Interatomic Potential Research Articles

Development and evaluation of machine-learned interatomic potentials for carbon nanotubes for molecular dynamics simulations

Development and evaluation of machine-learned interatomic potentials for carbon nanotubes for molecular dynamics simulations

Exploring the energy landscape of aluminas through machine learning interatomic potential

Exploring the energy landscape of aluminas through machine learning interatomic potential

Water Is Cool: Advanced Phonon Dynamics in Ice Ih and Ice XI via Machine Learning Potentials and Quantum Nuclear Vibrations.

Low-dimensional water, despite the relative simplicity of its constituents, exhibits a vast range of phenomena that are of central importance in natural sciences. A large number of bulk as well as nanoscale polymorphs offer engineering possibilities for technological applications such as desalinization, drug delivery, or biological interfacing. However, little is known about the stability of such structures. Therefore, in this study, we employ an array of state-of-the-art computational techniques to study the vibrational properties of ice Ih and XI in their bulk and thin film forms in order to elucidate their structural stability and dynamic behavior. An efficient workflow, consisting of quantum mechanical simulations (based on density functional theory) and machine learning interatomic potentials (MTPs) coupled to temperature-dependent effective potentials (TDEP) and classical molecular dynamics, was verified necessary to capture the temperature-dependent stabilization of the phonons in bulk ice Ih and XI. Anharmonicity and nuclear quantum effects, incorporated in an efficient way through a quantum thermal bath technique, were found crucial to dynamically stabilize low-frequency lattice modes and high-frequency vibrational stretching involving hydrogen. We have identified three novel thin film structures that retain their stability up to at least 250 K and have shed light on their phonon characteristics. In addition, our examination of the Raman spectrum of ice underscores the shortcomings of predicting vibrational properties when relying entirely on the harmonic approximation or purely anharmonic effects. The corrected redistribution of vibrational intensities is found to be achieved only upon inclusion of quantum nuclear vibrations. This was found to be even more crucial for low-dimensional thin film (2D) structures. Overall, our findings demonstrate the significance of joining advanced computational methodologies in unraveling the intricate vibrational dynamics of crystalline ice materials, offering valuable insights into their thermodynamic and structural properties. Furthermore, we suggest a procedure based on MTPs coupled to a quantum thermal bath for the computationally efficient probing of nuclear effects in ice structures, although equally applicable to any other system.

High temperature melting of dense molecular hydrogen from machine-learning interatomic potentials trained on quantum Monte Carlo.

We present results and discuss methods for computing the melting temperature of dense molecular hydrogen using a machine learned model trained on quantum Monte Carlo data. In this newly trained model, we emphasize the importance of accurate total energies in the training. We integrate a two phase method for estimating the melting temperature with estimates from the Clausius-Clapeyron relation to provide a more accurate melting curve from the model. We make detailed predictions of the melting temperature, solid and liquid volumes, latent heat, and internal energy from 50 to 180GPa for both classical hydrogen and quantum hydrogen. At pressures of roughly 173GPa and 1635K, we observe molecular dissociation in the liquid phase. We compare with previous simulations and experimental measurements.

Dislocation plasticity in c$c$‐axis nanopillar compression of wurtzite ceramics: A study using neural network potentials

AbstractCeramics typically exhibit brittle characteristics at room temperature. However, under high compressive stress conditions, such as during nanoindentation or compression tests on nanopillars, these materials can reach the necessary shear stress for dislocation nucleation and glide before fracture occurs. This allows for the observation of plastic deformation through dislocations even at room temperature. Yet, detailed insights into their atomic‐scale dislocation plasticity remain scarce. This study employs atomistic simulations to explore the dislocation plasticity in ‐oriented nanopillars of wurtzite ceramics (i.e., GaN and ZnO) under uniaxial [0001] compression, utilizing a specially developed first‐principles neural network interatomic potential. We observed activation of and dislocations, corroborating experimental findings from in‐situ compression tests. Moreover, a wurtzite‐to‐hexagonal phase transformation begins at facet edges and extends inward, forming wedge‐shaped regions. As compression progresses, dislocations nucleate at the wurtzite‐hexagonal phase boundary and spread throughout the nanopillar, contributing to a rougher surface texture. These quantitative results offer new insights into the plastic deformation behaviors of nanopillars under compressive loading, highlighting the potential of dislocation engineering in these ceramics.

Nanoindentation and deformation behaviors of α$\alpha$‐Al2O3${\rm Al}_2 {\rm O}_3$: A molecular dynamic study

AbstractStudying the nanoindentation behavior of materials at the atomic level is crucial for advancing technologies involving energetic atom bombardment, ion implantation, and nanomechanical testing. Using molecular dynamics simulations with two typical rigid ion interatomic potentials (fixed charges), we showed that the mechanical responses and dislocation nucleation of alumina (‐) are highly temperature‐ and orientation‐dependent. The complex behavior in dislocation dynamics has been rationalized by computing the distribution of stacking fault energy. It is further found that SHIK potential provides better accuracy in describing dislocation dynamics at a much lower computational cost.

Ab-initio trained machine learning potential for MAX compound Ti2AlC: construction, validation, and study of non linear elasticity

Abstract One of the intriguing features exhibited by the layered MAX phase compounds, is the nonlinear elastic behaviour. Since the experimental observation of this curious behaviour, the underlying micro-mechanism has been discussed to interpret experimental observations. However, the theoretical investigation remained a challenge due to the associated length and time scales of the phenomena. In the present work, we adopt a data driven approach to develop a machine learned interatomic potential for the MAX compound Ti2AlC following the moment tensor potential protocol. The constructed potential is validated in lattice constant, formation energy, elastic constant, and stacking fault energies. Finally, applying machine learned potential in classical molecular dynamics provides a faithful representation of the experimentally observed nonlinear elasticity for Ti2AlC. The generated atomic configurations confirm the proposal of formation of ripplocations which allow atomic layers to glide relative to each other without breaking the in-plane bonds. We find common defects, like Al vacancy, strongly influence the hysteresis properties of the stress–strain curve, paving the route to defect-engineered nonlinear elasticity.

Microscopic dynamics of enhanced glass-forming ability with minor oxygen addition in bulk metallic glasses.

Minor oxygen addition has been proposed as a promising strategy to enhance the performance of metallic glasses, particularly their glass-forming ability. In this work, we investigate the microscopic dynamics of a CuZr glass former with oxygen content up to 2 at. % using molecular dynamics simulations based on specially developed neural network interatomic potentials. Our findings indicate a gradual increase in the glass transition temperature with oxygen addition, with an anomalous peak at 0.4 at. % O. We reveal an anti-correlation of kinetic fragility and dynamic heterogeneity behind this unusual rise, where the system exhibits reduced kinetic fragility alongside more significant dynamic heterogeneity. Using the continuous time random walk method, we show that at 0.4 at. % O, a highly mobile Cu atomic layer forms around O-Zr clusters, resulting in notable dynamic heterogeneity. This dynamic behavior is closely linked to the bonding pattern within the O-Zr network, particularly favoring the configuration with edge and surface sharing. In addition, such structures contribute to a more compact O-Zr network, leading to lower kinetic fragility. These findings provide detailed insights into the microscopic dynamics behind the effects of minor oxygen additions.

Accelerating charge estimation in molecular dynamics simulations using physics-informed neural networks: corrosion applications

Molecular Dynamics (MD) simulations are used to understand the effects of corrosion on metallic materials in salt brine. Reactive force fields in classical MD enable accurate modeling of bond formation and breakage in the aqueous medium and at the metal-electrolyte interface, while also facilitating dynamic partial charge equilibration. However, MD simulations are computationally intensive and unsuitable for modeling the long time scales characteristic of corrosive phenomena. To address this, we develop reduced-order machine learning models that provide accurate and efficient predictions of charge density in corrosive environments. Specifically, we use Long Short-Term Memory (LSTM) networks to forecast charge density evolution based on atomic environments represented by Smooth Overlap of Atomic Positions (SOAP) descriptors. A physics-informed loss function enforces charge neutrality and electronegativity equivalence. The atomic charges predicted by the deep learning model trained on this work were obtained two orders of magnitude faster than those from molecular dynamics (MD) simulations, with an error of less than 3% compared to the MD-obtained charges, even in extrapolative scenarios, while adhering to physical constraints. This demonstrates the excellent accuracy, computational efficiency, and validity of the developed model. Lastly, even though developed for corrosion, these protocols are formulated in a phenomenon-agnostic manner, allowing application to various variable-charge interatomic potentials and related fields.

Prediction of thermal conductivity in CALF-20 with first-principles accuracy via machine learning interatomic potentials

Understanding the thermal transport properties of CALF-20, a recent addition to the metal-organic framework family, is crucial for its effective utilization in greenhouse gas capture. Here, we report the thermal transport study of CALF-20 using artificial neural network-based machine learning potentials. We use the Green-Kubo approach based on equilibrium molecular dynamics, with a heat-flux renormalization technique, to determine the thermal conductivity (κ) of CALF-20. We predict that the anisotropic thermal transport in CALF-20, with κ below 1 Wm−1K−1 at 300 K, is ideal for thermoelectric applications. Our analysis reveals a weak temperature dependence (κ ~ 1/T0.56) and near invariance with pressure in κ value of CALF-20, which stands out from the typical trend observed in crystalline materials. The outcome of the study, leveraging advanced computational techniques for predictive modeling, offers valuable insights into more suitable applications of CALF-20 with tailored thermal properties.

Chemical short-range order and its influence on selected properties of non-dilute random alloys

Chemical short-range order and its influence on selected properties of non-dilute random alloys

Developing interatomic potentials for complex concentrated alloys of Cu, Ti, Ni, Cr, Co, Al, Fe, and Mn

Developing interatomic potentials for complex concentrated alloys of Cu, Ti, Ni, Cr, Co, Al, Fe, and Mn

Crystallization of h-BN by molecular dynamics simulation using a machine learning interatomic potential

Crystallization of h-BN by molecular dynamics simulation using a machine learning interatomic potential

Machine learning interatomic potential for friction study in silicon and molybdenum disulfide

Machine learning interatomic potential for friction study in silicon and molybdenum disulfide

Discovery of new topological insulators and semimetals using deep generative models

Topological materials possess unique electronic properties and hold immense attraction to both fundamental physics research and practical applications. Over the past decades, the discovery of new topological materials has relied on the symmetry-based analysis of the quantum wave function. In this study, we propose an efficient inverse design method CTMT (CTMT: CDVAE, Topogivity, interatomic potentials (IAPs) as realized in M3GNet, and TQC) utilizing deep generative machine learning models to discover novel topological insulators and semimetals in a much-fast and low-cost manner. This method covers the entire process of new crystal structure generation, heuristic rule screening, fast stability estimation, and topology type diagnosis, resulting in 4 topological insulators and 16 topological semimetals. Especially, the newly discovered topological materials include several chiral Kramers-Weyl fermion semimetals and chiral materials with low symmetry, whose topology is previously considered challenging to discern. These findings demonstrate the capability of CTMT in discovering topological materials and its great potential for data-driven inverse design of advanced functional materials.

Machine Learning Interatomic Potential for Modeling the Mechanical and Thermal Properties of Naphthyl-Based Nanotubes.

Two-dimensional (2D) nanomaterials are at the forefront of potential technological advancements. Carbon-based materials have been extensively studied since synthesizing graphene, which revealed properties of great interest for novel applications across diverse scientific and technological domains. New carbon allotropes continue to be explored theoretically, with several successful synthesis processes for carbon-based materials recently achieved. In this context, this study investigates the mechanical and thermal properties of DHQ-based monolayers and nanotubes, a carbon allotrope characterized by 4-, 6-, and 10-membered carbon rings, with a potential synthesis route using naphthalene as a molecular precursor. A machine-learned interatomic potential (MLIP) was developed to explore this nanomaterial's mechanical and thermal behavior at larger scales than those accessible through the first-principles calculations. The MLIP was trained on data derived from the DFT/PBE (density functional theory/Perdew-Burke-Ernzerhof) level using ab initio molecular dynamics (AIMD). Classical molecular dynamics (CMD) simulations, employing the trained MLIP, revealed that Young's modulus of DHQ-based nanotubes ranges from 127 to 243 N/m, depending on chirality and diameter, with fracture occurring at strains between 13.6 and 17.4% of the initial length. Regarding thermal response, a critical temperature of 2200 K was identified, marking the onset of a transition to an amorphous phase at higher temperatures.

Improving Bond Dissociations of Reactive Machine Learning Potentials through Physics-Constrained Data Augmentation.

In the field of computational chemistry, predicting bond dissociation energies (BDEs) presents well-known challenges, particularly due to the multireference character of reactive systems. Many chemical reactions involve configurations where single-reference methods fall short, as the electronic structure can significantly change during bond breaking. As generating training data for partially broken bonds is a challenging task, even state-of-the-art reactive machine learning interatomic potentials (MLIPs) often fail to predict reliable BDEs and smooth dissociation curves. By contrast, simple and inexpensive physics-based models, such as the well-established Morse potential, do not suffer from any such limitations. This work leverages the Morse potential to improve reactive MLIPs by augmenting the training data set with inexpensive Morse data along the dissociation pathways. This physics-constrained data augmentation (PCDA) approach results in MLIPs with smooth bond dissociation curves as well as near coupled-cluster level BDEs, all without requiring any expensive multireference quantum mechanical calculations. A case study for methane combustion demonstrates how the PCDA approach can improve an existing reactive MLIP, namely, ANI-1xnr. Not only are the BDEs and bond dissociation curves for all radicals and molecules significantly improved compared to ANI-1xnr but the PCDA-trained MLIP retains the reliability of ANI-1xnr when performing reactive molecular dynamics simulations.

Exploring parameter dependence of atomic minima with implicit differentiation

Interatomic potentials are essential to go beyond ab initio size limitations, but simulation results depend sensitively on potential parameters. Forward propagation of parameter variation is key for uncertainty quantification, whilst backpropagation has found application for emerging inverse problems such as fine-tuning or targeted design. Here, the implicit derivative of functions defined as a fixed point is used to Taylor-expand the energy and structure of atomic minima in potential parameters, evaluating terms via automatic differentiation, dense linear algebra or a sparse operator approach. The latter allows efficient forward and backpropagation through relaxed structures of arbitrarily large systems. The implicit expansion accurately predicts lattice distortion and defect formation energies and volumes with classical and machine-learning potentials, enabling high-dimensional uncertainty propagation without prohibitive overhead. We then show how the implicit derivative can be used to solve challenging inverse problems, minimizing an implicit loss to fine-tune potentials and stabilize solute-induced structural rearrangements at dislocations in tungsten.

Optimal design of experiments in the context of machine-learning inter-atomic potentials: improving the efficiency and transferability of kernel based methods

Abstract Data-driven machine learning (ML) models of atomistic interactions are often based on flexible and non-physical functions that can relate nuanced aspects of atomic arrangements to predictions of energies and forces. As a result, these potentials are only as good as the training data (usually the results of so-called ab initio simulations), and we need to ensure that we have enough information to make a model sufficiently accurate, reliable and transferable. The main challenge stems from the fact that descriptors of chemical environments are often sparse, high-dimensional objects without a well-defined continuous metric. Therefore, it is rather unlikely that any ad hoc method for selecting training examples will be indiscriminate, and it is easy to fall into the trap of confirmation bias, where the same narrow and biased sampling is used to generate training and test sets. We will show that an approach derived from classical concepts of statistical planning of experiments and optimal design can help to mitigate such problems at a relatively low computational cost. The key feature of the method we will investigate is that it allows us to assess the quality of the data without obtaining reference energies and forces—a so-called offline approach. In other words, we are focusing on an approach that is easy to implement and does not require sophisticated frameworks that involve automated access to high performance computing.

Unifying the Description of Hydrocarbons and Hydrogenated Carbon Materials with a Chemically Reactive Machine Learning Interatomic Potential

Unifying the Description of Hydrocarbons and Hydrogenated Carbon Materials with a Chemically Reactive Machine Learning Interatomic Potential