According to Goldstein,
"We can define action-angle variables for [a separable Hamiltonian] system when the orbit equations for all of the $(q_i, p_i)$ pairs describe either closed orbits (libration or periodic functions of $q_i$)." (p. 457, 3rd Ed.)
I'm working through the following example: for a particle of mass $m$ in a generic central potential $V = kr^\alpha$, let us fix the total energy ($E$) and angular momentum ($L$) of the system. We would then have the following expression for the radial momentum: $$ p_r (r) = \sqrt{2m\bigg(E - kr^\alpha - \frac{L}{2mr^2}\bigg)}\ . $$ Is it correct then, that to determine the conditions under which action-angle variables for this system exist, I would have to set $p_r(r) = p_r(r + a)$, where $a$ would denote the period, and derive constraints on the constants $E, L, k$, and $\alpha$?