Ooookay, my question is: Why is the gradient operator different in non-orthonormal coordinate systems?
According to wikipedia (as well as Wolfram Alpha), the gradient in, say cylindrical coordinates is
grad(f)=(df/dr)*e_r + (1/r)*(df/d(phi))*e_(phi) + (df/dz)*e_z
Where there is a 1/r factor (which, as far as I remember is one of the non-vanishing Christoffel-symbols in cylindrical, but I dunno if that is a coincidence or not) that is not present in cartesian.
However, assuming an arbitrary inner product space, the gradient of a scalar function f, is basically the covariant derivative of f (resulting in a (0;1) tensor field).
But, the covariant derivative of a scalar field is the same as its partial derivative, as for scalars, there is no need for a metric connection, so no extra factors would need to appear anywhere.
I assume this probably has something to do with the fact that a f(x,y,z) scalar field in {u,v,w} coordinates will become f(x(u,v,w), y(u,v,w), z(u,v,w)) with composite functions, however I have no idea how to formulate this properly in a general case.
If someone could deduce this for me, I'd be terribly appreciative.