Dielectric Constant of Liquid Water Determined with Neural Network Quantum Molecular Dynamics.

The static dielectric constant ϵ_{0} and its temperature dependence for liquid water is investigated using neural network quantum molecular dynamics (NNQMD). We compute the exact dielectric constant in canonical ensemble from NNQMD trajectories using fluctuations in macroscopic polarization computed from maximally localized Wannier functions (MLWF). Two deep neural networks are constructed. The first, NNQMD, is trained on QMD configurations for liquid water under a variety of temperature and density conditions to learn potential energy surface and forces and then perform molecular dynamics simulations. The second network, NNMLWF, is trained to predict locations of MLWF of individual molecules using the atomic configurations from NNQMD. Training data for both the neural networks is produced using a highly accurate quantum-mechanical method, DFT-SCAN that yields an excellent description of liquid water. We produce 280×10^{6} configurations of water at 7 temperatures using NNQMD and predict MLWF centers using NNMLWF to compute the polarization fluctuations. The length of trajectories needed for a converged value of the dielectric constant at 0°C is found to be 20 ns (40×10^{6} configurations with 0.5 fs time step). The computed dielectric constants for 0, 15, 30, 45, 60, 75, and 90°C are in good agreement with experiments. Our scalable scheme to compute dielectric constants with quantum accuracy is also applicable to other polar molecular liquids.

Dielectric Polymer Property Prediction Using Recurrent Neural Networks with Optimizations.

Despite the growing success of machine learning for predicting structure-property relationships in molecules and materials, such as predicting the dielectric properties of polymers, it is still in its infancy. We report on the effectiveness of solving structure-property relationships for a computer-generated database of dielectric polymers using recurrent neural network (RNN) models. The implementation of a series of optimization strategies was crucial to achieving high learning speeds and sufficient accuracy: (1) binary and nonbinary representations of SMILES (Simplified Molecular Input Line System) fingerprints and (2) backpropagation with affine transformation of the input sequence (ATransformedBP) and resilient backpropagation with initial weight update parameter optimizations (iRPROP- optimized). For the investigated database of polymers, the binary SMILES representation was found to be superior to the decimal representation with respect to the training and prediction performance. All developed and optimized Elman-type RNN algorithms outperformed nonoptimized RNN models in the efficient prediction of nonlinear structure-activity relationships. The average relative standard deviation (RSD) remained well below 5%, and the maximum RSD did not exceed 30%. Moreover, we provide a C++ codebase as a testbed for a new generation of open programming languages that target increasingly diverse computer architectures.

Computational Synthesis of MoS2 Layers by Reactive Molecular Dynamics Simulations: Initial Sulfidation of MoO3 Surfaces.

Transition metal dichalcogenides (TMDC) like MoS2 are promising candidates for next-generation electric and optoelectronic devices. These TMDC monolayers are typically synthesized by chemical vapor deposition (CVD). However, despite significant amount of empirical work on this CVD growth of monolayered crystals, neither experiment nor theory has been able to decipher mechanisms of selection rules for different growth scenarios, or make predictions of optimized environmental parameters and growth factors. Here, we present an atomic-scale mechanistic analysis of the initial sulfidation process on MoO3 surfaces using first-principles-informed ReaxFF reactive molecular dynamics (RMD) simulations. We identify a three-step reaction process associated with synthesis of the MoS2 samples from MoO3 and S2 precursors: O2 evolution and self-reduction of the MoO3 surface; SO/SO2 formation and S2-assisted reduction; and sulfidation of the reduced surface and Mo-S bond formation. These atomic processes occurring during early stage MoS2 synthesis, which are consistent with experimental observations and existing theoretical literature, provide valuable input for guided rational synthesis of MoS2 and other TMDC crystals by the CVD process.

A divide-conquer-recombine algorithmic paradigm for large spatiotemporal quantum molecular dynamics simulations.

We introduce an extension of the divide-and-conquer (DC) algorithmic paradigm called divide-conquer-recombine (DCR) to perform large quantum molecular dynamics (QMD) simulations on massively parallel supercomputers, in which interatomic forces are computed quantum mechanically in the framework of density functional theory (DFT). In DCR, the DC phase constructs globally informed, overlapping local-domain solutions, which in the recombine phase are synthesized into a global solution encompassing large spatiotemporal scales. For the DC phase, we design a lean divide-and-conquer (LDC) DFT algorithm, which significantly reduces the prefactor of the O(N) computational cost for N electrons by applying a density-adaptive boundary condition at the peripheries of the DC domains. Our globally scalable and locally efficient solver is based on a hybrid real-reciprocal space approach that combines: (1) a highly scalable real-space multigrid to represent the global charge density; and (2) a numerically efficient plane-wave basis for local electronic wave functions and charge density within each domain. Hybrid space-band decomposition is used to implement the LDC-DFT algorithm on parallel computers. A benchmark test on an IBM Blue Gene/Q computer exhibits an isogranular parallel efficiency of 0.984 on 786 432 cores for a 50.3 × 10(6)-atom SiC system. As a test of production runs, LDC-DFT-based QMD simulation involving 16 661 atoms is performed on the Blue Gene/Q to study on-demand production of hydrogen gas from water using LiAl alloy particles. As an example of the recombine phase, LDC-DFT electronic structures are used as a basis set to describe global photoexcitation dynamics with nonadiabatic QMD (NAQMD) and kinetic Monte Carlo (KMC) methods. The NAQMD simulations are based on the linear response time-dependent density functional theory to describe electronic excited states and a surface-hopping approach to describe transitions between the excited states. A series of techniques are employed for efficiently calculating the long-range exact exchange correction and excited-state forces. The NAQMD trajectories are analyzed to extract the rates of various excitonic processes, which are then used in KMC simulation to study the dynamics of the global exciton flow network. This has allowed the study of large-scale photoexcitation dynamics in 6400-atom amorphous molecular solid, reaching the experimental time scales.

Exploring far-from-equilibrium ultrafast polarization control in ferroelectric oxides with excited-state neural network quantum molecular dynamics.

Ferroelectric materials exhibit a rich range of complex polar topologies, but their study under far-from-equilibrium optical excitation has been largely unexplored because of the difficulty in modeling the multiple spatiotemporal scales involved quantum-mechanically. To study optical excitation at spatiotemporal scales where these topologies emerge, we have performed multiscale excited-state neural network quantum molecular dynamics simulations that integrate quantum-mechanical description of electronic excitation and billion-atom machine learning molecular dynamics to describe ultrafast polarization control in an archetypal ferroelectric oxide, lead titanate. Far-from-equilibrium quantum simulations reveal a marked photo-induced change in the electronic energy landscape and resulting cross-over from ferroelectric to octahedral tilting topological dynamics within picoseconds. The coupling and frustration of these dynamics, in turn, create topological defects in the form of polar strings. The demonstrated nexus of multiscale quantum simulation and machine learning will boost not only the emerging field of ferroelectric topotronics but also broader optoelectronic applications.

Neural Network Quantum Molecular Dynamics, Intermediate Range Order in GeSe2, and Neutron Scattering Experiments.

A remarkable property of certain covalent glasses and their melts is intermediate range order, manifested as the first sharp diffraction peak (FSDP) in neutron-scattering experiments, as was exhaustively investigated by Price, Saboungi, and collaborators. Atomistic simulations thus far have relied on either quantum molecular dynamics (QMD), with systems too small to resolve FSDP, or classical molecular dynamics, without quantum-mechanical accuracy. We investigate prototypical FSDP in GeSe2 glass and melt using neural-network quantum molecular dynamics (NNQMD) based on machine learning, which allows large simulation sizes with validated quantum mechanical accuracy to make quantitative comparisons with neutron data. The system-size dependence of the FSDP height is determined by comparing QMD and NNQMD simulations with experimental data. Partial pair distribution functions, bond-angle distributions, partial and neutron structure factors, and ring-size distributions are presented. Calculated FSDP heights agree quantitatively with neutron scattering data for GeSe2 glass at 10 K and melt at 1100 K.

Defect Healing in Layered Materials: A Machine Learning-Assisted Characterization of MoS2 Crystal Phases.

Monolayer MoS2 is an outstanding candidate for a next-generation semiconducting material because of its exceptional physical, chemical, and mechanical properties. To make this promising layered material applicable to nanostructured electronic applications, synthesis of a highly crystalline MoS2 monolayer is vitally important. Among different types of synthesis methods, chemical vapor deposition (CVD) is the most practical way to synthesize few- or mono-layer MoS2 on the target substrate owing to its simplicity and scalability. However, synthesis of a highly crystalline MoS2 layer remains elusive. This is because of the number of grains and defects unavoidably generated during CVD synthesis. Here, we perform multimillion-atom reactive molecular dynamics (RMD) simulations to identify an origin of the grain growth, migration, and defect healing process on a CVD-grown MoS2 monolayer. RMD results reveal that grain boundaries could be successfully repaired by multiple heat treatments. Our work proposes a new way of controlling the grain growth and migration on a CVD-grown MoS2 monolayer.

Faceting, Grain Growth, and Crack Healing in Alumina.

Reactive molecular dynamics simulations are performed to study self-healing of cracks in Al2O3 containing core/shell SiC/SiO2 nanoparticles. These simulations are carried out in a precracked Al2O3 under mode 1 strain at 1426 °C. The nanoparticles are embedded ahead of the precrack in the Al2O3 matrix. When the crack begins to propagate at a strain of 2%, the nanoparticles closest to the advancing crack distort to create nanochannels through which silica flows toward the crack and stops its growth. At this strain, the Al2O3 matrix at the interface of SiC/SiO2 nanoparticles forms facets along the prismatic (A) ⟨2̅110⟩ and prismatic (M) ⟨1̅010⟩ planes. These facets act as nucleation sites for the growth of multiple secondary amorphous grains in the Al2O3 matrix. These grains grow with an increase in the applied strain. Voids and nanocracks form in the grain boundaries but are again healed by diffusion of silica from the nanoparticles.

Molecular Simulation of MoS2 Exfoliation.

A wide variety of two-dimensional layered materials are synthesized by liquid-phase exfoliation. Here we examine exfoliation of MoS2 into nanosheets in a mixture of water and isopropanol (IPA) containing cavitation bubbles. Using force fields optimized with experimental data on interfacial energies between MoS2 and the solvent, multimillion-atom molecular dynamics simulations are performed in conjunction with experiments to examine shock-induced collapse of cavitation bubbles and the resulting exfoliation of MoS2. The collapse of cavitation bubbles generates high-speed nanojets and shock waves in the solvent. Large shear stresses due to the nanojet impact on MoS2 surfaces initiate exfoliation, and shock waves reflected from MoS2 surfaces enhance exfoliation. Structural correlations in the solvent indicate that shock induces an ice VII like motif in the first solvation shell of water.

Structural Phase Transformation in Strained Monolayer MoWSe2 Alloy.

Two-dimensional (2D) materials exhibit different mechanical properties from their bulk counterparts owing to their monolayer atomic thickness. Here, we have examined the mechanical behavior of 2D molybdenum tungsten diselenide (MoWSe2) precipitation alloy grown using chemical vapor deposition and composed of numerous nanoscopic MoSe2 and WSe2 regions. Applying a bending strain blue-shifted the MoSe2 and WSe2 A1g Raman modes with the stress concentrated near the precipitate interfaces predominantly affecting the WSe2 modes. In situ local Raman measurements suggested that the crack propagated primarily thorough MoSe2-rich regions in the monolayer alloy. Molecular dynamics (MD) simulations were performed to study crack propagation in an MoSe2 monolayer containing nanoscopic WSe2 regions akin to the experiment. Raman spectra calculated from MD trajectories of crack propagation confirmed the emergence of intermediate peaks in the strained monolayer alloy, mirroring experimental results. The simulations revealed that the stress buildup around the crack tip caused an irreversible structural transformation from the 2H to 1T phase both in the MoSe2 matrix and WSe2 patches. This was corroborated by high-angle annular dark-field images. Crack branching and subsequent healing of a crack branch were also observed in WSe2, indicating the increased toughness and crack propagation resistance of the alloyed 2D MoWSe2 over the unalloyed counterparts.

Nanocarbon synthesis by high-temperature oxidation of nanoparticles.

High-temperature oxidation of silicon-carbide nanoparticles (nSiC) underlies a wide range of technologies from high-power electronic switches for efficient electrical grid and thermal protection of space vehicles to self-healing ceramic nanocomposites. Here, multimillion-atom reactive molecular dynamics simulations validated by ab initio quantum molecular dynamics simulations predict unexpected condensation of large graphene flakes during high-temperature oxidation of nSiC. Initial oxidation produces a molten silica shell that acts as an autocatalytic 'nanoreactor' by actively transporting oxygen reactants while protecting the nanocarbon product from harsh oxidizing environment. Percolation transition produces porous nanocarbon with fractal geometry, which consists of mostly sp(2) carbons with pentagonal and heptagonal defects. This work suggests a simple synthetic pathway to high surface-area, low-density nanocarbon with numerous energy, biomedical and mechanical-metamaterial applications, including the reinforcement of self-healing composites.