A study of how to compute electronic stopping power quickly using a combination of machine learning and Time-Dependent Density Functional Theory (TD-DFT).
See our paper for more details.
A version of this project with the datasets and outputs used when writing our paper is on the Materials Data Facility
The environment for this project is defined in environment.yml.
Install it using Conda:
conda env create --file environment.yml --forceThis will produce an environment named td_dft you must activate it
following the instructions given by conda after installation and
launch Jupyter from within this environment for the notebooks to function.
This project is broken in to several subfolders.
datasets contains all of the TD-DFT data associated with this project.
It is not tracked by git, so get the data from our two
datasets on the Materials Data Facility.
stopping_power_ml is a Python module that contains utility operations for this project.
Generally, these are methods that are used in more than one notebook.
single-velocity contains notebooks related to predicting the stopping power using only data
relating to a single projectile velocity. We explore whether these models can be used
to determine whether ML can be used to halt a stopping power calculation early, and
whether our model can predict stopping power in different directions than
what was included in the training set.
multiple-velocities contains notebooks for testing whether our models
can predict stopping powers in different directions and velocities.
You will notice that the name for each notebook starts with a number. To run the notebooks, execute them in the order indicated by this number because the output of some notebooks are used as inputs into the following notebooks.
A word of warning: the two notebooks in the root directory 0_parse_qbox and
1_generate_representation take a significant amount of computing time to complete.