Bulletin of the American Physical Society
APS March Meeting 2022
Volume 67, Number 3
Monday–Friday, March 14–18, 2022; Chicago
Session Q46: Emerging Trends in Molecular Dynamics Simulations and Machine Learning IIFocus Session Recordings Available
|
Hide Abstracts |
Sponsoring Units: DCOMP GDS DSOFT DPOLY Chair: Thomas Linker, University of Southern California Room: McCormick Place W-470A |
Wednesday, March 16, 2022 3:00PM - 3:36PM |
Q46.00001: Bio-inspired machine learning towards mechanistic insights, generative design, and discovery from the bottom up Invited Speaker: Markus Buehler Nature produces a variety of tough materials with many functions, often out of simple and abundant materials, and at low energy. Such systems - examples of which include spider silk, conch shells, nacre or bone - provide broad inspiration for engineering. Here we explore the translation of biomaterials to engineering designs, using a variety of tools including molecular modeling, AI and machine learning, and experimental synthesis using 3D printing, and characterization. We review a series of bottom-up studies focused on the mechanical behavior of bio-inspired composite materials, especially fracture , compression and impact, and how these phenomena can be modeled using a combination of molecular dynamics and machine learning. We present examples that involve deep convolutional neural networks, graph neural networks, transformers, and game theoretical approaches towards analysis and design of atomic-level material structures. One case study will cover a recent example that realizes a text-to-material design approach, developing new architected multimaterial composite designs based on human readable description and subsequent 3D printing - from word to matter, using transformer neural networks. Another case study will explore the use of deep learning to design synthetic diatom geometries, which are manufactured using multimaterial 3D printing. We conclude the talk with a series of case studies of material optimization using genetic algorithms focused on grain boundary architectures and gradients, novel 3D printed composites, as well as a translation of molecular structures to music and back to assess universal patterns through vibrational patterning. A particular case study will include an analysis of Bach's Goldberg variations and translation to proteins using deep learning, and new musical composition. The approach used in this example, the Deep Aria (https://soundcloud.com/user-275864738/aria-inceptionism-in-protein) will be explained. |
Wednesday, March 16, 2022 3:36PM - 3:48PM |
Q46.00002: Development of SNAP Interatomic Potentials for Gas-Metal Interactions for Fusion Energy Materials Mary Alice, Mitchell A Wood, Aidan P Thompson Tungsten is currently the candidate material for the divertor component of future fusion reactors. The divertor will be subject to high particle fluxes of a variety of plasma species including both hydrogen and nitrogen. This results in a variety of microstructural changes including hydrogen blister and tungsten-nitride formation that require further understanding to prevent material degradation. Modeling like molecular dynamics (MD) can provide insight into material deformation, but accurate interatomic potentials (IAPs) are limited for these types of material interactions especially given the complexity of accurately representing the range of chemical environments for gas-metal interactions. Machine learned interatomic potentials like the Spectral Neighbor Analysis Potential (SNAP) have been shown to have higher accuracy compared to traditional IAPs and may be well suited to modeling these types of gas-metal interactions. In this work, the development of W-H and W-N SNAP potentials that reproduce gas species behavior as well as gas impurity behavior on surfaces and in bulk will be discussed. MD simulations of hydrogen or nitrogen implantation in tungsten will be shown. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525. |
Wednesday, March 16, 2022 3:48PM - 4:00PM |
Q46.00003: Supervised Learning and the Finite-Temperature String Method for Computing Committor Functions and Reaction Rates Clay H Batton, Muhammad R Hasyim, Kranthi K Mandadapu The committor function, the probability that a system enters the product state before the reactant state, determines the reaction rates and transition pathways for a given rare-event problem. Recent work [1] constructed an algorithm where a neural network that represents the committor function is trained with data obtained from importance sampling. In this work [2], we extend their approach by combining it with supervised learning, where sample-mean estimates of the committor function obtained via short simulations are used to aid the training of the neural network, and the finite-temperature string (FTS) method, which enables homogeneous sampling across the transition pathway. We show that these modifications are crucial for obtaining accurate estimates of the committor function and reaction rates on systems with non-convex potential energy, where reference solutions are known. We also show that the sampling distribution of reaction rates estimated from algorithms employing the FTS method obeys a log-normal distribution, allowing accurate estimation of reaction rates with small sample size. |
Wednesday, March 16, 2022 4:00PM - 4:12PM |
Q46.00004: Learning to Simulate Time-averaged Coarse-grained Molecular Dynamics with Geometric Machine Learning Xiang Fu, Tian Xie, Nathan J Rebello, Bradley D Olsen, Tommi S Jaakkola Molecular dynamics (MD) simulation is the workhorse of various scientific domains. However, simulating a physical system with many particles is tremendously computationally expensive. Learning-based force fields have made major progress in accelerating ab-initio MD simulation but are still significantly slower than classical force fields. Complex systems such as battery and protein take weeks to months to simulate, even with classical force fields. We adopt a different ML approach by learning time-averaged acceleration at a coarse-grained level from trajectory data generated by traditional MD simulation. We coarse-grain a physical system using graph clustering and then use a deep graph neural network to model the time-averaged evolution. Our model can simulate complex systems at a lower spatial/temporal resolution and preserve key statistics of interest. Despite only trained to make single-step predictions, our model can rollout for 100,000 steps and recover properties related to 1-10ns level long-time dynamics. Our model applies to a range of estimation problems for complex systems, including predicting radius of gyration of single-chain coarse-grained polymers of more than 1000 beads in implicit solvent and Li diffusivity of multi-component Li-ion polymer electrolyte systems. |
Wednesday, March 16, 2022 4:12PM - 4:48PM |
Q46.00005: Simulating materials with quantum computers Invited Speaker: Lindsay Bassman Quantum materials exhibit a wide array of exotic phenomena and practically useful properties. A better understanding of these materials can provide deeper insights into fundamental physics in the quantum realm as well as advance information processing technology and sustainability. The emergence of digital quantum computers (DQCs), which can efficiently perform quantum simulations that are otherwise intractable on classical computers, provides a promising path forward for testing and analyzing the remarkable, and often counter-intuitive, behavior of quantum materials. Equipped with these new tools, scientists from diverse domains are racing towards achieving physical quantum advantage (i.e., using a quantum computer to learn new physics with a computation that cannot feasibly be run on any classical computer). In this talk, I will provide a summary of progress made towards this goal that is accessible to scientists across the physical sciences. I will review available technology and algorithms, discuss prospects for machine learning to aid quantum computing simulations, and showcase simulations that have been successfully performed on currently available DQCs, emphasizing the variety of properties that can be studied with this nascent technology. Ideally, this talk will serve as an organized overview of progress in the field for domain experts and an accessible introduction to scientists in related fields interested in beginning to perform their own simulations of quantum materials on DQCs. |
Wednesday, March 16, 2022 4:48PM - 5:00PM |
Q46.00006: Size-Dependent Melting Temperature of Rubidium: Thermodynamic Integration Based on First-principles Calculations Shogo Fukushima, Aiichiro Nakano, Rajiv K Kalia, Priya Vashishta, Fuyuki Shimojo, Hiroyuki Kumazoe, Masaaki Misawa, Kohei Shimamura, Akihide Koura It is well known that a maximum exists in the pressure dependence of melting temperature of alkali metals such as Rubidium (Rb). Thermodynamic integration (TI) based on quantum molecular dynamics (QMD) enables us to estimate the phase-transition temperature accurately, but with a very high computational cost. Therefore, we focused on Artificial Neural Network (ANN). By training QMD data using ANN, it is possible to create high accuracy interatomic potential (ANN potential). The computational cost of TI can be greatly reduced by using ANN potentials, while retaining first-principles accuracy. However, existing ANN potentials were trained with rather small QMD simulations, and it is imperative to systematically study the size dependence of melting temperature. |
Wednesday, March 16, 2022 5:00PM - 5:12PM |
Q46.00007: Asymmetric Carrier Transport in Two-Dimensional Ferroelectric Layers RURU MA, Anikeya Aditya, Thomas M Linker, Shogo Fukushima, Fuyuki Shimojo, Aiichiro Nakano, Rajiv K Kalia, Priya Vashishta Ferroelectric tunnel junctions can offer low power and non-destructive readout for memory based applications, but are currently limited by their low tunneling electro-resistance (TER). Recent experimental study has shown greatly enhanced tunneling barrier and TER of hetrostacks based on ferroelectric CuInP2S6 (CIPS). However, microscopic mechanisms underlying the observed asymmetric charge carrier transport in CIPS remain unknown. We perform nonadiabatic quantum molecular dynamics simulations to study carrier dynamics in a graphene-CIPS interface under high electric fields. Simulations results exhibit a high asymmetry in carrier transport due to CIPS polarization, which is likely related to the observed tunneling barrier enhancement. Better understanding of the asymmetry in the charge carrier transport will give valuable insight into the design of novel ferroelectric tunneling junctions. |
Wednesday, March 16, 2022 5:12PM - 5:24PM |
Q46.00008: Graph Neural Network with Inductive Bias for Robust Molecular Dynamics Pankaj Rajak, Aravind Krishnamoorthy, Rajiv K Kalia, Aiichiro Nakano, Priya Vashishta, Ekin D Cubuk Machine learning (ML) force field models for molecular dynamics (MD) simulations often suffer from poor system stability with instabilities such as atom clustering that must be corrected by active learning approaches. However, relationship between the structural and chemical complexity of multi-component systems and the robustness of long-time ML-based MD dynamics has not been studied in detail. We develop a graph neural network (GNN) model for a range of material systems to perform ML-MD simulations with quantum mechanical accuracy but orders-of-magnitude faster. A GNN model is sufficient to ensure robust long-time dynamics in a ‘simple’ system like SiC and Cu. However, we need additional inductive bias, in the form of energy decomposition into 2-body and 3-body terms to generate stable MD trajectories for complex systems such as GeSe2 and VO2, which can exist in multiple metastable atomic configurations. |
Wednesday, March 16, 2022 5:24PM - 5:36PM |
Q46.00009: Neural Network Potentials for CO2 Adsorption in Amine-Appended Metal-Organic Frameworks Yusuf Shaidu, Jeffrey B Neaton Metal-organic frameworks (MOFs), a class of porous materials consisting of metal ions and organic ligands, are promising candidates for CO2 capture and separations applications. Amine-appended variants of MOFs Mg2(dobpdc) (dobpdc4- = 4,4’-dioxidobiphenyl-3,3’-dicarboxylate) selectively bind CO2 via a novel cooperative adsorption mechanism that gives rise to step-shaped isotherms, allowing for high working capacities to be attained with small temperature-swings[1]. The equilibrium thermodynamics of CO2 is quantified experimentally with the differential enthalpy which can be compared to zero temperature density functional theory (DFT) calculations. However, finite temperature studies within DFT are hindered by their computational complexity due to the large number of atoms per primitive cell, making the energetics and details of the CO2 insertion/deinsertion dynamics not easily accessible. Here, we summarize the development of an accurate and transferable neural network potential for amine-appended MOFs via active learning approach to enable a detailed study of the energetics and dynamical nature of CO2 bound amine-appended MOFs at finite temperatures. |
Wednesday, March 16, 2022 5:36PM - 5:48PM |
Q46.00010: Machine learning strategies for potential development in highly concentrated/high-entropy driven Ni-based Superalloys Marium Mostafiz Mou, Tyler J McGilvry-James, Ridwan Sakidja In this current work, we developed the Deep Learning potentials for highly concentrated multi-component metallic system by utilizing the DeepMD as the Deep Learning/Machine Learning code. For this purpose, we employed the outcome from the ab-initio molecular dynamics simulations from the Vienna Ab-initio Software Package (VASP) following the Density Functional Theory (DFT) approximations including the data of energy, forces, and virial database. These data were strategically sampled from compositions close to the concentrated system and/or high-entropy alloy with the same constituents. The efficiency and validity of the developed potentials were verified through a series of predictive molecular dynamics simulations of thermomechanical properties that are of great interest to the development of advanced Ni-based Superalloys. The support from the National Energy Technology Laboratory (Grant No. FE0031554) is gratefully acknowledged. We would also like to express our gratitude to NERSC for providing the supercomputer resource. |
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. |
© 2025 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