Bulletin of the American Physical Society
21st Biennial Conference of the APS Topical Group on Shock Compression of Condensed Matter
Volume 64, Number 8
Sunday–Friday, June 16–21, 2019; Portland, Oregon
Session J6: TMS: First-principles and Molecular Dynamics III |
Hide Abstracts |
Chair: Brian Barnes, ARL Room: Broadway III/IV |
Tuesday, June 18, 2019 11:00AM - 11:30AM |
J6.00001: Machine Learning Reactive Force Fields for an Atomistically-Resolved View into Shockwave-Driven Carbon Condensation Invited Speaker: Rebecca Lindsey In this work, we discuss development and application of the Chebyshev Interaction Model for Efficient Simulations (ChIMES) to the problem of carbon condensation. Formation of soot particles resulting from strong shockwave propagation in carbon-rich energetic materials can have significant implications on material performance and sensitivity. However, the time and length scales associated with condensate growth (i.e. up to hundreds of ns and 10's of nm, respectively) make direct investigation highly challenging. From an experimental standpoint, these scales preclude formation tracking, leading to what appears to be instantaneous condensate growth, while for highly predictive quantum-based simulation methods, these scales are prohibitively large. Reactive force field-based approaches, capable of multi-million atom simulations, offer a viable alternative means of studying carbon condensation. To this effect, we discuss development of the Chebyshev Interaction Model for Efficient Simulation (ChIMES), a generalized many-body reactive force field machine-learned to Kohn-Sham Density Functional Theory (DFT) molecular dynamics trajectories. ChIMES models are linear-scaling with respect to system size and are thus highly suitable for large-scale problems requiring atomistic resolution. Strategies for achieving ``quantum-accurate'' descriptions of chemistry in complicated molecular systems are discussed and broad insights stemming from application to carbon condensation in liquid carbon monoxide under extreme conditions are presented. Our results indicate possible mechanisms, timescales, and chemistry for the ensuing condensate products. In collaboration with: L.E. Fried, N. Goldman, S. Bastea and M.R. Armstrong This work is performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. LLNL-ABS-768127 [Preview Abstract] |
Tuesday, June 18, 2019 11:30AM - 11:45AM |
J6.00002: Machine Learning of Interatomic Potentials for Shock Compression Phenomena Benjamin Nebgen, Kipton Barros, Leonid Burakovsky, Saryu Fensin, Timothy Germann, Nicholas Lubbers, Justin Smith As classical molecular dynamics simulations have become a widely used tool for investigating shock loading and release, the need for transferable interatomic potentials that are valid across a wide range of temperatures and pressures has become increasingly urgent. Machine learning is emerging as a powerful tool to emulate electronic structure calculations. Deep neural networks can now predict atomic interactions with accuracies exceeding density functional theory, at a tiny fraction of the computational cost. We will describe recent methods for building interatomic potentials relevant to chemistry, materials science, and biophysics applications. A key concept is active learning, which facilitates optimal training dataset generation using uncertainty quantification built into the neural network. Active learning fills in gaps of the machine learning model without saturating the dataset with similar structures, leading to a surprising level of transferability. Using the recently deployed Sierra supercomputer, we have applied these concepts to develop neural network-based interatomic potentials for aluminum and tin. We will describe these results, and their initial implementation in parallel large-scale non-equilibrium molecular dynamics simulations of shock compression and release. [Preview Abstract] |
Tuesday, June 18, 2019 11:45AM - 12:00PM |
J6.00003: Quantum-accurate SNAP carbon potential for MD shock simulations Jonathan Willman, Ashley Williams, Kien Nguyen Cong, Mitchell Wood, Aidan Thompson, Ivan Oleynik The ability of molecular dynamics (MD) to realistically simulate the high-strain-rate physics is critically dependent on the availability of high fidelity interatomic potentials capable of capturing the major physics of materials response to high temperatures and pressures. We have developed a Spectral Neighbor Analysis Potential (SNAP) machine-learning potential for high-pressure carbon. SNAP is formulated in terms of the bispectrum components, a set of general four-body geometric invariants that characterize the local neighborhood of each atom. Statistical data analysis is used to train the SNAP potential to reproduce a large set of first-principles training data. In this presentation we describe (1) the generation of the training database comprising the consistent and meaningful set of first-principles DFT calculations; (2) the robust and physically guided fit of the SNAP parameters; and (3) the validation of the SNAP potential in large-scale MD simulations of shock compression of carbon materials. [Preview Abstract] |
Tuesday, June 18, 2019 12:00PM - 12:15PM |
J6.00004: Transferable kinetic Monte Carlo models of condensed phase high temperature chemistry learned from molecular dynamics data. Qian Yang, Enze Chen, Vincent Dufour-Decieux, Carlos Sing-Long, Rodrigo Freitas, Evan Reed Complex chemical processes such as the decomposition of energetic materials are typically modeled at the atomistic level using large-scale molecular dynamics simulations (MD), generating a wealth of data. We propose that these rich and expensive datasets can be reused systematically to study related systems, reducing the number of new MD simulations required. We develop a statistical learning framework for extracting information about the fundamental underlying reaction pathways observed from MD data, using it to build kinetic Monte Carlo models (KMC) of the corresponding chemical reaction network. We show how and why our KMC models can predict the dynamics of entirely different chemical trajectories. We demonstrate our framework throughout on different systems of high temperature, high pressure liquid hydrocarbons. One can easily imagine a future in which MD simulations used for research are routinely archived and analyzed in order to add to and modify an existing repository of elementary chemical reactions and reaction rates. This repository would form a ``chemical genome'' that can then be used to quickly simulate all kinds of new chemical systems. [Preview Abstract] |
Tuesday, June 18, 2019 12:15PM - 12:30PM |
J6.00005: Machine-learning based multi-scale model for shock-particle interactions Oishik Sen, Soren Taverniers, Pratik Das, Gustaaf Jacobs, H.S. Udaykumar Macro-scale computational models for shock-particle interactions require closure laws for the drag and subgrid-scale pseudo-turbulent stresses on particles. Traditionally, closures for drag are provided via semi-empirical models developed from expensive physical experiments, which are often limited in parameter spaces. In recent years, drag and the pseudo-turbulent stresses on particle clusters have been learned using machine-learning techniques from \textit{in silico }experiments of high-fidelity particle-resolved computations of shock-particle interactions. In this work, we compare macro-scale simulations which employ machine-learned, simulation-derived drag and pseudo-turbulent stresses with those that use phenomenological drag laws for closure. The macro-scale simulations study the evolution of a cloud of aluminum particles with volume fraction of 4 {\%} under a shock of Mach number, \textit{Ma} $=$ 3. The results are compared against particle-resolved meso-scale simulations of shock-particle interactions. It is shown that the macro-scale computations which employ machine-learned drag and SGS laws capture the dominant flow features such as shocks, expansion fans, and the vortical structures accurately. Furthermore, the locus of the center of mass of the particle cloud obtained from the macro-scale simulations are also shown to be in good agreement with the meso-scale particle-resolved computations. [Preview Abstract] |
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