Symposium Organizers
Enrique Martinez, Los Alamos National Laboratory
Graeme Henkelman, University of Texas
Hannes Jonsson, University of Iceland
Steven Kenny, Loughborough University
Symposium Support
Modelling and Simulation in Materials Science and Engineering | IOP Publishing
TC04.01: First Principles I
Session Chairs
Enrique Martinez
Danny Perez
Monday PM, November 27, 2017
Hynes, Level 2, Room 202
8:30 AM - *TC04.01.01
Advances in Orbital-Free Density Functional Theory Simulations of Materials
Beatriz del Rio 1 , William Witt 1 , Johannes Dieterich 1 , Emily Carter 2
1 Department of Mechanical and Aerospace Engineering, Princeton University, Princeton, New Jersey, United States, 2 School of Engineering and Applied Science, Princeton University, Princeton, New Jersey, United States
Show AbstractOrbital-free density functional theory (OFDFT) simulations are orders of magnitude faster than those based on Kohn-Sham density functional theory because of the theoretical simplicity of the former that merely involves variationally optimizing the electron density rather than wavefunctions. However, the lack of one-electron wavefunctions (orbitals) in practice limits the accuracy of OFDFT because of the need to employ (i) a kinetic energy density functional (KEDF) to evaluate the kinetic energy of the electrons and (ii) normally a local pseudopotential (LPS) to account for the electron-ion (screened nucleus) interaction. We start with a bit of history of our earlier contributions to improve the accuracy, generality, and speed of OFDFT implementations, followed updates on recent advances, including a global optimization algorithm to efficiently construct LPSs, an open-source library of KEDFs that work with GPUs as well as CPUs, and new approaches to KEDFs and angular-momentum-dependent OFDFT.
9:00 AM - *TC04.01.02
High-Throughput Materials Discovery and Development—Breakthroughs and Challenges in the Mapping of the Materials Genome
Marco Buongiorno Nardelli 1
1 , University of North Texas, Denton, Texas, United States
Show AbstractHigh-Throughput Quantum-Mechanics computation of materials properties by ab initio methods has become the foundation of an effective approach to materials design, discovery and characterization. This data driven approach to materials science currently presents the most promising path to the development of advanced technological materials that could solve or mitigate important social and economic challenges of the 21st century. In particular, the rapid proliferation of computational data on materials properties presents the possibility to complement and extend materials property databases where the experimental data is lacking and difficult to obtain.
Enhanced repositories such as AFLOWLIB open novel opportunities for structure discovery and optimization, including uncovering of unsuspected compounds, metastable structures and correlations between various properties. The practical realization of these opportunities depends almost exclusively on the the design of efficient algorithms for electronic structure simulations of realistic material systems beyond the limitations of the current standard theories. In this talk, I will review recent progress in theoretical and computational tools for data generation and advanced characterization, and in particular, discuss the development and validation of novel functionals within Density Functional Theory and of local basis representations for effective ab-initio tight-binding schemes.
9:30 AM - TC04.01.03
A Quantum-Accurate Force Field for Molybdenum from Machine Learning of Large Materials Data
Chi Chen 1 , Zhi Deng 1 , Richard Tran 1 , Hanmei Tang 1 , Iek-Heng Chu 1 , Shyue Ping Ong 1
1 Nanoengineering, University of California, San Diego, La Jolla, California, United States
Show AbstractIn this talk, we will present a highly accurate force field for molybdenum (Mo) developed through the rigorous application of machine learning techniques on large materials data sets. Molybdenum is one of the most important structural metals, and is valued for its ability to withstand high temperatures, its high corrosion resistance, and its excellent strength-to-weight ratio. Despite its importance, currently available force-fields for Mo based on the embedded atom (EAM) and modified embedded atom methods (MEAM) still do not provide satisfactory accuracy on many properties. We will show that by fitting simultaneously the energies, forces and stress tensors of a large density functional theory (DFT)-computed dataset on a diverse set of Mo structures, e.g., lattice deformations, surface structures, grain boundaries, NVT/NPT ab initio molecular dynamics trajectories of solid and liquid Mo, etc., a highly accurate force field for Mo can be developed. We will outline a systematic approach to structural selection based on principal component analysis, as well as a novel differential evolution algorithm for optimizing the hyperparameters in the model fitting so that both the model error and the property prediction error can be simultaneously lowered. We will demonstrate that this force field can successfully predict a range of properties, such as energies, forces, elastic constants, melting point, phonon spectra, surface energies, etc. with accuracy close to that of DFT computations. We expect that this newly developed Mo force field will find broad application in large-scale, long-time scale simulations, for example, in the study of Mo grain boundaries.
9:45 AM - TC04.01.04
Advanced Understanding on Dielectric Breakdown from Quantum Kinetic Monte Carlo Simulations
Fei Lin 1 , Jianqiu Huang 1 , Celine Hin 1
1 , Virginia Tech, Blacksburg, Virginia, United States
Show AbstractWe propose for the first time an exact Quantum Kinetic Monte Carlo method to calculate electron transport for real materials. The method is formulated in a Hubbard model, which contains electron hopping integrals and interactions. The method allows us to calculate real time dynamics of electrons, such as electrical conductivity. When coupled with Density Functional Theory and Maximally Localized Wannier Functions, our method can be used to predict dielectric breakdown [1].
Our code has been applied to study the dielectric breakdown in three systems. In a crystalline α-quartz and amorphous quartz, we observe the increase of electron hopping integrals as we increase the applied electric field in the system. We also determine the critical hopping integral value, at which the α-quartz and silica undergo a dielectric breakdown and become conducting. In the amorphous quartz system, a larger dielectric breakdown voltage is expected, since our calculation shows that disorder in the system yields a smaller hopping integral compared to the clean crystalline quartz. Finally, we calculate electron transport property for an Al/amorphous-SiO2 interface system. The Al part transfers electron to amorphous SiO2, resulting in a virtual thinning of the oxide layer. Our increased hopping integrals compared with the amorphous system suggests that this oxide layer thinning close to the interface leads to increased tunneling of electrons in the system.
[1] Fei Lin, Jianqiu Huang, and Celine Hin, Electron transport from Quantum Kinetic Monte Carlo simulations, arXiv: 1704.07545
10:30 AM - *TC04.01.05
Atomic-Resolution Simulation of Real-Timescale Deformation Processes
Ju Li 1
1 , Massachusetts Institute of Technology, Cambridge, Massachusetts, United States
Show AbstractPoint defects and dislocations facilitate diffusive and displacive deformation processes in materials. We study their complex interactions and rate characteristics using advanced algorithms such as strain-boost hyperdynamics [Phys. Rev. B 82 (2010) 184114] and diffusive molecular dynamics (DMD) [PRB 84 (2011) 054103; PRB 86 (2012) 014115]. These simulations are checked against experimental observations of stress-driven processes using transmission electron microscopy and nanomechanical measurements. Diffusive-displacive transitions and coupling are revealed from the simulations.
11:00 AM - TC04.01.06
Density Functional Tight Binding Simulations of Reaction Chemistry in Nitromethane with LATTE-LAMMPS
Romain Perriot 1 , Marc Cawkwell 1 , Shawn McGrane 1
1 , Los Alamos National Laboratory, Los Alamos, New Mexico, United States
Show AbstractModeling fundamental chemistry is crucial for understanding the properties of organic materials under extreme conditions. Critical to the modeling of reaction chemistry is the description of electronic interactions, which dictate the processes of bond breaking and remaking. At the same time, we require large systems on sufficiently long timescales in order to compare to experimental results, which prevents the use of the expensive density functional theory (DFT). However, density functional tight binding (DFTB) allows to simulate systems with hundreds of atoms, for hundreds of picoseconds, while accurately modeling electronic interactions. Recently, the LATTE DFTB code, developed at LANL, was integrated to LAMMPS, allowing to perform electronic-level calculations while benefitting from the numerous resources available in LAMMPS. We demonstrate the use of LATTE-LAMMPS by performing nudged-elastic band (NEB) calculations for fundamental reaction pathways and Hugoniostat calculation on nitromethane, a prototypical and relatively simple organic compound.
11:15 AM - *TC04.01.07
From Linear Scaling Density Functional Theory Calculation to Atomic Force Field Fitting
Lin-Wang Wang 1
1 , Lawrence Berkeley National Lab, Berkeley, California, United States
Show AbstractIn this talk, I will present several of our recent methods used for large scale atomic simulations. The first is the linear scaling three dimensional fragment (LS3DF) method which has been used by our group for the last several years. Recently, we have implemented the GPU version of the code, which allows it to calculate systems with tens of thousands of atoms at the plane wave density functional theory (DFT) level within half hour. I will present a few examples of such calculations and discuss the challenges in such an approach. Secondly, I will discuss a method which uses classical force field to accelerate the DFT atomic relaxation. DFT based atomic relaxations consists a major portion of overall material simulations. Yet the relaxation can be agonizingly slow taking hundreds of steps especially when the system becomes large. I will show that, force field calculation can be used to accelerate the DFT atomic relaxations, improving the convergence by a factor of 3-4. This is at no additional cost in term of computational time. Finally, I will discuss our recent work on automatic fitting of force fields based on data generated from molecular dynamics. This follows the new trend of using machine learning to develop the force field. If such approach becomes reliable and accurate, many large systems can be simulated easily in the future.
This work was support by the Director, Office of Science, the Office of Basic Energy Science (BES), Materials Sciences and Engineering (DOE) through the theory of material (KC2301) program under contract DE-AC02-05CH11231.
11:45 AM - TC04.01.08
Hybrid Quantum-Classical Simulation of Interface of Epoxy Resin and Al Metal—Moisture Effects on Adhesion
Shuji Ogata 1 , Masayuki Uranagase 1
1 , Nagoya Inst of Technology, Nagoya Japan
Show AbstractDue to the relatively long computation timings of DFT, the hybrid quantum (QM)-classical (CL) simulation has been limited to the settings of small QM-regions (a few hundred atoms in 0D or 1D geometry). Recent development of the order-N Real-space grid DFT code combined with many-core supercomputer makes it possible to apply the hybrid QM-CL method to inhomogeneous organic-inorganic interfaces as resin-metal one with large (thousands of atoms in 2D) QM-regions.
Increasing industrial demand exists to bond dissimilar materials using epoxy resin for advanced manufacturing of various devices and systems as an automobile and its parts. One of the principal issues is to understand the interface bonding mechanisms between the epoxy resin and metal. Adhesion strength between metal and epoxy resin is well known to reduce significantly in a moist environment. The FTIR spectroscopy has shown that 15-30% of the original epoxide groups remain near the adherend in the resin. We therefore address the issue address the issue from two directions by considering two extreme models of resin: 100% of epoxide groups remain in the one model, while 0%, in the other model. In both models, we calculate the shear strength of the interfacial adhesion between the surface oxidized Al and bisphenol-A epoxy resin with a varying number of water molecules or hydroxide ions inserted inbetween using the hybrid QM-CL simulation method. The QM region, which is under the 2D-PBC, is composed of one to two thousand atoms at the interface.
Through the simulation runs, we find that the adhesion strength in the 100% model is an order of magnitude larger than that in the 0% model. In the 100% model, microscopic analyses find the following key features; (i) The inter-atomic Al-O bond between the Al atom of the oxide and the O atom of the epoxide group in the epoxy resin contributes substantially to the strength of the interfacial adhesion. (ii) Dissociation of the O atom of the epoxide group forming such an Al-O bond is enhanced when an H2O rather than an OH resides in close proximately to the bond. (iii) The epoxy resin becomes more plastic by incorporating H2O molecules. Both (ii) and (iii) act to weaken the shear strength of the interfacial adhesion.
TC04.02: First Principles II
Session Chairs
Steven Kenny
Romain Perriot
Monday PM, November 27, 2017
Hynes, Level 2, Room 202
1:30 PM - *TC04.02.01
Accurate and Efficient Molecular Dynamics with N-Body Machine Learning Force Fields
Aldo Glielmo 4 , Claudio Zeni 4 , Peter Sollich 4 , Federico Bianchini 3 , James Kermode 2 , Alessandro De Vita 4 1
4 , King's College London, London United Kingdom, 3 , University of Oslo, Oslo Norway, 2 , University of Warwick, Warwick United Kingdom, 1 Engineering and Architecture, University of Trieste, Trieste Italy
Show AbstractMachine learning (ML) force fields that try to reproduce reference QM forces in molecular and solid systems are becoming increasingly available, although their use is not as yet fully mainstream. Much of the appeal of these force fields is associated with their not being based on heavily parametrised functional forms that have to be carefully tuned for each system and may still not extrapolate well to chemically complex situations [1]. Another key feature of ML force fields is that they offer a natural way to tackle the validation problem, since the expected error on the predicted forces on atoms can be easily calculated. This makes it e.g., possible to compute new QM information (only) if needed, to avoid any extrapolation, using online learning (“on the fly”) techniques [2]. Finally, two key issues relevant for practical applications are (i) whether ML force fields could be at the same time as fast and more accurate than traditional force fields, or alternatively (ii) if ML force fields could be trained to be systematically more accurate than standard force fields without becoming impractically slower. In this talk I will review and elaborate on the above, and discuss how these issues described could be explored and possibly resolved by Gaussian process regression techniques using “covariant” force kernels [3] equivalent to “n-body” interatomic potentials, and/or more resolving kernels constructed from these.
[1] F. Bianchini, J.R. Kermode, and A. De Vita, Modell. Simul. Mater. Sci. Eng. 24, 045012 (2016).
[2] Z. Li, J. R. Kermode and A. De Vita, Phys. Rev. Lett., 114, 096405 (2015).
[3] A. Glielmo, P. Sollich, and A. De Vita, Physical Review B 95, 214302 (2017).
2:00 PM - TC04.02.02
Feature Engineering for First Principles Atomistic Hamiltonians
John Thomas 1 , Anton Van der Ven 1
1 , University of California, Santa Barbara, Santa Barbara, California, United States
Show AbstractAtomistic Hamiltonians form an important link between computationally intensive first principles calculations and predictive simulation methods at larger length scales. Atomistic Hamiltonians can be used within Monte Carlo or molecular dynamics simulations to calculate free energies and constitutive relations of materials, thus providing feed-forward inputs for predictive phase-field or finite element methods. Unfortunately, a significant bottleneck of this paradigm lies in constructing features that represent relevant atomistic degrees of freedom while also respecting all symmetries of the system.
In this talk, I will describe recent progress towards constructing atomistic Hamiltonians that express the potential energy or crystalline materials in terms of features that represent both conventional and unconventional degrees of freedom while employing group theoretical concepts to ensure that all crystal and intrinsic symmetries are obeyed. These features are capable of describing anharmonic lattice dynamical properties, dynamically-stabilized phases, and aggregated cooperative degrees of freedom. I will 1discuss the application of these approaches towards predicting structural phase transitions and constitutive relations in thermoelectric semiconductors, metal oxides and metal hydrides.
2:15 PM - *TC04.02.03
Charge Density Based Embedding for QM/MM Modeling of Metals
Gang Lu 1
1 , California State University, Northridge, Northridge, California, United States
Show AbstractMaterials properties are to a large extent determined by lattice defects. In order to model the defects, multiscale methods such as QM/MM are often required. However, QM/MM simulations for metals are particularly challenging owing to the existence of delocalized electronic states at QM/MM interface, which renders standard QM/MM coupling approaches such as cutting/saturating covalent bonds across the interface ineffective or invalid. To address this challenge, we have developed two QM/MM methods for metallic systems based on charge density embedding. In the first method, the QM/MM coupling is achieved by adding an embedding potential to the Kohn-Sham (KS) Hamiltonian of the QM subsystem. This embedding potential captures the coupling between QM and MM subsystems and is defined as the functional derivative of QM/MM interaction energy with respect to charge density, evaluated by orbital-free DFT (OFDFT). In the second method, the presence of the MM subsystem is formulated by a constraint potential added to the KS Hamiltonian of the QM subsystem. The QM/MM coupling problem is recast into a constraint DFT problem for the QM subsystem. The second QM/MM method is formally exact and does not depend on OFDFT. Both QM/MM methods will be reviewed and their applications including electron wind force on defects and catalysis for core/shell nanoparticles will be presented.
2:45 PM - TC04.02.04
Computing Energy Barriers for Rare Events from QM/MM Simulations through the Virtual Work Principle
Thomas Swinburne 1 2 , James Kermode 3
1 , T-1 Group, Los Alamos National Laboratory, Los Alamos, New Mexico, United States, 2 , CCFE, UKAEA, Abingdon United Kingdom, 3 , University of Warwick, Coventry United Kingdom
Show AbstractHybrid quantum/classical techniques can flexibly couple boundary DFT simulations to an em- pirical or elastic medium to model materials systems that cannot be contained in small periodic supercells. However, due to electronic non-locality a total energy cannot be defined, meaning energy barriers cannot be calculated. We provide a general solution using the principle of virtual work in a modified nudged elastic band algorithm. Our method enables the first ab initio calculations of the kink formation energy for 〈100〉 edge dislocations in molybdenum and lattice trapping barriers to brittle fracture in silicon.
3:30 PM - *TC04.02.05
Large-Scale Real-Space Electronic Structure Calculations
Vikram Gavini 1 , Phani Motamarri 1 , Bikash Kanungo 1
1 , University of Michigan, Ann Arbor, Michigan, United States
Show AbstractIn this talk, the development of a real-space formulation for Kohn-Sham density functional theory (DFT) and a finite-element discretization of this formulation [1,2], which can handle arbitrary boundary conditions and is amenable to adaptive coarse-graining, will be presented. In particular, the accuracy afforded by using higher-order and enriched finite-element discretizations, and the efficiency and scalability of the Chebyshev filtering algorithm in pseudopotential and all-electron Kohn-Sham DFT calculations will be demonstrated. Further, the development of a subquadratic-scaling approach (in the number of electrons) based on a subspace projection and Fermi-operator expansion will be discussed [3], which will be the basis for the future development of coarse-graining techniques for Kohn-Sham DFT. The developed techniques have enabled, to date, pseudopotential calculations on ~10,000 atoms, as well as all-electron calculations on systems containing ~10,000 electrons.
[1] P. Motamarri, M.R. Nowak, K. Leiter, J. Knap, V. Gavini, Higher-order adaptive finite-element methods for Kohn-Sham density functional theory, J. Comp. Phys. 253, 308-343 (2013).
[2] B. Kanungo, V. Gavini, Large-scale all-electron density functional theory calculations using an enriched finite element basis, Phys. Rev. B 95, 035112 (2017).
[3] P. Motamarri, V. Gavini, A subquadratic-scaling subspace projection method for large-scale Kohn-Sham DFT calculations using spectral finite-element discretization, Phys. Rev. B 90, 115127 (2014).
4:00 PM - TC04.02.06
Implications of the DFT+U Method on Polaron Properties in Energy Materials
Zi Wang 1 , Casey Brock 2 , Amina Matt 3 , Kirk Bevan 1
1 , McGill University, Montreal, Quebec, Canada, 2 , Vanderbilt University, Nashville, Tennessee, United States, 3 , Ecole Polytechnique Federale de Lausanne, Lausanne Switzerland
Show AbstractMany novel materials used in clean energy applications such as lithium iron phosphate (LiFePO4), hematite (Fe2O3), and titanium dioxide (TiO2) are known to exhibit polaronic behavior. Being transition metal oxides, the strongly correlated interaction of the d shell electrons opens a gap and localizes conduction electrons into polaronic states, leading to the hopping conduction behaviour observed in these materials. Therefore, to model localized electronic conductivities of such materials from a theoretical point of view, it is necessary to calculate activation energies and other polaron-specific properties. Using an ab initio framework of density functional theory (DFT) combined with Hubbard-like on-site coulomb corrections (DFT+U) to account for the localized d shell electrons, we study the effects of varying the DFT+U on-site projection scheme on polaronic properties in two battery cathode materials (LiFePO4 and LiMn2O4) and two photovoltaic materials (Fe2O3 and TiO2). Our results show a significant dependence of polaronic properties on the on-site projection radius, while bulk properties such as the band gap and crystal structure are less strongly affected. This suggests that, while determining the value for the single parameter on-site U-term is important to predict the band gap, additional parameters such as the on-site projection radius play a significant role as well, and extensive care should be taken to maintain consistency when one wishes to predict such polaronic properties.
4:15 PM - *TC04.02.07
Calculating Alloy Phase Diagrams with Nested Sampling
Robert Baldock 1 , Christopher Sutton 2 , Luca Ghiringhelli 2 , Michael Payne 3 , Gábor Csányi 4 , Nicola Marzari 1
1 Theory and Simulation of Materials (THEOS), and National Centre for Computational Design and Discovery of Novel Materials (MARVEL), EPFL, Lausanne Switzerland, 2 , Fritz-Haber-Institut der Max-Planck-Gesellschaft, Berlin Germany, 3 Department of Physics, University of Cambridge, Cambridge United Kingdom, 4 Department of Engineering, University of Cambridge, Cambridge United Kingdom
Show AbstractThe automated calculation of complete phase diagrams, directly from a first-principles or empirical potential energy function, is one of the outstanding challenges in computational materials science. Here, we show how nested sampling, a Bayesian Markov chain Monte Carlo algorithm, can be transformed into a powerful tool for exactly this task. Since our nested sampling algorithms do not require previous information about the nature or location of phase transitions, or the atomic structures of phases formed by the material, they can be used as agnostic, black-box tools for phase diagram calculation. In this way our approach is different from methods such as thermodynamic integration and self-consistent phonons, and closer in spirit to algorithms such as Wang-Landau sampling and parallel tempering. I will begin this talk by reviewing the theory of nested sampling, and describe our recent advances in the efficient calculation of pressure-temperature phase diagrams. Next, I will introduce a new nested sampling algorithm, which enables the one-shot calculation of composition-temperature phase diagrams for alloys, including alloys that exhibit a miscibility gap whereby the material separates into domains of different composition. I will showcase the efficacy of the approach by presenting the binary phase diagrams of a Lennard-Jones alloy (continuous atomistic state space) and GaxIn1-xP as described using a lattice model (discrete atomistic state space).
4:45 PM - TC04.02.08
Grand-Canonical Minima Hopping (GCMH) Potential Energy Surface Exploration for 2D Materials and Fullerenes
Brian McGuigan 1 , Pascal Pochet 2 , Harley Johnson 1
1 Mechanical Science and Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States, 2 Atomistic Simulation Laboratory, Alternative Energies and Atomic Energy Commission (CEA), Grenoble, Auvergne-Rhône-Alpes , France
Show AbstractIdentification of global energy minima is one of the persistent challenges in atomistic modeling of materials, particularly in structures and materials where the potential energy surfaces are “soft”, with numerous local minima near the global minimum energy configuration. A variety of methods have been proposed but one common shortcoming of most approaches is that systems are constrained to a fixed number of atoms during the search process. This makes it impossible to consider certain processes, such as vacancy or interstitial mediated dislocation climb; it also may make it difficult for a search process to avoid becoming trapped in unphysical parts of the energy landscape. Here we demonstrate a new variation on a recent potential energy surface exploration method known as the Minima Hopping Method1. The Minima Hopping Method finds successively lower energy configurations in an atomistic energy description by locally adding kinetic energy to structures in order to overcome barriers between local minima. In the new variation, (i) we allow for varying numbers of atoms in the system, and (ii) we parallelize a softening process by which search directions are adjusted. We demonstrate the method on fullerenes and defect structures in graphene, showing the ability to efficiently find both known minima and new defect structures, including climb and dissociation of dislocation configurations in graphene.
1. S. Goedecker, The Journal of Chemical Physics 120, 9911 (2004)
Symposium Organizers
Enrique Martinez, Los Alamos National Laboratory
Graeme Henkelman, University of Texas
Hannes Jonsson, University of Iceland
Steven Kenny, Loughborough University
Symposium Support
Modelling and Simulation in Materials Science and Engineering | IOP Publishing
TC04.03: Atomistic
Session Chairs
Tuesday AM, November 28, 2017
Hynes, Level 2, Room 202
8:00 AM - *TC04.03.01
Molecular Dynamics—From the First Principles of Quantum Mechanics to Computer Aided Materials Design
Anders Niklasson 1
1 Theoretical Division T-1, Los Alamos National Laboratory, Los Alamos, New Mexico, United States
Show AbstractA framework for quantum-based molecular dynamics simulations is presented that provides a significant increase in the accessible time and length scales compared to previous formulations. The new framework has been developed by means of a multidisciplinary, coordinated design approach, where we reformulate the underlying equations to allow for new integration algorithms, solvers and data structures that are easily adapted to run on emerging hybrid exascale architectures. For the first time, quantum-based Born-Oppenheimer molecular dynamics appears as a realistic alternative to classical force field methods in nanosecond simulations with 10,000-100,000 atoms. The new framework opens the door to a new generation computer aided materials design and analysis, which is applicable to a broad variety of systems in chemistry, materials science, and biology.
Refs: A.M.N. Niklasson, “Extended Born-Oppenheimer molecular dynamics”, Phys. Rev. Lett. 100, 123004 (2008); M.J. Cawkwell and A.M.N. Niklasson, “Energy conserving, linear scaling Born-Oppenheimer molecular dynamic“, J. Chem. Phys. 137, 134106 (2012); S.M. Mniszewski et. al, “Efficient Parallel Linear Scaling Construction of the Density Matrix for Born-Oppenheimer Molecular Dynamics”, J. Chem. Theory Comput. 11, 4644 (2015); A.M.N. Niklasson and M.J. Cawkwell “Generalized extended Lagrangian Born-Oppenheimer molecular dynamics”, J. Chem. Phys. 141, 164123 (2014); A.M.N. Niklasson et. al “Graph-based linear scaling electronic structure theory”, J. Chem. Phys. 144, 234101 (2016); A.M.N. Niklasson “Next generation extended Lagrangian first principles molecular dynamics”, arXiv.org > cond-mat > arXiv.1705.10845
8:30 AM - TC04.03.02
Artificial Driving Force Methods to Accelerate HCP Grain Boundary Motion in Molecular Dynamics
Shawn Coleman 1 , Matthew Guziewski 2 , Christopher Weinberger 2
1 , U.S. Army Research Laboratory, Aberdeen Proving Ground, Maryland, United States, 2 Mechanical Engineering, Colorado State University, Fort Collins, Colorado, United States
Show AbstractArtificial driving force methods are expanded to enable high throughput investigations of hexagonal close-packed grain boundary motion via molecular dynamics. To drive the motion of grain boundaries constructed from bicrystal models, an orientation dependent energy term to an existing potential to impose additional forces isolated on the grain boundary atoms. The artificial forces acting on boundary atoms promote their reorientation to the lower energy configuration, thus moving the boundary plane towards the higher energy grain. The artificial driving force methods show consistent results as compared to more physically based shear simulations, but are not restricted to shear coupled grain boundary geometries. Thus, high throughput calculations for Mg grain boundary motion are established and the resulting mobility values are incorporated into a novel grain boundary structure-property database.
8:45 AM - *TC04.03.03
Multiscale Modelling of Graphene from Nano to Micron Scales
Tapio Ala-Nissila 2 1
2 Department of Applied Physics, Aalto University School of Science, Espoo Finland, 1 Departments of Mathematical Sciences and Physics, Loughborough University, Loughborough United Kingdom
Show Abstract
Over the last few years novel two-dimensional materials and nanoscopically thin heteroepitaxial overlayers have attracted intense attention due to their unusual properties and important technological applications. Many physical properties of these systems such as thermal conductivity and electrical transport are intimately coupled to the large scale mechanical and structural properties of the materials. However, modeling such properties is a formidable challenge due to a wide span of length and time scales involved. In this talk, I will review recent significant progress in structural multi-scale modeling of two dimensional materials and thin heteroepitaxial overlayers [1], and graphene in particular [2], based to a large extent on the Phase Field Crystal (PFC) model combined with standard microscopic modeling methods (classical Molecular Dynamics and quantum Density Functional Theory). The PFC framework allows one to reach diffusive time scales for structural relaxation of the materials at the atomic scale, which facilitates quantitative characterisation of domain walls, dislocations, grain boundaries, and strain-driven self-organisation up to almost micron length scales. This allows one to study e.g. thermal conduction and electrical transport in realistic multi-grain systems [3].
References
1. K. R. Elder et al,. Phys. Rev. Lett. vol. 108, 226102 (2012); Phys. Rev. B vol. 88, 075423 (2013); J. Chem. Phys. 144, 174703 (2016).
2. P. Hirvonen et al., Phys. Rev. B 94, 035414 (2016).
3. Z. Fan et al., Phys. Rev. B vol. 95, 144309 (2017); to be published.
9:15 AM - TC04.03.04
Modelling Phonon Lifetimes in Classical Molecular Dynamics
Artur Tamm 1 , Magdalena Caro 2 , Alfredo Caro 3 , Alfredo Correa 1
1 , Lawrence Livermore National Laboratory, Livermore, California, United States, 2 , Virginia Polytechnic Institute and State University, Blacksburg, Virginia, United States, 3 , Los Alamos National Laboratory, Los Alamos, New Mexico, United States
Show AbstractA novel parameter free model for adding electron-ion interactions to a classical
molecular dynamics (MD) simulation is proposed. In the new model it is assumed
that the electrons behave as a thermal bath exerting non-conservative force on atoms.
The resulting stochastic dynamical system is casted as a generalisation of Langevin
dynamics with local correlations withing a cutoff radius. The cutoff radius is defined
on the basis of the extent of atomic densities. This locality of correlations is con-
structed so that total forces add up to zero conserving the center of mass motion.
Also, the total torques on the system is zero. Due to the local nature of the weight-
ing functions the same properties are retained when the system is split into smaller
part making the model useful the study of clusters, but more importantly it can
be utilised to study phonon lifetimes. The newly proposed model is applied to the
study of phonon modes in Ni crystal. The coupling strength between the ions and
electrons used in this study is computed ab initio with time-dependent density func-
tional theory (TDDFT). The simulations show that the phonon lifetimes depend on
both the frequency of the mode as well as the polarisation. It is observed that modes
with higher frequency have shorter lifetimes for all the branches. Furthermore, the
transverse phonon modes have a longer lifetime compared to longitudional modes. In
addition, the observed lifetimes agree qualitatively with the theoretical predictions
as well as with first-order perturbation theory density functional theory calculations.
In conclusion, the proposed model is able to describe electron-ion interaction with
the input data obtained from TDDFT. The predicted phonon lifetimes agree well
with higher order methods.
This work performed under the auspices of the U.S. Department of Energy
by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344
and funded by the Energy Dissipation to Defect Evolution Center, EDDE, an En-
ergy Frontier Research Center of the U.S. Department of Energy (Award Number
2014ORNL1026).
LLNL-ABS-733130
9:30 AM - TC04.03.05
A Deep Neural Network Atomic Potential for Investigation of Lattice Defects in Aluminum
Hideki Mori 1
1 , College of Industrial Technology, Amagasaki Japan
Show AbstractFor advanced material design, it is very important to clarify the extended lattice defect behavior, such as the crack propagation and the dislocation migration. The atomistic modeling is very suitable for this purpose, because the crack tip and dislocation core behavior are severely affected by discreteness of atomic spacing.
High accuracy density functional theory (DFT) calculations have established a leading position in atomistic modeling in decade. Unfortunately, for the high cost calculation of DFT, the number of atoms which DFT calculation can treat is limited to few hundreds, practically.
On the other hand, million atoms scale molecular dynamics (MD) simulations with empirical potentials enable us to realistic and flexible modeling of extended lattice defects. However, the accuracy of these potentials is limited by its function type in principle.
One solution of these problems is an atomic potential based on deep neural network (DNN) framework. By the universal approximation theorem, the DNN potential can compute the any function. Therefore, with enough DFT training data, DNN potential become sophisticated replica potential of DFT calculation. Moreover, by systematic selection of DFT training data, we can investigate how structural trends and simple lattice defect, such as vacancy and self-interstitial, affect the extended lattice defects. In this work, for one meaningful case study, we construct the atomic DNN potential for aluminum (Al).
The atomic radial and angle functions developed by Behler et.el. are used for the input descriptor of the atomic environment [1]. For DFT training date, we prepare the 3268 atomic structure of Al calculated by Quantum ESPRESSO package [2]. For crystal structure data, we select the FCC, HCP, BCC, simple cubic and diamond cubic structure. For defect data, we select the vacancy, divacancy, self-interstitial in T-site and O-site and (100), (110) and (111) surfaces model. For neural network training, we use the aenet package by Artrith et el [3].
The stacking fault energy by using constructed DNN potential is 139 mJ/m^2, which is very close to 131 mJ/m2 by DFT calculation. The edge dislocation core width by using DNN potential is 0.78 nm, which is consistent with other calculation. To investigate the relationship between the dislocation core property in Al and the structural trends, we construct the another DNN potential which is trained without HCP structure. The stacking fault energy decrease to 75 mJ/m2. The edge dislocation core width also increase to 1.81 nm. These results clearly demonstrate that the dislocation core property in Al is controlled by the structural trends between FCC and HCP of Al.
References: [1] J. Behler and M. Parrinello, Phys. Rev. Lett., 98 (2007) 146401, [2] P. Giannozzi, et al., J. Phys.:Condens. Matter., 21 (2009) 395502, [3] N. Artrith and A. Urban, Comput. Mater. Sci. 114 (2016) 135
9:45 AM - TC04.03.06
Multiscale Modeling of Realistic Two-Dimensional Materials
Petri Hirvonen 1 , Zheyong Fan 1 , Mikko Ervasti 1 , Morteza Jalalvand 2 3 , Khatereh Azizi 2 1 , Ville Vierimaa 1 , Matthew Seymour 5 , Luiz Pereira 4 , Davide Donadio 6 , S. Mehdi Vaez Allaei 2 , Nikolas Provatas 5 , Ari Harju 1 , Ken Elder 7 , Tapio Ala-Nissila 1 8
1 , Aalto University, Espoo Finland, 2 , University of Tehran, Tehran Iran (the Islamic Republic of), 3 , Institute for Advanced Studies in Basic Sciences, Zanjan Iran (the Islamic Republic of), 5 , McGill University, Montreal, Quebec, Canada, 4 , Universidade Federal do Rio Grande do Norte, Natal Brazil, 6 , University of California, Davis, Davis, California, United States, 7 , Oakland University, Rochester, Michigan, United States, 8 , Loughborough University, Loughborough United Kingdom
Show AbstractTwo-dimensional (2D) materials are typically polycrystalline, i.e., composed of different-sized and -shaped crystals in different orientations. The interfaces in between, grain boundaries, and other defects as well, greatly influence the properties of these materials, but modeling their formation is challenging due to the multiple length and time scales involved. We extend the phase field crystal (PFC) approach [Elder et al., Phys. Rev. Lett. 88, 245701 (2002)] to quantitative modeling of such defected microstructures, and compare the results to density functional theory (DFT) and molecular dynamics (MD). We have considered grain boundaries and grain boundary triple junctions in graphene, and recently in hexagonal boron nitride also. A one-mode PFC model is found to capture the atomic-level structure of graphene well topologically, whereas a three-mode model predicts the formation energies of grain boundaries quantitatively. [Hirvonen et al., Phys. Rev. B 94, 035414 (2016)] We find both positive and negative formation energies for triple junctions in graphene.
PFC models are able to capture the dynamics of large polycrystalline systems on diffusive time scales while retaining atomic resolution. We exploit these multiscale characteristics to construct large polycrystalline systems of 2D materials whose sizes and formation time scales are beyond the reach of DFT calculations and MD simulations, respectively. We use these systems as the starting point of MD simulations for investigating the thermal and electronic transport properties of these materials. From our large-scale MD simulations, we find bimodal grain size scaling for thermal conductivity.
10:30 AM - *TC04.03.07
The Inverse Problem in Atomistic Thermodynamics—A Method to Find Element Combinations that Achieve a Target Atomistic Structure
Arvind Kalidindi 1 , Peter Larsen 1 , Christopher Schuh 1
1 , Massachusetts Institute of Technology, Cambridge, Massachusetts, United States
Show AbstractIdentifying the right alloying elements to achieve a desired microstructure amounts to solving an inverse design problem that is central to materials chemistry. A number of atomistic frameworks exist for determining equilibrium microstructure given an alloy chemistry, or more generally, an interatomic potential. However, when the objective is to achieve a particular structure one would have to run simulations of a large number of alloys to determine which chemistries produce the desired equilibrium state. In this talk, a framework for solving such inverse problems is presented. For one specific context, namely, a lattice model with or without an interface, we show that this model determines all possible ground states considering all possible values of parameters in the interatomic potential, thus providing a full mapping between alloy chemistry and equilibrium microstructure. In the particular cases of interest to the present study, the model is used to study segregation to grain boundaries, revealing interesting effects of pairwise interactions on the complexions possible at a coherent interface.
11:00 AM - *TC04.03.08
Finding Reaction Coordinates during Phase Transformations in Solids
Jutta Rogal 1 2
1 ICAMS, Ruhr University Bochum, Bochum Germany, 2 Chemistry, New York University, New York, New York, United States
Show AbstractAtomistic modelling of the dynamics of phase transformations is a particularly challenging task. If the mechanism of the phase transformation is governed by so-called rare events then the timescale of interest will reach far beyond the capabilities of regular molecular dynamics simulations. In addition to the timescale problem the simulations provide a vast amount of data in a high-dimensional space that requires the projection into a low dimensional space and the identification of suitable reaction coordinates.
One example are the atomistic rearrangements during solid-solid phase transformations in bulk systems which involve massive structural changes including concerted multi-atom processes. The interface between two structurally different phases leads to a complex energy landscape that needs to be explored during the dynamical evolution of the interface. Here, we employ an adaptive kinetic Monte Carlo (AKMC) approach together with driven adiabatic free energy dynamics (d-AFED) to investigate such processes at the interface between the body-centered cubic and A15 phase in molybdenum.
A second example is the initial nucleation and growth during solidification in metals. Here, we investigate the atomistic mechanisms of nucleation in nickel for various undercoolings using transition path sampling (TPS). The analysis of the path ensemble reveals a non-classical behaviour with mainly non-spherical nuclei, and shows that the nucleation initiates in regions with high orientational order that also predetermine the selection of specific polymorphs.
Similarly, the formation of certain polymorphs during nucleation and growth as well as the transformation between various polymorphs is of particular interest in molecular crystals. If time permits new applications to this class of solids will also be presented.
11:30 AM - TC04.03.09
Hybrid Quantum-Classical Design and Experimental Confirmation of a Porous Silicon Multilayer Reflector
Alessio Palavicini 1 , Chumin Wang 1
1 , Universidad Nacional Autonoma de Mexico, Mexico City Mexico
Show AbstractWe report a multiscale design of omnidirectional Bragg mirrors based on dielectric multilayers by combining quantum mechanical and electromagnetic theories. This design begins with the calculation of electric permittivity for each layer using the density functional theory, followed by a transfer-matrix study of light propagation along the multilayer reflector [1]. The theoretical results were further validated by fabricating free‐standing porous silicon multilayer films obtained through electrochemical etching of p+‐type [100]‐oriented crystalline Si wafers, alternating two anodic current densities and finishing with a high current in order to separate the multilayer from the substrate [2]. The measured infrared transmittance from FTIR spectrophotometry confirms the position and width of the optical bandgap predicted by the hybrid design [3].
This work has been partially supported by UNAM-IN106317 and CONACyT-252943. Computations were performed at Miztli of DGTIC, UNAM. A. P. acknowledges the financial support from PAEP of UNAM.
[1] A. Palavicini and C. Wang, Optics and Photonics Journal 3, 20 (2013).
[2] P. Alfaro, A. Palavicini and C. Wang, Thin Solid Films 571, 206 (2014).
[3] A. Palavicini and C. Wang, to be published.
11:45 AM - TC04.03.10
Coupling MD with Phase Field Models—Atomistically Informed Free Energy Representation for Applications in Organic Electronics
Balaji Sesha Sarath Pokuri 1 , Shi Li 2 , Rabeh Ali 1 , Sean Ryno 2 , Chad Risko 2 , Baskar Ganapathysubramanian 1
1 , Iowa State University, Ames, Iowa, United States, 2 , University of Kentucky, Lexington, Kentucky, United States
Show AbstractPhase field based simulation strategies have been shown to be quite useful in exploring process-structure relationships for solvent based fabrication of organic electronics. The free energy functional predominately contains the specifications of the material system. Standard free energy functional representations are simplistic. However, with the increasing complexity of the molecular structure (conjugated, diverse side groups, anisotropic) of the materials that are currently being developed and utilized, standard flory-huggins type parametrization may be insufficient. This is further exacerbated by the need to accurately model multi component systems (consisting of donor, acceptor, solvent, solvent additives). In this work, we tackle this problem through a atomistically-driven construction of the free energy of representative material systems. The free energy construction is formally represented as a regression problem using a finite number of MD simulations. MD calculations of free energy are performed using the 2PT method with all-atom MD simulations. In order to minimize the number of MD simulations, we use a bayesian optimization approach to intelligently sample the configuration space to spawn MD simulations. Under assumptions of smoothness of the free energy, we can provide rigorous bounds on the number of MD simulations required to construct the free energy functional (given a error threshold). We illustrate the framework through a few examples of increasing complexity.
TC04.04: Atomistic-Continuum
Session Chairs
Tuesday PM, November 28, 2017
Hynes, Level 2, Room 202
1:30 PM - *TC04.04.01
Mathematical Underpinnings of Diffusive Molecular Dynamics
Brittan Farmer 3 , Petr Plechac 2 , Mitchell Luskin 3 , Gideon Simpson 1
3 Mathematics, University of Minnesota, Minneapolis, Minnesota, United States, 2 Mathematics, University of Delaware, Newark, Delaware, United States, 1 Mathematics, Drexel University, Philadelphia, Pennsylvania, United States
Show AbstractMetastable condensed matter typically fluctuates about local energy minima at the femtosecond time scale before transitioning between local minima after nanoseconds or microseconds. This vast scale separation limits the applicability of classical molecular dynamics methods and has spurned the development of a host of approximate algorithms. One recently proposed method is diffusive molecular dynamics which aims to integrate a system of ordinary differential equations describing the likelihood of occupancy by one of two species, in the case of a binary alloy, while quasistatically evolving the locations of the atoms. While diffusive molecular dynamics has shown to be efficient and provide agreement with observations, it is fundamentally a model, with unclear connections to classical molecular dynamics.
In this work, we formulate a spin-diffusion stochastic process and show how it can be connected to diffusive molecular dynamics. The spin-diffusion model couples a classical overdamped Langevin equation to a kinetic Monte Carlo model for exchange amongst the species of a binary alloy. Under suitable assumptions and approximations, spin-diffusion can be shown to lead to diffusive molecular molecular dynamics type models. The key assumptions and approximations include a well defined time scale separation, a choice of spin exchange rates, a low temperature approximation, and a mean field type approximation. We derive several models from different assumptions and show their relationship to diffusive molecular dynamics. Differences and similarities amongst the models are explored in a simple test problem.
2:00 PM - TC04.04.02
An Implementation for Mass and Heat Diffusion in Atomic Systems
Dingyi Sun 1 , Mauricio Ponga 2
1 , Brown University, Providence, Rhode Island, United States, 2 , University of British Columbia, Vancouver, British Columbia, Canada
Show AbstractModeling energy transport in nanoscale materials is a challenging problem and an active research area. One of the main limitations in the current state-of-the-art modeling techniques for energy transport at the nanoscale is that the time and size of systems that can be simulated is so small and often times, the simulation of experimental setups cannot be achieved. The limitation in modeling techniques comes from exceedingly large time and length scales involved in materials, that are needed in order to properly predict diffusive events, such as heat and mass transport. In this work, we used a mixed atomic-kinematic thermo-chemo-mechanical formulation to describe non-equilibrium system in order to simulate energy transport at the nanoscale. The formulation uses traditional force field methods in order to compute equilibrium configurations in addition to phenomenological kinematic laws for heat and mass transport. We describe the methodology and show different test cases and transport simulations of nanoscale materials using the proposed methodology.
2:15 PM - *TC04.04.03
Hyperdynamics + Quasicontinuum = Hyper-QC
Ellad Tadmor 1
1 , University of Minnesota, Minneapolis, Minnesota, United States
Show AbstractThe quasicontinuum (QC) method is a spatial multiscale method that extends the length scales accessible to fully-atomistic simulations (like molecular dynamics (MD)) by several orders of magnitude. While the recent development of the so-called "hot-QC method" [1] enables dynamic simulations at finite temperature, the times accessible to these simulations remain limited to the sub-microsecond time scale due to the small time step required for stability of the numerical integration. To address this limitation, we develop a novel finite-temperature QC method that can treat much longer time scales by coupling the hot-QC method with hyperdynamics – a method for accelerating time in MD simulations. We refer to the new approach as "hyper-QC" [2]. The temporal acceleration makes it possible to perform multiscale hot-QC simulations at near experimental loading rates for problems such as nanoindentation [3]. An interesting observation regarding the possibility of entropically stabilized dislocations [4], uncovered in a hyper-QC simulation, will also be discussed.
[1] E. B. Tadmor, F. Legoll, W. K. Kim. L. M. Dupuy, and R. E. Miller, "Finite-Temperature Quasicontinuum", Appl. Mech. Rev., 65:010803, 2013.
[2] W. K. Kim, M. Luskin, D. Perez, A. Voter, and E. B. Tadmor, "Hyper-QC: An Accelerated Finite-Temperature Quasicontinuum Method using Hyperdynamics", J. Mech. Phys. Solids, 63:94-112, 2014.
[3] W. K. Kim and E. B. Tadmor, "Accelerated Quasicontinuum: A Practical Perspective on Hyper-QC with Application to Nanoindentation", Phil. Mag., in press, 2017.
[4] W. K. Kim and E. B. Tadmor, "Entropically Stabilized Dislocations", Phys. Rev. Lett., 112:105501, 2014.
2:45 PM - TC04.04.04
An Atomistic Continuum Model to Investigate the Dynamic Evolution of Microstructure During Laser Shock Loading of Al Microstructures
Sergey Galitskiy 1 , Dmitry Ivanov 2 , Avinash Dongare 1
1 , University of Connecticut, Storrs, Connecticut, United States, 2 Theoretical Physics II, University of Kassel, Kassel Germany
Show AbstractLaser shock loading of materials has enabled the experimental investigation of the dynamic response of materials at strain rates up to 1010 s-1. The experimental investigation is largely focused on characterization of the shock pressures, shock velocities and the spall strength of metals using rear surface velocity profiles. However, the analysis of the microstructural evolution (melting, nucleation and evolution of defects) is limited due to the short time and the length scales of the processes. To address this challenge, an atomistic-continuum approach is used to simulate the laser induced shock loading and spall failure (attributed to nucleation, growth and coalescence of voids) in FCC metals. This approach combines the molecular dynamics (MD) simulations with a two-temperature model (TTM) to capture highly non-equilibrium electron-phonon states induced by ultra-fast laser energy absorption by electrons, as well as its subsequent relaxation. The combined MD-TTM is able to reproduce the absorption of laser energy, generation of shock waves, phase transformation (melting) as well as the spallation failure of FCC metals as observed experimentally. The combined MD-TTM model is also used to investigate the effect of laser shock loading parameters (laser fluence and pulse duration) to investigate the variations in the deformation behavior and spall failure behavior of Al microstructures (single crystal and nanocrystalline) with dimensions of 500 nm and 1 micron in the shock loading direction. The spall strength values compare very well with the experimental values reported in the literature. In addition, the role of microstructure on the dynamic evolution of defects (dislocation densities) during laser shock loading is used to define the decay of the elastic precursor and the strain rate dependence of the spall strength of metals. The framework of the MD-TTM method, the variations in the microstructural evolution with the loading parameters and the dynamic evolution of defect densities will be presented.
3:30 PM - TC04.04.05
Computing Phonon Dispersion Using Fast Zero-Point Correlations of Conjugate Variables
Anant Raj 1 , Jacob Eapen 1
1 , North Carolina State University, Raleigh, North Carolina, United States
Show AbstractTime correlations of dynamic variables in the reciprocal space offer a rich theoretical setting for computing the phonon dispersion curves, particularly for systems with marked anharmonic interactions. Currently, there are two general methods using the time correlation approach. In the first, the phonon dispersion is calculated using the frequencies extracted from the zero time expectation value of the correlation of displacements in the reciprocal space; equipartition of energy between the phonon modes is implicitly assumed. In the second method, the frequencies obtained from the Fourier transform of the time correlation of velocities are used to compute the dispersion of phonon modes. While the equipartition approach is computationally fast, since it requires only the zero time correlations, the assumption of equipartition of energy between the modes can result in tangible errors, especially at lower temperatures. In contrast, the Fourier method is more accurate but requires long correlation times for resolving the frequencies, making this approach computationally expensive.
In this work, we present a family of methods to compute the normal mode frequencies from the ratio of the expectation value of the correlation of conjugate variables at zero time. This approach is computationally as fast as the equipartition methods, but is also robust and works even in the absence of equipartition. We present two variants of this approach that are appropriate for use in atomistic simulations. In the first, the phonon dispersion curves are calculated using the ratio of the normal mode amplitudes for velocity and displacement while in the second they are computed analogously from velocity and acceleration. We further elucidate the attractiveness of the second approach entailing velocity and acceleration for systems exhibiting anisotropic and highly anharmonic vibrations. Although the approach involving velocity and displacement is known before, we derive a family of zero-point methods from statistical-mechanical first principles. Finally, we demonstrate the accuracy, efficiency and robustness of the zero-point methods on model linear chains and graphene using atomistic simulations.
3:45 PM - TC04.04.06
Comprehensively Integrated Environment for Advanced Materials Simulations (CINEMAS)
Kapil Gupta 1 , S. C. Lee 1
1 , Indo-Korea Science and Technology Center (IKST), Korea Institute of Science and Technology (KIST), Bangalore INDIA, Bangalore India
Show AbstractHigh-throughput computations have provided new pathway for advanced materials engineering and modelling. This is leading to a resurgent interest growing in recent time to develop computational environment for the preparation, submission and control of massive multiple calculations. Though so far there is no single platform which provides solution to all the requirement of a researcher performing density functional theory (DFT) calculations, on single canvas. In the purview of materials simulations most of the researchers still have been following traditional methods of performing these computations in split manner. A great deal of advantages can be reaped out by graphically interfacing them with the end user though a simplest personal computer.
We here present CINEMAS, a robust and comprehensive and integrated graphical user interface. This desktop application connects, the massive job controller handling multiple jobs, structural visualization as well as data plotting customized specifically for the requirements of DFT calculations and manipulation of analyzed graphics to publication standards. There are existing simulation environments which require knowledge of programming language such as python or C. However CINEMAS is based on user friendly graphical user interface, through which a user can rather focus on simulation itself. CINEMAS platform also provides automated calculations modules to simulate most of the material phenomena. This tool also enables user to execute several predefined and custom arithmetic on the output data. In summary this tool is aimed to provide solutions to all DFT simulation requirements at a single canvas, hence considerably reducing number of steps to perform such simulations.
This interface not only will enable user to perform large number of simulations seamlessly, but also to collect and process the output data into the form of a database. On which various modern machine learning and data mining techniques can be applied to achieve new scientific objectives.
4:00 PM - *TC04.04.07
Atomistic Growth Mechanisms and Property Optimization of 2D Materials
Zhenyu Zhang 1
1 , University of Science and Technology of China, Hefei, Anhui, China
Show AbstractThe 2D materials family keeps its amazing pace in expanding its family size, with more and more growing and outreaching branches in its family tree. Each member is this family has its uniqueness in both fabrication methods and intriguing properties. Many of the layered materials also share clear commonalities, most notably weak van der Waals (vdW) coupling between the layers. In this talk, we will review some of the latest developments in exploration of the atomistic growth mechanisms of several newcomers of the 2D materials family, including phosphorene, grown on metal or semiconductor substrates following a novel half-layer-by-half-layer mode, and tellurene, whose formation mechanism is rooted in the multi-valence nature of Te. We will also present convincing evidence for vdW-induced topological phase transition in certain layered materials, further demonstrating the vital importance of the seemingly weak dispersion force.
4:30 PM - TC04.04.08
Scaling of Molecular Dynamics Simulations to the Mesoscales Using Quasi Coarse Grained Dynamics
Sumit Suresh 1 , Avinash Dongare 1
1 Department of Materials Science and Engineering, Institute of Materials Science, University of Connecticut, Storrs, Connecticut, United States
Show AbstractClassical molecular dynamics (MD) simulations have helped in extracting critical information for phenomena where resolutions provided by experimental techniques either are incapable due to theoretical limits or are far too expensive or inefficient. For the past few decades, MD simulations have been employed to study the processes related to phase transformation and high strain rate deformation using computationally efficient interatomic potentials. Even though events like plastic deformation and heat transfer are well represented, this technique has limitations on modeling system sizes of more than a few hundreds of nanometers and timescales more than a few nanoseconds due to the computational expense.
A computationally efficient mesoscale model called Quasi-Coarse-Grained Dynamics (QCGD) is recently developed to address the challenge of time and length scales of MD simulations. The QCGD uses scaling relationships to model a given microstructure using a lesser number of “representative” atoms with the energetics defined by scaled interatomic potentials. This method uses the framework of molecular dynamics and distance based scaling to modify interatomic potentials, and thus aims to bridge the gap of length and time scales between atomistic and continuum level simulations. The applicability of this method is demonstrated by modeling the microstructural evolution during melting behavior as well as the deformation behavior of Al powder particles during single particle impact with dimensions of tens of microns. The kinetics of melting and the evolution of pressures and temperatures at various impact velocities and using various levels of coarse graining is investigated to investigate effects of system size and loading conditions. The QCGD simulations using the embedded atom model (EAM) potential are able to reproduce the experimentally observed phenomena called “jetting”, which is defined as the material ejection at the periphery of the particle/substrate interface during impact of Al powder particles onto a substrate. A critical velocity of 1100 m/s was identified as the criterion for jetting of pure polycrystalline aluminum modeled which is in excellent agreement with the experimentally observed values. The QCGD framework, the scaling relationships, the comparisons between MD and QCGD simulations will be presented.
4:45 PM - TC04.04.09
From the Atomistic to the Mesoscale—Modeling Precipitation in Alloys
Anirudh Raju Natarajan 1 , John Thomas 1 , Brian Puchala 2 , Anton Van der Ven 1
1 , University of California, Santa Barbara, Santa Barbara, California, United States, 2 Materials Science & Engineering, University of Michigan–Ann Arbor, Ann Arbor, Michigan, United States
Show AbstractPrecipitate formation within an alloy is extensively used to tune material properties in technological applications ranging from structural materials to thermoelectrics. In most applications, the formation of precipitates from a super-saturated solid solution is utilized to engineer properties such as mechanical strength and creep resistance. Such precipitate formation involves many different length-scales making it a complex problem to model. Models to understand the phenomena at specific length scales are well established, however rigorous connections bridging the atomistic and mesoscale have remained elusive. In this talk, we will describe how quantities required in mesoscale modeling techniques such as the phase-field method may be systematically obtained starting from first-principles electronic structure methods. Using effective Hamiltonians informed from density-functional theory calculations, we use Monte-Carlo techniques to calculate accurate thermodynamic quantities at elevated temperatures. Through the course of this talk, the methods will be applied to a variety of alloy systems including magnesium and nickel alloys to elucidate the rich variety of materials phenomena that can be described and understood starting from first-principles.
TC04.05: Poster Session: Advanced Atomistic Algorithms in Materials Science
Session Chairs
Wednesday AM, November 29, 2017
Hynes, Level 1, Hall B
8:00 PM - TC04.05.01
Temperature Dependent Atomic Segregation and Structural Evolution of (PdPt)923 Alloyed-Bimetallic Nanoparticles—A Molecular Dynamics Study
Carlos Rodriguez-Proenza 1 6 , Cristy Azanza-Ricardo 2 , José Rodríguez-López 3 , Ramiro Perez-Campos 4 , Rodrigo Alonso Esparza Muñoz 5
1 Nanotecnología, Posgrado en Ciencia e Ingeniería de Materiales, Centro de Física Aplicada y Tecnología Avanzada, Universidad Nacional Autónoma de México, Campus Juriquilla, Querétaro, Querétaro, Mexico, 6 Ciencias Básicas, Instituto Tecnológico de Querétaro, Querétaro, Querétaro, Mexico, 2 Biología Molecular, Centro de Física Aplicada y Tecnología Avanzada, UNAM, Querétaro, Querétaro, Mexico, 3 Materiales Avanzados, Instituto Potosino de Investigación Científica y Tecnológica, A.C., San Luís Potosí, San Luís Potosí, Mexico, 4 Nanotecnología, Centro de Física Aplicada y Tecnología Avanzada, UNAM, Querétaro, Querétaro, Mexico, 5 Nanotecnología, Centro de Física Aplicada y Tecnología Avanzada, UNAM, Querétaro, Querétaro, Mexico
Show AbstractUnderstanding the thermodynamic and structural properties of bimetallic nanoparticles is of great interest for their use in advanced catalysis and other potential applications. In this work, we carried out molecular dynamics simulations with many-body-embeded atom model (EAM) potential to examine the thermal and structural evolution of alloyed-(PtPd)923-bimetallic nanoparticle systems during continous heating and freezing processes. The effect of temperature on the atomic surface segregation in the nanoparticles from a random distribution of atoms, was studied using NVT estatistical assembly.
The results show that the Pd atoms tend to surface segregation on the nanoparticles while Pt atoms diffuse into the core as temperature increases. In addition, the nanopartiles undergo continous structural changes during heating and freezing processes. This trend of atomic segregation could be attributed to significantly higher surface energy of Pt atoms than those of Pd. As a result, a considerable amount of thermal energy is used by Pt and Pd atoms for migration. The obtained results are corroborated both from the radial distribution functions and temperature dependence of the number of Pd and Pt atoms on the surface, since they represent a significant measure of the disorder in the systems investigated. These observations are in good agreement with the previous studies. Moreover, for each system, both the corresponding caloric curves and the temperature dependance of heat capacity are plotted by considering that they are two significant criteria for identifying the melting and freezing points of nanoparticles. Significantly, at the end of the cooling processes, all the nanoparticles showed a tendency towards the formation of icosahedral structure.
8:00 PM - TC04.05.02
Screening of Guest Metals in Metal-Organic-Frameworks for the Oxygen Reduction Reaction
Wenrui Chai 1
1 Chemistry, University of Texas at Austin, Austin, Texas, United States
Show AbstractMetal-Organic-Frameworks (MOFs) is a class of porous material that has special properties unlike traditional porous materials due to the inclusion of inorganic metal sites. The properties of MOFs vary greatly, depending on the nature of the organic ligand, the identity of the guest metal as well as the identity of the connecting metal ion. Phosphorous-carbon-phosphorous (PCP) pincer MOFs is a subcategory of MOFs with a pincer ligand that can strongly bind to the guest metal to create a Lewis acid site and allow for stronger gas adsorption. Another important advantage of such materials is the access to post-synthetic modifications. These materials have enormous potential for gas separation and adsorption, and by extension in gas phase catalysis. However, given the experimental difficulty of synthesizing such materials, a survey of the many possible combinations of guest metals is intractable. DFT simulations, however, can be used to screen host metals to find potentially useful candidates for further investigation, as well as to gain an understanding of the reactivity of the MOFs.
In this study we started with an experimental crystal structure of a Pd-based PCP monomer [Junpeng He et al.], and replaced the Pd with group 6 to 12, 3d, 4d and 5d transition metals. From the relaxed structures, the binding energy of O, OH and OOH were calculated as reactivity descriptors for screen for the oxygen reduction reaction. Our results show the expected correlation between OH and OOH binding. The binding energy of these species in the MOFs were also correlated to binding energy trends on metallic slabs. The most promising metal hosts included, Rh, Ru, Ir and Os, for which the binding energy differences between OH and OOH (~2.7 eV) are lower than that of a Pt slab (~2.9 eV) suggesting that the PCP pincer ligand loaded with Rh, Ru, Ir and Os could reduce the overpotential for the ORR. Our calculations also show that through interactions with the ligand, which can be post-synthetically modified, the metal can be tuned to towards better catalytic performance. This work provides target PCP pincer MOFs for synthesis and testing as ORR catalysts.
8:00 PM - TC04.05.03
Free Energy Barriers in Large Crystalline Systems without Collective Coordinate Functions
Thomas Swinburne 1 3 , Mihai-Cosmin Marinica 2
1 , T-1 Group, Los Alamos National Laboratory, Los Alamos United Kingdom, 3 Materials Modelling, CCFE, UKAEA, Abingdon United Kingdom, 2 DEN-CEA, Université Paris-Saclay, Paris, Gif-sur-Yvette, France
Show AbstractFree energy calculations of structural transformations in crystalline systems are often hindered by the need to define suitable collective coordinate functions, which may be difficult or even impossible to define. We present an alternative solution, defining a reaction coordinate from the zero temperature transition pathway and deriving an expression for the free energy gradient that remains exact even if the system instead follows a wide range of finite temperature pathways. Our method, implementated in LAMMPS, allows free energy barrier calculations on systems of over 100,000 atoms, which we demonstrate with the problem of double kink nucleation on screw dislocations in tungsten and a complex structural transformation of C15 defects in bcc iron.
8:00 PM - TC04.05.05
Smart MD-Sampling Method for Interfacial Free Energy between Polymer-Grafted Substrate and Liquid
Masayuki Uranagase 1 , Shuji Ogata 1
1 , Nagoya Institute of Technology, Nagoya Japan
Show AbstractThe solid-liquid interfacial free energy is a principal physical quantity that determines the wettability and adhesive properties. However, substantial ambiguities exist in evaluating the quantity through experimental measurement of the contact-angle. Accurate theoretical evaluation of the interfacial free energy between decorated surface and liquid is highly desired for designing novel materials with unique surface characters, e.g., superwettable materials that may be used for self-cleaning, fluid separation, etc.
In calculating the interfacial free energy at a finite temperature, the thermodynamic integration technique can be used combined with a series of MD simulations. In fact, several methodologies in the thermodynamic integration technique have been proposed for those solid-liquid interfacial systems with flat, crystalline surfaces of solids.
Intending to change surface properties, solid surfaces are often coated by polymers after their chemical reactions with surface, e.g., self-assembled monolayer. Those solid generally assume inhomogenously polymer-grafted surfaces. Until now, there is no practical method to evaluate the interfacial free energy of such complex solid surfaces. In this presentation, we propose a novel method for evaluating the interfacial free energy between polymer-grafted surface and liquid based on the thermodynamic integration.
In former methods, the interfacial free energy, more precisely the work of adhesion, is obtained by integrating an external force exerted by liquid molecules when the liquid on the surface is separated from the surface by gradually shifting the origin of the plane-shape external potential. But, they are inefficient when they are applied to the polymer-grafted surface because the plane shape potential need be shifted by a long distance (comparable to the polymer length) to separate the liquid from the surface. To overcome this problem, spherical external potentials are introduced around selected atoms of the grafted-polymers. With the spherical external potentials, evaluation of the interfacial free energy can be performed adaptively for any setting of the surface decoration. In addition to the shape of the external potential, sophisticated higher-order integration scheme is adopted in order to reduce the number of sampling points of the external forces for further improvement in calculation efficiency.
Present method is applied to the interface between water and gold surface grafted by poly(ethylene glycol). This system attracts much attention in the field related to biomaterial because the surface density of poly(ethylene glycol) on gold surface affects resistance to the adsorption of proteins. We will also report the dependence of hydrophilicity of the surface on the density of poly(ethylene glycol) that grafts to the gold surface.
8:00 PM - TC04.05.08
Hydrogen Diffusion in Palladium under Hydrostatic Strain—Ab Initio Path Integral Molecular Dynamics Study
Hajime Kimizuka 1 , Shigenobu Ogata 1 2
1 , Osaka University, Osaka Japan, 2 , Kyoto University, Kyoto Japan
Show AbstractMany studies have been conducted to identify the characteristics of hydrogen (H) storage in palladium (Pd) nanoparticles and H purification using Pd membranes, for the purpose of utilizing H gas as a potential secondary energy carrier. Recently, much attention has been paid to a “defect engineering” approach for controlling H permeability and diffusivity by introducing surfaces, grain boundaries, and/or dislocations into bulk Pd. To characterize the net H diffusivity in such a system, it is essential to quantitatively assess the H-diffusion behavior not only in the vicinity of lattice defects but also in a Pd lattice under finite strains. Thus, we investigated the effect of lattice strain on the diffusion behavior of H in face-centered cubic Pd based on ab-initio path integral molecular dynamics (PIMD) modeling in the framework of density functional theory (DFT).
First, the minimum energy path and the transition state of H migration along the path between the octahedral (O) and the (T) interstitial sites were determined using the nudged elastic band (NEB) method based on DFT for a Pd32H supercell. Then, H-migration kinetics was analyzed by performing ab-initio PIMD calculations. This was necessary as the H atom has sufficiently low mass that it exhibits significant nuclear quantum delocalization and zero-point motion even at ambient temperature. The “classical” DFT result showed that the H atom was stable at O site and the corresponding H-migration barrier (Em) in Pd was 0.16 eV. The PIMD-based free energy profiles for H migration between the O site and T site were evaluated at 75-600 K. We confirmed that the nuclear quantum effects significantly affected the Em and the difference between the energies of the H atom at the O site and the T site (ET-O); The Em and ET-O values in Pd at 300 K increased by 32% and 98%, respectively, relative to the classical limit. The results suggested that the nuclear quantum effects play a crucial role in the process of H migration in Pd even at temperatures above 300 K. It is noted that the apparent activation barrier of 0.23 eV for Pd derived from the PIMD result was in excellent agreement with that from the experiments.
In addition, we investigated the H diffusion in the strained Pd lattice. The ET-O value decreased as hydrostatic (i.e., homogeneous triaxial) tensile strain was increased, and then the H atom at the T site became energetically more stable than that at the O site at the strain of 1-2%. Further, the H-migration barrier along the O-T path was decreased by 60%. We found that the strain significantly changed the shape of the free energy profile for the H migration, which resulted from the combination of internal relaxation of the atomic positions and the accompanying nuclear quantum effect. Our ab-initio results suggested a possibility that the H diffusivity in Pd can be controlled by imposing an elastic strain.
8:00 PM - TC04.05.09
Structural and Mechanic Properties of Gd2(TM)2O7 with TM = Mo, Ru and Ti Pyrochlores—A First-Principles Calculation
Amaranta Castro Espinosa 1 , Martin Romero Martinez 2 , Raúl Escamilla Guerrero 1 , Francisco Morales Leal 1 , Oscar Olicón Hernández 1
1 Instituto de Investigaciones en Materiales, Universidad Nacional Autónoma de México (UNAM), Ciudad de México Mexico, 2 Facultad de Ciencias, Universidad Nacional Autónoma de México (UNAM), Ciudad de México Mexico
Show AbstractThe structural and elastic properties of the Gd2(TM)2O7 with TM = Mo, Ru and Ti were investigated using first-principles, within of plane-wave pseudopotential density functional theory with the generalized gradient approximation (GGA) and local density approximation (LDA). The equilibrium cell constants agree well with the reported experimental results. We calculate a set of elastic parameters (bulk modulus (K), shear modulus (G), Young’s modulus (Y) and Poisson’s ratio (ν)) in the framework of the Voigt-Reuss-Hill approximation. The Debye temperature (θD) was calculated using these elastic moduli. The calculated elastic constants were positives and satisfy the well-known Born criteria, indicating that the cubic structure is stable.
Acknowledgements
The authors thank the projects DGAPA-UNAM IN106116/28. They also thank M.M.S. Alberto Lopez-Vivas, A. Pompa, and C. Gonzalez for providing technical help. Calculations were performed using resources from the Supercomputing Center DGTIC-UNAM. A. Castro Espinosa thank CONACYT for the scolarship awarded.
8:00 PM - TC04.05.10
Mechanical and Tribological Properties of Graphene Reinforced Natural Rubber Composites—A Molecular Dynamics Study
Raj Chawla 1
1 Mechanical Engineering, Lovely Professional University, Jalandhar, Punjab, India
Show AbstractGraphene reinforced natural rubber composites are developed to study the improvement in mechanical and tribological properties of composites by the introduction of graphene as reinforcement. Constant strain minimization method has been applied to calculate Young's modulus and shear modulus of the graphene-natural rubber composites. A three layer model containing Fe (Iron) atoms at the top and bottom and composite in middle has been constructed to calculate the tribological properties. A shear loading is applied to the top iron nanorod by sliding it to the surface of the composite for 600 ps with a velocity of 0.1Å/ps. The results show the increase of 185% in Young’s modulus,32% in shear modulus and 48% in hardness by reinforcing natural rubber with single layer graphene sheet. In addition, reduction of 28% and 36% in the average friction coefficient and abrasion rate obtained for graphene-polymer composites. Also, interaction energy between graphene and natural rubber, the angle, bond and kinetic energy of the polymer and composites are given and discussed.
8:00 PM - TC04.05.11
Theoretical Prediction of Glass Forming Ability of Cu-Zr Alloys Based on Nucleation Theory and Molecular Dynamics Simulation
Yuji Sato 1 , Chiaki Nakai 1 , Masato Wakeda 1 , Shigenobu Ogata 1 2
1 Department of Mechanical Science and Bioengineering, Osaka University, Toyonaka, Osaka, Japan, 2 Center for Elements Strategy Initiative for Structural Materials, Kyoto University, Kyoto, Kyoto, Japan
Show AbstractMetallic glasses possess brilliant properties as structural materials including high elastic limit, high toughness, and high corrosion resistance. Reachable bulk sample size is limited within centimeters even for highly selected alloy compositions, and thus the application of metallic glasses as structural materials is restricted. Alloys with a lower glass forming ability (GFA) require a higher cooling rate in the quenching process to reach a solid glass state to prevent crystallization since it is necessary to perform vitrification prior to a critical crystal nucleus, which corresponds to the minimum nucleus size required for continued crystal growth and is spontaneously nucleated in the molten alloy. In order to form a large-sized bulk metallic glass, exploration of metallic alloy compositions having excellent GFA is essential. However, it is difficult to prepare various alloy compositions in experiment. Hence, theoretical prediction of GFA of metallic alloys based on computer simulation is one of the keys to explore metallic alloy compositions having excellent GFA. GFA can be evaluated by the critical cooling rate, which is the minimum cooling rate without crystallization. The critical cooling rate is estimated from Time-Temperature-Transformation (TTT) diagram. In order to describe TTT diagram, information of crystal nucleation must be obtained. Molecular dynamics (MD) simulation is a promising tool to analyze crystal nucleation and predict TTT diagram, the critical cooling rate and GFA. However, direct MD prediction is still challenging because of the time-scale limitation of MD. For the practical bulk metallic glass alloys, the time necessary for quenching at typical cooling rate is five or more orders of magnitude longer than the MD time-scale.
In this study, therefore, I propose a method to depict TTT diagram of the alloys based on the classical nucleation theory, which is widely employed to understand nucleation phenomena, informed by MD simulations. Although the method is assuming the conventional classical nucleation theory, all the materials parameters appeared in the theory were determined by MD simulations using realistic Finnis-Sinclair (FS) type interatomic potentials. Using the method, the TTT diagrams and critical cooling rates of Cu-Zr alloy, which is one of the binary metallic glasses, were computed. The two composition (Cu50Zr50 and Cu20Zr80) were employed and the comparison of TTT diagram and the critical cooling rate was conducted. The former actually forms rod-shaped glass of 2 mm diameter in experiment, but the latter does not vitrify. As the results of this analysis, I found that the proposed method reasonably predicts the critical cooling rate.
8:00 PM - TC04.05.12
Heterogeneity in Homogeneous Nucleation from Billion-Atom Molecular Dynamics Simulation by Multi-GPUs Parallel Computation
Yasushi Shibuta 1 , Shinji Sakane 2 , Eisuke Miyoshi 2 , Shin Okita 1 , Tomohiro Takaki 2 , Munekazu Ohno 3
1 , The University of Tokyo, Tokyo Japan, 2 , Kyoto Institute of Technology, Kyoto Japan, 3 , Hokkaido University, Sapporo Japan
Show AbstractWe performed billion-atom molecular dynamics (MD) simulations of homogeneous nucleation from an undercooled iron melt by way of multi-GPUs parallel computation. [1]. We have developed a parallel GPU code for multiple GPUs to accelerate the large-scale MD simulations. The code is written using the CUDA based on the C/C++ language, and the MPI is used for internode communication. 512 GPUs are successfully parallelized for the large-scale MD simulation. From very-large MD simulations performed on the GPU-rich supercomputer TSUBAME2.5, we revealed that some satellite-like small grains surrounding previously formed large grains exist in the middle of the nucleation process, which are not distributed uniformly. At the same time, grains with a twin boundary are formed by heterogeneous nucleation from the surface of the previously formed grains. The local heterogeneity in the distribution of grains is caused by the local accumulation of the icosahedral structure in the undercooled melt near the previously formed grains.
[1] Y. Shibuta, S. Sakane, E. Miyoshi, S. Okita, M. Ohno, T. Takaki, Nature Communications, 8 (2017) 10.
8:00 PM - TC04.05.14
Linking Phase Field and Atomistic Simulation to Study Solidification in Undercooled Titanium
Sepideh Kavousi 1 , Brian Novak 1 , Mohammad Dodaran 1 , Dorel Moldovan 1
1 , Louisiana State University, Baton Rouge, Louisiana, United States
Show AbstractWe present a complementary phase field method (PFM) and molecular dynamics (MD) simulation to study the solidification and dendritic growth of pure titanium. Although phase field is a very powerful tool to study the solidification process qualitatively, it is unable to predict the microstructure evolution quantitatively. In order to have a more realistic phase field simulation, we first do MD calculations on pure titanium to obtain the bulk and interfacial properties, and then use them in the phase field model to predict the solidification. And then compare the growth rate calculated from PFM and MD. We investigate the effect of various parameters on the dendritic arm spacing and find a power law relation for predicting arm spacing. In our phase field simulations, we implement a numerical method that enables us to investigate solidification in larger domains. Specifically, the temperature and order parameter equations are solved only in the domains close to the solidification front.
8:00 PM - TC04.05.15
Kinetic Monte Carlo Simulations of Oxygen Diffusion in Environmental Barrier Coating Materials
Brian Good 1
1 , NASA Glenn Research Ctr, Cleveland, Ohio, United States
Show AbstractCeramic Matrix Composite (CMC) materials are of interest for use in next-generation turbine engines, offering a number of significant advantages, e.g. reduced weight and high operating temperatures. However, in the hot environment in which such components operate, the presence of water vapor can lead to corrosion and recession, limiting the useful life of the components. Such degradation can be reduced through the use of Environmental Barrier Coatings (EBCs) that limit the amount of oxygen and water vapor reaching the component. Candidate EBC materials include Yttrium and Ytterbium silicates. In this work we present results of kinetic Monte Carlo (kMC) simulations of oxygen diffusion, via the vacancy mechanism, in Yttrium and Ytterbium disilicates, along with a brief discussion of interstitial diffusion.
An EBC system typically includes a bond coat located between the EBC and the component surface. Bond coat materials are generally chosen for properties other than low oxygen diffusivity, but low oxygen diffusivity is nevertheless a desirable characteristic, as the bond coat could provide some additional component protection, particularly in the case where cracks in the EBC provide a direct path from the environment to the component surface. We have therefore performed similar kMC simulations of oxygen diffusion in Hf silicate, a proposed bond coat material.
8:00 PM - TC04.05.16
Phase-Field Crystal Modeling in Materials Science—From Fundamentals to Applications
Salvador Valtierra 1 , Kirk Bevan 1 , Nana Ofori-Opoku 2 3 , Nan Wang 1 , Nikolas Provatas 1
1 , McGill University, Montreal, Quebec, Canada, 2 Center for Hierarchical Materials Design, Northwestern University, Evanston, Illinois, United States, 3 Materials Measurement Laboratory, National Institute of Standards and Technology, Gaithersburg, Maryland, United States
Show AbstractWe present an in-depth examination of the fundamental postulates underlying the phase-field crystal (PFC) method for simulating microstructural defects. Beginning from statistical mechanics, we arrive a PFC model through a coarse-grained free energy constructed from classical functional density theory. The equations of motion are then constructed based on dissipative dynamics that lead to the minimization of a free energy functional. Through this bottom-up exploration, it is shown how PFC can produce liquid and periodic solid phases that display defects such as dislocations and grain boundaries. By further modifying this free energy, the basic model can incorporate different lattice structures such as FCC and BCC lattices. PFC has been successful in studying a wide range of phenomena such as solidification and electromigration. The aim of this study is to provide a bottom-up perspective on PFC, that will spur new theoretical insights and algorithm development in material science modeling.
8:00 PM - TC04.05.17
Computational Analysis on Defect Behavior Near Surface and Grain Boundaries in Metallic Systems Using Long-Time Atomistic Simulations
Shotaro Hara 1
1 , Chiba Institute of Technology, Chiba Japan
Show AbstractLong-term degradation phenomena in metallic systems under a high temperature, such as creep voiding, are basically governed by the vacancy diffusion, accumulation and growth processes at an atomistic scale around the material heterogeneities (surface or grain boundaries). However, the fundamentals of these diffusive behaviors are not still understood because a molecular dynamics simulation often suffers from tracking thermally-activated processes due to its limited time scale. Here, we have investigated the vacancy-mediated grain boundary diffusion using a bond-boost method for safely accelerating atomistic simulations. Using this scheme, we access the finite-temperature dynamical process of vacancy migration over timescale comparable to laboratory experiments. Our simulations allow us to estimate the rate for vacancy migration along the various types of symmetric tilt grain boundaries, indicating that the rates at grain boundaries are about several times larger than that of bulk. In addition to the vacancy migration process, the vacancy segregation and voiding at grain boundary has been analyzed using a diffusive molecular dynamics simulations, which is a novel approach for exploring the atomic level mass action along the chemical potential gradient at diffusive time scale. We show that under the tensile stress acting normal to the grain boundary, the void growth on the grain boundary is strongly anisotropic unlike the isotropic void growth in bulk, reflecting the fact that the grain boundary diffusion is much faster than bulk diffusion.
8:00 PM - TC04.05.18
Modelling the Deposition Process on the CdTe/CdS Interface
Miao Yu 1 , Steven Kenny 2
1 School of Mathematics and Statistics, Xidian University, Xi'an, Shaanxi, China, 2 Department of Materials, Loughborough University, Loughborough United Kingdom
Show AbstractCdTe is an excellent material for low-cost, high efficiency thin film solar cells and holds the record for Watts/$ performance. Defects such as grain boundaries and dislocations lower the efficiency of CdTe solar cells, thus it is important to do research on how these defects are formed during the growth process, especially on the interfaces of different materials.
In this work we use computer simulation to predict the growth of a sputter deposited CdTe thin film on the CdS surfaces. Single deposition tests have been performed, to study the behaviour of deposited clusters under different conditions.
We deposit a CdxTey (x,y = 0,1) cluster onto the wurtzite (111) Cd and S terminated CdS surfaces with energies ranging from 1 to 40 eV. More than 1,200 simulations have been performed for each of these cases so as to sample the possible deposition positions and to collect sufficient statistics. The results show that Cd atoms are more readily sputtered from the surface than Te atoms and the sticking probability is higher on S terminated surfaces than Cd terminated surfaces. They also show that increasing the deposition energy typically leads to an increase in the number of deposited atoms replace surface atoms and tends to decrease the number of atoms that sit on the surface layer, whilst increasing the number of interstitials observed.
8:00 PM - TC04.05.20
Molecular Dynamics Study of Polycondensation at a Polyimide/Cu(100) Interface
Eleanor Coyle 1 , Katherine Sebeck 1 , John Kieffer 1
1 , Univ of Michigan, Ann Arbor, Michigan, United States
Show AbstractAccurate modeling of structural developments during polyimide cure at a polymer/metal interfaces can provide critical information with regard to the incorporation of polycondensation water and how its presence influences structure, electrical and mechanical properties, and interfacial interactions. A previous study of bulk polyimide demonstrated that a newly-developed dynamic polymerization technique combining molecular dynamics and Monte Carlo methods can generate a more realistic bulk structure in the absence of an interface. In this work the algorithm is used to explore polyimide cure in the vicinity of a Cu(100) interface. The relevant interfacial effects, including spatial confinement, molecular dissociation, adsorption, and charge transfer, are explored. Results are compared with literature data for monomer/substrate and polymer/substrate interfacial interactions. Network structure and hydration behavior are tracked throughout the course of the polymerization reaction. This system is used to investigate the effects of water concentration and distance from the interface on the hydration behavior of this material.
8:00 PM - TC04.05.21
Ab Initio Band Structure of Ruthenium-Doped Gadolinium Orthotantalate
Jon Goldsby 1
1 , NASA, Cleveland, Ohio, United States
Show AbstractThere is growing momentum in the automotive industry to harvest energy from the exhaust using solid state power conversion processes, which can convert kilowatt levels of electrical power from the vehicle’s engine. Solid state power conversion devices, such as thermoelectrics, depend solely upon the temperature gradients for their operation. Aeronautic gas turbine engines have temperature gradients as well throughout due to the enthalpy processes of combustion, which offers the possibility of generating electrical power for use in primary and secondary electrical systems in the aircraft. However, currently available thermoelectric materials do not possess the environmental durability and performance levels necessary to realize these benefits. New materials must be developed that can meet the requirements to harvest waste enthalpy from gas-turbine engines. Computational methods offers an efficient and systematic manner to design new materials and guide there development. A computational -based material approach was used to determine the suitability of ruthenium-doped gadolinium orthotantalate (GdTa (1-x) RuxO4) as a practical thermoelectric material. The calculations were carried out using a projector augmented wave (PAW) method using a commercial code (Materials Design Inc.) MedeA incorporating the Vienna Ab-initio Simulation Package (VASP) as the computational engine. The calculation were based on density functional theory using the GGA-PBE exchange-correlation functional using and optimized mesh. This study makes predictions of the band structure and density of states properties using several popular DFT functionals.
Symposium Organizers
Enrique Martinez, Los Alamos National Laboratory
Graeme Henkelman, University of Texas
Hannes Jonsson, University of Iceland
Steven Kenny, Loughborough University
Symposium Support
Modelling and Simulation in Materials Science and Engineering | IOP Publishing
TC04.06: Accelerated MD
Session Chairs
Graeme Henkelman
Gideon Simpson
Wednesday AM, November 29, 2017
Hynes, Level 2, Room 202
8:15 AM - *TC04.06.01
Long-Timescale Simulation of Metallic Nanoparticles Using Parallel Trajectory Splicing
Danny Perez 1 , Rao Huang 2 , Arthur Voter 1
1 , Los Alamos National Laboratory, Los Alamos, New Mexico, United States, 2 Department of Physics, Xiamen University, Xiamen China
Show AbstractBecause the properties of nanoparticles depend sensitively on their structure, fine control over their shapes is paramount. However, owing to their complex energy landscapes, computational understanding and prediction of shape evolution and stability proves extremely challenging. We show how a recently introduced Accelerated Molecular Dynamics method, Parallel Trajectory Splicing, can efficiently leverage parallel computing resources to extend simulation timescales towards milliseconds. This enables the direct observation of complex phenomena that would remain out of reach of conventional molecular dynamics simulations, such as shape fluctuations or kinetic effects during growth.
8:45 AM - TC04.06.03
Parallel Temperature Accelerated Dynamics for the Exascale—Speculation, Replication and Splicing
Richard Zamora 1 , Danny Perez 1 , Arthur Voter 1
1 Theoretical Division, T-1, Los Alamos National Laboratory, Los Alamos, New Mexico, United States
Show AbstractAccelerated Molecular Dynamics (AMD) is comprised of several MD-based algorithms designed to discover the state-to-state evolution of a system of atoms at an accelerated rate compared to direct MD. Of these methods, Temperature Accelerated Dynamics (TAD) hastens the evolution by exploring possible transitions from each official state at an elevated temperature. The recently introduced Speculatively Parallel TAD (SpecTAD) method allows for the parallel scaling of TAD performance by executing a dynamically generated tree of speculative states on distinct computational cores. Although speculative parallelism can be very powerful, other algorithms are often more efficient. Here, we compare the performance of speculative parallelization with alternative parallelization techniques. Overall, this work motivates the use of parallel trajectory splicing whenever possible, as some combination of speculation and replica-based parallelization are typically most efficient.
9:00 AM - TC04.06.04
Long-Time Dynamics Simulations of Surface Segregation in Pd-Au Nanoparticles
Xinyu Li 1 , Zhiyao Duan 1 , Graeme Henkelman 1
1 Chemistry, University of Texas at Austin, Austin, Texas, United States
Show AbstractBimetallic nanoparticles are widely used as catalysts in chemical production, environmental cleaning, and renewable energy generation and utilization. A primary factor that influences the activity of bimetallic catalysts is the elemental distribution in the surface and subsurface region. The elemental distribution is strongly influenced by the chemical environment the catalysts are operating in. The surface structure of the catalyst may differ significantly under reaction conditions from as-synthesized due to differences in environmental conditions. To understand the structure-function relationships of bimetallic catalyst, Monte Carlo simulations have been employed to predict the equilibrium elemental distribution under experimentally relevant conditions. However, kinetics of the structural evolution, which provides important mechanistic insights and time scale information, is rarely simulated due to the timescale gap between what can be simulated with molecular dynamics (MD) and experimentally relevant time scales.. Adaptive kinetic Monte Carlo (aKMC) is a promising algorithm, based upon harmonic transition-sate theory, designed to overcome this time scale gap.
In the study, we applied the aKMC method to understand the structural evolution of Au@Pd core-shell nanoparticles with 201 atoms and a truncated octahedral shape. Thermodynamically, Au surface segregation is favorable in Au@Pd nanoparticles, with all of the Au atoms on the surface. Using aKMC, we have simulated ~0.02 s of dynamics, a significantly longer timescale than the 10-6 s achievable by MD. Au surface segregation was observed in the aKMC simulation. The mechanism of Au surface segregation involves surface defects formation, concerted multiple atom rearrangements, and long-distance diffusion events. These remarkably dynamic surface rearrangements facilitate initial Au surface segregation. After the initial Au surface segregation, when the Au atoms occupy corner and edge sites of the nanoparticle, further Au surface segregation is inhibited. For a large portion of the simulated time, the nanoparticle remains in a partially Au surface-segregated state with Au atoms decorating the corners and edges. These simulations reveal differences between the thermodynamically favorable structure and more realistic structures which are kinetically accessible. In summary, our study, employing the aKMC method, effectively bridges the timescale gap and gives insights into the kinetics of surface segregation in Au@Pd bimetallic nanoparticles.
9:15 AM - TC04.06.05
Accelerated Simulations of Long-Term Relaxation in Glasses
Mathieu Bauchy 1
1 , University of California, Los Angeles, Los Angeles, California, United States
Show AbstractAlthough it is indeed commonly believed that, as frozen supercooled liquids, glasses should continue to flow over the years (e.g., in the case of the stained-glass windows of medieval cathedrals), the dramatic increase of their viscosity below the glass transition temperature suggests, on the contrary, that their relaxation time is on the order of 1032 years at room temperature. However, a recent study reported the intriguing dynamics of the relaxation of a commercial Corning® Gorilla Glass® at room temperature, over 1.5 years. Here, we report a novel atomistic simulation method allowing us to directly access the long-term (years) dynamics of glass relaxation at room temperature. Based on the simulation of a series of mixed alkali silicate glasses, we demonstrate that room-temperature relaxation is a direct consequence of the mixed alkali effect. Although both volume and energy feature a stretched exponential relaxation, our results reveal a bifurcation of the stretching exponents, with β = 3/5 and 3/7 for energy and volume relaxation, respectively. Relaxation is found to occur through the diffusion of local stressed structural instabilities inside the atomic network, which anneal each other when a compressed atomic unit meets one that is under tension. The driving force for such diffusion-trap relaxation mechanism is found to be at a maximum when the concentrations of each alkali atom equals each other, which arises from a balance between the concentration of each alkali atom and the magnitude of the local stress that they undergo.
10:00 AM - *TC04.06.06
Towards Multiscale Modeling of Thin-Film Growth Processes
Shree Ram Acharya 1 , Talat Rahman 1 2 3
1 , University of Central Florida, Manhattan, Kansas, United States, 2 Applied Physics, Aalto University, Helsinki Finland, 3 , Donostia International Physics Center, San Sebastian Spain
Show AbstractSome years back a Self-Learning Kinetic Monte Carlo (SLKMC) method [1] was introduced to examine temporal and spatial evolution of two-dimensional adatom islands containing 2-50 atoms on fcc(111) surfaces, unbiased by any set of diffusion processes chosen apriori. The usage of pattern recognition and diffusion path finding schemes enabled collection of large database for each island size, whose saturation signified its completeness. A variety of diffusion mechanisms involving single and multiple atoms and concerted motion of the islands were uncovered in long time (comparable to those in experimental observations) KMC simulations. Kinetically driven island shape changes and characteristics of island coarsening were also revealed in these large length and time scale simulations [2]. In this talk, after review of present status, we will present results for the diffusion kinetics of two dimensional adatoms islands of Ag and Pd on the Ag(111) and Pd(111), and of Cu and Ni islands on Cu(111) and Ni(111) [3], considering for each set both homo and hetro systems, with special attention to the role of lateral interactions as revealed from calculations of energetic based on density functional theory. For example, lateral interactions help explain why irrespective of the substrate, concerted processes dominate in the kinetics of Pd clusters whereas competition among concerted, multi-atom and single-atom processes is obtained in the diffusion of Ag islands. We present prediction of activation energy barriers of single-atom diffusion processes of the above islands, based on linear and non-linear statistical approach. System encoded in terms of easily accessible geometrical and ground state energetic features and the calculated barriers of the single-atom processes are used to train and test the model. It is shown that such a simplified model can make ultra-fast, yet accurate, barrier prediction which otherwise require intensive computational resources. These results are very promising for application of tools suitable for multiscale modeling of thin film growth and morphological evolution of nanostructured systems.
[1] O. Trushin, A. Karim, A. Kara, T. S. Rahman, PRB, 72 (2005).
[2] G. Nandipati, A. Kara, S. I. Shah, and T. S. Rahman, Phys. Rev. B 88, 115402 (2013).
[3] S. R. Acharya, S. I. Shah, and T. S. Rahman, Surface Science, 662, 42 (2017).
* Work supported in part by grants from NSF
10:30 AM - TC04.06.07
Multiscale QM/MM Modelling of Rare Events in Materials Chemomechanics
James Kermode 1
1 , University of Warwick, Coventry United Kingdom
Show AbstractFracture is the dominant failure process underlying many materials reliability issues. At the same time, it remains one of the most challenging ‘multi-scale’ modelling problems, requiring both an accurate description of the chemical processes occurring in the near tip region and the inclusion of a much larger region in the model systems. I will explain how these requirements can be met simultaneously by combining a quantum mechanical description of the crack tip with a classical atomistic model that captures the long-range elastic behaviour of the surrounding crystal matrix, using a QM/MM (quantum mechanics/molecular mechanics) approach such as the `Learn on the Fly’ (LOTF) scheme [1]. I will review recent developments to this scheme, along with applications to slow crack growth [2] and chemically activated fracture [3]. Recent extensions to metallic systems enable processes such as dislocation motion in Ni-based superalloys [4] to be described.
A key limitation of most QM/MM approaches for materials systems is the restriction to dynamical simulations on DFT timescales; due to electronic non-locality a total energy cannot be defined, meaning energy barriers for rare events cannot be calculated. I will discuss a general solution using the principle of virtual work in a modified nudged elastic band (NEB) algorithm [5]. The new method has been tested with ab initio calculations of the kink formation energy for edge dislocations in molybdenum and lattice trapping barriers to brittle fracture in silicon.
[1] G. Csányi et al., Phys. Rev. Lett. 93, 175503 (2004)
[2] J. R. Kermode et al., Phys. Rev. Lett. 115, 135501 (2015)
[3] A. Gleizer et al., Phys. Rev. Lett. 112, 115501 (2014)
[4] F. Bianchini et al., Modell. Simul. Mater. Sci. Eng. 24, 045012 (2016)
[5] T. Swinburne and J.R. Kermode, Submitted (2017)
10:45 AM - TC04.06.08
Cataloging Isomers of Nanopores in Graphene through Multiscale Atomistic Simulations
Ananth Govind Rajan 1 , Daniel Blankschtein 1 , Michael Strano 1
1 , Massachusetts Institute of Technology, Cambridge, Massachusetts, United States
Show AbstractExtended defects or nanopores in graphene can be used to tailor graphene’s electronic, magnetic, and barrier properties. However, the existence of a large number of isomers of nanopores make their study an intractable problem, while confounding the interpretation of experimental data. Herein, we combine extensive electronic-structure density functional theory calculations and kinetic Monte Carlo simulations, in a multiscale modeling approach, to generate a catalog of unique, most probable isomers of graphene nanopores. The calculated first-principles data set provides rates for transitions between nanopore shapes in the graphene lattice consistent with experimental data, and offers a hitherto unexplored reason for the stability of zigzag edges in graphene. The results demonstrate remarkable agreement of our model with precise nanopore shapes observed in experimental microscopy data, such that the methodology presented here is likely to open up several new avenues in the study and molecular design of nanoporous 2D materials for optoelectronics, magnetism, gas separations, desalination, and biological applications.
11:00 AM - TC04.06.09
Finding Minimum Energy Pathways Using Distortion Symmetries—Applications to Ferroelectric Switching
Jason Munro 2 , Haricharan Padmanabhan 2 , Vincent Liu 2 , Long-Qing Chen 2 , Brian VanLeewen 2 , Venkatraman Gopalan 2 , Hirofumi Akamatsu 1 , Ismaila Dabo 2
2 Materials Science and Engineering, The Pennsylvania State University, University Park, Pennsylvania, United States, 1 Institute for Materials Chemistry and Engineering, Kyushu University, Fukuoka Japan
Show AbstractComplex systems often evolve through rare events with time scales largely inaccessible by conventional simulation techniques. The nudged elastic band method is a commonly used algorithm to address this problem, allowing the minimum energy path between the initial and final states of a kinetic process to be determined. However, these calculated pathways are critically dependent on the choice of the initial path, necessitating many trials with different starting points in order to obtain accurate minimum energy pathway predictions. Traditionally, these have been constructed by applying perturbations to the structure at the transition point of the initial path. More recently, a novel approach to the problem has been formulated by considering a path’s distortion symmetry group. This new class of symmetry groups consists of the composition of conventional spatial symmetry operations with an operation called distortion reversal, all of which have been fully listed. [1,2] Using these, exploration of additional pathways that are otherwise inaccessible is enabled through the generation of symmetry-adapted perturbations to the initial path. This new approach has been implemented into a Python module, as well as the open-source Quantum-ESPRESSO software package. [3] It has then been used in the calculation of minimum energy pathways for bulk polarization switching and domain-wall motion in various ferroelectric materials including Ca3Ti2O7, BiFeO3, and LiNbO3. The method not only successfully reproduces previously reported paths, but has also led to the discovery of hidden pathways that reveal a variety of notable physical phenomena.
[1] VanLeeuwen, B. K. & Gopalan, V. Nat. Commun. 6, 8818 (2015).
[2] VanLeeuwen, B. K., Gopalan, V. & Litvin, D. B. Acta Cryst. A 70, 24–38 (2014).
[3] Paolo Giannozzi et al. J. Phys.: Condens. Matter 21, 395502 (2009).
11:15 AM - TC04.06.10
Method for Modeling Spin, Orbital and Charge Ordering and Application to Jahn-Teller Distortions in Layered Battery Materials
Maxwell Radin 1 , John Thomas 1 , Anton Van der Ven 1
1 , University of California, Santa Barbara, Ann Arbor, Michigan, United States
Show AbstractThe electronic degrees of freedom associated with spin, orbital, and charge ordering can lead to complex behavior in many crystalline solids. For example, spin, orbital, and charge ordering can occur simultaneously in many common Li- and Na-ion battery materials upon (de)intercalation. (In these systems, orbital ordering often arises because of the Jahn-Teller activity of Ni3+ and Mn3+ ions in octahedral environments.) The ubiquity of such ordering phenomena combined with the advent of efficient first-principles simulation methods motivates the development of tools to systematically explore electronic degrees of freedom. This work reports on a computational framework extending the Clusters Approach to Statistical Mechanics (CASM) to the ordering of electronic degrees of freedom, and presents a case study on spin, orbital, and charge ordering in layered oxides relevant to Li- and Na-ion batteries. The results provide insight into the shape of the energy landscape, as well as qualitative differences between Li and Na compounds.
11:30 AM - TC04.06.11
Accelerated Green's Function Molecular Dynamics for Materials Science Simulations
Fabio Andrijauskas 1 , Vitor Coluci 1
1 School of Technology, UNICAMP, Limeira Brazil
Show AbstractLimitation of the accessible simulation time in molecular dynamics (MD) simulations is an important issue for studying materials properties in realistic time scales. Techniques such as hyperdynamics, parallel-replica dynamics, and temperature accelerated dynamics have tackled this issue to extend the MD simulation time [1]. In a different approach, the Green's function formalism has been applied to solve the equations of motion in classical MD simulations and has allowed to extend by eight orders of magnitude the time scale of vibration processes in carbon nanomaterials [2]. In Green's function molecular dynamics (GFMD), the interatomic potential is expanded up to the quadratic terms which allows exact solution for problems within the harmonic approximation, larger time steps for non-harmonic potentials, reasonable (10-4%) energy conservation, and fast temporal convergence. However, GFMD exhibits high computational cost (O(N3)) because it uses matrix multiplication and diagonalization within the main MD loop. This high cost has limited the use of GFMD for materials science simulations. In this work, we combined high performance computing techniques to accelerate GFMD simulations. Parallel GFMD versions were implemented to use CPU multithreading (OpenMP) and co-processing (accelerators as Xeon Phi and Nvidia Tesla). The parallelism techniques we used were relatively simple and they were applied mainly to the matrix operations by using typical linear algebra libraries (Magma and cuBlas) and to the integration main loop by using OpenAcc and OpenMP (offload) for Xeon Phi. The performance tests were carried out to one-dimensional lattices comprised of up to 15000 atomic oscillators, limited by the total memory of the available co-processor. The parallel versions were validated with the available analytical solution for the atomic displacements for the linear chain of harmonic oscillators. The speedup scales linearly with the number of atoms for a fixed number of CPU cores. By combining OpenMP and co-processors, a small number of CPU cores (<10) is necessary to reach the maximum speedup. We obtained a maximum speedup of 12 for the OpenMP version, 28 for the combination of Xeon Phi and OpenMP, and 30 for the combination of Nvidia Tesla and OpenMP. These high speedups indicate that relatively small computational resources allow the acceleration of GFMD. For large systems where the interactomic interactions are short range and the matrices are sparse, strategies such as O(N) diagonalization should be applied to make GFMD more competitive.
References
[1] A.F. Voter, F. Montalenti, T. C. Germann, Extending the time scale in atomistic simulation of materials, Annu. Rev. Mater. Res. 32, 321-346 (2002)
[2] V. K. Tewary, Extending the time scale in molecular dynamics simulations: Propagation of ripples in graphene, Phys. Rev. B 80 161409 (2009)
11:45 AM - TC04.06.12
Development of a Simple Molecular Reactive Force Field (SMRFF) Applied to the Nucleation of Hybrid Organic-Inorganic Perovskites in Solution
Henry Herbol 1 , James Stevenson 1 , Paulette Clancy 1
1 , Cornell University, Ithaca, New York, United States
Show AbstractMolecular Dynamics (MD) is a cornerstone of computational materials science, offering the significant advantage over ab initio methods of the ability to conduct large-scale, atomistic studies of complex systems, e.g., crystal nucleation, solvent properties, bulk properties, etc. Unfortunately, the force fields used in MD calculations to represent the functional forms defining the inter- and intra- molecular forces of the system are almost exclusively non-reactive in nature. That is, they cannot handle bond-forming and -breaking on-the-fly. This becomes an issue when trying to study evolving reactive systems, such as those processes that occur during the nucleation of Hybrid Organic Inorganic Perovskites (HOIPs) in solution. Currently, there exist two main contenders for reactive force fields: the Reactive Force Field (ReaxFF; developed by the van Duin group) and the Charge-Optimized Many-Body force field (COMB; developed by the Sinnott group). While these reactive force fields are rooted in fundamental physical science principles, they require extensive efforts to parameterize and are generally complex with large numbers of variables to optimize. There is, thus, some impetus to develop a simpler, more readily accessible reactive force field for the community to help develop.
In 2016, our group developed the first version of a so-called Simple Molecular Reactive Force Field (SMRFF). This force field was a combination of simple potentials: Lennard-Jones (LJ), Morse, Coulomb, and Jorgenson’s all-atom Optimized Potential for Liquid Simulations (OPLS-AA). For a test case to predict energetically optimized structures of PbS nanocrystals, we found that the SMRFF predictions were comparable to those obtained using an ab initio Density Functional Theory (DFT), with errors within those found when using different DFT methods. However, the potential form for that test case did not include three-body interactions, which rendered the original SMRFF formulation unsuitable for application to more complex systems that require directionally constrained geometries, such as the HOIP (hybrid organic-inorganic perovskites) family of materials. In this talk, we present a different functional form for SMRFF that uses a Tersoff potential for close-range interactions, with a smooth transition into Lennard-Jones and Coulombic interactions for longer ranged interactions. We will present preliminary results to showcase the ability of this Tersoff-based SMRFF reactive force field to study HOIP nucleation in solution. Our new approach accomplishes a study that would be computationally prohibitive by other techniques.
TC04.07/ES08.09: Joint Session: Advanced Nuclear Modeling
Session Chairs
Wednesday PM, November 29, 2017
Hynes, Level 2, Room 206
1:30 PM - *TC04.07.01/ES08.09.01
Highlights of Advanced Nuclear Fuel Research within the Nuclear Energy Advanced Modeling and Simulation (NEAMS) Program
Christopher Stanek 1
1 , Los Alamos National Laboratory, Los Alamos, New Mexico, United States
Show AbstractThe US Department of Energy – Office of Nuclear Energy program Nuclear Energy Advanced Modeling and Simulation (NEAMS) is developing a mechanistic computational toolset for nuclear fuel design and/or analysis. Multiscale materials modeling of advanced fuels is an important element of this approach in order to allow the transition from empirical to more mechanistic models. By design, atomic and mesoscale models are necessarily connected to the development of an advanced fuel performance code. In this talk, several highlights of advanced nuclear fuel research within this approach will be provided, including insights in to doped-uranium dioxide and uranium silicide-based fuels.
2:00 PM - TC04.07.02/ES08.09.02
Temperature Accelerated Rate Matrix Construction in the ParSplice Framework
Thomas Swinburne 1 , Danny Perez 1
1 , T-1 Group, Los Alamos National Laboratory, Los Alamos, New Mexico, United States
Show AbstractAtomistic simulations provide essential information to higher order simulation schemes by discovering new system states and evaluating the rate of interstate transitions. However, as the majority of interstate transitions are very rare on typical simulation timescales accelerated techniques are required in order to explore nontrivial regions of state space. Here, we combine the recently developed ParSplice simulation framework with the temperature accelerated dynamics method to construct low temperature rate matrices, optimizing the use of massively-parallel computational resources through an uncertainty quantification scheme. The key concepts will be presented and applications relevant to nuclear materials science will be discussed.
2:15 PM - TC04.07.03/ES08.09.03
Understanding the Amorphization Resistance of Complex Oxides via Machine Learning
Ghanshyam Pilania 1 , Karl Whittle 2 , Chao Jiang 3 , Robin Grimes 4 , Christopher Stanek 1 , Kurt Sickafus 5 , Blas Uberuaga 1
1 , Los Alamos National Laboratory, Los Alamos, New Mexico, United States, 2 , University of Liverpool, Liverpool United Kingdom, 3 , Idaho National Laboratory, Idaho Falls, Idaho, United States, 4 , Imperial College London, London United Kingdom, 5 , University of Tennessee, Knoxville, Knoxville, Tennessee, United States
Show AbstractThe response of complex oxides to irradiation is dictated by a number of factors that are challenging to connect to experimental observables. For example, the critical amorphization temperature Tc, the temperature at which a compound can no longer be amorphized, is an inherently kinetic property that depends on the behavior of multiple defect types in a chemically disordered system. Developing a predictive capability based on first principles for such properties is daunting. Here, we use machine learning to relate Tc to fundamental properties of pyrochlores (A2B2O7). We use basic properties of the elemental constituents as well as DFT-computed phase energetics as features in a kernel-based ridge regression learning framework. We identify the energy of amorphization as a critical feature. This framework enables design maps that estimate Tc as a function of pyrochlore chemistry. This work highlights the utility of machine learning to provide fundamental insight into inherently complex materials problems.
2:30 PM - TC04.07/ES08.09
BREAK
3:30 PM - *TC04.07.04/ES08.09.04
Increasing the Power of Accelerated Molecular Dynamics Methods and Plans to Exploit the Coming Exascale
Arthur Voter 1
1 , Los Alamos National Laboratory, Los Alamos, New Mexico, United States
Show AbstractMany important materials processes take place on time scales that far exceed the roughly one microsecond accessible to molecular dynamics simulation. Typically, this long-time evolution is characterized by a succession of thermally activated infrequent events involving defects in the material. In the accelerated molecular dynamics (AMD) methodology, known characteristics of infrequent-event systems are exploited to make reactive events take place more frequently, in a dynamically correct way. For certain processes, this approach has been remarkably successful, offering a view of complex dynamical evolution on time scales of microseconds, milliseconds, and sometimes beyond. We have recently made advances in all three of the basic AMD methods (hyperdynamics, parallel replica dynamics, and temperature accelerated dynamics (TAD)), exploiting both algorithmic advances and novel parallelization approaches. I will describe these advances, present some examples of our latest results, and discuss what should be possible when exascale computing arrives in roughly four years.
4:00 PM - TC04.07.05/ES08.09.05
Modeling Point Defects in Alloys with DFT, Cluster Expansions and KMC
Normand Modine 1 , Alan Wright 1 , Stephen Lee 1 , Stephen Foiles 1 , Corbett Battaile 1 , John Thomas 2 , Anton Van der Ven 2
1 , Sandia National Laboratories, Albuquerque, New Mexico, United States, 2 Materials Department, University of California, Santa Barbara, California, United States
Show AbstractModeling defects in alloys is a challenging problem because defect properties are sensitive to the occupations of nearby atomic sites and thus vary with location in the alloy. This leads each defect species to have an entire distribution of formation energies, defect levels, activation energies for diffusion, etc. in an alloy. Furthermore, defects can form, diffuse, and annihilate by slow, activated processes with time scales of seconds or even years. Unless the defects are in equilibrium, the distributions of defect properties will change over these same time scales. Density Functional Theory (DFT) allows the accurate determination of ground state and transition state energies for a defect in a particular local environment in the alloy but requires thousands of processing hours for each such calculation. Kinetic Monte-Carlo (KMC) can be used to model the relevant slow, activated processes and the changing distribution of defect properties but requires energy evaluations for millions or billions of local environments. We have used the Cluster Expansion (CE) formalism to “glue” together these seemingly incompatible methods in order to model defect diffusion in alloys. In the CE approach, the occupation of each alloy site is represented by an Ising-like variable, and products of these variables are used to expand quantities of interest. Once a CE is fit to a training set of DFT energies, it allows very rapid evaluation of the energy for an arbitrary configuration, while maintaining the accuracy of the underlying DFT calculations. These energy evaluations are then used to drive our KMC simulations. We will demonstrate the application of our DFT/MC/KMC approach to model thermal and carrier-induced diffusion of radiation-induced point defects in III-V alloys and show that trapping in energetically favorable regions of the alloy leads to a diffusion rate the slows dramatically with time. We will also discuss application of our approach to doping and the formation of compensating defects in semiconductors.
Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC., a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-NA-0003525.
4:15 PM - TC04.07.06/ES08.09.06
Electronic Stopping Power in a Non-Uniform Electron Gas
Magdalena Caro 1 , Artur Tamm 2 , Alfredo Correa 2 , Alfredo Caro 3
1 , Department of Mechanical Engineering, Virginia Polytechnic Institute and State University, Falls Church, Virginia, United States, 2 , Lawrence Livermore National Laboratory, Livermore, California, United States, 3 , Los Alamos National Laboratory, Los Alamos, New Mexico, United States
Show AbstractTheoretical predictions of energy losses for projectiles traveling through solid targets are usually based on results for a uniform electron gas, jellium. In those models, dissipation is a function of the electron gas density.
In this work, we use Time Dependent Density Functional Theory, TD-DFT, to calculate the energy dissipated by energetic projectiles on non-uniform electron density targets, namely the case of binary collisions, and the case of a projectile traveling along a high symmetry direction in an fcc crystal. In particular, we study the case of a Ni projectile traveling in a Ni target along channeling trajectories. We relate the instantaneous dissipation, β, to the local electron density, ρ, experienced by the projectile, and find that β is a multivalued function of the host electronic density, represented by loops in the β-ρ plane, in contrast to all published results describing dissipation in jellium.
We conclude that real inhomogeneous electron gases have a significantly different effect on dissipation, and that jellium results represent an average approximation for the actual dissipation.
This work was supported as part of the Energy Dissipation to Defect Evolution (EDDE), an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences (Award Number 2014ORNL1026).The authors acknowledge computing support from the Lawrence Livermore National Laboratory Institutional Computing Grand Challenge program.
4:30 PM - TC04.07.07/ES08.09.07
Modeling Radiation Damage Using SRIM, MD and AKMC
Steven Kenny 1 , Mark Wootton 1
1 , Loughborough University, Loughborough United Kingdom
Show AbstractA model of an experiment which uses proton based radiation damage to simulate material damage in a nuclear power plant has been created using a combination of SRIM, molecular dynamics and adaptive kinetic Monte Carlo. Through the use of this model multiple radiation damage events in an iron chrome material have been simulated with collision cascade energies and rates chosen from a distribution consistent with bombardment by 3 MeV protons. These results will be compared and contrasted with a model where the material is simulated purely using molecular dynamics, where although the collision cascade energies can be matched the rate of bombardment is many orders of magnitude too high. The results show that a significant amount of annealing of defects take place in the material when realistic timescales are modelled between cascade events. This would be expected to lead to significant changes in the damage evolution in the material.
4:45 PM - TC04.07.08/ES08.09.08
Accelerated Quantum Molecular Dynamics
Enrique Martinez 1 , Christian Negre 1 , Danny Perez 1 , Marc Cawkwell 1 , Arthur Voter 1 , Anders Niklasson 1
1 , Los Alamos National Laboratory, Los Alamos, New Mexico, United States
Show AbstractThe accurate study of the long-term evolution of rare events is extraordinarily challenging as computations are arduous and quantum-based molecular dynamics simulation times are limited to, at most, hundreds of ps. Here, the Extended Lagrangian Born-Oppenheimer molecular dynamics formalism is used in conjunction with Parallel Replica Dynamics to obtain an accurate tool to describe the long-term dynamics of reactive benzene. Langevin dynamics has been employed at different temperatures to calculate the first reaction times in a periodic benzene sample at different pressures. Our coupled engine run for times on the order of the ns (two to three orders of magnitude longer than traditional techniques) and is capable of detecting reactions characterized by rates significantly lower than we could study before.
Symposium Organizers
Enrique Martinez, Los Alamos National Laboratory
Graeme Henkelman, University of Texas
Hannes Jonsson, University of Iceland
Steven Kenny, Loughborough University
Symposium Support
Modelling and Simulation in Materials Science and Engineering | IOP Publishing
TC04.08: Monte Carlo
Session Chairs
Tapio Ala-Nissila
Enrique Martinez
Thursday AM, November 30, 2017
Hynes, Level 2, Room 202
8:00 AM - *TC04.08.01
Adaptive Kinetic Monte Carlo Methods for Thin-Film Growth
Adam Lloyd 1 , Steven Kenny 1 , Roger Smith 1
1 , Loughborough University, Loughborough United Kingdom
Show AbstractWe present off and on lattice adaptive kinetic Monte Carlo (AKMC) methods with application to thin film growth. A super basin method is used to extend time-scales achievable in AKMC when small energy barrier transitions dominate a long time-scale simulation. As finding a set of possible transitions from each state is the most computationally expensive part of the AKMC method, transition searches are concentrated on canonically labelled regions (defect volumes) surrounding surface defects and ad-atoms. These regions are then stored along with successful transition searches for reuse later in a simulation.
Off-lattice methods can determine complicated growth mechanisms which may include concerted motions of atoms. These methods can also be used to model thin film growth on defective surfaces which may include phase boundaries, step edges and amorphous regions. The method works well in the initial stages of growth but can require excessive numbers of saddle point searches if the surface grows in a complex manner. In these cases, and especially if the growth is dominated by single atoms moves between lattice sites, then it is relatively straightforward to implement a lattice based adaptive approach to model growth. An on-lattice AKMC model can extend time-scales that can be simulated in the off-lattice method and evolve much larger systems. Examples of thin film growth of Ag on ZnO surfaces (an interface common in low-e window production) are shown which use both approaches.
8:30 AM - *TC04.08.02
The Kinetic Activation-Relaxation Technique—An Off-Lattice Kinetic Monte Carlo Algorithm with On-the-Fly Catalog Building for Complex Materials
Normand Mousseau 1
1 Department of Physics, Université de Montréal, Montreal, Quebec, Canada
Show AbstractIn spite of considerable advances in computational capacities over the last decades, there remains a considerable gap between experimentally relevant time scales and those accessible to atomistic simulations. This gap reflects the fundamentally multi scale nature of atomistic kinetics that can only be lifted partially through approximate methods that attempt to capture the most important aspect of specific phenomena. Among those approaches, the kinetic activation-relaxation technique (k-ART) is an off-lattice kinetic Monte Carlo with on-the-fly cataloging capabilities that allows fully atomistic second-long and more simulations of complex alloys and amorphous systems such as amorphous silicon and steels, while incorporating exactly elastic effects. In this talk, I'll present the k-ART method, recent applications to various systems and its advantages and limitations in the study of complex materials.
The work presented here was done in collaboration with C. Becquart, L. K. Béland, O. Bouhali, P. Brommer, F. El-Mellouhi, A. Hemeryck, A. Jay, J.-F. Joly, S. Mahmoud, O. Restrepo, and M. Trochet. It was supported, in part, by NSERC, the Canada Research Chair Foundation, the Université de Montréal and the Qatar National Research Fund.
9:00 AM - TC04.08.03
Predicting the Morphology of Silver Nanoparticles from Integrated Approach of First-Principles Calculations and Monte Carlo Simulation
Hosna Sultana 1 , Eunseok Lee 2
1 Optical Science and Engineering, University of Alabama, Huntsville, Huntsville, Alabama, United States, 2 Mechanical and Aerospace Engineering, University of Alabama, Huntsville, Huntsville, Alabama, United States
Show AbstractNoble-metal nanoparticles (NP) attract attentions for their excellent material properties, such as optical properties, catalytic behaviors, etc. These properties should be tailorable by controlling the morphology of NP, for their size, shape, and crystallinity. However, obtaining the desired morphology is challenging due to the lack of comprehensive understanding of the nucleation and growth of NP. In this study, we present an integrated approach of first-principles calculations and Monte Carlo method to simulate the nucleation and growth of NP at multi-scales and elaborate the underlying mechanisms. The nucleation and growth process was modeled as a consecutive series of Monte Carlo (MC) event, in which two particles are selected and coalescence into one particle. We calculated the formation energy of small size atomic clusters (the number of atoms less than 600) using the density functional theory (DFT) calculations. The energy of larger-size particles was calculated by parameterization of the DFT data. Based on the resultant energy profile, MC simulation was performed. Our model also accounts for the kinetic effects: e.g. the site-dependent activation energy barrier of the adsorption will deviate the adsorption site from the one in the equilibrium state. We implemented these kinetic factors in the kinetic MC simulations to predict the morphology and size distribution of NP. The comparison with classical theories and their modification is also presented.
9:15 AM - TC04.08.04
GPU-Accelerated Kinetic Lattice Monte-Carlo for Experimental-Scale Studies
Jeffrey Kelling 1 , Karl-Heinz Heinig 1 , Martin Weigel 3 , Gemming Sibylle 1 2
1 , Helmholtz-Zentrum Dresden-Rossendorf, Dresden Germany, 3 Applied Mathematics Research Centre, Coventry University, Coventry United Kingdom, 2 Institute of Physics, TU Chemnitz, Chemnitz Germany
Show AbstractMicro- and nano-structured materials, including composites, are crucial for
future energy technologies. Key processes during production and life-time are
governed by self-organization in phase separation processes at the micro and
nano scale. Examples include nano-structured Silicon thin film absorber
layers in solar cells providing tailored band-gaps [1] on top of reduced
production cost. In the case of micro-patterned electrolyte-matrices, used in
a range of fuel cell technologies, both production and aging are governed by
phase separation and affect the efficiency and lifetime of large industrial
installations.
Simulations of these out-of-equilibrium, inhomogeneous real world systems
provide important insights, finding reaction pathways for self-organization
and self-alignment of nanostructures. To this end, 3D kinetic Metropolis
lattice Monte Carlo simulations can be used to model physical systems at
experimental scales in an atomistic way, thereby side-stepping many caveats
connected with the alternative phase-field simulations.
These same long-time and large-scale simulations also provide important
insights into more fundamental physical problems. The question of
super-universality, that is if and how different types of quenched disorder
affect universal properties, is still under investigation even for
fundamental models like Ising [2], realizations of which can be found in
important complex magnetic systems, apart from binary mixtures.
We propose massively parallel simulation techniques using the architecture of
modern graphics processing units (GPUs) to address these problems, ranging
from the kinetic Metropolis Monte Carlo to any Potts models with quenched
disorder. While pioneering work in this area [3] focused on efficient but
correlated stochastic cellular automaton implementations, our simulations can
be virtually correlation-free [4].
Here, we present two implementations for large-scale simulations on GPUs: One
is optimized to offer fast time-to-solution on experimental-scale simulations
[5], the other provides highly efficient parameter studies or large sample
sizes for large-scale simulations [6]. Harnessing the compute power of
modern (multi-)GPU installations leads to increased energy efficiency as well
as reduced time-to-solution.
[1] Apl. Phys. Lett. 103, 133106 (2013); Appl. Phys. Lett. 103, 203103 (2013);
Nanolett. 16, 1942 (2016)
[2] e.g. EPL 117, 10012 (2017); Phys. Rev. B 88 042129 (2013);
Phys. Rev. B 78 224419 (2012); J. Phys. A 24 L1087 (1991)
[3] J. Comp. Phys. 228, 4469 (2009); Phys. Proc. 15 92 (2001)
[4] http://arxiv.org/abs/1705.01022 ; https://arxiv.org/abs/1701.03638
[5] EPJST 210, 175 (2012)
[6] Phys. Rev. E 94 022107 (2016)
9:30 AM - TC04.08.05
Quantitative Calculation of Reaction Rates of Two Classes of Chemical Reactions by Infrequent Metadynamics Simulations
Luiz Fernando Lopes Oliveira 1 , Christopher Fu 1 , Jim Pfaendtner 1
1 , University of Washington, Seattle, Washington, United States
Show AbstractIt is common that interesting physicochemical phenomena present high free energy barriers. Examples exist in different fields such as protein folding, prediction of crystal structures and chemical reactions. Observation of such rare events by means of standard molecular dynamics simulations is usually quite difficult due to time-scale limitations. One means to address this issue has been through the use of so-called enhanced sampling methods. Such methods are helpful for accelerating the mapping of the free energy landscape but, by themselves, typically do not provide kinetics.
Our approach uses the metadynamics (MetaD) family of methods combined with the infrequent metadynamics algorithm to post-process the results. Infrequent MetaD has previously been used to estimate the rates of chemical reactions, protein-ligand unbinding, Kramers’ turnover and the unbinding of ligand-substrate. After presenting the method, we will show its application to unraveling kinetic properties of two reactions. First, infrequent MetaD is applied to the study of intramolecular hydrogen transfer that takes place in peroxide species. In comparison with previous studies, carried out by infrequent metadynamics, it presents an additional challenge: high entropic contributions in the formation of ring-structures in the TS. We performed our study using, as a Hamiltonian, the self-consistent-charge density-functional based tight-binding (DFTB) method. By modeling four reactions, we are able to demonstrate that we can qualitatively reproduce the phenomena observed in experiments and those predicted by transition state theory modeled by higher levels of theory. The second class of reactions we will present is the mechanism of cyclohexaphenylene (CHP) dehydrogenation leading to the formation of tribenzocoronene (TBC) on a Cu(111) surface. The feasibility of the method for reactions on surfaces is demonstrated using a QM/MM type of approach in which the surface is described by many-body classical EAM potential and the molecules being adsorbed are described by the PM6 method. We present the potential energy landscape of this mechanism calculated solely from infrequent MetaD simulations and study the time and temperature dependence of the system by pairing the rates from infrequent MetaD with a kinetic Monte Carlo scheme.
10:15 AM - *TC04.08.06
Temperature Programmed Molecular Dynamics—Theory and Applications
Abhijit Chatterjee 1
1 Department of Chemical Engineering, Indian Institute of Technology Bombay, Mumbai, Maharashtra, India
Show AbstractTemperature programmed molecular dynamics (TPMD) method is a recent addition to the class of temperature accelerated schemes. TPMD can be used to efficiently study the long timescale dynamics of a material system. A situation commonly encountered in several complex materials is that the system remains trapped for long periods of time in a collection of potential basins in the energy landscape, called a superbasin. Transitions between superbasin states can be rare at molecular dynamics timescales, however, escapes from the superbasin involving larger activation barriers are even rarer. Often, it is the latter type of moves that are of interest when long timescales of seconds and beyond need to be accessed. TPMD method employs a temperature program with state-constrained molecular dynamics calculations that allows transitions of interest to happen more frequently. Separation of timescales is exploited for identifying the superbasin and selecting a superbasin escape for a move. Using TPMD method one can accurately study superbasin-to-superbasin transitions while disregarding low-barrier pathways that have been traditionally difficult to handle with rare-event simulations. I will describe some of the other key features of the TPMD method such as estimation of Arrhenius parameters for kinetic pathways. Application of the technique to various materials problems will be discussed.
10:45 AM - *TC04.08.07
A Rejection Scheme for off Lattice Kinetic Monte Carlo
Tim Schulze 1 , Hamza Ruzayqat 1
1 , University of Tennessee, Knoxville, Tennessee, United States
Show Abstract
While most Kinetic Monte Carlo (KMC) simulations are lattice based,
many important technological applications involve multi-component
systems where lattice mismatch leads to elastic strain and crystal
defects, neither of which can be accurately modeled with a lattice
based approach. Off-lattice Kinetic Monte Carlo (OLKMC) is aimed at overcoming
these limitations. Fully general off-lattice simulations make use of
either an empirical potential or an even more costly density
functional theory calculation, seeking to exhaustively calculate the
transition path to all of the neighboring states within the
multi-particle configuration space.
Thus, a fully implemented off-lattice simulation is an enormously complex
task when compared to lattice based simulations, where rates can be
precomputed and stored. So much so that KMC simulation loses much of
its utility and applications of these methods are often limited to systems
with only a few hundred atoms, simulated for much shorter times, and
at much greater computational cost.
The principal aim of this talk is to discuss a rejection-based Monte Carlo procedure that uses local environment information to create rate estimates. These estimates can be computed in a relatively inexpensive way and allow one to select a candidate move without computing detailed rates for the entire system.
11:15 AM - TC04.08.08
Chemically Transferable Kinetic Monte Carlo Models with Thousands of Reactions Learned from Molecular Dynamics Simulation of Hydrocarbon Chemistry
Enze Chen 1 , Qian Yang 1 , Carlos Sing-Long 2 , Evan Reed 1
1 , Stanford University, Stanford, California, United States, 2 , Pontificia Universidad Católica de Chile, Santiago Chile
Show Abstract
Molecular dynamics (MD) simulations of complex chemical systems at high temperatures and pressures are typically analyzed by considering only a few dominant processes. However, we discover that these simulations can contain a wealth of kinetic information that can be leveraged to construct truly predictive models. Towards this goal, we develop a method for statistical learning of kinetic Monte Carlo (KMC) models from a MD simulation initialized with 64 molecules of isobutane at 3300 K and 40.53 GPa. We discover that the trained KMC models can extrapolate to chemical systems with substantially different initial compositions by testing the KMC models on a separate chemical system initialized with 216 molecules of methane at the same temperature and pressure. Remarkably, we discover that the molecular concentration trajectories of the dominant species are comparable between simulations of our KMC model and the MD test data.
Furthermore, we find that we can reduce the total number of 8376 reactions in the trained KMC model by 92.5% down to 631 reactions without increasing the simulation error of the dominant species. The reduction in the complexity of our model allows us to focus on only the salient reactions in our system. Multiple simulations of our KMC model can be performed by simply adjusting initial molecular concentrations without the generation of extra MD data. Since our fast KMC model can simulate the same chemical system as MD in a matter of minutes rather than weeks and is transferable to different chemical systems, this suggests a path forward for reusing MD simulations to quickly simulate related chemical systems with KMC methods.
11:30 AM - TC04.08.09
Time Extrapolation of Molecular Dynamics Simulations via Statistically Learned Kinetic Monte Carlo Models
Qian Yang 1 , Carlos Sing-Long 2 , Evan Reed 1
1 , Stanford University, Stanford, California, United States, 2 , Pontificia Universidad Catolica de Chile, Santiago Chile
Show AbstractWe develop a novel method for extending the timescales of atomistic simulation by using statistical learning to train a kinetic Monte Carlo (kMC) of chemical reactivity from a high temperature and pressure molecular dynamics simulation (MD). We find that the learned kMC models are able to extrapolate the concentrations of the largest population species for an order of magnitude longer in time than the MD data that the kMC model was trained on. For example, we show that using only 25 picoseconds of molecular dynamics data for a system initialized with 216 molecules of methane, our learned kMC model is able to extrapolate the concentration of methane in the system over more than 200 picoseconds. Similar time extrapolation results are achieved for the growth of large carbon clusters in the same system. Furthermore, we find that model reduction using a new data-driven L1-regularization method that we developed improves the accuracy of our learned kMC models when extrapolated over time, for both the largest population species and large carbon clusters.
Molecular dynamics simulations on the nanosecond timescale of large systems with thousands of atoms currently require weeks of computation on high performance parallel machines, while the corresponding learned kMC models require merely minutes on a laptop computer. Our results suggest a path forward for significantly extending the timescales on which complex chemistry can be simulated by using short MD simulations to build kMC models that can then rapidly extrapolate the system out in time.
11:45 AM - TC04.08.10
Phase Field Crystal Modeling Using Transformation Matrices—An Application to Lithium Battery Electrodes
Ananya Renuka Balakrishna 1 , W Craig Carter 1
1 , Massachusetts Institute of Technology, Boston, Massachusetts, United States
Show AbstractThe crystal structures of materials typically span a wide range of irregular lattice geometries. Modeling these irregular lattice features using phase field crystal (PFC) techniques is a challenge. At present, PFC techniques to stabilize irregular crystal geometries include, tuning non-linear resonances [1] and introducing higher order derivatives in the free energy functional [2]. Both these methods make the computation intensive and lack a systematic framework to model a variety of crystal symmetries. Alternatively, the free energy equation can be constructed using correlational terms [3], however this approach is limited to model a narrow/finite set of crystal geometries. Furthermore, the current PFC techniques cannot be applied directly to model phase transformation in ordered materials.
In the present work, we propose a new technique based on the concept of affine lattice transformations [4], to solve a PFC model in its simplest form [5]. In this technique, we employ a 2D transformation matrix to deform a regular hexagon cell structure to an irregular crystal geometry. This technique does not require additional energy terms, and provides a systematic approach to stabilize a wide class of irregular crystal geometries. We next extend the transformation matrix as a function of an order parameter, to describe different lattice structures in multi-phase systems. Using this technique, we simulate phase transformtion with dynamic lattice reorientations [6], and apply the model to investigate the structural changes in lithium battery electrodes during an electrochemical cycle. The results provide insights on how lattices distort to maintain coherency between different phases, and how lattice defects/misfits interact with a moving phase boundary. The present work establishes a simple theoretical framework to investigate, in-situ, the role of material crystallography during phase transformation.
References
[1] K.-A. Wu, M. Plapp and P. W. Voorhees, J. Phys.: Condens. Matter 22, 364102 (2010).
[2] R. Prieler, J. Hubert, D. Li, B. Verleye, R. Haberkern and H. Emmerich, J. Phys.: Condens. Matter 21, 464110 (2009).
[3] M. Greenwood, N. Provatas and J. Rottler, Phys. Rev. Lett. 105, 045702 (2010).
[4] J. M. Ball and R. D. James, Archive for Rational Mechanics and Analysis 100, 13 (1987).
[5] K. R. Elder and M. Grant, Phys. Rev. E 70, 051605 (2004).
[6] A. Renuka Balakrishna and W. C. Carter, Phase field crystal modeling using transformation matrices, Manuscript in preparation, (2017).
TC04.09: Machine Learning
Session Chairs
Thursday PM, November 30, 2017
Hynes, Level 2, Room 202
1:30 PM - TC04.09.01
A Machine Learning Approach to Predict the Energy Levels of a Material from Its Atomic Structure
Javad Hashemi 1 , Alireza Khorshidi 1
1 , Brown University, Providence, Rhode Island, United States
Show AbstractWhile accurate quantum physical calculations are becoming more feasible through ever increasing computational capability, performing hundreds of calculations for a mid size system in a short amount of time is still a far-fetched dream for scientists. This is where machine learning methods step in to predict the outcomes of new calculations based on the previous ones in a fraction of their computational cost. Although, atomic structure is the most definitive indicator of the physical and chemical properties of a compound, it only has been used to predict the total energy thus far, and its potential to predict other physical properties has not been exploited. Here we propose a novel approach to predict energy levels of materials (HOMO, LUMO, defect levels, etc.) based on thier atomic structure. This approach will be specially useful to be trained against more evolved methods where computational cost is significant even for small systems.
1:45 PM - TC04.09.02
A Hybrid Quantum-Mechanics/Machine-Learning Scheme for Large-Scale Atomistic Simulations
Yin-Jia Zhang 1 , Alireza Khorshidi 1 , Georg Kastlunger 1 , Andrew Peterson 1
1 , Brown University, Providence, Rhode Island, United States
Show AbstractElectronic-structure calculations such as those employing Kohn-Sham density-functional theory have allowed for accurate atomistic-level simulations in materials science. However, with the currently existing computational facilities, these methods tend to be limited to systems of at most a few hundred atoms, necessitating the need for the development of new computational methods for large length- and time-scale simulations. Machine-learning techniques can provide accurate potentials that can, in principle, match the quality of electronic-structure calculations, provided sufficient training data. In this talk we will review the theory of machine-learning interatomic potentials and the implementation within the open-source Atomistic Machine-learning Package Amp [1]. We then discuss how machine-learning potentials provide particular advantages as the molecular mechanics simulator in the well-known framework of hybrid quantum-mechanics / molecular-mechanics methods.
References
[1] A. Khorshidi, A.A. Peterson; Amp: A modular approach to machine learning in atomistic simulations, Computer Physics Communications, 2016.
[2] Y. Zhang, A. Khorshidi, G. Kastlunger, A.A. Peterson; QM/ML: A simple quantum-mechanics / machine-learning algorithm, under review.
2:00 PM - TC04.09.03
Graph Representation of Periodic Systems for Accurate and Explainable Prediction of Material Properties via Machine Learning
Tian Xie 1 , Jeffrey Grossman 1
1 , Massachusetts Institute of Technology, Cambridge, Massachusetts, United States
Show AbstractMachine learning (ML) models are being developed with the aim of designing new materials with accuracy close to ab-initio calculations, but speed orders of magnitude faster. Due to the arbitrary size of periodic crystal systems, crystals need to be represented in fixed length to be compatible with most ML algorithms. In existing methods, this problem is resolved by representing crystals with a fixed-length vector converted from atom coordinates under required symmetry invariance. However, one drawback of such an approach is that chemical insights are lacking because of the “black-box” nature of ML models.
In this work, we develop a framework for representing periodic crystal systems by directly building neural networks on top of periodic graphs generated from crystal structures, which provides both material property prediction with DFT accuracy and atomic level chemical insights. The accuracy of our approach is demonstrated by predicting the formation energy, band gap, and fermi energy of inorganic crystals, using calculated data from the open Materials Project database which includes compounds from simple metal oxides to complex minerals. The method shows mean absolute error (MAE) on crystals outside the training set close to the MAE of DFT calculated properties compared to experimental results. More importantly, we demonstrate the ability of this method to provide chemical insights by considering an example of perovskites. We are able to predict the relative stability of different sites in the perovskite structure despite the fact that the model is only trained with total formation energies of the perovskite crystals. The trends given by our model are in good agreement with chemical intuition.
2:15 PM - TC04.09.04
Machine Learning Configurational Energy of Multicomponent Solid-Solutions—Ab Initio Calculation of Material Properties and Parameters
Mahesh Chandran 1 , S. C. Lee 1 , JaeHyuk Shim 2
1 , Indo-Korea Science and Technology Centre (IKST), Bangalore India, 2 High Temperature Energy Materials Research Centre, Korea Institute of Science and Technology, Seoul Korea (the Republic of)
Show AbstractAlloys used in engineering applications are multicomponent and multiphase solid solutions with no unique ground state configuration of atoms in the lattice. This makes it computationally challenging to perform abinitio calculations using Density Functional Theory (DFT). The current state-of-the-art approach to this problem is the cluster expansion (CE) combined with DFT [1,2], which, though accurate, is computationally expensive for routine calculations when number of components in the material is greater than 3 or 4 [3].
Recently, a new approach based on machine learning (ML) of the configurational energy E(σ) to calculate material properties and parameters of a multicomponent solid solution have been proposed [4]. The ML model EML(σ*) trained with DFT-calculated energies predicts E(σ) with accuracy comparable to CE. The feature vector σ* for ML is constructed by concatenating histograms of spatial correlation functions of σ for all combinations of the components. The predictive accuracy of the ML model obtained using only pair correlation functions is improved by adding triplet correlation functions (calculated for only equilateral triangles). The EML(σ*) is used to sample the configurational space rapidly to generate full distribution P(E(σ))≡P(E) of the configurational energy. The P(E) contains all necessary information of the solid solution and can be used for further calculations, for e.g., P(E) can be selectively sampled for targeted DFT calculations. The ML+DFT framework have been employed to calculate (100) interface energy σIE in Ni-based superalloy, Alloy 617, reduced to 5 components [4]. The ML is performed using a dataset of 300 and 900 γ and γ' energies calculated using DFT. The EML(σ*) for γ' predicted total energies with maximum error of ≈25 meV (for a 64-atom system) whereas the total energies for γ is predicted with maximum errors ≈90 meV (higher errors in γ energy is due to much larger configurational space compared to γ'). From a set of interface structures, constructed by sampling P(E) for γ and γ' energies independently, the most-probable interface gave σIE ≈25.95 mJ/m2 which is in good agreement with the value from precipitation model fit to experimental data.
The proposed new ML+DFT framework is a computationally less expensive approach and can be easily extended to multicomponent alloys with any number of components. This opens up route to abinitio calculation of material properties and parameters required for higher length-scale models.
[1] J. M. Sanchez, F. Ducastelle, D. Gratias, Physica A 128, 334 (1984).
[2] G. L. W. Hart, V. Blum, M. J. Walorski, and A. Zunger, Nature Materials 4, 391 (2005).
[3] Mahesh Chandran, Comp. Mat. Sci. 108, 192 (2015).
[4] Mahesh Chandran, S. C. Lee, and J. H. Shim, (submitted).
2:30 PM - TC04.09.05
Determination of Force-Matched Machine Learning Potentials with Explicit Three-Body Interactions
Nir Goldman 1 , Rebecca Lindsey 1 , Larry Fried 1
1 , Lawrence Livermore National Laboratory, Livermore, California, United States
Show AbstractMolecular dynamics simulations (MD) hold promise as an independent route to determining physical and chemical properties of materials under reactive conditions. Such studies can provide simple chemical pictures of ionized intermediates and reaction mechanisms, and can help identify atomic-scale properties that determine observed macroscopic kinetics. Difficulty arises in determining models for chemical bonding that are both accurate and computationally efficient. Kohn-Sham Density Functional Theory (DFT) has been shown to accurately reproduce the phase boundaries and thermal decomposition of many materials, particularly at higher thermodynamic conditions. However, due to its extreme computational cost, simulations are generally limited to 10’s of picoseconds and nanometer system sizes, frequently far less than the time and length scales probed by experiments. Classical force fields usually exhibit many orders of magnitude increase in computational efficiency and scalability. However, these methods are generally fit to the energetics of small molecular clusters or very specific bulk data, and consequently can yield poor results outside of their fitting regime.
To address these issues, we present a method for creating machine learning models for molecular dynamics simulations of materials under extreme conditions by determining linear combinations of Chebyshev polynomials though force matching to trajectories from DFT. We apply our method to liquid carbon near the diamond/graphite/liquid triple point and beyond, where the material exhibits metallization and many-body effects can be substantial. We show that inclusion of explicit terms for three-body interactions yields improved descriptions of both dynamic and structural properties over previous empirical potential efforts, and exhibits transferability to nearby state points. The simplicity of our functional form combined with the rapidity with which new potentials can be determined allows for DFT to be extended to experimental time and length scales while retaining most of its accuracy. Ultimately, our simulation approach is completely general and will have particular impact in research areas where there is traditionally a reliance on expensive DFT calculations for interpretation of imaging and spectroscopy experiments, and prediction of properties to guide materials synthesis.
3:00 PM - TC04.09.06
A Universal Strategy for the Creation of Machine Learning Based Atomistic Force Fields
Rohit Batra 1 , Huan Tran 1 , James Chapman 1 , Sridevi Krishnan 1 , Lihua Chen 1 , Rampi Ramprasad 1
1 , University of Connecticut, Storrs, Connecticut, United States
Show AbstractEmerging machine learning (ML) based approaches provide powerful and novel tools to study a variety of physical and chemical problems.[1-5] In this contribution, we outline a universal strategy to create ML based atomistic force fields, which can be used to perform high-fidelity molecular dynamics simulations. This scheme involves (1) preparing a big reference dataset of atomic environments and forces with sufficiently low noise, e.g., using density functional theory or higher-level methods, (2) utilizing a generalizable class of structural fingerprints for representing atomic environments, (3) optimally selecting diverse and non-redundant training datasets from the reference data, and (4) proposing various learning approaches to predict atomic forces directly (and rapidly) from atomic configurations. From the atomistic forces, accurate potential energies can then be obtained by appropriate integration along a reaction coordinate or along a molecular dynamics trajectory.[3,5] Based on this strategy, we have created model ML force fields for six elemental bulk solids, including Al, Cu, Ti, W, Si, and C, and show that all of them can reach chemical accuracy. The proposed procedure is general and universal, in that it can potentially be used to generate ML force fields for any material using the same unified workflow with little human intervention. Moreover, the force fields can be systematically improved by adding new training data progressively to represent atomic environments not encountered previously.
References:
[1] Behler et al., Phys. Rev. Lett. 98:146401 (2007).
[2] Bartok et al., Phys. Rev. Lett. 104:136403 (2010).
[3] Botu et al., Phys. Rev. B 92:094306 (2015).
[4] Li et al., Phys. Rev. Lett. 114:096405 (2015).
[5] Botu et al., J. Phys. Chem. C 121:511-522 (2017).
TC04.10: Beyond Atomistic Methods
Session Chairs
Thursday PM, November 30, 2017
Hynes, Level 2, Room 202
3:15 PM - *TC04.10.01
Step Instabilities in Fe/Cu(100) Growth
Jacques Amar 1 , Yunsic Shim 1
1 , Univ of Toledo, Toledo, Ohio, United States
Show AbstractWe examine the step instability observed in Fe growth on a Cu(100) vicinal substrate [1] in which deposition of a relatively low coverage of Fe leads to a dramatic increase in the step roughness, with a change in the step morphology from relatively straight [100] steps to a mixture of [100] and [110] steps. Our temperature-accelerated dynamics (TAD) simulations and energetics calculations [2] indicate that the dramatic change in the step morphology is due to a variety of unexpected complex multiatom interlayer diffusion (MID) processes near step-edges which lead to a competition between [100] and [110] steps, and whose barriers are significantly reduced due to the existence of strong Fe-Fe and Fe-Cu interactions as well as strain effects. In contrast, TAD simulations of vicinal Cu/Cu(100) growth do not lead to an instability, in good agreement with experiments. Our results also indicate that while the instability is driven by energetics, kinetics plays a crucial role. The tendency of small Fe clusters to form bcc-like structures will also be discussed along with the effects of low barriers on our simulations [2-4]. We also present the results of TAD simulations and energetics calculations for the case of Fe/Cu(100) growth on vicinal surfaces with [110] steps. Our results in this case indicate that a variety of effects, including MID processes as well as pinning of edge-diffusing atoms, lead to the formation of protrusions with [110] steps perpendicular to the original [110] steps. These results may also explain the instabilities observed experimentally in Ni/Cu(100) and Co/Cu(100) growth on vicinal surfaces with [110] steps. Our results also indicate that in this case the instability is purely kinetic and is not driven by energetics.
[1]. F. Dulot, B. Kierren, and D. Malterre, Surf. Sci. 494, 229 (2001).
[2]. Y. Shim and J.G. Amar, Phys. Rev. Materials 1, 043403 (2017).
[3]. Y. Shim, N.B. Callahan, and J.G. Amar, J. Chem. Phys. 138, 094101 (2013).
[4]. Y. Shim and J.G. Amar, J. Chem. Phys. 145, 014105 (2016).
* Supported by NSF DMR-1410840
3:45 PM - *TC04.10.02
Kinetics of Fivefold-Twinned Nanowire Growth in Colloidal Syntheses
Kristen Fichthorn 1 , Xin Qi 1
1 , Pennsylvania State University, University Park, Pennsylvania, United States
Show AbstractThere has been significant emphasis recently on the synthesis of fivefold-twinned Ag and Cu nanowires, which are considered to be excellent candidates for transparent conductors in flexible and stretchable electronic devices. A fundamental understanding of nanowire growth is important in achieving optimal syntheses. Nanowires grow from fivefold-twinned seeds, which are likely Marks decahedra, with {111} end facets, {100} side facets, and small {111} notches at the corners of the pentagonal cross-section. High-resolution transmission electron microscopy studies have shown that the {100} side facets can achieve a stepped structure, that runs parallel to the nanowire axis and is likely associated with the relief of strain in these structures. Our climbing-image nudged-elastic band method calculations of diffusion barriers based on embedded-atom method potentials indicate that diffusion in the {111} notches and along step edges is significantly faster than diffusion on {100} facets. Thus, these structures become “superhighways” that channel atom diffusion to the wire ends to increase wire aspect ratios. We use kinetic Monte Carlo simulations and finite Markov chains to model nanowire growth and to predict net atom fluxes from nanowire “sides” to the “ends”. These simulations show that wire aspect ratios depend on the degree of faceting that can be achieved on the {100} side facets and this is consistent with experiments. Large-scale molecular dynamics simulations of solution-phase nanowires covered with typical experimental capping molecules indicate that these molecules likely stabilize nanowire faceting to facilitate anisotropic growth.
4:15 PM - TC04.10.03
Incorporating Hydrogen-Diffusion and Dislocation-Hydrogen Interactions into Large Scale Discrete Dislocation Dynamics Simulations of Metals
Yejun Gu 1 , Jaafar El-Awady 1
1 , Johns Hopkins University, Baltimore, Maryland, United States
Show AbstractHydrogen (H) embrittlement remains one of the critical failure mechanisms in metals. However, a comprehensive understanding of the effects of H-diffusion and H-dislocation interactions on dislocation microstructure evolution, damage accumulation, and subsequent H-induced intergranular fracture is still lacking. One of the challenges to study this problem at the atomic scale is the significantly large time scales of H-diffusion compared to those accessible by molecular dynamics simulations. In addition, the length scales required to capture dislocation microstructure evolution are typically orders of magnitude larger than those comprehensible by atomistic simulations. As such, we present here a new three-dimensional (3D) discrete dislocation dynamics (DDD) framework to quantify the effect of hydrogen on the evolution of dislocation plasticity in metals. The first order elastic interaction energy associated with the hydrogen induced volume change due to presence of hydrogen is accounted for. Furthermore, the diffusion of hydrogen is computed by diffusion equations at the continuum level. We apply different proper approaches to solve these diffusion equations in different scale regimes, i.e., transient diffusion and steady-state distributions. Several simulation results using this H-induced DDD framework are performed and the effect of the hydrogen on plastic deformation is quantified. This framework is shown to provide a powerful tool to better understand the influence of H-atoms on dislocation microstructure evolution, damage accumulation, and subsequent H-induced intergranular fracture.
4:30 PM - *TC04.10.04
Cross-Scale Simulations of Crystal Plasticity
Luis Zepeda Ruiz 1 , Alexander Stukowski 2 , Tomas Oppelstrup 1 , Vasily Bulatov 1
1 , Lawrence Livermore National Laboratory, Livermore, California, United States, 2 , Technische Universität Darmstadt, Darmstadt Germany
Show AbstractCreative ideas and methods for multi-scaling have been and continue to be advanced at a rapid pace aiming to overcome notorious length- and time-scale limits of atomistic simulations. Predictions of crystal plasticity response from the underlying atomic dynamics has been one prominent context where multiscale simulations are both acutely necessary and efficient owing to the existence of reduced variables for coarse-graining – by dislocation lines – suggested by Nature itself. Discrete Dislocation Dynamics (DDD) is a mesoscopic approach that focuses its undivided attention on dislocation lines and attempts to subsume the rest of material dynamics into a few parameters defining dislocation motion. DDD is a scale-bridging method connecting macroscopic crystal plasticity to atomistic dynamics, the latter studied in and imported from small-scale fully atomistic simulations of single dislocations or small groups of dislocations. Given its atomistic input, DDD tracks simultaneous motion and interactions of dislocation ensembles that are sufficiently large to be statistically representative of macroscopic plasticity response. However, as any multi-scale method, DDD adds considerable uncertainties to its resulting predictions, “errors in translation” so to say (symposium TC05 at this Annual Meeting focuses on uncertainties in multiscale materials simulations).
Being long-time practitioners and developers of the mesoscopic DDD method, we have recently demonstrated that direct MD simulations of crystal plasticity (side-stepping DDD altogether) are not only feasible, but deliver wealth of important observations on fundamental mechanisms of dynamic response that define plasticity and strength of tantalum metal [1]. Our MD simulations are cross-scale rather than multi-scale, i.e. simultaneously large enough to be representative of macroscopic crystal plasticity and yet fully resolved tracing every “jiggle” of atomic motion. Unabashedly brute force, in many important aspects our MD simulations surpass existing capabilities of and present themselves as a practical and forward-looking alternative to DDD for quantitative predictions of crystal plasticity and strength. Where feasible, direct MD simulations of crystal plasticity are already more efficient than DDD on the balance of their prediction uncertainties and costs. We expect the massively parallel supercomputers of the near future to further shift this balance in favor of the good old honest Molecular Dynamics.
[1] L. A. Zepeda-Ruiz, A. Stukowski, T. Oppelstrup, V.V. Bulatov, Probing the limits of metal plasticity with molecular dynamics simulations, Nature, in print (2017).