Machine Learning to infer improved closure relationships for the Ornstein-Zernike equation
The following collections of code were used to first generate then analyse data on the structures of simple liquids.
The data was generated by performing forward molecular dynamics simulations for specified potentials. From this the radial distribution function and then the direct correlation function were extracted.
This data was then used as an input for the inverse problem of solving to find the interaction potential given structural infomation about the system.
Various machine learning algorithms were applied in order to do this.
- Install comprhys/espresso.git
- Run tables.py to generate the tables
- Run system.py to build the input file
- Use input="
awk "NR==<row>" <input_file>" to grab a row from the input file - Run bulk.py to get the bulk properties
That work is availiable on arxiv here: Coarse-graining and designing liquids with the Ornstein-Zernike equation and machine learnt closures
An early abstract for this work was published as a workshop submission at the Neurips Machine Learning for Physical Sciences Workshop. Since this additional work has been carried out exploring the problem in more depth. The notebooks here have been updated to reflect the more recent work not as it stood at the time of the workshop submission.