Bulletin of the American Physical Society
APS March Meeting 2023
Volume 68, Number 3
Las Vegas, Nevada (March 5-10)
Virtual (March 20-22); Time Zone: Pacific Time
Session N60: Emerging Trends in Molecular Dynamics Simulations and Machine Learning IVFocus
|
Hide Abstracts |
Sponsoring Units: DCOMP DSOFT GDS DPOLY Chair: Cheng Chin, University of Chicago Room: Room 419 |
Wednesday, March 8, 2023 11:30AM - 12:06PM |
N60.00001: Refinement of Training Schemes for Machine-Learning Interatomic Potentials and Its Applications Invited Speaker: Kohei Shimamura Machine learning interatomic potentials (MLIPs) have provided significant contributions to the field of materials science. Through training with First-Principles (FP) data, MLIPs have achieved high accuracy, which is similar to that obtained using FP calculations and low computational costs comparable to that for empirical interatomic potentials. Combining MLIPs with molecular dynamics (MD) simulations provides sufficient statistics on fundamental physical quantities, as well as the ability to investigate atomic dynamics in large systems. We have successfully calculated important physical quantities such as free energy, specific heat, dielectric constant, structure factor, and thermal conductivity with high accuracy. |
Wednesday, March 8, 2023 12:06PM - 12:18PM |
N60.00002: Benchmarking machine-learned interatomic potential methods for reactive molecular dynamics at metal surfaces Wojciech G Stark, Julia Westermayr, Cas van der Oord, Gabor Csanyi, Reinhard J Maurer Machine-learned interatomic potentials (MLIP) have become widely used tools to accelerate ab initio molecular dynamics simulations in materials science. Many promising MLIPs emerged recently, from simple linear models to deep neural networks (DNN), differing in stability, accuracy, and inference time. The field of reactive dynamics at surfaces has specific requirements on potentials, which make it an interesting area to benchmark different MLIP approaches. Reactive scattering dynamics are highly sensitive to potential corrugation and low reaction probabilities require extensive ensemble averaging. Therefore, MLIPs need to combine smooth and accurate landscapes with extremely efficient inference. In this study, we compare different families of MLIPs, from atomic cluster expansion (ACE), invariant DNN-based SchNet to novel equivariant neural networks such as PaiNN and MACE on the example of reactive molecular hydrogen scattering on copper. We compare these diverse methods by measuring accuracy and inference performance directly on dynamical observables. This provides a detailed picture of MLIP smoothness and corrugation accuracy that goes beyond basic train/test error analysis. |
Wednesday, March 8, 2023 12:18PM - 12:30PM |
N60.00003: Fast and Scalable Uncertainty Estimates in Deep Learning Interatomic Potentials Albert Zhu, Simon L Batzner, Albert Musaelian, Boris Kozinsky Deep Learning has emerged as a promising and computationally-efficient tool for achieving highly accurate predictions of molecular and materials properties. However, a common short-coming shared by deep learning approaches is that they only obtain point estimates of their predictions without any associated predictive uncertainties, which are important for detecting rare events in molecular dynamics simulations. Existing attempts to quantify uncertainty in deep learning models have focused primarily on the use of ensembles of independently trained neural networks, the sample distribution of which is used to estimate model uncertainty. This process, however, results in a large computational overhead in both training and prediction, often orders of magnitude more expensive than a single deep learning model. In this project, we propose a method to estimate predictive uncertainty using only a single neural network along with a computationally-inexpensive Gaussian Mixture Model, eliminating the need for an ensemble. We compare the quality of the uncertainty estimates obtained with our method to those of ensembles. Furthermore, we study the effectiveness of our method in active learning and discover the results to be comparable to active learning with ensembles. |
Wednesday, March 8, 2023 12:30PM - 12:42PM |
N60.00004: Predicting Vapor-Liquid Equilibria and Phase Transitions with Machine-Learned Interatomic Potentials Mitchell A Wood, Normand A Modine, Dionysios Sema, Ember Sikorski, Stan Moore, Nicolas G Hadjiconstantinou A long-standing goal in classical molecular dynamics is to achieve ‘transferability’ of an interatomic potential, meaning the model can remain accurate in out-of-domain applications and reproduce physical properties over a wide range of phase () space. With the increasing popularity and promise of machine learned interatomic potentials (ML-IAP), this goal is finally within reach and has been recently demonstrated in extreme environments of carbon and iron. The present work has trained state-of-the-art ML-IAPs (SNAP, POD, Allegro) for Al over a wide range of phase regimes (0.2-3.0 and 933-10,000 ) that are challenging to model with any other simulation method. We demonstrate the efficiency of these atomic representations by reaching quantum-mechanical accuracy in small dataset and then perform large-scale molecular dynamics simulations to predict the vapor-liquid phase equilibrium and the critical point. Furthermore, these predictions can be validated against carefully crafted DFT simulations. These demonstrations show how ML tools complement and bolster first-principles calculations for equation of state models by allowing for dynamic predictions of materials in extreme environments. |
Wednesday, March 8, 2023 12:42PM - 12:54PM |
N60.00005: Numerical modeling of hydrogen absorption in metal hydrides Olivier Nadeau, Gabriel Antonius One of the main challenges to fully utilize hydrogen as a green and renewable energy vector is its storage. We study the absorption of hydrogen in ternary compounds of type M-Mg-Ni with a combination of ab initio molecular dynamics [1] and classical molecular dynamics [2] using machine learning interatomic potentials (MLIP). Our goal is to accurately predict the enthalpy of absorption, the desorption temperature and the entropy of absorption. We employ the newly developed Machine Learning Assisted Canonical Sampling (MLACS) method [3] to generate on-the-fly interatomic potentials throughout the molecular dynamics simulation. This approach allows us to compute the phonon spectrum of the materials taking into account the anharmonicity of the potential at a reduced computational cost. We will present preliminary results to evaluate the accuracy and speedup enabled by this approach. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700