When and why should particle redistribute() with nGrow be used?

[spring-png]

In the example AmrCoreAdv or AmrLevelAdv, the redistribute() is called posttimestep, and with nGrow=iteration for the refined mesh. When the particle is in the grown cell, how can the source be transferred to the liquid phase correctly?

[WeiqunZhang]

I am not exactly sure what you mean by "the source be transferred to the liquid phase correctly". I suppose it can be handled by having enough ghost cells in the mesh data. In fact, even if particles are in valid cells, you may still need to have ghost cells if the "transfer" is not limit to the cell that contains that particle.

[atmyers]

The nGrow argument is used any time you want to leave particles where they are if they are within some number of cells from their current box. It is mostly used to support subcycling in time, which is the case in the tutorials you mention. For that Redistribute() call, if the particles on the fine level were all inside their boxes at the beginning of the coarse time step, then they can be at most iteration cells away after iteration fine level steps. You don't want to transfer those particles to the coarse level until all the fine steps have been completed, because they haven't reached the proper time yet. If you have some particle-mesh interpolation you want to do, then those Multifabs need to have enough ghost cells to cover the support of particles up to nGrow cells outside the box.