Algorithm Development in Materials Science and Engineering: Interatomic Potentials and Their Applications
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS: Computational Materials Science and Engineering Committee, TMS: Integrated Computational Materials Engineering Committee, TMS: Phase Transformations Committee, TMS: Solidification Committee
Program Organizers: Mohsen Asle Zaeem, Colorado School of Mines; Mikhail Mendelev, NASA ARC; Garritt Tucker, Colorado School of Mines; Ebrahim Asadi, University of Memphis; Bryan Wong, University of California, Riverside; Sam Reeve, Oak Ridge National Laboratory; Enrique Martinez Saez, Clemson University; Adrian Sabau, Oak Ridge National Laboratory
Tuesday 8:00 AM
March 1, 2022
Room: 253A
Location: Anaheim Convention Center
Session Chair: Mikhail Mendelev, KBR, Inc./NASA Ames Research Center; Garritt Tucker, Colorado School of Mines
8:00 AM Invited
Addressing Variability in Atomistic Predictions: Lucas Hale1; 1National Institute of Standards and Technology
A primary challenge associated with atomistic calculations is that the choice of interatomic potential can result in substantially, even qualitatively, different predictions for basic properties. Appropriately selecting a potential for a particular investigation requires knowing how all available potentials predict properties relevant to that use case. The NIST Interatomic Potentials Repository has worked towards addressing this by developing a high throughput framework allowing for calculations to be performed across hosted interatomic potentials and the results compiled into a database. The calculation framework also provides a means of changing this issue into a research asset as running simulations across different potentials allows for more possible mechanisms to be explored, and for relationships between properties to be observed. A particular focus of the project is working towards making the data, tools, and property calculation methods consistent with FAIR and open practices.
8:30 AM
NOW ON-DEMAND ONLY - A Generalization of the Universal Equation of States to Develop Magnetic Interatomic Potentials: Isaac Toda-Caraballo1; Jan Wróbel2; Duc Nguyen-Manh3; 1CENIM-CSIC; 2Warsaw University of Technology; 3Culham Centre for Fusion Energy, United Kingdom Atomic Energy Authority
The use of atomistic simulations is extremely helpful in analysing the thermodynamics in metallic systems. DFT-based methods provide a powerful tool to describe phase stability, enthalpy and/or magnetic configurations, but are constrained to small systems due to a large computational cost. On the other hand, Molecular Dynamics can simulate large systems, although the incorporation of magnetism into the formulation is still an open matter. In this work, we present a Generalized Universal Equation of States (GUESs) which enables the formulation of novel approach to develop new Magnetic Interatomic Potentials (MIP). The GUES has been tested in a large dataset of DFT-base energies in ferromagnetic and antiferromagnetic iron with outstanding accuracy. This paved the way to develop a MIP for ferromagnetic iron, which has shown good performance in predicting a wide range of crystal lattices, elastic constants, forces in the lattice, vacancies, interstitials, transformation paths as well as 𝛾-surfaces.
8:50 AM Invited
An Entropy-maximization Approach for the Generation of Training Sets for Machine-learned Potentials: Joshua Brown1; Mariia Karabin2; Danny Perez1; 1Los Alamos National Laboratory; 2Oak Ridge National Laboratory
The last few years have seen considerable advances in the development of machine-learned interatomic potentials. A very important aspect of the parameterization of transferable potentials is the generation of training sets that are sufficiently diverse, yet compact enough to be affordably characterized with high-fidelity reference methods. We formulate the generation of a training set as an optimization problem where the figure of merit is the entropy of the distribution of atom-wise descriptors. This can be used to create a fictitious potential to explicitly drive the generation of new configurations that maximally improve the diversity of the training set. I will show how this strategy can provide an automated and scalable solution to generate large training sets without human intervention.
9:20 AM Break
9:40 AM
NOW ON-DEMAND ONLY - Interatomic Potentials for Materials Science and Beyond; Advances in Machine Learned Spectral Neighborhood Analysis Potentials: Mitchell Wood1; Mary Alice Cusentino1; Ivan Oleynik2; Aidan Thompson1; 1Sandia National Laboratories; 2University of South Florida
With exascale super computers arriving in the near future, it is timely to ask whether our simulation software is capable of matching this unprecedented computing capability. While many research challenges in material physics, chemistry and biology lie just out of reach on peta-scale machines due to length and time restrictions inherent to Molecular Dynamics(MD), questions of the accuracy of our simulations will continue to linger. This is particularly true for complex alloys, composites of disparate components as well as materials in extremes of temperature, pressure and radiation exposure. This talk will overview advances made in machine learned Spectral Neighborhood Analysis Potential(SNAP) for both their physical accuracy and computational performance on leadership platforms. Exemplar problems include plasma facing materials, phase transitions of carbon and metals near their triple-point. Additionally, a discussion will be presented of best practices for assembling training data and model form selection for SNAP and related ML potentials.
10:00 AM
Refinements to the Production of Machine Learning Interatomic Potentials: Jared Stimac1; Jeremy Mason1; 1University of California, Davis
Machine learning potentials (MLP) have the potential to allow dramatically accelerated simulations of atomic systems with the accuracy of quantum mechanical techniques through the use of supervised regression algorithms. One of the related open questions is how to optimally construct the MLP's training set, since expanding the training set increases the computational cost of both MLP construction and potential energy or force evaluations. In pursuit of reducing these costs and alleviating the necessity for enormous training sets, our framework combines an efficient implementation of a sparse Gaussian process algorithm with a novel set of descriptors for atomic environments. These are specifically designed to help the sparse Gaussian process select as few inducing points—which dominate the computational complexity in all respects—as necessary. To this end, we aim to produce better performing potentials with less training and data than competing frameworks.