@ Aleksei Bykov: Right, the equation is simplified by introducing x=cos(theta). However, the resulting equation for f does not seem solvable with an arbitrary function g(t, r).
I am mostly interested in the behavior of f at t >> \tau, where the equation can be reduced to the diffusion equation. However, a higher accuracy is desired.