As for the time step, I don't know what's optimal. This is problem dependent. My first choice is to pick so many time steps as I want results. If that is too coarse for the solver, then the solver automatically adjust the time step, but you get faster convergence if you choose a time step in the first place, that is sufficiently small for the solver to converge. Then you save the time it takes for the solver to try those too big steps. Note also that some time steps may be easier for the solver than others; that depends on what happens in the model.
↧