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 T60: Emerging Trends in Molecular Dynamics Simulations and Machine Learning VI |
Hide Abstracts |
Sponsoring Units: DCOMP Chair: Pankaj Rajak, Argonne National Laboratory Room: Room 419 |
Thursday, March 9, 2023 11:30AM - 11:42AM |
T60.00001: Prediction of Metal-Organic Framework Adsorption Isotherms using Ab Initio derived Neural Network Potentials Pedro Guimarães Martins, Yusuf Shaidu, Eric Taw, Alex Smith, Jeffrey B Neaton Metal-organic frameworks (MOFs) form an extensive class of porous materials of interest for new applications, and therefore, predicting their properties computationally is crucial in advancing the exploration of this wide chemical space. A predictive and generalizable method for calculations of isotherms could accelerate the discovery of new promising MOFs for gas adsorption applications. While empirical force fields can be combined with grand canonical Monte Carlo (GCMC) simulations to estimate adsorption isotherms in specific MOFs [1], this approach relies on fixed functional forms as well as an extensive system-specific parameterization, limiting its applicability to a broader set of MOFs or to automated force field generation. Ab initio neural network potentials (NNPs), derived from density functional theory calculations, are a promising and generalizable alternative for accurate binding energetics and isotherm prediction. In this work, we combine ab initio NNPs and GCMC simulations to predict CO2 adsorption isotherms for well-studied MOF, Mg2(dobpdc) (dobpdc4- = 4,4’-dioxidobiphenyl-3,3’-dicarboxylate), comparing to prior simulations and experimentals. |
Thursday, March 9, 2023 11:42AM - 11:54AM |
T60.00002: Representing Fly Behavior with Recurrent Neural Networks Ishan Saran Behavior is a complex process that operates across many time and length |
Thursday, March 9, 2023 11:54AM - 12:06PM |
T60.00003: Thermal Conductivity of Amine-Appended Metal-Organic Frameworks for Carbon Capture using Neural Network Interatomic Potentials Yusuf Shaidu, Jeffrey B Neaton Diamine-appended Mg2(dobpdc) (dobpdc 4-=4,4’-dioxidobiphenyl-3,3’-dicarboxylate) metal-organic frameworks (MOFs) capture CO2 via cooperative adsorption mechanisms, enabling a small temperature or pressure swing to reach full working capacity. As the adsorption/desorption process is temperature driven, understanding thermal transport in these materials is vital to their use in carbon capture applications. Previous molecular dynamics simulations of Zn2(dobpdc) MOFs suggest they are in the ultra-low thermal conductivity regime[1]. In this work, we combine neural network potentials[2] with a recently-introduced general framework[3] to predict the thermal conductivity of amine-appended Mg2(dobpdc). We explain its origin and identify the responsible phonon modes, comparing the case of the diamine-appended MOF with the unfunctionalized MOF with and without CO2. Implications for thermal management of these materials in carbon capture applications are discussed. |
Thursday, March 9, 2023 12:06PM - 12:18PM |
T60.00004: Simulation of High-Generation Phytoglycogen Dendrimers Interacting with Charged, Hydrophobic Molecules Benjamin E Morling, Sylvia M Luyben, Robert A Wickham, John R Dutcher Phytoglycogen (PG) is a naturally occurring, highly branched, polysaccharide that is extracted from sweet corn as a soft, hydrated, dendritic nanoparticle. Chemical modification of native PG with small, charged and/or hydrophobic molecular species allows for a wider range of applications for PG in personal care, nutrition, and biomedicine. We efficiently simulate this 11-generation (18,424 monomer) dendrimer in water using dynamical self-consistent field theory (dSCFT) by implementing an operator decomposition scheme. By coupling the Poisson-Boltzmann equation to our neutral dSCFT model for native PG nanoparticles, we can account for the interaction of charged, hydrophobic molecules with PG. These molecules are also capable of irreversibly binding to the dendrimer. We examine the effects of varying the hydrophobicity of the molecules, their concentration, and the concentration of salt on the distribution of molecules throughout the dendrimer, and on the dendrimer radius and hydration. |
Thursday, March 9, 2023 12:18PM - 12:30PM |
T60.00005: Modeling Excited-State Dynamics for Polariton Chemistry with Hierarchically Interacting Particle Neural Network Xinyang Li, Yu Zhang, Sergei Tretiak, Kipton M Barros, Nicholas E Lubbers, Alice Allen, Ben T Nebgen, Sakib Matin Chemical reactions inside an optical cavity happen intrinsically in the collective coupling regimes, where many molecules couple to the same cavity simultaneously. However, modeling such a complicated system with an assemble of molecules is computationally expensive. Recently, machine learning (ML) techniques, especially neural networks (NN), have been widely used to predict quantities, like energies and charges. As such, it is a natural choice to combine NN with polariton chemistry simulations. Unfortunately, even for a single molecule, predicting excited states quantities with NN remains a challenging task. In this talk, we present a general protocol to predict excited-state properties, such as energies, transition dipoles, and non-adiabatic coupling vectors (NACR) with the hierarchically interacting particle neural network (HIP-NN), and applying these predictions to excited-state polariton chemistry calculation in the collective coupling regime. |
Thursday, March 9, 2023 12:30PM - 12:42PM |
T60.00006: Optimal Development of Transferable Machine Learning Interatomic Potentials using Active Learning David O Montes de Oca Zapiain, Mitchell A Wood, Dionysios Sema, Aidan P Thompson Automated methods for generating atomistic configurations make possible the creation of vast and diverse datasets where potentials that exhibit consistent accuracy across diverse configurations can be trained. However, automated generation frameworks struggle to create configurations of physical relevance to the potential development task of interest. For that reason, a training dataset that combines configurations generated using automated frameworks and domain expertise is expected to yield better performing potentials. Nevertheless, integrating configurations from these two frameworks is not a trivial task given the fact that data-driven methods often yield a very large number of configurations, which places a severe computational burden for calculating the results, and domain-expertise methods are not capable of scaling in order to generate a vast number of configurations. Consequently, there is a critical need for an enhanced training protocol that can integrate both types of configurations in an optimized and data-driven manner. This work addresses this challenge by establishing an automated protocol to train a potential using an ensemble of neural network based potentials and active learning. The developed protocol trains the potential iteratively by automatically incorporating a fixed set of configurations that maximize the information gained. |
Thursday, March 9, 2023 12:42PM - 12:54PM |
T60.00007: Physics-Guided Problem Decomposition for Scaling Deep Learning of Quantum Spin Models Wei-Cheng Lee, Sangeeta Srivastava, Samuel Olin, Viktor A Podolskiy, Anuj Karpatne, Anish Arora Given their ability to effectively learn non-linear mappings and perform fast inference, deep neural networks (NNs) have been proposed as a viable alternative to traditional simulation-driven approaches for solving high-dimensional eigenvalue equations (HDEs), which are the foundation for many scientific applications. Unfortunately, for the learned models in these scientific applications to achieve generalization, a large, diverse, and preferably annotated dataset is typically needed and is computationally expensive to obtain. Furthermore, the learned models tend to be memory- and compute-intensive primarily due to the size of the output layer. While generalization, especially extrapolation, with scarce data has been attempted by imposing physical constraints in the form of physics loss, the problem of model scalability has remained. |
Thursday, March 9, 2023 12:54PM - 1:06PM |
T60.00008: Sample Size Determination for Machine Learning Surrogates of Molecular Dynamics Simulations Fanbo Sun, Kadupitiya JCS, Vikram Jadhao The performance promise of machine learning (ML) surrogates of molecular dynamics (MD) simulations of soft materials is significant, but generally comes at the cost of acquiring large training datasets to learn the complex relationships between input soft material attributes and output properties. Under the constraint of limited high-performance computing resources, optimizing the selection of the simulations for creating well-represented training datasets becomes paramount. Using an artificial neural network based, well-trained ML surrogate for MD simulations of confined electrolytes, we explore the possibility of balancing surrogate accuracy with the cost of training. The dependence of performance metrics such as accuracy, mean-squared error, and speed-up on the training data volumes is investigated. We show that a decrease in the training dataset size by 8x leads to a drop in the surrogate accuracy by ~4%. We determine the relative importance of different input features, and introduce a sample size reduction strategy to further reduce the training size while maintaining the desired levels of accuracy and robustness in surrogate predictions. The link between the uncertainties present in the ground truth data and the surrogate performance is also examined. |
Thursday, March 9, 2023 1:06PM - 1:18PM |
T60.00009: Interpretable machine learning analysis of quantum gas microscopy data of doped Fermi-Hubbard model YANJUN LIU, Yao Wang, Henning Schloemer, Annabelle Bohrdt, Timon A Hilker, Fabian Grusdt, Immanuel Bloch, Eun-Ah Kim, Joannis Koepsell, Dominik Bourgund, Sarah Hirthe, Guillaume Salomon, Christian Gross, Pimonpan Sompet Exploring the phase diagram of the Fermi-Hubbard model is among the key motivations of quantum simulation experiments. We apply the Hybrid-CCNN [1] approach to quantum gas microscopy data of the Fermi Hubbard model across a large doping range. The Hybrid-CCNN approach combines unbiased unsupervised machine learning with feature revealing supervised machine learning. The unsupervised learning stage identifies three different regimes: the magnetic polaron regime at low to intermediate dopings, and the Fermi liquid regime at high dopings, consistent with the manual analysis based on target correlation functions[2]. Moreover, unsupervised learning identifies the cross-over regime [2] as a distinct region of the phase space. The feature analysis using interpretable supervised learning techniques reveals characteristics unique to this intermediate phase. We discuss theoretical implications of the machine learning findings. |
Thursday, March 9, 2023 1:18PM - 1:30PM |
T60.00010: Bitstring-ChiFc: machine learning Fisher Information Metric from bitstrings Victor Kasatkin, Itay Hen, Nic Ezzell, Utkarsh Mishra, Daniel A Lidar, Lalit Gupta, Evgeny Mozgunov We consider the task of identifying phase transitions in low temperature states in systems on a lattice described by a parameterised family of spin Hamiltonians with no known order parameters given access to an oracle producing bit strings representing the Z-basis measurement of such states. For fixed finite system size such phase transitions are typically reflected by maxima of fidelity susceptibility. Given only the access to bit strings we can only hope to estimate the classical fidelity susceptibility. Here we show the conditions under which classical fidelity susceptibility matches the full fidelity susceptibility and demonstrate a machine learning based method for determining the classical fidelity susceptibility from a dataset of bit strings or an access to an oracle able to produce such bit strings. We investigated numerically the performance of the methods on a few toy models including the Transverse Field Ising Model, Frustrated Ising Ladder, and MaxCut problem represented as an Ising system on a random graph. In this talk we present the numerical results on one of these toy models. |
Thursday, March 9, 2023 1:30PM - 1:42PM |
T60.00011: Kinetic isotope effects for excited-state gas-phase reactions: A surface-hopping ab initio molecular dynamics study Richard A Messerly, Brendan Gifford, Ivana Gonzales, Troy Holland Dissociative recombination (DR) reactions are important when modeling charged species in the presence of free electrons. While experimental measurements of DR reaction rates are challenging, surface hopping ab initio molecular dynamics (SH-AIMD) simulations provide an attractive alternative. SH-AIMD is especially well-suited for estimating branching ratios, i.e., the relative rates of competing production channels, for DR reactions. Although the radiolysis of diatomic tritium has been studied experimentally, previous attempts to model these systems have failed to account for isotope effects in DR reactions. Previous SH-AIMD studies have also not investigated tritium isotope effects for the branching ratios of DR reactions. In this study, we compute the DR branching ratios of the protiated and tritiated ketenyl ion. Comparison with literature values for the protiated branching ratios provides confidence in the reliability of our SH-AIMD results. Our simulations predict a significant increase of the HC + CO branching ratio for the tritiated system. |
Thursday, March 9, 2023 1:42PM - 1:54PM |
T60.00012: Investigation on Multi-Pass Ultra-Precision Cutting of Sapphire Using Molecular Dynamic Simulations Yiyang Du, Dalei Xi, Aditya Nagaraj, Suk Bum Kwon, Dae Nyoung Kim, Sangkee Min, Woo Kyun Kim Sapphire (α-Al2O3) is a material with numerous preferable mechanical properties such as high mechanical strength, high thermal and electrical resistivities, and light transmittance, thus having a wide range of applications. However, due to its inherent brittleness, its machinability has been proven to be a challenging roadblock in its versatility. To avoid fracture during the machining process, the ductile mode cutting mechanisms activated through ultra-precision cutting are investigated via the applications of molecular dynamic simulation. Specifically, this presentation discusses the effects of the number of passes on the cutting process and how the deformation mechanisms change when multi-pass cutting is applied to different crystallographic orientations of sapphire. The results suggest that the optimization of multi-pass cuts can significantly increase the critical depth of cut below which the crystal deforms in a ductile manner without significant fractures initiating. |
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