Reduction of simulation cost and error for particle simulations of rarefied flows

Abstract The computational cost in simulating a steady, rarefied gas flow by use of a particle method is shown to be greatly reduced if the simulation employs an average number of particles per cell that is greater than a certain (approximate) minimum. The minimum value depends on the rms level of statistical fluctuations judged acceptable in the resultant data and is found to be significantly greater than values often used in practice when two- and three-dimensional flows are simulated. The computational cost is found to remain fixed for values greater than the minimum, showing that the ensemble- and time-averaging operations become interchangeable. For the problem studied, it is shown that a regime exists for which computational cost may be reduced by a factor of 10 by merely increasing the size of the simulation by a factor of five, when holding the rms level of statistical fluctuations fixed.