PerQueue: managing complex and dynamic workflows.

Area of Science:

• Computational Materials Science
• Workflow Management Systems
• Scientific Computing

Background:

• Existing workflow managers for computational materials discovery lack dynamic capabilities.
• Efficient planning and execution of complex workloads are crucial in materials science.

Purpose of the Study:

• Introduce PerQueue, a novel workflow manager designed for dynamic capabilities in computational materials discovery.
• To provide a better overview, flexibility, and high dynamism in defining and executing workflows.

Main Methods:

• Utilizing modular and dynamic building blocks to explicitly define workflows.
• Demonstrating PerQueue's application through four diverse use cases in computational materials discovery.
• Implementing high-throughput screening with Density Functional Theory (DFT).

Main Results:

• PerQueue successfully managed complex workflows, including active learning for Machine-Learning Interatomic Potentials (MLIPs) with Molecular Dynamics (MD).
• Demonstrated reuse of MLIPs for kinetic Monte Carlo (kMC) simulations.
• Showcased active-learning-accelerated image segmentation with human-in-the-loop integration.

Conclusions:

• PerQueue offers a flexible and dynamic solution for computational materials discovery workflows.
• The modular approach of PerQueue enhances the management of complex scientific computing tasks.
• PerQueue facilitates advanced applications from DFT screening to active learning and image analysis.