Back
to the home page
The numerical properties of a particle-ion, fluid-electron computer simulation code, used in the study of the parallel-propagating electromagnetic Alfvén ion-cyclotron (AIC) instability, are examined. A numerical odd-even mode is suppressed by means of a two-timestep averaging method. Excellent energy conservation is obtained by using a method similar to the Boris particle mover to advance the transverse fields. Linear growth rates obtained from the code differ substantially from those predicted by uniform Vlasov theory, here dervied using a multifluid model. Short wavelengths in particular show substantial growth rates when damping is predicted, and additionally show strong linear mode coupling. Positive growth rates are even observed in the case of a Maxwellian ion distrbution. Disagreement is also generally found among short-wavelength mode frequencies. These inconsistencies are resolved by taking into consideration general grid and discrete-particle effects of the simulation model. A theoretical study reveals a real, physical process by which an ion distribution may collisionlessly relax via short-wavelength AIC instabilites acting resonantly on small portions of the distribution function. This process is combined with a linear mode coupling theory and other characteristics of the AIC instability to explain all observed differences. Nonlinear short-wavelength saturation levels are also obtained and their relevance to other field-aligned, electromagnetic simulations is discussed.
Particle-in-cell (PIC) simulations are useful in several areas of plasma and semiconductor device physics. The algorithms tested here have, for example, been used in the study of the propagation of electromagnetic waves in the Earth's magnetosphere. A simulation of this type, run on a vector machine such as the Cray, spends nearly all of its time weighting particle quantities to the grid or vice versa (gather-scatter algorithms). The speed with which parallel machines can execute gather-scatter tasks is therefore of significant interest. These tasks may be partitioned in two ways for parallel processing. The first method is to partition the grid so that each processor takes care of only particles within the subgrid assigned to it. The number of particles within a PE can vary significantly from one time step to the next. A second method, then, is to partition particles evenly so that each PE always keeps track of the same set of particles. The first approach would require less memory since only a portion of the grid is stored in a processor. However, from the viewpoint of load balancing, the second scheme would be preferred since the same amount of computation is carried out by all processors. The performance of the second approach was compared with the first on the iPSC/2 hypercube with promising results. With a uniform particle distribution, speedups relative to a single processor are found to be 14.5 with 16 processors and 25.5 with 32 processors. The performance with 32 basic processors and 8192 particles in the system is found to be 1/7th that of a Cray 1 or 1S and 1/15th that of a Cray X-MP or Cray 2 running with one processor.
Artificial velocity-space instabilities excited by the discreteness of particles are a nuisance in plasma particle simulations. The suppression of these instabilities is considered for the case of a uniformly magnetized electrostatic particle simulation using a combination of analytical and numerical techniques. It is found that, for a representative "rings-and-spokes" perpendicular velocity distribution modeling a Maxwellian, the instabilities are suppressed if the number of perpendicular velocity "rings" exceeds (1/3)(wp/wc)^2 and the number of gyrophase "spokes" exceeds 8k_{max}v_{th}/wc.
Basic plasma and fusion plasma abstracts.
Ionospheric, magnetospheric, and solar abstracts.
Computational biology abstracts.
Back to the publication list.
Back
to the home page