What Is AutoPot?

AutoPot is a collection of workflows with the aim to generate machine-learning interatomic potentials (MLIPs) in an automated fashion.

One type of algorithm that automates generating MLIPs is an active learning algorithm. AutoPot presently implements a number of such algorithms using the BlackDynamite (BD) framework.

BD has been designed to manage a large number, for example more than 10000, of interconnected simulations in a massively parallelized fashion. This allows simulations from which newly selected configurations are extracted for training, and simulations that generate new training data, to run efficiently in parallel.

BlackDynamite consists of two components:

BlackDynamiteCore

A Python-based package for managing parametric studies, for example running many molecular dynamics simulations in parallel with different parameters such as temperature or atomic configuration.

Motoko

A workflow management system based on BD studies.