A generalizable machine learning potential of Ag-Au nanoalloys and its application t surface reconstruction, segregation and diffusion

Owing to the excellent catalytic properties of Ag-Au binary nanoalloys, nanostructured Ag-Au, such as Ag-Au nanoparticles and nanopillars, has been under intense investigation. To achieve high accuracy in molecular simulations of Ag-Au nanoalloys, the surface properties must be modeled with first-principles precision. In this work, we constructed a generalizable machine learning interatomic potential for Ag-Au nanoalloys based on deep neural networks trained from a database constructed with first-principles calculations. This potential is highlighted by the accurate prediction of Au (111) surface reconstruction and the segregation of Au toward the Ag-Au nanoalloy surface, where the empirical force field failed in both cases. Moreover, regarding the adsorption and diffusion of adatoms on surfaces, the overall performance of our potential is better than the empirical force fields. We stress that the reported surface properties are blind to the potential modeling in the sense that none of the surface configurations is explicitly included in the training database; therefore, the reported potential is expected to have a strong generalization ability to a wide range of properties and to play a key role in investigating nanostructured Ag-Au evolution, where accurate descriptions of free surfaces are necessary.

Accelerated Discovery of CH4 Uptake Capacity MOFs using Bayesian Optimization

High-throughput computational studies for discovery of metal-organic frameworks (MOFs) for separations and storage applications are often limited by the costs of computing thermodynamic quantities, with recent studies reliant ab initio results for a narrow selection of MOFs and empirical force-field methods for larger selections. Here, we conduct a proof-of-concept study using Bayesian optimization on CH4 uptake capacity of hypothetical MOFs for an existing dataset (Wilmer et al, Nature Chem. 2012, 4, 83). We show that less than 0.1% of the database needs to be screened with our Bayesian optimization approach to recover the top candidate MOFs. This opens the possibility of efficient screening of MOF databases using accurate ab-initio calculations for future adsorption studies on a minimal subset of MOFs. Furthermore, Bayesian optimization and the surrogate model presented here can offer interpretable material design insights and our framework will be applicable in the context of other target properties.

Comparison of Methods for Bulk Automated Simulation of Glycosidic Bond Conformations

Six empirical force fields were tested for applicability to calculations for automated carbohydrate database filling. They were probed on eleven disaccharide molecules containing representative structural features from widespread classes of carbohydrates. The accuracy of each method was queried by predictions of nuclear Overhauser effects (NOEs) from conformational ensembles obtained from 50 to 100 ns molecular dynamics (MD) trajectories and their comparison to the published experimental data. Using various ranking schemes, it was concluded that explicit solvent MM3 MD yielded non-inferior NOE accuracy with newer GLYCAM-06, and ultimately PBE0-D3/def2-TZVP (Triple-Zeta Valence Polarized) Density Functional Theory (DFT) simulations. For seven of eleven molecules, at least one empirical force field with explicit solvent outperformed DFT in NOE prediction. The aggregate of characteristics (accuracy, speed, and compatibility) made MM3 dynamics with explicit solvent at 300 K the most favorable method for bulk generation of disaccharide conformation maps for massive database filling.

C13 – a new empirical force field to characterize the mechanical behavior of carbyne chains

The C13 empirical potential is developed for accurate modeling of mechanical properties of carbyne specifically taking in account bond alternation.

Semi-Empirical Force-Field Model for the Ti1−xAlxN (0 ≤ x ≤ 1) System

We present a modified embedded atom method (MEAM) semi-empirical force-field model for the Ti1−xAlxN (0 ≤ x ≤ 1) alloy system. The MEAM parameters, determined via an adaptive simulated-annealing (ASA) minimization scheme, optimize the model’s predictions with respect to 0 K equilibrium volumes, elastic constants, cohesive energies, enthalpies of mixing, and point-defect formation energies, for a set of ≈40 elemental, binary, and ternary Ti-Al-N structures and configurations. Subsequently, the reliability of the model is thoroughly verified against known finite-temperature thermodynamic and kinetic properties of key binary Ti-N and Al-N phases, as well as properties of Ti1−xAlxN (0 < x < 1) alloys. The successful outcome of the validation underscores the transferability of our model, opening the way for large-scale molecular dynamics simulations of, e.g., phase evolution, interfacial processes, and mechanical response in Ti-Al-N-based alloys, superlattices, and nanostructures.

Solvation dynamics: improved reproduction of the time-dependent Stokes shift with polarizable empirical force field chromophore models

The inclusion of polarizability slows down the computed solvation dynamics due to interactions of induced dipoles, improving agreement to experiment.