|
Description
|
TREX is a European Union (EU) Center of Excellence (CoE) in the field of high-accuracy quantum chemical and materials simulations. The CoE focuses on QMC approaches for solving the quantum many-body problem at the heart of atomistic physics, chemistry and materials science. Due to their inherent parallelizability and high computational cost, QMC approaches, and thus TREX, are uniquely positioned to fully exploit the massive parallelism of the upcoming exascale supercomputer architectures. TREX thus focuses on implementations of QMC approaches optimized for exascale HPC. Specifically, this includes the development and promotion of an open-source, high-performance software platform of interoperable flagship codes and exascale-ready libraries for QMC calculations.
This scope includes the development, validation, and application of ML methods to accelerate QMC calculations in Work Package (WP) 4. This was done primarily by constructing MLPs, which are data-driven surrogate models of QMC potential energy surfaces. Trained on a set of reference QMC calculations for a specific atomistic system, an MLP accurately approximates that systems’ potential energy surface at a fraction of the computational cost of QMC calculations. The resulting acceleration by multiple orders of magnitude greatly extends the reach of QMC approaches, enabling running more and longer simulations with larger unit cells, a task that will remain computationally unfeasible using QMC calculations alone for the foreseeable future. Other ways to integrate ML approaches with QMC calculations were explored as well, e.g., learning the Jastrow factor.
This Periodic Activity Report D4.4 “Report on release of transferable QMC-quality ML models” centers on the MLPs developed by TREX. It is based on task T4.3 on “workflows to machine-learn QMC accuracy” of work package 4, which includes the development, validation, and application of MLPs trained on QMC reference data generated by TREX codes. This report is also related to the deliverables D4.1 (Report on workflow implementation also in an ensemble mode), D5.2 (Report on ML results delivered for water systems.), and D5.4 (Datasets made available for benchmarking and ML modelling).
Visit the TREX website to learn more. (2024-07-16)
***This entry has been automatically imported via Datacite by LIST harvest scripts. Please refer to https://doi.org/10.5281/zenodo.12748832 for the original and latest version of the dataset and data downloads*** (2025-02-12)
|
