Nearest-Grid-Point Interpolation in Gyrokinetic Particle-in-Cell Simulation

Linear interpolation or other higher order interpolation schemes are almost always used in particle-in-cell simulation because of their lower noise characteristics and accuracy relative to the nearest-grid-point method. The higher order interpolation schemes are chosen because of their optimal performance, balancing a smaller number of particles against more computer operations per particle per time step. However, this is not the case in large-scale gyrokinetic particle simulations, where sometimes nearest-grid-point interpolation is used with results virtually identical to those of linear interpolation using the same number of particles. Here, a comparison and analysis of nearest-grid-point and linear interpolation schemes is given showing why nearest-grid-point interpolation can be optimal. A simple two-dimensional slab model of the ion-temperature-gradient instability and associated generation of zonal flows is used to compare the interpolation schemes. Significant improvements in performance are possible by using nearest-grid-point interpolation because of the large number of gathers/scatters associated with gyroaveraging.