Renormalization of wave function during MCWF evolution

[dehond]

I noticed that when I do a Monte Carlo Wave Function evolution with classical variables (using semiclassical.mcwf_dynamic(T, psi0, fquantum, fclassical, fjump_classical)) the wave function that gets passed into the classical DE's as specified by fclassical(t, psi, u, du) is not normalized. This influences the expectation values I calculate to update du; namely, they're too small. I imagine this happens because of the non-Hermitian nature of the problem.

My workaround is to divide every expectation value I calculate within the code by dagger(psi) * psi, so as to enforce normalization. This gives me the right result when I for instance compare my calculation of Doppler cooling to those one can find in the literature.

For me, this is fine, but I imagine it might affect some examples that are included in the documentation (this one, for instance).

[david-pl]

Thanks for the report! This is actually a mistake in the example of Doppler cooling. I will fix that! During a time evolution with the MCWF method, the state is never normalized since the decreasing norm is a criterion for when a jump occurs. So you always need to normalize when computing expectation values in dynamic functions. I'm not sure if there would be a better way to do this. I also noticed that the documentation does a poor job in pointing this out, so I will improve this as well.

[dehond]

I see, thanks for acknowledging this! I think it would indeed be a good idea to make this more explicit in the documentation and in the examples.

[david-pl]

On second thought, it might be a better idea to internally compute a normalized version of psi at each time step, which then gets passed to any dynamical function. This could then be turned off with a kwarg, but the default should be true.
Otherwise this can be quite an annoying trap for most users...

[dehond]

Indeed I think that would be helpful, but perhaps that's more invasive than simply updating the documentation and some of the examples.

[david-pl]

I haven't found a good way to do this. It's not possible to know when (or if) the user needs a normalized state. Normalizing regardless of this hurts performance. So documenting it will have to do. Thanks again for pointing it out, though!