The relating code of the paper "SEDONA-GesaRaT: a Deep Learning Accelerated Radiative Transfer Program for 3-D Supernova Simulations".

This repository contains the relating python and C++ code of the paper "SEDONA-GesaRaT: a Deep Learning Accelerated Radiative Transfer Program for 3-D Supernova Simulations". In the repository, the four APNN neural networks, corresponding to the four LTE or NLTE atmoic physics solver recipes, are available as ".pt" file format, they are "CMF100_LTE.pt", "CMFAll_Ca_Nlte.pt", "CMFAll_Si_Ca_Nlte.pt", "CMFAll_Si_S_Ca_Nlte.pt". The python script to train the neural networks is "NLTEOpacEmisTrain.py", an example neural network training result is in the "NLTE_3El_Net" folder, note that the training result is in ".h5" file format, please use the python notebook "2_APNN.ipynb" to convert the ".h5" file into ".pt" file. Two examples of the SEDONA radiative transfer simulation, are available in the "HesmaDownload/" folder, only a few files are provided here to illustrate the data structure, the full data set including the SEDONA simulation results and SEDONA-GesaRaT validational simulation results, are about 100GB. Please contact the authors to access the full data set. The script of converting the HESMA downloaded supernova ejecta structure is in "1_HesmaToSedona.ipynb". There are two folders for the SEDONA-GesaRaT C++ program, the "sedonaGesaRaT/" is for the time-dependent 3-D simulations, the "sedonaGesaRaTIFU/" is upgraded to retrieve the IFU data cubes of supernovae. The installation processes should be broadly similar to the original SEDONA program, the major difference is the extra dependency of LibTorch (which enables the PyTorch neural network in C++).