Bulletin of the American Physical Society
APS March Meeting 2019
Volume 64, Number 2
Monday–Friday, March 4–8, 2019; Boston, Massachusetts
Session K21: Emerging Trends in Molecular Dynamics Simulations and Data Analytics IFocus
|
Hide Abstracts |
Sponsoring Units: DCOMP Chair: Rajiv Kalia, University of Southern California Room: BCEC 157B |
Wednesday, March 6, 2019 8:00AM - 8:36AM |
K21.00001: Reactive molecular dynamics simulations and machine learning Invited Speaker: Priya Vashishta Machine learning (ML) is revolutionizing scientific and engineering disciplines owing to its ability to capture hidden patterns in large amounts of data. The recent success of ML can be attributed to increasing amount of data, simulation resources, and improving understanding of statistical inference. For these reasons computational materials science is undergoing a paradigm shift. The main reason is that trial-and-error approach to materials design is inefficient: laboratory trials require a lot of time, and the results of previous trials are not utilized in a systematic fashion. A data-driven approach, which draws upon all relevant data from experiments, and reactive and quantum molecular dynamics simulations, can address these issues. The MAGICS (Materials Genome Innovation for Computational Software) Center develops to aid the synthesis of stacked layered materials by chemical vapor deposition, exfoliation, and intercalation. The identification of different phases can be formulated as a classification problem and can be solved using ML techniques. We have used feed-forward neural network with three hidden layers to identify the different phases present during computational synthesis of MoSe2. Work reported here was carried out in collaboration with Rajiv K. Kalia, Aiichiro Nakano, Lindsay Bassman, Sungwook Hong, Aravind Krishnamoorthy, Kuang Liu, Ankit Mishra, Ken-ichi Nomura, and Pankaj Rajak, |
Wednesday, March 6, 2019 8:36AM - 9:12AM |
K21.00002: High-Dimensional Neural Network Potentials for Complex Systems Invited Speaker: Jörg Behler In recent years, machine learning potentials have become a promising new approach for the representation of high-dimensional reactive potential-energy surfaces [1]. After training to a set of reference energies and forces obtained from electronic structure calculations they allow to perform large-scale simulations with first-principles accuracy at a fraction of the computational costs. As they do not contain any system-specific terms, they are applicable to a wide range of problems in chemistry, physics and materials science. |
Wednesday, March 6, 2019 9:12AM - 9:24AM |
K21.00003: Nanoindentation on Monolayer Kirigami MoS2 Beibei Wang, Rajiv Kalia, Aiichiro Nakano, Priya Vashishta Kirigami is an ancient technique that alters the structure and flexibility of materials. This presentation will focus on molecular dynamics (MD) simulations of mechanical properties of atomically thin kirigami MoS2 membranes. Two kirigami structures — rectangular and hexagonal patterns — are studied in indentation simulations. Our simulations reveal dramatic changes in the ductility of monolayer kirigami MoS2 compared with regular MoS2 monolayer. We observe the inelastic response of kirigami MoS2 membranes under indentation. We also identify the role of defects in the inelasticity of kirigami MoS2. Results for other mechanical properties of kirigami MoS2 will also be presented. |
Wednesday, March 6, 2019 9:24AM - 9:36AM |
K21.00004: Theoretical Studies of Water by Climbing Jacob’s Ladder with Deep Learning Mohan Chen, Linfeng Zhang, Han Wang, Jianhang Xu, Hsin-Yu Ko, Biswajit Santra, John P Perdew, Weinan E, Xifan Wu
|
Wednesday, March 6, 2019 9:36AM - 9:48AM |
K21.00005: Machine Learning Polarizable Force Field Parameters Ying Li, Hui Li, Frank Pickard, Badri Narayanan, Subramanian Sankaranarayanan, Maria Chan, Benard Brooks, Benoit Roux Machine learning (ML) techniques with the genetic algorithm (GA) have been applied to determine a polarizable force field parameters using quantum mechanics (QM) data of molecular clusters at the MP2/6-31G(d,p), DFMP2(fc)/jul-cc-pVDZ, and DFMP2(fc)/jul-cc-pVTZ levels to predict experimental condensed phase properties (i.e., density and heat of vaporization). The performance of this ML/GA approach is demonstrated on 4943 dimer electrostatic potentials and 1250 cluster interaction energies for methanol. Excellent agreement between the training data set from QM calculations and the optimized force field model was achieved. The present effort shows the possibility of using machine learning techniques to develop descriptive polarizable force field using only QM data. The ML/GA strategy to optimize force fields parameters described here could easily be extended to other molecular systems. |
Wednesday, March 6, 2019 9:48AM - 10:00AM |
K21.00006: Hot spot formation and shock initiation of RDX Ankit Mishra, Ken-ichi Nomura, Aiichiro Nakano, Rajiv Kalia, Priya Vashishta Hot spot formation mechanism in energetic materials (EM) is crucial to handling and design of safer explosives. Presence of defects such as cracks, voids and grain boundaries are mainly responsible for creation of extreme conditions leading to hot spots. Recent experimental studies have shown that the presence of filled and empty void in EMs are responsible for creation of these regions. Here, we present a million-atom reactive molecular dynamics study to investigate the nature of hot spots during shock compression of RDX in gas filled and empty voids. We observe higher and sustained temperature rise in gas filled voids as compared to empty voids at particle velocities ranging from 2-4 km/sec. Furthermore, we observe that the higher amount of heat release on account of high potential energy drop in gas-filled voids correlate positively to more stable fragment formation, hence validating its role in high impact shock initiation of RDX. |
Wednesday, March 6, 2019 10:00AM - 10:12AM |
K21.00007: Towards Exact Molecular Dynamics Simulations with Machine-Learned Force Fields Stefan Chmiela, Huziel Sauceda, Klaus-Robert Müller, Alexandre Tkatchenko The predictive power of molecular dynamics (MD) simulations hinges on the accuracy of the underlying interatomic potential, however ubiquitous classical force fields are typically challenged by quantum effects. |
Wednesday, March 6, 2019 10:12AM - 10:24AM |
K21.00008: Atomic Scale Response of Electrochemical Systems Under Potential Bias James Goff In the past decade, there has been considerable progress in the simulation of solid-solution |
Wednesday, March 6, 2019 10:24AM - 10:36AM |
K21.00009: Constructing Accurate Machine Learning Force Fields for Flexible Molecules Valentin Vassilev Galindo, Igor Poltavskyi, Alexandre Tkatchenko Machine learning (ML) models can reproduce potential energy surfaces (PES) for molecules containing up to a few tens of atoms with an accuracy comparable to the most exact ab initio methods. This provides a tool for computing thermodynamic properties that would require millions of CPU years otherwise. For instance, a recently developed sGDML1 model predicts forces and energy with CCSD(T) accuracy using just a few hundreds of configurations for training. However, up to now ML has been mainly applied to rather rigid molecules. In this regard, our objective is to test ML models for flexible molecules and out-of-equilibrium configurations along transition paths. For this, we select molecules (e.g. azobenzene, stilbene) with relatively complex transition paths, which result from an interplay between long- and short-range interactions. Then, different paths connecting PES minima are tested using sGDML. This allows us to define optimal descriptors and the appropriate strategies for choosing the training sets, which is crucial for ML models relying on a limited number of training points. Our results open an avenue for calculating transport paths, transition rates and other "out-of-equilibrium” properties with previously unattained accuracy. |
Wednesday, March 6, 2019 10:36AM - 10:48AM |
K21.00010: Machine Learning for Auto-tuning of Simulation Parameters in Car-Parrinello Molecular Dynamics Jayanath Chamindu Kadupitige, Geoffrey C Fox, Vikram Jadhao Simulating the dynamics of ions near polarizable nanoparticles (NPs) is challenging due to the need to solve the Poisson equation at every simulation timestep. Car-Parrinello Molecular dynamics (CPMD) simulations based on a dynamical optimization framework can bypass this obstacle by representing the polarization charge density as virtual dynamic variables, and evolving them in parallel with the physical dynamics of ions. Using these CPMD simulations of ions near polarizable NPs, we demonstrate the computational gains accessible by integrating machine learning (ML) for parameter prediction in CPMD simulations. An artificial neural network based regression model was integrated with CPMD and it predicted the optimal simulation timestep and critical parameters characterizing the virtual system on-the-fly with 94.3% accuracy. The ML-enhanced, hybrid OpenMP/MPI parallelized, CPMD simulations generated stable and accurate dynamics of thousands of ions in the presence of polarizable NPs for over 10 million steps (over 30 ns) with walltime reducing from thousands of hours to tens of hours yielding a maximum speedup of ~600. |
Wednesday, March 6, 2019 10:48AM - 11:00AM |
K21.00011: Parallel Trajectory Splicing and the Value of Information Andrew Garmon, Danny Perez A range of specialized Molecular Dynamics (MD) methods have been developed in order to overcome the challenge of reaching longer timescales in systems that evolve through sequences of rare events. In this talk, we consider Parallel Trajectory Splicing (ParSplice) which works by generating large number of trajectory segments in parallel in such a way that they can later be assembled into a single statistically correct state-to-state trajectory, enabling parallel speedups up to the number of parallel workers. In practice, the ability for ParSplice to scale significantly improves when it is possible to predict where the trajectory will be found in the future. With this insight in mind, we develop a maximum likelihood transition model that is updated on the fly and show how the value of the information gained from generating a segment can rigorously be taken into consideration in order to improve performance. |
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