Supplementary Data: Learning collision operators from plasma phase space data using differentiable simulators

This repository contains some of the simulation data and post-processing tools for the paper "Learning collision operators from plasma phase space data using differentiable simulators". We provide the following files: simulations.zip: Dataset of 95 simulations produced, as detailed in Supplemental Material 2 of the paper. The zip file contains: database.csv: Table summarizing relevant simulation parameters used for each entry of the dataset. The row of the table matches the corresponding simulation index. data/: Folder containing all the simulation data. Each subfolder corresponds to a simulation (the name of the subfolder is the simulation index) and contains: AD/: Advection-Diffusion models computed from particle track data. FDIST/: Phase space evolution of subpopulations. Subfolder names correspond to the sampled PDFs (e.g., normal_0_0 is a normal centered at zero in v_x and v_y) as described in Supplemental Material 3 of the paper. energy.png: Plot with the evolution of the energy of the system over the simulation. os-stdin: OSIRIS input deck used to generate the data. This input deck should allow anyone to reproduce the simulation and is compatible with the open-source version of OSIRIS: https://osiris-code.github.io. Hardware-specific input parameters (e.g., node_conf) might need to be altered depending on the HPC system being used. run-info: File automatically generated by OSIRIS when the simulation was launched. Contains all the relevant information about the submitted input deck, git repo commit used, etc. utils.zip: Python scripts used to post-process raw particle track data. These were the tools used to generate the phase space evolution of subpopulations, compute advection-diffusion models from particle track information, and check for energy conservation. We do not provide any particle track data since the storage cost would be too high. This data can be regenerated by re-running the provided input decks in simulations.zip. The differentiable simulator framework developed to extract collision operators and perform the analysis presented in the paper is available at: https://github.com/diogodcarvalho/ml-pic-collision-operators