace 5-1 with 4 and "cat"-1 is an error.
In Prolog dash is not arithmetic, it's a term connecting Left/Right with a dash. You can write "cat"-1 ("cat dash one") to pair two things up and pass them around together, and people do:
?- "cat"-1 = N-1. % unify against it,
N="cat" % reason that N must be "cat" for this to be true.
?- N-1 = -(N,1) % rewrite it in normal name(X,Y) format,
true % where dash is the name.
% make a list of pairs joined with dash:
?- pairs_keys_values(Ps, [cat,dog,cow,fish], [1,2,chicken,N-1])
Ps = [cat-1, dog-2, cow-chicken, fish-(N-1)]
So people wanted a way to wedge math into a logical relation language, and came up with a predicate which knows how to do calculation on terms with the names + - / * and so on. A predicate like math_evaluate(Result, Term) and if you had it you could write it like Result math_evaluate N-1 so it looks neater. That math_evaluate predicate is is(Result, Term):
?- N=5, is(Result, N-1)
N = 5,
Result = 4
which you can write Result is N-1 in a similar way that you can rewrite -(N,1) as N - 1.