SODAS

Toolkit for the characterization of atomistic phase trajectories

• sodas allows for the conversion of atomic graphs (in the form of the graph + line graph method known as ALIGNN) to a spatio-temperally resolved latent space. Useful for understanding structural trnaisitons during atomistic simulations. The projection scheme allows for the spatial and temporal characterization of structure during a transition (otherwise known as a reaction coordinate). sodas can tell you how similar structures are to one another, as well as quantify their evolution through time by labelling each structure during a transition based on how far it is located in the latent space from know end points. Note, each latent space projection scheme you choose will vary, ex: PCA may give different results than UMAP.

Installation

The following dependencies need to be installed before installing sodas. The installation time is typically within 10 minutes on a normal local machine.

• PyTorch (pytorch>=1.8.1)
• PyTorch-Geometric (pyg>=2.0.1): for implementing graph representations
• Networkx (networkx>=2.8.6)
• Scipy (scipy>=1.9.0)
• Numpy (numpy>=1.21.1)
• Atomic Simulation Environment (ase>= 3.22.1): for reading/writing atomic structures

To install sodas, clone this repo and run:

pip install -e /path/to/the/repo

The -e option signifies an editable install, which is well suited for development; this allows you to edit the source code without having to re-install.

To uninstall:

pip uninstall sodas

How to use

sodas is intended to be a plug-and-play framework where you provide data in the form as an ase atoms object and sodas++ does the rest. You have full control over the ALIGNN and the data projections through the sodas class.

• The src folder contains the source code.
• The example folder contains an example for how to use SODAS++ to characterize an Al melt molecular dynamics simulation.

Citing SODAS

Please use the following citiation to cite the SODAS toolkit: Bamidele Aroboto, Shaohua Chen, Tim Hsu, Brandon C. Wood, Yang Jiao, James Chapman; Universal and interpretable classification of atomistic structural transitions via unsupervised graph learning. Appl. Phys. Lett. 28 August 2023; 123 (9): 094103.

Or cite directly from the manuscript at: https://pubs.aip.org/aip/apl/article/123/9/094103/2909293/Universal-and-interpretable-classification-of