What question did this study set out to answer?

The aim is to report on the design and implementation of scalable workflows for materials science.

April 23, 2026Open Access

First report on exascale workflow infrastructure and demonstrators. Deliverable D2.4 of the HORIZON-EUROHPC-JU-2021-COE-01 project MaX (101093374)

Key Points

The aim is to report on the design and implementation of scalable workflows for materials science.
Developed workflows within the MAX Centre of Excellence focusing on materials science applications.
Demonstrated high-throughput simulations, automated neural network training, and electronic structure calculations.
Collaborated with supercomputing facilities to validate efficiency and stability of the workflows.
Implemented workflows enabled seamless catalyst screening and materials simulations.
Validated neural network interatomic potential workflows on high-performance computing resources.
Supported accurate predictive simulations through first-principles electronic structure corrections.

Abstract

This deliverable reports the current status of the design and implementation of exascaleready computational workflows developed within Work Package 2 (WP2), “Exascale workflows and extreme data for materials” of the MAX Centre of Excellence. These workflows demonstrate scalable, modular, and fully reproducible automation pipelines tailored for materials science applications. The report highlights several workflow demonstrators that leverage MaX codes such as Quantum ESPRESSO and Yambo, and related simulation tools.One demonstrator focuses on high-throughput simulations for catalyst screening, employing QUANTUM ESPRESSO and AiiDA’s WorkGraph extension to study the dehydrogenation of formic acid on metal surfaces. This use case integrates surface relaxation,adsorption energy evaluation, and reaction pathway analysis via NEB calculations into a single, traceable workflow.A second demonstrator addresses the growing role of machine learning in atomistic simulations by introducing AiiDA-TrainsPot, a newly developed workflow for the automated training of neural network interatomic potentials (NNIPs). Combining DFT, molecular dynamics, and active learning, AiiDA-TrainsPot has been validated through a collaboration with CINECA by a hero run on the full Leonardo supercomputer Booster partition, demonstrating its efficiency and stability.The third major demonstrator is AiiDA-hubbard, a workflow for computing Hubbard +U +V parameters from first principles using DFPT. Applied to over 100 Li-based materials, this workflow enables accurate electronic structure corrections in transitionmetal compounds, supporting predictive simulations without empirical fitting. In addition, we present a snapshot of the implementation work on other substantial workflows for non-adiabatic molecular dynamics, exciton-phonon coupling, and photo-catalysis.By combining automation, provenance tracking, and interoperability, these workflows mark a significant step toward scalable, data-centric computational science and demonstrate the scientific and industrial value of MAX software in the exascale computing landscape.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper