Bulletin of the American Physical Society
APS March Meeting 2021
Volume 66, Number 1
Monday–Friday, March 15–19, 2021; Virtual; Time Zone: Central Daylight Time, USA
Session C22: Emerging Trends in MD Simulations and Machine Learning IIFocus Live
|
Hide Abstracts |
Sponsoring Units: DCOMP GDS DSOFT DPOLY Chair: Priya Vashishta, Univ of Southern California |
Monday, March 15, 2021 3:00PM - 3:12PM Live |
C22.00001: Backmapping of Equilibrated Condensed-Phase Molecular Structures with Generative Adversarial Networks Marc Stieffenhofer, Michael Wand, Tristan Bereau A tight and consistent link between resolutions is crucial to further expand the impact of multiscale modeling for complex materials. We herein tackle the generation of condensed molecular structures as a refinement---backmapping---of a coarse-grained structure. Traditional schemes start from a rough coarse-to-fine mapping and perform further energy minimization and molecular dynamics simulations to equilibrate the system. In this study we introduce DeepBackmap: A deep neural |
Monday, March 15, 2021 3:12PM - 3:24PM Live |
C22.00002: Frequency dependence of W made simple using a multi-pole approximation Dario Leon Valido, Claudia Cardoso, Daniele Varsano, Elisa Molinari, Andrea Ferretti
|
Monday, March 15, 2021 3:24PM - 3:36PM Live |
C22.00003: Learning electron densities in condensed-phase space Alan Lewis, Andrea Grisafi, Michele Ceriotti, Mariana Rossi The electron density is a fundamental quantity for modelling and understanding physical phenomena in materials. Not only is it the central quantity of theories like density-functional theory, but it also allows the calculation of a wide range of observables that are either directly or indirectly connected to it, like total energies, dipole moments, the electrostatic potential, work functions, and others. In this work, we present a model that is able to learn and predict the electronic density of diverse materials, ranging from liquids to solid semiconductors and metals. This is achieved by extending the framework presented by Fabrizio et al (Chem. Sci., 10, 9424, 2019) to work with periodic boundary conditions and numeric atom-centred orbitals in the FHI-aims code, where a resolution of the identity is used in order to obtain coefficients for the expansion of the periodic density. This density is learned using a Gaussian process regression model with local symmetry-adapted representations of the atomic structure, making our method both data-efficient and highly transferable. We discuss the applicability of this model for large-scale periodic systems and its transferability across the periodic table. |
Monday, March 15, 2021 3:36PM - 3:48PM Live |
C22.00004: Biased, efficient sampling of polymer conformations using Brownian bridges Vivek Narsimhan, Shiyan WANG, Doraiswami Ramkrishna In this talk, we introduce a mathematical concept known as a Brownian bridge, and describe how it can be utilized in many areas of polymer physics. A Brownian bridge is a random process whose start and end regions in phase space are specified. Such processes naturally find utility in situations when one wants to sample polymer conformations with a given topology and/or energy. In the first part of the talk, we expand upon our previous work and discuss how one can systematically generate bridge processes for continuous polymer chains described by a stochastic differential equation. We will then discuss how to use such ideas to generate polymer conformations of a given topology (e.g., rings, polymers with fixed winding or twist, etc.). We will then study a polymer under an external field and show how a bridge formulation allows one to exactly sample polymer conformations in a given range of total energy. This latter idea thus allows one to sample rare events (i.e., high energy) efficiently, or conversely most probable configurations (i.e., low energy). We will conclude on how to to scale these ideas to larger dimensional systems through approximation methods, and discuss some advantages and disadvantages of using bridge processes compared to other biased sampling strategies. |
Monday, March 15, 2021 3:48PM - 4:00PM Live |
C22.00005: Flexible Molecules Need More Flexible Machine Learning Force Fields Valentin Vassilev Galindo, Grgory Cordeiro Fonseca, Igor Poltavskyi, Alexandre Tkatchenko Robust machine learning (ML) models should reliably predict molecular potential-energy surfaces (PES), including equilibrium and transition regions, using limited amounts of reference data. We assess the performance of state-of-the-art ML models, namely sGDML, SchNet, GAP/SOAP, and BPNN, in solving this task on an example of cis to trans thermal relaxation in molecular switches exemplified by the azobenzene molecule. Our results demonstrate difficulties all the employed methods face in this task. The local ML models, GAP/SOAP, and BPNN show large errors caused by the limitations of descriptors in learning long-range interactions. The global model, sGDML, demonstrates the best accuracy, but it strongly depends upon the training set and the choice of descriptor. Moreover, the optimal descriptor is different for different transition mechanisms. Finally, SchNet is the most overall reliable model, but its prediction errors vary for different parts of the PES. Our findings reveal that constructing accurate and data-efficient ML force fields is still an open challenge, requiring further developments. To resolve this, we propose moving from learning the entire PES within a single ML model to the employment of local models that are combined into a global force field. |
Monday, March 15, 2021 4:00PM - 4:12PM Live |
C22.00006: Network structure of non-equilibrium quantum transport models. Abigail Poteshman, Lee Bassett, Danielle Bassett Different models of a physical system are often employed based on requirements for accuracy and which specific phenomena are being examined. In the emerging field of quantum network science, a key challenge is understanding the relationships between network structure and underlying physical descriptions of many-body quantum systems. In this work, we construct networks representing the energy landscape of non-equilibrium transport through quantum antidots—an example of an open, many-body quantum system—corresponding to two distinct models of internal quantum states: a single-particle, non-interacting model and a mean-field model including interactions. We find that the cycle structure reflects common spin conservation rules, but the two models result in different minimum lengths of cycle basis elements, reflecting the (in)distinguishability of particles in the models. Furthermore, spin-conserving, internal energy relaxation produces topological discrepancies in the degree distribution and the length distribution of cycles in the cycle bases across the two models. Our approach motivates future efforts to use network science to understand the dynamics of quantum systems with potential applications to quantum information technologies. |
Monday, March 15, 2021 4:12PM - 4:24PM Live |
C22.00007: Two-tier machine learning acceleration of molecular dynamics with enhanced sampling: surface reactions and restructuring on metal catalysts Lixin Sun, Simon Batzner, Wei Che, Jin Soo Lim, Yu Xie, Steven Torrisi, Jonathan Vandermause, Boris Kozinsky Efficient molecular dynamics(MD) are critical for energy landscape exploration and reaction free energy computation. For heterogeneous catalysis, it is prohibitive to directly compute these reactions with ab initio molecular dynamics. To solve this problem, we introduce a two-tier machine learning approach to accelerate MD simulations. First, a single point calculation is accelerated by replacing DFT force calculations with the Tensor-Field Neural Network force field. Second, reaction coordinates learned are learned with multi-task neural networks and are employed to guide enhanced sampling to further accelerate the estimation of free energy barriers. This framework is applied to model formate dehydrogenation, a key reaction in fuel cells running with formic acid. Au(110) and Cu(110) surfaces are chosen as the model catalysts. The simulations sample free energy landscape and reveal how different initial formate coverages affect surface restructuring of the catalysts. |
Monday, March 15, 2021 4:24PM - 4:36PM Live |
C22.00008: Generalization of SNAP to arbitrary machine-learning interatomic potentials in LAMMPS Aidan Thompson SNAP is an automated methodology for generating accurate and robust application-specific machine-learning interatomic potentials (MLIAPs) in LAMMPS. The MLIAP package generalizes SNAP to arbitrary MLIAPs. This is accomplished by separating the energy model (e.g. linear, non-linear, Gaussian process, neural network) from the local atomic neighborhood descriptors (e.g. ACE, Behler-Parrinello, DeepPot, SNAP, SOAP). Any new model added to the MLIAP package can be combined with any existing descriptor to compute energy and forces, and vice versa. Gradients of energy and forces w.r.t. model parameters can also be computed for training MLIAPs against ab initio data. I will discuss the underlying algorithms and describe some interesting applications. |
Monday, March 15, 2021 4:36PM - 4:48PM Not Participating |
C22.00009: Quantum Dynamics Made Fast: Achieving Linear Time-Scaling for Nonequilibrium Green Functions Niclas Schlünzen, Jan-Philip Joost, Michael Bonitz The accurate description of the nonequilibrium dynamics in correlated quantum-many-body systems remains to be a driving force for current research in condensed-matter physics and beyond. |
Monday, March 15, 2021 4:48PM - 5:24PM Live |
C22.00010: Modeling the Dynamics of Complex Energy Materials with Machine Learning Invited Speaker: Nongnuch Artrith Materials for energy applications, e.g., heterogeneous catalysts and battery materials, often exhibit complicated chemical compositions, defects, and disorder, making the direct modeling with first principles methods challenging. Machine-learning (ML) potentials trained on first principles data enable computationally efficient linear-scaling atomistic molecular dynamic simulations with an accuracy close to the reference method. Here, I will give an overview of recent methodological advancements of ML potentials based on artificial neural networks (ANNs) [1-5] and applications of the method to challenging materials classes including metal and oxide nanoparticles and amorphous phases. |
Monday, March 15, 2021 5:24PM - 5:36PM Live |
C22.00011: Accurate many-body repulsive potentials for density-functional tight binding from deep tensor neural networks Leonardo Medrano Sandonas, Martin Stoehr, Alexandre Tkatchenko Machine learning (ML) has been proven to be an extremely valuable tool for simulations with ab initio accuracy at the computational cost between classical interatomic potentials and density-functional approximations. Similar efficiency can only be achieved by semi-empirical methods, such as density-functional tight-binding (DFTB). One of the limiting factors in terms of the accuracy and transferability of DFTB parametrizations is the so-called repulsive potential, which plays a considerable role for the prediction of energetic, structural, and dynamical properties. Few attempts of using ML-techniques to address this issue have been proposed recently but, up to now, evidence of transferability and scalability is still scarce. Hence, we combine DFTB with deep tensor neural networks (DTNN) to maximize the strengths of both approaches. The DTNN is used to construct a non-linear model for the localized many-body interatomic repulsive energy, substantially improving upon standard DFTB and DTNN. The resulting DFTB+DTNN model yields accurate predictions of several physicochemical properties for a large variety of organic molecules compared to the hybrid DFT-PBE0 functional. DFTB+DTNN thus opens a route to the fast access to reliable property calculation of diverse molecular systems. |
Monday, March 15, 2021 5:36PM - 5:48PM Live |
C22.00012: Improving Molecular Force Fields Across Configurational Space by Combining Supervised and Unsupervised Machine Learning Grgory Cordeiro Fonseca, Igor Poltavskyi, Valentin Vassilev Galindo, Alexandre Tkatchenko The training set is as key as the choice of Machine Learning (ML) model itself for the range of applicability and accuracy of the ML model. However, most atomistic reference datasets inherit the inhomogeneous distribution across configurational space (CS) from the MD trajectories. Thus, choosing the training set randomly or according to the probability distribution of the data leads to biased models, whose prediction errors on specific regions of CS can easily exceed the mean value by a factor of three. |
Monday, March 15, 2021 5:48PM - 6:00PM Live |
C22.00013: Central Moment Lattice Boltzmann Schemes with Fokker-Planck Guided Collision for Simulation of Multiphase Flows with Surfactant Effects and Turbulence William Schupbach, Kannan Premnath We present central moment lattice Boltzmann (LB) schemes, whose collision steps are represented by a novel Fokker-Planck (FP) kinetic model, for computations of multiphase hydrodynamics, interface tracking and surfactant evolution. Our approach involves matching the changes in different discrete central moments under collision to those of the continuous central moments as given by the drift-diffusion based FP model of the Boltzmann equation. This effectively results in relaxations to “equilibria”, the Markovian central moment attractors, which depend on the lower order moments. Based on this, a LB scheme for the pressure and velocity fields in multiphase flows at high density ratios is constructed. The interface motions are captured using the Allen-Cahn equation computed by another FP-guided LB scheme. The surfactant concentration is evolved by a model that accounts its preferential adsorption on interfaces, which is solved by a third LB scheme using FP collision. The Langmuir isotherm then parameterizes the effect of surfactant concentration on surface tension and its tangential gradients. Various simulations involving the effect of surfactant concentrations and turbulence on multiphase systems are presented and the advantages of using the FP model are shown. |
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