If you are normalizing the uniform like this: phi_e/phi_e.sum(), you should use a Dirichlet distribution as prior instead.
phi_e/phi_e.sum()