A recent report details the extraordinary work carried out at the Idaho National Laboratory (INL) High Performance Computing (HPC) facility. In excess of one billion core hours, delivered in less than 12 months, provided by this leading supercomputer facility have driven radically improved understanding in nuclear materials, energy storage, and renewable energy.
A highlight of the report is the contribution of Jonathan Wormald, Richard Smith, Michael Zerkle, and Thomas Webb of the US Naval Nuclear Laboratory who studied the temperature dependent properties of zirconium hydride. This study employed a Machine Learned Potential (MLP) generated with the Materials Design automated MedeA MLPG (Machine Learning Potential Generator) using MedeA VASP calculations for a training set of hydride structures. The new MLP represents a dramatic breakthrough in simulation fidelity over prior generations of potentials. Understanding the properties of materials such as zirconium hydride, and their response to external factors, such as temperature and stress, are critical in designing, operating, and maintaining nuclear reactors. This study demonstrates, for the first time, that first-principles accuracy can be obtained for nuclear materials and processes using machine-learned potentials. Many congratulations to the team from the Naval Nuclear Laboratory.
The article is reproduced below.
A.3. Advanced Materials Simulation Engineering Tool
Jonathan Wormald1, Richard Smith1, Michael Zerkle1, Thomas Webb1
1 Naval Nuclear Laboratory
The Advanced Materials Simulation Engineering Tool (AMSET) project is a multi-institutional effort to develop a module in the MedeA simulation environment (Materials Design, Inc.) that generates machine learning interatomic potentials (MLPs) trained to ab initio calculations. A key part of this project is the validation of MLPs. Currently, a Zr-H MLP is being tested for the lattice expansion and the associated ε to δ phase transition of ZrH2 using the LAMMPS molecular dynamics software. Present results have demonstrated the novel ability to predict the phase transition in ZrH2 associated with lattice expansion using the newly developed Zr-H MLP.
Figure 1. Lattice Expansion of ZrH2 predicted with a Zr-H MLP and compared to experiment (E. Zuzek et al. 1990. Bulletin of Alloy Phase Diagrams, 111(4) 385-395). The tetragonal ε to cubic δ phase transition and associated behavior of the lattice is observed with a transition temperature of 725 K for the MLP, compared to experimental value of more than 900 K.
Atomistic simulations are the first stage of constructing multiscale models of materials behavior. Due to their predictive accuracy, ab initio calculations are the ideal technique to quantitatively inform multiscale materials models; however, computational restrictions limit the applicability of these calculations to systems too small to study materials degradation. Despite their success in modeling sufficiently large systems, conventional interatomic potentials are often limited in accuracy and lack transferability to a broad range of chemistries, stoichiometries and crystal structures needed to provide flexibility for studying complex phenomena. MLPs are a new form of interatomic potential that provides the quantitative accuracy of ab initio calculations at a length-scale sufficient for materials degradation but with the flexibility in chemistry, stoichiometry, and crystal structure lacking in conventional potentials. As a first step in AMSET, a Zr-H MLP was generated to study corrosion effects in Zr-based materials as well as the thermal neutron scattering law of Zirconium Hydride (ZrHx) used in TRIGA reactors. The initial validation of the MLP was the study of the lattice expansion of ZrH2 and the associated tetragonal ε to cubic δ phase transition. The validation of the lattice expansion and phase transition will help determine the fitness of the MLP for modeling the δ phase corrosion product in Zr-based materials and its ability to model the material behavior of TRIGA reactor fuel elements.
Materials Design Inc. 2020. “The Advanced Materials Simulation Engineering Tool (AMSET) Project: Ushering in a new age of atomic potentials with engineering accuracy,” https://www.materialsdesign.com/post/2020/05/06/the-advanced-materials-simulation-engineering-tool-amset-project
Nuclear Science and Technology
Footnote: The original article can be found online here: https://inldigitallibrary.inl.gov/sites/sti/sti/Sort_58555.pdf