I am thinking about the question of the title..
Suppose we know $Av = \lambda v$ and all eigenvalues of $A$ are different (so diagonalizable). What do the eigenvectors and eigenvalues look like of $A^TP-PA$?
I know that $A^TP - PA$ is a linear map.
Obviously, we can let $A = V\Lambda V^{-1}$, where $\Lambda$ is diagonal. So we have $$(V\Lambda V^{-1})^TP-P(V\Lambda V^{-1}) = V^{-T}\Lambda V^TP - PV\Lambda V^{-1}$$
then how do I go further? I suppose to get a form like $M\tilde{\Lambda}M^{-1}$ and I can say something. How to get that point?