virp: VIRtual cell generation by Permutation

virp is a code for the fast generation of a virtual cell from a crystal structure (in CIF format) containing site disorder. It is named after Singapore's first superhero, VR Man, whose superpower is "Virping". The show was a flop, but we are still proud of him.

This project is inspired by the Supercell code of Okhotnikov, Charpentier and Cadars (J. Cheminform. 8, 17), which formed the basis of our fast virtual cell generation algorithm, as well as the aflow++ framework (Comput. Mater. Sci. 217, 111889), for the statistical postprocessing of materials properties.

Theory

First principles materials simulations and computational materials databases are adapted to represent ideal, ordered crystalline materials. However, to treat stoichiometric tuning in chemical compositions, defect-ordered compounds, metal alloys, or many naturally-occuring minerals, site disorder has to be considered. This has been variously treated with special quasirandom structures, cluster expansion, or statistical combination of simulation results from a library of quasirandom structures (here, we call them virtual cells). The configurational space of virtual cells and their simulated properties can be sampled with a relatively small sample set (proof incoming). The prediction of structural parameters and materials properties is expedited using machine-learned force fields and band gap prediction models (matgl).

Installation

Install github instance: pip install git+https://github.com/andypaulchen/virp.git
Update to latest release: uninstall and re-install
Install PyPI release: pip install virp

Building a database

The root directory has a folder (session) which holds the python scripts which build a library of virtual cells (generate.py) and postprocessing scripts (connectivity.py and properties.py). After each script is run, the results are saved as .csv files.

1. To prepare for a session, copy the session folder in your workspace and place the .cif files you want to process (make virtual cells + postprocessing) in the subfolder _disordered_cifs. Feel free to rename session folder to something more identifiable

2. Run generate.py to create a supercell and (by default) 400 virtual cells.

   • after this step, a structure subfolder (e.g. structure) is created in session for each structure.cif file in _disordered_cifs, with the same name. Inside this folder is a supercell CIF and folders for structure-optimized (stropt) and non-structure-optimized virtual cells (no_stropt). The details of this run is recorded in virp_session_summary.csv.

3. Run properties.py to predict materials properties. This is performed on stropt subfolders only.

   • after this step, the results are written to virtual_properties.csv in the structure subfolder.

In summary, this is what a session looks like after all three routines have completed:

The folder database_demo contains Postprocessing Demo.ipynb, which concatenates individual sessions into a database (database.csv), and conducts postprocessing routines.

Versions and changelog

v0.1.1: first workable code, with function to generate a virtual cell.
v0.2.1: added enumeration function
v0.2.2: enumeration can be imported now (fix)
v0.3.0: you can now make a batch of virtual cells
v0.4.3: added tools to build a database
v1.1.0: major user-friendliness updates, database compilation
v1.2.x: you can now store log files
v1.3.x: you can now choose the relaxer algorithm
v1.4.x: you can choose to not relax structures
v2.0.x: supercells are orthogonalised, treats correlated disorder (under testing)

Debugging and support

The virp code has been tested on a limited number of platforms, so far Windows and Linux. If you are running into any problems during operation, please hound me (Andy Paul Chen) at la.vache.qui.vit(at)gmail.com, and I will try my best to help.