Comparison of macroscopic particle‐in‐cell and semikinetic models of the polar wind

Abstract

The outflow of the polar wind along diverging geomagnetic field lines has been the subject of many modeling studies for the past 25 years. As the plasma drifts up and out of the topside ionosphere, it undergoes several transitions; for instance, its velocity changes from subsonic to supersonic and its velocity distribution changes from Maxwellian to non‐Maxwellian. The complexity of the flow led to the development of several modeling approaches, such as the generalized moment, the kinetic, and the semikinetic models. Recently, a “macroscopic” particle‐in‐cell (PIC) model was adopted to study the polar wind. However, because one is always restricted to a finite number of particles, the validity of the approach must be established when it is applied to macroscopic flows. In this study the polar wind predictions obtained from a macroscopic PIC simulation were compared to those obtained from the more rigorous semikinetic model for steady state conditions. The study also shows the sensitivity of the PIC simulation to the adopted modeling parameters for both time‐dependent and steady state conditions, including the number of simulation particles, the time step, the spatial bin size, etc. The study indicates that (1) the PIC model can be a powerful simulation tool if special attention is given to its potential pitfalls; (2) because of the finite number of particles, the PIC technique is subject to a considerable amount of noise; (3) the noise level is higher for the higher‐order moments, such as heat flow, and for the velocity distribution function; (4) the use of a time step that is too large leads to a modulation of the results; (5) an insufficient number of spatial bins yields a poor spatial resolution, while too many spatial bins leads to more noise; (6) the noise level can be reduced by averaging over time and/or space, but this affects the spatial and/or temporal resolution; (7) the bin size in velocity space must be carefully chosen to balance numerical noise and velocity space resolution; (8) in the steady state the PIC technique can achieve the same accuracy as the semikinetic model if all of the PIC modeling parameters are optimized; and (9) a few hundred thousand simulation particles, as used in some previous studies, are not adequate to resolve the tail of the velocity distribution, even in the steady state when time averaging is possible. For 106 particles the noise in the tail is still appreciable; and (10) when poor spatial resolution is used, important features can be missed, as was the case in some previous studies.