Learning in continuous action space for developing high dimensional potential energy models

Manna, Sukriti; Loeffler, Troy D.; Batra, Rohit; Banik, Suvo; Chan, Henry; Varughese, Bilvin; Sasikumar, Kiran; Sternberg, Michael; Peterka, Tom; Cherukara, Mathew J.; Gray, Stephen K.; Sumpter, Bobby G.; Sankaranarayanan, Subramanian K. R. S.

doi:10.1038/s41467-021-27849-6

Cited by 41 publications

(41 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…All the calculations were performed petascale supercomputer, Theta , at Argonne Leadership Computing Facility (ALCF). While performing high throughput DFT calculations can quickly become expensive, when such computing resources are not readily available, one can use cheap models like semi-empirical 91 or machine learnt classical force fields 33 , 42 which offers a reasonable compromise between computational cost and accuracy.…”

Section: Methodsmentioning

confidence: 99%

Machine learning the metastable phase diagram of covalently bonded carbon

et al. 2022

Self Cite

View full text Add to dashboard Cite

Conventional phase diagram generation involves experimentation to provide an initial estimate of the set of thermodynamically accessible phases and their boundaries, followed by use of phenomenological models to interpolate between the available experimental data points and extrapolate to experimentally inaccessible regions. Such an approach, combined with high throughput first-principles calculations and data-mining techniques, has led to exhaustive thermodynamic databases (e.g. compatible with the CALPHAD method), albeit focused on the reduced set of phases observed at distinct thermodynamic equilibria. In contrast, materials during their synthesis, operation, or processing, may not reach their thermodynamic equilibrium state but, instead, remain trapped in a local (metastable) free energy minimum, which may exhibit desirable properties. Here, we introduce an automated workflow that integrates first-principles physics and atomistic simulations with machine learning (ML), and high-performance computing to allow rapid exploration of the metastable phases to construct “metastable” phase diagrams for materials far-from-equilibrium. Using carbon as a prototypical system, we demonstrate automated metastable phase diagram construction to map hundreds of metastable states ranging from near equilibrium to far-from-equilibrium (400 meV/atom). We incorporate the free energy calculations into a neural-network-based learning of the equations of state that allows for efficient construction of metastable phase diagrams. We use the metastable phase diagram and identify domains of relative stability and synthesizability of metastable materials. High temperature high pressure experiments using a diamond anvil cell on graphite sample coupled with high-resolution transmission electron microscopy (HRTEM) confirm our metastable phase predictions. In particular, we identify the previously ambiguous structure of n-diamond as a cubic-analog of diaphite-like lonsdaelite phase.

show abstract

Section: Methodsmentioning

confidence: 99%

Machine learning the metastable phase diagram of covalently bonded carbon

et al. 2022

Self Cite

View full text Add to dashboard Cite

show abstract

“…In general, DL and its underlying technology, deep neural networks, are deemed suitable for tasks that require efficient detection of representative patterns in large data sets. In materials science, for instance, DL with decision tree algorithms has been used to generate high dimensional potential energy models for 54 elemental systems and alloys ( Manna et al, 2022 ). Algorithms learn models from large sets of annotated data (the “training data”), from which they can extract and learn the relation between the data patterns and the output classes.…”

Section: Limitations Of MLmentioning

confidence: 99%

Materials Discovery With Machine Learning and Knowledge Discovery

Oliveira¹,

Oliveira

2022

Front. Chem.

View full text Add to dashboard Cite

Machine learning and other artificial intelligence methods are gaining increasing prominence in chemistry and materials sciences, especially for materials design and discovery, and in data analysis of results generated by sensors and biosensors. In this paper, we present a perspective on this current use of machine learning, and discuss the prospects of the future impact of extending the use of machine learning to encompass knowledge discovery as an essential step towards a new paradigm of machine-generated knowledge. The reasons why results so far have been limited are given with a discussion of the limitations of machine learning in tasks requiring interpretation. Also discussed is the need to adapt the training of students and scientists in chemistry and materials sciences, to better explore the potential of artificial intelligence capabilities.

show abstract

“…Spin polarized calculations with the Brillioun zone sampled only at the Γ-point were performed. To handle errors that may arise during the structural relaxation or static DFT calculations, our high throughput workflow used an in-house python wrapper [68,69] around VASP along with a robust set of error handling tools.…”

Section: Dft Calculationsmentioning

confidence: 99%

Transfer and Active Learning of High Dimensional Neural Network Potentials for Transition Metal Clusters and Bulk

Sankaranarayanan

Varughese

Manna

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Classical molecular dynamics (MD) simulations represent a very popular and powerful tool for materials modeling and design. The predictive power of MD hinges on the ability of the interatomic potential to capture the underlying physics and chemistry. There have been decades of seminal work on developing interatomic potentials albeit with a focus predominantly on capturing the properties of bulk materials. Such physics-based models, while extensively deployed for predicting dynamics and properties of nanoscale systems over the past two decades, tend to perform poorly in predicting nanoscale potential energy surface when compared to high-fidelity first-principles calculations. These limitations stem from the lack of flexibility in such models, which rely on a pre-defined functional form. Machine learning models and approaches have emerged as a viable alternative to capture the diverse size-dependent cluster geometries, nanoscale dynamics and the complex nanoscale potential energy surfaces (PES), without sacrificing the bulk properties. Here, we introduce an ML workflow that combines transfer and active learning strategies to develop high-dimensional neural networks (NN) for capturing cluster and bulk properties for several different transition metals with applications in catalysis, microelectronics, and energy storage to name a few. Our NN first learns the bulk PES from the high-quality physics-based models in literature and subsequently augments this learning via retraining with a higher fidelity first-principles training dataset to concurrently capture both the nanoscale and bulk PES. Our workflow departs from status-quo in its ability to learn from a sparsely sampled dataset that nonetheless covers a diverse range of cluster configurations from near-equilibrium to highly non-equilibrium as well as learning strategies that iteratively improves the fingerprinting depending on model fidelity. All the developed models are rigorously tested against an extensive first-principles dataset of energies and forces of cluster configurations as well as several properties of bulk configurations for 10 different transition metals. Our approach is material agnostic and provides a methodology to transfer and build upon the learnings from decades of seminal work in molecular simulations on to a new generation of ML trained potentials to accelerate materials discovery and design.

show abstract

Learning in continuous action space for developing high dimensional potential energy models

Cited by 41 publications

References 43 publications

Machine learning the metastable phase diagram of covalently bonded carbon

Machine learning the metastable phase diagram of covalently bonded carbon

Materials Discovery With Machine Learning and Knowledge Discovery

Transfer and Active Learning of High Dimensional Neural Network Potentials for Transition Metal Clusters and Bulk

Contact Info

Product

Resources

About