Molecular Autonomous Pathfinder Using Deep Reinforcement Learning.

Diffusion in solids is a slow process that dictates rate-limiting processes in key chemical reactions. Unlike crystalline solids that offer well-defined diffusion pathways, the lack of similar structural motifs in amorphous or glassy materials poses great challenges in bridging the slow diffusion process and material failures. To tackle this problem, we propose an AI-guided long-term atomistic simulation approach: molecular autonomous pathfinder (MAP) framework based on deep reinforcement learning (DRL), where the RL agent is trained to uncover energy efficient diffusion pathways. We employ a Deep Q-Network architecture with distributed prioritized replay buffer, enabling fully online agent training with accelerated experience sampling by an ensemble of asynchronous agents. After training, the agents provide atomistic configurations of diffusion pathways with their energy profile. We use a piecewise nudged elastic band to refine the energy profile of the obtained pathway and the corresponding diffusion time on the basis of transition-state theory. With the MAP framework, we demonstrate atomistic diffusion mechanisms in amorphous silica with time scales comparable to experiments.

Alkali hydroxide (LiOH, NaOH, KOH) in water: Structural and vibrational properties, including neutron scattering results.

Structural and vibrational properties of aqueous solutions of alkali hydroxides (LiOH, NaOH, and KOH) are computed using quantum molecular dynamics simulations for solute concentrations ranging between 1 and 10M. Element-resolved partial radial distribution functions, neutron and x-ray structure factors, and angular distribution functions are computed for the three hydroxide solutions as a function of concentration. The vibrational spectra and frequency-dependent conductivity are computed from the Fourier transforms of velocity autocorrelation and current autocorrelation functions. Our results for the structure are validated with the available neutron data for 17M concentration of NaOH in water [Semrouni et al., Phys. Chem. Chem. Phys. 21, 6828 (2019)]. We found that the larger ionic radius [rLi+

Surface Transfer Doping in MoO3-x/Hydrogenated Diamond Heterostructure.

Surface transfer doping is proposed to be a potential solution for doping diamond, which is hard to dope for applications in high-power electronics. While MoO3 is found to be an effective surface electron acceptor for hydrogen-terminated diamond with a negative electron affinity, the effects of commonly existing oxygen vacancies remain elusive. We have performed reactive molecular dynamics simulations to study the deposition of MoO3-x on a hydrogenated diamond (111) surface and used first-principles calculations based on density functional theory to investigate the electronic structures and charge transfer mechanisms. We find that MoO3-x is an effective surface electron acceptor and the spatial extent of doped holes in hydrogenated diamond is extended, promoting excellent transport properties. Charge transfer is found to monotonically decrease with the level of oxygen vacancy, providing guidance for engineering of the surface transfer doping process.

Inelastic Neutron Scattering Study of Phonon Density of States of Iodine Oxides and First-Principles Calculations.

Iodine oxides I2Oy (y = 4, 5, 6) crystallize into atypical structures that fall between molecular- and framework-base types and exhibit high reactivity in an ambient environment, a property highly desired in the so-called "agent defeat materials". Inelastic neutron scattering experiments were performed to determine the phonon density of states of the newly synthesized I2O5 and I2O6 samples. First-principles calculations were carried out for I2O4, I2O5, and I2O6 to predict their thermodynamic properties and phonon density of states. Comparison of the INS data with the Raman and infrared measurements as well as the first-principles calculations sheds light on their distinctive, anisotropic thermomechanical properties.

Induction and Ferroelectric Switching of Flux Closure Domains in Strained PbTiO3 with Neural Network Quantum Molecular Dynamics.

We have developed an extension of the Neural Network Quantum Molecular Dynamics (NNQMD) simulation method to incorporate electric-field dynamics based on Born effective charge (BEC), called NNQMD-BEC. We first validate NNQMD-BEC for the switching mechanisms of archetypal ferroelectric PbTiO3 bulk crystal and 180° domain walls (DWs). NNQMD-BEC simulations correctly describe the nucleation-and-growth mechanism during DW switching. In triaxially strained PbTiO3 with strain conditions commonly seen in many superlattice configurations, we find that flux-closure texture can be induced with application of an electric field perpendicular to the original polarization direction. Upon field reversal, the flux-closure texture switches via a pair of transient vortices as the intermediate state, indicating an energy-efficient switching pathway. Our NNQMD-BEC method provides a theoretical guidance to study electro-mechano effects with existing machine learning force fields using a simple BEC extension, which will be relevant for engineering applications such as field-controlled switching in mechanically strained ferroelectric devices.

Pressure-Controlled Layer-by-Layer to Continuous Oxidation of ZrS2(001) Surface.

Understanding oxidation mechanisms of layered semiconducting transition-metal dichalcogenides (TMDC) is important not only for controlling native oxide formation but also for synthesis of oxide and oxysulfide products. Here, reactive molecular dynamics simulations show that oxygen partial pressure controls not only the ZrS2 oxidation rate but also the oxide morphology and quality. We find a transition from layer-by-layer oxidation to amorphous-oxide-mediated continuous oxidation as the oxidation progresses, where different pressures selectively expose different oxidation stages within a given time window. While the kinetics of the fast continuous oxidation stage is well described by the conventional Deal-Grove model, the layer-by-layer oxidation stage is dictated by reactive bond-switching mechanisms. This work provides atomistic details and a potential foundation for rational pressure-controlled oxidation of TMDC materials.

Wrinkles, Ridges, Miura-Ori, and Moiré Patterns in MoSe2 Using Neural Networks.

Effects of lateral compression on out-of-plane deformation of two-dimensional MoSe2 layers are investigated. A MoSe2 monolayer develops periodic wrinkles under uniaxial compression and Miura-Ori patterns under biaxial compression. When a flat MoSe2 monolayer is placed on top of a wrinkled MoSe2 layer, the van der Waals (vdW) interaction transforms wrinkles into ridges and generates mixed 2H and 1T phases and chain-like defects. Under a biaxial strain, the vdW interaction induces regions of Miura-Ori patterns in bilayers. Strained systems analyzed using a convolutional neural network show that the compressed system consists of semiconducting 2H and metallic 1T phases. The energetics, mechanical response, defect structure, and dynamics are analyzed as bilayers undergo wrinkle-ridge transformations under uniaxial compression and moiré transformations under biaxial compression. Our results indicate that in-plane compression can induce self-assembly of out-of-plane metasurfaces with controllable semiconducting and metallic phases and moiré patterns with unique optoelectronic properties.

Squishing Skyrmions: Symmetry-Guided Dynamic Transformation of Polar Topologies Under Compression.

Mechanical controllability of recently discovered topological defects (e.g., skyrmions) in ferroelectric materials is of interest for the development of ultralow-power mechano-electronics that are protected against thermal noise. However, fundamental understanding is hindered by the "multiscale quantum challenge" to describe topological switching encompassing large spatiotemporal scales with quantum mechanical accuracy. Here, we overcome this challenge by developing a machine-learning-based multiscale simulation framework─a hybrid neural network quantum molecular dynamics (NNQMD) and molecular mechanics (MM) method. For nanostructures composed of SrTiO3 and PbTiO3, we find how the symmetry of mechanical loading essentially controls polar topological switching. We find under symmetry-breaking uniaxial compression a squishing-to-annihilation pathway versus formation of a topological composite named skyrmionium under symmetry-preserving isotropic compression. The distinct pathways are explained in terms of the underlying materials' elasticity and symmetry, as well as the Landau-Lifshitz-Kittel scaling law. Such rational control of ferroelectric topologies will likely facilitate exploration of the rich ferroelectric "topotronics" design space.

Defect-free and crystallinity-preserving ductile deformation in semiconducting Ag2S.

Typical ductile materials are metals, which deform by the motion of defects like dislocations in association with non-directional metallic bonds. Unfortunately, this textbook mechanism does not operate in most inorganic semiconductors at ambient temperature, thus severely limiting the development of much-needed flexible electronic devices. We found a shear-deformation mechanism in a recently discovered ductile semiconductor, monoclinic-silver sulfide (Ag2S), which is defect-free, omni-directional, and preserving perfect crystallinity. Our first-principles molecular dynamics simulations elucidate the ductile deformation mechanism in monoclinic-Ag2S under six types of shear systems. Planer mass movement of sulfur atoms plays an important role for the remarkable structural recovery of sulfur-sublattice. This in turn arises from a distinctively high symmetry of the anion-sublattice in Ag2S, which is not seen in other brittle silver chalcogenides. Such mechanistic and lattice-symmetric understanding provides a guideline for designing even higher-performance ductile inorganic semiconductors.

Photoexcitation-Induced Nonthermal Ultrafast Loss of Long-Range Order in GeTe.

Nonadiabatic quantum molecular dynamics is used to investigate the evolution of GeTe photoexcited states. Results reveal a photoexcitation-induced picosecond nonthermal path for the loss of long-range order. A valence electron excitation threshold of 4% is found to trigger local disorder by switching Ge atoms from octahedral to tetrahedral sites and promoting Ge-Ge bonding. The resulting loss of long-range order for a higher valence electron excitation fraction is achieved without fulfilling the Lindemann criterion for melting, therefore utilizing a nonthermal path. The photoexcitation-induced structural disorder is accompanied by charge transfer from Te to Ge, Ge-Te bonding-to-antibonding, and Ge-Ge antibonding-to-bonding change, triggering Ge-Te bond breaking and promoting the formation of Ge-Ge wrong bonds. These results provide an electronic-structure basis to understand the photoexcitation-induced ultrafast changes in the structure and properties of GeTe and other phase-change materials.

Anisotropic atomistic shock response mechanisms of aramid crystals.

Aramid fibers composed of poly(p-phenylene terephthalamide) (PPTA) polymers are attractive materials due to their high strength, low weight, and high shock resilience. Even though they have widely been utilized as a basic ingredient in Kevlar, Twaron, and other fabrics and applications, their intrinsic behavior under intense shock loading is still to be understood. In this work, we characterize the anisotropic shock response of PPTA crystals by performing reactive molecular dynamics simulations. Results from shock loading along the two perpendicular directions to the polymer backbones, [100] and [010], indicate distinct shock release mechanisms that preserve and destroy the hydrogen bond network. Shocks along the [100] direction for particle velocity Up < 2.46 km/s indicate the formation of a plastic regime composed of shear bands, where the PPTA structure is planarized. Shocks along the [010] direction for particle velocity Up < 2.18 km/s indicate a complex response regime, where elastic compression shifts to amorphization as the shock is intensified. While hydrogen bonds are mostly preserved for shocks along the [100] direction, hydrogen bonds are continuously destroyed with the amorphization of the crystal for shocks along the [010] direction. Decomposition of the polymer chains by cross-linking is triggered at the threshold particle velocity Up = 2.18 km/s for the [010] direction and Up = 2.46 km/s for the [100] direction. These atomistic insights based on large-scale simulations highlight the intricate and anisotropic mechanisms underpinning the shock response of PPTA polymers and are expected to support the enhancement of their applications.

Hydrogen Bonding in Liquid Ammonia.

The nature of hydrogen bonding in condensed ammonia phases, liquid and crystalline ammonia has been a topic of much investigation. Here, we use quantum molecular dynamics simulations to investigate hydrogen bond structure and lifetimes in two ammonia phases: liquid ammonia and crystalline ammonia-I. Unlike liquid water, which has two covalently bonded hydrogen and two hydrogen bonds per oxygen atom, each nitrogen atom in liquid ammonia is found to have only one hydrogen bond at 2.24 Å. The computed lifetime of the hydrogen bond is t ≅ 0.1 ps. In contrast to crystalline water-ice, we find that hydrogen bonding is practically nonexistent in crystalline ammonia-I.

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.

AI-driven quantification of ground glass opacities in lungs of COVID-19 patients using 3D computed tomography imaging.

OBJECTIVES: Ground-glass opacity (GGO)-a hazy, gray appearing density on computed tomography (CT) of lungs-is one of the hallmark features of SARS-CoV-2 in COVID-19 patients. This AI-driven study is focused on segmentation, morphology, and distribution patterns of GGOs. METHOD: We use an AI-driven unsupervised machine learning approach called PointNet++ to detect and quantify GGOs in CT scans of COVID-19 patients and to assess the severity of the disease. We have conducted our study on the "MosMedData", which contains CT lung scans of 1110 patients with or without COVID-19 infections. We quantify the morphologies of GGOs using Minkowski tensors and compute the abnormality score of individual regions of segmented lung and GGOs. RESULTS: PointNet++ detects GGOs with the highest evaluation accuracy (98%), average class accuracy (95%), and intersection over union (92%) using only a fraction of 3D data. On average, the shapes of GGOs in the COVID-19 datasets deviate from sphericity by 15% and anisotropies in GGOs are dominated by dipole and hexapole components. These anisotropies may help to quantitatively delineate GGOs of COVID-19 from other lung diseases. CONCLUSION: The PointNet++ and the Minkowski tensor based morphological approach together with abnormality analysis will provide radiologists and clinicians with a valuable set of tools when interpreting CT lung scans of COVID-19 patients. Implementation would be particularly useful in countries severely devastated by COVID-19 such as India, where the number of cases has outstripped available resources creating delays or even breakdowns in patient care. This AI-driven approach synthesizes both the unique GGO distribution pattern and severity of the disease to allow for more efficient diagnosis, triaging and conservation of limited resources.

Deep Well Trapping of Hot Carriers in a Hexagonal Boron Nitride Coating of Polymer Dielectrics.

Polymer dielectrics can be cost-effective alternatives to conventional inorganic dielectric materials, but their practical application is critically hindered by their breakdown under high electric fields driven by excited hot charge carriers. Using a joint experiment-simulation approach, we show that a 2D nanocoating of hexagonal boron nitride (hBN) mitigates the damage done by hot carriers, thereby increasing the breakdown strength. Surface potential decay and dielectric breakdown measurements of hBN-coated Kapton show the carrier-trapping effect in the hBN nanocoating, which leads to an increased breakdown strength. Nonadiabatic quantum molecular dynamics simulations demonstrate that hBN layers at the polymer-electrode interfaces can trap hot carriers, elucidating the observed increase in the breakdown field. The trapping of hot carriers is due to a deep potential well formed in the hBN layers at the polymer-electrode interface. Searching for materials with similar deep well potential profiles could lead to a computationally efficient way to design good polymer coatings that can mitigate breakdown.

Electric-field-induced crossover of polarization reversal mechanisms in Al1-xScxN ferroelectrics.

Scandium-doped aluminum nitride, Al1-xScxN, represents a new class of displacive ferroelectric materials with high polarization and sharp hysteresis along with high-temperature resilience, facile synthesizability and compatibility with standard CMOS fabrication techniques. The fundamental physics behind the transformation of unswitchable piezoelectric AlN into switchable Al-Sc-N ferroelectrics depends upon important atomic properties such as local structure, dopant distributions and the presence of competing mechanism of polarization switching in the presence of an applied electric-field that have not been understood. We computationally synthesize Al1-xScxN to quantify the inhomogeneity of Sc distribution and phase segregation, and characterize its crystal and electronic structure as a function of Sc-doping. Nudged elastic band calculations of the potential energy surface and quantum molecular dynamics simulations of direct electric-field-driven ferroelectric switching reveal a crossover between two polarization reversal mechanisms-inhomogeneous nucleation-and-growth mechanism originating near Sc-rich regions in the limit of low applied fields and nucleation-limited-switching in the high-field regime. Understanding polarization reversal pathways for these two mechanisms as well as the role of local Sc concentration on activation barriers provides design rules to identify other combinations of dopant elements, such as Zr, Mg etc. to synthesize superior AlN-based ferroelectric materials.

AI-DRIVEN QUANTIFICATION OF GROUND GLASS OPACITIES IN LUNGS OF COVID-19 PATIENTS USING 3D COMPUTED TOMOGRAPHY IMAGING.

OBJECTIVES: Ground-glass opacity (GGO) - a hazy, gray appearing density on computed tomography (CT) of lungs - is one of the hallmark features of SARS-CoV-2 in COVID-19 patients. This AI-driven study is focused on segmentation, morphology, and distribution patterns of GGOs. METHOD: We use an AI-driven unsupervised machine learning approach called PointNet++ to detect and quantify GGOs in CT scans of COVID-19 patients and to assess the severity of the disease. We have conducted our study on the "MosMedData", which contains CT lung scans of 1110 patients with or without COVID-19 infections. We quantify the morphologies of GGOs using Minkowski tensors and compute the abnormality score of individual regions of segmented lung and GGOs. RESULTS: PointNet++ detects GGOs with the highest evaluation accuracy (98%), average class accuracy (95%), and intersection over union (92%) using only a fraction of 3D data. On average, the shapes of GGOs in the COVID-19 datasets deviate from sphericity by 15% and anisotropies in GGOs are dominated by dipole and hexapole components. These anisotropies may help to quantitatively delineate GGOs of COVID-19 from other lung diseases. CONCLUSION: The PointNet++ and the Minkowski tensor based morphological approach together with abnormality analysis will provide radiologists and clinicians with a valuable set of tools when interpreting CT lung scans of COVID-19 patients. Implementation would be particularly useful in countries severely devastated by COVID-19 such as India, where the number of cases has outstripped available resources creating delays or even breakdowns in patient care. This AI-driven approach synthesizes both the unique GGO distribution pattern and severity of the disease to allow for more efficient diagnosis, triaging and conservation of limited resources. KEY POINTS: Our approach to GGO analysis has four distinguishing features:We combine an unsupervised computer vision approach with convex hull and convex points algorithms to segment and preserve the actual structure of the lung.To the best of our knowledge, we are the first group to use PointNet++ architecture for 3D visualization, segmentation, classification, and pattern analysis of GGOs.We make abnormality predictions using a deep network and Cox proportional hazards model using lung CT images of COVID-19 patients.We quantify the shapes and sizes of GGOs using Minkowski tensors to understand the morphological variations of GGOs within the COVID-19 cohort.

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.

Sulfurization of MoO3 in the Chemical Vapor Deposition Synthesis of MoS2 Enhanced by an H2S/H2 Mixture.

The typical layered transition metal dichalcogenide (TMDC) material, MoS2, is considered a promising candidate for the next-generation electronic device due to its exceptional physical and chemical properties. In chemical vapor deposition synthesis, the sulfurization of MoO3 powders is an essential reaction step in which the MoO3 reactants are converted into MoS2 products. Recent studies have suggested using an H2S/H2 mixture to reduce MoO3 powders in an effective way. However, reaction mechanisms associated with the sulfurization of MoO3 by the H2S/H2 mixture are yet to be fully understood. Here, we perform quantum molecular dynamics (QMD) simulations to investigate the sulfurization of MoO3 flakes using two different gaseous environments: pure H2S precursors and a H2S/H2 mixture. Our QMD results reveal that the H2S/H2 mixture could effectively reduce and sulfurize the MoO3 reactants through additional reactions of H2 and MoO3, thereby providing valuable input for experimental synthesis of higher-quality TMDC materials.

Lattice thermal transport in two-dimensional alloys and fractal heterostructures.

Engineering thermal transport in two dimensional materials, alloys and heterostructures is critical for the design of next-generation flexible optoelectronic and energy harvesting devices. Direct experimental characterization of lattice thermal conductivity in these ultra-thin systems is challenging and the impact of dopant atoms and hetero-phase interfaces, introduced unintentionally during synthesis or as part of deliberate material design, on thermal transport properties is not understood. Here, we use non-equilibrium molecular dynamics simulations to calculate lattice thermal conductivity of [Formula: see text] monolayer crystals including [Formula: see text] alloys with substitutional point defects, periodic [Formula: see text] heterostructures with characteristic length scales and scale-free fractal [Formula: see text] heterostructures. Each of these features has a distinct effect on phonon propagation in the crystal, which can be used to design fractal and periodic alloy structures with highly tunable thermal conductivities. This control over lattice thermal conductivity will enable applications ranging from thermal barriers to thermoelectrics.