CHAOTIC BEHAVIOR IN PIC SIMULATION AND ITS RELATION TO COMPUTATIONAL ERRORS

For a parallelized one-dimensional electrostatic PIC simulation code, an increase of separation of test particle orbits has been observed be tween numerical results with a single node and multiple nodes under th e same initial and boundary conditions. The PIC simulation treats the many body Hamiltonian system which is characterized by positive Lyapun ov exponents. In order to check the accuracy of simulation results at microscopic level, we have evaluated the chaotic property of the dynam ical system realized in these two calculations. By measuring the maxim um Lyapunov exponent, it is shown that the separation of particle orbi ts is not caused by accumulation of numerical errors, but caused by th e origin of chaos that the solution is sensitive to initial conditions . Therefore, both results obtained with the single node and the multip le nodes are considered as the realization of an ensemble and equally reliable. We also discuss the relation between breakdown of energy con servation and stochasticity in the particle simulation.