An issue with separation of variables is that with a nonlinear PDE, there is no way to construct the general solution by summing separated solutions. But if the goal is to find just one solution, then looking for a separated one is a good idea. If both sides in
$$\sqrt{g(r)r^{2} - r^{2} (\partial_{r} f)^{2} } = \partial_{\varphi} f$$
are constant, then the constant has to be zero because a nonzero constant value of $\partial_{\varphi} f$ would make $f$ multi-valued. From here $f$ can be found...
But it would be simpler to just assume from the outset that $f$ is radial (looking for a radial solution with radial data is natural). So $f=f(r)$, and the equation is simply $f'(r)=\sqrt{g(r)}$.
$$ \sqrt{g(r)r^{2} - r^{2} (\partial_{r} f)^{2} } = \partial_{\varphi} f. $$ So, the left and the right side is equal to some constant a.
– John Taylor Feb 24 '13 at 19:17