Explore the vast range of titles available.

Although the application of force to induce chemical transformations is an active area of research, detailed understanding of these mechanochemical pathways is still lacking. Now, the mechanochemical activation of [4]-ladderane has been studied and found to exhibit unique non-equilibrium dynamic effects.

An accurate description of information is relevant for a range of problems in atomistic machine learning (ML), such as crafting training sets, performing uncertainty quantification (UQ), or extracting physical insights from large datasets. However, atomistic ML often relies on unsupervised learning or model predictions to analyze information contents from simulation or training data. Here, we introduce a theoretical framework that provides a rigorous, model-free tool to quantify information contents in atomistic simulations. We demonstrate that the information entropy of a distribution of atom-centered environments explains known heuristics in ML potential developments, from training set sizes to dataset optimality. Using this tool, we propose a model-free UQ method that reliably predicts epistemic uncertainty and detects out-of-distribution samples, including rare events in systems such as nucleation. This method provides a general tool for data-driven atomistic modeling and combines efforts in ML, simulations, and physical explainability. Dataset analysis or uncertainty quantification (UQ) in atomistic machine learning (ML) often rely on model-based heuristics. Here, the authors present a model-free, information theoretical approach to estimate errors, UQ, and outliers for ML-driven simulations.

Interface diffusion along a metal/ceramic interface present in numerous energy and electronic devices can critically affect their performance and stability. Hole formation in a polycrystalline Ni film on an α -Al 2 O 3 substrate coupled with a continuum diffusion analysis demonstrates that Ni diffusion along the Ni/ α -Al 2 O 3 interface is surprisingly fast. Ab initio calculations demonstrate that both Ni vacancy formation and migration energies at the coherent Ni/ α -Al 2 O 3 interface are much smaller than in bulk Ni, suggesting that the activation energy for diffusion along coherent Ni/ α -Al 2 O 3 interfaces is comparable to that along (incoherent/high angle) grain boundaries. Based on these results, we develop a simple model for diffusion along metal/ceramic interfaces, apply it to a wide range of metal/ceramic systems and validate it with several ab initio calculations. These results suggest that fast metal diffusion along metal/ceramic interfaces should be common, but is not universal. Little is known about diffusion along metal/ceramic interfaces even though it controls the physical behavior and lifetimes of many devices (including batteries, microelectronics, and jet engines). Here, the authors show that diffusion along a nickel/sapphire interface is abnormally fast due to nickel vacancies and generalise their findings to a wide-range of metal/ceramic systems.

Machine-learned interatomic potentials (ML-IAPs) continue to gain popularity as accurate, computationally efficient replacements for traditional, physics-based interatomic potentials and expensive ab initio methods. Uncertainty quantification (UQ) of ML-IAPs is a growing area of research as UQ is critical in many applications of IAPs, such as developing curated datasets, active learning-based data augmentation, self-improving models, and estimating the uncertainty of molecular dynamics simulations. In this paper, we construct and benchmark a series of different neural network potentials (NNPs) with varying network architectures to determine the performance of these models with respect to both the mean and uncertainty calibration error. Each NNP method is specifically designed to predict either epistemic or aleatoric uncertainty with particular focus on the differences in behavior between the epistemic and aleatoric uncertainty estimates. We benchmark these methods using multiple datasets common in the ML-IAP literature. The results show that the aleatoric uncertainty from single-shot model architectures is a competitive alternative to ensemble-based epistemic uncertainty predictions in regions of sufficient data-density. However, in regions where the representative data is sparse, aleatoric uncertainty models tend to overpredict and epistemic methods tend to underpredict the actual model error. We conclude that the type of UQ is crucial when discussing performance of probabilistic model results as different methods have different performance characteristics depending on the regime in which they are evaluated. Therefore, the type of UQ method should be carefully evaluated against both the data characteristics and requirements for the intended application. Scientific contribution In this study, we present a systematic approach for benchmarking uncertainty quantifi cation (UQ) strategies for neural network interatomic potentials evaluated by both accuracy and uncertainty calibration metrics across multiple datasets. We present a robust hyperparameter tuning framework that allows for equitable comparison of different UQ methods and demonstrate that model performance is highly sensitive to hyperparameter settings. Finally, we conclude that when properly trained, single-shot aleatoric approaches can provide performance competitive with standard ensemble-based methods in data-rich regimes but tend to overestimate error in data-sparse regions, thereby providing guidance for UQ method selection.

We present new parameterizations of the ChIMES physics informed machine-learned interatomic model for simulating carbon under conditions ranging from 300 K and 0 GPa to 10,000 K and 100 GPa, along with a new multi-fidelity active learning strategy. The resulting models show significant improvement in accuracy and temperature/pressure transferability relative to the original ChIMES carbon model developed in 2017 and can serve as a foundation for future transfer-learned ChIMES parameter sets. Applications to carbon melting point prediction, shockwave-driven conversion of graphite to diamond, and thermal conversion of nanodiamond to graphitic nanoonion are provided. Ultimately, we find the new models to be robust, accurate, and well-suited for modeling evolution in carbon systems under extreme conditions.

In order to realize the full potential of ion trap quantum computers, an improved understanding is required of the motional heating that trapped ions experience. Experimental studies of the temperature-, frequency-, and ion-electrode distance-dependence of the electric field noise responsible for motional heating, as well as the noise before and after ion bombardment cleaning of trap electrodes, suggest that fluctuations of adsorbate dipoles are a likely source of so-called 'anomalous heating,' or motional heating of the trapped ions at a rate much higher than the Johnson noise limit. Previous computational studies have investigated how the fluctuation of model adsorbate dipoles affects anomalous heating. However, the way in which specific adsorbates affect the electric field noise has not yet been examined, and an electric dipole model employed in previous studies is only accurate for a small subset of possible adsorbates. Here, we analyze the behavior of both in-plane and out-of-plane vibrational modes of twenty-one adsorbate-electrode combinations within the independent fluctuating dipole model, utilizing accurate first principles computational methods to determine the surface-induced dipole moments. We find the chemical specificity of the adsorbate can change the electric field noise by seven orders of magnitude and specifically that soft in-plane modes of weakly-adsorbed hydrocarbons produce the greatest noise and ion heating. We discuss the dynamics captured by the fluctuating dipole model, namely the adsorbate-dependent turn-on temperature and electric field noise magnitude, and also discuss the model's failure to reproduce the measured 1/ noise frequency scaling with a single adsorbate species. We suggest future research directions for improved, quantitatively predictive models based on extensions of the present framework to multiple interacting adsorbates.

Model ensembles are effective tools for estimating prediction uncertainty in deep learning atomistic force fields. However, their widespread adoption is hindered by high computational costs and overconfident error estimates. In this work, we address these challenges by leveraging distributions of per-sample errors obtained during training and employing a distance-based similarity search in the model latent space. Our method, which we call LTAU (Loss Trajectory Analysis for Uncertainty), efficiently estimates the full probability distribution function of errors for any test point using the logged training errors, achieving speeds that are 2–3 orders of magnitudes faster than typical ensemble methods and allowing it to be used for tasks where training or evaluating multiple models would be infeasible. We apply LTAU towards estimating parametric uncertainty in atomistic force fields ( LTAU-FF ), demonstrating that it produces well-calibrated confidence intervals and predicts errors that correlate strongly with the true errors for data near the training domain. Furthermore, we show that the errors predicted by LTAU-FF can be used in practical applications for detecting out-of-domain data, tuning model performance, and predicting failure during simulations. We believe that LTAU will be a valuable tool for uncertainty quantification in atomistic force fields and is a promising method that should be further explored in other domains of machine learning.

Constructing a chemically diverse dataset while avoiding sampling bias is critical to training efficient and generalizable force fields. However, in computational chemistry and materials science, many common dataset generation techniques are prone to oversampling regions of the potential energy surface. Furthermore, these regions can be difficult to identify and isolate from each other or may not align well with human intuition, making it challenging to systematically remove bias in the dataset. While traditional clustering and pruning (down-sampling) approaches can be useful for this, they can often lead to information loss or a failure to properly identify distinct regions of the potential energy surface due to difficulties associated with the high dimensionality of atomic descriptors. In this work, we introduce the Multi-kernel Edge Attention-based Graph Autoencoder (MEAGraph) model, an unsupervised approach for analyzing atomic datasets. MEAGraph combines multiple linear kernel transformations with attention-based message passing to capture geometric sensitivity and enable effective dataset pruning without relying on labels or extensive training. Demonstrated applications on niobium, tantalum, and iron datasets show that MEAGraph efficiently groups similar atomic environments, allowing for the use of basic pruning techniques for removing sampling bias. This approach provides an effective method for representation learning and clustering that can be used for data analysis, outlier detection, and dataset optimization.

A simulation can stand its ground against an experiment only if its prediction uncertainty is known. The unknown accuracy of interatomic potentials (IPs) is a major source of prediction uncertainty, severely limiting the use of large-scale classical atomistic simulations in a wide range of scientific and engineering applications. Here we explore covariance between predictions of metal plasticity, from 178 large-scale (~10 8 atoms) molecular dynamics (MD) simulations, and a variety of indicator properties computed at small-scales (≤10 2 atoms). All simulations use the same 178 IPs. In a manner similar to statistical studies in public health, we analyze correlations of strength with indicators, identify the best predictor properties, and build a cross-scale “strength-on-predictors” regression model. This model is then used to estimate regression error over the statistical pool of IPs. Small-scale predictors found to be highly covariant with strength are computed using expensive quantum-accurate calculations and used to predict flow strength, within the statistical error bounds established in our study.

Nanoscale composites have been a technological dream for many years. Recently, increased interest has arisen in using carbon nanotubes as a filler for polymer composites, owing to their very small diameters on the order of 1 nm, very high aspect ratios of 1000 or more, and exceptional strength with Young's modulus of approximately 1 TPa. A key issue for realizing these composites is obtaining good interfacial adhesion between the phases. In this work, we used force-field based molecular mechanics calculations to determine binding energies and sliding frictional stresses between pristine carbon nanotubes and a range of polymer substrates, in an effort to understand the factors governing interfacial adhesion. The particular polymers studied were chosen to correspond to reported composites in the literature. We also examined polymer morphologies by performing energy-minimizations in a vacuum. Hydrogen bond interactions with the ∏-bond network of pristine carbon nanotubes were found to bond most strongly to the surface, in the absence of chemically altered nanotubes. Surprisingly, we found that binding energies and frictional forces play only a minor role in determining the strength of the interface, but that helical polymer conformations are essential.