Reproducibility is required in Monte Carlo programs. When I use OpenMC to solve a fixed-source problem, the results are reproducible if the weight window method is not used.However, when the weight window method is applied and multiple threads are used for computation, reproducibility is lost the results differ between two separate runs. Below is how i set the weight window:
After my test, I find that the particle tracks are the same in the first few batches and are different in the last few batches. For example, if 10 batches are set, the tracks may start to differ by the fifth batch, so the results are different.
Does anyone know the reason? I need the weight windows method to be stable. Many thanks for any pointers.