Particle number increases dramatically in DSMC

[ancientflow]

I conducted a simulation of gas being ejected at 1000 Pa into a vacuum. In the case, there is only elastic collision. I tried two gas intake method. The first one is to use Nfluxpercell provided by WarpX, and the second is to insert a piece of code which ensures the absolutely same particle weight. In the first case, the particle number increases dramatically. However, in the second case, the particle number increases gradually, which is basically consistent with the flow rate. I found that when injected by Nfluxpercell without collision, the average weight of all particles have a small change at different time steps. Is it the slight difference in particle weights that led to the splitting during the collision process?

[RemiLehe]

Hi @ancientflow

The fact that the number of particles increases dramatically when particles have different weights is expected because elastic collision splits particles (It uses a similar algorithm as in this paper: https://www.sciencedirect.com/science/article/pii/S0021999119302037). To remedy this, you can use particle merging, as implemented here: #4820

Regarding the fact that the average weight changes at different time steps: this is somewhat unexpected. Could you share a script with which we could reproduce this behavior on our side?

[ancientflow]

tube_fluid.txt
xe_elastic.txt
These are the script and cross section file that I used in the first situation. The result I obtained is that the average weight is fluctuates around 1.412e8. In addition, when I enabled the collision and particle merging at the same time, although the particle number was limited, the calculation time will be extremely long after a few times steps.