[Archive] linear vs non-linear Poisson

Message by Henning Wulf:
is there any major disadvantage of the linear Poisson solver compared to the non-linear one, the latter one being the default setting?
I have the impression that in many cases the non-linear one fails to reach convergence, while the linear one seems to converge usually.
Is there specific common cases for which the linear Poisson solver would clearly lead to wrong results?
I have played with many parameters to try to have the non-linear Poisson solver converging, including csat, iterNewton, tolNewton, the initial satellite potential, and the particle dt’s, but up to now without success.
Any suggestion would be appreciated.
Thanks and best regards,

Message by Sébastien Hess:
Hi Henning,
Non-linear Poisson solver uses the analytical form of some electron distributions (e.g. Global Maxwell Boltzmann distribution) to implicit the plasma oscillation (so you do not need to solve the Debye length and plasma pulsation). It only works when the main electron population can be described as an analytical distribution
Linear Poisson solver simply solves the Poisson equation. If you only have analytical populations of electrons, this should gradually converge toward the solution that is reach in one step by the Non-linear solver. Note that in this case the time evolution is purely numerical,no physics inside. If there are physical variation at a time scale of the same order than the convergence time scale, it may not converge at all.
If you have only (or in a vast majority) PIC populations of electrons, you must use it and resolve the Debye length and plasma pulsation.
If you are in between (balanced mix of PIC and analytical), good luck.
What do you mean by “does not converge?” is it a real divergence or just this annoying warning message?
If you are simulating an electric thruster plume, note that you MUST NOT set the environment electron population to “GlobalMaxwellBoltzmannVolDistrib” use “UnlimitedGlobalMaxwellBoltzmannVolDistrib” instead.

Message by Henning Wulf:
Hi Sébastien,
many thanks for your reply.
I should probably say that I’m simulating using the standard WC GEO plasma characteristics, without active plasma sources currently. As such, I have a global MB electron distribution, with the secondaries defined as PIC (standard SPIS-GEO settings). I have tried with a maximum Newton iteration number of 1000 also, but without success.
I’m not sure how to distinguish between the real divergence and a false warning message. But I noticed also that the time step with the non-linear solver keeps at ~1E-6 seconds, which makes it also not feasible from the computing time point of view.
With kind regards,