Discrepancy between fields driven in first and subsequent timesteps for `from_file` injection

[jameschappell]

Hi,

I've noticed that the fields driven by a particle bunch injected via the from_file option differs significantly in its first timestep compared to subsequent timesteps. The same behaviour is not true for the fixed_weight_pdf or fixed_weight injection options and I attach plots below to illustrate this behaviour. In the attached minimal example input decks I purposely make dt very small to minimise evolution of the bunch between timesteps, but have seen the same behaviour for much larger timesteps (both fixed and adaptive).

Thanks in advance for your help. Please let me know if there is any other information you need from me.

• I am using HiPACE++ (v24.03-23-gcb94a319796c) running in double precision with AMReX (24.05-2-gee11254ffc7c)
• HiPACE++ is run on Nvidia A100s, compiled using CUDA v11.8.89
• I am not expecting the results of from_file and fixed_weight_pdf/fixed_weight to agree with each other as the input beams are slightly different, only that the fields driven by the from_file injected bunch should be consistent for all timesteps.

input_files.zip

from_file:

fixed_weight_pdf:

fixed_weight:

[AlexanderSinn]

Thanks for reporting this and providing the input scripts. It looks like when initializing a beam from a file, the MR level tag was not initialized, causing some beam particles to not deposit their current. The MR level tag is re-initialized before each subsequent time step, which worked. I added proper initialization in #1117 which fixed the issue in my testing.

[jameschappell]

Thanks for the speedy fix! Can confirm this has fixed it in my testing too.