We generate machine-learned interatomic potentials for comprehensive simulations of energy materials. The basis of our efforts is the Python-based open source software project libatoms/workflow which we are co-developing.
Workflow provides users with tools to create general-purpose routines with a focus on machine-learned interatomic potentials. Applications range from building procedures for their generation to conducting atomistic simulations. Workflow thereby stongly supports incorporating a high degree of automatization and efficiency while requiring limited user input. In that Workflow has the capability to call (and write) functions that parallelize operations over a set of atomic configurations in a user-friendly way. Moreover, computationally intense operations — density-function theory calculations for instance — can readily be outsourced to a high-performance computing infrastructure due to its compatibility with the ExPyRe package.
The Workflow project benefits from a highly active community consistenly extending and improving its capabilities. A comprehensive overview of all features currently available is provided via the online documentation which also comprises various educational examples allowing to easily explore its functionalities. Continuous integration with extensive testing ensures functioning of these features and examples while simultaneously further developing the Workflow project.