Force maxwell reinit seems to lead to suspicious output on restarted runs

[David Dickinson]

In testing a separate issue with a small NL cyclone example I noticed that the phi2 value written as the first output of a restarted job appears to give a large jump in the value before recovering to more reasonable values. It appears this is associated with the force_maxwell_reinit = T recalculation of the potentials during a restart. Below is a rough plot showing phi2 vs time from an original run (blue), a default restart (orange) and a restart with force_maxwell_reinit = F (green). Whilst orange appears to recover to green after the initial jump we can note that the actual values are different, even if fairly similar.

This sort of behaviour may also occur during a time step change as by default we handle this by writing the current state to a restart file and then going through the restart code to load this back in after the time step has been changed. Any jumps caused by this will typically be less visible than at true restarts as we do not typically write out the fields at every time step so may miss a discrete jump and decay.

[David Dickinson]

Repeating the initial run with/without force_maxwell_reinit active and nwrite = 1 I do not see any significant jumps in the potential during the time step change (which occurs pre-overshoot). This may suggest that there’s something inconsistent with our restart initialisation and force_maxwell_reinit.

[David Dickinson]

Here’s some data with nwrite=1 to show how the restart seems to impact a few timesteps after restart

Here we see how the issue appears in all ky but to varying degree, with the lower ky showing a larger jump

‌

[David Dickinson]

Interestingly the above results are with the code currently in PR #491 whilst the related code in PR #530 does not show a similar jump at the restart (with force_maxwell T and F differing at a 1e-10 level). It’s worth noting that the initial conditions in PR #530 aren’t currently consistent with the hybrid model.

[David Dickinson]

Simulations using kinetic electrons do not show this sensitivity to force_maxwell_reinit so it appears unique to the code in PR #491 with hybrid electrons.