Bulletin of the American Physical Society
APS March Meeting 2022
Volume 67, Number 3
Monday–Friday, March 14–18, 2022; Chicago
Session N49: Emerging Trends in Molecular Dynamics Simulations and Machine Learning IIIFocus Recordings Available
|
Hide Abstracts |
Sponsoring Units: DCOMP GDS DSOFT DPOLY Chair: Aravind Krishnamoorthy, University of Southern California Room: McCormick Place W-471B |
Wednesday, March 16, 2022 11:30AM - 12:06PM |
N49.00001: Modeling Earth's interior from atomic to global scale Invited Speaker: Renata M Wentzcovitch Geophysics stands on a synergistic tripod consisting of seismology, geodynamics, and mineral physics. It advances by close cooperation between these fields that can be computationally very intensive. The role of mineral physics is to provide information on mineral properties to interpret seismic tomography and provide input for advanced and more refined geodynamics simulations. Computational mineral physics has contributed significantly to the integration of these fields. It has complemented experiments by expanding the pressure and temperature range in which properties can be obtained and has offered access to atomic-scale phenomena that suggested new interpretations of experimental and seismological data. |
Wednesday, March 16, 2022 12:06PM - 12:18PM Withdrawn |
N49.00002: Molecular Dynamics Simulations of Solid Electrolytes with NequIP Equivariant Machine Learning Models Juan F Gomez, Liwen Wan, Simon L Batzner, Albert Musaelian, Brandon Wood, Boris Kozinsky Optimizing ion transport kinetics in solid-state energy storage systems is critical for performance. These devices contain interfaces, such as internal grain boundaries, which appear unavoidably from synthesis/processing, and electrolyte/cathode interfaces. Owing to the experimental difficulty in probing the complex chemical reactions and structural rearrangement that happens during operation, the details of how these microstructures impact ion transport remain elusive. A computational model that predicts ionic transport across these interfaces would be invaluable to the design of these solid-state energy storage systems. |
Wednesday, March 16, 2022 12:18PM - 12:30PM |
N49.00003: Obtaining vibronic excitation spectra of small organic molecules using machine learning simulations and power spectra techniques Andrew M Johannesen, Jason D Goodpaster Vibrationally-electronically excited spectra for small organic molecules are obtained from machine learning driven molecular dynamics trajectories in order to elucidate surface enhanced Raman spectroscopy (SERS) spectra. Vibronic excitation spectra can be obtained from molecular dynamics trajectories through autocorrelation and power spectrum based methods. Direct calculation of vibronic excitation spectra with computationally expensive wavefunction methods are usually impractical, and are often plagued by pragmatic considerations (such as the harmonic oscillator approximation or numerical differentiation techniques) which limit their applicability. Machine learning potentials offer a path toward the practical production of these through molecular dynamics trajectories. Using the ANI method for constructing machine learning potentials and the TRAVIS molecular analysis package for calculating trajectory autocorrelation, we produce vibronic excited state spectra while avoiding both the harmonic oscillator approximation and numerical differentiation. Suitable for higher temperature analysis and applicable across a wide range of electronic structure methods, this work demonstrates the predictive power machine learning models offer for vibronic excited state spectra in SERS phenomena. |
Wednesday, March 16, 2022 12:30PM - 12:42PM |
N49.00004: Improved, Reliable Uncertainty Quantification of Interatomic Models using Sloppy Model Analysis Yonatan Kurniawan Interatomic models (IMs) are widely used in molecular modeling to circumvent the computational cost of quantum calculations. These IMs are often designed for specific applications of interest, and they are used to predict other materials properties that are not used in the development process. Uncertainty quantification (UQ) is relevant for assessing the reliability of these out-of-sample predictions. Previous studies have shown that many IMs are insensitive to large, coordinated changes in many of their parameters, a phenomenon known as sloppiness. Furthermore, our previous work has shown that sloppiness poses challenges both for the implementation and interpretation of traditional UQ analysis. We propose a systematic UQ process for sloppy models, utilizing the Manifold Boundary Approximation Method (MBAM) to identify sloppy parameters and find the reduced, less-sloppy model. We demonstrate this process using the Stillinger-Weber (SW) potential, calibrating it to the atomic forces of a molybdenum disulfide system. We find that the parametric uncertainty of the reduced model is less sensitive to the choice of confidence level and leads to non-diverging uncertainty in higher statistical confidence level. |
Wednesday, March 16, 2022 12:42PM - 12:54PM |
N49.00005: Large-scale dynamics simulations of complex liquid electrolytes with NequIP equivariant machine learning models. Nicola Molinari, Albert Musaelian, Simon L Batzner, Boris Kozinsky Electrolytes control efficiency, anode/cathode stability, battery power as well as safety, thus their optimization is crucial for the design of next-generation energy storage devices. In this work, we focus on ionic liquid electrolytes and demonstrate the application of state-of-the-art equivariant graph neural network models for interatomic interactions (NequIP [1]), trained on DFT energies and forces. Ionic liquid electrolytes exhibit a unique challenge due to their strong interactions and viscous dynamics. Additionally, substantially diverse inter-atomic environments are often present as a function of lithium-salt doping [2], raising the subtle question of model transferability. In summary, we examine the tradeoffs between computational speed and accuracy for large-scale ionic liquid molecular dynamics investigations with state-of-the-art machine learning models. |
Wednesday, March 16, 2022 12:54PM - 1:06PM |
N49.00006: Application of Machine Learning to the Development of Ti Interatomic Potentials Sean J O'Connor, Volker Eyert, Jörg-Rüdiger Hill, David Reith, Erich Wimmer, Patrick R Thomas, Ben Sikora, Paul Rulis Within computational physics and material science an important area of research is that of identifying potential function parameters that tie electronic scale bonding phenomena to atomic scale nanostructures. Force-matching makes use of quantum mechanical ab initio electronic structure calculations to produce computationally efficient analytic interatomic potentials (IP) for use in classical molecular dynamics (CMD) simulations of nanoscale defects. Although the principle of force-matching is straightforward, the practical application of the method and others like it is labor intensive and computationally prohibitive. For this reason the creation of useful and accurate IPs is a cumbersome process. The more sophisticated the IP, the larger the associated computational cost, as more fitted parameters are required. Furthermore, even sophisticated IPs tend to have a limited scope of usefulness. |
Wednesday, March 16, 2022 1:06PM - 1:18PM |
N49.00007: Artifactual Liquid-Liquid Hydrogen Phase Transition from a Machine-Learnt Potential Samuel B Trickey, Valentin Karasiev, Joshua Hinz, Suxing Hu Condensed hydrogen is an intrinsically extreme system of great importance. Cheng et al. [ Nature 585, 217 (2020) ] trained a Hydrogen machine-learning potential (MLP) mostly on small system ab initio MD (AIMD) using DFT. In MD on larger systems (≤ 1728 atoms), the MLP gives a continuous liquid-liquid phase transition and supercriticality, at odds with all prior conventional AIMD. They claimed the prior calculations are erroneous because of finite-size effects exacerbated by use of the NVT ensemble. Our AIMD NPT simulations up through 2,048 atoms do not sustain that. Consistent with our earlier NVT work at smaller sizes [ Phys. Rev. Res. 2, 032065(R) (2020) ], we find a first-order transition. We conclude that the MLP-MD results are artifactual, because the MLP-MD does not systematically reproduce the DFT AIMD from which it supposedly comes. Comparison suggests, but does not prove, that the MLP is a smooth interpolation across the phases. |
Wednesday, March 16, 2022 1:18PM - 1:30PM |
N49.00008: Machine Learned Interatomic Potential Development for W-ZrC Ember L Sikorski, Julien Tranchida, Mary Alice Cusentino, Mitchell A Wood, Aidan P Thompson While tungsten is a leading candidate for the divertor material in future fusion reactors, its performance is limited by its thermo-mechanical properties, such as high ductile to brittle transition temperature and substantial recrystallization at ITER operating conditions. To improve mechanical performance, dispersoids like zirconium carbide can be added during manufacturing. Molecular dynamics (MD) can be leveraged to better understand how such microstructural changes will impact divertor material performance. However, there is a lack of accurate W-ZrC interatomic potentials. In this work, we will describe a machine learned Spectral Neighbor Analysis Potential (SNAP) developed for the W-ZrC system. The SNAP potential is trained on Density Functional Theory data for the respective pure materials, surfaces, and W-ZrC interfaces. Ab initio Molecular Dynamics data is additionally included in the training set to improve applicability of the potential to simulations at temperature. We will discuss the use and accuracy of the newly developed W-ZrC potential in running MD simulations of ZrC dispersoids in W. |
Wednesday, March 16, 2022 1:30PM - 1:42PM |
N49.00009: Deep Potential Development of Highly Concentrated/High Entropy-driven Carbides Tyler J McGilvry-James, Marium Mostafiz Mou, Ridwan Sakidja In this study, DeepMD (Deep Potential Molecular Dynamics) code was utilized to develop Deep potentials for highly concentrated/high entropy driven TM-rich carbides (TM= transition metals) as the key precipitates in Ni-based Superalloys. The deep learning algorithm has been trained against ab-initio molecular dynamics data generated in VASP following Density Functional Theory (DFT) approximations. The data sets include the energy, force, and virial of corresponding supplied trajectories and atoms. The accuracy of the Deep potentials was then tested using classical molecular dynamics simulations with the focus on the elastic and thermo-mechanical property validations. The use of high-entropy alloy (HEA) compositional strategy to maximize the interaction statistics within the carbide phases allowed us to construct the multi-component potentials. 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. |
Wednesday, March 16, 2022 1:42PM - 1:54PM |
N49.00010: Generative Coarse-Graining Wujie Wang In theoretical and computational science, order reduction or coarse-graining (CG) is a popular approach to describe complex systems with lower-dimensional representations that capture the key physics. In modeling the dynamics of large biological and materials systems at the molecular level, CG simulations unify selected groups of atoms into CG particles described by simpler equations of motions than the all-atom system. However, the restoration of fine-grained (FG) coordinates from CG representations is challenging due to the inherent information loss in the CG procedure. Traditional backmapping methods do not fully capture the geometrical constraints of the all-atom distribution and therefore tend to reconstruct unrealistic geometries. Inspired by the recent progress in deep generative models and equivariant neural nets, we propose a unified probabilistic formulation of backmapping, where the conditional likelihood of FG positions solely depends on the CG conformations. We design a conditional variational autoencoder that encodes the missing FG information into an invariant latent space and decodes back to FG geometries via convolutions over equivariant vector basis obtained from CG geometries. To evaluate both the diversity and quality of generated geometries, we propose three metrics and apply them to two molecular dynamics datasets: Alanine Dipeptide and Chignolin. Extensive experiments on these benchmarks demonstrate that our approach significantly outperforms all related baseline methods. Interestingly, the results suggest that most FG information can be robustly recovered even for ultra-low resolution CG models. |
Wednesday, March 16, 2022 1:54PM - 2:06PM |
N49.00011: Dynamical strengthening of covalent and non-covalent molecular interactions by nuclear quantum effects at finite temperature Huziel E Sauceda, Valentin Vassilev Galindo, Stefan Chmiela, Klaus-Robert Müller, Alexandre Tkatchenko Nuclear quantum effects (NQE) tend to generate delocalized dynamics due to the inclusion of the zero-point energy and its coupling to interatomic interactions' anharmonicities.[1] Here, we present evidence that NQE often enhance electronic interactions and can result in dynamical molecular stabilization at finite temperature. The underlying physical mechanism promoted by NQE depends on the particular system under consideration. First, the effective reduction of interatomic distances between functional groups within a molecule can enhance the n → π* interaction by increasing the overlap between molecular orbitals. Second, NQE can localize methyl rotors by temporarily changing molecular bond orders and leading to the emergence of localized transient states. Third, the nuclear quantum dilation induced by the NQE increases the molecular polarizability which hence results in a strengthening of van der Waals interactions.[1] The implications of these boosted interactions include counterintuitive hydroxyl–hydroxyl bonding, hindered methyl rotor dynamics, stronger molecule-surface interactions, and molecular stiffening which generates smoother free-energy surfaces. Our findings yield new insights into the paramount role of nuclear quantum fluctuations in molecules and materials. |
Wednesday, March 16, 2022 2:06PM - 2:18PM |
N49.00012: Quantum paraelectricity and structural phase transitions of SrTiO3 by on-the-fly machine-learned interatomic potentials Carla Verdi, Luigi Ranalli, Cesare Franchini, Georg Kresse Strontium titanate (SrTiO3) is a versatile building block for a variety of technologies and a ubiquitous playground for studying emergent phenomena in complex oxide materials and heterostructures. Central to many of its remarkable properties is the strongly anharmonic lattice dynamics, including competing ferroelectric and antiferrodistortive instabilities and the suppression of ferroelectricity by quantum fluctuations at low temperature. Here we employ machine-learned interatomic potentials trained on the fly [1] in combination with the stochastic self-consistent harmonic approximation (SSCHA) [2] to fully capture quantum and anharmonic effects and their temperature dependence, with the accuracy of the underlying first-principles description of the potential energy surface. We investigate the cubic to tetragonal transition characterized by the softening of the antiferrodistortive mode, and we further show that the anharmonic quantum fluctuations stabilize the paraelectric phase. This approach enables detailed studies of emergent properties in anharmonic materials including quantum paraelectrics. |
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