On the elimination of numerical Čerenkov radiation in PIC simulations (Q703723)

Particle-in-cell (PIC) simulations are a useful tool in modeling plasma in physical devices. The Yee finite difference time domain (FDTD) method is commonly used in PIC simulations to model the electromagnetic fields. However, in the Yee FDTD method, poorly resolved waves at frequencies near the cut-off frequency of the grid travel more slowly than the physical speed of light. These slowly traveling, poorely resolved waves are not a problem in many simulations because the physics of interest occurs at much lower frequencies. However, when high energy particles are present, the particles may travel faster than the numerical speed of their own radiation, leading to non-physical, numerical Čerenkov radiation. Due to nonlinear interaction between the particles and the fields, the numerical Čerenkov radiation couples into the frequency band of physical interest and corrupts the PIC simulation. There are two methods of mitigating the effects of the numerical Čerenkov radiation. The computational stencil to approximate the curl operator can be altered to improve the high frequency physics, or a filtering scheme can be introduced to attenuate the waves that cause the numerical Čerenkov radiation. In examining different computational stencils for approximating the curl operator, it is found that these stencils are larger than the standard Yee computational stencil. Thus, boundary conditions and particle current weighting are extremely difficult to implement while maintaining a charge conservation numerical scheme. Therefore, filtering is more commonly applied. But using filtering schemes, care must be taken not to alter the physics of interest in the numerical simulation. Examining the effect of filters on the wave dispersion equation, it is shown that the ideal filter is characterized by an all pass magnitude response with zero phase in the frequency band of physical interest and increasing phase as the slowly traveling high frequencies near the grid cut off point are approached. Further, in the present paper, two previously published filters by Godfrey and Friedman are analyzed and compared to ideally desired filter properties.