Space charge grid sizing

[austin-hoover]

Hi ImpactX developers,

I'm comparing ImpactX to PyORBIT in some simple cases. For example, this simulation tracks a spherical Gaussian distribution through a drift with 3D space charge forces: https://github.com/austin-hoover/pyorbit-benchmarks/blob/main/free-expansion/impactx/run.py.

For PyORBIT space charge calculations, I'm using a grid of constant size ($N_x \times N_y \times N_z$), with limits set based on the min/max particle coordinates. The two codes basically agree in this case. However, I'm not sure how the grid is being defined in ImpactX. I know ImpactX can vary the grid size adaptively during the simulation; is this feature turned on by default? Is it possible to fix the grid size for testing?

Here are some settings which were adapted from the examples repo. If I understand blocking_factor correctly, this should be set equal to n_cell to fix the grid size? I'm not sure what dynamic_size does. (I get an error if I set that to False.)

sim = impactx.ImpactX()

sim.max_level = 1
sim.n_cell = [16, 16, 20]
sim.blocking_factor_x = [16]  # grid size must be divisible by this
sim.blocking_factor_y = [16]
sim.blocking_factor_z = [4]

sim.particle_shape = 2  # B-spline order
sim.space_charge = "3D"
sim.poisson_solver = "fft"  
sim.dynamic_size = True  # ?
sim.prob_relative = [1.05, 1.05, 1.05]  # mesh size relative to beam extent

Also, any references to help understand the adaptive grid sizing would be appreciated.

Thanks,
Austin

[cemitch99]

@austin-hoover Thanks for your question. A high-level description of parameters related to the grid size (and other space charge inputs) appears in the Usage > Parameters > Collective Effects section of the documentation. Several of these parameters are related to the multi-level multi-grid (MLMG) solver and the support for adaptive mesh refinement.

As a start, to avoid confusion, I recommend using the FFT solver, as you do above, and disabling adaptive mesh refinement by setting sim.max_level = 0 (or commenting this input line entirely, since this setting is the internal default). The number of grid cells in each dimension is then set by sim.n_cell = [N_x,N_y,N_z]. The physical dimensions of the mesh relative to the beam can be set using sim.prob_relative, as above. (The setting above means that the mesh is +-5% larger than the bunch extent in each direction.) You can comment the lines related to dynamic_size and blocking_factor to allow these to take default values. In this case, the number of grid cells will be fixed, and the dimensions of the grid will be set by the beam min/max coordinates. This should emulate the functionality of PyORBIT (for 3D space charge with open boundary conditions).

[austin-hoover]

@cemitch99 Thanks for the help!