Considering Schrödinger operators $$ H(\hbar) = \hbar \Delta + V $$ where $V$ is some potential, perturbation theory tells that the eigenvalues of $H(\hbar)$ are holomorphic on some region containing the positive real axis, but don't necessarily have an analytic continuation to the origin.
However, one tries to find asymptotic expansions in $\hbar$ in $0$, thus hopes that $$ \lambda(\hbar) \sim \lambda_0 + \hbar \lambda_1 + \hbar^2 \lambda_2 + \dots.$$ Usually one finds those expansions by WKB-methods, belonging to eigenfunctions that concentrate at the minimums of $V$.
Now, there are a lot of theorems about the behavior of this and statements that tell you that the expansions you get by WKB-methods actually belong to a "true" eigenvalue, but to me it seems that nobody ever tries to answer the question, if every eigenvalue of $H(\hbar)$ (considered as holomorphic function in $\hbar$ on some region $U$ with $0 \in \partial U$) actually admits an asymptotic expansion at $0$. Instead, it seems, that this is always automatically assumed.
But this is per se is not clear to me at all; it means for example that $0$ is not a branch point of $\lambda(\hbar)$.
Do you know about this question? I can't even have a guess if it is true that every eigenvalue admits an asymptotic expansion in the general case or if that is true only if $V$ is "well-behaved", like for example having no nondegenerate minima (and always assuming that it is such that the spectrum is discrete). Neither do I have an idea how one would prove something like that so far.